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                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 528                         if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
 529                                 if (__allocate_data_block(&dn))
 530                                         goto sync_out;
 531                                 allocated = true;
 532                         }
 533                         len--;
 534                         start++;
 535                         dn.ofs_in_node++;
 536                 }
 537
 538                 if (allocated)
 539                         sync_inode_page(&dn);
 540
 541                 f2fs_put_dnode(&dn);
 542                 f2fs_unlock_op(sbi);
 543         }
 544         return;
 545
 546 sync_out:
 547         if (allocated)
 548                 sync_inode_page(&dn);
 549         f2fs_put_dnode(&dn);
 550 out:
 551         f2fs_unlock_op(sbi);
 552         return;
 553 }
 554
 555 /*
 556  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 557  * f2fs_map_blocks structure.
 558  * If original data blocks are allocated, then give them to blockdev.
 559  * Otherwise,
 560  *     a. preallocate requested block addresses
 561  *     b. do not use extent cache for better performance
 562  *     c. give the block addresses to blockdev
 563  */
 564 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 565                                                 int create, int flag)
 566 {
 567         unsigned int maxblocks = map->m_len;
 568         struct dnode_of_data dn;
 569         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 570         pgoff_t pgofs, end_offset;
 571         int err = 0, ofs = 1;
 572         struct extent_info ei;
 573         bool allocated = false;
 574
 575         map->m_len = 0;
 576         map->m_flags = 0;
 577
 578         /* it only supports block size == page size */
 579         pgofs = (pgoff_t)map->m_lblk;
 580
 581         if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 582                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 583                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 584                 map->m_flags = F2FS_MAP_MAPPED;
 585                 goto out;
 586         }
 587
 588         if (create)
 589                 f2fs_lock_op(F2FS_I_SB(inode));
 590
 591         /* When reading holes, we need its node page */
 592         set_new_dnode(&dn, inode, NULL, NULL, 0);
 593         err = get_dnode_of_data(&dn, pgofs, mode);
 594         if (err) {
 595                 if (err == -ENOENT)
 596                         err = 0;
 597                 goto unlock_out;
 598         }
 599         if (dn.data_blkaddr == NEW_ADDR) {
 600                 if (flag == F2FS_GET_BLOCK_BMAP) {
 601                         err = -ENOENT;
 602                         goto put_out;
 603                 } else if (flag == F2FS_GET_BLOCK_READ ||
 604                                 flag == F2FS_GET_BLOCK_DIO) {
 605                         goto put_out;
 606                 }
 607                 /*
 608                  * if it is in fiemap call path (flag = F2FS_GET_BLOCK_FIEMAP),
 609                  * mark it as mapped and unwritten block.
 610                  */
 611         }
 612
 613         if (dn.data_blkaddr != NULL_ADDR) {
 614                 map->m_flags = F2FS_MAP_MAPPED;
 615                 map->m_pblk = dn.data_blkaddr;
 616                 if (dn.data_blkaddr == NEW_ADDR)
 617                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 618         } else if (create) {
 619                 err = __allocate_data_block(&dn);
 620                 if (err)
 621                         goto put_out;
 622                 allocated = true;
 623                 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
 624                 map->m_pblk = dn.data_blkaddr;
 625         } else {
 626                 if (flag == F2FS_GET_BLOCK_BMAP)
 627                         err = -ENOENT;
 628                 goto put_out;
 629         }
 630
 631         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 632         map->m_len = 1;
 633         dn.ofs_in_node++;
 634         pgofs++;
 635
 636 get_next:
 637         if (dn.ofs_in_node >= end_offset) {
 638                 if (allocated)
 639                         sync_inode_page(&dn);
 640                 allocated = false;
 641                 f2fs_put_dnode(&dn);
 642
 643                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 644                 err = get_dnode_of_data(&dn, pgofs, mode);
 645                 if (err) {
 646                         if (err == -ENOENT)
 647                                 err = 0;
 648                         goto unlock_out;
 649                 }
 650
 651                 if (dn.data_blkaddr == NEW_ADDR &&
 652                                 flag != F2FS_GET_BLOCK_FIEMAP)
 653                         goto put_out;
 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                 if (blkaddr == NULL_ADDR && create) {
 661                         err = __allocate_data_block(&dn);
 662                         if (err)
 663                                 goto sync_out;
 664                         allocated = true;
 665                         map->m_flags |= F2FS_MAP_NEW;
 666                         blkaddr = dn.data_blkaddr;
 667                 }
 668                 /* Give more consecutive addresses for the readahead */
 669                 if ((map->m_pblk != NEW_ADDR &&
 670                                 blkaddr == (map->m_pblk + ofs)) ||
 671                                 (map->m_pblk == NEW_ADDR &&
 672                                 blkaddr == NEW_ADDR)) {
 673                         ofs++;
 674                         dn.ofs_in_node++;
 675                         pgofs++;
 676                         map->m_len++;
 677                         goto get_next;
 678                 }
 679         }
 680 sync_out:
 681         if (allocated)
 682                 sync_inode_page(&dn);
 683 put_out:
 684         f2fs_put_dnode(&dn);
 685 unlock_out:
 686         if (create)
 687                 f2fs_unlock_op(F2FS_I_SB(inode));
 688 out:
 689         trace_f2fs_map_blocks(inode, map, err);
 690         return err;
 691 }
 692
 693 static int __get_data_block(struct inode *inode, sector_t iblock,
 694                         struct buffer_head *bh, int create, int flag)
 695 {
 696         struct f2fs_map_blocks map;
 697         int ret;
 698
 699         map.m_lblk = iblock;
 700         map.m_len = bh->b_size >> inode->i_blkbits;
 701
 702         ret = f2fs_map_blocks(inode, &map, create, flag);
 703         if (!ret) {
 704                 map_bh(bh, inode->i_sb, map.m_pblk);
 705                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 706                 bh->b_size = map.m_len << inode->i_blkbits;
 707         }
 708         return ret;
 709 }
 710
 711 static int get_data_block(struct inode *inode, sector_t iblock,
 712                         struct buffer_head *bh_result, int create, int flag)
 713 {
 714         return __get_data_block(inode, iblock, bh_result, create, flag);
 715 }
 716
 717 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 718                         struct buffer_head *bh_result, int create)
 719 {
 720         return __get_data_block(inode, iblock, bh_result, create,
 721                                                 F2FS_GET_BLOCK_DIO);
 722 }
 723
 724 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 725                         struct buffer_head *bh_result, int create)
 726 {
 727         return __get_data_block(inode, iblock, bh_result, create,
 728                                                 F2FS_GET_BLOCK_BMAP);
 729 }
 730
 731 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 732 {
 733         return (offset >> inode->i_blkbits);
 734 }
 735
 736 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 737 {
 738         return (blk << inode->i_blkbits);
 739 }
 740
 741 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 742                 u64 start, u64 len)
 743 {
 744         struct buffer_head map_bh;
 745         sector_t start_blk, last_blk;
 746         loff_t isize = i_size_read(inode);
 747         u64 logical = 0, phys = 0, size = 0;
 748         u32 flags = 0;
 749         bool past_eof = false, whole_file = false;
 750         int ret = 0;
 751
 752         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 753         if (ret)
 754                 return ret;
 755
 756         mutex_lock(&inode->i_mutex);
 757
 758         if (len >= isize) {
 759                 whole_file = true;
 760                 len = isize;
 761         }
 762
 763         if (logical_to_blk(inode, len) == 0)
 764                 len = blk_to_logical(inode, 1);
 765
 766         start_blk = logical_to_blk(inode, start);
 767         last_blk = logical_to_blk(inode, start + len - 1);
 768 next:
 769         memset(&map_bh, 0, sizeof(struct buffer_head));
 770         map_bh.b_size = len;
 771
 772         ret = get_data_block(inode, start_blk, &map_bh, 0,
 773                                         F2FS_GET_BLOCK_FIEMAP);
 774         if (ret)
 775                 goto out;
 776
 777         /* HOLE */
 778         if (!buffer_mapped(&map_bh)) {
 779                 start_blk++;
 780
 781                 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
 782                         past_eof = 1;
 783
 784                 if (past_eof && size) {
 785                         flags |= FIEMAP_EXTENT_LAST;
 786                         ret = fiemap_fill_next_extent(fieinfo, logical,
 787                                         phys, size, flags);
 788                 } else if (size) {
 789                         ret = fiemap_fill_next_extent(fieinfo, logical,
 790                                         phys, size, flags);
 791                         size = 0;
 792                 }
 793
 794                 /* if we have holes up to/past EOF then we're done */
 795                 if (start_blk > last_blk || past_eof || ret)
 796                         goto out;
 797         } else {
 798                 if (start_blk > last_blk && !whole_file) {
 799                         ret = fiemap_fill_next_extent(fieinfo, logical,
 800                                         phys, size, flags);
 801                         goto out;
 802                 }
 803
 804                 /*
 805                  * if size != 0 then we know we already have an extent
 806                  * to add, so add it.
 807                  */
 808                 if (size) {
 809                         ret = fiemap_fill_next_extent(fieinfo, logical,
 810                                         phys, size, flags);
 811                         if (ret)
 812                                 goto out;
 813                 }
 814
 815                 logical = blk_to_logical(inode, start_blk);
 816                 phys = blk_to_logical(inode, map_bh.b_blocknr);
 817                 size = map_bh.b_size;
 818                 flags = 0;
 819                 if (buffer_unwritten(&map_bh))
 820                         flags = FIEMAP_EXTENT_UNWRITTEN;
 821
 822                 start_blk += logical_to_blk(inode, size);
 823
 824                 /*
 825                  * If we are past the EOF, then we need to make sure as
 826                  * soon as we find a hole that the last extent we found
 827                  * is marked with FIEMAP_EXTENT_LAST
 828                  */
 829                 if (!past_eof && logical + size >= isize)
 830                         past_eof = true;
 831         }
 832         cond_resched();
 833         if (fatal_signal_pending(current))
 834                 ret = -EINTR;
 835         else
 836                 goto next;
 837 out:
 838         if (ret == 1)
 839                 ret = 0;
 840
 841         mutex_unlock(&inode->i_mutex);
 842         return ret;
 843 }
 844
 845 /*
 846  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 847  * Major change was from block_size == page_size in f2fs by default.
 848  */
 849 static int f2fs_mpage_readpages(struct address_space *mapping,
 850                         struct list_head *pages, struct page *page,
 851                         unsigned nr_pages)
 852 {
 853         struct bio *bio = NULL;
 854         unsigned page_idx;
 855         sector_t last_block_in_bio = 0;
 856         struct inode *inode = mapping->host;
 857         const unsigned blkbits = inode->i_blkbits;
 858         const unsigned blocksize = 1 << blkbits;
 859         sector_t block_in_file;
 860         sector_t last_block;
 861         sector_t last_block_in_file;
 862         sector_t block_nr;
 863         struct block_device *bdev = inode->i_sb->s_bdev;
 864         struct f2fs_map_blocks map;
 865
 866         map.m_pblk = 0;
 867         map.m_lblk = 0;
 868         map.m_len = 0;
 869         map.m_flags = 0;
 870
 871         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 872
 873                 prefetchw(&page->flags);
 874                 if (pages) {
 875                         page = list_entry(pages->prev, struct page, lru);
 876                         list_del(&page->lru);
 877                         if (add_to_page_cache_lru(page, mapping,
 878                                                   page->index, GFP_KERNEL))
 879                                 goto next_page;
 880                 }
 881
 882                 block_in_file = (sector_t)page->index;
 883                 last_block = block_in_file + nr_pages;
 884                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 885                                                                 blkbits;
 886                 if (last_block > last_block_in_file)
 887                         last_block = last_block_in_file;
 888
 889                 /*
 890                  * Map blocks using the previous result first.
 891                  */
 892                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 893                                 block_in_file > map.m_lblk &&
 894                                 block_in_file < (map.m_lblk + map.m_len))
 895                         goto got_it;
 896
 897                 /*
 898                  * Then do more f2fs_map_blocks() calls until we are
 899                  * done with this page.
 900                  */
 901                 map.m_flags = 0;
 902
 903                 if (block_in_file < last_block) {
 904                         map.m_lblk = block_in_file;
 905                         map.m_len = last_block - block_in_file;
 906
 907                         if (f2fs_map_blocks(inode, &map, 0, false))
 908                                 goto set_error_page;
 909                 }
 910 got_it:
 911                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 912                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 913                         SetPageMappedToDisk(page);
 914
 915                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 916                                 SetPageUptodate(page);
 917                                 goto confused;
 918                         }
 919                 } else {
 920                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 921                         SetPageUptodate(page);
 922                         unlock_page(page);
 923                         goto next_page;
 924                 }
 925
 926                 /*
 927                  * This page will go to BIO.  Do we need to send this
 928                  * BIO off first?
 929                  */
 930                 if (bio && (last_block_in_bio != block_nr - 1)) {
 931 submit_and_realloc:
 932                         submit_bio(READ, bio);
 933                         bio = NULL;
 934                 }
 935                 if (bio == NULL) {
 936                         struct f2fs_crypto_ctx *ctx = NULL;
 937
 938                         if (f2fs_encrypted_inode(inode) &&
 939                                         S_ISREG(inode->i_mode)) {
 940                                 struct page *cpage;
 941
 942                                 ctx = f2fs_get_crypto_ctx(inode);
 943                                 if (IS_ERR(ctx))
 944                                         goto set_error_page;
 945
 946                                 /* wait the page to be moved by cleaning */
 947                                 cpage = find_lock_page(
 948                                                 META_MAPPING(F2FS_I_SB(inode)),
 949                                                 block_nr);
 950                                 if (cpage) {
 951                                         f2fs_wait_on_page_writeback(cpage,
 952                                                                         DATA);
 953                                         f2fs_put_page(cpage, 1);
 954                                 }
 955                         }
 956
 957                         bio = bio_alloc(GFP_KERNEL,
 958                                 min_t(int, nr_pages, BIO_MAX_PAGES));
 959                         if (!bio) {
 960                                 if (ctx)
 961                                         f2fs_release_crypto_ctx(ctx);
 962                                 goto set_error_page;
 963                         }
 964                         bio->bi_bdev = bdev;
 965                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
 966                         bio->bi_end_io = f2fs_read_end_io;
 967                         bio->bi_private = ctx;
 968                 }
 969
 970                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
 971                         goto submit_and_realloc;
 972
 973                 last_block_in_bio = block_nr;
 974                 goto next_page;
 975 set_error_page:
 976                 SetPageError(page);
 977                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 978                 unlock_page(page);
 979                 goto next_page;
 980 confused:
 981                 if (bio) {
 982                         submit_bio(READ, bio);
 983                         bio = NULL;
 984                 }
 985                 unlock_page(page);
 986 next_page:
 987                 if (pages)
 988                         page_cache_release(page);
 989         }
 990         BUG_ON(pages && !list_empty(pages));
 991         if (bio)
 992                 submit_bio(READ, bio);
 993         return 0;
 994 }
 995
 996 static int f2fs_read_data_page(struct file *file, struct page *page)
 997 {
 998         struct inode *inode = page->mapping->host;
 999         int ret = -EAGAIN;
1000
1001         trace_f2fs_readpage(page, DATA);
1002
1003         /* If the file has inline data, try to read it directly */
1004         if (f2fs_has_inline_data(inode))
1005                 ret = f2fs_read_inline_data(inode, page);
1006         if (ret == -EAGAIN)
1007                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1008         return ret;
1009 }
1010
1011 static int f2fs_read_data_pages(struct file *file,
1012                         struct address_space *mapping,
1013                         struct list_head *pages, unsigned nr_pages)
1014 {
1015         struct inode *inode = file->f_mapping->host;
1016
1017         /* If the file has inline data, skip readpages */
1018         if (f2fs_has_inline_data(inode))
1019                 return 0;
1020
1021         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1022 }
1023
1024 int do_write_data_page(struct f2fs_io_info *fio)
1025 {
1026         struct page *page = fio->page;
1027         struct inode *inode = page->mapping->host;
1028         struct dnode_of_data dn;
1029         int err = 0;
1030
1031         set_new_dnode(&dn, inode, NULL, NULL, 0);
1032         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1033         if (err)
1034                 return err;
1035
1036         fio->blk_addr = dn.data_blkaddr;
1037
1038         /* This page is already truncated */
1039         if (fio->blk_addr == NULL_ADDR) {
1040                 ClearPageUptodate(page);
1041                 goto out_writepage;
1042         }
1043
1044         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1045                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1046                 if (IS_ERR(fio->encrypted_page)) {
1047                         err = PTR_ERR(fio->encrypted_page);
1048                         goto out_writepage;
1049                 }
1050         }
1051
1052         set_page_writeback(page);
1053
1054         /*
1055          * If current allocation needs SSR,
1056          * it had better in-place writes for updated data.
1057          */
1058         if (unlikely(fio->blk_addr != NEW_ADDR &&
1059                         !is_cold_data(page) &&
1060                         need_inplace_update(inode))) {
1061                 rewrite_data_page(fio);
1062                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1063                 trace_f2fs_do_write_data_page(page, IPU);
1064         } else {
1065                 write_data_page(&dn, fio);
1066                 set_data_blkaddr(&dn);
1067                 f2fs_update_extent_cache(&dn);
1068                 trace_f2fs_do_write_data_page(page, OPU);
1069                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1070                 if (page->index == 0)
1071                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1072         }
1073 out_writepage:
1074         f2fs_put_dnode(&dn);
1075         return err;
1076 }
1077
1078 static int f2fs_write_data_page(struct page *page,
1079                                         struct writeback_control *wbc)
1080 {
1081         struct inode *inode = page->mapping->host;
1082         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1083         loff_t i_size = i_size_read(inode);
1084         const pgoff_t end_index = ((unsigned long long) i_size)
1085                                                         >> PAGE_CACHE_SHIFT;
1086         unsigned offset = 0;
1087         bool need_balance_fs = false;
1088         int err = 0;
1089         struct f2fs_io_info fio = {
1090                 .sbi = sbi,
1091                 .type = DATA,
1092                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1093                 .page = page,
1094                 .encrypted_page = NULL,
1095         };
1096
1097         trace_f2fs_writepage(page, DATA);
1098
1099         if (page->index < end_index)
1100                 goto write;
1101
1102         /*
1103          * If the offset is out-of-range of file size,
1104          * this page does not have to be written to disk.
1105          */
1106         offset = i_size & (PAGE_CACHE_SIZE - 1);
1107         if ((page->index >= end_index + 1) || !offset)
1108                 goto out;
1109
1110         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1111 write:
1112         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1113                 goto redirty_out;
1114         if (f2fs_is_drop_cache(inode))
1115                 goto out;
1116         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1117                         available_free_memory(sbi, BASE_CHECK))
1118                 goto redirty_out;
1119
1120         /* Dentry blocks are controlled by checkpoint */
1121         if (S_ISDIR(inode->i_mode)) {
1122                 if (unlikely(f2fs_cp_error(sbi)))
1123                         goto redirty_out;
1124                 err = do_write_data_page(&fio);
1125                 goto done;
1126         }
1127
1128         /* we should bypass data pages to proceed the kworkder jobs */
1129         if (unlikely(f2fs_cp_error(sbi))) {
1130                 SetPageError(page);
1131                 goto out;
1132         }
1133
1134         if (!wbc->for_reclaim)
1135                 need_balance_fs = true;
1136         else if (has_not_enough_free_secs(sbi, 0))
1137                 goto redirty_out;
1138
1139         err = -EAGAIN;
1140         f2fs_lock_op(sbi);
1141         if (f2fs_has_inline_data(inode))
1142                 err = f2fs_write_inline_data(inode, page);
1143         if (err == -EAGAIN)
1144                 err = do_write_data_page(&fio);
1145         f2fs_unlock_op(sbi);
1146 done:
1147         if (err && err != -ENOENT)
1148                 goto redirty_out;
1149
1150         clear_cold_data(page);
1151 out:
1152         inode_dec_dirty_pages(inode);
1153         if (err)
1154                 ClearPageUptodate(page);
1155         unlock_page(page);
1156         if (need_balance_fs)
1157                 f2fs_balance_fs(sbi);
1158         if (wbc->for_reclaim)
1159                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1160         return 0;
1161
1162 redirty_out:
1163         redirty_page_for_writepage(wbc, page);
1164         return AOP_WRITEPAGE_ACTIVATE;
1165 }
1166
1167 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1168                         void *data)
1169 {
1170         struct address_space *mapping = data;
1171         int ret = mapping->a_ops->writepage(page, wbc);
1172         mapping_set_error(mapping, ret);
1173         return ret;
1174 }
1175
1176 /*
1177  * This function was copied from write_cche_pages from mm/page-writeback.c.
1178  * The major change is making write step of cold data page separately from
1179  * warm/hot data page.
1180  */
1181 static int f2fs_write_cache_pages(struct address_space *mapping,
1182                         struct writeback_control *wbc, writepage_t writepage,
1183                         void *data)
1184 {
1185         int ret = 0;
1186         int done = 0;
1187         struct pagevec pvec;
1188         int nr_pages;
1189         pgoff_t uninitialized_var(writeback_index);
1190         pgoff_t index;
1191         pgoff_t end;            /* Inclusive */
1192         pgoff_t done_index;
1193         int cycled;
1194         int range_whole = 0;
1195         int tag;
1196         int step = 0;
1197
1198         pagevec_init(&pvec, 0);
1199 next:
1200         if (wbc->range_cyclic) {
1201                 writeback_index = mapping->writeback_index; /* prev offset */
1202                 index = writeback_index;
1203                 if (index == 0)
1204                         cycled = 1;
1205                 else
1206                         cycled = 0;
1207                 end = -1;
1208         } else {
1209                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1210                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1211                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1212                         range_whole = 1;
1213                 cycled = 1; /* ignore range_cyclic tests */
1214         }
1215         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1216                 tag = PAGECACHE_TAG_TOWRITE;
1217         else
1218                 tag = PAGECACHE_TAG_DIRTY;
1219 retry:
1220         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1221                 tag_pages_for_writeback(mapping, index, end);
1222         done_index = index;
1223         while (!done && (index <= end)) {
1224                 int i;
1225
1226                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1227                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1228                 if (nr_pages == 0)
1229                         break;
1230
1231                 for (i = 0; i < nr_pages; i++) {
1232                         struct page *page = pvec.pages[i];
1233
1234                         if (page->index > end) {
1235                                 done = 1;
1236                                 break;
1237                         }
1238
1239                         done_index = page->index;
1240
1241                         lock_page(page);
1242
1243                         if (unlikely(page->mapping != mapping)) {
1244 continue_unlock:
1245                                 unlock_page(page);
1246                                 continue;
1247                         }
1248
1249                         if (!PageDirty(page)) {
1250                                 /* someone wrote it for us */
1251                                 goto continue_unlock;
1252                         }
1253
1254                         if (step == is_cold_data(page))
1255                                 goto continue_unlock;
1256
1257                         if (PageWriteback(page)) {
1258                                 if (wbc->sync_mode != WB_SYNC_NONE)
1259                                         f2fs_wait_on_page_writeback(page, DATA);
1260                                 else
1261                                         goto continue_unlock;
1262                         }
1263
1264                         BUG_ON(PageWriteback(page));
1265                         if (!clear_page_dirty_for_io(page))
1266                                 goto continue_unlock;
1267
1268                         ret = (*writepage)(page, wbc, data);
1269                         if (unlikely(ret)) {
1270                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1271                                         unlock_page(page);
1272                                         ret = 0;
1273                                 } else {
1274                                         done_index = page->index + 1;
1275                                         done = 1;
1276                                         break;
1277                                 }
1278                         }
1279
1280                         if (--wbc->nr_to_write <= 0 &&
1281                             wbc->sync_mode == WB_SYNC_NONE) {
1282                                 done = 1;
1283                                 break;
1284                         }
1285                 }
1286                 pagevec_release(&pvec);
1287                 cond_resched();
1288         }
1289
1290         if (step < 1) {
1291                 step++;
1292                 goto next;
1293         }
1294
1295         if (!cycled && !done) {
1296                 cycled = 1;
1297                 index = 0;
1298                 end = writeback_index - 1;
1299                 goto retry;
1300         }
1301         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1302                 mapping->writeback_index = done_index;
1303
1304         return ret;
1305 }
1306
1307 static int f2fs_write_data_pages(struct address_space *mapping,
1308                             struct writeback_control *wbc)
1309 {
1310         struct inode *inode = mapping->host;
1311         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1312         bool locked = false;
1313         int ret;
1314         long diff;
1315
1316         trace_f2fs_writepages(mapping->host, wbc, DATA);
1317
1318         /* deal with chardevs and other special file */
1319         if (!mapping->a_ops->writepage)
1320                 return 0;
1321
1322         /* skip writing if there is no dirty page in this inode */
1323         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1324                 return 0;
1325
1326         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1327                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1328                         available_free_memory(sbi, DIRTY_DENTS))
1329                 goto skip_write;
1330
1331         /* during POR, we don't need to trigger writepage at all. */
1332         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1333                 goto skip_write;
1334
1335         diff = nr_pages_to_write(sbi, DATA, wbc);
1336
1337         if (!S_ISDIR(inode->i_mode)) {
1338                 mutex_lock(&sbi->writepages);
1339                 locked = true;
1340         }
1341         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1342         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1343         if (locked)
1344                 mutex_unlock(&sbi->writepages);
1345
1346         remove_dirty_dir_inode(inode);
1347
1348         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1349         return ret;
1350
1351 skip_write:
1352         wbc->pages_skipped += get_dirty_pages(inode);
1353         return 0;
1354 }
1355
1356 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1357 {
1358         struct inode *inode = mapping->host;
1359
1360         if (to > inode->i_size) {
1361                 truncate_pagecache(inode, inode->i_size);
1362                 truncate_blocks(inode, inode->i_size, true);
1363         }
1364 }
1365
1366 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1367                 loff_t pos, unsigned len, unsigned flags,
1368                 struct page **pagep, void **fsdata)
1369 {
1370         struct inode *inode = mapping->host;
1371         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1372         struct page *page = NULL;
1373         struct page *ipage;
1374         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1375         struct dnode_of_data dn;
1376         int err = 0;
1377
1378         trace_f2fs_write_begin(inode, pos, len, flags);
1379
1380         f2fs_balance_fs(sbi);
1381
1382         /*
1383          * We should check this at this moment to avoid deadlock on inode page
1384          * and #0 page. The locking rule for inline_data conversion should be:
1385          * lock_page(page #0) -> lock_page(inode_page)
1386          */
1387         if (index != 0) {
1388                 err = f2fs_convert_inline_inode(inode);
1389                 if (err)
1390                         goto fail;
1391         }
1392 repeat:
1393         page = grab_cache_page_write_begin(mapping, index, flags);
1394         if (!page) {
1395                 err = -ENOMEM;
1396                 goto fail;
1397         }
1398
1399         *pagep = page;
1400
1401         f2fs_lock_op(sbi);
1402
1403         /* check inline_data */
1404         ipage = get_node_page(sbi, inode->i_ino);
1405         if (IS_ERR(ipage)) {
1406                 err = PTR_ERR(ipage);
1407                 goto unlock_fail;
1408         }
1409
1410         set_new_dnode(&dn, inode, ipage, ipage, 0);
1411
1412         if (f2fs_has_inline_data(inode)) {
1413                 if (pos + len <= MAX_INLINE_DATA) {
1414                         read_inline_data(page, ipage);
1415                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1416                         sync_inode_page(&dn);
1417                         goto put_next;
1418                 }
1419                 err = f2fs_convert_inline_page(&dn, page);
1420                 if (err)
1421                         goto put_fail;
1422         }
1423
1424         err = f2fs_get_block(&dn, index);
1425         if (err)
1426                 goto put_fail;
1427 put_next:
1428         f2fs_put_dnode(&dn);
1429         f2fs_unlock_op(sbi);
1430
1431         f2fs_wait_on_page_writeback(page, DATA);
1432
1433         if (len == PAGE_CACHE_SIZE)
1434                 goto out_update;
1435         if (PageUptodate(page))
1436                 goto out_clear;
1437
1438         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1439                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1440                 unsigned end = start + len;
1441
1442                 /* Reading beyond i_size is simple: memset to zero */
1443                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1444                 goto out_update;
1445         }
1446
1447         if (dn.data_blkaddr == NEW_ADDR) {
1448                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1449         } else {
1450                 struct f2fs_io_info fio = {
1451                         .sbi = sbi,
1452                         .type = DATA,
1453                         .rw = READ_SYNC,
1454                         .blk_addr = dn.data_blkaddr,
1455                         .page = page,
1456                         .encrypted_page = NULL,
1457                 };
1458                 err = f2fs_submit_page_bio(&fio);
1459                 if (err)
1460                         goto fail;
1461
1462                 lock_page(page);
1463                 if (unlikely(!PageUptodate(page))) {
1464                         err = -EIO;
1465                         goto fail;
1466                 }
1467                 if (unlikely(page->mapping != mapping)) {
1468                         f2fs_put_page(page, 1);
1469                         goto repeat;
1470                 }
1471
1472                 /* avoid symlink page */
1473                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1474                         err = f2fs_decrypt_one(inode, page);
1475                         if (err)
1476                                 goto fail;
1477                 }
1478         }
1479 out_update:
1480         SetPageUptodate(page);
1481 out_clear:
1482         clear_cold_data(page);
1483         return 0;
1484
1485 put_fail:
1486         f2fs_put_dnode(&dn);
1487 unlock_fail:
1488         f2fs_unlock_op(sbi);
1489 fail:
1490         f2fs_put_page(page, 1);
1491         f2fs_write_failed(mapping, pos + len);
1492         return err;
1493 }
1494
1495 static int f2fs_write_end(struct file *file,
1496                         struct address_space *mapping,
1497                         loff_t pos, unsigned len, unsigned copied,
1498                         struct page *page, void *fsdata)
1499 {
1500         struct inode *inode = page->mapping->host;
1501
1502         trace_f2fs_write_end(inode, pos, len, copied);
1503
1504         set_page_dirty(page);
1505
1506         if (pos + copied > i_size_read(inode)) {
1507                 i_size_write(inode, pos + copied);
1508                 mark_inode_dirty(inode);
1509                 update_inode_page(inode);
1510         }
1511
1512         f2fs_put_page(page, 1);
1513         return copied;
1514 }
1515
1516 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1517                            loff_t offset)
1518 {
1519         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1520
1521         if (offset & blocksize_mask)
1522                 return -EINVAL;
1523
1524         if (iov_iter_alignment(iter) & blocksize_mask)
1525                 return -EINVAL;
1526
1527         return 0;
1528 }
1529
1530 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1531                               loff_t offset)
1532 {
1533         struct file *file = iocb->ki_filp;
1534         struct address_space *mapping = file->f_mapping;
1535         struct inode *inode = mapping->host;
1536         size_t count = iov_iter_count(iter);
1537         int err;
1538
1539         /* we don't need to use inline_data strictly */
1540         if (f2fs_has_inline_data(inode)) {
1541                 err = f2fs_convert_inline_inode(inode);
1542                 if (err)
1543                         return err;
1544         }
1545
1546         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1547                 return 0;
1548
1549         err = check_direct_IO(inode, iter, offset);
1550         if (err)
1551                 return err;
1552
1553         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1554
1555         if (iov_iter_rw(iter) == WRITE)
1556                 __allocate_data_blocks(inode, offset, count);
1557
1558         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1559         if (err < 0 && iov_iter_rw(iter) == WRITE)
1560                 f2fs_write_failed(mapping, offset + count);
1561
1562         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1563
1564         return err;
1565 }
1566
1567 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1568                                                         unsigned int length)
1569 {
1570         struct inode *inode = page->mapping->host;
1571         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1572
1573         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1574                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1575                 return;
1576
1577         if (PageDirty(page)) {
1578                 if (inode->i_ino == F2FS_META_INO(sbi))
1579                         dec_page_count(sbi, F2FS_DIRTY_META);
1580                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1581                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1582                 else
1583                         inode_dec_dirty_pages(inode);
1584         }
1585
1586         /* This is atomic written page, keep Private */
1587         if (IS_ATOMIC_WRITTEN_PAGE(page))
1588                 return;
1589
1590         ClearPagePrivate(page);
1591 }
1592
1593 int f2fs_release_page(struct page *page, gfp_t wait)
1594 {
1595         /* If this is dirty page, keep PagePrivate */
1596         if (PageDirty(page))
1597                 return 0;
1598
1599         /* This is atomic written page, keep Private */
1600         if (IS_ATOMIC_WRITTEN_PAGE(page))
1601                 return 0;
1602
1603         ClearPagePrivate(page);
1604         return 1;
1605 }
1606
1607 static int f2fs_set_data_page_dirty(struct page *page)
1608 {
1609         struct address_space *mapping = page->mapping;
1610         struct inode *inode = mapping->host;
1611
1612         trace_f2fs_set_page_dirty(page, DATA);
1613
1614         SetPageUptodate(page);
1615
1616         if (f2fs_is_atomic_file(inode)) {
1617                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1618                         register_inmem_page(inode, page);
1619                         return 1;
1620                 }
1621                 /*
1622                  * Previously, this page has been registered, we just
1623                  * return here.
1624                  */
1625                 return 0;
1626         }
1627
1628         if (!PageDirty(page)) {
1629                 __set_page_dirty_nobuffers(page);
1630                 update_dirty_page(inode, page);
1631                 return 1;
1632         }
1633         return 0;
1634 }
1635
1636 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1637 {
1638         struct inode *inode = mapping->host;
1639
1640         /* we don't need to use inline_data strictly */
1641         if (f2fs_has_inline_data(inode)) {
1642                 int err = f2fs_convert_inline_inode(inode);
1643                 if (err)
1644                         return err;
1645         }
1646         return generic_block_bmap(mapping, block, get_data_block_bmap);
1647 }
1648
1649 const struct address_space_operations f2fs_dblock_aops = {
1650         .readpage       = f2fs_read_data_page,
1651         .readpages      = f2fs_read_data_pages,
1652         .writepage      = f2fs_write_data_page,
1653         .writepages     = f2fs_write_data_pages,
1654         .write_begin    = f2fs_write_begin,
1655         .write_end      = f2fs_write_end,
1656         .set_page_dirty = f2fs_set_data_page_dirty,
1657         .invalidatepage = f2fs_invalidate_page,
1658         .releasepage    = f2fs_release_page,
1659         .direct_IO      = f2fs_direct_IO,
1660         .bmap           = f2fs_bmap,
1661 };