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/aio.h>
  16 #include <linux/writeback.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include <trace/events/f2fs.h>
  26
  27 /*
  28  * Lock ordering for the change of data block address:
  29  * ->data_page
  30  *  ->node_page
  31  *    update block addresses in the node page
  32  */
  33 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
  34 {
  35         struct f2fs_node *rn;
  36         __le32 *addr_array;
  37         struct page *node_page = dn->node_page;
  38         unsigned int ofs_in_node = dn->ofs_in_node;
  39
  40         wait_on_page_writeback(node_page);
  41
  42         rn = (struct f2fs_node *)page_address(node_page);
  43
  44         /* Get physical address of data block */
  45         addr_array = blkaddr_in_node(rn);
  46         addr_array[ofs_in_node] = cpu_to_le32(new_addr);
  47         set_page_dirty(node_page);
  48 }
  49
  50 int reserve_new_block(struct dnode_of_data *dn)
  51 {
  52         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
  53
  54         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
  55                 return -EPERM;
  56         if (!inc_valid_block_count(sbi, dn->inode, 1))
  57                 return -ENOSPC;
  58
  59         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
  60
  61         __set_data_blkaddr(dn, NEW_ADDR);
  62         dn->data_blkaddr = NEW_ADDR;
  63         sync_inode_page(dn);
  64         return 0;
  65 }
  66
  67 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
  68                                         struct buffer_head *bh_result)
  69 {
  70         struct f2fs_inode_info *fi = F2FS_I(inode);
  71 #ifdef CONFIG_F2FS_STAT_FS
  72         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
  73 #endif
  74         pgoff_t start_fofs, end_fofs;
  75         block_t start_blkaddr;
  76
  77         read_lock(&fi->ext.ext_lock);
  78         if (fi->ext.len == 0) {
  79                 read_unlock(&fi->ext.ext_lock);
  80                 return 0;
  81         }
  82
  83 #ifdef CONFIG_F2FS_STAT_FS
  84         sbi->total_hit_ext++;
  85 #endif
  86         start_fofs = fi->ext.fofs;
  87         end_fofs = fi->ext.fofs + fi->ext.len - 1;
  88         start_blkaddr = fi->ext.blk_addr;
  89
  90         if (pgofs >= start_fofs && pgofs <= end_fofs) {
  91                 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
  92                 size_t count;
  93
  94                 clear_buffer_new(bh_result);
  95                 map_bh(bh_result, inode->i_sb,
  96                                 start_blkaddr + pgofs - start_fofs);
  97                 count = end_fofs - pgofs + 1;
  98                 if (count < (UINT_MAX >> blkbits))
  99                         bh_result->b_size = (count << blkbits);
 100                 else
 101                         bh_result->b_size = UINT_MAX;
 102
 103 #ifdef CONFIG_F2FS_STAT_FS
 104                 sbi->read_hit_ext++;
 105 #endif
 106                 read_unlock(&fi->ext.ext_lock);
 107                 return 1;
 108         }
 109         read_unlock(&fi->ext.ext_lock);
 110         return 0;
 111 }
 112
 113 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
 114 {
 115         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 116         pgoff_t fofs, start_fofs, end_fofs;
 117         block_t start_blkaddr, end_blkaddr;
 118
 119         BUG_ON(blk_addr == NEW_ADDR);
 120         fofs = start_bidx_of_node(ofs_of_node(dn->node_page)) + dn->ofs_in_node;
 121
 122         /* Update the page address in the parent node */
 123         __set_data_blkaddr(dn, blk_addr);
 124
 125         write_lock(&fi->ext.ext_lock);
 126
 127         start_fofs = fi->ext.fofs;
 128         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 129         start_blkaddr = fi->ext.blk_addr;
 130         end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
 131
 132         /* Drop and initialize the matched extent */
 133         if (fi->ext.len == 1 && fofs == start_fofs)
 134                 fi->ext.len = 0;
 135
 136         /* Initial extent */
 137         if (fi->ext.len == 0) {
 138                 if (blk_addr != NULL_ADDR) {
 139                         fi->ext.fofs = fofs;
 140                         fi->ext.blk_addr = blk_addr;
 141                         fi->ext.len = 1;
 142                 }
 143                 goto end_update;
 144         }
 145
 146         /* Front merge */
 147         if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
 148                 fi->ext.fofs--;
 149                 fi->ext.blk_addr--;
 150                 fi->ext.len++;
 151                 goto end_update;
 152         }
 153
 154         /* Back merge */
 155         if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
 156                 fi->ext.len++;
 157                 goto end_update;
 158         }
 159
 160         /* Split the existing extent */
 161         if (fi->ext.len > 1 &&
 162                 fofs >= start_fofs && fofs <= end_fofs) {
 163                 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
 164                         fi->ext.len = fofs - start_fofs;
 165                 } else {
 166                         fi->ext.fofs = fofs + 1;
 167                         fi->ext.blk_addr = start_blkaddr +
 168                                         fofs - start_fofs + 1;
 169                         fi->ext.len -= fofs - start_fofs + 1;
 170                 }
 171                 goto end_update;
 172         }
 173         write_unlock(&fi->ext.ext_lock);
 174         return;
 175
 176 end_update:
 177         write_unlock(&fi->ext.ext_lock);
 178         sync_inode_page(dn);
 179         return;
 180 }
 181
 182 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
 183 {
 184         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 185         struct address_space *mapping = inode->i_mapping;
 186         struct dnode_of_data dn;
 187         struct page *page;
 188         int err;
 189
 190         page = find_get_page(mapping, index);
 191         if (page && PageUptodate(page))
 192                 return page;
 193         f2fs_put_page(page, 0);
 194
 195         set_new_dnode(&dn, inode, NULL, NULL, 0);
 196         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 197         if (err)
 198                 return ERR_PTR(err);
 199         f2fs_put_dnode(&dn);
 200
 201         if (dn.data_blkaddr == NULL_ADDR)
 202                 return ERR_PTR(-ENOENT);
 203
 204         /* By fallocate(), there is no cached page, but with NEW_ADDR */
 205         if (dn.data_blkaddr == NEW_ADDR)
 206                 return ERR_PTR(-EINVAL);
 207
 208         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
 209         if (!page)
 210                 return ERR_PTR(-ENOMEM);
 211
 212         if (PageUptodate(page)) {
 213                 unlock_page(page);
 214                 return page;
 215         }
 216
 217         err = f2fs_readpage(sbi, page, dn.data_blkaddr,
 218                                         sync ? READ_SYNC : READA);
 219         if (sync) {
 220                 wait_on_page_locked(page);
 221                 if (!PageUptodate(page)) {
 222                         f2fs_put_page(page, 0);
 223                         return ERR_PTR(-EIO);
 224                 }
 225         }
 226         return page;
 227 }
 228
 229 /*
 230  * If it tries to access a hole, return an error.
 231  * Because, the callers, functions in dir.c and GC, should be able to know
 232  * whether this page exists or not.
 233  */
 234 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 235 {
 236         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 237         struct address_space *mapping = inode->i_mapping;
 238         struct dnode_of_data dn;
 239         struct page *page;
 240         int err;
 241
 242 repeat:
 243         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
 244         if (!page)
 245                 return ERR_PTR(-ENOMEM);
 246
 247         set_new_dnode(&dn, inode, NULL, NULL, 0);
 248         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 249         if (err) {
 250                 f2fs_put_page(page, 1);
 251                 return ERR_PTR(err);
 252         }
 253         f2fs_put_dnode(&dn);
 254
 255         if (dn.data_blkaddr == NULL_ADDR) {
 256                 f2fs_put_page(page, 1);
 257                 return ERR_PTR(-ENOENT);
 258         }
 259
 260         if (PageUptodate(page))
 261                 return page;
 262
 263         BUG_ON(dn.data_blkaddr == NEW_ADDR);
 264         BUG_ON(dn.data_blkaddr == NULL_ADDR);
 265
 266         err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 267         if (err)
 268                 return ERR_PTR(err);
 269
 270         lock_page(page);
 271         if (!PageUptodate(page)) {
 272                 f2fs_put_page(page, 1);
 273                 return ERR_PTR(-EIO);
 274         }
 275         if (page->mapping != mapping) {
 276                 f2fs_put_page(page, 1);
 277                 goto repeat;
 278         }
 279         return page;
 280 }
 281
 282 /*
 283  * Caller ensures that this data page is never allocated.
 284  * A new zero-filled data page is allocated in the page cache.
 285  *
 286  * Also, caller should grab and release a mutex by calling mutex_lock_op() and
 287  * mutex_unlock_op().
 288  * Note that, npage is set only by make_empty_dir.
 289  */
 290 struct page *get_new_data_page(struct inode *inode,
 291                 struct page *npage, pgoff_t index, bool new_i_size)
 292 {
 293         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 294         struct address_space *mapping = inode->i_mapping;
 295         struct page *page;
 296         struct dnode_of_data dn;
 297         int err;
 298
 299         set_new_dnode(&dn, inode, npage, npage, 0);
 300         err = get_dnode_of_data(&dn, index, ALLOC_NODE);
 301         if (err)
 302                 return ERR_PTR(err);
 303
 304         if (dn.data_blkaddr == NULL_ADDR) {
 305                 if (reserve_new_block(&dn)) {
 306                         if (!npage)
 307                                 f2fs_put_dnode(&dn);
 308                         return ERR_PTR(-ENOSPC);
 309                 }
 310         }
 311         if (!npage)
 312                 f2fs_put_dnode(&dn);
 313 repeat:
 314         page = grab_cache_page(mapping, index);
 315         if (!page)
 316                 return ERR_PTR(-ENOMEM);
 317
 318         if (PageUptodate(page))
 319                 return page;
 320
 321         if (dn.data_blkaddr == NEW_ADDR) {
 322                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 323                 SetPageUptodate(page);
 324         } else {
 325                 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 326                 if (err)
 327                         return ERR_PTR(err);
 328                 lock_page(page);
 329                 if (!PageUptodate(page)) {
 330                         f2fs_put_page(page, 1);
 331                         return ERR_PTR(-EIO);
 332                 }
 333                 if (page->mapping != mapping) {
 334                         f2fs_put_page(page, 1);
 335                         goto repeat;
 336                 }
 337         }
 338
 339         if (new_i_size &&
 340                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 341                 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
 342                 /* Only the directory inode sets new_i_size */
 343                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 344                 mark_inode_dirty_sync(inode);
 345         }
 346         return page;
 347 }
 348
 349 static void read_end_io(struct bio *bio, int err)
 350 {
 351         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 352         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 353
 354         do {
 355                 struct page *page = bvec->bv_page;
 356
 357                 if (--bvec >= bio->bi_io_vec)
 358                         prefetchw(&bvec->bv_page->flags);
 359
 360                 if (uptodate) {
 361                         SetPageUptodate(page);
 362                 } else {
 363                         ClearPageUptodate(page);
 364                         SetPageError(page);
 365                 }
 366                 unlock_page(page);
 367         } while (bvec >= bio->bi_io_vec);
 368         kfree(bio->bi_private);
 369         bio_put(bio);
 370 }
 371
 372 /*
 373  * Fill the locked page with data located in the block address.
 374  * Return unlocked page.
 375  */
 376 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
 377                                         block_t blk_addr, int type)
 378 {
 379         struct block_device *bdev = sbi->sb->s_bdev;
 380         struct bio *bio;
 381
 382         trace_f2fs_readpage(page, blk_addr, type);
 383
 384         down_read(&sbi->bio_sem);
 385
 386         /* Allocate a new bio */
 387         bio = f2fs_bio_alloc(bdev, 1);
 388
 389         /* Initialize the bio */
 390         bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 391         bio->bi_end_io = read_end_io;
 392
 393         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 394                 kfree(bio->bi_private);
 395                 bio_put(bio);
 396                 up_read(&sbi->bio_sem);
 397                 f2fs_put_page(page, 1);
 398                 return -EFAULT;
 399         }
 400
 401         submit_bio(type, bio);
 402         up_read(&sbi->bio_sem);
 403         return 0;
 404 }
 405
 406 /*
 407  * This function should be used by the data read flow only where it
 408  * does not check the "create" flag that indicates block allocation.
 409  * The reason for this special functionality is to exploit VFS readahead
 410  * mechanism.
 411  */
 412 static int get_data_block_ro(struct inode *inode, sector_t iblock,
 413                         struct buffer_head *bh_result, int create)
 414 {
 415         unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 416         unsigned maxblocks = bh_result->b_size >> blkbits;
 417         struct dnode_of_data dn;
 418         pgoff_t pgofs;
 419         int err;
 420
 421         /* Get the page offset from the block offset(iblock) */
 422         pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
 423
 424         if (check_extent_cache(inode, pgofs, bh_result)) {
 425                 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
 426                 return 0;
 427         }
 428
 429         /* When reading holes, we need its node page */
 430         set_new_dnode(&dn, inode, NULL, NULL, 0);
 431         err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
 432         if (err) {
 433                 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
 434                 return (err == -ENOENT) ? 0 : err;
 435         }
 436
 437         /* It does not support data allocation */
 438         BUG_ON(create);
 439
 440         if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
 441                 int i;
 442                 unsigned int end_offset;
 443
 444                 end_offset = IS_INODE(dn.node_page) ?
 445                                 ADDRS_PER_INODE :
 446                                 ADDRS_PER_BLOCK;
 447
 448                 clear_buffer_new(bh_result);
 449
 450                 /* Give more consecutive addresses for the read ahead */
 451                 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
 452                         if (((datablock_addr(dn.node_page,
 453                                                         dn.ofs_in_node + i))
 454                                 != (dn.data_blkaddr + i)) || maxblocks == i)
 455                                 break;
 456                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 457                 bh_result->b_size = (i << blkbits);
 458         }
 459         f2fs_put_dnode(&dn);
 460         trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
 461         return 0;
 462 }
 463
 464 static int f2fs_read_data_page(struct file *file, struct page *page)
 465 {
 466         return mpage_readpage(page, get_data_block_ro);
 467 }
 468
 469 static int f2fs_read_data_pages(struct file *file,
 470                         struct address_space *mapping,
 471                         struct list_head *pages, unsigned nr_pages)
 472 {
 473         return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
 474 }
 475
 476 int do_write_data_page(struct page *page)
 477 {
 478         struct inode *inode = page->mapping->host;
 479         block_t old_blk_addr, new_blk_addr;
 480         struct dnode_of_data dn;
 481         int err = 0;
 482
 483         set_new_dnode(&dn, inode, NULL, NULL, 0);
 484         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 485         if (err)
 486                 return err;
 487
 488         old_blk_addr = dn.data_blkaddr;
 489
 490         /* This page is already truncated */
 491         if (old_blk_addr == NULL_ADDR)
 492                 goto out_writepage;
 493
 494         set_page_writeback(page);
 495
 496         /*
 497          * If current allocation needs SSR,
 498          * it had better in-place writes for updated data.
 499          */
 500         if (unlikely(old_blk_addr != NEW_ADDR &&
 501                         !is_cold_data(page) &&
 502                         need_inplace_update(inode))) {
 503                 rewrite_data_page(F2FS_SB(inode->i_sb), page,
 504                                                 old_blk_addr);
 505         } else {
 506                 write_data_page(inode, page, &dn,
 507                                 old_blk_addr, &new_blk_addr);
 508                 update_extent_cache(new_blk_addr, &dn);
 509         }
 510 out_writepage:
 511         f2fs_put_dnode(&dn);
 512         return err;
 513 }
 514
 515 static int f2fs_write_data_page(struct page *page,
 516                                         struct writeback_control *wbc)
 517 {
 518         struct inode *inode = page->mapping->host;
 519         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 520         loff_t i_size = i_size_read(inode);
 521         const pgoff_t end_index = ((unsigned long long) i_size)
 522                                                         >> PAGE_CACHE_SHIFT;
 523         unsigned offset;
 524         bool need_balance_fs = false;
 525         int err = 0;
 526
 527         if (page->index < end_index)
 528                 goto write;
 529
 530         /*
 531          * If the offset is out-of-range of file size,
 532          * this page does not have to be written to disk.
 533          */
 534         offset = i_size & (PAGE_CACHE_SIZE - 1);
 535         if ((page->index >= end_index + 1) || !offset) {
 536                 if (S_ISDIR(inode->i_mode)) {
 537                         dec_page_count(sbi, F2FS_DIRTY_DENTS);
 538                         inode_dec_dirty_dents(inode);
 539                 }
 540                 goto out;
 541         }
 542
 543         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 544 write:
 545         if (sbi->por_doing) {
 546                 err = AOP_WRITEPAGE_ACTIVATE;
 547                 goto redirty_out;
 548         }
 549
 550         /* Dentry blocks are controlled by checkpoint */
 551         if (S_ISDIR(inode->i_mode)) {
 552                 dec_page_count(sbi, F2FS_DIRTY_DENTS);
 553                 inode_dec_dirty_dents(inode);
 554                 err = do_write_data_page(page);
 555         } else {
 556                 int ilock = mutex_lock_op(sbi);
 557                 err = do_write_data_page(page);
 558                 mutex_unlock_op(sbi, ilock);
 559                 need_balance_fs = true;
 560         }
 561         if (err == -ENOENT)
 562                 goto out;
 563         else if (err)
 564                 goto redirty_out;
 565
 566         if (wbc->for_reclaim)
 567                 f2fs_submit_bio(sbi, DATA, true);
 568
 569         clear_cold_data(page);
 570 out:
 571         unlock_page(page);
 572         if (need_balance_fs)
 573                 f2fs_balance_fs(sbi);
 574         return 0;
 575
 576 redirty_out:
 577         wbc->pages_skipped++;
 578         set_page_dirty(page);
 579         return err;
 580 }
 581
 582 #define MAX_DESIRED_PAGES_WP    4096
 583
 584 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
 585                         void *data)
 586 {
 587         struct address_space *mapping = data;
 588         int ret = mapping->a_ops->writepage(page, wbc);
 589         mapping_set_error(mapping, ret);
 590         return ret;
 591 }
 592
 593 static int f2fs_write_data_pages(struct address_space *mapping,
 594                             struct writeback_control *wbc)
 595 {
 596         struct inode *inode = mapping->host;
 597         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 598         bool locked = false;
 599         int ret;
 600         long excess_nrtw = 0, desired_nrtw;
 601
 602         /* deal with chardevs and other special file */
 603         if (!mapping->a_ops->writepage)
 604                 return 0;
 605
 606         if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
 607                 desired_nrtw = MAX_DESIRED_PAGES_WP;
 608                 excess_nrtw = desired_nrtw - wbc->nr_to_write;
 609                 wbc->nr_to_write = desired_nrtw;
 610         }
 611
 612         if (!S_ISDIR(inode->i_mode)) {
 613                 mutex_lock(&sbi->writepages);
 614                 locked = true;
 615         }
 616         ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
 617         if (locked)
 618                 mutex_unlock(&sbi->writepages);
 619         f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
 620
 621         remove_dirty_dir_inode(inode);
 622
 623         wbc->nr_to_write -= excess_nrtw;
 624         return ret;
 625 }
 626
 627 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 628                 loff_t pos, unsigned len, unsigned flags,
 629                 struct page **pagep, void **fsdata)
 630 {
 631         struct inode *inode = mapping->host;
 632         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 633         struct page *page;
 634         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
 635         struct dnode_of_data dn;
 636         int err = 0;
 637         int ilock;
 638
 639         /* for nobh_write_end */
 640         *fsdata = NULL;
 641
 642         f2fs_balance_fs(sbi);
 643 repeat:
 644         page = grab_cache_page_write_begin(mapping, index, flags);
 645         if (!page)
 646                 return -ENOMEM;
 647         *pagep = page;
 648
 649         ilock = mutex_lock_op(sbi);
 650
 651         set_new_dnode(&dn, inode, NULL, NULL, 0);
 652         err = get_dnode_of_data(&dn, index, ALLOC_NODE);
 653         if (err)
 654                 goto err;
 655
 656         if (dn.data_blkaddr == NULL_ADDR)
 657                 err = reserve_new_block(&dn);
 658
 659         f2fs_put_dnode(&dn);
 660         if (err)
 661                 goto err;
 662
 663         mutex_unlock_op(sbi, ilock);
 664
 665         if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
 666                 return 0;
 667
 668         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
 669                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 670                 unsigned end = start + len;
 671
 672                 /* Reading beyond i_size is simple: memset to zero */
 673                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
 674                 goto out;
 675         }
 676
 677         if (dn.data_blkaddr == NEW_ADDR) {
 678                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 679         } else {
 680                 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 681                 if (err)
 682                         return err;
 683                 lock_page(page);
 684                 if (!PageUptodate(page)) {
 685                         f2fs_put_page(page, 1);
 686                         return -EIO;
 687                 }
 688                 if (page->mapping != mapping) {
 689                         f2fs_put_page(page, 1);
 690                         goto repeat;
 691                 }
 692         }
 693 out:
 694         SetPageUptodate(page);
 695         clear_cold_data(page);
 696         return 0;
 697
 698 err:
 699         mutex_unlock_op(sbi, ilock);
 700         f2fs_put_page(page, 1);
 701         return err;
 702 }
 703
 704 static int f2fs_write_end(struct file *file,
 705                         struct address_space *mapping,
 706                         loff_t pos, unsigned len, unsigned copied,
 707                         struct page *page, void *fsdata)
 708 {
 709         struct inode *inode = page->mapping->host;
 710
 711         SetPageUptodate(page);
 712         set_page_dirty(page);
 713
 714         if (pos + copied > i_size_read(inode)) {
 715                 i_size_write(inode, pos + copied);
 716                 mark_inode_dirty(inode);
 717                 update_inode_page(inode);
 718         }
 719
 720         unlock_page(page);
 721         page_cache_release(page);
 722         return copied;
 723 }
 724
 725 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
 726                 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 727 {
 728         struct file *file = iocb->ki_filp;
 729         struct inode *inode = file->f_mapping->host;
 730
 731         if (rw == WRITE)
 732                 return 0;
 733
 734         /* Needs synchronization with the cleaner */
 735         return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 736                                                   get_data_block_ro);
 737 }
 738
 739 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
 740                                       unsigned int length)
 741 {
 742         struct inode *inode = page->mapping->host;
 743         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 744         if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
 745                 dec_page_count(sbi, F2FS_DIRTY_DENTS);
 746                 inode_dec_dirty_dents(inode);
 747         }
 748         ClearPagePrivate(page);
 749 }
 750
 751 static int f2fs_release_data_page(struct page *page, gfp_t wait)
 752 {
 753         ClearPagePrivate(page);
 754         return 1;
 755 }
 756
 757 static int f2fs_set_data_page_dirty(struct page *page)
 758 {
 759         struct address_space *mapping = page->mapping;
 760         struct inode *inode = mapping->host;
 761
 762         SetPageUptodate(page);
 763         if (!PageDirty(page)) {
 764                 __set_page_dirty_nobuffers(page);
 765                 set_dirty_dir_page(inode, page);
 766                 return 1;
 767         }
 768         return 0;
 769 }
 770
 771 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
 772 {
 773         return generic_block_bmap(mapping, block, get_data_block_ro);
 774 }
 775
 776 const struct address_space_operations f2fs_dblock_aops = {
 777         .readpage       = f2fs_read_data_page,
 778         .readpages      = f2fs_read_data_pages,
 779         .writepage      = f2fs_write_data_page,
 780         .writepages     = f2fs_write_data_pages,
 781         .write_begin    = f2fs_write_begin,
 782         .write_end      = f2fs_write_end,
 783         .set_page_dirty = f2fs_set_data_page_dirty,
 784         .invalidatepage = f2fs_invalidate_data_page,
 785         .releasepage    = f2fs_release_data_page,
 786         .direct_IO      = f2fs_direct_IO,
 787         .bmap           = f2fs_bmap,
 788 };