fs/nilfs2/inode.c

   1 /*
   2  * inode.c - NILFS inode operations.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/uio.h>
  29 #include "nilfs.h"
  30 #include "btnode.h"
  31 #include "segment.h"
  32 #include "page.h"
  33 #include "mdt.h"
  34 #include "cpfile.h"
  35 #include "ifile.h"
  36
  37 /**
  38  * struct nilfs_iget_args - arguments used during comparison between inodes
  39  * @ino: inode number
  40  * @cno: checkpoint number
  41  * @root: pointer on NILFS root object (mounted checkpoint)
  42  * @for_gc: inode for GC flag
  43  */
  44 struct nilfs_iget_args {
  45         u64 ino;
  46         __u64 cno;
  47         struct nilfs_root *root;
  48         int for_gc;
  49 };
  50
  51 void nilfs_inode_add_blocks(struct inode *inode, int n)
  52 {
  53         struct nilfs_root *root = NILFS_I(inode)->i_root;
  54
  55         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  56         if (root)
  57                 atomic_add(n, &root->blocks_count);
  58 }
  59
  60 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  61 {
  62         struct nilfs_root *root = NILFS_I(inode)->i_root;
  63
  64         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  65         if (root)
  66                 atomic_sub(n, &root->blocks_count);
  67 }
  68
  69 /**
  70  * nilfs_get_block() - get a file block on the filesystem (callback function)
  71  * @inode - inode struct of the target file
  72  * @blkoff - file block number
  73  * @bh_result - buffer head to be mapped on
  74  * @create - indicate whether allocating the block or not when it has not
  75  *      been allocated yet.
  76  *
  77  * This function does not issue actual read request of the specified data
  78  * block. It is done by VFS.
  79  */
  80 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  81                     struct buffer_head *bh_result, int create)
  82 {
  83         struct nilfs_inode_info *ii = NILFS_I(inode);
  84         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  85         __u64 blknum = 0;
  86         int err = 0, ret;
  87         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  88
  89         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  90         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  91         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  92         if (ret >= 0) { /* found */
  93                 map_bh(bh_result, inode->i_sb, blknum);
  94                 if (ret > 0)
  95                         bh_result->b_size = (ret << inode->i_blkbits);
  96                 goto out;
  97         }
  98         /* data block was not found */
  99         if (ret == -ENOENT && create) {
 100                 struct nilfs_transaction_info ti;
 101
 102                 bh_result->b_blocknr = 0;
 103                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 104                 if (unlikely(err))
 105                         goto out;
 106                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 107                                         (unsigned long)bh_result);
 108                 if (unlikely(err != 0)) {
 109                         if (err == -EEXIST) {
 110                                 /*
 111                                  * The get_block() function could be called
 112                                  * from multiple callers for an inode.
 113                                  * However, the page having this block must
 114                                  * be locked in this case.
 115                                  */
 116                                 printk(KERN_WARNING
 117                                        "nilfs_get_block: a race condition "
 118                                        "while inserting a data block. "
 119                                        "(inode number=%lu, file block "
 120                                        "offset=%llu)\n",
 121                                        inode->i_ino,
 122                                        (unsigned long long)blkoff);
 123                                 err = 0;
 124                         }
 125                         nilfs_transaction_abort(inode->i_sb);
 126                         goto out;
 127                 }
 128                 nilfs_mark_inode_dirty(inode);
 129                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 130                 /* Error handling should be detailed */
 131                 set_buffer_new(bh_result);
 132                 set_buffer_delay(bh_result);
 133                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 134                                                       to proper value */
 135         } else if (ret == -ENOENT) {
 136                 /* not found is not error (e.g. hole); must return without
 137                    the mapped state flag. */
 138                 ;
 139         } else {
 140                 err = ret;
 141         }
 142
 143  out:
 144         return err;
 145 }
 146
 147 /**
 148  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 149  * address_space_operations.
 150  * @file - file struct of the file to be read
 151  * @page - the page to be read
 152  */
 153 static int nilfs_readpage(struct file *file, struct page *page)
 154 {
 155         return mpage_readpage(page, nilfs_get_block);
 156 }
 157
 158 /**
 159  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 160  * address_space_operations.
 161  * @file - file struct of the file to be read
 162  * @mapping - address_space struct used for reading multiple pages
 163  * @pages - the pages to be read
 164  * @nr_pages - number of pages to be read
 165  */
 166 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 167                            struct list_head *pages, unsigned nr_pages)
 168 {
 169         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 170 }
 171
 172 static int nilfs_writepages(struct address_space *mapping,
 173                             struct writeback_control *wbc)
 174 {
 175         struct inode *inode = mapping->host;
 176         int err = 0;
 177
 178         if (wbc->sync_mode == WB_SYNC_ALL)
 179                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 180                                                     wbc->range_start,
 181                                                     wbc->range_end);
 182         return err;
 183 }
 184
 185 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 186 {
 187         struct inode *inode = page->mapping->host;
 188         int err;
 189
 190         redirty_page_for_writepage(wbc, page);
 191         unlock_page(page);
 192
 193         if (wbc->sync_mode == WB_SYNC_ALL) {
 194                 err = nilfs_construct_segment(inode->i_sb);
 195                 if (unlikely(err))
 196                         return err;
 197         } else if (wbc->for_reclaim)
 198                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 199
 200         return 0;
 201 }
 202
 203 static int nilfs_set_page_dirty(struct page *page)
 204 {
 205         int ret = __set_page_dirty_buffers(page);
 206
 207         if (ret) {
 208                 struct inode *inode = page->mapping->host;
 209                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 210
 211                 nilfs_set_file_dirty(inode, nr_dirty);
 212         }
 213         return ret;
 214 }
 215
 216 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 217                              loff_t pos, unsigned len, unsigned flags,
 218                              struct page **pagep, void **fsdata)
 219
 220 {
 221         struct inode *inode = mapping->host;
 222         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 223
 224         if (unlikely(err))
 225                 return err;
 226
 227         err = block_write_begin(mapping, pos, len, flags, pagep,
 228                                 nilfs_get_block);
 229         if (unlikely(err)) {
 230                 loff_t isize = mapping->host->i_size;
 231                 if (pos + len > isize)
 232                         vmtruncate(mapping->host, isize);
 233
 234                 nilfs_transaction_abort(inode->i_sb);
 235         }
 236         return err;
 237 }
 238
 239 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 240                            loff_t pos, unsigned len, unsigned copied,
 241                            struct page *page, void *fsdata)
 242 {
 243         struct inode *inode = mapping->host;
 244         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 245         unsigned nr_dirty;
 246         int err;
 247
 248         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 249                                                   start + copied);
 250         copied = generic_write_end(file, mapping, pos, len, copied, page,
 251                                    fsdata);
 252         nilfs_set_file_dirty(inode, nr_dirty);
 253         err = nilfs_transaction_commit(inode->i_sb);
 254         return err ? : copied;
 255 }
 256
 257 static ssize_t
 258 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 259                 loff_t offset, unsigned long nr_segs)
 260 {
 261         struct file *file = iocb->ki_filp;
 262         struct inode *inode = file->f_mapping->host;
 263         ssize_t size;
 264
 265         if (rw == WRITE)
 266                 return 0;
 267
 268         /* Needs synchronization with the cleaner */
 269         size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 270                                   nilfs_get_block);
 271
 272         /*
 273          * In case of error extending write may have instantiated a few
 274          * blocks outside i_size. Trim these off again.
 275          */
 276         if (unlikely((rw & WRITE) && size < 0)) {
 277                 loff_t isize = i_size_read(inode);
 278                 loff_t end = offset + iov_length(iov, nr_segs);
 279
 280                 if (end > isize)
 281                         vmtruncate(inode, isize);
 282         }
 283
 284         return size;
 285 }
 286
 287 const struct address_space_operations nilfs_aops = {
 288         .writepage              = nilfs_writepage,
 289         .readpage               = nilfs_readpage,
 290         .writepages             = nilfs_writepages,
 291         .set_page_dirty         = nilfs_set_page_dirty,
 292         .readpages              = nilfs_readpages,
 293         .write_begin            = nilfs_write_begin,
 294         .write_end              = nilfs_write_end,
 295         /* .releasepage         = nilfs_releasepage, */
 296         .invalidatepage         = block_invalidatepage,
 297         .direct_IO              = nilfs_direct_IO,
 298         .is_partially_uptodate  = block_is_partially_uptodate,
 299 };
 300
 301 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 302 {
 303         struct super_block *sb = dir->i_sb;
 304         struct the_nilfs *nilfs = sb->s_fs_info;
 305         struct inode *inode;
 306         struct nilfs_inode_info *ii;
 307         struct nilfs_root *root;
 308         int err = -ENOMEM;
 309         ino_t ino;
 310
 311         inode = new_inode(sb);
 312         if (unlikely(!inode))
 313                 goto failed;
 314
 315         mapping_set_gfp_mask(inode->i_mapping,
 316                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 317
 318         root = NILFS_I(dir)->i_root;
 319         ii = NILFS_I(inode);
 320         ii->i_state = 1 << NILFS_I_NEW;
 321         ii->i_root = root;
 322
 323         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 324         if (unlikely(err))
 325                 goto failed_ifile_create_inode;
 326         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 327
 328         atomic_inc(&root->inodes_count);
 329         inode_init_owner(inode, dir, mode);
 330         inode->i_ino = ino;
 331         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 332
 333         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 334                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 335                 if (err < 0)
 336                         goto failed_bmap;
 337
 338                 set_bit(NILFS_I_BMAP, &ii->i_state);
 339                 /* No lock is needed; iget() ensures it. */
 340         }
 341
 342         ii->i_flags = nilfs_mask_flags(
 343                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 344
 345         /* ii->i_file_acl = 0; */
 346         /* ii->i_dir_acl = 0; */
 347         ii->i_dir_start_lookup = 0;
 348         nilfs_set_inode_flags(inode);
 349         spin_lock(&nilfs->ns_next_gen_lock);
 350         inode->i_generation = nilfs->ns_next_generation++;
 351         spin_unlock(&nilfs->ns_next_gen_lock);
 352         insert_inode_hash(inode);
 353
 354         err = nilfs_init_acl(inode, dir);
 355         if (unlikely(err))
 356                 goto failed_acl; /* never occur. When supporting
 357                                     nilfs_init_acl(), proper cancellation of
 358                                     above jobs should be considered */
 359
 360         return inode;
 361
 362  failed_acl:
 363  failed_bmap:
 364         clear_nlink(inode);
 365         iput(inode);  /* raw_inode will be deleted through
 366                          generic_delete_inode() */
 367         goto failed;
 368
 369  failed_ifile_create_inode:
 370         make_bad_inode(inode);
 371         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 372                          called */
 373  failed:
 374         return ERR_PTR(err);
 375 }
 376
 377 void nilfs_set_inode_flags(struct inode *inode)
 378 {
 379         unsigned int flags = NILFS_I(inode)->i_flags;
 380
 381         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 382                             S_DIRSYNC);
 383         if (flags & FS_SYNC_FL)
 384                 inode->i_flags |= S_SYNC;
 385         if (flags & FS_APPEND_FL)
 386                 inode->i_flags |= S_APPEND;
 387         if (flags & FS_IMMUTABLE_FL)
 388                 inode->i_flags |= S_IMMUTABLE;
 389         if (flags & FS_NOATIME_FL)
 390                 inode->i_flags |= S_NOATIME;
 391         if (flags & FS_DIRSYNC_FL)
 392                 inode->i_flags |= S_DIRSYNC;
 393         mapping_set_gfp_mask(inode->i_mapping,
 394                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 395 }
 396
 397 int nilfs_read_inode_common(struct inode *inode,
 398                             struct nilfs_inode *raw_inode)
 399 {
 400         struct nilfs_inode_info *ii = NILFS_I(inode);
 401         int err;
 402
 403         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 404         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 405         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 406         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 407         inode->i_size = le64_to_cpu(raw_inode->i_size);
 408         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 409         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 410         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 411         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 412         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 413         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 414         if (inode->i_nlink == 0 && inode->i_mode == 0)
 415                 return -EINVAL; /* this inode is deleted */
 416
 417         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 418         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 419 #if 0
 420         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 421         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 422                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 423 #endif
 424         ii->i_dir_start_lookup = 0;
 425         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 426
 427         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 428             S_ISLNK(inode->i_mode)) {
 429                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 430                 if (err < 0)
 431                         return err;
 432                 set_bit(NILFS_I_BMAP, &ii->i_state);
 433                 /* No lock is needed; iget() ensures it. */
 434         }
 435         return 0;
 436 }
 437
 438 static int __nilfs_read_inode(struct super_block *sb,
 439                               struct nilfs_root *root, unsigned long ino,
 440                               struct inode *inode)
 441 {
 442         struct the_nilfs *nilfs = sb->s_fs_info;
 443         struct buffer_head *bh;
 444         struct nilfs_inode *raw_inode;
 445         int err;
 446
 447         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 448         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 449         if (unlikely(err))
 450                 goto bad_inode;
 451
 452         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 453
 454         err = nilfs_read_inode_common(inode, raw_inode);
 455         if (err)
 456                 goto failed_unmap;
 457
 458         if (S_ISREG(inode->i_mode)) {
 459                 inode->i_op = &nilfs_file_inode_operations;
 460                 inode->i_fop = &nilfs_file_operations;
 461                 inode->i_mapping->a_ops = &nilfs_aops;
 462         } else if (S_ISDIR(inode->i_mode)) {
 463                 inode->i_op = &nilfs_dir_inode_operations;
 464                 inode->i_fop = &nilfs_dir_operations;
 465                 inode->i_mapping->a_ops = &nilfs_aops;
 466         } else if (S_ISLNK(inode->i_mode)) {
 467                 inode->i_op = &nilfs_symlink_inode_operations;
 468                 inode->i_mapping->a_ops = &nilfs_aops;
 469         } else {
 470                 inode->i_op = &nilfs_special_inode_operations;
 471                 init_special_inode(
 472                         inode, inode->i_mode,
 473                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 474         }
 475         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 476         brelse(bh);
 477         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 478         nilfs_set_inode_flags(inode);
 479         return 0;
 480
 481  failed_unmap:
 482         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 483         brelse(bh);
 484
 485  bad_inode:
 486         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 487         return err;
 488 }
 489
 490 static int nilfs_iget_test(struct inode *inode, void *opaque)
 491 {
 492         struct nilfs_iget_args *args = opaque;
 493         struct nilfs_inode_info *ii;
 494
 495         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 496                 return 0;
 497
 498         ii = NILFS_I(inode);
 499         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 500                 return !args->for_gc;
 501
 502         return args->for_gc && args->cno == ii->i_cno;
 503 }
 504
 505 static int nilfs_iget_set(struct inode *inode, void *opaque)
 506 {
 507         struct nilfs_iget_args *args = opaque;
 508
 509         inode->i_ino = args->ino;
 510         if (args->for_gc) {
 511                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 512                 NILFS_I(inode)->i_cno = args->cno;
 513                 NILFS_I(inode)->i_root = NULL;
 514         } else {
 515                 if (args->root && args->ino == NILFS_ROOT_INO)
 516                         nilfs_get_root(args->root);
 517                 NILFS_I(inode)->i_root = args->root;
 518         }
 519         return 0;
 520 }
 521
 522 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 523                             unsigned long ino)
 524 {
 525         struct nilfs_iget_args args = {
 526                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 527         };
 528
 529         return ilookup5(sb, ino, nilfs_iget_test, &args);
 530 }
 531
 532 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 533                                 unsigned long ino)
 534 {
 535         struct nilfs_iget_args args = {
 536                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 537         };
 538
 539         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 540 }
 541
 542 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 543                          unsigned long ino)
 544 {
 545         struct inode *inode;
 546         int err;
 547
 548         inode = nilfs_iget_locked(sb, root, ino);
 549         if (unlikely(!inode))
 550                 return ERR_PTR(-ENOMEM);
 551         if (!(inode->i_state & I_NEW))
 552                 return inode;
 553
 554         err = __nilfs_read_inode(sb, root, ino, inode);
 555         if (unlikely(err)) {
 556                 iget_failed(inode);
 557                 return ERR_PTR(err);
 558         }
 559         unlock_new_inode(inode);
 560         return inode;
 561 }
 562
 563 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 564                                 __u64 cno)
 565 {
 566         struct nilfs_iget_args args = {
 567                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 568         };
 569         struct inode *inode;
 570         int err;
 571
 572         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 573         if (unlikely(!inode))
 574                 return ERR_PTR(-ENOMEM);
 575         if (!(inode->i_state & I_NEW))
 576                 return inode;
 577
 578         err = nilfs_init_gcinode(inode);
 579         if (unlikely(err)) {
 580                 iget_failed(inode);
 581                 return ERR_PTR(err);
 582         }
 583         unlock_new_inode(inode);
 584         return inode;
 585 }
 586
 587 void nilfs_write_inode_common(struct inode *inode,
 588                               struct nilfs_inode *raw_inode, int has_bmap)
 589 {
 590         struct nilfs_inode_info *ii = NILFS_I(inode);
 591
 592         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 593         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 594         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 595         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 596         raw_inode->i_size = cpu_to_le64(inode->i_size);
 597         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 598         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 599         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 600         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 601         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 602
 603         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 604         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 605
 606         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 607                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 608
 609                 /* zero-fill unused portion in the case of super root block */
 610                 raw_inode->i_xattr = 0;
 611                 raw_inode->i_pad = 0;
 612                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 613                        nilfs->ns_inode_size - sizeof(*raw_inode));
 614         }
 615
 616         if (has_bmap)
 617                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 618         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 619                 raw_inode->i_device_code =
 620                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 621         /* When extending inode, nilfs->ns_inode_size should be checked
 622            for substitutions of appended fields */
 623 }
 624
 625 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 626 {
 627         ino_t ino = inode->i_ino;
 628         struct nilfs_inode_info *ii = NILFS_I(inode);
 629         struct inode *ifile = ii->i_root->ifile;
 630         struct nilfs_inode *raw_inode;
 631
 632         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 633
 634         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 635                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 636         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 637
 638         nilfs_write_inode_common(inode, raw_inode, 0);
 639                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 640                    deadlock of bmap. This delays update of i_bmap to just
 641                    before writing */
 642         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 643 }
 644
 645 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 646
 647 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 648                                 unsigned long from)
 649 {
 650         unsigned long b;
 651         int ret;
 652
 653         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 654                 return;
 655 repeat:
 656         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 657         if (ret == -ENOENT)
 658                 return;
 659         else if (ret < 0)
 660                 goto failed;
 661
 662         if (b < from)
 663                 return;
 664
 665         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 666         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 667         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 668         if (!ret || (ret == -ENOMEM &&
 669                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 670                 goto repeat;
 671
 672 failed:
 673         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 674                       "failed to truncate bmap (ino=%lu, err=%d)",
 675                       ii->vfs_inode.i_ino, ret);
 676 }
 677
 678 void nilfs_truncate(struct inode *inode)
 679 {
 680         unsigned long blkoff;
 681         unsigned int blocksize;
 682         struct nilfs_transaction_info ti;
 683         struct super_block *sb = inode->i_sb;
 684         struct nilfs_inode_info *ii = NILFS_I(inode);
 685
 686         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 687                 return;
 688         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 689                 return;
 690
 691         blocksize = sb->s_blocksize;
 692         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 693         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 694
 695         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 696
 697         nilfs_truncate_bmap(ii, blkoff);
 698
 699         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 700         if (IS_SYNC(inode))
 701                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 702
 703         nilfs_mark_inode_dirty(inode);
 704         nilfs_set_file_dirty(inode, 0);
 705         nilfs_transaction_commit(sb);
 706         /* May construct a logical segment and may fail in sync mode.
 707            But truncate has no return value. */
 708 }
 709
 710 static void nilfs_clear_inode(struct inode *inode)
 711 {
 712         struct nilfs_inode_info *ii = NILFS_I(inode);
 713         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 714
 715         /*
 716          * Free resources allocated in nilfs_read_inode(), here.
 717          */
 718         BUG_ON(!list_empty(&ii->i_dirty));
 719         brelse(ii->i_bh);
 720         ii->i_bh = NULL;
 721
 722         if (mdi && mdi->mi_palloc_cache)
 723                 nilfs_palloc_destroy_cache(inode);
 724
 725         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 726                 nilfs_bmap_clear(ii->i_bmap);
 727
 728         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 729
 730         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 731                 nilfs_put_root(ii->i_root);
 732 }
 733
 734 void nilfs_evict_inode(struct inode *inode)
 735 {
 736         struct nilfs_transaction_info ti;
 737         struct super_block *sb = inode->i_sb;
 738         struct nilfs_inode_info *ii = NILFS_I(inode);
 739         int ret;
 740
 741         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 742                 if (inode->i_data.nrpages)
 743                         truncate_inode_pages(&inode->i_data, 0);
 744                 clear_inode(inode);
 745                 nilfs_clear_inode(inode);
 746                 return;
 747         }
 748         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 749
 750         if (inode->i_data.nrpages)
 751                 truncate_inode_pages(&inode->i_data, 0);
 752
 753         /* TODO: some of the following operations may fail.  */
 754         nilfs_truncate_bmap(ii, 0);
 755         nilfs_mark_inode_dirty(inode);
 756         clear_inode(inode);
 757
 758         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 759         if (!ret)
 760                 atomic_dec(&ii->i_root->inodes_count);
 761
 762         nilfs_clear_inode(inode);
 763
 764         if (IS_SYNC(inode))
 765                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 766         nilfs_transaction_commit(sb);
 767         /* May construct a logical segment and may fail in sync mode.
 768            But delete_inode has no return value. */
 769 }
 770
 771 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 772 {
 773         struct nilfs_transaction_info ti;
 774         struct inode *inode = dentry->d_inode;
 775         struct super_block *sb = inode->i_sb;
 776         int err;
 777
 778         err = inode_change_ok(inode, iattr);
 779         if (err)
 780                 return err;
 781
 782         err = nilfs_transaction_begin(sb, &ti, 0);
 783         if (unlikely(err))
 784                 return err;
 785
 786         if ((iattr->ia_valid & ATTR_SIZE) &&
 787             iattr->ia_size != i_size_read(inode)) {
 788                 inode_dio_wait(inode);
 789
 790                 err = vmtruncate(inode, iattr->ia_size);
 791                 if (unlikely(err))
 792                         goto out_err;
 793         }
 794
 795         setattr_copy(inode, iattr);
 796         mark_inode_dirty(inode);
 797
 798         if (iattr->ia_valid & ATTR_MODE) {
 799                 err = nilfs_acl_chmod(inode);
 800                 if (unlikely(err))
 801                         goto out_err;
 802         }
 803
 804         return nilfs_transaction_commit(sb);
 805
 806 out_err:
 807         nilfs_transaction_abort(sb);
 808         return err;
 809 }
 810
 811 int nilfs_permission(struct inode *inode, int mask)
 812 {
 813         struct nilfs_root *root = NILFS_I(inode)->i_root;
 814         if ((mask & MAY_WRITE) && root &&
 815             root->cno != NILFS_CPTREE_CURRENT_CNO)
 816                 return -EROFS; /* snapshot is not writable */
 817
 818         return generic_permission(inode, mask);
 819 }
 820
 821 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 822 {
 823         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 824         struct nilfs_inode_info *ii = NILFS_I(inode);
 825         int err;
 826
 827         spin_lock(&nilfs->ns_inode_lock);
 828         if (ii->i_bh == NULL) {
 829                 spin_unlock(&nilfs->ns_inode_lock);
 830                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 831                                                   inode->i_ino, pbh);
 832                 if (unlikely(err))
 833                         return err;
 834                 spin_lock(&nilfs->ns_inode_lock);
 835                 if (ii->i_bh == NULL)
 836                         ii->i_bh = *pbh;
 837                 else {
 838                         brelse(*pbh);
 839                         *pbh = ii->i_bh;
 840                 }
 841         } else
 842                 *pbh = ii->i_bh;
 843
 844         get_bh(*pbh);
 845         spin_unlock(&nilfs->ns_inode_lock);
 846         return 0;
 847 }
 848
 849 int nilfs_inode_dirty(struct inode *inode)
 850 {
 851         struct nilfs_inode_info *ii = NILFS_I(inode);
 852         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 853         int ret = 0;
 854
 855         if (!list_empty(&ii->i_dirty)) {
 856                 spin_lock(&nilfs->ns_inode_lock);
 857                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 858                         test_bit(NILFS_I_BUSY, &ii->i_state);
 859                 spin_unlock(&nilfs->ns_inode_lock);
 860         }
 861         return ret;
 862 }
 863
 864 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 865 {
 866         struct nilfs_inode_info *ii = NILFS_I(inode);
 867         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 868
 869         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 870
 871         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 872                 return 0;
 873
 874         spin_lock(&nilfs->ns_inode_lock);
 875         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 876             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 877                 /* Because this routine may race with nilfs_dispose_list(),
 878                    we have to check NILFS_I_QUEUED here, too. */
 879                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 880                         /* This will happen when somebody is freeing
 881                            this inode. */
 882                         nilfs_warning(inode->i_sb, __func__,
 883                                       "cannot get inode (ino=%lu)\n",
 884                                       inode->i_ino);
 885                         spin_unlock(&nilfs->ns_inode_lock);
 886                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 887                                            freeing inode */
 888                 }
 889                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 890                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 891         }
 892         spin_unlock(&nilfs->ns_inode_lock);
 893         return 0;
 894 }
 895
 896 int nilfs_mark_inode_dirty(struct inode *inode)
 897 {
 898         struct buffer_head *ibh;
 899         int err;
 900
 901         err = nilfs_load_inode_block(inode, &ibh);
 902         if (unlikely(err)) {
 903                 nilfs_warning(inode->i_sb, __func__,
 904                               "failed to reget inode block.\n");
 905                 return err;
 906         }
 907         nilfs_update_inode(inode, ibh);
 908         mark_buffer_dirty(ibh);
 909         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 910         brelse(ibh);
 911         return 0;
 912 }
 913
 914 /**
 915  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 916  * @inode: inode of the file to be registered.
 917  *
 918  * nilfs_dirty_inode() loads a inode block containing the specified
 919  * @inode and copies data from a nilfs_inode to a corresponding inode
 920  * entry in the inode block. This operation is excluded from the segment
 921  * construction. This function can be called both as a single operation
 922  * and as a part of indivisible file operations.
 923  */
 924 void nilfs_dirty_inode(struct inode *inode, int flags)
 925 {
 926         struct nilfs_transaction_info ti;
 927         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 928
 929         if (is_bad_inode(inode)) {
 930                 nilfs_warning(inode->i_sb, __func__,
 931                               "tried to mark bad_inode dirty. ignored.\n");
 932                 dump_stack();
 933                 return;
 934         }
 935         if (mdi) {
 936                 nilfs_mdt_mark_dirty(inode);
 937                 return;
 938         }
 939         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 940         nilfs_mark_inode_dirty(inode);
 941         nilfs_transaction_commit(inode->i_sb); /* never fails */
 942 }
 943
 944 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 945                  __u64 start, __u64 len)
 946 {
 947         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 948         __u64 logical = 0, phys = 0, size = 0;
 949         __u32 flags = 0;
 950         loff_t isize;
 951         sector_t blkoff, end_blkoff;
 952         sector_t delalloc_blkoff;
 953         unsigned long delalloc_blklen;
 954         unsigned int blkbits = inode->i_blkbits;
 955         int ret, n;
 956
 957         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 958         if (ret)
 959                 return ret;
 960
 961         mutex_lock(&inode->i_mutex);
 962
 963         isize = i_size_read(inode);
 964
 965         blkoff = start >> blkbits;
 966         end_blkoff = (start + len - 1) >> blkbits;
 967
 968         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 969                                                         &delalloc_blkoff);
 970
 971         do {
 972                 __u64 blkphy;
 973                 unsigned int maxblocks;
 974
 975                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
 976                         if (size) {
 977                                 /* End of the current extent */
 978                                 ret = fiemap_fill_next_extent(
 979                                         fieinfo, logical, phys, size, flags);
 980                                 if (ret)
 981                                         break;
 982                         }
 983                         if (blkoff > end_blkoff)
 984                                 break;
 985
 986                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 987                         logical = blkoff << blkbits;
 988                         phys = 0;
 989                         size = delalloc_blklen << blkbits;
 990
 991                         blkoff = delalloc_blkoff + delalloc_blklen;
 992                         delalloc_blklen = nilfs_find_uncommitted_extent(
 993                                 inode, blkoff, &delalloc_blkoff);
 994                         continue;
 995                 }
 996
 997                 /*
 998                  * Limit the number of blocks that we look up so as
 999                  * not to get into the next delayed allocation extent.
1000                  */
1001                 maxblocks = INT_MAX;
1002                 if (delalloc_blklen)
1003                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1004                                           maxblocks);
1005                 blkphy = 0;
1006
1007                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1008                 n = nilfs_bmap_lookup_contig(
1009                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1010                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1011
1012                 if (n < 0) {
1013                         int past_eof;
1014
1015                         if (unlikely(n != -ENOENT))
1016                                 break; /* error */
1017
1018                         /* HOLE */
1019                         blkoff++;
1020                         past_eof = ((blkoff << blkbits) >= isize);
1021
1022                         if (size) {
1023                                 /* End of the current extent */
1024
1025                                 if (past_eof)
1026                                         flags |= FIEMAP_EXTENT_LAST;
1027
1028                                 ret = fiemap_fill_next_extent(
1029                                         fieinfo, logical, phys, size, flags);
1030                                 if (ret)
1031                                         break;
1032                                 size = 0;
1033                         }
1034                         if (blkoff > end_blkoff || past_eof)
1035                                 break;
1036                 } else {
1037                         if (size) {
1038                                 if (phys && blkphy << blkbits == phys + size) {
1039                                         /* The current extent goes on */
1040                                         size += n << blkbits;
1041                                 } else {
1042                                         /* Terminate the current extent */
1043                                         ret = fiemap_fill_next_extent(
1044                                                 fieinfo, logical, phys, size,
1045                                                 flags);
1046                                         if (ret || blkoff > end_blkoff)
1047                                                 break;
1048
1049                                         /* Start another extent */
1050                                         flags = FIEMAP_EXTENT_MERGED;
1051                                         logical = blkoff << blkbits;
1052                                         phys = blkphy << blkbits;
1053                                         size = n << blkbits;
1054                                 }
1055                         } else {
1056                                 /* Start a new extent */
1057                                 flags = FIEMAP_EXTENT_MERGED;
1058                                 logical = blkoff << blkbits;
1059                                 phys = blkphy << blkbits;
1060                                 size = n << blkbits;
1061                         }
1062                         blkoff += n;
1063                 }
1064                 cond_resched();
1065         } while (true);
1066
1067         /* If ret is 1 then we just hit the end of the extent array */
1068         if (ret == 1)
1069                 ret = 0;
1070
1071         mutex_unlock(&inode->i_mutex);
1072         return ret;
1073 }