fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98
  99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 100                                      struct inode *inode,  struct inode *dir,
 101                                      const struct qstr *qstr)
 102 {
 103         int err;
 104
 105         err = btrfs_init_acl(trans, inode, dir);
 106         if (!err)
 107                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 108         return err;
 109 }
 110
 111 /*
 112  * this does all the hard work for inserting an inline extent into
 113  * the btree.  The caller should have done a btrfs_drop_extents so that
 114  * no overlapping inline items exist in the btree
 115  */
 116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 117                                 struct btrfs_root *root, struct inode *inode,
 118                                 u64 start, size_t size, size_t compressed_size,
 119                                 int compress_type,
 120                                 struct page **compressed_pages)
 121 {
 122         struct btrfs_key key;
 123         struct btrfs_path *path;
 124         struct extent_buffer *leaf;
 125         struct page *page = NULL;
 126         char *kaddr;
 127         unsigned long ptr;
 128         struct btrfs_file_extent_item *ei;
 129         int err = 0;
 130         int ret;
 131         size_t cur_size = size;
 132         size_t datasize;
 133         unsigned long offset;
 134
 135         if (compressed_size && compressed_pages)
 136                 cur_size = compressed_size;
 137
 138         path = btrfs_alloc_path();
 139         if (!path)
 140                 return -ENOMEM;
 141
 142         path->leave_spinning = 1;
 143
 144         key.objectid = btrfs_ino(inode);
 145         key.offset = start;
 146         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 147         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 148
 149         inode_add_bytes(inode, size);
 150         ret = btrfs_insert_empty_item(trans, root, path, &key,
 151                                       datasize);
 152         if (ret) {
 153                 err = ret;
 154                 goto fail;
 155         }
 156         leaf = path->nodes[0];
 157         ei = btrfs_item_ptr(leaf, path->slots[0],
 158                             struct btrfs_file_extent_item);
 159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 160         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 161         btrfs_set_file_extent_encryption(leaf, ei, 0);
 162         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 163         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 164         ptr = btrfs_file_extent_inline_start(ei);
 165
 166         if (compress_type != BTRFS_COMPRESS_NONE) {
 167                 struct page *cpage;
 168                 int i = 0;
 169                 while (compressed_size > 0) {
 170                         cpage = compressed_pages[i];
 171                         cur_size = min_t(unsigned long, compressed_size,
 172                                        PAGE_CACHE_SIZE);
 173
 174                         kaddr = kmap_atomic(cpage);
 175                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 176                         kunmap_atomic(kaddr);
 177
 178                         i++;
 179                         ptr += cur_size;
 180                         compressed_size -= cur_size;
 181                 }
 182                 btrfs_set_file_extent_compression(leaf, ei,
 183                                                   compress_type);
 184         } else {
 185                 page = find_get_page(inode->i_mapping,
 186                                      start >> PAGE_CACHE_SHIFT);
 187                 btrfs_set_file_extent_compression(leaf, ei, 0);
 188                 kaddr = kmap_atomic(page);
 189                 offset = start & (PAGE_CACHE_SIZE - 1);
 190                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 191                 kunmap_atomic(kaddr);
 192                 page_cache_release(page);
 193         }
 194         btrfs_mark_buffer_dirty(leaf);
 195         btrfs_free_path(path);
 196
 197         /*
 198          * we're an inline extent, so nobody can
 199          * extend the file past i_size without locking
 200          * a page we already have locked.
 201          *
 202          * We must do any isize and inode updates
 203          * before we unlock the pages.  Otherwise we
 204          * could end up racing with unlink.
 205          */
 206         BTRFS_I(inode)->disk_i_size = inode->i_size;
 207         ret = btrfs_update_inode(trans, root, inode);
 208
 209         return ret;
 210 fail:
 211         btrfs_free_path(path);
 212         return err;
 213 }
 214
 215
 216 /*
 217  * conditionally insert an inline extent into the file.  This
 218  * does the checks required to make sure the data is small enough
 219  * to fit as an inline extent.
 220  */
 221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 222                                  struct btrfs_root *root,
 223                                  struct inode *inode, u64 start, u64 end,
 224                                  size_t compressed_size, int compress_type,
 225                                  struct page **compressed_pages)
 226 {
 227         u64 isize = i_size_read(inode);
 228         u64 actual_end = min(end + 1, isize);
 229         u64 inline_len = actual_end - start;
 230         u64 aligned_end = (end + root->sectorsize - 1) &
 231                         ~((u64)root->sectorsize - 1);
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 249         if (ret)
 250                 return ret;
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         if (ret && ret != -ENOSPC) {
 258                 btrfs_abort_transaction(trans, root, ret);
 259                 return ret;
 260         } else if (ret == -ENOSPC) {
 261                 return 1;
 262         }
 263
 264         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 265         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 266         return 0;
 267 }
 268
 269 struct async_extent {
 270         u64 start;
 271         u64 ram_size;
 272         u64 compressed_size;
 273         struct page **pages;
 274         unsigned long nr_pages;
 275         int compress_type;
 276         struct list_head list;
 277 };
 278
 279 struct async_cow {
 280         struct inode *inode;
 281         struct btrfs_root *root;
 282         struct page *locked_page;
 283         u64 start;
 284         u64 end;
 285         struct list_head extents;
 286         struct btrfs_work work;
 287 };
 288
 289 static noinline int add_async_extent(struct async_cow *cow,
 290                                      u64 start, u64 ram_size,
 291                                      u64 compressed_size,
 292                                      struct page **pages,
 293                                      unsigned long nr_pages,
 294                                      int compress_type)
 295 {
 296         struct async_extent *async_extent;
 297
 298         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 299         BUG_ON(!async_extent); /* -ENOMEM */
 300         async_extent->start = start;
 301         async_extent->ram_size = ram_size;
 302         async_extent->compressed_size = compressed_size;
 303         async_extent->pages = pages;
 304         async_extent->nr_pages = nr_pages;
 305         async_extent->compress_type = compress_type;
 306         list_add_tail(&async_extent->list, &cow->extents);
 307         return 0;
 308 }
 309
 310 /*
 311  * we create compressed extents in two phases.  The first
 312  * phase compresses a range of pages that have already been
 313  * locked (both pages and state bits are locked).
 314  *
 315  * This is done inside an ordered work queue, and the compression
 316  * is spread across many cpus.  The actual IO submission is step
 317  * two, and the ordered work queue takes care of making sure that
 318  * happens in the same order things were put onto the queue by
 319  * writepages and friends.
 320  *
 321  * If this code finds it can't get good compression, it puts an
 322  * entry onto the work queue to write the uncompressed bytes.  This
 323  * makes sure that both compressed inodes and uncompressed inodes
 324  * are written in the same order that the flusher thread sent them
 325  * down.
 326  */
 327 static noinline int compress_file_range(struct inode *inode,
 328                                         struct page *locked_page,
 329                                         u64 start, u64 end,
 330                                         struct async_cow *async_cow,
 331                                         int *num_added)
 332 {
 333         struct btrfs_root *root = BTRFS_I(inode)->root;
 334         struct btrfs_trans_handle *trans;
 335         u64 num_bytes;
 336         u64 blocksize = root->sectorsize;
 337         u64 actual_end;
 338         u64 isize = i_size_read(inode);
 339         int ret = 0;
 340         struct page **pages = NULL;
 341         unsigned long nr_pages;
 342         unsigned long nr_pages_ret = 0;
 343         unsigned long total_compressed = 0;
 344         unsigned long total_in = 0;
 345         unsigned long max_compressed = 128 * 1024;
 346         unsigned long max_uncompressed = 128 * 1024;
 347         int i;
 348         int will_compress;
 349         int compress_type = root->fs_info->compress_type;
 350
 351         /* if this is a small write inside eof, kick off a defrag */
 352         if ((end - start + 1) < 16 * 1024 &&
 353             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 354                 btrfs_add_inode_defrag(NULL, inode);
 355
 356         actual_end = min_t(u64, isize, end + 1);
 357 again:
 358         will_compress = 0;
 359         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 360         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 361
 362         /*
 363          * we don't want to send crud past the end of i_size through
 364          * compression, that's just a waste of CPU time.  So, if the
 365          * end of the file is before the start of our current
 366          * requested range of bytes, we bail out to the uncompressed
 367          * cleanup code that can deal with all of this.
 368          *
 369          * It isn't really the fastest way to fix things, but this is a
 370          * very uncommon corner.
 371          */
 372         if (actual_end <= start)
 373                 goto cleanup_and_bail_uncompressed;
 374
 375         total_compressed = actual_end - start;
 376
 377         /* we want to make sure that amount of ram required to uncompress
 378          * an extent is reasonable, so we limit the total size in ram
 379          * of a compressed extent to 128k.  This is a crucial number
 380          * because it also controls how easily we can spread reads across
 381          * cpus for decompression.
 382          *
 383          * We also want to make sure the amount of IO required to do
 384          * a random read is reasonably small, so we limit the size of
 385          * a compressed extent to 128k.
 386          */
 387         total_compressed = min(total_compressed, max_uncompressed);
 388         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 389         num_bytes = max(blocksize,  num_bytes);
 390         total_in = 0;
 391         ret = 0;
 392
 393         /*
 394          * we do compression for mount -o compress and when the
 395          * inode has not been flagged as nocompress.  This flag can
 396          * change at any time if we discover bad compression ratios.
 397          */
 398         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 399             (btrfs_test_opt(root, COMPRESS) ||
 400              (BTRFS_I(inode)->force_compress) ||
 401              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 402                 WARN_ON(pages);
 403                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 404                 if (!pages) {
 405                         /* just bail out to the uncompressed code */
 406                         goto cont;
 407                 }
 408
 409                 if (BTRFS_I(inode)->force_compress)
 410                         compress_type = BTRFS_I(inode)->force_compress;
 411
 412                 ret = btrfs_compress_pages(compress_type,
 413                                            inode->i_mapping, start,
 414                                            total_compressed, pages,
 415                                            nr_pages, &nr_pages_ret,
 416                                            &total_in,
 417                                            &total_compressed,
 418                                            max_compressed);
 419
 420                 if (!ret) {
 421                         unsigned long offset = total_compressed &
 422                                 (PAGE_CACHE_SIZE - 1);
 423                         struct page *page = pages[nr_pages_ret - 1];
 424                         char *kaddr;
 425
 426                         /* zero the tail end of the last page, we might be
 427                          * sending it down to disk
 428                          */
 429                         if (offset) {
 430                                 kaddr = kmap_atomic(page);
 431                                 memset(kaddr + offset, 0,
 432                                        PAGE_CACHE_SIZE - offset);
 433                                 kunmap_atomic(kaddr);
 434                         }
 435                         will_compress = 1;
 436                 }
 437         }
 438 cont:
 439         if (start == 0) {
 440                 trans = btrfs_join_transaction(root);
 441                 if (IS_ERR(trans)) {
 442                         ret = PTR_ERR(trans);
 443                         trans = NULL;
 444                         goto cleanup_and_out;
 445                 }
 446                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 447
 448                 /* lets try to make an inline extent */
 449                 if (ret || total_in < (actual_end - start)) {
 450                         /* we didn't compress the entire range, try
 451                          * to make an uncompressed inline extent.
 452                          */
 453                         ret = cow_file_range_inline(trans, root, inode,
 454                                                     start, end, 0, 0, NULL);
 455                 } else {
 456                         /* try making a compressed inline extent */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end,
 459                                                     total_compressed,
 460                                                     compress_type, pages);
 461                 }
 462                 if (ret <= 0) {
 463                         /*
 464                          * inline extent creation worked or returned error,
 465                          * we don't need to create any more async work items.
 466                          * Unlock and free up our temp pages.
 467                          */
 468                         extent_clear_unlock_delalloc(inode,
 469                              &BTRFS_I(inode)->io_tree,
 470                              start, end, NULL,
 471                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 472                              EXTENT_CLEAR_DELALLOC |
 473                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 474
 475                         btrfs_end_transaction(trans, root);
 476                         goto free_pages_out;
 477                 }
 478                 btrfs_end_transaction(trans, root);
 479         }
 480
 481         if (will_compress) {
 482                 /*
 483                  * we aren't doing an inline extent round the compressed size
 484                  * up to a block size boundary so the allocator does sane
 485                  * things
 486                  */
 487                 total_compressed = (total_compressed + blocksize - 1) &
 488                         ~(blocksize - 1);
 489
 490                 /*
 491                  * one last check to make sure the compression is really a
 492                  * win, compare the page count read with the blocks on disk
 493                  */
 494                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 495                         ~(PAGE_CACHE_SIZE - 1);
 496                 if (total_compressed >= total_in) {
 497                         will_compress = 0;
 498                 } else {
 499                         num_bytes = total_in;
 500                 }
 501         }
 502         if (!will_compress && pages) {
 503                 /*
 504                  * the compression code ran but failed to make things smaller,
 505                  * free any pages it allocated and our page pointer array
 506                  */
 507                 for (i = 0; i < nr_pages_ret; i++) {
 508                         WARN_ON(pages[i]->mapping);
 509                         page_cache_release(pages[i]);
 510                 }
 511                 kfree(pages);
 512                 pages = NULL;
 513                 total_compressed = 0;
 514                 nr_pages_ret = 0;
 515
 516                 /* flag the file so we don't compress in the future */
 517                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 518                     !(BTRFS_I(inode)->force_compress)) {
 519                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 520                 }
 521         }
 522         if (will_compress) {
 523                 *num_added += 1;
 524
 525                 /* the async work queues will take care of doing actual
 526                  * allocation on disk for these compressed pages,
 527                  * and will submit them to the elevator.
 528                  */
 529                 add_async_extent(async_cow, start, num_bytes,
 530                                  total_compressed, pages, nr_pages_ret,
 531                                  compress_type);
 532
 533                 if (start + num_bytes < end) {
 534                         start += num_bytes;
 535                         pages = NULL;
 536                         cond_resched();
 537                         goto again;
 538                 }
 539         } else {
 540 cleanup_and_bail_uncompressed:
 541                 /*
 542                  * No compression, but we still need to write the pages in
 543                  * the file we've been given so far.  redirty the locked
 544                  * page if it corresponds to our extent and set things up
 545                  * for the async work queue to run cow_file_range to do
 546                  * the normal delalloc dance
 547                  */
 548                 if (page_offset(locked_page) >= start &&
 549                     page_offset(locked_page) <= end) {
 550                         __set_page_dirty_nobuffers(locked_page);
 551                         /* unlocked later on in the async handlers */
 552                 }
 553                 add_async_extent(async_cow, start, end - start + 1,
 554                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 555                 *num_added += 1;
 556         }
 557
 558 out:
 559         return ret;
 560
 561 free_pages_out:
 562         for (i = 0; i < nr_pages_ret; i++) {
 563                 WARN_ON(pages[i]->mapping);
 564                 page_cache_release(pages[i]);
 565         }
 566         kfree(pages);
 567
 568         goto out;
 569
 570 cleanup_and_out:
 571         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 572                                      start, end, NULL,
 573                                      EXTENT_CLEAR_UNLOCK_PAGE |
 574                                      EXTENT_CLEAR_DIRTY |
 575                                      EXTENT_CLEAR_DELALLOC |
 576                                      EXTENT_SET_WRITEBACK |
 577                                      EXTENT_END_WRITEBACK);
 578         if (!trans || IS_ERR(trans))
 579                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 580         else
 581                 btrfs_abort_transaction(trans, root, ret);
 582         goto free_pages_out;
 583 }
 584
 585 /*
 586  * phase two of compressed writeback.  This is the ordered portion
 587  * of the code, which only gets called in the order the work was
 588  * queued.  We walk all the async extents created by compress_file_range
 589  * and send them down to the disk.
 590  */
 591 static noinline int submit_compressed_extents(struct inode *inode,
 592                                               struct async_cow *async_cow)
 593 {
 594         struct async_extent *async_extent;
 595         u64 alloc_hint = 0;
 596         struct btrfs_trans_handle *trans;
 597         struct btrfs_key ins;
 598         struct extent_map *em;
 599         struct btrfs_root *root = BTRFS_I(inode)->root;
 600         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 601         struct extent_io_tree *io_tree;
 602         int ret = 0;
 603
 604         if (list_empty(&async_cow->extents))
 605                 return 0;
 606
 607
 608         while (!list_empty(&async_cow->extents)) {
 609                 async_extent = list_entry(async_cow->extents.next,
 610                                           struct async_extent, list);
 611                 list_del(&async_extent->list);
 612
 613                 io_tree = &BTRFS_I(inode)->io_tree;
 614
 615 retry:
 616                 /* did the compression code fall back to uncompressed IO? */
 617                 if (!async_extent->pages) {
 618                         int page_started = 0;
 619                         unsigned long nr_written = 0;
 620
 621                         lock_extent(io_tree, async_extent->start,
 622                                          async_extent->start +
 623                                          async_extent->ram_size - 1);
 624
 625                         /* allocate blocks */
 626                         ret = cow_file_range(inode, async_cow->locked_page,
 627                                              async_extent->start,
 628                                              async_extent->start +
 629                                              async_extent->ram_size - 1,
 630                                              &page_started, &nr_written, 0);
 631
 632                         /* JDM XXX */
 633
 634                         /*
 635                          * if page_started, cow_file_range inserted an
 636                          * inline extent and took care of all the unlocking
 637                          * and IO for us.  Otherwise, we need to submit
 638                          * all those pages down to the drive.
 639                          */
 640                         if (!page_started && !ret)
 641                                 extent_write_locked_range(io_tree,
 642                                                   inode, async_extent->start,
 643                                                   async_extent->start +
 644                                                   async_extent->ram_size - 1,
 645                                                   btrfs_get_extent,
 646                                                   WB_SYNC_ALL);
 647                         kfree(async_extent);
 648                         cond_resched();
 649                         continue;
 650                 }
 651
 652                 lock_extent(io_tree, async_extent->start,
 653                             async_extent->start + async_extent->ram_size - 1);
 654
 655                 trans = btrfs_join_transaction(root);
 656                 if (IS_ERR(trans)) {
 657                         ret = PTR_ERR(trans);
 658                 } else {
 659                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 660                         ret = btrfs_reserve_extent(trans, root,
 661                                            async_extent->compressed_size,
 662                                            async_extent->compressed_size,
 663                                            0, alloc_hint, &ins, 1);
 664                         if (ret && ret != -ENOSPC)
 665                                 btrfs_abort_transaction(trans, root, ret);
 666                         btrfs_end_transaction(trans, root);
 667                 }
 668
 669                 if (ret) {
 670                         int i;
 671                         for (i = 0; i < async_extent->nr_pages; i++) {
 672                                 WARN_ON(async_extent->pages[i]->mapping);
 673                                 page_cache_release(async_extent->pages[i]);
 674                         }
 675                         kfree(async_extent->pages);
 676                         async_extent->nr_pages = 0;
 677                         async_extent->pages = NULL;
 678                         unlock_extent(io_tree, async_extent->start,
 679                                       async_extent->start +
 680                                       async_extent->ram_size - 1);
 681                         if (ret == -ENOSPC)
 682                                 goto retry;
 683                         goto out_free; /* JDM: Requeue? */
 684                 }
 685
 686                 /*
 687                  * here we're doing allocation and writeback of the
 688                  * compressed pages
 689                  */
 690                 btrfs_drop_extent_cache(inode, async_extent->start,
 691                                         async_extent->start +
 692                                         async_extent->ram_size - 1, 0);
 693
 694                 em = alloc_extent_map();
 695                 BUG_ON(!em); /* -ENOMEM */
 696                 em->start = async_extent->start;
 697                 em->len = async_extent->ram_size;
 698                 em->orig_start = em->start;
 699
 700                 em->block_start = ins.objectid;
 701                 em->block_len = ins.offset;
 702                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 703                 em->compress_type = async_extent->compress_type;
 704                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 705                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 706
 707                 while (1) {
 708                         write_lock(&em_tree->lock);
 709                         ret = add_extent_mapping(em_tree, em);
 710                         write_unlock(&em_tree->lock);
 711                         if (ret != -EEXIST) {
 712                                 free_extent_map(em);
 713                                 break;
 714                         }
 715                         btrfs_drop_extent_cache(inode, async_extent->start,
 716                                                 async_extent->start +
 717                                                 async_extent->ram_size - 1, 0);
 718                 }
 719
 720                 ret = btrfs_add_ordered_extent_compress(inode,
 721                                                 async_extent->start,
 722                                                 ins.objectid,
 723                                                 async_extent->ram_size,
 724                                                 ins.offset,
 725                                                 BTRFS_ORDERED_COMPRESSED,
 726                                                 async_extent->compress_type);
 727                 BUG_ON(ret); /* -ENOMEM */
 728
 729                 /*
 730                  * clear dirty, set writeback and unlock the pages.
 731                  */
 732                 extent_clear_unlock_delalloc(inode,
 733                                 &BTRFS_I(inode)->io_tree,
 734                                 async_extent->start,
 735                                 async_extent->start +
 736                                 async_extent->ram_size - 1,
 737                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 738                                 EXTENT_CLEAR_UNLOCK |
 739                                 EXTENT_CLEAR_DELALLOC |
 740                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 741
 742                 ret = btrfs_submit_compressed_write(inode,
 743                                     async_extent->start,
 744                                     async_extent->ram_size,
 745                                     ins.objectid,
 746                                     ins.offset, async_extent->pages,
 747                                     async_extent->nr_pages);
 748
 749                 BUG_ON(ret); /* -ENOMEM */
 750                 alloc_hint = ins.objectid + ins.offset;
 751                 kfree(async_extent);
 752                 cond_resched();
 753         }
 754         ret = 0;
 755 out:
 756         return ret;
 757 out_free:
 758         kfree(async_extent);
 759         goto out;
 760 }
 761
 762 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 763                                       u64 num_bytes)
 764 {
 765         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 766         struct extent_map *em;
 767         u64 alloc_hint = 0;
 768
 769         read_lock(&em_tree->lock);
 770         em = search_extent_mapping(em_tree, start, num_bytes);
 771         if (em) {
 772                 /*
 773                  * if block start isn't an actual block number then find the
 774                  * first block in this inode and use that as a hint.  If that
 775                  * block is also bogus then just don't worry about it.
 776                  */
 777                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 778                         free_extent_map(em);
 779                         em = search_extent_mapping(em_tree, 0, 0);
 780                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 781                                 alloc_hint = em->block_start;
 782                         if (em)
 783                                 free_extent_map(em);
 784                 } else {
 785                         alloc_hint = em->block_start;
 786                         free_extent_map(em);
 787                 }
 788         }
 789         read_unlock(&em_tree->lock);
 790
 791         return alloc_hint;
 792 }
 793
 794 /*
 795  * when extent_io.c finds a delayed allocation range in the file,
 796  * the call backs end up in this code.  The basic idea is to
 797  * allocate extents on disk for the range, and create ordered data structs
 798  * in ram to track those extents.
 799  *
 800  * locked_page is the page that writepage had locked already.  We use
 801  * it to make sure we don't do extra locks or unlocks.
 802  *
 803  * *page_started is set to one if we unlock locked_page and do everything
 804  * required to start IO on it.  It may be clean and already done with
 805  * IO when we return.
 806  */
 807 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 808                                      struct inode *inode,
 809                                      struct btrfs_root *root,
 810                                      struct page *locked_page,
 811                                      u64 start, u64 end, int *page_started,
 812                                      unsigned long *nr_written,
 813                                      int unlock)
 814 {
 815         u64 alloc_hint = 0;
 816         u64 num_bytes;
 817         unsigned long ram_size;
 818         u64 disk_num_bytes;
 819         u64 cur_alloc_size;
 820         u64 blocksize = root->sectorsize;
 821         struct btrfs_key ins;
 822         struct extent_map *em;
 823         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 824         int ret = 0;
 825
 826         BUG_ON(btrfs_is_free_space_inode(inode));
 827
 828         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 829         num_bytes = max(blocksize,  num_bytes);
 830         disk_num_bytes = num_bytes;
 831
 832         /* if this is a small write inside eof, kick off defrag */
 833         if (num_bytes < 64 * 1024 &&
 834             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 835                 btrfs_add_inode_defrag(trans, inode);
 836
 837         if (start == 0) {
 838                 /* lets try to make an inline extent */
 839                 ret = cow_file_range_inline(trans, root, inode,
 840                                             start, end, 0, 0, NULL);
 841                 if (ret == 0) {
 842                         extent_clear_unlock_delalloc(inode,
 843                                      &BTRFS_I(inode)->io_tree,
 844                                      start, end, NULL,
 845                                      EXTENT_CLEAR_UNLOCK_PAGE |
 846                                      EXTENT_CLEAR_UNLOCK |
 847                                      EXTENT_CLEAR_DELALLOC |
 848                                      EXTENT_CLEAR_DIRTY |
 849                                      EXTENT_SET_WRITEBACK |
 850                                      EXTENT_END_WRITEBACK);
 851
 852                         *nr_written = *nr_written +
 853                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 854                         *page_started = 1;
 855                         goto out;
 856                 } else if (ret < 0) {
 857                         btrfs_abort_transaction(trans, root, ret);
 858                         goto out_unlock;
 859                 }
 860         }
 861
 862         BUG_ON(disk_num_bytes >
 863                btrfs_super_total_bytes(root->fs_info->super_copy));
 864
 865         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 866         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 867
 868         while (disk_num_bytes > 0) {
 869                 unsigned long op;
 870
 871                 cur_alloc_size = disk_num_bytes;
 872                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 873                                            root->sectorsize, 0, alloc_hint,
 874                                            &ins, 1);
 875                 if (ret < 0) {
 876                         btrfs_abort_transaction(trans, root, ret);
 877                         goto out_unlock;
 878                 }
 879
 880                 em = alloc_extent_map();
 881                 BUG_ON(!em); /* -ENOMEM */
 882                 em->start = start;
 883                 em->orig_start = em->start;
 884                 ram_size = ins.offset;
 885                 em->len = ins.offset;
 886
 887                 em->block_start = ins.objectid;
 888                 em->block_len = ins.offset;
 889                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 890                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 891
 892                 while (1) {
 893                         write_lock(&em_tree->lock);
 894                         ret = add_extent_mapping(em_tree, em);
 895                         write_unlock(&em_tree->lock);
 896                         if (ret != -EEXIST) {
 897                                 free_extent_map(em);
 898                                 break;
 899                         }
 900                         btrfs_drop_extent_cache(inode, start,
 901                                                 start + ram_size - 1, 0);
 902                 }
 903
 904                 cur_alloc_size = ins.offset;
 905                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 906                                                ram_size, cur_alloc_size, 0);
 907                 BUG_ON(ret); /* -ENOMEM */
 908
 909                 if (root->root_key.objectid ==
 910                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 911                         ret = btrfs_reloc_clone_csums(inode, start,
 912                                                       cur_alloc_size);
 913                         if (ret) {
 914                                 btrfs_abort_transaction(trans, root, ret);
 915                                 goto out_unlock;
 916                         }
 917                 }
 918
 919                 if (disk_num_bytes < cur_alloc_size)
 920                         break;
 921
 922                 /* we're not doing compressed IO, don't unlock the first
 923                  * page (which the caller expects to stay locked), don't
 924                  * clear any dirty bits and don't set any writeback bits
 925                  *
 926                  * Do set the Private2 bit so we know this page was properly
 927                  * setup for writepage
 928                  */
 929                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 930                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 931                         EXTENT_SET_PRIVATE2;
 932
 933                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 934                                              start, start + ram_size - 1,
 935                                              locked_page, op);
 936                 disk_num_bytes -= cur_alloc_size;
 937                 num_bytes -= cur_alloc_size;
 938                 alloc_hint = ins.objectid + ins.offset;
 939                 start += cur_alloc_size;
 940         }
 941 out:
 942         return ret;
 943
 944 out_unlock:
 945         extent_clear_unlock_delalloc(inode,
 946                      &BTRFS_I(inode)->io_tree,
 947                      start, end, locked_page,
 948                      EXTENT_CLEAR_UNLOCK_PAGE |
 949                      EXTENT_CLEAR_UNLOCK |
 950                      EXTENT_CLEAR_DELALLOC |
 951                      EXTENT_CLEAR_DIRTY |
 952                      EXTENT_SET_WRITEBACK |
 953                      EXTENT_END_WRITEBACK);
 954
 955         goto out;
 956 }
 957
 958 static noinline int cow_file_range(struct inode *inode,
 959                                    struct page *locked_page,
 960                                    u64 start, u64 end, int *page_started,
 961                                    unsigned long *nr_written,
 962                                    int unlock)
 963 {
 964         struct btrfs_trans_handle *trans;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         int ret;
 967
 968         trans = btrfs_join_transaction(root);
 969         if (IS_ERR(trans)) {
 970                 extent_clear_unlock_delalloc(inode,
 971                              &BTRFS_I(inode)->io_tree,
 972                              start, end, locked_page,
 973                              EXTENT_CLEAR_UNLOCK_PAGE |
 974                              EXTENT_CLEAR_UNLOCK |
 975                              EXTENT_CLEAR_DELALLOC |
 976                              EXTENT_CLEAR_DIRTY |
 977                              EXTENT_SET_WRITEBACK |
 978                              EXTENT_END_WRITEBACK);
 979                 return PTR_ERR(trans);
 980         }
 981         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 982
 983         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
 984                                page_started, nr_written, unlock);
 985
 986         btrfs_end_transaction(trans, root);
 987
 988         return ret;
 989 }
 990
 991 /*
 992  * work queue call back to started compression on a file and pages
 993  */
 994 static noinline void async_cow_start(struct btrfs_work *work)
 995 {
 996         struct async_cow *async_cow;
 997         int num_added = 0;
 998         async_cow = container_of(work, struct async_cow, work);
 999
1000         compress_file_range(async_cow->inode, async_cow->locked_page,
1001                             async_cow->start, async_cow->end, async_cow,
1002                             &num_added);
1003         if (num_added == 0) {
1004                 btrfs_add_delayed_iput(async_cow->inode);
1005                 async_cow->inode = NULL;
1006         }
1007 }
1008
1009 /*
1010  * work queue call back to submit previously compressed pages
1011  */
1012 static noinline void async_cow_submit(struct btrfs_work *work)
1013 {
1014         struct async_cow *async_cow;
1015         struct btrfs_root *root;
1016         unsigned long nr_pages;
1017
1018         async_cow = container_of(work, struct async_cow, work);
1019
1020         root = async_cow->root;
1021         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1022                 PAGE_CACHE_SHIFT;
1023
1024         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1025             5 * 1024 * 1024 &&
1026             waitqueue_active(&root->fs_info->async_submit_wait))
1027                 wake_up(&root->fs_info->async_submit_wait);
1028
1029         if (async_cow->inode)
1030                 submit_compressed_extents(async_cow->inode, async_cow);
1031 }
1032
1033 static noinline void async_cow_free(struct btrfs_work *work)
1034 {
1035         struct async_cow *async_cow;
1036         async_cow = container_of(work, struct async_cow, work);
1037         if (async_cow->inode)
1038                 btrfs_add_delayed_iput(async_cow->inode);
1039         kfree(async_cow);
1040 }
1041
1042 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043                                 u64 start, u64 end, int *page_started,
1044                                 unsigned long *nr_written)
1045 {
1046         struct async_cow *async_cow;
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         unsigned long nr_pages;
1049         u64 cur_end;
1050         int limit = 10 * 1024 * 1024;
1051
1052         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053                          1, 0, NULL, GFP_NOFS);
1054         while (start < end) {
1055                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1056                 BUG_ON(!async_cow); /* -ENOMEM */
1057                 async_cow->inode = igrab(inode);
1058                 async_cow->root = root;
1059                 async_cow->locked_page = locked_page;
1060                 async_cow->start = start;
1061
1062                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1063                         cur_end = end;
1064                 else
1065                         cur_end = min(end, start + 512 * 1024 - 1);
1066
1067                 async_cow->end = cur_end;
1068                 INIT_LIST_HEAD(&async_cow->extents);
1069
1070                 async_cow->work.func = async_cow_start;
1071                 async_cow->work.ordered_func = async_cow_submit;
1072                 async_cow->work.ordered_free = async_cow_free;
1073                 async_cow->work.flags = 0;
1074
1075                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1076                         PAGE_CACHE_SHIFT;
1077                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1078
1079                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1080                                    &async_cow->work);
1081
1082                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083                         wait_event(root->fs_info->async_submit_wait,
1084                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1085                             limit));
1086                 }
1087
1088                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1089                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1090                         wait_event(root->fs_info->async_submit_wait,
1091                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1092                            0));
1093                 }
1094
1095                 *nr_written += nr_pages;
1096                 start = cur_end + 1;
1097         }
1098         *page_started = 1;
1099         return 0;
1100 }
1101
1102 static noinline int csum_exist_in_range(struct btrfs_root *root,
1103                                         u64 bytenr, u64 num_bytes)
1104 {
1105         int ret;
1106         struct btrfs_ordered_sum *sums;
1107         LIST_HEAD(list);
1108
1109         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1110                                        bytenr + num_bytes - 1, &list, 0);
1111         if (ret == 0 && list_empty(&list))
1112                 return 0;
1113
1114         while (!list_empty(&list)) {
1115                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116                 list_del(&sums->list);
1117                 kfree(sums);
1118         }
1119         return 1;
1120 }
1121
1122 /*
1123  * when nowcow writeback call back.  This checks for snapshots or COW copies
1124  * of the extents that exist in the file, and COWs the file as required.
1125  *
1126  * If no cow copies or snapshots exist, we write directly to the existing
1127  * blocks on disk
1128  */
1129 static noinline int run_delalloc_nocow(struct inode *inode,
1130                                        struct page *locked_page,
1131                               u64 start, u64 end, int *page_started, int force,
1132                               unsigned long *nr_written)
1133 {
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135         struct btrfs_trans_handle *trans;
1136         struct extent_buffer *leaf;
1137         struct btrfs_path *path;
1138         struct btrfs_file_extent_item *fi;
1139         struct btrfs_key found_key;
1140         u64 cow_start;
1141         u64 cur_offset;
1142         u64 extent_end;
1143         u64 extent_offset;
1144         u64 disk_bytenr;
1145         u64 num_bytes;
1146         int extent_type;
1147         int ret, err;
1148         int type;
1149         int nocow;
1150         int check_prev = 1;
1151         bool nolock;
1152         u64 ino = btrfs_ino(inode);
1153
1154         path = btrfs_alloc_path();
1155         if (!path) {
1156                 extent_clear_unlock_delalloc(inode,
1157                              &BTRFS_I(inode)->io_tree,
1158                              start, end, locked_page,
1159                              EXTENT_CLEAR_UNLOCK_PAGE |
1160                              EXTENT_CLEAR_UNLOCK |
1161                              EXTENT_CLEAR_DELALLOC |
1162                              EXTENT_CLEAR_DIRTY |
1163                              EXTENT_SET_WRITEBACK |
1164                              EXTENT_END_WRITEBACK);
1165                 return -ENOMEM;
1166         }
1167
1168         nolock = btrfs_is_free_space_inode(inode);
1169
1170         if (nolock)
1171                 trans = btrfs_join_transaction_nolock(root);
1172         else
1173                 trans = btrfs_join_transaction(root);
1174
1175         if (IS_ERR(trans)) {
1176                 extent_clear_unlock_delalloc(inode,
1177                              &BTRFS_I(inode)->io_tree,
1178                              start, end, locked_page,
1179                              EXTENT_CLEAR_UNLOCK_PAGE |
1180                              EXTENT_CLEAR_UNLOCK |
1181                              EXTENT_CLEAR_DELALLOC |
1182                              EXTENT_CLEAR_DIRTY |
1183                              EXTENT_SET_WRITEBACK |
1184                              EXTENT_END_WRITEBACK);
1185                 btrfs_free_path(path);
1186                 return PTR_ERR(trans);
1187         }
1188
1189         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1190
1191         cow_start = (u64)-1;
1192         cur_offset = start;
1193         while (1) {
1194                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1195                                                cur_offset, 0);
1196                 if (ret < 0) {
1197                         btrfs_abort_transaction(trans, root, ret);
1198                         goto error;
1199                 }
1200                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201                         leaf = path->nodes[0];
1202                         btrfs_item_key_to_cpu(leaf, &found_key,
1203                                               path->slots[0] - 1);
1204                         if (found_key.objectid == ino &&
1205                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1206                                 path->slots[0]--;
1207                 }
1208                 check_prev = 0;
1209 next_slot:
1210                 leaf = path->nodes[0];
1211                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212                         ret = btrfs_next_leaf(root, path);
1213                         if (ret < 0) {
1214                                 btrfs_abort_transaction(trans, root, ret);
1215                                 goto error;
1216                         }
1217                         if (ret > 0)
1218                                 break;
1219                         leaf = path->nodes[0];
1220                 }
1221
1222                 nocow = 0;
1223                 disk_bytenr = 0;
1224                 num_bytes = 0;
1225                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1226
1227                 if (found_key.objectid > ino ||
1228                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229                     found_key.offset > end)
1230                         break;
1231
1232                 if (found_key.offset > cur_offset) {
1233                         extent_end = found_key.offset;
1234                         extent_type = 0;
1235                         goto out_check;
1236                 }
1237
1238                 fi = btrfs_item_ptr(leaf, path->slots[0],
1239                                     struct btrfs_file_extent_item);
1240                 extent_type = btrfs_file_extent_type(leaf, fi);
1241
1242                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1244                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1245                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1246                         extent_end = found_key.offset +
1247                                 btrfs_file_extent_num_bytes(leaf, fi);
1248                         if (extent_end <= start) {
1249                                 path->slots[0]++;
1250                                 goto next_slot;
1251                         }
1252                         if (disk_bytenr == 0)
1253                                 goto out_check;
1254                         if (btrfs_file_extent_compression(leaf, fi) ||
1255                             btrfs_file_extent_encryption(leaf, fi) ||
1256                             btrfs_file_extent_other_encoding(leaf, fi))
1257                                 goto out_check;
1258                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1259                                 goto out_check;
1260                         if (btrfs_extent_readonly(root, disk_bytenr))
1261                                 goto out_check;
1262                         if (btrfs_cross_ref_exist(trans, root, ino,
1263                                                   found_key.offset -
1264                                                   extent_offset, disk_bytenr))
1265                                 goto out_check;
1266                         disk_bytenr += extent_offset;
1267                         disk_bytenr += cur_offset - found_key.offset;
1268                         num_bytes = min(end + 1, extent_end) - cur_offset;
1269                         /*
1270                          * force cow if csum exists in the range.
1271                          * this ensure that csum for a given extent are
1272                          * either valid or do not exist.
1273                          */
1274                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1275                                 goto out_check;
1276                         nocow = 1;
1277                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278                         extent_end = found_key.offset +
1279                                 btrfs_file_extent_inline_len(leaf, fi);
1280                         extent_end = ALIGN(extent_end, root->sectorsize);
1281                 } else {
1282                         BUG_ON(1);
1283                 }
1284 out_check:
1285                 if (extent_end <= start) {
1286                         path->slots[0]++;
1287                         goto next_slot;
1288                 }
1289                 if (!nocow) {
1290                         if (cow_start == (u64)-1)
1291                                 cow_start = cur_offset;
1292                         cur_offset = extent_end;
1293                         if (cur_offset > end)
1294                                 break;
1295                         path->slots[0]++;
1296                         goto next_slot;
1297                 }
1298
1299                 btrfs_release_path(path);
1300                 if (cow_start != (u64)-1) {
1301                         ret = __cow_file_range(trans, inode, root, locked_page,
1302                                                cow_start, found_key.offset - 1,
1303                                                page_started, nr_written, 1);
1304                         if (ret) {
1305                                 btrfs_abort_transaction(trans, root, ret);
1306                                 goto error;
1307                         }
1308                         cow_start = (u64)-1;
1309                 }
1310
1311                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312                         struct extent_map *em;
1313                         struct extent_map_tree *em_tree;
1314                         em_tree = &BTRFS_I(inode)->extent_tree;
1315                         em = alloc_extent_map();
1316                         BUG_ON(!em); /* -ENOMEM */
1317                         em->start = cur_offset;
1318                         em->orig_start = em->start;
1319                         em->len = num_bytes;
1320                         em->block_len = num_bytes;
1321                         em->block_start = disk_bytenr;
1322                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1323                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1324                         set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1325                         while (1) {
1326                                 write_lock(&em_tree->lock);
1327                                 ret = add_extent_mapping(em_tree, em);
1328                                 write_unlock(&em_tree->lock);
1329                                 if (ret != -EEXIST) {
1330                                         free_extent_map(em);
1331                                         break;
1332                                 }
1333                                 btrfs_drop_extent_cache(inode, em->start,
1334                                                 em->start + em->len - 1, 0);
1335                         }
1336                         type = BTRFS_ORDERED_PREALLOC;
1337                 } else {
1338                         type = BTRFS_ORDERED_NOCOW;
1339                 }
1340
1341                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1342                                                num_bytes, num_bytes, type);
1343                 BUG_ON(ret); /* -ENOMEM */
1344
1345                 if (root->root_key.objectid ==
1346                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1348                                                       num_bytes);
1349                         if (ret) {
1350                                 btrfs_abort_transaction(trans, root, ret);
1351                                 goto error;
1352                         }
1353                 }
1354
1355                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1356                                 cur_offset, cur_offset + num_bytes - 1,
1357                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359                                 EXTENT_SET_PRIVATE2);
1360                 cur_offset = extent_end;
1361                 if (cur_offset > end)
1362                         break;
1363         }
1364         btrfs_release_path(path);
1365
1366         if (cur_offset <= end && cow_start == (u64)-1) {
1367                 cow_start = cur_offset;
1368                 cur_offset = end;
1369         }
1370
1371         if (cow_start != (u64)-1) {
1372                 ret = __cow_file_range(trans, inode, root, locked_page,
1373                                        cow_start, end,
1374                                        page_started, nr_written, 1);
1375                 if (ret) {
1376                         btrfs_abort_transaction(trans, root, ret);
1377                         goto error;
1378                 }
1379         }
1380
1381 error:
1382         err = btrfs_end_transaction(trans, root);
1383         if (!ret)
1384                 ret = err;
1385
1386         if (ret && cur_offset < end)
1387                 extent_clear_unlock_delalloc(inode,
1388                              &BTRFS_I(inode)->io_tree,
1389                              cur_offset, end, locked_page,
1390                              EXTENT_CLEAR_UNLOCK_PAGE |
1391                              EXTENT_CLEAR_UNLOCK |
1392                              EXTENT_CLEAR_DELALLOC |
1393                              EXTENT_CLEAR_DIRTY |
1394                              EXTENT_SET_WRITEBACK |
1395                              EXTENT_END_WRITEBACK);
1396
1397         btrfs_free_path(path);
1398         return ret;
1399 }
1400
1401 /*
1402  * extent_io.c call back to do delayed allocation processing
1403  */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405                               u64 start, u64 end, int *page_started,
1406                               unsigned long *nr_written)
1407 {
1408         int ret;
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410
1411         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413                                          page_started, 1, nr_written);
1414         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416                                          page_started, 0, nr_written);
1417         } else if (!btrfs_test_opt(root, COMPRESS) &&
1418                    !(BTRFS_I(inode)->force_compress) &&
1419                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420                 ret = cow_file_range(inode, locked_page, start, end,
1421                                       page_started, nr_written, 1);
1422         } else {
1423                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424                         &BTRFS_I(inode)->runtime_flags);
1425                 ret = cow_file_range_async(inode, locked_page, start, end,
1426                                            page_started, nr_written);
1427         }
1428         return ret;
1429 }
1430
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432                                     struct extent_state *orig, u64 split)
1433 {
1434         /* not delalloc, ignore it */
1435         if (!(orig->state & EXTENT_DELALLOC))
1436                 return;
1437
1438         spin_lock(&BTRFS_I(inode)->lock);
1439         BTRFS_I(inode)->outstanding_extents++;
1440         spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 /*
1444  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445  * extents so we can keep track of new extents that are just merged onto old
1446  * extents, such as when we are doing sequential writes, so we can properly
1447  * account for the metadata space we'll need.
1448  */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450                                     struct extent_state *new,
1451                                     struct extent_state *other)
1452 {
1453         /* not delalloc, ignore it */
1454         if (!(other->state & EXTENT_DELALLOC))
1455                 return;
1456
1457         spin_lock(&BTRFS_I(inode)->lock);
1458         BTRFS_I(inode)->outstanding_extents--;
1459         spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461
1462 /*
1463  * extent_io.c set_bit_hook, used to track delayed allocation
1464  * bytes in this file, and to maintain the list of inodes that
1465  * have pending delalloc work to be done.
1466  */
1467 static void btrfs_set_bit_hook(struct inode *inode,
1468                                struct extent_state *state, int *bits)
1469 {
1470
1471         /*
1472          * set_bit and clear bit hooks normally require _irqsave/restore
1473          * but in this case, we are only testing for the DELALLOC
1474          * bit, which is only set or cleared with irqs on
1475          */
1476         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1477                 struct btrfs_root *root = BTRFS_I(inode)->root;
1478                 u64 len = state->end + 1 - state->start;
1479                 bool do_list = !btrfs_is_free_space_inode(inode);
1480
1481                 if (*bits & EXTENT_FIRST_DELALLOC) {
1482                         *bits &= ~EXTENT_FIRST_DELALLOC;
1483                 } else {
1484                         spin_lock(&BTRFS_I(inode)->lock);
1485                         BTRFS_I(inode)->outstanding_extents++;
1486                         spin_unlock(&BTRFS_I(inode)->lock);
1487                 }
1488
1489                 spin_lock(&root->fs_info->delalloc_lock);
1490                 BTRFS_I(inode)->delalloc_bytes += len;
1491                 root->fs_info->delalloc_bytes += len;
1492                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1493                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494                                       &root->fs_info->delalloc_inodes);
1495                 }
1496                 spin_unlock(&root->fs_info->delalloc_lock);
1497         }
1498 }
1499
1500 /*
1501  * extent_io.c clear_bit_hook, see set_bit_hook for why
1502  */
1503 static void btrfs_clear_bit_hook(struct inode *inode,
1504                                  struct extent_state *state, int *bits)
1505 {
1506         /*
1507          * set_bit and clear bit hooks normally require _irqsave/restore
1508          * but in this case, we are only testing for the DELALLOC
1509          * bit, which is only set or cleared with irqs on
1510          */
1511         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1512                 struct btrfs_root *root = BTRFS_I(inode)->root;
1513                 u64 len = state->end + 1 - state->start;
1514                 bool do_list = !btrfs_is_free_space_inode(inode);
1515
1516                 if (*bits & EXTENT_FIRST_DELALLOC) {
1517                         *bits &= ~EXTENT_FIRST_DELALLOC;
1518                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519                         spin_lock(&BTRFS_I(inode)->lock);
1520                         BTRFS_I(inode)->outstanding_extents--;
1521                         spin_unlock(&BTRFS_I(inode)->lock);
1522                 }
1523
1524                 if (*bits & EXTENT_DO_ACCOUNTING)
1525                         btrfs_delalloc_release_metadata(inode, len);
1526
1527                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1528                     && do_list)
1529                         btrfs_free_reserved_data_space(inode, len);
1530
1531                 spin_lock(&root->fs_info->delalloc_lock);
1532                 root->fs_info->delalloc_bytes -= len;
1533                 BTRFS_I(inode)->delalloc_bytes -= len;
1534
1535                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1536                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1538                 }
1539                 spin_unlock(&root->fs_info->delalloc_lock);
1540         }
1541 }
1542
1543 /*
1544  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545  * we don't create bios that span stripes or chunks
1546  */
1547 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1548                          size_t size, struct bio *bio,
1549                          unsigned long bio_flags)
1550 {
1551         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552         struct btrfs_mapping_tree *map_tree;
1553         u64 logical = (u64)bio->bi_sector << 9;
1554         u64 length = 0;
1555         u64 map_length;
1556         int ret;
1557
1558         if (bio_flags & EXTENT_BIO_COMPRESSED)
1559                 return 0;
1560
1561         length = bio->bi_size;
1562         map_tree = &root->fs_info->mapping_tree;
1563         map_length = length;
1564         ret = btrfs_map_block(map_tree, READ, logical,
1565                               &map_length, NULL, 0);
1566         /* Will always return 0 or 1 with map_multi == NULL */
1567         BUG_ON(ret < 0);
1568         if (map_length < length + size)
1569                 return 1;
1570         return 0;
1571 }
1572
1573 /*
1574  * in order to insert checksums into the metadata in large chunks,
1575  * we wait until bio submission time.   All the pages in the bio are
1576  * checksummed and sums are attached onto the ordered extent record.
1577  *
1578  * At IO completion time the cums attached on the ordered extent record
1579  * are inserted into the btree
1580  */
1581 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1582                                     struct bio *bio, int mirror_num,
1583                                     unsigned long bio_flags,
1584                                     u64 bio_offset)
1585 {
1586         struct btrfs_root *root = BTRFS_I(inode)->root;
1587         int ret = 0;
1588
1589         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1590         BUG_ON(ret); /* -ENOMEM */
1591         return 0;
1592 }
1593
1594 /*
1595  * in order to insert checksums into the metadata in large chunks,
1596  * we wait until bio submission time.   All the pages in the bio are
1597  * checksummed and sums are attached onto the ordered extent record.
1598  *
1599  * At IO completion time the cums attached on the ordered extent record
1600  * are inserted into the btree
1601  */
1602 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1603                           int mirror_num, unsigned long bio_flags,
1604                           u64 bio_offset)
1605 {
1606         struct btrfs_root *root = BTRFS_I(inode)->root;
1607         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1608 }
1609
1610 /*
1611  * extent_io.c submission hook. This does the right thing for csum calculation
1612  * on write, or reading the csums from the tree before a read
1613  */
1614 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1615                           int mirror_num, unsigned long bio_flags,
1616                           u64 bio_offset)
1617 {
1618         struct btrfs_root *root = BTRFS_I(inode)->root;
1619         int ret = 0;
1620         int skip_sum;
1621         int metadata = 0;
1622
1623         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1624
1625         if (btrfs_is_free_space_inode(inode))
1626                 metadata = 2;
1627
1628         if (!(rw & REQ_WRITE)) {
1629                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1630                 if (ret)
1631                         return ret;
1632
1633                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1634                         return btrfs_submit_compressed_read(inode, bio,
1635                                                     mirror_num, bio_flags);
1636                 } else if (!skip_sum) {
1637                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1638                         if (ret)
1639                                 return ret;
1640                 }
1641                 goto mapit;
1642         } else if (!skip_sum) {
1643                 /* csum items have already been cloned */
1644                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1645                         goto mapit;
1646                 /* we're doing a write, do the async checksumming */
1647                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1648                                    inode, rw, bio, mirror_num,
1649                                    bio_flags, bio_offset,
1650                                    __btrfs_submit_bio_start,
1651                                    __btrfs_submit_bio_done);
1652         }
1653
1654 mapit:
1655         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1656 }
1657
1658 /*
1659  * given a list of ordered sums record them in the inode.  This happens
1660  * at IO completion time based on sums calculated at bio submission time.
1661  */
1662 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1663                              struct inode *inode, u64 file_offset,
1664                              struct list_head *list)
1665 {
1666         struct btrfs_ordered_sum *sum;
1667
1668         list_for_each_entry(sum, list, list) {
1669                 btrfs_csum_file_blocks(trans,
1670                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1671         }
1672         return 0;
1673 }
1674
1675 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1676                               struct extent_state **cached_state)
1677 {
1678         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1679         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1680                                    cached_state, GFP_NOFS);
1681 }
1682
1683 /* see btrfs_writepage_start_hook for details on why this is required */
1684 struct btrfs_writepage_fixup {
1685         struct page *page;
1686         struct btrfs_work work;
1687 };
1688
1689 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1690 {
1691         struct btrfs_writepage_fixup *fixup;
1692         struct btrfs_ordered_extent *ordered;
1693         struct extent_state *cached_state = NULL;
1694         struct page *page;
1695         struct inode *inode;
1696         u64 page_start;
1697         u64 page_end;
1698         int ret;
1699
1700         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1701         page = fixup->page;
1702 again:
1703         lock_page(page);
1704         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1705                 ClearPageChecked(page);
1706                 goto out_page;
1707         }
1708
1709         inode = page->mapping->host;
1710         page_start = page_offset(page);
1711         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1712
1713         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1714                          &cached_state);
1715
1716         /* already ordered? We're done */
1717         if (PagePrivate2(page))
1718                 goto out;
1719
1720         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1721         if (ordered) {
1722                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1723                                      page_end, &cached_state, GFP_NOFS);
1724                 unlock_page(page);
1725                 btrfs_start_ordered_extent(inode, ordered, 1);
1726                 btrfs_put_ordered_extent(ordered);
1727                 goto again;
1728         }
1729
1730         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1731         if (ret) {
1732                 mapping_set_error(page->mapping, ret);
1733                 end_extent_writepage(page, ret, page_start, page_end);
1734                 ClearPageChecked(page);
1735                 goto out;
1736          }
1737
1738         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1739         ClearPageChecked(page);
1740         set_page_dirty(page);
1741 out:
1742         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1743                              &cached_state, GFP_NOFS);
1744 out_page:
1745         unlock_page(page);
1746         page_cache_release(page);
1747         kfree(fixup);
1748 }
1749
1750 /*
1751  * There are a few paths in the higher layers of the kernel that directly
1752  * set the page dirty bit without asking the filesystem if it is a
1753  * good idea.  This causes problems because we want to make sure COW
1754  * properly happens and the data=ordered rules are followed.
1755  *
1756  * In our case any range that doesn't have the ORDERED bit set
1757  * hasn't been properly setup for IO.  We kick off an async process
1758  * to fix it up.  The async helper will wait for ordered extents, set
1759  * the delalloc bit and make it safe to write the page.
1760  */
1761 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1762 {
1763         struct inode *inode = page->mapping->host;
1764         struct btrfs_writepage_fixup *fixup;
1765         struct btrfs_root *root = BTRFS_I(inode)->root;
1766
1767         /* this page is properly in the ordered list */
1768         if (TestClearPagePrivate2(page))
1769                 return 0;
1770
1771         if (PageChecked(page))
1772                 return -EAGAIN;
1773
1774         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1775         if (!fixup)
1776                 return -EAGAIN;
1777
1778         SetPageChecked(page);
1779         page_cache_get(page);
1780         fixup->work.func = btrfs_writepage_fixup_worker;
1781         fixup->page = page;
1782         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1783         return -EBUSY;
1784 }
1785
1786 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1787                                        struct inode *inode, u64 file_pos,
1788                                        u64 disk_bytenr, u64 disk_num_bytes,
1789                                        u64 num_bytes, u64 ram_bytes,
1790                                        u8 compression, u8 encryption,
1791                                        u16 other_encoding, int extent_type)
1792 {
1793         struct btrfs_root *root = BTRFS_I(inode)->root;
1794         struct btrfs_file_extent_item *fi;
1795         struct btrfs_path *path;
1796         struct extent_buffer *leaf;
1797         struct btrfs_key ins;
1798         int ret;
1799
1800         path = btrfs_alloc_path();
1801         if (!path)
1802                 return -ENOMEM;
1803
1804         path->leave_spinning = 1;
1805
1806         /*
1807          * we may be replacing one extent in the tree with another.
1808          * The new extent is pinned in the extent map, and we don't want
1809          * to drop it from the cache until it is completely in the btree.
1810          *
1811          * So, tell btrfs_drop_extents to leave this extent in the cache.
1812          * the caller is expected to unpin it and allow it to be merged
1813          * with the others.
1814          */
1815         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1816                                  file_pos + num_bytes, 0);
1817         if (ret)
1818                 goto out;
1819
1820         ins.objectid = btrfs_ino(inode);
1821         ins.offset = file_pos;
1822         ins.type = BTRFS_EXTENT_DATA_KEY;
1823         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1824         if (ret)
1825                 goto out;
1826         leaf = path->nodes[0];
1827         fi = btrfs_item_ptr(leaf, path->slots[0],
1828                             struct btrfs_file_extent_item);
1829         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1830         btrfs_set_file_extent_type(leaf, fi, extent_type);
1831         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1832         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1833         btrfs_set_file_extent_offset(leaf, fi, 0);
1834         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1835         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1836         btrfs_set_file_extent_compression(leaf, fi, compression);
1837         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1838         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1839
1840         btrfs_mark_buffer_dirty(leaf);
1841         btrfs_release_path(path);
1842
1843         inode_add_bytes(inode, num_bytes);
1844
1845         ins.objectid = disk_bytenr;
1846         ins.offset = disk_num_bytes;
1847         ins.type = BTRFS_EXTENT_ITEM_KEY;
1848         ret = btrfs_alloc_reserved_file_extent(trans, root,
1849                                         root->root_key.objectid,
1850                                         btrfs_ino(inode), file_pos, &ins);
1851 out:
1852         btrfs_free_path(path);
1853
1854         return ret;
1855 }
1856
1857 /*
1858  * helper function for btrfs_finish_ordered_io, this
1859  * just reads in some of the csum leaves to prime them into ram
1860  * before we start the transaction.  It limits the amount of btree
1861  * reads required while inside the transaction.
1862  */
1863 /* as ordered data IO finishes, this gets called so we can finish
1864  * an ordered extent if the range of bytes in the file it covers are
1865  * fully written.
1866  */
1867 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1868 {
1869         struct inode *inode = ordered_extent->inode;
1870         struct btrfs_root *root = BTRFS_I(inode)->root;
1871         struct btrfs_trans_handle *trans = NULL;
1872         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1873         struct extent_state *cached_state = NULL;
1874         int compress_type = 0;
1875         int ret;
1876         bool nolock;
1877
1878         nolock = btrfs_is_free_space_inode(inode);
1879
1880         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1881                 ret = -EIO;
1882                 goto out;
1883         }
1884
1885         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1886                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1887                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1888                 if (!ret) {
1889                         if (nolock)
1890                                 trans = btrfs_join_transaction_nolock(root);
1891                         else
1892                                 trans = btrfs_join_transaction(root);
1893                         if (IS_ERR(trans)) {
1894                                 ret = PTR_ERR(trans);
1895                                 trans = NULL;
1896                                 goto out;
1897                         }
1898                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1899                         ret = btrfs_update_inode_fallback(trans, root, inode);
1900                         if (ret) /* -ENOMEM or corruption */
1901                                 btrfs_abort_transaction(trans, root, ret);
1902                 }
1903                 goto out;
1904         }
1905
1906         lock_extent_bits(io_tree, ordered_extent->file_offset,
1907                          ordered_extent->file_offset + ordered_extent->len - 1,
1908                          0, &cached_state);
1909
1910         if (nolock)
1911                 trans = btrfs_join_transaction_nolock(root);
1912         else
1913                 trans = btrfs_join_transaction(root);
1914         if (IS_ERR(trans)) {
1915                 ret = PTR_ERR(trans);
1916                 trans = NULL;
1917                 goto out_unlock;
1918         }
1919         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1920
1921         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1922                 compress_type = ordered_extent->compress_type;
1923         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1924                 BUG_ON(compress_type);
1925                 ret = btrfs_mark_extent_written(trans, inode,
1926                                                 ordered_extent->file_offset,
1927                                                 ordered_extent->file_offset +
1928                                                 ordered_extent->len);
1929         } else {
1930                 BUG_ON(root == root->fs_info->tree_root);
1931                 ret = insert_reserved_file_extent(trans, inode,
1932                                                 ordered_extent->file_offset,
1933                                                 ordered_extent->start,
1934                                                 ordered_extent->disk_len,
1935                                                 ordered_extent->len,
1936                                                 ordered_extent->len,
1937                                                 compress_type, 0, 0,
1938                                                 BTRFS_FILE_EXTENT_REG);
1939         }
1940         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1941                            ordered_extent->file_offset, ordered_extent->len,
1942                            trans->transid);
1943         if (ret < 0) {
1944                 btrfs_abort_transaction(trans, root, ret);
1945                 goto out_unlock;
1946         }
1947
1948         add_pending_csums(trans, inode, ordered_extent->file_offset,
1949                           &ordered_extent->list);
1950
1951         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1952         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1953                 ret = btrfs_update_inode_fallback(trans, root, inode);
1954                 if (ret) { /* -ENOMEM or corruption */
1955                         btrfs_abort_transaction(trans, root, ret);
1956                         goto out_unlock;
1957                 }
1958         } else {
1959                 btrfs_set_inode_last_trans(trans, inode);
1960         }
1961         ret = 0;
1962 out_unlock:
1963         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1964                              ordered_extent->file_offset +
1965                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1966 out:
1967         if (root != root->fs_info->tree_root)
1968                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1969         if (trans)
1970                 btrfs_end_transaction(trans, root);
1971
1972         if (ret)
1973                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1974                                       ordered_extent->file_offset +
1975                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1976
1977         /*
1978          * This needs to be done to make sure anybody waiting knows we are done
1979          * updating everything for this ordered extent.
1980          */
1981         btrfs_remove_ordered_extent(inode, ordered_extent);
1982
1983         /* once for us */
1984         btrfs_put_ordered_extent(ordered_extent);
1985         /* once for the tree */
1986         btrfs_put_ordered_extent(ordered_extent);
1987
1988         return ret;
1989 }
1990
1991 static void finish_ordered_fn(struct btrfs_work *work)
1992 {
1993         struct btrfs_ordered_extent *ordered_extent;
1994         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1995         btrfs_finish_ordered_io(ordered_extent);
1996 }
1997
1998 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1999                                 struct extent_state *state, int uptodate)
2000 {
2001         struct inode *inode = page->mapping->host;
2002         struct btrfs_root *root = BTRFS_I(inode)->root;
2003         struct btrfs_ordered_extent *ordered_extent = NULL;
2004         struct btrfs_workers *workers;
2005
2006         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2007
2008         ClearPagePrivate2(page);
2009         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2010                                             end - start + 1, uptodate))
2011                 return 0;
2012
2013         ordered_extent->work.func = finish_ordered_fn;
2014         ordered_extent->work.flags = 0;
2015
2016         if (btrfs_is_free_space_inode(inode))
2017                 workers = &root->fs_info->endio_freespace_worker;
2018         else
2019                 workers = &root->fs_info->endio_write_workers;
2020         btrfs_queue_worker(workers, &ordered_extent->work);
2021
2022         return 0;
2023 }
2024
2025 /*
2026  * when reads are done, we need to check csums to verify the data is correct
2027  * if there's a match, we allow the bio to finish.  If not, the code in
2028  * extent_io.c will try to find good copies for us.
2029  */
2030 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2031                                struct extent_state *state, int mirror)
2032 {
2033         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2034         struct inode *inode = page->mapping->host;
2035         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2036         char *kaddr;
2037         u64 private = ~(u32)0;
2038         int ret;
2039         struct btrfs_root *root = BTRFS_I(inode)->root;
2040         u32 csum = ~(u32)0;
2041
2042         if (PageChecked(page)) {
2043                 ClearPageChecked(page);
2044                 goto good;
2045         }
2046
2047         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2048                 goto good;
2049
2050         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2051             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2052                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2053                                   GFP_NOFS);
2054                 return 0;
2055         }
2056
2057         if (state && state->start == start) {
2058                 private = state->private;
2059                 ret = 0;
2060         } else {
2061                 ret = get_state_private(io_tree, start, &private);
2062         }
2063         kaddr = kmap_atomic(page);
2064         if (ret)
2065                 goto zeroit;
2066
2067         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2068         btrfs_csum_final(csum, (char *)&csum);
2069         if (csum != private)
2070                 goto zeroit;
2071
2072         kunmap_atomic(kaddr);
2073 good:
2074         return 0;
2075
2076 zeroit:
2077         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2078                        "private %llu\n",
2079                        (unsigned long long)btrfs_ino(page->mapping->host),
2080                        (unsigned long long)start, csum,
2081                        (unsigned long long)private);
2082         memset(kaddr + offset, 1, end - start + 1);
2083         flush_dcache_page(page);
2084         kunmap_atomic(kaddr);
2085         if (private == 0)
2086                 return 0;
2087         return -EIO;
2088 }
2089
2090 struct delayed_iput {
2091         struct list_head list;
2092         struct inode *inode;
2093 };
2094
2095 /* JDM: If this is fs-wide, why can't we add a pointer to
2096  * btrfs_inode instead and avoid the allocation? */
2097 void btrfs_add_delayed_iput(struct inode *inode)
2098 {
2099         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2100         struct delayed_iput *delayed;
2101
2102         if (atomic_add_unless(&inode->i_count, -1, 1))
2103                 return;
2104
2105         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2106         delayed->inode = inode;
2107
2108         spin_lock(&fs_info->delayed_iput_lock);
2109         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2110         spin_unlock(&fs_info->delayed_iput_lock);
2111 }
2112
2113 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2114 {
2115         LIST_HEAD(list);
2116         struct btrfs_fs_info *fs_info = root->fs_info;
2117         struct delayed_iput *delayed;
2118         int empty;
2119
2120         spin_lock(&fs_info->delayed_iput_lock);
2121         empty = list_empty(&fs_info->delayed_iputs);
2122         spin_unlock(&fs_info->delayed_iput_lock);
2123         if (empty)
2124                 return;
2125
2126         spin_lock(&fs_info->delayed_iput_lock);
2127         list_splice_init(&fs_info->delayed_iputs, &list);
2128         spin_unlock(&fs_info->delayed_iput_lock);
2129
2130         while (!list_empty(&list)) {
2131                 delayed = list_entry(list.next, struct delayed_iput, list);
2132                 list_del(&delayed->list);
2133                 iput(delayed->inode);
2134                 kfree(delayed);
2135         }
2136 }
2137
2138 enum btrfs_orphan_cleanup_state {
2139         ORPHAN_CLEANUP_STARTED  = 1,
2140         ORPHAN_CLEANUP_DONE     = 2,
2141 };
2142
2143 /*
2144  * This is called in transaction commit time. If there are no orphan
2145  * files in the subvolume, it removes orphan item and frees block_rsv
2146  * structure.
2147  */
2148 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2149                               struct btrfs_root *root)
2150 {
2151         struct btrfs_block_rsv *block_rsv;
2152         int ret;
2153
2154         if (atomic_read(&root->orphan_inodes) ||
2155             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2156                 return;
2157
2158         spin_lock(&root->orphan_lock);
2159         if (atomic_read(&root->orphan_inodes)) {
2160                 spin_unlock(&root->orphan_lock);
2161                 return;
2162         }
2163
2164         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2165                 spin_unlock(&root->orphan_lock);
2166                 return;
2167         }
2168
2169         block_rsv = root->orphan_block_rsv;
2170         root->orphan_block_rsv = NULL;
2171         spin_unlock(&root->orphan_lock);
2172
2173         if (root->orphan_item_inserted &&
2174             btrfs_root_refs(&root->root_item) > 0) {
2175                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2176                                             root->root_key.objectid);
2177                 BUG_ON(ret);
2178                 root->orphan_item_inserted = 0;
2179         }
2180
2181         if (block_rsv) {
2182                 WARN_ON(block_rsv->size > 0);
2183                 btrfs_free_block_rsv(root, block_rsv);
2184         }
2185 }
2186
2187 /*
2188  * This creates an orphan entry for the given inode in case something goes
2189  * wrong in the middle of an unlink/truncate.
2190  *
2191  * NOTE: caller of this function should reserve 5 units of metadata for
2192  *       this function.
2193  */
2194 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2195 {
2196         struct btrfs_root *root = BTRFS_I(inode)->root;
2197         struct btrfs_block_rsv *block_rsv = NULL;
2198         int reserve = 0;
2199         int insert = 0;
2200         int ret;
2201
2202         if (!root->orphan_block_rsv) {
2203                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2204                 if (!block_rsv)
2205                         return -ENOMEM;
2206         }
2207
2208         spin_lock(&root->orphan_lock);
2209         if (!root->orphan_block_rsv) {
2210                 root->orphan_block_rsv = block_rsv;
2211         } else if (block_rsv) {
2212                 btrfs_free_block_rsv(root, block_rsv);
2213                 block_rsv = NULL;
2214         }
2215
2216         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2217                               &BTRFS_I(inode)->runtime_flags)) {
2218 #if 0
2219                 /*
2220                  * For proper ENOSPC handling, we should do orphan
2221                  * cleanup when mounting. But this introduces backward
2222                  * compatibility issue.
2223                  */
2224                 if (!xchg(&root->orphan_item_inserted, 1))
2225                         insert = 2;
2226                 else
2227                         insert = 1;
2228 #endif
2229                 insert = 1;
2230                 atomic_inc(&root->orphan_inodes);
2231         }
2232
2233         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2234                               &BTRFS_I(inode)->runtime_flags))
2235                 reserve = 1;
2236         spin_unlock(&root->orphan_lock);
2237
2238         /* grab metadata reservation from transaction handle */
2239         if (reserve) {
2240                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2241                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2242         }
2243
2244         /* insert an orphan item to track this unlinked/truncated file */
2245         if (insert >= 1) {
2246                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2247                 if (ret && ret != -EEXIST) {
2248                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249                                   &BTRFS_I(inode)->runtime_flags);
2250                         btrfs_abort_transaction(trans, root, ret);
2251                         return ret;
2252                 }
2253                 ret = 0;
2254         }
2255
2256         /* insert an orphan item to track subvolume contains orphan files */
2257         if (insert >= 2) {
2258                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2259                                                root->root_key.objectid);
2260                 if (ret && ret != -EEXIST) {
2261                         btrfs_abort_transaction(trans, root, ret);
2262                         return ret;
2263                 }
2264         }
2265         return 0;
2266 }
2267
2268 /*
2269  * We have done the truncate/delete so we can go ahead and remove the orphan
2270  * item for this particular inode.
2271  */
2272 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2273 {
2274         struct btrfs_root *root = BTRFS_I(inode)->root;
2275         int delete_item = 0;
2276         int release_rsv = 0;
2277         int ret = 0;
2278
2279         spin_lock(&root->orphan_lock);
2280         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281                                &BTRFS_I(inode)->runtime_flags))
2282                 delete_item = 1;
2283
2284         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2285                                &BTRFS_I(inode)->runtime_flags))
2286                 release_rsv = 1;
2287         spin_unlock(&root->orphan_lock);
2288
2289         if (trans && delete_item) {
2290                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2291                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2292         }
2293
2294         if (release_rsv) {
2295                 btrfs_orphan_release_metadata(inode);
2296                 atomic_dec(&root->orphan_inodes);
2297         }
2298
2299         return 0;
2300 }
2301
2302 /*
2303  * this cleans up any orphans that may be left on the list from the last use
2304  * of this root.
2305  */
2306 int btrfs_orphan_cleanup(struct btrfs_root *root)
2307 {
2308         struct btrfs_path *path;
2309         struct extent_buffer *leaf;
2310         struct btrfs_key key, found_key;
2311         struct btrfs_trans_handle *trans;
2312         struct inode *inode;
2313         u64 last_objectid = 0;
2314         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2315
2316         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2317                 return 0;
2318
2319         path = btrfs_alloc_path();
2320         if (!path) {
2321                 ret = -ENOMEM;
2322                 goto out;
2323         }
2324         path->reada = -1;
2325
2326         key.objectid = BTRFS_ORPHAN_OBJECTID;
2327         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2328         key.offset = (u64)-1;
2329
2330         while (1) {
2331                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2332                 if (ret < 0)
2333                         goto out;
2334
2335                 /*
2336                  * if ret == 0 means we found what we were searching for, which
2337                  * is weird, but possible, so only screw with path if we didn't
2338                  * find the key and see if we have stuff that matches
2339                  */
2340                 if (ret > 0) {
2341                         ret = 0;
2342                         if (path->slots[0] == 0)
2343                                 break;
2344                         path->slots[0]--;
2345                 }
2346
2347                 /* pull out the item */
2348                 leaf = path->nodes[0];
2349                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2350
2351                 /* make sure the item matches what we want */
2352                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2353                         break;
2354                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2355                         break;
2356
2357                 /* release the path since we're done with it */
2358                 btrfs_release_path(path);
2359
2360                 /*
2361                  * this is where we are basically btrfs_lookup, without the
2362                  * crossing root thing.  we store the inode number in the
2363                  * offset of the orphan item.
2364                  */
2365
2366                 if (found_key.offset == last_objectid) {
2367                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2368                                "stopping orphan cleanup\n");
2369                         ret = -EINVAL;
2370                         goto out;
2371                 }
2372
2373                 last_objectid = found_key.offset;
2374
2375                 found_key.objectid = found_key.offset;
2376                 found_key.type = BTRFS_INODE_ITEM_KEY;
2377                 found_key.offset = 0;
2378                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2379                 ret = PTR_RET(inode);
2380                 if (ret && ret != -ESTALE)
2381                         goto out;
2382
2383                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2384                         struct btrfs_root *dead_root;
2385                         struct btrfs_fs_info *fs_info = root->fs_info;
2386                         int is_dead_root = 0;
2387
2388                         /*
2389                          * this is an orphan in the tree root. Currently these
2390                          * could come from 2 sources:
2391                          *  a) a snapshot deletion in progress
2392                          *  b) a free space cache inode
2393                          * We need to distinguish those two, as the snapshot
2394                          * orphan must not get deleted.
2395                          * find_dead_roots already ran before us, so if this
2396                          * is a snapshot deletion, we should find the root
2397                          * in the dead_roots list
2398                          */
2399                         spin_lock(&fs_info->trans_lock);
2400                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2401                                             root_list) {
2402                                 if (dead_root->root_key.objectid ==
2403                                     found_key.objectid) {
2404                                         is_dead_root = 1;
2405                                         break;
2406                                 }
2407                         }
2408                         spin_unlock(&fs_info->trans_lock);
2409                         if (is_dead_root) {
2410                                 /* prevent this orphan from being found again */
2411                                 key.offset = found_key.objectid - 1;
2412                                 continue;
2413                         }
2414                 }
2415                 /*
2416                  * Inode is already gone but the orphan item is still there,
2417                  * kill the orphan item.
2418                  */
2419                 if (ret == -ESTALE) {
2420                         trans = btrfs_start_transaction(root, 1);
2421                         if (IS_ERR(trans)) {
2422                                 ret = PTR_ERR(trans);
2423                                 goto out;
2424                         }
2425                         printk(KERN_ERR "auto deleting %Lu\n",
2426                                found_key.objectid);
2427                         ret = btrfs_del_orphan_item(trans, root,
2428                                                     found_key.objectid);
2429                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2430                         btrfs_end_transaction(trans, root);
2431                         continue;
2432                 }
2433
2434                 /*
2435                  * add this inode to the orphan list so btrfs_orphan_del does
2436                  * the proper thing when we hit it
2437                  */
2438                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2439                         &BTRFS_I(inode)->runtime_flags);
2440
2441                 /* if we have links, this was a truncate, lets do that */
2442                 if (inode->i_nlink) {
2443                         if (!S_ISREG(inode->i_mode)) {
2444                                 WARN_ON(1);
2445                                 iput(inode);
2446                                 continue;
2447                         }
2448                         nr_truncate++;
2449                         ret = btrfs_truncate(inode);
2450                 } else {
2451                         nr_unlink++;
2452                 }
2453
2454                 /* this will do delete_inode and everything for us */
2455                 iput(inode);
2456                 if (ret)
2457                         goto out;
2458         }
2459         /* release the path since we're done with it */
2460         btrfs_release_path(path);
2461
2462         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2463
2464         if (root->orphan_block_rsv)
2465                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2466                                         (u64)-1);
2467
2468         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2469                 trans = btrfs_join_transaction(root);
2470                 if (!IS_ERR(trans))
2471                         btrfs_end_transaction(trans, root);
2472         }
2473
2474         if (nr_unlink)
2475                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2476         if (nr_truncate)
2477                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2478
2479 out:
2480         if (ret)
2481                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2482         btrfs_free_path(path);
2483         return ret;
2484 }
2485
2486 /*
2487  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2488  * don't find any xattrs, we know there can't be any acls.
2489  *
2490  * slot is the slot the inode is in, objectid is the objectid of the inode
2491  */
2492 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2493                                           int slot, u64 objectid)
2494 {
2495         u32 nritems = btrfs_header_nritems(leaf);
2496         struct btrfs_key found_key;
2497         int scanned = 0;
2498
2499         slot++;
2500         while (slot < nritems) {
2501                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2502
2503                 /* we found a different objectid, there must not be acls */
2504                 if (found_key.objectid != objectid)
2505                         return 0;
2506
2507                 /* we found an xattr, assume we've got an acl */
2508                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2509                         return 1;
2510
2511                 /*
2512                  * we found a key greater than an xattr key, there can't
2513                  * be any acls later on
2514                  */
2515                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2516                         return 0;
2517
2518                 slot++;
2519                 scanned++;
2520
2521                 /*
2522                  * it goes inode, inode backrefs, xattrs, extents,
2523                  * so if there are a ton of hard links to an inode there can
2524                  * be a lot of backrefs.  Don't waste time searching too hard,
2525                  * this is just an optimization
2526                  */
2527                 if (scanned >= 8)
2528                         break;
2529         }
2530         /* we hit the end of the leaf before we found an xattr or
2531          * something larger than an xattr.  We have to assume the inode
2532          * has acls
2533          */
2534         return 1;
2535 }
2536
2537 /*
2538  * read an inode from the btree into the in-memory inode
2539  */
2540 static void btrfs_read_locked_inode(struct inode *inode)
2541 {
2542         struct btrfs_path *path;
2543         struct extent_buffer *leaf;
2544         struct btrfs_inode_item *inode_item;
2545         struct btrfs_timespec *tspec;
2546         struct btrfs_root *root = BTRFS_I(inode)->root;
2547         struct btrfs_key location;
2548         int maybe_acls;
2549         u32 rdev;
2550         int ret;
2551         bool filled = false;
2552
2553         ret = btrfs_fill_inode(inode, &rdev);
2554         if (!ret)
2555                 filled = true;
2556
2557         path = btrfs_alloc_path();
2558         if (!path)
2559                 goto make_bad;
2560
2561         path->leave_spinning = 1;
2562         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2563
2564         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2565         if (ret)
2566                 goto make_bad;
2567
2568         leaf = path->nodes[0];
2569
2570         if (filled)
2571                 goto cache_acl;
2572
2573         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2574                                     struct btrfs_inode_item);
2575         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2576         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2577         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2578         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2579         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2580
2581         tspec = btrfs_inode_atime(inode_item);
2582         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2584
2585         tspec = btrfs_inode_mtime(inode_item);
2586         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2588
2589         tspec = btrfs_inode_ctime(inode_item);
2590         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2591         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2592
2593         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2594         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2595         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2596
2597         /*
2598          * If we were modified in the current generation and evicted from memory
2599          * and then re-read we need to do a full sync since we don't have any
2600          * idea about which extents were modified before we were evicted from
2601          * cache.
2602          */
2603         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2604                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2605                         &BTRFS_I(inode)->runtime_flags);
2606
2607         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2608         inode->i_generation = BTRFS_I(inode)->generation;
2609         inode->i_rdev = 0;
2610         rdev = btrfs_inode_rdev(leaf, inode_item);
2611
2612         BTRFS_I(inode)->index_cnt = (u64)-1;
2613         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2614 cache_acl:
2615         /*
2616          * try to precache a NULL acl entry for files that don't have
2617          * any xattrs or acls
2618          */
2619         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2620                                            btrfs_ino(inode));
2621         if (!maybe_acls)
2622                 cache_no_acl(inode);
2623
2624         btrfs_free_path(path);
2625
2626         switch (inode->i_mode & S_IFMT) {
2627         case S_IFREG:
2628                 inode->i_mapping->a_ops = &btrfs_aops;
2629                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2630                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2631                 inode->i_fop = &btrfs_file_operations;
2632                 inode->i_op = &btrfs_file_inode_operations;
2633                 break;
2634         case S_IFDIR:
2635                 inode->i_fop = &btrfs_dir_file_operations;
2636                 if (root == root->fs_info->tree_root)
2637                         inode->i_op = &btrfs_dir_ro_inode_operations;
2638                 else
2639                         inode->i_op = &btrfs_dir_inode_operations;
2640                 break;
2641         case S_IFLNK:
2642                 inode->i_op = &btrfs_symlink_inode_operations;
2643                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2644                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2645                 break;
2646         default:
2647                 inode->i_op = &btrfs_special_inode_operations;
2648                 init_special_inode(inode, inode->i_mode, rdev);
2649                 break;
2650         }
2651
2652         btrfs_update_iflags(inode);
2653         return;
2654
2655 make_bad:
2656         btrfs_free_path(path);
2657         make_bad_inode(inode);
2658 }
2659
2660 /*
2661  * given a leaf and an inode, copy the inode fields into the leaf
2662  */
2663 static void fill_inode_item(struct btrfs_trans_handle *trans,
2664                             struct extent_buffer *leaf,
2665                             struct btrfs_inode_item *item,
2666                             struct inode *inode)
2667 {
2668         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2669         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2670         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2671         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2672         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2673
2674         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2675                                inode->i_atime.tv_sec);
2676         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2677                                 inode->i_atime.tv_nsec);
2678
2679         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2680                                inode->i_mtime.tv_sec);
2681         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2682                                 inode->i_mtime.tv_nsec);
2683
2684         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2685                                inode->i_ctime.tv_sec);
2686         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2687                                 inode->i_ctime.tv_nsec);
2688
2689         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2690         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2691         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2692         btrfs_set_inode_transid(leaf, item, trans->transid);
2693         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2694         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2695         btrfs_set_inode_block_group(leaf, item, 0);
2696 }
2697
2698 /*
2699  * copy everything in the in-memory inode into the btree.
2700  */
2701 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2702                                 struct btrfs_root *root, struct inode *inode)
2703 {
2704         struct btrfs_inode_item *inode_item;
2705         struct btrfs_path *path;
2706         struct extent_buffer *leaf;
2707         int ret;
2708
2709         path = btrfs_alloc_path();
2710         if (!path)
2711                 return -ENOMEM;
2712
2713         path->leave_spinning = 1;
2714         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2715                                  1);
2716         if (ret) {
2717                 if (ret > 0)
2718                         ret = -ENOENT;
2719                 goto failed;
2720         }
2721
2722         btrfs_unlock_up_safe(path, 1);
2723         leaf = path->nodes[0];
2724         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2725                                     struct btrfs_inode_item);
2726
2727         fill_inode_item(trans, leaf, inode_item, inode);
2728         btrfs_mark_buffer_dirty(leaf);
2729         btrfs_set_inode_last_trans(trans, inode);
2730         ret = 0;
2731 failed:
2732         btrfs_free_path(path);
2733         return ret;
2734 }
2735
2736 /*
2737  * copy everything in the in-memory inode into the btree.
2738  */
2739 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2740                                 struct btrfs_root *root, struct inode *inode)
2741 {
2742         int ret;
2743
2744         /*
2745          * If the inode is a free space inode, we can deadlock during commit
2746          * if we put it into the delayed code.
2747          *
2748          * The data relocation inode should also be directly updated
2749          * without delay
2750          */
2751         if (!btrfs_is_free_space_inode(inode)
2752             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2753                 btrfs_update_root_times(trans, root);
2754
2755                 ret = btrfs_delayed_update_inode(trans, root, inode);
2756                 if (!ret)
2757                         btrfs_set_inode_last_trans(trans, inode);
2758                 return ret;
2759         }
2760
2761         return btrfs_update_inode_item(trans, root, inode);
2762 }
2763
2764 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2765                                          struct btrfs_root *root,
2766                                          struct inode *inode)
2767 {
2768         int ret;
2769
2770         ret = btrfs_update_inode(trans, root, inode);
2771         if (ret == -ENOSPC)
2772                 return btrfs_update_inode_item(trans, root, inode);
2773         return ret;
2774 }
2775
2776 /*
2777  * unlink helper that gets used here in inode.c and in the tree logging
2778  * recovery code.  It remove a link in a directory with a given name, and
2779  * also drops the back refs in the inode to the directory
2780  */
2781 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2782                                 struct btrfs_root *root,
2783                                 struct inode *dir, struct inode *inode,
2784                                 const char *name, int name_len)
2785 {
2786         struct btrfs_path *path;
2787         int ret = 0;
2788         struct extent_buffer *leaf;
2789         struct btrfs_dir_item *di;
2790         struct btrfs_key key;
2791         u64 index;
2792         u64 ino = btrfs_ino(inode);
2793         u64 dir_ino = btrfs_ino(dir);
2794
2795         path = btrfs_alloc_path();
2796         if (!path) {
2797                 ret = -ENOMEM;
2798                 goto out;
2799         }
2800
2801         path->leave_spinning = 1;
2802         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2803                                     name, name_len, -1);
2804         if (IS_ERR(di)) {
2805                 ret = PTR_ERR(di);
2806                 goto err;
2807         }
2808         if (!di) {
2809                 ret = -ENOENT;
2810                 goto err;
2811         }
2812         leaf = path->nodes[0];
2813         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2814         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2815         if (ret)
2816                 goto err;
2817         btrfs_release_path(path);
2818
2819         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2820                                   dir_ino, &index);
2821         if (ret) {
2822                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2823                        "inode %llu parent %llu\n", name_len, name,
2824                        (unsigned long long)ino, (unsigned long long)dir_ino);
2825                 btrfs_abort_transaction(trans, root, ret);
2826                 goto err;
2827         }
2828
2829         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2830         if (ret) {
2831                 btrfs_abort_transaction(trans, root, ret);
2832                 goto err;
2833         }
2834
2835         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2836                                          inode, dir_ino);
2837         if (ret != 0 && ret != -ENOENT) {
2838                 btrfs_abort_transaction(trans, root, ret);
2839                 goto err;
2840         }
2841
2842         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2843                                            dir, index);
2844         if (ret == -ENOENT)
2845                 ret = 0;
2846 err:
2847         btrfs_free_path(path);
2848         if (ret)
2849                 goto out;
2850
2851         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2852         inode_inc_iversion(inode);
2853         inode_inc_iversion(dir);
2854         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2855         ret = btrfs_update_inode(trans, root, dir);
2856 out:
2857         return ret;
2858 }
2859
2860 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2861                        struct btrfs_root *root,
2862                        struct inode *dir, struct inode *inode,
2863                        const char *name, int name_len)
2864 {
2865         int ret;
2866         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2867         if (!ret) {
2868                 btrfs_drop_nlink(inode);
2869                 ret = btrfs_update_inode(trans, root, inode);
2870         }
2871         return ret;
2872 }
2873
2874
2875 /* helper to check if there is any shared block in the path */
2876 static int check_path_shared(struct btrfs_root *root,
2877                              struct btrfs_path *path)
2878 {
2879         struct extent_buffer *eb;
2880         int level;
2881         u64 refs = 1;
2882
2883         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2884                 int ret;
2885
2886                 if (!path->nodes[level])
2887                         break;
2888                 eb = path->nodes[level];
2889                 if (!btrfs_block_can_be_shared(root, eb))
2890                         continue;
2891                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2892                                                &refs, NULL);
2893                 if (refs > 1)
2894                         return 1;
2895         }
2896         return 0;
2897 }
2898
2899 /*
2900  * helper to start transaction for unlink and rmdir.
2901  *
2902  * unlink and rmdir are special in btrfs, they do not always free space.
2903  * so in enospc case, we should make sure they will free space before
2904  * allowing them to use the global metadata reservation.
2905  */
2906 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2907                                                        struct dentry *dentry)
2908 {
2909         struct btrfs_trans_handle *trans;
2910         struct btrfs_root *root = BTRFS_I(dir)->root;
2911         struct btrfs_path *path;
2912         struct btrfs_dir_item *di;
2913         struct inode *inode = dentry->d_inode;
2914         u64 index;
2915         int check_link = 1;
2916         int err = -ENOSPC;
2917         int ret;
2918         u64 ino = btrfs_ino(inode);
2919         u64 dir_ino = btrfs_ino(dir);
2920
2921         /*
2922          * 1 for the possible orphan item
2923          * 1 for the dir item
2924          * 1 for the dir index
2925          * 1 for the inode ref
2926          * 1 for the inode ref in the tree log
2927          * 2 for the dir entries in the log
2928          * 1 for the inode
2929          */
2930         trans = btrfs_start_transaction(root, 8);
2931         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2932                 return trans;
2933
2934         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2935                 return ERR_PTR(-ENOSPC);
2936
2937         /* check if there is someone else holds reference */
2938         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2939                 return ERR_PTR(-ENOSPC);
2940
2941         if (atomic_read(&inode->i_count) > 2)
2942                 return ERR_PTR(-ENOSPC);
2943
2944         if (xchg(&root->fs_info->enospc_unlink, 1))
2945                 return ERR_PTR(-ENOSPC);
2946
2947         path = btrfs_alloc_path();
2948         if (!path) {
2949                 root->fs_info->enospc_unlink = 0;
2950                 return ERR_PTR(-ENOMEM);
2951         }
2952
2953         /* 1 for the orphan item */
2954         trans = btrfs_start_transaction(root, 1);
2955         if (IS_ERR(trans)) {
2956                 btrfs_free_path(path);
2957                 root->fs_info->enospc_unlink = 0;
2958                 return trans;
2959         }
2960
2961         path->skip_locking = 1;
2962         path->search_commit_root = 1;
2963
2964         ret = btrfs_lookup_inode(trans, root, path,
2965                                 &BTRFS_I(dir)->location, 0);
2966         if (ret < 0) {
2967                 err = ret;
2968                 goto out;
2969         }
2970         if (ret == 0) {
2971                 if (check_path_shared(root, path))
2972                         goto out;
2973         } else {
2974                 check_link = 0;
2975         }
2976         btrfs_release_path(path);
2977
2978         ret = btrfs_lookup_inode(trans, root, path,
2979                                 &BTRFS_I(inode)->location, 0);
2980         if (ret < 0) {
2981                 err = ret;
2982                 goto out;
2983         }
2984         if (ret == 0) {
2985                 if (check_path_shared(root, path))
2986                         goto out;
2987         } else {
2988                 check_link = 0;
2989         }
2990         btrfs_release_path(path);
2991
2992         if (ret == 0 && S_ISREG(inode->i_mode)) {
2993                 ret = btrfs_lookup_file_extent(trans, root, path,
2994                                                ino, (u64)-1, 0);
2995                 if (ret < 0) {
2996                         err = ret;
2997                         goto out;
2998                 }
2999                 BUG_ON(ret == 0); /* Corruption */
3000                 if (check_path_shared(root, path))
3001                         goto out;
3002                 btrfs_release_path(path);
3003         }
3004
3005         if (!check_link) {
3006                 err = 0;
3007                 goto out;
3008         }
3009
3010         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3011                                 dentry->d_name.name, dentry->d_name.len, 0);
3012         if (IS_ERR(di)) {
3013                 err = PTR_ERR(di);
3014                 goto out;
3015         }
3016         if (di) {
3017                 if (check_path_shared(root, path))
3018                         goto out;
3019         } else {
3020                 err = 0;
3021                 goto out;
3022         }
3023         btrfs_release_path(path);
3024
3025         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3026                                         dentry->d_name.len, ino, dir_ino, 0,
3027                                         &index);
3028         if (ret) {
3029                 err = ret;
3030                 goto out;
3031         }
3032
3033         if (check_path_shared(root, path))
3034                 goto out;
3035
3036         btrfs_release_path(path);
3037
3038         /*
3039          * This is a commit root search, if we can lookup inode item and other
3040          * relative items in the commit root, it means the transaction of
3041          * dir/file creation has been committed, and the dir index item that we
3042          * delay to insert has also been inserted into the commit root. So
3043          * we needn't worry about the delayed insertion of the dir index item
3044          * here.
3045          */
3046         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3047                                 dentry->d_name.name, dentry->d_name.len, 0);
3048         if (IS_ERR(di)) {
3049                 err = PTR_ERR(di);
3050                 goto out;
3051         }
3052         BUG_ON(ret == -ENOENT);
3053         if (check_path_shared(root, path))
3054                 goto out;
3055
3056         err = 0;
3057 out:
3058         btrfs_free_path(path);
3059         /* Migrate the orphan reservation over */
3060         if (!err)
3061                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3062                                 &root->fs_info->global_block_rsv,
3063                                 trans->bytes_reserved);
3064
3065         if (err) {
3066                 btrfs_end_transaction(trans, root);
3067                 root->fs_info->enospc_unlink = 0;
3068                 return ERR_PTR(err);
3069         }
3070
3071         trans->block_rsv = &root->fs_info->global_block_rsv;
3072         return trans;
3073 }
3074
3075 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3076                                struct btrfs_root *root)
3077 {
3078         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3079                 btrfs_block_rsv_release(root, trans->block_rsv,
3080                                         trans->bytes_reserved);
3081                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3082                 BUG_ON(!root->fs_info->enospc_unlink);
3083                 root->fs_info->enospc_unlink = 0;
3084         }
3085         btrfs_end_transaction(trans, root);
3086 }
3087
3088 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3089 {
3090         struct btrfs_root *root = BTRFS_I(dir)->root;
3091         struct btrfs_trans_handle *trans;
3092         struct inode *inode = dentry->d_inode;
3093         int ret;
3094
3095         trans = __unlink_start_trans(dir, dentry);
3096         if (IS_ERR(trans))
3097                 return PTR_ERR(trans);
3098
3099         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3100
3101         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3102                                  dentry->d_name.name, dentry->d_name.len);
3103         if (ret)
3104                 goto out;
3105
3106         if (inode->i_nlink == 0) {
3107                 ret = btrfs_orphan_add(trans, inode);
3108                 if (ret)
3109                         goto out;
3110         }
3111
3112 out:
3113         __unlink_end_trans(trans, root);
3114         btrfs_btree_balance_dirty(root);
3115         return ret;
3116 }
3117
3118 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3119                         struct btrfs_root *root,
3120                         struct inode *dir, u64 objectid,
3121                         const char *name, int name_len)
3122 {
3123         struct btrfs_path *path;
3124         struct extent_buffer *leaf;
3125         struct btrfs_dir_item *di;
3126         struct btrfs_key key;
3127         u64 index;
3128         int ret;
3129         u64 dir_ino = btrfs_ino(dir);
3130
3131         path = btrfs_alloc_path();
3132         if (!path)
3133                 return -ENOMEM;
3134
3135         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3136                                    name, name_len, -1);
3137         if (IS_ERR_OR_NULL(di)) {
3138                 if (!di)
3139                         ret = -ENOENT;
3140                 else
3141                         ret = PTR_ERR(di);
3142                 goto out;
3143         }
3144
3145         leaf = path->nodes[0];
3146         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3147         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3148         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3149         if (ret) {
3150                 btrfs_abort_transaction(trans, root, ret);
3151                 goto out;
3152         }
3153         btrfs_release_path(path);
3154
3155         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3156                                  objectid, root->root_key.objectid,
3157                                  dir_ino, &index, name, name_len);
3158         if (ret < 0) {
3159                 if (ret != -ENOENT) {
3160                         btrfs_abort_transaction(trans, root, ret);
3161                         goto out;
3162                 }
3163                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3164                                                  name, name_len);
3165                 if (IS_ERR_OR_NULL(di)) {
3166                         if (!di)
3167                                 ret = -ENOENT;
3168                         else
3169                                 ret = PTR_ERR(di);
3170                         btrfs_abort_transaction(trans, root, ret);
3171                         goto out;
3172                 }
3173
3174                 leaf = path->nodes[0];
3175                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3176                 btrfs_release_path(path);
3177                 index = key.offset;
3178         }
3179         btrfs_release_path(path);
3180
3181         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3182         if (ret) {
3183                 btrfs_abort_transaction(trans, root, ret);
3184                 goto out;
3185         }
3186
3187         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3188         inode_inc_iversion(dir);
3189         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3190         ret = btrfs_update_inode_fallback(trans, root, dir);
3191         if (ret)
3192                 btrfs_abort_transaction(trans, root, ret);
3193 out:
3194         btrfs_free_path(path);
3195         return ret;
3196 }
3197
3198 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3199 {
3200         struct inode *inode = dentry->d_inode;
3201         int err = 0;
3202         struct btrfs_root *root = BTRFS_I(dir)->root;
3203         struct btrfs_trans_handle *trans;
3204
3205         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3206                 return -ENOTEMPTY;
3207         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3208                 return -EPERM;
3209
3210         trans = __unlink_start_trans(dir, dentry);
3211         if (IS_ERR(trans))
3212                 return PTR_ERR(trans);
3213
3214         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3215                 err = btrfs_unlink_subvol(trans, root, dir,
3216                                           BTRFS_I(inode)->location.objectid,
3217                                           dentry->d_name.name,
3218                                           dentry->d_name.len);
3219                 goto out;
3220         }
3221
3222         err = btrfs_orphan_add(trans, inode);
3223         if (err)
3224                 goto out;
3225
3226         /* now the directory is empty */
3227         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3228                                  dentry->d_name.name, dentry->d_name.len);
3229         if (!err)
3230                 btrfs_i_size_write(inode, 0);
3231 out:
3232         __unlink_end_trans(trans, root);
3233         btrfs_btree_balance_dirty(root);
3234
3235         return err;
3236 }
3237
3238 /*
3239  * this can truncate away extent items, csum items and directory items.
3240  * It starts at a high offset and removes keys until it can't find
3241  * any higher than new_size
3242  *
3243  * csum items that cross the new i_size are truncated to the new size
3244  * as well.
3245  *
3246  * min_type is the minimum key type to truncate down to.  If set to 0, this
3247  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3248  */
3249 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3250                                struct btrfs_root *root,
3251                                struct inode *inode,
3252                                u64 new_size, u32 min_type)
3253 {
3254         struct btrfs_path *path;
3255         struct extent_buffer *leaf;
3256         struct btrfs_file_extent_item *fi;
3257         struct btrfs_key key;
3258         struct btrfs_key found_key;
3259         u64 extent_start = 0;
3260         u64 extent_num_bytes = 0;
3261         u64 extent_offset = 0;
3262         u64 item_end = 0;
3263         u64 mask = root->sectorsize - 1;
3264         u32 found_type = (u8)-1;
3265         int found_extent;
3266         int del_item;
3267         int pending_del_nr = 0;
3268         int pending_del_slot = 0;
3269         int extent_type = -1;
3270         int ret;
3271         int err = 0;
3272         u64 ino = btrfs_ino(inode);
3273
3274         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3275
3276         path = btrfs_alloc_path();
3277         if (!path)
3278                 return -ENOMEM;
3279         path->reada = -1;
3280
3281         /*
3282          * We want to drop from the next block forward in case this new size is
3283          * not block aligned since we will be keeping the last block of the
3284          * extent just the way it is.
3285          */
3286         if (root->ref_cows || root == root->fs_info->tree_root)
3287                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3288
3289         /*
3290          * This function is also used to drop the items in the log tree before
3291          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3292          * it is used to drop the loged items. So we shouldn't kill the delayed
3293          * items.
3294          */
3295         if (min_type == 0 && root == BTRFS_I(inode)->root)
3296                 btrfs_kill_delayed_inode_items(inode);
3297
3298         key.objectid = ino;
3299         key.offset = (u64)-1;
3300         key.type = (u8)-1;
3301
3302 search_again:
3303         path->leave_spinning = 1;
3304         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3305         if (ret < 0) {
3306                 err = ret;
3307                 goto out;
3308         }
3309
3310         if (ret > 0) {
3311                 /* there are no items in the tree for us to truncate, we're
3312                  * done
3313                  */
3314                 if (path->slots[0] == 0)
3315                         goto out;
3316                 path->slots[0]--;
3317         }
3318
3319         while (1) {
3320                 fi = NULL;
3321                 leaf = path->nodes[0];
3322                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3323                 found_type = btrfs_key_type(&found_key);
3324
3325                 if (found_key.objectid != ino)
3326                         break;
3327
3328                 if (found_type < min_type)
3329                         break;
3330
3331                 item_end = found_key.offset;
3332                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3333                         fi = btrfs_item_ptr(leaf, path->slots[0],
3334                                             struct btrfs_file_extent_item);
3335                         extent_type = btrfs_file_extent_type(leaf, fi);
3336                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3337                                 item_end +=
3338                                     btrfs_file_extent_num_bytes(leaf, fi);
3339                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3340                                 item_end += btrfs_file_extent_inline_len(leaf,
3341                                                                          fi);
3342                         }
3343                         item_end--;
3344                 }
3345                 if (found_type > min_type) {
3346                         del_item = 1;
3347                 } else {
3348                         if (item_end < new_size)
3349                                 break;
3350                         if (found_key.offset >= new_size)
3351                                 del_item = 1;
3352                         else
3353                                 del_item = 0;
3354                 }
3355                 found_extent = 0;
3356                 /* FIXME, shrink the extent if the ref count is only 1 */
3357                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3358                         goto delete;
3359
3360                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3361                         u64 num_dec;
3362                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3363                         if (!del_item) {
3364                                 u64 orig_num_bytes =
3365                                         btrfs_file_extent_num_bytes(leaf, fi);
3366                                 extent_num_bytes = new_size -
3367                                         found_key.offset + root->sectorsize - 1;
3368                                 extent_num_bytes = extent_num_bytes &
3369                                         ~((u64)root->sectorsize - 1);
3370                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3371                                                          extent_num_bytes);
3372                                 num_dec = (orig_num_bytes -
3373                                            extent_num_bytes);
3374                                 if (root->ref_cows && extent_start != 0)
3375                                         inode_sub_bytes(inode, num_dec);
3376                                 btrfs_mark_buffer_dirty(leaf);
3377                         } else {
3378                                 extent_num_bytes =
3379                                         btrfs_file_extent_disk_num_bytes(leaf,
3380                                                                          fi);
3381                                 extent_offset = found_key.offset -
3382                                         btrfs_file_extent_offset(leaf, fi);
3383
3384                                 /* FIXME blocksize != 4096 */
3385                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3386                                 if (extent_start != 0) {
3387                                         found_extent = 1;
3388                                         if (root->ref_cows)
3389                                                 inode_sub_bytes(inode, num_dec);
3390                                 }
3391                         }
3392                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3393                         /*
3394                          * we can't truncate inline items that have had
3395                          * special encodings
3396                          */
3397                         if (!del_item &&
3398                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3399                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3400                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3401                                 u32 size = new_size - found_key.offset;
3402
3403                                 if (root->ref_cows) {
3404                                         inode_sub_bytes(inode, item_end + 1 -
3405                                                         new_size);
3406                                 }
3407                                 size =
3408                                     btrfs_file_extent_calc_inline_size(size);
3409                                 btrfs_truncate_item(trans, root, path,
3410                                                     size, 1);
3411                         } else if (root->ref_cows) {
3412                                 inode_sub_bytes(inode, item_end + 1 -
3413                                                 found_key.offset);
3414                         }
3415                 }
3416 delete:
3417                 if (del_item) {
3418                         if (!pending_del_nr) {
3419                                 /* no pending yet, add ourselves */
3420                                 pending_del_slot = path->slots[0];
3421                                 pending_del_nr = 1;
3422                         } else if (pending_del_nr &&
3423                                    path->slots[0] + 1 == pending_del_slot) {
3424                                 /* hop on the pending chunk */
3425                                 pending_del_nr++;
3426                                 pending_del_slot = path->slots[0];
3427                         } else {
3428                                 BUG();
3429                         }
3430                 } else {
3431                         break;
3432                 }
3433                 if (found_extent && (root->ref_cows ||
3434                                      root == root->fs_info->tree_root)) {
3435                         btrfs_set_path_blocking(path);
3436                         ret = btrfs_free_extent(trans, root, extent_start,
3437                                                 extent_num_bytes, 0,
3438                                                 btrfs_header_owner(leaf),
3439                                                 ino, extent_offset, 0);
3440                         BUG_ON(ret);
3441                 }
3442
3443                 if (found_type == BTRFS_INODE_ITEM_KEY)
3444                         break;
3445
3446                 if (path->slots[0] == 0 ||
3447                     path->slots[0] != pending_del_slot) {
3448                         if (pending_del_nr) {
3449                                 ret = btrfs_del_items(trans, root, path,
3450                                                 pending_del_slot,
3451                                                 pending_del_nr);
3452                                 if (ret) {
3453                                         btrfs_abort_transaction(trans,
3454                                                                 root, ret);
3455                                         goto error;
3456                                 }
3457                                 pending_del_nr = 0;
3458                         }
3459                         btrfs_release_path(path);
3460                         goto search_again;
3461                 } else {
3462                         path->slots[0]--;
3463                 }
3464         }
3465 out:
3466         if (pending_del_nr) {
3467                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3468                                       pending_del_nr);
3469                 if (ret)
3470                         btrfs_abort_transaction(trans, root, ret);
3471         }
3472 error:
3473         btrfs_free_path(path);
3474         return err;
3475 }
3476
3477 /*
3478  * btrfs_truncate_page - read, zero a chunk and write a page
3479  * @inode - inode that we're zeroing
3480  * @from - the offset to start zeroing
3481  * @len - the length to zero, 0 to zero the entire range respective to the
3482  *      offset
3483  * @front - zero up to the offset instead of from the offset on
3484  *
3485  * This will find the page for the "from" offset and cow the page and zero the
3486  * part we want to zero.  This is used with truncate and hole punching.
3487  */
3488 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3489                         int front)
3490 {
3491         struct address_space *mapping = inode->i_mapping;
3492         struct btrfs_root *root = BTRFS_I(inode)->root;
3493         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494         struct btrfs_ordered_extent *ordered;
3495         struct extent_state *cached_state = NULL;
3496         char *kaddr;
3497         u32 blocksize = root->sectorsize;
3498         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3499         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3500         struct page *page;
3501         gfp_t mask = btrfs_alloc_write_mask(mapping);
3502         int ret = 0;
3503         u64 page_start;
3504         u64 page_end;
3505
3506         if ((offset & (blocksize - 1)) == 0 &&
3507             (!len || ((len & (blocksize - 1)) == 0)))
3508                 goto out;
3509         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3510         if (ret)
3511                 goto out;
3512
3513         ret = -ENOMEM;
3514 again:
3515         page = find_or_create_page(mapping, index, mask);
3516         if (!page) {
3517                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3518                 goto out;
3519         }
3520
3521         page_start = page_offset(page);
3522         page_end = page_start + PAGE_CACHE_SIZE - 1;
3523
3524         if (!PageUptodate(page)) {
3525                 ret = btrfs_readpage(NULL, page);
3526                 lock_page(page);
3527                 if (page->mapping != mapping) {
3528                         unlock_page(page);
3529                         page_cache_release(page);
3530                         goto again;
3531                 }
3532                 if (!PageUptodate(page)) {
3533                         ret = -EIO;
3534                         goto out_unlock;
3535                 }
3536         }
3537         wait_on_page_writeback(page);
3538
3539         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3540         set_page_extent_mapped(page);
3541
3542         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3543         if (ordered) {
3544                 unlock_extent_cached(io_tree, page_start, page_end,
3545                                      &cached_state, GFP_NOFS);
3546                 unlock_page(page);
3547                 page_cache_release(page);
3548                 btrfs_start_ordered_extent(inode, ordered, 1);
3549                 btrfs_put_ordered_extent(ordered);
3550                 goto again;
3551         }
3552
3553         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3554                           EXTENT_DIRTY | EXTENT_DELALLOC |
3555                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3556                           0, 0, &cached_state, GFP_NOFS);
3557
3558         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3559                                         &cached_state);
3560         if (ret) {
3561                 unlock_extent_cached(io_tree, page_start, page_end,
3562                                      &cached_state, GFP_NOFS);
3563                 goto out_unlock;
3564         }
3565
3566         ret = 0;
3567         if (offset != PAGE_CACHE_SIZE) {
3568                 if (!len)
3569                         len = PAGE_CACHE_SIZE - offset;
3570                 kaddr = kmap(page);
3571                 if (front)
3572                         memset(kaddr, 0, offset);
3573                 else
3574                         memset(kaddr + offset, 0, len);
3575                 flush_dcache_page(page);
3576                 kunmap(page);
3577         }
3578         ClearPageChecked(page);
3579         set_page_dirty(page);
3580         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3581                              GFP_NOFS);
3582
3583 out_unlock:
3584         if (ret)
3585                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3586         unlock_page(page);
3587         page_cache_release(page);
3588 out:
3589         return ret;
3590 }
3591
3592 /*
3593  * This function puts in dummy file extents for the area we're creating a hole
3594  * for.  So if we are truncating this file to a larger size we need to insert
3595  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3596  * the range between oldsize and size
3597  */
3598 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3599 {
3600         struct btrfs_trans_handle *trans;
3601         struct btrfs_root *root = BTRFS_I(inode)->root;
3602         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3603         struct extent_map *em = NULL;
3604         struct extent_state *cached_state = NULL;
3605         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3606         u64 mask = root->sectorsize - 1;
3607         u64 hole_start = (oldsize + mask) & ~mask;
3608         u64 block_end = (size + mask) & ~mask;
3609         u64 last_byte;
3610         u64 cur_offset;
3611         u64 hole_size;
3612         int err = 0;
3613
3614         if (size <= hole_start)
3615                 return 0;
3616
3617         while (1) {
3618                 struct btrfs_ordered_extent *ordered;
3619                 btrfs_wait_ordered_range(inode, hole_start,
3620                                          block_end - hole_start);
3621                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3622                                  &cached_state);
3623                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3624                 if (!ordered)
3625                         break;
3626                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3627                                      &cached_state, GFP_NOFS);
3628                 btrfs_put_ordered_extent(ordered);
3629         }
3630
3631         cur_offset = hole_start;
3632         while (1) {
3633                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3634                                 block_end - cur_offset, 0);
3635                 if (IS_ERR(em)) {
3636                         err = PTR_ERR(em);
3637                         break;
3638                 }
3639                 last_byte = min(extent_map_end(em), block_end);
3640                 last_byte = (last_byte + mask) & ~mask;
3641                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3642                         struct extent_map *hole_em;
3643                         hole_size = last_byte - cur_offset;
3644
3645                         trans = btrfs_start_transaction(root, 3);
3646                         if (IS_ERR(trans)) {
3647                                 err = PTR_ERR(trans);
3648                                 break;
3649                         }
3650
3651                         err = btrfs_drop_extents(trans, root, inode,
3652                                                  cur_offset,
3653                                                  cur_offset + hole_size, 1);
3654                         if (err) {
3655                                 btrfs_abort_transaction(trans, root, err);
3656                                 btrfs_end_transaction(trans, root);
3657                                 break;
3658                         }
3659
3660                         err = btrfs_insert_file_extent(trans, root,
3661                                         btrfs_ino(inode), cur_offset, 0,
3662                                         0, hole_size, 0, hole_size,
3663                                         0, 0, 0);
3664                         if (err) {
3665                                 btrfs_abort_transaction(trans, root, err);
3666                                 btrfs_end_transaction(trans, root);
3667                                 break;
3668                         }
3669
3670                         btrfs_drop_extent_cache(inode, cur_offset,
3671                                                 cur_offset + hole_size - 1, 0);
3672                         hole_em = alloc_extent_map();
3673                         if (!hole_em) {
3674                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3675                                         &BTRFS_I(inode)->runtime_flags);
3676                                 goto next;
3677                         }
3678                         hole_em->start = cur_offset;
3679                         hole_em->len = hole_size;
3680                         hole_em->orig_start = cur_offset;
3681
3682                         hole_em->block_start = EXTENT_MAP_HOLE;
3683                         hole_em->block_len = 0;
3684                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3685                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3686                         hole_em->generation = trans->transid;
3687
3688                         while (1) {
3689                                 write_lock(&em_tree->lock);
3690                                 err = add_extent_mapping(em_tree, hole_em);
3691                                 if (!err)
3692                                         list_move(&hole_em->list,
3693                                                   &em_tree->modified_extents);
3694                                 write_unlock(&em_tree->lock);
3695                                 if (err != -EEXIST)
3696                                         break;
3697                                 btrfs_drop_extent_cache(inode, cur_offset,
3698                                                         cur_offset +
3699                                                         hole_size - 1, 0);
3700                         }
3701                         free_extent_map(hole_em);
3702 next:
3703                         btrfs_update_inode(trans, root, inode);
3704                         btrfs_end_transaction(trans, root);
3705                 }
3706                 free_extent_map(em);
3707                 em = NULL;
3708                 cur_offset = last_byte;
3709                 if (cur_offset >= block_end)
3710                         break;
3711         }
3712
3713         free_extent_map(em);
3714         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3715                              GFP_NOFS);
3716         return err;
3717 }
3718
3719 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3720 {
3721         struct btrfs_root *root = BTRFS_I(inode)->root;
3722         struct btrfs_trans_handle *trans;
3723         loff_t oldsize = i_size_read(inode);
3724         int ret;
3725
3726         if (newsize == oldsize)
3727                 return 0;
3728
3729         if (newsize > oldsize) {
3730                 truncate_pagecache(inode, oldsize, newsize);
3731                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3732                 if (ret)
3733                         return ret;
3734
3735                 trans = btrfs_start_transaction(root, 1);
3736                 if (IS_ERR(trans))
3737                         return PTR_ERR(trans);
3738
3739                 i_size_write(inode, newsize);
3740                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3741                 ret = btrfs_update_inode(trans, root, inode);
3742                 btrfs_end_transaction(trans, root);
3743         } else {
3744
3745                 /*
3746                  * We're truncating a file that used to have good data down to
3747                  * zero. Make sure it gets into the ordered flush list so that
3748                  * any new writes get down to disk quickly.
3749                  */
3750                 if (newsize == 0)
3751                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3752                                 &BTRFS_I(inode)->runtime_flags);
3753
3754                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3755                 truncate_setsize(inode, newsize);
3756                 ret = btrfs_truncate(inode);
3757         }
3758
3759         return ret;
3760 }
3761
3762 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3763 {
3764         struct inode *inode = dentry->d_inode;
3765         struct btrfs_root *root = BTRFS_I(inode)->root;
3766         int err;
3767
3768         if (btrfs_root_readonly(root))
3769                 return -EROFS;
3770
3771         err = inode_change_ok(inode, attr);
3772         if (err)
3773                 return err;
3774
3775         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3776                 err = btrfs_setsize(inode, attr->ia_size);
3777                 if (err)
3778                         return err;
3779         }
3780
3781         if (attr->ia_valid) {
3782                 setattr_copy(inode, attr);
3783                 inode_inc_iversion(inode);
3784                 err = btrfs_dirty_inode(inode);
3785
3786                 if (!err && attr->ia_valid & ATTR_MODE)
3787                         err = btrfs_acl_chmod(inode);
3788         }
3789
3790         return err;
3791 }
3792
3793 void btrfs_evict_inode(struct inode *inode)
3794 {
3795         struct btrfs_trans_handle *trans;
3796         struct btrfs_root *root = BTRFS_I(inode)->root;
3797         struct btrfs_block_rsv *rsv, *global_rsv;
3798         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3799         int ret;
3800
3801         trace_btrfs_inode_evict(inode);
3802
3803         truncate_inode_pages(&inode->i_data, 0);
3804         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3805                                btrfs_is_free_space_inode(inode)))
3806                 goto no_delete;
3807
3808         if (is_bad_inode(inode)) {
3809                 btrfs_orphan_del(NULL, inode);
3810                 goto no_delete;
3811         }
3812         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3813         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3814
3815         if (root->fs_info->log_root_recovering) {
3816                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3817                                  &BTRFS_I(inode)->runtime_flags));
3818                 goto no_delete;
3819         }
3820
3821         if (inode->i_nlink > 0) {
3822                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3823                 goto no_delete;
3824         }
3825
3826         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3827         if (!rsv) {
3828                 btrfs_orphan_del(NULL, inode);
3829                 goto no_delete;
3830         }
3831         rsv->size = min_size;
3832         rsv->failfast = 1;
3833         global_rsv = &root->fs_info->global_block_rsv;
3834
3835         btrfs_i_size_write(inode, 0);
3836
3837         /*
3838          * This is a bit simpler than btrfs_truncate since we've already
3839          * reserved our space for our orphan item in the unlink, so we just
3840          * need to reserve some slack space in case we add bytes and update
3841          * inode item when doing the truncate.
3842          */
3843         while (1) {
3844                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3845                                              BTRFS_RESERVE_FLUSH_LIMIT);
3846
3847                 /*
3848                  * Try and steal from the global reserve since we will
3849                  * likely not use this space anyway, we want to try as
3850                  * hard as possible to get this to work.
3851                  */
3852                 if (ret)
3853                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3854
3855                 if (ret) {
3856                         printk(KERN_WARNING "Could not get space for a "
3857                                "delete, will truncate on mount %d\n", ret);
3858                         btrfs_orphan_del(NULL, inode);
3859                         btrfs_free_block_rsv(root, rsv);
3860                         goto no_delete;
3861                 }
3862
3863                 trans = btrfs_start_transaction_lflush(root, 1);
3864                 if (IS_ERR(trans)) {
3865                         btrfs_orphan_del(NULL, inode);
3866                         btrfs_free_block_rsv(root, rsv);
3867                         goto no_delete;
3868                 }
3869
3870                 trans->block_rsv = rsv;
3871
3872                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3873                 if (ret != -ENOSPC)
3874                         break;
3875
3876                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3877                 ret = btrfs_update_inode(trans, root, inode);
3878                 BUG_ON(ret);
3879
3880                 btrfs_end_transaction(trans, root);
3881                 trans = NULL;
3882                 btrfs_btree_balance_dirty(root);
3883         }
3884
3885         btrfs_free_block_rsv(root, rsv);
3886
3887         if (ret == 0) {
3888                 trans->block_rsv = root->orphan_block_rsv;
3889                 ret = btrfs_orphan_del(trans, inode);
3890                 BUG_ON(ret);
3891         }
3892
3893         trans->block_rsv = &root->fs_info->trans_block_rsv;
3894         if (!(root == root->fs_info->tree_root ||
3895               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3896                 btrfs_return_ino(root, btrfs_ino(inode));
3897
3898         btrfs_end_transaction(trans, root);
3899         btrfs_btree_balance_dirty(root);
3900 no_delete:
3901         clear_inode(inode);
3902         return;
3903 }
3904
3905 /*
3906  * this returns the key found in the dir entry in the location pointer.
3907  * If no dir entries were found, location->objectid is 0.
3908  */
3909 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3910                                struct btrfs_key *location)
3911 {
3912         const char *name = dentry->d_name.name;
3913         int namelen = dentry->d_name.len;
3914         struct btrfs_dir_item *di;
3915         struct btrfs_path *path;
3916         struct btrfs_root *root = BTRFS_I(dir)->root;
3917         int ret = 0;
3918
3919         path = btrfs_alloc_path();
3920         if (!path)
3921                 return -ENOMEM;
3922
3923         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3924                                     namelen, 0);
3925         if (IS_ERR(di))
3926                 ret = PTR_ERR(di);
3927
3928         if (IS_ERR_OR_NULL(di))
3929                 goto out_err;
3930
3931         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3932 out:
3933         btrfs_free_path(path);
3934         return ret;
3935 out_err:
3936         location->objectid = 0;
3937         goto out;
3938 }
3939
3940 /*
3941  * when we hit a tree root in a directory, the btrfs part of the inode
3942  * needs to be changed to reflect the root directory of the tree root.  This
3943  * is kind of like crossing a mount point.
3944  */
3945 static int fixup_tree_root_location(struct btrfs_root *root,
3946                                     struct inode *dir,
3947                                     struct dentry *dentry,
3948                                     struct btrfs_key *location,
3949                                     struct btrfs_root **sub_root)
3950 {
3951         struct btrfs_path *path;
3952         struct btrfs_root *new_root;
3953         struct btrfs_root_ref *ref;
3954         struct extent_buffer *leaf;
3955         int ret;
3956         int err = 0;
3957
3958         path = btrfs_alloc_path();
3959         if (!path) {
3960                 err = -ENOMEM;
3961                 goto out;
3962         }
3963
3964         err = -ENOENT;
3965         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3966                                   BTRFS_I(dir)->root->root_key.objectid,
3967                                   location->objectid);
3968         if (ret) {
3969                 if (ret < 0)
3970                         err = ret;
3971                 goto out;
3972         }
3973
3974         leaf = path->nodes[0];
3975         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3976         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3977             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3978                 goto out;
3979
3980         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3981                                    (unsigned long)(ref + 1),
3982                                    dentry->d_name.len);
3983         if (ret)
3984                 goto out;
3985
3986         btrfs_release_path(path);
3987
3988         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3989         if (IS_ERR(new_root)) {
3990                 err = PTR_ERR(new_root);
3991                 goto out;
3992         }
3993
3994         if (btrfs_root_refs(&new_root->root_item) == 0) {
3995                 err = -ENOENT;
3996                 goto out;
3997         }
3998
3999         *sub_root = new_root;
4000         location->objectid = btrfs_root_dirid(&new_root->root_item);
4001         location->type = BTRFS_INODE_ITEM_KEY;
4002         location->offset = 0;
4003         err = 0;
4004 out:
4005         btrfs_free_path(path);
4006         return err;
4007 }
4008
4009 static void inode_tree_add(struct inode *inode)
4010 {
4011         struct btrfs_root *root = BTRFS_I(inode)->root;
4012         struct btrfs_inode *entry;
4013         struct rb_node **p;
4014         struct rb_node *parent;
4015         u64 ino = btrfs_ino(inode);
4016 again:
4017         p = &root->inode_tree.rb_node;
4018         parent = NULL;
4019
4020         if (inode_unhashed(inode))
4021                 return;
4022
4023         spin_lock(&root->inode_lock);
4024         while (*p) {
4025                 parent = *p;
4026                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4027
4028                 if (ino < btrfs_ino(&entry->vfs_inode))
4029                         p = &parent->rb_left;
4030                 else if (ino > btrfs_ino(&entry->vfs_inode))
4031                         p = &parent->rb_right;
4032                 else {
4033                         WARN_ON(!(entry->vfs_inode.i_state &
4034                                   (I_WILL_FREE | I_FREEING)));
4035                         rb_erase(parent, &root->inode_tree);
4036                         RB_CLEAR_NODE(parent);
4037                         spin_unlock(&root->inode_lock);
4038                         goto again;
4039                 }
4040         }
4041         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4042         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4043         spin_unlock(&root->inode_lock);
4044 }
4045
4046 static void inode_tree_del(struct inode *inode)
4047 {
4048         struct btrfs_root *root = BTRFS_I(inode)->root;
4049         int empty = 0;
4050
4051         spin_lock(&root->inode_lock);
4052         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4053                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4054                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4055                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4056         }
4057         spin_unlock(&root->inode_lock);
4058
4059         /*
4060          * Free space cache has inodes in the tree root, but the tree root has a
4061          * root_refs of 0, so this could end up dropping the tree root as a
4062          * snapshot, so we need the extra !root->fs_info->tree_root check to
4063          * make sure we don't drop it.
4064          */
4065         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4066             root != root->fs_info->tree_root) {
4067                 synchronize_srcu(&root->fs_info->subvol_srcu);
4068                 spin_lock(&root->inode_lock);
4069                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4070                 spin_unlock(&root->inode_lock);
4071                 if (empty)
4072                         btrfs_add_dead_root(root);
4073         }
4074 }
4075
4076 void btrfs_invalidate_inodes(struct btrfs_root *root)
4077 {
4078         struct rb_node *node;
4079         struct rb_node *prev;
4080         struct btrfs_inode *entry;
4081         struct inode *inode;
4082         u64 objectid = 0;
4083
4084         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4085
4086         spin_lock(&root->inode_lock);
4087 again:
4088         node = root->inode_tree.rb_node;
4089         prev = NULL;
4090         while (node) {
4091                 prev = node;
4092                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4093
4094                 if (objectid < btrfs_ino(&entry->vfs_inode))
4095                         node = node->rb_left;
4096                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4097                         node = node->rb_right;
4098                 else
4099                         break;
4100         }
4101         if (!node) {
4102                 while (prev) {
4103                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4104                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4105                                 node = prev;
4106                                 break;
4107                         }
4108                         prev = rb_next(prev);
4109                 }
4110         }
4111         while (node) {
4112                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4113                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4114                 inode = igrab(&entry->vfs_inode);
4115                 if (inode) {
4116                         spin_unlock(&root->inode_lock);
4117                         if (atomic_read(&inode->i_count) > 1)
4118                                 d_prune_aliases(inode);
4119                         /*
4120                          * btrfs_drop_inode will have it removed from
4121                          * the inode cache when its usage count
4122                          * hits zero.
4123                          */
4124                         iput(inode);
4125                         cond_resched();
4126                         spin_lock(&root->inode_lock);
4127                         goto again;
4128                 }
4129
4130                 if (cond_resched_lock(&root->inode_lock))
4131                         goto again;
4132
4133                 node = rb_next(node);
4134         }
4135         spin_unlock(&root->inode_lock);
4136 }
4137
4138 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4139 {
4140         struct btrfs_iget_args *args = p;
4141         inode->i_ino = args->ino;
4142         BTRFS_I(inode)->root = args->root;
4143         return 0;
4144 }
4145
4146 static int btrfs_find_actor(struct inode *inode, void *opaque)
4147 {
4148         struct btrfs_iget_args *args = opaque;
4149         return args->ino == btrfs_ino(inode) &&
4150                 args->root == BTRFS_I(inode)->root;
4151 }
4152
4153 static struct inode *btrfs_iget_locked(struct super_block *s,
4154                                        u64 objectid,
4155                                        struct btrfs_root *root)
4156 {
4157         struct inode *inode;
4158         struct btrfs_iget_args args;
4159         args.ino = objectid;
4160         args.root = root;
4161
4162         inode = iget5_locked(s, objectid, btrfs_find_actor,
4163                              btrfs_init_locked_inode,
4164                              (void *)&args);
4165         return inode;
4166 }
4167
4168 /* Get an inode object given its location and corresponding root.
4169  * Returns in *is_new if the inode was read from disk
4170  */
4171 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4172                          struct btrfs_root *root, int *new)
4173 {
4174         struct inode *inode;
4175
4176         inode = btrfs_iget_locked(s, location->objectid, root);
4177         if (!inode)
4178                 return ERR_PTR(-ENOMEM);
4179
4180         if (inode->i_state & I_NEW) {
4181                 BTRFS_I(inode)->root = root;
4182                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4183                 btrfs_read_locked_inode(inode);
4184                 if (!is_bad_inode(inode)) {
4185                         inode_tree_add(inode);
4186                         unlock_new_inode(inode);
4187                         if (new)
4188                                 *new = 1;
4189                 } else {
4190                         unlock_new_inode(inode);
4191                         iput(inode);
4192                         inode = ERR_PTR(-ESTALE);
4193                 }
4194         }
4195
4196         return inode;
4197 }
4198
4199 static struct inode *new_simple_dir(struct super_block *s,
4200                                     struct btrfs_key *key,
4201                                     struct btrfs_root *root)
4202 {
4203         struct inode *inode = new_inode(s);
4204
4205         if (!inode)
4206                 return ERR_PTR(-ENOMEM);
4207
4208         BTRFS_I(inode)->root = root;
4209         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4210         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4211
4212         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4213         inode->i_op = &btrfs_dir_ro_inode_operations;
4214         inode->i_fop = &simple_dir_operations;
4215         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4216         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4217
4218         return inode;
4219 }
4220
4221 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4222 {
4223         struct inode *inode;
4224         struct btrfs_root *root = BTRFS_I(dir)->root;
4225         struct btrfs_root *sub_root = root;
4226         struct btrfs_key location;
4227         int index;
4228         int ret = 0;
4229
4230         if (dentry->d_name.len > BTRFS_NAME_LEN)
4231                 return ERR_PTR(-ENAMETOOLONG);
4232
4233         if (unlikely(d_need_lookup(dentry))) {
4234                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4235                 kfree(dentry->d_fsdata);
4236                 dentry->d_fsdata = NULL;
4237                 /* This thing is hashed, drop it for now */
4238                 d_drop(dentry);
4239         } else {
4240                 ret = btrfs_inode_by_name(dir, dentry, &location);
4241         }
4242
4243         if (ret < 0)
4244                 return ERR_PTR(ret);
4245
4246         if (location.objectid == 0)
4247                 return NULL;
4248
4249         if (location.type == BTRFS_INODE_ITEM_KEY) {
4250                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4251                 return inode;
4252         }
4253
4254         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4255
4256         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4257         ret = fixup_tree_root_location(root, dir, dentry,
4258                                        &location, &sub_root);
4259         if (ret < 0) {
4260                 if (ret != -ENOENT)
4261                         inode = ERR_PTR(ret);
4262                 else
4263                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4264         } else {
4265                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4266         }
4267         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4268
4269         if (!IS_ERR(inode) && root != sub_root) {
4270                 down_read(&root->fs_info->cleanup_work_sem);
4271                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4272                         ret = btrfs_orphan_cleanup(sub_root);
4273                 up_read(&root->fs_info->cleanup_work_sem);
4274                 if (ret)
4275                         inode = ERR_PTR(ret);
4276         }
4277
4278         return inode;
4279 }
4280
4281 static int btrfs_dentry_delete(const struct dentry *dentry)
4282 {
4283         struct btrfs_root *root;
4284         struct inode *inode = dentry->d_inode;
4285
4286         if (!inode && !IS_ROOT(dentry))
4287                 inode = dentry->d_parent->d_inode;
4288
4289         if (inode) {
4290                 root = BTRFS_I(inode)->root;
4291                 if (btrfs_root_refs(&root->root_item) == 0)
4292                         return 1;
4293
4294                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4295                         return 1;
4296         }
4297         return 0;
4298 }
4299
4300 static void btrfs_dentry_release(struct dentry *dentry)
4301 {
4302         if (dentry->d_fsdata)
4303                 kfree(dentry->d_fsdata);
4304 }
4305
4306 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4307                                    unsigned int flags)
4308 {
4309         struct dentry *ret;
4310
4311         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4312         if (unlikely(d_need_lookup(dentry))) {
4313                 spin_lock(&dentry->d_lock);
4314                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4315                 spin_unlock(&dentry->d_lock);
4316         }
4317         return ret;
4318 }
4319
4320 unsigned char btrfs_filetype_table[] = {
4321         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4322 };
4323
4324 static int btrfs_real_readdir(struct file *filp, void *dirent,
4325                               filldir_t filldir)
4326 {
4327         struct inode *inode = filp->f_dentry->d_inode;
4328         struct btrfs_root *root = BTRFS_I(inode)->root;
4329         struct btrfs_item *item;
4330         struct btrfs_dir_item *di;
4331         struct btrfs_key key;
4332         struct btrfs_key found_key;
4333         struct btrfs_path *path;
4334         struct list_head ins_list;
4335         struct list_head del_list;
4336         int ret;
4337         struct extent_buffer *leaf;
4338         int slot;
4339         unsigned char d_type;
4340         int over = 0;
4341         u32 di_cur;
4342         u32 di_total;
4343         u32 di_len;
4344         int key_type = BTRFS_DIR_INDEX_KEY;
4345         char tmp_name[32];
4346         char *name_ptr;
4347         int name_len;
4348         int is_curr = 0;        /* filp->f_pos points to the current index? */
4349
4350         /* FIXME, use a real flag for deciding about the key type */
4351         if (root->fs_info->tree_root == root)
4352                 key_type = BTRFS_DIR_ITEM_KEY;
4353
4354         /* special case for "." */
4355         if (filp->f_pos == 0) {
4356                 over = filldir(dirent, ".", 1,
4357                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4358                 if (over)
4359                         return 0;
4360                 filp->f_pos = 1;
4361         }
4362         /* special case for .., just use the back ref */
4363         if (filp->f_pos == 1) {
4364                 u64 pino = parent_ino(filp->f_path.dentry);
4365                 over = filldir(dirent, "..", 2,
4366                                filp->f_pos, pino, DT_DIR);
4367                 if (over)
4368                         return 0;
4369                 filp->f_pos = 2;
4370         }
4371         path = btrfs_alloc_path();
4372         if (!path)
4373                 return -ENOMEM;
4374
4375         path->reada = 1;
4376
4377         if (key_type == BTRFS_DIR_INDEX_KEY) {
4378                 INIT_LIST_HEAD(&ins_list);
4379                 INIT_LIST_HEAD(&del_list);
4380                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4381         }
4382
4383         btrfs_set_key_type(&key, key_type);
4384         key.offset = filp->f_pos;
4385         key.objectid = btrfs_ino(inode);
4386
4387         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4388         if (ret < 0)
4389                 goto err;
4390
4391         while (1) {
4392                 leaf = path->nodes[0];
4393                 slot = path->slots[0];
4394                 if (slot >= btrfs_header_nritems(leaf)) {
4395                         ret = btrfs_next_leaf(root, path);
4396                         if (ret < 0)
4397                                 goto err;
4398                         else if (ret > 0)
4399                                 break;
4400                         continue;
4401                 }
4402
4403                 item = btrfs_item_nr(leaf, slot);
4404                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4405
4406                 if (found_key.objectid != key.objectid)
4407                         break;
4408                 if (btrfs_key_type(&found_key) != key_type)
4409                         break;
4410                 if (found_key.offset < filp->f_pos)
4411                         goto next;
4412                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4413                     btrfs_should_delete_dir_index(&del_list,
4414                                                   found_key.offset))
4415                         goto next;
4416
4417                 filp->f_pos = found_key.offset;
4418                 is_curr = 1;
4419
4420                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4421                 di_cur = 0;
4422                 di_total = btrfs_item_size(leaf, item);
4423
4424                 while (di_cur < di_total) {
4425                         struct btrfs_key location;
4426
4427                         if (verify_dir_item(root, leaf, di))
4428                                 break;
4429
4430                         name_len = btrfs_dir_name_len(leaf, di);
4431                         if (name_len <= sizeof(tmp_name)) {
4432                                 name_ptr = tmp_name;
4433                         } else {
4434                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4435                                 if (!name_ptr) {
4436                                         ret = -ENOMEM;
4437                                         goto err;
4438                                 }
4439                         }
4440                         read_extent_buffer(leaf, name_ptr,
4441                                            (unsigned long)(di + 1), name_len);
4442
4443                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4444                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4445
4446
4447                         /* is this a reference to our own snapshot? If so
4448                          * skip it.
4449                          *
4450                          * In contrast to old kernels, we insert the snapshot's
4451                          * dir item and dir index after it has been created, so
4452                          * we won't find a reference to our own snapshot. We
4453                          * still keep the following code for backward
4454                          * compatibility.
4455                          */
4456                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4457                             location.objectid == root->root_key.objectid) {
4458                                 over = 0;
4459                                 goto skip;
4460                         }
4461                         over = filldir(dirent, name_ptr, name_len,
4462                                        found_key.offset, location.objectid,
4463                                        d_type);
4464
4465 skip:
4466                         if (name_ptr != tmp_name)
4467                                 kfree(name_ptr);
4468
4469                         if (over)
4470                                 goto nopos;
4471                         di_len = btrfs_dir_name_len(leaf, di) +
4472                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4473                         di_cur += di_len;
4474                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4475                 }
4476 next:
4477                 path->slots[0]++;
4478         }
4479
4480         if (key_type == BTRFS_DIR_INDEX_KEY) {
4481                 if (is_curr)
4482                         filp->f_pos++;
4483                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4484                                                       &ins_list);
4485                 if (ret)
4486                         goto nopos;
4487         }
4488
4489         /* Reached end of directory/root. Bump pos past the last item. */
4490         if (key_type == BTRFS_DIR_INDEX_KEY)
4491                 /*
4492                  * 32-bit glibc will use getdents64, but then strtol -
4493                  * so the last number we can serve is this.
4494                  */
4495                 filp->f_pos = 0x7fffffff;
4496         else
4497                 filp->f_pos++;
4498 nopos:
4499         ret = 0;
4500 err:
4501         if (key_type == BTRFS_DIR_INDEX_KEY)
4502                 btrfs_put_delayed_items(&ins_list, &del_list);
4503         btrfs_free_path(path);
4504         return ret;
4505 }
4506
4507 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4508 {
4509         struct btrfs_root *root = BTRFS_I(inode)->root;
4510         struct btrfs_trans_handle *trans;
4511         int ret = 0;
4512         bool nolock = false;
4513
4514         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4515                 return 0;
4516
4517         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4518                 nolock = true;
4519
4520         if (wbc->sync_mode == WB_SYNC_ALL) {
4521                 if (nolock)
4522                         trans = btrfs_join_transaction_nolock(root);
4523                 else
4524                         trans = btrfs_join_transaction(root);
4525                 if (IS_ERR(trans))
4526                         return PTR_ERR(trans);
4527                 ret = btrfs_commit_transaction(trans, root);
4528         }
4529         return ret;
4530 }
4531
4532 /*
4533  * This is somewhat expensive, updating the tree every time the
4534  * inode changes.  But, it is most likely to find the inode in cache.
4535  * FIXME, needs more benchmarking...there are no reasons other than performance
4536  * to keep or drop this code.
4537  */
4538 int btrfs_dirty_inode(struct inode *inode)
4539 {
4540         struct btrfs_root *root = BTRFS_I(inode)->root;
4541         struct btrfs_trans_handle *trans;
4542         int ret;
4543
4544         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4545                 return 0;
4546
4547         trans = btrfs_join_transaction(root);
4548         if (IS_ERR(trans))
4549                 return PTR_ERR(trans);
4550
4551         ret = btrfs_update_inode(trans, root, inode);
4552         if (ret && ret == -ENOSPC) {
4553                 /* whoops, lets try again with the full transaction */
4554                 btrfs_end_transaction(trans, root);
4555                 trans = btrfs_start_transaction(root, 1);
4556                 if (IS_ERR(trans))
4557                         return PTR_ERR(trans);
4558
4559                 ret = btrfs_update_inode(trans, root, inode);
4560         }
4561         btrfs_end_transaction(trans, root);
4562         if (BTRFS_I(inode)->delayed_node)
4563                 btrfs_balance_delayed_items(root);
4564
4565         return ret;
4566 }
4567
4568 /*
4569  * This is a copy of file_update_time.  We need this so we can return error on
4570  * ENOSPC for updating the inode in the case of file write and mmap writes.
4571  */
4572 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4573                              int flags)
4574 {
4575         struct btrfs_root *root = BTRFS_I(inode)->root;
4576
4577         if (btrfs_root_readonly(root))
4578                 return -EROFS;
4579
4580         if (flags & S_VERSION)
4581                 inode_inc_iversion(inode);
4582         if (flags & S_CTIME)
4583                 inode->i_ctime = *now;
4584         if (flags & S_MTIME)
4585                 inode->i_mtime = *now;
4586         if (flags & S_ATIME)
4587                 inode->i_atime = *now;
4588         return btrfs_dirty_inode(inode);
4589 }
4590
4591 /*
4592  * find the highest existing sequence number in a directory
4593  * and then set the in-memory index_cnt variable to reflect
4594  * free sequence numbers
4595  */
4596 static int btrfs_set_inode_index_count(struct inode *inode)
4597 {
4598         struct btrfs_root *root = BTRFS_I(inode)->root;
4599         struct btrfs_key key, found_key;
4600         struct btrfs_path *path;
4601         struct extent_buffer *leaf;
4602         int ret;
4603
4604         key.objectid = btrfs_ino(inode);
4605         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4606         key.offset = (u64)-1;
4607
4608         path = btrfs_alloc_path();
4609         if (!path)
4610                 return -ENOMEM;
4611
4612         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4613         if (ret < 0)
4614                 goto out;
4615         /* FIXME: we should be able to handle this */
4616         if (ret == 0)
4617                 goto out;
4618         ret = 0;
4619
4620         /*
4621          * MAGIC NUMBER EXPLANATION:
4622          * since we search a directory based on f_pos we have to start at 2
4623          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4624          * else has to start at 2
4625          */
4626         if (path->slots[0] == 0) {
4627                 BTRFS_I(inode)->index_cnt = 2;
4628                 goto out;
4629         }
4630
4631         path->slots[0]--;
4632
4633         leaf = path->nodes[0];
4634         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4635
4636         if (found_key.objectid != btrfs_ino(inode) ||
4637             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4638                 BTRFS_I(inode)->index_cnt = 2;
4639                 goto out;
4640         }
4641
4642         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4643 out:
4644         btrfs_free_path(path);
4645         return ret;
4646 }
4647
4648 /*
4649  * helper to find a free sequence number in a given directory.  This current
4650  * code is very simple, later versions will do smarter things in the btree
4651  */
4652 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4653 {
4654         int ret = 0;
4655
4656         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4657                 ret = btrfs_inode_delayed_dir_index_count(dir);
4658                 if (ret) {
4659                         ret = btrfs_set_inode_index_count(dir);
4660                         if (ret)
4661                                 return ret;
4662                 }
4663         }
4664
4665         *index = BTRFS_I(dir)->index_cnt;
4666         BTRFS_I(dir)->index_cnt++;
4667
4668         return ret;
4669 }
4670
4671 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4672                                      struct btrfs_root *root,
4673                                      struct inode *dir,
4674                                      const char *name, int name_len,
4675                                      u64 ref_objectid, u64 objectid,
4676                                      umode_t mode, u64 *index)
4677 {
4678         struct inode *inode;
4679         struct btrfs_inode_item *inode_item;
4680         struct btrfs_key *location;
4681         struct btrfs_path *path;
4682         struct btrfs_inode_ref *ref;
4683         struct btrfs_key key[2];
4684         u32 sizes[2];
4685         unsigned long ptr;
4686         int ret;
4687         int owner;
4688
4689         path = btrfs_alloc_path();
4690         if (!path)
4691                 return ERR_PTR(-ENOMEM);
4692
4693         inode = new_inode(root->fs_info->sb);
4694         if (!inode) {
4695                 btrfs_free_path(path);
4696                 return ERR_PTR(-ENOMEM);
4697         }
4698
4699         /*
4700          * we have to initialize this early, so we can reclaim the inode
4701          * number if we fail afterwards in this function.
4702          */
4703         inode->i_ino = objectid;
4704
4705         if (dir) {
4706                 trace_btrfs_inode_request(dir);
4707
4708                 ret = btrfs_set_inode_index(dir, index);
4709                 if (ret) {
4710                         btrfs_free_path(path);
4711                         iput(inode);
4712                         return ERR_PTR(ret);
4713                 }
4714         }
4715         /*
4716          * index_cnt is ignored for everything but a dir,
4717          * btrfs_get_inode_index_count has an explanation for the magic
4718          * number
4719          */
4720         BTRFS_I(inode)->index_cnt = 2;
4721         BTRFS_I(inode)->root = root;
4722         BTRFS_I(inode)->generation = trans->transid;
4723         inode->i_generation = BTRFS_I(inode)->generation;
4724
4725         /*
4726          * We could have gotten an inode number from somebody who was fsynced
4727          * and then removed in this same transaction, so let's just set full
4728          * sync since it will be a full sync anyway and this will blow away the
4729          * old info in the log.
4730          */
4731         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4732
4733         if (S_ISDIR(mode))
4734                 owner = 0;
4735         else
4736                 owner = 1;
4737
4738         key[0].objectid = objectid;
4739         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4740         key[0].offset = 0;
4741
4742         /*
4743          * Start new inodes with an inode_ref. This is slightly more
4744          * efficient for small numbers of hard links since they will
4745          * be packed into one item. Extended refs will kick in if we
4746          * add more hard links than can fit in the ref item.
4747          */
4748         key[1].objectid = objectid;
4749         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4750         key[1].offset = ref_objectid;
4751
4752         sizes[0] = sizeof(struct btrfs_inode_item);
4753         sizes[1] = name_len + sizeof(*ref);
4754
4755         path->leave_spinning = 1;
4756         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4757         if (ret != 0)
4758                 goto fail;
4759
4760         inode_init_owner(inode, dir, mode);
4761         inode_set_bytes(inode, 0);
4762         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4763         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4764                                   struct btrfs_inode_item);
4765         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4766                              sizeof(*inode_item));
4767         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4768
4769         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4770                              struct btrfs_inode_ref);
4771         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4772         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4773         ptr = (unsigned long)(ref + 1);
4774         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4775
4776         btrfs_mark_buffer_dirty(path->nodes[0]);
4777         btrfs_free_path(path);
4778
4779         location = &BTRFS_I(inode)->location;
4780         location->objectid = objectid;
4781         location->offset = 0;
4782         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4783
4784         btrfs_inherit_iflags(inode, dir);
4785
4786         if (S_ISREG(mode)) {
4787                 if (btrfs_test_opt(root, NODATASUM))
4788                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4789                 if (btrfs_test_opt(root, NODATACOW) ||
4790                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4791                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4792         }
4793
4794         insert_inode_hash(inode);
4795         inode_tree_add(inode);
4796
4797         trace_btrfs_inode_new(inode);
4798         btrfs_set_inode_last_trans(trans, inode);
4799
4800         btrfs_update_root_times(trans, root);
4801
4802         return inode;
4803 fail:
4804         if (dir)
4805                 BTRFS_I(dir)->index_cnt--;
4806         btrfs_free_path(path);
4807         iput(inode);
4808         return ERR_PTR(ret);
4809 }
4810
4811 static inline u8 btrfs_inode_type(struct inode *inode)
4812 {
4813         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4814 }
4815
4816 /*
4817  * utility function to add 'inode' into 'parent_inode' with
4818  * a give name and a given sequence number.
4819  * if 'add_backref' is true, also insert a backref from the
4820  * inode to the parent directory.
4821  */
4822 int btrfs_add_link(struct btrfs_trans_handle *trans,
4823                    struct inode *parent_inode, struct inode *inode,
4824                    const char *name, int name_len, int add_backref, u64 index)
4825 {
4826         int ret = 0;
4827         struct btrfs_key key;
4828         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4829         u64 ino = btrfs_ino(inode);
4830         u64 parent_ino = btrfs_ino(parent_inode);
4831
4832         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4833                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4834         } else {
4835                 key.objectid = ino;
4836                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4837                 key.offset = 0;
4838         }
4839
4840         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4841                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4842                                          key.objectid, root->root_key.objectid,
4843                                          parent_ino, index, name, name_len);
4844         } else if (add_backref) {
4845                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4846                                              parent_ino, index);
4847         }
4848
4849         /* Nothing to clean up yet */
4850         if (ret)
4851                 return ret;
4852
4853         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4854                                     parent_inode, &key,
4855                                     btrfs_inode_type(inode), index);
4856         if (ret == -EEXIST)
4857                 goto fail_dir_item;
4858         else if (ret) {
4859                 btrfs_abort_transaction(trans, root, ret);
4860                 return ret;
4861         }
4862
4863         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4864                            name_len * 2);
4865         inode_inc_iversion(parent_inode);
4866         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4867         ret = btrfs_update_inode(trans, root, parent_inode);
4868         if (ret)
4869                 btrfs_abort_transaction(trans, root, ret);
4870         return ret;
4871
4872 fail_dir_item:
4873         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4874                 u64 local_index;
4875                 int err;
4876                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4877                                  key.objectid, root->root_key.objectid,
4878                                  parent_ino, &local_index, name, name_len);
4879
4880         } else if (add_backref) {
4881                 u64 local_index;
4882                 int err;
4883
4884                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4885                                           ino, parent_ino, &local_index);
4886         }
4887         return ret;
4888 }
4889
4890 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4891                             struct inode *dir, struct dentry *dentry,
4892                             struct inode *inode, int backref, u64 index)
4893 {
4894         int err = btrfs_add_link(trans, dir, inode,
4895                                  dentry->d_name.name, dentry->d_name.len,
4896                                  backref, index);
4897         if (err > 0)
4898                 err = -EEXIST;
4899         return err;
4900 }
4901
4902 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4903                         umode_t mode, dev_t rdev)
4904 {
4905         struct btrfs_trans_handle *trans;
4906         struct btrfs_root *root = BTRFS_I(dir)->root;
4907         struct inode *inode = NULL;
4908         int err;
4909         int drop_inode = 0;
4910         u64 objectid;
4911         u64 index = 0;
4912
4913         if (!new_valid_dev(rdev))
4914                 return -EINVAL;
4915
4916         /*
4917          * 2 for inode item and ref
4918          * 2 for dir items
4919          * 1 for xattr if selinux is on
4920          */
4921         trans = btrfs_start_transaction(root, 5);
4922         if (IS_ERR(trans))
4923                 return PTR_ERR(trans);
4924
4925         err = btrfs_find_free_ino(root, &objectid);
4926         if (err)
4927                 goto out_unlock;
4928
4929         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4930                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4931                                 mode, &index);
4932         if (IS_ERR(inode)) {
4933                 err = PTR_ERR(inode);
4934                 goto out_unlock;
4935         }
4936
4937         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4938         if (err) {
4939                 drop_inode = 1;
4940                 goto out_unlock;
4941         }
4942
4943         /*
4944         * If the active LSM wants to access the inode during
4945         * d_instantiate it needs these. Smack checks to see
4946         * if the filesystem supports xattrs by looking at the
4947         * ops vector.
4948         */
4949
4950         inode->i_op = &btrfs_special_inode_operations;
4951         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4952         if (err)
4953                 drop_inode = 1;
4954         else {
4955                 init_special_inode(inode, inode->i_mode, rdev);
4956                 btrfs_update_inode(trans, root, inode);
4957                 d_instantiate(dentry, inode);
4958         }
4959 out_unlock:
4960         btrfs_end_transaction(trans, root);
4961         btrfs_btree_balance_dirty(root);
4962         if (drop_inode) {
4963                 inode_dec_link_count(inode);
4964                 iput(inode);
4965         }
4966         return err;
4967 }
4968
4969 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4970                         umode_t mode, bool excl)
4971 {
4972         struct btrfs_trans_handle *trans;
4973         struct btrfs_root *root = BTRFS_I(dir)->root;
4974         struct inode *inode = NULL;
4975         int drop_inode = 0;
4976         int err;
4977         u64 objectid;
4978         u64 index = 0;
4979
4980         /*
4981          * 2 for inode item and ref
4982          * 2 for dir items
4983          * 1 for xattr if selinux is on
4984          */
4985         trans = btrfs_start_transaction(root, 5);
4986         if (IS_ERR(trans))
4987                 return PTR_ERR(trans);
4988
4989         err = btrfs_find_free_ino(root, &objectid);
4990         if (err)
4991                 goto out_unlock;
4992
4993         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4994                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4995                                 mode, &index);
4996         if (IS_ERR(inode)) {
4997                 err = PTR_ERR(inode);
4998                 goto out_unlock;
4999         }
5000
5001         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5002         if (err) {
5003                 drop_inode = 1;
5004                 goto out_unlock;
5005         }
5006
5007         /*
5008         * If the active LSM wants to access the inode during
5009         * d_instantiate it needs these. Smack checks to see
5010         * if the filesystem supports xattrs by looking at the
5011         * ops vector.
5012         */
5013         inode->i_fop = &btrfs_file_operations;
5014         inode->i_op = &btrfs_file_inode_operations;
5015
5016         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5017         if (err)
5018                 drop_inode = 1;
5019         else {
5020                 inode->i_mapping->a_ops = &btrfs_aops;
5021                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5022                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5023                 d_instantiate(dentry, inode);
5024         }
5025 out_unlock:
5026         btrfs_end_transaction(trans, root);
5027         if (drop_inode) {
5028                 inode_dec_link_count(inode);
5029                 iput(inode);
5030         }
5031         btrfs_btree_balance_dirty(root);
5032         return err;
5033 }
5034
5035 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5036                       struct dentry *dentry)
5037 {
5038         struct btrfs_trans_handle *trans;
5039         struct btrfs_root *root = BTRFS_I(dir)->root;
5040         struct inode *inode = old_dentry->d_inode;
5041         u64 index;
5042         int err;
5043         int drop_inode = 0;
5044
5045         /* do not allow sys_link's with other subvols of the same device */
5046         if (root->objectid != BTRFS_I(inode)->root->objectid)
5047                 return -EXDEV;
5048
5049         if (inode->i_nlink >= BTRFS_LINK_MAX)
5050                 return -EMLINK;
5051
5052         err = btrfs_set_inode_index(dir, &index);
5053         if (err)
5054                 goto fail;
5055
5056         /*
5057          * 2 items for inode and inode ref
5058          * 2 items for dir items
5059          * 1 item for parent inode
5060          */
5061         trans = btrfs_start_transaction(root, 5);
5062         if (IS_ERR(trans)) {
5063                 err = PTR_ERR(trans);
5064                 goto fail;
5065         }
5066
5067         btrfs_inc_nlink(inode);
5068         inode_inc_iversion(inode);
5069         inode->i_ctime = CURRENT_TIME;
5070         ihold(inode);
5071
5072         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5073
5074         if (err) {
5075                 drop_inode = 1;
5076         } else {
5077                 struct dentry *parent = dentry->d_parent;
5078                 err = btrfs_update_inode(trans, root, inode);
5079                 if (err)
5080                         goto fail;
5081                 d_instantiate(dentry, inode);
5082                 btrfs_log_new_name(trans, inode, NULL, parent);
5083         }
5084
5085         btrfs_end_transaction(trans, root);
5086 fail:
5087         if (drop_inode) {
5088                 inode_dec_link_count(inode);
5089                 iput(inode);
5090         }
5091         btrfs_btree_balance_dirty(root);
5092         return err;
5093 }
5094
5095 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5096 {
5097         struct inode *inode = NULL;
5098         struct btrfs_trans_handle *trans;
5099         struct btrfs_root *root = BTRFS_I(dir)->root;
5100         int err = 0;
5101         int drop_on_err = 0;
5102         u64 objectid = 0;
5103         u64 index = 0;
5104
5105         /*
5106          * 2 items for inode and ref
5107          * 2 items for dir items
5108          * 1 for xattr if selinux is on
5109          */
5110         trans = btrfs_start_transaction(root, 5);
5111         if (IS_ERR(trans))
5112                 return PTR_ERR(trans);
5113
5114         err = btrfs_find_free_ino(root, &objectid);
5115         if (err)
5116                 goto out_fail;
5117
5118         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5119                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5120                                 S_IFDIR | mode, &index);
5121         if (IS_ERR(inode)) {
5122                 err = PTR_ERR(inode);
5123                 goto out_fail;
5124         }
5125
5126         drop_on_err = 1;
5127
5128         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5129         if (err)
5130                 goto out_fail;
5131
5132         inode->i_op = &btrfs_dir_inode_operations;
5133         inode->i_fop = &btrfs_dir_file_operations;
5134
5135         btrfs_i_size_write(inode, 0);
5136         err = btrfs_update_inode(trans, root, inode);
5137         if (err)
5138                 goto out_fail;
5139
5140         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5141                              dentry->d_name.len, 0, index);
5142         if (err)
5143                 goto out_fail;
5144
5145         d_instantiate(dentry, inode);
5146         drop_on_err = 0;
5147
5148 out_fail:
5149         btrfs_end_transaction(trans, root);
5150         if (drop_on_err)
5151                 iput(inode);
5152         btrfs_btree_balance_dirty(root);
5153         return err;
5154 }
5155
5156 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5157  * and an extent that you want to insert, deal with overlap and insert
5158  * the new extent into the tree.
5159  */
5160 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5161                                 struct extent_map *existing,
5162                                 struct extent_map *em,
5163                                 u64 map_start, u64 map_len)
5164 {
5165         u64 start_diff;
5166
5167         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5168         start_diff = map_start - em->start;
5169         em->start = map_start;
5170         em->len = map_len;
5171         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5172             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5173                 em->block_start += start_diff;
5174                 em->block_len -= start_diff;
5175         }
5176         return add_extent_mapping(em_tree, em);
5177 }
5178
5179 static noinline int uncompress_inline(struct btrfs_path *path,
5180                                       struct inode *inode, struct page *page,
5181                                       size_t pg_offset, u64 extent_offset,
5182                                       struct btrfs_file_extent_item *item)
5183 {
5184         int ret;
5185         struct extent_buffer *leaf = path->nodes[0];
5186         char *tmp;
5187         size_t max_size;
5188         unsigned long inline_size;
5189         unsigned long ptr;
5190         int compress_type;
5191
5192         WARN_ON(pg_offset != 0);
5193         compress_type = btrfs_file_extent_compression(leaf, item);
5194         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5195         inline_size = btrfs_file_extent_inline_item_len(leaf,
5196                                         btrfs_item_nr(leaf, path->slots[0]));
5197         tmp = kmalloc(inline_size, GFP_NOFS);
5198         if (!tmp)
5199                 return -ENOMEM;
5200         ptr = btrfs_file_extent_inline_start(item);
5201
5202         read_extent_buffer(leaf, tmp, ptr, inline_size);
5203
5204         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5205         ret = btrfs_decompress(compress_type, tmp, page,
5206                                extent_offset, inline_size, max_size);
5207         if (ret) {
5208                 char *kaddr = kmap_atomic(page);
5209                 unsigned long copy_size = min_t(u64,
5210                                   PAGE_CACHE_SIZE - pg_offset,
5211                                   max_size - extent_offset);
5212                 memset(kaddr + pg_offset, 0, copy_size);
5213                 kunmap_atomic(kaddr);
5214         }
5215         kfree(tmp);
5216         return 0;
5217 }
5218
5219 /*
5220  * a bit scary, this does extent mapping from logical file offset to the disk.
5221  * the ugly parts come from merging extents from the disk with the in-ram
5222  * representation.  This gets more complex because of the data=ordered code,
5223  * where the in-ram extents might be locked pending data=ordered completion.
5224  *
5225  * This also copies inline extents directly into the page.
5226  */
5227
5228 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5229                                     size_t pg_offset, u64 start, u64 len,
5230                                     int create)
5231 {
5232         int ret;
5233         int err = 0;
5234         u64 bytenr;
5235         u64 extent_start = 0;
5236         u64 extent_end = 0;
5237         u64 objectid = btrfs_ino(inode);
5238         u32 found_type;
5239         struct btrfs_path *path = NULL;
5240         struct btrfs_root *root = BTRFS_I(inode)->root;
5241         struct btrfs_file_extent_item *item;
5242         struct extent_buffer *leaf;
5243         struct btrfs_key found_key;
5244         struct extent_map *em = NULL;
5245         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5246         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5247         struct btrfs_trans_handle *trans = NULL;
5248         int compress_type;
5249
5250 again:
5251         read_lock(&em_tree->lock);
5252         em = lookup_extent_mapping(em_tree, start, len);
5253         if (em)
5254                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5255         read_unlock(&em_tree->lock);
5256
5257         if (em) {
5258                 if (em->start > start || em->start + em->len <= start)
5259                         free_extent_map(em);
5260                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5261                         free_extent_map(em);
5262                 else
5263                         goto out;
5264         }
5265         em = alloc_extent_map();
5266         if (!em) {
5267                 err = -ENOMEM;
5268                 goto out;
5269         }
5270         em->bdev = root->fs_info->fs_devices->latest_bdev;
5271         em->start = EXTENT_MAP_HOLE;
5272         em->orig_start = EXTENT_MAP_HOLE;
5273         em->len = (u64)-1;
5274         em->block_len = (u64)-1;
5275
5276         if (!path) {
5277                 path = btrfs_alloc_path();
5278                 if (!path) {
5279                         err = -ENOMEM;
5280                         goto out;
5281                 }
5282                 /*
5283                  * Chances are we'll be called again, so go ahead and do
5284                  * readahead
5285                  */
5286                 path->reada = 1;
5287         }
5288
5289         ret = btrfs_lookup_file_extent(trans, root, path,
5290                                        objectid, start, trans != NULL);
5291         if (ret < 0) {
5292                 err = ret;
5293                 goto out;
5294         }
5295
5296         if (ret != 0) {
5297                 if (path->slots[0] == 0)
5298                         goto not_found;
5299                 path->slots[0]--;
5300         }
5301
5302         leaf = path->nodes[0];
5303         item = btrfs_item_ptr(leaf, path->slots[0],
5304                               struct btrfs_file_extent_item);
5305         /* are we inside the extent that was found? */
5306         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5307         found_type = btrfs_key_type(&found_key);
5308         if (found_key.objectid != objectid ||
5309             found_type != BTRFS_EXTENT_DATA_KEY) {
5310                 goto not_found;
5311         }
5312
5313         found_type = btrfs_file_extent_type(leaf, item);
5314         extent_start = found_key.offset;
5315         compress_type = btrfs_file_extent_compression(leaf, item);
5316         if (found_type == BTRFS_FILE_EXTENT_REG ||
5317             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5318                 extent_end = extent_start +
5319                        btrfs_file_extent_num_bytes(leaf, item);
5320         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5321                 size_t size;
5322                 size = btrfs_file_extent_inline_len(leaf, item);
5323                 extent_end = (extent_start + size + root->sectorsize - 1) &
5324                         ~((u64)root->sectorsize - 1);
5325         }
5326
5327         if (start >= extent_end) {
5328                 path->slots[0]++;
5329                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5330                         ret = btrfs_next_leaf(root, path);
5331                         if (ret < 0) {
5332                                 err = ret;
5333                                 goto out;
5334                         }
5335                         if (ret > 0)
5336                                 goto not_found;
5337                         leaf = path->nodes[0];
5338                 }
5339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5340                 if (found_key.objectid != objectid ||
5341                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5342                         goto not_found;
5343                 if (start + len <= found_key.offset)
5344                         goto not_found;
5345                 em->start = start;
5346                 em->len = found_key.offset - start;
5347                 goto not_found_em;
5348         }
5349
5350         if (found_type == BTRFS_FILE_EXTENT_REG ||
5351             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5352                 em->start = extent_start;
5353                 em->len = extent_end - extent_start;
5354                 em->orig_start = extent_start -
5355                                  btrfs_file_extent_offset(leaf, item);
5356                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5357                 if (bytenr == 0) {
5358                         em->block_start = EXTENT_MAP_HOLE;
5359                         goto insert;
5360                 }
5361                 if (compress_type != BTRFS_COMPRESS_NONE) {
5362                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5363                         em->compress_type = compress_type;
5364                         em->block_start = bytenr;
5365                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5366                                                                          item);
5367                 } else {
5368                         bytenr += btrfs_file_extent_offset(leaf, item);
5369                         em->block_start = bytenr;
5370                         em->block_len = em->len;
5371                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5372                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5373                 }
5374                 goto insert;
5375         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5376                 unsigned long ptr;
5377                 char *map;
5378                 size_t size;
5379                 size_t extent_offset;
5380                 size_t copy_size;
5381
5382                 em->block_start = EXTENT_MAP_INLINE;
5383                 if (!page || create) {
5384                         em->start = extent_start;
5385                         em->len = extent_end - extent_start;
5386                         goto out;
5387                 }
5388
5389                 size = btrfs_file_extent_inline_len(leaf, item);
5390                 extent_offset = page_offset(page) + pg_offset - extent_start;
5391                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5392                                 size - extent_offset);
5393                 em->start = extent_start + extent_offset;
5394                 em->len = (copy_size + root->sectorsize - 1) &
5395                         ~((u64)root->sectorsize - 1);
5396                 em->orig_start = EXTENT_MAP_INLINE;
5397                 if (compress_type) {
5398                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5399                         em->compress_type = compress_type;
5400                 }
5401                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5402                 if (create == 0 && !PageUptodate(page)) {
5403                         if (btrfs_file_extent_compression(leaf, item) !=
5404                             BTRFS_COMPRESS_NONE) {
5405                                 ret = uncompress_inline(path, inode, page,
5406                                                         pg_offset,
5407                                                         extent_offset, item);
5408                                 BUG_ON(ret); /* -ENOMEM */
5409                         } else {
5410                                 map = kmap(page);
5411                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5412                                                    copy_size);
5413                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5414                                         memset(map + pg_offset + copy_size, 0,
5415                                                PAGE_CACHE_SIZE - pg_offset -
5416                                                copy_size);
5417                                 }
5418                                 kunmap(page);
5419                         }
5420                         flush_dcache_page(page);
5421                 } else if (create && PageUptodate(page)) {
5422                         BUG();
5423                         if (!trans) {
5424                                 kunmap(page);
5425                                 free_extent_map(em);
5426                                 em = NULL;
5427
5428                                 btrfs_release_path(path);
5429                                 trans = btrfs_join_transaction(root);
5430
5431                                 if (IS_ERR(trans))
5432                                         return ERR_CAST(trans);
5433                                 goto again;
5434                         }
5435                         map = kmap(page);
5436                         write_extent_buffer(leaf, map + pg_offset, ptr,
5437                                             copy_size);
5438                         kunmap(page);
5439                         btrfs_mark_buffer_dirty(leaf);
5440                 }
5441                 set_extent_uptodate(io_tree, em->start,
5442                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5443                 goto insert;
5444         } else {
5445                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5446         }
5447 not_found:
5448         em->start = start;
5449         em->len = len;
5450 not_found_em:
5451         em->block_start = EXTENT_MAP_HOLE;
5452         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5453 insert:
5454         btrfs_release_path(path);
5455         if (em->start > start || extent_map_end(em) <= start) {
5456                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5457                        "[%llu %llu]\n", (unsigned long long)em->start,
5458                        (unsigned long long)em->len,
5459                        (unsigned long long)start,
5460                        (unsigned long long)len);
5461                 err = -EIO;
5462                 goto out;
5463         }
5464
5465         err = 0;
5466         write_lock(&em_tree->lock);
5467         ret = add_extent_mapping(em_tree, em);
5468         /* it is possible that someone inserted the extent into the tree
5469          * while we had the lock dropped.  It is also possible that
5470          * an overlapping map exists in the tree
5471          */
5472         if (ret == -EEXIST) {
5473                 struct extent_map *existing;
5474
5475                 ret = 0;
5476
5477                 existing = lookup_extent_mapping(em_tree, start, len);
5478                 if (existing && (existing->start > start ||
5479                     existing->start + existing->len <= start)) {
5480                         free_extent_map(existing);
5481                         existing = NULL;
5482                 }
5483                 if (!existing) {
5484                         existing = lookup_extent_mapping(em_tree, em->start,
5485                                                          em->len);
5486                         if (existing) {
5487                                 err = merge_extent_mapping(em_tree, existing,
5488                                                            em, start,
5489                                                            root->sectorsize);
5490                                 free_extent_map(existing);
5491                                 if (err) {
5492                                         free_extent_map(em);
5493                                         em = NULL;
5494                                 }
5495                         } else {
5496                                 err = -EIO;
5497                                 free_extent_map(em);
5498                                 em = NULL;
5499                         }
5500                 } else {
5501                         free_extent_map(em);
5502                         em = existing;
5503                         err = 0;
5504                 }
5505         }
5506         write_unlock(&em_tree->lock);
5507 out:
5508
5509         if (em)
5510                 trace_btrfs_get_extent(root, em);
5511
5512         if (path)
5513                 btrfs_free_path(path);
5514         if (trans) {
5515                 ret = btrfs_end_transaction(trans, root);
5516                 if (!err)
5517                         err = ret;
5518         }
5519         if (err) {
5520                 free_extent_map(em);
5521                 return ERR_PTR(err);
5522         }
5523         BUG_ON(!em); /* Error is always set */
5524         return em;
5525 }
5526
5527 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5528                                            size_t pg_offset, u64 start, u64 len,
5529                                            int create)
5530 {
5531         struct extent_map *em;
5532         struct extent_map *hole_em = NULL;
5533         u64 range_start = start;
5534         u64 end;
5535         u64 found;
5536         u64 found_end;
5537         int err = 0;
5538
5539         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5540         if (IS_ERR(em))
5541                 return em;
5542         if (em) {
5543                 /*
5544                  * if our em maps to a hole, there might
5545                  * actually be delalloc bytes behind it
5546                  */
5547                 if (em->block_start != EXTENT_MAP_HOLE)
5548                         return em;
5549                 else
5550                         hole_em = em;
5551         }
5552
5553         /* check to see if we've wrapped (len == -1 or similar) */
5554         end = start + len;
5555         if (end < start)
5556                 end = (u64)-1;
5557         else
5558                 end -= 1;
5559
5560         em = NULL;
5561
5562         /* ok, we didn't find anything, lets look for delalloc */
5563         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5564                                  end, len, EXTENT_DELALLOC, 1);
5565         found_end = range_start + found;
5566         if (found_end < range_start)
5567                 found_end = (u64)-1;
5568
5569         /*
5570          * we didn't find anything useful, return
5571          * the original results from get_extent()
5572          */
5573         if (range_start > end || found_end <= start) {
5574                 em = hole_em;
5575                 hole_em = NULL;
5576                 goto out;
5577         }
5578
5579         /* adjust the range_start to make sure it doesn't
5580          * go backwards from the start they passed in
5581          */
5582         range_start = max(start,range_start);
5583         found = found_end - range_start;
5584
5585         if (found > 0) {
5586                 u64 hole_start = start;
5587                 u64 hole_len = len;
5588
5589                 em = alloc_extent_map();
5590                 if (!em) {
5591                         err = -ENOMEM;
5592                         goto out;
5593                 }
5594                 /*
5595                  * when btrfs_get_extent can't find anything it
5596                  * returns one huge hole
5597                  *
5598                  * make sure what it found really fits our range, and
5599                  * adjust to make sure it is based on the start from
5600                  * the caller
5601                  */
5602                 if (hole_em) {
5603                         u64 calc_end = extent_map_end(hole_em);
5604
5605                         if (calc_end <= start || (hole_em->start > end)) {
5606                                 free_extent_map(hole_em);
5607                                 hole_em = NULL;
5608                         } else {
5609                                 hole_start = max(hole_em->start, start);
5610                                 hole_len = calc_end - hole_start;
5611                         }
5612                 }
5613                 em->bdev = NULL;
5614                 if (hole_em && range_start > hole_start) {
5615                         /* our hole starts before our delalloc, so we
5616                          * have to return just the parts of the hole
5617                          * that go until  the delalloc starts
5618                          */
5619                         em->len = min(hole_len,
5620                                       range_start - hole_start);
5621                         em->start = hole_start;
5622                         em->orig_start = hole_start;
5623                         /*
5624                          * don't adjust block start at all,
5625                          * it is fixed at EXTENT_MAP_HOLE
5626                          */
5627                         em->block_start = hole_em->block_start;
5628                         em->block_len = hole_len;
5629                 } else {
5630                         em->start = range_start;
5631                         em->len = found;
5632                         em->orig_start = range_start;
5633                         em->block_start = EXTENT_MAP_DELALLOC;
5634                         em->block_len = found;
5635                 }
5636         } else if (hole_em) {
5637                 return hole_em;
5638         }
5639 out:
5640
5641         free_extent_map(hole_em);
5642         if (err) {
5643                 free_extent_map(em);
5644                 return ERR_PTR(err);
5645         }
5646         return em;
5647 }
5648
5649 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5650                                                   struct extent_map *em,
5651                                                   u64 start, u64 len)
5652 {
5653         struct btrfs_root *root = BTRFS_I(inode)->root;
5654         struct btrfs_trans_handle *trans;
5655         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5656         struct btrfs_key ins;
5657         u64 alloc_hint;
5658         int ret;
5659         bool insert = false;
5660
5661         /*
5662          * Ok if the extent map we looked up is a hole and is for the exact
5663          * range we want, there is no reason to allocate a new one, however if
5664          * it is not right then we need to free this one and drop the cache for
5665          * our range.
5666          */
5667         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5668             em->len != len) {
5669                 free_extent_map(em);
5670                 em = NULL;
5671                 insert = true;
5672                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5673         }
5674
5675         trans = btrfs_join_transaction(root);
5676         if (IS_ERR(trans))
5677                 return ERR_CAST(trans);
5678
5679         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5680                 btrfs_add_inode_defrag(trans, inode);
5681
5682         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5683
5684         alloc_hint = get_extent_allocation_hint(inode, start, len);
5685         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5686                                    alloc_hint, &ins, 1);
5687         if (ret) {
5688                 em = ERR_PTR(ret);
5689                 goto out;
5690         }
5691
5692         if (!em) {
5693                 em = alloc_extent_map();
5694                 if (!em) {
5695                         em = ERR_PTR(-ENOMEM);
5696                         goto out;
5697                 }
5698         }
5699
5700         em->start = start;
5701         em->orig_start = em->start;
5702         em->len = ins.offset;
5703
5704         em->block_start = ins.objectid;
5705         em->block_len = ins.offset;
5706         em->bdev = root->fs_info->fs_devices->latest_bdev;
5707
5708         /*
5709          * We need to do this because if we're using the original em we searched
5710          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5711          */
5712         em->flags = 0;
5713         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5714
5715         while (insert) {
5716                 write_lock(&em_tree->lock);
5717                 ret = add_extent_mapping(em_tree, em);
5718                 write_unlock(&em_tree->lock);
5719                 if (ret != -EEXIST)
5720                         break;
5721                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5722         }
5723
5724         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5725                                            ins.offset, ins.offset, 0);
5726         if (ret) {
5727                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5728                 em = ERR_PTR(ret);
5729         }
5730 out:
5731         btrfs_end_transaction(trans, root);
5732         return em;
5733 }
5734
5735 /*
5736  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5737  * block must be cow'd
5738  */
5739 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5740                                       struct inode *inode, u64 offset, u64 len)
5741 {
5742         struct btrfs_path *path;
5743         int ret;
5744         struct extent_buffer *leaf;
5745         struct btrfs_root *root = BTRFS_I(inode)->root;
5746         struct btrfs_file_extent_item *fi;
5747         struct btrfs_key key;
5748         u64 disk_bytenr;
5749         u64 backref_offset;
5750         u64 extent_end;
5751         u64 num_bytes;
5752         int slot;
5753         int found_type;
5754
5755         path = btrfs_alloc_path();
5756         if (!path)
5757                 return -ENOMEM;
5758
5759         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5760                                        offset, 0);
5761         if (ret < 0)
5762                 goto out;
5763
5764         slot = path->slots[0];
5765         if (ret == 1) {
5766                 if (slot == 0) {
5767                         /* can't find the item, must cow */
5768                         ret = 0;
5769                         goto out;
5770                 }
5771                 slot--;
5772         }
5773         ret = 0;
5774         leaf = path->nodes[0];
5775         btrfs_item_key_to_cpu(leaf, &key, slot);
5776         if (key.objectid != btrfs_ino(inode) ||
5777             key.type != BTRFS_EXTENT_DATA_KEY) {
5778                 /* not our file or wrong item type, must cow */
5779                 goto out;
5780         }
5781
5782         if (key.offset > offset) {
5783                 /* Wrong offset, must cow */
5784                 goto out;
5785         }
5786
5787         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5788         found_type = btrfs_file_extent_type(leaf, fi);
5789         if (found_type != BTRFS_FILE_EXTENT_REG &&
5790             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5791                 /* not a regular extent, must cow */
5792                 goto out;
5793         }
5794         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5795         backref_offset = btrfs_file_extent_offset(leaf, fi);
5796
5797         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5798         if (extent_end < offset + len) {
5799                 /* extent doesn't include our full range, must cow */
5800                 goto out;
5801         }
5802
5803         if (btrfs_extent_readonly(root, disk_bytenr))
5804                 goto out;
5805
5806         /*
5807          * look for other files referencing this extent, if we
5808          * find any we must cow
5809          */
5810         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5811                                   key.offset - backref_offset, disk_bytenr))
5812                 goto out;
5813
5814         /*
5815          * adjust disk_bytenr and num_bytes to cover just the bytes
5816          * in this extent we are about to write.  If there
5817          * are any csums in that range we have to cow in order
5818          * to keep the csums correct
5819          */
5820         disk_bytenr += backref_offset;
5821         disk_bytenr += offset - key.offset;
5822         num_bytes = min(offset + len, extent_end) - offset;
5823         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5824                                 goto out;
5825         /*
5826          * all of the above have passed, it is safe to overwrite this extent
5827          * without cow
5828          */
5829         ret = 1;
5830 out:
5831         btrfs_free_path(path);
5832         return ret;
5833 }
5834
5835 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5836                               struct extent_state **cached_state, int writing)
5837 {
5838         struct btrfs_ordered_extent *ordered;
5839         int ret = 0;
5840
5841         while (1) {
5842                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5843                                  0, cached_state);
5844                 /*
5845                  * We're concerned with the entire range that we're going to be
5846                  * doing DIO to, so we need to make sure theres no ordered
5847                  * extents in this range.
5848                  */
5849                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5850                                                      lockend - lockstart + 1);
5851
5852                 /*
5853                  * We need to make sure there are no buffered pages in this
5854                  * range either, we could have raced between the invalidate in
5855                  * generic_file_direct_write and locking the extent.  The
5856                  * invalidate needs to happen so that reads after a write do not
5857                  * get stale data.
5858                  */
5859                 if (!ordered && (!writing ||
5860                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5861                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5862                                     *cached_state)))
5863                         break;
5864
5865                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5866                                      cached_state, GFP_NOFS);
5867
5868                 if (ordered) {
5869                         btrfs_start_ordered_extent(inode, ordered, 1);
5870                         btrfs_put_ordered_extent(ordered);
5871                 } else {
5872                         /* Screw you mmap */
5873                         ret = filemap_write_and_wait_range(inode->i_mapping,
5874                                                            lockstart,
5875                                                            lockend);
5876                         if (ret)
5877                                 break;
5878
5879                         /*
5880                          * If we found a page that couldn't be invalidated just
5881                          * fall back to buffered.
5882                          */
5883                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5884                                         lockstart >> PAGE_CACHE_SHIFT,
5885                                         lockend >> PAGE_CACHE_SHIFT);
5886                         if (ret)
5887                                 break;
5888                 }
5889
5890                 cond_resched();
5891         }
5892
5893         return ret;
5894 }
5895
5896 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5897                                            u64 len, u64 orig_start,
5898                                            u64 block_start, u64 block_len,
5899                                            int type)
5900 {
5901         struct extent_map_tree *em_tree;
5902         struct extent_map *em;
5903         struct btrfs_root *root = BTRFS_I(inode)->root;
5904         int ret;
5905
5906         em_tree = &BTRFS_I(inode)->extent_tree;
5907         em = alloc_extent_map();
5908         if (!em)
5909                 return ERR_PTR(-ENOMEM);
5910
5911         em->start = start;
5912         em->orig_start = orig_start;
5913         em->len = len;
5914         em->block_len = block_len;
5915         em->block_start = block_start;
5916         em->bdev = root->fs_info->fs_devices->latest_bdev;
5917         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5918         if (type == BTRFS_ORDERED_PREALLOC)
5919                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5920
5921         do {
5922                 btrfs_drop_extent_cache(inode, em->start,
5923                                 em->start + em->len - 1, 0);
5924                 write_lock(&em_tree->lock);
5925                 ret = add_extent_mapping(em_tree, em);
5926                 write_unlock(&em_tree->lock);
5927         } while (ret == -EEXIST);
5928
5929         if (ret) {
5930                 free_extent_map(em);
5931                 return ERR_PTR(ret);
5932         }
5933
5934         return em;
5935 }
5936
5937
5938 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5939                                    struct buffer_head *bh_result, int create)
5940 {
5941         struct extent_map *em;
5942         struct btrfs_root *root = BTRFS_I(inode)->root;
5943         struct extent_state *cached_state = NULL;
5944         u64 start = iblock << inode->i_blkbits;
5945         u64 lockstart, lockend;
5946         u64 len = bh_result->b_size;
5947         struct btrfs_trans_handle *trans;
5948         int unlock_bits = EXTENT_LOCKED;
5949         int ret;
5950
5951         if (create) {
5952                 ret = btrfs_delalloc_reserve_space(inode, len);
5953                 if (ret)
5954                         return ret;
5955                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5956         } else {
5957                 len = min_t(u64, len, root->sectorsize);
5958         }
5959
5960         lockstart = start;
5961         lockend = start + len - 1;
5962
5963         /*
5964          * If this errors out it's because we couldn't invalidate pagecache for
5965          * this range and we need to fallback to buffered.
5966          */
5967         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5968                 return -ENOTBLK;
5969
5970         if (create) {
5971                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5972                                      lockend, EXTENT_DELALLOC, NULL,
5973                                      &cached_state, GFP_NOFS);
5974                 if (ret)
5975                         goto unlock_err;
5976         }
5977
5978         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5979         if (IS_ERR(em)) {
5980                 ret = PTR_ERR(em);
5981                 goto unlock_err;
5982         }
5983
5984         /*
5985          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5986          * io.  INLINE is special, and we could probably kludge it in here, but
5987          * it's still buffered so for safety lets just fall back to the generic
5988          * buffered path.
5989          *
5990          * For COMPRESSED we _have_ to read the entire extent in so we can
5991          * decompress it, so there will be buffering required no matter what we
5992          * do, so go ahead and fallback to buffered.
5993          *
5994          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5995          * to buffered IO.  Don't blame me, this is the price we pay for using
5996          * the generic code.
5997          */
5998         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5999             em->block_start == EXTENT_MAP_INLINE) {
6000                 free_extent_map(em);
6001                 ret = -ENOTBLK;
6002                 goto unlock_err;
6003         }
6004
6005         /* Just a good old fashioned hole, return */
6006         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6007                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6008                 free_extent_map(em);
6009                 ret = 0;
6010                 goto unlock_err;
6011         }
6012
6013         /*
6014          * We don't allocate a new extent in the following cases
6015          *
6016          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6017          * existing extent.
6018          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6019          * just use the extent.
6020          *
6021          */
6022         if (!create) {
6023                 len = min(len, em->len - (start - em->start));
6024                 lockstart = start + len;
6025                 goto unlock;
6026         }
6027
6028         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6029             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6030              em->block_start != EXTENT_MAP_HOLE)) {
6031                 int type;
6032                 int ret;
6033                 u64 block_start;
6034
6035                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6036                         type = BTRFS_ORDERED_PREALLOC;
6037                 else
6038                         type = BTRFS_ORDERED_NOCOW;
6039                 len = min(len, em->len - (start - em->start));
6040                 block_start = em->block_start + (start - em->start);
6041
6042                 /*
6043                  * we're not going to log anything, but we do need
6044                  * to make sure the current transaction stays open
6045                  * while we look for nocow cross refs
6046                  */
6047                 trans = btrfs_join_transaction(root);
6048                 if (IS_ERR(trans))
6049                         goto must_cow;
6050
6051                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6052                         u64 orig_start = em->start;
6053
6054                         if (type == BTRFS_ORDERED_PREALLOC) {
6055                                 free_extent_map(em);
6056                                 em = create_pinned_em(inode, start, len,
6057                                                        orig_start,
6058                                                        block_start, len, type);
6059                                 if (IS_ERR(em)) {
6060                                         btrfs_end_transaction(trans, root);
6061                                         goto unlock_err;
6062                                 }
6063                         }
6064
6065                         ret = btrfs_add_ordered_extent_dio(inode, start,
6066                                            block_start, len, len, type);
6067                         btrfs_end_transaction(trans, root);
6068                         if (ret) {
6069                                 free_extent_map(em);
6070                                 goto unlock_err;
6071                         }
6072                         goto unlock;
6073                 }
6074                 btrfs_end_transaction(trans, root);
6075         }
6076 must_cow:
6077         /*
6078          * this will cow the extent, reset the len in case we changed
6079          * it above
6080          */
6081         len = bh_result->b_size;
6082         em = btrfs_new_extent_direct(inode, em, start, len);
6083         if (IS_ERR(em)) {
6084                 ret = PTR_ERR(em);
6085                 goto unlock_err;
6086         }
6087         len = min(len, em->len - (start - em->start));
6088 unlock:
6089         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6090                 inode->i_blkbits;
6091         bh_result->b_size = len;
6092         bh_result->b_bdev = em->bdev;
6093         set_buffer_mapped(bh_result);
6094         if (create) {
6095                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6096                         set_buffer_new(bh_result);
6097
6098                 /*
6099                  * Need to update the i_size under the extent lock so buffered
6100                  * readers will get the updated i_size when we unlock.
6101                  */
6102                 if (start + len > i_size_read(inode))
6103                         i_size_write(inode, start + len);
6104         }
6105
6106         /*
6107          * In the case of write we need to clear and unlock the entire range,
6108          * in the case of read we need to unlock only the end area that we
6109          * aren't using if there is any left over space.
6110          */
6111         if (lockstart < lockend) {
6112                 if (create && len < lockend - lockstart) {
6113                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6114                                          lockstart + len - 1,
6115                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6116                                          &cached_state, GFP_NOFS);
6117                         /*
6118                          * Beside unlock, we also need to cleanup reserved space
6119                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6120                          */
6121                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6122                                          lockstart + len, lockend,
6123                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6124                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6125                 } else {
6126                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6127                                          lockend, unlock_bits, 1, 0,
6128                                          &cached_state, GFP_NOFS);
6129                 }
6130         } else {
6131                 free_extent_state(cached_state);
6132         }
6133
6134         free_extent_map(em);
6135
6136         return 0;
6137
6138 unlock_err:
6139         if (create)
6140                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6141
6142         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6143                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6144         return ret;
6145 }
6146
6147 struct btrfs_dio_private {
6148         struct inode *inode;
6149         u64 logical_offset;
6150         u64 disk_bytenr;
6151         u64 bytes;
6152         void *private;
6153
6154         /* number of bios pending for this dio */
6155         atomic_t pending_bios;
6156
6157         /* IO errors */
6158         int errors;
6159
6160         struct bio *orig_bio;
6161 };
6162
6163 static void btrfs_endio_direct_read(struct bio *bio, int err)
6164 {
6165         struct btrfs_dio_private *dip = bio->bi_private;
6166         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6167         struct bio_vec *bvec = bio->bi_io_vec;
6168         struct inode *inode = dip->inode;
6169         struct btrfs_root *root = BTRFS_I(inode)->root;
6170         u64 start;
6171
6172         start = dip->logical_offset;
6173         do {
6174                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6175                         struct page *page = bvec->bv_page;
6176                         char *kaddr;
6177                         u32 csum = ~(u32)0;
6178                         u64 private = ~(u32)0;
6179                         unsigned long flags;
6180
6181                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6182                                               start, &private))
6183                                 goto failed;
6184                         local_irq_save(flags);
6185                         kaddr = kmap_atomic(page);
6186                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6187                                                csum, bvec->bv_len);
6188                         btrfs_csum_final(csum, (char *)&csum);
6189                         kunmap_atomic(kaddr);
6190                         local_irq_restore(flags);
6191
6192                         flush_dcache_page(bvec->bv_page);
6193                         if (csum != private) {
6194 failed:
6195                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6196                                       " %llu csum %u private %u\n",
6197                                       (unsigned long long)btrfs_ino(inode),
6198                                       (unsigned long long)start,
6199                                       csum, (unsigned)private);
6200                                 err = -EIO;
6201                         }
6202                 }
6203
6204                 start += bvec->bv_len;
6205                 bvec++;
6206         } while (bvec <= bvec_end);
6207
6208         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6209                       dip->logical_offset + dip->bytes - 1);
6210         bio->bi_private = dip->private;
6211
6212         kfree(dip);
6213
6214         /* If we had a csum failure make sure to clear the uptodate flag */
6215         if (err)
6216                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6217         dio_end_io(bio, err);
6218 }
6219
6220 static void btrfs_endio_direct_write(struct bio *bio, int err)
6221 {
6222         struct btrfs_dio_private *dip = bio->bi_private;
6223         struct inode *inode = dip->inode;
6224         struct btrfs_root *root = BTRFS_I(inode)->root;
6225         struct btrfs_ordered_extent *ordered = NULL;
6226         u64 ordered_offset = dip->logical_offset;
6227         u64 ordered_bytes = dip->bytes;
6228         int ret;
6229
6230         if (err)
6231                 goto out_done;
6232 again:
6233         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6234                                                    &ordered_offset,
6235                                                    ordered_bytes, !err);
6236         if (!ret)
6237                 goto out_test;
6238
6239         ordered->work.func = finish_ordered_fn;
6240         ordered->work.flags = 0;
6241         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6242                            &ordered->work);
6243 out_test:
6244         /*
6245          * our bio might span multiple ordered extents.  If we haven't
6246          * completed the accounting for the whole dio, go back and try again
6247          */
6248         if (ordered_offset < dip->logical_offset + dip->bytes) {
6249                 ordered_bytes = dip->logical_offset + dip->bytes -
6250                         ordered_offset;
6251                 ordered = NULL;
6252                 goto again;
6253         }
6254 out_done:
6255         bio->bi_private = dip->private;
6256
6257         kfree(dip);
6258
6259         /* If we had an error make sure to clear the uptodate flag */
6260         if (err)
6261                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6262         dio_end_io(bio, err);
6263 }
6264
6265 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6266                                     struct bio *bio, int mirror_num,
6267                                     unsigned long bio_flags, u64 offset)
6268 {
6269         int ret;
6270         struct btrfs_root *root = BTRFS_I(inode)->root;
6271         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6272         BUG_ON(ret); /* -ENOMEM */
6273         return 0;
6274 }
6275
6276 static void btrfs_end_dio_bio(struct bio *bio, int err)
6277 {
6278         struct btrfs_dio_private *dip = bio->bi_private;
6279
6280         if (err) {
6281                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6282                       "sector %#Lx len %u err no %d\n",
6283                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6284                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6285                 dip->errors = 1;
6286
6287                 /*
6288                  * before atomic variable goto zero, we must make sure
6289                  * dip->errors is perceived to be set.
6290                  */
6291                 smp_mb__before_atomic_dec();
6292         }
6293
6294         /* if there are more bios still pending for this dio, just exit */
6295         if (!atomic_dec_and_test(&dip->pending_bios))
6296                 goto out;
6297
6298         if (dip->errors)
6299                 bio_io_error(dip->orig_bio);
6300         else {
6301                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6302                 bio_endio(dip->orig_bio, 0);
6303         }
6304 out:
6305         bio_put(bio);
6306 }
6307
6308 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6309                                        u64 first_sector, gfp_t gfp_flags)
6310 {
6311         int nr_vecs = bio_get_nr_vecs(bdev);
6312         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6313 }
6314
6315 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6316                                          int rw, u64 file_offset, int skip_sum,
6317                                          int async_submit)
6318 {
6319         int write = rw & REQ_WRITE;
6320         struct btrfs_root *root = BTRFS_I(inode)->root;
6321         int ret;
6322
6323         bio_get(bio);
6324
6325         if (!write) {
6326                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6327                 if (ret)
6328                         goto err;
6329         }
6330
6331         if (skip_sum)
6332                 goto map;
6333
6334         if (write && async_submit) {
6335                 ret = btrfs_wq_submit_bio(root->fs_info,
6336                                    inode, rw, bio, 0, 0,
6337                                    file_offset,
6338                                    __btrfs_submit_bio_start_direct_io,
6339                                    __btrfs_submit_bio_done);
6340                 goto err;
6341         } else if (write) {
6342                 /*
6343                  * If we aren't doing async submit, calculate the csum of the
6344                  * bio now.
6345                  */
6346                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6347                 if (ret)
6348                         goto err;
6349         } else if (!skip_sum) {
6350                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6351                 if (ret)
6352                         goto err;
6353         }
6354
6355 map:
6356         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6357 err:
6358         bio_put(bio);
6359         return ret;
6360 }
6361
6362 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6363                                     int skip_sum)
6364 {
6365         struct inode *inode = dip->inode;
6366         struct btrfs_root *root = BTRFS_I(inode)->root;
6367         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6368         struct bio *bio;
6369         struct bio *orig_bio = dip->orig_bio;
6370         struct bio_vec *bvec = orig_bio->bi_io_vec;
6371         u64 start_sector = orig_bio->bi_sector;
6372         u64 file_offset = dip->logical_offset;
6373         u64 submit_len = 0;
6374         u64 map_length;
6375         int nr_pages = 0;
6376         int ret = 0;
6377         int async_submit = 0;
6378
6379         map_length = orig_bio->bi_size;
6380         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6381                               &map_length, NULL, 0);
6382         if (ret) {
6383                 bio_put(orig_bio);
6384                 return -EIO;
6385         }
6386
6387         if (map_length >= orig_bio->bi_size) {
6388                 bio = orig_bio;
6389                 goto submit;
6390         }
6391
6392         async_submit = 1;
6393         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6394         if (!bio)
6395                 return -ENOMEM;
6396         bio->bi_private = dip;
6397         bio->bi_end_io = btrfs_end_dio_bio;
6398         atomic_inc(&dip->pending_bios);
6399
6400         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6401                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6402                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6403                                  bvec->bv_offset) < bvec->bv_len)) {
6404                         /*
6405                          * inc the count before we submit the bio so
6406                          * we know the end IO handler won't happen before
6407                          * we inc the count. Otherwise, the dip might get freed
6408                          * before we're done setting it up
6409                          */
6410                         atomic_inc(&dip->pending_bios);
6411                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6412                                                      file_offset, skip_sum,
6413                                                      async_submit);
6414                         if (ret) {
6415                                 bio_put(bio);
6416                                 atomic_dec(&dip->pending_bios);
6417                                 goto out_err;
6418                         }
6419
6420                         start_sector += submit_len >> 9;
6421                         file_offset += submit_len;
6422
6423                         submit_len = 0;
6424                         nr_pages = 0;
6425
6426                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6427                                                   start_sector, GFP_NOFS);
6428                         if (!bio)
6429                                 goto out_err;
6430                         bio->bi_private = dip;
6431                         bio->bi_end_io = btrfs_end_dio_bio;
6432
6433                         map_length = orig_bio->bi_size;
6434                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6435                                               &map_length, NULL, 0);
6436                         if (ret) {
6437                                 bio_put(bio);
6438                                 goto out_err;
6439                         }
6440                 } else {
6441                         submit_len += bvec->bv_len;
6442                         nr_pages ++;
6443                         bvec++;
6444                 }
6445         }
6446
6447 submit:
6448         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6449                                      async_submit);
6450         if (!ret)
6451                 return 0;
6452
6453         bio_put(bio);
6454 out_err:
6455         dip->errors = 1;
6456         /*
6457          * before atomic variable goto zero, we must
6458          * make sure dip->errors is perceived to be set.
6459          */
6460         smp_mb__before_atomic_dec();
6461         if (atomic_dec_and_test(&dip->pending_bios))
6462                 bio_io_error(dip->orig_bio);
6463
6464         /* bio_end_io() will handle error, so we needn't return it */
6465         return 0;
6466 }
6467
6468 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6469                                 loff_t file_offset)
6470 {
6471         struct btrfs_root *root = BTRFS_I(inode)->root;
6472         struct btrfs_dio_private *dip;
6473         struct bio_vec *bvec = bio->bi_io_vec;
6474         int skip_sum;
6475         int write = rw & REQ_WRITE;
6476         int ret = 0;
6477
6478         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6479
6480         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6481         if (!dip) {
6482                 ret = -ENOMEM;
6483                 goto free_ordered;
6484         }
6485
6486         dip->private = bio->bi_private;
6487         dip->inode = inode;
6488         dip->logical_offset = file_offset;
6489
6490         dip->bytes = 0;
6491         do {
6492                 dip->bytes += bvec->bv_len;
6493                 bvec++;
6494         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6495
6496         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6497         bio->bi_private = dip;
6498         dip->errors = 0;
6499         dip->orig_bio = bio;
6500         atomic_set(&dip->pending_bios, 0);
6501
6502         if (write)
6503                 bio->bi_end_io = btrfs_endio_direct_write;
6504         else
6505                 bio->bi_end_io = btrfs_endio_direct_read;
6506
6507         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6508         if (!ret)
6509                 return;
6510 free_ordered:
6511         /*
6512          * If this is a write, we need to clean up the reserved space and kill
6513          * the ordered extent.
6514          */
6515         if (write) {
6516                 struct btrfs_ordered_extent *ordered;
6517                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6518                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6519                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6520                         btrfs_free_reserved_extent(root, ordered->start,
6521                                                    ordered->disk_len);
6522                 btrfs_put_ordered_extent(ordered);
6523                 btrfs_put_ordered_extent(ordered);
6524         }
6525         bio_endio(bio, ret);
6526 }
6527
6528 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6529                         const struct iovec *iov, loff_t offset,
6530                         unsigned long nr_segs)
6531 {
6532         int seg;
6533         int i;
6534         size_t size;
6535         unsigned long addr;
6536         unsigned blocksize_mask = root->sectorsize - 1;
6537         ssize_t retval = -EINVAL;
6538         loff_t end = offset;
6539
6540         if (offset & blocksize_mask)
6541                 goto out;
6542
6543         /* Check the memory alignment.  Blocks cannot straddle pages */
6544         for (seg = 0; seg < nr_segs; seg++) {
6545                 addr = (unsigned long)iov[seg].iov_base;
6546                 size = iov[seg].iov_len;
6547                 end += size;
6548                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6549                         goto out;
6550
6551                 /* If this is a write we don't need to check anymore */
6552                 if (rw & WRITE)
6553                         continue;
6554
6555                 /*
6556                  * Check to make sure we don't have duplicate iov_base's in this
6557                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6558                  * when reading back.
6559                  */
6560                 for (i = seg + 1; i < nr_segs; i++) {
6561                         if (iov[seg].iov_base == iov[i].iov_base)
6562                                 goto out;
6563                 }
6564         }
6565         retval = 0;
6566 out:
6567         return retval;
6568 }
6569
6570 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6571                         const struct iovec *iov, loff_t offset,
6572                         unsigned long nr_segs)
6573 {
6574         struct file *file = iocb->ki_filp;
6575         struct inode *inode = file->f_mapping->host;
6576
6577         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6578                             offset, nr_segs))
6579                 return 0;
6580
6581         return __blockdev_direct_IO(rw, iocb, inode,
6582                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6583                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6584                    btrfs_submit_direct, 0);
6585 }
6586
6587 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6588                 __u64 start, __u64 len)
6589 {
6590         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6591 }
6592
6593 int btrfs_readpage(struct file *file, struct page *page)
6594 {
6595         struct extent_io_tree *tree;
6596         tree = &BTRFS_I(page->mapping->host)->io_tree;
6597         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6598 }
6599
6600 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6601 {
6602         struct extent_io_tree *tree;
6603
6604
6605         if (current->flags & PF_MEMALLOC) {
6606                 redirty_page_for_writepage(wbc, page);
6607                 unlock_page(page);
6608                 return 0;
6609         }
6610         tree = &BTRFS_I(page->mapping->host)->io_tree;
6611         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6612 }
6613
6614 int btrfs_writepages(struct address_space *mapping,
6615                      struct writeback_control *wbc)
6616 {
6617         struct extent_io_tree *tree;
6618
6619         tree = &BTRFS_I(mapping->host)->io_tree;
6620         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6621 }
6622
6623 static int
6624 btrfs_readpages(struct file *file, struct address_space *mapping,
6625                 struct list_head *pages, unsigned nr_pages)
6626 {
6627         struct extent_io_tree *tree;
6628         tree = &BTRFS_I(mapping->host)->io_tree;
6629         return extent_readpages(tree, mapping, pages, nr_pages,
6630                                 btrfs_get_extent);
6631 }
6632 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6633 {
6634         struct extent_io_tree *tree;
6635         struct extent_map_tree *map;
6636         int ret;
6637
6638         tree = &BTRFS_I(page->mapping->host)->io_tree;
6639         map = &BTRFS_I(page->mapping->host)->extent_tree;
6640         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6641         if (ret == 1) {
6642                 ClearPagePrivate(page);
6643                 set_page_private(page, 0);
6644                 page_cache_release(page);
6645         }
6646         return ret;
6647 }
6648
6649 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6650 {
6651         if (PageWriteback(page) || PageDirty(page))
6652                 return 0;
6653         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6654 }
6655
6656 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6657 {
6658         struct inode *inode = page->mapping->host;
6659         struct extent_io_tree *tree;
6660         struct btrfs_ordered_extent *ordered;
6661         struct extent_state *cached_state = NULL;
6662         u64 page_start = page_offset(page);
6663         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6664
6665         /*
6666          * we have the page locked, so new writeback can't start,
6667          * and the dirty bit won't be cleared while we are here.
6668          *
6669          * Wait for IO on this page so that we can safely clear
6670          * the PagePrivate2 bit and do ordered accounting
6671          */
6672         wait_on_page_writeback(page);
6673
6674         tree = &BTRFS_I(inode)->io_tree;
6675         if (offset) {
6676                 btrfs_releasepage(page, GFP_NOFS);
6677                 return;
6678         }
6679         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6680         ordered = btrfs_lookup_ordered_extent(inode,
6681                                            page_offset(page));
6682         if (ordered) {
6683                 /*
6684                  * IO on this page will never be started, so we need
6685                  * to account for any ordered extents now
6686                  */
6687                 clear_extent_bit(tree, page_start, page_end,
6688                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6689                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6690                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6691                 /*
6692                  * whoever cleared the private bit is responsible
6693                  * for the finish_ordered_io
6694                  */
6695                 if (TestClearPagePrivate2(page) &&
6696                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6697                                                    PAGE_CACHE_SIZE, 1)) {
6698                         btrfs_finish_ordered_io(ordered);
6699                 }
6700                 btrfs_put_ordered_extent(ordered);
6701                 cached_state = NULL;
6702                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6703         }
6704         clear_extent_bit(tree, page_start, page_end,
6705                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6706                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6707                  &cached_state, GFP_NOFS);
6708         __btrfs_releasepage(page, GFP_NOFS);
6709
6710         ClearPageChecked(page);
6711         if (PagePrivate(page)) {
6712                 ClearPagePrivate(page);
6713                 set_page_private(page, 0);
6714                 page_cache_release(page);
6715         }
6716 }
6717
6718 /*
6719  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6720  * called from a page fault handler when a page is first dirtied. Hence we must
6721  * be careful to check for EOF conditions here. We set the page up correctly
6722  * for a written page which means we get ENOSPC checking when writing into
6723  * holes and correct delalloc and unwritten extent mapping on filesystems that
6724  * support these features.
6725  *
6726  * We are not allowed to take the i_mutex here so we have to play games to
6727  * protect against truncate races as the page could now be beyond EOF.  Because
6728  * vmtruncate() writes the inode size before removing pages, once we have the
6729  * page lock we can determine safely if the page is beyond EOF. If it is not
6730  * beyond EOF, then the page is guaranteed safe against truncation until we
6731  * unlock the page.
6732  */
6733 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6734 {
6735         struct page *page = vmf->page;
6736         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6737         struct btrfs_root *root = BTRFS_I(inode)->root;
6738         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6739         struct btrfs_ordered_extent *ordered;
6740         struct extent_state *cached_state = NULL;
6741         char *kaddr;
6742         unsigned long zero_start;
6743         loff_t size;
6744         int ret;
6745         int reserved = 0;
6746         u64 page_start;
6747         u64 page_end;
6748
6749         sb_start_pagefault(inode->i_sb);
6750         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6751         if (!ret) {
6752                 ret = file_update_time(vma->vm_file);
6753                 reserved = 1;
6754         }
6755         if (ret) {
6756                 if (ret == -ENOMEM)
6757                         ret = VM_FAULT_OOM;
6758                 else /* -ENOSPC, -EIO, etc */
6759                         ret = VM_FAULT_SIGBUS;
6760                 if (reserved)
6761                         goto out;
6762                 goto out_noreserve;
6763         }
6764
6765         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6766 again:
6767         lock_page(page);
6768         size = i_size_read(inode);
6769         page_start = page_offset(page);
6770         page_end = page_start + PAGE_CACHE_SIZE - 1;
6771
6772         if ((page->mapping != inode->i_mapping) ||
6773             (page_start >= size)) {
6774                 /* page got truncated out from underneath us */
6775                 goto out_unlock;
6776         }
6777         wait_on_page_writeback(page);
6778
6779         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6780         set_page_extent_mapped(page);
6781
6782         /*
6783          * we can't set the delalloc bits if there are pending ordered
6784          * extents.  Drop our locks and wait for them to finish
6785          */
6786         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6787         if (ordered) {
6788                 unlock_extent_cached(io_tree, page_start, page_end,
6789                                      &cached_state, GFP_NOFS);
6790                 unlock_page(page);
6791                 btrfs_start_ordered_extent(inode, ordered, 1);
6792                 btrfs_put_ordered_extent(ordered);
6793                 goto again;
6794         }
6795
6796         /*
6797          * XXX - page_mkwrite gets called every time the page is dirtied, even
6798          * if it was already dirty, so for space accounting reasons we need to
6799          * clear any delalloc bits for the range we are fixing to save.  There
6800          * is probably a better way to do this, but for now keep consistent with
6801          * prepare_pages in the normal write path.
6802          */
6803         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6804                           EXTENT_DIRTY | EXTENT_DELALLOC |
6805                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6806                           0, 0, &cached_state, GFP_NOFS);
6807
6808         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6809                                         &cached_state);
6810         if (ret) {
6811                 unlock_extent_cached(io_tree, page_start, page_end,
6812                                      &cached_state, GFP_NOFS);
6813                 ret = VM_FAULT_SIGBUS;
6814                 goto out_unlock;
6815         }
6816         ret = 0;
6817
6818         /* page is wholly or partially inside EOF */
6819         if (page_start + PAGE_CACHE_SIZE > size)
6820                 zero_start = size & ~PAGE_CACHE_MASK;
6821         else
6822                 zero_start = PAGE_CACHE_SIZE;
6823
6824         if (zero_start != PAGE_CACHE_SIZE) {
6825                 kaddr = kmap(page);
6826                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6827                 flush_dcache_page(page);
6828                 kunmap(page);
6829         }
6830         ClearPageChecked(page);
6831         set_page_dirty(page);
6832         SetPageUptodate(page);
6833
6834         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6835         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6836         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6837
6838         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6839
6840 out_unlock:
6841         if (!ret) {
6842                 sb_end_pagefault(inode->i_sb);
6843                 return VM_FAULT_LOCKED;
6844         }
6845         unlock_page(page);
6846 out:
6847         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6848 out_noreserve:
6849         sb_end_pagefault(inode->i_sb);
6850         return ret;
6851 }
6852
6853 static int btrfs_truncate(struct inode *inode)
6854 {
6855         struct btrfs_root *root = BTRFS_I(inode)->root;
6856         struct btrfs_block_rsv *rsv;
6857         int ret;
6858         int err = 0;
6859         struct btrfs_trans_handle *trans;
6860         u64 mask = root->sectorsize - 1;
6861         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6862
6863         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6864         if (ret)
6865                 return ret;
6866
6867         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6868         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6869
6870         /*
6871          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6872          * 3 things going on here
6873          *
6874          * 1) We need to reserve space for our orphan item and the space to
6875          * delete our orphan item.  Lord knows we don't want to have a dangling
6876          * orphan item because we didn't reserve space to remove it.
6877          *
6878          * 2) We need to reserve space to update our inode.
6879          *
6880          * 3) We need to have something to cache all the space that is going to
6881          * be free'd up by the truncate operation, but also have some slack
6882          * space reserved in case it uses space during the truncate (thank you
6883          * very much snapshotting).
6884          *
6885          * And we need these to all be seperate.  The fact is we can use alot of
6886          * space doing the truncate, and we have no earthly idea how much space
6887          * we will use, so we need the truncate reservation to be seperate so it
6888          * doesn't end up using space reserved for updating the inode or
6889          * removing the orphan item.  We also need to be able to stop the
6890          * transaction and start a new one, which means we need to be able to
6891          * update the inode several times, and we have no idea of knowing how
6892          * many times that will be, so we can't just reserve 1 item for the
6893          * entirety of the opration, so that has to be done seperately as well.
6894          * Then there is the orphan item, which does indeed need to be held on
6895          * to for the whole operation, and we need nobody to touch this reserved
6896          * space except the orphan code.
6897          *
6898          * So that leaves us with
6899          *
6900          * 1) root->orphan_block_rsv - for the orphan deletion.
6901          * 2) rsv - for the truncate reservation, which we will steal from the
6902          * transaction reservation.
6903          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6904          * updating the inode.
6905          */
6906         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6907         if (!rsv)
6908                 return -ENOMEM;
6909         rsv->size = min_size;
6910         rsv->failfast = 1;
6911
6912         /*
6913          * 1 for the truncate slack space
6914          * 1 for the orphan item we're going to add
6915          * 1 for the orphan item deletion
6916          * 1 for updating the inode.
6917          */
6918         trans = btrfs_start_transaction(root, 4);
6919         if (IS_ERR(trans)) {
6920                 err = PTR_ERR(trans);
6921                 goto out;
6922         }
6923
6924         /* Migrate the slack space for the truncate to our reserve */
6925         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6926                                       min_size);
6927         BUG_ON(ret);
6928
6929         ret = btrfs_orphan_add(trans, inode);
6930         if (ret) {
6931                 btrfs_end_transaction(trans, root);
6932                 goto out;
6933         }
6934
6935         /*
6936          * setattr is responsible for setting the ordered_data_close flag,
6937          * but that is only tested during the last file release.  That
6938          * could happen well after the next commit, leaving a great big
6939          * window where new writes may get lost if someone chooses to write
6940          * to this file after truncating to zero
6941          *
6942          * The inode doesn't have any dirty data here, and so if we commit
6943          * this is a noop.  If someone immediately starts writing to the inode
6944          * it is very likely we'll catch some of their writes in this
6945          * transaction, and the commit will find this file on the ordered
6946          * data list with good things to send down.
6947          *
6948          * This is a best effort solution, there is still a window where
6949          * using truncate to replace the contents of the file will
6950          * end up with a zero length file after a crash.
6951          */
6952         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6953                                            &BTRFS_I(inode)->runtime_flags))
6954                 btrfs_add_ordered_operation(trans, root, inode);
6955
6956         /*
6957          * So if we truncate and then write and fsync we normally would just
6958          * write the extents that changed, which is a problem if we need to
6959          * first truncate that entire inode.  So set this flag so we write out
6960          * all of the extents in the inode to the sync log so we're completely
6961          * safe.
6962          */
6963         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6964         trans->block_rsv = rsv;
6965
6966         while (1) {
6967                 ret = btrfs_truncate_inode_items(trans, root, inode,
6968                                                  inode->i_size,
6969                                                  BTRFS_EXTENT_DATA_KEY);
6970                 if (ret != -ENOSPC) {
6971                         err = ret;
6972                         break;
6973                 }
6974
6975                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6976                 ret = btrfs_update_inode(trans, root, inode);
6977                 if (ret) {
6978                         err = ret;
6979                         break;
6980                 }
6981
6982                 btrfs_end_transaction(trans, root);
6983                 btrfs_btree_balance_dirty(root);
6984
6985                 trans = btrfs_start_transaction(root, 2);
6986                 if (IS_ERR(trans)) {
6987                         ret = err = PTR_ERR(trans);
6988                         trans = NULL;
6989                         break;
6990                 }
6991
6992                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
6993                                               rsv, min_size);
6994                 BUG_ON(ret);    /* shouldn't happen */
6995                 trans->block_rsv = rsv;
6996         }
6997
6998         if (ret == 0 && inode->i_nlink > 0) {
6999                 trans->block_rsv = root->orphan_block_rsv;
7000                 ret = btrfs_orphan_del(trans, inode);
7001                 if (ret)
7002                         err = ret;
7003         } else if (ret && inode->i_nlink > 0) {
7004                 /*
7005                  * Failed to do the truncate, remove us from the in memory
7006                  * orphan list.
7007                  */
7008                 ret = btrfs_orphan_del(NULL, inode);
7009         }
7010
7011         if (trans) {
7012                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7013                 ret = btrfs_update_inode(trans, root, inode);
7014                 if (ret && !err)
7015                         err = ret;
7016
7017                 ret = btrfs_end_transaction(trans, root);
7018                 btrfs_btree_balance_dirty(root);
7019         }
7020
7021 out:
7022         btrfs_free_block_rsv(root, rsv);
7023
7024         if (ret && !err)
7025                 err = ret;
7026
7027         return err;
7028 }
7029
7030 /*
7031  * create a new subvolume directory/inode (helper for the ioctl).
7032  */
7033 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7034                              struct btrfs_root *new_root, u64 new_dirid)
7035 {
7036         struct inode *inode;
7037         int err;
7038         u64 index = 0;
7039
7040         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7041                                 new_dirid, new_dirid,
7042                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7043                                 &index);
7044         if (IS_ERR(inode))
7045                 return PTR_ERR(inode);
7046         inode->i_op = &btrfs_dir_inode_operations;
7047         inode->i_fop = &btrfs_dir_file_operations;
7048
7049         set_nlink(inode, 1);
7050         btrfs_i_size_write(inode, 0);
7051
7052         err = btrfs_update_inode(trans, new_root, inode);
7053
7054         iput(inode);
7055         return err;
7056 }
7057
7058 struct inode *btrfs_alloc_inode(struct super_block *sb)
7059 {
7060         struct btrfs_inode *ei;
7061         struct inode *inode;
7062
7063         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7064         if (!ei)
7065                 return NULL;
7066
7067         ei->root = NULL;
7068         ei->generation = 0;
7069         ei->last_trans = 0;
7070         ei->last_sub_trans = 0;
7071         ei->logged_trans = 0;
7072         ei->delalloc_bytes = 0;
7073         ei->disk_i_size = 0;
7074         ei->flags = 0;
7075         ei->csum_bytes = 0;
7076         ei->index_cnt = (u64)-1;
7077         ei->last_unlink_trans = 0;
7078         ei->last_log_commit = 0;
7079
7080         spin_lock_init(&ei->lock);
7081         ei->outstanding_extents = 0;
7082         ei->reserved_extents = 0;
7083
7084         ei->runtime_flags = 0;
7085         ei->force_compress = BTRFS_COMPRESS_NONE;
7086
7087         ei->delayed_node = NULL;
7088
7089         inode = &ei->vfs_inode;
7090         extent_map_tree_init(&ei->extent_tree);
7091         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7092         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7093         ei->io_tree.track_uptodate = 1;
7094         ei->io_failure_tree.track_uptodate = 1;
7095         mutex_init(&ei->log_mutex);
7096         mutex_init(&ei->delalloc_mutex);
7097         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7098         INIT_LIST_HEAD(&ei->delalloc_inodes);
7099         INIT_LIST_HEAD(&ei->ordered_operations);
7100         RB_CLEAR_NODE(&ei->rb_node);
7101
7102         return inode;
7103 }
7104
7105 static void btrfs_i_callback(struct rcu_head *head)
7106 {
7107         struct inode *inode = container_of(head, struct inode, i_rcu);
7108         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7109 }
7110
7111 void btrfs_destroy_inode(struct inode *inode)
7112 {
7113         struct btrfs_ordered_extent *ordered;
7114         struct btrfs_root *root = BTRFS_I(inode)->root;
7115
7116         WARN_ON(!hlist_empty(&inode->i_dentry));
7117         WARN_ON(inode->i_data.nrpages);
7118         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7119         WARN_ON(BTRFS_I(inode)->reserved_extents);
7120         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7121         WARN_ON(BTRFS_I(inode)->csum_bytes);
7122
7123         /*
7124          * This can happen where we create an inode, but somebody else also
7125          * created the same inode and we need to destroy the one we already
7126          * created.
7127          */
7128         if (!root)
7129                 goto free;
7130
7131         /*
7132          * Make sure we're properly removed from the ordered operation
7133          * lists.
7134          */
7135         smp_mb();
7136         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7137                 spin_lock(&root->fs_info->ordered_extent_lock);
7138                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7139                 spin_unlock(&root->fs_info->ordered_extent_lock);
7140         }
7141
7142         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7143                      &BTRFS_I(inode)->runtime_flags)) {
7144                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7145                        (unsigned long long)btrfs_ino(inode));
7146                 atomic_dec(&root->orphan_inodes);
7147         }
7148
7149         while (1) {
7150                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7151                 if (!ordered)
7152                         break;
7153                 else {
7154                         printk(KERN_ERR "btrfs found ordered "
7155                                "extent %llu %llu on inode cleanup\n",
7156                                (unsigned long long)ordered->file_offset,
7157                                (unsigned long long)ordered->len);
7158                         btrfs_remove_ordered_extent(inode, ordered);
7159                         btrfs_put_ordered_extent(ordered);
7160                         btrfs_put_ordered_extent(ordered);
7161                 }
7162         }
7163         inode_tree_del(inode);
7164         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7165 free:
7166         btrfs_remove_delayed_node(inode);
7167         call_rcu(&inode->i_rcu, btrfs_i_callback);
7168 }
7169
7170 int btrfs_drop_inode(struct inode *inode)
7171 {
7172         struct btrfs_root *root = BTRFS_I(inode)->root;
7173
7174         if (btrfs_root_refs(&root->root_item) == 0 &&
7175             !btrfs_is_free_space_inode(inode))
7176                 return 1;
7177         else
7178                 return generic_drop_inode(inode);
7179 }
7180
7181 static void init_once(void *foo)
7182 {
7183         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7184
7185         inode_init_once(&ei->vfs_inode);
7186 }
7187
7188 void btrfs_destroy_cachep(void)
7189 {
7190         /*
7191          * Make sure all delayed rcu free inodes are flushed before we
7192          * destroy cache.
7193          */
7194         rcu_barrier();
7195         if (btrfs_inode_cachep)
7196                 kmem_cache_destroy(btrfs_inode_cachep);
7197         if (btrfs_trans_handle_cachep)
7198                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7199         if (btrfs_transaction_cachep)
7200                 kmem_cache_destroy(btrfs_transaction_cachep);
7201         if (btrfs_path_cachep)
7202                 kmem_cache_destroy(btrfs_path_cachep);
7203         if (btrfs_free_space_cachep)
7204                 kmem_cache_destroy(btrfs_free_space_cachep);
7205         if (btrfs_delalloc_work_cachep)
7206                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7207 }
7208
7209 int btrfs_init_cachep(void)
7210 {
7211         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7212                         sizeof(struct btrfs_inode), 0,
7213                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7214         if (!btrfs_inode_cachep)
7215                 goto fail;
7216
7217         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7218                         sizeof(struct btrfs_trans_handle), 0,
7219                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7220         if (!btrfs_trans_handle_cachep)
7221                 goto fail;
7222
7223         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7224                         sizeof(struct btrfs_transaction), 0,
7225                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7226         if (!btrfs_transaction_cachep)
7227                 goto fail;
7228
7229         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7230                         sizeof(struct btrfs_path), 0,
7231                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7232         if (!btrfs_path_cachep)
7233                 goto fail;
7234
7235         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7236                         sizeof(struct btrfs_free_space), 0,
7237                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7238         if (!btrfs_free_space_cachep)
7239                 goto fail;
7240
7241         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7242                         sizeof(struct btrfs_delalloc_work), 0,
7243                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7244                         NULL);
7245         if (!btrfs_delalloc_work_cachep)
7246                 goto fail;
7247
7248         return 0;
7249 fail:
7250         btrfs_destroy_cachep();
7251         return -ENOMEM;
7252 }
7253
7254 static int btrfs_getattr(struct vfsmount *mnt,
7255                          struct dentry *dentry, struct kstat *stat)
7256 {
7257         struct inode *inode = dentry->d_inode;
7258         u32 blocksize = inode->i_sb->s_blocksize;
7259
7260         generic_fillattr(inode, stat);
7261         stat->dev = BTRFS_I(inode)->root->anon_dev;
7262         stat->blksize = PAGE_CACHE_SIZE;
7263         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7264                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7265         return 0;
7266 }
7267
7268 /*
7269  * If a file is moved, it will inherit the cow and compression flags of the new
7270  * directory.
7271  */
7272 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7273 {
7274         struct btrfs_inode *b_dir = BTRFS_I(dir);
7275         struct btrfs_inode *b_inode = BTRFS_I(inode);
7276
7277         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7278                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7279         else
7280                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7281
7282         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7283                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7284                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7285         } else {
7286                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7287                                     BTRFS_INODE_NOCOMPRESS);
7288         }
7289 }
7290
7291 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7292                            struct inode *new_dir, struct dentry *new_dentry)
7293 {
7294         struct btrfs_trans_handle *trans;
7295         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7296         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7297         struct inode *new_inode = new_dentry->d_inode;
7298         struct inode *old_inode = old_dentry->d_inode;
7299         struct timespec ctime = CURRENT_TIME;
7300         u64 index = 0;
7301         u64 root_objectid;
7302         int ret;
7303         u64 old_ino = btrfs_ino(old_inode);
7304
7305         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7306                 return -EPERM;
7307
7308         /* we only allow rename subvolume link between subvolumes */
7309         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7310                 return -EXDEV;
7311
7312         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7313             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7314                 return -ENOTEMPTY;
7315
7316         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7317             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7318                 return -ENOTEMPTY;
7319         /*
7320          * we're using rename to replace one file with another.
7321          * and the replacement file is large.  Start IO on it now so
7322          * we don't add too much work to the end of the transaction
7323          */
7324         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7325             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7326                 filemap_flush(old_inode->i_mapping);
7327
7328         /* close the racy window with snapshot create/destroy ioctl */
7329         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7330                 down_read(&root->fs_info->subvol_sem);
7331         /*
7332          * We want to reserve the absolute worst case amount of items.  So if
7333          * both inodes are subvols and we need to unlink them then that would
7334          * require 4 item modifications, but if they are both normal inodes it
7335          * would require 5 item modifications, so we'll assume their normal
7336          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7337          * should cover the worst case number of items we'll modify.
7338          */
7339         trans = btrfs_start_transaction(root, 20);
7340         if (IS_ERR(trans)) {
7341                 ret = PTR_ERR(trans);
7342                 goto out_notrans;
7343         }
7344
7345         if (dest != root)
7346                 btrfs_record_root_in_trans(trans, dest);
7347
7348         ret = btrfs_set_inode_index(new_dir, &index);
7349         if (ret)
7350                 goto out_fail;
7351
7352         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7353                 /* force full log commit if subvolume involved. */
7354                 root->fs_info->last_trans_log_full_commit = trans->transid;
7355         } else {
7356                 ret = btrfs_insert_inode_ref(trans, dest,
7357                                              new_dentry->d_name.name,
7358                                              new_dentry->d_name.len,
7359                                              old_ino,
7360                                              btrfs_ino(new_dir), index);
7361                 if (ret)
7362                         goto out_fail;
7363                 /*
7364                  * this is an ugly little race, but the rename is required
7365                  * to make sure that if we crash, the inode is either at the
7366                  * old name or the new one.  pinning the log transaction lets
7367                  * us make sure we don't allow a log commit to come in after
7368                  * we unlink the name but before we add the new name back in.
7369                  */
7370                 btrfs_pin_log_trans(root);
7371         }
7372         /*
7373          * make sure the inode gets flushed if it is replacing
7374          * something.
7375          */
7376         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7377                 btrfs_add_ordered_operation(trans, root, old_inode);
7378
7379         inode_inc_iversion(old_dir);
7380         inode_inc_iversion(new_dir);
7381         inode_inc_iversion(old_inode);
7382         old_dir->i_ctime = old_dir->i_mtime = ctime;
7383         new_dir->i_ctime = new_dir->i_mtime = ctime;
7384         old_inode->i_ctime = ctime;
7385
7386         if (old_dentry->d_parent != new_dentry->d_parent)
7387                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7388
7389         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7390                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7391                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7392                                         old_dentry->d_name.name,
7393                                         old_dentry->d_name.len);
7394         } else {
7395                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7396                                         old_dentry->d_inode,
7397                                         old_dentry->d_name.name,
7398                                         old_dentry->d_name.len);
7399                 if (!ret)
7400                         ret = btrfs_update_inode(trans, root, old_inode);
7401         }
7402         if (ret) {
7403                 btrfs_abort_transaction(trans, root, ret);
7404                 goto out_fail;
7405         }
7406
7407         if (new_inode) {
7408                 inode_inc_iversion(new_inode);
7409                 new_inode->i_ctime = CURRENT_TIME;
7410                 if (unlikely(btrfs_ino(new_inode) ==
7411                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7412                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7413                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7414                                                 root_objectid,
7415                                                 new_dentry->d_name.name,
7416                                                 new_dentry->d_name.len);
7417                         BUG_ON(new_inode->i_nlink == 0);
7418                 } else {
7419                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7420                                                  new_dentry->d_inode,
7421                                                  new_dentry->d_name.name,
7422                                                  new_dentry->d_name.len);
7423                 }
7424                 if (!ret && new_inode->i_nlink == 0) {
7425                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7426                         BUG_ON(ret);
7427                 }
7428                 if (ret) {
7429                         btrfs_abort_transaction(trans, root, ret);
7430                         goto out_fail;
7431                 }
7432         }
7433
7434         fixup_inode_flags(new_dir, old_inode);
7435
7436         ret = btrfs_add_link(trans, new_dir, old_inode,
7437                              new_dentry->d_name.name,
7438                              new_dentry->d_name.len, 0, index);
7439         if (ret) {
7440                 btrfs_abort_transaction(trans, root, ret);
7441                 goto out_fail;
7442         }
7443
7444         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7445                 struct dentry *parent = new_dentry->d_parent;
7446                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7447                 btrfs_end_log_trans(root);
7448         }
7449 out_fail:
7450         btrfs_end_transaction(trans, root);
7451 out_notrans:
7452         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7453                 up_read(&root->fs_info->subvol_sem);
7454
7455         return ret;
7456 }
7457
7458 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7459 {
7460         struct btrfs_delalloc_work *delalloc_work;
7461
7462         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7463                                      work);
7464         if (delalloc_work->wait)
7465                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7466         else
7467                 filemap_flush(delalloc_work->inode->i_mapping);
7468
7469         if (delalloc_work->delay_iput)
7470                 btrfs_add_delayed_iput(delalloc_work->inode);
7471         else
7472                 iput(delalloc_work->inode);
7473         complete(&delalloc_work->completion);
7474 }
7475
7476 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7477                                                     int wait, int delay_iput)
7478 {
7479         struct btrfs_delalloc_work *work;
7480
7481         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7482         if (!work)
7483                 return NULL;
7484
7485         init_completion(&work->completion);
7486         INIT_LIST_HEAD(&work->list);
7487         work->inode = inode;
7488         work->wait = wait;
7489         work->delay_iput = delay_iput;
7490         work->work.func = btrfs_run_delalloc_work;
7491
7492         return work;
7493 }
7494
7495 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7496 {
7497         wait_for_completion(&work->completion);
7498         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7499 }
7500
7501 /*
7502  * some fairly slow code that needs optimization. This walks the list
7503  * of all the inodes with pending delalloc and forces them to disk.
7504  */
7505 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7506 {
7507         struct list_head *head = &root->fs_info->delalloc_inodes;
7508         struct btrfs_inode *binode;
7509         struct inode *inode;
7510         struct btrfs_delalloc_work *work, *next;
7511         struct list_head works;
7512         int ret = 0;
7513
7514         if (root->fs_info->sb->s_flags & MS_RDONLY)
7515                 return -EROFS;
7516
7517         INIT_LIST_HEAD(&works);
7518
7519         spin_lock(&root->fs_info->delalloc_lock);
7520         while (!list_empty(head)) {
7521                 binode = list_entry(head->next, struct btrfs_inode,
7522                                     delalloc_inodes);
7523                 inode = igrab(&binode->vfs_inode);
7524                 if (!inode)
7525                         list_del_init(&binode->delalloc_inodes);
7526                 spin_unlock(&root->fs_info->delalloc_lock);
7527                 if (inode) {
7528                         work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7529                         if (!work) {
7530                                 ret = -ENOMEM;
7531                                 goto out;
7532                         }
7533                         list_add_tail(&work->list, &works);
7534                         btrfs_queue_worker(&root->fs_info->flush_workers,
7535                                            &work->work);
7536                 }
7537                 cond_resched();
7538                 spin_lock(&root->fs_info->delalloc_lock);
7539         }
7540         spin_unlock(&root->fs_info->delalloc_lock);
7541
7542         /* the filemap_flush will queue IO into the worker threads, but
7543          * we have to make sure the IO is actually started and that
7544          * ordered extents get created before we return
7545          */
7546         atomic_inc(&root->fs_info->async_submit_draining);
7547         while (atomic_read(&root->fs_info->nr_async_submits) ||
7548               atomic_read(&root->fs_info->async_delalloc_pages)) {
7549                 wait_event(root->fs_info->async_submit_wait,
7550                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7551                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7552         }
7553         atomic_dec(&root->fs_info->async_submit_draining);
7554 out:
7555         list_for_each_entry_safe(work, next, &works, list) {
7556                 list_del_init(&work->list);
7557                 btrfs_wait_and_free_delalloc_work(work);
7558         }
7559         return ret;
7560 }
7561
7562 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7563                          const char *symname)
7564 {
7565         struct btrfs_trans_handle *trans;
7566         struct btrfs_root *root = BTRFS_I(dir)->root;
7567         struct btrfs_path *path;
7568         struct btrfs_key key;
7569         struct inode *inode = NULL;
7570         int err;
7571         int drop_inode = 0;
7572         u64 objectid;
7573         u64 index = 0 ;
7574         int name_len;
7575         int datasize;
7576         unsigned long ptr;
7577         struct btrfs_file_extent_item *ei;
7578         struct extent_buffer *leaf;
7579
7580         name_len = strlen(symname) + 1;
7581         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7582                 return -ENAMETOOLONG;
7583
7584         /*
7585          * 2 items for inode item and ref
7586          * 2 items for dir items
7587          * 1 item for xattr if selinux is on
7588          */
7589         trans = btrfs_start_transaction(root, 5);
7590         if (IS_ERR(trans))
7591                 return PTR_ERR(trans);
7592
7593         err = btrfs_find_free_ino(root, &objectid);
7594         if (err)
7595                 goto out_unlock;
7596
7597         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7598                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7599                                 S_IFLNK|S_IRWXUGO, &index);
7600         if (IS_ERR(inode)) {
7601                 err = PTR_ERR(inode);
7602                 goto out_unlock;
7603         }
7604
7605         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7606         if (err) {
7607                 drop_inode = 1;
7608                 goto out_unlock;
7609         }
7610
7611         /*
7612         * If the active LSM wants to access the inode during
7613         * d_instantiate it needs these. Smack checks to see
7614         * if the filesystem supports xattrs by looking at the
7615         * ops vector.
7616         */
7617         inode->i_fop = &btrfs_file_operations;
7618         inode->i_op = &btrfs_file_inode_operations;
7619
7620         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7621         if (err)
7622                 drop_inode = 1;
7623         else {
7624                 inode->i_mapping->a_ops = &btrfs_aops;
7625                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7626                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7627         }
7628         if (drop_inode)
7629                 goto out_unlock;
7630
7631         path = btrfs_alloc_path();
7632         if (!path) {
7633                 err = -ENOMEM;
7634                 drop_inode = 1;
7635                 goto out_unlock;
7636         }
7637         key.objectid = btrfs_ino(inode);
7638         key.offset = 0;
7639         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7640         datasize = btrfs_file_extent_calc_inline_size(name_len);
7641         err = btrfs_insert_empty_item(trans, root, path, &key,
7642                                       datasize);
7643         if (err) {
7644                 drop_inode = 1;
7645                 btrfs_free_path(path);
7646                 goto out_unlock;
7647         }
7648         leaf = path->nodes[0];
7649         ei = btrfs_item_ptr(leaf, path->slots[0],
7650                             struct btrfs_file_extent_item);
7651         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7652         btrfs_set_file_extent_type(leaf, ei,
7653                                    BTRFS_FILE_EXTENT_INLINE);
7654         btrfs_set_file_extent_encryption(leaf, ei, 0);
7655         btrfs_set_file_extent_compression(leaf, ei, 0);
7656         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7657         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7658
7659         ptr = btrfs_file_extent_inline_start(ei);
7660         write_extent_buffer(leaf, symname, ptr, name_len);
7661         btrfs_mark_buffer_dirty(leaf);
7662         btrfs_free_path(path);
7663
7664         inode->i_op = &btrfs_symlink_inode_operations;
7665         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7666         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7667         inode_set_bytes(inode, name_len);
7668         btrfs_i_size_write(inode, name_len - 1);
7669         err = btrfs_update_inode(trans, root, inode);
7670         if (err)
7671                 drop_inode = 1;
7672
7673 out_unlock:
7674         if (!err)
7675                 d_instantiate(dentry, inode);
7676         btrfs_end_transaction(trans, root);
7677         if (drop_inode) {
7678                 inode_dec_link_count(inode);
7679                 iput(inode);
7680         }
7681         btrfs_btree_balance_dirty(root);
7682         return err;
7683 }
7684
7685 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7686                                        u64 start, u64 num_bytes, u64 min_size,
7687                                        loff_t actual_len, u64 *alloc_hint,
7688                                        struct btrfs_trans_handle *trans)
7689 {
7690         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7691         struct extent_map *em;
7692         struct btrfs_root *root = BTRFS_I(inode)->root;
7693         struct btrfs_key ins;
7694         u64 cur_offset = start;
7695         u64 i_size;
7696         int ret = 0;
7697         bool own_trans = true;
7698
7699         if (trans)
7700                 own_trans = false;
7701         while (num_bytes > 0) {
7702                 if (own_trans) {
7703                         trans = btrfs_start_transaction(root, 3);
7704                         if (IS_ERR(trans)) {
7705                                 ret = PTR_ERR(trans);
7706                                 break;
7707                         }
7708                 }
7709
7710                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7711                                            0, *alloc_hint, &ins, 1);
7712                 if (ret) {
7713                         if (own_trans)
7714                                 btrfs_end_transaction(trans, root);
7715                         break;
7716                 }
7717
7718                 ret = insert_reserved_file_extent(trans, inode,
7719                                                   cur_offset, ins.objectid,
7720                                                   ins.offset, ins.offset,
7721                                                   ins.offset, 0, 0, 0,
7722                                                   BTRFS_FILE_EXTENT_PREALLOC);
7723                 if (ret) {
7724                         btrfs_abort_transaction(trans, root, ret);
7725                         if (own_trans)
7726                                 btrfs_end_transaction(trans, root);
7727                         break;
7728                 }
7729                 btrfs_drop_extent_cache(inode, cur_offset,
7730                                         cur_offset + ins.offset -1, 0);
7731
7732                 em = alloc_extent_map();
7733                 if (!em) {
7734                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7735                                 &BTRFS_I(inode)->runtime_flags);
7736                         goto next;
7737                 }
7738
7739                 em->start = cur_offset;
7740                 em->orig_start = cur_offset;
7741                 em->len = ins.offset;
7742                 em->block_start = ins.objectid;
7743                 em->block_len = ins.offset;
7744                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7745                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7746                 em->generation = trans->transid;
7747
7748                 while (1) {
7749                         write_lock(&em_tree->lock);
7750                         ret = add_extent_mapping(em_tree, em);
7751                         if (!ret)
7752                                 list_move(&em->list,
7753                                           &em_tree->modified_extents);
7754                         write_unlock(&em_tree->lock);
7755                         if (ret != -EEXIST)
7756                                 break;
7757                         btrfs_drop_extent_cache(inode, cur_offset,
7758                                                 cur_offset + ins.offset - 1,
7759                                                 0);
7760                 }
7761                 free_extent_map(em);
7762 next:
7763                 num_bytes -= ins.offset;
7764                 cur_offset += ins.offset;
7765                 *alloc_hint = ins.objectid + ins.offset;
7766
7767                 inode_inc_iversion(inode);
7768                 inode->i_ctime = CURRENT_TIME;
7769                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7770                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7771                     (actual_len > inode->i_size) &&
7772                     (cur_offset > inode->i_size)) {
7773                         if (cur_offset > actual_len)
7774                                 i_size = actual_len;
7775                         else
7776                                 i_size = cur_offset;
7777                         i_size_write(inode, i_size);
7778                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7779                 }
7780
7781                 ret = btrfs_update_inode(trans, root, inode);
7782
7783                 if (ret) {
7784                         btrfs_abort_transaction(trans, root, ret);
7785                         if (own_trans)
7786                                 btrfs_end_transaction(trans, root);
7787                         break;
7788                 }
7789
7790                 if (own_trans)
7791                         btrfs_end_transaction(trans, root);
7792         }
7793         return ret;
7794 }
7795
7796 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7797                               u64 start, u64 num_bytes, u64 min_size,
7798                               loff_t actual_len, u64 *alloc_hint)
7799 {
7800         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7801                                            min_size, actual_len, alloc_hint,
7802                                            NULL);
7803 }
7804
7805 int btrfs_prealloc_file_range_trans(struct inode *inode,
7806                                     struct btrfs_trans_handle *trans, int mode,
7807                                     u64 start, u64 num_bytes, u64 min_size,
7808                                     loff_t actual_len, u64 *alloc_hint)
7809 {
7810         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7811                                            min_size, actual_len, alloc_hint, trans);
7812 }
7813
7814 static int btrfs_set_page_dirty(struct page *page)
7815 {
7816         return __set_page_dirty_nobuffers(page);
7817 }
7818
7819 static int btrfs_permission(struct inode *inode, int mask)
7820 {
7821         struct btrfs_root *root = BTRFS_I(inode)->root;
7822         umode_t mode = inode->i_mode;
7823
7824         if (mask & MAY_WRITE &&
7825             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7826                 if (btrfs_root_readonly(root))
7827                         return -EROFS;
7828                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7829                         return -EACCES;
7830         }
7831         return generic_permission(inode, mask);
7832 }
7833
7834 static const struct inode_operations btrfs_dir_inode_operations = {
7835         .getattr        = btrfs_getattr,
7836         .lookup         = btrfs_lookup,
7837         .create         = btrfs_create,
7838         .unlink         = btrfs_unlink,
7839         .link           = btrfs_link,
7840         .mkdir          = btrfs_mkdir,
7841         .rmdir          = btrfs_rmdir,
7842         .rename         = btrfs_rename,
7843         .symlink        = btrfs_symlink,
7844         .setattr        = btrfs_setattr,
7845         .mknod          = btrfs_mknod,
7846         .setxattr       = btrfs_setxattr,
7847         .getxattr       = btrfs_getxattr,
7848         .listxattr      = btrfs_listxattr,
7849         .removexattr    = btrfs_removexattr,
7850         .permission     = btrfs_permission,
7851         .get_acl        = btrfs_get_acl,
7852 };
7853 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7854         .lookup         = btrfs_lookup,
7855         .permission     = btrfs_permission,
7856         .get_acl        = btrfs_get_acl,
7857 };
7858
7859 static const struct file_operations btrfs_dir_file_operations = {
7860         .llseek         = generic_file_llseek,
7861         .read           = generic_read_dir,
7862         .readdir        = btrfs_real_readdir,
7863         .unlocked_ioctl = btrfs_ioctl,
7864 #ifdef CONFIG_COMPAT
7865         .compat_ioctl   = btrfs_ioctl,
7866 #endif
7867         .release        = btrfs_release_file,
7868         .fsync          = btrfs_sync_file,
7869 };
7870
7871 static struct extent_io_ops btrfs_extent_io_ops = {
7872         .fill_delalloc = run_delalloc_range,
7873         .submit_bio_hook = btrfs_submit_bio_hook,
7874         .merge_bio_hook = btrfs_merge_bio_hook,
7875         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7876         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7877         .writepage_start_hook = btrfs_writepage_start_hook,
7878         .set_bit_hook = btrfs_set_bit_hook,
7879         .clear_bit_hook = btrfs_clear_bit_hook,
7880         .merge_extent_hook = btrfs_merge_extent_hook,
7881         .split_extent_hook = btrfs_split_extent_hook,
7882 };
7883
7884 /*
7885  * btrfs doesn't support the bmap operation because swapfiles
7886  * use bmap to make a mapping of extents in the file.  They assume
7887  * these extents won't change over the life of the file and they
7888  * use the bmap result to do IO directly to the drive.
7889  *
7890  * the btrfs bmap call would return logical addresses that aren't
7891  * suitable for IO and they also will change frequently as COW
7892  * operations happen.  So, swapfile + btrfs == corruption.
7893  *
7894  * For now we're avoiding this by dropping bmap.
7895  */
7896 static const struct address_space_operations btrfs_aops = {
7897         .readpage       = btrfs_readpage,
7898         .writepage      = btrfs_writepage,
7899         .writepages     = btrfs_writepages,
7900         .readpages      = btrfs_readpages,
7901         .direct_IO      = btrfs_direct_IO,
7902         .invalidatepage = btrfs_invalidatepage,
7903         .releasepage    = btrfs_releasepage,
7904         .set_page_dirty = btrfs_set_page_dirty,
7905         .error_remove_page = generic_error_remove_page,
7906 };
7907
7908 static const struct address_space_operations btrfs_symlink_aops = {
7909         .readpage       = btrfs_readpage,
7910         .writepage      = btrfs_writepage,
7911         .invalidatepage = btrfs_invalidatepage,
7912         .releasepage    = btrfs_releasepage,
7913 };
7914
7915 static const struct inode_operations btrfs_file_inode_operations = {
7916         .getattr        = btrfs_getattr,
7917         .setattr        = btrfs_setattr,
7918         .setxattr       = btrfs_setxattr,
7919         .getxattr       = btrfs_getxattr,
7920         .listxattr      = btrfs_listxattr,
7921         .removexattr    = btrfs_removexattr,
7922         .permission     = btrfs_permission,
7923         .fiemap         = btrfs_fiemap,
7924         .get_acl        = btrfs_get_acl,
7925         .update_time    = btrfs_update_time,
7926 };
7927 static const struct inode_operations btrfs_special_inode_operations = {
7928         .getattr        = btrfs_getattr,
7929         .setattr        = btrfs_setattr,
7930         .permission     = btrfs_permission,
7931         .setxattr       = btrfs_setxattr,
7932         .getxattr       = btrfs_getxattr,
7933         .listxattr      = btrfs_listxattr,
7934         .removexattr    = btrfs_removexattr,
7935         .get_acl        = btrfs_get_acl,
7936         .update_time    = btrfs_update_time,
7937 };
7938 static const struct inode_operations btrfs_symlink_inode_operations = {
7939         .readlink       = generic_readlink,
7940         .follow_link    = page_follow_link_light,
7941         .put_link       = page_put_link,
7942         .getattr        = btrfs_getattr,
7943         .setattr        = btrfs_setattr,
7944         .permission     = btrfs_permission,
7945         .setxattr       = btrfs_setxattr,
7946         .getxattr       = btrfs_getxattr,
7947         .listxattr      = btrfs_listxattr,
7948         .removexattr    = btrfs_removexattr,
7949         .get_acl        = btrfs_get_acl,
7950         .update_time    = btrfs_update_time,
7951 };
7952
7953 const struct dentry_operations btrfs_dentry_operations = {
7954         .d_delete       = btrfs_dentry_delete,
7955         .d_release      = btrfs_dentry_release,
7956 };