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         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->blk_addr,
 151                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 152                 return -EFAULT;
 153
 154         trace_f2fs_submit_page_bio(page, fio);
 155         f2fs_trace_ios(fio, 0);
 156
 157         /* Allocate a new bio */
 158         bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
 159
 160         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 161                 bio_put(bio);
 162                 return -EFAULT;
 163         }
 164
 165         submit_bio(fio->rw, bio);
 166         return 0;
 167 }
 168
 169 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 170 {
 171         struct f2fs_sb_info *sbi = fio->sbi;
 172         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 173         struct f2fs_bio_info *io;
 174         bool is_read = is_read_io(fio->rw);
 175         struct page *bio_page;
 176
 177         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 178
 179         verify_block_addr(fio, fio->blk_addr);
 180
 181         down_write(&io->io_rwsem);
 182
 183         if (!is_read)
 184                 inc_page_count(sbi, F2FS_WRITEBACK);
 185
 186         if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
 187                                                 io->fio.rw != fio->rw))
 188                 __submit_merged_bio(io);
 189 alloc_new:
 190         if (io->bio == NULL) {
 191                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 192
 193                 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
 194                 io->fio = *fio;
 195         }
 196
 197         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 198
 199         if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
 200                                                         PAGE_CACHE_SIZE) {
 201                 __submit_merged_bio(io);
 202                 goto alloc_new;
 203         }
 204
 205         io->last_block_in_bio = fio->blk_addr;
 206         f2fs_trace_ios(fio, 0);
 207
 208         up_write(&io->io_rwsem);
 209         trace_f2fs_submit_page_mbio(fio->page, fio);
 210 }
 211
 212 /*
 213  * Lock ordering for the change of data block address:
 214  * ->data_page
 215  *  ->node_page
 216  *    update block addresses in the node page
 217  */
 218 void set_data_blkaddr(struct dnode_of_data *dn)
 219 {
 220         struct f2fs_node *rn;
 221         __le32 *addr_array;
 222         struct page *node_page = dn->node_page;
 223         unsigned int ofs_in_node = dn->ofs_in_node;
 224
 225         f2fs_wait_on_page_writeback(node_page, NODE);
 226
 227         rn = F2FS_NODE(node_page);
 228
 229         /* Get physical address of data block */
 230         addr_array = blkaddr_in_node(rn);
 231         addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 232         set_page_dirty(node_page);
 233 }
 234
 235 int reserve_new_block(struct dnode_of_data *dn)
 236 {
 237         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 238
 239         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 240                 return -EPERM;
 241         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 242                 return -ENOSPC;
 243
 244         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 245
 246         dn->data_blkaddr = NEW_ADDR;
 247         set_data_blkaddr(dn);
 248         mark_inode_dirty(dn->inode);
 249         sync_inode_page(dn);
 250         return 0;
 251 }
 252
 253 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 254 {
 255         bool need_put = dn->inode_page ? false : true;
 256         int err;
 257
 258         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 259         if (err)
 260                 return err;
 261
 262         if (dn->data_blkaddr == NULL_ADDR)
 263                 err = reserve_new_block(dn);
 264         if (err || need_put)
 265                 f2fs_put_dnode(dn);
 266         return err;
 267 }
 268
 269 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 270 {
 271         struct extent_info ei;
 272         struct inode *inode = dn->inode;
 273
 274         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 275                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 276                 return 0;
 277         }
 278
 279         return f2fs_reserve_block(dn, index);
 280 }
 281
 282 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 283                                                 int rw, bool for_write)
 284 {
 285         struct address_space *mapping = inode->i_mapping;
 286         struct dnode_of_data dn;
 287         struct page *page;
 288         struct extent_info ei;
 289         int err;
 290         struct f2fs_io_info fio = {
 291                 .sbi = F2FS_I_SB(inode),
 292                 .type = DATA,
 293                 .rw = rw,
 294                 .encrypted_page = NULL,
 295         };
 296
 297         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 298                 return read_mapping_page(mapping, index, NULL);
 299
 300         page = f2fs_grab_cache_page(mapping, index, for_write);
 301         if (!page)
 302                 return ERR_PTR(-ENOMEM);
 303
 304         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 305                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 306                 goto got_it;
 307         }
 308
 309         set_new_dnode(&dn, inode, NULL, NULL, 0);
 310         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 311         if (err)
 312                 goto put_err;
 313         f2fs_put_dnode(&dn);
 314
 315         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 316                 err = -ENOENT;
 317                 goto put_err;
 318         }
 319 got_it:
 320         if (PageUptodate(page)) {
 321                 unlock_page(page);
 322                 return page;
 323         }
 324
 325         /*
 326          * A new dentry page is allocated but not able to be written, since its
 327          * new inode page couldn't be allocated due to -ENOSPC.
 328          * In such the case, its blkaddr can be remained as NEW_ADDR.
 329          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 330          */
 331         if (dn.data_blkaddr == NEW_ADDR) {
 332                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 333                 SetPageUptodate(page);
 334                 unlock_page(page);
 335                 return page;
 336         }
 337
 338         fio.blk_addr = dn.data_blkaddr;
 339         fio.page = page;
 340         err = f2fs_submit_page_bio(&fio);
 341         if (err)
 342                 goto put_err;
 343         return page;
 344
 345 put_err:
 346         f2fs_put_page(page, 1);
 347         return ERR_PTR(err);
 348 }
 349
 350 struct page *find_data_page(struct inode *inode, pgoff_t index)
 351 {
 352         struct address_space *mapping = inode->i_mapping;
 353         struct page *page;
 354
 355         page = find_get_page(mapping, index);
 356         if (page && PageUptodate(page))
 357                 return page;
 358         f2fs_put_page(page, 0);
 359
 360         page = get_read_data_page(inode, index, READ_SYNC, false);
 361         if (IS_ERR(page))
 362                 return page;
 363
 364         if (PageUptodate(page))
 365                 return page;
 366
 367         wait_on_page_locked(page);
 368         if (unlikely(!PageUptodate(page))) {
 369                 f2fs_put_page(page, 0);
 370                 return ERR_PTR(-EIO);
 371         }
 372         return page;
 373 }
 374
 375 /*
 376  * If it tries to access a hole, return an error.
 377  * Because, the callers, functions in dir.c and GC, should be able to know
 378  * whether this page exists or not.
 379  */
 380 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 381                                                         bool for_write)
 382 {
 383         struct address_space *mapping = inode->i_mapping;
 384         struct page *page;
 385 repeat:
 386         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 387         if (IS_ERR(page))
 388                 return page;
 389
 390         /* wait for read completion */
 391         lock_page(page);
 392         if (unlikely(!PageUptodate(page))) {
 393                 f2fs_put_page(page, 1);
 394                 return ERR_PTR(-EIO);
 395         }
 396         if (unlikely(page->mapping != mapping)) {
 397                 f2fs_put_page(page, 1);
 398                 goto repeat;
 399         }
 400         return page;
 401 }
 402
 403 /*
 404  * Caller ensures that this data page is never allocated.
 405  * A new zero-filled data page is allocated in the page cache.
 406  *
 407  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 408  * f2fs_unlock_op().
 409  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 410  * ipage should be released by this function.
 411  */
 412 struct page *get_new_data_page(struct inode *inode,
 413                 struct page *ipage, pgoff_t index, bool new_i_size)
 414 {
 415         struct address_space *mapping = inode->i_mapping;
 416         struct page *page;
 417         struct dnode_of_data dn;
 418         int err;
 419 repeat:
 420         page = f2fs_grab_cache_page(mapping, index, true);
 421         if (!page) {
 422                 /*
 423                  * before exiting, we should make sure ipage will be released
 424                  * if any error occur.
 425                  */
 426                 f2fs_put_page(ipage, 1);
 427                 return ERR_PTR(-ENOMEM);
 428         }
 429
 430         set_new_dnode(&dn, inode, ipage, NULL, 0);
 431         err = f2fs_reserve_block(&dn, index);
 432         if (err) {
 433                 f2fs_put_page(page, 1);
 434                 return ERR_PTR(err);
 435         }
 436         if (!ipage)
 437                 f2fs_put_dnode(&dn);
 438
 439         if (PageUptodate(page))
 440                 goto got_it;
 441
 442         if (dn.data_blkaddr == NEW_ADDR) {
 443                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 444                 SetPageUptodate(page);
 445         } else {
 446                 f2fs_put_page(page, 1);
 447
 448                 page = get_read_data_page(inode, index, READ_SYNC, true);
 449                 if (IS_ERR(page))
 450                         goto repeat;
 451
 452                 /* wait for read completion */
 453                 lock_page(page);
 454         }
 455 got_it:
 456         if (new_i_size && i_size_read(inode) <
 457                                 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
 458                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
 459                 /* Only the directory inode sets new_i_size */
 460                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 461         }
 462         return page;
 463 }
 464
 465 static int __allocate_data_block(struct dnode_of_data *dn)
 466 {
 467         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 468         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 469         struct f2fs_summary sum;
 470         struct node_info ni;
 471         int seg = CURSEG_WARM_DATA;
 472         pgoff_t fofs;
 473
 474         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 475                 return -EPERM;
 476
 477         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 478         if (dn->data_blkaddr == NEW_ADDR)
 479                 goto alloc;
 480
 481         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 482                 return -ENOSPC;
 483
 484 alloc:
 485         get_node_info(sbi, dn->nid, &ni);
 486         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 487
 488         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 489                 seg = CURSEG_DIRECT_IO;
 490
 491         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 492                                                                 &sum, seg);
 493         set_data_blkaddr(dn);
 494
 495         /* update i_size */
 496         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 497                                                         dn->ofs_in_node;
 498         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
 499                 i_size_write(dn->inode,
 500                                 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
 501
 502         /* direct IO doesn't use extent cache to maximize the performance */
 503         f2fs_drop_largest_extent(dn->inode, fofs);
 504
 505         return 0;
 506 }
 507
 508 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
 509                                                         size_t count)
 510 {
 511         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 512         struct dnode_of_data dn;
 513         u64 start = F2FS_BYTES_TO_BLK(offset);
 514         u64 len = F2FS_BYTES_TO_BLK(count);
 515         bool allocated;
 516         u64 end_offset;
 517
 518         while (len) {
 519                 f2fs_balance_fs(sbi);
 520                 f2fs_lock_op(sbi);
 521
 522                 /* When reading holes, we need its node page */
 523                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 524                 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
 525                         goto out;
 526
 527                 allocated = false;
 528                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 529
 530                 while (dn.ofs_in_node < end_offset && len) {
 531                         block_t blkaddr;
 532
 533                         if (unlikely(f2fs_cp_error(sbi)))
 534                                 goto sync_out;
 535
 536                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 537                         if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
 538                                 if (__allocate_data_block(&dn))
 539                                         goto sync_out;
 540                                 allocated = true;
 541                         }
 542                         len--;
 543                         start++;
 544                         dn.ofs_in_node++;
 545                 }
 546
 547                 if (allocated)
 548                         sync_inode_page(&dn);
 549
 550                 f2fs_put_dnode(&dn);
 551                 f2fs_unlock_op(sbi);
 552         }
 553         return;
 554
 555 sync_out:
 556         if (allocated)
 557                 sync_inode_page(&dn);
 558         f2fs_put_dnode(&dn);
 559 out:
 560         f2fs_unlock_op(sbi);
 561         return;
 562 }
 563
 564 /*
 565  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 566  * f2fs_map_blocks structure.
 567  * If original data blocks are allocated, then give them to blockdev.
 568  * Otherwise,
 569  *     a. preallocate requested block addresses
 570  *     b. do not use extent cache for better performance
 571  *     c. give the block addresses to blockdev
 572  */
 573 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 574                                                 int create, int flag)
 575 {
 576         unsigned int maxblocks = map->m_len;
 577         struct dnode_of_data dn;
 578         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 579         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 580         pgoff_t pgofs, end_offset;
 581         int err = 0, ofs = 1;
 582         struct extent_info ei;
 583         bool allocated = false;
 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(F2FS_I_SB(inode));
 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 (__is_valid_data_blkaddr(dn.data_blkaddr) &&
 611                 !f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, DATA_GENERIC)) {
 612                 err = -EFAULT;
 613                 goto sync_out;
 614         }
 615
 616         if (!is_valid_data_blkaddr(sbi, dn.data_blkaddr)) {
 617                 if (create) {
 618                         if (unlikely(f2fs_cp_error(sbi))) {
 619                                 err = -EIO;
 620                                 goto put_out;
 621                         }
 622                         err = __allocate_data_block(&dn);
 623                         if (err)
 624                                 goto put_out;
 625                         allocated = true;
 626                         map->m_flags = F2FS_MAP_NEW;
 627                 } else {
 628                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 629                                                 dn.data_blkaddr != NEW_ADDR) {
 630                                 if (flag == F2FS_GET_BLOCK_BMAP)
 631                                         err = -ENOENT;
 632                                 goto put_out;
 633                         }
 634
 635                         /*
 636                          * preallocated unwritten block should be mapped
 637                          * for fiemap.
 638                          */
 639                         if (dn.data_blkaddr == NEW_ADDR)
 640                                 map->m_flags = F2FS_MAP_UNWRITTEN;
 641                 }
 642         }
 643
 644         map->m_flags |= F2FS_MAP_MAPPED;
 645         map->m_pblk = dn.data_blkaddr;
 646         map->m_len = 1;
 647
 648         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 649         dn.ofs_in_node++;
 650         pgofs++;
 651
 652 get_next:
 653         if (dn.ofs_in_node >= end_offset) {
 654                 if (allocated)
 655                         sync_inode_page(&dn);
 656                 allocated = false;
 657                 f2fs_put_dnode(&dn);
 658
 659                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 660                 err = get_dnode_of_data(&dn, pgofs, mode);
 661                 if (err) {
 662                         if (err == -ENOENT)
 663                                 err = 0;
 664                         goto unlock_out;
 665                 }
 666
 667                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 668         }
 669
 670         if (maxblocks > map->m_len) {
 671                 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 672
 673                 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 674                         if (create) {
 675                                 if (unlikely(f2fs_cp_error(sbi))) {
 676                                         err = -EIO;
 677                                         goto sync_out;
 678                                 }
 679                                 err = __allocate_data_block(&dn);
 680                                 if (err)
 681                                         goto sync_out;
 682                                 allocated = true;
 683                                 map->m_flags |= F2FS_MAP_NEW;
 684                                 blkaddr = dn.data_blkaddr;
 685                         } else {
 686                                 /*
 687                                  * we only merge preallocated unwritten blocks
 688                                  * for fiemap.
 689                                  */
 690                                 if (flag != F2FS_GET_BLOCK_FIEMAP ||
 691                                                 blkaddr != NEW_ADDR)
 692                                         goto sync_out;
 693                         }
 694                 }
 695
 696                 /* Give more consecutive addresses for the readahead */
 697                 if ((map->m_pblk != NEW_ADDR &&
 698                                 blkaddr == (map->m_pblk + ofs)) ||
 699                                 (map->m_pblk == NEW_ADDR &&
 700                                 blkaddr == NEW_ADDR)) {
 701                         ofs++;
 702                         dn.ofs_in_node++;
 703                         pgofs++;
 704                         map->m_len++;
 705                         goto get_next;
 706                 }
 707         }
 708 sync_out:
 709         if (allocated)
 710                 sync_inode_page(&dn);
 711 put_out:
 712         f2fs_put_dnode(&dn);
 713 unlock_out:
 714         if (create)
 715                 f2fs_unlock_op(F2FS_I_SB(inode));
 716 out:
 717         trace_f2fs_map_blocks(inode, map, err);
 718         return err;
 719 }
 720
 721 static int __get_data_block(struct inode *inode, sector_t iblock,
 722                         struct buffer_head *bh, int create, int flag)
 723 {
 724         struct f2fs_map_blocks map;
 725         int ret;
 726
 727         map.m_lblk = iblock;
 728         map.m_len = bh->b_size >> inode->i_blkbits;
 729
 730         ret = f2fs_map_blocks(inode, &map, create, flag);
 731         if (!ret) {
 732                 map_bh(bh, inode->i_sb, map.m_pblk);
 733                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 734                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
 735         }
 736         return ret;
 737 }
 738
 739 static int get_data_block(struct inode *inode, sector_t iblock,
 740                         struct buffer_head *bh_result, int create, int flag)
 741 {
 742         return __get_data_block(inode, iblock, bh_result, create, flag);
 743 }
 744
 745 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 746                         struct buffer_head *bh_result, int create)
 747 {
 748         return __get_data_block(inode, iblock, bh_result, create,
 749                                                 F2FS_GET_BLOCK_DIO);
 750 }
 751
 752 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 753                         struct buffer_head *bh_result, int create)
 754 {
 755         return __get_data_block(inode, iblock, bh_result, create,
 756                                                 F2FS_GET_BLOCK_BMAP);
 757 }
 758
 759 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 760 {
 761         return (offset >> inode->i_blkbits);
 762 }
 763
 764 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 765 {
 766         return (blk << inode->i_blkbits);
 767 }
 768
 769 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 770                 u64 start, u64 len)
 771 {
 772         struct buffer_head map_bh;
 773         sector_t start_blk, last_blk;
 774         loff_t isize = i_size_read(inode);
 775         u64 logical = 0, phys = 0, size = 0;
 776         u32 flags = 0;
 777         bool past_eof = false, whole_file = false;
 778         int ret = 0;
 779
 780         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 781         if (ret)
 782                 return ret;
 783
 784         if (f2fs_has_inline_data(inode)) {
 785                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 786                 if (ret != -EAGAIN)
 787                         return ret;
 788         }
 789
 790         mutex_lock(&inode->i_mutex);
 791
 792         if (len >= isize) {
 793                 whole_file = true;
 794                 len = isize;
 795         }
 796
 797         if (logical_to_blk(inode, len) == 0)
 798                 len = blk_to_logical(inode, 1);
 799
 800         start_blk = logical_to_blk(inode, start);
 801         last_blk = logical_to_blk(inode, start + len - 1);
 802 next:
 803         memset(&map_bh, 0, sizeof(struct buffer_head));
 804         map_bh.b_size = len;
 805
 806         ret = get_data_block(inode, start_blk, &map_bh, 0,
 807                                         F2FS_GET_BLOCK_FIEMAP);
 808         if (ret)
 809                 goto out;
 810
 811         /* HOLE */
 812         if (!buffer_mapped(&map_bh)) {
 813                 start_blk++;
 814
 815                 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
 816                         past_eof = 1;
 817
 818                 if (past_eof && size) {
 819                         flags |= FIEMAP_EXTENT_LAST;
 820                         ret = fiemap_fill_next_extent(fieinfo, logical,
 821                                         phys, size, flags);
 822                 } else if (size) {
 823                         ret = fiemap_fill_next_extent(fieinfo, logical,
 824                                         phys, size, flags);
 825                         size = 0;
 826                 }
 827
 828                 /* if we have holes up to/past EOF then we're done */
 829                 if (start_blk > last_blk || past_eof || ret)
 830                         goto out;
 831         } else {
 832                 if (start_blk > last_blk && !whole_file) {
 833                         ret = fiemap_fill_next_extent(fieinfo, logical,
 834                                         phys, size, flags);
 835                         goto out;
 836                 }
 837
 838                 /*
 839                  * if size != 0 then we know we already have an extent
 840                  * to add, so add it.
 841                  */
 842                 if (size) {
 843                         ret = fiemap_fill_next_extent(fieinfo, logical,
 844                                         phys, size, flags);
 845                         if (ret)
 846                                 goto out;
 847                 }
 848
 849                 logical = blk_to_logical(inode, start_blk);
 850                 phys = blk_to_logical(inode, map_bh.b_blocknr);
 851                 size = map_bh.b_size;
 852                 flags = 0;
 853                 if (buffer_unwritten(&map_bh))
 854                         flags = FIEMAP_EXTENT_UNWRITTEN;
 855
 856                 start_blk += logical_to_blk(inode, size);
 857
 858                 /*
 859                  * If we are past the EOF, then we need to make sure as
 860                  * soon as we find a hole that the last extent we found
 861                  * is marked with FIEMAP_EXTENT_LAST
 862                  */
 863                 if (!past_eof && logical + size >= isize)
 864                         past_eof = true;
 865         }
 866         cond_resched();
 867         if (fatal_signal_pending(current))
 868                 ret = -EINTR;
 869         else
 870                 goto next;
 871 out:
 872         if (ret == 1)
 873                 ret = 0;
 874
 875         mutex_unlock(&inode->i_mutex);
 876         return ret;
 877 }
 878
 879 struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
 880                                                         unsigned nr_pages)
 881 {
 882         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 883         struct f2fs_crypto_ctx *ctx = NULL;
 884         struct block_device *bdev = sbi->sb->s_bdev;
 885         struct bio *bio;
 886
 887         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
 888                 return ERR_PTR(-EFAULT);
 889
 890         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
 891                 ctx = f2fs_get_crypto_ctx(inode);
 892                 if (IS_ERR(ctx))
 893                         return ERR_CAST(ctx);
 894
 895                 /* wait the page to be moved by cleaning */
 896                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
 897         }
 898
 899         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
 900         if (!bio) {
 901                 if (ctx)
 902                         f2fs_release_crypto_ctx(ctx);
 903                 return ERR_PTR(-ENOMEM);
 904         }
 905         bio->bi_bdev = bdev;
 906         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
 907         bio->bi_end_io = f2fs_read_end_io;
 908         bio->bi_private = ctx;
 909
 910         return bio;
 911 }
 912
 913 /*
 914  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 915  * Major change was from block_size == page_size in f2fs by default.
 916  */
 917 static int f2fs_mpage_readpages(struct address_space *mapping,
 918                         struct list_head *pages, struct page *page,
 919                         unsigned nr_pages)
 920 {
 921         struct bio *bio = NULL;
 922         unsigned page_idx;
 923         sector_t last_block_in_bio = 0;
 924         struct inode *inode = mapping->host;
 925         const unsigned blkbits = inode->i_blkbits;
 926         const unsigned blocksize = 1 << blkbits;
 927         sector_t block_in_file;
 928         sector_t last_block;
 929         sector_t last_block_in_file;
 930         sector_t block_nr;
 931         struct f2fs_map_blocks map;
 932
 933         map.m_pblk = 0;
 934         map.m_lblk = 0;
 935         map.m_len = 0;
 936         map.m_flags = 0;
 937
 938         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 939
 940                 prefetchw(&page->flags);
 941                 if (pages) {
 942                         page = list_entry(pages->prev, struct page, lru);
 943                         list_del(&page->lru);
 944                         if (add_to_page_cache_lru(page, mapping,
 945                                                   page->index, GFP_KERNEL))
 946                                 goto next_page;
 947                 }
 948
 949                 block_in_file = (sector_t)page->index;
 950                 last_block = block_in_file + nr_pages;
 951                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 952                                                                 blkbits;
 953                 if (last_block > last_block_in_file)
 954                         last_block = last_block_in_file;
 955
 956                 /*
 957                  * Map blocks using the previous result first.
 958                  */
 959                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 960                                 block_in_file > map.m_lblk &&
 961                                 block_in_file < (map.m_lblk + map.m_len))
 962                         goto got_it;
 963
 964                 /*
 965                  * Then do more f2fs_map_blocks() calls until we are
 966                  * done with this page.
 967                  */
 968                 map.m_flags = 0;
 969
 970                 if (block_in_file < last_block) {
 971                         map.m_lblk = block_in_file;
 972                         map.m_len = last_block - block_in_file;
 973
 974                         if (f2fs_map_blocks(inode, &map, 0,
 975                                                         F2FS_GET_BLOCK_READ))
 976                                 goto set_error_page;
 977                 }
 978 got_it:
 979                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 980                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 981                         SetPageMappedToDisk(page);
 982
 983                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 984                                 SetPageUptodate(page);
 985                                 goto confused;
 986                         }
 987
 988                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
 989                                                                 DATA_GENERIC))
 990                                 goto set_error_page;
 991                 } else {
 992                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 993                         SetPageUptodate(page);
 994                         unlock_page(page);
 995                         goto next_page;
 996                 }
 997
 998                 /*
 999                  * This page will go to BIO.  Do we need to send this
1000                  * BIO off first?
1001                  */
1002                 if (bio && (last_block_in_bio != block_nr - 1)) {
1003 submit_and_realloc:
1004                         submit_bio(READ, bio);
1005                         bio = NULL;
1006                 }
1007                 if (bio == NULL) {
1008                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1009                         if (IS_ERR(bio))
1010                                 goto set_error_page;
1011                 }
1012
1013                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1014                         goto submit_and_realloc;
1015
1016                 last_block_in_bio = block_nr;
1017                 goto next_page;
1018 set_error_page:
1019                 SetPageError(page);
1020                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1021                 unlock_page(page);
1022                 goto next_page;
1023 confused:
1024                 if (bio) {
1025                         submit_bio(READ, bio);
1026                         bio = NULL;
1027                 }
1028                 unlock_page(page);
1029 next_page:
1030                 if (pages)
1031                         page_cache_release(page);
1032         }
1033         BUG_ON(pages && !list_empty(pages));
1034         if (bio)
1035                 submit_bio(READ, bio);
1036         return 0;
1037 }
1038
1039 static int f2fs_read_data_page(struct file *file, struct page *page)
1040 {
1041         struct inode *inode = page->mapping->host;
1042         int ret = -EAGAIN;
1043
1044         trace_f2fs_readpage(page, DATA);
1045
1046         /* If the file has inline data, try to read it directly */
1047         if (f2fs_has_inline_data(inode))
1048                 ret = f2fs_read_inline_data(inode, page);
1049         if (ret == -EAGAIN)
1050                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1051         return ret;
1052 }
1053
1054 static int f2fs_read_data_pages(struct file *file,
1055                         struct address_space *mapping,
1056                         struct list_head *pages, unsigned nr_pages)
1057 {
1058         struct inode *inode = file->f_mapping->host;
1059         struct page *page = list_entry(pages->prev, struct page, lru);
1060
1061         trace_f2fs_readpages(inode, page, nr_pages);
1062
1063         /* If the file has inline data, skip readpages */
1064         if (f2fs_has_inline_data(inode))
1065                 return 0;
1066
1067         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1068 }
1069
1070 int do_write_data_page(struct f2fs_io_info *fio)
1071 {
1072         struct page *page = fio->page;
1073         struct inode *inode = page->mapping->host;
1074         struct dnode_of_data dn;
1075         int err = 0;
1076
1077         set_new_dnode(&dn, inode, NULL, NULL, 0);
1078         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1079         if (err)
1080                 return err;
1081
1082         fio->blk_addr = dn.data_blkaddr;
1083
1084         /* This page is already truncated */
1085         if (fio->blk_addr == NULL_ADDR) {
1086                 ClearPageUptodate(page);
1087                 goto out_writepage;
1088         }
1089
1090         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1091
1092                 /* wait for GCed encrypted page writeback */
1093                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1094                                                         fio->blk_addr);
1095
1096                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1097                 if (IS_ERR(fio->encrypted_page)) {
1098                         err = PTR_ERR(fio->encrypted_page);
1099                         goto out_writepage;
1100                 }
1101         }
1102
1103         set_page_writeback(page);
1104
1105         if (__is_valid_data_blkaddr(fio->blk_addr) &&
1106                 !f2fs_is_valid_blkaddr(fio->sbi, fio->blk_addr,
1107                                                         DATA_GENERIC)) {
1108                 err = -EFAULT;
1109                 goto out_writepage;
1110         }
1111         /*
1112          * If current allocation needs SSR,
1113          * it had better in-place writes for updated data.
1114          */
1115         if (unlikely(is_valid_data_blkaddr(fio->sbi, fio->blk_addr) &&
1116                         !is_cold_data(page) &&
1117                         need_inplace_update(inode))) {
1118                 rewrite_data_page(fio);
1119                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1120                 trace_f2fs_do_write_data_page(page, IPU);
1121         } else {
1122                 write_data_page(&dn, fio);
1123                 set_data_blkaddr(&dn);
1124                 f2fs_update_extent_cache(&dn);
1125                 trace_f2fs_do_write_data_page(page, OPU);
1126                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1127                 if (page->index == 0)
1128                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1129         }
1130 out_writepage:
1131         f2fs_put_dnode(&dn);
1132         return err;
1133 }
1134
1135 static int f2fs_write_data_page(struct page *page,
1136                                         struct writeback_control *wbc)
1137 {
1138         struct inode *inode = page->mapping->host;
1139         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1140         loff_t i_size = i_size_read(inode);
1141         const pgoff_t end_index = ((unsigned long long) i_size)
1142                                                         >> PAGE_CACHE_SHIFT;
1143         unsigned offset = 0;
1144         bool need_balance_fs = false;
1145         int err = 0;
1146         struct f2fs_io_info fio = {
1147                 .sbi = sbi,
1148                 .type = DATA,
1149                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1150                 .page = page,
1151                 .encrypted_page = NULL,
1152         };
1153
1154         trace_f2fs_writepage(page, DATA);
1155
1156         if (page->index < end_index)
1157                 goto write;
1158
1159         /*
1160          * If the offset is out-of-range of file size,
1161          * this page does not have to be written to disk.
1162          */
1163         offset = i_size & (PAGE_CACHE_SIZE - 1);
1164         if ((page->index >= end_index + 1) || !offset)
1165                 goto out;
1166
1167         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1168 write:
1169         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1170                 goto redirty_out;
1171         if (f2fs_is_drop_cache(inode))
1172                 goto out;
1173         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1174                         available_free_memory(sbi, BASE_CHECK))
1175                 goto redirty_out;
1176
1177         /* Dentry blocks are controlled by checkpoint */
1178         if (S_ISDIR(inode->i_mode)) {
1179                 if (unlikely(f2fs_cp_error(sbi)))
1180                         goto redirty_out;
1181                 err = do_write_data_page(&fio);
1182                 goto done;
1183         }
1184
1185         /* we should bypass data pages to proceed the kworkder jobs */
1186         if (unlikely(f2fs_cp_error(sbi))) {
1187                 SetPageError(page);
1188                 goto out;
1189         }
1190
1191         if (!wbc->for_reclaim)
1192                 need_balance_fs = true;
1193         else if (has_not_enough_free_secs(sbi, 0))
1194                 goto redirty_out;
1195
1196         err = -EAGAIN;
1197         f2fs_lock_op(sbi);
1198         if (f2fs_has_inline_data(inode))
1199                 err = f2fs_write_inline_data(inode, page);
1200         if (err == -EAGAIN)
1201                 err = do_write_data_page(&fio);
1202         f2fs_unlock_op(sbi);
1203 done:
1204         if (err && err != -ENOENT)
1205                 goto redirty_out;
1206
1207         clear_cold_data(page);
1208 out:
1209         inode_dec_dirty_pages(inode);
1210         if (err)
1211                 ClearPageUptodate(page);
1212         unlock_page(page);
1213         if (need_balance_fs)
1214                 f2fs_balance_fs(sbi);
1215         if (wbc->for_reclaim)
1216                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1217         return 0;
1218
1219 redirty_out:
1220         redirty_page_for_writepage(wbc, page);
1221         return AOP_WRITEPAGE_ACTIVATE;
1222 }
1223
1224 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1225                         void *data)
1226 {
1227         struct address_space *mapping = data;
1228         int ret = mapping->a_ops->writepage(page, wbc);
1229         mapping_set_error(mapping, ret);
1230         return ret;
1231 }
1232
1233 /*
1234  * This function was copied from write_cche_pages from mm/page-writeback.c.
1235  * The major change is making write step of cold data page separately from
1236  * warm/hot data page.
1237  */
1238 static int f2fs_write_cache_pages(struct address_space *mapping,
1239                         struct writeback_control *wbc, writepage_t writepage,
1240                         void *data)
1241 {
1242         int ret = 0;
1243         int done = 0;
1244         struct pagevec pvec;
1245         int nr_pages;
1246         pgoff_t uninitialized_var(writeback_index);
1247         pgoff_t index;
1248         pgoff_t end;            /* Inclusive */
1249         pgoff_t done_index;
1250         int cycled;
1251         int range_whole = 0;
1252         int tag;
1253         int step = 0;
1254
1255         pagevec_init(&pvec, 0);
1256 next:
1257         if (wbc->range_cyclic) {
1258                 writeback_index = mapping->writeback_index; /* prev offset */
1259                 index = writeback_index;
1260                 if (index == 0)
1261                         cycled = 1;
1262                 else
1263                         cycled = 0;
1264                 end = -1;
1265         } else {
1266                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1267                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1268                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1269                         range_whole = 1;
1270                 cycled = 1; /* ignore range_cyclic tests */
1271         }
1272         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1273                 tag = PAGECACHE_TAG_TOWRITE;
1274         else
1275                 tag = PAGECACHE_TAG_DIRTY;
1276 retry:
1277         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1278                 tag_pages_for_writeback(mapping, index, end);
1279         done_index = index;
1280         while (!done && (index <= end)) {
1281                 int i;
1282
1283                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1284                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1285                 if (nr_pages == 0)
1286                         break;
1287
1288                 for (i = 0; i < nr_pages; i++) {
1289                         struct page *page = pvec.pages[i];
1290
1291                         if (page->index > end) {
1292                                 done = 1;
1293                                 break;
1294                         }
1295
1296                         done_index = page->index;
1297
1298                         lock_page(page);
1299
1300                         if (unlikely(page->mapping != mapping)) {
1301 continue_unlock:
1302                                 unlock_page(page);
1303                                 continue;
1304                         }
1305
1306                         if (!PageDirty(page)) {
1307                                 /* someone wrote it for us */
1308                                 goto continue_unlock;
1309                         }
1310
1311                         if (step == is_cold_data(page))
1312                                 goto continue_unlock;
1313
1314                         if (PageWriteback(page)) {
1315                                 if (wbc->sync_mode != WB_SYNC_NONE)
1316                                         f2fs_wait_on_page_writeback(page, DATA);
1317                                 else
1318                                         goto continue_unlock;
1319                         }
1320
1321                         BUG_ON(PageWriteback(page));
1322                         if (!clear_page_dirty_for_io(page))
1323                                 goto continue_unlock;
1324
1325                         ret = (*writepage)(page, wbc, data);
1326                         if (unlikely(ret)) {
1327                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1328                                         unlock_page(page);
1329                                         ret = 0;
1330                                 } else {
1331                                         done_index = page->index + 1;
1332                                         done = 1;
1333                                         break;
1334                                 }
1335                         }
1336
1337                         if (--wbc->nr_to_write <= 0 &&
1338                             wbc->sync_mode == WB_SYNC_NONE) {
1339                                 done = 1;
1340                                 break;
1341                         }
1342                 }
1343                 pagevec_release(&pvec);
1344                 cond_resched();
1345         }
1346
1347         if (step < 1) {
1348                 step++;
1349                 goto next;
1350         }
1351
1352         if (!cycled && !done) {
1353                 cycled = 1;
1354                 index = 0;
1355                 end = writeback_index - 1;
1356                 goto retry;
1357         }
1358         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1359                 mapping->writeback_index = done_index;
1360
1361         return ret;
1362 }
1363
1364 static int f2fs_write_data_pages(struct address_space *mapping,
1365                             struct writeback_control *wbc)
1366 {
1367         struct inode *inode = mapping->host;
1368         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1369         bool locked = false;
1370         int ret;
1371         long diff;
1372
1373         trace_f2fs_writepages(mapping->host, wbc, DATA);
1374
1375         /* deal with chardevs and other special file */
1376         if (!mapping->a_ops->writepage)
1377                 return 0;
1378
1379         /* skip writing if there is no dirty page in this inode */
1380         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1381                 return 0;
1382
1383         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1384                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1385                         available_free_memory(sbi, DIRTY_DENTS))
1386                 goto skip_write;
1387
1388         /* during POR, we don't need to trigger writepage at all. */
1389         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1390                 goto skip_write;
1391
1392         diff = nr_pages_to_write(sbi, DATA, wbc);
1393
1394         if (!S_ISDIR(inode->i_mode)) {
1395                 mutex_lock(&sbi->writepages);
1396                 locked = true;
1397         }
1398         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1399         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1400         if (locked)
1401                 mutex_unlock(&sbi->writepages);
1402
1403         remove_dirty_dir_inode(inode);
1404
1405         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1406         return ret;
1407
1408 skip_write:
1409         wbc->pages_skipped += get_dirty_pages(inode);
1410         return 0;
1411 }
1412
1413 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1414 {
1415         struct inode *inode = mapping->host;
1416
1417         if (to > inode->i_size) {
1418                 truncate_pagecache(inode, inode->i_size);
1419                 truncate_blocks(inode, inode->i_size, true);
1420         }
1421 }
1422
1423 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1424                 loff_t pos, unsigned len, unsigned flags,
1425                 struct page **pagep, void **fsdata)
1426 {
1427         struct inode *inode = mapping->host;
1428         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1429         struct page *page = NULL;
1430         struct page *ipage;
1431         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1432         struct dnode_of_data dn;
1433         int err = 0;
1434
1435         trace_f2fs_write_begin(inode, pos, len, flags);
1436
1437         f2fs_balance_fs(sbi);
1438
1439         /*
1440          * We should check this at this moment to avoid deadlock on inode page
1441          * and #0 page. The locking rule for inline_data conversion should be:
1442          * lock_page(page #0) -> lock_page(inode_page)
1443          */
1444         if (index != 0) {
1445                 err = f2fs_convert_inline_inode(inode);
1446                 if (err)
1447                         goto fail;
1448         }
1449 repeat:
1450         /*
1451          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1452          * wait_for_stable_page. Will wait that below with our IO control.
1453          */
1454         page = pagecache_get_page(mapping, index,
1455                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1456         if (!page) {
1457                 err = -ENOMEM;
1458                 goto fail;
1459         }
1460
1461         *pagep = page;
1462
1463         f2fs_lock_op(sbi);
1464
1465         /* check inline_data */
1466         ipage = get_node_page(sbi, inode->i_ino);
1467         if (IS_ERR(ipage)) {
1468                 err = PTR_ERR(ipage);
1469                 goto unlock_fail;
1470         }
1471
1472         set_new_dnode(&dn, inode, ipage, ipage, 0);
1473
1474         if (f2fs_has_inline_data(inode)) {
1475                 if (pos + len <= MAX_INLINE_DATA) {
1476                         read_inline_data(page, ipage);
1477                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1478                         sync_inode_page(&dn);
1479                         goto put_next;
1480                 }
1481                 err = f2fs_convert_inline_page(&dn, page);
1482                 if (err)
1483                         goto put_fail;
1484         }
1485
1486         err = f2fs_get_block(&dn, index);
1487         if (err)
1488                 goto put_fail;
1489 put_next:
1490         f2fs_put_dnode(&dn);
1491         f2fs_unlock_op(sbi);
1492
1493         f2fs_wait_on_page_writeback(page, DATA);
1494
1495         /* wait for GCed encrypted page writeback */
1496         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1497                 f2fs_wait_on_encrypted_page_writeback(sbi, dn.data_blkaddr);
1498
1499         if (len == PAGE_CACHE_SIZE)
1500                 goto out_update;
1501         if (PageUptodate(page))
1502                 goto out_clear;
1503
1504         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1505                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1506                 unsigned end = start + len;
1507
1508                 /* Reading beyond i_size is simple: memset to zero */
1509                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1510                 goto out_update;
1511         }
1512
1513         if (dn.data_blkaddr == NEW_ADDR) {
1514                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1515         } else {
1516                 struct bio *bio;
1517
1518                 bio = f2fs_grab_bio(inode, dn.data_blkaddr, 1);
1519                 if (IS_ERR(bio)) {
1520                         err = PTR_ERR(bio);
1521                         goto fail;
1522                 }
1523
1524                 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
1525                         bio_put(bio);
1526                         err = -EFAULT;
1527                         goto fail;
1528                 }
1529
1530                 submit_bio(READ_SYNC, bio);
1531
1532                 lock_page(page);
1533                 if (unlikely(!PageUptodate(page))) {
1534                         err = -EIO;
1535                         goto fail;
1536                 }
1537                 if (unlikely(page->mapping != mapping)) {
1538                         f2fs_put_page(page, 1);
1539                         goto repeat;
1540                 }
1541         }
1542 out_update:
1543         SetPageUptodate(page);
1544 out_clear:
1545         clear_cold_data(page);
1546         return 0;
1547
1548 put_fail:
1549         f2fs_put_dnode(&dn);
1550 unlock_fail:
1551         f2fs_unlock_op(sbi);
1552 fail:
1553         f2fs_put_page(page, 1);
1554         f2fs_write_failed(mapping, pos + len);
1555         return err;
1556 }
1557
1558 static int f2fs_write_end(struct file *file,
1559                         struct address_space *mapping,
1560                         loff_t pos, unsigned len, unsigned copied,
1561                         struct page *page, void *fsdata)
1562 {
1563         struct inode *inode = page->mapping->host;
1564
1565         trace_f2fs_write_end(inode, pos, len, copied);
1566
1567         set_page_dirty(page);
1568
1569         if (pos + copied > i_size_read(inode)) {
1570                 i_size_write(inode, pos + copied);
1571                 mark_inode_dirty(inode);
1572                 update_inode_page(inode);
1573         }
1574
1575         f2fs_put_page(page, 1);
1576         return copied;
1577 }
1578
1579 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1580                            loff_t offset)
1581 {
1582         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1583
1584         if (offset & blocksize_mask)
1585                 return -EINVAL;
1586
1587         if (iov_iter_alignment(iter) & blocksize_mask)
1588                 return -EINVAL;
1589
1590         return 0;
1591 }
1592
1593 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1594                               loff_t offset)
1595 {
1596         struct file *file = iocb->ki_filp;
1597         struct address_space *mapping = file->f_mapping;
1598         struct inode *inode = mapping->host;
1599         size_t count = iov_iter_count(iter);
1600         int err;
1601
1602         /* we don't need to use inline_data strictly */
1603         if (f2fs_has_inline_data(inode)) {
1604                 err = f2fs_convert_inline_inode(inode);
1605                 if (err)
1606                         return err;
1607         }
1608
1609         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1610                 return 0;
1611
1612         err = check_direct_IO(inode, iter, offset);
1613         if (err)
1614                 return err;
1615
1616         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1617
1618         if (iov_iter_rw(iter) == WRITE) {
1619                 __allocate_data_blocks(inode, offset, count);
1620                 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1621                         err = -EIO;
1622                         goto out;
1623                 }
1624         }
1625
1626         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1627 out:
1628         if (err < 0 && iov_iter_rw(iter) == WRITE)
1629                 f2fs_write_failed(mapping, offset + count);
1630
1631         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1632
1633         return err;
1634 }
1635
1636 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1637                                                         unsigned int length)
1638 {
1639         struct inode *inode = page->mapping->host;
1640         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1641
1642         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1643                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1644                 return;
1645
1646         if (PageDirty(page)) {
1647                 if (inode->i_ino == F2FS_META_INO(sbi))
1648                         dec_page_count(sbi, F2FS_DIRTY_META);
1649                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1650                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1651                 else
1652                         inode_dec_dirty_pages(inode);
1653         }
1654
1655         /* This is atomic written page, keep Private */
1656         if (IS_ATOMIC_WRITTEN_PAGE(page))
1657                 return;
1658
1659         ClearPagePrivate(page);
1660 }
1661
1662 int f2fs_release_page(struct page *page, gfp_t wait)
1663 {
1664         /* If this is dirty page, keep PagePrivate */
1665         if (PageDirty(page))
1666                 return 0;
1667
1668         /* This is atomic written page, keep Private */
1669         if (IS_ATOMIC_WRITTEN_PAGE(page))
1670                 return 0;
1671
1672         ClearPagePrivate(page);
1673         return 1;
1674 }
1675
1676 static int f2fs_set_data_page_dirty(struct page *page)
1677 {
1678         struct address_space *mapping = page->mapping;
1679         struct inode *inode = mapping->host;
1680
1681         trace_f2fs_set_page_dirty(page, DATA);
1682
1683         SetPageUptodate(page);
1684
1685         if (f2fs_is_atomic_file(inode)) {
1686                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1687                         register_inmem_page(inode, page);
1688                         return 1;
1689                 }
1690                 /*
1691                  * Previously, this page has been registered, we just
1692                  * return here.
1693                  */
1694                 return 0;
1695         }
1696
1697         if (!PageDirty(page)) {
1698                 __set_page_dirty_nobuffers(page);
1699                 update_dirty_page(inode, page);
1700                 return 1;
1701         }
1702         return 0;
1703 }
1704
1705 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1706 {
1707         struct inode *inode = mapping->host;
1708
1709         if (f2fs_has_inline_data(inode))
1710                 return 0;
1711
1712         /* make sure allocating whole blocks */
1713         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1714                 filemap_write_and_wait(mapping);
1715
1716         return generic_block_bmap(mapping, block, get_data_block_bmap);
1717 }
1718
1719 const struct address_space_operations f2fs_dblock_aops = {
1720         .readpage       = f2fs_read_data_page,
1721         .readpages      = f2fs_read_data_pages,
1722         .writepage      = f2fs_write_data_page,
1723         .writepages     = f2fs_write_data_pages,
1724         .write_begin    = f2fs_write_begin,
1725         .write_end      = f2fs_write_end,
1726         .set_page_dirty = f2fs_set_data_page_dirty,
1727         .invalidatepage = f2fs_invalidate_page,
1728         .releasepage    = f2fs_release_page,
1729         .direct_IO      = f2fs_direct_IO,
1730         .bmap           = f2fs_bmap,
1731 };