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.h"
  26 #include <trace/events/f2fs.h>
  27
  28 static void f2fs_read_end_io(struct bio *bio, int err)
  29 {
  30         struct bio_vec *bvec;
  31         int i;
  32
  33         bio_for_each_segment_all(bvec, bio, i) {
  34                 struct page *page = bvec->bv_page;
  35
  36                 if (!err) {
  37                         SetPageUptodate(page);
  38                 } else {
  39                         ClearPageUptodate(page);
  40                         SetPageError(page);
  41                 }
  42                 unlock_page(page);
  43         }
  44         bio_put(bio);
  45 }
  46
  47 static void f2fs_write_end_io(struct bio *bio, int err)
  48 {
  49         struct f2fs_sb_info *sbi = bio->bi_private;
  50         struct bio_vec *bvec;
  51         int i;
  52
  53         bio_for_each_segment_all(bvec, bio, i) {
  54                 struct page *page = bvec->bv_page;
  55
  56                 if (unlikely(err)) {
  57                         set_page_dirty(page);
  58                         set_bit(AS_EIO, &page->mapping->flags);
  59                         f2fs_stop_checkpoint(sbi);
  60                 }
  61                 end_page_writeback(page);
  62                 dec_page_count(sbi, F2FS_WRITEBACK);
  63         }
  64
  65         if (!get_pages(sbi, F2FS_WRITEBACK) &&
  66                         !list_empty(&sbi->cp_wait.task_list))
  67                 wake_up(&sbi->cp_wait);
  68
  69         bio_put(bio);
  70 }
  71
  72 /*
  73  * Low-level block read/write IO operations.
  74  */
  75 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  76                                 int npages, bool is_read)
  77 {
  78         struct bio *bio;
  79
  80         /* No failure on bio allocation */
  81         bio = bio_alloc(GFP_NOIO, npages);
  82
  83         bio->bi_bdev = sbi->sb->s_bdev;
  84         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  85         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  86         bio->bi_private = sbi;
  87
  88         return bio;
  89 }
  90
  91 static void __submit_merged_bio(struct f2fs_bio_info *io)
  92 {
  93         struct f2fs_io_info *fio = &io->fio;
  94
  95         if (!io->bio)
  96                 return;
  97
  98         if (is_read_io(fio->rw))
  99                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 100         else
 101                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 102
 103         submit_bio(fio->rw, io->bio);
 104         io->bio = NULL;
 105 }
 106
 107 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 108                                 enum page_type type, int rw)
 109 {
 110         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 111         struct f2fs_bio_info *io;
 112
 113         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 114
 115         down_write(&io->io_rwsem);
 116
 117         /* change META to META_FLUSH in the checkpoint procedure */
 118         if (type >= META_FLUSH) {
 119                 io->fio.type = META_FLUSH;
 120                 if (test_opt(sbi, NOBARRIER))
 121                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 122                 else
 123                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 124         }
 125         __submit_merged_bio(io);
 126         up_write(&io->io_rwsem);
 127 }
 128
 129 /*
 130  * Fill the locked page with data located in the block address.
 131  * Return unlocked page.
 132  */
 133 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
 134                                         struct f2fs_io_info *fio)
 135 {
 136         struct bio *bio;
 137
 138         trace_f2fs_submit_page_bio(page, fio);
 139         f2fs_trace_ios(page, fio, 0);
 140
 141         /* Allocate a new bio */
 142         bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw));
 143
 144         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 145                 bio_put(bio);
 146                 f2fs_put_page(page, 1);
 147                 return -EFAULT;
 148         }
 149
 150         submit_bio(fio->rw, bio);
 151         return 0;
 152 }
 153
 154 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
 155                                         struct f2fs_io_info *fio)
 156 {
 157         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 158         struct f2fs_bio_info *io;
 159         bool is_read = is_read_io(fio->rw);
 160
 161         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 162
 163         verify_block_addr(sbi, fio->blk_addr);
 164
 165         down_write(&io->io_rwsem);
 166
 167         if (!is_read)
 168                 inc_page_count(sbi, F2FS_WRITEBACK);
 169
 170         if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
 171                                                 io->fio.rw != fio->rw))
 172                 __submit_merged_bio(io);
 173 alloc_new:
 174         if (io->bio == NULL) {
 175                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 176
 177                 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
 178                 io->fio = *fio;
 179         }
 180
 181         if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
 182                                                         PAGE_CACHE_SIZE) {
 183                 __submit_merged_bio(io);
 184                 goto alloc_new;
 185         }
 186
 187         io->last_block_in_bio = fio->blk_addr;
 188         f2fs_trace_ios(page, fio, 0);
 189
 190         up_write(&io->io_rwsem);
 191         trace_f2fs_submit_page_mbio(page, fio);
 192 }
 193
 194 /*
 195  * Lock ordering for the change of data block address:
 196  * ->data_page
 197  *  ->node_page
 198  *    update block addresses in the node page
 199  */
 200 static void __set_data_blkaddr(struct dnode_of_data *dn)
 201 {
 202         struct f2fs_node *rn;
 203         __le32 *addr_array;
 204         struct page *node_page = dn->node_page;
 205         unsigned int ofs_in_node = dn->ofs_in_node;
 206
 207         f2fs_wait_on_page_writeback(node_page, NODE);
 208
 209         rn = F2FS_NODE(node_page);
 210
 211         /* Get physical address of data block */
 212         addr_array = blkaddr_in_node(rn);
 213         addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 214         set_page_dirty(node_page);
 215 }
 216
 217 int reserve_new_block(struct dnode_of_data *dn)
 218 {
 219         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 220
 221         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 222                 return -EPERM;
 223         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 224                 return -ENOSPC;
 225
 226         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 227
 228         dn->data_blkaddr = NEW_ADDR;
 229         __set_data_blkaddr(dn);
 230         mark_inode_dirty(dn->inode);
 231         sync_inode_page(dn);
 232         return 0;
 233 }
 234
 235 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 236 {
 237         bool need_put = dn->inode_page ? false : true;
 238         int err;
 239
 240         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 241         if (err)
 242                 return err;
 243
 244         if (dn->data_blkaddr == NULL_ADDR)
 245                 err = reserve_new_block(dn);
 246         if (err || need_put)
 247                 f2fs_put_dnode(dn);
 248         return err;
 249 }
 250
 251 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
 252                                         struct buffer_head *bh_result)
 253 {
 254         struct f2fs_inode_info *fi = F2FS_I(inode);
 255         pgoff_t start_fofs, end_fofs;
 256         block_t start_blkaddr;
 257
 258         if (is_inode_flag_set(fi, FI_NO_EXTENT))
 259                 return 0;
 260
 261         read_lock(&fi->ext_lock);
 262         if (fi->ext.len == 0) {
 263                 read_unlock(&fi->ext_lock);
 264                 return 0;
 265         }
 266
 267         stat_inc_total_hit(inode->i_sb);
 268
 269         start_fofs = fi->ext.fofs;
 270         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 271         start_blkaddr = fi->ext.blk;
 272
 273         if (pgofs >= start_fofs && pgofs <= end_fofs) {
 274                 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 275                 size_t count;
 276
 277                 set_buffer_new(bh_result);
 278                 map_bh(bh_result, inode->i_sb,
 279                                 start_blkaddr + pgofs - start_fofs);
 280                 count = end_fofs - pgofs + 1;
 281                 if (count < (UINT_MAX >> blkbits))
 282                         bh_result->b_size = (count << blkbits);
 283                 else
 284                         bh_result->b_size = UINT_MAX;
 285
 286                 stat_inc_read_hit(inode->i_sb);
 287                 read_unlock(&fi->ext_lock);
 288                 return 1;
 289         }
 290         read_unlock(&fi->ext_lock);
 291         return 0;
 292 }
 293
 294 void update_extent_cache(struct dnode_of_data *dn)
 295 {
 296         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 297         pgoff_t fofs, start_fofs, end_fofs;
 298         block_t start_blkaddr, end_blkaddr;
 299         int need_update = true;
 300
 301         f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
 302
 303         /* Update the page address in the parent node */
 304         __set_data_blkaddr(dn);
 305
 306         if (is_inode_flag_set(fi, FI_NO_EXTENT))
 307                 return;
 308
 309         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 310                                                         dn->ofs_in_node;
 311
 312         write_lock(&fi->ext_lock);
 313
 314         start_fofs = fi->ext.fofs;
 315         end_fofs = fi->ext.fofs + fi->ext.len - 1;
 316         start_blkaddr = fi->ext.blk;
 317         end_blkaddr = fi->ext.blk + fi->ext.len - 1;
 318
 319         /* Drop and initialize the matched extent */
 320         if (fi->ext.len == 1 && fofs == start_fofs)
 321                 fi->ext.len = 0;
 322
 323         /* Initial extent */
 324         if (fi->ext.len == 0) {
 325                 if (dn->data_blkaddr != NULL_ADDR) {
 326                         fi->ext.fofs = fofs;
 327                         fi->ext.blk = dn->data_blkaddr;
 328                         fi->ext.len = 1;
 329                 }
 330                 goto end_update;
 331         }
 332
 333         /* Front merge */
 334         if (fofs == start_fofs - 1 && dn->data_blkaddr == start_blkaddr - 1) {
 335                 fi->ext.fofs--;
 336                 fi->ext.blk--;
 337                 fi->ext.len++;
 338                 goto end_update;
 339         }
 340
 341         /* Back merge */
 342         if (fofs == end_fofs + 1 && dn->data_blkaddr == end_blkaddr + 1) {
 343                 fi->ext.len++;
 344                 goto end_update;
 345         }
 346
 347         /* Split the existing extent */
 348         if (fi->ext.len > 1 &&
 349                 fofs >= start_fofs && fofs <= end_fofs) {
 350                 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
 351                         fi->ext.len = fofs - start_fofs;
 352                 } else {
 353                         fi->ext.fofs = fofs + 1;
 354                         fi->ext.blk = start_blkaddr + fofs - start_fofs + 1;
 355                         fi->ext.len -= fofs - start_fofs + 1;
 356                 }
 357         } else {
 358                 need_update = false;
 359         }
 360
 361         /* Finally, if the extent is very fragmented, let's drop the cache. */
 362         if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
 363                 fi->ext.len = 0;
 364                 set_inode_flag(fi, FI_NO_EXTENT);
 365                 need_update = true;
 366         }
 367 end_update:
 368         write_unlock(&fi->ext_lock);
 369         if (need_update)
 370                 sync_inode_page(dn);
 371         return;
 372 }
 373
 374 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
 375 {
 376         struct address_space *mapping = inode->i_mapping;
 377         struct dnode_of_data dn;
 378         struct page *page;
 379         int err;
 380         struct f2fs_io_info fio = {
 381                 .type = DATA,
 382                 .rw = sync ? READ_SYNC : READA,
 383         };
 384
 385         page = find_get_page(mapping, index);
 386         if (page && PageUptodate(page))
 387                 return page;
 388         f2fs_put_page(page, 0);
 389
 390         set_new_dnode(&dn, inode, NULL, NULL, 0);
 391         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 392         if (err)
 393                 return ERR_PTR(err);
 394         f2fs_put_dnode(&dn);
 395
 396         if (dn.data_blkaddr == NULL_ADDR)
 397                 return ERR_PTR(-ENOENT);
 398
 399         /* By fallocate(), there is no cached page, but with NEW_ADDR */
 400         if (unlikely(dn.data_blkaddr == NEW_ADDR))
 401                 return ERR_PTR(-EINVAL);
 402
 403         page = grab_cache_page(mapping, index);
 404         if (!page)
 405                 return ERR_PTR(-ENOMEM);
 406
 407         if (PageUptodate(page)) {
 408                 unlock_page(page);
 409                 return page;
 410         }
 411
 412         fio.blk_addr = dn.data_blkaddr;
 413         err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
 414         if (err)
 415                 return ERR_PTR(err);
 416
 417         if (sync) {
 418                 wait_on_page_locked(page);
 419                 if (unlikely(!PageUptodate(page))) {
 420                         f2fs_put_page(page, 0);
 421                         return ERR_PTR(-EIO);
 422                 }
 423         }
 424         return page;
 425 }
 426
 427 /*
 428  * If it tries to access a hole, return an error.
 429  * Because, the callers, functions in dir.c and GC, should be able to know
 430  * whether this page exists or not.
 431  */
 432 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
 433 {
 434         struct address_space *mapping = inode->i_mapping;
 435         struct dnode_of_data dn;
 436         struct page *page;
 437         int err;
 438         struct f2fs_io_info fio = {
 439                 .type = DATA,
 440                 .rw = READ_SYNC,
 441         };
 442 repeat:
 443         page = grab_cache_page(mapping, index);
 444         if (!page)
 445                 return ERR_PTR(-ENOMEM);
 446
 447         set_new_dnode(&dn, inode, NULL, NULL, 0);
 448         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 449         if (err) {
 450                 f2fs_put_page(page, 1);
 451                 return ERR_PTR(err);
 452         }
 453         f2fs_put_dnode(&dn);
 454
 455         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 456                 f2fs_put_page(page, 1);
 457                 return ERR_PTR(-ENOENT);
 458         }
 459
 460         if (PageUptodate(page))
 461                 return page;
 462
 463         /*
 464          * A new dentry page is allocated but not able to be written, since its
 465          * new inode page couldn't be allocated due to -ENOSPC.
 466          * In such the case, its blkaddr can be remained as NEW_ADDR.
 467          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 468          */
 469         if (dn.data_blkaddr == NEW_ADDR) {
 470                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 471                 SetPageUptodate(page);
 472                 return page;
 473         }
 474
 475         fio.blk_addr = dn.data_blkaddr;
 476         err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
 477         if (err)
 478                 return ERR_PTR(err);
 479
 480         lock_page(page);
 481         if (unlikely(!PageUptodate(page))) {
 482                 f2fs_put_page(page, 1);
 483                 return ERR_PTR(-EIO);
 484         }
 485         if (unlikely(page->mapping != mapping)) {
 486                 f2fs_put_page(page, 1);
 487                 goto repeat;
 488         }
 489         return page;
 490 }
 491
 492 /*
 493  * Caller ensures that this data page is never allocated.
 494  * A new zero-filled data page is allocated in the page cache.
 495  *
 496  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 497  * f2fs_unlock_op().
 498  * Note that, ipage is set only by make_empty_dir.
 499  */
 500 struct page *get_new_data_page(struct inode *inode,
 501                 struct page *ipage, pgoff_t index, bool new_i_size)
 502 {
 503         struct address_space *mapping = inode->i_mapping;
 504         struct page *page;
 505         struct dnode_of_data dn;
 506         int err;
 507
 508         set_new_dnode(&dn, inode, ipage, NULL, 0);
 509         err = f2fs_reserve_block(&dn, index);
 510         if (err)
 511                 return ERR_PTR(err);
 512 repeat:
 513         page = grab_cache_page(mapping, index);
 514         if (!page) {
 515                 err = -ENOMEM;
 516                 goto put_err;
 517         }
 518
 519         if (PageUptodate(page))
 520                 return page;
 521
 522         if (dn.data_blkaddr == NEW_ADDR) {
 523                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 524                 SetPageUptodate(page);
 525         } else {
 526                 struct f2fs_io_info fio = {
 527                         .type = DATA,
 528                         .rw = READ_SYNC,
 529                         .blk_addr = dn.data_blkaddr,
 530                 };
 531                 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
 532                 if (err)
 533                         goto put_err;
 534
 535                 lock_page(page);
 536                 if (unlikely(!PageUptodate(page))) {
 537                         f2fs_put_page(page, 1);
 538                         err = -EIO;
 539                         goto put_err;
 540                 }
 541                 if (unlikely(page->mapping != mapping)) {
 542                         f2fs_put_page(page, 1);
 543                         goto repeat;
 544                 }
 545         }
 546
 547         if (new_i_size &&
 548                 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
 549                 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
 550                 /* Only the directory inode sets new_i_size */
 551                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 552         }
 553         return page;
 554
 555 put_err:
 556         f2fs_put_dnode(&dn);
 557         return ERR_PTR(err);
 558 }
 559
 560 static int __allocate_data_block(struct dnode_of_data *dn)
 561 {
 562         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 563         struct f2fs_inode_info *fi = F2FS_I(dn->inode);
 564         struct f2fs_summary sum;
 565         struct node_info ni;
 566         int seg = CURSEG_WARM_DATA;
 567         pgoff_t fofs;
 568
 569         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 570                 return -EPERM;
 571         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 572                 return -ENOSPC;
 573
 574         get_node_info(sbi, dn->nid, &ni);
 575         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 576
 577         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 578                 seg = CURSEG_DIRECT_IO;
 579
 580         allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg);
 581
 582         /* direct IO doesn't use extent cache to maximize the performance */
 583         __set_data_blkaddr(dn);
 584
 585         /* update i_size */
 586         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
 587                                                         dn->ofs_in_node;
 588         if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
 589                 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
 590
 591         return 0;
 592 }
 593
 594 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
 595                                                         size_t count)
 596 {
 597         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 598         struct dnode_of_data dn;
 599         u64 start = F2FS_BYTES_TO_BLK(offset);
 600         u64 len = F2FS_BYTES_TO_BLK(count);
 601         bool allocated;
 602         u64 end_offset;
 603
 604         while (len) {
 605                 f2fs_balance_fs(sbi);
 606                 f2fs_lock_op(sbi);
 607
 608                 /* When reading holes, we need its node page */
 609                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 610                 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
 611                         goto out;
 612
 613                 allocated = false;
 614                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 615
 616                 while (dn.ofs_in_node < end_offset && len) {
 617                         if (dn.data_blkaddr == NULL_ADDR) {
 618                                 if (__allocate_data_block(&dn))
 619                                         goto sync_out;
 620                                 allocated = true;
 621                         }
 622                         len--;
 623                         start++;
 624                         dn.ofs_in_node++;
 625                 }
 626
 627                 if (allocated)
 628                         sync_inode_page(&dn);
 629
 630                 f2fs_put_dnode(&dn);
 631                 f2fs_unlock_op(sbi);
 632         }
 633         return;
 634
 635 sync_out:
 636         if (allocated)
 637                 sync_inode_page(&dn);
 638         f2fs_put_dnode(&dn);
 639 out:
 640         f2fs_unlock_op(sbi);
 641         return;
 642 }
 643
 644 /*
 645  * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
 646  * If original data blocks are allocated, then give them to blockdev.
 647  * Otherwise,
 648  *     a. preallocate requested block addresses
 649  *     b. do not use extent cache for better performance
 650  *     c. give the block addresses to blockdev
 651  */
 652 static int __get_data_block(struct inode *inode, sector_t iblock,
 653                         struct buffer_head *bh_result, int create, bool fiemap)
 654 {
 655         unsigned int blkbits = inode->i_sb->s_blocksize_bits;
 656         unsigned maxblocks = bh_result->b_size >> blkbits;
 657         struct dnode_of_data dn;
 658         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 659         pgoff_t pgofs, end_offset;
 660         int err = 0, ofs = 1;
 661         bool allocated = false;
 662
 663         /* Get the page offset from the block offset(iblock) */
 664         pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
 665
 666         if (check_extent_cache(inode, pgofs, bh_result))
 667                 goto out;
 668
 669         if (create)
 670                 f2fs_lock_op(F2FS_I_SB(inode));
 671
 672         /* When reading holes, we need its node page */
 673         set_new_dnode(&dn, inode, NULL, NULL, 0);
 674         err = get_dnode_of_data(&dn, pgofs, mode);
 675         if (err) {
 676                 if (err == -ENOENT)
 677                         err = 0;
 678                 goto unlock_out;
 679         }
 680         if (dn.data_blkaddr == NEW_ADDR && !fiemap)
 681                 goto put_out;
 682
 683         if (dn.data_blkaddr != NULL_ADDR) {
 684                 set_buffer_new(bh_result);
 685                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 686         } else if (create) {
 687                 err = __allocate_data_block(&dn);
 688                 if (err)
 689                         goto put_out;
 690                 allocated = true;
 691                 set_buffer_new(bh_result);
 692                 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
 693         } else {
 694                 goto put_out;
 695         }
 696
 697         end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 698         bh_result->b_size = (((size_t)1) << blkbits);
 699         dn.ofs_in_node++;
 700         pgofs++;
 701
 702 get_next:
 703         if (dn.ofs_in_node >= end_offset) {
 704                 if (allocated)
 705                         sync_inode_page(&dn);
 706                 allocated = false;
 707                 f2fs_put_dnode(&dn);
 708
 709                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 710                 err = get_dnode_of_data(&dn, pgofs, mode);
 711                 if (err) {
 712                         if (err == -ENOENT)
 713                                 err = 0;
 714                         goto unlock_out;
 715                 }
 716                 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
 717                         goto put_out;
 718
 719                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 720         }
 721
 722         if (maxblocks > (bh_result->b_size >> blkbits)) {
 723                 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 724                 if (blkaddr == NULL_ADDR && create) {
 725                         err = __allocate_data_block(&dn);
 726                         if (err)
 727                                 goto sync_out;
 728                         allocated = true;
 729                         blkaddr = dn.data_blkaddr;
 730                 }
 731                 /* Give more consecutive addresses for the readahead */
 732                 if (blkaddr == (bh_result->b_blocknr + ofs)) {
 733                         ofs++;
 734                         dn.ofs_in_node++;
 735                         pgofs++;
 736                         bh_result->b_size += (((size_t)1) << blkbits);
 737                         goto get_next;
 738                 }
 739         }
 740 sync_out:
 741         if (allocated)
 742                 sync_inode_page(&dn);
 743 put_out:
 744         f2fs_put_dnode(&dn);
 745 unlock_out:
 746         if (create)
 747                 f2fs_unlock_op(F2FS_I_SB(inode));
 748 out:
 749         trace_f2fs_get_data_block(inode, iblock, bh_result, err);
 750         return err;
 751 }
 752
 753 static int get_data_block(struct inode *inode, sector_t iblock,
 754                         struct buffer_head *bh_result, int create)
 755 {
 756         return __get_data_block(inode, iblock, bh_result, create, false);
 757 }
 758
 759 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
 760                         struct buffer_head *bh_result, int create)
 761 {
 762         return __get_data_block(inode, iblock, bh_result, create, true);
 763 }
 764
 765 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 766                 u64 start, u64 len)
 767 {
 768         return generic_block_fiemap(inode, fieinfo,
 769                                 start, len, get_data_block_fiemap);
 770 }
 771
 772 static int f2fs_read_data_page(struct file *file, struct page *page)
 773 {
 774         struct inode *inode = page->mapping->host;
 775         int ret = -EAGAIN;
 776
 777         trace_f2fs_readpage(page, DATA);
 778
 779         /* If the file has inline data, try to read it directly */
 780         if (f2fs_has_inline_data(inode))
 781                 ret = f2fs_read_inline_data(inode, page);
 782         if (ret == -EAGAIN)
 783                 ret = mpage_readpage(page, get_data_block);
 784
 785         return ret;
 786 }
 787
 788 static int f2fs_read_data_pages(struct file *file,
 789                         struct address_space *mapping,
 790                         struct list_head *pages, unsigned nr_pages)
 791 {
 792         struct inode *inode = file->f_mapping->host;
 793
 794         /* If the file has inline data, skip readpages */
 795         if (f2fs_has_inline_data(inode))
 796                 return 0;
 797
 798         return mpage_readpages(mapping, pages, nr_pages, get_data_block);
 799 }
 800
 801 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
 802 {
 803         struct inode *inode = page->mapping->host;
 804         struct dnode_of_data dn;
 805         int err = 0;
 806
 807         set_new_dnode(&dn, inode, NULL, NULL, 0);
 808         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
 809         if (err)
 810                 return err;
 811
 812         fio->blk_addr = dn.data_blkaddr;
 813
 814         /* This page is already truncated */
 815         if (fio->blk_addr == NULL_ADDR)
 816                 goto out_writepage;
 817
 818         set_page_writeback(page);
 819
 820         /*
 821          * If current allocation needs SSR,
 822          * it had better in-place writes for updated data.
 823          */
 824         if (unlikely(fio->blk_addr != NEW_ADDR &&
 825                         !is_cold_data(page) &&
 826                         need_inplace_update(inode))) {
 827                 rewrite_data_page(page, fio);
 828                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
 829         } else {
 830                 write_data_page(page, &dn, fio);
 831                 update_extent_cache(&dn);
 832                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
 833         }
 834 out_writepage:
 835         f2fs_put_dnode(&dn);
 836         return err;
 837 }
 838
 839 static int f2fs_write_data_page(struct page *page,
 840                                         struct writeback_control *wbc)
 841 {
 842         struct inode *inode = page->mapping->host;
 843         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 844         loff_t i_size = i_size_read(inode);
 845         const pgoff_t end_index = ((unsigned long long) i_size)
 846                                                         >> PAGE_CACHE_SHIFT;
 847         unsigned offset = 0;
 848         bool need_balance_fs = false;
 849         int err = 0;
 850         struct f2fs_io_info fio = {
 851                 .type = DATA,
 852                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
 853         };
 854
 855         trace_f2fs_writepage(page, DATA);
 856
 857         if (page->index < end_index)
 858                 goto write;
 859
 860         /*
 861          * If the offset is out-of-range of file size,
 862          * this page does not have to be written to disk.
 863          */
 864         offset = i_size & (PAGE_CACHE_SIZE - 1);
 865         if ((page->index >= end_index + 1) || !offset)
 866                 goto out;
 867
 868         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 869 write:
 870         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 871                 goto redirty_out;
 872         if (f2fs_is_drop_cache(inode))
 873                 goto out;
 874         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
 875                         available_free_memory(sbi, BASE_CHECK))
 876                 goto redirty_out;
 877
 878         /* Dentry blocks are controlled by checkpoint */
 879         if (S_ISDIR(inode->i_mode)) {
 880                 if (unlikely(f2fs_cp_error(sbi)))
 881                         goto redirty_out;
 882                 err = do_write_data_page(page, &fio);
 883                 goto done;
 884         }
 885
 886         /* we should bypass data pages to proceed the kworkder jobs */
 887         if (unlikely(f2fs_cp_error(sbi))) {
 888                 SetPageError(page);
 889                 goto out;
 890         }
 891
 892         if (!wbc->for_reclaim)
 893                 need_balance_fs = true;
 894         else if (has_not_enough_free_secs(sbi, 0))
 895                 goto redirty_out;
 896
 897         err = -EAGAIN;
 898         f2fs_lock_op(sbi);
 899         if (f2fs_has_inline_data(inode))
 900                 err = f2fs_write_inline_data(inode, page);
 901         if (err == -EAGAIN)
 902                 err = do_write_data_page(page, &fio);
 903         f2fs_unlock_op(sbi);
 904 done:
 905         if (err && err != -ENOENT)
 906                 goto redirty_out;
 907
 908         clear_cold_data(page);
 909 out:
 910         inode_dec_dirty_pages(inode);
 911         unlock_page(page);
 912         if (need_balance_fs)
 913                 f2fs_balance_fs(sbi);
 914         if (wbc->for_reclaim)
 915                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 916         return 0;
 917
 918 redirty_out:
 919         redirty_page_for_writepage(wbc, page);
 920         return AOP_WRITEPAGE_ACTIVATE;
 921 }
 922
 923 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
 924                         void *data)
 925 {
 926         struct address_space *mapping = data;
 927         int ret = mapping->a_ops->writepage(page, wbc);
 928         mapping_set_error(mapping, ret);
 929         return ret;
 930 }
 931
 932 static int f2fs_write_data_pages(struct address_space *mapping,
 933                             struct writeback_control *wbc)
 934 {
 935         struct inode *inode = mapping->host;
 936         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 937         bool locked = false;
 938         int ret;
 939         long diff;
 940
 941         trace_f2fs_writepages(mapping->host, wbc, DATA);
 942
 943         /* deal with chardevs and other special file */
 944         if (!mapping->a_ops->writepage)
 945                 return 0;
 946
 947         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
 948                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
 949                         available_free_memory(sbi, DIRTY_DENTS))
 950                 goto skip_write;
 951
 952         diff = nr_pages_to_write(sbi, DATA, wbc);
 953
 954         if (!S_ISDIR(inode->i_mode)) {
 955                 mutex_lock(&sbi->writepages);
 956                 locked = true;
 957         }
 958         ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
 959         if (locked)
 960                 mutex_unlock(&sbi->writepages);
 961
 962         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 963
 964         remove_dirty_dir_inode(inode);
 965
 966         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
 967         return ret;
 968
 969 skip_write:
 970         wbc->pages_skipped += get_dirty_pages(inode);
 971         return 0;
 972 }
 973
 974 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
 975 {
 976         struct inode *inode = mapping->host;
 977
 978         if (to > inode->i_size) {
 979                 truncate_pagecache(inode, inode->i_size);
 980                 truncate_blocks(inode, inode->i_size, true);
 981         }
 982 }
 983
 984 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
 985                 loff_t pos, unsigned len, unsigned flags,
 986                 struct page **pagep, void **fsdata)
 987 {
 988         struct inode *inode = mapping->host;
 989         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 990         struct page *page, *ipage;
 991         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
 992         struct dnode_of_data dn;
 993         int err = 0;
 994
 995         trace_f2fs_write_begin(inode, pos, len, flags);
 996
 997         f2fs_balance_fs(sbi);
 998
 999         /*
1000          * We should check this at this moment to avoid deadlock on inode page
1001          * and #0 page. The locking rule for inline_data conversion should be:
1002          * lock_page(page #0) -> lock_page(inode_page)
1003          */
1004         if (index != 0) {
1005                 err = f2fs_convert_inline_inode(inode);
1006                 if (err)
1007                         goto fail;
1008         }
1009 repeat:
1010         page = grab_cache_page_write_begin(mapping, index, flags);
1011         if (!page) {
1012                 err = -ENOMEM;
1013                 goto fail;
1014         }
1015
1016         *pagep = page;
1017
1018         f2fs_lock_op(sbi);
1019
1020         /* check inline_data */
1021         ipage = get_node_page(sbi, inode->i_ino);
1022         if (IS_ERR(ipage)) {
1023                 err = PTR_ERR(ipage);
1024                 goto unlock_fail;
1025         }
1026
1027         set_new_dnode(&dn, inode, ipage, ipage, 0);
1028
1029         if (f2fs_has_inline_data(inode)) {
1030                 if (pos + len <= MAX_INLINE_DATA) {
1031                         read_inline_data(page, ipage);
1032                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1033                         sync_inode_page(&dn);
1034                         goto put_next;
1035                 }
1036                 err = f2fs_convert_inline_page(&dn, page);
1037                 if (err)
1038                         goto put_fail;
1039         }
1040         err = f2fs_reserve_block(&dn, index);
1041         if (err)
1042                 goto put_fail;
1043 put_next:
1044         f2fs_put_dnode(&dn);
1045         f2fs_unlock_op(sbi);
1046
1047         if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1048                 return 0;
1049
1050         f2fs_wait_on_page_writeback(page, DATA);
1051
1052         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1053                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1054                 unsigned end = start + len;
1055
1056                 /* Reading beyond i_size is simple: memset to zero */
1057                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1058                 goto out;
1059         }
1060
1061         if (dn.data_blkaddr == NEW_ADDR) {
1062                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1063         } else {
1064                 struct f2fs_io_info fio = {
1065                         .type = DATA,
1066                         .rw = READ_SYNC,
1067                         .blk_addr = dn.data_blkaddr,
1068                 };
1069                 err = f2fs_submit_page_bio(sbi, page, &fio);
1070                 if (err)
1071                         goto fail;
1072
1073                 lock_page(page);
1074                 if (unlikely(!PageUptodate(page))) {
1075                         f2fs_put_page(page, 1);
1076                         err = -EIO;
1077                         goto fail;
1078                 }
1079                 if (unlikely(page->mapping != mapping)) {
1080                         f2fs_put_page(page, 1);
1081                         goto repeat;
1082                 }
1083         }
1084 out:
1085         SetPageUptodate(page);
1086         clear_cold_data(page);
1087         return 0;
1088
1089 put_fail:
1090         f2fs_put_dnode(&dn);
1091 unlock_fail:
1092         f2fs_unlock_op(sbi);
1093         f2fs_put_page(page, 1);
1094 fail:
1095         f2fs_write_failed(mapping, pos + len);
1096         return err;
1097 }
1098
1099 static int f2fs_write_end(struct file *file,
1100                         struct address_space *mapping,
1101                         loff_t pos, unsigned len, unsigned copied,
1102                         struct page *page, void *fsdata)
1103 {
1104         struct inode *inode = page->mapping->host;
1105
1106         trace_f2fs_write_end(inode, pos, len, copied);
1107
1108         set_page_dirty(page);
1109
1110         if (pos + copied > i_size_read(inode)) {
1111                 i_size_write(inode, pos + copied);
1112                 mark_inode_dirty(inode);
1113                 update_inode_page(inode);
1114         }
1115
1116         f2fs_put_page(page, 1);
1117         return copied;
1118 }
1119
1120 static int check_direct_IO(struct inode *inode, int rw,
1121                 struct iov_iter *iter, loff_t offset)
1122 {
1123         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1124
1125         if (rw == READ)
1126                 return 0;
1127
1128         if (offset & blocksize_mask)
1129                 return -EINVAL;
1130
1131         if (iov_iter_alignment(iter) & blocksize_mask)
1132                 return -EINVAL;
1133
1134         return 0;
1135 }
1136
1137 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1138                 struct iov_iter *iter, loff_t offset)
1139 {
1140         struct file *file = iocb->ki_filp;
1141         struct address_space *mapping = file->f_mapping;
1142         struct inode *inode = mapping->host;
1143         size_t count = iov_iter_count(iter);
1144         int err;
1145
1146         /* we don't need to use inline_data strictly */
1147         if (f2fs_has_inline_data(inode)) {
1148                 err = f2fs_convert_inline_inode(inode);
1149                 if (err)
1150                         return err;
1151         }
1152
1153         if (check_direct_IO(inode, rw, iter, offset))
1154                 return 0;
1155
1156         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1157
1158         if (rw & WRITE)
1159                 __allocate_data_blocks(inode, offset, count);
1160
1161         err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block);
1162         if (err < 0 && (rw & WRITE))
1163                 f2fs_write_failed(mapping, offset + count);
1164
1165         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1166
1167         return err;
1168 }
1169
1170 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1171                                                         unsigned int length)
1172 {
1173         struct inode *inode = page->mapping->host;
1174         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1175
1176         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1177                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1178                 return;
1179
1180         if (PageDirty(page)) {
1181                 if (inode->i_ino == F2FS_META_INO(sbi))
1182                         dec_page_count(sbi, F2FS_DIRTY_META);
1183                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1184                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1185                 else
1186                         inode_dec_dirty_pages(inode);
1187         }
1188         ClearPagePrivate(page);
1189 }
1190
1191 int f2fs_release_page(struct page *page, gfp_t wait)
1192 {
1193         /* If this is dirty page, keep PagePrivate */
1194         if (PageDirty(page))
1195                 return 0;
1196
1197         ClearPagePrivate(page);
1198         return 1;
1199 }
1200
1201 static int f2fs_set_data_page_dirty(struct page *page)
1202 {
1203         struct address_space *mapping = page->mapping;
1204         struct inode *inode = mapping->host;
1205
1206         trace_f2fs_set_page_dirty(page, DATA);
1207
1208         SetPageUptodate(page);
1209
1210         if (f2fs_is_atomic_file(inode)) {
1211                 register_inmem_page(inode, page);
1212                 return 1;
1213         }
1214
1215         mark_inode_dirty(inode);
1216
1217         if (!PageDirty(page)) {
1218                 __set_page_dirty_nobuffers(page);
1219                 update_dirty_page(inode, page);
1220                 return 1;
1221         }
1222         return 0;
1223 }
1224
1225 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1226 {
1227         struct inode *inode = mapping->host;
1228
1229         /* we don't need to use inline_data strictly */
1230         if (f2fs_has_inline_data(inode)) {
1231                 int err = f2fs_convert_inline_inode(inode);
1232                 if (err)
1233                         return err;
1234         }
1235         return generic_block_bmap(mapping, block, get_data_block);
1236 }
1237
1238 const struct address_space_operations f2fs_dblock_aops = {
1239         .readpage       = f2fs_read_data_page,
1240         .readpages      = f2fs_read_data_pages,
1241         .writepage      = f2fs_write_data_page,
1242         .writepages     = f2fs_write_data_pages,
1243         .write_begin    = f2fs_write_begin,
1244         .write_end      = f2fs_write_end,
1245         .set_page_dirty = f2fs_set_data_page_dirty,
1246         .invalidatepage = f2fs_invalidate_page,
1247         .releasepage    = f2fs_release_page,
1248         .direct_IO      = f2fs_direct_IO,
1249         .bmap           = f2fs_bmap,
1250 };