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, int err)
  31 {
  32         struct bio_vec *bvec;
  33         int i;
  34
  35         if (f2fs_bio_encrypted(bio)) {
  36                 if (err) {
  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 (!err) {
  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, int err)
  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(err)) {
  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         /* No failure on bio allocation */
  94         bio = bio_alloc(GFP_NOIO, npages);
  95
  96         bio->bi_bdev = sbi->sb->s_bdev;
  97         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  98         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  99         bio->bi_private = is_read ? NULL : sbi;
 100
 101         return bio;
 102 }
 103
 104 static void __submit_merged_bio(struct f2fs_bio_info *io)
 105 {
 106         struct f2fs_io_info *fio = &io->fio;
 107
 108         if (!io->bio)
 109                 return;
 110
 111         if (is_read_io(fio->rw))
 112                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 113         else
 114                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 115
 116         submit_bio(fio->rw, io->bio);
 117         io->bio = NULL;
 118 }
 119
 120 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 121                                 enum page_type type, int rw)
 122 {
 123         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 124         struct f2fs_bio_info *io;
 125
 126         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 127
 128         down_write(&io->io_rwsem);
 129
 130         /* change META to META_FLUSH in the checkpoint procedure */
 131         if (type >= META_FLUSH) {
 132                 io->fio.type = META_FLUSH;
 133                 if (test_opt(sbi, NOBARRIER))
 134                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 135                 else
 136                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 137         }
 138         __submit_merged_bio(io);
 139         up_write(&io->io_rwsem);
 140 }
 141
 142 /*
 143  * Fill the locked page with data located in the block address.
 144  * Return unlocked page.
 145  */
 146 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 147 {
 148         struct bio *bio;
 149         struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 150
 151         trace_f2fs_submit_page_bio(page, fio);
 152         f2fs_trace_ios(fio, 0);
 153
 154         /* Allocate a new bio */
 155         bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
 156
 157         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 158                 bio_put(bio);
 159                 return -EFAULT;
 160         }
 161
 162         submit_bio(fio->rw, bio);
 163         return 0;
 164 }
 165
 166 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 167 {
 168         struct f2fs_sb_info *sbi = fio->sbi;
 169         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 170         struct f2fs_bio_info *io;
 171         bool is_read = is_read_io(fio->rw);
 172         struct page *bio_page;
 173
 174         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 175
 176         verify_block_addr(sbi, fio->blk_addr);
 177
 178         down_write(&io->io_rwsem);
 179
 180         if (!is_read)
 181                 inc_page_count(sbi, F2FS_WRITEBACK);
 182
 183         if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
 184                                                 io->fio.rw != fio->rw))
 185                 __submit_merged_bio(io);
 186 alloc_new:
 187         if (io->bio == NULL) {
 188                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 189
 190                 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
 191                 io->fio = *fio;
 192         }
 193
 194         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 195
 196         if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
 197                                                         PAGE_CACHE_SIZE) {
 198                 __submit_merged_bio(io);
 199                 goto alloc_new;
 200         }
 201
 202         io->last_block_in_bio = fio->blk_addr;
 203         f2fs_trace_ios(fio, 0);
 204
 205         up_write(&io->io_rwsem);
 206         trace_f2fs_submit_page_mbio(fio->page, fio);
 207 }
 208
 209 /*
 210  * Lock ordering for the change of data block address:
 211  * ->data_page
 212  *  ->node_page
 213  *    update block addresses in the node page
 214  */
 215 void set_data_blkaddr(struct dnode_of_data *dn)
 216 {
 217         struct f2fs_node *rn;
 218         __le32 *addr_array;
 219         struct page *node_page = dn->node_page;
 220         unsigned int ofs_in_node = dn->ofs_in_node;
 221
 222         f2fs_wait_on_page_writeback(node_page, NODE);
 223
 224         rn = F2FS_NODE(node_page);
 225
 226         /* Get physical address of data block */
 227         addr_array = blkaddr_in_node(rn);
 228         addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 229         set_page_dirty(node_page);
 230 }
 231
 232 int reserve_new_block(struct dnode_of_data *dn)
 233 {
 234         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 235
 236         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 237                 return -EPERM;
 238         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 239                 return -ENOSPC;
 240
 241         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 242
 243         dn->data_blkaddr = NEW_ADDR;
 244         set_data_blkaddr(dn);
 245         mark_inode_dirty(dn->inode);
 246         sync_inode_page(dn);
 247         return 0;
 248 }
 249
 250 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 251 {
 252         bool need_put = dn->inode_page ? false : true;
 253         int err;
 254
 255         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 256         if (err)
 257                 return err;
 258
 259         if (dn->data_blkaddr == NULL_ADDR)
 260                 err = reserve_new_block(dn);
 261         if (err || need_put)
 262                 f2fs_put_dnode(dn);
 263         return err;
 264 }
 265
 266 struct page *get_read_data_page(struct inode *inode, pgoff_t index, int rw)
 267 {
 268         struct address_space *mapping = inode->i_mapping;
 269         struct dnode_of_data dn;
 270         struct page *page;
 271         struct extent_info ei;
 272         int err;
 273         struct f2fs_io_info fio = {
 274                 .sbi = F2FS_I_SB(inode),
 275                 .type = DATA,
 276                 .rw = rw,
 277                 .encrypted_page = NULL,
 278         };
 279
 280         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 281                 return read_mapping_page(mapping, index, NULL);
 282
 283         page = grab_cache_page(mapping, index);
 284         if (!page)
 285                 return ERR_PTR(-ENOMEM);
 286
 287         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 288                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 289                 goto got_it;
 290         }
 291
 292         set_new_dnode(&dn, inode, NULL, NULL, 0);
 293         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 294         if (err)
 295                 goto put_err;
 296         f2fs_put_dnode(&dn);
 297
 298         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 299                 err = -ENOENT;
 300                 goto put_err;
 301         }
 302 got_it:
 303         if (PageUptodate(page)) {
 304                 unlock_page(page);
 305                 return page;
 306         }
 307
 308         /*
 309          * A new dentry page is allocated but not able to be written, since its
 310          * new inode page couldn't be allocated due to -ENOSPC.
 311          * In such the case, its blkaddr can be remained as NEW_ADDR.
 312          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 313          */
 314         if (dn.data_blkaddr == NEW_ADDR) {
 315                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 316                 SetPageUptodate(page);
 317                 unlock_page(page);
 318                 return page;
 319         }
 320
 321         fio.blk_addr = dn.data_blkaddr;
 322         fio.page = page;
 323         err = f2fs_submit_page_bio(&fio);
 324         if (err)
 325                 goto put_err;
 326         return page;
 327
 328 put_err:
 329         f2fs_put_page(page, 1);
 330         return ERR_PTR(err);
 331 }
 332
 333 struct page *find_data_page(struct inode *inode, pgoff_t index)
 334 {
 335         struct address_space *mapping = inode->i_mapping;
 336         struct page *page;
 337
 338         page = find_get_page(mapping, index);
 339         if (page && PageUptodate(page))
 340                 return page;
 341         f2fs_put_page(page, 0);
 342
 343         page = get_read_data_page(inode, index, READ_SYNC);
 344         if (IS_ERR(page))
 345                 return page;
 346
 347         if (PageUptodate(page))
 348                 return page;
 349
 350         wait_on_page_locked(page);
 351         if (unlikely(!PageUptodate(page))) {
 352                 f2fs_put_page(page, 0);
 353                 return ERR_PTR(-EIO);
 354         }
 355         return page;
 356 }
 357
 358 /*
 359  * If it tries to access a hole, return an error.
 360  * Because, the callers, functions in dir.c and GC, should be able to know
 361  * whether this page exists or not.
 362  */
 363 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 364 {
 365         struct address_space *mapping = inode->i_mapping;
 366         struct page *page;
 367 repeat:
 368         page = get_read_data_page(inode, index, READ_SYNC);
 369         if (IS_ERR(page))
 370                 return page;
 371
 372         /* wait for read completion */
 373         lock_page(page);
 374         if (unlikely(!PageUptodate(page))) {
 375                 f2fs_put_page(page, 1);
 376                 return ERR_PTR(-EIO);
 377         }
 378         if (unlikely(page->mapping != mapping)) {
 379                 f2fs_put_page(page, 1);
 380                 goto repeat;
 381         }
 382         return page;
 383 }
 384
 385 /*
 386  * Caller ensures that this data page is never allocated.
 387  * A new zero-filled data page is allocated in the page cache.
 388  *
 389  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 390  * f2fs_unlock_op().
 391  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 392  * ipage should be released by this function.
 393  */
 394 struct page *get_new_data_page(struct inode *inode,
 395                 struct page *ipage, pgoff_t index, bool new_i_size)
 396 {
 397         struct address_space *mapping = inode->i_mapping;
 398         struct page *page;
 399         struct dnode_of_data dn;
 400         int err;
 401 repeat:
 402         page = grab_cache_page(mapping, index);
 403         if (!page) {
 404                 /*
 405                  * before exiting, we should make sure ipage will be released
 406                  * if any error occur.
 407                  */
 408                 f2fs_put_page(ipage, 1);
 409                 return ERR_PTR(-ENOMEM);
 410         }
 411
 412         set_new_dnode(&dn, inode, ipage, NULL, 0);
 413         err = f2fs_reserve_block(&dn, index);
 414         if (err) {
 415                 f2fs_put_page(page, 1);
 416                 return ERR_PTR(err);
 417         }
 418         if (!ipage)
 419                 f2fs_put_dnode(&dn);
 420
 421         if (PageUptodate(page))
 422                 goto got_it;
 423
 424         if (dn.data_blkaddr == NEW_ADDR) {
 425                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 426                 SetPageUptodate(page);
 427         } else {
 428                 f2fs_put_page(page, 1);
 429
 430                 page = get_read_data_page(inode, index, READ_SYNC);
 431                 if (IS_ERR(page))
 432                         goto repeat;
 433
 434                 /* wait for read completion */
 435                 lock_page(page);
 436         }
 437 got_it:
 438         if (new_i_size &&
 439                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 440                 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
 441                 /* Only the directory inode sets new_i_size */
 442                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 443         }
 444         return page;
 445 }
 446
 447 static int __allocate_data_block(struct dnode_of_data *dn)
 448 {
 449         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 450         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 451         struct f2fs_summary sum;
 452         struct node_info ni;
 453         int seg = CURSEG_WARM_DATA;
 454         pgoff_t fofs;
 455
 456         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 457                 return -EPERM;
 458
 459         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 460         if (dn->data_blkaddr == NEW_ADDR)
 461                 goto alloc;
 462
 463         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 464                 return -ENOSPC;
 465
 466 alloc:
 467         get_node_info(sbi, dn->nid, &ni);
 468         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 469
 470         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 471                 seg = CURSEG_DIRECT_IO;
 472
 473         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 474                                                                 &sum, seg);
 475         set_data_blkaddr(dn);
 476
 477         /* update i_size */
 478         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 479                                                         dn->ofs_in_node;
 480         if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
 481                 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
 482
 483         /* direct IO doesn't use extent cache to maximize the performance */
 484         f2fs_drop_largest_extent(dn->inode, fofs);
 485
 486         return 0;
 487 }
 488
 489 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
 490                                                         size_t count)
 491 {
 492         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 493         struct dnode_of_data dn;
 494         u64 start = F2FS_BYTES_TO_BLK(offset);
 495         u64 len = F2FS_BYTES_TO_BLK(count);
 496         bool allocated;
 497         u64 end_offset;
 498
 499         while (len) {
 500                 f2fs_balance_fs(sbi);
 501                 f2fs_lock_op(sbi);
 502
 503                 /* When reading holes, we need its node page */
 504                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 505                 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
 506                         goto out;
 507
 508                 allocated = false;
 509                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 510
 511                 while (dn.ofs_in_node < end_offset && len) {
 512                         block_t blkaddr;
 513
 514                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 515                         if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
 516                                 if (__allocate_data_block(&dn))
 517                                         goto sync_out;
 518                                 allocated = true;
 519                         }
 520                         len--;
 521                         start++;
 522                         dn.ofs_in_node++;
 523                 }
 524
 525                 if (allocated)
 526                         sync_inode_page(&dn);
 527
 528                 f2fs_put_dnode(&dn);
 529                 f2fs_unlock_op(sbi);
 530         }
 531         return;
 532
 533 sync_out:
 534         if (allocated)
 535                 sync_inode_page(&dn);
 536         f2fs_put_dnode(&dn);
 537 out:
 538         f2fs_unlock_op(sbi);
 539         return;
 540 }
 541
 542 /*
 543  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 544  * f2fs_map_blocks structure.
 545  * If original data blocks are allocated, then give them to blockdev.
 546  * Otherwise,
 547  *     a. preallocate requested block addresses
 548  *     b. do not use extent cache for better performance
 549  *     c. give the block addresses to blockdev
 550  */
 551 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 552                         int create, bool fiemap)
 553 {
 554         unsigned int maxblocks = map->m_len;
 555         struct dnode_of_data dn;
 556         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 557         pgoff_t pgofs, end_offset;
 558         int err = 0, ofs = 1;
 559         struct extent_info ei;
 560         bool allocated = false;
 561
 562         map->m_len = 0;
 563         map->m_flags = 0;
 564
 565         /* it only supports block size == page size */
 566         pgofs = (pgoff_t)map->m_lblk;
 567
 568         if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 569                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 570                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 571                 map->m_flags = F2FS_MAP_MAPPED;
 572                 goto out;
 573         }
 574
 575         if (create)
 576                 f2fs_lock_op(F2FS_I_SB(inode));
 577
 578         /* When reading holes, we need its node page */
 579         set_new_dnode(&dn, inode, NULL, NULL, 0);
 580         err = get_dnode_of_data(&dn, pgofs, mode);
 581         if (err) {
 582                 if (err == -ENOENT)
 583                         err = 0;
 584                 goto unlock_out;
 585         }
 586         if (dn.data_blkaddr == NEW_ADDR && !fiemap)
 587                 goto put_out;
 588
 589         if (dn.data_blkaddr != NULL_ADDR) {
 590                 map->m_flags = F2FS_MAP_MAPPED;
 591                 map->m_pblk = dn.data_blkaddr;
 592                 if (dn.data_blkaddr == NEW_ADDR)
 593                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 594         } else if (create) {
 595                 err = __allocate_data_block(&dn);
 596                 if (err)
 597                         goto put_out;
 598                 allocated = true;
 599                 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
 600                 map->m_pblk = dn.data_blkaddr;
 601         } else {
 602                 goto put_out;
 603         }
 604
 605         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 606         map->m_len = 1;
 607         dn.ofs_in_node++;
 608         pgofs++;
 609
 610 get_next:
 611         if (dn.ofs_in_node >= end_offset) {
 612                 if (allocated)
 613                         sync_inode_page(&dn);
 614                 allocated = false;
 615                 f2fs_put_dnode(&dn);
 616
 617                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 618                 err = get_dnode_of_data(&dn, pgofs, mode);
 619                 if (err) {
 620                         if (err == -ENOENT)
 621                                 err = 0;
 622                         goto unlock_out;
 623                 }
 624                 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
 625                         goto put_out;
 626
 627                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 628         }
 629
 630         if (maxblocks > map->m_len) {
 631                 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 632                 if (blkaddr == NULL_ADDR && create) {
 633                         err = __allocate_data_block(&dn);
 634                         if (err)
 635                                 goto sync_out;
 636                         allocated = true;
 637                         map->m_flags |= F2FS_MAP_NEW;
 638                         blkaddr = dn.data_blkaddr;
 639                 }
 640                 /* Give more consecutive addresses for the readahead */
 641                 if ((map->m_pblk != NEW_ADDR &&
 642                                 blkaddr == (map->m_pblk + ofs)) ||
 643                                 (map->m_pblk == NEW_ADDR &&
 644                                 blkaddr == NEW_ADDR)) {
 645                         ofs++;
 646                         dn.ofs_in_node++;
 647                         pgofs++;
 648                         map->m_len++;
 649                         goto get_next;
 650                 }
 651         }
 652 sync_out:
 653         if (allocated)
 654                 sync_inode_page(&dn);
 655 put_out:
 656         f2fs_put_dnode(&dn);
 657 unlock_out:
 658         if (create)
 659                 f2fs_unlock_op(F2FS_I_SB(inode));
 660 out:
 661         trace_f2fs_map_blocks(inode, map, err);
 662         return err;
 663 }
 664
 665 static int __get_data_block(struct inode *inode, sector_t iblock,
 666                         struct buffer_head *bh, int create, bool fiemap)
 667 {
 668         struct f2fs_map_blocks map;
 669         int ret;
 670
 671         map.m_lblk = iblock;
 672         map.m_len = bh->b_size >> inode->i_blkbits;
 673
 674         ret = f2fs_map_blocks(inode, &map, create, fiemap);
 675         if (!ret) {
 676                 map_bh(bh, inode->i_sb, map.m_pblk);
 677                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 678                 bh->b_size = map.m_len << inode->i_blkbits;
 679         }
 680         return ret;
 681 }
 682
 683 static int get_data_block(struct inode *inode, sector_t iblock,
 684                         struct buffer_head *bh_result, int create)
 685 {
 686         return __get_data_block(inode, iblock, bh_result, create, false);
 687 }
 688
 689 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
 690                         struct buffer_head *bh_result, int create)
 691 {
 692         return __get_data_block(inode, iblock, bh_result, create, true);
 693 }
 694
 695 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 696 {
 697         return (offset >> inode->i_blkbits);
 698 }
 699
 700 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 701 {
 702         return (blk << inode->i_blkbits);
 703 }
 704
 705 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 706                 u64 start, u64 len)
 707 {
 708         struct buffer_head map_bh;
 709         sector_t start_blk, last_blk;
 710         loff_t isize = i_size_read(inode);
 711         u64 logical = 0, phys = 0, size = 0;
 712         u32 flags = 0;
 713         bool past_eof = false, whole_file = false;
 714         int ret = 0;
 715
 716         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 717         if (ret)
 718                 return ret;
 719
 720         mutex_lock(&inode->i_mutex);
 721
 722         if (len >= isize) {
 723                 whole_file = true;
 724                 len = isize;
 725         }
 726
 727         if (logical_to_blk(inode, len) == 0)
 728                 len = blk_to_logical(inode, 1);
 729
 730         start_blk = logical_to_blk(inode, start);
 731         last_blk = logical_to_blk(inode, start + len - 1);
 732 next:
 733         memset(&map_bh, 0, sizeof(struct buffer_head));
 734         map_bh.b_size = len;
 735
 736         ret = get_data_block_fiemap(inode, start_blk, &map_bh, 0);
 737         if (ret)
 738                 goto out;
 739
 740         /* HOLE */
 741         if (!buffer_mapped(&map_bh)) {
 742                 start_blk++;
 743
 744                 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
 745                         past_eof = 1;
 746
 747                 if (past_eof && size) {
 748                         flags |= FIEMAP_EXTENT_LAST;
 749                         ret = fiemap_fill_next_extent(fieinfo, logical,
 750                                         phys, size, flags);
 751                 } else if (size) {
 752                         ret = fiemap_fill_next_extent(fieinfo, logical,
 753                                         phys, size, flags);
 754                         size = 0;
 755                 }
 756
 757                 /* if we have holes up to/past EOF then we're done */
 758                 if (start_blk > last_blk || past_eof || ret)
 759                         goto out;
 760         } else {
 761                 if (start_blk > last_blk && !whole_file) {
 762                         ret = fiemap_fill_next_extent(fieinfo, logical,
 763                                         phys, size, flags);
 764                         goto out;
 765                 }
 766
 767                 /*
 768                  * if size != 0 then we know we already have an extent
 769                  * to add, so add it.
 770                  */
 771                 if (size) {
 772                         ret = fiemap_fill_next_extent(fieinfo, logical,
 773                                         phys, size, flags);
 774                         if (ret)
 775                                 goto out;
 776                 }
 777
 778                 logical = blk_to_logical(inode, start_blk);
 779                 phys = blk_to_logical(inode, map_bh.b_blocknr);
 780                 size = map_bh.b_size;
 781                 flags = 0;
 782                 if (buffer_unwritten(&map_bh))
 783                         flags = FIEMAP_EXTENT_UNWRITTEN;
 784
 785                 start_blk += logical_to_blk(inode, size);
 786
 787                 /*
 788                  * If we are past the EOF, then we need to make sure as
 789                  * soon as we find a hole that the last extent we found
 790                  * is marked with FIEMAP_EXTENT_LAST
 791                  */
 792                 if (!past_eof && logical + size >= isize)
 793                         past_eof = true;
 794         }
 795         cond_resched();
 796         if (fatal_signal_pending(current))
 797                 ret = -EINTR;
 798         else
 799                 goto next;
 800 out:
 801         if (ret == 1)
 802                 ret = 0;
 803
 804         mutex_unlock(&inode->i_mutex);
 805         return ret;
 806 }
 807
 808 /*
 809  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 810  * Major change was from block_size == page_size in f2fs by default.
 811  */
 812 static int f2fs_mpage_readpages(struct address_space *mapping,
 813                         struct list_head *pages, struct page *page,
 814                         unsigned nr_pages)
 815 {
 816         struct bio *bio = NULL;
 817         unsigned page_idx;
 818         sector_t last_block_in_bio = 0;
 819         struct inode *inode = mapping->host;
 820         const unsigned blkbits = inode->i_blkbits;
 821         const unsigned blocksize = 1 << blkbits;
 822         sector_t block_in_file;
 823         sector_t last_block;
 824         sector_t last_block_in_file;
 825         sector_t block_nr;
 826         struct block_device *bdev = inode->i_sb->s_bdev;
 827         struct f2fs_map_blocks map;
 828
 829         map.m_pblk = 0;
 830         map.m_lblk = 0;
 831         map.m_len = 0;
 832         map.m_flags = 0;
 833
 834         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 835
 836                 prefetchw(&page->flags);
 837                 if (pages) {
 838                         page = list_entry(pages->prev, struct page, lru);
 839                         list_del(&page->lru);
 840                         if (add_to_page_cache_lru(page, mapping,
 841                                                   page->index, GFP_KERNEL))
 842                                 goto next_page;
 843                 }
 844
 845                 block_in_file = (sector_t)page->index;
 846                 last_block = block_in_file + nr_pages;
 847                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 848                                                                 blkbits;
 849                 if (last_block > last_block_in_file)
 850                         last_block = last_block_in_file;
 851
 852                 /*
 853                  * Map blocks using the previous result first.
 854                  */
 855                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 856                                 block_in_file > map.m_lblk &&
 857                                 block_in_file < (map.m_lblk + map.m_len))
 858                         goto got_it;
 859
 860                 /*
 861                  * Then do more f2fs_map_blocks() calls until we are
 862                  * done with this page.
 863                  */
 864                 map.m_flags = 0;
 865
 866                 if (block_in_file < last_block) {
 867                         map.m_lblk = block_in_file;
 868                         map.m_len = last_block - block_in_file;
 869
 870                         if (f2fs_map_blocks(inode, &map, 0, false))
 871                                 goto set_error_page;
 872                 }
 873 got_it:
 874                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 875                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 876                         SetPageMappedToDisk(page);
 877
 878                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 879                                 SetPageUptodate(page);
 880                                 goto confused;
 881                         }
 882                 } else {
 883                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 884                         SetPageUptodate(page);
 885                         unlock_page(page);
 886                         goto next_page;
 887                 }
 888
 889                 /*
 890                  * This page will go to BIO.  Do we need to send this
 891                  * BIO off first?
 892                  */
 893                 if (bio && (last_block_in_bio != block_nr - 1)) {
 894 submit_and_realloc:
 895                         submit_bio(READ, bio);
 896                         bio = NULL;
 897                 }
 898                 if (bio == NULL) {
 899                         struct f2fs_crypto_ctx *ctx = NULL;
 900
 901                         if (f2fs_encrypted_inode(inode) &&
 902                                         S_ISREG(inode->i_mode)) {
 903                                 struct page *cpage;
 904
 905                                 ctx = f2fs_get_crypto_ctx(inode);
 906                                 if (IS_ERR(ctx))
 907                                         goto set_error_page;
 908
 909                                 /* wait the page to be moved by cleaning */
 910                                 cpage = find_lock_page(
 911                                                 META_MAPPING(F2FS_I_SB(inode)),
 912                                                 block_nr);
 913                                 if (cpage) {
 914                                         f2fs_wait_on_page_writeback(cpage,
 915                                                                         DATA);
 916                                         f2fs_put_page(cpage, 1);
 917                                 }
 918                         }
 919
 920                         bio = bio_alloc(GFP_KERNEL,
 921                                 min_t(int, nr_pages, bio_get_nr_vecs(bdev)));
 922                         if (!bio) {
 923                                 if (ctx)
 924                                         f2fs_release_crypto_ctx(ctx);
 925                                 goto set_error_page;
 926                         }
 927                         bio->bi_bdev = bdev;
 928                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
 929                         bio->bi_end_io = f2fs_read_end_io;
 930                         bio->bi_private = ctx;
 931                 }
 932
 933                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
 934                         goto submit_and_realloc;
 935
 936                 last_block_in_bio = block_nr;
 937                 goto next_page;
 938 set_error_page:
 939                 SetPageError(page);
 940                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 941                 unlock_page(page);
 942                 goto next_page;
 943 confused:
 944                 if (bio) {
 945                         submit_bio(READ, bio);
 946                         bio = NULL;
 947                 }
 948                 unlock_page(page);
 949 next_page:
 950                 if (pages)
 951                         page_cache_release(page);
 952         }
 953         BUG_ON(pages && !list_empty(pages));
 954         if (bio)
 955                 submit_bio(READ, bio);
 956         return 0;
 957 }
 958
 959 static int f2fs_read_data_page(struct file *file, struct page *page)
 960 {
 961         struct inode *inode = page->mapping->host;
 962         int ret = -EAGAIN;
 963
 964         trace_f2fs_readpage(page, DATA);
 965
 966         /* If the file has inline data, try to read it directly */
 967         if (f2fs_has_inline_data(inode))
 968                 ret = f2fs_read_inline_data(inode, page);
 969         if (ret == -EAGAIN)
 970                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
 971         return ret;
 972 }
 973
 974 static int f2fs_read_data_pages(struct file *file,
 975                         struct address_space *mapping,
 976                         struct list_head *pages, unsigned nr_pages)
 977 {
 978         struct inode *inode = file->f_mapping->host;
 979
 980         /* If the file has inline data, skip readpages */
 981         if (f2fs_has_inline_data(inode))
 982                 return 0;
 983
 984         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
 985 }
 986
 987 int do_write_data_page(struct f2fs_io_info *fio)
 988 {
 989         struct page *page = fio->page;
 990         struct inode *inode = page->mapping->host;
 991         struct dnode_of_data dn;
 992         int err = 0;
 993
 994         set_new_dnode(&dn, inode, NULL, NULL, 0);
 995         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 996         if (err)
 997                 return err;
 998
 999         fio->blk_addr = dn.data_blkaddr;
1000
1001         /* This page is already truncated */
1002         if (fio->blk_addr == NULL_ADDR) {
1003                 ClearPageUptodate(page);
1004                 goto out_writepage;
1005         }
1006
1007         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1008                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1009                 if (IS_ERR(fio->encrypted_page)) {
1010                         err = PTR_ERR(fio->encrypted_page);
1011                         goto out_writepage;
1012                 }
1013         }
1014
1015         set_page_writeback(page);
1016
1017         /*
1018          * If current allocation needs SSR,
1019          * it had better in-place writes for updated data.
1020          */
1021         if (unlikely(fio->blk_addr != NEW_ADDR &&
1022                         !is_cold_data(page) &&
1023                         need_inplace_update(inode))) {
1024                 rewrite_data_page(fio);
1025                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1026                 trace_f2fs_do_write_data_page(page, IPU);
1027         } else {
1028                 write_data_page(&dn, fio);
1029                 set_data_blkaddr(&dn);
1030                 f2fs_update_extent_cache(&dn);
1031                 trace_f2fs_do_write_data_page(page, OPU);
1032                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1033                 if (page->index == 0)
1034                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1035         }
1036 out_writepage:
1037         f2fs_put_dnode(&dn);
1038         return err;
1039 }
1040
1041 static int f2fs_write_data_page(struct page *page,
1042                                         struct writeback_control *wbc)
1043 {
1044         struct inode *inode = page->mapping->host;
1045         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1046         loff_t i_size = i_size_read(inode);
1047         const pgoff_t end_index = ((unsigned long long) i_size)
1048                                                         >> PAGE_CACHE_SHIFT;
1049         unsigned offset = 0;
1050         bool need_balance_fs = false;
1051         int err = 0;
1052         struct f2fs_io_info fio = {
1053                 .sbi = sbi,
1054                 .type = DATA,
1055                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1056                 .page = page,
1057                 .encrypted_page = NULL,
1058         };
1059
1060         trace_f2fs_writepage(page, DATA);
1061
1062         if (page->index < end_index)
1063                 goto write;
1064
1065         /*
1066          * If the offset is out-of-range of file size,
1067          * this page does not have to be written to disk.
1068          */
1069         offset = i_size & (PAGE_CACHE_SIZE - 1);
1070         if ((page->index >= end_index + 1) || !offset)
1071                 goto out;
1072
1073         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1074 write:
1075         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1076                 goto redirty_out;
1077         if (f2fs_is_drop_cache(inode))
1078                 goto out;
1079         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1080                         available_free_memory(sbi, BASE_CHECK))
1081                 goto redirty_out;
1082
1083         /* Dentry blocks are controlled by checkpoint */
1084         if (S_ISDIR(inode->i_mode)) {
1085                 if (unlikely(f2fs_cp_error(sbi)))
1086                         goto redirty_out;
1087                 err = do_write_data_page(&fio);
1088                 goto done;
1089         }
1090
1091         /* we should bypass data pages to proceed the kworkder jobs */
1092         if (unlikely(f2fs_cp_error(sbi))) {
1093                 SetPageError(page);
1094                 goto out;
1095         }
1096
1097         if (!wbc->for_reclaim)
1098                 need_balance_fs = true;
1099         else if (has_not_enough_free_secs(sbi, 0))
1100                 goto redirty_out;
1101
1102         err = -EAGAIN;
1103         f2fs_lock_op(sbi);
1104         if (f2fs_has_inline_data(inode))
1105                 err = f2fs_write_inline_data(inode, page);
1106         if (err == -EAGAIN)
1107                 err = do_write_data_page(&fio);
1108         f2fs_unlock_op(sbi);
1109 done:
1110         if (err && err != -ENOENT)
1111                 goto redirty_out;
1112
1113         clear_cold_data(page);
1114 out:
1115         inode_dec_dirty_pages(inode);
1116         if (err)
1117                 ClearPageUptodate(page);
1118         unlock_page(page);
1119         if (need_balance_fs)
1120                 f2fs_balance_fs(sbi);
1121         if (wbc->for_reclaim)
1122                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1123         return 0;
1124
1125 redirty_out:
1126         redirty_page_for_writepage(wbc, page);
1127         return AOP_WRITEPAGE_ACTIVATE;
1128 }
1129
1130 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1131                         void *data)
1132 {
1133         struct address_space *mapping = data;
1134         int ret = mapping->a_ops->writepage(page, wbc);
1135         mapping_set_error(mapping, ret);
1136         return ret;
1137 }
1138
1139 /*
1140  * This function was copied from write_cche_pages from mm/page-writeback.c.
1141  * The major change is making write step of cold data page separately from
1142  * warm/hot data page.
1143  */
1144 static int f2fs_write_cache_pages(struct address_space *mapping,
1145                         struct writeback_control *wbc, writepage_t writepage,
1146                         void *data)
1147 {
1148         int ret = 0;
1149         int done = 0;
1150         struct pagevec pvec;
1151         int nr_pages;
1152         pgoff_t uninitialized_var(writeback_index);
1153         pgoff_t index;
1154         pgoff_t end;            /* Inclusive */
1155         pgoff_t done_index;
1156         int cycled;
1157         int range_whole = 0;
1158         int tag;
1159         int step = 0;
1160
1161         pagevec_init(&pvec, 0);
1162 next:
1163         if (wbc->range_cyclic) {
1164                 writeback_index = mapping->writeback_index; /* prev offset */
1165                 index = writeback_index;
1166                 if (index == 0)
1167                         cycled = 1;
1168                 else
1169                         cycled = 0;
1170                 end = -1;
1171         } else {
1172                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1173                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1174                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1175                         range_whole = 1;
1176                 cycled = 1; /* ignore range_cyclic tests */
1177         }
1178         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1179                 tag = PAGECACHE_TAG_TOWRITE;
1180         else
1181                 tag = PAGECACHE_TAG_DIRTY;
1182 retry:
1183         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1184                 tag_pages_for_writeback(mapping, index, end);
1185         done_index = index;
1186         while (!done && (index <= end)) {
1187                 int i;
1188
1189                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1190                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1191                 if (nr_pages == 0)
1192                         break;
1193
1194                 for (i = 0; i < nr_pages; i++) {
1195                         struct page *page = pvec.pages[i];
1196
1197                         if (page->index > end) {
1198                                 done = 1;
1199                                 break;
1200                         }
1201
1202                         done_index = page->index;
1203
1204                         lock_page(page);
1205
1206                         if (unlikely(page->mapping != mapping)) {
1207 continue_unlock:
1208                                 unlock_page(page);
1209                                 continue;
1210                         }
1211
1212                         if (!PageDirty(page)) {
1213                                 /* someone wrote it for us */
1214                                 goto continue_unlock;
1215                         }
1216
1217                         if (step == is_cold_data(page))
1218                                 goto continue_unlock;
1219
1220                         if (PageWriteback(page)) {
1221                                 if (wbc->sync_mode != WB_SYNC_NONE)
1222                                         f2fs_wait_on_page_writeback(page, DATA);
1223                                 else
1224                                         goto continue_unlock;
1225                         }
1226
1227                         BUG_ON(PageWriteback(page));
1228                         if (!clear_page_dirty_for_io(page))
1229                                 goto continue_unlock;
1230
1231                         ret = (*writepage)(page, wbc, data);
1232                         if (unlikely(ret)) {
1233                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1234                                         unlock_page(page);
1235                                         ret = 0;
1236                                 } else {
1237                                         done_index = page->index + 1;
1238                                         done = 1;
1239                                         break;
1240                                 }
1241                         }
1242
1243                         if (--wbc->nr_to_write <= 0 &&
1244                             wbc->sync_mode == WB_SYNC_NONE) {
1245                                 done = 1;
1246                                 break;
1247                         }
1248                 }
1249                 pagevec_release(&pvec);
1250                 cond_resched();
1251         }
1252
1253         if (step < 1) {
1254                 step++;
1255                 goto next;
1256         }
1257
1258         if (!cycled && !done) {
1259                 cycled = 1;
1260                 index = 0;
1261                 end = writeback_index - 1;
1262                 goto retry;
1263         }
1264         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1265                 mapping->writeback_index = done_index;
1266
1267         return ret;
1268 }
1269
1270 static int f2fs_write_data_pages(struct address_space *mapping,
1271                             struct writeback_control *wbc)
1272 {
1273         struct inode *inode = mapping->host;
1274         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1275         bool locked = false;
1276         int ret;
1277         long diff;
1278
1279         trace_f2fs_writepages(mapping->host, wbc, DATA);
1280
1281         /* deal with chardevs and other special file */
1282         if (!mapping->a_ops->writepage)
1283                 return 0;
1284
1285         /* skip writing if there is no dirty page in this inode */
1286         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1287                 return 0;
1288
1289         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1290                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1291                         available_free_memory(sbi, DIRTY_DENTS))
1292                 goto skip_write;
1293
1294         /* during POR, we don't need to trigger writepage at all. */
1295         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1296                 goto skip_write;
1297
1298         diff = nr_pages_to_write(sbi, DATA, wbc);
1299
1300         if (!S_ISDIR(inode->i_mode)) {
1301                 mutex_lock(&sbi->writepages);
1302                 locked = true;
1303         }
1304         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1305         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1306         if (locked)
1307                 mutex_unlock(&sbi->writepages);
1308
1309         remove_dirty_dir_inode(inode);
1310
1311         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1312         return ret;
1313
1314 skip_write:
1315         wbc->pages_skipped += get_dirty_pages(inode);
1316         return 0;
1317 }
1318
1319 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1320 {
1321         struct inode *inode = mapping->host;
1322
1323         if (to > inode->i_size) {
1324                 truncate_pagecache(inode, inode->i_size);
1325                 truncate_blocks(inode, inode->i_size, true);
1326         }
1327 }
1328
1329 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1330                 loff_t pos, unsigned len, unsigned flags,
1331                 struct page **pagep, void **fsdata)
1332 {
1333         struct inode *inode = mapping->host;
1334         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1335         struct page *page = NULL;
1336         struct page *ipage;
1337         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1338         struct dnode_of_data dn;
1339         int err = 0;
1340
1341         trace_f2fs_write_begin(inode, pos, len, flags);
1342
1343         f2fs_balance_fs(sbi);
1344
1345         /*
1346          * We should check this at this moment to avoid deadlock on inode page
1347          * and #0 page. The locking rule for inline_data conversion should be:
1348          * lock_page(page #0) -> lock_page(inode_page)
1349          */
1350         if (index != 0) {
1351                 err = f2fs_convert_inline_inode(inode);
1352                 if (err)
1353                         goto fail;
1354         }
1355 repeat:
1356         page = grab_cache_page_write_begin(mapping, index, flags);
1357         if (!page) {
1358                 err = -ENOMEM;
1359                 goto fail;
1360         }
1361
1362         *pagep = page;
1363
1364         f2fs_lock_op(sbi);
1365
1366         /* check inline_data */
1367         ipage = get_node_page(sbi, inode->i_ino);
1368         if (IS_ERR(ipage)) {
1369                 err = PTR_ERR(ipage);
1370                 goto unlock_fail;
1371         }
1372
1373         set_new_dnode(&dn, inode, ipage, ipage, 0);
1374
1375         if (f2fs_has_inline_data(inode)) {
1376                 if (pos + len <= MAX_INLINE_DATA) {
1377                         read_inline_data(page, ipage);
1378                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1379                         sync_inode_page(&dn);
1380                         goto put_next;
1381                 }
1382                 err = f2fs_convert_inline_page(&dn, page);
1383                 if (err)
1384                         goto put_fail;
1385         }
1386         err = f2fs_reserve_block(&dn, index);
1387         if (err)
1388                 goto put_fail;
1389 put_next:
1390         f2fs_put_dnode(&dn);
1391         f2fs_unlock_op(sbi);
1392
1393         f2fs_wait_on_page_writeback(page, DATA);
1394
1395         if (len == PAGE_CACHE_SIZE)
1396                 goto out_update;
1397         if (PageUptodate(page))
1398                 goto out_clear;
1399
1400         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1401                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1402                 unsigned end = start + len;
1403
1404                 /* Reading beyond i_size is simple: memset to zero */
1405                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1406                 goto out_update;
1407         }
1408
1409         if (dn.data_blkaddr == NEW_ADDR) {
1410                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1411         } else {
1412                 struct f2fs_io_info fio = {
1413                         .sbi = sbi,
1414                         .type = DATA,
1415                         .rw = READ_SYNC,
1416                         .blk_addr = dn.data_blkaddr,
1417                         .page = page,
1418                         .encrypted_page = NULL,
1419                 };
1420                 err = f2fs_submit_page_bio(&fio);
1421                 if (err)
1422                         goto fail;
1423
1424                 lock_page(page);
1425                 if (unlikely(!PageUptodate(page))) {
1426                         err = -EIO;
1427                         goto fail;
1428                 }
1429                 if (unlikely(page->mapping != mapping)) {
1430                         f2fs_put_page(page, 1);
1431                         goto repeat;
1432                 }
1433
1434                 /* avoid symlink page */
1435                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1436                         err = f2fs_decrypt_one(inode, page);
1437                         if (err)
1438                                 goto fail;
1439                 }
1440         }
1441 out_update:
1442         SetPageUptodate(page);
1443 out_clear:
1444         clear_cold_data(page);
1445         return 0;
1446
1447 put_fail:
1448         f2fs_put_dnode(&dn);
1449 unlock_fail:
1450         f2fs_unlock_op(sbi);
1451 fail:
1452         f2fs_put_page(page, 1);
1453         f2fs_write_failed(mapping, pos + len);
1454         return err;
1455 }
1456
1457 static int f2fs_write_end(struct file *file,
1458                         struct address_space *mapping,
1459                         loff_t pos, unsigned len, unsigned copied,
1460                         struct page *page, void *fsdata)
1461 {
1462         struct inode *inode = page->mapping->host;
1463
1464         trace_f2fs_write_end(inode, pos, len, copied);
1465
1466         set_page_dirty(page);
1467
1468         if (pos + copied > i_size_read(inode)) {
1469                 i_size_write(inode, pos + copied);
1470                 mark_inode_dirty(inode);
1471                 update_inode_page(inode);
1472         }
1473
1474         f2fs_put_page(page, 1);
1475         return copied;
1476 }
1477
1478 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1479                            loff_t offset)
1480 {
1481         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1482
1483         if (iov_iter_rw(iter) == READ)
1484                 return 0;
1485
1486         if (offset & blocksize_mask)
1487                 return -EINVAL;
1488
1489         if (iov_iter_alignment(iter) & blocksize_mask)
1490                 return -EINVAL;
1491
1492         return 0;
1493 }
1494
1495 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1496                               loff_t offset)
1497 {
1498         struct file *file = iocb->ki_filp;
1499         struct address_space *mapping = file->f_mapping;
1500         struct inode *inode = mapping->host;
1501         size_t count = iov_iter_count(iter);
1502         int err;
1503
1504         /* we don't need to use inline_data strictly */
1505         if (f2fs_has_inline_data(inode)) {
1506                 err = f2fs_convert_inline_inode(inode);
1507                 if (err)
1508                         return err;
1509         }
1510
1511         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1512                 return 0;
1513
1514         if (check_direct_IO(inode, iter, offset))
1515                 return 0;
1516
1517         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1518
1519         if (iov_iter_rw(iter) == WRITE)
1520                 __allocate_data_blocks(inode, offset, count);
1521
1522         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block);
1523         if (err < 0 && iov_iter_rw(iter) == WRITE)
1524                 f2fs_write_failed(mapping, offset + count);
1525
1526         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1527
1528         return err;
1529 }
1530
1531 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1532                                                         unsigned int length)
1533 {
1534         struct inode *inode = page->mapping->host;
1535         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1536
1537         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1538                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1539                 return;
1540
1541         if (PageDirty(page)) {
1542                 if (inode->i_ino == F2FS_META_INO(sbi))
1543                         dec_page_count(sbi, F2FS_DIRTY_META);
1544                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1545                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1546                 else
1547                         inode_dec_dirty_pages(inode);
1548         }
1549         ClearPagePrivate(page);
1550 }
1551
1552 int f2fs_release_page(struct page *page, gfp_t wait)
1553 {
1554         /* If this is dirty page, keep PagePrivate */
1555         if (PageDirty(page))
1556                 return 0;
1557
1558         ClearPagePrivate(page);
1559         return 1;
1560 }
1561
1562 static int f2fs_set_data_page_dirty(struct page *page)
1563 {
1564         struct address_space *mapping = page->mapping;
1565         struct inode *inode = mapping->host;
1566
1567         trace_f2fs_set_page_dirty(page, DATA);
1568
1569         SetPageUptodate(page);
1570
1571         if (f2fs_is_atomic_file(inode)) {
1572                 register_inmem_page(inode, page);
1573                 return 1;
1574         }
1575
1576         if (!PageDirty(page)) {
1577                 __set_page_dirty_nobuffers(page);
1578                 update_dirty_page(inode, page);
1579                 return 1;
1580         }
1581         return 0;
1582 }
1583
1584 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1585 {
1586         struct inode *inode = mapping->host;
1587
1588         /* we don't need to use inline_data strictly */
1589         if (f2fs_has_inline_data(inode)) {
1590                 int err = f2fs_convert_inline_inode(inode);
1591                 if (err)
1592                         return err;
1593         }
1594         return generic_block_bmap(mapping, block, get_data_block);
1595 }
1596
1597 const struct address_space_operations f2fs_dblock_aops = {
1598         .readpage       = f2fs_read_data_page,
1599         .readpages      = f2fs_read_data_pages,
1600         .writepage      = f2fs_write_data_page,
1601         .writepages     = f2fs_write_data_pages,
1602         .write_begin    = f2fs_write_begin,
1603         .write_end      = f2fs_write_end,
1604         .set_page_dirty = f2fs_set_data_page_dirty,
1605         .invalidatepage = f2fs_invalidate_page,
1606         .releasepage    = f2fs_release_page,
1607         .direct_IO      = f2fs_direct_IO,
1608         .bmap           = f2fs_bmap,
1609 };