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 &&
 451                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 452                 i_size_write(inode, ((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) < ((fofs + 1) << PAGE_CACHE_SHIFT))
 493                 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
 494
 495         /* direct IO doesn't use extent cache to maximize the performance */
 496         f2fs_drop_largest_extent(dn->inode, fofs);
 497
 498         return 0;
 499 }
 500
 501 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
 502                                                         size_t count)
 503 {
 504         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 505         struct dnode_of_data dn;
 506         u64 start = F2FS_BYTES_TO_BLK(offset);
 507         u64 len = F2FS_BYTES_TO_BLK(count);
 508         bool allocated;
 509         u64 end_offset;
 510
 511         while (len) {
 512                 f2fs_balance_fs(sbi);
 513                 f2fs_lock_op(sbi);
 514
 515                 /* When reading holes, we need its node page */
 516                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 517                 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
 518                         goto out;
 519
 520                 allocated = false;
 521                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 522
 523                 while (dn.ofs_in_node < end_offset && len) {
 524                         block_t blkaddr;
 525
 526                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 527                         if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
 528                                 if (__allocate_data_block(&dn))
 529                                         goto sync_out;
 530                                 allocated = true;
 531                         }
 532                         len--;
 533                         start++;
 534                         dn.ofs_in_node++;
 535                 }
 536
 537                 if (allocated)
 538                         sync_inode_page(&dn);
 539
 540                 f2fs_put_dnode(&dn);
 541                 f2fs_unlock_op(sbi);
 542         }
 543         return;
 544
 545 sync_out:
 546         if (allocated)
 547                 sync_inode_page(&dn);
 548         f2fs_put_dnode(&dn);
 549 out:
 550         f2fs_unlock_op(sbi);
 551         return;
 552 }
 553
 554 /*
 555  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 556  * f2fs_map_blocks structure.
 557  * If original data blocks are allocated, then give them to blockdev.
 558  * Otherwise,
 559  *     a. preallocate requested block addresses
 560  *     b. do not use extent cache for better performance
 561  *     c. give the block addresses to blockdev
 562  */
 563 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 564                                                 int create, int flag)
 565 {
 566         unsigned int maxblocks = map->m_len;
 567         struct dnode_of_data dn;
 568         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 569         pgoff_t pgofs, end_offset;
 570         int err = 0, ofs = 1;
 571         struct extent_info ei;
 572         bool allocated = false;
 573
 574         map->m_len = 0;
 575         map->m_flags = 0;
 576
 577         /* it only supports block size == page size */
 578         pgofs = (pgoff_t)map->m_lblk;
 579
 580         if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 581                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 582                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 583                 map->m_flags = F2FS_MAP_MAPPED;
 584                 goto out;
 585         }
 586
 587         if (create)
 588                 f2fs_lock_op(F2FS_I_SB(inode));
 589
 590         /* When reading holes, we need its node page */
 591         set_new_dnode(&dn, inode, NULL, NULL, 0);
 592         err = get_dnode_of_data(&dn, pgofs, mode);
 593         if (err) {
 594                 if (err == -ENOENT)
 595                         err = 0;
 596                 goto unlock_out;
 597         }
 598         if (dn.data_blkaddr == NEW_ADDR) {
 599                 if (flag == F2FS_GET_BLOCK_BMAP) {
 600                         err = -ENOENT;
 601                         goto put_out;
 602                 } else if (flag == F2FS_GET_BLOCK_READ ||
 603                                 flag == F2FS_GET_BLOCK_DIO) {
 604                         goto put_out;
 605                 }
 606                 /*
 607                  * if it is in fiemap call path (flag = F2FS_GET_BLOCK_FIEMAP),
 608                  * mark it as mapped and unwritten block.
 609                  */
 610         }
 611
 612         if (dn.data_blkaddr != NULL_ADDR) {
 613                 map->m_flags = F2FS_MAP_MAPPED;
 614                 map->m_pblk = dn.data_blkaddr;
 615                 if (dn.data_blkaddr == NEW_ADDR)
 616                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 617         } else if (create) {
 618                 err = __allocate_data_block(&dn);
 619                 if (err)
 620                         goto put_out;
 621                 allocated = true;
 622                 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
 623                 map->m_pblk = dn.data_blkaddr;
 624         } else {
 625                 if (flag == F2FS_GET_BLOCK_BMAP)
 626                         err = -ENOENT;
 627                 goto put_out;
 628         }
 629
 630         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 631         map->m_len = 1;
 632         dn.ofs_in_node++;
 633         pgofs++;
 634
 635 get_next:
 636         if (dn.ofs_in_node >= end_offset) {
 637                 if (allocated)
 638                         sync_inode_page(&dn);
 639                 allocated = false;
 640                 f2fs_put_dnode(&dn);
 641
 642                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 643                 err = get_dnode_of_data(&dn, pgofs, mode);
 644                 if (err) {
 645                         if (err == -ENOENT)
 646                                 err = 0;
 647                         goto unlock_out;
 648                 }
 649
 650                 if (dn.data_blkaddr == NEW_ADDR &&
 651                                 flag != F2FS_GET_BLOCK_FIEMAP)
 652                         goto put_out;
 653
 654                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 655         }
 656
 657         if (maxblocks > map->m_len) {
 658                 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 659                 if (blkaddr == NULL_ADDR && create) {
 660                         err = __allocate_data_block(&dn);
 661                         if (err)
 662                                 goto sync_out;
 663                         allocated = true;
 664                         map->m_flags |= F2FS_MAP_NEW;
 665                         blkaddr = dn.data_blkaddr;
 666                 }
 667                 /* Give more consecutive addresses for the readahead */
 668                 if ((map->m_pblk != NEW_ADDR &&
 669                                 blkaddr == (map->m_pblk + ofs)) ||
 670                                 (map->m_pblk == NEW_ADDR &&
 671                                 blkaddr == NEW_ADDR)) {
 672                         ofs++;
 673                         dn.ofs_in_node++;
 674                         pgofs++;
 675                         map->m_len++;
 676                         goto get_next;
 677                 }
 678         }
 679 sync_out:
 680         if (allocated)
 681                 sync_inode_page(&dn);
 682 put_out:
 683         f2fs_put_dnode(&dn);
 684 unlock_out:
 685         if (create)
 686                 f2fs_unlock_op(F2FS_I_SB(inode));
 687 out:
 688         trace_f2fs_map_blocks(inode, map, err);
 689         return err;
 690 }
 691
 692 static int __get_data_block(struct inode *inode, sector_t iblock,
 693                         struct buffer_head *bh, int create, int flag)
 694 {
 695         struct f2fs_map_blocks map;
 696         int ret;
 697
 698         map.m_lblk = iblock;
 699         map.m_len = bh->b_size >> inode->i_blkbits;
 700
 701         ret = f2fs_map_blocks(inode, &map, create, flag);
 702         if (!ret) {
 703                 map_bh(bh, inode->i_sb, map.m_pblk);
 704                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 705                 bh->b_size = map.m_len << inode->i_blkbits;
 706         }
 707         return ret;
 708 }
 709
 710 static int get_data_block(struct inode *inode, sector_t iblock,
 711                         struct buffer_head *bh_result, int create, int flag)
 712 {
 713         return __get_data_block(inode, iblock, bh_result, create, flag);
 714 }
 715
 716 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 717                         struct buffer_head *bh_result, int create)
 718 {
 719         return __get_data_block(inode, iblock, bh_result, create,
 720                                                 F2FS_GET_BLOCK_DIO);
 721 }
 722
 723 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 724                         struct buffer_head *bh_result, int create)
 725 {
 726         return __get_data_block(inode, iblock, bh_result, create,
 727                                                 F2FS_GET_BLOCK_BMAP);
 728 }
 729
 730 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 731 {
 732         return (offset >> inode->i_blkbits);
 733 }
 734
 735 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 736 {
 737         return (blk << inode->i_blkbits);
 738 }
 739
 740 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 741                 u64 start, u64 len)
 742 {
 743         struct buffer_head map_bh;
 744         sector_t start_blk, last_blk;
 745         loff_t isize = i_size_read(inode);
 746         u64 logical = 0, phys = 0, size = 0;
 747         u32 flags = 0;
 748         bool past_eof = false, whole_file = false;
 749         int ret = 0;
 750
 751         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 752         if (ret)
 753                 return ret;
 754
 755         mutex_lock(&inode->i_mutex);
 756
 757         if (len >= isize) {
 758                 whole_file = true;
 759                 len = isize;
 760         }
 761
 762         if (logical_to_blk(inode, len) == 0)
 763                 len = blk_to_logical(inode, 1);
 764
 765         start_blk = logical_to_blk(inode, start);
 766         last_blk = logical_to_blk(inode, start + len - 1);
 767 next:
 768         memset(&map_bh, 0, sizeof(struct buffer_head));
 769         map_bh.b_size = len;
 770
 771         ret = get_data_block(inode, start_blk, &map_bh, 0,
 772                                         F2FS_GET_BLOCK_FIEMAP);
 773         if (ret)
 774                 goto out;
 775
 776         /* HOLE */
 777         if (!buffer_mapped(&map_bh)) {
 778                 start_blk++;
 779
 780                 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
 781                         past_eof = 1;
 782
 783                 if (past_eof && size) {
 784                         flags |= FIEMAP_EXTENT_LAST;
 785                         ret = fiemap_fill_next_extent(fieinfo, logical,
 786                                         phys, size, flags);
 787                 } else if (size) {
 788                         ret = fiemap_fill_next_extent(fieinfo, logical,
 789                                         phys, size, flags);
 790                         size = 0;
 791                 }
 792
 793                 /* if we have holes up to/past EOF then we're done */
 794                 if (start_blk > last_blk || past_eof || ret)
 795                         goto out;
 796         } else {
 797                 if (start_blk > last_blk && !whole_file) {
 798                         ret = fiemap_fill_next_extent(fieinfo, logical,
 799                                         phys, size, flags);
 800                         goto out;
 801                 }
 802
 803                 /*
 804                  * if size != 0 then we know we already have an extent
 805                  * to add, so add it.
 806                  */
 807                 if (size) {
 808                         ret = fiemap_fill_next_extent(fieinfo, logical,
 809                                         phys, size, flags);
 810                         if (ret)
 811                                 goto out;
 812                 }
 813
 814                 logical = blk_to_logical(inode, start_blk);
 815                 phys = blk_to_logical(inode, map_bh.b_blocknr);
 816                 size = map_bh.b_size;
 817                 flags = 0;
 818                 if (buffer_unwritten(&map_bh))
 819                         flags = FIEMAP_EXTENT_UNWRITTEN;
 820
 821                 start_blk += logical_to_blk(inode, size);
 822
 823                 /*
 824                  * If we are past the EOF, then we need to make sure as
 825                  * soon as we find a hole that the last extent we found
 826                  * is marked with FIEMAP_EXTENT_LAST
 827                  */
 828                 if (!past_eof && logical + size >= isize)
 829                         past_eof = true;
 830         }
 831         cond_resched();
 832         if (fatal_signal_pending(current))
 833                 ret = -EINTR;
 834         else
 835                 goto next;
 836 out:
 837         if (ret == 1)
 838                 ret = 0;
 839
 840         mutex_unlock(&inode->i_mutex);
 841         return ret;
 842 }
 843
 844 /*
 845  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 846  * Major change was from block_size == page_size in f2fs by default.
 847  */
 848 static int f2fs_mpage_readpages(struct address_space *mapping,
 849                         struct list_head *pages, struct page *page,
 850                         unsigned nr_pages)
 851 {
 852         struct bio *bio = NULL;
 853         unsigned page_idx;
 854         sector_t last_block_in_bio = 0;
 855         struct inode *inode = mapping->host;
 856         const unsigned blkbits = inode->i_blkbits;
 857         const unsigned blocksize = 1 << blkbits;
 858         sector_t block_in_file;
 859         sector_t last_block;
 860         sector_t last_block_in_file;
 861         sector_t block_nr;
 862         struct block_device *bdev = inode->i_sb->s_bdev;
 863         struct f2fs_map_blocks map;
 864
 865         map.m_pblk = 0;
 866         map.m_lblk = 0;
 867         map.m_len = 0;
 868         map.m_flags = 0;
 869
 870         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 871
 872                 prefetchw(&page->flags);
 873                 if (pages) {
 874                         page = list_entry(pages->prev, struct page, lru);
 875                         list_del(&page->lru);
 876                         if (add_to_page_cache_lru(page, mapping,
 877                                                   page->index, GFP_KERNEL))
 878                                 goto next_page;
 879                 }
 880
 881                 block_in_file = (sector_t)page->index;
 882                 last_block = block_in_file + nr_pages;
 883                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 884                                                                 blkbits;
 885                 if (last_block > last_block_in_file)
 886                         last_block = last_block_in_file;
 887
 888                 /*
 889                  * Map blocks using the previous result first.
 890                  */
 891                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 892                                 block_in_file > map.m_lblk &&
 893                                 block_in_file < (map.m_lblk + map.m_len))
 894                         goto got_it;
 895
 896                 /*
 897                  * Then do more f2fs_map_blocks() calls until we are
 898                  * done with this page.
 899                  */
 900                 map.m_flags = 0;
 901
 902                 if (block_in_file < last_block) {
 903                         map.m_lblk = block_in_file;
 904                         map.m_len = last_block - block_in_file;
 905
 906                         if (f2fs_map_blocks(inode, &map, 0, false))
 907                                 goto set_error_page;
 908                 }
 909 got_it:
 910                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 911                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 912                         SetPageMappedToDisk(page);
 913
 914                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 915                                 SetPageUptodate(page);
 916                                 goto confused;
 917                         }
 918                 } else {
 919                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 920                         SetPageUptodate(page);
 921                         unlock_page(page);
 922                         goto next_page;
 923                 }
 924
 925                 /*
 926                  * This page will go to BIO.  Do we need to send this
 927                  * BIO off first?
 928                  */
 929                 if (bio && (last_block_in_bio != block_nr - 1)) {
 930 submit_and_realloc:
 931                         submit_bio(READ, bio);
 932                         bio = NULL;
 933                 }
 934                 if (bio == NULL) {
 935                         struct f2fs_crypto_ctx *ctx = NULL;
 936
 937                         if (f2fs_encrypted_inode(inode) &&
 938                                         S_ISREG(inode->i_mode)) {
 939                                 struct page *cpage;
 940
 941                                 ctx = f2fs_get_crypto_ctx(inode);
 942                                 if (IS_ERR(ctx))
 943                                         goto set_error_page;
 944
 945                                 /* wait the page to be moved by cleaning */
 946                                 cpage = find_lock_page(
 947                                                 META_MAPPING(F2FS_I_SB(inode)),
 948                                                 block_nr);
 949                                 if (cpage) {
 950                                         f2fs_wait_on_page_writeback(cpage,
 951                                                                         DATA);
 952                                         f2fs_put_page(cpage, 1);
 953                                 }
 954                         }
 955
 956                         bio = bio_alloc(GFP_KERNEL,
 957                                 min_t(int, nr_pages, BIO_MAX_PAGES));
 958                         if (!bio) {
 959                                 if (ctx)
 960                                         f2fs_release_crypto_ctx(ctx);
 961                                 goto set_error_page;
 962                         }
 963                         bio->bi_bdev = bdev;
 964                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
 965                         bio->bi_end_io = f2fs_read_end_io;
 966                         bio->bi_private = ctx;
 967                 }
 968
 969                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
 970                         goto submit_and_realloc;
 971
 972                 last_block_in_bio = block_nr;
 973                 goto next_page;
 974 set_error_page:
 975                 SetPageError(page);
 976                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 977                 unlock_page(page);
 978                 goto next_page;
 979 confused:
 980                 if (bio) {
 981                         submit_bio(READ, bio);
 982                         bio = NULL;
 983                 }
 984                 unlock_page(page);
 985 next_page:
 986                 if (pages)
 987                         page_cache_release(page);
 988         }
 989         BUG_ON(pages && !list_empty(pages));
 990         if (bio)
 991                 submit_bio(READ, bio);
 992         return 0;
 993 }
 994
 995 static int f2fs_read_data_page(struct file *file, struct page *page)
 996 {
 997         struct inode *inode = page->mapping->host;
 998         int ret = -EAGAIN;
 999
1000         trace_f2fs_readpage(page, DATA);
1001
1002         /* If the file has inline data, try to read it directly */
1003         if (f2fs_has_inline_data(inode))
1004                 ret = f2fs_read_inline_data(inode, page);
1005         if (ret == -EAGAIN)
1006                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1007         return ret;
1008 }
1009
1010 static int f2fs_read_data_pages(struct file *file,
1011                         struct address_space *mapping,
1012                         struct list_head *pages, unsigned nr_pages)
1013 {
1014         struct inode *inode = file->f_mapping->host;
1015
1016         /* If the file has inline data, skip readpages */
1017         if (f2fs_has_inline_data(inode))
1018                 return 0;
1019
1020         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1021 }
1022
1023 int do_write_data_page(struct f2fs_io_info *fio)
1024 {
1025         struct page *page = fio->page;
1026         struct inode *inode = page->mapping->host;
1027         struct dnode_of_data dn;
1028         int err = 0;
1029
1030         set_new_dnode(&dn, inode, NULL, NULL, 0);
1031         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1032         if (err)
1033                 return err;
1034
1035         fio->blk_addr = dn.data_blkaddr;
1036
1037         /* This page is already truncated */
1038         if (fio->blk_addr == NULL_ADDR) {
1039                 ClearPageUptodate(page);
1040                 goto out_writepage;
1041         }
1042
1043         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1044                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1045                 if (IS_ERR(fio->encrypted_page)) {
1046                         err = PTR_ERR(fio->encrypted_page);
1047                         goto out_writepage;
1048                 }
1049         }
1050
1051         set_page_writeback(page);
1052
1053         /*
1054          * If current allocation needs SSR,
1055          * it had better in-place writes for updated data.
1056          */
1057         if (unlikely(fio->blk_addr != NEW_ADDR &&
1058                         !is_cold_data(page) &&
1059                         need_inplace_update(inode))) {
1060                 rewrite_data_page(fio);
1061                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1062                 trace_f2fs_do_write_data_page(page, IPU);
1063         } else {
1064                 write_data_page(&dn, fio);
1065                 set_data_blkaddr(&dn);
1066                 f2fs_update_extent_cache(&dn);
1067                 trace_f2fs_do_write_data_page(page, OPU);
1068                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1069                 if (page->index == 0)
1070                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1071         }
1072 out_writepage:
1073         f2fs_put_dnode(&dn);
1074         return err;
1075 }
1076
1077 static int f2fs_write_data_page(struct page *page,
1078                                         struct writeback_control *wbc)
1079 {
1080         struct inode *inode = page->mapping->host;
1081         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1082         loff_t i_size = i_size_read(inode);
1083         const pgoff_t end_index = ((unsigned long long) i_size)
1084                                                         >> PAGE_CACHE_SHIFT;
1085         unsigned offset = 0;
1086         bool need_balance_fs = false;
1087         int err = 0;
1088         struct f2fs_io_info fio = {
1089                 .sbi = sbi,
1090                 .type = DATA,
1091                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1092                 .page = page,
1093                 .encrypted_page = NULL,
1094         };
1095
1096         trace_f2fs_writepage(page, DATA);
1097
1098         if (page->index < end_index)
1099                 goto write;
1100
1101         /*
1102          * If the offset is out-of-range of file size,
1103          * this page does not have to be written to disk.
1104          */
1105         offset = i_size & (PAGE_CACHE_SIZE - 1);
1106         if ((page->index >= end_index + 1) || !offset)
1107                 goto out;
1108
1109         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1110 write:
1111         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1112                 goto redirty_out;
1113         if (f2fs_is_drop_cache(inode))
1114                 goto out;
1115         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1116                         available_free_memory(sbi, BASE_CHECK))
1117                 goto redirty_out;
1118
1119         /* Dentry blocks are controlled by checkpoint */
1120         if (S_ISDIR(inode->i_mode)) {
1121                 if (unlikely(f2fs_cp_error(sbi)))
1122                         goto redirty_out;
1123                 err = do_write_data_page(&fio);
1124                 goto done;
1125         }
1126
1127         /* we should bypass data pages to proceed the kworkder jobs */
1128         if (unlikely(f2fs_cp_error(sbi))) {
1129                 SetPageError(page);
1130                 goto out;
1131         }
1132
1133         if (!wbc->for_reclaim)
1134                 need_balance_fs = true;
1135         else if (has_not_enough_free_secs(sbi, 0))
1136                 goto redirty_out;
1137
1138         err = -EAGAIN;
1139         f2fs_lock_op(sbi);
1140         if (f2fs_has_inline_data(inode))
1141                 err = f2fs_write_inline_data(inode, page);
1142         if (err == -EAGAIN)
1143                 err = do_write_data_page(&fio);
1144         f2fs_unlock_op(sbi);
1145 done:
1146         if (err && err != -ENOENT)
1147                 goto redirty_out;
1148
1149         clear_cold_data(page);
1150 out:
1151         inode_dec_dirty_pages(inode);
1152         if (err)
1153                 ClearPageUptodate(page);
1154         unlock_page(page);
1155         if (need_balance_fs)
1156                 f2fs_balance_fs(sbi);
1157         if (wbc->for_reclaim)
1158                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1159         return 0;
1160
1161 redirty_out:
1162         redirty_page_for_writepage(wbc, page);
1163         return AOP_WRITEPAGE_ACTIVATE;
1164 }
1165
1166 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1167                         void *data)
1168 {
1169         struct address_space *mapping = data;
1170         int ret = mapping->a_ops->writepage(page, wbc);
1171         mapping_set_error(mapping, ret);
1172         return ret;
1173 }
1174
1175 /*
1176  * This function was copied from write_cche_pages from mm/page-writeback.c.
1177  * The major change is making write step of cold data page separately from
1178  * warm/hot data page.
1179  */
1180 static int f2fs_write_cache_pages(struct address_space *mapping,
1181                         struct writeback_control *wbc, writepage_t writepage,
1182                         void *data)
1183 {
1184         int ret = 0;
1185         int done = 0;
1186         struct pagevec pvec;
1187         int nr_pages;
1188         pgoff_t uninitialized_var(writeback_index);
1189         pgoff_t index;
1190         pgoff_t end;            /* Inclusive */
1191         pgoff_t done_index;
1192         int cycled;
1193         int range_whole = 0;
1194         int tag;
1195         int step = 0;
1196
1197         pagevec_init(&pvec, 0);
1198 next:
1199         if (wbc->range_cyclic) {
1200                 writeback_index = mapping->writeback_index; /* prev offset */
1201                 index = writeback_index;
1202                 if (index == 0)
1203                         cycled = 1;
1204                 else
1205                         cycled = 0;
1206                 end = -1;
1207         } else {
1208                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1209                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1210                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1211                         range_whole = 1;
1212                 cycled = 1; /* ignore range_cyclic tests */
1213         }
1214         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1215                 tag = PAGECACHE_TAG_TOWRITE;
1216         else
1217                 tag = PAGECACHE_TAG_DIRTY;
1218 retry:
1219         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1220                 tag_pages_for_writeback(mapping, index, end);
1221         done_index = index;
1222         while (!done && (index <= end)) {
1223                 int i;
1224
1225                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1226                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1227                 if (nr_pages == 0)
1228                         break;
1229
1230                 for (i = 0; i < nr_pages; i++) {
1231                         struct page *page = pvec.pages[i];
1232
1233                         if (page->index > end) {
1234                                 done = 1;
1235                                 break;
1236                         }
1237
1238                         done_index = page->index;
1239
1240                         lock_page(page);
1241
1242                         if (unlikely(page->mapping != mapping)) {
1243 continue_unlock:
1244                                 unlock_page(page);
1245                                 continue;
1246                         }
1247
1248                         if (!PageDirty(page)) {
1249                                 /* someone wrote it for us */
1250                                 goto continue_unlock;
1251                         }
1252
1253                         if (step == is_cold_data(page))
1254                                 goto continue_unlock;
1255
1256                         if (PageWriteback(page)) {
1257                                 if (wbc->sync_mode != WB_SYNC_NONE)
1258                                         f2fs_wait_on_page_writeback(page, DATA);
1259                                 else
1260                                         goto continue_unlock;
1261                         }
1262
1263                         BUG_ON(PageWriteback(page));
1264                         if (!clear_page_dirty_for_io(page))
1265                                 goto continue_unlock;
1266
1267                         ret = (*writepage)(page, wbc, data);
1268                         if (unlikely(ret)) {
1269                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1270                                         unlock_page(page);
1271                                         ret = 0;
1272                                 } else {
1273                                         done_index = page->index + 1;
1274                                         done = 1;
1275                                         break;
1276                                 }
1277                         }
1278
1279                         if (--wbc->nr_to_write <= 0 &&
1280                             wbc->sync_mode == WB_SYNC_NONE) {
1281                                 done = 1;
1282                                 break;
1283                         }
1284                 }
1285                 pagevec_release(&pvec);
1286                 cond_resched();
1287         }
1288
1289         if (step < 1) {
1290                 step++;
1291                 goto next;
1292         }
1293
1294         if (!cycled && !done) {
1295                 cycled = 1;
1296                 index = 0;
1297                 end = writeback_index - 1;
1298                 goto retry;
1299         }
1300         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1301                 mapping->writeback_index = done_index;
1302
1303         return ret;
1304 }
1305
1306 static int f2fs_write_data_pages(struct address_space *mapping,
1307                             struct writeback_control *wbc)
1308 {
1309         struct inode *inode = mapping->host;
1310         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1311         bool locked = false;
1312         int ret;
1313         long diff;
1314
1315         trace_f2fs_writepages(mapping->host, wbc, DATA);
1316
1317         /* deal with chardevs and other special file */
1318         if (!mapping->a_ops->writepage)
1319                 return 0;
1320
1321         /* skip writing if there is no dirty page in this inode */
1322         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1323                 return 0;
1324
1325         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1326                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1327                         available_free_memory(sbi, DIRTY_DENTS))
1328                 goto skip_write;
1329
1330         /* during POR, we don't need to trigger writepage at all. */
1331         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1332                 goto skip_write;
1333
1334         diff = nr_pages_to_write(sbi, DATA, wbc);
1335
1336         if (!S_ISDIR(inode->i_mode)) {
1337                 mutex_lock(&sbi->writepages);
1338                 locked = true;
1339         }
1340         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1341         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1342         if (locked)
1343                 mutex_unlock(&sbi->writepages);
1344
1345         remove_dirty_dir_inode(inode);
1346
1347         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1348         return ret;
1349
1350 skip_write:
1351         wbc->pages_skipped += get_dirty_pages(inode);
1352         return 0;
1353 }
1354
1355 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1356 {
1357         struct inode *inode = mapping->host;
1358
1359         if (to > inode->i_size) {
1360                 truncate_pagecache(inode, inode->i_size);
1361                 truncate_blocks(inode, inode->i_size, true);
1362         }
1363 }
1364
1365 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1366                 loff_t pos, unsigned len, unsigned flags,
1367                 struct page **pagep, void **fsdata)
1368 {
1369         struct inode *inode = mapping->host;
1370         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1371         struct page *page = NULL;
1372         struct page *ipage;
1373         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1374         struct dnode_of_data dn;
1375         int err = 0;
1376
1377         trace_f2fs_write_begin(inode, pos, len, flags);
1378
1379         f2fs_balance_fs(sbi);
1380
1381         /*
1382          * We should check this at this moment to avoid deadlock on inode page
1383          * and #0 page. The locking rule for inline_data conversion should be:
1384          * lock_page(page #0) -> lock_page(inode_page)
1385          */
1386         if (index != 0) {
1387                 err = f2fs_convert_inline_inode(inode);
1388                 if (err)
1389                         goto fail;
1390         }
1391 repeat:
1392         page = grab_cache_page_write_begin(mapping, index, flags);
1393         if (!page) {
1394                 err = -ENOMEM;
1395                 goto fail;
1396         }
1397
1398         *pagep = page;
1399
1400         f2fs_lock_op(sbi);
1401
1402         /* check inline_data */
1403         ipage = get_node_page(sbi, inode->i_ino);
1404         if (IS_ERR(ipage)) {
1405                 err = PTR_ERR(ipage);
1406                 goto unlock_fail;
1407         }
1408
1409         set_new_dnode(&dn, inode, ipage, ipage, 0);
1410
1411         if (f2fs_has_inline_data(inode)) {
1412                 if (pos + len <= MAX_INLINE_DATA) {
1413                         read_inline_data(page, ipage);
1414                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1415                         sync_inode_page(&dn);
1416                         goto put_next;
1417                 }
1418                 err = f2fs_convert_inline_page(&dn, page);
1419                 if (err)
1420                         goto put_fail;
1421         }
1422
1423         err = f2fs_get_block(&dn, index);
1424         if (err)
1425                 goto put_fail;
1426 put_next:
1427         f2fs_put_dnode(&dn);
1428         f2fs_unlock_op(sbi);
1429
1430         f2fs_wait_on_page_writeback(page, DATA);
1431
1432         if (len == PAGE_CACHE_SIZE)
1433                 goto out_update;
1434         if (PageUptodate(page))
1435                 goto out_clear;
1436
1437         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1438                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1439                 unsigned end = start + len;
1440
1441                 /* Reading beyond i_size is simple: memset to zero */
1442                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1443                 goto out_update;
1444         }
1445
1446         if (dn.data_blkaddr == NEW_ADDR) {
1447                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1448         } else {
1449                 struct f2fs_io_info fio = {
1450                         .sbi = sbi,
1451                         .type = DATA,
1452                         .rw = READ_SYNC,
1453                         .blk_addr = dn.data_blkaddr,
1454                         .page = page,
1455                         .encrypted_page = NULL,
1456                 };
1457                 err = f2fs_submit_page_bio(&fio);
1458                 if (err)
1459                         goto fail;
1460
1461                 lock_page(page);
1462                 if (unlikely(!PageUptodate(page))) {
1463                         err = -EIO;
1464                         goto fail;
1465                 }
1466                 if (unlikely(page->mapping != mapping)) {
1467                         f2fs_put_page(page, 1);
1468                         goto repeat;
1469                 }
1470
1471                 /* avoid symlink page */
1472                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1473                         err = f2fs_decrypt_one(inode, page);
1474                         if (err)
1475                                 goto fail;
1476                 }
1477         }
1478 out_update:
1479         SetPageUptodate(page);
1480 out_clear:
1481         clear_cold_data(page);
1482         return 0;
1483
1484 put_fail:
1485         f2fs_put_dnode(&dn);
1486 unlock_fail:
1487         f2fs_unlock_op(sbi);
1488 fail:
1489         f2fs_put_page(page, 1);
1490         f2fs_write_failed(mapping, pos + len);
1491         return err;
1492 }
1493
1494 static int f2fs_write_end(struct file *file,
1495                         struct address_space *mapping,
1496                         loff_t pos, unsigned len, unsigned copied,
1497                         struct page *page, void *fsdata)
1498 {
1499         struct inode *inode = page->mapping->host;
1500
1501         trace_f2fs_write_end(inode, pos, len, copied);
1502
1503         set_page_dirty(page);
1504
1505         if (pos + copied > i_size_read(inode)) {
1506                 i_size_write(inode, pos + copied);
1507                 mark_inode_dirty(inode);
1508                 update_inode_page(inode);
1509         }
1510
1511         f2fs_put_page(page, 1);
1512         return copied;
1513 }
1514
1515 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1516                            loff_t offset)
1517 {
1518         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1519
1520         if (offset & blocksize_mask)
1521                 return -EINVAL;
1522
1523         if (iov_iter_alignment(iter) & blocksize_mask)
1524                 return -EINVAL;
1525
1526         return 0;
1527 }
1528
1529 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1530                               loff_t offset)
1531 {
1532         struct file *file = iocb->ki_filp;
1533         struct address_space *mapping = file->f_mapping;
1534         struct inode *inode = mapping->host;
1535         size_t count = iov_iter_count(iter);
1536         int err;
1537
1538         /* we don't need to use inline_data strictly */
1539         if (f2fs_has_inline_data(inode)) {
1540                 err = f2fs_convert_inline_inode(inode);
1541                 if (err)
1542                         return err;
1543         }
1544
1545         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1546                 return 0;
1547
1548         err = check_direct_IO(inode, iter, offset);
1549         if (err)
1550                 return err;
1551
1552         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1553
1554         if (iov_iter_rw(iter) == WRITE)
1555                 __allocate_data_blocks(inode, offset, count);
1556
1557         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1558         if (err < 0 && iov_iter_rw(iter) == WRITE)
1559                 f2fs_write_failed(mapping, offset + count);
1560
1561         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1562
1563         return err;
1564 }
1565
1566 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1567                                                         unsigned int length)
1568 {
1569         struct inode *inode = page->mapping->host;
1570         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1571
1572         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1573                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1574                 return;
1575
1576         if (PageDirty(page)) {
1577                 if (inode->i_ino == F2FS_META_INO(sbi))
1578                         dec_page_count(sbi, F2FS_DIRTY_META);
1579                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1580                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1581                 else
1582                         inode_dec_dirty_pages(inode);
1583         }
1584
1585         /* This is atomic written page, keep Private */
1586         if (IS_ATOMIC_WRITTEN_PAGE(page))
1587                 return;
1588
1589         ClearPagePrivate(page);
1590 }
1591
1592 int f2fs_release_page(struct page *page, gfp_t wait)
1593 {
1594         /* If this is dirty page, keep PagePrivate */
1595         if (PageDirty(page))
1596                 return 0;
1597
1598         /* This is atomic written page, keep Private */
1599         if (IS_ATOMIC_WRITTEN_PAGE(page))
1600                 return 0;
1601
1602         ClearPagePrivate(page);
1603         return 1;
1604 }
1605
1606 static int f2fs_set_data_page_dirty(struct page *page)
1607 {
1608         struct address_space *mapping = page->mapping;
1609         struct inode *inode = mapping->host;
1610
1611         trace_f2fs_set_page_dirty(page, DATA);
1612
1613         SetPageUptodate(page);
1614
1615         if (f2fs_is_atomic_file(inode)) {
1616                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1617                         register_inmem_page(inode, page);
1618                         return 1;
1619                 }
1620                 /*
1621                  * Previously, this page has been registered, we just
1622                  * return here.
1623                  */
1624                 return 0;
1625         }
1626
1627         if (!PageDirty(page)) {
1628                 __set_page_dirty_nobuffers(page);
1629                 update_dirty_page(inode, page);
1630                 return 1;
1631         }
1632         return 0;
1633 }
1634
1635 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1636 {
1637         struct inode *inode = mapping->host;
1638
1639         /* we don't need to use inline_data strictly */
1640         if (f2fs_has_inline_data(inode)) {
1641                 int err = f2fs_convert_inline_inode(inode);
1642                 if (err)
1643                         return err;
1644         }
1645         return generic_block_bmap(mapping, block, get_data_block_bmap);
1646 }
1647
1648 const struct address_space_operations f2fs_dblock_aops = {
1649         .readpage       = f2fs_read_data_page,
1650         .readpages      = f2fs_read_data_pages,
1651         .writepage      = f2fs_write_data_page,
1652         .writepages     = f2fs_write_data_pages,
1653         .write_begin    = f2fs_write_begin,
1654         .write_end      = f2fs_write_end,
1655         .set_page_dirty = f2fs_set_data_page_dirty,
1656         .invalidatepage = f2fs_invalidate_page,
1657         .releasepage    = f2fs_release_page,
1658         .direct_IO      = f2fs_direct_IO,
1659         .bmap           = f2fs_bmap,
1660 };