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/sched/mm.h>
  12 #include <linux/mpage.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/blk-crypto.h>
  18 #include <linux/swap.h>
  19 #include <linux/prefetch.h>
  20 #include <linux/uio.h>
  21 #include <linux/sched/signal.h>
  22 #include <linux/fiemap.h>
  23 #include <linux/iomap.h>
  24
  25 #include "f2fs.h"
  26 #include "node.h"
  27 #include "segment.h"
  28 #include "iostat.h"
  29 #include <trace/events/f2fs.h>
  30
  31 #define NUM_PREALLOC_POST_READ_CTXS     128
  32
  33 static struct kmem_cache *bio_post_read_ctx_cache;
  34 static struct kmem_cache *bio_entry_slab;
  35 static mempool_t *bio_post_read_ctx_pool;
  36 static struct bio_set f2fs_bioset;
  37
  38 #define F2FS_BIO_POOL_SIZE      NR_CURSEG_TYPE
  39
  40 int __init f2fs_init_bioset(void)
  41 {
  42         if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  43                                         0, BIOSET_NEED_BVECS))
  44                 return -ENOMEM;
  45         return 0;
  46 }
  47
  48 void f2fs_destroy_bioset(void)
  49 {
  50         bioset_exit(&f2fs_bioset);
  51 }
  52
  53 static bool __is_cp_guaranteed(struct page *page)
  54 {
  55         struct address_space *mapping = page->mapping;
  56         struct inode *inode;
  57         struct f2fs_sb_info *sbi;
  58
  59         if (!mapping)
  60                 return false;
  61
  62         inode = mapping->host;
  63         sbi = F2FS_I_SB(inode);
  64
  65         if (inode->i_ino == F2FS_META_INO(sbi) ||
  66                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  67                         S_ISDIR(inode->i_mode))
  68                 return true;
  69
  70         if (f2fs_is_compressed_page(page))
  71                 return false;
  72         if ((S_ISREG(inode->i_mode) &&
  73                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  74                         page_private_gcing(page))
  75                 return true;
  76         return false;
  77 }
  78
  79 static enum count_type __read_io_type(struct page *page)
  80 {
  81         struct address_space *mapping = page_file_mapping(page);
  82
  83         if (mapping) {
  84                 struct inode *inode = mapping->host;
  85                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  86
  87                 if (inode->i_ino == F2FS_META_INO(sbi))
  88                         return F2FS_RD_META;
  89
  90                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  91                         return F2FS_RD_NODE;
  92         }
  93         return F2FS_RD_DATA;
  94 }
  95
  96 /* postprocessing steps for read bios */
  97 enum bio_post_read_step {
  98 #ifdef CONFIG_FS_ENCRYPTION
  99         STEP_DECRYPT    = 1 << 0,
 100 #else
 101         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
 102 #endif
 103 #ifdef CONFIG_F2FS_FS_COMPRESSION
 104         STEP_DECOMPRESS = 1 << 1,
 105 #else
 106         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 107 #endif
 108 #ifdef CONFIG_FS_VERITY
 109         STEP_VERITY     = 1 << 2,
 110 #else
 111         STEP_VERITY     = 0,    /* compile out the verity-related code */
 112 #endif
 113 };
 114
 115 struct bio_post_read_ctx {
 116         struct bio *bio;
 117         struct f2fs_sb_info *sbi;
 118         struct work_struct work;
 119         unsigned int enabled_steps;
 120         block_t fs_blkaddr;
 121 };
 122
 123 static void f2fs_finish_read_bio(struct bio *bio)
 124 {
 125         struct bio_vec *bv;
 126         struct bvec_iter_all iter_all;
 127
 128         /*
 129          * Update and unlock the bio's pagecache pages, and put the
 130          * decompression context for any compressed pages.
 131          */
 132         bio_for_each_segment_all(bv, bio, iter_all) {
 133                 struct page *page = bv->bv_page;
 134
 135                 if (f2fs_is_compressed_page(page)) {
 136                         if (bio->bi_status)
 137                                 f2fs_end_read_compressed_page(page, true, 0);
 138                         f2fs_put_page_dic(page);
 139                         continue;
 140                 }
 141
 142                 /* PG_error was set if decryption or verity failed. */
 143                 if (bio->bi_status || PageError(page)) {
 144                         ClearPageUptodate(page);
 145                         /* will re-read again later */
 146                         ClearPageError(page);
 147                 } else {
 148                         SetPageUptodate(page);
 149                 }
 150                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 151                 unlock_page(page);
 152         }
 153
 154         if (bio->bi_private)
 155                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 156         bio_put(bio);
 157 }
 158
 159 static void f2fs_verify_bio(struct work_struct *work)
 160 {
 161         struct bio_post_read_ctx *ctx =
 162                 container_of(work, struct bio_post_read_ctx, work);
 163         struct bio *bio = ctx->bio;
 164         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 165
 166         /*
 167          * fsverity_verify_bio() may call readpages() again, and while verity
 168          * will be disabled for this, decryption and/or decompression may still
 169          * be needed, resulting in another bio_post_read_ctx being allocated.
 170          * So to prevent deadlocks we need to release the current ctx to the
 171          * mempool first.  This assumes that verity is the last post-read step.
 172          */
 173         mempool_free(ctx, bio_post_read_ctx_pool);
 174         bio->bi_private = NULL;
 175
 176         /*
 177          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 178          * as those were handled separately by f2fs_end_read_compressed_page().
 179          */
 180         if (may_have_compressed_pages) {
 181                 struct bio_vec *bv;
 182                 struct bvec_iter_all iter_all;
 183
 184                 bio_for_each_segment_all(bv, bio, iter_all) {
 185                         struct page *page = bv->bv_page;
 186
 187                         if (!f2fs_is_compressed_page(page) &&
 188                             !PageError(page) && !fsverity_verify_page(page))
 189                                 SetPageError(page);
 190                 }
 191         } else {
 192                 fsverity_verify_bio(bio);
 193         }
 194
 195         f2fs_finish_read_bio(bio);
 196 }
 197
 198 /*
 199  * If the bio's data needs to be verified with fs-verity, then enqueue the
 200  * verity work for the bio.  Otherwise finish the bio now.
 201  *
 202  * Note that to avoid deadlocks, the verity work can't be done on the
 203  * decryption/decompression workqueue.  This is because verifying the data pages
 204  * can involve reading verity metadata pages from the file, and these verity
 205  * metadata pages may be encrypted and/or compressed.
 206  */
 207 static void f2fs_verify_and_finish_bio(struct bio *bio)
 208 {
 209         struct bio_post_read_ctx *ctx = bio->bi_private;
 210
 211         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 212                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 213                 fsverity_enqueue_verify_work(&ctx->work);
 214         } else {
 215                 f2fs_finish_read_bio(bio);
 216         }
 217 }
 218
 219 /*
 220  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 221  * remaining page was read by @ctx->bio.
 222  *
 223  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 224  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 225  * that the bio includes at least one compressed page.  The actual decompression
 226  * is done on a per-cluster basis, not a per-bio basis.
 227  */
 228 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
 229 {
 230         struct bio_vec *bv;
 231         struct bvec_iter_all iter_all;
 232         bool all_compressed = true;
 233         block_t blkaddr = ctx->fs_blkaddr;
 234
 235         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 236                 struct page *page = bv->bv_page;
 237
 238                 /* PG_error was set if decryption failed. */
 239                 if (f2fs_is_compressed_page(page))
 240                         f2fs_end_read_compressed_page(page, PageError(page),
 241                                                 blkaddr);
 242                 else
 243                         all_compressed = false;
 244
 245                 blkaddr++;
 246         }
 247
 248         /*
 249          * Optimization: if all the bio's pages are compressed, then scheduling
 250          * the per-bio verity work is unnecessary, as verity will be fully
 251          * handled at the compression cluster level.
 252          */
 253         if (all_compressed)
 254                 ctx->enabled_steps &= ~STEP_VERITY;
 255 }
 256
 257 static void f2fs_post_read_work(struct work_struct *work)
 258 {
 259         struct bio_post_read_ctx *ctx =
 260                 container_of(work, struct bio_post_read_ctx, work);
 261
 262         if (ctx->enabled_steps & STEP_DECRYPT)
 263                 fscrypt_decrypt_bio(ctx->bio);
 264
 265         if (ctx->enabled_steps & STEP_DECOMPRESS)
 266                 f2fs_handle_step_decompress(ctx);
 267
 268         f2fs_verify_and_finish_bio(ctx->bio);
 269 }
 270
 271 static void f2fs_read_end_io(struct bio *bio)
 272 {
 273         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 274         struct bio_post_read_ctx *ctx;
 275
 276         iostat_update_and_unbind_ctx(bio, 0);
 277         ctx = bio->bi_private;
 278
 279         if (time_to_inject(sbi, FAULT_READ_IO)) {
 280                 f2fs_show_injection_info(sbi, FAULT_READ_IO);
 281                 bio->bi_status = BLK_STS_IOERR;
 282         }
 283
 284         if (bio->bi_status) {
 285                 f2fs_finish_read_bio(bio);
 286                 return;
 287         }
 288
 289         if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
 290                 INIT_WORK(&ctx->work, f2fs_post_read_work);
 291                 queue_work(ctx->sbi->post_read_wq, &ctx->work);
 292         } else {
 293                 f2fs_verify_and_finish_bio(bio);
 294         }
 295 }
 296
 297 static void f2fs_write_end_io(struct bio *bio)
 298 {
 299         struct f2fs_sb_info *sbi;
 300         struct bio_vec *bvec;
 301         struct bvec_iter_all iter_all;
 302
 303         iostat_update_and_unbind_ctx(bio, 1);
 304         sbi = bio->bi_private;
 305
 306         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 307                 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
 308                 bio->bi_status = BLK_STS_IOERR;
 309         }
 310
 311         bio_for_each_segment_all(bvec, bio, iter_all) {
 312                 struct page *page = bvec->bv_page;
 313                 enum count_type type = WB_DATA_TYPE(page);
 314
 315                 if (page_private_dummy(page)) {
 316                         clear_page_private_dummy(page);
 317                         unlock_page(page);
 318                         mempool_free(page, sbi->write_io_dummy);
 319
 320                         if (unlikely(bio->bi_status))
 321                                 f2fs_stop_checkpoint(sbi, true);
 322                         continue;
 323                 }
 324
 325                 fscrypt_finalize_bounce_page(&page);
 326
 327 #ifdef CONFIG_F2FS_FS_COMPRESSION
 328                 if (f2fs_is_compressed_page(page)) {
 329                         f2fs_compress_write_end_io(bio, page);
 330                         continue;
 331                 }
 332 #endif
 333
 334                 if (unlikely(bio->bi_status)) {
 335                         mapping_set_error(page->mapping, -EIO);
 336                         if (type == F2FS_WB_CP_DATA)
 337                                 f2fs_stop_checkpoint(sbi, true);
 338                 }
 339
 340                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 341                                         page->index != nid_of_node(page));
 342
 343                 dec_page_count(sbi, type);
 344                 if (f2fs_in_warm_node_list(sbi, page))
 345                         f2fs_del_fsync_node_entry(sbi, page);
 346                 clear_page_private_gcing(page);
 347                 end_page_writeback(page);
 348         }
 349         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 350                                 wq_has_sleeper(&sbi->cp_wait))
 351                 wake_up(&sbi->cp_wait);
 352
 353         bio_put(bio);
 354 }
 355
 356 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 357                 block_t blk_addr, sector_t *sector)
 358 {
 359         struct block_device *bdev = sbi->sb->s_bdev;
 360         int i;
 361
 362         if (f2fs_is_multi_device(sbi)) {
 363                 for (i = 0; i < sbi->s_ndevs; i++) {
 364                         if (FDEV(i).start_blk <= blk_addr &&
 365                             FDEV(i).end_blk >= blk_addr) {
 366                                 blk_addr -= FDEV(i).start_blk;
 367                                 bdev = FDEV(i).bdev;
 368                                 break;
 369                         }
 370                 }
 371         }
 372
 373         if (sector)
 374                 *sector = SECTOR_FROM_BLOCK(blk_addr);
 375         return bdev;
 376 }
 377
 378 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 379 {
 380         int i;
 381
 382         if (!f2fs_is_multi_device(sbi))
 383                 return 0;
 384
 385         for (i = 0; i < sbi->s_ndevs; i++)
 386                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 387                         return i;
 388         return 0;
 389 }
 390
 391 static void __attach_io_flag(struct f2fs_io_info *fio, unsigned int io_flag)
 392 {
 393         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 394         unsigned int fua_flag = io_flag & temp_mask;
 395         unsigned int meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 396
 397         /*
 398          * data/node io flag bits per temp:
 399          *      REQ_META     |      REQ_FUA      |
 400          *    5 |    4 |   3 |    2 |    1 |   0 |
 401          * Cold | Warm | Hot | Cold | Warm | Hot |
 402          */
 403         if ((1 << fio->temp) & meta_flag)
 404                 fio->op_flags |= REQ_META;
 405         if ((1 << fio->temp) & fua_flag)
 406                 fio->op_flags |= REQ_FUA;
 407 }
 408
 409 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 410 {
 411         struct f2fs_sb_info *sbi = fio->sbi;
 412         struct block_device *bdev;
 413         sector_t sector;
 414         struct bio *bio;
 415
 416         if (fio->type == DATA)
 417                 __attach_io_flag(fio, sbi->data_io_flag);
 418         else if (fio->type == NODE)
 419                 __attach_io_flag(fio, sbi->node_io_flag);
 420
 421         bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 422         bio = bio_alloc_bioset(bdev, npages, fio->op | fio->op_flags, GFP_NOIO,
 423                                &f2fs_bioset);
 424         bio->bi_iter.bi_sector = sector;
 425         if (is_read_io(fio->op)) {
 426                 bio->bi_end_io = f2fs_read_end_io;
 427                 bio->bi_private = NULL;
 428         } else {
 429                 bio->bi_end_io = f2fs_write_end_io;
 430                 bio->bi_private = sbi;
 431                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 432                                                 fio->type, fio->temp);
 433         }
 434         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 435
 436         if (fio->io_wbc)
 437                 wbc_init_bio(fio->io_wbc, bio);
 438
 439         return bio;
 440 }
 441
 442 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 443                                   pgoff_t first_idx,
 444                                   const struct f2fs_io_info *fio,
 445                                   gfp_t gfp_mask)
 446 {
 447         /*
 448          * The f2fs garbage collector sets ->encrypted_page when it wants to
 449          * read/write raw data without encryption.
 450          */
 451         if (!fio || !fio->encrypted_page)
 452                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 453 }
 454
 455 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 456                                      pgoff_t next_idx,
 457                                      const struct f2fs_io_info *fio)
 458 {
 459         /*
 460          * The f2fs garbage collector sets ->encrypted_page when it wants to
 461          * read/write raw data without encryption.
 462          */
 463         if (fio && fio->encrypted_page)
 464                 return !bio_has_crypt_ctx(bio);
 465
 466         return fscrypt_mergeable_bio(bio, inode, next_idx);
 467 }
 468
 469 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 470                                 struct bio *bio, enum page_type type)
 471 {
 472         if (!is_read_io(bio_op(bio))) {
 473                 unsigned int start;
 474
 475                 if (type != DATA && type != NODE)
 476                         goto submit_io;
 477
 478                 if (f2fs_lfs_mode(sbi) && current->plug)
 479                         blk_finish_plug(current->plug);
 480
 481                 if (!F2FS_IO_ALIGNED(sbi))
 482                         goto submit_io;
 483
 484                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 485                 start %= F2FS_IO_SIZE(sbi);
 486
 487                 if (start == 0)
 488                         goto submit_io;
 489
 490                 /* fill dummy pages */
 491                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 492                         struct page *page =
 493                                 mempool_alloc(sbi->write_io_dummy,
 494                                               GFP_NOIO | __GFP_NOFAIL);
 495                         f2fs_bug_on(sbi, !page);
 496
 497                         lock_page(page);
 498
 499                         zero_user_segment(page, 0, PAGE_SIZE);
 500                         set_page_private_dummy(page);
 501
 502                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 503                                 f2fs_bug_on(sbi, 1);
 504                 }
 505                 /*
 506                  * In the NODE case, we lose next block address chain. So, we
 507                  * need to do checkpoint in f2fs_sync_file.
 508                  */
 509                 if (type == NODE)
 510                         set_sbi_flag(sbi, SBI_NEED_CP);
 511         }
 512 submit_io:
 513         if (is_read_io(bio_op(bio)))
 514                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 515         else
 516                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 517
 518         iostat_update_submit_ctx(bio, type);
 519         submit_bio(bio);
 520 }
 521
 522 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 523                                 struct bio *bio, enum page_type type)
 524 {
 525         __submit_bio(sbi, bio, type);
 526 }
 527
 528 static void __submit_merged_bio(struct f2fs_bio_info *io)
 529 {
 530         struct f2fs_io_info *fio = &io->fio;
 531
 532         if (!io->bio)
 533                 return;
 534
 535         if (is_read_io(fio->op))
 536                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 537         else
 538                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 539
 540         __submit_bio(io->sbi, io->bio, fio->type);
 541         io->bio = NULL;
 542 }
 543
 544 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 545                                                 struct page *page, nid_t ino)
 546 {
 547         struct bio_vec *bvec;
 548         struct bvec_iter_all iter_all;
 549
 550         if (!bio)
 551                 return false;
 552
 553         if (!inode && !page && !ino)
 554                 return true;
 555
 556         bio_for_each_segment_all(bvec, bio, iter_all) {
 557                 struct page *target = bvec->bv_page;
 558
 559                 if (fscrypt_is_bounce_page(target)) {
 560                         target = fscrypt_pagecache_page(target);
 561                         if (IS_ERR(target))
 562                                 continue;
 563                 }
 564                 if (f2fs_is_compressed_page(target)) {
 565                         target = f2fs_compress_control_page(target);
 566                         if (IS_ERR(target))
 567                                 continue;
 568                 }
 569
 570                 if (inode && inode == target->mapping->host)
 571                         return true;
 572                 if (page && page == target)
 573                         return true;
 574                 if (ino && ino == ino_of_node(target))
 575                         return true;
 576         }
 577
 578         return false;
 579 }
 580
 581 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 582                                 enum page_type type, enum temp_type temp)
 583 {
 584         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 585         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 586
 587         f2fs_down_write(&io->io_rwsem);
 588
 589         /* change META to META_FLUSH in the checkpoint procedure */
 590         if (type >= META_FLUSH) {
 591                 io->fio.type = META_FLUSH;
 592                 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 593                 if (!test_opt(sbi, NOBARRIER))
 594                         io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 595         }
 596         __submit_merged_bio(io);
 597         f2fs_up_write(&io->io_rwsem);
 598 }
 599
 600 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 601                                 struct inode *inode, struct page *page,
 602                                 nid_t ino, enum page_type type, bool force)
 603 {
 604         enum temp_type temp;
 605         bool ret = true;
 606
 607         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 608                 if (!force)     {
 609                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 610                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 611
 612                         f2fs_down_read(&io->io_rwsem);
 613                         ret = __has_merged_page(io->bio, inode, page, ino);
 614                         f2fs_up_read(&io->io_rwsem);
 615                 }
 616                 if (ret)
 617                         __f2fs_submit_merged_write(sbi, type, temp);
 618
 619                 /* TODO: use HOT temp only for meta pages now. */
 620                 if (type >= META)
 621                         break;
 622         }
 623 }
 624
 625 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 626 {
 627         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 628 }
 629
 630 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 631                                 struct inode *inode, struct page *page,
 632                                 nid_t ino, enum page_type type)
 633 {
 634         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 635 }
 636
 637 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 638 {
 639         f2fs_submit_merged_write(sbi, DATA);
 640         f2fs_submit_merged_write(sbi, NODE);
 641         f2fs_submit_merged_write(sbi, META);
 642 }
 643
 644 /*
 645  * Fill the locked page with data located in the block address.
 646  * A caller needs to unlock the page on failure.
 647  */
 648 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 649 {
 650         struct bio *bio;
 651         struct page *page = fio->encrypted_page ?
 652                         fio->encrypted_page : fio->page;
 653
 654         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 655                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 656                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 657                 return -EFSCORRUPTED;
 658
 659         trace_f2fs_submit_page_bio(page, fio);
 660
 661         /* Allocate a new bio */
 662         bio = __bio_alloc(fio, 1);
 663
 664         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 665                                fio->page->index, fio, GFP_NOIO);
 666
 667         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 668                 bio_put(bio);
 669                 return -EFAULT;
 670         }
 671
 672         if (fio->io_wbc && !is_read_io(fio->op))
 673                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 674
 675         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 676                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 677
 678         __submit_bio(fio->sbi, bio, fio->type);
 679         return 0;
 680 }
 681
 682 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 683                                 block_t last_blkaddr, block_t cur_blkaddr)
 684 {
 685         if (unlikely(sbi->max_io_bytes &&
 686                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 687                 return false;
 688         if (last_blkaddr + 1 != cur_blkaddr)
 689                 return false;
 690         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 691 }
 692
 693 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 694                                                 struct f2fs_io_info *fio)
 695 {
 696         if (io->fio.op != fio->op)
 697                 return false;
 698         return io->fio.op_flags == fio->op_flags;
 699 }
 700
 701 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 702                                         struct f2fs_bio_info *io,
 703                                         struct f2fs_io_info *fio,
 704                                         block_t last_blkaddr,
 705                                         block_t cur_blkaddr)
 706 {
 707         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 708                 unsigned int filled_blocks =
 709                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 710                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 711                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 712
 713                 /* IOs in bio is aligned and left space of vectors is not enough */
 714                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 715                         return false;
 716         }
 717         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 718                 return false;
 719         return io_type_is_mergeable(io, fio);
 720 }
 721
 722 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 723                                 struct page *page, enum temp_type temp)
 724 {
 725         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 726         struct bio_entry *be;
 727
 728         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 729         be->bio = bio;
 730         bio_get(bio);
 731
 732         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 733                 f2fs_bug_on(sbi, 1);
 734
 735         f2fs_down_write(&io->bio_list_lock);
 736         list_add_tail(&be->list, &io->bio_list);
 737         f2fs_up_write(&io->bio_list_lock);
 738 }
 739
 740 static void del_bio_entry(struct bio_entry *be)
 741 {
 742         list_del(&be->list);
 743         kmem_cache_free(bio_entry_slab, be);
 744 }
 745
 746 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 747                                                         struct page *page)
 748 {
 749         struct f2fs_sb_info *sbi = fio->sbi;
 750         enum temp_type temp;
 751         bool found = false;
 752         int ret = -EAGAIN;
 753
 754         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 755                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 756                 struct list_head *head = &io->bio_list;
 757                 struct bio_entry *be;
 758
 759                 f2fs_down_write(&io->bio_list_lock);
 760                 list_for_each_entry(be, head, list) {
 761                         if (be->bio != *bio)
 762                                 continue;
 763
 764                         found = true;
 765
 766                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 767                                                             *fio->last_block,
 768                                                             fio->new_blkaddr));
 769                         if (f2fs_crypt_mergeable_bio(*bio,
 770                                         fio->page->mapping->host,
 771                                         fio->page->index, fio) &&
 772                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 773                                         PAGE_SIZE) {
 774                                 ret = 0;
 775                                 break;
 776                         }
 777
 778                         /* page can't be merged into bio; submit the bio */
 779                         del_bio_entry(be);
 780                         __submit_bio(sbi, *bio, DATA);
 781                         break;
 782                 }
 783                 f2fs_up_write(&io->bio_list_lock);
 784         }
 785
 786         if (ret) {
 787                 bio_put(*bio);
 788                 *bio = NULL;
 789         }
 790
 791         return ret;
 792 }
 793
 794 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 795                                         struct bio **bio, struct page *page)
 796 {
 797         enum temp_type temp;
 798         bool found = false;
 799         struct bio *target = bio ? *bio : NULL;
 800
 801         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 802                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 803                 struct list_head *head = &io->bio_list;
 804                 struct bio_entry *be;
 805
 806                 if (list_empty(head))
 807                         continue;
 808
 809                 f2fs_down_read(&io->bio_list_lock);
 810                 list_for_each_entry(be, head, list) {
 811                         if (target)
 812                                 found = (target == be->bio);
 813                         else
 814                                 found = __has_merged_page(be->bio, NULL,
 815                                                                 page, 0);
 816                         if (found)
 817                                 break;
 818                 }
 819                 f2fs_up_read(&io->bio_list_lock);
 820
 821                 if (!found)
 822                         continue;
 823
 824                 found = false;
 825
 826                 f2fs_down_write(&io->bio_list_lock);
 827                 list_for_each_entry(be, head, list) {
 828                         if (target)
 829                                 found = (target == be->bio);
 830                         else
 831                                 found = __has_merged_page(be->bio, NULL,
 832                                                                 page, 0);
 833                         if (found) {
 834                                 target = be->bio;
 835                                 del_bio_entry(be);
 836                                 break;
 837                         }
 838                 }
 839                 f2fs_up_write(&io->bio_list_lock);
 840         }
 841
 842         if (found)
 843                 __submit_bio(sbi, target, DATA);
 844         if (bio && *bio) {
 845                 bio_put(*bio);
 846                 *bio = NULL;
 847         }
 848 }
 849
 850 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 851 {
 852         struct bio *bio = *fio->bio;
 853         struct page *page = fio->encrypted_page ?
 854                         fio->encrypted_page : fio->page;
 855
 856         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 857                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 858                 return -EFSCORRUPTED;
 859
 860         trace_f2fs_submit_page_bio(page, fio);
 861
 862         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 863                                                 fio->new_blkaddr))
 864                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 865 alloc_new:
 866         if (!bio) {
 867                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 868                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 869                                        fio->page->index, fio, GFP_NOIO);
 870
 871                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 872         } else {
 873                 if (add_ipu_page(fio, &bio, page))
 874                         goto alloc_new;
 875         }
 876
 877         if (fio->io_wbc)
 878                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 879
 880         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 881
 882         *fio->last_block = fio->new_blkaddr;
 883         *fio->bio = bio;
 884
 885         return 0;
 886 }
 887
 888 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 889 {
 890         struct f2fs_sb_info *sbi = fio->sbi;
 891         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 892         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 893         struct page *bio_page;
 894
 895         f2fs_bug_on(sbi, is_read_io(fio->op));
 896
 897         f2fs_down_write(&io->io_rwsem);
 898 next:
 899         if (fio->in_list) {
 900                 spin_lock(&io->io_lock);
 901                 if (list_empty(&io->io_list)) {
 902                         spin_unlock(&io->io_lock);
 903                         goto out;
 904                 }
 905                 fio = list_first_entry(&io->io_list,
 906                                                 struct f2fs_io_info, list);
 907                 list_del(&fio->list);
 908                 spin_unlock(&io->io_lock);
 909         }
 910
 911         verify_fio_blkaddr(fio);
 912
 913         if (fio->encrypted_page)
 914                 bio_page = fio->encrypted_page;
 915         else if (fio->compressed_page)
 916                 bio_page = fio->compressed_page;
 917         else
 918                 bio_page = fio->page;
 919
 920         /* set submitted = true as a return value */
 921         fio->submitted = true;
 922
 923         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 924
 925         if (io->bio &&
 926             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 927                               fio->new_blkaddr) ||
 928              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 929                                        bio_page->index, fio)))
 930                 __submit_merged_bio(io);
 931 alloc_new:
 932         if (io->bio == NULL) {
 933                 if (F2FS_IO_ALIGNED(sbi) &&
 934                                 (fio->type == DATA || fio->type == NODE) &&
 935                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 936                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 937                         fio->retry = true;
 938                         goto skip;
 939                 }
 940                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
 941                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 942                                        bio_page->index, fio, GFP_NOIO);
 943                 io->fio = *fio;
 944         }
 945
 946         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 947                 __submit_merged_bio(io);
 948                 goto alloc_new;
 949         }
 950
 951         if (fio->io_wbc)
 952                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
 953
 954         io->last_block_in_bio = fio->new_blkaddr;
 955
 956         trace_f2fs_submit_page_write(fio->page, fio);
 957 skip:
 958         if (fio->in_list)
 959                 goto next;
 960 out:
 961         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 962                                 !f2fs_is_checkpoint_ready(sbi))
 963                 __submit_merged_bio(io);
 964         f2fs_up_write(&io->io_rwsem);
 965 }
 966
 967 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 968                                       unsigned nr_pages, unsigned op_flag,
 969                                       pgoff_t first_idx, bool for_write)
 970 {
 971         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 972         struct bio *bio;
 973         struct bio_post_read_ctx *ctx = NULL;
 974         unsigned int post_read_steps = 0;
 975         sector_t sector;
 976         struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
 977
 978         bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
 979                                REQ_OP_READ | op_flag,
 980                                for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
 981         if (!bio)
 982                 return ERR_PTR(-ENOMEM);
 983         bio->bi_iter.bi_sector = sector;
 984         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
 985         bio->bi_end_io = f2fs_read_end_io;
 986
 987         if (fscrypt_inode_uses_fs_layer_crypto(inode))
 988                 post_read_steps |= STEP_DECRYPT;
 989
 990         if (f2fs_need_verity(inode, first_idx))
 991                 post_read_steps |= STEP_VERITY;
 992
 993         /*
 994          * STEP_DECOMPRESS is handled specially, since a compressed file might
 995          * contain both compressed and uncompressed clusters.  We'll allocate a
 996          * bio_post_read_ctx if the file is compressed, but the caller is
 997          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
 998          */
 999
1000         if (post_read_steps || f2fs_compressed_file(inode)) {
1001                 /* Due to the mempool, this never fails. */
1002                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1003                 ctx->bio = bio;
1004                 ctx->sbi = sbi;
1005                 ctx->enabled_steps = post_read_steps;
1006                 ctx->fs_blkaddr = blkaddr;
1007                 bio->bi_private = ctx;
1008         }
1009         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1010
1011         return bio;
1012 }
1013
1014 /* This can handle encryption stuffs */
1015 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1016                                  block_t blkaddr, int op_flags, bool for_write)
1017 {
1018         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1019         struct bio *bio;
1020
1021         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1022                                         page->index, for_write);
1023         if (IS_ERR(bio))
1024                 return PTR_ERR(bio);
1025
1026         /* wait for GCed page writeback via META_MAPPING */
1027         f2fs_wait_on_block_writeback(inode, blkaddr);
1028
1029         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1030                 bio_put(bio);
1031                 return -EFAULT;
1032         }
1033         ClearPageError(page);
1034         inc_page_count(sbi, F2FS_RD_DATA);
1035         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1036         __submit_bio(sbi, bio, DATA);
1037         return 0;
1038 }
1039
1040 static void __set_data_blkaddr(struct dnode_of_data *dn)
1041 {
1042         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1043         __le32 *addr_array;
1044         int base = 0;
1045
1046         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1047                 base = get_extra_isize(dn->inode);
1048
1049         /* Get physical address of data block */
1050         addr_array = blkaddr_in_node(rn);
1051         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1052 }
1053
1054 /*
1055  * Lock ordering for the change of data block address:
1056  * ->data_page
1057  *  ->node_page
1058  *    update block addresses in the node page
1059  */
1060 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1061 {
1062         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1063         __set_data_blkaddr(dn);
1064         if (set_page_dirty(dn->node_page))
1065                 dn->node_changed = true;
1066 }
1067
1068 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1069 {
1070         dn->data_blkaddr = blkaddr;
1071         f2fs_set_data_blkaddr(dn);
1072         f2fs_update_extent_cache(dn);
1073 }
1074
1075 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1076 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1077 {
1078         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1079         int err;
1080
1081         if (!count)
1082                 return 0;
1083
1084         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1085                 return -EPERM;
1086         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1087                 return err;
1088
1089         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1090                                                 dn->ofs_in_node, count);
1091
1092         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1093
1094         for (; count > 0; dn->ofs_in_node++) {
1095                 block_t blkaddr = f2fs_data_blkaddr(dn);
1096
1097                 if (blkaddr == NULL_ADDR) {
1098                         dn->data_blkaddr = NEW_ADDR;
1099                         __set_data_blkaddr(dn);
1100                         count--;
1101                 }
1102         }
1103
1104         if (set_page_dirty(dn->node_page))
1105                 dn->node_changed = true;
1106         return 0;
1107 }
1108
1109 /* Should keep dn->ofs_in_node unchanged */
1110 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1111 {
1112         unsigned int ofs_in_node = dn->ofs_in_node;
1113         int ret;
1114
1115         ret = f2fs_reserve_new_blocks(dn, 1);
1116         dn->ofs_in_node = ofs_in_node;
1117         return ret;
1118 }
1119
1120 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1121 {
1122         bool need_put = dn->inode_page ? false : true;
1123         int err;
1124
1125         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1126         if (err)
1127                 return err;
1128
1129         if (dn->data_blkaddr == NULL_ADDR)
1130                 err = f2fs_reserve_new_block(dn);
1131         if (err || need_put)
1132                 f2fs_put_dnode(dn);
1133         return err;
1134 }
1135
1136 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1137 {
1138         struct extent_info ei = {0, };
1139         struct inode *inode = dn->inode;
1140
1141         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1142                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1143                 return 0;
1144         }
1145
1146         return f2fs_reserve_block(dn, index);
1147 }
1148
1149 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1150                                                 int op_flags, bool for_write)
1151 {
1152         struct address_space *mapping = inode->i_mapping;
1153         struct dnode_of_data dn;
1154         struct page *page;
1155         struct extent_info ei = {0, };
1156         int err;
1157
1158         page = f2fs_grab_cache_page(mapping, index, for_write);
1159         if (!page)
1160                 return ERR_PTR(-ENOMEM);
1161
1162         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1163                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1164                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1165                                                 DATA_GENERIC_ENHANCE_READ)) {
1166                         err = -EFSCORRUPTED;
1167                         goto put_err;
1168                 }
1169                 goto got_it;
1170         }
1171
1172         set_new_dnode(&dn, inode, NULL, NULL, 0);
1173         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1174         if (err)
1175                 goto put_err;
1176         f2fs_put_dnode(&dn);
1177
1178         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1179                 err = -ENOENT;
1180                 goto put_err;
1181         }
1182         if (dn.data_blkaddr != NEW_ADDR &&
1183                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1184                                                 dn.data_blkaddr,
1185                                                 DATA_GENERIC_ENHANCE)) {
1186                 err = -EFSCORRUPTED;
1187                 goto put_err;
1188         }
1189 got_it:
1190         if (PageUptodate(page)) {
1191                 unlock_page(page);
1192                 return page;
1193         }
1194
1195         /*
1196          * A new dentry page is allocated but not able to be written, since its
1197          * new inode page couldn't be allocated due to -ENOSPC.
1198          * In such the case, its blkaddr can be remained as NEW_ADDR.
1199          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1200          * f2fs_init_inode_metadata.
1201          */
1202         if (dn.data_blkaddr == NEW_ADDR) {
1203                 zero_user_segment(page, 0, PAGE_SIZE);
1204                 if (!PageUptodate(page))
1205                         SetPageUptodate(page);
1206                 unlock_page(page);
1207                 return page;
1208         }
1209
1210         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1211                                                 op_flags, for_write);
1212         if (err)
1213                 goto put_err;
1214         return page;
1215
1216 put_err:
1217         f2fs_put_page(page, 1);
1218         return ERR_PTR(err);
1219 }
1220
1221 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1222 {
1223         struct address_space *mapping = inode->i_mapping;
1224         struct page *page;
1225
1226         page = find_get_page(mapping, index);
1227         if (page && PageUptodate(page))
1228                 return page;
1229         f2fs_put_page(page, 0);
1230
1231         page = f2fs_get_read_data_page(inode, index, 0, false);
1232         if (IS_ERR(page))
1233                 return page;
1234
1235         if (PageUptodate(page))
1236                 return page;
1237
1238         wait_on_page_locked(page);
1239         if (unlikely(!PageUptodate(page))) {
1240                 f2fs_put_page(page, 0);
1241                 return ERR_PTR(-EIO);
1242         }
1243         return page;
1244 }
1245
1246 /*
1247  * If it tries to access a hole, return an error.
1248  * Because, the callers, functions in dir.c and GC, should be able to know
1249  * whether this page exists or not.
1250  */
1251 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1252                                                         bool for_write)
1253 {
1254         struct address_space *mapping = inode->i_mapping;
1255         struct page *page;
1256 repeat:
1257         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1258         if (IS_ERR(page))
1259                 return page;
1260
1261         /* wait for read completion */
1262         lock_page(page);
1263         if (unlikely(page->mapping != mapping)) {
1264                 f2fs_put_page(page, 1);
1265                 goto repeat;
1266         }
1267         if (unlikely(!PageUptodate(page))) {
1268                 f2fs_put_page(page, 1);
1269                 return ERR_PTR(-EIO);
1270         }
1271         return page;
1272 }
1273
1274 /*
1275  * Caller ensures that this data page is never allocated.
1276  * A new zero-filled data page is allocated in the page cache.
1277  *
1278  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1279  * f2fs_unlock_op().
1280  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1281  * ipage should be released by this function.
1282  */
1283 struct page *f2fs_get_new_data_page(struct inode *inode,
1284                 struct page *ipage, pgoff_t index, bool new_i_size)
1285 {
1286         struct address_space *mapping = inode->i_mapping;
1287         struct page *page;
1288         struct dnode_of_data dn;
1289         int err;
1290
1291         page = f2fs_grab_cache_page(mapping, index, true);
1292         if (!page) {
1293                 /*
1294                  * before exiting, we should make sure ipage will be released
1295                  * if any error occur.
1296                  */
1297                 f2fs_put_page(ipage, 1);
1298                 return ERR_PTR(-ENOMEM);
1299         }
1300
1301         set_new_dnode(&dn, inode, ipage, NULL, 0);
1302         err = f2fs_reserve_block(&dn, index);
1303         if (err) {
1304                 f2fs_put_page(page, 1);
1305                 return ERR_PTR(err);
1306         }
1307         if (!ipage)
1308                 f2fs_put_dnode(&dn);
1309
1310         if (PageUptodate(page))
1311                 goto got_it;
1312
1313         if (dn.data_blkaddr == NEW_ADDR) {
1314                 zero_user_segment(page, 0, PAGE_SIZE);
1315                 if (!PageUptodate(page))
1316                         SetPageUptodate(page);
1317         } else {
1318                 f2fs_put_page(page, 1);
1319
1320                 /* if ipage exists, blkaddr should be NEW_ADDR */
1321                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1322                 page = f2fs_get_lock_data_page(inode, index, true);
1323                 if (IS_ERR(page))
1324                         return page;
1325         }
1326 got_it:
1327         if (new_i_size && i_size_read(inode) <
1328                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1329                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1330         return page;
1331 }
1332
1333 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1334 {
1335         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1336         struct f2fs_summary sum;
1337         struct node_info ni;
1338         block_t old_blkaddr;
1339         blkcnt_t count = 1;
1340         int err;
1341
1342         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1343                 return -EPERM;
1344
1345         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1346         if (err)
1347                 return err;
1348
1349         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1350         if (dn->data_blkaddr != NULL_ADDR)
1351                 goto alloc;
1352
1353         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1354                 return err;
1355
1356 alloc:
1357         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1358         old_blkaddr = dn->data_blkaddr;
1359         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1360                                 &sum, seg_type, NULL);
1361         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1362                 invalidate_mapping_pages(META_MAPPING(sbi),
1363                                         old_blkaddr, old_blkaddr);
1364                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1365         }
1366         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1367         return 0;
1368 }
1369
1370 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1371 {
1372         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1373                 if (lock)
1374                         f2fs_down_read(&sbi->node_change);
1375                 else
1376                         f2fs_up_read(&sbi->node_change);
1377         } else {
1378                 if (lock)
1379                         f2fs_lock_op(sbi);
1380                 else
1381                         f2fs_unlock_op(sbi);
1382         }
1383 }
1384
1385 /*
1386  * f2fs_map_blocks() tries to find or build mapping relationship which
1387  * maps continuous logical blocks to physical blocks, and return such
1388  * info via f2fs_map_blocks structure.
1389  */
1390 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1391                                                 int create, int flag)
1392 {
1393         unsigned int maxblocks = map->m_len;
1394         struct dnode_of_data dn;
1395         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1396         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1397         pgoff_t pgofs, end_offset, end;
1398         int err = 0, ofs = 1;
1399         unsigned int ofs_in_node, last_ofs_in_node;
1400         blkcnt_t prealloc;
1401         struct extent_info ei = {0, };
1402         block_t blkaddr;
1403         unsigned int start_pgofs;
1404         int bidx = 0;
1405
1406         if (!maxblocks)
1407                 return 0;
1408
1409         map->m_bdev = inode->i_sb->s_bdev;
1410         map->m_multidev_dio =
1411                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1412
1413         map->m_len = 0;
1414         map->m_flags = 0;
1415
1416         /* it only supports block size == page size */
1417         pgofs = (pgoff_t)map->m_lblk;
1418         end = pgofs + maxblocks;
1419
1420         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1421                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1422                                                         map->m_may_create)
1423                         goto next_dnode;
1424
1425                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1426                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1427                 map->m_flags = F2FS_MAP_MAPPED;
1428                 if (map->m_next_extent)
1429                         *map->m_next_extent = pgofs + map->m_len;
1430
1431                 /* for hardware encryption, but to avoid potential issue in future */
1432                 if (flag == F2FS_GET_BLOCK_DIO)
1433                         f2fs_wait_on_block_writeback_range(inode,
1434                                                 map->m_pblk, map->m_len);
1435
1436                 if (map->m_multidev_dio) {
1437                         block_t blk_addr = map->m_pblk;
1438
1439                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1440
1441                         map->m_bdev = FDEV(bidx).bdev;
1442                         map->m_pblk -= FDEV(bidx).start_blk;
1443                         map->m_len = min(map->m_len,
1444                                 FDEV(bidx).end_blk + 1 - map->m_pblk);
1445
1446                         if (map->m_may_create)
1447                                 f2fs_update_device_state(sbi, inode->i_ino,
1448                                                         blk_addr, map->m_len);
1449                 }
1450                 goto out;
1451         }
1452
1453 next_dnode:
1454         if (map->m_may_create)
1455                 f2fs_do_map_lock(sbi, flag, true);
1456
1457         /* When reading holes, we need its node page */
1458         set_new_dnode(&dn, inode, NULL, NULL, 0);
1459         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1460         if (err) {
1461                 if (flag == F2FS_GET_BLOCK_BMAP)
1462                         map->m_pblk = 0;
1463
1464                 if (err == -ENOENT) {
1465                         /*
1466                          * There is one exceptional case that read_node_page()
1467                          * may return -ENOENT due to filesystem has been
1468                          * shutdown or cp_error, so force to convert error
1469                          * number to EIO for such case.
1470                          */
1471                         if (map->m_may_create &&
1472                                 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1473                                 f2fs_cp_error(sbi))) {
1474                                 err = -EIO;
1475                                 goto unlock_out;
1476                         }
1477
1478                         err = 0;
1479                         if (map->m_next_pgofs)
1480                                 *map->m_next_pgofs =
1481                                         f2fs_get_next_page_offset(&dn, pgofs);
1482                         if (map->m_next_extent)
1483                                 *map->m_next_extent =
1484                                         f2fs_get_next_page_offset(&dn, pgofs);
1485                 }
1486                 goto unlock_out;
1487         }
1488
1489         start_pgofs = pgofs;
1490         prealloc = 0;
1491         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1492         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1493
1494 next_block:
1495         blkaddr = f2fs_data_blkaddr(&dn);
1496
1497         if (__is_valid_data_blkaddr(blkaddr) &&
1498                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1499                 err = -EFSCORRUPTED;
1500                 goto sync_out;
1501         }
1502
1503         if (__is_valid_data_blkaddr(blkaddr)) {
1504                 /* use out-place-update for driect IO under LFS mode */
1505                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1506                                                         map->m_may_create) {
1507                         err = __allocate_data_block(&dn, map->m_seg_type);
1508                         if (err)
1509                                 goto sync_out;
1510                         blkaddr = dn.data_blkaddr;
1511                         set_inode_flag(inode, FI_APPEND_WRITE);
1512                 }
1513         } else {
1514                 if (create) {
1515                         if (unlikely(f2fs_cp_error(sbi))) {
1516                                 err = -EIO;
1517                                 goto sync_out;
1518                         }
1519                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1520                                 if (blkaddr == NULL_ADDR) {
1521                                         prealloc++;
1522                                         last_ofs_in_node = dn.ofs_in_node;
1523                                 }
1524                         } else {
1525                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1526                                         flag != F2FS_GET_BLOCK_DIO);
1527                                 err = __allocate_data_block(&dn,
1528                                                         map->m_seg_type);
1529                                 if (!err) {
1530                                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1531                                                 file_need_truncate(inode);
1532                                         set_inode_flag(inode, FI_APPEND_WRITE);
1533                                 }
1534                         }
1535                         if (err)
1536                                 goto sync_out;
1537                         map->m_flags |= F2FS_MAP_NEW;
1538                         blkaddr = dn.data_blkaddr;
1539                 } else {
1540                         if (f2fs_compressed_file(inode) &&
1541                                         f2fs_sanity_check_cluster(&dn) &&
1542                                         (flag != F2FS_GET_BLOCK_FIEMAP ||
1543                                         IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1544                                 err = -EFSCORRUPTED;
1545                                 goto sync_out;
1546                         }
1547                         if (flag == F2FS_GET_BLOCK_BMAP) {
1548                                 map->m_pblk = 0;
1549                                 goto sync_out;
1550                         }
1551                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1552                                 goto sync_out;
1553                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1554                                                 blkaddr == NULL_ADDR) {
1555                                 if (map->m_next_pgofs)
1556                                         *map->m_next_pgofs = pgofs + 1;
1557                                 goto sync_out;
1558                         }
1559                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1560                                 /* for defragment case */
1561                                 if (map->m_next_pgofs)
1562                                         *map->m_next_pgofs = pgofs + 1;
1563                                 goto sync_out;
1564                         }
1565                 }
1566         }
1567
1568         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1569                 goto skip;
1570
1571         if (map->m_multidev_dio)
1572                 bidx = f2fs_target_device_index(sbi, blkaddr);
1573
1574         if (map->m_len == 0) {
1575                 /* preallocated unwritten block should be mapped for fiemap. */
1576                 if (blkaddr == NEW_ADDR)
1577                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1578                 map->m_flags |= F2FS_MAP_MAPPED;
1579
1580                 map->m_pblk = blkaddr;
1581                 map->m_len = 1;
1582
1583                 if (map->m_multidev_dio)
1584                         map->m_bdev = FDEV(bidx).bdev;
1585         } else if ((map->m_pblk != NEW_ADDR &&
1586                         blkaddr == (map->m_pblk + ofs)) ||
1587                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1588                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1589                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1590                         goto sync_out;
1591                 ofs++;
1592                 map->m_len++;
1593         } else {
1594                 goto sync_out;
1595         }
1596
1597 skip:
1598         dn.ofs_in_node++;
1599         pgofs++;
1600
1601         /* preallocate blocks in batch for one dnode page */
1602         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1603                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1604
1605                 dn.ofs_in_node = ofs_in_node;
1606                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1607                 if (err)
1608                         goto sync_out;
1609
1610                 map->m_len += dn.ofs_in_node - ofs_in_node;
1611                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1612                         err = -ENOSPC;
1613                         goto sync_out;
1614                 }
1615                 dn.ofs_in_node = end_offset;
1616         }
1617
1618         if (pgofs >= end)
1619                 goto sync_out;
1620         else if (dn.ofs_in_node < end_offset)
1621                 goto next_block;
1622
1623         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1624                 if (map->m_flags & F2FS_MAP_MAPPED) {
1625                         unsigned int ofs = start_pgofs - map->m_lblk;
1626
1627                         f2fs_update_extent_cache_range(&dn,
1628                                 start_pgofs, map->m_pblk + ofs,
1629                                 map->m_len - ofs);
1630                 }
1631         }
1632
1633         f2fs_put_dnode(&dn);
1634
1635         if (map->m_may_create) {
1636                 f2fs_do_map_lock(sbi, flag, false);
1637                 f2fs_balance_fs(sbi, dn.node_changed);
1638         }
1639         goto next_dnode;
1640
1641 sync_out:
1642
1643         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1644                 /*
1645                  * for hardware encryption, but to avoid potential issue
1646                  * in future
1647                  */
1648                 f2fs_wait_on_block_writeback_range(inode,
1649                                                 map->m_pblk, map->m_len);
1650                 invalidate_mapping_pages(META_MAPPING(sbi),
1651                                                 map->m_pblk, map->m_pblk);
1652
1653                 if (map->m_multidev_dio) {
1654                         block_t blk_addr = map->m_pblk;
1655
1656                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1657
1658                         map->m_bdev = FDEV(bidx).bdev;
1659                         map->m_pblk -= FDEV(bidx).start_blk;
1660
1661                         if (map->m_may_create)
1662                                 f2fs_update_device_state(sbi, inode->i_ino,
1663                                                         blk_addr, map->m_len);
1664
1665                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1666                                                 FDEV(bidx).end_blk + 1);
1667                 }
1668         }
1669
1670         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1671                 if (map->m_flags & F2FS_MAP_MAPPED) {
1672                         unsigned int ofs = start_pgofs - map->m_lblk;
1673
1674                         f2fs_update_extent_cache_range(&dn,
1675                                 start_pgofs, map->m_pblk + ofs,
1676                                 map->m_len - ofs);
1677                 }
1678                 if (map->m_next_extent)
1679                         *map->m_next_extent = pgofs + 1;
1680         }
1681         f2fs_put_dnode(&dn);
1682 unlock_out:
1683         if (map->m_may_create) {
1684                 f2fs_do_map_lock(sbi, flag, false);
1685                 f2fs_balance_fs(sbi, dn.node_changed);
1686         }
1687 out:
1688         trace_f2fs_map_blocks(inode, map, create, flag, err);
1689         return err;
1690 }
1691
1692 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1693 {
1694         struct f2fs_map_blocks map;
1695         block_t last_lblk;
1696         int err;
1697
1698         if (pos + len > i_size_read(inode))
1699                 return false;
1700
1701         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1702         map.m_next_pgofs = NULL;
1703         map.m_next_extent = NULL;
1704         map.m_seg_type = NO_CHECK_TYPE;
1705         map.m_may_create = false;
1706         last_lblk = F2FS_BLK_ALIGN(pos + len);
1707
1708         while (map.m_lblk < last_lblk) {
1709                 map.m_len = last_lblk - map.m_lblk;
1710                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1711                 if (err || map.m_len == 0)
1712                         return false;
1713                 map.m_lblk += map.m_len;
1714         }
1715         return true;
1716 }
1717
1718 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1719 {
1720         return (bytes >> inode->i_blkbits);
1721 }
1722
1723 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1724 {
1725         return (blks << inode->i_blkbits);
1726 }
1727
1728 static int f2fs_xattr_fiemap(struct inode *inode,
1729                                 struct fiemap_extent_info *fieinfo)
1730 {
1731         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1732         struct page *page;
1733         struct node_info ni;
1734         __u64 phys = 0, len;
1735         __u32 flags;
1736         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1737         int err = 0;
1738
1739         if (f2fs_has_inline_xattr(inode)) {
1740                 int offset;
1741
1742                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1743                                                 inode->i_ino, false);
1744                 if (!page)
1745                         return -ENOMEM;
1746
1747                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1748                 if (err) {
1749                         f2fs_put_page(page, 1);
1750                         return err;
1751                 }
1752
1753                 phys = blks_to_bytes(inode, ni.blk_addr);
1754                 offset = offsetof(struct f2fs_inode, i_addr) +
1755                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1756                                         get_inline_xattr_addrs(inode));
1757
1758                 phys += offset;
1759                 len = inline_xattr_size(inode);
1760
1761                 f2fs_put_page(page, 1);
1762
1763                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1764
1765                 if (!xnid)
1766                         flags |= FIEMAP_EXTENT_LAST;
1767
1768                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1769                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1770                 if (err || err == 1)
1771                         return err;
1772         }
1773
1774         if (xnid) {
1775                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1776                 if (!page)
1777                         return -ENOMEM;
1778
1779                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1780                 if (err) {
1781                         f2fs_put_page(page, 1);
1782                         return err;
1783                 }
1784
1785                 phys = blks_to_bytes(inode, ni.blk_addr);
1786                 len = inode->i_sb->s_blocksize;
1787
1788                 f2fs_put_page(page, 1);
1789
1790                 flags = FIEMAP_EXTENT_LAST;
1791         }
1792
1793         if (phys) {
1794                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1795                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1796         }
1797
1798         return (err < 0 ? err : 0);
1799 }
1800
1801 static loff_t max_inode_blocks(struct inode *inode)
1802 {
1803         loff_t result = ADDRS_PER_INODE(inode);
1804         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1805
1806         /* two direct node blocks */
1807         result += (leaf_count * 2);
1808
1809         /* two indirect node blocks */
1810         leaf_count *= NIDS_PER_BLOCK;
1811         result += (leaf_count * 2);
1812
1813         /* one double indirect node block */
1814         leaf_count *= NIDS_PER_BLOCK;
1815         result += leaf_count;
1816
1817         return result;
1818 }
1819
1820 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1821                 u64 start, u64 len)
1822 {
1823         struct f2fs_map_blocks map;
1824         sector_t start_blk, last_blk;
1825         pgoff_t next_pgofs;
1826         u64 logical = 0, phys = 0, size = 0;
1827         u32 flags = 0;
1828         int ret = 0;
1829         bool compr_cluster = false, compr_appended;
1830         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1831         unsigned int count_in_cluster = 0;
1832         loff_t maxbytes;
1833
1834         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1835                 ret = f2fs_precache_extents(inode);
1836                 if (ret)
1837                         return ret;
1838         }
1839
1840         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1841         if (ret)
1842                 return ret;
1843
1844         inode_lock(inode);
1845
1846         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1847         if (start > maxbytes) {
1848                 ret = -EFBIG;
1849                 goto out;
1850         }
1851
1852         if (len > maxbytes || (maxbytes - len) < start)
1853                 len = maxbytes - start;
1854
1855         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1856                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1857                 goto out;
1858         }
1859
1860         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1861                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1862                 if (ret != -EAGAIN)
1863                         goto out;
1864         }
1865
1866         if (bytes_to_blks(inode, len) == 0)
1867                 len = blks_to_bytes(inode, 1);
1868
1869         start_blk = bytes_to_blks(inode, start);
1870         last_blk = bytes_to_blks(inode, start + len - 1);
1871
1872 next:
1873         memset(&map, 0, sizeof(map));
1874         map.m_lblk = start_blk;
1875         map.m_len = bytes_to_blks(inode, len);
1876         map.m_next_pgofs = &next_pgofs;
1877         map.m_seg_type = NO_CHECK_TYPE;
1878
1879         if (compr_cluster) {
1880                 map.m_lblk += 1;
1881                 map.m_len = cluster_size - count_in_cluster;
1882         }
1883
1884         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1885         if (ret)
1886                 goto out;
1887
1888         /* HOLE */
1889         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1890                 start_blk = next_pgofs;
1891
1892                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1893                                                 max_inode_blocks(inode)))
1894                         goto prep_next;
1895
1896                 flags |= FIEMAP_EXTENT_LAST;
1897         }
1898
1899         compr_appended = false;
1900         /* In a case of compressed cluster, append this to the last extent */
1901         if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1902                         !(map.m_flags & F2FS_MAP_FLAGS))) {
1903                 compr_appended = true;
1904                 goto skip_fill;
1905         }
1906
1907         if (size) {
1908                 flags |= FIEMAP_EXTENT_MERGED;
1909                 if (IS_ENCRYPTED(inode))
1910                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1911
1912                 ret = fiemap_fill_next_extent(fieinfo, logical,
1913                                 phys, size, flags);
1914                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1915                 if (ret)
1916                         goto out;
1917                 size = 0;
1918         }
1919
1920         if (start_blk > last_blk)
1921                 goto out;
1922
1923 skip_fill:
1924         if (map.m_pblk == COMPRESS_ADDR) {
1925                 compr_cluster = true;
1926                 count_in_cluster = 1;
1927         } else if (compr_appended) {
1928                 unsigned int appended_blks = cluster_size -
1929                                                 count_in_cluster + 1;
1930                 size += blks_to_bytes(inode, appended_blks);
1931                 start_blk += appended_blks;
1932                 compr_cluster = false;
1933         } else {
1934                 logical = blks_to_bytes(inode, start_blk);
1935                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
1936                         blks_to_bytes(inode, map.m_pblk) : 0;
1937                 size = blks_to_bytes(inode, map.m_len);
1938                 flags = 0;
1939
1940                 if (compr_cluster) {
1941                         flags = FIEMAP_EXTENT_ENCODED;
1942                         count_in_cluster += map.m_len;
1943                         if (count_in_cluster == cluster_size) {
1944                                 compr_cluster = false;
1945                                 size += blks_to_bytes(inode, 1);
1946                         }
1947                 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
1948                         flags = FIEMAP_EXTENT_UNWRITTEN;
1949                 }
1950
1951                 start_blk += bytes_to_blks(inode, size);
1952         }
1953
1954 prep_next:
1955         cond_resched();
1956         if (fatal_signal_pending(current))
1957                 ret = -EINTR;
1958         else
1959                 goto next;
1960 out:
1961         if (ret == 1)
1962                 ret = 0;
1963
1964         inode_unlock(inode);
1965         return ret;
1966 }
1967
1968 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1969 {
1970         if (IS_ENABLED(CONFIG_FS_VERITY) &&
1971             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1972                 return inode->i_sb->s_maxbytes;
1973
1974         return i_size_read(inode);
1975 }
1976
1977 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1978                                         unsigned nr_pages,
1979                                         struct f2fs_map_blocks *map,
1980                                         struct bio **bio_ret,
1981                                         sector_t *last_block_in_bio,
1982                                         bool is_readahead)
1983 {
1984         struct bio *bio = *bio_ret;
1985         const unsigned blocksize = blks_to_bytes(inode, 1);
1986         sector_t block_in_file;
1987         sector_t last_block;
1988         sector_t last_block_in_file;
1989         sector_t block_nr;
1990         int ret = 0;
1991
1992         block_in_file = (sector_t)page_index(page);
1993         last_block = block_in_file + nr_pages;
1994         last_block_in_file = bytes_to_blks(inode,
1995                         f2fs_readpage_limit(inode) + blocksize - 1);
1996         if (last_block > last_block_in_file)
1997                 last_block = last_block_in_file;
1998
1999         /* just zeroing out page which is beyond EOF */
2000         if (block_in_file >= last_block)
2001                 goto zero_out;
2002         /*
2003          * Map blocks using the previous result first.
2004          */
2005         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2006                         block_in_file > map->m_lblk &&
2007                         block_in_file < (map->m_lblk + map->m_len))
2008                 goto got_it;
2009
2010         /*
2011          * Then do more f2fs_map_blocks() calls until we are
2012          * done with this page.
2013          */
2014         map->m_lblk = block_in_file;
2015         map->m_len = last_block - block_in_file;
2016
2017         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2018         if (ret)
2019                 goto out;
2020 got_it:
2021         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2022                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2023                 SetPageMappedToDisk(page);
2024
2025                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2026                                                 DATA_GENERIC_ENHANCE_READ)) {
2027                         ret = -EFSCORRUPTED;
2028                         goto out;
2029                 }
2030         } else {
2031 zero_out:
2032                 zero_user_segment(page, 0, PAGE_SIZE);
2033                 if (f2fs_need_verity(inode, page->index) &&
2034                     !fsverity_verify_page(page)) {
2035                         ret = -EIO;
2036                         goto out;
2037                 }
2038                 if (!PageUptodate(page))
2039                         SetPageUptodate(page);
2040                 unlock_page(page);
2041                 goto out;
2042         }
2043
2044         /*
2045          * This page will go to BIO.  Do we need to send this
2046          * BIO off first?
2047          */
2048         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2049                                        *last_block_in_bio, block_nr) ||
2050                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2051 submit_and_realloc:
2052                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2053                 bio = NULL;
2054         }
2055         if (bio == NULL) {
2056                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2057                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2058                                 false);
2059                 if (IS_ERR(bio)) {
2060                         ret = PTR_ERR(bio);
2061                         bio = NULL;
2062                         goto out;
2063                 }
2064         }
2065
2066         /*
2067          * If the page is under writeback, we need to wait for
2068          * its completion to see the correct decrypted data.
2069          */
2070         f2fs_wait_on_block_writeback(inode, block_nr);
2071
2072         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2073                 goto submit_and_realloc;
2074
2075         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2076         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2077         ClearPageError(page);
2078         *last_block_in_bio = block_nr;
2079         goto out;
2080 out:
2081         *bio_ret = bio;
2082         return ret;
2083 }
2084
2085 #ifdef CONFIG_F2FS_FS_COMPRESSION
2086 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2087                                 unsigned nr_pages, sector_t *last_block_in_bio,
2088                                 bool is_readahead, bool for_write)
2089 {
2090         struct dnode_of_data dn;
2091         struct inode *inode = cc->inode;
2092         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2093         struct bio *bio = *bio_ret;
2094         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2095         sector_t last_block_in_file;
2096         const unsigned blocksize = blks_to_bytes(inode, 1);
2097         struct decompress_io_ctx *dic = NULL;
2098         struct extent_info ei = {0, };
2099         bool from_dnode = true;
2100         int i;
2101         int ret = 0;
2102
2103         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2104
2105         last_block_in_file = bytes_to_blks(inode,
2106                         f2fs_readpage_limit(inode) + blocksize - 1);
2107
2108         /* get rid of pages beyond EOF */
2109         for (i = 0; i < cc->cluster_size; i++) {
2110                 struct page *page = cc->rpages[i];
2111
2112                 if (!page)
2113                         continue;
2114                 if ((sector_t)page->index >= last_block_in_file) {
2115                         zero_user_segment(page, 0, PAGE_SIZE);
2116                         if (!PageUptodate(page))
2117                                 SetPageUptodate(page);
2118                 } else if (!PageUptodate(page)) {
2119                         continue;
2120                 }
2121                 unlock_page(page);
2122                 if (for_write)
2123                         put_page(page);
2124                 cc->rpages[i] = NULL;
2125                 cc->nr_rpages--;
2126         }
2127
2128         /* we are done since all pages are beyond EOF */
2129         if (f2fs_cluster_is_empty(cc))
2130                 goto out;
2131
2132         if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2133                 from_dnode = false;
2134
2135         if (!from_dnode)
2136                 goto skip_reading_dnode;
2137
2138         set_new_dnode(&dn, inode, NULL, NULL, 0);
2139         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2140         if (ret)
2141                 goto out;
2142
2143         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2144
2145 skip_reading_dnode:
2146         for (i = 1; i < cc->cluster_size; i++) {
2147                 block_t blkaddr;
2148
2149                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2150                                         dn.ofs_in_node + i) :
2151                                         ei.blk + i - 1;
2152
2153                 if (!__is_valid_data_blkaddr(blkaddr))
2154                         break;
2155
2156                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2157                         ret = -EFAULT;
2158                         goto out_put_dnode;
2159                 }
2160                 cc->nr_cpages++;
2161
2162                 if (!from_dnode && i >= ei.c_len)
2163                         break;
2164         }
2165
2166         /* nothing to decompress */
2167         if (cc->nr_cpages == 0) {
2168                 ret = 0;
2169                 goto out_put_dnode;
2170         }
2171
2172         dic = f2fs_alloc_dic(cc);
2173         if (IS_ERR(dic)) {
2174                 ret = PTR_ERR(dic);
2175                 goto out_put_dnode;
2176         }
2177
2178         for (i = 0; i < cc->nr_cpages; i++) {
2179                 struct page *page = dic->cpages[i];
2180                 block_t blkaddr;
2181                 struct bio_post_read_ctx *ctx;
2182
2183                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2184                                         dn.ofs_in_node + i + 1) :
2185                                         ei.blk + i;
2186
2187                 f2fs_wait_on_block_writeback(inode, blkaddr);
2188
2189                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2190                         if (atomic_dec_and_test(&dic->remaining_pages))
2191                                 f2fs_decompress_cluster(dic);
2192                         continue;
2193                 }
2194
2195                 if (bio && (!page_is_mergeable(sbi, bio,
2196                                         *last_block_in_bio, blkaddr) ||
2197                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2198 submit_and_realloc:
2199                         __submit_bio(sbi, bio, DATA);
2200                         bio = NULL;
2201                 }
2202
2203                 if (!bio) {
2204                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2205                                         is_readahead ? REQ_RAHEAD : 0,
2206                                         page->index, for_write);
2207                         if (IS_ERR(bio)) {
2208                                 ret = PTR_ERR(bio);
2209                                 f2fs_decompress_end_io(dic, ret);
2210                                 f2fs_put_dnode(&dn);
2211                                 *bio_ret = NULL;
2212                                 return ret;
2213                         }
2214                 }
2215
2216                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2217                         goto submit_and_realloc;
2218
2219                 ctx = get_post_read_ctx(bio);
2220                 ctx->enabled_steps |= STEP_DECOMPRESS;
2221                 refcount_inc(&dic->refcnt);
2222
2223                 inc_page_count(sbi, F2FS_RD_DATA);
2224                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2225                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2226                 ClearPageError(page);
2227                 *last_block_in_bio = blkaddr;
2228         }
2229
2230         if (from_dnode)
2231                 f2fs_put_dnode(&dn);
2232
2233         *bio_ret = bio;
2234         return 0;
2235
2236 out_put_dnode:
2237         if (from_dnode)
2238                 f2fs_put_dnode(&dn);
2239 out:
2240         for (i = 0; i < cc->cluster_size; i++) {
2241                 if (cc->rpages[i]) {
2242                         ClearPageUptodate(cc->rpages[i]);
2243                         ClearPageError(cc->rpages[i]);
2244                         unlock_page(cc->rpages[i]);
2245                 }
2246         }
2247         *bio_ret = bio;
2248         return ret;
2249 }
2250 #endif
2251
2252 /*
2253  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2254  * Major change was from block_size == page_size in f2fs by default.
2255  */
2256 static int f2fs_mpage_readpages(struct inode *inode,
2257                 struct readahead_control *rac, struct page *page)
2258 {
2259         struct bio *bio = NULL;
2260         sector_t last_block_in_bio = 0;
2261         struct f2fs_map_blocks map;
2262 #ifdef CONFIG_F2FS_FS_COMPRESSION
2263         struct compress_ctx cc = {
2264                 .inode = inode,
2265                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2266                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2267                 .cluster_idx = NULL_CLUSTER,
2268                 .rpages = NULL,
2269                 .cpages = NULL,
2270                 .nr_rpages = 0,
2271                 .nr_cpages = 0,
2272         };
2273         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2274 #endif
2275         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2276         unsigned max_nr_pages = nr_pages;
2277         int ret = 0;
2278
2279         map.m_pblk = 0;
2280         map.m_lblk = 0;
2281         map.m_len = 0;
2282         map.m_flags = 0;
2283         map.m_next_pgofs = NULL;
2284         map.m_next_extent = NULL;
2285         map.m_seg_type = NO_CHECK_TYPE;
2286         map.m_may_create = false;
2287
2288         for (; nr_pages; nr_pages--) {
2289                 if (rac) {
2290                         page = readahead_page(rac);
2291                         prefetchw(&page->flags);
2292                 }
2293
2294 #ifdef CONFIG_F2FS_FS_COMPRESSION
2295                 if (f2fs_compressed_file(inode)) {
2296                         /* there are remained comressed pages, submit them */
2297                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2298                                 ret = f2fs_read_multi_pages(&cc, &bio,
2299                                                         max_nr_pages,
2300                                                         &last_block_in_bio,
2301                                                         rac != NULL, false);
2302                                 f2fs_destroy_compress_ctx(&cc, false);
2303                                 if (ret)
2304                                         goto set_error_page;
2305                         }
2306                         if (cc.cluster_idx == NULL_CLUSTER) {
2307                                 if (nc_cluster_idx ==
2308                                         page->index >> cc.log_cluster_size) {
2309                                         goto read_single_page;
2310                                 }
2311
2312                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2313                                 if (ret < 0)
2314                                         goto set_error_page;
2315                                 else if (!ret) {
2316                                         nc_cluster_idx =
2317                                                 page->index >> cc.log_cluster_size;
2318                                         goto read_single_page;
2319                                 }
2320
2321                                 nc_cluster_idx = NULL_CLUSTER;
2322                         }
2323                         ret = f2fs_init_compress_ctx(&cc);
2324                         if (ret)
2325                                 goto set_error_page;
2326
2327                         f2fs_compress_ctx_add_page(&cc, page);
2328
2329                         goto next_page;
2330                 }
2331 read_single_page:
2332 #endif
2333
2334                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2335                                         &bio, &last_block_in_bio, rac);
2336                 if (ret) {
2337 #ifdef CONFIG_F2FS_FS_COMPRESSION
2338 set_error_page:
2339 #endif
2340                         SetPageError(page);
2341                         zero_user_segment(page, 0, PAGE_SIZE);
2342                         unlock_page(page);
2343                 }
2344 #ifdef CONFIG_F2FS_FS_COMPRESSION
2345 next_page:
2346 #endif
2347                 if (rac)
2348                         put_page(page);
2349
2350 #ifdef CONFIG_F2FS_FS_COMPRESSION
2351                 if (f2fs_compressed_file(inode)) {
2352                         /* last page */
2353                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2354                                 ret = f2fs_read_multi_pages(&cc, &bio,
2355                                                         max_nr_pages,
2356                                                         &last_block_in_bio,
2357                                                         rac != NULL, false);
2358                                 f2fs_destroy_compress_ctx(&cc, false);
2359                         }
2360                 }
2361 #endif
2362         }
2363         if (bio)
2364                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2365         return ret;
2366 }
2367
2368 static int f2fs_read_data_page(struct file *file, struct page *page)
2369 {
2370         struct inode *inode = page_file_mapping(page)->host;
2371         int ret = -EAGAIN;
2372
2373         trace_f2fs_readpage(page, DATA);
2374
2375         if (!f2fs_is_compress_backend_ready(inode)) {
2376                 unlock_page(page);
2377                 return -EOPNOTSUPP;
2378         }
2379
2380         /* If the file has inline data, try to read it directly */
2381         if (f2fs_has_inline_data(inode))
2382                 ret = f2fs_read_inline_data(inode, page);
2383         if (ret == -EAGAIN)
2384                 ret = f2fs_mpage_readpages(inode, NULL, page);
2385         return ret;
2386 }
2387
2388 static void f2fs_readahead(struct readahead_control *rac)
2389 {
2390         struct inode *inode = rac->mapping->host;
2391
2392         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2393
2394         if (!f2fs_is_compress_backend_ready(inode))
2395                 return;
2396
2397         /* If the file has inline data, skip readpages */
2398         if (f2fs_has_inline_data(inode))
2399                 return;
2400
2401         f2fs_mpage_readpages(inode, rac, NULL);
2402 }
2403
2404 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2405 {
2406         struct inode *inode = fio->page->mapping->host;
2407         struct page *mpage, *page;
2408         gfp_t gfp_flags = GFP_NOFS;
2409
2410         if (!f2fs_encrypted_file(inode))
2411                 return 0;
2412
2413         page = fio->compressed_page ? fio->compressed_page : fio->page;
2414
2415         /* wait for GCed page writeback via META_MAPPING */
2416         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2417
2418         if (fscrypt_inode_uses_inline_crypto(inode))
2419                 return 0;
2420
2421 retry_encrypt:
2422         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2423                                         PAGE_SIZE, 0, gfp_flags);
2424         if (IS_ERR(fio->encrypted_page)) {
2425                 /* flush pending IOs and wait for a while in the ENOMEM case */
2426                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2427                         f2fs_flush_merged_writes(fio->sbi);
2428                         memalloc_retry_wait(GFP_NOFS);
2429                         gfp_flags |= __GFP_NOFAIL;
2430                         goto retry_encrypt;
2431                 }
2432                 return PTR_ERR(fio->encrypted_page);
2433         }
2434
2435         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2436         if (mpage) {
2437                 if (PageUptodate(mpage))
2438                         memcpy(page_address(mpage),
2439                                 page_address(fio->encrypted_page), PAGE_SIZE);
2440                 f2fs_put_page(mpage, 1);
2441         }
2442         return 0;
2443 }
2444
2445 static inline bool check_inplace_update_policy(struct inode *inode,
2446                                 struct f2fs_io_info *fio)
2447 {
2448         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2449         unsigned int policy = SM_I(sbi)->ipu_policy;
2450
2451         if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
2452                         is_inode_flag_set(inode, FI_OPU_WRITE))
2453                 return false;
2454         if (policy & (0x1 << F2FS_IPU_FORCE))
2455                 return true;
2456         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2457                 return true;
2458         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2459                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2460                 return true;
2461         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2462                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2463                 return true;
2464
2465         /*
2466          * IPU for rewrite async pages
2467          */
2468         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2469                         fio && fio->op == REQ_OP_WRITE &&
2470                         !(fio->op_flags & REQ_SYNC) &&
2471                         !IS_ENCRYPTED(inode))
2472                 return true;
2473
2474         /* this is only set during fdatasync */
2475         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2476                         is_inode_flag_set(inode, FI_NEED_IPU))
2477                 return true;
2478
2479         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2480                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2481                 return true;
2482
2483         return false;
2484 }
2485
2486 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2487 {
2488         /* swap file is migrating in aligned write mode */
2489         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2490                 return false;
2491
2492         if (f2fs_is_pinned_file(inode))
2493                 return true;
2494
2495         /* if this is cold file, we should overwrite to avoid fragmentation */
2496         if (file_is_cold(inode))
2497                 return true;
2498
2499         return check_inplace_update_policy(inode, fio);
2500 }
2501
2502 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2503 {
2504         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2505
2506         /* The below cases were checked when setting it. */
2507         if (f2fs_is_pinned_file(inode))
2508                 return false;
2509         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2510                 return true;
2511         if (f2fs_lfs_mode(sbi))
2512                 return true;
2513         if (S_ISDIR(inode->i_mode))
2514                 return true;
2515         if (IS_NOQUOTA(inode))
2516                 return true;
2517         if (f2fs_is_atomic_file(inode))
2518                 return true;
2519
2520         /* swap file is migrating in aligned write mode */
2521         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2522                 return true;
2523
2524         if (is_inode_flag_set(inode, FI_OPU_WRITE))
2525                 return true;
2526
2527         if (fio) {
2528                 if (page_private_gcing(fio->page))
2529                         return true;
2530                 if (page_private_dummy(fio->page))
2531                         return true;
2532                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2533                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2534                         return true;
2535         }
2536         return false;
2537 }
2538
2539 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2540 {
2541         struct inode *inode = fio->page->mapping->host;
2542
2543         if (f2fs_should_update_outplace(inode, fio))
2544                 return false;
2545
2546         return f2fs_should_update_inplace(inode, fio);
2547 }
2548
2549 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2550 {
2551         struct page *page = fio->page;
2552         struct inode *inode = page->mapping->host;
2553         struct dnode_of_data dn;
2554         struct extent_info ei = {0, };
2555         struct node_info ni;
2556         bool ipu_force = false;
2557         int err = 0;
2558
2559         set_new_dnode(&dn, inode, NULL, NULL, 0);
2560         if (need_inplace_update(fio) &&
2561                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2562                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2563
2564                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2565                                                 DATA_GENERIC_ENHANCE))
2566                         return -EFSCORRUPTED;
2567
2568                 ipu_force = true;
2569                 fio->need_lock = LOCK_DONE;
2570                 goto got_it;
2571         }
2572
2573         /* Deadlock due to between page->lock and f2fs_lock_op */
2574         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2575                 return -EAGAIN;
2576
2577         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2578         if (err)
2579                 goto out;
2580
2581         fio->old_blkaddr = dn.data_blkaddr;
2582
2583         /* This page is already truncated */
2584         if (fio->old_blkaddr == NULL_ADDR) {
2585                 ClearPageUptodate(page);
2586                 clear_page_private_gcing(page);
2587                 goto out_writepage;
2588         }
2589 got_it:
2590         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2591                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2592                                                 DATA_GENERIC_ENHANCE)) {
2593                 err = -EFSCORRUPTED;
2594                 goto out_writepage;
2595         }
2596         /*
2597          * If current allocation needs SSR,
2598          * it had better in-place writes for updated data.
2599          */
2600         if (ipu_force ||
2601                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2602                                         need_inplace_update(fio))) {
2603                 err = f2fs_encrypt_one_page(fio);
2604                 if (err)
2605                         goto out_writepage;
2606
2607                 set_page_writeback(page);
2608                 ClearPageError(page);
2609                 f2fs_put_dnode(&dn);
2610                 if (fio->need_lock == LOCK_REQ)
2611                         f2fs_unlock_op(fio->sbi);
2612                 err = f2fs_inplace_write_data(fio);
2613                 if (err) {
2614                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2615                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2616                         if (PageWriteback(page))
2617                                 end_page_writeback(page);
2618                 } else {
2619                         set_inode_flag(inode, FI_UPDATE_WRITE);
2620                 }
2621                 trace_f2fs_do_write_data_page(fio->page, IPU);
2622                 return err;
2623         }
2624
2625         if (fio->need_lock == LOCK_RETRY) {
2626                 if (!f2fs_trylock_op(fio->sbi)) {
2627                         err = -EAGAIN;
2628                         goto out_writepage;
2629                 }
2630                 fio->need_lock = LOCK_REQ;
2631         }
2632
2633         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2634         if (err)
2635                 goto out_writepage;
2636
2637         fio->version = ni.version;
2638
2639         err = f2fs_encrypt_one_page(fio);
2640         if (err)
2641                 goto out_writepage;
2642
2643         set_page_writeback(page);
2644         ClearPageError(page);
2645
2646         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2647                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2648
2649         /* LFS mode write path */
2650         f2fs_outplace_write_data(&dn, fio);
2651         trace_f2fs_do_write_data_page(page, OPU);
2652         set_inode_flag(inode, FI_APPEND_WRITE);
2653         if (page->index == 0)
2654                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2655 out_writepage:
2656         f2fs_put_dnode(&dn);
2657 out:
2658         if (fio->need_lock == LOCK_REQ)
2659                 f2fs_unlock_op(fio->sbi);
2660         return err;
2661 }
2662
2663 int f2fs_write_single_data_page(struct page *page, int *submitted,
2664                                 struct bio **bio,
2665                                 sector_t *last_block,
2666                                 struct writeback_control *wbc,
2667                                 enum iostat_type io_type,
2668                                 int compr_blocks,
2669                                 bool allow_balance)
2670 {
2671         struct inode *inode = page->mapping->host;
2672         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2673         loff_t i_size = i_size_read(inode);
2674         const pgoff_t end_index = ((unsigned long long)i_size)
2675                                                         >> PAGE_SHIFT;
2676         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2677         unsigned offset = 0;
2678         bool need_balance_fs = false;
2679         int err = 0;
2680         struct f2fs_io_info fio = {
2681                 .sbi = sbi,
2682                 .ino = inode->i_ino,
2683                 .type = DATA,
2684                 .op = REQ_OP_WRITE,
2685                 .op_flags = wbc_to_write_flags(wbc),
2686                 .old_blkaddr = NULL_ADDR,
2687                 .page = page,
2688                 .encrypted_page = NULL,
2689                 .submitted = false,
2690                 .compr_blocks = compr_blocks,
2691                 .need_lock = LOCK_RETRY,
2692                 .io_type = io_type,
2693                 .io_wbc = wbc,
2694                 .bio = bio,
2695                 .last_block = last_block,
2696         };
2697
2698         trace_f2fs_writepage(page, DATA);
2699
2700         /* we should bypass data pages to proceed the kworkder jobs */
2701         if (unlikely(f2fs_cp_error(sbi))) {
2702                 mapping_set_error(page->mapping, -EIO);
2703                 /*
2704                  * don't drop any dirty dentry pages for keeping lastest
2705                  * directory structure.
2706                  */
2707                 if (S_ISDIR(inode->i_mode))
2708                         goto redirty_out;
2709                 goto out;
2710         }
2711
2712         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2713                 goto redirty_out;
2714
2715         if (page->index < end_index ||
2716                         f2fs_verity_in_progress(inode) ||
2717                         compr_blocks)
2718                 goto write;
2719
2720         /*
2721          * If the offset is out-of-range of file size,
2722          * this page does not have to be written to disk.
2723          */
2724         offset = i_size & (PAGE_SIZE - 1);
2725         if ((page->index >= end_index + 1) || !offset)
2726                 goto out;
2727
2728         zero_user_segment(page, offset, PAGE_SIZE);
2729 write:
2730         if (f2fs_is_drop_cache(inode))
2731                 goto out;
2732         /* we should not write 0'th page having journal header */
2733         if (f2fs_is_volatile_file(inode) && (!page->index ||
2734                         (!wbc->for_reclaim &&
2735                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2736                 goto redirty_out;
2737
2738         /* Dentry/quota blocks are controlled by checkpoint */
2739         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2740                 /*
2741                  * We need to wait for node_write to avoid block allocation during
2742                  * checkpoint. This can only happen to quota writes which can cause
2743                  * the below discard race condition.
2744                  */
2745                 if (IS_NOQUOTA(inode))
2746                         f2fs_down_read(&sbi->node_write);
2747
2748                 fio.need_lock = LOCK_DONE;
2749                 err = f2fs_do_write_data_page(&fio);
2750
2751                 if (IS_NOQUOTA(inode))
2752                         f2fs_up_read(&sbi->node_write);
2753
2754                 goto done;
2755         }
2756
2757         if (!wbc->for_reclaim)
2758                 need_balance_fs = true;
2759         else if (has_not_enough_free_secs(sbi, 0, 0))
2760                 goto redirty_out;
2761         else
2762                 set_inode_flag(inode, FI_HOT_DATA);
2763
2764         err = -EAGAIN;
2765         if (f2fs_has_inline_data(inode)) {
2766                 err = f2fs_write_inline_data(inode, page);
2767                 if (!err)
2768                         goto out;
2769         }
2770
2771         if (err == -EAGAIN) {
2772                 err = f2fs_do_write_data_page(&fio);
2773                 if (err == -EAGAIN) {
2774                         fio.need_lock = LOCK_REQ;
2775                         err = f2fs_do_write_data_page(&fio);
2776                 }
2777         }
2778
2779         if (err) {
2780                 file_set_keep_isize(inode);
2781         } else {
2782                 spin_lock(&F2FS_I(inode)->i_size_lock);
2783                 if (F2FS_I(inode)->last_disk_size < psize)
2784                         F2FS_I(inode)->last_disk_size = psize;
2785                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2786         }
2787
2788 done:
2789         if (err && err != -ENOENT)
2790                 goto redirty_out;
2791
2792 out:
2793         inode_dec_dirty_pages(inode);
2794         if (err) {
2795                 ClearPageUptodate(page);
2796                 clear_page_private_gcing(page);
2797         }
2798
2799         if (wbc->for_reclaim) {
2800                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2801                 clear_inode_flag(inode, FI_HOT_DATA);
2802                 f2fs_remove_dirty_inode(inode);
2803                 submitted = NULL;
2804         }
2805         unlock_page(page);
2806         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2807                         !F2FS_I(inode)->cp_task && allow_balance)
2808                 f2fs_balance_fs(sbi, need_balance_fs);
2809
2810         if (unlikely(f2fs_cp_error(sbi))) {
2811                 f2fs_submit_merged_write(sbi, DATA);
2812                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2813                 submitted = NULL;
2814         }
2815
2816         if (submitted)
2817                 *submitted = fio.submitted ? 1 : 0;
2818
2819         return 0;
2820
2821 redirty_out:
2822         redirty_page_for_writepage(wbc, page);
2823         /*
2824          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2825          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2826          * file_write_and_wait_range() will see EIO error, which is critical
2827          * to return value of fsync() followed by atomic_write failure to user.
2828          */
2829         if (!err || wbc->for_reclaim)
2830                 return AOP_WRITEPAGE_ACTIVATE;
2831         unlock_page(page);
2832         return err;
2833 }
2834
2835 static int f2fs_write_data_page(struct page *page,
2836                                         struct writeback_control *wbc)
2837 {
2838 #ifdef CONFIG_F2FS_FS_COMPRESSION
2839         struct inode *inode = page->mapping->host;
2840
2841         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2842                 goto out;
2843
2844         if (f2fs_compressed_file(inode)) {
2845                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2846                         redirty_page_for_writepage(wbc, page);
2847                         return AOP_WRITEPAGE_ACTIVATE;
2848                 }
2849         }
2850 out:
2851 #endif
2852
2853         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2854                                                 wbc, FS_DATA_IO, 0, true);
2855 }
2856
2857 /*
2858  * This function was copied from write_cche_pages from mm/page-writeback.c.
2859  * The major change is making write step of cold data page separately from
2860  * warm/hot data page.
2861  */
2862 static int f2fs_write_cache_pages(struct address_space *mapping,
2863                                         struct writeback_control *wbc,
2864                                         enum iostat_type io_type)
2865 {
2866         int ret = 0;
2867         int done = 0, retry = 0;
2868         struct pagevec pvec;
2869         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2870         struct bio *bio = NULL;
2871         sector_t last_block;
2872 #ifdef CONFIG_F2FS_FS_COMPRESSION
2873         struct inode *inode = mapping->host;
2874         struct compress_ctx cc = {
2875                 .inode = inode,
2876                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2877                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2878                 .cluster_idx = NULL_CLUSTER,
2879                 .rpages = NULL,
2880                 .nr_rpages = 0,
2881                 .cpages = NULL,
2882                 .valid_nr_cpages = 0,
2883                 .rbuf = NULL,
2884                 .cbuf = NULL,
2885                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2886                 .private = NULL,
2887         };
2888 #endif
2889         int nr_pages;
2890         pgoff_t index;
2891         pgoff_t end;            /* Inclusive */
2892         pgoff_t done_index;
2893         int range_whole = 0;
2894         xa_mark_t tag;
2895         int nwritten = 0;
2896         int submitted = 0;
2897         int i;
2898
2899         pagevec_init(&pvec);
2900
2901         if (get_dirty_pages(mapping->host) <=
2902                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2903                 set_inode_flag(mapping->host, FI_HOT_DATA);
2904         else
2905                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2906
2907         if (wbc->range_cyclic) {
2908                 index = mapping->writeback_index; /* prev offset */
2909                 end = -1;
2910         } else {
2911                 index = wbc->range_start >> PAGE_SHIFT;
2912                 end = wbc->range_end >> PAGE_SHIFT;
2913                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2914                         range_whole = 1;
2915         }
2916         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2917                 tag = PAGECACHE_TAG_TOWRITE;
2918         else
2919                 tag = PAGECACHE_TAG_DIRTY;
2920 retry:
2921         retry = 0;
2922         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2923                 tag_pages_for_writeback(mapping, index, end);
2924         done_index = index;
2925         while (!done && !retry && (index <= end)) {
2926                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2927                                 tag);
2928                 if (nr_pages == 0)
2929                         break;
2930
2931                 for (i = 0; i < nr_pages; i++) {
2932                         struct page *page = pvec.pages[i];
2933                         bool need_readd;
2934 readd:
2935                         need_readd = false;
2936 #ifdef CONFIG_F2FS_FS_COMPRESSION
2937                         if (f2fs_compressed_file(inode)) {
2938                                 void *fsdata = NULL;
2939                                 struct page *pagep;
2940                                 int ret2;
2941
2942                                 ret = f2fs_init_compress_ctx(&cc);
2943                                 if (ret) {
2944                                         done = 1;
2945                                         break;
2946                                 }
2947
2948                                 if (!f2fs_cluster_can_merge_page(&cc,
2949                                                                 page->index)) {
2950                                         ret = f2fs_write_multi_pages(&cc,
2951                                                 &submitted, wbc, io_type);
2952                                         if (!ret)
2953                                                 need_readd = true;
2954                                         goto result;
2955                                 }
2956
2957                                 if (unlikely(f2fs_cp_error(sbi)))
2958                                         goto lock_page;
2959
2960                                 if (!f2fs_cluster_is_empty(&cc))
2961                                         goto lock_page;
2962
2963                                 ret2 = f2fs_prepare_compress_overwrite(
2964                                                         inode, &pagep,
2965                                                         page->index, &fsdata);
2966                                 if (ret2 < 0) {
2967                                         ret = ret2;
2968                                         done = 1;
2969                                         break;
2970                                 } else if (ret2 &&
2971                                         (!f2fs_compress_write_end(inode,
2972                                                 fsdata, page->index, 1) ||
2973                                          !f2fs_all_cluster_page_loaded(&cc,
2974                                                 &pvec, i, nr_pages))) {
2975                                         retry = 1;
2976                                         break;
2977                                 }
2978                         }
2979 #endif
2980                         /* give a priority to WB_SYNC threads */
2981                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2982                                         wbc->sync_mode == WB_SYNC_NONE) {
2983                                 done = 1;
2984                                 break;
2985                         }
2986 #ifdef CONFIG_F2FS_FS_COMPRESSION
2987 lock_page:
2988 #endif
2989                         done_index = page->index;
2990 retry_write:
2991                         lock_page(page);
2992
2993                         if (unlikely(page->mapping != mapping)) {
2994 continue_unlock:
2995                                 unlock_page(page);
2996                                 continue;
2997                         }
2998
2999                         if (!PageDirty(page)) {
3000                                 /* someone wrote it for us */
3001                                 goto continue_unlock;
3002                         }
3003
3004                         if (PageWriteback(page)) {
3005                                 if (wbc->sync_mode != WB_SYNC_NONE)
3006                                         f2fs_wait_on_page_writeback(page,
3007                                                         DATA, true, true);
3008                                 else
3009                                         goto continue_unlock;
3010                         }
3011
3012                         if (!clear_page_dirty_for_io(page))
3013                                 goto continue_unlock;
3014
3015 #ifdef CONFIG_F2FS_FS_COMPRESSION
3016                         if (f2fs_compressed_file(inode)) {
3017                                 get_page(page);
3018                                 f2fs_compress_ctx_add_page(&cc, page);
3019                                 continue;
3020                         }
3021 #endif
3022                         ret = f2fs_write_single_data_page(page, &submitted,
3023                                         &bio, &last_block, wbc, io_type,
3024                                         0, true);
3025                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3026                                 unlock_page(page);
3027 #ifdef CONFIG_F2FS_FS_COMPRESSION
3028 result:
3029 #endif
3030                         nwritten += submitted;
3031                         wbc->nr_to_write -= submitted;
3032
3033                         if (unlikely(ret)) {
3034                                 /*
3035                                  * keep nr_to_write, since vfs uses this to
3036                                  * get # of written pages.
3037                                  */
3038                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3039                                         ret = 0;
3040                                         goto next;
3041                                 } else if (ret == -EAGAIN) {
3042                                         ret = 0;
3043                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3044                                                 f2fs_io_schedule_timeout(
3045                                                         DEFAULT_IO_TIMEOUT);
3046                                                 goto retry_write;
3047                                         }
3048                                         goto next;
3049                                 }
3050                                 done_index = page->index + 1;
3051                                 done = 1;
3052                                 break;
3053                         }
3054
3055                         if (wbc->nr_to_write <= 0 &&
3056                                         wbc->sync_mode == WB_SYNC_NONE) {
3057                                 done = 1;
3058                                 break;
3059                         }
3060 next:
3061                         if (need_readd)
3062                                 goto readd;
3063                 }
3064                 pagevec_release(&pvec);
3065                 cond_resched();
3066         }
3067 #ifdef CONFIG_F2FS_FS_COMPRESSION
3068         /* flush remained pages in compress cluster */
3069         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3070                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3071                 nwritten += submitted;
3072                 wbc->nr_to_write -= submitted;
3073                 if (ret) {
3074                         done = 1;
3075                         retry = 0;
3076                 }
3077         }
3078         if (f2fs_compressed_file(inode))
3079                 f2fs_destroy_compress_ctx(&cc, false);
3080 #endif
3081         if (retry) {
3082                 index = 0;
3083                 end = -1;
3084                 goto retry;
3085         }
3086         if (wbc->range_cyclic && !done)
3087                 done_index = 0;
3088         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3089                 mapping->writeback_index = done_index;
3090
3091         if (nwritten)
3092                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3093                                                                 NULL, 0, DATA);
3094         /* submit cached bio of IPU write */
3095         if (bio)
3096                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3097
3098         return ret;
3099 }
3100
3101 static inline bool __should_serialize_io(struct inode *inode,
3102                                         struct writeback_control *wbc)
3103 {
3104         /* to avoid deadlock in path of data flush */
3105         if (F2FS_I(inode)->cp_task)
3106                 return false;
3107
3108         if (!S_ISREG(inode->i_mode))
3109                 return false;
3110         if (IS_NOQUOTA(inode))
3111                 return false;
3112
3113         if (f2fs_need_compress_data(inode))
3114                 return true;
3115         if (wbc->sync_mode != WB_SYNC_ALL)
3116                 return true;
3117         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3118                 return true;
3119         return false;
3120 }
3121
3122 static int __f2fs_write_data_pages(struct address_space *mapping,
3123                                                 struct writeback_control *wbc,
3124                                                 enum iostat_type io_type)
3125 {
3126         struct inode *inode = mapping->host;
3127         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3128         struct blk_plug plug;
3129         int ret;
3130         bool locked = false;
3131
3132         /* deal with chardevs and other special file */
3133         if (!mapping->a_ops->writepage)
3134                 return 0;
3135
3136         /* skip writing if there is no dirty page in this inode */
3137         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3138                 return 0;
3139
3140         /* during POR, we don't need to trigger writepage at all. */
3141         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3142                 goto skip_write;
3143
3144         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3145                         wbc->sync_mode == WB_SYNC_NONE &&
3146                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3147                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3148                 goto skip_write;
3149
3150         /* skip writing in file defragment preparing stage */
3151         if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3152                 goto skip_write;
3153
3154         trace_f2fs_writepages(mapping->host, wbc, DATA);
3155
3156         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3157         if (wbc->sync_mode == WB_SYNC_ALL)
3158                 atomic_inc(&sbi->wb_sync_req[DATA]);
3159         else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3160                 /* to avoid potential deadlock */
3161                 if (current->plug)
3162                         blk_finish_plug(current->plug);
3163                 goto skip_write;
3164         }
3165
3166         if (__should_serialize_io(inode, wbc)) {
3167                 mutex_lock(&sbi->writepages);
3168                 locked = true;
3169         }
3170
3171         blk_start_plug(&plug);
3172         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3173         blk_finish_plug(&plug);
3174
3175         if (locked)
3176                 mutex_unlock(&sbi->writepages);
3177
3178         if (wbc->sync_mode == WB_SYNC_ALL)
3179                 atomic_dec(&sbi->wb_sync_req[DATA]);
3180         /*
3181          * if some pages were truncated, we cannot guarantee its mapping->host
3182          * to detect pending bios.
3183          */
3184
3185         f2fs_remove_dirty_inode(inode);
3186         return ret;
3187
3188 skip_write:
3189         wbc->pages_skipped += get_dirty_pages(inode);
3190         trace_f2fs_writepages(mapping->host, wbc, DATA);
3191         return 0;
3192 }
3193
3194 static int f2fs_write_data_pages(struct address_space *mapping,
3195                             struct writeback_control *wbc)
3196 {
3197         struct inode *inode = mapping->host;
3198
3199         return __f2fs_write_data_pages(mapping, wbc,
3200                         F2FS_I(inode)->cp_task == current ?
3201                         FS_CP_DATA_IO : FS_DATA_IO);
3202 }
3203
3204 void f2fs_write_failed(struct inode *inode, loff_t to)
3205 {
3206         loff_t i_size = i_size_read(inode);
3207
3208         if (IS_NOQUOTA(inode))
3209                 return;
3210
3211         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3212         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3213                 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3214                 filemap_invalidate_lock(inode->i_mapping);
3215
3216                 truncate_pagecache(inode, i_size);
3217                 f2fs_truncate_blocks(inode, i_size, true);
3218
3219                 filemap_invalidate_unlock(inode->i_mapping);
3220                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3221         }
3222 }
3223
3224 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3225                         struct page *page, loff_t pos, unsigned len,
3226                         block_t *blk_addr, bool *node_changed)
3227 {
3228         struct inode *inode = page->mapping->host;
3229         pgoff_t index = page->index;
3230         struct dnode_of_data dn;
3231         struct page *ipage;
3232         bool locked = false;
3233         struct extent_info ei = {0, };
3234         int err = 0;
3235         int flag;
3236
3237         /*
3238          * If a whole page is being written and we already preallocated all the
3239          * blocks, then there is no need to get a block address now.
3240          */
3241         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3242                 return 0;
3243
3244         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3245         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3246                 flag = F2FS_GET_BLOCK_DEFAULT;
3247         else
3248                 flag = F2FS_GET_BLOCK_PRE_AIO;
3249
3250         if (f2fs_has_inline_data(inode) ||
3251                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3252                 f2fs_do_map_lock(sbi, flag, true);
3253                 locked = true;
3254         }
3255
3256 restart:
3257         /* check inline_data */
3258         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3259         if (IS_ERR(ipage)) {
3260                 err = PTR_ERR(ipage);
3261                 goto unlock_out;
3262         }
3263
3264         set_new_dnode(&dn, inode, ipage, ipage, 0);
3265
3266         if (f2fs_has_inline_data(inode)) {
3267                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3268                         f2fs_do_read_inline_data(page, ipage);
3269                         set_inode_flag(inode, FI_DATA_EXIST);
3270                         if (inode->i_nlink)
3271                                 set_page_private_inline(ipage);
3272                 } else {
3273                         err = f2fs_convert_inline_page(&dn, page);
3274                         if (err)
3275                                 goto out;
3276                         if (dn.data_blkaddr == NULL_ADDR)
3277                                 err = f2fs_get_block(&dn, index);
3278                 }
3279         } else if (locked) {
3280                 err = f2fs_get_block(&dn, index);
3281         } else {
3282                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3283                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3284                 } else {
3285                         /* hole case */
3286                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3287                         if (err || dn.data_blkaddr == NULL_ADDR) {
3288                                 f2fs_put_dnode(&dn);
3289                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3290                                                                 true);
3291                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3292                                 locked = true;
3293                                 goto restart;
3294                         }
3295                 }
3296         }
3297
3298         /* convert_inline_page can make node_changed */
3299         *blk_addr = dn.data_blkaddr;
3300         *node_changed = dn.node_changed;
3301 out:
3302         f2fs_put_dnode(&dn);
3303 unlock_out:
3304         if (locked)
3305                 f2fs_do_map_lock(sbi, flag, false);
3306         return err;
3307 }
3308
3309 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3310                 loff_t pos, unsigned len, unsigned flags,
3311                 struct page **pagep, void **fsdata)
3312 {
3313         struct inode *inode = mapping->host;
3314         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3315         struct page *page = NULL;
3316         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3317         bool need_balance = false, drop_atomic = false;
3318         block_t blkaddr = NULL_ADDR;
3319         int err = 0;
3320
3321         trace_f2fs_write_begin(inode, pos, len, flags);
3322
3323         if (!f2fs_is_checkpoint_ready(sbi)) {
3324                 err = -ENOSPC;
3325                 goto fail;
3326         }
3327
3328         if ((f2fs_is_atomic_file(inode) &&
3329                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3330                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3331                 err = -ENOMEM;
3332                 drop_atomic = true;
3333                 goto fail;
3334         }
3335
3336         /*
3337          * We should check this at this moment to avoid deadlock on inode page
3338          * and #0 page. The locking rule for inline_data conversion should be:
3339          * lock_page(page #0) -> lock_page(inode_page)
3340          */
3341         if (index != 0) {
3342                 err = f2fs_convert_inline_inode(inode);
3343                 if (err)
3344                         goto fail;
3345         }
3346
3347 #ifdef CONFIG_F2FS_FS_COMPRESSION
3348         if (f2fs_compressed_file(inode)) {
3349                 int ret;
3350
3351                 *fsdata = NULL;
3352
3353                 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3354                         goto repeat;
3355
3356                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3357                                                         index, fsdata);
3358                 if (ret < 0) {
3359                         err = ret;
3360                         goto fail;
3361                 } else if (ret) {
3362                         return 0;
3363                 }
3364         }
3365 #endif
3366
3367 repeat:
3368         /*
3369          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3370          * wait_for_stable_page. Will wait that below with our IO control.
3371          */
3372         page = f2fs_pagecache_get_page(mapping, index,
3373                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3374         if (!page) {
3375                 err = -ENOMEM;
3376                 goto fail;
3377         }
3378
3379         /* TODO: cluster can be compressed due to race with .writepage */
3380
3381         *pagep = page;
3382
3383         err = prepare_write_begin(sbi, page, pos, len,
3384                                         &blkaddr, &need_balance);
3385         if (err)
3386                 goto fail;
3387
3388         if (need_balance && !IS_NOQUOTA(inode) &&
3389                         has_not_enough_free_secs(sbi, 0, 0)) {
3390                 unlock_page(page);
3391                 f2fs_balance_fs(sbi, true);
3392                 lock_page(page);
3393                 if (page->mapping != mapping) {
3394                         /* The page got truncated from under us */
3395                         f2fs_put_page(page, 1);
3396                         goto repeat;
3397                 }
3398         }
3399
3400         f2fs_wait_on_page_writeback(page, DATA, false, true);
3401
3402         if (len == PAGE_SIZE || PageUptodate(page))
3403                 return 0;
3404
3405         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3406             !f2fs_verity_in_progress(inode)) {
3407                 zero_user_segment(page, len, PAGE_SIZE);
3408                 return 0;
3409         }
3410
3411         if (blkaddr == NEW_ADDR) {
3412                 zero_user_segment(page, 0, PAGE_SIZE);
3413                 SetPageUptodate(page);
3414         } else {
3415                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3416                                 DATA_GENERIC_ENHANCE_READ)) {
3417                         err = -EFSCORRUPTED;
3418                         goto fail;
3419                 }
3420                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3421                 if (err)
3422                         goto fail;
3423
3424                 lock_page(page);
3425                 if (unlikely(page->mapping != mapping)) {
3426                         f2fs_put_page(page, 1);
3427                         goto repeat;
3428                 }
3429                 if (unlikely(!PageUptodate(page))) {
3430                         err = -EIO;
3431                         goto fail;
3432                 }
3433         }
3434         return 0;
3435
3436 fail:
3437         f2fs_put_page(page, 1);
3438         f2fs_write_failed(inode, pos + len);
3439         if (drop_atomic)
3440                 f2fs_drop_inmem_pages_all(sbi, false);
3441         return err;
3442 }
3443
3444 static int f2fs_write_end(struct file *file,
3445                         struct address_space *mapping,
3446                         loff_t pos, unsigned len, unsigned copied,
3447                         struct page *page, void *fsdata)
3448 {
3449         struct inode *inode = page->mapping->host;
3450
3451         trace_f2fs_write_end(inode, pos, len, copied);
3452
3453         /*
3454          * This should be come from len == PAGE_SIZE, and we expect copied
3455          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3456          * let generic_perform_write() try to copy data again through copied=0.
3457          */
3458         if (!PageUptodate(page)) {
3459                 if (unlikely(copied != len))
3460                         copied = 0;
3461                 else
3462                         SetPageUptodate(page);
3463         }
3464
3465 #ifdef CONFIG_F2FS_FS_COMPRESSION
3466         /* overwrite compressed file */
3467         if (f2fs_compressed_file(inode) && fsdata) {
3468                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3469                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3470
3471                 if (pos + copied > i_size_read(inode) &&
3472                                 !f2fs_verity_in_progress(inode))
3473                         f2fs_i_size_write(inode, pos + copied);
3474                 return copied;
3475         }
3476 #endif
3477
3478         if (!copied)
3479                 goto unlock_out;
3480
3481         set_page_dirty(page);
3482
3483         if (pos + copied > i_size_read(inode) &&
3484             !f2fs_verity_in_progress(inode))
3485                 f2fs_i_size_write(inode, pos + copied);
3486 unlock_out:
3487         f2fs_put_page(page, 1);
3488         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3489         return copied;
3490 }
3491
3492 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3493 {
3494         struct inode *inode = folio->mapping->host;
3495         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3496
3497         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3498                                 (offset || length != folio_size(folio)))
3499                 return;
3500
3501         if (folio_test_dirty(folio)) {
3502                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3503                         dec_page_count(sbi, F2FS_DIRTY_META);
3504                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3505                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3506                 } else {
3507                         inode_dec_dirty_pages(inode);
3508                         f2fs_remove_dirty_inode(inode);
3509                 }
3510         }
3511
3512         clear_page_private_gcing(&folio->page);
3513
3514         if (test_opt(sbi, COMPRESS_CACHE) &&
3515                         inode->i_ino == F2FS_COMPRESS_INO(sbi))
3516                 clear_page_private_data(&folio->page);
3517
3518         if (page_private_atomic(&folio->page))
3519                 return f2fs_drop_inmem_page(inode, &folio->page);
3520
3521         folio_detach_private(folio);
3522 }
3523
3524 int f2fs_release_page(struct page *page, gfp_t wait)
3525 {
3526         /* If this is dirty page, keep PagePrivate */
3527         if (PageDirty(page))
3528                 return 0;
3529
3530         /* This is atomic written page, keep Private */
3531         if (page_private_atomic(page))
3532                 return 0;
3533
3534         if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3535                 struct inode *inode = page->mapping->host;
3536
3537                 if (inode->i_ino == F2FS_COMPRESS_INO(F2FS_I_SB(inode)))
3538                         clear_page_private_data(page);
3539         }
3540
3541         clear_page_private_gcing(page);
3542
3543         detach_page_private(page);
3544         set_page_private(page, 0);
3545         return 1;
3546 }
3547
3548 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3549                 struct folio *folio)
3550 {
3551         struct inode *inode = mapping->host;
3552
3553         trace_f2fs_set_page_dirty(&folio->page, DATA);
3554
3555         if (!folio_test_uptodate(folio))
3556                 folio_mark_uptodate(folio);
3557         BUG_ON(folio_test_swapcache(folio));
3558
3559         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3560                 if (!page_private_atomic(&folio->page)) {
3561                         f2fs_register_inmem_page(inode, &folio->page);
3562                         return true;
3563                 }
3564                 /*
3565                  * Previously, this page has been registered, we just
3566                  * return here.
3567                  */
3568                 return false;
3569         }
3570
3571         if (!folio_test_dirty(folio)) {
3572                 filemap_dirty_folio(mapping, folio);
3573                 f2fs_update_dirty_folio(inode, folio);
3574                 return true;
3575         }
3576         return true;
3577 }
3578
3579
3580 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3581 {
3582 #ifdef CONFIG_F2FS_FS_COMPRESSION
3583         struct dnode_of_data dn;
3584         sector_t start_idx, blknr = 0;
3585         int ret;
3586
3587         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3588
3589         set_new_dnode(&dn, inode, NULL, NULL, 0);
3590         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3591         if (ret)
3592                 return 0;
3593
3594         if (dn.data_blkaddr != COMPRESS_ADDR) {
3595                 dn.ofs_in_node += block - start_idx;
3596                 blknr = f2fs_data_blkaddr(&dn);
3597                 if (!__is_valid_data_blkaddr(blknr))
3598                         blknr = 0;
3599         }
3600
3601         f2fs_put_dnode(&dn);
3602         return blknr;
3603 #else
3604         return 0;
3605 #endif
3606 }
3607
3608
3609 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3610 {
3611         struct inode *inode = mapping->host;
3612         sector_t blknr = 0;
3613
3614         if (f2fs_has_inline_data(inode))
3615                 goto out;
3616
3617         /* make sure allocating whole blocks */
3618         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3619                 filemap_write_and_wait(mapping);
3620
3621         /* Block number less than F2FS MAX BLOCKS */
3622         if (unlikely(block >= max_file_blocks(inode)))
3623                 goto out;
3624
3625         if (f2fs_compressed_file(inode)) {
3626                 blknr = f2fs_bmap_compress(inode, block);
3627         } else {
3628                 struct f2fs_map_blocks map;
3629
3630                 memset(&map, 0, sizeof(map));
3631                 map.m_lblk = block;
3632                 map.m_len = 1;
3633                 map.m_next_pgofs = NULL;
3634                 map.m_seg_type = NO_CHECK_TYPE;
3635
3636                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3637                         blknr = map.m_pblk;
3638         }
3639 out:
3640         trace_f2fs_bmap(inode, block, blknr);
3641         return blknr;
3642 }
3643
3644 #ifdef CONFIG_MIGRATION
3645 #include <linux/migrate.h>
3646
3647 int f2fs_migrate_page(struct address_space *mapping,
3648                 struct page *newpage, struct page *page, enum migrate_mode mode)
3649 {
3650         int rc, extra_count;
3651         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3652         bool atomic_written = page_private_atomic(page);
3653
3654         BUG_ON(PageWriteback(page));
3655
3656         /* migrating an atomic written page is safe with the inmem_lock hold */
3657         if (atomic_written) {
3658                 if (mode != MIGRATE_SYNC)
3659                         return -EBUSY;
3660                 if (!mutex_trylock(&fi->inmem_lock))
3661                         return -EAGAIN;
3662         }
3663
3664         /* one extra reference was held for atomic_write page */
3665         extra_count = atomic_written ? 1 : 0;
3666         rc = migrate_page_move_mapping(mapping, newpage,
3667                                 page, extra_count);
3668         if (rc != MIGRATEPAGE_SUCCESS) {
3669                 if (atomic_written)
3670                         mutex_unlock(&fi->inmem_lock);
3671                 return rc;
3672         }
3673
3674         if (atomic_written) {
3675                 struct inmem_pages *cur;
3676
3677                 list_for_each_entry(cur, &fi->inmem_pages, list)
3678                         if (cur->page == page) {
3679                                 cur->page = newpage;
3680                                 break;
3681                         }
3682                 mutex_unlock(&fi->inmem_lock);
3683                 put_page(page);
3684                 get_page(newpage);
3685         }
3686
3687         /* guarantee to start from no stale private field */
3688         set_page_private(newpage, 0);
3689         if (PagePrivate(page)) {
3690                 set_page_private(newpage, page_private(page));
3691                 SetPagePrivate(newpage);
3692                 get_page(newpage);
3693
3694                 set_page_private(page, 0);
3695                 ClearPagePrivate(page);
3696                 put_page(page);
3697         }
3698
3699         if (mode != MIGRATE_SYNC_NO_COPY)
3700                 migrate_page_copy(newpage, page);
3701         else
3702                 migrate_page_states(newpage, page);
3703
3704         return MIGRATEPAGE_SUCCESS;
3705 }
3706 #endif
3707
3708 #ifdef CONFIG_SWAP
3709 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3710                                                         unsigned int blkcnt)
3711 {
3712         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3713         unsigned int blkofs;
3714         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3715         unsigned int secidx = start_blk / blk_per_sec;
3716         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3717         int ret = 0;
3718
3719         f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3720         filemap_invalidate_lock(inode->i_mapping);
3721
3722         set_inode_flag(inode, FI_ALIGNED_WRITE);
3723         set_inode_flag(inode, FI_OPU_WRITE);
3724
3725         for (; secidx < end_sec; secidx++) {
3726                 f2fs_down_write(&sbi->pin_sem);
3727
3728                 f2fs_lock_op(sbi);
3729                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3730                 f2fs_unlock_op(sbi);
3731
3732                 set_inode_flag(inode, FI_SKIP_WRITES);
3733
3734                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3735                         struct page *page;
3736                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3737
3738                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3739                         if (IS_ERR(page)) {
3740                                 f2fs_up_write(&sbi->pin_sem);
3741                                 ret = PTR_ERR(page);
3742                                 goto done;
3743                         }
3744
3745                         set_page_dirty(page);
3746                         f2fs_put_page(page, 1);
3747                 }
3748
3749                 clear_inode_flag(inode, FI_SKIP_WRITES);
3750
3751                 ret = filemap_fdatawrite(inode->i_mapping);
3752
3753                 f2fs_up_write(&sbi->pin_sem);
3754
3755                 if (ret)
3756                         break;
3757         }
3758
3759 done:
3760         clear_inode_flag(inode, FI_SKIP_WRITES);
3761         clear_inode_flag(inode, FI_OPU_WRITE);
3762         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3763
3764         filemap_invalidate_unlock(inode->i_mapping);
3765         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3766
3767         return ret;
3768 }
3769
3770 static int check_swap_activate(struct swap_info_struct *sis,
3771                                 struct file *swap_file, sector_t *span)
3772 {
3773         struct address_space *mapping = swap_file->f_mapping;
3774         struct inode *inode = mapping->host;
3775         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3776         sector_t cur_lblock;
3777         sector_t last_lblock;
3778         sector_t pblock;
3779         sector_t lowest_pblock = -1;
3780         sector_t highest_pblock = 0;
3781         int nr_extents = 0;
3782         unsigned long nr_pblocks;
3783         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3784         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3785         unsigned int not_aligned = 0;
3786         int ret = 0;
3787
3788         /*
3789          * Map all the blocks into the extent list.  This code doesn't try
3790          * to be very smart.
3791          */
3792         cur_lblock = 0;
3793         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3794
3795         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3796                 struct f2fs_map_blocks map;
3797 retry:
3798                 cond_resched();
3799
3800                 memset(&map, 0, sizeof(map));
3801                 map.m_lblk = cur_lblock;
3802                 map.m_len = last_lblock - cur_lblock;
3803                 map.m_next_pgofs = NULL;
3804                 map.m_next_extent = NULL;
3805                 map.m_seg_type = NO_CHECK_TYPE;
3806                 map.m_may_create = false;
3807
3808                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3809                 if (ret)
3810                         goto out;
3811
3812                 /* hole */
3813                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3814                         f2fs_err(sbi, "Swapfile has holes");
3815                         ret = -EINVAL;
3816                         goto out;
3817                 }
3818
3819                 pblock = map.m_pblk;
3820                 nr_pblocks = map.m_len;
3821
3822                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3823                                 nr_pblocks & sec_blks_mask) {
3824                         not_aligned++;
3825
3826                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3827                         if (cur_lblock + nr_pblocks > sis->max)
3828                                 nr_pblocks -= blks_per_sec;
3829
3830                         if (!nr_pblocks) {
3831                                 /* this extent is last one */
3832                                 nr_pblocks = map.m_len;
3833                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3834                                 goto next;
3835                         }
3836
3837                         ret = f2fs_migrate_blocks(inode, cur_lblock,
3838                                                         nr_pblocks);
3839                         if (ret)
3840                                 goto out;
3841                         goto retry;
3842                 }
3843 next:
3844                 if (cur_lblock + nr_pblocks >= sis->max)
3845                         nr_pblocks = sis->max - cur_lblock;
3846
3847                 if (cur_lblock) {       /* exclude the header page */
3848                         if (pblock < lowest_pblock)
3849                                 lowest_pblock = pblock;
3850                         if (pblock + nr_pblocks - 1 > highest_pblock)
3851                                 highest_pblock = pblock + nr_pblocks - 1;
3852                 }
3853
3854                 /*
3855                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3856                  */
3857                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3858                 if (ret < 0)
3859                         goto out;
3860                 nr_extents += ret;
3861                 cur_lblock += nr_pblocks;
3862         }
3863         ret = nr_extents;
3864         *span = 1 + highest_pblock - lowest_pblock;
3865         if (cur_lblock == 0)
3866                 cur_lblock = 1; /* force Empty message */
3867         sis->max = cur_lblock;
3868         sis->pages = cur_lblock - 1;
3869         sis->highest_bit = cur_lblock - 1;
3870 out:
3871         if (not_aligned)
3872                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3873                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
3874         return ret;
3875 }
3876
3877 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3878                                 sector_t *span)
3879 {
3880         struct inode *inode = file_inode(file);
3881         int ret;
3882
3883         if (!S_ISREG(inode->i_mode))
3884                 return -EINVAL;
3885
3886         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3887                 return -EROFS;
3888
3889         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3890                 f2fs_err(F2FS_I_SB(inode),
3891                         "Swapfile not supported in LFS mode");
3892                 return -EINVAL;
3893         }
3894
3895         ret = f2fs_convert_inline_inode(inode);
3896         if (ret)
3897                 return ret;
3898
3899         if (!f2fs_disable_compressed_file(inode))
3900                 return -EINVAL;
3901
3902         f2fs_precache_extents(inode);
3903
3904         ret = check_swap_activate(sis, file, span);
3905         if (ret < 0)
3906                 return ret;
3907
3908         set_inode_flag(inode, FI_PIN_FILE);
3909         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3910         return ret;
3911 }
3912
3913 static void f2fs_swap_deactivate(struct file *file)
3914 {
3915         struct inode *inode = file_inode(file);
3916
3917         clear_inode_flag(inode, FI_PIN_FILE);
3918 }
3919 #else
3920 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3921                                 sector_t *span)
3922 {
3923         return -EOPNOTSUPP;
3924 }
3925
3926 static void f2fs_swap_deactivate(struct file *file)
3927 {
3928 }
3929 #endif
3930
3931 const struct address_space_operations f2fs_dblock_aops = {
3932         .readpage       = f2fs_read_data_page,
3933         .readahead      = f2fs_readahead,
3934         .writepage      = f2fs_write_data_page,
3935         .writepages     = f2fs_write_data_pages,
3936         .write_begin    = f2fs_write_begin,
3937         .write_end      = f2fs_write_end,
3938         .dirty_folio    = f2fs_dirty_data_folio,
3939         .invalidate_folio = f2fs_invalidate_folio,
3940         .releasepage    = f2fs_release_page,
3941         .direct_IO      = noop_direct_IO,
3942         .bmap           = f2fs_bmap,
3943         .swap_activate  = f2fs_swap_activate,
3944         .swap_deactivate = f2fs_swap_deactivate,
3945 #ifdef CONFIG_MIGRATION
3946         .migratepage    = f2fs_migrate_page,
3947 #endif
3948 };
3949
3950 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3951 {
3952         struct address_space *mapping = page_mapping(page);
3953         unsigned long flags;
3954
3955         xa_lock_irqsave(&mapping->i_pages, flags);
3956         __xa_clear_mark(&mapping->i_pages, page_index(page),
3957                                                 PAGECACHE_TAG_DIRTY);
3958         xa_unlock_irqrestore(&mapping->i_pages, flags);
3959 }
3960
3961 int __init f2fs_init_post_read_processing(void)
3962 {
3963         bio_post_read_ctx_cache =
3964                 kmem_cache_create("f2fs_bio_post_read_ctx",
3965                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
3966         if (!bio_post_read_ctx_cache)
3967                 goto fail;
3968         bio_post_read_ctx_pool =
3969                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3970                                          bio_post_read_ctx_cache);
3971         if (!bio_post_read_ctx_pool)
3972                 goto fail_free_cache;
3973         return 0;
3974
3975 fail_free_cache:
3976         kmem_cache_destroy(bio_post_read_ctx_cache);
3977 fail:
3978         return -ENOMEM;
3979 }
3980
3981 void f2fs_destroy_post_read_processing(void)
3982 {
3983         mempool_destroy(bio_post_read_ctx_pool);
3984         kmem_cache_destroy(bio_post_read_ctx_cache);
3985 }
3986
3987 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
3988 {
3989         if (!f2fs_sb_has_encrypt(sbi) &&
3990                 !f2fs_sb_has_verity(sbi) &&
3991                 !f2fs_sb_has_compression(sbi))
3992                 return 0;
3993
3994         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
3995                                                  WQ_UNBOUND | WQ_HIGHPRI,
3996                                                  num_online_cpus());
3997         if (!sbi->post_read_wq)
3998                 return -ENOMEM;
3999         return 0;
4000 }
4001
4002 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4003 {
4004         if (sbi->post_read_wq)
4005                 destroy_workqueue(sbi->post_read_wq);
4006 }
4007
4008 int __init f2fs_init_bio_entry_cache(void)
4009 {
4010         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4011                         sizeof(struct bio_entry));
4012         if (!bio_entry_slab)
4013                 return -ENOMEM;
4014         return 0;
4015 }
4016
4017 void f2fs_destroy_bio_entry_cache(void)
4018 {
4019         kmem_cache_destroy(bio_entry_slab);
4020 }
4021
4022 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4023                             unsigned int flags, struct iomap *iomap,
4024                             struct iomap *srcmap)
4025 {
4026         struct f2fs_map_blocks map = {};
4027         pgoff_t next_pgofs = 0;
4028         int err;
4029
4030         map.m_lblk = bytes_to_blks(inode, offset);
4031         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4032         map.m_next_pgofs = &next_pgofs;
4033         map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4034         if (flags & IOMAP_WRITE)
4035                 map.m_may_create = true;
4036
4037         err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4038                               F2FS_GET_BLOCK_DIO);
4039         if (err)
4040                 return err;
4041
4042         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4043
4044         /*
4045          * When inline encryption is enabled, sometimes I/O to an encrypted file
4046          * has to be broken up to guarantee DUN contiguity.  Handle this by
4047          * limiting the length of the mapping returned.
4048          */
4049         map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4050
4051         if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4052                 iomap->length = blks_to_bytes(inode, map.m_len);
4053                 if (map.m_flags & F2FS_MAP_MAPPED) {
4054                         iomap->type = IOMAP_MAPPED;
4055                         iomap->flags |= IOMAP_F_MERGED;
4056                 } else {
4057                         iomap->type = IOMAP_UNWRITTEN;
4058                 }
4059                 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4060                         return -EINVAL;
4061
4062                 iomap->bdev = map.m_bdev;
4063                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4064         } else {
4065                 iomap->length = blks_to_bytes(inode, next_pgofs) -
4066                                 iomap->offset;
4067                 iomap->type = IOMAP_HOLE;
4068                 iomap->addr = IOMAP_NULL_ADDR;
4069         }
4070
4071         if (map.m_flags & F2FS_MAP_NEW)
4072                 iomap->flags |= IOMAP_F_NEW;
4073         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4074             offset + length > i_size_read(inode))
4075                 iomap->flags |= IOMAP_F_DIRTY;
4076
4077         return 0;
4078 }
4079
4080 const struct iomap_ops f2fs_iomap_ops = {
4081         .iomap_begin    = f2fs_iomap_begin,
4082 };