fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/swap.h>
  18 #include <linux/prefetch.h>
  19 #include <linux/uio.h>
  20 #include <linux/cleancache.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 #include <trace/events/android_fs.h>
  28
  29 #define NUM_PREALLOC_POST_READ_CTXS     128
  30
  31 static struct kmem_cache *bio_post_read_ctx_cache;
  32 static mempool_t *bio_post_read_ctx_pool;
  33
  34 static bool __is_cp_guaranteed(struct page *page)
  35 {
  36         struct address_space *mapping = page->mapping;
  37         struct inode *inode;
  38         struct f2fs_sb_info *sbi;
  39
  40         if (!mapping)
  41                 return false;
  42
  43         inode = mapping->host;
  44         sbi = F2FS_I_SB(inode);
  45
  46         if (inode->i_ino == F2FS_META_INO(sbi) ||
  47                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  48                         S_ISDIR(inode->i_mode) ||
  49                         (S_ISREG(inode->i_mode) &&
  50                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  51                         is_cold_data(page))
  52                 return true;
  53         return false;
  54 }
  55
  56 static enum count_type __read_io_type(struct page *page)
  57 {
  58         struct address_space *mapping = page_file_mapping(page);
  59
  60         if (mapping) {
  61                 struct inode *inode = mapping->host;
  62                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  63
  64                 if (inode->i_ino == F2FS_META_INO(sbi))
  65                         return F2FS_RD_META;
  66
  67                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  68                         return F2FS_RD_NODE;
  69         }
  70         return F2FS_RD_DATA;
  71 }
  72
  73 /* postprocessing steps for read bios */
  74 enum bio_post_read_step {
  75         STEP_INITIAL = 0,
  76         STEP_DECRYPT,
  77 };
  78
  79 struct bio_post_read_ctx {
  80         struct bio *bio;
  81         struct work_struct work;
  82         unsigned int cur_step;
  83         unsigned int enabled_steps;
  84 };
  85
  86 static void __read_end_io(struct bio *bio)
  87 {
  88         struct page *page;
  89         struct bio_vec *bv;
  90         int i;
  91
  92         bio_for_each_segment_all(bv, bio, i) {
  93                 page = bv->bv_page;
  94
  95                 /* PG_error was set if any post_read step failed */
  96                 if (bio->bi_error || PageError(page)) {
  97                         ClearPageUptodate(page);
  98                         /* will re-read again later */
  99                         ClearPageError(page);
 100                 } else {
 101                         SetPageUptodate(page);
 102                 }
 103                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 104                 unlock_page(page);
 105         }
 106         if (bio->bi_private)
 107                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 108         bio_put(bio);
 109 }
 110
 111 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 112
 113 static void decrypt_work(struct work_struct *work)
 114 {
 115         struct bio_post_read_ctx *ctx =
 116                 container_of(work, struct bio_post_read_ctx, work);
 117
 118         fscrypt_decrypt_bio(ctx->bio);
 119
 120         bio_post_read_processing(ctx);
 121 }
 122
 123 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 124 {
 125         switch (++ctx->cur_step) {
 126         case STEP_DECRYPT:
 127                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 128                         INIT_WORK(&ctx->work, decrypt_work);
 129                         fscrypt_enqueue_decrypt_work(&ctx->work);
 130                         return;
 131                 }
 132                 ctx->cur_step++;
 133                 /* fall-through */
 134         default:
 135                 __read_end_io(ctx->bio);
 136         }
 137 }
 138
 139 static bool f2fs_bio_post_read_required(struct bio *bio)
 140 {
 141         return bio->bi_private && !bio->bi_error;
 142 }
 143
 144 static void f2fs_read_end_io(struct bio *bio)
 145 {
 146         struct page *first_page = bio->bi_io_vec[0].bv_page;
 147
 148         if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_READ_IO)) {
 149                 f2fs_show_injection_info(FAULT_READ_IO);
 150                 bio->bi_error = -EIO;
 151         }
 152
 153         if (f2fs_bio_post_read_required(bio)) {
 154                 struct bio_post_read_ctx *ctx = bio->bi_private;
 155
 156                 ctx->cur_step = STEP_INITIAL;
 157                 bio_post_read_processing(ctx);
 158                 return;
 159         }
 160
 161         if (first_page != NULL &&
 162                 __read_io_type(first_page) == F2FS_RD_DATA) {
 163                 trace_android_fs_dataread_end(first_page->mapping->host,
 164                                                 page_offset(first_page),
 165                                                 bio->bi_iter.bi_size);
 166         }
 167
 168         __read_end_io(bio);
 169 }
 170
 171 static void f2fs_write_end_io(struct bio *bio)
 172 {
 173         struct f2fs_sb_info *sbi = bio->bi_private;
 174         struct bio_vec *bvec;
 175         int i;
 176
 177         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 178                 f2fs_show_injection_info(FAULT_WRITE_IO);
 179                 bio->bi_error = -EIO;
 180         }
 181
 182         bio_for_each_segment_all(bvec, bio, i) {
 183                 struct page *page = bvec->bv_page;
 184                 enum count_type type = WB_DATA_TYPE(page);
 185
 186                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 187                         set_page_private(page, (unsigned long)NULL);
 188                         ClearPagePrivate(page);
 189                         unlock_page(page);
 190                         mempool_free(page, sbi->write_io_dummy);
 191
 192                         if (unlikely(bio->bi_error))
 193                                 f2fs_stop_checkpoint(sbi, true);
 194                         continue;
 195                 }
 196
 197                 fscrypt_pullback_bio_page(&page, true);
 198
 199                 if (unlikely(bio->bi_error)) {
 200                         set_bit(AS_EIO, &page->mapping->flags);
 201                         if (type == F2FS_WB_CP_DATA)
 202                                 f2fs_stop_checkpoint(sbi, true);
 203                 }
 204
 205                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 206                                         page->index != nid_of_node(page));
 207
 208                 dec_page_count(sbi, type);
 209                 if (f2fs_in_warm_node_list(sbi, page))
 210                         f2fs_del_fsync_node_entry(sbi, page);
 211                 clear_cold_data(page);
 212                 end_page_writeback(page);
 213         }
 214         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 215                                 wq_has_sleeper(&sbi->cp_wait))
 216                 wake_up(&sbi->cp_wait);
 217
 218         bio_put(bio);
 219 }
 220
 221 /*
 222  * Return true, if pre_bio's bdev is same as its target device.
 223  */
 224 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 225                                 block_t blk_addr, struct bio *bio)
 226 {
 227         struct block_device *bdev = sbi->sb->s_bdev;
 228         int i;
 229
 230         if (f2fs_is_multi_device(sbi)) {
 231                 for (i = 0; i < sbi->s_ndevs; i++) {
 232                         if (FDEV(i).start_blk <= blk_addr &&
 233                             FDEV(i).end_blk >= blk_addr) {
 234                                 blk_addr -= FDEV(i).start_blk;
 235                                 bdev = FDEV(i).bdev;
 236                                 break;
 237                         }
 238                 }
 239         }
 240         if (bio) {
 241                 bio->bi_bdev = bdev;
 242                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 243         }
 244         return bdev;
 245 }
 246
 247 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 248 {
 249         int i;
 250
 251         if (!f2fs_is_multi_device(sbi))
 252                 return 0;
 253
 254         for (i = 0; i < sbi->s_ndevs; i++)
 255                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 256                         return i;
 257         return 0;
 258 }
 259
 260 static bool __same_bdev(struct f2fs_sb_info *sbi,
 261                                 block_t blk_addr, struct bio *bio)
 262 {
 263         return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
 264 }
 265
 266 /*
 267  * Low-level block read/write IO operations.
 268  */
 269 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 270 {
 271         struct f2fs_sb_info *sbi = fio->sbi;
 272         struct bio *bio;
 273
 274         bio = f2fs_bio_alloc(sbi, npages, true);
 275
 276         f2fs_target_device(sbi, fio->new_blkaddr, bio);
 277         if (is_read_io(fio->op)) {
 278                 bio->bi_end_io = f2fs_read_end_io;
 279                 bio->bi_private = NULL;
 280         } else {
 281                 bio->bi_end_io = f2fs_write_end_io;
 282                 bio->bi_private = sbi;
 283                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 284                                                 fio->type, fio->temp);
 285         }
 286         if (fio->io_wbc)
 287                 wbc_init_bio(fio->io_wbc, bio);
 288
 289         return bio;
 290 }
 291
 292 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 293                                 struct bio *bio, enum page_type type)
 294 {
 295         if (!is_read_io(bio_op(bio))) {
 296                 unsigned int start;
 297
 298                 if (type != DATA && type != NODE)
 299                         goto submit_io;
 300
 301                 if (test_opt(sbi, LFS) && current->plug)
 302                         blk_finish_plug(current->plug);
 303
 304                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 305                 start %= F2FS_IO_SIZE(sbi);
 306
 307                 if (start == 0)
 308                         goto submit_io;
 309
 310                 /* fill dummy pages */
 311                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 312                         struct page *page =
 313                                 mempool_alloc(sbi->write_io_dummy,
 314                                               GFP_NOIO | __GFP_NOFAIL);
 315                         f2fs_bug_on(sbi, !page);
 316
 317                         zero_user_segment(page, 0, PAGE_SIZE);
 318                         SetPagePrivate(page);
 319                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 320                         lock_page(page);
 321                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 322                                 f2fs_bug_on(sbi, 1);
 323                 }
 324                 /*
 325                  * In the NODE case, we lose next block address chain. So, we
 326                  * need to do checkpoint in f2fs_sync_file.
 327                  */
 328                 if (type == NODE)
 329                         set_sbi_flag(sbi, SBI_NEED_CP);
 330         }
 331 submit_io:
 332         if (is_read_io(bio_op(bio)))
 333                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 334         else
 335                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 336         submit_bio(bio_op(bio), bio);
 337 }
 338
 339 static void __f2fs_submit_read_bio(struct f2fs_sb_info *sbi,
 340                                 struct bio *bio, enum page_type type)
 341 {
 342         if (trace_android_fs_dataread_start_enabled() && (type == DATA)) {
 343                 struct page *first_page = bio->bi_io_vec[0].bv_page;
 344
 345                 if (first_page != NULL &&
 346                         __read_io_type(first_page) == F2FS_RD_DATA) {
 347                         char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
 348
 349                         path = android_fstrace_get_pathname(pathbuf,
 350                                                 MAX_TRACE_PATHBUF_LEN,
 351                                                 first_page->mapping->host);
 352
 353                         trace_android_fs_dataread_start(
 354                                 first_page->mapping->host,
 355                                 page_offset(first_page),
 356                                 bio->bi_iter.bi_size,
 357                                 current->pid,
 358                                 path,
 359                                 current->comm);
 360                 }
 361         }
 362         __submit_bio(sbi, bio, type);
 363 }
 364
 365 static void __submit_merged_bio(struct f2fs_bio_info *io)
 366 {
 367         struct f2fs_io_info *fio = &io->fio;
 368
 369         if (!io->bio)
 370                 return;
 371
 372         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 373
 374         if (is_read_io(fio->op))
 375                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 376         else
 377                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 378
 379         __submit_bio(io->sbi, io->bio, fio->type);
 380         io->bio = NULL;
 381 }
 382
 383 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 384                                                 struct page *page, nid_t ino)
 385 {
 386         struct bio_vec *bvec;
 387         struct page *target;
 388         int i;
 389
 390         if (!bio)
 391                 return false;
 392
 393         if (!inode && !page && !ino)
 394                 return true;
 395
 396         bio_for_each_segment_all(bvec, bio, i) {
 397
 398                 if (bvec->bv_page->mapping)
 399                         target = bvec->bv_page;
 400                 else
 401                         target = fscrypt_control_page(bvec->bv_page);
 402
 403                 if (inode && inode == target->mapping->host)
 404                         return true;
 405                 if (page && page == target)
 406                         return true;
 407                 if (ino && ino == ino_of_node(target))
 408                         return true;
 409         }
 410
 411         return false;
 412 }
 413
 414 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 415                                 enum page_type type, enum temp_type temp)
 416 {
 417         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 418         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 419
 420         down_write(&io->io_rwsem);
 421
 422         /* change META to META_FLUSH in the checkpoint procedure */
 423         if (type >= META_FLUSH) {
 424                 io->fio.type = META_FLUSH;
 425                 io->fio.op = REQ_OP_WRITE;
 426                 io->fio.op_flags = REQ_META | REQ_PRIO;
 427                 if (!test_opt(sbi, NOBARRIER))
 428                         io->fio.op_flags |= WRITE_FLUSH | REQ_FUA;
 429         }
 430         __submit_merged_bio(io);
 431         up_write(&io->io_rwsem);
 432 }
 433
 434 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 435                                 struct inode *inode, struct page *page,
 436                                 nid_t ino, enum page_type type, bool force)
 437 {
 438         enum temp_type temp;
 439         bool ret = true;
 440
 441         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 442                 if (!force)     {
 443                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 444                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 445
 446                         down_read(&io->io_rwsem);
 447                         ret = __has_merged_page(io->bio, inode, page, ino);
 448                         up_read(&io->io_rwsem);
 449                 }
 450                 if (ret)
 451                         __f2fs_submit_merged_write(sbi, type, temp);
 452
 453                 /* TODO: use HOT temp only for meta pages now. */
 454                 if (type >= META)
 455                         break;
 456         }
 457 }
 458
 459 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 460 {
 461         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 462 }
 463
 464 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 465                                 struct inode *inode, struct page *page,
 466                                 nid_t ino, enum page_type type)
 467 {
 468         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 469 }
 470
 471 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 472 {
 473         f2fs_submit_merged_write(sbi, DATA);
 474         f2fs_submit_merged_write(sbi, NODE);
 475         f2fs_submit_merged_write(sbi, META);
 476 }
 477
 478 /*
 479  * Fill the locked page with data located in the block address.
 480  * A caller needs to unlock the page on failure.
 481  */
 482 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 483 {
 484         struct bio *bio;
 485         struct page *page = fio->encrypted_page ?
 486                         fio->encrypted_page : fio->page;
 487
 488         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 489                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 490                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 491                 return -EFSCORRUPTED;
 492
 493         trace_f2fs_submit_page_bio(page, fio);
 494         f2fs_trace_ios(fio, 0);
 495
 496         /* Allocate a new bio */
 497         bio = __bio_alloc(fio, 1);
 498
 499         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 500                 bio_put(bio);
 501                 return -EFAULT;
 502         }
 503
 504         if (fio->io_wbc && !is_read_io(fio->op))
 505                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 506
 507         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 508
 509         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 510                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 511
 512         __f2fs_submit_read_bio(fio->sbi, bio, fio->type);
 513         return 0;
 514 }
 515
 516 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 517                                 block_t last_blkaddr, block_t cur_blkaddr)
 518 {
 519         if (last_blkaddr + 1 != cur_blkaddr)
 520                 return false;
 521         return __same_bdev(sbi, cur_blkaddr, bio);
 522 }
 523
 524 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 525                                                 struct f2fs_io_info *fio)
 526 {
 527         if (io->fio.op != fio->op)
 528                 return false;
 529         return io->fio.op_flags == fio->op_flags;
 530 }
 531
 532 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 533                                         struct f2fs_bio_info *io,
 534                                         struct f2fs_io_info *fio,
 535                                         block_t last_blkaddr,
 536                                         block_t cur_blkaddr)
 537 {
 538         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 539                 unsigned int filled_blocks =
 540                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 541                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 542                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 543
 544                 /* IOs in bio is aligned and left space of vectors is not enough */
 545                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 546                         return false;
 547         }
 548         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 549                 return false;
 550         return io_type_is_mergeable(io, fio);
 551 }
 552
 553 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 554 {
 555         struct bio *bio = *fio->bio;
 556         struct page *page = fio->encrypted_page ?
 557                         fio->encrypted_page : fio->page;
 558
 559         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 560                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 561                 return -EFSCORRUPTED;
 562
 563         trace_f2fs_submit_page_bio(page, fio);
 564         f2fs_trace_ios(fio, 0);
 565
 566         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 567                                                 fio->new_blkaddr)) {
 568                 __submit_bio(fio->sbi, bio, fio->type);
 569                 bio = NULL;
 570         }
 571 alloc_new:
 572         if (!bio) {
 573                 bio = __bio_alloc(fio, BIO_MAX_PAGES);
 574                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 575         }
 576
 577         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 578                 __submit_bio(fio->sbi, bio, fio->type);
 579                 bio = NULL;
 580                 goto alloc_new;
 581         }
 582
 583         if (fio->io_wbc)
 584                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 585
 586         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 587
 588         *fio->last_block = fio->new_blkaddr;
 589         *fio->bio = bio;
 590
 591         return 0;
 592 }
 593
 594 static void f2fs_submit_ipu_bio(struct f2fs_sb_info *sbi, struct bio **bio,
 595                                                         struct page *page)
 596 {
 597         if (!bio)
 598                 return;
 599
 600         if (!__has_merged_page(*bio, NULL, page, 0))
 601                 return;
 602
 603         __submit_bio(sbi, *bio, DATA);
 604         *bio = NULL;
 605 }
 606
 607 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 608 {
 609         struct f2fs_sb_info *sbi = fio->sbi;
 610         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 611         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 612         struct page *bio_page;
 613
 614         f2fs_bug_on(sbi, is_read_io(fio->op));
 615
 616         down_write(&io->io_rwsem);
 617 next:
 618         if (fio->in_list) {
 619                 spin_lock(&io->io_lock);
 620                 if (list_empty(&io->io_list)) {
 621                         spin_unlock(&io->io_lock);
 622                         goto out;
 623                 }
 624                 fio = list_first_entry(&io->io_list,
 625                                                 struct f2fs_io_info, list);
 626                 list_del(&fio->list);
 627                 spin_unlock(&io->io_lock);
 628         }
 629
 630         verify_fio_blkaddr(fio);
 631
 632         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 633
 634         /* set submitted = true as a return value */
 635         fio->submitted = true;
 636
 637         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 638
 639         if (io->bio && !io_is_mergeable(sbi, io->bio, io, fio,
 640                         io->last_block_in_bio, fio->new_blkaddr))
 641                 __submit_merged_bio(io);
 642 alloc_new:
 643         if (io->bio == NULL) {
 644                 if ((fio->type == DATA || fio->type == NODE) &&
 645                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 646                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 647                         fio->retry = true;
 648                         goto skip;
 649                 }
 650                 io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
 651                 io->fio = *fio;
 652         }
 653
 654         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 655                 __submit_merged_bio(io);
 656                 goto alloc_new;
 657         }
 658
 659         if (fio->io_wbc)
 660                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 661
 662         io->last_block_in_bio = fio->new_blkaddr;
 663         f2fs_trace_ios(fio, 0);
 664
 665         trace_f2fs_submit_page_write(fio->page, fio);
 666 skip:
 667         if (fio->in_list)
 668                 goto next;
 669 out:
 670         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 671                                 !f2fs_is_checkpoint_ready(sbi))
 672                 __submit_merged_bio(io);
 673         up_write(&io->io_rwsem);
 674 }
 675
 676 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 677                                         unsigned nr_pages, unsigned op_flag)
 678 {
 679         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 680         struct bio *bio;
 681         struct bio_post_read_ctx *ctx;
 682         unsigned int post_read_steps = 0;
 683
 684         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 685         if (!bio)
 686                 return ERR_PTR(-ENOMEM);
 687         f2fs_target_device(sbi, blkaddr, bio);
 688         bio->bi_end_io = f2fs_read_end_io;
 689         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 690
 691         if (f2fs_encrypted_file(inode))
 692                 post_read_steps |= 1 << STEP_DECRYPT;
 693         if (post_read_steps) {
 694                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 695                 if (!ctx) {
 696                         bio_put(bio);
 697                         return ERR_PTR(-ENOMEM);
 698                 }
 699                 ctx->bio = bio;
 700                 ctx->enabled_steps = post_read_steps;
 701                 bio->bi_private = ctx;
 702         }
 703
 704         return bio;
 705 }
 706
 707 /* This can handle encryption stuffs */
 708 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 709                                                         block_t blkaddr)
 710 {
 711         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 712         struct bio *bio;
 713
 714         bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 715         if (IS_ERR(bio))
 716                 return PTR_ERR(bio);
 717
 718         /* wait for GCed page writeback via META_MAPPING */
 719         f2fs_wait_on_block_writeback(inode, blkaddr);
 720
 721         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 722                 bio_put(bio);
 723                 return -EFAULT;
 724         }
 725         ClearPageError(page);
 726         inc_page_count(sbi, F2FS_RD_DATA);
 727         __f2fs_submit_read_bio(sbi, bio, DATA);
 728         return 0;
 729 }
 730
 731 static void __set_data_blkaddr(struct dnode_of_data *dn)
 732 {
 733         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 734         __le32 *addr_array;
 735         int base = 0;
 736
 737         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 738                 base = get_extra_isize(dn->inode);
 739
 740         /* Get physical address of data block */
 741         addr_array = blkaddr_in_node(rn);
 742         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 743 }
 744
 745 /*
 746  * Lock ordering for the change of data block address:
 747  * ->data_page
 748  *  ->node_page
 749  *    update block addresses in the node page
 750  */
 751 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 752 {
 753         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 754         __set_data_blkaddr(dn);
 755         if (set_page_dirty(dn->node_page))
 756                 dn->node_changed = true;
 757 }
 758
 759 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 760 {
 761         dn->data_blkaddr = blkaddr;
 762         f2fs_set_data_blkaddr(dn);
 763         f2fs_update_extent_cache(dn);
 764 }
 765
 766 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 767 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 768 {
 769         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 770         int err;
 771
 772         if (!count)
 773                 return 0;
 774
 775         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 776                 return -EPERM;
 777         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 778                 return err;
 779
 780         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 781                                                 dn->ofs_in_node, count);
 782
 783         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 784
 785         for (; count > 0; dn->ofs_in_node++) {
 786                 block_t blkaddr = datablock_addr(dn->inode,
 787                                         dn->node_page, dn->ofs_in_node);
 788                 if (blkaddr == NULL_ADDR) {
 789                         dn->data_blkaddr = NEW_ADDR;
 790                         __set_data_blkaddr(dn);
 791                         count--;
 792                 }
 793         }
 794
 795         if (set_page_dirty(dn->node_page))
 796                 dn->node_changed = true;
 797         return 0;
 798 }
 799
 800 /* Should keep dn->ofs_in_node unchanged */
 801 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 802 {
 803         unsigned int ofs_in_node = dn->ofs_in_node;
 804         int ret;
 805
 806         ret = f2fs_reserve_new_blocks(dn, 1);
 807         dn->ofs_in_node = ofs_in_node;
 808         return ret;
 809 }
 810
 811 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 812 {
 813         bool need_put = dn->inode_page ? false : true;
 814         int err;
 815
 816         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 817         if (err)
 818                 return err;
 819
 820         if (dn->data_blkaddr == NULL_ADDR)
 821                 err = f2fs_reserve_new_block(dn);
 822         if (err || need_put)
 823                 f2fs_put_dnode(dn);
 824         return err;
 825 }
 826
 827 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 828 {
 829         struct extent_info ei  = {0,0,0};
 830         struct inode *inode = dn->inode;
 831
 832         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 833                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 834                 return 0;
 835         }
 836
 837         return f2fs_reserve_block(dn, index);
 838 }
 839
 840 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 841                                                 int op_flags, bool for_write)
 842 {
 843         struct address_space *mapping = inode->i_mapping;
 844         struct dnode_of_data dn;
 845         struct page *page;
 846         struct extent_info ei = {0,0,0};
 847         int err;
 848
 849         page = f2fs_grab_cache_page(mapping, index, for_write);
 850         if (!page)
 851                 return ERR_PTR(-ENOMEM);
 852
 853         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 854                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 855                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
 856                                                 DATA_GENERIC_ENHANCE_READ)) {
 857                         err = -EFSCORRUPTED;
 858                         goto put_err;
 859                 }
 860                 goto got_it;
 861         }
 862
 863         set_new_dnode(&dn, inode, NULL, NULL, 0);
 864         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 865         if (err)
 866                 goto put_err;
 867         f2fs_put_dnode(&dn);
 868
 869         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 870                 err = -ENOENT;
 871                 goto put_err;
 872         }
 873         if (dn.data_blkaddr != NEW_ADDR &&
 874                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
 875                                                 dn.data_blkaddr,
 876                                                 DATA_GENERIC_ENHANCE)) {
 877                 err = -EFSCORRUPTED;
 878                 goto put_err;
 879         }
 880 got_it:
 881         if (PageUptodate(page)) {
 882                 unlock_page(page);
 883                 return page;
 884         }
 885
 886         /*
 887          * A new dentry page is allocated but not able to be written, since its
 888          * new inode page couldn't be allocated due to -ENOSPC.
 889          * In such the case, its blkaddr can be remained as NEW_ADDR.
 890          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 891          * f2fs_init_inode_metadata.
 892          */
 893         if (dn.data_blkaddr == NEW_ADDR) {
 894                 zero_user_segment(page, 0, PAGE_SIZE);
 895                 if (!PageUptodate(page))
 896                         SetPageUptodate(page);
 897                 unlock_page(page);
 898                 return page;
 899         }
 900
 901         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 902         if (err)
 903                 goto put_err;
 904         return page;
 905
 906 put_err:
 907         f2fs_put_page(page, 1);
 908         return ERR_PTR(err);
 909 }
 910
 911 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 912 {
 913         struct address_space *mapping = inode->i_mapping;
 914         struct page *page;
 915
 916         page = find_get_page(mapping, index);
 917         if (page && PageUptodate(page))
 918                 return page;
 919         f2fs_put_page(page, 0);
 920
 921         page = f2fs_get_read_data_page(inode, index, REQ_SYNC, false);
 922         if (IS_ERR(page))
 923                 return page;
 924
 925         if (PageUptodate(page))
 926                 return page;
 927
 928         wait_on_page_locked(page);
 929         if (unlikely(!PageUptodate(page))) {
 930                 f2fs_put_page(page, 0);
 931                 return ERR_PTR(-EIO);
 932         }
 933         return page;
 934 }
 935
 936 /*
 937  * If it tries to access a hole, return an error.
 938  * Because, the callers, functions in dir.c and GC, should be able to know
 939  * whether this page exists or not.
 940  */
 941 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 942                                                         bool for_write)
 943 {
 944         struct address_space *mapping = inode->i_mapping;
 945         struct page *page;
 946 repeat:
 947         page = f2fs_get_read_data_page(inode, index, REQ_SYNC, for_write);
 948         if (IS_ERR(page))
 949                 return page;
 950
 951         /* wait for read completion */
 952         lock_page(page);
 953         if (unlikely(page->mapping != mapping)) {
 954                 f2fs_put_page(page, 1);
 955                 goto repeat;
 956         }
 957         if (unlikely(!PageUptodate(page))) {
 958                 f2fs_put_page(page, 1);
 959                 return ERR_PTR(-EIO);
 960         }
 961         return page;
 962 }
 963
 964 /*
 965  * Caller ensures that this data page is never allocated.
 966  * A new zero-filled data page is allocated in the page cache.
 967  *
 968  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 969  * f2fs_unlock_op().
 970  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 971  * ipage should be released by this function.
 972  */
 973 struct page *f2fs_get_new_data_page(struct inode *inode,
 974                 struct page *ipage, pgoff_t index, bool new_i_size)
 975 {
 976         struct address_space *mapping = inode->i_mapping;
 977         struct page *page;
 978         struct dnode_of_data dn;
 979         int err;
 980
 981         page = f2fs_grab_cache_page(mapping, index, true);
 982         if (!page) {
 983                 /*
 984                  * before exiting, we should make sure ipage will be released
 985                  * if any error occur.
 986                  */
 987                 f2fs_put_page(ipage, 1);
 988                 return ERR_PTR(-ENOMEM);
 989         }
 990
 991         set_new_dnode(&dn, inode, ipage, NULL, 0);
 992         err = f2fs_reserve_block(&dn, index);
 993         if (err) {
 994                 f2fs_put_page(page, 1);
 995                 return ERR_PTR(err);
 996         }
 997         if (!ipage)
 998                 f2fs_put_dnode(&dn);
 999
1000         if (PageUptodate(page))
1001                 goto got_it;
1002
1003         if (dn.data_blkaddr == NEW_ADDR) {
1004                 zero_user_segment(page, 0, PAGE_SIZE);
1005                 if (!PageUptodate(page))
1006                         SetPageUptodate(page);
1007         } else {
1008                 f2fs_put_page(page, 1);
1009
1010                 /* if ipage exists, blkaddr should be NEW_ADDR */
1011                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1012                 page = f2fs_get_lock_data_page(inode, index, true);
1013                 if (IS_ERR(page))
1014                         return page;
1015         }
1016 got_it:
1017         if (new_i_size && i_size_read(inode) <
1018                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1019                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1020         return page;
1021 }
1022
1023 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1024 {
1025         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1026         struct f2fs_summary sum;
1027         struct node_info ni;
1028         block_t old_blkaddr;
1029         blkcnt_t count = 1;
1030         int err;
1031
1032         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1033                 return -EPERM;
1034
1035         err = f2fs_get_node_info(sbi, dn->nid, &ni);
1036         if (err)
1037                 return err;
1038
1039         dn->data_blkaddr = datablock_addr(dn->inode,
1040                                 dn->node_page, dn->ofs_in_node);
1041         if (dn->data_blkaddr != NULL_ADDR)
1042                 goto alloc;
1043
1044         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1045                 return err;
1046
1047 alloc:
1048         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1049         old_blkaddr = dn->data_blkaddr;
1050         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1051                                         &sum, seg_type, NULL, false);
1052         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1053                 invalidate_mapping_pages(META_MAPPING(sbi),
1054                                         old_blkaddr, old_blkaddr);
1055         f2fs_set_data_blkaddr(dn);
1056
1057         /*
1058          * i_size will be updated by direct_IO. Otherwise, we'll get stale
1059          * data from unwritten block via dio_read.
1060          */
1061         return 0;
1062 }
1063
1064 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1065 {
1066         struct inode *inode = file_inode(iocb->ki_filp);
1067         struct f2fs_map_blocks map;
1068         int flag;
1069         int err = 0;
1070         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1071
1072         /* convert inline data for Direct I/O*/
1073         if (direct_io) {
1074                 err = f2fs_convert_inline_inode(inode);
1075                 if (err)
1076                         return err;
1077         }
1078
1079         if (direct_io && allow_outplace_dio(inode, iocb, from))
1080                 return 0;
1081
1082         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
1083                 return 0;
1084
1085         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1086         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1087         if (map.m_len > map.m_lblk)
1088                 map.m_len -= map.m_lblk;
1089         else
1090                 map.m_len = 0;
1091
1092         map.m_next_pgofs = NULL;
1093         map.m_next_extent = NULL;
1094         map.m_seg_type = NO_CHECK_TYPE;
1095         map.m_may_create = true;
1096
1097         if (direct_io) {
1098                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1099                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1100                                         F2FS_GET_BLOCK_PRE_AIO :
1101                                         F2FS_GET_BLOCK_PRE_DIO;
1102                 goto map_blocks;
1103         }
1104         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1105                 err = f2fs_convert_inline_inode(inode);
1106                 if (err)
1107                         return err;
1108         }
1109         if (f2fs_has_inline_data(inode))
1110                 return err;
1111
1112         flag = F2FS_GET_BLOCK_PRE_AIO;
1113
1114 map_blocks:
1115         err = f2fs_map_blocks(inode, &map, 1, flag);
1116         if (map.m_len > 0 && err == -ENOSPC) {
1117                 if (!direct_io)
1118                         set_inode_flag(inode, FI_NO_PREALLOC);
1119                 err = 0;
1120         }
1121         return err;
1122 }
1123
1124 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1125 {
1126         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1127                 if (lock)
1128                         down_read(&sbi->node_change);
1129                 else
1130                         up_read(&sbi->node_change);
1131         } else {
1132                 if (lock)
1133                         f2fs_lock_op(sbi);
1134                 else
1135                         f2fs_unlock_op(sbi);
1136         }
1137 }
1138
1139 /*
1140  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1141  * f2fs_map_blocks structure.
1142  * If original data blocks are allocated, then give them to blockdev.
1143  * Otherwise,
1144  *     a. preallocate requested block addresses
1145  *     b. do not use extent cache for better performance
1146  *     c. give the block addresses to blockdev
1147  */
1148 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1149                                                 int create, int flag)
1150 {
1151         unsigned int maxblocks = map->m_len;
1152         struct dnode_of_data dn;
1153         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1154         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1155         pgoff_t pgofs, end_offset, end;
1156         int err = 0, ofs = 1;
1157         unsigned int ofs_in_node, last_ofs_in_node;
1158         blkcnt_t prealloc;
1159         struct extent_info ei = {0,0,0};
1160         block_t blkaddr;
1161         unsigned int start_pgofs;
1162
1163         if (!maxblocks)
1164                 return 0;
1165
1166         map->m_len = 0;
1167         map->m_flags = 0;
1168
1169         /* it only supports block size == page size */
1170         pgofs = (pgoff_t)map->m_lblk;
1171         end = pgofs + maxblocks;
1172
1173         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1174                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1175                                                         map->m_may_create)
1176                         goto next_dnode;
1177
1178                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1179                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1180                 map->m_flags = F2FS_MAP_MAPPED;
1181                 if (map->m_next_extent)
1182                         *map->m_next_extent = pgofs + map->m_len;
1183
1184                 /* for hardware encryption, but to avoid potential issue in future */
1185                 if (flag == F2FS_GET_BLOCK_DIO)
1186                         f2fs_wait_on_block_writeback_range(inode,
1187                                                 map->m_pblk, map->m_len);
1188                 goto out;
1189         }
1190
1191 next_dnode:
1192         if (map->m_may_create)
1193                 __do_map_lock(sbi, flag, true);
1194
1195         /* When reading holes, we need its node page */
1196         set_new_dnode(&dn, inode, NULL, NULL, 0);
1197         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1198         if (err) {
1199                 if (flag == F2FS_GET_BLOCK_BMAP)
1200                         map->m_pblk = 0;
1201                 if (err == -ENOENT) {
1202                         err = 0;
1203                         if (map->m_next_pgofs)
1204                                 *map->m_next_pgofs =
1205                                         f2fs_get_next_page_offset(&dn, pgofs);
1206                         if (map->m_next_extent)
1207                                 *map->m_next_extent =
1208                                         f2fs_get_next_page_offset(&dn, pgofs);
1209                 }
1210                 goto unlock_out;
1211         }
1212
1213         start_pgofs = pgofs;
1214         prealloc = 0;
1215         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1216         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1217
1218 next_block:
1219         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1220
1221         if (__is_valid_data_blkaddr(blkaddr) &&
1222                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1223                 err = -EFSCORRUPTED;
1224                 goto sync_out;
1225         }
1226
1227         if (__is_valid_data_blkaddr(blkaddr)) {
1228                 /* use out-place-update for driect IO under LFS mode */
1229                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1230                                                         map->m_may_create) {
1231                         err = __allocate_data_block(&dn, map->m_seg_type);
1232                         if (!err) {
1233                                 blkaddr = dn.data_blkaddr;
1234                                 set_inode_flag(inode, FI_APPEND_WRITE);
1235                         }
1236                 }
1237         } else {
1238                 if (create) {
1239                         if (unlikely(f2fs_cp_error(sbi))) {
1240                                 err = -EIO;
1241                                 goto sync_out;
1242                         }
1243                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1244                                 if (blkaddr == NULL_ADDR) {
1245                                         prealloc++;
1246                                         last_ofs_in_node = dn.ofs_in_node;
1247                                 }
1248                         } else {
1249                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1250                                         flag != F2FS_GET_BLOCK_DIO);
1251                                 err = __allocate_data_block(&dn,
1252                                                         map->m_seg_type);
1253                                 if (!err)
1254                                         set_inode_flag(inode, FI_APPEND_WRITE);
1255                         }
1256                         if (err)
1257                                 goto sync_out;
1258                         map->m_flags |= F2FS_MAP_NEW;
1259                         blkaddr = dn.data_blkaddr;
1260                 } else {
1261                         if (flag == F2FS_GET_BLOCK_BMAP) {
1262                                 map->m_pblk = 0;
1263                                 goto sync_out;
1264                         }
1265                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1266                                 goto sync_out;
1267                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1268                                                 blkaddr == NULL_ADDR) {
1269                                 if (map->m_next_pgofs)
1270                                         *map->m_next_pgofs = pgofs + 1;
1271                                 goto sync_out;
1272                         }
1273                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1274                                 /* for defragment case */
1275                                 if (map->m_next_pgofs)
1276                                         *map->m_next_pgofs = pgofs + 1;
1277                                 goto sync_out;
1278                         }
1279                 }
1280         }
1281
1282         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1283                 goto skip;
1284
1285         if (map->m_len == 0) {
1286                 /* preallocated unwritten block should be mapped for fiemap. */
1287                 if (blkaddr == NEW_ADDR)
1288                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1289                 map->m_flags |= F2FS_MAP_MAPPED;
1290
1291                 map->m_pblk = blkaddr;
1292                 map->m_len = 1;
1293         } else if ((map->m_pblk != NEW_ADDR &&
1294                         blkaddr == (map->m_pblk + ofs)) ||
1295                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1296                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1297                 ofs++;
1298                 map->m_len++;
1299         } else {
1300                 goto sync_out;
1301         }
1302
1303 skip:
1304         dn.ofs_in_node++;
1305         pgofs++;
1306
1307         /* preallocate blocks in batch for one dnode page */
1308         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1309                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1310
1311                 dn.ofs_in_node = ofs_in_node;
1312                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1313                 if (err)
1314                         goto sync_out;
1315
1316                 map->m_len += dn.ofs_in_node - ofs_in_node;
1317                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1318                         err = -ENOSPC;
1319                         goto sync_out;
1320                 }
1321                 dn.ofs_in_node = end_offset;
1322         }
1323
1324         if (pgofs >= end)
1325                 goto sync_out;
1326         else if (dn.ofs_in_node < end_offset)
1327                 goto next_block;
1328
1329         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1330                 if (map->m_flags & F2FS_MAP_MAPPED) {
1331                         unsigned int ofs = start_pgofs - map->m_lblk;
1332
1333                         f2fs_update_extent_cache_range(&dn,
1334                                 start_pgofs, map->m_pblk + ofs,
1335                                 map->m_len - ofs);
1336                 }
1337         }
1338
1339         f2fs_put_dnode(&dn);
1340
1341         if (map->m_may_create) {
1342                 __do_map_lock(sbi, flag, false);
1343                 f2fs_balance_fs(sbi, dn.node_changed);
1344         }
1345         goto next_dnode;
1346
1347 sync_out:
1348
1349         /* for hardware encryption, but to avoid potential issue in future */
1350         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1351                 f2fs_wait_on_block_writeback_range(inode,
1352                                                 map->m_pblk, map->m_len);
1353
1354         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1355                 if (map->m_flags & F2FS_MAP_MAPPED) {
1356                         unsigned int ofs = start_pgofs - map->m_lblk;
1357
1358                         f2fs_update_extent_cache_range(&dn,
1359                                 start_pgofs, map->m_pblk + ofs,
1360                                 map->m_len - ofs);
1361                 }
1362                 if (map->m_next_extent)
1363                         *map->m_next_extent = pgofs + 1;
1364         }
1365         f2fs_put_dnode(&dn);
1366 unlock_out:
1367         if (map->m_may_create) {
1368                 __do_map_lock(sbi, flag, false);
1369                 f2fs_balance_fs(sbi, dn.node_changed);
1370         }
1371 out:
1372         trace_f2fs_map_blocks(inode, map, err);
1373         return err;
1374 }
1375
1376 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1377 {
1378         struct f2fs_map_blocks map;
1379         block_t last_lblk;
1380         int err;
1381
1382         if (pos + len > i_size_read(inode))
1383                 return false;
1384
1385         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1386         map.m_next_pgofs = NULL;
1387         map.m_next_extent = NULL;
1388         map.m_seg_type = NO_CHECK_TYPE;
1389         map.m_may_create = false;
1390         last_lblk = F2FS_BLK_ALIGN(pos + len);
1391
1392         while (map.m_lblk < last_lblk) {
1393                 map.m_len = last_lblk - map.m_lblk;
1394                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1395                 if (err || map.m_len == 0)
1396                         return false;
1397                 map.m_lblk += map.m_len;
1398         }
1399         return true;
1400 }
1401
1402 static int __get_data_block(struct inode *inode, sector_t iblock,
1403                         struct buffer_head *bh, int create, int flag,
1404                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1405 {
1406         struct f2fs_map_blocks map;
1407         int err;
1408
1409         map.m_lblk = iblock;
1410         map.m_len = bh->b_size >> inode->i_blkbits;
1411         map.m_next_pgofs = next_pgofs;
1412         map.m_next_extent = NULL;
1413         map.m_seg_type = seg_type;
1414         map.m_may_create = may_write;
1415
1416         err = f2fs_map_blocks(inode, &map, create, flag);
1417         if (!err) {
1418                 map_bh(bh, inode->i_sb, map.m_pblk);
1419                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1420                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1421         }
1422         return err;
1423 }
1424
1425 static int get_data_block(struct inode *inode, sector_t iblock,
1426                         struct buffer_head *bh_result, int create, int flag,
1427                         pgoff_t *next_pgofs)
1428 {
1429         return __get_data_block(inode, iblock, bh_result, create,
1430                                                         flag, next_pgofs,
1431                                                         NO_CHECK_TYPE, create);
1432 }
1433
1434 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1435                         struct buffer_head *bh_result, int create)
1436 {
1437         return __get_data_block(inode, iblock, bh_result, create,
1438                                 F2FS_GET_BLOCK_DIO, NULL,
1439                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1440                                 IS_SWAPFILE(inode) ? false : true);
1441 }
1442
1443 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1444                         struct buffer_head *bh_result, int create)
1445 {
1446         return __get_data_block(inode, iblock, bh_result, create,
1447                                 F2FS_GET_BLOCK_DIO, NULL,
1448                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1449                                 false);
1450 }
1451
1452 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1453                         struct buffer_head *bh_result, int create)
1454 {
1455         /* Block number less than F2FS MAX BLOCKS */
1456         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1457                 return -EFBIG;
1458
1459         return __get_data_block(inode, iblock, bh_result, create,
1460                                                 F2FS_GET_BLOCK_BMAP, NULL,
1461                                                 NO_CHECK_TYPE, create);
1462 }
1463
1464 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1465 {
1466         return (offset >> inode->i_blkbits);
1467 }
1468
1469 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1470 {
1471         return (blk << inode->i_blkbits);
1472 }
1473
1474 static int f2fs_xattr_fiemap(struct inode *inode,
1475                                 struct fiemap_extent_info *fieinfo)
1476 {
1477         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1478         struct page *page;
1479         struct node_info ni;
1480         __u64 phys = 0, len;
1481         __u32 flags;
1482         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1483         int err = 0;
1484
1485         if (f2fs_has_inline_xattr(inode)) {
1486                 int offset;
1487
1488                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1489                                                 inode->i_ino, false);
1490                 if (!page)
1491                         return -ENOMEM;
1492
1493                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1494                 if (err) {
1495                         f2fs_put_page(page, 1);
1496                         return err;
1497                 }
1498
1499                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1500                 offset = offsetof(struct f2fs_inode, i_addr) +
1501                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1502                                         get_inline_xattr_addrs(inode));
1503
1504                 phys += offset;
1505                 len = inline_xattr_size(inode);
1506
1507                 f2fs_put_page(page, 1);
1508
1509                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1510
1511                 if (!xnid)
1512                         flags |= FIEMAP_EXTENT_LAST;
1513
1514                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1515                 if (err || err == 1)
1516                         return err;
1517         }
1518
1519         if (xnid) {
1520                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1521                 if (!page)
1522                         return -ENOMEM;
1523
1524                 err = f2fs_get_node_info(sbi, xnid, &ni);
1525                 if (err) {
1526                         f2fs_put_page(page, 1);
1527                         return err;
1528                 }
1529
1530                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1531                 len = inode->i_sb->s_blocksize;
1532
1533                 f2fs_put_page(page, 1);
1534
1535                 flags = FIEMAP_EXTENT_LAST;
1536         }
1537
1538         if (phys)
1539                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1540
1541         return (err < 0 ? err : 0);
1542 }
1543
1544 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1545                 u64 start, u64 len)
1546 {
1547         struct buffer_head map_bh;
1548         sector_t start_blk, last_blk;
1549         pgoff_t next_pgofs;
1550         u64 logical = 0, phys = 0, size = 0;
1551         u32 flags = 0;
1552         int ret = 0;
1553
1554         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1555                 ret = f2fs_precache_extents(inode);
1556                 if (ret)
1557                         return ret;
1558         }
1559
1560         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1561         if (ret)
1562                 return ret;
1563
1564         inode_lock(inode);
1565
1566         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1567                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1568                 goto out;
1569         }
1570
1571         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1572                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1573                 if (ret != -EAGAIN)
1574                         goto out;
1575         }
1576
1577         if (logical_to_blk(inode, len) == 0)
1578                 len = blk_to_logical(inode, 1);
1579
1580         start_blk = logical_to_blk(inode, start);
1581         last_blk = logical_to_blk(inode, start + len - 1);
1582
1583 next:
1584         memset(&map_bh, 0, sizeof(struct buffer_head));
1585         map_bh.b_size = len;
1586
1587         ret = get_data_block(inode, start_blk, &map_bh, 0,
1588                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1589         if (ret)
1590                 goto out;
1591
1592         /* HOLE */
1593         if (!buffer_mapped(&map_bh)) {
1594                 start_blk = next_pgofs;
1595
1596                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1597                                         F2FS_I_SB(inode)->max_file_blocks))
1598                         goto prep_next;
1599
1600                 flags |= FIEMAP_EXTENT_LAST;
1601         }
1602
1603         if (size) {
1604                 if (f2fs_encrypted_inode(inode))
1605                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1606
1607                 ret = fiemap_fill_next_extent(fieinfo, logical,
1608                                 phys, size, flags);
1609         }
1610
1611         if (start_blk > last_blk || ret)
1612                 goto out;
1613
1614         logical = blk_to_logical(inode, start_blk);
1615         phys = blk_to_logical(inode, map_bh.b_blocknr);
1616         size = map_bh.b_size;
1617         flags = 0;
1618         if (buffer_unwritten(&map_bh))
1619                 flags = FIEMAP_EXTENT_UNWRITTEN;
1620
1621         start_blk += logical_to_blk(inode, size);
1622
1623 prep_next:
1624         cond_resched();
1625         if (fatal_signal_pending(current))
1626                 ret = -EINTR;
1627         else
1628                 goto next;
1629 out:
1630         if (ret == 1)
1631                 ret = 0;
1632
1633         inode_unlock(inode);
1634         return ret;
1635 }
1636
1637 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1638                                         unsigned nr_pages,
1639                                         struct f2fs_map_blocks *map,
1640                                         struct bio **bio_ret,
1641                                         sector_t *last_block_in_bio,
1642                                         bool is_readahead)
1643 {
1644         struct bio *bio = *bio_ret;
1645         const unsigned blkbits = inode->i_blkbits;
1646         const unsigned blocksize = 1 << blkbits;
1647         sector_t block_in_file;
1648         sector_t last_block;
1649         sector_t last_block_in_file;
1650         sector_t block_nr;
1651         int ret = 0;
1652
1653         block_in_file = (sector_t)page_index(page);
1654         last_block = block_in_file + nr_pages;
1655         last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1656                                                         blkbits;
1657         if (last_block > last_block_in_file)
1658                 last_block = last_block_in_file;
1659
1660         /* just zeroing out page which is beyond EOF */
1661         if (block_in_file >= last_block)
1662                 goto zero_out;
1663         /*
1664          * Map blocks using the previous result first.
1665          */
1666         if ((map->m_flags & F2FS_MAP_MAPPED) &&
1667                         block_in_file > map->m_lblk &&
1668                         block_in_file < (map->m_lblk + map->m_len))
1669                 goto got_it;
1670
1671         /*
1672          * Then do more f2fs_map_blocks() calls until we are
1673          * done with this page.
1674          */
1675         map->m_lblk = block_in_file;
1676         map->m_len = last_block - block_in_file;
1677
1678         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1679         if (ret)
1680                 goto out;
1681 got_it:
1682         if ((map->m_flags & F2FS_MAP_MAPPED)) {
1683                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1684                 SetPageMappedToDisk(page);
1685
1686                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
1687                                         !cleancache_get_page(page))) {
1688                         SetPageUptodate(page);
1689                         goto confused;
1690                 }
1691
1692                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1693                                                 DATA_GENERIC_ENHANCE_READ)) {
1694                         ret = -EFSCORRUPTED;
1695                         goto out;
1696                 }
1697         } else {
1698 zero_out:
1699                 zero_user_segment(page, 0, PAGE_SIZE);
1700                 if (!PageUptodate(page))
1701                         SetPageUptodate(page);
1702                 unlock_page(page);
1703                 goto out;
1704         }
1705
1706         /*
1707          * This page will go to BIO.  Do we need to send this
1708          * BIO off first?
1709          */
1710         if (bio && !page_is_mergeable(F2FS_I_SB(inode), bio,
1711                                 *last_block_in_bio, block_nr)) {
1712 submit_and_realloc:
1713                 __f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
1714                 bio = NULL;
1715         }
1716         if (bio == NULL) {
1717                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1718                                 is_readahead ? REQ_RAHEAD : 0);
1719                 if (IS_ERR(bio)) {
1720                         ret = PTR_ERR(bio);
1721                         bio = NULL;
1722                         goto out;
1723                 }
1724         }
1725
1726         /*
1727          * If the page is under writeback, we need to wait for
1728          * its completion to see the correct decrypted data.
1729          */
1730         f2fs_wait_on_block_writeback(inode, block_nr);
1731
1732         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1733                 goto submit_and_realloc;
1734
1735         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1736         ClearPageError(page);
1737         *last_block_in_bio = block_nr;
1738         goto out;
1739 confused:
1740         if (bio) {
1741                 __f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
1742                 bio = NULL;
1743         }
1744         unlock_page(page);
1745 out:
1746         *bio_ret = bio;
1747         return ret;
1748 }
1749
1750 /*
1751  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1752  * Major change was from block_size == page_size in f2fs by default.
1753  *
1754  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1755  * this function ever deviates from doing just read-ahead, it should either
1756  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1757  * from read-ahead.
1758  */
1759 static int f2fs_mpage_readpages(struct address_space *mapping,
1760                         struct list_head *pages, struct page *page,
1761                         unsigned nr_pages, bool is_readahead)
1762 {
1763         struct bio *bio = NULL;
1764         sector_t last_block_in_bio = 0;
1765         struct inode *inode = mapping->host;
1766         struct f2fs_map_blocks map;
1767         int ret = 0;
1768
1769         map.m_pblk = 0;
1770         map.m_lblk = 0;
1771         map.m_len = 0;
1772         map.m_flags = 0;
1773         map.m_next_pgofs = NULL;
1774         map.m_next_extent = NULL;
1775         map.m_seg_type = NO_CHECK_TYPE;
1776         map.m_may_create = false;
1777
1778         for (; nr_pages; nr_pages--) {
1779                 if (pages) {
1780                         page = list_last_entry(pages, struct page, lru);
1781
1782                         prefetchw(&page->flags);
1783                         list_del(&page->lru);
1784                         if (add_to_page_cache_lru(page, mapping,
1785                                                   page_index(page), GFP_KERNEL))
1786                                 goto next_page;
1787                 }
1788
1789                 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1790                                         &last_block_in_bio, is_readahead);
1791                 if (ret) {
1792                         SetPageError(page);
1793                         zero_user_segment(page, 0, PAGE_SIZE);
1794                         unlock_page(page);
1795                 }
1796 next_page:
1797                 if (pages)
1798                         put_page(page);
1799         }
1800         BUG_ON(pages && !list_empty(pages));
1801         if (bio)
1802                 __f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
1803         return pages ? 0 : ret;
1804 }
1805
1806 static int f2fs_read_data_page(struct file *file, struct page *page)
1807 {
1808         struct inode *inode = page_file_mapping(page)->host;
1809         int ret = -EAGAIN;
1810
1811         trace_f2fs_readpage(page, DATA);
1812
1813         /* If the file has inline data, try to read it directly */
1814         if (f2fs_has_inline_data(inode))
1815                 ret = f2fs_read_inline_data(inode, page);
1816         if (ret == -EAGAIN)
1817                 ret = f2fs_mpage_readpages(page_file_mapping(page),
1818                                                 NULL, page, 1, false);
1819         return ret;
1820 }
1821
1822 static int f2fs_read_data_pages(struct file *file,
1823                         struct address_space *mapping,
1824                         struct list_head *pages, unsigned nr_pages)
1825 {
1826         struct inode *inode = mapping->host;
1827         struct page *page = list_last_entry(pages, struct page, lru);
1828
1829         trace_f2fs_readpages(inode, page, nr_pages);
1830
1831         /* If the file has inline data, skip readpages */
1832         if (f2fs_has_inline_data(inode))
1833                 return 0;
1834
1835         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1836 }
1837
1838 static int encrypt_one_page(struct f2fs_io_info *fio)
1839 {
1840         struct inode *inode = fio->page->mapping->host;
1841         struct page *mpage;
1842         gfp_t gfp_flags = GFP_NOFS;
1843
1844         if (!f2fs_encrypted_file(inode))
1845                 return 0;
1846
1847         /* wait for GCed page writeback via META_MAPPING */
1848         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1849
1850 retry_encrypt:
1851         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1852                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1853         if (IS_ERR(fio->encrypted_page)) {
1854                 /* flush pending IOs and wait for a while in the ENOMEM case */
1855                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1856                         f2fs_flush_merged_writes(fio->sbi);
1857                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1858                         gfp_flags |= __GFP_NOFAIL;
1859                         goto retry_encrypt;
1860                 }
1861                 return PTR_ERR(fio->encrypted_page);
1862         }
1863
1864         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1865         if (mpage) {
1866                 if (PageUptodate(mpage))
1867                         memcpy(page_address(mpage),
1868                                 page_address(fio->encrypted_page), PAGE_SIZE);
1869                 f2fs_put_page(mpage, 1);
1870         }
1871         return 0;
1872 }
1873
1874 static inline bool check_inplace_update_policy(struct inode *inode,
1875                                 struct f2fs_io_info *fio)
1876 {
1877         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1878         unsigned int policy = SM_I(sbi)->ipu_policy;
1879
1880         if (policy & (0x1 << F2FS_IPU_FORCE))
1881                 return true;
1882         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1883                 return true;
1884         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1885                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1886                 return true;
1887         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1888                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1889                 return true;
1890
1891         /*
1892          * IPU for rewrite async pages
1893          */
1894         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1895                         fio && fio->op == REQ_OP_WRITE &&
1896                         !(fio->op_flags & REQ_SYNC) &&
1897                         !f2fs_encrypted_inode(inode))
1898                 return true;
1899
1900         /* this is only set during fdatasync */
1901         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1902                         is_inode_flag_set(inode, FI_NEED_IPU))
1903                 return true;
1904
1905         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1906                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1907                 return true;
1908
1909         return false;
1910 }
1911
1912 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1913 {
1914         if (f2fs_is_pinned_file(inode))
1915                 return true;
1916
1917         /* if this is cold file, we should overwrite to avoid fragmentation */
1918         if (file_is_cold(inode))
1919                 return true;
1920
1921         return check_inplace_update_policy(inode, fio);
1922 }
1923
1924 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1925 {
1926         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1927
1928         if (test_opt(sbi, LFS))
1929                 return true;
1930         if (S_ISDIR(inode->i_mode))
1931                 return true;
1932         if (IS_NOQUOTA(inode))
1933                 return true;
1934         if (f2fs_is_atomic_file(inode))
1935                 return true;
1936         if (fio) {
1937                 if (is_cold_data(fio->page))
1938                         return true;
1939                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1940                         return true;
1941                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1942                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1943                         return true;
1944         }
1945         return false;
1946 }
1947
1948 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1949 {
1950         struct inode *inode = fio->page->mapping->host;
1951
1952         if (f2fs_should_update_outplace(inode, fio))
1953                 return false;
1954
1955         return f2fs_should_update_inplace(inode, fio);
1956 }
1957
1958 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1959 {
1960         struct page *page = fio->page;
1961         struct inode *inode = page->mapping->host;
1962         struct dnode_of_data dn;
1963         struct extent_info ei = {0,0,0};
1964         struct node_info ni;
1965         bool ipu_force = false;
1966         int err = 0;
1967
1968         set_new_dnode(&dn, inode, NULL, NULL, 0);
1969         if (need_inplace_update(fio) &&
1970                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1971                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1972
1973                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1974                                                 DATA_GENERIC_ENHANCE))
1975                         return -EFSCORRUPTED;
1976
1977                 ipu_force = true;
1978                 fio->need_lock = LOCK_DONE;
1979                 goto got_it;
1980         }
1981
1982         /* Deadlock due to between page->lock and f2fs_lock_op */
1983         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1984                 return -EAGAIN;
1985
1986         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1987         if (err)
1988                 goto out;
1989
1990         fio->old_blkaddr = dn.data_blkaddr;
1991
1992         /* This page is already truncated */
1993         if (fio->old_blkaddr == NULL_ADDR) {
1994                 ClearPageUptodate(page);
1995                 clear_cold_data(page);
1996                 goto out_writepage;
1997         }
1998 got_it:
1999         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2000                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2001                                                 DATA_GENERIC_ENHANCE)) {
2002                 err = -EFSCORRUPTED;
2003                 goto out_writepage;
2004         }
2005         /*
2006          * If current allocation needs SSR,
2007          * it had better in-place writes for updated data.
2008          */
2009         if (ipu_force ||
2010                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2011                                         need_inplace_update(fio))) {
2012                 err = encrypt_one_page(fio);
2013                 if (err)
2014                         goto out_writepage;
2015
2016                 set_page_writeback(page);
2017                 ClearPageError(page);
2018                 f2fs_put_dnode(&dn);
2019                 if (fio->need_lock == LOCK_REQ)
2020                         f2fs_unlock_op(fio->sbi);
2021                 err = f2fs_inplace_write_data(fio);
2022                 if (err) {
2023                         if (f2fs_encrypted_file(inode))
2024                                 fscrypt_pullback_bio_page(&fio->encrypted_page,
2025                                                                         true);
2026                         if (PageWriteback(page))
2027                                 end_page_writeback(page);
2028                 } else {
2029                         set_inode_flag(inode, FI_UPDATE_WRITE);
2030                 }
2031                 trace_f2fs_do_write_data_page(fio->page, IPU);
2032                 return err;
2033         }
2034
2035         if (fio->need_lock == LOCK_RETRY) {
2036                 if (!f2fs_trylock_op(fio->sbi)) {
2037                         err = -EAGAIN;
2038                         goto out_writepage;
2039                 }
2040                 fio->need_lock = LOCK_REQ;
2041         }
2042
2043         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2044         if (err)
2045                 goto out_writepage;
2046
2047         fio->version = ni.version;
2048
2049         err = encrypt_one_page(fio);
2050         if (err)
2051                 goto out_writepage;
2052
2053         set_page_writeback(page);
2054         ClearPageError(page);
2055
2056         /* LFS mode write path */
2057         f2fs_outplace_write_data(&dn, fio);
2058         trace_f2fs_do_write_data_page(page, OPU);
2059         set_inode_flag(inode, FI_APPEND_WRITE);
2060         if (page->index == 0)
2061                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2062 out_writepage:
2063         f2fs_put_dnode(&dn);
2064 out:
2065         if (fio->need_lock == LOCK_REQ)
2066                 f2fs_unlock_op(fio->sbi);
2067         return err;
2068 }
2069
2070 static int __write_data_page(struct page *page, bool *submitted,
2071                                 struct bio **bio,
2072                                 sector_t *last_block,
2073                                 struct writeback_control *wbc,
2074                                 enum iostat_type io_type)
2075 {
2076         struct inode *inode = page->mapping->host;
2077         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2078         loff_t i_size = i_size_read(inode);
2079         const pgoff_t end_index = ((unsigned long long) i_size)
2080                                                         >> PAGE_SHIFT;
2081         loff_t psize = (page->index + 1) << PAGE_SHIFT;
2082         unsigned offset = 0;
2083         bool need_balance_fs = false;
2084         int err = 0;
2085         struct f2fs_io_info fio = {
2086                 .sbi = sbi,
2087                 .ino = inode->i_ino,
2088                 .type = DATA,
2089                 .op = REQ_OP_WRITE,
2090                 .op_flags = wbc_to_write_flags(wbc),
2091                 .old_blkaddr = NULL_ADDR,
2092                 .page = page,
2093                 .encrypted_page = NULL,
2094                 .submitted = false,
2095                 .need_lock = LOCK_RETRY,
2096                 .io_type = io_type,
2097                 .io_wbc = wbc,
2098                 .bio = bio,
2099                 .last_block = last_block,
2100         };
2101
2102         trace_f2fs_writepage(page, DATA);
2103
2104         /* we should bypass data pages to proceed the kworkder jobs */
2105         if (unlikely(f2fs_cp_error(sbi))) {
2106                 mapping_set_error(page->mapping, -EIO);
2107                 /*
2108                  * don't drop any dirty dentry pages for keeping lastest
2109                  * directory structure.
2110                  */
2111                 if (S_ISDIR(inode->i_mode))
2112                         goto redirty_out;
2113                 goto out;
2114         }
2115
2116         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2117                 goto redirty_out;
2118
2119         if (page->index < end_index)
2120                 goto write;
2121
2122         /*
2123          * If the offset is out-of-range of file size,
2124          * this page does not have to be written to disk.
2125          */
2126         offset = i_size & (PAGE_SIZE - 1);
2127         if ((page->index >= end_index + 1) || !offset)
2128                 goto out;
2129
2130         zero_user_segment(page, offset, PAGE_SIZE);
2131 write:
2132         if (f2fs_is_drop_cache(inode))
2133                 goto out;
2134         /* we should not write 0'th page having journal header */
2135         if (f2fs_is_volatile_file(inode) && (!page->index ||
2136                         (!wbc->for_reclaim &&
2137                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2138                 goto redirty_out;
2139
2140         /* Dentry blocks are controlled by checkpoint */
2141         if (S_ISDIR(inode->i_mode)) {
2142                 fio.need_lock = LOCK_DONE;
2143                 err = f2fs_do_write_data_page(&fio);
2144                 goto done;
2145         }
2146
2147         if (!wbc->for_reclaim)
2148                 need_balance_fs = true;
2149         else if (has_not_enough_free_secs(sbi, 0, 0))
2150                 goto redirty_out;
2151         else
2152                 set_inode_flag(inode, FI_HOT_DATA);
2153
2154         err = -EAGAIN;
2155         if (f2fs_has_inline_data(inode)) {
2156                 err = f2fs_write_inline_data(inode, page);
2157                 if (!err)
2158                         goto out;
2159         }
2160
2161         if (err == -EAGAIN) {
2162                 err = f2fs_do_write_data_page(&fio);
2163                 if (err == -EAGAIN) {
2164                         fio.need_lock = LOCK_REQ;
2165                         err = f2fs_do_write_data_page(&fio);
2166                 }
2167         }
2168
2169         if (err) {
2170                 file_set_keep_isize(inode);
2171         } else {
2172                 down_write(&F2FS_I(inode)->i_sem);
2173                 if (F2FS_I(inode)->last_disk_size < psize)
2174                         F2FS_I(inode)->last_disk_size = psize;
2175                 up_write(&F2FS_I(inode)->i_sem);
2176         }
2177
2178 done:
2179         if (err && err != -ENOENT)
2180                 goto redirty_out;
2181
2182 out:
2183         inode_dec_dirty_pages(inode);
2184         if (err) {
2185                 ClearPageUptodate(page);
2186                 clear_cold_data(page);
2187         }
2188
2189         if (wbc->for_reclaim) {
2190                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2191                 clear_inode_flag(inode, FI_HOT_DATA);
2192                 f2fs_remove_dirty_inode(inode);
2193                 submitted = NULL;
2194         }
2195
2196         unlock_page(page);
2197         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2198                                         !F2FS_I(inode)->cp_task) {
2199                 f2fs_submit_ipu_bio(sbi, bio, page);
2200                 f2fs_balance_fs(sbi, need_balance_fs);
2201         }
2202
2203         if (unlikely(f2fs_cp_error(sbi))) {
2204                 f2fs_submit_ipu_bio(sbi, bio, page);
2205                 f2fs_submit_merged_write(sbi, DATA);
2206                 submitted = NULL;
2207         }
2208
2209         if (submitted)
2210                 *submitted = fio.submitted;
2211
2212         return 0;
2213
2214 redirty_out:
2215         redirty_page_for_writepage(wbc, page);
2216         /*
2217          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2218          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2219          * file_write_and_wait_range() will see EIO error, which is critical
2220          * to return value of fsync() followed by atomic_write failure to user.
2221          */
2222         if (!err || wbc->for_reclaim)
2223                 return AOP_WRITEPAGE_ACTIVATE;
2224         unlock_page(page);
2225         return err;
2226 }
2227
2228 static int f2fs_write_data_page(struct page *page,
2229                                         struct writeback_control *wbc)
2230 {
2231         return __write_data_page(page, NULL, NULL, NULL, wbc, FS_DATA_IO);
2232 }
2233
2234 /*
2235  * This function was copied from write_cche_pages from mm/page-writeback.c.
2236  * The major change is making write step of cold data page separately from
2237  * warm/hot data page.
2238  */
2239 static int f2fs_write_cache_pages(struct address_space *mapping,
2240                                         struct writeback_control *wbc,
2241                                         enum iostat_type io_type)
2242 {
2243         int ret = 0;
2244         int done = 0;
2245         struct pagevec pvec;
2246         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2247         struct bio *bio = NULL;
2248         sector_t last_block;
2249         int nr_pages;
2250         pgoff_t uninitialized_var(writeback_index);
2251         pgoff_t index;
2252         pgoff_t end;            /* Inclusive */
2253         pgoff_t done_index;
2254         int cycled;
2255         int range_whole = 0;
2256         int tag;
2257         int nwritten = 0;
2258
2259         pagevec_init(&pvec, 0);
2260
2261         if (get_dirty_pages(mapping->host) <=
2262                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2263                 set_inode_flag(mapping->host, FI_HOT_DATA);
2264         else
2265                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2266
2267         if (wbc->range_cyclic) {
2268                 writeback_index = mapping->writeback_index; /* prev offset */
2269                 index = writeback_index;
2270                 if (index == 0)
2271                         cycled = 1;
2272                 else
2273                         cycled = 0;
2274                 end = -1;
2275         } else {
2276                 index = wbc->range_start >> PAGE_SHIFT;
2277                 end = wbc->range_end >> PAGE_SHIFT;
2278                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2279                         range_whole = 1;
2280                 cycled = 1; /* ignore range_cyclic tests */
2281         }
2282         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2283                 tag = PAGECACHE_TAG_TOWRITE;
2284         else
2285                 tag = PAGECACHE_TAG_DIRTY;
2286 retry:
2287         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2288                 tag_pages_for_writeback(mapping, index, end);
2289         done_index = index;
2290         while (!done && (index <= end)) {
2291                 int i;
2292
2293                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2294                                 tag);
2295                 if (nr_pages == 0)
2296                         break;
2297
2298                 for (i = 0; i < nr_pages; i++) {
2299                         struct page *page = pvec.pages[i];
2300                         bool submitted = false;
2301
2302                         /* give a priority to WB_SYNC threads */
2303                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2304                                         wbc->sync_mode == WB_SYNC_NONE) {
2305                                 done = 1;
2306                                 break;
2307                         }
2308
2309                         done_index = page->index;
2310 retry_write:
2311                         lock_page(page);
2312
2313                         if (unlikely(page->mapping != mapping)) {
2314 continue_unlock:
2315                                 unlock_page(page);
2316                                 continue;
2317                         }
2318
2319                         if (!PageDirty(page)) {
2320                                 /* someone wrote it for us */
2321                                 goto continue_unlock;
2322                         }
2323
2324                         if (PageWriteback(page)) {
2325                                 if (wbc->sync_mode != WB_SYNC_NONE) {
2326                                         f2fs_wait_on_page_writeback(page,
2327                                                         DATA, true, true);
2328                                         f2fs_submit_ipu_bio(sbi, &bio, page);
2329                                 } else {
2330                                         goto continue_unlock;
2331                                 }
2332                         }
2333
2334                         if (!clear_page_dirty_for_io(page))
2335                                 goto continue_unlock;
2336
2337                         ret = __write_data_page(page, &submitted, &bio,
2338                                         &last_block, wbc, io_type);
2339                         if (unlikely(ret)) {
2340                                 /*
2341                                  * keep nr_to_write, since vfs uses this to
2342                                  * get # of written pages.
2343                                  */
2344                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2345                                         unlock_page(page);
2346                                         ret = 0;
2347                                         continue;
2348                                 } else if (ret == -EAGAIN) {
2349                                         ret = 0;
2350                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2351                                                 cond_resched();
2352                                                 congestion_wait(BLK_RW_ASYNC, msecs_to_jiffies(6));
2353                                                 goto retry_write;
2354                                         }
2355                                         continue;
2356                                 }
2357                                 done_index = page->index + 1;
2358                                 done = 1;
2359                                 break;
2360                         } else if (submitted) {
2361                                 nwritten++;
2362                         }
2363
2364                         if (--wbc->nr_to_write <= 0 &&
2365                                         wbc->sync_mode == WB_SYNC_NONE) {
2366                                 done = 1;
2367                                 break;
2368                         }
2369                 }
2370                 pagevec_release(&pvec);
2371                 cond_resched();
2372         }
2373
2374         if (!cycled && !done) {
2375                 cycled = 1;
2376                 index = 0;
2377                 end = writeback_index - 1;
2378                 goto retry;
2379         }
2380         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2381                 mapping->writeback_index = done_index;
2382
2383         if (nwritten)
2384                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2385                                                                 NULL, 0, DATA);
2386         /* submit cached bio of IPU write */
2387         if (bio)
2388                 __submit_bio(sbi, bio, DATA);
2389
2390         return ret;
2391 }
2392
2393 static inline bool __should_serialize_io(struct inode *inode,
2394                                         struct writeback_control *wbc)
2395 {
2396         if (!S_ISREG(inode->i_mode))
2397                 return false;
2398         if (IS_NOQUOTA(inode))
2399                 return false;
2400         /* to avoid deadlock in path of data flush */
2401         if (F2FS_I(inode)->cp_task)
2402                 return false;
2403         if (wbc->sync_mode != WB_SYNC_ALL)
2404                 return true;
2405         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2406                 return true;
2407         return false;
2408 }
2409
2410 static int __f2fs_write_data_pages(struct address_space *mapping,
2411                                                 struct writeback_control *wbc,
2412                                                 enum iostat_type io_type)
2413 {
2414         struct inode *inode = mapping->host;
2415         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2416         struct blk_plug plug;
2417         int ret;
2418         bool locked = false;
2419
2420         /* deal with chardevs and other special file */
2421         if (!mapping->a_ops->writepage)
2422                 return 0;
2423
2424         /* skip writing if there is no dirty page in this inode */
2425         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2426                 return 0;
2427
2428         /* during POR, we don't need to trigger writepage at all. */
2429         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2430                 goto skip_write;
2431
2432         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2433                         wbc->sync_mode == WB_SYNC_NONE &&
2434                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2435                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2436                 goto skip_write;
2437
2438         /* skip writing during file defragment */
2439         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2440                 goto skip_write;
2441
2442         trace_f2fs_writepages(mapping->host, wbc, DATA);
2443
2444         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2445         if (wbc->sync_mode == WB_SYNC_ALL)
2446                 atomic_inc(&sbi->wb_sync_req[DATA]);
2447         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2448                 goto skip_write;
2449
2450         if (__should_serialize_io(inode, wbc)) {
2451                 mutex_lock(&sbi->writepages);
2452                 locked = true;
2453         }
2454
2455         blk_start_plug(&plug);
2456         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2457         blk_finish_plug(&plug);
2458
2459         if (locked)
2460                 mutex_unlock(&sbi->writepages);
2461
2462         if (wbc->sync_mode == WB_SYNC_ALL)
2463                 atomic_dec(&sbi->wb_sync_req[DATA]);
2464         /*
2465          * if some pages were truncated, we cannot guarantee its mapping->host
2466          * to detect pending bios.
2467          */
2468
2469         f2fs_remove_dirty_inode(inode);
2470         return ret;
2471
2472 skip_write:
2473         wbc->pages_skipped += get_dirty_pages(inode);
2474         trace_f2fs_writepages(mapping->host, wbc, DATA);
2475         return 0;
2476 }
2477
2478 static int f2fs_write_data_pages(struct address_space *mapping,
2479                             struct writeback_control *wbc)
2480 {
2481         struct inode *inode = mapping->host;
2482
2483         return __f2fs_write_data_pages(mapping, wbc,
2484                         F2FS_I(inode)->cp_task == current ?
2485                         FS_CP_DATA_IO : FS_DATA_IO);
2486 }
2487
2488 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2489 {
2490         struct inode *inode = mapping->host;
2491         loff_t i_size = i_size_read(inode);
2492
2493         if (to > i_size) {
2494                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2495                 down_write(&F2FS_I(inode)->i_mmap_sem);
2496
2497                 truncate_pagecache(inode, i_size);
2498                 if (!IS_NOQUOTA(inode))
2499                         f2fs_truncate_blocks(inode, i_size, true);
2500
2501                 up_write(&F2FS_I(inode)->i_mmap_sem);
2502                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2503         }
2504 }
2505
2506 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2507                         struct page *page, loff_t pos, unsigned len,
2508                         block_t *blk_addr, bool *node_changed)
2509 {
2510         struct inode *inode = page->mapping->host;
2511         pgoff_t index = page->index;
2512         struct dnode_of_data dn;
2513         struct page *ipage;
2514         bool locked = false;
2515         struct extent_info ei = {0,0,0};
2516         int err = 0;
2517         int flag;
2518
2519         /*
2520          * we already allocated all the blocks, so we don't need to get
2521          * the block addresses when there is no need to fill the page.
2522          */
2523         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2524                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2525                 return 0;
2526
2527         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2528         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2529                 flag = F2FS_GET_BLOCK_DEFAULT;
2530         else
2531                 flag = F2FS_GET_BLOCK_PRE_AIO;
2532
2533         if (f2fs_has_inline_data(inode) ||
2534                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2535                 __do_map_lock(sbi, flag, true);
2536                 locked = true;
2537         }
2538 restart:
2539         /* check inline_data */
2540         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2541         if (IS_ERR(ipage)) {
2542                 err = PTR_ERR(ipage);
2543                 goto unlock_out;
2544         }
2545
2546         set_new_dnode(&dn, inode, ipage, ipage, 0);
2547
2548         if (f2fs_has_inline_data(inode)) {
2549                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2550                         f2fs_do_read_inline_data(page, ipage);
2551                         set_inode_flag(inode, FI_DATA_EXIST);
2552                         if (inode->i_nlink)
2553                                 set_inline_node(ipage);
2554                 } else {
2555                         err = f2fs_convert_inline_page(&dn, page);
2556                         if (err)
2557                                 goto out;
2558                         if (dn.data_blkaddr == NULL_ADDR)
2559                                 err = f2fs_get_block(&dn, index);
2560                 }
2561         } else if (locked) {
2562                 err = f2fs_get_block(&dn, index);
2563         } else {
2564                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2565                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2566                 } else {
2567                         /* hole case */
2568                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2569                         if (err || dn.data_blkaddr == NULL_ADDR) {
2570                                 f2fs_put_dnode(&dn);
2571                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2572                                                                 true);
2573                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2574                                 locked = true;
2575                                 goto restart;
2576                         }
2577                 }
2578         }
2579
2580         /* convert_inline_page can make node_changed */
2581         *blk_addr = dn.data_blkaddr;
2582         *node_changed = dn.node_changed;
2583 out:
2584         f2fs_put_dnode(&dn);
2585 unlock_out:
2586         if (locked)
2587                 __do_map_lock(sbi, flag, false);
2588         return err;
2589 }
2590
2591 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2592                 loff_t pos, unsigned len, unsigned flags,
2593                 struct page **pagep, void **fsdata)
2594 {
2595         struct inode *inode = mapping->host;
2596         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2597         struct page *page = NULL;
2598         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2599         bool need_balance = false, drop_atomic = false;
2600         block_t blkaddr = NULL_ADDR;
2601         int err = 0;
2602
2603         if (trace_android_fs_datawrite_start_enabled()) {
2604                 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2605
2606                 path = android_fstrace_get_pathname(pathbuf,
2607                                                     MAX_TRACE_PATHBUF_LEN,
2608                                                     inode);
2609                 trace_android_fs_datawrite_start(inode, pos, len,
2610                                                  current->pid, path,
2611                                                  current->comm);
2612         }
2613         trace_f2fs_write_begin(inode, pos, len, flags);
2614
2615         if (!f2fs_is_checkpoint_ready(sbi)) {
2616                 err = -ENOSPC;
2617                 goto fail;
2618         }
2619
2620         if ((f2fs_is_atomic_file(inode) &&
2621                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2622                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2623                 err = -ENOMEM;
2624                 drop_atomic = true;
2625                 goto fail;
2626         }
2627
2628         /*
2629          * We should check this at this moment to avoid deadlock on inode page
2630          * and #0 page. The locking rule for inline_data conversion should be:
2631          * lock_page(page #0) -> lock_page(inode_page)
2632          */
2633         if (index != 0) {
2634                 err = f2fs_convert_inline_inode(inode);
2635                 if (err)
2636                         goto fail;
2637         }
2638 repeat:
2639         /*
2640          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2641          * wait_for_stable_page. Will wait that below with our IO control.
2642          */
2643         page = f2fs_pagecache_get_page(mapping, index,
2644                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2645         if (!page) {
2646                 err = -ENOMEM;
2647                 goto fail;
2648         }
2649
2650         *pagep = page;
2651
2652         err = prepare_write_begin(sbi, page, pos, len,
2653                                         &blkaddr, &need_balance);
2654         if (err)
2655                 goto fail;
2656
2657         if (need_balance && !IS_NOQUOTA(inode) &&
2658                         has_not_enough_free_secs(sbi, 0, 0)) {
2659                 unlock_page(page);
2660                 f2fs_balance_fs(sbi, true);
2661                 lock_page(page);
2662                 if (page->mapping != mapping) {
2663                         /* The page got truncated from under us */
2664                         f2fs_put_page(page, 1);
2665                         goto repeat;
2666                 }
2667         }
2668
2669         f2fs_wait_on_page_writeback(page, DATA, false, true);
2670
2671         if (len == PAGE_SIZE || PageUptodate(page))
2672                 return 0;
2673
2674         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2675                 zero_user_segment(page, len, PAGE_SIZE);
2676                 return 0;
2677         }
2678
2679         if (blkaddr == NEW_ADDR) {
2680                 zero_user_segment(page, 0, PAGE_SIZE);
2681                 SetPageUptodate(page);
2682         } else {
2683                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2684                                 DATA_GENERIC_ENHANCE_READ)) {
2685                         err = -EFSCORRUPTED;
2686                         goto fail;
2687                 }
2688                 err = f2fs_submit_page_read(inode, page, blkaddr);
2689                 if (err)
2690                         goto fail;
2691
2692                 lock_page(page);
2693                 if (unlikely(page->mapping != mapping)) {
2694                         f2fs_put_page(page, 1);
2695                         goto repeat;
2696                 }
2697                 if (unlikely(!PageUptodate(page))) {
2698                         err = -EIO;
2699                         goto fail;
2700                 }
2701         }
2702         return 0;
2703
2704 fail:
2705         f2fs_put_page(page, 1);
2706         f2fs_write_failed(mapping, pos + len);
2707         if (drop_atomic)
2708                 f2fs_drop_inmem_pages_all(sbi, false);
2709         return err;
2710 }
2711
2712 static int f2fs_write_end(struct file *file,
2713                         struct address_space *mapping,
2714                         loff_t pos, unsigned len, unsigned copied,
2715                         struct page *page, void *fsdata)
2716 {
2717         struct inode *inode = page->mapping->host;
2718
2719         trace_android_fs_datawrite_end(inode, pos, len);
2720         trace_f2fs_write_end(inode, pos, len, copied);
2721
2722         /*
2723          * This should be come from len == PAGE_SIZE, and we expect copied
2724          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2725          * let generic_perform_write() try to copy data again through copied=0.
2726          */
2727         if (!PageUptodate(page)) {
2728                 if (unlikely(copied != len))
2729                         copied = 0;
2730                 else
2731                         SetPageUptodate(page);
2732         }
2733         if (!copied)
2734                 goto unlock_out;
2735
2736         set_page_dirty(page);
2737
2738         if (pos + copied > i_size_read(inode))
2739                 f2fs_i_size_write(inode, pos + copied);
2740 unlock_out:
2741         f2fs_put_page(page, 1);
2742         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2743         return copied;
2744 }
2745
2746 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2747                            loff_t offset)
2748 {
2749         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2750         unsigned blkbits = i_blkbits;
2751         unsigned blocksize_mask = (1 << blkbits) - 1;
2752         unsigned long align = offset | iov_iter_alignment(iter);
2753         struct block_device *bdev = inode->i_sb->s_bdev;
2754
2755         if (align & blocksize_mask) {
2756                 if (bdev)
2757                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2758                 blocksize_mask = (1 << blkbits) - 1;
2759                 if (align & blocksize_mask)
2760                         return -EINVAL;
2761                 return 1;
2762         }
2763         return 0;
2764 }
2765
2766 static void f2fs_dio_end_io(struct bio *bio)
2767 {
2768         struct f2fs_private_dio *dio = bio->bi_private;
2769
2770         dec_page_count(F2FS_I_SB(dio->inode),
2771                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2772
2773         bio->bi_private = dio->orig_private;
2774         bio->bi_end_io = dio->orig_end_io;
2775
2776         kvfree(dio);
2777
2778         bio_endio(bio);
2779 }
2780
2781 static void f2fs_dio_submit_bio(int rw, struct bio *bio, struct inode *inode,
2782                                                         loff_t file_offset)
2783 {
2784         struct f2fs_private_dio *dio;
2785         bool write = (rw == REQ_OP_WRITE);
2786
2787         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2788                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2789         if (!dio)
2790                 goto out;
2791
2792         dio->inode = inode;
2793         dio->orig_end_io = bio->bi_end_io;
2794         dio->orig_private = bio->bi_private;
2795         dio->write = write;
2796
2797         bio->bi_end_io = f2fs_dio_end_io;
2798         bio->bi_private = dio;
2799
2800         inc_page_count(F2FS_I_SB(inode),
2801                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2802
2803         submit_bio(rw, bio);
2804         return;
2805 out:
2806         bio->bi_error = -EIO;
2807         bio_endio(bio);
2808 }
2809
2810 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
2811                                 loff_t offset)
2812 {
2813         struct address_space *mapping = iocb->ki_filp->f_mapping;
2814         struct inode *inode = mapping->host;
2815         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2816         struct f2fs_inode_info *fi = F2FS_I(inode);
2817         size_t count = iov_iter_count(iter);
2818         int rw = iov_iter_rw(iter);
2819         int err;
2820         enum rw_hint hint = iocb->ki_hint;
2821         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2822         bool do_opu;
2823
2824         err = check_direct_IO(inode, iter, offset);
2825         if (err)
2826                 return err < 0 ? err : 0;
2827
2828         if (f2fs_force_buffered_io(inode, iocb, iter))
2829                 return 0;
2830
2831         if (trace_android_fs_dataread_start_enabled() &&
2832             (iov_iter_rw(iter) == READ)) {
2833                 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2834
2835                 path = android_fstrace_get_pathname(pathbuf,
2836                                                     MAX_TRACE_PATHBUF_LEN,
2837                                                     inode);
2838                 trace_android_fs_dataread_start(inode, offset,
2839                                                 count, current->pid, path,
2840                                                 current->comm);
2841         }
2842         if (trace_android_fs_datawrite_start_enabled() &&
2843             (iov_iter_rw(iter) == WRITE)) {
2844                 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2845
2846                 path = android_fstrace_get_pathname(pathbuf,
2847                                                     MAX_TRACE_PATHBUF_LEN,
2848                                                     inode);
2849                 trace_android_fs_datawrite_start(inode, offset, count,
2850                                                  current->pid, path,
2851                                                  current->comm);
2852         }
2853
2854         do_opu = allow_outplace_dio(inode, iocb, iter);
2855
2856         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2857
2858         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2859                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2860
2861         if (iocb->ki_flags & IOCB_NOWAIT) {
2862                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2863                         iocb->ki_hint = hint;
2864                         err = -EAGAIN;
2865                         goto out;
2866                 }
2867                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2868                         up_read(&fi->i_gc_rwsem[rw]);
2869                         iocb->ki_hint = hint;
2870                         err = -EAGAIN;
2871                         goto out;
2872                 }
2873         } else {
2874                 down_read(&fi->i_gc_rwsem[rw]);
2875                 if (do_opu)
2876                         down_read(&fi->i_gc_rwsem[READ]);
2877         }
2878
2879         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2880                         iter, offset,
2881                         rw == WRITE ? get_data_block_dio_write :
2882                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2883                         DIO_LOCKING | DIO_SKIP_HOLES);
2884
2885         if (do_opu)
2886                 up_read(&fi->i_gc_rwsem[READ]);
2887
2888         up_read(&fi->i_gc_rwsem[rw]);
2889
2890         if (rw == WRITE) {
2891                 if (whint_mode == WHINT_MODE_OFF)
2892                         iocb->ki_hint = hint;
2893                 if (err > 0) {
2894                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2895                                                                         err);
2896                         if (!do_opu)
2897                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2898                 } else if (err < 0) {
2899                         f2fs_write_failed(mapping, offset + count);
2900                 }
2901         }
2902 out:
2903         if (trace_android_fs_dataread_start_enabled() &&
2904             (iov_iter_rw(iter) == READ))
2905                 trace_android_fs_dataread_end(inode, offset, count);
2906         if (trace_android_fs_datawrite_start_enabled() &&
2907             (iov_iter_rw(iter) == WRITE))
2908                 trace_android_fs_datawrite_end(inode, offset, count);
2909
2910         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2911
2912         return err;
2913 }
2914
2915 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2916                                                         unsigned int length)
2917 {
2918         struct inode *inode = page->mapping->host;
2919         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2920
2921         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2922                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2923                 return;
2924
2925         if (PageDirty(page)) {
2926                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2927                         dec_page_count(sbi, F2FS_DIRTY_META);
2928                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2929                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2930                 } else {
2931                         inode_dec_dirty_pages(inode);
2932                         f2fs_remove_dirty_inode(inode);
2933                 }
2934         }
2935
2936         clear_cold_data(page);
2937
2938         if (IS_ATOMIC_WRITTEN_PAGE(page))
2939                 return f2fs_drop_inmem_page(inode, page);
2940
2941         f2fs_clear_page_private(page);
2942 }
2943
2944 int f2fs_release_page(struct page *page, gfp_t wait)
2945 {
2946         /* If this is dirty page, keep PagePrivate */
2947         if (PageDirty(page))
2948                 return 0;
2949
2950         /* This is atomic written page, keep Private */
2951         if (IS_ATOMIC_WRITTEN_PAGE(page))
2952                 return 0;
2953
2954         clear_cold_data(page);
2955         f2fs_clear_page_private(page);
2956         return 1;
2957 }
2958
2959 static int f2fs_set_data_page_dirty(struct page *page)
2960 {
2961         struct inode *inode = page_file_mapping(page)->host;
2962
2963         trace_f2fs_set_page_dirty(page, DATA);
2964
2965         if (!PageUptodate(page))
2966                 SetPageUptodate(page);
2967         if (PageSwapCache(page))
2968                 return __set_page_dirty_nobuffers(page);
2969
2970         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2971                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2972                         f2fs_register_inmem_page(inode, page);
2973                         return 1;
2974                 }
2975                 /*
2976                  * Previously, this page has been registered, we just
2977                  * return here.
2978                  */
2979                 return 0;
2980         }
2981
2982         if (!PageDirty(page)) {
2983                 __set_page_dirty_nobuffers(page);
2984                 f2fs_update_dirty_page(inode, page);
2985                 return 1;
2986         }
2987         return 0;
2988 }
2989
2990 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2991 {
2992         struct inode *inode = mapping->host;
2993
2994         if (f2fs_has_inline_data(inode))
2995                 return 0;
2996
2997         /* make sure allocating whole blocks */
2998         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2999                 filemap_write_and_wait(mapping);
3000
3001         return generic_block_bmap(mapping, block, get_data_block_bmap);
3002 }
3003
3004 #ifdef CONFIG_MIGRATION
3005 #include <linux/migrate.h>
3006
3007 int f2fs_migrate_page(struct address_space *mapping,
3008                 struct page *newpage, struct page *page, enum migrate_mode mode)
3009 {
3010         int rc, extra_count;
3011         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3012         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
3013
3014         BUG_ON(PageWriteback(page));
3015
3016         /* migrating an atomic written page is safe with the inmem_lock hold */
3017         if (atomic_written) {
3018                 if (mode != MIGRATE_SYNC)
3019                         return -EBUSY;
3020                 if (!mutex_trylock(&fi->inmem_lock))
3021                         return -EAGAIN;
3022         }
3023
3024         /* one extra reference was held for atomic_write page */
3025         extra_count = atomic_written ? 1 : 0;
3026         rc = migrate_page_move_mapping(mapping, newpage,
3027                                 page, NULL, mode, extra_count);
3028         if (rc != MIGRATEPAGE_SUCCESS) {
3029                 if (atomic_written)
3030                         mutex_unlock(&fi->inmem_lock);
3031                 return rc;
3032         }
3033
3034         if (atomic_written) {
3035                 struct inmem_pages *cur;
3036                 list_for_each_entry(cur, &fi->inmem_pages, list)
3037                         if (cur->page == page) {
3038                                 cur->page = newpage;
3039                                 break;
3040                         }
3041                 mutex_unlock(&fi->inmem_lock);
3042                 put_page(page);
3043                 get_page(newpage);
3044         }
3045
3046         if (PagePrivate(page)) {
3047                 f2fs_set_page_private(newpage, page_private(page));
3048                 f2fs_clear_page_private(page);
3049         }
3050
3051         migrate_page_copy(newpage, page);
3052
3053         return MIGRATEPAGE_SUCCESS;
3054 }
3055 #endif
3056
3057 #ifdef CONFIG_SWAP
3058 /* Copied from generic_swapfile_activate() to check any holes */
3059 static int check_swap_activate(struct file *swap_file, unsigned int max)
3060 {
3061         struct address_space *mapping = swap_file->f_mapping;
3062         struct inode *inode = mapping->host;
3063         unsigned blocks_per_page;
3064         unsigned long page_no;
3065         unsigned blkbits;
3066         sector_t probe_block;
3067         sector_t last_block;
3068         sector_t lowest_block = -1;
3069         sector_t highest_block = 0;
3070
3071         blkbits = inode->i_blkbits;
3072         blocks_per_page = PAGE_SIZE >> blkbits;
3073
3074         /*
3075          * Map all the blocks into the extent list.  This code doesn't try
3076          * to be very smart.
3077          */
3078         probe_block = 0;
3079         page_no = 0;
3080         last_block = i_size_read(inode) >> blkbits;
3081         while ((probe_block + blocks_per_page) <= last_block && page_no < max) {
3082                 unsigned block_in_page;
3083                 sector_t first_block;
3084
3085                 cond_resched();
3086
3087                 first_block = bmap(inode, probe_block);
3088                 if (first_block == 0)
3089                         goto bad_bmap;
3090
3091                 /*
3092                  * It must be PAGE_SIZE aligned on-disk
3093                  */
3094                 if (first_block & (blocks_per_page - 1)) {
3095                         probe_block++;
3096                         goto reprobe;
3097                 }
3098
3099                 for (block_in_page = 1; block_in_page < blocks_per_page;
3100                                         block_in_page++) {
3101                         sector_t block;
3102
3103                         block = bmap(inode, probe_block + block_in_page);
3104                         if (block == 0)
3105                                 goto bad_bmap;
3106                         if (block != first_block + block_in_page) {
3107                                 /* Discontiguity */
3108                                 probe_block++;
3109                                 goto reprobe;
3110                         }
3111                 }
3112
3113                 first_block >>= (PAGE_SHIFT - blkbits);
3114                 if (page_no) {  /* exclude the header page */
3115                         if (first_block < lowest_block)
3116                                 lowest_block = first_block;
3117                         if (first_block > highest_block)
3118                                 highest_block = first_block;
3119                 }
3120
3121                 page_no++;
3122                 probe_block += blocks_per_page;
3123 reprobe:
3124                 continue;
3125         }
3126         return 0;
3127
3128 bad_bmap:
3129         pr_err("swapon: swapfile has holes\n");
3130         return -EINVAL;
3131 }
3132
3133 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3134                                 sector_t *span)
3135 {
3136         struct inode *inode = file_inode(file);
3137         int ret;
3138
3139         if (!S_ISREG(inode->i_mode))
3140                 return -EINVAL;
3141
3142         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3143                 return -EROFS;
3144
3145         ret = f2fs_convert_inline_inode(inode);
3146         if (ret)
3147                 return ret;
3148
3149         ret = check_swap_activate(file, sis->max);
3150         if (ret)
3151                 return ret;
3152
3153         set_inode_flag(inode, FI_PIN_FILE);
3154         f2fs_precache_extents(inode);
3155         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3156         return 0;
3157 }
3158
3159 static void f2fs_swap_deactivate(struct file *file)
3160 {
3161         struct inode *inode = file_inode(file);
3162
3163         clear_inode_flag(inode, FI_PIN_FILE);
3164 }
3165 #else
3166 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3167                                 sector_t *span)
3168 {
3169         return -EOPNOTSUPP;
3170 }
3171
3172 static void f2fs_swap_deactivate(struct file *file)
3173 {
3174 }
3175 #endif
3176
3177 const struct address_space_operations f2fs_dblock_aops = {
3178         .readpage       = f2fs_read_data_page,
3179         .readpages      = f2fs_read_data_pages,
3180         .writepage      = f2fs_write_data_page,
3181         .writepages     = f2fs_write_data_pages,
3182         .write_begin    = f2fs_write_begin,
3183         .write_end      = f2fs_write_end,
3184         .set_page_dirty = f2fs_set_data_page_dirty,
3185         .invalidatepage = f2fs_invalidate_page,
3186         .releasepage    = f2fs_release_page,
3187         .direct_IO      = f2fs_direct_IO,
3188         .bmap           = f2fs_bmap,
3189         .swap_activate  = f2fs_swap_activate,
3190         .swap_deactivate = f2fs_swap_deactivate,
3191 #ifdef CONFIG_MIGRATION
3192         .migratepage    = f2fs_migrate_page,
3193 #endif
3194 };
3195
3196 void f2fs_clear_radix_tree_dirty_tag(struct page *page)
3197 {
3198         struct address_space *mapping = page_mapping(page);
3199         unsigned long flags;
3200
3201         spin_lock_irqsave(&mapping->tree_lock, flags);
3202         radix_tree_tag_clear(&mapping->page_tree, page_index(page),
3203                                         PAGECACHE_TAG_DIRTY);
3204         spin_unlock_irqrestore(&mapping->tree_lock, flags);
3205 }
3206
3207 int __init f2fs_init_post_read_processing(void)
3208 {
3209         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
3210         if (!bio_post_read_ctx_cache)
3211                 goto fail;
3212         bio_post_read_ctx_pool =
3213                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3214                                          bio_post_read_ctx_cache);
3215         if (!bio_post_read_ctx_pool)
3216                 goto fail_free_cache;
3217         return 0;
3218
3219 fail_free_cache:
3220         kmem_cache_destroy(bio_post_read_ctx_cache);
3221 fail:
3222         return -ENOMEM;
3223 }
3224
3225 void __exit f2fs_destroy_post_read_processing(void)
3226 {
3227         mempool_destroy(bio_post_read_ctx_pool);
3228         kmem_cache_destroy(bio_post_read_ctx_cache);
3229 }