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/aio.h>
  16 #include <linux/writeback.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include <trace/events/f2fs.h>
  26
  27 static void f2fs_read_end_io(struct bio *bio, int err)
  28 {
  29         struct bio_vec *bvec;
  30         int i;
  31
  32         bio_for_each_segment_all(bvec, bio, i) {
  33                 struct page *page = bvec->bv_page;
  34
  35                 if (!err) {
  36                         SetPageUptodate(page);
  37                 } else {
  38                         ClearPageUptodate(page);
  39                         SetPageError(page);
  40                 }
  41                 unlock_page(page);
  42         }
  43         bio_put(bio);
  44 }
  45
  46 static void f2fs_write_end_io(struct bio *bio, int err)
  47 {
  48         struct f2fs_sb_info *sbi = bio->bi_private;
  49         struct bio_vec *bvec;
  50         int i;
  51
  52         bio_for_each_segment_all(bvec, bio, i) {
  53                 struct page *page = bvec->bv_page;
  54
  55                 if (unlikely(err)) {
  56                         SetPageError(page);
  57                         set_bit(AS_EIO, &page->mapping->flags);
  58                         f2fs_stop_checkpoint(sbi);
  59                 }
  60                 end_page_writeback(page);
  61                 dec_page_count(sbi, F2FS_WRITEBACK);
  62         }
  63
  64         if (sbi->wait_io) {
  65                 complete(sbi->wait_io);
  66                 sbi->wait_io = NULL;
  67         }
  68
  69         if (!get_pages(sbi, F2FS_WRITEBACK) &&
  70                         !list_empty(&sbi->cp_wait.task_list))
  71                 wake_up(&sbi->cp_wait);
  72
  73         bio_put(bio);
  74 }
  75
  76 /*
  77  * Low-level block read/write IO operations.
  78  */
  79 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  80                                 int npages, bool is_read)
  81 {
  82         struct bio *bio;
  83
  84         /* No failure on bio allocation */
  85         bio = bio_alloc(GFP_NOIO, npages);
  86
  87         bio->bi_bdev = sbi->sb->s_bdev;
  88         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
  89         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  90         bio->bi_private = sbi;
  91
  92         return bio;
  93 }
  94
  95 static void __submit_merged_bio(struct f2fs_bio_info *io)
  96 {
  97         struct f2fs_io_info *fio = &io->fio;
  98         int rw;
  99
 100         if (!io->bio)
 101                 return;
 102
 103         rw = fio->rw;
 104
 105         if (is_read_io(rw)) {
 106                 trace_f2fs_submit_read_bio(io->sbi->sb, rw,
 107                                                 fio->type, io->bio);
 108                 submit_bio(rw, io->bio);
 109         } else {
 110                 trace_f2fs_submit_write_bio(io->sbi->sb, rw,
 111                                                 fio->type, io->bio);
 112                 /*
 113                  * META_FLUSH is only from the checkpoint procedure, and we
 114                  * should wait this metadata bio for FS consistency.
 115                  */
 116                 if (fio->type == META_FLUSH) {
 117                         DECLARE_COMPLETION_ONSTACK(wait);
 118                         io->sbi->wait_io = &wait;
 119                         submit_bio(rw, io->bio);
 120                         wait_for_completion(&wait);
 121                 } else {
 122                         submit_bio(rw, io->bio);
 123                 }
 124         }
 125
 126         io->bio = NULL;
 127 }
 128
 129 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 130                                 enum page_type type, int rw)
 131 {
 132         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 133         struct f2fs_bio_info *io;
 134
 135         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 136
 137         down_write(&io->io_rwsem);
 138
 139         /* change META to META_FLUSH in the checkpoint procedure */
 140         if (type >= META_FLUSH) {
 141                 io->fio.type = META_FLUSH;
 142                 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 143         }
 144         __submit_merged_bio(io);
 145         up_write(&io->io_rwsem);
 146 }
 147
 148 /*
 149  * Fill the locked page with data located in the block address.
 150  * Return unlocked page.
 151  */
 152 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
 153                                         block_t blk_addr, int rw)
 154 {
 155         struct bio *bio;
 156
 157         trace_f2fs_submit_page_bio(page, blk_addr, rw);
 158
 159         /* Allocate a new bio */
 160         bio = __bio_alloc(sbi, blk_addr, 1, is_read_io(rw));
 161
 162         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 163                 bio_put(bio);
 164                 f2fs_put_page(page, 1);
 165                 return -EFAULT;
 166         }
 167
 168         submit_bio(rw, bio);
 169         return 0;
 170 }
 171
 172 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
 173                         block_t blk_addr, struct f2fs_io_info *fio)
 174 {
 175         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 176         struct f2fs_bio_info *io;
 177         bool is_read = is_read_io(fio->rw);
 178
 179         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 180
 181         verify_block_addr(sbi, blk_addr);
 182
 183         down_write(&io->io_rwsem);
 184
 185         if (!is_read)
 186                 inc_page_count(sbi, F2FS_WRITEBACK);
 187
 188         if (io->bio && (io->last_block_in_bio != blk_addr - 1 ||
 189                                                 io->fio.rw != fio->rw))
 190                 __submit_merged_bio(io);
 191 alloc_new:
 192         if (io->bio == NULL) {
 193                 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
 194
 195                 io->bio = __bio_alloc(sbi, blk_addr, bio_blocks, is_read);
 196                 io->fio = *fio;
 197         }
 198
 199         if (bio_add_page(io->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 = blk_addr;
 206
 207         up_write(&io->io_rwsem);
 208         trace_f2fs_submit_page_mbio(page, fio->rw, fio->type, blk_addr);
 209 }
 210
 211 /*
 212  * Lock ordering for the change of data block address:
 213  * ->data_page
 214  *  ->node_page
 215  *    update block addresses in the node page
 216  */
 217 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
 218 {
 219         struct f2fs_node *rn;
 220         __le32 *addr_array;
 221         struct page *node_page = dn->node_page;
 222         unsigned int ofs_in_node = dn->ofs_in_node;
 223
 224         f2fs_wait_on_page_writeback(node_page, NODE);
 225
 226         rn = F2FS_NODE(node_page);
 227
 228         /* Get physical address of data block */
 229         addr_array = blkaddr_in_node(rn);
 230         addr_array[ofs_in_node] = cpu_to_le32(new_addr);
 231         set_page_dirty(node_page);
 232 }
 233
 234 int reserve_new_block(struct dnode_of_data *dn)
 235 {
 236         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 237
 238         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 239                 return -EPERM;
 240         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 241                 return -ENOSPC;
 242
 243         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 244
 245         __set_data_blkaddr(dn, NEW_ADDR);
 246         dn->data_blkaddr = NEW_ADDR;
 247         mark_inode_dirty(dn->inode);
 248         sync_inode_page(dn);
 249         return 0;
 250 }
 251
 252 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 253 {
 254         bool need_put = dn->inode_page ? false : true;
 255         int err;
 256
 257         /* if inode_page exists, index should be zero */
 258         f2fs_bug_on(!need_put && index);
 259
 260         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 261         if (err)
 262                 return err;
 263
 264         if (dn->data_blkaddr == NULL_ADDR)
 265                 err = reserve_new_block(dn);
 266         if (err || need_put)
 267                 f2fs_put_dnode(dn);
 268         return err;
 269 }
 270
 271 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
 272                                         struct buffer_head *bh_result)
 273 {
 274         struct f2fs_inode_info *fi = F2FS_I(inode);
 275         pgoff_t start_fofs, end_fofs;
 276         block_t start_blkaddr;
 277
 278         if (is_inode_flag_set(fi, FI_NO_EXTENT))
 279                 return 0;
 280
 281         read_lock(&fi->ext.ext_lock);
 282         if (fi->ext.len == 0) {
 283                 read_unlock(&fi->ext.ext_lock);
 284                 return 0;
 285         }
 286
 287         stat_inc_total_hit(inode->i_sb);
 288
 289         start_fofs = fi->ext.fofs;
 290         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 291         start_blkaddr = fi->ext.blk_addr;
 292
 293         if (pgofs >= start_fofs && pgofs <= end_fofs) {
 294                 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 295                 size_t count;
 296
 297                 clear_buffer_new(bh_result);
 298                 map_bh(bh_result, inode->i_sb,
 299                                 start_blkaddr + pgofs - start_fofs);
 300                 count = end_fofs - pgofs + 1;
 301                 if (count < (UINT_MAX >> blkbits))
 302                         bh_result->b_size = (count << blkbits);
 303                 else
 304                         bh_result->b_size = UINT_MAX;
 305
 306                 stat_inc_read_hit(inode->i_sb);
 307                 read_unlock(&fi->ext.ext_lock);
 308                 return 1;
 309         }
 310         read_unlock(&fi->ext.ext_lock);
 311         return 0;
 312 }
 313
 314 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
 315 {
 316         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 317         pgoff_t fofs, start_fofs, end_fofs;
 318         block_t start_blkaddr, end_blkaddr;
 319         int need_update = true;
 320
 321         f2fs_bug_on(blk_addr == NEW_ADDR);
 322         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 323                                                         dn->ofs_in_node;
 324
 325         /* Update the page address in the parent node */
 326         __set_data_blkaddr(dn, blk_addr);
 327
 328         if (is_inode_flag_set(fi, FI_NO_EXTENT))
 329                 return;
 330
 331         write_lock(&fi->ext.ext_lock);
 332
 333         start_fofs = fi->ext.fofs;
 334         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 335         start_blkaddr = fi->ext.blk_addr;
 336         end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
 337
 338         /* Drop and initialize the matched extent */
 339         if (fi->ext.len == 1 && fofs == start_fofs)
 340                 fi->ext.len = 0;
 341
 342         /* Initial extent */
 343         if (fi->ext.len == 0) {
 344                 if (blk_addr != NULL_ADDR) {
 345                         fi->ext.fofs = fofs;
 346                         fi->ext.blk_addr = blk_addr;
 347                         fi->ext.len = 1;
 348                 }
 349                 goto end_update;
 350         }
 351
 352         /* Front merge */
 353         if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
 354                 fi->ext.fofs--;
 355                 fi->ext.blk_addr--;
 356                 fi->ext.len++;
 357                 goto end_update;
 358         }
 359
 360         /* Back merge */
 361         if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
 362                 fi->ext.len++;
 363                 goto end_update;
 364         }
 365
 366         /* Split the existing extent */
 367         if (fi->ext.len > 1 &&
 368                 fofs >= start_fofs && fofs <= end_fofs) {
 369                 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
 370                         fi->ext.len = fofs - start_fofs;
 371                 } else {
 372                         fi->ext.fofs = fofs + 1;
 373                         fi->ext.blk_addr = start_blkaddr +
 374                                         fofs - start_fofs + 1;
 375                         fi->ext.len -= fofs - start_fofs + 1;
 376                 }
 377         } else {
 378                 need_update = false;
 379         }
 380
 381         /* Finally, if the extent is very fragmented, let's drop the cache. */
 382         if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
 383                 fi->ext.len = 0;
 384                 set_inode_flag(fi, FI_NO_EXTENT);
 385                 need_update = true;
 386         }
 387 end_update:
 388         write_unlock(&fi->ext.ext_lock);
 389         if (need_update)
 390                 sync_inode_page(dn);
 391         return;
 392 }
 393
 394 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
 395 {
 396         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 397         struct address_space *mapping = inode->i_mapping;
 398         struct dnode_of_data dn;
 399         struct page *page;
 400         int err;
 401
 402         page = find_get_page(mapping, index);
 403         if (page && PageUptodate(page))
 404                 return page;
 405         f2fs_put_page(page, 0);
 406
 407         set_new_dnode(&dn, inode, NULL, NULL, 0);
 408         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 409         if (err)
 410                 return ERR_PTR(err);
 411         f2fs_put_dnode(&dn);
 412
 413         if (dn.data_blkaddr == NULL_ADDR)
 414                 return ERR_PTR(-ENOENT);
 415
 416         /* By fallocate(), there is no cached page, but with NEW_ADDR */
 417         if (unlikely(dn.data_blkaddr == NEW_ADDR))
 418                 return ERR_PTR(-EINVAL);
 419
 420         page = grab_cache_page(mapping, index);
 421         if (!page)
 422                 return ERR_PTR(-ENOMEM);
 423
 424         if (PageUptodate(page)) {
 425                 unlock_page(page);
 426                 return page;
 427         }
 428
 429         err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
 430                                         sync ? READ_SYNC : READA);
 431         if (err)
 432                 return ERR_PTR(err);
 433
 434         if (sync) {
 435                 wait_on_page_locked(page);
 436                 if (unlikely(!PageUptodate(page))) {
 437                         f2fs_put_page(page, 0);
 438                         return ERR_PTR(-EIO);
 439                 }
 440         }
 441         return page;
 442 }
 443
 444 /*
 445  * If it tries to access a hole, return an error.
 446  * Because, the callers, functions in dir.c and GC, should be able to know
 447  * whether this page exists or not.
 448  */
 449 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 450 {
 451         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 452         struct address_space *mapping = inode->i_mapping;
 453         struct dnode_of_data dn;
 454         struct page *page;
 455         int err;
 456
 457 repeat:
 458         page = grab_cache_page(mapping, index);
 459         if (!page)
 460                 return ERR_PTR(-ENOMEM);
 461
 462         set_new_dnode(&dn, inode, NULL, NULL, 0);
 463         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 464         if (err) {
 465                 f2fs_put_page(page, 1);
 466                 return ERR_PTR(err);
 467         }
 468         f2fs_put_dnode(&dn);
 469
 470         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 471                 f2fs_put_page(page, 1);
 472                 return ERR_PTR(-ENOENT);
 473         }
 474
 475         if (PageUptodate(page))
 476                 return page;
 477
 478         /*
 479          * A new dentry page is allocated but not able to be written, since its
 480          * new inode page couldn't be allocated due to -ENOSPC.
 481          * In such the case, its blkaddr can be remained as NEW_ADDR.
 482          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 483          */
 484         if (dn.data_blkaddr == NEW_ADDR) {
 485                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 486                 SetPageUptodate(page);
 487                 return page;
 488         }
 489
 490         err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC);
 491         if (err)
 492                 return ERR_PTR(err);
 493
 494         lock_page(page);
 495         if (unlikely(!PageUptodate(page))) {
 496                 f2fs_put_page(page, 1);
 497                 return ERR_PTR(-EIO);
 498         }
 499         if (unlikely(page->mapping != mapping)) {
 500                 f2fs_put_page(page, 1);
 501                 goto repeat;
 502         }
 503         return page;
 504 }
 505
 506 /*
 507  * Caller ensures that this data page is never allocated.
 508  * A new zero-filled data page is allocated in the page cache.
 509  *
 510  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 511  * f2fs_unlock_op().
 512  * Note that, ipage is set only by make_empty_dir.
 513  */
 514 struct page *get_new_data_page(struct inode *inode,
 515                 struct page *ipage, pgoff_t index, bool new_i_size)
 516 {
 517         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 518         struct address_space *mapping = inode->i_mapping;
 519         struct page *page;
 520         struct dnode_of_data dn;
 521         int err;
 522
 523         set_new_dnode(&dn, inode, ipage, NULL, 0);
 524         err = f2fs_reserve_block(&dn, index);
 525         if (err)
 526                 return ERR_PTR(err);
 527 repeat:
 528         page = grab_cache_page(mapping, index);
 529         if (!page) {
 530                 err = -ENOMEM;
 531                 goto put_err;
 532         }
 533
 534         if (PageUptodate(page))
 535                 return page;
 536
 537         if (dn.data_blkaddr == NEW_ADDR) {
 538                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 539                 SetPageUptodate(page);
 540         } else {
 541                 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
 542                                                                 READ_SYNC);
 543                 if (err)
 544                         goto put_err;
 545
 546                 lock_page(page);
 547                 if (unlikely(!PageUptodate(page))) {
 548                         f2fs_put_page(page, 1);
 549                         err = -EIO;
 550                         goto put_err;
 551                 }
 552                 if (unlikely(page->mapping != mapping)) {
 553                         f2fs_put_page(page, 1);
 554                         goto repeat;
 555                 }
 556         }
 557
 558         if (new_i_size &&
 559                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 560                 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
 561                 /* Only the directory inode sets new_i_size */
 562                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 563         }
 564         return page;
 565
 566 put_err:
 567         f2fs_put_dnode(&dn);
 568         return ERR_PTR(err);
 569 }
 570
 571 static int __allocate_data_block(struct dnode_of_data *dn)
 572 {
 573         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 574         struct f2fs_summary sum;
 575         block_t new_blkaddr;
 576         struct node_info ni;
 577         int type;
 578
 579         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 580                 return -EPERM;
 581         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 582                 return -ENOSPC;
 583
 584         __set_data_blkaddr(dn, NEW_ADDR);
 585         dn->data_blkaddr = NEW_ADDR;
 586
 587         get_node_info(sbi, dn->nid, &ni);
 588         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 589
 590         type = CURSEG_WARM_DATA;
 591
 592         allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
 593
 594         /* direct IO doesn't use extent cache to maximize the performance */
 595         set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
 596         update_extent_cache(new_blkaddr, dn);
 597         clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
 598
 599         dn->data_blkaddr = new_blkaddr;
 600         return 0;
 601 }
 602
 603 /*
 604  * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
 605  * If original data blocks are allocated, then give them to blockdev.
 606  * Otherwise,
 607  *     a. preallocate requested block addresses
 608  *     b. do not use extent cache for better performance
 609  *     c. give the block addresses to blockdev
 610  */
 611 static int __get_data_block(struct inode *inode, sector_t iblock,
 612                         struct buffer_head *bh_result, int create, bool fiemap)
 613 {
 614         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 615         unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 616         unsigned maxblocks = bh_result->b_size >> blkbits;
 617         struct dnode_of_data dn;
 618         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 619         pgoff_t pgofs, end_offset;
 620         int err = 0, ofs = 1;
 621         bool allocated = false;
 622
 623         /* Get the page offset from the block offset(iblock) */
 624         pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
 625
 626         if (check_extent_cache(inode, pgofs, bh_result))
 627                 goto out;
 628
 629         if (create)
 630                 f2fs_lock_op(sbi);
 631
 632         /* When reading holes, we need its node page */
 633         set_new_dnode(&dn, inode, NULL, NULL, 0);
 634         err = get_dnode_of_data(&dn, pgofs, mode);
 635         if (err) {
 636                 if (err == -ENOENT)
 637                         err = 0;
 638                 goto unlock_out;
 639         }
 640         if (dn.data_blkaddr == NEW_ADDR && !fiemap)
 641                 goto put_out;
 642
 643         if (dn.data_blkaddr != NULL_ADDR) {
 644                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 645         } else if (create) {
 646                 err = __allocate_data_block(&dn);
 647                 if (err)
 648                         goto put_out;
 649                 allocated = true;
 650                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 651         } else {
 652                 goto put_out;
 653         }
 654
 655         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 656         bh_result->b_size = (((size_t)1) << blkbits);
 657         dn.ofs_in_node++;
 658         pgofs++;
 659
 660 get_next:
 661         if (dn.ofs_in_node >= end_offset) {
 662                 if (allocated)
 663                         sync_inode_page(&dn);
 664                 allocated = false;
 665                 f2fs_put_dnode(&dn);
 666
 667                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 668                 err = get_dnode_of_data(&dn, pgofs, mode);
 669                 if (err) {
 670                         if (err == -ENOENT)
 671                                 err = 0;
 672                         goto unlock_out;
 673                 }
 674                 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
 675                         goto put_out;
 676
 677                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 678         }
 679
 680         if (maxblocks > (bh_result->b_size >> blkbits)) {
 681                 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 682                 if (blkaddr == NULL_ADDR && create) {
 683                         err = __allocate_data_block(&dn);
 684                         if (err)
 685                                 goto sync_out;
 686                         allocated = true;
 687                         blkaddr = dn.data_blkaddr;
 688                 }
 689                 /* Give more consecutive addresses for the read ahead */
 690                 if (blkaddr == (bh_result->b_blocknr + ofs)) {
 691                         ofs++;
 692                         dn.ofs_in_node++;
 693                         pgofs++;
 694                         bh_result->b_size += (((size_t)1) << blkbits);
 695                         goto get_next;
 696                 }
 697         }
 698 sync_out:
 699         if (allocated)
 700                 sync_inode_page(&dn);
 701 put_out:
 702         f2fs_put_dnode(&dn);
 703 unlock_out:
 704         if (create)
 705                 f2fs_unlock_op(sbi);
 706 out:
 707         trace_f2fs_get_data_block(inode, iblock, bh_result, err);
 708         return err;
 709 }
 710
 711 static int get_data_block(struct inode *inode, sector_t iblock,
 712                         struct buffer_head *bh_result, int create)
 713 {
 714         return __get_data_block(inode, iblock, bh_result, create, false);
 715 }
 716
 717 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
 718                         struct buffer_head *bh_result, int create)
 719 {
 720         return __get_data_block(inode, iblock, bh_result, create, true);
 721 }
 722
 723 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 724                 u64 start, u64 len)
 725 {
 726         return generic_block_fiemap(inode, fieinfo,
 727                                 start, len, get_data_block_fiemap);
 728 }
 729
 730 static int f2fs_read_data_page(struct file *file, struct page *page)
 731 {
 732         struct inode *inode = page->mapping->host;
 733         int ret;
 734
 735         trace_f2fs_readpage(page, DATA);
 736
 737         /* If the file has inline data, try to read it directlly */
 738         if (f2fs_has_inline_data(inode))
 739                 ret = f2fs_read_inline_data(inode, page);
 740         else
 741                 ret = mpage_readpage(page, get_data_block);
 742
 743         return ret;
 744 }
 745
 746 static int f2fs_read_data_pages(struct file *file,
 747                         struct address_space *mapping,
 748                         struct list_head *pages, unsigned nr_pages)
 749 {
 750         struct inode *inode = file->f_mapping->host;
 751
 752         /* If the file has inline data, skip readpages */
 753         if (f2fs_has_inline_data(inode))
 754                 return 0;
 755
 756         return mpage_readpages(mapping, pages, nr_pages, get_data_block);
 757 }
 758
 759 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
 760 {
 761         struct inode *inode = page->mapping->host;
 762         block_t old_blkaddr, new_blkaddr;
 763         struct dnode_of_data dn;
 764         int err = 0;
 765
 766         set_new_dnode(&dn, inode, NULL, NULL, 0);
 767         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 768         if (err)
 769                 return err;
 770
 771         old_blkaddr = dn.data_blkaddr;
 772
 773         /* This page is already truncated */
 774         if (old_blkaddr == NULL_ADDR)
 775                 goto out_writepage;
 776
 777         set_page_writeback(page);
 778
 779         /*
 780          * If current allocation needs SSR,
 781          * it had better in-place writes for updated data.
 782          */
 783         if (unlikely(old_blkaddr != NEW_ADDR &&
 784                         !is_cold_data(page) &&
 785                         need_inplace_update(inode))) {
 786                 rewrite_data_page(page, old_blkaddr, fio);
 787         } else {
 788                 write_data_page(page, &dn, &new_blkaddr, fio);
 789                 update_extent_cache(new_blkaddr, &dn);
 790         }
 791 out_writepage:
 792         f2fs_put_dnode(&dn);
 793         return err;
 794 }
 795
 796 static int f2fs_write_data_page(struct page *page,
 797                                         struct writeback_control *wbc)
 798 {
 799         struct inode *inode = page->mapping->host;
 800         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 801         loff_t i_size = i_size_read(inode);
 802         const pgoff_t end_index = ((unsigned long long) i_size)
 803                                                         >> PAGE_CACHE_SHIFT;
 804         unsigned offset = 0;
 805         bool need_balance_fs = false;
 806         int err = 0;
 807         struct f2fs_io_info fio = {
 808                 .type = DATA,
 809                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
 810         };
 811
 812         trace_f2fs_writepage(page, DATA);
 813
 814         if (page->index < end_index)
 815                 goto write;
 816
 817         /*
 818          * If the offset is out-of-range of file size,
 819          * this page does not have to be written to disk.
 820          */
 821         offset = i_size & (PAGE_CACHE_SIZE - 1);
 822         if ((page->index >= end_index + 1) || !offset)
 823                 goto out;
 824
 825         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 826 write:
 827         if (unlikely(sbi->por_doing))
 828                 goto redirty_out;
 829
 830         /* Dentry blocks are controlled by checkpoint */
 831         if (S_ISDIR(inode->i_mode)) {
 832                 err = do_write_data_page(page, &fio);
 833                 goto done;
 834         }
 835
 836         if (!wbc->for_reclaim)
 837                 need_balance_fs = true;
 838         else if (has_not_enough_free_secs(sbi, 0))
 839                 goto redirty_out;
 840
 841         f2fs_lock_op(sbi);
 842         if (f2fs_has_inline_data(inode) || f2fs_may_inline(inode))
 843                 err = f2fs_write_inline_data(inode, page, offset);
 844         else
 845                 err = do_write_data_page(page, &fio);
 846         f2fs_unlock_op(sbi);
 847 done:
 848         if (err && err != -ENOENT)
 849                 goto redirty_out;
 850
 851         clear_cold_data(page);
 852 out:
 853         inode_dec_dirty_dents(inode);
 854         unlock_page(page);
 855         if (need_balance_fs)
 856                 f2fs_balance_fs(sbi);
 857         if (wbc->for_reclaim)
 858                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 859         return 0;
 860
 861 redirty_out:
 862         redirty_page_for_writepage(wbc, page);
 863         return AOP_WRITEPAGE_ACTIVATE;
 864 }
 865
 866 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
 867                         void *data)
 868 {
 869         struct address_space *mapping = data;
 870         int ret = mapping->a_ops->writepage(page, wbc);
 871         mapping_set_error(mapping, ret);
 872         return ret;
 873 }
 874
 875 static int f2fs_write_data_pages(struct address_space *mapping,
 876                             struct writeback_control *wbc)
 877 {
 878         struct inode *inode = mapping->host;
 879         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 880         bool locked = false;
 881         int ret;
 882         long diff;
 883
 884         trace_f2fs_writepages(mapping->host, wbc, DATA);
 885
 886         /* deal with chardevs and other special file */
 887         if (!mapping->a_ops->writepage)
 888                 return 0;
 889
 890         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
 891                         get_dirty_dents(inode) < nr_pages_to_skip(sbi, DATA) &&
 892                         available_free_memory(sbi, DIRTY_DENTS))
 893                 goto skip_write;
 894
 895         diff = nr_pages_to_write(sbi, DATA, wbc);
 896
 897         if (!S_ISDIR(inode->i_mode)) {
 898                 mutex_lock(&sbi->writepages);
 899                 locked = true;
 900         }
 901         ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
 902         if (locked)
 903                 mutex_unlock(&sbi->writepages);
 904
 905         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 906
 907         remove_dirty_dir_inode(inode);
 908
 909         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
 910         return ret;
 911
 912 skip_write:
 913         wbc->pages_skipped += get_dirty_dents(inode);
 914         return 0;
 915 }
 916
 917 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 918                 loff_t pos, unsigned len, unsigned flags,
 919                 struct page **pagep, void **fsdata)
 920 {
 921         struct inode *inode = mapping->host;
 922         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 923         struct page *page;
 924         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
 925         struct dnode_of_data dn;
 926         int err = 0;
 927
 928         trace_f2fs_write_begin(inode, pos, len, flags);
 929
 930         f2fs_balance_fs(sbi);
 931 repeat:
 932         err = f2fs_convert_inline_data(inode, pos + len);
 933         if (err)
 934                 return err;
 935
 936         page = grab_cache_page_write_begin(mapping, index, flags);
 937         if (!page)
 938                 return -ENOMEM;
 939
 940         /* to avoid latency during memory pressure */
 941         unlock_page(page);
 942
 943         *pagep = page;
 944
 945         if (f2fs_has_inline_data(inode) && (pos + len) <= MAX_INLINE_DATA)
 946                 goto inline_data;
 947
 948         f2fs_lock_op(sbi);
 949         set_new_dnode(&dn, inode, NULL, NULL, 0);
 950         err = f2fs_reserve_block(&dn, index);
 951         f2fs_unlock_op(sbi);
 952
 953         if (err) {
 954                 f2fs_put_page(page, 0);
 955                 return err;
 956         }
 957 inline_data:
 958         lock_page(page);
 959         if (unlikely(page->mapping != mapping)) {
 960                 f2fs_put_page(page, 1);
 961                 goto repeat;
 962         }
 963
 964         f2fs_wait_on_page_writeback(page, DATA);
 965
 966         if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
 967                 return 0;
 968
 969         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
 970                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 971                 unsigned end = start + len;
 972
 973                 /* Reading beyond i_size is simple: memset to zero */
 974                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
 975                 goto out;
 976         }
 977
 978         if (dn.data_blkaddr == NEW_ADDR) {
 979                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 980         } else {
 981                 if (f2fs_has_inline_data(inode)) {
 982                         err = f2fs_read_inline_data(inode, page);
 983                         if (err) {
 984                                 page_cache_release(page);
 985                                 return err;
 986                         }
 987                 } else {
 988                         err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
 989                                                         READ_SYNC);
 990                         if (err)
 991                                 return err;
 992                 }
 993
 994                 lock_page(page);
 995                 if (unlikely(!PageUptodate(page))) {
 996                         f2fs_put_page(page, 1);
 997                         return -EIO;
 998                 }
 999                 if (unlikely(page->mapping != mapping)) {
1000                         f2fs_put_page(page, 1);
1001                         goto repeat;
1002                 }
1003         }
1004 out:
1005         SetPageUptodate(page);
1006         clear_cold_data(page);
1007         return 0;
1008 }
1009
1010 static int f2fs_write_end(struct file *file,
1011                         struct address_space *mapping,
1012                         loff_t pos, unsigned len, unsigned copied,
1013                         struct page *page, void *fsdata)
1014 {
1015         struct inode *inode = page->mapping->host;
1016
1017         trace_f2fs_write_end(inode, pos, len, copied);
1018
1019         SetPageUptodate(page);
1020         set_page_dirty(page);
1021
1022         if (pos + copied > i_size_read(inode)) {
1023                 i_size_write(inode, pos + copied);
1024                 mark_inode_dirty(inode);
1025                 update_inode_page(inode);
1026         }
1027
1028         f2fs_put_page(page, 1);
1029         return copied;
1030 }
1031
1032 static int check_direct_IO(struct inode *inode, int rw,
1033                 struct iov_iter *iter, loff_t offset)
1034 {
1035         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1036
1037         if (rw == READ)
1038                 return 0;
1039
1040         if (offset & blocksize_mask)
1041                 return -EINVAL;
1042
1043         if (iov_iter_alignment(iter) & blocksize_mask)
1044                 return -EINVAL;
1045
1046         return 0;
1047 }
1048
1049 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1050                 struct iov_iter *iter, loff_t offset)
1051 {
1052         struct file *file = iocb->ki_filp;
1053         struct inode *inode = file->f_mapping->host;
1054
1055         /* Let buffer I/O handle the inline data case. */
1056         if (f2fs_has_inline_data(inode))
1057                 return 0;
1058
1059         if (check_direct_IO(inode, rw, iter, offset))
1060                 return 0;
1061
1062         /* clear fsync mark to recover these blocks */
1063         fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
1064
1065         return blockdev_direct_IO(rw, iocb, inode, iter, offset,
1066                                   get_data_block);
1067 }
1068
1069 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
1070                                       unsigned int length)
1071 {
1072         struct inode *inode = page->mapping->host;
1073         if (PageDirty(page))
1074                 inode_dec_dirty_dents(inode);
1075         ClearPagePrivate(page);
1076 }
1077
1078 static int f2fs_release_data_page(struct page *page, gfp_t wait)
1079 {
1080         ClearPagePrivate(page);
1081         return 1;
1082 }
1083
1084 static int f2fs_set_data_page_dirty(struct page *page)
1085 {
1086         struct address_space *mapping = page->mapping;
1087         struct inode *inode = mapping->host;
1088
1089         trace_f2fs_set_page_dirty(page, DATA);
1090
1091         SetPageUptodate(page);
1092         mark_inode_dirty(inode);
1093
1094         if (!PageDirty(page)) {
1095                 __set_page_dirty_nobuffers(page);
1096                 set_dirty_dir_page(inode, page);
1097                 return 1;
1098         }
1099         return 0;
1100 }
1101
1102 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1103 {
1104         struct inode *inode = mapping->host;
1105
1106         if (f2fs_has_inline_data(inode))
1107                 return 0;
1108
1109         return generic_block_bmap(mapping, block, get_data_block);
1110 }
1111
1112 const struct address_space_operations f2fs_dblock_aops = {
1113         .readpage       = f2fs_read_data_page,
1114         .readpages      = f2fs_read_data_pages,
1115         .writepage      = f2fs_write_data_page,
1116         .writepages     = f2fs_write_data_pages,
1117         .write_begin    = f2fs_write_begin,
1118         .write_end      = f2fs_write_end,
1119         .set_page_dirty = f2fs_set_data_page_dirty,
1120         .invalidatepage = f2fs_invalidate_data_page,
1121         .releasepage    = f2fs_release_data_page,
1122         .direct_IO      = f2fs_direct_IO,
1123         .bmap           = f2fs_bmap,
1124 };