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         if (set_page_dirty(node_page))
 229                 dn->node_changed = true;
 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 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 267 {
 268         struct extent_info ei;
 269         struct inode *inode = dn->inode;
 270
 271         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 272                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 273                 return 0;
 274         }
 275
 276         return f2fs_reserve_block(dn, index);
 277 }
 278
 279 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 280                                                 int rw, bool for_write)
 281 {
 282         struct address_space *mapping = inode->i_mapping;
 283         struct dnode_of_data dn;
 284         struct page *page;
 285         struct extent_info ei;
 286         int err;
 287         struct f2fs_io_info fio = {
 288                 .sbi = F2FS_I_SB(inode),
 289                 .type = DATA,
 290                 .rw = rw,
 291                 .encrypted_page = NULL,
 292         };
 293
 294         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 295                 return read_mapping_page(mapping, index, NULL);
 296
 297         page = f2fs_grab_cache_page(mapping, index, for_write);
 298         if (!page)
 299                 return ERR_PTR(-ENOMEM);
 300
 301         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 302                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 303                 goto got_it;
 304         }
 305
 306         set_new_dnode(&dn, inode, NULL, NULL, 0);
 307         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 308         if (err)
 309                 goto put_err;
 310         f2fs_put_dnode(&dn);
 311
 312         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 313                 err = -ENOENT;
 314                 goto put_err;
 315         }
 316 got_it:
 317         if (PageUptodate(page)) {
 318                 unlock_page(page);
 319                 return page;
 320         }
 321
 322         /*
 323          * A new dentry page is allocated but not able to be written, since its
 324          * new inode page couldn't be allocated due to -ENOSPC.
 325          * In such the case, its blkaddr can be remained as NEW_ADDR.
 326          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 327          */
 328         if (dn.data_blkaddr == NEW_ADDR) {
 329                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 330                 SetPageUptodate(page);
 331                 unlock_page(page);
 332                 return page;
 333         }
 334
 335         fio.blk_addr = dn.data_blkaddr;
 336         fio.page = page;
 337         err = f2fs_submit_page_bio(&fio);
 338         if (err)
 339                 goto put_err;
 340         return page;
 341
 342 put_err:
 343         f2fs_put_page(page, 1);
 344         return ERR_PTR(err);
 345 }
 346
 347 struct page *find_data_page(struct inode *inode, pgoff_t index)
 348 {
 349         struct address_space *mapping = inode->i_mapping;
 350         struct page *page;
 351
 352         page = find_get_page(mapping, index);
 353         if (page && PageUptodate(page))
 354                 return page;
 355         f2fs_put_page(page, 0);
 356
 357         page = get_read_data_page(inode, index, READ_SYNC, false);
 358         if (IS_ERR(page))
 359                 return page;
 360
 361         if (PageUptodate(page))
 362                 return page;
 363
 364         wait_on_page_locked(page);
 365         if (unlikely(!PageUptodate(page))) {
 366                 f2fs_put_page(page, 0);
 367                 return ERR_PTR(-EIO);
 368         }
 369         return page;
 370 }
 371
 372 /*
 373  * If it tries to access a hole, return an error.
 374  * Because, the callers, functions in dir.c and GC, should be able to know
 375  * whether this page exists or not.
 376  */
 377 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 378                                                         bool for_write)
 379 {
 380         struct address_space *mapping = inode->i_mapping;
 381         struct page *page;
 382 repeat:
 383         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 384         if (IS_ERR(page))
 385                 return page;
 386
 387         /* wait for read completion */
 388         lock_page(page);
 389         if (unlikely(!PageUptodate(page))) {
 390                 f2fs_put_page(page, 1);
 391                 return ERR_PTR(-EIO);
 392         }
 393         if (unlikely(page->mapping != mapping)) {
 394                 f2fs_put_page(page, 1);
 395                 goto repeat;
 396         }
 397         return page;
 398 }
 399
 400 /*
 401  * Caller ensures that this data page is never allocated.
 402  * A new zero-filled data page is allocated in the page cache.
 403  *
 404  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 405  * f2fs_unlock_op().
 406  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 407  * ipage should be released by this function.
 408  */
 409 struct page *get_new_data_page(struct inode *inode,
 410                 struct page *ipage, pgoff_t index, bool new_i_size)
 411 {
 412         struct address_space *mapping = inode->i_mapping;
 413         struct page *page;
 414         struct dnode_of_data dn;
 415         int err;
 416
 417         page = f2fs_grab_cache_page(mapping, index, true);
 418         if (!page) {
 419                 /*
 420                  * before exiting, we should make sure ipage will be released
 421                  * if any error occur.
 422                  */
 423                 f2fs_put_page(ipage, 1);
 424                 return ERR_PTR(-ENOMEM);
 425         }
 426
 427         set_new_dnode(&dn, inode, ipage, NULL, 0);
 428         err = f2fs_reserve_block(&dn, index);
 429         if (err) {
 430                 f2fs_put_page(page, 1);
 431                 return ERR_PTR(err);
 432         }
 433         if (!ipage)
 434                 f2fs_put_dnode(&dn);
 435
 436         if (PageUptodate(page))
 437                 goto got_it;
 438
 439         if (dn.data_blkaddr == NEW_ADDR) {
 440                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 441                 SetPageUptodate(page);
 442         } else {
 443                 f2fs_put_page(page, 1);
 444
 445                 /* if ipage exists, blkaddr should be NEW_ADDR */
 446                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 447                 page = get_lock_data_page(inode, index, true);
 448                 if (IS_ERR(page))
 449                         return page;
 450         }
 451 got_it:
 452         if (new_i_size && i_size_read(inode) <
 453                                 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
 454                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
 455                 /* Only the directory inode sets new_i_size */
 456                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 457         }
 458         return page;
 459 }
 460
 461 static int __allocate_data_block(struct dnode_of_data *dn)
 462 {
 463         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 464         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 465         struct f2fs_summary sum;
 466         struct node_info ni;
 467         int seg = CURSEG_WARM_DATA;
 468         pgoff_t fofs;
 469
 470         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 471                 return -EPERM;
 472
 473         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 474         if (dn->data_blkaddr == NEW_ADDR)
 475                 goto alloc;
 476
 477         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 478                 return -ENOSPC;
 479
 480 alloc:
 481         get_node_info(sbi, dn->nid, &ni);
 482         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 483
 484         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 485                 seg = CURSEG_DIRECT_IO;
 486
 487         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 488                                                                 &sum, seg);
 489         set_data_blkaddr(dn);
 490
 491         /* update i_size */
 492         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 493                                                         dn->ofs_in_node;
 494         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
 495                 i_size_write(dn->inode,
 496                                 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
 497         return 0;
 498 }
 499
 500 static int __allocate_data_blocks(struct inode *inode, loff_t offset,
 501                                                         size_t count)
 502 {
 503         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 504         struct dnode_of_data dn;
 505         u64 start = F2FS_BYTES_TO_BLK(offset);
 506         u64 len = F2FS_BYTES_TO_BLK(count);
 507         bool allocated;
 508         u64 end_offset;
 509         int err = 0;
 510
 511         while (len) {
 512                 f2fs_lock_op(sbi);
 513
 514                 /* When reading holes, we need its node page */
 515                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 516                 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 517                 if (err)
 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                         if (unlikely(f2fs_cp_error(sbi))) {
 527                                 err = -EIO;
 528                                 goto sync_out;
 529                         }
 530
 531                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 532                         if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
 533                                 err = __allocate_data_block(&dn);
 534                                 if (err)
 535                                         goto sync_out;
 536                                 allocated = true;
 537                         }
 538                         len--;
 539                         start++;
 540                         dn.ofs_in_node++;
 541                 }
 542
 543                 if (allocated)
 544                         sync_inode_page(&dn);
 545
 546                 f2fs_put_dnode(&dn);
 547                 f2fs_unlock_op(sbi);
 548
 549                 f2fs_balance_fs(sbi, dn.node_changed);
 550         }
 551         return err;
 552
 553 sync_out:
 554         if (allocated)
 555                 sync_inode_page(&dn);
 556         f2fs_put_dnode(&dn);
 557 out:
 558         f2fs_unlock_op(sbi);
 559         f2fs_balance_fs(sbi, dn.node_changed);
 560         return err;
 561 }
 562
 563 /*
 564  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 565  * f2fs_map_blocks structure.
 566  * If original data blocks are allocated, then give them to blockdev.
 567  * Otherwise,
 568  *     a. preallocate requested block addresses
 569  *     b. do not use extent cache for better performance
 570  *     c. give the block addresses to blockdev
 571  */
 572 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 573                                                 int create, int flag)
 574 {
 575         unsigned int maxblocks = map->m_len;
 576         struct dnode_of_data dn;
 577         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 578         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 579         pgoff_t pgofs, end_offset;
 580         int err = 0, ofs = 1;
 581         struct extent_info ei;
 582         bool allocated = false;
 583         block_t blkaddr;
 584
 585         map->m_len = 0;
 586         map->m_flags = 0;
 587
 588         /* it only supports block size == page size */
 589         pgofs = (pgoff_t)map->m_lblk;
 590
 591         if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 592                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 593                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 594                 map->m_flags = F2FS_MAP_MAPPED;
 595                 goto out;
 596         }
 597
 598         if (create)
 599                 f2fs_lock_op(sbi);
 600
 601         /* When reading holes, we need its node page */
 602         set_new_dnode(&dn, inode, NULL, NULL, 0);
 603         err = get_dnode_of_data(&dn, pgofs, mode);
 604         if (err) {
 605                 if (err == -ENOENT)
 606                         err = 0;
 607                 goto unlock_out;
 608         }
 609
 610         if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
 611                 if (create) {
 612                         if (unlikely(f2fs_cp_error(sbi))) {
 613                                 err = -EIO;
 614                                 goto put_out;
 615                         }
 616                         err = __allocate_data_block(&dn);
 617                         if (err)
 618                                 goto put_out;
 619                         allocated = true;
 620                         map->m_flags = F2FS_MAP_NEW;
 621                 } else {
 622                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 623                                                 dn.data_blkaddr != NEW_ADDR) {
 624                                 if (flag == F2FS_GET_BLOCK_BMAP)
 625                                         err = -ENOENT;
 626                                 goto put_out;
 627                         }
 628
 629                         /*
 630                          * preallocated unwritten block should be mapped
 631                          * for fiemap.
 632                          */
 633                         if (dn.data_blkaddr == NEW_ADDR)
 634                                 map->m_flags = F2FS_MAP_UNWRITTEN;
 635                 }
 636         }
 637
 638         map->m_flags |= F2FS_MAP_MAPPED;
 639         map->m_pblk = dn.data_blkaddr;
 640         map->m_len = 1;
 641
 642         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 643         dn.ofs_in_node++;
 644         pgofs++;
 645
 646 get_next:
 647         if (map->m_len >= maxblocks)
 648                 goto sync_out;
 649
 650         if (dn.ofs_in_node >= end_offset) {
 651                 if (allocated)
 652                         sync_inode_page(&dn);
 653                 allocated = false;
 654                 f2fs_put_dnode(&dn);
 655
 656                 if (create) {
 657                         f2fs_unlock_op(sbi);
 658                         f2fs_balance_fs(sbi, dn.node_changed);
 659                         f2fs_lock_op(sbi);
 660                 }
 661
 662                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 663                 err = get_dnode_of_data(&dn, pgofs, mode);
 664                 if (err) {
 665                         if (err == -ENOENT)
 666                                 err = 0;
 667                         goto unlock_out;
 668                 }
 669
 670                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 671         }
 672
 673         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 674
 675         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 676                 if (create) {
 677                         if (unlikely(f2fs_cp_error(sbi))) {
 678                                 err = -EIO;
 679                                 goto sync_out;
 680                         }
 681                         err = __allocate_data_block(&dn);
 682                         if (err)
 683                                 goto sync_out;
 684                         allocated = true;
 685                         map->m_flags |= F2FS_MAP_NEW;
 686                         blkaddr = dn.data_blkaddr;
 687                 } else {
 688                         /*
 689                          * we only merge preallocated unwritten blocks
 690                          * for fiemap.
 691                          */
 692                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 693                                         blkaddr != NEW_ADDR)
 694                                 goto sync_out;
 695                 }
 696         }
 697
 698         /* Give more consecutive addresses for the readahead */
 699         if ((map->m_pblk != NEW_ADDR &&
 700                         blkaddr == (map->m_pblk + ofs)) ||
 701                         (map->m_pblk == NEW_ADDR &&
 702                         blkaddr == NEW_ADDR)) {
 703                 ofs++;
 704                 dn.ofs_in_node++;
 705                 pgofs++;
 706                 map->m_len++;
 707                 goto get_next;
 708         }
 709
 710 sync_out:
 711         if (allocated)
 712                 sync_inode_page(&dn);
 713 put_out:
 714         f2fs_put_dnode(&dn);
 715 unlock_out:
 716         if (create) {
 717                 f2fs_unlock_op(sbi);
 718                 f2fs_balance_fs(sbi, dn.node_changed);
 719         }
 720 out:
 721         trace_f2fs_map_blocks(inode, map, err);
 722         return err;
 723 }
 724
 725 static int __get_data_block(struct inode *inode, sector_t iblock,
 726                         struct buffer_head *bh, int create, int flag)
 727 {
 728         struct f2fs_map_blocks map;
 729         int ret;
 730
 731         map.m_lblk = iblock;
 732         map.m_len = bh->b_size >> inode->i_blkbits;
 733
 734         ret = f2fs_map_blocks(inode, &map, create, flag);
 735         if (!ret) {
 736                 map_bh(bh, inode->i_sb, map.m_pblk);
 737                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 738                 bh->b_size = map.m_len << inode->i_blkbits;
 739         }
 740         return ret;
 741 }
 742
 743 static int get_data_block(struct inode *inode, sector_t iblock,
 744                         struct buffer_head *bh_result, int create, int flag)
 745 {
 746         return __get_data_block(inode, iblock, bh_result, create, flag);
 747 }
 748
 749 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 750                         struct buffer_head *bh_result, int create)
 751 {
 752         return __get_data_block(inode, iblock, bh_result, create,
 753                                                 F2FS_GET_BLOCK_DIO);
 754 }
 755
 756 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 757                         struct buffer_head *bh_result, int create)
 758 {
 759         /* Block number less than F2FS MAX BLOCKS */
 760         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 761                 return -EFBIG;
 762
 763         return __get_data_block(inode, iblock, bh_result, create,
 764                                                 F2FS_GET_BLOCK_BMAP);
 765 }
 766
 767 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 768 {
 769         return (offset >> inode->i_blkbits);
 770 }
 771
 772 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 773 {
 774         return (blk << inode->i_blkbits);
 775 }
 776
 777 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 778                 u64 start, u64 len)
 779 {
 780         struct buffer_head map_bh;
 781         sector_t start_blk, last_blk;
 782         loff_t isize;
 783         u64 logical = 0, phys = 0, size = 0;
 784         u32 flags = 0;
 785         int ret = 0;
 786
 787         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 788         if (ret)
 789                 return ret;
 790
 791         if (f2fs_has_inline_data(inode)) {
 792                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 793                 if (ret != -EAGAIN)
 794                         return ret;
 795         }
 796
 797         mutex_lock(&inode->i_mutex);
 798
 799         isize = i_size_read(inode);
 800         if (start >= isize)
 801                 goto out;
 802
 803         if (start + len > isize)
 804                 len = isize - start;
 805
 806         if (logical_to_blk(inode, len) == 0)
 807                 len = blk_to_logical(inode, 1);
 808
 809         start_blk = logical_to_blk(inode, start);
 810         last_blk = logical_to_blk(inode, start + len - 1);
 811
 812 next:
 813         memset(&map_bh, 0, sizeof(struct buffer_head));
 814         map_bh.b_size = len;
 815
 816         ret = get_data_block(inode, start_blk, &map_bh, 0,
 817                                         F2FS_GET_BLOCK_FIEMAP);
 818         if (ret)
 819                 goto out;
 820
 821         /* HOLE */
 822         if (!buffer_mapped(&map_bh)) {
 823                 /* Go through holes util pass the EOF */
 824                 if (blk_to_logical(inode, start_blk++) < isize)
 825                         goto prep_next;
 826                 /* Found a hole beyond isize means no more extents.
 827                  * Note that the premise is that filesystems don't
 828                  * punch holes beyond isize and keep size unchanged.
 829                  */
 830                 flags |= FIEMAP_EXTENT_LAST;
 831         }
 832
 833         if (size)
 834                 ret = fiemap_fill_next_extent(fieinfo, logical,
 835                                 phys, size, flags);
 836
 837         if (start_blk > last_blk || ret)
 838                 goto out;
 839
 840         logical = blk_to_logical(inode, start_blk);
 841         phys = blk_to_logical(inode, map_bh.b_blocknr);
 842         size = map_bh.b_size;
 843         flags = 0;
 844         if (buffer_unwritten(&map_bh))
 845                 flags = FIEMAP_EXTENT_UNWRITTEN;
 846
 847         start_blk += logical_to_blk(inode, size);
 848
 849 prep_next:
 850         cond_resched();
 851         if (fatal_signal_pending(current))
 852                 ret = -EINTR;
 853         else
 854                 goto next;
 855 out:
 856         if (ret == 1)
 857                 ret = 0;
 858
 859         mutex_unlock(&inode->i_mutex);
 860         return ret;
 861 }
 862
 863 /*
 864  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 865  * Major change was from block_size == page_size in f2fs by default.
 866  */
 867 static int f2fs_mpage_readpages(struct address_space *mapping,
 868                         struct list_head *pages, struct page *page,
 869                         unsigned nr_pages)
 870 {
 871         struct bio *bio = NULL;
 872         unsigned page_idx;
 873         sector_t last_block_in_bio = 0;
 874         struct inode *inode = mapping->host;
 875         const unsigned blkbits = inode->i_blkbits;
 876         const unsigned blocksize = 1 << blkbits;
 877         sector_t block_in_file;
 878         sector_t last_block;
 879         sector_t last_block_in_file;
 880         sector_t block_nr;
 881         struct block_device *bdev = inode->i_sb->s_bdev;
 882         struct f2fs_map_blocks map;
 883
 884         map.m_pblk = 0;
 885         map.m_lblk = 0;
 886         map.m_len = 0;
 887         map.m_flags = 0;
 888
 889         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 890
 891                 prefetchw(&page->flags);
 892                 if (pages) {
 893                         page = list_entry(pages->prev, struct page, lru);
 894                         list_del(&page->lru);
 895                         if (add_to_page_cache_lru(page, mapping,
 896                                                   page->index, GFP_KERNEL))
 897                                 goto next_page;
 898                 }
 899
 900                 block_in_file = (sector_t)page->index;
 901                 last_block = block_in_file + nr_pages;
 902                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 903                                                                 blkbits;
 904                 if (last_block > last_block_in_file)
 905                         last_block = last_block_in_file;
 906
 907                 /*
 908                  * Map blocks using the previous result first.
 909                  */
 910                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 911                                 block_in_file > map.m_lblk &&
 912                                 block_in_file < (map.m_lblk + map.m_len))
 913                         goto got_it;
 914
 915                 /*
 916                  * Then do more f2fs_map_blocks() calls until we are
 917                  * done with this page.
 918                  */
 919                 map.m_flags = 0;
 920
 921                 if (block_in_file < last_block) {
 922                         map.m_lblk = block_in_file;
 923                         map.m_len = last_block - block_in_file;
 924
 925                         if (f2fs_map_blocks(inode, &map, 0,
 926                                                         F2FS_GET_BLOCK_READ))
 927                                 goto set_error_page;
 928                 }
 929 got_it:
 930                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 931                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 932                         SetPageMappedToDisk(page);
 933
 934                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 935                                 SetPageUptodate(page);
 936                                 goto confused;
 937                         }
 938                 } else {
 939                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 940                         SetPageUptodate(page);
 941                         unlock_page(page);
 942                         goto next_page;
 943                 }
 944
 945                 /*
 946                  * This page will go to BIO.  Do we need to send this
 947                  * BIO off first?
 948                  */
 949                 if (bio && (last_block_in_bio != block_nr - 1)) {
 950 submit_and_realloc:
 951                         submit_bio(READ, bio);
 952                         bio = NULL;
 953                 }
 954                 if (bio == NULL) {
 955                         struct f2fs_crypto_ctx *ctx = NULL;
 956
 957                         if (f2fs_encrypted_inode(inode) &&
 958                                         S_ISREG(inode->i_mode)) {
 959
 960                                 ctx = f2fs_get_crypto_ctx(inode);
 961                                 if (IS_ERR(ctx))
 962                                         goto set_error_page;
 963
 964                                 /* wait the page to be moved by cleaning */
 965                                 f2fs_wait_on_encrypted_page_writeback(
 966                                                 F2FS_I_SB(inode), block_nr);
 967                         }
 968
 969                         bio = bio_alloc(GFP_KERNEL,
 970                                 min_t(int, nr_pages, BIO_MAX_PAGES));
 971                         if (!bio) {
 972                                 if (ctx)
 973                                         f2fs_release_crypto_ctx(ctx);
 974                                 goto set_error_page;
 975                         }
 976                         bio->bi_bdev = bdev;
 977                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
 978                         bio->bi_end_io = f2fs_read_end_io;
 979                         bio->bi_private = ctx;
 980                 }
 981
 982                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
 983                         goto submit_and_realloc;
 984
 985                 last_block_in_bio = block_nr;
 986                 goto next_page;
 987 set_error_page:
 988                 SetPageError(page);
 989                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 990                 unlock_page(page);
 991                 goto next_page;
 992 confused:
 993                 if (bio) {
 994                         submit_bio(READ, bio);
 995                         bio = NULL;
 996                 }
 997                 unlock_page(page);
 998 next_page:
 999                 if (pages)
1000                         page_cache_release(page);
1001         }
1002         BUG_ON(pages && !list_empty(pages));
1003         if (bio)
1004                 submit_bio(READ, bio);
1005         return 0;
1006 }
1007
1008 static int f2fs_read_data_page(struct file *file, struct page *page)
1009 {
1010         struct inode *inode = page->mapping->host;
1011         int ret = -EAGAIN;
1012
1013         trace_f2fs_readpage(page, DATA);
1014
1015         /* If the file has inline data, try to read it directly */
1016         if (f2fs_has_inline_data(inode))
1017                 ret = f2fs_read_inline_data(inode, page);
1018         if (ret == -EAGAIN)
1019                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1020         return ret;
1021 }
1022
1023 static int f2fs_read_data_pages(struct file *file,
1024                         struct address_space *mapping,
1025                         struct list_head *pages, unsigned nr_pages)
1026 {
1027         struct inode *inode = file->f_mapping->host;
1028         struct page *page = list_entry(pages->prev, struct page, lru);
1029
1030         trace_f2fs_readpages(inode, page, nr_pages);
1031
1032         /* If the file has inline data, skip readpages */
1033         if (f2fs_has_inline_data(inode))
1034                 return 0;
1035
1036         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1037 }
1038
1039 int do_write_data_page(struct f2fs_io_info *fio)
1040 {
1041         struct page *page = fio->page;
1042         struct inode *inode = page->mapping->host;
1043         struct dnode_of_data dn;
1044         int err = 0;
1045
1046         set_new_dnode(&dn, inode, NULL, NULL, 0);
1047         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1048         if (err)
1049                 return err;
1050
1051         fio->blk_addr = dn.data_blkaddr;
1052
1053         /* This page is already truncated */
1054         if (fio->blk_addr == NULL_ADDR) {
1055                 ClearPageUptodate(page);
1056                 goto out_writepage;
1057         }
1058
1059         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1060
1061                 /* wait for GCed encrypted page writeback */
1062                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1063                                                         fio->blk_addr);
1064
1065                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1066                 if (IS_ERR(fio->encrypted_page)) {
1067                         err = PTR_ERR(fio->encrypted_page);
1068                         goto out_writepage;
1069                 }
1070         }
1071
1072         set_page_writeback(page);
1073
1074         /*
1075          * If current allocation needs SSR,
1076          * it had better in-place writes for updated data.
1077          */
1078         if (unlikely(fio->blk_addr != NEW_ADDR &&
1079                         !is_cold_data(page) &&
1080                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1081                         need_inplace_update(inode))) {
1082                 rewrite_data_page(fio);
1083                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1084                 trace_f2fs_do_write_data_page(page, IPU);
1085         } else {
1086                 write_data_page(&dn, fio);
1087                 set_data_blkaddr(&dn);
1088                 f2fs_update_extent_cache(&dn);
1089                 trace_f2fs_do_write_data_page(page, OPU);
1090                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1091                 if (page->index == 0)
1092                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1093         }
1094 out_writepage:
1095         f2fs_put_dnode(&dn);
1096         return err;
1097 }
1098
1099 static int f2fs_write_data_page(struct page *page,
1100                                         struct writeback_control *wbc)
1101 {
1102         struct inode *inode = page->mapping->host;
1103         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1104         loff_t i_size = i_size_read(inode);
1105         const pgoff_t end_index = ((unsigned long long) i_size)
1106                                                         >> PAGE_CACHE_SHIFT;
1107         unsigned offset = 0;
1108         bool need_balance_fs = false;
1109         int err = 0;
1110         struct f2fs_io_info fio = {
1111                 .sbi = sbi,
1112                 .type = DATA,
1113                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1114                 .page = page,
1115                 .encrypted_page = NULL,
1116         };
1117
1118         trace_f2fs_writepage(page, DATA);
1119
1120         if (page->index < end_index)
1121                 goto write;
1122
1123         /*
1124          * If the offset is out-of-range of file size,
1125          * this page does not have to be written to disk.
1126          */
1127         offset = i_size & (PAGE_CACHE_SIZE - 1);
1128         if ((page->index >= end_index + 1) || !offset)
1129                 goto out;
1130
1131         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1132 write:
1133         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1134                 goto redirty_out;
1135         if (f2fs_is_drop_cache(inode))
1136                 goto out;
1137         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1138                         available_free_memory(sbi, BASE_CHECK))
1139                 goto redirty_out;
1140
1141         /* Dentry blocks are controlled by checkpoint */
1142         if (S_ISDIR(inode->i_mode)) {
1143                 if (unlikely(f2fs_cp_error(sbi)))
1144                         goto redirty_out;
1145                 err = do_write_data_page(&fio);
1146                 goto done;
1147         }
1148
1149         /* we should bypass data pages to proceed the kworkder jobs */
1150         if (unlikely(f2fs_cp_error(sbi))) {
1151                 SetPageError(page);
1152                 goto out;
1153         }
1154
1155         if (!wbc->for_reclaim)
1156                 need_balance_fs = true;
1157         else if (has_not_enough_free_secs(sbi, 0))
1158                 goto redirty_out;
1159
1160         err = -EAGAIN;
1161         f2fs_lock_op(sbi);
1162         if (f2fs_has_inline_data(inode))
1163                 err = f2fs_write_inline_data(inode, page);
1164         if (err == -EAGAIN)
1165                 err = do_write_data_page(&fio);
1166         f2fs_unlock_op(sbi);
1167 done:
1168         if (err && err != -ENOENT)
1169                 goto redirty_out;
1170
1171         clear_cold_data(page);
1172 out:
1173         inode_dec_dirty_pages(inode);
1174         if (err)
1175                 ClearPageUptodate(page);
1176         unlock_page(page);
1177         f2fs_balance_fs(sbi, need_balance_fs);
1178         if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi))) {
1179                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1180                 remove_dirty_inode(inode);
1181         }
1182         return 0;
1183
1184 redirty_out:
1185         redirty_page_for_writepage(wbc, page);
1186         return AOP_WRITEPAGE_ACTIVATE;
1187 }
1188
1189 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1190                         void *data)
1191 {
1192         struct address_space *mapping = data;
1193         int ret = mapping->a_ops->writepage(page, wbc);
1194         mapping_set_error(mapping, ret);
1195         return ret;
1196 }
1197
1198 /*
1199  * This function was copied from write_cche_pages from mm/page-writeback.c.
1200  * The major change is making write step of cold data page separately from
1201  * warm/hot data page.
1202  */
1203 static int f2fs_write_cache_pages(struct address_space *mapping,
1204                         struct writeback_control *wbc, writepage_t writepage,
1205                         void *data)
1206 {
1207         int ret = 0;
1208         int done = 0;
1209         struct pagevec pvec;
1210         int nr_pages;
1211         pgoff_t uninitialized_var(writeback_index);
1212         pgoff_t index;
1213         pgoff_t end;            /* Inclusive */
1214         pgoff_t done_index;
1215         int cycled;
1216         int range_whole = 0;
1217         int tag;
1218         int step = 0;
1219
1220         pagevec_init(&pvec, 0);
1221 next:
1222         if (wbc->range_cyclic) {
1223                 writeback_index = mapping->writeback_index; /* prev offset */
1224                 index = writeback_index;
1225                 if (index == 0)
1226                         cycled = 1;
1227                 else
1228                         cycled = 0;
1229                 end = -1;
1230         } else {
1231                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1232                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1233                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1234                         range_whole = 1;
1235                 cycled = 1; /* ignore range_cyclic tests */
1236         }
1237         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1238                 tag = PAGECACHE_TAG_TOWRITE;
1239         else
1240                 tag = PAGECACHE_TAG_DIRTY;
1241 retry:
1242         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1243                 tag_pages_for_writeback(mapping, index, end);
1244         done_index = index;
1245         while (!done && (index <= end)) {
1246                 int i;
1247
1248                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1249                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1250                 if (nr_pages == 0)
1251                         break;
1252
1253                 for (i = 0; i < nr_pages; i++) {
1254                         struct page *page = pvec.pages[i];
1255
1256                         if (page->index > end) {
1257                                 done = 1;
1258                                 break;
1259                         }
1260
1261                         done_index = page->index;
1262
1263                         lock_page(page);
1264
1265                         if (unlikely(page->mapping != mapping)) {
1266 continue_unlock:
1267                                 unlock_page(page);
1268                                 continue;
1269                         }
1270
1271                         if (!PageDirty(page)) {
1272                                 /* someone wrote it for us */
1273                                 goto continue_unlock;
1274                         }
1275
1276                         if (step == is_cold_data(page))
1277                                 goto continue_unlock;
1278
1279                         if (PageWriteback(page)) {
1280                                 if (wbc->sync_mode != WB_SYNC_NONE)
1281                                         f2fs_wait_on_page_writeback(page, DATA);
1282                                 else
1283                                         goto continue_unlock;
1284                         }
1285
1286                         BUG_ON(PageWriteback(page));
1287                         if (!clear_page_dirty_for_io(page))
1288                                 goto continue_unlock;
1289
1290                         ret = (*writepage)(page, wbc, data);
1291                         if (unlikely(ret)) {
1292                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1293                                         unlock_page(page);
1294                                         ret = 0;
1295                                 } else {
1296                                         done_index = page->index + 1;
1297                                         done = 1;
1298                                         break;
1299                                 }
1300                         }
1301
1302                         if (--wbc->nr_to_write <= 0 &&
1303                             wbc->sync_mode == WB_SYNC_NONE) {
1304                                 done = 1;
1305                                 break;
1306                         }
1307                 }
1308                 pagevec_release(&pvec);
1309                 cond_resched();
1310         }
1311
1312         if (step < 1) {
1313                 step++;
1314                 goto next;
1315         }
1316
1317         if (!cycled && !done) {
1318                 cycled = 1;
1319                 index = 0;
1320                 end = writeback_index - 1;
1321                 goto retry;
1322         }
1323         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1324                 mapping->writeback_index = done_index;
1325
1326         return ret;
1327 }
1328
1329 static int f2fs_write_data_pages(struct address_space *mapping,
1330                             struct writeback_control *wbc)
1331 {
1332         struct inode *inode = mapping->host;
1333         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1334         bool locked = false;
1335         int ret;
1336         long diff;
1337
1338         trace_f2fs_writepages(mapping->host, wbc, DATA);
1339
1340         /* deal with chardevs and other special file */
1341         if (!mapping->a_ops->writepage)
1342                 return 0;
1343
1344         /* skip writing if there is no dirty page in this inode */
1345         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1346                 return 0;
1347
1348         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1349                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1350                         available_free_memory(sbi, DIRTY_DENTS))
1351                 goto skip_write;
1352
1353         /* skip writing during file defragment */
1354         if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1355                 goto skip_write;
1356
1357         /* during POR, we don't need to trigger writepage at all. */
1358         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1359                 goto skip_write;
1360
1361         diff = nr_pages_to_write(sbi, DATA, wbc);
1362
1363         if (!S_ISDIR(inode->i_mode)) {
1364                 mutex_lock(&sbi->writepages);
1365                 locked = true;
1366         }
1367         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1368         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1369         if (locked)
1370                 mutex_unlock(&sbi->writepages);
1371
1372         remove_dirty_inode(inode);
1373
1374         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1375         return ret;
1376
1377 skip_write:
1378         wbc->pages_skipped += get_dirty_pages(inode);
1379         return 0;
1380 }
1381
1382 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1383 {
1384         struct inode *inode = mapping->host;
1385         loff_t i_size = i_size_read(inode);
1386
1387         if (to > i_size) {
1388                 truncate_pagecache(inode, i_size);
1389                 truncate_blocks(inode, i_size, true);
1390         }
1391 }
1392
1393 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1394                         struct page *page, loff_t pos, unsigned len,
1395                         block_t *blk_addr, bool *node_changed)
1396 {
1397         struct inode *inode = page->mapping->host;
1398         pgoff_t index = page->index;
1399         struct dnode_of_data dn;
1400         struct page *ipage;
1401         bool locked = false;
1402         struct extent_info ei;
1403         int err = 0;
1404
1405         if (f2fs_has_inline_data(inode) ||
1406                         (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1407                 f2fs_lock_op(sbi);
1408                 locked = true;
1409         }
1410 restart:
1411         /* check inline_data */
1412         ipage = get_node_page(sbi, inode->i_ino);
1413         if (IS_ERR(ipage)) {
1414                 err = PTR_ERR(ipage);
1415                 goto unlock_out;
1416         }
1417
1418         set_new_dnode(&dn, inode, ipage, ipage, 0);
1419
1420         if (f2fs_has_inline_data(inode)) {
1421                 if (pos + len <= MAX_INLINE_DATA) {
1422                         read_inline_data(page, ipage);
1423                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1424                         sync_inode_page(&dn);
1425                 } else {
1426                         err = f2fs_convert_inline_page(&dn, page);
1427                         if (err)
1428                                 goto out;
1429                         if (dn.data_blkaddr == NULL_ADDR)
1430                                 err = f2fs_get_block(&dn, index);
1431                 }
1432         } else if (locked) {
1433                 err = f2fs_get_block(&dn, index);
1434         } else {
1435                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1436                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1437                 } else {
1438                         bool restart = false;
1439
1440                         /* hole case */
1441                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1442                         if (err || (!err && dn.data_blkaddr == NULL_ADDR))
1443                                 restart = true;
1444                         if (restart) {
1445                                 f2fs_put_dnode(&dn);
1446                                 f2fs_lock_op(sbi);
1447                                 locked = true;
1448                                 goto restart;
1449                         }
1450                 }
1451         }
1452
1453         /* convert_inline_page can make node_changed */
1454         *blk_addr = dn.data_blkaddr;
1455         *node_changed = dn.node_changed;
1456 out:
1457         f2fs_put_dnode(&dn);
1458 unlock_out:
1459         if (locked)
1460                 f2fs_unlock_op(sbi);
1461         return err;
1462 }
1463
1464 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1465                 loff_t pos, unsigned len, unsigned flags,
1466                 struct page **pagep, void **fsdata)
1467 {
1468         struct inode *inode = mapping->host;
1469         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1470         struct page *page = NULL;
1471         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1472         bool need_balance = false;
1473         block_t blkaddr = NULL_ADDR;
1474         int err = 0;
1475
1476         trace_f2fs_write_begin(inode, pos, len, flags);
1477
1478         /*
1479          * We should check this at this moment to avoid deadlock on inode page
1480          * and #0 page. The locking rule for inline_data conversion should be:
1481          * lock_page(page #0) -> lock_page(inode_page)
1482          */
1483         if (index != 0) {
1484                 err = f2fs_convert_inline_inode(inode);
1485                 if (err)
1486                         goto fail;
1487         }
1488 repeat:
1489         page = grab_cache_page_write_begin(mapping, index, flags);
1490         if (!page) {
1491                 err = -ENOMEM;
1492                 goto fail;
1493         }
1494
1495         *pagep = page;
1496
1497         err = prepare_write_begin(sbi, page, pos, len,
1498                                         &blkaddr, &need_balance);
1499         if (err)
1500                 goto fail;
1501
1502         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1503                 unlock_page(page);
1504                 f2fs_balance_fs(sbi, true);
1505                 lock_page(page);
1506                 if (page->mapping != mapping) {
1507                         /* The page got truncated from under us */
1508                         f2fs_put_page(page, 1);
1509                         goto repeat;
1510                 }
1511         }
1512
1513         f2fs_wait_on_page_writeback(page, DATA);
1514
1515         /* wait for GCed encrypted page writeback */
1516         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1517                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1518
1519         if (len == PAGE_CACHE_SIZE)
1520                 goto out_update;
1521         if (PageUptodate(page))
1522                 goto out_clear;
1523
1524         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1525                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1526                 unsigned end = start + len;
1527
1528                 /* Reading beyond i_size is simple: memset to zero */
1529                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1530                 goto out_update;
1531         }
1532
1533         if (blkaddr == NEW_ADDR) {
1534                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1535         } else {
1536                 struct f2fs_io_info fio = {
1537                         .sbi = sbi,
1538                         .type = DATA,
1539                         .rw = READ_SYNC,
1540                         .blk_addr = blkaddr,
1541                         .page = page,
1542                         .encrypted_page = NULL,
1543                 };
1544                 err = f2fs_submit_page_bio(&fio);
1545                 if (err)
1546                         goto fail;
1547
1548                 lock_page(page);
1549                 if (unlikely(!PageUptodate(page))) {
1550                         err = -EIO;
1551                         goto fail;
1552                 }
1553                 if (unlikely(page->mapping != mapping)) {
1554                         f2fs_put_page(page, 1);
1555                         goto repeat;
1556                 }
1557
1558                 /* avoid symlink page */
1559                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1560                         err = f2fs_decrypt_one(inode, page);
1561                         if (err)
1562                                 goto fail;
1563                 }
1564         }
1565 out_update:
1566         SetPageUptodate(page);
1567 out_clear:
1568         clear_cold_data(page);
1569         return 0;
1570
1571 fail:
1572         f2fs_put_page(page, 1);
1573         f2fs_write_failed(mapping, pos + len);
1574         return err;
1575 }
1576
1577 static int f2fs_write_end(struct file *file,
1578                         struct address_space *mapping,
1579                         loff_t pos, unsigned len, unsigned copied,
1580                         struct page *page, void *fsdata)
1581 {
1582         struct inode *inode = page->mapping->host;
1583
1584         trace_f2fs_write_end(inode, pos, len, copied);
1585
1586         set_page_dirty(page);
1587
1588         if (pos + copied > i_size_read(inode)) {
1589                 i_size_write(inode, pos + copied);
1590                 mark_inode_dirty(inode);
1591                 update_inode_page(inode);
1592         }
1593
1594         f2fs_put_page(page, 1);
1595         return copied;
1596 }
1597
1598 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1599                            loff_t offset)
1600 {
1601         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1602
1603         if (offset & blocksize_mask)
1604                 return -EINVAL;
1605
1606         if (iov_iter_alignment(iter) & blocksize_mask)
1607                 return -EINVAL;
1608
1609         return 0;
1610 }
1611
1612 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1613                               loff_t offset)
1614 {
1615         struct file *file = iocb->ki_filp;
1616         struct address_space *mapping = file->f_mapping;
1617         struct inode *inode = mapping->host;
1618         size_t count = iov_iter_count(iter);
1619         int err;
1620
1621         /* we don't need to use inline_data strictly */
1622         err = f2fs_convert_inline_inode(inode);
1623         if (err)
1624                 return err;
1625
1626         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1627                 return 0;
1628
1629         err = check_direct_IO(inode, iter, offset);
1630         if (err)
1631                 return err;
1632
1633         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1634
1635         if (iov_iter_rw(iter) == WRITE) {
1636                 err = __allocate_data_blocks(inode, offset, count);
1637                 if (err)
1638                         goto out;
1639         }
1640
1641         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1642 out:
1643         if (err < 0 && iov_iter_rw(iter) == WRITE)
1644                 f2fs_write_failed(mapping, offset + count);
1645
1646         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1647
1648         return err;
1649 }
1650
1651 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1652                                                         unsigned int length)
1653 {
1654         struct inode *inode = page->mapping->host;
1655         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1656
1657         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1658                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1659                 return;
1660
1661         if (PageDirty(page)) {
1662                 if (inode->i_ino == F2FS_META_INO(sbi))
1663                         dec_page_count(sbi, F2FS_DIRTY_META);
1664                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1665                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1666                 else
1667                         inode_dec_dirty_pages(inode);
1668         }
1669
1670         /* This is atomic written page, keep Private */
1671         if (IS_ATOMIC_WRITTEN_PAGE(page))
1672                 return;
1673
1674         ClearPagePrivate(page);
1675 }
1676
1677 int f2fs_release_page(struct page *page, gfp_t wait)
1678 {
1679         /* If this is dirty page, keep PagePrivate */
1680         if (PageDirty(page))
1681                 return 0;
1682
1683         /* This is atomic written page, keep Private */
1684         if (IS_ATOMIC_WRITTEN_PAGE(page))
1685                 return 0;
1686
1687         ClearPagePrivate(page);
1688         return 1;
1689 }
1690
1691 static int f2fs_set_data_page_dirty(struct page *page)
1692 {
1693         struct address_space *mapping = page->mapping;
1694         struct inode *inode = mapping->host;
1695
1696         trace_f2fs_set_page_dirty(page, DATA);
1697
1698         SetPageUptodate(page);
1699
1700         if (f2fs_is_atomic_file(inode)) {
1701                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1702                         register_inmem_page(inode, page);
1703                         return 1;
1704                 }
1705                 /*
1706                  * Previously, this page has been registered, we just
1707                  * return here.
1708                  */
1709                 return 0;
1710         }
1711
1712         if (!PageDirty(page)) {
1713                 __set_page_dirty_nobuffers(page);
1714                 update_dirty_page(inode, page);
1715                 return 1;
1716         }
1717         return 0;
1718 }
1719
1720 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1721 {
1722         struct inode *inode = mapping->host;
1723
1724         if (f2fs_has_inline_data(inode))
1725                 return 0;
1726
1727         /* make sure allocating whole blocks */
1728         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1729                 filemap_write_and_wait(mapping);
1730
1731         return generic_block_bmap(mapping, block, get_data_block_bmap);
1732 }
1733
1734 const struct address_space_operations f2fs_dblock_aops = {
1735         .readpage       = f2fs_read_data_page,
1736         .readpages      = f2fs_read_data_pages,
1737         .writepage      = f2fs_write_data_page,
1738         .writepages     = f2fs_write_data_pages,
1739         .write_begin    = f2fs_write_begin,
1740         .write_end      = f2fs_write_end,
1741         .set_page_dirty = f2fs_set_data_page_dirty,
1742         .invalidatepage = f2fs_invalidate_page,
1743         .releasepage    = f2fs_release_page,
1744         .direct_IO      = f2fs_direct_IO,
1745         .bmap           = f2fs_bmap,
1746 };