fs/f2fs/data.c

   1 /*
   2  * fs/f2fs/data.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/mpage.h>
  15 #include <linux/writeback.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21 #include <linux/uio.h>
  22 #include <linux/cleancache.h>
  23
  24 #include "f2fs.h"
  25 #include "node.h"
  26 #include "segment.h"
  27 #include "trace.h"
  28 #include <trace/events/f2fs.h>
  29
  30 static void f2fs_read_end_io(struct bio *bio)
  31 {
  32         struct bio_vec *bvec;
  33         int i;
  34
  35         if (f2fs_bio_encrypted(bio)) {
  36                 if (bio->bi_error) {
  37                         f2fs_release_crypto_ctx(bio->bi_private);
  38                 } else {
  39                         f2fs_end_io_crypto_work(bio->bi_private, bio);
  40                         return;
  41                 }
  42         }
  43
  44         bio_for_each_segment_all(bvec, bio, i) {
  45                 struct page *page = bvec->bv_page;
  46
  47                 if (!bio->bi_error) {
  48                         SetPageUptodate(page);
  49                 } else {
  50                         ClearPageUptodate(page);
  51                         SetPageError(page);
  52                 }
  53                 unlock_page(page);
  54         }
  55         bio_put(bio);
  56 }
  57
  58 static void f2fs_write_end_io(struct bio *bio)
  59 {
  60         struct f2fs_sb_info *sbi = bio->bi_private;
  61         struct bio_vec *bvec;
  62         int i;
  63
  64         bio_for_each_segment_all(bvec, bio, i) {
  65                 struct page *page = bvec->bv_page;
  66
  67                 f2fs_restore_and_release_control_page(&page);
  68
  69                 if (unlikely(bio->bi_error)) {
  70                         set_page_dirty(page);
  71                         set_bit(AS_EIO, &page->mapping->flags);
  72                         f2fs_stop_checkpoint(sbi);
  73                 }
  74                 end_page_writeback(page);
  75                 dec_page_count(sbi, F2FS_WRITEBACK);
  76         }
  77
  78         if (!get_pages(sbi, F2FS_WRITEBACK) &&
  79                         !list_empty(&sbi->cp_wait.task_list))
  80                 wake_up(&sbi->cp_wait);
  81
  82         bio_put(bio);
  83 }
  84
  85 /*
  86  * Low-level block read/write IO operations.
  87  */
  88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  89                                 int npages, bool is_read)
  90 {
  91         struct bio *bio;
  92
  93         bio = f2fs_bio_alloc(npages);
  94
  95         bio->bi_bdev = sbi->sb->s_bdev;
  96         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  97         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  98         bio->bi_private = is_read ? NULL : sbi;
  99
 100         return bio;
 101 }
 102
 103 static void __submit_merged_bio(struct f2fs_bio_info *io)
 104 {
 105         struct f2fs_io_info *fio = &io->fio;
 106
 107         if (!io->bio)
 108                 return;
 109
 110         if (is_read_io(fio->rw))
 111                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 112         else
 113                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 114
 115         submit_bio(fio->rw, io->bio);
 116         io->bio = NULL;
 117 }
 118
 119 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 120                                 enum page_type type, int rw)
 121 {
 122         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 123         struct f2fs_bio_info *io;
 124
 125         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 126
 127         down_write(&io->io_rwsem);
 128
 129         /* change META to META_FLUSH in the checkpoint procedure */
 130         if (type >= META_FLUSH) {
 131                 io->fio.type = META_FLUSH;
 132                 if (test_opt(sbi, NOBARRIER))
 133                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 134                 else
 135                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 136         }
 137         __submit_merged_bio(io);
 138         up_write(&io->io_rwsem);
 139 }
 140
 141 /*
 142  * Fill the locked page with data located in the block address.
 143  * Return unlocked page.
 144  */
 145 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 146 {
 147         struct bio *bio;
 148         struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 149
 150         trace_f2fs_submit_page_bio(page, fio);
 151         f2fs_trace_ios(fio, 0);
 152
 153         /* Allocate a new bio */
 154         bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
 155
 156         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 157                 bio_put(bio);
 158                 return -EFAULT;
 159         }
 160
 161         submit_bio(fio->rw, bio);
 162         return 0;
 163 }
 164
 165 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 166 {
 167         struct f2fs_sb_info *sbi = fio->sbi;
 168         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 169         struct f2fs_bio_info *io;
 170         bool is_read = is_read_io(fio->rw);
 171         struct page *bio_page;
 172
 173         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 174
 175         verify_block_addr(sbi, fio->blk_addr);
 176
 177         down_write(&io->io_rwsem);
 178
 179         if (!is_read)
 180                 inc_page_count(sbi, F2FS_WRITEBACK);
 181
 182         if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
 183                                                 io->fio.rw != fio->rw))
 184                 __submit_merged_bio(io);
 185 alloc_new:
 186         if (io->bio == NULL) {
 187                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 188
 189                 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
 190                 io->fio = *fio;
 191         }
 192
 193         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 194
 195         if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
 196                                                         PAGE_CACHE_SIZE) {
 197                 __submit_merged_bio(io);
 198                 goto alloc_new;
 199         }
 200
 201         io->last_block_in_bio = fio->blk_addr;
 202         f2fs_trace_ios(fio, 0);
 203
 204         up_write(&io->io_rwsem);
 205         trace_f2fs_submit_page_mbio(fio->page, fio);
 206 }
 207
 208 /*
 209  * Lock ordering for the change of data block address:
 210  * ->data_page
 211  *  ->node_page
 212  *    update block addresses in the node page
 213  */
 214 void set_data_blkaddr(struct dnode_of_data *dn)
 215 {
 216         struct f2fs_node *rn;
 217         __le32 *addr_array;
 218         struct page *node_page = dn->node_page;
 219         unsigned int ofs_in_node = dn->ofs_in_node;
 220
 221         f2fs_wait_on_page_writeback(node_page, NODE);
 222
 223         rn = F2FS_NODE(node_page);
 224
 225         /* Get physical address of data block */
 226         addr_array = blkaddr_in_node(rn);
 227         addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 228         set_page_dirty(node_page);
 229 }
 230
 231 int reserve_new_block(struct dnode_of_data *dn)
 232 {
 233         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 234
 235         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 236                 return -EPERM;
 237         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 238                 return -ENOSPC;
 239
 240         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 241
 242         dn->data_blkaddr = NEW_ADDR;
 243         set_data_blkaddr(dn);
 244         mark_inode_dirty(dn->inode);
 245         sync_inode_page(dn);
 246         return 0;
 247 }
 248
 249 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 250 {
 251         bool need_put = dn->inode_page ? false : true;
 252         int err;
 253
 254         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 255         if (err)
 256                 return err;
 257
 258         if (dn->data_blkaddr == NULL_ADDR)
 259                 err = reserve_new_block(dn);
 260         if (err || need_put)
 261                 f2fs_put_dnode(dn);
 262         return err;
 263 }
 264
 265 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 266 {
 267         struct extent_info ei;
 268         struct inode *inode = dn->inode;
 269
 270         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 271                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 272                 return 0;
 273         }
 274
 275         return f2fs_reserve_block(dn, index);
 276 }
 277
 278 struct page *get_read_data_page(struct inode *inode, pgoff_t index, int rw)
 279 {
 280         struct address_space *mapping = inode->i_mapping;
 281         struct dnode_of_data dn;
 282         struct page *page;
 283         struct extent_info ei;
 284         int err;
 285         struct f2fs_io_info fio = {
 286                 .sbi = F2FS_I_SB(inode),
 287                 .type = DATA,
 288                 .rw = rw,
 289                 .encrypted_page = NULL,
 290         };
 291
 292         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 293                 return read_mapping_page(mapping, index, NULL);
 294
 295         page = grab_cache_page(mapping, index);
 296         if (!page)
 297                 return ERR_PTR(-ENOMEM);
 298
 299         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 300                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 301                 goto got_it;
 302         }
 303
 304         set_new_dnode(&dn, inode, NULL, NULL, 0);
 305         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 306         if (err)
 307                 goto put_err;
 308         f2fs_put_dnode(&dn);
 309
 310         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 311                 err = -ENOENT;
 312                 goto put_err;
 313         }
 314 got_it:
 315         if (PageUptodate(page)) {
 316                 unlock_page(page);
 317                 return page;
 318         }
 319
 320         /*
 321          * A new dentry page is allocated but not able to be written, since its
 322          * new inode page couldn't be allocated due to -ENOSPC.
 323          * In such the case, its blkaddr can be remained as NEW_ADDR.
 324          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 325          */
 326         if (dn.data_blkaddr == NEW_ADDR) {
 327                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 328                 SetPageUptodate(page);
 329                 unlock_page(page);
 330                 return page;
 331         }
 332
 333         fio.blk_addr = dn.data_blkaddr;
 334         fio.page = page;
 335         err = f2fs_submit_page_bio(&fio);
 336         if (err)
 337                 goto put_err;
 338         return page;
 339
 340 put_err:
 341         f2fs_put_page(page, 1);
 342         return ERR_PTR(err);
 343 }
 344
 345 struct page *find_data_page(struct inode *inode, pgoff_t index)
 346 {
 347         struct address_space *mapping = inode->i_mapping;
 348         struct page *page;
 349
 350         page = find_get_page(mapping, index);
 351         if (page && PageUptodate(page))
 352                 return page;
 353         f2fs_put_page(page, 0);
 354
 355         page = get_read_data_page(inode, index, READ_SYNC);
 356         if (IS_ERR(page))
 357                 return page;
 358
 359         if (PageUptodate(page))
 360                 return page;
 361
 362         wait_on_page_locked(page);
 363         if (unlikely(!PageUptodate(page))) {
 364                 f2fs_put_page(page, 0);
 365                 return ERR_PTR(-EIO);
 366         }
 367         return page;
 368 }
 369
 370 /*
 371  * If it tries to access a hole, return an error.
 372  * Because, the callers, functions in dir.c and GC, should be able to know
 373  * whether this page exists or not.
 374  */
 375 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 376 {
 377         struct address_space *mapping = inode->i_mapping;
 378         struct page *page;
 379 repeat:
 380         page = get_read_data_page(inode, index, READ_SYNC);
 381         if (IS_ERR(page))
 382                 return page;
 383
 384         /* wait for read completion */
 385         lock_page(page);
 386         if (unlikely(!PageUptodate(page))) {
 387                 f2fs_put_page(page, 1);
 388                 return ERR_PTR(-EIO);
 389         }
 390         if (unlikely(page->mapping != mapping)) {
 391                 f2fs_put_page(page, 1);
 392                 goto repeat;
 393         }
 394         return page;
 395 }
 396
 397 /*
 398  * Caller ensures that this data page is never allocated.
 399  * A new zero-filled data page is allocated in the page cache.
 400  *
 401  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 402  * f2fs_unlock_op().
 403  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 404  * ipage should be released by this function.
 405  */
 406 struct page *get_new_data_page(struct inode *inode,
 407                 struct page *ipage, pgoff_t index, bool new_i_size)
 408 {
 409         struct address_space *mapping = inode->i_mapping;
 410         struct page *page;
 411         struct dnode_of_data dn;
 412         int err;
 413 repeat:
 414         page = grab_cache_page(mapping, index);
 415         if (!page) {
 416                 /*
 417                  * before exiting, we should make sure ipage will be released
 418                  * if any error occur.
 419                  */
 420                 f2fs_put_page(ipage, 1);
 421                 return ERR_PTR(-ENOMEM);
 422         }
 423
 424         set_new_dnode(&dn, inode, ipage, NULL, 0);
 425         err = f2fs_reserve_block(&dn, index);
 426         if (err) {
 427                 f2fs_put_page(page, 1);
 428                 return ERR_PTR(err);
 429         }
 430         if (!ipage)
 431                 f2fs_put_dnode(&dn);
 432
 433         if (PageUptodate(page))
 434                 goto got_it;
 435
 436         if (dn.data_blkaddr == NEW_ADDR) {
 437                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 438                 SetPageUptodate(page);
 439         } else {
 440                 f2fs_put_page(page, 1);
 441
 442                 page = get_read_data_page(inode, index, READ_SYNC);
 443                 if (IS_ERR(page))
 444                         goto repeat;
 445
 446                 /* wait for read completion */
 447                 lock_page(page);
 448         }
 449 got_it:
 450         if (new_i_size && i_size_read(inode) <
 451                                 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
 452                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
 453                 /* Only the directory inode sets new_i_size */
 454                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 455         }
 456         return page;
 457 }
 458
 459 static int __allocate_data_block(struct dnode_of_data *dn)
 460 {
 461         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 462         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 463         struct f2fs_summary sum;
 464         struct node_info ni;
 465         int seg = CURSEG_WARM_DATA;
 466         pgoff_t fofs;
 467
 468         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 469                 return -EPERM;
 470
 471         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 472         if (dn->data_blkaddr == NEW_ADDR)
 473                 goto alloc;
 474
 475         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 476                 return -ENOSPC;
 477
 478 alloc:
 479         get_node_info(sbi, dn->nid, &ni);
 480         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 481
 482         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 483                 seg = CURSEG_DIRECT_IO;
 484
 485         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 486                                                                 &sum, seg);
 487         set_data_blkaddr(dn);
 488
 489         /* update i_size */
 490         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 491                                                         dn->ofs_in_node;
 492         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
 493                 i_size_write(dn->inode,
 494                                 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
 495
 496         /* direct IO doesn't use extent cache to maximize the performance */
 497         f2fs_drop_largest_extent(dn->inode, fofs);
 498
 499         return 0;
 500 }
 501
 502 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
 503                                                         size_t count)
 504 {
 505         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 506         struct dnode_of_data dn;
 507         u64 start = F2FS_BYTES_TO_BLK(offset);
 508         u64 len = F2FS_BYTES_TO_BLK(count);
 509         bool allocated;
 510         u64 end_offset;
 511
 512         while (len) {
 513                 f2fs_balance_fs(sbi);
 514                 f2fs_lock_op(sbi);
 515
 516                 /* When reading holes, we need its node page */
 517                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 518                 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
 519                         goto out;
 520
 521                 allocated = false;
 522                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 523
 524                 while (dn.ofs_in_node < end_offset && len) {
 525                         block_t blkaddr;
 526
 527                         if (unlikely(f2fs_cp_error(sbi)))
 528                                 goto sync_out;
 529
 530                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 531                         if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
 532                                 if (__allocate_data_block(&dn))
 533                                         goto sync_out;
 534                                 allocated = true;
 535                         }
 536                         len--;
 537                         start++;
 538                         dn.ofs_in_node++;
 539                 }
 540
 541                 if (allocated)
 542                         sync_inode_page(&dn);
 543
 544                 f2fs_put_dnode(&dn);
 545                 f2fs_unlock_op(sbi);
 546         }
 547         return;
 548
 549 sync_out:
 550         if (allocated)
 551                 sync_inode_page(&dn);
 552         f2fs_put_dnode(&dn);
 553 out:
 554         f2fs_unlock_op(sbi);
 555         return;
 556 }
 557
 558 /*
 559  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 560  * f2fs_map_blocks structure.
 561  * If original data blocks are allocated, then give them to blockdev.
 562  * Otherwise,
 563  *     a. preallocate requested block addresses
 564  *     b. do not use extent cache for better performance
 565  *     c. give the block addresses to blockdev
 566  */
 567 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 568                                                 int create, int flag)
 569 {
 570         unsigned int maxblocks = map->m_len;
 571         struct dnode_of_data dn;
 572         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 573         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 574         pgoff_t pgofs, end_offset;
 575         int err = 0, ofs = 1;
 576         struct extent_info ei;
 577         bool allocated = false;
 578
 579         map->m_len = 0;
 580         map->m_flags = 0;
 581
 582         /* it only supports block size == page size */
 583         pgofs = (pgoff_t)map->m_lblk;
 584
 585         if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 586                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 587                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 588                 map->m_flags = F2FS_MAP_MAPPED;
 589                 goto out;
 590         }
 591
 592         if (create)
 593                 f2fs_lock_op(F2FS_I_SB(inode));
 594
 595         /* When reading holes, we need its node page */
 596         set_new_dnode(&dn, inode, NULL, NULL, 0);
 597         err = get_dnode_of_data(&dn, pgofs, mode);
 598         if (err) {
 599                 if (err == -ENOENT)
 600                         err = 0;
 601                 goto unlock_out;
 602         }
 603
 604         if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
 605                 if (create) {
 606                         if (unlikely(f2fs_cp_error(sbi))) {
 607                                 err = -EIO;
 608                                 goto put_out;
 609                         }
 610                         err = __allocate_data_block(&dn);
 611                         if (err)
 612                                 goto put_out;
 613                         allocated = true;
 614                         map->m_flags = F2FS_MAP_NEW;
 615                 } else {
 616                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 617                                                 dn.data_blkaddr != NEW_ADDR) {
 618                                 if (flag == F2FS_GET_BLOCK_BMAP)
 619                                         err = -ENOENT;
 620                                 goto put_out;
 621                         }
 622
 623                         /*
 624                          * preallocated unwritten block should be mapped
 625                          * for fiemap.
 626                          */
 627                         if (dn.data_blkaddr == NEW_ADDR)
 628                                 map->m_flags = F2FS_MAP_UNWRITTEN;
 629                 }
 630         }
 631
 632         map->m_flags |= F2FS_MAP_MAPPED;
 633         map->m_pblk = dn.data_blkaddr;
 634         map->m_len = 1;
 635
 636         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 637         dn.ofs_in_node++;
 638         pgofs++;
 639
 640 get_next:
 641         if (dn.ofs_in_node >= end_offset) {
 642                 if (allocated)
 643                         sync_inode_page(&dn);
 644                 allocated = false;
 645                 f2fs_put_dnode(&dn);
 646
 647                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 648                 err = get_dnode_of_data(&dn, pgofs, mode);
 649                 if (err) {
 650                         if (err == -ENOENT)
 651                                 err = 0;
 652                         goto unlock_out;
 653                 }
 654
 655                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 656         }
 657
 658         if (maxblocks > map->m_len) {
 659                 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 660
 661                 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 662                         if (create) {
 663                                 if (unlikely(f2fs_cp_error(sbi))) {
 664                                         err = -EIO;
 665                                         goto sync_out;
 666                                 }
 667                                 err = __allocate_data_block(&dn);
 668                                 if (err)
 669                                         goto sync_out;
 670                                 allocated = true;
 671                                 map->m_flags |= F2FS_MAP_NEW;
 672                                 blkaddr = dn.data_blkaddr;
 673                         } else {
 674                                 /*
 675                                  * we only merge preallocated unwritten blocks
 676                                  * for fiemap.
 677                                  */
 678                                 if (flag != F2FS_GET_BLOCK_FIEMAP ||
 679                                                 blkaddr != NEW_ADDR)
 680                                         goto sync_out;
 681                         }
 682                 }
 683
 684                 /* Give more consecutive addresses for the readahead */
 685                 if ((map->m_pblk != NEW_ADDR &&
 686                                 blkaddr == (map->m_pblk + ofs)) ||
 687                                 (map->m_pblk == NEW_ADDR &&
 688                                 blkaddr == NEW_ADDR)) {
 689                         ofs++;
 690                         dn.ofs_in_node++;
 691                         pgofs++;
 692                         map->m_len++;
 693                         goto get_next;
 694                 }
 695         }
 696 sync_out:
 697         if (allocated)
 698                 sync_inode_page(&dn);
 699 put_out:
 700         f2fs_put_dnode(&dn);
 701 unlock_out:
 702         if (create)
 703                 f2fs_unlock_op(F2FS_I_SB(inode));
 704 out:
 705         trace_f2fs_map_blocks(inode, map, err);
 706         return err;
 707 }
 708
 709 static int __get_data_block(struct inode *inode, sector_t iblock,
 710                         struct buffer_head *bh, int create, int flag)
 711 {
 712         struct f2fs_map_blocks map;
 713         int ret;
 714
 715         map.m_lblk = iblock;
 716         map.m_len = bh->b_size >> inode->i_blkbits;
 717
 718         ret = f2fs_map_blocks(inode, &map, create, flag);
 719         if (!ret) {
 720                 map_bh(bh, inode->i_sb, map.m_pblk);
 721                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 722                 bh->b_size = map.m_len << inode->i_blkbits;
 723         }
 724         return ret;
 725 }
 726
 727 static int get_data_block(struct inode *inode, sector_t iblock,
 728                         struct buffer_head *bh_result, int create, int flag)
 729 {
 730         return __get_data_block(inode, iblock, bh_result, create, flag);
 731 }
 732
 733 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 734                         struct buffer_head *bh_result, int create)
 735 {
 736         return __get_data_block(inode, iblock, bh_result, create,
 737                                                 F2FS_GET_BLOCK_DIO);
 738 }
 739
 740 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 741                         struct buffer_head *bh_result, int create)
 742 {
 743         return __get_data_block(inode, iblock, bh_result, create,
 744                                                 F2FS_GET_BLOCK_BMAP);
 745 }
 746
 747 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 748 {
 749         return (offset >> inode->i_blkbits);
 750 }
 751
 752 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 753 {
 754         return (blk << inode->i_blkbits);
 755 }
 756
 757 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 758                 u64 start, u64 len)
 759 {
 760         struct buffer_head map_bh;
 761         sector_t start_blk, last_blk;
 762         loff_t isize = i_size_read(inode);
 763         u64 logical = 0, phys = 0, size = 0;
 764         u32 flags = 0;
 765         bool past_eof = false, whole_file = false;
 766         int ret = 0;
 767
 768         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 769         if (ret)
 770                 return ret;
 771
 772         mutex_lock(&inode->i_mutex);
 773
 774         if (len >= isize) {
 775                 whole_file = true;
 776                 len = isize;
 777         }
 778
 779         if (logical_to_blk(inode, len) == 0)
 780                 len = blk_to_logical(inode, 1);
 781
 782         start_blk = logical_to_blk(inode, start);
 783         last_blk = logical_to_blk(inode, start + len - 1);
 784 next:
 785         memset(&map_bh, 0, sizeof(struct buffer_head));
 786         map_bh.b_size = len;
 787
 788         ret = get_data_block(inode, start_blk, &map_bh, 0,
 789                                         F2FS_GET_BLOCK_FIEMAP);
 790         if (ret)
 791                 goto out;
 792
 793         /* HOLE */
 794         if (!buffer_mapped(&map_bh)) {
 795                 start_blk++;
 796
 797                 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
 798                         past_eof = 1;
 799
 800                 if (past_eof && size) {
 801                         flags |= FIEMAP_EXTENT_LAST;
 802                         ret = fiemap_fill_next_extent(fieinfo, logical,
 803                                         phys, size, flags);
 804                 } else if (size) {
 805                         ret = fiemap_fill_next_extent(fieinfo, logical,
 806                                         phys, size, flags);
 807                         size = 0;
 808                 }
 809
 810                 /* if we have holes up to/past EOF then we're done */
 811                 if (start_blk > last_blk || past_eof || ret)
 812                         goto out;
 813         } else {
 814                 if (start_blk > last_blk && !whole_file) {
 815                         ret = fiemap_fill_next_extent(fieinfo, logical,
 816                                         phys, size, flags);
 817                         goto out;
 818                 }
 819
 820                 /*
 821                  * if size != 0 then we know we already have an extent
 822                  * to add, so add it.
 823                  */
 824                 if (size) {
 825                         ret = fiemap_fill_next_extent(fieinfo, logical,
 826                                         phys, size, flags);
 827                         if (ret)
 828                                 goto out;
 829                 }
 830
 831                 logical = blk_to_logical(inode, start_blk);
 832                 phys = blk_to_logical(inode, map_bh.b_blocknr);
 833                 size = map_bh.b_size;
 834                 flags = 0;
 835                 if (buffer_unwritten(&map_bh))
 836                         flags = FIEMAP_EXTENT_UNWRITTEN;
 837
 838                 start_blk += logical_to_blk(inode, size);
 839
 840                 /*
 841                  * If we are past the EOF, then we need to make sure as
 842                  * soon as we find a hole that the last extent we found
 843                  * is marked with FIEMAP_EXTENT_LAST
 844                  */
 845                 if (!past_eof && logical + size >= isize)
 846                         past_eof = true;
 847         }
 848         cond_resched();
 849         if (fatal_signal_pending(current))
 850                 ret = -EINTR;
 851         else
 852                 goto next;
 853 out:
 854         if (ret == 1)
 855                 ret = 0;
 856
 857         mutex_unlock(&inode->i_mutex);
 858         return ret;
 859 }
 860
 861 /*
 862  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 863  * Major change was from block_size == page_size in f2fs by default.
 864  */
 865 static int f2fs_mpage_readpages(struct address_space *mapping,
 866                         struct list_head *pages, struct page *page,
 867                         unsigned nr_pages)
 868 {
 869         struct bio *bio = NULL;
 870         unsigned page_idx;
 871         sector_t last_block_in_bio = 0;
 872         struct inode *inode = mapping->host;
 873         const unsigned blkbits = inode->i_blkbits;
 874         const unsigned blocksize = 1 << blkbits;
 875         sector_t block_in_file;
 876         sector_t last_block;
 877         sector_t last_block_in_file;
 878         sector_t block_nr;
 879         struct block_device *bdev = inode->i_sb->s_bdev;
 880         struct f2fs_map_blocks map;
 881
 882         map.m_pblk = 0;
 883         map.m_lblk = 0;
 884         map.m_len = 0;
 885         map.m_flags = 0;
 886
 887         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 888
 889                 prefetchw(&page->flags);
 890                 if (pages) {
 891                         page = list_entry(pages->prev, struct page, lru);
 892                         list_del(&page->lru);
 893                         if (add_to_page_cache_lru(page, mapping,
 894                                                   page->index, GFP_KERNEL))
 895                                 goto next_page;
 896                 }
 897
 898                 block_in_file = (sector_t)page->index;
 899                 last_block = block_in_file + nr_pages;
 900                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 901                                                                 blkbits;
 902                 if (last_block > last_block_in_file)
 903                         last_block = last_block_in_file;
 904
 905                 /*
 906                  * Map blocks using the previous result first.
 907                  */
 908                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 909                                 block_in_file > map.m_lblk &&
 910                                 block_in_file < (map.m_lblk + map.m_len))
 911                         goto got_it;
 912
 913                 /*
 914                  * Then do more f2fs_map_blocks() calls until we are
 915                  * done with this page.
 916                  */
 917                 map.m_flags = 0;
 918
 919                 if (block_in_file < last_block) {
 920                         map.m_lblk = block_in_file;
 921                         map.m_len = last_block - block_in_file;
 922
 923                         if (f2fs_map_blocks(inode, &map, 0,
 924                                                         F2FS_GET_BLOCK_READ))
 925                                 goto set_error_page;
 926                 }
 927 got_it:
 928                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 929                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 930                         SetPageMappedToDisk(page);
 931
 932                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 933                                 SetPageUptodate(page);
 934                                 goto confused;
 935                         }
 936                 } else {
 937                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 938                         SetPageUptodate(page);
 939                         unlock_page(page);
 940                         goto next_page;
 941                 }
 942
 943                 /*
 944                  * This page will go to BIO.  Do we need to send this
 945                  * BIO off first?
 946                  */
 947                 if (bio && (last_block_in_bio != block_nr - 1)) {
 948 submit_and_realloc:
 949                         submit_bio(READ, bio);
 950                         bio = NULL;
 951                 }
 952                 if (bio == NULL) {
 953                         struct f2fs_crypto_ctx *ctx = NULL;
 954
 955                         if (f2fs_encrypted_inode(inode) &&
 956                                         S_ISREG(inode->i_mode)) {
 957                                 struct page *cpage;
 958
 959                                 ctx = f2fs_get_crypto_ctx(inode);
 960                                 if (IS_ERR(ctx))
 961                                         goto set_error_page;
 962
 963                                 /* wait the page to be moved by cleaning */
 964                                 cpage = find_lock_page(
 965                                                 META_MAPPING(F2FS_I_SB(inode)),
 966                                                 block_nr);
 967                                 if (cpage) {
 968                                         f2fs_wait_on_page_writeback(cpage,
 969                                                                         DATA);
 970                                         f2fs_put_page(cpage, 1);
 971                                 }
 972                         }
 973
 974                         bio = bio_alloc(GFP_KERNEL,
 975                                 min_t(int, nr_pages, BIO_MAX_PAGES));
 976                         if (!bio) {
 977                                 if (ctx)
 978                                         f2fs_release_crypto_ctx(ctx);
 979                                 goto set_error_page;
 980                         }
 981                         bio->bi_bdev = bdev;
 982                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
 983                         bio->bi_end_io = f2fs_read_end_io;
 984                         bio->bi_private = ctx;
 985                 }
 986
 987                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
 988                         goto submit_and_realloc;
 989
 990                 last_block_in_bio = block_nr;
 991                 goto next_page;
 992 set_error_page:
 993                 SetPageError(page);
 994                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 995                 unlock_page(page);
 996                 goto next_page;
 997 confused:
 998                 if (bio) {
 999                         submit_bio(READ, bio);
1000                         bio = NULL;
1001                 }
1002                 unlock_page(page);
1003 next_page:
1004                 if (pages)
1005                         page_cache_release(page);
1006         }
1007         BUG_ON(pages && !list_empty(pages));
1008         if (bio)
1009                 submit_bio(READ, bio);
1010         return 0;
1011 }
1012
1013 static int f2fs_read_data_page(struct file *file, struct page *page)
1014 {
1015         struct inode *inode = page->mapping->host;
1016         int ret = -EAGAIN;
1017
1018         trace_f2fs_readpage(page, DATA);
1019
1020         /* If the file has inline data, try to read it directly */
1021         if (f2fs_has_inline_data(inode))
1022                 ret = f2fs_read_inline_data(inode, page);
1023         if (ret == -EAGAIN)
1024                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1025         return ret;
1026 }
1027
1028 static int f2fs_read_data_pages(struct file *file,
1029                         struct address_space *mapping,
1030                         struct list_head *pages, unsigned nr_pages)
1031 {
1032         struct inode *inode = file->f_mapping->host;
1033
1034         /* If the file has inline data, skip readpages */
1035         if (f2fs_has_inline_data(inode))
1036                 return 0;
1037
1038         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1039 }
1040
1041 int do_write_data_page(struct f2fs_io_info *fio)
1042 {
1043         struct page *page = fio->page;
1044         struct inode *inode = page->mapping->host;
1045         struct dnode_of_data dn;
1046         int err = 0;
1047
1048         set_new_dnode(&dn, inode, NULL, NULL, 0);
1049         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1050         if (err)
1051                 return err;
1052
1053         fio->blk_addr = dn.data_blkaddr;
1054
1055         /* This page is already truncated */
1056         if (fio->blk_addr == NULL_ADDR) {
1057                 ClearPageUptodate(page);
1058                 goto out_writepage;
1059         }
1060
1061         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1062                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1063                 if (IS_ERR(fio->encrypted_page)) {
1064                         err = PTR_ERR(fio->encrypted_page);
1065                         goto out_writepage;
1066                 }
1067         }
1068
1069         set_page_writeback(page);
1070
1071         /*
1072          * If current allocation needs SSR,
1073          * it had better in-place writes for updated data.
1074          */
1075         if (unlikely(fio->blk_addr != NEW_ADDR &&
1076                         !is_cold_data(page) &&
1077                         need_inplace_update(inode))) {
1078                 rewrite_data_page(fio);
1079                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1080                 trace_f2fs_do_write_data_page(page, IPU);
1081         } else {
1082                 write_data_page(&dn, fio);
1083                 set_data_blkaddr(&dn);
1084                 f2fs_update_extent_cache(&dn);
1085                 trace_f2fs_do_write_data_page(page, OPU);
1086                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1087                 if (page->index == 0)
1088                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1089         }
1090 out_writepage:
1091         f2fs_put_dnode(&dn);
1092         return err;
1093 }
1094
1095 static int f2fs_write_data_page(struct page *page,
1096                                         struct writeback_control *wbc)
1097 {
1098         struct inode *inode = page->mapping->host;
1099         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1100         loff_t i_size = i_size_read(inode);
1101         const pgoff_t end_index = ((unsigned long long) i_size)
1102                                                         >> PAGE_CACHE_SHIFT;
1103         unsigned offset = 0;
1104         bool need_balance_fs = false;
1105         int err = 0;
1106         struct f2fs_io_info fio = {
1107                 .sbi = sbi,
1108                 .type = DATA,
1109                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1110                 .page = page,
1111                 .encrypted_page = NULL,
1112         };
1113
1114         trace_f2fs_writepage(page, DATA);
1115
1116         if (page->index < end_index)
1117                 goto write;
1118
1119         /*
1120          * If the offset is out-of-range of file size,
1121          * this page does not have to be written to disk.
1122          */
1123         offset = i_size & (PAGE_CACHE_SIZE - 1);
1124         if ((page->index >= end_index + 1) || !offset)
1125                 goto out;
1126
1127         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1128 write:
1129         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1130                 goto redirty_out;
1131         if (f2fs_is_drop_cache(inode))
1132                 goto out;
1133         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1134                         available_free_memory(sbi, BASE_CHECK))
1135                 goto redirty_out;
1136
1137         /* Dentry blocks are controlled by checkpoint */
1138         if (S_ISDIR(inode->i_mode)) {
1139                 if (unlikely(f2fs_cp_error(sbi)))
1140                         goto redirty_out;
1141                 err = do_write_data_page(&fio);
1142                 goto done;
1143         }
1144
1145         /* we should bypass data pages to proceed the kworkder jobs */
1146         if (unlikely(f2fs_cp_error(sbi))) {
1147                 SetPageError(page);
1148                 goto out;
1149         }
1150
1151         if (!wbc->for_reclaim)
1152                 need_balance_fs = true;
1153         else if (has_not_enough_free_secs(sbi, 0))
1154                 goto redirty_out;
1155
1156         err = -EAGAIN;
1157         f2fs_lock_op(sbi);
1158         if (f2fs_has_inline_data(inode))
1159                 err = f2fs_write_inline_data(inode, page);
1160         if (err == -EAGAIN)
1161                 err = do_write_data_page(&fio);
1162         f2fs_unlock_op(sbi);
1163 done:
1164         if (err && err != -ENOENT)
1165                 goto redirty_out;
1166
1167         clear_cold_data(page);
1168 out:
1169         inode_dec_dirty_pages(inode);
1170         if (err)
1171                 ClearPageUptodate(page);
1172         unlock_page(page);
1173         if (need_balance_fs)
1174                 f2fs_balance_fs(sbi);
1175         if (wbc->for_reclaim)
1176                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1177         return 0;
1178
1179 redirty_out:
1180         redirty_page_for_writepage(wbc, page);
1181         return AOP_WRITEPAGE_ACTIVATE;
1182 }
1183
1184 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1185                         void *data)
1186 {
1187         struct address_space *mapping = data;
1188         int ret = mapping->a_ops->writepage(page, wbc);
1189         mapping_set_error(mapping, ret);
1190         return ret;
1191 }
1192
1193 /*
1194  * This function was copied from write_cche_pages from mm/page-writeback.c.
1195  * The major change is making write step of cold data page separately from
1196  * warm/hot data page.
1197  */
1198 static int f2fs_write_cache_pages(struct address_space *mapping,
1199                         struct writeback_control *wbc, writepage_t writepage,
1200                         void *data)
1201 {
1202         int ret = 0;
1203         int done = 0;
1204         struct pagevec pvec;
1205         int nr_pages;
1206         pgoff_t uninitialized_var(writeback_index);
1207         pgoff_t index;
1208         pgoff_t end;            /* Inclusive */
1209         pgoff_t done_index;
1210         int cycled;
1211         int range_whole = 0;
1212         int tag;
1213         int step = 0;
1214
1215         pagevec_init(&pvec, 0);
1216 next:
1217         if (wbc->range_cyclic) {
1218                 writeback_index = mapping->writeback_index; /* prev offset */
1219                 index = writeback_index;
1220                 if (index == 0)
1221                         cycled = 1;
1222                 else
1223                         cycled = 0;
1224                 end = -1;
1225         } else {
1226                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1227                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1228                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1229                         range_whole = 1;
1230                 cycled = 1; /* ignore range_cyclic tests */
1231         }
1232         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1233                 tag = PAGECACHE_TAG_TOWRITE;
1234         else
1235                 tag = PAGECACHE_TAG_DIRTY;
1236 retry:
1237         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1238                 tag_pages_for_writeback(mapping, index, end);
1239         done_index = index;
1240         while (!done && (index <= end)) {
1241                 int i;
1242
1243                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1244                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1245                 if (nr_pages == 0)
1246                         break;
1247
1248                 for (i = 0; i < nr_pages; i++) {
1249                         struct page *page = pvec.pages[i];
1250
1251                         if (page->index > end) {
1252                                 done = 1;
1253                                 break;
1254                         }
1255
1256                         done_index = page->index;
1257
1258                         lock_page(page);
1259
1260                         if (unlikely(page->mapping != mapping)) {
1261 continue_unlock:
1262                                 unlock_page(page);
1263                                 continue;
1264                         }
1265
1266                         if (!PageDirty(page)) {
1267                                 /* someone wrote it for us */
1268                                 goto continue_unlock;
1269                         }
1270
1271                         if (step == is_cold_data(page))
1272                                 goto continue_unlock;
1273
1274                         if (PageWriteback(page)) {
1275                                 if (wbc->sync_mode != WB_SYNC_NONE)
1276                                         f2fs_wait_on_page_writeback(page, DATA);
1277                                 else
1278                                         goto continue_unlock;
1279                         }
1280
1281                         BUG_ON(PageWriteback(page));
1282                         if (!clear_page_dirty_for_io(page))
1283                                 goto continue_unlock;
1284
1285                         ret = (*writepage)(page, wbc, data);
1286                         if (unlikely(ret)) {
1287                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1288                                         unlock_page(page);
1289                                         ret = 0;
1290                                 } else {
1291                                         done_index = page->index + 1;
1292                                         done = 1;
1293                                         break;
1294                                 }
1295                         }
1296
1297                         if (--wbc->nr_to_write <= 0 &&
1298                             wbc->sync_mode == WB_SYNC_NONE) {
1299                                 done = 1;
1300                                 break;
1301                         }
1302                 }
1303                 pagevec_release(&pvec);
1304                 cond_resched();
1305         }
1306
1307         if (step < 1) {
1308                 step++;
1309                 goto next;
1310         }
1311
1312         if (!cycled && !done) {
1313                 cycled = 1;
1314                 index = 0;
1315                 end = writeback_index - 1;
1316                 goto retry;
1317         }
1318         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1319                 mapping->writeback_index = done_index;
1320
1321         return ret;
1322 }
1323
1324 static int f2fs_write_data_pages(struct address_space *mapping,
1325                             struct writeback_control *wbc)
1326 {
1327         struct inode *inode = mapping->host;
1328         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1329         bool locked = false;
1330         int ret;
1331         long diff;
1332
1333         trace_f2fs_writepages(mapping->host, wbc, DATA);
1334
1335         /* deal with chardevs and other special file */
1336         if (!mapping->a_ops->writepage)
1337                 return 0;
1338
1339         /* skip writing if there is no dirty page in this inode */
1340         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1341                 return 0;
1342
1343         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1344                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1345                         available_free_memory(sbi, DIRTY_DENTS))
1346                 goto skip_write;
1347
1348         /* during POR, we don't need to trigger writepage at all. */
1349         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1350                 goto skip_write;
1351
1352         diff = nr_pages_to_write(sbi, DATA, wbc);
1353
1354         if (!S_ISDIR(inode->i_mode)) {
1355                 mutex_lock(&sbi->writepages);
1356                 locked = true;
1357         }
1358         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1359         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1360         if (locked)
1361                 mutex_unlock(&sbi->writepages);
1362
1363         remove_dirty_dir_inode(inode);
1364
1365         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1366         return ret;
1367
1368 skip_write:
1369         wbc->pages_skipped += get_dirty_pages(inode);
1370         return 0;
1371 }
1372
1373 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1374 {
1375         struct inode *inode = mapping->host;
1376
1377         if (to > inode->i_size) {
1378                 truncate_pagecache(inode, inode->i_size);
1379                 truncate_blocks(inode, inode->i_size, true);
1380         }
1381 }
1382
1383 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1384                 loff_t pos, unsigned len, unsigned flags,
1385                 struct page **pagep, void **fsdata)
1386 {
1387         struct inode *inode = mapping->host;
1388         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1389         struct page *page = NULL;
1390         struct page *ipage;
1391         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1392         struct dnode_of_data dn;
1393         int err = 0;
1394
1395         trace_f2fs_write_begin(inode, pos, len, flags);
1396
1397         f2fs_balance_fs(sbi);
1398
1399         /*
1400          * We should check this at this moment to avoid deadlock on inode page
1401          * and #0 page. The locking rule for inline_data conversion should be:
1402          * lock_page(page #0) -> lock_page(inode_page)
1403          */
1404         if (index != 0) {
1405                 err = f2fs_convert_inline_inode(inode);
1406                 if (err)
1407                         goto fail;
1408         }
1409 repeat:
1410         page = grab_cache_page_write_begin(mapping, index, flags);
1411         if (!page) {
1412                 err = -ENOMEM;
1413                 goto fail;
1414         }
1415
1416         *pagep = page;
1417
1418         f2fs_lock_op(sbi);
1419
1420         /* check inline_data */
1421         ipage = get_node_page(sbi, inode->i_ino);
1422         if (IS_ERR(ipage)) {
1423                 err = PTR_ERR(ipage);
1424                 goto unlock_fail;
1425         }
1426
1427         set_new_dnode(&dn, inode, ipage, ipage, 0);
1428
1429         if (f2fs_has_inline_data(inode)) {
1430                 if (pos + len <= MAX_INLINE_DATA) {
1431                         read_inline_data(page, ipage);
1432                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1433                         sync_inode_page(&dn);
1434                         goto put_next;
1435                 }
1436                 err = f2fs_convert_inline_page(&dn, page);
1437                 if (err)
1438                         goto put_fail;
1439         }
1440
1441         err = f2fs_get_block(&dn, index);
1442         if (err)
1443                 goto put_fail;
1444 put_next:
1445         f2fs_put_dnode(&dn);
1446         f2fs_unlock_op(sbi);
1447
1448         f2fs_wait_on_page_writeback(page, DATA);
1449
1450         if (len == PAGE_CACHE_SIZE)
1451                 goto out_update;
1452         if (PageUptodate(page))
1453                 goto out_clear;
1454
1455         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1456                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1457                 unsigned end = start + len;
1458
1459                 /* Reading beyond i_size is simple: memset to zero */
1460                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1461                 goto out_update;
1462         }
1463
1464         if (dn.data_blkaddr == NEW_ADDR) {
1465                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1466         } else {
1467                 struct f2fs_io_info fio = {
1468                         .sbi = sbi,
1469                         .type = DATA,
1470                         .rw = READ_SYNC,
1471                         .blk_addr = dn.data_blkaddr,
1472                         .page = page,
1473                         .encrypted_page = NULL,
1474                 };
1475                 err = f2fs_submit_page_bio(&fio);
1476                 if (err)
1477                         goto fail;
1478
1479                 lock_page(page);
1480                 if (unlikely(!PageUptodate(page))) {
1481                         err = -EIO;
1482                         goto fail;
1483                 }
1484                 if (unlikely(page->mapping != mapping)) {
1485                         f2fs_put_page(page, 1);
1486                         goto repeat;
1487                 }
1488
1489                 /* avoid symlink page */
1490                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1491                         err = f2fs_decrypt_one(inode, page);
1492                         if (err)
1493                                 goto fail;
1494                 }
1495         }
1496 out_update:
1497         SetPageUptodate(page);
1498 out_clear:
1499         clear_cold_data(page);
1500         return 0;
1501
1502 put_fail:
1503         f2fs_put_dnode(&dn);
1504 unlock_fail:
1505         f2fs_unlock_op(sbi);
1506 fail:
1507         f2fs_put_page(page, 1);
1508         f2fs_write_failed(mapping, pos + len);
1509         return err;
1510 }
1511
1512 static int f2fs_write_end(struct file *file,
1513                         struct address_space *mapping,
1514                         loff_t pos, unsigned len, unsigned copied,
1515                         struct page *page, void *fsdata)
1516 {
1517         struct inode *inode = page->mapping->host;
1518
1519         trace_f2fs_write_end(inode, pos, len, copied);
1520
1521         set_page_dirty(page);
1522
1523         if (pos + copied > i_size_read(inode)) {
1524                 i_size_write(inode, pos + copied);
1525                 mark_inode_dirty(inode);
1526                 update_inode_page(inode);
1527         }
1528
1529         f2fs_put_page(page, 1);
1530         return copied;
1531 }
1532
1533 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1534                            loff_t offset)
1535 {
1536         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1537
1538         if (offset & blocksize_mask)
1539                 return -EINVAL;
1540
1541         if (iov_iter_alignment(iter) & blocksize_mask)
1542                 return -EINVAL;
1543
1544         return 0;
1545 }
1546
1547 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1548                               loff_t offset)
1549 {
1550         struct file *file = iocb->ki_filp;
1551         struct address_space *mapping = file->f_mapping;
1552         struct inode *inode = mapping->host;
1553         size_t count = iov_iter_count(iter);
1554         int err;
1555
1556         /* we don't need to use inline_data strictly */
1557         if (f2fs_has_inline_data(inode)) {
1558                 err = f2fs_convert_inline_inode(inode);
1559                 if (err)
1560                         return err;
1561         }
1562
1563         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1564                 return 0;
1565
1566         err = check_direct_IO(inode, iter, offset);
1567         if (err)
1568                 return err;
1569
1570         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1571
1572         if (iov_iter_rw(iter) == WRITE) {
1573                 __allocate_data_blocks(inode, offset, count);
1574                 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1575                         err = -EIO;
1576                         goto out;
1577                 }
1578         }
1579
1580         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1581 out:
1582         if (err < 0 && iov_iter_rw(iter) == WRITE)
1583                 f2fs_write_failed(mapping, offset + count);
1584
1585         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1586
1587         return err;
1588 }
1589
1590 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1591                                                         unsigned int length)
1592 {
1593         struct inode *inode = page->mapping->host;
1594         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1595
1596         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1597                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1598                 return;
1599
1600         if (PageDirty(page)) {
1601                 if (inode->i_ino == F2FS_META_INO(sbi))
1602                         dec_page_count(sbi, F2FS_DIRTY_META);
1603                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1604                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1605                 else
1606                         inode_dec_dirty_pages(inode);
1607         }
1608
1609         /* This is atomic written page, keep Private */
1610         if (IS_ATOMIC_WRITTEN_PAGE(page))
1611                 return;
1612
1613         ClearPagePrivate(page);
1614 }
1615
1616 int f2fs_release_page(struct page *page, gfp_t wait)
1617 {
1618         /* If this is dirty page, keep PagePrivate */
1619         if (PageDirty(page))
1620                 return 0;
1621
1622         /* This is atomic written page, keep Private */
1623         if (IS_ATOMIC_WRITTEN_PAGE(page))
1624                 return 0;
1625
1626         ClearPagePrivate(page);
1627         return 1;
1628 }
1629
1630 static int f2fs_set_data_page_dirty(struct page *page)
1631 {
1632         struct address_space *mapping = page->mapping;
1633         struct inode *inode = mapping->host;
1634
1635         trace_f2fs_set_page_dirty(page, DATA);
1636
1637         SetPageUptodate(page);
1638
1639         if (f2fs_is_atomic_file(inode)) {
1640                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1641                         register_inmem_page(inode, page);
1642                         return 1;
1643                 }
1644                 /*
1645                  * Previously, this page has been registered, we just
1646                  * return here.
1647                  */
1648                 return 0;
1649         }
1650
1651         if (!PageDirty(page)) {
1652                 __set_page_dirty_nobuffers(page);
1653                 update_dirty_page(inode, page);
1654                 return 1;
1655         }
1656         return 0;
1657 }
1658
1659 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1660 {
1661         struct inode *inode = mapping->host;
1662
1663         /* we don't need to use inline_data strictly */
1664         if (f2fs_has_inline_data(inode)) {
1665                 int err = f2fs_convert_inline_inode(inode);
1666                 if (err)
1667                         return err;
1668         }
1669         return generic_block_bmap(mapping, block, get_data_block_bmap);
1670 }
1671
1672 const struct address_space_operations f2fs_dblock_aops = {
1673         .readpage       = f2fs_read_data_page,
1674         .readpages      = f2fs_read_data_pages,
1675         .writepage      = f2fs_write_data_page,
1676         .writepages     = f2fs_write_data_pages,
1677         .write_begin    = f2fs_write_begin,
1678         .write_end      = f2fs_write_end,
1679         .set_page_dirty = f2fs_set_data_page_dirty,
1680         .invalidatepage = f2fs_invalidate_page,
1681         .releasepage    = f2fs_release_page,
1682         .direct_IO      = f2fs_direct_IO,
1683         .bmap           = f2fs_bmap,
1684 };