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         f2fs_wait_on_page_writeback(node_page, NODE, false);
  41
  42         rn = F2FS_NODE(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         pgoff_t start_fofs, end_fofs;
  72         block_t start_blkaddr;
  73
  74         read_lock(&fi->ext.ext_lock);
  75         if (fi->ext.len == 0) {
  76                 read_unlock(&fi->ext.ext_lock);
  77                 return 0;
  78         }
  79
  80         stat_inc_total_hit(inode->i_sb);
  81
  82         start_fofs = fi->ext.fofs;
  83         end_fofs = fi->ext.fofs + fi->ext.len - 1;
  84         start_blkaddr = fi->ext.blk_addr;
  85
  86         if (pgofs >= start_fofs && pgofs <= end_fofs) {
  87                 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
  88                 size_t count;
  89
  90                 clear_buffer_new(bh_result);
  91                 map_bh(bh_result, inode->i_sb,
  92                                 start_blkaddr + pgofs - start_fofs);
  93                 count = end_fofs - pgofs + 1;
  94                 if (count < (UINT_MAX >> blkbits))
  95                         bh_result->b_size = (count << blkbits);
  96                 else
  97                         bh_result->b_size = UINT_MAX;
  98
  99                 stat_inc_read_hit(inode->i_sb);
 100                 read_unlock(&fi->ext.ext_lock);
 101                 return 1;
 102         }
 103         read_unlock(&fi->ext.ext_lock);
 104         return 0;
 105 }
 106
 107 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
 108 {
 109         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 110         pgoff_t fofs, start_fofs, end_fofs;
 111         block_t start_blkaddr, end_blkaddr;
 112
 113         f2fs_bug_on(blk_addr == NEW_ADDR);
 114         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 115                                                         dn->ofs_in_node;
 116
 117         /* Update the page address in the parent node */
 118         __set_data_blkaddr(dn, blk_addr);
 119
 120         write_lock(&fi->ext.ext_lock);
 121
 122         start_fofs = fi->ext.fofs;
 123         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 124         start_blkaddr = fi->ext.blk_addr;
 125         end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
 126
 127         /* Drop and initialize the matched extent */
 128         if (fi->ext.len == 1 && fofs == start_fofs)
 129                 fi->ext.len = 0;
 130
 131         /* Initial extent */
 132         if (fi->ext.len == 0) {
 133                 if (blk_addr != NULL_ADDR) {
 134                         fi->ext.fofs = fofs;
 135                         fi->ext.blk_addr = blk_addr;
 136                         fi->ext.len = 1;
 137                 }
 138                 goto end_update;
 139         }
 140
 141         /* Front merge */
 142         if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
 143                 fi->ext.fofs--;
 144                 fi->ext.blk_addr--;
 145                 fi->ext.len++;
 146                 goto end_update;
 147         }
 148
 149         /* Back merge */
 150         if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
 151                 fi->ext.len++;
 152                 goto end_update;
 153         }
 154
 155         /* Split the existing extent */
 156         if (fi->ext.len > 1 &&
 157                 fofs >= start_fofs && fofs <= end_fofs) {
 158                 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
 159                         fi->ext.len = fofs - start_fofs;
 160                 } else {
 161                         fi->ext.fofs = fofs + 1;
 162                         fi->ext.blk_addr = start_blkaddr +
 163                                         fofs - start_fofs + 1;
 164                         fi->ext.len -= fofs - start_fofs + 1;
 165                 }
 166                 goto end_update;
 167         }
 168         write_unlock(&fi->ext.ext_lock);
 169         return;
 170
 171 end_update:
 172         write_unlock(&fi->ext.ext_lock);
 173         sync_inode_page(dn);
 174 }
 175
 176 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
 177 {
 178         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 179         struct address_space *mapping = inode->i_mapping;
 180         struct dnode_of_data dn;
 181         struct page *page;
 182         int err;
 183
 184         page = find_get_page(mapping, index);
 185         if (page && PageUptodate(page))
 186                 return page;
 187         f2fs_put_page(page, 0);
 188
 189         set_new_dnode(&dn, inode, NULL, NULL, 0);
 190         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 191         if (err)
 192                 return ERR_PTR(err);
 193         f2fs_put_dnode(&dn);
 194
 195         if (dn.data_blkaddr == NULL_ADDR)
 196                 return ERR_PTR(-ENOENT);
 197
 198         /* By fallocate(), there is no cached page, but with NEW_ADDR */
 199         if (dn.data_blkaddr == NEW_ADDR)
 200                 return ERR_PTR(-EINVAL);
 201
 202         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
 203         if (!page)
 204                 return ERR_PTR(-ENOMEM);
 205
 206         if (PageUptodate(page)) {
 207                 unlock_page(page);
 208                 return page;
 209         }
 210
 211         err = f2fs_readpage(sbi, page, dn.data_blkaddr,
 212                                         sync ? READ_SYNC : READA);
 213         if (sync) {
 214                 wait_on_page_locked(page);
 215                 if (!PageUptodate(page)) {
 216                         f2fs_put_page(page, 0);
 217                         return ERR_PTR(-EIO);
 218                 }
 219         }
 220         return page;
 221 }
 222
 223 /*
 224  * If it tries to access a hole, return an error.
 225  * Because, the callers, functions in dir.c and GC, should be able to know
 226  * whether this page exists or not.
 227  */
 228 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 229 {
 230         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 231         struct address_space *mapping = inode->i_mapping;
 232         struct dnode_of_data dn;
 233         struct page *page;
 234         int err;
 235
 236 repeat:
 237         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
 238         if (!page)
 239                 return ERR_PTR(-ENOMEM);
 240
 241         set_new_dnode(&dn, inode, NULL, NULL, 0);
 242         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 243         if (err) {
 244                 f2fs_put_page(page, 1);
 245                 return ERR_PTR(err);
 246         }
 247         f2fs_put_dnode(&dn);
 248
 249         if (dn.data_blkaddr == NULL_ADDR) {
 250                 f2fs_put_page(page, 1);
 251                 return ERR_PTR(-ENOENT);
 252         }
 253
 254         if (PageUptodate(page))
 255                 return page;
 256
 257         /*
 258          * A new dentry page is allocated but not able to be written, since its
 259          * new inode page couldn't be allocated due to -ENOSPC.
 260          * In such the case, its blkaddr can be remained as NEW_ADDR.
 261          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 262          */
 263         if (dn.data_blkaddr == NEW_ADDR) {
 264                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 265                 SetPageUptodate(page);
 266                 return page;
 267         }
 268
 269         err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 270         if (err)
 271                 return ERR_PTR(err);
 272
 273         lock_page(page);
 274         if (!PageUptodate(page)) {
 275                 f2fs_put_page(page, 1);
 276                 return ERR_PTR(-EIO);
 277         }
 278         if (page->mapping != mapping) {
 279                 f2fs_put_page(page, 1);
 280                 goto repeat;
 281         }
 282         return page;
 283 }
 284
 285 /*
 286  * Caller ensures that this data page is never allocated.
 287  * A new zero-filled data page is allocated in the page cache.
 288  *
 289  * Also, caller should grab and release a mutex by calling mutex_lock_op() and
 290  * mutex_unlock_op().
 291  * Note that, npage is set only by make_empty_dir.
 292  */
 293 struct page *get_new_data_page(struct inode *inode,
 294                 struct page *npage, pgoff_t index, bool new_i_size)
 295 {
 296         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 297         struct address_space *mapping = inode->i_mapping;
 298         struct page *page;
 299         struct dnode_of_data dn;
 300         int err;
 301
 302         set_new_dnode(&dn, inode, npage, npage, 0);
 303         err = get_dnode_of_data(&dn, index, ALLOC_NODE);
 304         if (err)
 305                 return ERR_PTR(err);
 306
 307         if (dn.data_blkaddr == NULL_ADDR) {
 308                 if (reserve_new_block(&dn)) {
 309                         if (!npage)
 310                                 f2fs_put_dnode(&dn);
 311                         return ERR_PTR(-ENOSPC);
 312                 }
 313         }
 314         if (!npage)
 315                 f2fs_put_dnode(&dn);
 316 repeat:
 317         page = grab_cache_page(mapping, index);
 318         if (!page)
 319                 return ERR_PTR(-ENOMEM);
 320
 321         if (PageUptodate(page))
 322                 return page;
 323
 324         if (dn.data_blkaddr == NEW_ADDR) {
 325                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 326                 SetPageUptodate(page);
 327         } else {
 328                 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 329                 if (err)
 330                         return ERR_PTR(err);
 331                 lock_page(page);
 332                 if (!PageUptodate(page)) {
 333                         f2fs_put_page(page, 1);
 334                         return ERR_PTR(-EIO);
 335                 }
 336                 if (page->mapping != mapping) {
 337                         f2fs_put_page(page, 1);
 338                         goto repeat;
 339                 }
 340         }
 341
 342         if (new_i_size &&
 343                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 344                 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
 345                 /* Only the directory inode sets new_i_size */
 346                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 347                 mark_inode_dirty_sync(inode);
 348         }
 349         return page;
 350 }
 351
 352 static void read_end_io(struct bio *bio, int err)
 353 {
 354         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 355         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 356
 357         do {
 358                 struct page *page = bvec->bv_page;
 359
 360                 if (--bvec >= bio->bi_io_vec)
 361                         prefetchw(&bvec->bv_page->flags);
 362
 363                 if (uptodate) {
 364                         SetPageUptodate(page);
 365                 } else {
 366                         ClearPageUptodate(page);
 367                         SetPageError(page);
 368                 }
 369                 unlock_page(page);
 370         } while (bvec >= bio->bi_io_vec);
 371         bio_put(bio);
 372 }
 373
 374 /*
 375  * Fill the locked page with data located in the block address.
 376  * Return unlocked page.
 377  */
 378 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
 379                                         block_t blk_addr, int type)
 380 {
 381         struct block_device *bdev = sbi->sb->s_bdev;
 382         struct bio *bio;
 383
 384         trace_f2fs_readpage(page, blk_addr, type);
 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                 bio_put(bio);
 395                 f2fs_put_page(page, 1);
 396                 return -EFAULT;
 397         }
 398
 399         submit_bio(type, bio);
 400         return 0;
 401 }
 402
 403 /*
 404  * This function should be used by the data read flow only where it
 405  * does not check the "create" flag that indicates block allocation.
 406  * The reason for this special functionality is to exploit VFS readahead
 407  * mechanism.
 408  */
 409 static int get_data_block_ro(struct inode *inode, sector_t iblock,
 410                         struct buffer_head *bh_result, int create)
 411 {
 412         unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 413         unsigned maxblocks = bh_result->b_size >> blkbits;
 414         struct dnode_of_data dn;
 415         pgoff_t pgofs;
 416         int err;
 417
 418         /* Get the page offset from the block offset(iblock) */
 419         pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
 420
 421         if (check_extent_cache(inode, pgofs, bh_result)) {
 422                 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
 423                 return 0;
 424         }
 425
 426         /* When reading holes, we need its node page */
 427         set_new_dnode(&dn, inode, NULL, NULL, 0);
 428         err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
 429         if (err) {
 430                 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
 431                 return (err == -ENOENT) ? 0 : err;
 432         }
 433
 434         /* It does not support data allocation */
 435         f2fs_bug_on(create);
 436
 437         if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
 438                 int i;
 439                 unsigned int end_offset;
 440
 441                 end_offset = IS_INODE(dn.node_page) ?
 442                                 ADDRS_PER_INODE(F2FS_I(inode)) :
 443                                 ADDRS_PER_BLOCK;
 444
 445                 clear_buffer_new(bh_result);
 446
 447                 /* Give more consecutive addresses for the read ahead */
 448                 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
 449                         if (((datablock_addr(dn.node_page,
 450                                                         dn.ofs_in_node + i))
 451                                 != (dn.data_blkaddr + i)) || maxblocks == i)
 452                                 break;
 453                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 454                 bh_result->b_size = (i << blkbits);
 455         }
 456         f2fs_put_dnode(&dn);
 457         trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
 458         return 0;
 459 }
 460
 461 static int f2fs_read_data_page(struct file *file, struct page *page)
 462 {
 463         return mpage_readpage(page, get_data_block_ro);
 464 }
 465
 466 static int f2fs_read_data_pages(struct file *file,
 467                         struct address_space *mapping,
 468                         struct list_head *pages, unsigned nr_pages)
 469 {
 470         return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
 471 }
 472
 473 int do_write_data_page(struct page *page)
 474 {
 475         struct inode *inode = page->mapping->host;
 476         block_t old_blk_addr, new_blk_addr;
 477         struct dnode_of_data dn;
 478         int err = 0;
 479
 480         set_new_dnode(&dn, inode, NULL, NULL, 0);
 481         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 482         if (err)
 483                 return err;
 484
 485         old_blk_addr = dn.data_blkaddr;
 486
 487         /* This page is already truncated */
 488         if (old_blk_addr == NULL_ADDR)
 489                 goto out_writepage;
 490
 491         set_page_writeback(page);
 492
 493         /*
 494          * If current allocation needs SSR,
 495          * it had better in-place writes for updated data.
 496          */
 497         if (unlikely(old_blk_addr != NEW_ADDR &&
 498                         !is_cold_data(page) &&
 499                         need_inplace_update(inode))) {
 500                 rewrite_data_page(F2FS_SB(inode->i_sb), page,
 501                                                 old_blk_addr);
 502         } else {
 503                 write_data_page(inode, page, &dn,
 504                                 old_blk_addr, &new_blk_addr);
 505                 update_extent_cache(new_blk_addr, &dn);
 506         }
 507 out_writepage:
 508         f2fs_put_dnode(&dn);
 509         return err;
 510 }
 511
 512 static int f2fs_write_data_page(struct page *page,
 513                                         struct writeback_control *wbc)
 514 {
 515         struct inode *inode = page->mapping->host;
 516         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 517         loff_t i_size = i_size_read(inode);
 518         const pgoff_t end_index = ((unsigned long long) i_size)
 519                                                         >> PAGE_CACHE_SHIFT;
 520         unsigned offset;
 521         bool need_balance_fs = false;
 522         int err = 0;
 523
 524         if (page->index < end_index)
 525                 goto write;
 526
 527         /*
 528          * If the offset is out-of-range of file size,
 529          * this page does not have to be written to disk.
 530          */
 531         offset = i_size & (PAGE_CACHE_SIZE - 1);
 532         if ((page->index >= end_index + 1) || !offset) {
 533                 if (S_ISDIR(inode->i_mode)) {
 534                         dec_page_count(sbi, F2FS_DIRTY_DENTS);
 535                         inode_dec_dirty_dents(inode);
 536                 }
 537                 goto out;
 538         }
 539
 540         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 541 write:
 542         if (sbi->por_doing) {
 543                 err = AOP_WRITEPAGE_ACTIVATE;
 544                 goto redirty_out;
 545         }
 546
 547         /* Dentry blocks are controlled by checkpoint */
 548         if (S_ISDIR(inode->i_mode)) {
 549                 dec_page_count(sbi, F2FS_DIRTY_DENTS);
 550                 inode_dec_dirty_dents(inode);
 551                 err = do_write_data_page(page);
 552         } else {
 553                 f2fs_lock_op(sbi);
 554                 err = do_write_data_page(page);
 555                 f2fs_unlock_op(sbi);
 556                 need_balance_fs = true;
 557         }
 558         if (err == -ENOENT)
 559                 goto out;
 560         else if (err)
 561                 goto redirty_out;
 562
 563         if (wbc->for_reclaim)
 564                 f2fs_submit_bio(sbi, DATA, true);
 565
 566         clear_cold_data(page);
 567 out:
 568         unlock_page(page);
 569         if (need_balance_fs)
 570                 f2fs_balance_fs(sbi);
 571         return 0;
 572
 573 redirty_out:
 574         wbc->pages_skipped++;
 575         set_page_dirty(page);
 576         return err;
 577 }
 578
 579 #define MAX_DESIRED_PAGES_WP    4096
 580
 581 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
 582                         void *data)
 583 {
 584         struct address_space *mapping = data;
 585         int ret = mapping->a_ops->writepage(page, wbc);
 586         mapping_set_error(mapping, ret);
 587         return ret;
 588 }
 589
 590 static int f2fs_write_data_pages(struct address_space *mapping,
 591                             struct writeback_control *wbc)
 592 {
 593         struct inode *inode = mapping->host;
 594         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 595         bool locked = false;
 596         int ret;
 597         long excess_nrtw = 0, desired_nrtw;
 598
 599         /* deal with chardevs and other special file */
 600         if (!mapping->a_ops->writepage)
 601                 return 0;
 602
 603         if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
 604                 desired_nrtw = MAX_DESIRED_PAGES_WP;
 605                 excess_nrtw = desired_nrtw - wbc->nr_to_write;
 606                 wbc->nr_to_write = desired_nrtw;
 607         }
 608
 609         if (!S_ISDIR(inode->i_mode)) {
 610                 mutex_lock(&sbi->writepages);
 611                 locked = true;
 612         }
 613         ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
 614         if (locked)
 615                 mutex_unlock(&sbi->writepages);
 616         f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
 617
 618         remove_dirty_dir_inode(inode);
 619
 620         wbc->nr_to_write -= excess_nrtw;
 621         return ret;
 622 }
 623
 624 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 625                 loff_t pos, unsigned len, unsigned flags,
 626                 struct page **pagep, void **fsdata)
 627 {
 628         struct inode *inode = mapping->host;
 629         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 630         struct page *page;
 631         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
 632         struct dnode_of_data dn;
 633         int err = 0;
 634
 635         f2fs_balance_fs(sbi);
 636 repeat:
 637         page = grab_cache_page_write_begin(mapping, index, flags);
 638         if (!page)
 639                 return -ENOMEM;
 640         *pagep = page;
 641
 642         f2fs_lock_op(sbi);
 643
 644         set_new_dnode(&dn, inode, NULL, NULL, 0);
 645         err = get_dnode_of_data(&dn, index, ALLOC_NODE);
 646         if (err)
 647                 goto err;
 648
 649         if (dn.data_blkaddr == NULL_ADDR)
 650                 err = reserve_new_block(&dn);
 651
 652         f2fs_put_dnode(&dn);
 653         if (err)
 654                 goto err;
 655
 656         f2fs_unlock_op(sbi);
 657
 658         if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
 659                 return 0;
 660
 661         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
 662                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 663                 unsigned end = start + len;
 664
 665                 /* Reading beyond i_size is simple: memset to zero */
 666                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
 667                 goto out;
 668         }
 669
 670         if (dn.data_blkaddr == NEW_ADDR) {
 671                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 672         } else {
 673                 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 674                 if (err)
 675                         return err;
 676                 lock_page(page);
 677                 if (!PageUptodate(page)) {
 678                         f2fs_put_page(page, 1);
 679                         return -EIO;
 680                 }
 681                 if (page->mapping != mapping) {
 682                         f2fs_put_page(page, 1);
 683                         goto repeat;
 684                 }
 685         }
 686 out:
 687         SetPageUptodate(page);
 688         clear_cold_data(page);
 689         return 0;
 690
 691 err:
 692         f2fs_unlock_op(sbi);
 693         f2fs_put_page(page, 1);
 694         return err;
 695 }
 696
 697 static int f2fs_write_end(struct file *file,
 698                         struct address_space *mapping,
 699                         loff_t pos, unsigned len, unsigned copied,
 700                         struct page *page, void *fsdata)
 701 {
 702         struct inode *inode = page->mapping->host;
 703
 704         SetPageUptodate(page);
 705         set_page_dirty(page);
 706
 707         if (pos + copied > i_size_read(inode)) {
 708                 i_size_write(inode, pos + copied);
 709                 mark_inode_dirty(inode);
 710                 update_inode_page(inode);
 711         }
 712
 713         f2fs_put_page(page, 1);
 714         return copied;
 715 }
 716
 717 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
 718                 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 719 {
 720         struct file *file = iocb->ki_filp;
 721         struct inode *inode = file->f_mapping->host;
 722
 723         if (rw == WRITE)
 724                 return 0;
 725
 726         /* Needs synchronization with the cleaner */
 727         return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 728                                                   get_data_block_ro);
 729 }
 730
 731 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
 732                                       unsigned int length)
 733 {
 734         struct inode *inode = page->mapping->host;
 735         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 736         if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
 737                 dec_page_count(sbi, F2FS_DIRTY_DENTS);
 738                 inode_dec_dirty_dents(inode);
 739         }
 740         ClearPagePrivate(page);
 741 }
 742
 743 static int f2fs_release_data_page(struct page *page, gfp_t wait)
 744 {
 745         ClearPagePrivate(page);
 746         return 1;
 747 }
 748
 749 static int f2fs_set_data_page_dirty(struct page *page)
 750 {
 751         struct address_space *mapping = page->mapping;
 752         struct inode *inode = mapping->host;
 753
 754         trace_f2fs_set_page_dirty(page, DATA);
 755
 756         SetPageUptodate(page);
 757         if (!PageDirty(page)) {
 758                 __set_page_dirty_nobuffers(page);
 759                 set_dirty_dir_page(inode, page);
 760                 return 1;
 761         }
 762         return 0;
 763 }
 764
 765 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
 766 {
 767         return generic_block_bmap(mapping, block, get_data_block_ro);
 768 }
 769
 770 const struct address_space_operations f2fs_dblock_aops = {
 771         .readpage       = f2fs_read_data_page,
 772         .readpages      = f2fs_read_data_pages,
 773         .writepage      = f2fs_write_data_page,
 774         .writepages     = f2fs_write_data_pages,
 775         .write_begin    = f2fs_write_begin,
 776         .write_end      = f2fs_write_end,
 777         .set_page_dirty = f2fs_set_data_page_dirty,
 778         .invalidatepage = f2fs_invalidate_data_page,
 779         .releasepage    = f2fs_release_data_page,
 780         .direct_IO      = f2fs_direct_IO,
 781         .bmap           = f2fs_bmap,
 782 };