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         if (is_inode_flag_set(fi, FI_NO_EXTENT))
  75                 return 0;
  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         stat_inc_total_hit(inode->i_sb);
  84
  85         start_fofs = fi->ext.fofs;
  86         end_fofs = fi->ext.fofs + fi->ext.len - 1;
  87         start_blkaddr = fi->ext.blk_addr;
  88
  89         if (pgofs >= start_fofs && pgofs <= end_fofs) {
  90                 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
  91                 size_t count;
  92
  93                 clear_buffer_new(bh_result);
  94                 map_bh(bh_result, inode->i_sb,
  95                                 start_blkaddr + pgofs - start_fofs);
  96                 count = end_fofs - pgofs + 1;
  97                 if (count < (UINT_MAX >> blkbits))
  98                         bh_result->b_size = (count << blkbits);
  99                 else
 100                         bh_result->b_size = UINT_MAX;
 101
 102                 stat_inc_read_hit(inode->i_sb);
 103                 read_unlock(&fi->ext.ext_lock);
 104                 return 1;
 105         }
 106         read_unlock(&fi->ext.ext_lock);
 107         return 0;
 108 }
 109
 110 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
 111 {
 112         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 113         pgoff_t fofs, start_fofs, end_fofs;
 114         block_t start_blkaddr, end_blkaddr;
 115         int need_update = true;
 116
 117         f2fs_bug_on(blk_addr == NEW_ADDR);
 118         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 119                                                         dn->ofs_in_node;
 120
 121         /* Update the page address in the parent node */
 122         __set_data_blkaddr(dn, blk_addr);
 123
 124         if (is_inode_flag_set(fi, FI_NO_EXTENT))
 125                 return;
 126
 127         write_lock(&fi->ext.ext_lock);
 128
 129         start_fofs = fi->ext.fofs;
 130         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 131         start_blkaddr = fi->ext.blk_addr;
 132         end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
 133
 134         /* Drop and initialize the matched extent */
 135         if (fi->ext.len == 1 && fofs == start_fofs)
 136                 fi->ext.len = 0;
 137
 138         /* Initial extent */
 139         if (fi->ext.len == 0) {
 140                 if (blk_addr != NULL_ADDR) {
 141                         fi->ext.fofs = fofs;
 142                         fi->ext.blk_addr = blk_addr;
 143                         fi->ext.len = 1;
 144                 }
 145                 goto end_update;
 146         }
 147
 148         /* Front merge */
 149         if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
 150                 fi->ext.fofs--;
 151                 fi->ext.blk_addr--;
 152                 fi->ext.len++;
 153                 goto end_update;
 154         }
 155
 156         /* Back merge */
 157         if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
 158                 fi->ext.len++;
 159                 goto end_update;
 160         }
 161
 162         /* Split the existing extent */
 163         if (fi->ext.len > 1 &&
 164                 fofs >= start_fofs && fofs <= end_fofs) {
 165                 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
 166                         fi->ext.len = fofs - start_fofs;
 167                 } else {
 168                         fi->ext.fofs = fofs + 1;
 169                         fi->ext.blk_addr = start_blkaddr +
 170                                         fofs - start_fofs + 1;
 171                         fi->ext.len -= fofs - start_fofs + 1;
 172                 }
 173         } else {
 174                 need_update = false;
 175         }
 176
 177         /* Finally, if the extent is very fragmented, let's drop the cache. */
 178         if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
 179                 fi->ext.len = 0;
 180                 set_inode_flag(fi, FI_NO_EXTENT);
 181                 need_update = true;
 182         }
 183 end_update:
 184         write_unlock(&fi->ext.ext_lock);
 185         if (need_update)
 186                 sync_inode_page(dn);
 187         return;
 188 }
 189
 190 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
 191 {
 192         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 193         struct address_space *mapping = inode->i_mapping;
 194         struct dnode_of_data dn;
 195         struct page *page;
 196         int err;
 197
 198         page = find_get_page(mapping, index);
 199         if (page && PageUptodate(page))
 200                 return page;
 201         f2fs_put_page(page, 0);
 202
 203         set_new_dnode(&dn, inode, NULL, NULL, 0);
 204         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 205         if (err)
 206                 return ERR_PTR(err);
 207         f2fs_put_dnode(&dn);
 208
 209         if (dn.data_blkaddr == NULL_ADDR)
 210                 return ERR_PTR(-ENOENT);
 211
 212         /* By fallocate(), there is no cached page, but with NEW_ADDR */
 213         if (dn.data_blkaddr == NEW_ADDR)
 214                 return ERR_PTR(-EINVAL);
 215
 216         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
 217         if (!page)
 218                 return ERR_PTR(-ENOMEM);
 219
 220         if (PageUptodate(page)) {
 221                 unlock_page(page);
 222                 return page;
 223         }
 224
 225         err = f2fs_readpage(sbi, page, dn.data_blkaddr,
 226                                         sync ? READ_SYNC : READA);
 227         if (sync) {
 228                 wait_on_page_locked(page);
 229                 if (!PageUptodate(page)) {
 230                         f2fs_put_page(page, 0);
 231                         return ERR_PTR(-EIO);
 232                 }
 233         }
 234         return page;
 235 }
 236
 237 /*
 238  * If it tries to access a hole, return an error.
 239  * Because, the callers, functions in dir.c and GC, should be able to know
 240  * whether this page exists or not.
 241  */
 242 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 243 {
 244         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 245         struct address_space *mapping = inode->i_mapping;
 246         struct dnode_of_data dn;
 247         struct page *page;
 248         int err;
 249
 250 repeat:
 251         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
 252         if (!page)
 253                 return ERR_PTR(-ENOMEM);
 254
 255         set_new_dnode(&dn, inode, NULL, NULL, 0);
 256         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 257         if (err) {
 258                 f2fs_put_page(page, 1);
 259                 return ERR_PTR(err);
 260         }
 261         f2fs_put_dnode(&dn);
 262
 263         if (dn.data_blkaddr == NULL_ADDR) {
 264                 f2fs_put_page(page, 1);
 265                 return ERR_PTR(-ENOENT);
 266         }
 267
 268         if (PageUptodate(page))
 269                 return page;
 270
 271         /*
 272          * A new dentry page is allocated but not able to be written, since its
 273          * new inode page couldn't be allocated due to -ENOSPC.
 274          * In such the case, its blkaddr can be remained as NEW_ADDR.
 275          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 276          */
 277         if (dn.data_blkaddr == NEW_ADDR) {
 278                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 279                 SetPageUptodate(page);
 280                 return page;
 281         }
 282
 283         err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 284         if (err)
 285                 return ERR_PTR(err);
 286
 287         lock_page(page);
 288         if (!PageUptodate(page)) {
 289                 f2fs_put_page(page, 1);
 290                 return ERR_PTR(-EIO);
 291         }
 292         if (page->mapping != mapping) {
 293                 f2fs_put_page(page, 1);
 294                 goto repeat;
 295         }
 296         return page;
 297 }
 298
 299 /*
 300  * Caller ensures that this data page is never allocated.
 301  * A new zero-filled data page is allocated in the page cache.
 302  *
 303  * Also, caller should grab and release a mutex by calling mutex_lock_op() and
 304  * mutex_unlock_op().
 305  * Note that, npage is set only by make_empty_dir.
 306  */
 307 struct page *get_new_data_page(struct inode *inode,
 308                 struct page *npage, pgoff_t index, bool new_i_size)
 309 {
 310         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 311         struct address_space *mapping = inode->i_mapping;
 312         struct page *page;
 313         struct dnode_of_data dn;
 314         int err;
 315
 316         set_new_dnode(&dn, inode, npage, npage, 0);
 317         err = get_dnode_of_data(&dn, index, ALLOC_NODE);
 318         if (err)
 319                 return ERR_PTR(err);
 320
 321         if (dn.data_blkaddr == NULL_ADDR) {
 322                 if (reserve_new_block(&dn)) {
 323                         if (!npage)
 324                                 f2fs_put_dnode(&dn);
 325                         return ERR_PTR(-ENOSPC);
 326                 }
 327         }
 328         if (!npage)
 329                 f2fs_put_dnode(&dn);
 330 repeat:
 331         page = grab_cache_page(mapping, index);
 332         if (!page)
 333                 return ERR_PTR(-ENOMEM);
 334
 335         if (PageUptodate(page))
 336                 return page;
 337
 338         if (dn.data_blkaddr == NEW_ADDR) {
 339                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 340                 SetPageUptodate(page);
 341         } else {
 342                 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 343                 if (err)
 344                         return ERR_PTR(err);
 345                 lock_page(page);
 346                 if (!PageUptodate(page)) {
 347                         f2fs_put_page(page, 1);
 348                         return ERR_PTR(-EIO);
 349                 }
 350                 if (page->mapping != mapping) {
 351                         f2fs_put_page(page, 1);
 352                         goto repeat;
 353                 }
 354         }
 355
 356         if (new_i_size &&
 357                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 358                 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
 359                 /* Only the directory inode sets new_i_size */
 360                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 361                 mark_inode_dirty_sync(inode);
 362         }
 363         return page;
 364 }
 365
 366 static void read_end_io(struct bio *bio, int err)
 367 {
 368         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 369         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 370
 371         do {
 372                 struct page *page = bvec->bv_page;
 373
 374                 if (--bvec >= bio->bi_io_vec)
 375                         prefetchw(&bvec->bv_page->flags);
 376
 377                 if (uptodate) {
 378                         SetPageUptodate(page);
 379                 } else {
 380                         ClearPageUptodate(page);
 381                         SetPageError(page);
 382                 }
 383                 unlock_page(page);
 384         } while (bvec >= bio->bi_io_vec);
 385         bio_put(bio);
 386 }
 387
 388 /*
 389  * Fill the locked page with data located in the block address.
 390  * Return unlocked page.
 391  */
 392 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
 393                                         block_t blk_addr, int type)
 394 {
 395         struct block_device *bdev = sbi->sb->s_bdev;
 396         struct bio *bio;
 397
 398         trace_f2fs_readpage(page, blk_addr, type);
 399
 400         /* Allocate a new bio */
 401         bio = f2fs_bio_alloc(bdev, 1);
 402
 403         /* Initialize the bio */
 404         bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
 405         bio->bi_end_io = read_end_io;
 406
 407         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 408                 bio_put(bio);
 409                 f2fs_put_page(page, 1);
 410                 return -EFAULT;
 411         }
 412
 413         submit_bio(type, bio);
 414         return 0;
 415 }
 416
 417 /*
 418  * This function should be used by the data read flow only where it
 419  * does not check the "create" flag that indicates block allocation.
 420  * The reason for this special functionality is to exploit VFS readahead
 421  * mechanism.
 422  */
 423 static int get_data_block_ro(struct inode *inode, sector_t iblock,
 424                         struct buffer_head *bh_result, int create)
 425 {
 426         unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 427         unsigned maxblocks = bh_result->b_size >> blkbits;
 428         struct dnode_of_data dn;
 429         pgoff_t pgofs;
 430         int err;
 431
 432         /* Get the page offset from the block offset(iblock) */
 433         pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
 434
 435         if (check_extent_cache(inode, pgofs, bh_result)) {
 436                 trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
 437                 return 0;
 438         }
 439
 440         /* When reading holes, we need its node page */
 441         set_new_dnode(&dn, inode, NULL, NULL, 0);
 442         err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
 443         if (err) {
 444                 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
 445                 return (err == -ENOENT) ? 0 : err;
 446         }
 447
 448         /* It does not support data allocation */
 449         f2fs_bug_on(create);
 450
 451         if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
 452                 int i;
 453                 unsigned int end_offset;
 454
 455                 end_offset = IS_INODE(dn.node_page) ?
 456                                 ADDRS_PER_INODE(F2FS_I(inode)) :
 457                                 ADDRS_PER_BLOCK;
 458
 459                 clear_buffer_new(bh_result);
 460
 461                 /* Give more consecutive addresses for the read ahead */
 462                 for (i = 0; i < end_offset - dn.ofs_in_node; i++)
 463                         if (((datablock_addr(dn.node_page,
 464                                                         dn.ofs_in_node + i))
 465                                 != (dn.data_blkaddr + i)) || maxblocks == i)
 466                                 break;
 467                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 468                 bh_result->b_size = (i << blkbits);
 469         }
 470         f2fs_put_dnode(&dn);
 471         trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
 472         return 0;
 473 }
 474
 475 static int f2fs_read_data_page(struct file *file, struct page *page)
 476 {
 477         return mpage_readpage(page, get_data_block_ro);
 478 }
 479
 480 static int f2fs_read_data_pages(struct file *file,
 481                         struct address_space *mapping,
 482                         struct list_head *pages, unsigned nr_pages)
 483 {
 484         return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
 485 }
 486
 487 int do_write_data_page(struct page *page)
 488 {
 489         struct inode *inode = page->mapping->host;
 490         block_t old_blk_addr, new_blk_addr;
 491         struct dnode_of_data dn;
 492         int err = 0;
 493
 494         set_new_dnode(&dn, inode, NULL, NULL, 0);
 495         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 496         if (err)
 497                 return err;
 498
 499         old_blk_addr = dn.data_blkaddr;
 500
 501         /* This page is already truncated */
 502         if (old_blk_addr == NULL_ADDR)
 503                 goto out_writepage;
 504
 505         set_page_writeback(page);
 506
 507         /*
 508          * If current allocation needs SSR,
 509          * it had better in-place writes for updated data.
 510          */
 511         if (unlikely(old_blk_addr != NEW_ADDR &&
 512                         !is_cold_data(page) &&
 513                         need_inplace_update(inode))) {
 514                 rewrite_data_page(F2FS_SB(inode->i_sb), page,
 515                                                 old_blk_addr);
 516         } else {
 517                 write_data_page(inode, page, &dn,
 518                                 old_blk_addr, &new_blk_addr);
 519                 update_extent_cache(new_blk_addr, &dn);
 520         }
 521 out_writepage:
 522         f2fs_put_dnode(&dn);
 523         return err;
 524 }
 525
 526 static int f2fs_write_data_page(struct page *page,
 527                                         struct writeback_control *wbc)
 528 {
 529         struct inode *inode = page->mapping->host;
 530         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 531         loff_t i_size = i_size_read(inode);
 532         const pgoff_t end_index = ((unsigned long long) i_size)
 533                                                         >> PAGE_CACHE_SHIFT;
 534         unsigned offset;
 535         bool need_balance_fs = false;
 536         int err = 0;
 537
 538         if (page->index < end_index)
 539                 goto write;
 540
 541         /*
 542          * If the offset is out-of-range of file size,
 543          * this page does not have to be written to disk.
 544          */
 545         offset = i_size & (PAGE_CACHE_SIZE - 1);
 546         if ((page->index >= end_index + 1) || !offset) {
 547                 if (S_ISDIR(inode->i_mode)) {
 548                         dec_page_count(sbi, F2FS_DIRTY_DENTS);
 549                         inode_dec_dirty_dents(inode);
 550                 }
 551                 goto out;
 552         }
 553
 554         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 555 write:
 556         if (sbi->por_doing) {
 557                 err = AOP_WRITEPAGE_ACTIVATE;
 558                 goto redirty_out;
 559         }
 560
 561         /* Dentry blocks are controlled by checkpoint */
 562         if (S_ISDIR(inode->i_mode)) {
 563                 dec_page_count(sbi, F2FS_DIRTY_DENTS);
 564                 inode_dec_dirty_dents(inode);
 565                 err = do_write_data_page(page);
 566         } else {
 567                 f2fs_lock_op(sbi);
 568                 err = do_write_data_page(page);
 569                 f2fs_unlock_op(sbi);
 570                 need_balance_fs = true;
 571         }
 572         if (err == -ENOENT)
 573                 goto out;
 574         else if (err)
 575                 goto redirty_out;
 576
 577         if (wbc->for_reclaim)
 578                 f2fs_submit_bio(sbi, DATA, true);
 579
 580         clear_cold_data(page);
 581 out:
 582         unlock_page(page);
 583         if (need_balance_fs)
 584                 f2fs_balance_fs(sbi);
 585         return 0;
 586
 587 redirty_out:
 588         wbc->pages_skipped++;
 589         set_page_dirty(page);
 590         return err;
 591 }
 592
 593 #define MAX_DESIRED_PAGES_WP    4096
 594
 595 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
 596                         void *data)
 597 {
 598         struct address_space *mapping = data;
 599         int ret = mapping->a_ops->writepage(page, wbc);
 600         mapping_set_error(mapping, ret);
 601         return ret;
 602 }
 603
 604 static int f2fs_write_data_pages(struct address_space *mapping,
 605                             struct writeback_control *wbc)
 606 {
 607         struct inode *inode = mapping->host;
 608         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 609         bool locked = false;
 610         int ret;
 611         long excess_nrtw = 0, desired_nrtw;
 612
 613         /* deal with chardevs and other special file */
 614         if (!mapping->a_ops->writepage)
 615                 return 0;
 616
 617         if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
 618                 desired_nrtw = MAX_DESIRED_PAGES_WP;
 619                 excess_nrtw = desired_nrtw - wbc->nr_to_write;
 620                 wbc->nr_to_write = desired_nrtw;
 621         }
 622
 623         if (!S_ISDIR(inode->i_mode)) {
 624                 mutex_lock(&sbi->writepages);
 625                 locked = true;
 626         }
 627         ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
 628         if (locked)
 629                 mutex_unlock(&sbi->writepages);
 630         f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
 631
 632         remove_dirty_dir_inode(inode);
 633
 634         wbc->nr_to_write -= excess_nrtw;
 635         return ret;
 636 }
 637
 638 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 639                 loff_t pos, unsigned len, unsigned flags,
 640                 struct page **pagep, void **fsdata)
 641 {
 642         struct inode *inode = mapping->host;
 643         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 644         struct page *page;
 645         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
 646         struct dnode_of_data dn;
 647         int err = 0;
 648
 649         f2fs_balance_fs(sbi);
 650 repeat:
 651         page = grab_cache_page_write_begin(mapping, index, flags);
 652         if (!page)
 653                 return -ENOMEM;
 654         *pagep = page;
 655
 656         f2fs_lock_op(sbi);
 657
 658         set_new_dnode(&dn, inode, NULL, NULL, 0);
 659         err = get_dnode_of_data(&dn, index, ALLOC_NODE);
 660         if (err)
 661                 goto err;
 662
 663         if (dn.data_blkaddr == NULL_ADDR)
 664                 err = reserve_new_block(&dn);
 665
 666         f2fs_put_dnode(&dn);
 667         if (err)
 668                 goto err;
 669
 670         f2fs_unlock_op(sbi);
 671
 672         if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
 673                 return 0;
 674
 675         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
 676                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 677                 unsigned end = start + len;
 678
 679                 /* Reading beyond i_size is simple: memset to zero */
 680                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
 681                 goto out;
 682         }
 683
 684         if (dn.data_blkaddr == NEW_ADDR) {
 685                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 686         } else {
 687                 err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
 688                 if (err)
 689                         return err;
 690                 lock_page(page);
 691                 if (!PageUptodate(page)) {
 692                         f2fs_put_page(page, 1);
 693                         return -EIO;
 694                 }
 695                 if (page->mapping != mapping) {
 696                         f2fs_put_page(page, 1);
 697                         goto repeat;
 698                 }
 699         }
 700 out:
 701         SetPageUptodate(page);
 702         clear_cold_data(page);
 703         return 0;
 704
 705 err:
 706         f2fs_unlock_op(sbi);
 707         f2fs_put_page(page, 1);
 708         return err;
 709 }
 710
 711 static int f2fs_write_end(struct file *file,
 712                         struct address_space *mapping,
 713                         loff_t pos, unsigned len, unsigned copied,
 714                         struct page *page, void *fsdata)
 715 {
 716         struct inode *inode = page->mapping->host;
 717
 718         SetPageUptodate(page);
 719         set_page_dirty(page);
 720
 721         if (pos + copied > i_size_read(inode)) {
 722                 i_size_write(inode, pos + copied);
 723                 mark_inode_dirty(inode);
 724                 update_inode_page(inode);
 725         }
 726
 727         f2fs_put_page(page, 1);
 728         return copied;
 729 }
 730
 731 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
 732                 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
 733 {
 734         struct file *file = iocb->ki_filp;
 735         struct inode *inode = file->f_mapping->host;
 736
 737         if (rw == WRITE)
 738                 return 0;
 739
 740         /* Needs synchronization with the cleaner */
 741         return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 742                                                   get_data_block_ro);
 743 }
 744
 745 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
 746                                       unsigned int length)
 747 {
 748         struct inode *inode = page->mapping->host;
 749         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 750         if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
 751                 dec_page_count(sbi, F2FS_DIRTY_DENTS);
 752                 inode_dec_dirty_dents(inode);
 753         }
 754         ClearPagePrivate(page);
 755 }
 756
 757 static int f2fs_release_data_page(struct page *page, gfp_t wait)
 758 {
 759         ClearPagePrivate(page);
 760         return 1;
 761 }
 762
 763 static int f2fs_set_data_page_dirty(struct page *page)
 764 {
 765         struct address_space *mapping = page->mapping;
 766         struct inode *inode = mapping->host;
 767
 768         trace_f2fs_set_page_dirty(page, DATA);
 769
 770         SetPageUptodate(page);
 771         if (!PageDirty(page)) {
 772                 __set_page_dirty_nobuffers(page);
 773                 set_dirty_dir_page(inode, page);
 774                 return 1;
 775         }
 776         return 0;
 777 }
 778
 779 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
 780 {
 781         return generic_block_bmap(mapping, block, get_data_block_ro);
 782 }
 783
 784 const struct address_space_operations f2fs_dblock_aops = {
 785         .readpage       = f2fs_read_data_page,
 786         .readpages      = f2fs_read_data_pages,
 787         .writepage      = f2fs_write_data_page,
 788         .writepages     = f2fs_write_data_pages,
 789         .write_begin    = f2fs_write_begin,
 790         .write_end      = f2fs_write_end,
 791         .set_page_dirty = f2fs_set_data_page_dirty,
 792         .invalidatepage = f2fs_invalidate_data_page,
 793         .releasepage    = f2fs_release_data_page,
 794         .direct_IO      = f2fs_direct_IO,
 795         .bmap           = f2fs_bmap,
 796 };