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         return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  43                                         0, BIOSET_NEED_BVECS);
  44 }
  45
  46 void f2fs_destroy_bioset(void)
  47 {
  48         bioset_exit(&f2fs_bioset);
  49 }
  50
  51 static bool __is_cp_guaranteed(struct page *page)
  52 {
  53         struct address_space *mapping = page->mapping;
  54         struct inode *inode;
  55         struct f2fs_sb_info *sbi;
  56
  57         if (!mapping)
  58                 return false;
  59
  60         inode = mapping->host;
  61         sbi = F2FS_I_SB(inode);
  62
  63         if (inode->i_ino == F2FS_META_INO(sbi) ||
  64                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  65                         S_ISDIR(inode->i_mode))
  66                 return true;
  67
  68         if (f2fs_is_compressed_page(page))
  69                 return false;
  70         if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
  71                         page_private_gcing(page))
  72                 return true;
  73         return false;
  74 }
  75
  76 static enum count_type __read_io_type(struct page *page)
  77 {
  78         struct address_space *mapping = page_file_mapping(page);
  79
  80         if (mapping) {
  81                 struct inode *inode = mapping->host;
  82                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  83
  84                 if (inode->i_ino == F2FS_META_INO(sbi))
  85                         return F2FS_RD_META;
  86
  87                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  88                         return F2FS_RD_NODE;
  89         }
  90         return F2FS_RD_DATA;
  91 }
  92
  93 /* postprocessing steps for read bios */
  94 enum bio_post_read_step {
  95 #ifdef CONFIG_FS_ENCRYPTION
  96         STEP_DECRYPT    = BIT(0),
  97 #else
  98         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
  99 #endif
 100 #ifdef CONFIG_F2FS_FS_COMPRESSION
 101         STEP_DECOMPRESS = BIT(1),
 102 #else
 103         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 104 #endif
 105 #ifdef CONFIG_FS_VERITY
 106         STEP_VERITY     = BIT(2),
 107 #else
 108         STEP_VERITY     = 0,    /* compile out the verity-related code */
 109 #endif
 110 };
 111
 112 struct bio_post_read_ctx {
 113         struct bio *bio;
 114         struct f2fs_sb_info *sbi;
 115         struct work_struct work;
 116         unsigned int enabled_steps;
 117         /*
 118          * decompression_attempted keeps track of whether
 119          * f2fs_end_read_compressed_page() has been called on the pages in the
 120          * bio that belong to a compressed cluster yet.
 121          */
 122         bool decompression_attempted;
 123         block_t fs_blkaddr;
 124 };
 125
 126 /*
 127  * Update and unlock a bio's pages, and free the bio.
 128  *
 129  * This marks pages up-to-date only if there was no error in the bio (I/O error,
 130  * decryption error, or verity error), as indicated by bio->bi_status.
 131  *
 132  * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk)
 133  * aren't marked up-to-date here, as decompression is done on a per-compression-
 134  * cluster basis rather than a per-bio basis.  Instead, we only must do two
 135  * things for each compressed page here: call f2fs_end_read_compressed_page()
 136  * with failed=true if an error occurred before it would have normally gotten
 137  * called (i.e., I/O error or decryption error, but *not* verity error), and
 138  * release the bio's reference to the decompress_io_ctx of the page's cluster.
 139  */
 140 static void f2fs_finish_read_bio(struct bio *bio, bool in_task)
 141 {
 142         struct bio_vec *bv;
 143         struct bvec_iter_all iter_all;
 144         struct bio_post_read_ctx *ctx = bio->bi_private;
 145
 146         bio_for_each_segment_all(bv, bio, iter_all) {
 147                 struct page *page = bv->bv_page;
 148
 149                 if (f2fs_is_compressed_page(page)) {
 150                         if (ctx && !ctx->decompression_attempted)
 151                                 f2fs_end_read_compressed_page(page, true, 0,
 152                                                         in_task);
 153                         f2fs_put_page_dic(page, in_task);
 154                         continue;
 155                 }
 156
 157                 if (bio->bi_status)
 158                         ClearPageUptodate(page);
 159                 else
 160                         SetPageUptodate(page);
 161                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 162                 unlock_page(page);
 163         }
 164
 165         if (ctx)
 166                 mempool_free(ctx, bio_post_read_ctx_pool);
 167         bio_put(bio);
 168 }
 169
 170 static void f2fs_verify_bio(struct work_struct *work)
 171 {
 172         struct bio_post_read_ctx *ctx =
 173                 container_of(work, struct bio_post_read_ctx, work);
 174         struct bio *bio = ctx->bio;
 175         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 176
 177         /*
 178          * fsverity_verify_bio() may call readahead() again, and while verity
 179          * will be disabled for this, decryption and/or decompression may still
 180          * be needed, resulting in another bio_post_read_ctx being allocated.
 181          * So to prevent deadlocks we need to release the current ctx to the
 182          * mempool first.  This assumes that verity is the last post-read step.
 183          */
 184         mempool_free(ctx, bio_post_read_ctx_pool);
 185         bio->bi_private = NULL;
 186
 187         /*
 188          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 189          * as those were handled separately by f2fs_end_read_compressed_page().
 190          */
 191         if (may_have_compressed_pages) {
 192                 struct bio_vec *bv;
 193                 struct bvec_iter_all iter_all;
 194
 195                 bio_for_each_segment_all(bv, bio, iter_all) {
 196                         struct page *page = bv->bv_page;
 197
 198                         if (!f2fs_is_compressed_page(page) &&
 199                             !fsverity_verify_page(page)) {
 200                                 bio->bi_status = BLK_STS_IOERR;
 201                                 break;
 202                         }
 203                 }
 204         } else {
 205                 fsverity_verify_bio(bio);
 206         }
 207
 208         f2fs_finish_read_bio(bio, true);
 209 }
 210
 211 /*
 212  * If the bio's data needs to be verified with fs-verity, then enqueue the
 213  * verity work for the bio.  Otherwise finish the bio now.
 214  *
 215  * Note that to avoid deadlocks, the verity work can't be done on the
 216  * decryption/decompression workqueue.  This is because verifying the data pages
 217  * can involve reading verity metadata pages from the file, and these verity
 218  * metadata pages may be encrypted and/or compressed.
 219  */
 220 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task)
 221 {
 222         struct bio_post_read_ctx *ctx = bio->bi_private;
 223
 224         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 225                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 226                 fsverity_enqueue_verify_work(&ctx->work);
 227         } else {
 228                 f2fs_finish_read_bio(bio, in_task);
 229         }
 230 }
 231
 232 /*
 233  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 234  * remaining page was read by @ctx->bio.
 235  *
 236  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 237  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 238  * that the bio includes at least one compressed page.  The actual decompression
 239  * is done on a per-cluster basis, not a per-bio basis.
 240  */
 241 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx,
 242                 bool in_task)
 243 {
 244         struct bio_vec *bv;
 245         struct bvec_iter_all iter_all;
 246         bool all_compressed = true;
 247         block_t blkaddr = ctx->fs_blkaddr;
 248
 249         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 250                 struct page *page = bv->bv_page;
 251
 252                 if (f2fs_is_compressed_page(page))
 253                         f2fs_end_read_compressed_page(page, false, blkaddr,
 254                                                       in_task);
 255                 else
 256                         all_compressed = false;
 257
 258                 blkaddr++;
 259         }
 260
 261         ctx->decompression_attempted = true;
 262
 263         /*
 264          * Optimization: if all the bio's pages are compressed, then scheduling
 265          * the per-bio verity work is unnecessary, as verity will be fully
 266          * handled at the compression cluster level.
 267          */
 268         if (all_compressed)
 269                 ctx->enabled_steps &= ~STEP_VERITY;
 270 }
 271
 272 static void f2fs_post_read_work(struct work_struct *work)
 273 {
 274         struct bio_post_read_ctx *ctx =
 275                 container_of(work, struct bio_post_read_ctx, work);
 276         struct bio *bio = ctx->bio;
 277
 278         if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) {
 279                 f2fs_finish_read_bio(bio, true);
 280                 return;
 281         }
 282
 283         if (ctx->enabled_steps & STEP_DECOMPRESS)
 284                 f2fs_handle_step_decompress(ctx, true);
 285
 286         f2fs_verify_and_finish_bio(bio, true);
 287 }
 288
 289 static void f2fs_read_end_io(struct bio *bio)
 290 {
 291         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 292         struct bio_post_read_ctx *ctx;
 293         bool intask = in_task();
 294
 295         iostat_update_and_unbind_ctx(bio);
 296         ctx = bio->bi_private;
 297
 298         if (time_to_inject(sbi, FAULT_READ_IO))
 299                 bio->bi_status = BLK_STS_IOERR;
 300
 301         if (bio->bi_status) {
 302                 f2fs_finish_read_bio(bio, intask);
 303                 return;
 304         }
 305
 306         if (ctx) {
 307                 unsigned int enabled_steps = ctx->enabled_steps &
 308                                         (STEP_DECRYPT | STEP_DECOMPRESS);
 309
 310                 /*
 311                  * If we have only decompression step between decompression and
 312                  * decrypt, we don't need post processing for this.
 313                  */
 314                 if (enabled_steps == STEP_DECOMPRESS &&
 315                                 !f2fs_low_mem_mode(sbi)) {
 316                         f2fs_handle_step_decompress(ctx, intask);
 317                 } else if (enabled_steps) {
 318                         INIT_WORK(&ctx->work, f2fs_post_read_work);
 319                         queue_work(ctx->sbi->post_read_wq, &ctx->work);
 320                         return;
 321                 }
 322         }
 323
 324         f2fs_verify_and_finish_bio(bio, intask);
 325 }
 326
 327 static void f2fs_write_end_io(struct bio *bio)
 328 {
 329         struct f2fs_sb_info *sbi;
 330         struct bio_vec *bvec;
 331         struct bvec_iter_all iter_all;
 332
 333         iostat_update_and_unbind_ctx(bio);
 334         sbi = bio->bi_private;
 335
 336         if (time_to_inject(sbi, FAULT_WRITE_IO))
 337                 bio->bi_status = BLK_STS_IOERR;
 338
 339         bio_for_each_segment_all(bvec, bio, iter_all) {
 340                 struct page *page = bvec->bv_page;
 341                 enum count_type type = WB_DATA_TYPE(page);
 342
 343                 if (page_private_dummy(page)) {
 344                         clear_page_private_dummy(page);
 345                         unlock_page(page);
 346                         mempool_free(page, sbi->write_io_dummy);
 347
 348                         if (unlikely(bio->bi_status))
 349                                 f2fs_stop_checkpoint(sbi, true,
 350                                                 STOP_CP_REASON_WRITE_FAIL);
 351                         continue;
 352                 }
 353
 354                 fscrypt_finalize_bounce_page(&page);
 355
 356 #ifdef CONFIG_F2FS_FS_COMPRESSION
 357                 if (f2fs_is_compressed_page(page)) {
 358                         f2fs_compress_write_end_io(bio, page);
 359                         continue;
 360                 }
 361 #endif
 362
 363                 if (unlikely(bio->bi_status)) {
 364                         mapping_set_error(page->mapping, -EIO);
 365                         if (type == F2FS_WB_CP_DATA)
 366                                 f2fs_stop_checkpoint(sbi, true,
 367                                                 STOP_CP_REASON_WRITE_FAIL);
 368                 }
 369
 370                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 371                                         page->index != nid_of_node(page));
 372
 373                 dec_page_count(sbi, type);
 374                 if (f2fs_in_warm_node_list(sbi, page))
 375                         f2fs_del_fsync_node_entry(sbi, page);
 376                 clear_page_private_gcing(page);
 377                 end_page_writeback(page);
 378         }
 379         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 380                                 wq_has_sleeper(&sbi->cp_wait))
 381                 wake_up(&sbi->cp_wait);
 382
 383         bio_put(bio);
 384 }
 385
 386 #ifdef CONFIG_BLK_DEV_ZONED
 387 static void f2fs_zone_write_end_io(struct bio *bio)
 388 {
 389         struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
 390
 391         bio->bi_private = io->bi_private;
 392         complete(&io->zone_wait);
 393         f2fs_write_end_io(bio);
 394 }
 395 #endif
 396
 397 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 398                 block_t blk_addr, sector_t *sector)
 399 {
 400         struct block_device *bdev = sbi->sb->s_bdev;
 401         int i;
 402
 403         if (f2fs_is_multi_device(sbi)) {
 404                 for (i = 0; i < sbi->s_ndevs; i++) {
 405                         if (FDEV(i).start_blk <= blk_addr &&
 406                             FDEV(i).end_blk >= blk_addr) {
 407                                 blk_addr -= FDEV(i).start_blk;
 408                                 bdev = FDEV(i).bdev;
 409                                 break;
 410                         }
 411                 }
 412         }
 413
 414         if (sector)
 415                 *sector = SECTOR_FROM_BLOCK(blk_addr);
 416         return bdev;
 417 }
 418
 419 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 420 {
 421         int i;
 422
 423         if (!f2fs_is_multi_device(sbi))
 424                 return 0;
 425
 426         for (i = 0; i < sbi->s_ndevs; i++)
 427                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 428                         return i;
 429         return 0;
 430 }
 431
 432 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
 433 {
 434         unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0);
 435         unsigned int fua_flag, meta_flag, io_flag;
 436         blk_opf_t op_flags = 0;
 437
 438         if (fio->op != REQ_OP_WRITE)
 439                 return 0;
 440         if (fio->type == DATA)
 441                 io_flag = fio->sbi->data_io_flag;
 442         else if (fio->type == NODE)
 443                 io_flag = fio->sbi->node_io_flag;
 444         else
 445                 return 0;
 446
 447         fua_flag = io_flag & temp_mask;
 448         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 449
 450         /*
 451          * data/node io flag bits per temp:
 452          *      REQ_META     |      REQ_FUA      |
 453          *    5 |    4 |   3 |    2 |    1 |   0 |
 454          * Cold | Warm | Hot | Cold | Warm | Hot |
 455          */
 456         if (BIT(fio->temp) & meta_flag)
 457                 op_flags |= REQ_META;
 458         if (BIT(fio->temp) & fua_flag)
 459                 op_flags |= REQ_FUA;
 460         return op_flags;
 461 }
 462
 463 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 464 {
 465         struct f2fs_sb_info *sbi = fio->sbi;
 466         struct block_device *bdev;
 467         sector_t sector;
 468         struct bio *bio;
 469
 470         bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 471         bio = bio_alloc_bioset(bdev, npages,
 472                                 fio->op | fio->op_flags | f2fs_io_flags(fio),
 473                                 GFP_NOIO, &f2fs_bioset);
 474         bio->bi_iter.bi_sector = sector;
 475         if (is_read_io(fio->op)) {
 476                 bio->bi_end_io = f2fs_read_end_io;
 477                 bio->bi_private = NULL;
 478         } else {
 479                 bio->bi_end_io = f2fs_write_end_io;
 480                 bio->bi_private = sbi;
 481         }
 482         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 483
 484         if (fio->io_wbc)
 485                 wbc_init_bio(fio->io_wbc, bio);
 486
 487         return bio;
 488 }
 489
 490 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 491                                   pgoff_t first_idx,
 492                                   const struct f2fs_io_info *fio,
 493                                   gfp_t gfp_mask)
 494 {
 495         /*
 496          * The f2fs garbage collector sets ->encrypted_page when it wants to
 497          * read/write raw data without encryption.
 498          */
 499         if (!fio || !fio->encrypted_page)
 500                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 501 }
 502
 503 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 504                                      pgoff_t next_idx,
 505                                      const struct f2fs_io_info *fio)
 506 {
 507         /*
 508          * The f2fs garbage collector sets ->encrypted_page when it wants to
 509          * read/write raw data without encryption.
 510          */
 511         if (fio && fio->encrypted_page)
 512                 return !bio_has_crypt_ctx(bio);
 513
 514         return fscrypt_mergeable_bio(bio, inode, next_idx);
 515 }
 516
 517 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 518                                  enum page_type type)
 519 {
 520         WARN_ON_ONCE(!is_read_io(bio_op(bio)));
 521         trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 522
 523         iostat_update_submit_ctx(bio, type);
 524         submit_bio(bio);
 525 }
 526
 527 static void f2fs_align_write_bio(struct f2fs_sb_info *sbi, struct bio *bio)
 528 {
 529         unsigned int start =
 530                 (bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS) % F2FS_IO_SIZE(sbi);
 531
 532         if (start == 0)
 533                 return;
 534
 535         /* fill dummy pages */
 536         for (; start < F2FS_IO_SIZE(sbi); start++) {
 537                 struct page *page =
 538                         mempool_alloc(sbi->write_io_dummy,
 539                                       GFP_NOIO | __GFP_NOFAIL);
 540                 f2fs_bug_on(sbi, !page);
 541
 542                 lock_page(page);
 543
 544                 zero_user_segment(page, 0, PAGE_SIZE);
 545                 set_page_private_dummy(page);
 546
 547                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 548                         f2fs_bug_on(sbi, 1);
 549         }
 550 }
 551
 552 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 553                                   enum page_type type)
 554 {
 555         WARN_ON_ONCE(is_read_io(bio_op(bio)));
 556
 557         if (type == DATA || type == NODE) {
 558                 if (f2fs_lfs_mode(sbi) && current->plug)
 559                         blk_finish_plug(current->plug);
 560
 561                 if (F2FS_IO_ALIGNED(sbi)) {
 562                         f2fs_align_write_bio(sbi, bio);
 563                         /*
 564                          * In the NODE case, we lose next block address chain.
 565                          * So, we need to do checkpoint in f2fs_sync_file.
 566                          */
 567                         if (type == NODE)
 568                                 set_sbi_flag(sbi, SBI_NEED_CP);
 569                 }
 570         }
 571
 572         trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 573         iostat_update_submit_ctx(bio, type);
 574         submit_bio(bio);
 575 }
 576
 577 static void __submit_merged_bio(struct f2fs_bio_info *io)
 578 {
 579         struct f2fs_io_info *fio = &io->fio;
 580
 581         if (!io->bio)
 582                 return;
 583
 584         if (is_read_io(fio->op)) {
 585                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 586                 f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
 587         } else {
 588                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 589                 f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
 590         }
 591         io->bio = NULL;
 592 }
 593
 594 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 595                                                 struct page *page, nid_t ino)
 596 {
 597         struct bio_vec *bvec;
 598         struct bvec_iter_all iter_all;
 599
 600         if (!bio)
 601                 return false;
 602
 603         if (!inode && !page && !ino)
 604                 return true;
 605
 606         bio_for_each_segment_all(bvec, bio, iter_all) {
 607                 struct page *target = bvec->bv_page;
 608
 609                 if (fscrypt_is_bounce_page(target)) {
 610                         target = fscrypt_pagecache_page(target);
 611                         if (IS_ERR(target))
 612                                 continue;
 613                 }
 614                 if (f2fs_is_compressed_page(target)) {
 615                         target = f2fs_compress_control_page(target);
 616                         if (IS_ERR(target))
 617                                 continue;
 618                 }
 619
 620                 if (inode && inode == target->mapping->host)
 621                         return true;
 622                 if (page && page == target)
 623                         return true;
 624                 if (ino && ino == ino_of_node(target))
 625                         return true;
 626         }
 627
 628         return false;
 629 }
 630
 631 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
 632 {
 633         int i;
 634
 635         for (i = 0; i < NR_PAGE_TYPE; i++) {
 636                 int n = (i == META) ? 1 : NR_TEMP_TYPE;
 637                 int j;
 638
 639                 sbi->write_io[i] = f2fs_kmalloc(sbi,
 640                                 array_size(n, sizeof(struct f2fs_bio_info)),
 641                                 GFP_KERNEL);
 642                 if (!sbi->write_io[i])
 643                         return -ENOMEM;
 644
 645                 for (j = HOT; j < n; j++) {
 646                         init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
 647                         sbi->write_io[i][j].sbi = sbi;
 648                         sbi->write_io[i][j].bio = NULL;
 649                         spin_lock_init(&sbi->write_io[i][j].io_lock);
 650                         INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
 651                         INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
 652                         init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
 653 #ifdef CONFIG_BLK_DEV_ZONED
 654                         init_completion(&sbi->write_io[i][j].zone_wait);
 655                         sbi->write_io[i][j].zone_pending_bio = NULL;
 656                         sbi->write_io[i][j].bi_private = NULL;
 657 #endif
 658                 }
 659         }
 660
 661         return 0;
 662 }
 663
 664 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 665                                 enum page_type type, enum temp_type temp)
 666 {
 667         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 668         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 669
 670         f2fs_down_write(&io->io_rwsem);
 671
 672         if (!io->bio)
 673                 goto unlock_out;
 674
 675         /* change META to META_FLUSH in the checkpoint procedure */
 676         if (type >= META_FLUSH) {
 677                 io->fio.type = META_FLUSH;
 678                 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 679                 if (!test_opt(sbi, NOBARRIER))
 680                         io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 681         }
 682         __submit_merged_bio(io);
 683 unlock_out:
 684         f2fs_up_write(&io->io_rwsem);
 685 }
 686
 687 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 688                                 struct inode *inode, struct page *page,
 689                                 nid_t ino, enum page_type type, bool force)
 690 {
 691         enum temp_type temp;
 692         bool ret = true;
 693
 694         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 695                 if (!force)     {
 696                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 697                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 698
 699                         f2fs_down_read(&io->io_rwsem);
 700                         ret = __has_merged_page(io->bio, inode, page, ino);
 701                         f2fs_up_read(&io->io_rwsem);
 702                 }
 703                 if (ret)
 704                         __f2fs_submit_merged_write(sbi, type, temp);
 705
 706                 /* TODO: use HOT temp only for meta pages now. */
 707                 if (type >= META)
 708                         break;
 709         }
 710 }
 711
 712 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 713 {
 714         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 715 }
 716
 717 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 718                                 struct inode *inode, struct page *page,
 719                                 nid_t ino, enum page_type type)
 720 {
 721         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 722 }
 723
 724 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 725 {
 726         f2fs_submit_merged_write(sbi, DATA);
 727         f2fs_submit_merged_write(sbi, NODE);
 728         f2fs_submit_merged_write(sbi, META);
 729 }
 730
 731 /*
 732  * Fill the locked page with data located in the block address.
 733  * A caller needs to unlock the page on failure.
 734  */
 735 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 736 {
 737         struct bio *bio;
 738         struct page *page = fio->encrypted_page ?
 739                         fio->encrypted_page : fio->page;
 740
 741         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 742                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 743                         META_GENERIC : DATA_GENERIC_ENHANCE))) {
 744                 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
 745                 return -EFSCORRUPTED;
 746         }
 747
 748         trace_f2fs_submit_page_bio(page, fio);
 749
 750         /* Allocate a new bio */
 751         bio = __bio_alloc(fio, 1);
 752
 753         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 754                                fio->page->index, fio, GFP_NOIO);
 755
 756         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 757                 bio_put(bio);
 758                 return -EFAULT;
 759         }
 760
 761         if (fio->io_wbc && !is_read_io(fio->op))
 762                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 763
 764         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 765                         __read_io_type(page) : WB_DATA_TYPE(fio->page));
 766
 767         if (is_read_io(bio_op(bio)))
 768                 f2fs_submit_read_bio(fio->sbi, bio, fio->type);
 769         else
 770                 f2fs_submit_write_bio(fio->sbi, bio, fio->type);
 771         return 0;
 772 }
 773
 774 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 775                                 block_t last_blkaddr, block_t cur_blkaddr)
 776 {
 777         if (unlikely(sbi->max_io_bytes &&
 778                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 779                 return false;
 780         if (last_blkaddr + 1 != cur_blkaddr)
 781                 return false;
 782         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 783 }
 784
 785 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 786                                                 struct f2fs_io_info *fio)
 787 {
 788         if (io->fio.op != fio->op)
 789                 return false;
 790         return io->fio.op_flags == fio->op_flags;
 791 }
 792
 793 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 794                                         struct f2fs_bio_info *io,
 795                                         struct f2fs_io_info *fio,
 796                                         block_t last_blkaddr,
 797                                         block_t cur_blkaddr)
 798 {
 799         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 800                 unsigned int filled_blocks =
 801                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 802                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 803                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 804
 805                 /* IOs in bio is aligned and left space of vectors is not enough */
 806                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 807                         return false;
 808         }
 809         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 810                 return false;
 811         return io_type_is_mergeable(io, fio);
 812 }
 813
 814 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 815                                 struct page *page, enum temp_type temp)
 816 {
 817         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 818         struct bio_entry *be;
 819
 820         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 821         be->bio = bio;
 822         bio_get(bio);
 823
 824         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 825                 f2fs_bug_on(sbi, 1);
 826
 827         f2fs_down_write(&io->bio_list_lock);
 828         list_add_tail(&be->list, &io->bio_list);
 829         f2fs_up_write(&io->bio_list_lock);
 830 }
 831
 832 static void del_bio_entry(struct bio_entry *be)
 833 {
 834         list_del(&be->list);
 835         kmem_cache_free(bio_entry_slab, be);
 836 }
 837
 838 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 839                                                         struct page *page)
 840 {
 841         struct f2fs_sb_info *sbi = fio->sbi;
 842         enum temp_type temp;
 843         bool found = false;
 844         int ret = -EAGAIN;
 845
 846         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 847                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 848                 struct list_head *head = &io->bio_list;
 849                 struct bio_entry *be;
 850
 851                 f2fs_down_write(&io->bio_list_lock);
 852                 list_for_each_entry(be, head, list) {
 853                         if (be->bio != *bio)
 854                                 continue;
 855
 856                         found = true;
 857
 858                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 859                                                             *fio->last_block,
 860                                                             fio->new_blkaddr));
 861                         if (f2fs_crypt_mergeable_bio(*bio,
 862                                         fio->page->mapping->host,
 863                                         fio->page->index, fio) &&
 864                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 865                                         PAGE_SIZE) {
 866                                 ret = 0;
 867                                 break;
 868                         }
 869
 870                         /* page can't be merged into bio; submit the bio */
 871                         del_bio_entry(be);
 872                         f2fs_submit_write_bio(sbi, *bio, DATA);
 873                         break;
 874                 }
 875                 f2fs_up_write(&io->bio_list_lock);
 876         }
 877
 878         if (ret) {
 879                 bio_put(*bio);
 880                 *bio = NULL;
 881         }
 882
 883         return ret;
 884 }
 885
 886 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 887                                         struct bio **bio, struct page *page)
 888 {
 889         enum temp_type temp;
 890         bool found = false;
 891         struct bio *target = bio ? *bio : NULL;
 892
 893         f2fs_bug_on(sbi, !target && !page);
 894
 895         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 896                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 897                 struct list_head *head = &io->bio_list;
 898                 struct bio_entry *be;
 899
 900                 if (list_empty(head))
 901                         continue;
 902
 903                 f2fs_down_read(&io->bio_list_lock);
 904                 list_for_each_entry(be, head, list) {
 905                         if (target)
 906                                 found = (target == be->bio);
 907                         else
 908                                 found = __has_merged_page(be->bio, NULL,
 909                                                                 page, 0);
 910                         if (found)
 911                                 break;
 912                 }
 913                 f2fs_up_read(&io->bio_list_lock);
 914
 915                 if (!found)
 916                         continue;
 917
 918                 found = false;
 919
 920                 f2fs_down_write(&io->bio_list_lock);
 921                 list_for_each_entry(be, head, list) {
 922                         if (target)
 923                                 found = (target == be->bio);
 924                         else
 925                                 found = __has_merged_page(be->bio, NULL,
 926                                                                 page, 0);
 927                         if (found) {
 928                                 target = be->bio;
 929                                 del_bio_entry(be);
 930                                 break;
 931                         }
 932                 }
 933                 f2fs_up_write(&io->bio_list_lock);
 934         }
 935
 936         if (found)
 937                 f2fs_submit_write_bio(sbi, target, DATA);
 938         if (bio && *bio) {
 939                 bio_put(*bio);
 940                 *bio = NULL;
 941         }
 942 }
 943
 944 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 945 {
 946         struct bio *bio = *fio->bio;
 947         struct page *page = fio->encrypted_page ?
 948                         fio->encrypted_page : fio->page;
 949
 950         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 951                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) {
 952                 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
 953                 return -EFSCORRUPTED;
 954         }
 955
 956         trace_f2fs_submit_page_bio(page, fio);
 957
 958         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 959                                                 fio->new_blkaddr))
 960                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 961 alloc_new:
 962         if (!bio) {
 963                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 964                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 965                                        fio->page->index, fio, GFP_NOIO);
 966
 967                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 968         } else {
 969                 if (add_ipu_page(fio, &bio, page))
 970                         goto alloc_new;
 971         }
 972
 973         if (fio->io_wbc)
 974                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 975
 976         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 977
 978         *fio->last_block = fio->new_blkaddr;
 979         *fio->bio = bio;
 980
 981         return 0;
 982 }
 983
 984 #ifdef CONFIG_BLK_DEV_ZONED
 985 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
 986 {
 987         int devi = 0;
 988
 989         if (f2fs_is_multi_device(sbi)) {
 990                 devi = f2fs_target_device_index(sbi, blkaddr);
 991                 if (blkaddr < FDEV(devi).start_blk ||
 992                     blkaddr > FDEV(devi).end_blk) {
 993                         f2fs_err(sbi, "Invalid block %x", blkaddr);
 994                         return false;
 995                 }
 996                 blkaddr -= FDEV(devi).start_blk;
 997         }
 998         return bdev_zoned_model(FDEV(devi).bdev) == BLK_ZONED_HM &&
 999                 f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
1000                 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
1001 }
1002 #endif
1003
1004 void f2fs_submit_page_write(struct f2fs_io_info *fio)
1005 {
1006         struct f2fs_sb_info *sbi = fio->sbi;
1007         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
1008         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
1009         struct page *bio_page;
1010
1011         f2fs_bug_on(sbi, is_read_io(fio->op));
1012
1013         f2fs_down_write(&io->io_rwsem);
1014
1015 #ifdef CONFIG_BLK_DEV_ZONED
1016         if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
1017                 wait_for_completion_io(&io->zone_wait);
1018                 bio_put(io->zone_pending_bio);
1019                 io->zone_pending_bio = NULL;
1020                 io->bi_private = NULL;
1021         }
1022 #endif
1023
1024 next:
1025         if (fio->in_list) {
1026                 spin_lock(&io->io_lock);
1027                 if (list_empty(&io->io_list)) {
1028                         spin_unlock(&io->io_lock);
1029                         goto out;
1030                 }
1031                 fio = list_first_entry(&io->io_list,
1032                                                 struct f2fs_io_info, list);
1033                 list_del(&fio->list);
1034                 spin_unlock(&io->io_lock);
1035         }
1036
1037         verify_fio_blkaddr(fio);
1038
1039         if (fio->encrypted_page)
1040                 bio_page = fio->encrypted_page;
1041         else if (fio->compressed_page)
1042                 bio_page = fio->compressed_page;
1043         else
1044                 bio_page = fio->page;
1045
1046         /* set submitted = true as a return value */
1047         fio->submitted = 1;
1048
1049         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
1050
1051         if (io->bio &&
1052             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
1053                               fio->new_blkaddr) ||
1054              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
1055                                        bio_page->index, fio)))
1056                 __submit_merged_bio(io);
1057 alloc_new:
1058         if (io->bio == NULL) {
1059                 if (F2FS_IO_ALIGNED(sbi) &&
1060                                 (fio->type == DATA || fio->type == NODE) &&
1061                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
1062                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
1063                         fio->retry = 1;
1064                         goto skip;
1065                 }
1066                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1067                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1068                                        bio_page->index, fio, GFP_NOIO);
1069                 io->fio = *fio;
1070         }
1071
1072         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1073                 __submit_merged_bio(io);
1074                 goto alloc_new;
1075         }
1076
1077         if (fio->io_wbc)
1078                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1079
1080         io->last_block_in_bio = fio->new_blkaddr;
1081
1082         trace_f2fs_submit_page_write(fio->page, fio);
1083 skip:
1084         if (fio->in_list)
1085                 goto next;
1086 out:
1087 #ifdef CONFIG_BLK_DEV_ZONED
1088         if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1089                         is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1090                 bio_get(io->bio);
1091                 reinit_completion(&io->zone_wait);
1092                 io->bi_private = io->bio->bi_private;
1093                 io->bio->bi_private = io;
1094                 io->bio->bi_end_io = f2fs_zone_write_end_io;
1095                 io->zone_pending_bio = io->bio;
1096                 __submit_merged_bio(io);
1097         }
1098 #endif
1099         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1100                                 !f2fs_is_checkpoint_ready(sbi))
1101                 __submit_merged_bio(io);
1102         f2fs_up_write(&io->io_rwsem);
1103 }
1104
1105 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1106                                       unsigned nr_pages, blk_opf_t op_flag,
1107                                       pgoff_t first_idx, bool for_write)
1108 {
1109         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1110         struct bio *bio;
1111         struct bio_post_read_ctx *ctx = NULL;
1112         unsigned int post_read_steps = 0;
1113         sector_t sector;
1114         struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
1115
1116         bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1117                                REQ_OP_READ | op_flag,
1118                                for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1119         if (!bio)
1120                 return ERR_PTR(-ENOMEM);
1121         bio->bi_iter.bi_sector = sector;
1122         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1123         bio->bi_end_io = f2fs_read_end_io;
1124
1125         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1126                 post_read_steps |= STEP_DECRYPT;
1127
1128         if (f2fs_need_verity(inode, first_idx))
1129                 post_read_steps |= STEP_VERITY;
1130
1131         /*
1132          * STEP_DECOMPRESS is handled specially, since a compressed file might
1133          * contain both compressed and uncompressed clusters.  We'll allocate a
1134          * bio_post_read_ctx if the file is compressed, but the caller is
1135          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1136          */
1137
1138         if (post_read_steps || f2fs_compressed_file(inode)) {
1139                 /* Due to the mempool, this never fails. */
1140                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1141                 ctx->bio = bio;
1142                 ctx->sbi = sbi;
1143                 ctx->enabled_steps = post_read_steps;
1144                 ctx->fs_blkaddr = blkaddr;
1145                 ctx->decompression_attempted = false;
1146                 bio->bi_private = ctx;
1147         }
1148         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1149
1150         return bio;
1151 }
1152
1153 /* This can handle encryption stuffs */
1154 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1155                                  block_t blkaddr, blk_opf_t op_flags,
1156                                  bool for_write)
1157 {
1158         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1159         struct bio *bio;
1160
1161         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1162                                         page->index, for_write);
1163         if (IS_ERR(bio))
1164                 return PTR_ERR(bio);
1165
1166         /* wait for GCed page writeback via META_MAPPING */
1167         f2fs_wait_on_block_writeback(inode, blkaddr);
1168
1169         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1170                 iostat_update_and_unbind_ctx(bio);
1171                 if (bio->bi_private)
1172                         mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1173                 bio_put(bio);
1174                 return -EFAULT;
1175         }
1176         inc_page_count(sbi, F2FS_RD_DATA);
1177         f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1178         f2fs_submit_read_bio(sbi, bio, DATA);
1179         return 0;
1180 }
1181
1182 static void __set_data_blkaddr(struct dnode_of_data *dn)
1183 {
1184         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1185         __le32 *addr_array;
1186         int base = 0;
1187
1188         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1189                 base = get_extra_isize(dn->inode);
1190
1191         /* Get physical address of data block */
1192         addr_array = blkaddr_in_node(rn);
1193         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1194 }
1195
1196 /*
1197  * Lock ordering for the change of data block address:
1198  * ->data_page
1199  *  ->node_page
1200  *    update block addresses in the node page
1201  */
1202 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1203 {
1204         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1205         __set_data_blkaddr(dn);
1206         if (set_page_dirty(dn->node_page))
1207                 dn->node_changed = true;
1208 }
1209
1210 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1211 {
1212         dn->data_blkaddr = blkaddr;
1213         f2fs_set_data_blkaddr(dn);
1214         f2fs_update_read_extent_cache(dn);
1215 }
1216
1217 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1218 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1219 {
1220         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1221         int err;
1222
1223         if (!count)
1224                 return 0;
1225
1226         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1227                 return -EPERM;
1228         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1229                 return err;
1230
1231         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1232                                                 dn->ofs_in_node, count);
1233
1234         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1235
1236         for (; count > 0; dn->ofs_in_node++) {
1237                 block_t blkaddr = f2fs_data_blkaddr(dn);
1238
1239                 if (blkaddr == NULL_ADDR) {
1240                         dn->data_blkaddr = NEW_ADDR;
1241                         __set_data_blkaddr(dn);
1242                         count--;
1243                 }
1244         }
1245
1246         if (set_page_dirty(dn->node_page))
1247                 dn->node_changed = true;
1248         return 0;
1249 }
1250
1251 /* Should keep dn->ofs_in_node unchanged */
1252 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1253 {
1254         unsigned int ofs_in_node = dn->ofs_in_node;
1255         int ret;
1256
1257         ret = f2fs_reserve_new_blocks(dn, 1);
1258         dn->ofs_in_node = ofs_in_node;
1259         return ret;
1260 }
1261
1262 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1263 {
1264         bool need_put = dn->inode_page ? false : true;
1265         int err;
1266
1267         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1268         if (err)
1269                 return err;
1270
1271         if (dn->data_blkaddr == NULL_ADDR)
1272                 err = f2fs_reserve_new_block(dn);
1273         if (err || need_put)
1274                 f2fs_put_dnode(dn);
1275         return err;
1276 }
1277
1278 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1279                                      blk_opf_t op_flags, bool for_write,
1280                                      pgoff_t *next_pgofs)
1281 {
1282         struct address_space *mapping = inode->i_mapping;
1283         struct dnode_of_data dn;
1284         struct page *page;
1285         int err;
1286
1287         page = f2fs_grab_cache_page(mapping, index, for_write);
1288         if (!page)
1289                 return ERR_PTR(-ENOMEM);
1290
1291         if (f2fs_lookup_read_extent_cache_block(inode, index,
1292                                                 &dn.data_blkaddr)) {
1293                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1294                                                 DATA_GENERIC_ENHANCE_READ)) {
1295                         err = -EFSCORRUPTED;
1296                         f2fs_handle_error(F2FS_I_SB(inode),
1297                                                 ERROR_INVALID_BLKADDR);
1298                         goto put_err;
1299                 }
1300                 goto got_it;
1301         }
1302
1303         set_new_dnode(&dn, inode, NULL, NULL, 0);
1304         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1305         if (err) {
1306                 if (err == -ENOENT && next_pgofs)
1307                         *next_pgofs = f2fs_get_next_page_offset(&dn, index);
1308                 goto put_err;
1309         }
1310         f2fs_put_dnode(&dn);
1311
1312         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1313                 err = -ENOENT;
1314                 if (next_pgofs)
1315                         *next_pgofs = index + 1;
1316                 goto put_err;
1317         }
1318         if (dn.data_blkaddr != NEW_ADDR &&
1319                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1320                                                 dn.data_blkaddr,
1321                                                 DATA_GENERIC_ENHANCE)) {
1322                 err = -EFSCORRUPTED;
1323                 f2fs_handle_error(F2FS_I_SB(inode),
1324                                         ERROR_INVALID_BLKADDR);
1325                 goto put_err;
1326         }
1327 got_it:
1328         if (PageUptodate(page)) {
1329                 unlock_page(page);
1330                 return page;
1331         }
1332
1333         /*
1334          * A new dentry page is allocated but not able to be written, since its
1335          * new inode page couldn't be allocated due to -ENOSPC.
1336          * In such the case, its blkaddr can be remained as NEW_ADDR.
1337          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1338          * f2fs_init_inode_metadata.
1339          */
1340         if (dn.data_blkaddr == NEW_ADDR) {
1341                 zero_user_segment(page, 0, PAGE_SIZE);
1342                 if (!PageUptodate(page))
1343                         SetPageUptodate(page);
1344                 unlock_page(page);
1345                 return page;
1346         }
1347
1348         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1349                                                 op_flags, for_write);
1350         if (err)
1351                 goto put_err;
1352         return page;
1353
1354 put_err:
1355         f2fs_put_page(page, 1);
1356         return ERR_PTR(err);
1357 }
1358
1359 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1360                                         pgoff_t *next_pgofs)
1361 {
1362         struct address_space *mapping = inode->i_mapping;
1363         struct page *page;
1364
1365         page = find_get_page(mapping, index);
1366         if (page && PageUptodate(page))
1367                 return page;
1368         f2fs_put_page(page, 0);
1369
1370         page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1371         if (IS_ERR(page))
1372                 return page;
1373
1374         if (PageUptodate(page))
1375                 return page;
1376
1377         wait_on_page_locked(page);
1378         if (unlikely(!PageUptodate(page))) {
1379                 f2fs_put_page(page, 0);
1380                 return ERR_PTR(-EIO);
1381         }
1382         return page;
1383 }
1384
1385 /*
1386  * If it tries to access a hole, return an error.
1387  * Because, the callers, functions in dir.c and GC, should be able to know
1388  * whether this page exists or not.
1389  */
1390 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1391                                                         bool for_write)
1392 {
1393         struct address_space *mapping = inode->i_mapping;
1394         struct page *page;
1395
1396         page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1397         if (IS_ERR(page))
1398                 return page;
1399
1400         /* wait for read completion */
1401         lock_page(page);
1402         if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1403                 f2fs_put_page(page, 1);
1404                 return ERR_PTR(-EIO);
1405         }
1406         return page;
1407 }
1408
1409 /*
1410  * Caller ensures that this data page is never allocated.
1411  * A new zero-filled data page is allocated in the page cache.
1412  *
1413  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1414  * f2fs_unlock_op().
1415  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1416  * ipage should be released by this function.
1417  */
1418 struct page *f2fs_get_new_data_page(struct inode *inode,
1419                 struct page *ipage, pgoff_t index, bool new_i_size)
1420 {
1421         struct address_space *mapping = inode->i_mapping;
1422         struct page *page;
1423         struct dnode_of_data dn;
1424         int err;
1425
1426         page = f2fs_grab_cache_page(mapping, index, true);
1427         if (!page) {
1428                 /*
1429                  * before exiting, we should make sure ipage will be released
1430                  * if any error occur.
1431                  */
1432                 f2fs_put_page(ipage, 1);
1433                 return ERR_PTR(-ENOMEM);
1434         }
1435
1436         set_new_dnode(&dn, inode, ipage, NULL, 0);
1437         err = f2fs_reserve_block(&dn, index);
1438         if (err) {
1439                 f2fs_put_page(page, 1);
1440                 return ERR_PTR(err);
1441         }
1442         if (!ipage)
1443                 f2fs_put_dnode(&dn);
1444
1445         if (PageUptodate(page))
1446                 goto got_it;
1447
1448         if (dn.data_blkaddr == NEW_ADDR) {
1449                 zero_user_segment(page, 0, PAGE_SIZE);
1450                 if (!PageUptodate(page))
1451                         SetPageUptodate(page);
1452         } else {
1453                 f2fs_put_page(page, 1);
1454
1455                 /* if ipage exists, blkaddr should be NEW_ADDR */
1456                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1457                 page = f2fs_get_lock_data_page(inode, index, true);
1458                 if (IS_ERR(page))
1459                         return page;
1460         }
1461 got_it:
1462         if (new_i_size && i_size_read(inode) <
1463                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1464                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1465         return page;
1466 }
1467
1468 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1469 {
1470         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1471         struct f2fs_summary sum;
1472         struct node_info ni;
1473         block_t old_blkaddr;
1474         blkcnt_t count = 1;
1475         int err;
1476
1477         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1478                 return -EPERM;
1479
1480         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1481         if (err)
1482                 return err;
1483
1484         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1485         if (dn->data_blkaddr == NULL_ADDR) {
1486                 err = inc_valid_block_count(sbi, dn->inode, &count);
1487                 if (unlikely(err))
1488                         return err;
1489         }
1490
1491         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1492         old_blkaddr = dn->data_blkaddr;
1493         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1494                                 &sum, seg_type, NULL);
1495         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1496                 invalidate_mapping_pages(META_MAPPING(sbi),
1497                                         old_blkaddr, old_blkaddr);
1498                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1499         }
1500         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1501         return 0;
1502 }
1503
1504 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1505 {
1506         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1507                 f2fs_down_read(&sbi->node_change);
1508         else
1509                 f2fs_lock_op(sbi);
1510 }
1511
1512 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1513 {
1514         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1515                 f2fs_up_read(&sbi->node_change);
1516         else
1517                 f2fs_unlock_op(sbi);
1518 }
1519
1520 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1521 {
1522         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1523         int err = 0;
1524
1525         f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1526         if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1527                                                 &dn->data_blkaddr))
1528                 err = f2fs_reserve_block(dn, index);
1529         f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1530
1531         return err;
1532 }
1533
1534 static int f2fs_map_no_dnode(struct inode *inode,
1535                 struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1536                 pgoff_t pgoff)
1537 {
1538         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1539
1540         /*
1541          * There is one exceptional case that read_node_page() may return
1542          * -ENOENT due to filesystem has been shutdown or cp_error, return
1543          * -EIO in that case.
1544          */
1545         if (map->m_may_create &&
1546             (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1547                 return -EIO;
1548
1549         if (map->m_next_pgofs)
1550                 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1551         if (map->m_next_extent)
1552                 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1553         return 0;
1554 }
1555
1556 static bool f2fs_map_blocks_cached(struct inode *inode,
1557                 struct f2fs_map_blocks *map, int flag)
1558 {
1559         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1560         unsigned int maxblocks = map->m_len;
1561         pgoff_t pgoff = (pgoff_t)map->m_lblk;
1562         struct extent_info ei = {};
1563
1564         if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1565                 return false;
1566
1567         map->m_pblk = ei.blk + pgoff - ei.fofs;
1568         map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1569         map->m_flags = F2FS_MAP_MAPPED;
1570         if (map->m_next_extent)
1571                 *map->m_next_extent = pgoff + map->m_len;
1572
1573         /* for hardware encryption, but to avoid potential issue in future */
1574         if (flag == F2FS_GET_BLOCK_DIO)
1575                 f2fs_wait_on_block_writeback_range(inode,
1576                                         map->m_pblk, map->m_len);
1577
1578         if (f2fs_allow_multi_device_dio(sbi, flag)) {
1579                 int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1580                 struct f2fs_dev_info *dev = &sbi->devs[bidx];
1581
1582                 map->m_bdev = dev->bdev;
1583                 map->m_pblk -= dev->start_blk;
1584                 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1585         } else {
1586                 map->m_bdev = inode->i_sb->s_bdev;
1587         }
1588         return true;
1589 }
1590
1591 /*
1592  * f2fs_map_blocks() tries to find or build mapping relationship which
1593  * maps continuous logical blocks to physical blocks, and return such
1594  * info via f2fs_map_blocks structure.
1595  */
1596 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1597 {
1598         unsigned int maxblocks = map->m_len;
1599         struct dnode_of_data dn;
1600         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1601         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1602         pgoff_t pgofs, end_offset, end;
1603         int err = 0, ofs = 1;
1604         unsigned int ofs_in_node, last_ofs_in_node;
1605         blkcnt_t prealloc;
1606         block_t blkaddr;
1607         unsigned int start_pgofs;
1608         int bidx = 0;
1609         bool is_hole;
1610
1611         if (!maxblocks)
1612                 return 0;
1613
1614         if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1615                 goto out;
1616
1617         map->m_bdev = inode->i_sb->s_bdev;
1618         map->m_multidev_dio =
1619                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1620
1621         map->m_len = 0;
1622         map->m_flags = 0;
1623
1624         /* it only supports block size == page size */
1625         pgofs = (pgoff_t)map->m_lblk;
1626         end = pgofs + maxblocks;
1627
1628 next_dnode:
1629         if (map->m_may_create)
1630                 f2fs_map_lock(sbi, flag);
1631
1632         /* When reading holes, we need its node page */
1633         set_new_dnode(&dn, inode, NULL, NULL, 0);
1634         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1635         if (err) {
1636                 if (flag == F2FS_GET_BLOCK_BMAP)
1637                         map->m_pblk = 0;
1638                 if (err == -ENOENT)
1639                         err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1640                 goto unlock_out;
1641         }
1642
1643         start_pgofs = pgofs;
1644         prealloc = 0;
1645         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1646         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1647
1648 next_block:
1649         blkaddr = f2fs_data_blkaddr(&dn);
1650         is_hole = !__is_valid_data_blkaddr(blkaddr);
1651         if (!is_hole &&
1652             !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1653                 err = -EFSCORRUPTED;
1654                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1655                 goto sync_out;
1656         }
1657
1658         /* use out-place-update for direct IO under LFS mode */
1659         if (map->m_may_create &&
1660             (is_hole || (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO))) {
1661                 if (unlikely(f2fs_cp_error(sbi))) {
1662                         err = -EIO;
1663                         goto sync_out;
1664                 }
1665
1666                 switch (flag) {
1667                 case F2FS_GET_BLOCK_PRE_AIO:
1668                         if (blkaddr == NULL_ADDR) {
1669                                 prealloc++;
1670                                 last_ofs_in_node = dn.ofs_in_node;
1671                         }
1672                         break;
1673                 case F2FS_GET_BLOCK_PRE_DIO:
1674                 case F2FS_GET_BLOCK_DIO:
1675                         err = __allocate_data_block(&dn, map->m_seg_type);
1676                         if (err)
1677                                 goto sync_out;
1678                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1679                                 file_need_truncate(inode);
1680                         set_inode_flag(inode, FI_APPEND_WRITE);
1681                         break;
1682                 default:
1683                         WARN_ON_ONCE(1);
1684                         err = -EIO;
1685                         goto sync_out;
1686                 }
1687
1688                 blkaddr = dn.data_blkaddr;
1689                 if (is_hole)
1690                         map->m_flags |= F2FS_MAP_NEW;
1691         } else if (is_hole) {
1692                 if (f2fs_compressed_file(inode) &&
1693                     f2fs_sanity_check_cluster(&dn) &&
1694                     (flag != F2FS_GET_BLOCK_FIEMAP ||
1695                      IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1696                         err = -EFSCORRUPTED;
1697                         f2fs_handle_error(sbi,
1698                                         ERROR_CORRUPTED_CLUSTER);
1699                         goto sync_out;
1700                 }
1701
1702                 switch (flag) {
1703                 case F2FS_GET_BLOCK_PRECACHE:
1704                         goto sync_out;
1705                 case F2FS_GET_BLOCK_BMAP:
1706                         map->m_pblk = 0;
1707                         goto sync_out;
1708                 case F2FS_GET_BLOCK_FIEMAP:
1709                         if (blkaddr == NULL_ADDR) {
1710                                 if (map->m_next_pgofs)
1711                                         *map->m_next_pgofs = pgofs + 1;
1712                                 goto sync_out;
1713                         }
1714                         break;
1715                 default:
1716                         /* for defragment case */
1717                         if (map->m_next_pgofs)
1718                                 *map->m_next_pgofs = pgofs + 1;
1719                         goto sync_out;
1720                 }
1721         }
1722
1723         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1724                 goto skip;
1725
1726         if (map->m_multidev_dio)
1727                 bidx = f2fs_target_device_index(sbi, blkaddr);
1728
1729         if (map->m_len == 0) {
1730                 /* reserved delalloc block should be mapped for fiemap. */
1731                 if (blkaddr == NEW_ADDR)
1732                         map->m_flags |= F2FS_MAP_DELALLOC;
1733                 map->m_flags |= F2FS_MAP_MAPPED;
1734
1735                 map->m_pblk = blkaddr;
1736                 map->m_len = 1;
1737
1738                 if (map->m_multidev_dio)
1739                         map->m_bdev = FDEV(bidx).bdev;
1740         } else if ((map->m_pblk != NEW_ADDR &&
1741                         blkaddr == (map->m_pblk + ofs)) ||
1742                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1743                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1744                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1745                         goto sync_out;
1746                 ofs++;
1747                 map->m_len++;
1748         } else {
1749                 goto sync_out;
1750         }
1751
1752 skip:
1753         dn.ofs_in_node++;
1754         pgofs++;
1755
1756         /* preallocate blocks in batch for one dnode page */
1757         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1758                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1759
1760                 dn.ofs_in_node = ofs_in_node;
1761                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1762                 if (err)
1763                         goto sync_out;
1764
1765                 map->m_len += dn.ofs_in_node - ofs_in_node;
1766                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1767                         err = -ENOSPC;
1768                         goto sync_out;
1769                 }
1770                 dn.ofs_in_node = end_offset;
1771         }
1772
1773         if (pgofs >= end)
1774                 goto sync_out;
1775         else if (dn.ofs_in_node < end_offset)
1776                 goto next_block;
1777
1778         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1779                 if (map->m_flags & F2FS_MAP_MAPPED) {
1780                         unsigned int ofs = start_pgofs - map->m_lblk;
1781
1782                         f2fs_update_read_extent_cache_range(&dn,
1783                                 start_pgofs, map->m_pblk + ofs,
1784                                 map->m_len - ofs);
1785                 }
1786         }
1787
1788         f2fs_put_dnode(&dn);
1789
1790         if (map->m_may_create) {
1791                 f2fs_map_unlock(sbi, flag);
1792                 f2fs_balance_fs(sbi, dn.node_changed);
1793         }
1794         goto next_dnode;
1795
1796 sync_out:
1797
1798         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1799                 /*
1800                  * for hardware encryption, but to avoid potential issue
1801                  * in future
1802                  */
1803                 f2fs_wait_on_block_writeback_range(inode,
1804                                                 map->m_pblk, map->m_len);
1805
1806                 if (map->m_multidev_dio) {
1807                         block_t blk_addr = map->m_pblk;
1808
1809                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1810
1811                         map->m_bdev = FDEV(bidx).bdev;
1812                         map->m_pblk -= FDEV(bidx).start_blk;
1813
1814                         if (map->m_may_create)
1815                                 f2fs_update_device_state(sbi, inode->i_ino,
1816                                                         blk_addr, map->m_len);
1817
1818                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1819                                                 FDEV(bidx).end_blk + 1);
1820                 }
1821         }
1822
1823         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1824                 if (map->m_flags & F2FS_MAP_MAPPED) {
1825                         unsigned int ofs = start_pgofs - map->m_lblk;
1826
1827                         f2fs_update_read_extent_cache_range(&dn,
1828                                 start_pgofs, map->m_pblk + ofs,
1829                                 map->m_len - ofs);
1830                 }
1831                 if (map->m_next_extent)
1832                         *map->m_next_extent = pgofs + 1;
1833         }
1834         f2fs_put_dnode(&dn);
1835 unlock_out:
1836         if (map->m_may_create) {
1837                 f2fs_map_unlock(sbi, flag);
1838                 f2fs_balance_fs(sbi, dn.node_changed);
1839         }
1840 out:
1841         trace_f2fs_map_blocks(inode, map, flag, err);
1842         return err;
1843 }
1844
1845 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1846 {
1847         struct f2fs_map_blocks map;
1848         block_t last_lblk;
1849         int err;
1850
1851         if (pos + len > i_size_read(inode))
1852                 return false;
1853
1854         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1855         map.m_next_pgofs = NULL;
1856         map.m_next_extent = NULL;
1857         map.m_seg_type = NO_CHECK_TYPE;
1858         map.m_may_create = false;
1859         last_lblk = F2FS_BLK_ALIGN(pos + len);
1860
1861         while (map.m_lblk < last_lblk) {
1862                 map.m_len = last_lblk - map.m_lblk;
1863                 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1864                 if (err || map.m_len == 0)
1865                         return false;
1866                 map.m_lblk += map.m_len;
1867         }
1868         return true;
1869 }
1870
1871 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1872 {
1873         return (bytes >> inode->i_blkbits);
1874 }
1875
1876 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1877 {
1878         return (blks << inode->i_blkbits);
1879 }
1880
1881 static int f2fs_xattr_fiemap(struct inode *inode,
1882                                 struct fiemap_extent_info *fieinfo)
1883 {
1884         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1885         struct page *page;
1886         struct node_info ni;
1887         __u64 phys = 0, len;
1888         __u32 flags;
1889         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1890         int err = 0;
1891
1892         if (f2fs_has_inline_xattr(inode)) {
1893                 int offset;
1894
1895                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1896                                                 inode->i_ino, false);
1897                 if (!page)
1898                         return -ENOMEM;
1899
1900                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1901                 if (err) {
1902                         f2fs_put_page(page, 1);
1903                         return err;
1904                 }
1905
1906                 phys = blks_to_bytes(inode, ni.blk_addr);
1907                 offset = offsetof(struct f2fs_inode, i_addr) +
1908                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1909                                         get_inline_xattr_addrs(inode));
1910
1911                 phys += offset;
1912                 len = inline_xattr_size(inode);
1913
1914                 f2fs_put_page(page, 1);
1915
1916                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1917
1918                 if (!xnid)
1919                         flags |= FIEMAP_EXTENT_LAST;
1920
1921                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1922                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1923                 if (err)
1924                         return err;
1925         }
1926
1927         if (xnid) {
1928                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1929                 if (!page)
1930                         return -ENOMEM;
1931
1932                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1933                 if (err) {
1934                         f2fs_put_page(page, 1);
1935                         return err;
1936                 }
1937
1938                 phys = blks_to_bytes(inode, ni.blk_addr);
1939                 len = inode->i_sb->s_blocksize;
1940
1941                 f2fs_put_page(page, 1);
1942
1943                 flags = FIEMAP_EXTENT_LAST;
1944         }
1945
1946         if (phys) {
1947                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1948                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1949         }
1950
1951         return (err < 0 ? err : 0);
1952 }
1953
1954 static loff_t max_inode_blocks(struct inode *inode)
1955 {
1956         loff_t result = ADDRS_PER_INODE(inode);
1957         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1958
1959         /* two direct node blocks */
1960         result += (leaf_count * 2);
1961
1962         /* two indirect node blocks */
1963         leaf_count *= NIDS_PER_BLOCK;
1964         result += (leaf_count * 2);
1965
1966         /* one double indirect node block */
1967         leaf_count *= NIDS_PER_BLOCK;
1968         result += leaf_count;
1969
1970         return result;
1971 }
1972
1973 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1974                 u64 start, u64 len)
1975 {
1976         struct f2fs_map_blocks map;
1977         sector_t start_blk, last_blk;
1978         pgoff_t next_pgofs;
1979         u64 logical = 0, phys = 0, size = 0;
1980         u32 flags = 0;
1981         int ret = 0;
1982         bool compr_cluster = false, compr_appended;
1983         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1984         unsigned int count_in_cluster = 0;
1985         loff_t maxbytes;
1986
1987         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1988                 ret = f2fs_precache_extents(inode);
1989                 if (ret)
1990                         return ret;
1991         }
1992
1993         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1994         if (ret)
1995                 return ret;
1996
1997         inode_lock(inode);
1998
1999         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
2000         if (start > maxbytes) {
2001                 ret = -EFBIG;
2002                 goto out;
2003         }
2004
2005         if (len > maxbytes || (maxbytes - len) < start)
2006                 len = maxbytes - start;
2007
2008         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
2009                 ret = f2fs_xattr_fiemap(inode, fieinfo);
2010                 goto out;
2011         }
2012
2013         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
2014                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
2015                 if (ret != -EAGAIN)
2016                         goto out;
2017         }
2018
2019         if (bytes_to_blks(inode, len) == 0)
2020                 len = blks_to_bytes(inode, 1);
2021
2022         start_blk = bytes_to_blks(inode, start);
2023         last_blk = bytes_to_blks(inode, start + len - 1);
2024
2025 next:
2026         memset(&map, 0, sizeof(map));
2027         map.m_lblk = start_blk;
2028         map.m_len = bytes_to_blks(inode, len);
2029         map.m_next_pgofs = &next_pgofs;
2030         map.m_seg_type = NO_CHECK_TYPE;
2031
2032         if (compr_cluster) {
2033                 map.m_lblk += 1;
2034                 map.m_len = cluster_size - count_in_cluster;
2035         }
2036
2037         ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
2038         if (ret)
2039                 goto out;
2040
2041         /* HOLE */
2042         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
2043                 start_blk = next_pgofs;
2044
2045                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
2046                                                 max_inode_blocks(inode)))
2047                         goto prep_next;
2048
2049                 flags |= FIEMAP_EXTENT_LAST;
2050         }
2051
2052         compr_appended = false;
2053         /* In a case of compressed cluster, append this to the last extent */
2054         if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
2055                         !(map.m_flags & F2FS_MAP_FLAGS))) {
2056                 compr_appended = true;
2057                 goto skip_fill;
2058         }
2059
2060         if (size) {
2061                 flags |= FIEMAP_EXTENT_MERGED;
2062                 if (IS_ENCRYPTED(inode))
2063                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2064
2065                 ret = fiemap_fill_next_extent(fieinfo, logical,
2066                                 phys, size, flags);
2067                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2068                 if (ret)
2069                         goto out;
2070                 size = 0;
2071         }
2072
2073         if (start_blk > last_blk)
2074                 goto out;
2075
2076 skip_fill:
2077         if (map.m_pblk == COMPRESS_ADDR) {
2078                 compr_cluster = true;
2079                 count_in_cluster = 1;
2080         } else if (compr_appended) {
2081                 unsigned int appended_blks = cluster_size -
2082                                                 count_in_cluster + 1;
2083                 size += blks_to_bytes(inode, appended_blks);
2084                 start_blk += appended_blks;
2085                 compr_cluster = false;
2086         } else {
2087                 logical = blks_to_bytes(inode, start_blk);
2088                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2089                         blks_to_bytes(inode, map.m_pblk) : 0;
2090                 size = blks_to_bytes(inode, map.m_len);
2091                 flags = 0;
2092
2093                 if (compr_cluster) {
2094                         flags = FIEMAP_EXTENT_ENCODED;
2095                         count_in_cluster += map.m_len;
2096                         if (count_in_cluster == cluster_size) {
2097                                 compr_cluster = false;
2098                                 size += blks_to_bytes(inode, 1);
2099                         }
2100                 } else if (map.m_flags & F2FS_MAP_DELALLOC) {
2101                         flags = FIEMAP_EXTENT_UNWRITTEN;
2102                 }
2103
2104                 start_blk += bytes_to_blks(inode, size);
2105         }
2106
2107 prep_next:
2108         cond_resched();
2109         if (fatal_signal_pending(current))
2110                 ret = -EINTR;
2111         else
2112                 goto next;
2113 out:
2114         if (ret == 1)
2115                 ret = 0;
2116
2117         inode_unlock(inode);
2118         return ret;
2119 }
2120
2121 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2122 {
2123         if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2124                 return inode->i_sb->s_maxbytes;
2125
2126         return i_size_read(inode);
2127 }
2128
2129 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2130                                         unsigned nr_pages,
2131                                         struct f2fs_map_blocks *map,
2132                                         struct bio **bio_ret,
2133                                         sector_t *last_block_in_bio,
2134                                         bool is_readahead)
2135 {
2136         struct bio *bio = *bio_ret;
2137         const unsigned blocksize = blks_to_bytes(inode, 1);
2138         sector_t block_in_file;
2139         sector_t last_block;
2140         sector_t last_block_in_file;
2141         sector_t block_nr;
2142         int ret = 0;
2143
2144         block_in_file = (sector_t)page_index(page);
2145         last_block = block_in_file + nr_pages;
2146         last_block_in_file = bytes_to_blks(inode,
2147                         f2fs_readpage_limit(inode) + blocksize - 1);
2148         if (last_block > last_block_in_file)
2149                 last_block = last_block_in_file;
2150
2151         /* just zeroing out page which is beyond EOF */
2152         if (block_in_file >= last_block)
2153                 goto zero_out;
2154         /*
2155          * Map blocks using the previous result first.
2156          */
2157         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2158                         block_in_file > map->m_lblk &&
2159                         block_in_file < (map->m_lblk + map->m_len))
2160                 goto got_it;
2161
2162         /*
2163          * Then do more f2fs_map_blocks() calls until we are
2164          * done with this page.
2165          */
2166         map->m_lblk = block_in_file;
2167         map->m_len = last_block - block_in_file;
2168
2169         ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2170         if (ret)
2171                 goto out;
2172 got_it:
2173         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2174                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2175                 SetPageMappedToDisk(page);
2176
2177                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2178                                                 DATA_GENERIC_ENHANCE_READ)) {
2179                         ret = -EFSCORRUPTED;
2180                         f2fs_handle_error(F2FS_I_SB(inode),
2181                                                 ERROR_INVALID_BLKADDR);
2182                         goto out;
2183                 }
2184         } else {
2185 zero_out:
2186                 zero_user_segment(page, 0, PAGE_SIZE);
2187                 if (f2fs_need_verity(inode, page->index) &&
2188                     !fsverity_verify_page(page)) {
2189                         ret = -EIO;
2190                         goto out;
2191                 }
2192                 if (!PageUptodate(page))
2193                         SetPageUptodate(page);
2194                 unlock_page(page);
2195                 goto out;
2196         }
2197
2198         /*
2199          * This page will go to BIO.  Do we need to send this
2200          * BIO off first?
2201          */
2202         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2203                                        *last_block_in_bio, block_nr) ||
2204                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2205 submit_and_realloc:
2206                 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2207                 bio = NULL;
2208         }
2209         if (bio == NULL) {
2210                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2211                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2212                                 false);
2213                 if (IS_ERR(bio)) {
2214                         ret = PTR_ERR(bio);
2215                         bio = NULL;
2216                         goto out;
2217                 }
2218         }
2219
2220         /*
2221          * If the page is under writeback, we need to wait for
2222          * its completion to see the correct decrypted data.
2223          */
2224         f2fs_wait_on_block_writeback(inode, block_nr);
2225
2226         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2227                 goto submit_and_realloc;
2228
2229         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2230         f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2231                                                         F2FS_BLKSIZE);
2232         *last_block_in_bio = block_nr;
2233 out:
2234         *bio_ret = bio;
2235         return ret;
2236 }
2237
2238 #ifdef CONFIG_F2FS_FS_COMPRESSION
2239 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2240                                 unsigned nr_pages, sector_t *last_block_in_bio,
2241                                 bool is_readahead, bool for_write)
2242 {
2243         struct dnode_of_data dn;
2244         struct inode *inode = cc->inode;
2245         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2246         struct bio *bio = *bio_ret;
2247         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2248         sector_t last_block_in_file;
2249         const unsigned blocksize = blks_to_bytes(inode, 1);
2250         struct decompress_io_ctx *dic = NULL;
2251         struct extent_info ei = {};
2252         bool from_dnode = true;
2253         int i;
2254         int ret = 0;
2255
2256         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2257
2258         last_block_in_file = bytes_to_blks(inode,
2259                         f2fs_readpage_limit(inode) + blocksize - 1);
2260
2261         /* get rid of pages beyond EOF */
2262         for (i = 0; i < cc->cluster_size; i++) {
2263                 struct page *page = cc->rpages[i];
2264
2265                 if (!page)
2266                         continue;
2267                 if ((sector_t)page->index >= last_block_in_file) {
2268                         zero_user_segment(page, 0, PAGE_SIZE);
2269                         if (!PageUptodate(page))
2270                                 SetPageUptodate(page);
2271                 } else if (!PageUptodate(page)) {
2272                         continue;
2273                 }
2274                 unlock_page(page);
2275                 if (for_write)
2276                         put_page(page);
2277                 cc->rpages[i] = NULL;
2278                 cc->nr_rpages--;
2279         }
2280
2281         /* we are done since all pages are beyond EOF */
2282         if (f2fs_cluster_is_empty(cc))
2283                 goto out;
2284
2285         if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2286                 from_dnode = false;
2287
2288         if (!from_dnode)
2289                 goto skip_reading_dnode;
2290
2291         set_new_dnode(&dn, inode, NULL, NULL, 0);
2292         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2293         if (ret)
2294                 goto out;
2295
2296         if (unlikely(f2fs_cp_error(sbi))) {
2297                 ret = -EIO;
2298                 goto out_put_dnode;
2299         }
2300         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2301
2302 skip_reading_dnode:
2303         for (i = 1; i < cc->cluster_size; i++) {
2304                 block_t blkaddr;
2305
2306                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2307                                         dn.ofs_in_node + i) :
2308                                         ei.blk + i - 1;
2309
2310                 if (!__is_valid_data_blkaddr(blkaddr))
2311                         break;
2312
2313                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2314                         ret = -EFAULT;
2315                         goto out_put_dnode;
2316                 }
2317                 cc->nr_cpages++;
2318
2319                 if (!from_dnode && i >= ei.c_len)
2320                         break;
2321         }
2322
2323         /* nothing to decompress */
2324         if (cc->nr_cpages == 0) {
2325                 ret = 0;
2326                 goto out_put_dnode;
2327         }
2328
2329         dic = f2fs_alloc_dic(cc);
2330         if (IS_ERR(dic)) {
2331                 ret = PTR_ERR(dic);
2332                 goto out_put_dnode;
2333         }
2334
2335         for (i = 0; i < cc->nr_cpages; i++) {
2336                 struct page *page = dic->cpages[i];
2337                 block_t blkaddr;
2338                 struct bio_post_read_ctx *ctx;
2339
2340                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2341                                         dn.ofs_in_node + i + 1) :
2342                                         ei.blk + i;
2343
2344                 f2fs_wait_on_block_writeback(inode, blkaddr);
2345
2346                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2347                         if (atomic_dec_and_test(&dic->remaining_pages))
2348                                 f2fs_decompress_cluster(dic, true);
2349                         continue;
2350                 }
2351
2352                 if (bio && (!page_is_mergeable(sbi, bio,
2353                                         *last_block_in_bio, blkaddr) ||
2354                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2355 submit_and_realloc:
2356                         f2fs_submit_read_bio(sbi, bio, DATA);
2357                         bio = NULL;
2358                 }
2359
2360                 if (!bio) {
2361                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2362                                         is_readahead ? REQ_RAHEAD : 0,
2363                                         page->index, for_write);
2364                         if (IS_ERR(bio)) {
2365                                 ret = PTR_ERR(bio);
2366                                 f2fs_decompress_end_io(dic, ret, true);
2367                                 f2fs_put_dnode(&dn);
2368                                 *bio_ret = NULL;
2369                                 return ret;
2370                         }
2371                 }
2372
2373                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2374                         goto submit_and_realloc;
2375
2376                 ctx = get_post_read_ctx(bio);
2377                 ctx->enabled_steps |= STEP_DECOMPRESS;
2378                 refcount_inc(&dic->refcnt);
2379
2380                 inc_page_count(sbi, F2FS_RD_DATA);
2381                 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2382                 *last_block_in_bio = blkaddr;
2383         }
2384
2385         if (from_dnode)
2386                 f2fs_put_dnode(&dn);
2387
2388         *bio_ret = bio;
2389         return 0;
2390
2391 out_put_dnode:
2392         if (from_dnode)
2393                 f2fs_put_dnode(&dn);
2394 out:
2395         for (i = 0; i < cc->cluster_size; i++) {
2396                 if (cc->rpages[i]) {
2397                         ClearPageUptodate(cc->rpages[i]);
2398                         unlock_page(cc->rpages[i]);
2399                 }
2400         }
2401         *bio_ret = bio;
2402         return ret;
2403 }
2404 #endif
2405
2406 /*
2407  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2408  * Major change was from block_size == page_size in f2fs by default.
2409  */
2410 static int f2fs_mpage_readpages(struct inode *inode,
2411                 struct readahead_control *rac, struct page *page)
2412 {
2413         struct bio *bio = NULL;
2414         sector_t last_block_in_bio = 0;
2415         struct f2fs_map_blocks map;
2416 #ifdef CONFIG_F2FS_FS_COMPRESSION
2417         struct compress_ctx cc = {
2418                 .inode = inode,
2419                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2420                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2421                 .cluster_idx = NULL_CLUSTER,
2422                 .rpages = NULL,
2423                 .cpages = NULL,
2424                 .nr_rpages = 0,
2425                 .nr_cpages = 0,
2426         };
2427         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2428 #endif
2429         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2430         unsigned max_nr_pages = nr_pages;
2431         int ret = 0;
2432
2433         map.m_pblk = 0;
2434         map.m_lblk = 0;
2435         map.m_len = 0;
2436         map.m_flags = 0;
2437         map.m_next_pgofs = NULL;
2438         map.m_next_extent = NULL;
2439         map.m_seg_type = NO_CHECK_TYPE;
2440         map.m_may_create = false;
2441
2442         for (; nr_pages; nr_pages--) {
2443                 if (rac) {
2444                         page = readahead_page(rac);
2445                         prefetchw(&page->flags);
2446                 }
2447
2448 #ifdef CONFIG_F2FS_FS_COMPRESSION
2449                 if (f2fs_compressed_file(inode)) {
2450                         /* there are remained compressed pages, submit them */
2451                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2452                                 ret = f2fs_read_multi_pages(&cc, &bio,
2453                                                         max_nr_pages,
2454                                                         &last_block_in_bio,
2455                                                         rac != NULL, false);
2456                                 f2fs_destroy_compress_ctx(&cc, false);
2457                                 if (ret)
2458                                         goto set_error_page;
2459                         }
2460                         if (cc.cluster_idx == NULL_CLUSTER) {
2461                                 if (nc_cluster_idx ==
2462                                         page->index >> cc.log_cluster_size) {
2463                                         goto read_single_page;
2464                                 }
2465
2466                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2467                                 if (ret < 0)
2468                                         goto set_error_page;
2469                                 else if (!ret) {
2470                                         nc_cluster_idx =
2471                                                 page->index >> cc.log_cluster_size;
2472                                         goto read_single_page;
2473                                 }
2474
2475                                 nc_cluster_idx = NULL_CLUSTER;
2476                         }
2477                         ret = f2fs_init_compress_ctx(&cc);
2478                         if (ret)
2479                                 goto set_error_page;
2480
2481                         f2fs_compress_ctx_add_page(&cc, page);
2482
2483                         goto next_page;
2484                 }
2485 read_single_page:
2486 #endif
2487
2488                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2489                                         &bio, &last_block_in_bio, rac);
2490                 if (ret) {
2491 #ifdef CONFIG_F2FS_FS_COMPRESSION
2492 set_error_page:
2493 #endif
2494                         zero_user_segment(page, 0, PAGE_SIZE);
2495                         unlock_page(page);
2496                 }
2497 #ifdef CONFIG_F2FS_FS_COMPRESSION
2498 next_page:
2499 #endif
2500                 if (rac)
2501                         put_page(page);
2502
2503 #ifdef CONFIG_F2FS_FS_COMPRESSION
2504                 if (f2fs_compressed_file(inode)) {
2505                         /* last page */
2506                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2507                                 ret = f2fs_read_multi_pages(&cc, &bio,
2508                                                         max_nr_pages,
2509                                                         &last_block_in_bio,
2510                                                         rac != NULL, false);
2511                                 f2fs_destroy_compress_ctx(&cc, false);
2512                         }
2513                 }
2514 #endif
2515         }
2516         if (bio)
2517                 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2518         return ret;
2519 }
2520
2521 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2522 {
2523         struct page *page = &folio->page;
2524         struct inode *inode = page_file_mapping(page)->host;
2525         int ret = -EAGAIN;
2526
2527         trace_f2fs_readpage(page, DATA);
2528
2529         if (!f2fs_is_compress_backend_ready(inode)) {
2530                 unlock_page(page);
2531                 return -EOPNOTSUPP;
2532         }
2533
2534         /* If the file has inline data, try to read it directly */
2535         if (f2fs_has_inline_data(inode))
2536                 ret = f2fs_read_inline_data(inode, page);
2537         if (ret == -EAGAIN)
2538                 ret = f2fs_mpage_readpages(inode, NULL, page);
2539         return ret;
2540 }
2541
2542 static void f2fs_readahead(struct readahead_control *rac)
2543 {
2544         struct inode *inode = rac->mapping->host;
2545
2546         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2547
2548         if (!f2fs_is_compress_backend_ready(inode))
2549                 return;
2550
2551         /* If the file has inline data, skip readahead */
2552         if (f2fs_has_inline_data(inode))
2553                 return;
2554
2555         f2fs_mpage_readpages(inode, rac, NULL);
2556 }
2557
2558 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2559 {
2560         struct inode *inode = fio->page->mapping->host;
2561         struct page *mpage, *page;
2562         gfp_t gfp_flags = GFP_NOFS;
2563
2564         if (!f2fs_encrypted_file(inode))
2565                 return 0;
2566
2567         page = fio->compressed_page ? fio->compressed_page : fio->page;
2568
2569         /* wait for GCed page writeback via META_MAPPING */
2570         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2571
2572         if (fscrypt_inode_uses_inline_crypto(inode))
2573                 return 0;
2574
2575 retry_encrypt:
2576         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2577                                         PAGE_SIZE, 0, gfp_flags);
2578         if (IS_ERR(fio->encrypted_page)) {
2579                 /* flush pending IOs and wait for a while in the ENOMEM case */
2580                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2581                         f2fs_flush_merged_writes(fio->sbi);
2582                         memalloc_retry_wait(GFP_NOFS);
2583                         gfp_flags |= __GFP_NOFAIL;
2584                         goto retry_encrypt;
2585                 }
2586                 return PTR_ERR(fio->encrypted_page);
2587         }
2588
2589         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2590         if (mpage) {
2591                 if (PageUptodate(mpage))
2592                         memcpy(page_address(mpage),
2593                                 page_address(fio->encrypted_page), PAGE_SIZE);
2594                 f2fs_put_page(mpage, 1);
2595         }
2596         return 0;
2597 }
2598
2599 static inline bool check_inplace_update_policy(struct inode *inode,
2600                                 struct f2fs_io_info *fio)
2601 {
2602         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2603
2604         if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2605             is_inode_flag_set(inode, FI_OPU_WRITE))
2606                 return false;
2607         if (IS_F2FS_IPU_FORCE(sbi))
2608                 return true;
2609         if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2610                 return true;
2611         if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2612                 return true;
2613         if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2614             utilization(sbi) > SM_I(sbi)->min_ipu_util)
2615                 return true;
2616
2617         /*
2618          * IPU for rewrite async pages
2619          */
2620         if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2621             !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2622                 return true;
2623
2624         /* this is only set during fdatasync */
2625         if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2626                 return true;
2627
2628         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2629                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2630                 return true;
2631
2632         return false;
2633 }
2634
2635 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2636 {
2637         /* swap file is migrating in aligned write mode */
2638         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2639                 return false;
2640
2641         if (f2fs_is_pinned_file(inode))
2642                 return true;
2643
2644         /* if this is cold file, we should overwrite to avoid fragmentation */
2645         if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2646                 return true;
2647
2648         return check_inplace_update_policy(inode, fio);
2649 }
2650
2651 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2652 {
2653         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2654
2655         /* The below cases were checked when setting it. */
2656         if (f2fs_is_pinned_file(inode))
2657                 return false;
2658         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2659                 return true;
2660         if (f2fs_lfs_mode(sbi))
2661                 return true;
2662         if (S_ISDIR(inode->i_mode))
2663                 return true;
2664         if (IS_NOQUOTA(inode))
2665                 return true;
2666         if (f2fs_is_atomic_file(inode))
2667                 return true;
2668
2669         /* swap file is migrating in aligned write mode */
2670         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2671                 return true;
2672
2673         if (is_inode_flag_set(inode, FI_OPU_WRITE))
2674                 return true;
2675
2676         if (fio) {
2677                 if (page_private_gcing(fio->page))
2678                         return true;
2679                 if (page_private_dummy(fio->page))
2680                         return true;
2681                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2682                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2683                         return true;
2684         }
2685         return false;
2686 }
2687
2688 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2689 {
2690         struct inode *inode = fio->page->mapping->host;
2691
2692         if (f2fs_should_update_outplace(inode, fio))
2693                 return false;
2694
2695         return f2fs_should_update_inplace(inode, fio);
2696 }
2697
2698 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2699 {
2700         struct page *page = fio->page;
2701         struct inode *inode = page->mapping->host;
2702         struct dnode_of_data dn;
2703         struct node_info ni;
2704         bool ipu_force = false;
2705         int err = 0;
2706
2707         /* Use COW inode to make dnode_of_data for atomic write */
2708         if (f2fs_is_atomic_file(inode))
2709                 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2710         else
2711                 set_new_dnode(&dn, inode, NULL, NULL, 0);
2712
2713         if (need_inplace_update(fio) &&
2714             f2fs_lookup_read_extent_cache_block(inode, page->index,
2715                                                 &fio->old_blkaddr)) {
2716                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2717                                                 DATA_GENERIC_ENHANCE)) {
2718                         f2fs_handle_error(fio->sbi,
2719                                                 ERROR_INVALID_BLKADDR);
2720                         return -EFSCORRUPTED;
2721                 }
2722
2723                 ipu_force = true;
2724                 fio->need_lock = LOCK_DONE;
2725                 goto got_it;
2726         }
2727
2728         /* Deadlock due to between page->lock and f2fs_lock_op */
2729         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2730                 return -EAGAIN;
2731
2732         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2733         if (err)
2734                 goto out;
2735
2736         fio->old_blkaddr = dn.data_blkaddr;
2737
2738         /* This page is already truncated */
2739         if (fio->old_blkaddr == NULL_ADDR) {
2740                 ClearPageUptodate(page);
2741                 clear_page_private_gcing(page);
2742                 goto out_writepage;
2743         }
2744 got_it:
2745         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2746                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2747                                                 DATA_GENERIC_ENHANCE)) {
2748                 err = -EFSCORRUPTED;
2749                 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
2750                 goto out_writepage;
2751         }
2752
2753         /*
2754          * If current allocation needs SSR,
2755          * it had better in-place writes for updated data.
2756          */
2757         if (ipu_force ||
2758                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2759                                         need_inplace_update(fio))) {
2760                 err = f2fs_encrypt_one_page(fio);
2761                 if (err)
2762                         goto out_writepage;
2763
2764                 set_page_writeback(page);
2765                 f2fs_put_dnode(&dn);
2766                 if (fio->need_lock == LOCK_REQ)
2767                         f2fs_unlock_op(fio->sbi);
2768                 err = f2fs_inplace_write_data(fio);
2769                 if (err) {
2770                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2771                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2772                         if (PageWriteback(page))
2773                                 end_page_writeback(page);
2774                 } else {
2775                         set_inode_flag(inode, FI_UPDATE_WRITE);
2776                 }
2777                 trace_f2fs_do_write_data_page(fio->page, IPU);
2778                 return err;
2779         }
2780
2781         if (fio->need_lock == LOCK_RETRY) {
2782                 if (!f2fs_trylock_op(fio->sbi)) {
2783                         err = -EAGAIN;
2784                         goto out_writepage;
2785                 }
2786                 fio->need_lock = LOCK_REQ;
2787         }
2788
2789         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2790         if (err)
2791                 goto out_writepage;
2792
2793         fio->version = ni.version;
2794
2795         err = f2fs_encrypt_one_page(fio);
2796         if (err)
2797                 goto out_writepage;
2798
2799         set_page_writeback(page);
2800
2801         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2802                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2803
2804         /* LFS mode write path */
2805         f2fs_outplace_write_data(&dn, fio);
2806         trace_f2fs_do_write_data_page(page, OPU);
2807         set_inode_flag(inode, FI_APPEND_WRITE);
2808         if (page->index == 0)
2809                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2810 out_writepage:
2811         f2fs_put_dnode(&dn);
2812 out:
2813         if (fio->need_lock == LOCK_REQ)
2814                 f2fs_unlock_op(fio->sbi);
2815         return err;
2816 }
2817
2818 int f2fs_write_single_data_page(struct page *page, int *submitted,
2819                                 struct bio **bio,
2820                                 sector_t *last_block,
2821                                 struct writeback_control *wbc,
2822                                 enum iostat_type io_type,
2823                                 int compr_blocks,
2824                                 bool allow_balance)
2825 {
2826         struct inode *inode = page->mapping->host;
2827         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2828         loff_t i_size = i_size_read(inode);
2829         const pgoff_t end_index = ((unsigned long long)i_size)
2830                                                         >> PAGE_SHIFT;
2831         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2832         unsigned offset = 0;
2833         bool need_balance_fs = false;
2834         bool quota_inode = IS_NOQUOTA(inode);
2835         int err = 0;
2836         struct f2fs_io_info fio = {
2837                 .sbi = sbi,
2838                 .ino = inode->i_ino,
2839                 .type = DATA,
2840                 .op = REQ_OP_WRITE,
2841                 .op_flags = wbc_to_write_flags(wbc),
2842                 .old_blkaddr = NULL_ADDR,
2843                 .page = page,
2844                 .encrypted_page = NULL,
2845                 .submitted = 0,
2846                 .compr_blocks = compr_blocks,
2847                 .need_lock = LOCK_RETRY,
2848                 .post_read = f2fs_post_read_required(inode) ? 1 : 0,
2849                 .io_type = io_type,
2850                 .io_wbc = wbc,
2851                 .bio = bio,
2852                 .last_block = last_block,
2853         };
2854
2855         trace_f2fs_writepage(page, DATA);
2856
2857         /* we should bypass data pages to proceed the kworker jobs */
2858         if (unlikely(f2fs_cp_error(sbi))) {
2859                 mapping_set_error(page->mapping, -EIO);
2860                 /*
2861                  * don't drop any dirty dentry pages for keeping lastest
2862                  * directory structure.
2863                  */
2864                 if (S_ISDIR(inode->i_mode) &&
2865                                 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2866                         goto redirty_out;
2867
2868                 /* keep data pages in remount-ro mode */
2869                 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2870                         goto redirty_out;
2871                 goto out;
2872         }
2873
2874         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2875                 goto redirty_out;
2876
2877         if (page->index < end_index ||
2878                         f2fs_verity_in_progress(inode) ||
2879                         compr_blocks)
2880                 goto write;
2881
2882         /*
2883          * If the offset is out-of-range of file size,
2884          * this page does not have to be written to disk.
2885          */
2886         offset = i_size & (PAGE_SIZE - 1);
2887         if ((page->index >= end_index + 1) || !offset)
2888                 goto out;
2889
2890         zero_user_segment(page, offset, PAGE_SIZE);
2891 write:
2892         if (f2fs_is_drop_cache(inode))
2893                 goto out;
2894
2895         /* Dentry/quota blocks are controlled by checkpoint */
2896         if (S_ISDIR(inode->i_mode) || quota_inode) {
2897                 /*
2898                  * We need to wait for node_write to avoid block allocation during
2899                  * checkpoint. This can only happen to quota writes which can cause
2900                  * the below discard race condition.
2901                  */
2902                 if (quota_inode)
2903                         f2fs_down_read(&sbi->node_write);
2904
2905                 fio.need_lock = LOCK_DONE;
2906                 err = f2fs_do_write_data_page(&fio);
2907
2908                 if (quota_inode)
2909                         f2fs_up_read(&sbi->node_write);
2910
2911                 goto done;
2912         }
2913
2914         if (!wbc->for_reclaim)
2915                 need_balance_fs = true;
2916         else if (has_not_enough_free_secs(sbi, 0, 0))
2917                 goto redirty_out;
2918         else
2919                 set_inode_flag(inode, FI_HOT_DATA);
2920
2921         err = -EAGAIN;
2922         if (f2fs_has_inline_data(inode)) {
2923                 err = f2fs_write_inline_data(inode, page);
2924                 if (!err)
2925                         goto out;
2926         }
2927
2928         if (err == -EAGAIN) {
2929                 err = f2fs_do_write_data_page(&fio);
2930                 if (err == -EAGAIN) {
2931                         fio.need_lock = LOCK_REQ;
2932                         err = f2fs_do_write_data_page(&fio);
2933                 }
2934         }
2935
2936         if (err) {
2937                 file_set_keep_isize(inode);
2938         } else {
2939                 spin_lock(&F2FS_I(inode)->i_size_lock);
2940                 if (F2FS_I(inode)->last_disk_size < psize)
2941                         F2FS_I(inode)->last_disk_size = psize;
2942                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2943         }
2944
2945 done:
2946         if (err && err != -ENOENT)
2947                 goto redirty_out;
2948
2949 out:
2950         inode_dec_dirty_pages(inode);
2951         if (err) {
2952                 ClearPageUptodate(page);
2953                 clear_page_private_gcing(page);
2954         }
2955
2956         if (wbc->for_reclaim) {
2957                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2958                 clear_inode_flag(inode, FI_HOT_DATA);
2959                 f2fs_remove_dirty_inode(inode);
2960                 submitted = NULL;
2961         }
2962         unlock_page(page);
2963         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2964                         !F2FS_I(inode)->wb_task && allow_balance)
2965                 f2fs_balance_fs(sbi, need_balance_fs);
2966
2967         if (unlikely(f2fs_cp_error(sbi))) {
2968                 f2fs_submit_merged_write(sbi, DATA);
2969                 if (bio && *bio)
2970                         f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2971                 submitted = NULL;
2972         }
2973
2974         if (submitted)
2975                 *submitted = fio.submitted;
2976
2977         return 0;
2978
2979 redirty_out:
2980         redirty_page_for_writepage(wbc, page);
2981         /*
2982          * pageout() in MM translates EAGAIN, so calls handle_write_error()
2983          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2984          * file_write_and_wait_range() will see EIO error, which is critical
2985          * to return value of fsync() followed by atomic_write failure to user.
2986          */
2987         if (!err || wbc->for_reclaim)
2988                 return AOP_WRITEPAGE_ACTIVATE;
2989         unlock_page(page);
2990         return err;
2991 }
2992
2993 static int f2fs_write_data_page(struct page *page,
2994                                         struct writeback_control *wbc)
2995 {
2996 #ifdef CONFIG_F2FS_FS_COMPRESSION
2997         struct inode *inode = page->mapping->host;
2998
2999         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
3000                 goto out;
3001
3002         if (f2fs_compressed_file(inode)) {
3003                 if (f2fs_is_compressed_cluster(inode, page->index)) {
3004                         redirty_page_for_writepage(wbc, page);
3005                         return AOP_WRITEPAGE_ACTIVATE;
3006                 }
3007         }
3008 out:
3009 #endif
3010
3011         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
3012                                                 wbc, FS_DATA_IO, 0, true);
3013 }
3014
3015 /*
3016  * This function was copied from write_cache_pages from mm/page-writeback.c.
3017  * The major change is making write step of cold data page separately from
3018  * warm/hot data page.
3019  */
3020 static int f2fs_write_cache_pages(struct address_space *mapping,
3021                                         struct writeback_control *wbc,
3022                                         enum iostat_type io_type)
3023 {
3024         int ret = 0;
3025         int done = 0, retry = 0;
3026         struct page *pages[F2FS_ONSTACK_PAGES];
3027         struct folio_batch fbatch;
3028         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
3029         struct bio *bio = NULL;
3030         sector_t last_block;
3031 #ifdef CONFIG_F2FS_FS_COMPRESSION
3032         struct inode *inode = mapping->host;
3033         struct compress_ctx cc = {
3034                 .inode = inode,
3035                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
3036                 .cluster_size = F2FS_I(inode)->i_cluster_size,
3037                 .cluster_idx = NULL_CLUSTER,
3038                 .rpages = NULL,
3039                 .nr_rpages = 0,
3040                 .cpages = NULL,
3041                 .valid_nr_cpages = 0,
3042                 .rbuf = NULL,
3043                 .cbuf = NULL,
3044                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
3045                 .private = NULL,
3046         };
3047 #endif
3048         int nr_folios, p, idx;
3049         int nr_pages;
3050         pgoff_t index;
3051         pgoff_t end;            /* Inclusive */
3052         pgoff_t done_index;
3053         int range_whole = 0;
3054         xa_mark_t tag;
3055         int nwritten = 0;
3056         int submitted = 0;
3057         int i;
3058
3059         folio_batch_init(&fbatch);
3060
3061         if (get_dirty_pages(mapping->host) <=
3062                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3063                 set_inode_flag(mapping->host, FI_HOT_DATA);
3064         else
3065                 clear_inode_flag(mapping->host, FI_HOT_DATA);
3066
3067         if (wbc->range_cyclic) {
3068                 index = mapping->writeback_index; /* prev offset */
3069                 end = -1;
3070         } else {
3071                 index = wbc->range_start >> PAGE_SHIFT;
3072                 end = wbc->range_end >> PAGE_SHIFT;
3073                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3074                         range_whole = 1;
3075         }
3076         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3077                 tag = PAGECACHE_TAG_TOWRITE;
3078         else
3079                 tag = PAGECACHE_TAG_DIRTY;
3080 retry:
3081         retry = 0;
3082         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3083                 tag_pages_for_writeback(mapping, index, end);
3084         done_index = index;
3085         while (!done && !retry && (index <= end)) {
3086                 nr_pages = 0;
3087 again:
3088                 nr_folios = filemap_get_folios_tag(mapping, &index, end,
3089                                 tag, &fbatch);
3090                 if (nr_folios == 0) {
3091                         if (nr_pages)
3092                                 goto write;
3093                         break;
3094                 }
3095
3096                 for (i = 0; i < nr_folios; i++) {
3097                         struct folio *folio = fbatch.folios[i];
3098
3099                         idx = 0;
3100                         p = folio_nr_pages(folio);
3101 add_more:
3102                         pages[nr_pages] = folio_page(folio, idx);
3103                         folio_get(folio);
3104                         if (++nr_pages == F2FS_ONSTACK_PAGES) {
3105                                 index = folio->index + idx + 1;
3106                                 folio_batch_release(&fbatch);
3107                                 goto write;
3108                         }
3109                         if (++idx < p)
3110                                 goto add_more;
3111                 }
3112                 folio_batch_release(&fbatch);
3113                 goto again;
3114 write:
3115                 for (i = 0; i < nr_pages; i++) {
3116                         struct page *page = pages[i];
3117                         struct folio *folio = page_folio(page);
3118                         bool need_readd;
3119 readd:
3120                         need_readd = false;
3121 #ifdef CONFIG_F2FS_FS_COMPRESSION
3122                         if (f2fs_compressed_file(inode)) {
3123                                 void *fsdata = NULL;
3124                                 struct page *pagep;
3125                                 int ret2;
3126
3127                                 ret = f2fs_init_compress_ctx(&cc);
3128                                 if (ret) {
3129                                         done = 1;
3130                                         break;
3131                                 }
3132
3133                                 if (!f2fs_cluster_can_merge_page(&cc,
3134                                                                 folio->index)) {
3135                                         ret = f2fs_write_multi_pages(&cc,
3136                                                 &submitted, wbc, io_type);
3137                                         if (!ret)
3138                                                 need_readd = true;
3139                                         goto result;
3140                                 }
3141
3142                                 if (unlikely(f2fs_cp_error(sbi)))
3143                                         goto lock_folio;
3144
3145                                 if (!f2fs_cluster_is_empty(&cc))
3146                                         goto lock_folio;
3147
3148                                 if (f2fs_all_cluster_page_ready(&cc,
3149                                         pages, i, nr_pages, true))
3150                                         goto lock_folio;
3151
3152                                 ret2 = f2fs_prepare_compress_overwrite(
3153                                                         inode, &pagep,
3154                                                         folio->index, &fsdata);
3155                                 if (ret2 < 0) {
3156                                         ret = ret2;
3157                                         done = 1;
3158                                         break;
3159                                 } else if (ret2 &&
3160                                         (!f2fs_compress_write_end(inode,
3161                                                 fsdata, folio->index, 1) ||
3162                                          !f2fs_all_cluster_page_ready(&cc,
3163                                                 pages, i, nr_pages,
3164                                                 false))) {
3165                                         retry = 1;
3166                                         break;
3167                                 }
3168                         }
3169 #endif
3170                         /* give a priority to WB_SYNC threads */
3171                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3172                                         wbc->sync_mode == WB_SYNC_NONE) {
3173                                 done = 1;
3174                                 break;
3175                         }
3176 #ifdef CONFIG_F2FS_FS_COMPRESSION
3177 lock_folio:
3178 #endif
3179                         done_index = folio->index;
3180 retry_write:
3181                         folio_lock(folio);
3182
3183                         if (unlikely(folio->mapping != mapping)) {
3184 continue_unlock:
3185                                 folio_unlock(folio);
3186                                 continue;
3187                         }
3188
3189                         if (!folio_test_dirty(folio)) {
3190                                 /* someone wrote it for us */
3191                                 goto continue_unlock;
3192                         }
3193
3194                         if (folio_test_writeback(folio)) {
3195                                 if (wbc->sync_mode == WB_SYNC_NONE)
3196                                         goto continue_unlock;
3197                                 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3198                         }
3199
3200                         if (!folio_clear_dirty_for_io(folio))
3201                                 goto continue_unlock;
3202
3203 #ifdef CONFIG_F2FS_FS_COMPRESSION
3204                         if (f2fs_compressed_file(inode)) {
3205                                 folio_get(folio);
3206                                 f2fs_compress_ctx_add_page(&cc, &folio->page);
3207                                 continue;
3208                         }
3209 #endif
3210                         ret = f2fs_write_single_data_page(&folio->page,
3211                                         &submitted, &bio, &last_block,
3212                                         wbc, io_type, 0, true);
3213                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3214                                 folio_unlock(folio);
3215 #ifdef CONFIG_F2FS_FS_COMPRESSION
3216 result:
3217 #endif
3218                         nwritten += submitted;
3219                         wbc->nr_to_write -= submitted;
3220
3221                         if (unlikely(ret)) {
3222                                 /*
3223                                  * keep nr_to_write, since vfs uses this to
3224                                  * get # of written pages.
3225                                  */
3226                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3227                                         ret = 0;
3228                                         goto next;
3229                                 } else if (ret == -EAGAIN) {
3230                                         ret = 0;
3231                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3232                                                 f2fs_io_schedule_timeout(
3233                                                         DEFAULT_IO_TIMEOUT);
3234                                                 goto retry_write;
3235                                         }
3236                                         goto next;
3237                                 }
3238                                 done_index = folio_next_index(folio);
3239                                 done = 1;
3240                                 break;
3241                         }
3242
3243                         if (wbc->nr_to_write <= 0 &&
3244                                         wbc->sync_mode == WB_SYNC_NONE) {
3245                                 done = 1;
3246                                 break;
3247                         }
3248 next:
3249                         if (need_readd)
3250                                 goto readd;
3251                 }
3252                 release_pages(pages, nr_pages);
3253                 cond_resched();
3254         }
3255 #ifdef CONFIG_F2FS_FS_COMPRESSION
3256         /* flush remained pages in compress cluster */
3257         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3258                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3259                 nwritten += submitted;
3260                 wbc->nr_to_write -= submitted;
3261                 if (ret) {
3262                         done = 1;
3263                         retry = 0;
3264                 }
3265         }
3266         if (f2fs_compressed_file(inode))
3267                 f2fs_destroy_compress_ctx(&cc, false);
3268 #endif
3269         if (retry) {
3270                 index = 0;
3271                 end = -1;
3272                 goto retry;
3273         }
3274         if (wbc->range_cyclic && !done)
3275                 done_index = 0;
3276         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3277                 mapping->writeback_index = done_index;
3278
3279         if (nwritten)
3280                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3281                                                                 NULL, 0, DATA);
3282         /* submit cached bio of IPU write */
3283         if (bio)
3284                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3285
3286         return ret;
3287 }
3288
3289 static inline bool __should_serialize_io(struct inode *inode,
3290                                         struct writeback_control *wbc)
3291 {
3292         /* to avoid deadlock in path of data flush */
3293         if (F2FS_I(inode)->wb_task)
3294                 return false;
3295
3296         if (!S_ISREG(inode->i_mode))
3297                 return false;
3298         if (IS_NOQUOTA(inode))
3299                 return false;
3300
3301         if (f2fs_need_compress_data(inode))
3302                 return true;
3303         if (wbc->sync_mode != WB_SYNC_ALL)
3304                 return true;
3305         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3306                 return true;
3307         return false;
3308 }
3309
3310 static int __f2fs_write_data_pages(struct address_space *mapping,
3311                                                 struct writeback_control *wbc,
3312                                                 enum iostat_type io_type)
3313 {
3314         struct inode *inode = mapping->host;
3315         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3316         struct blk_plug plug;
3317         int ret;
3318         bool locked = false;
3319
3320         /* deal with chardevs and other special file */
3321         if (!mapping->a_ops->writepage)
3322                 return 0;
3323
3324         /* skip writing if there is no dirty page in this inode */
3325         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3326                 return 0;
3327
3328         /* during POR, we don't need to trigger writepage at all. */
3329         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3330                 goto skip_write;
3331
3332         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3333                         wbc->sync_mode == WB_SYNC_NONE &&
3334                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3335                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3336                 goto skip_write;
3337
3338         /* skip writing in file defragment preparing stage */
3339         if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3340                 goto skip_write;
3341
3342         trace_f2fs_writepages(mapping->host, wbc, DATA);
3343
3344         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3345         if (wbc->sync_mode == WB_SYNC_ALL)
3346                 atomic_inc(&sbi->wb_sync_req[DATA]);
3347         else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3348                 /* to avoid potential deadlock */
3349                 if (current->plug)
3350                         blk_finish_plug(current->plug);
3351                 goto skip_write;
3352         }
3353
3354         if (__should_serialize_io(inode, wbc)) {
3355                 mutex_lock(&sbi->writepages);
3356                 locked = true;
3357         }
3358
3359         blk_start_plug(&plug);
3360         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3361         blk_finish_plug(&plug);
3362
3363         if (locked)
3364                 mutex_unlock(&sbi->writepages);
3365
3366         if (wbc->sync_mode == WB_SYNC_ALL)
3367                 atomic_dec(&sbi->wb_sync_req[DATA]);
3368         /*
3369          * if some pages were truncated, we cannot guarantee its mapping->host
3370          * to detect pending bios.
3371          */
3372
3373         f2fs_remove_dirty_inode(inode);
3374         return ret;
3375
3376 skip_write:
3377         wbc->pages_skipped += get_dirty_pages(inode);
3378         trace_f2fs_writepages(mapping->host, wbc, DATA);
3379         return 0;
3380 }
3381
3382 static int f2fs_write_data_pages(struct address_space *mapping,
3383                             struct writeback_control *wbc)
3384 {
3385         struct inode *inode = mapping->host;
3386
3387         return __f2fs_write_data_pages(mapping, wbc,
3388                         F2FS_I(inode)->cp_task == current ?
3389                         FS_CP_DATA_IO : FS_DATA_IO);
3390 }
3391
3392 void f2fs_write_failed(struct inode *inode, loff_t to)
3393 {
3394         loff_t i_size = i_size_read(inode);
3395
3396         if (IS_NOQUOTA(inode))
3397                 return;
3398
3399         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3400         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3401                 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3402                 filemap_invalidate_lock(inode->i_mapping);
3403
3404                 truncate_pagecache(inode, i_size);
3405                 f2fs_truncate_blocks(inode, i_size, true);
3406
3407                 filemap_invalidate_unlock(inode->i_mapping);
3408                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3409         }
3410 }
3411
3412 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3413                         struct page *page, loff_t pos, unsigned len,
3414                         block_t *blk_addr, bool *node_changed)
3415 {
3416         struct inode *inode = page->mapping->host;
3417         pgoff_t index = page->index;
3418         struct dnode_of_data dn;
3419         struct page *ipage;
3420         bool locked = false;
3421         int flag = F2FS_GET_BLOCK_PRE_AIO;
3422         int err = 0;
3423
3424         /*
3425          * If a whole page is being written and we already preallocated all the
3426          * blocks, then there is no need to get a block address now.
3427          */
3428         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3429                 return 0;
3430
3431         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3432         if (f2fs_has_inline_data(inode)) {
3433                 if (pos + len > MAX_INLINE_DATA(inode))
3434                         flag = F2FS_GET_BLOCK_DEFAULT;
3435                 f2fs_map_lock(sbi, flag);
3436                 locked = true;
3437         } else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3438                 f2fs_map_lock(sbi, flag);
3439                 locked = true;
3440         }
3441
3442 restart:
3443         /* check inline_data */
3444         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3445         if (IS_ERR(ipage)) {
3446                 err = PTR_ERR(ipage);
3447                 goto unlock_out;
3448         }
3449
3450         set_new_dnode(&dn, inode, ipage, ipage, 0);
3451
3452         if (f2fs_has_inline_data(inode)) {
3453                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3454                         f2fs_do_read_inline_data(page, ipage);
3455                         set_inode_flag(inode, FI_DATA_EXIST);
3456                         if (inode->i_nlink)
3457                                 set_page_private_inline(ipage);
3458                         goto out;
3459                 }
3460                 err = f2fs_convert_inline_page(&dn, page);
3461                 if (err || dn.data_blkaddr != NULL_ADDR)
3462                         goto out;
3463         }
3464
3465         if (!f2fs_lookup_read_extent_cache_block(inode, index,
3466                                                  &dn.data_blkaddr)) {
3467                 if (locked) {
3468                         err = f2fs_reserve_block(&dn, index);
3469                         goto out;
3470                 }
3471
3472                 /* hole case */
3473                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3474                 if (!err && dn.data_blkaddr != NULL_ADDR)
3475                         goto out;
3476                 f2fs_put_dnode(&dn);
3477                 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3478                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3479                 locked = true;
3480                 goto restart;
3481         }
3482 out:
3483         if (!err) {
3484                 /* convert_inline_page can make node_changed */
3485                 *blk_addr = dn.data_blkaddr;
3486                 *node_changed = dn.node_changed;
3487         }
3488         f2fs_put_dnode(&dn);
3489 unlock_out:
3490         if (locked)
3491                 f2fs_map_unlock(sbi, flag);
3492         return err;
3493 }
3494
3495 static int __find_data_block(struct inode *inode, pgoff_t index,
3496                                 block_t *blk_addr)
3497 {
3498         struct dnode_of_data dn;
3499         struct page *ipage;
3500         int err = 0;
3501
3502         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3503         if (IS_ERR(ipage))
3504                 return PTR_ERR(ipage);
3505
3506         set_new_dnode(&dn, inode, ipage, ipage, 0);
3507
3508         if (!f2fs_lookup_read_extent_cache_block(inode, index,
3509                                                  &dn.data_blkaddr)) {
3510                 /* hole case */
3511                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3512                 if (err) {
3513                         dn.data_blkaddr = NULL_ADDR;
3514                         err = 0;
3515                 }
3516         }
3517         *blk_addr = dn.data_blkaddr;
3518         f2fs_put_dnode(&dn);
3519         return err;
3520 }
3521
3522 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3523                                 block_t *blk_addr, bool *node_changed)
3524 {
3525         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3526         struct dnode_of_data dn;
3527         struct page *ipage;
3528         int err = 0;
3529
3530         f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3531
3532         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3533         if (IS_ERR(ipage)) {
3534                 err = PTR_ERR(ipage);
3535                 goto unlock_out;
3536         }
3537         set_new_dnode(&dn, inode, ipage, ipage, 0);
3538
3539         if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3540                                                 &dn.data_blkaddr))
3541                 err = f2fs_reserve_block(&dn, index);
3542
3543         *blk_addr = dn.data_blkaddr;
3544         *node_changed = dn.node_changed;
3545         f2fs_put_dnode(&dn);
3546
3547 unlock_out:
3548         f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3549         return err;
3550 }
3551
3552 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3553                         struct page *page, loff_t pos, unsigned int len,
3554                         block_t *blk_addr, bool *node_changed, bool *use_cow)
3555 {
3556         struct inode *inode = page->mapping->host;
3557         struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3558         pgoff_t index = page->index;
3559         int err = 0;
3560         block_t ori_blk_addr = NULL_ADDR;
3561
3562         /* If pos is beyond the end of file, reserve a new block in COW inode */
3563         if ((pos & PAGE_MASK) >= i_size_read(inode))
3564                 goto reserve_block;
3565
3566         /* Look for the block in COW inode first */
3567         err = __find_data_block(cow_inode, index, blk_addr);
3568         if (err) {
3569                 return err;
3570         } else if (*blk_addr != NULL_ADDR) {
3571                 *use_cow = true;
3572                 return 0;
3573         }
3574
3575         if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3576                 goto reserve_block;
3577
3578         /* Look for the block in the original inode */
3579         err = __find_data_block(inode, index, &ori_blk_addr);
3580         if (err)
3581                 return err;
3582
3583 reserve_block:
3584         /* Finally, we should reserve a new block in COW inode for the update */
3585         err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3586         if (err)
3587                 return err;
3588         inc_atomic_write_cnt(inode);
3589
3590         if (ori_blk_addr != NULL_ADDR)
3591                 *blk_addr = ori_blk_addr;
3592         return 0;
3593 }
3594
3595 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3596                 loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3597 {
3598         struct inode *inode = mapping->host;
3599         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3600         struct page *page = NULL;
3601         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3602         bool need_balance = false;
3603         bool use_cow = false;
3604         block_t blkaddr = NULL_ADDR;
3605         int err = 0;
3606
3607         trace_f2fs_write_begin(inode, pos, len);
3608
3609         if (!f2fs_is_checkpoint_ready(sbi)) {
3610                 err = -ENOSPC;
3611                 goto fail;
3612         }
3613
3614         /*
3615          * We should check this at this moment to avoid deadlock on inode page
3616          * and #0 page. The locking rule for inline_data conversion should be:
3617          * lock_page(page #0) -> lock_page(inode_page)
3618          */
3619         if (index != 0) {
3620                 err = f2fs_convert_inline_inode(inode);
3621                 if (err)
3622                         goto fail;
3623         }
3624
3625 #ifdef CONFIG_F2FS_FS_COMPRESSION
3626         if (f2fs_compressed_file(inode)) {
3627                 int ret;
3628
3629                 *fsdata = NULL;
3630
3631                 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3632                         goto repeat;
3633
3634                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3635                                                         index, fsdata);
3636                 if (ret < 0) {
3637                         err = ret;
3638                         goto fail;
3639                 } else if (ret) {
3640                         return 0;
3641                 }
3642         }
3643 #endif
3644
3645 repeat:
3646         /*
3647          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3648          * wait_for_stable_page. Will wait that below with our IO control.
3649          */
3650         page = f2fs_pagecache_get_page(mapping, index,
3651                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3652         if (!page) {
3653                 err = -ENOMEM;
3654                 goto fail;
3655         }
3656
3657         /* TODO: cluster can be compressed due to race with .writepage */
3658
3659         *pagep = page;
3660
3661         if (f2fs_is_atomic_file(inode))
3662                 err = prepare_atomic_write_begin(sbi, page, pos, len,
3663                                         &blkaddr, &need_balance, &use_cow);
3664         else
3665                 err = prepare_write_begin(sbi, page, pos, len,
3666                                         &blkaddr, &need_balance);
3667         if (err)
3668                 goto fail;
3669
3670         if (need_balance && !IS_NOQUOTA(inode) &&
3671                         has_not_enough_free_secs(sbi, 0, 0)) {
3672                 unlock_page(page);
3673                 f2fs_balance_fs(sbi, true);
3674                 lock_page(page);
3675                 if (page->mapping != mapping) {
3676                         /* The page got truncated from under us */
3677                         f2fs_put_page(page, 1);
3678                         goto repeat;
3679                 }
3680         }
3681
3682         f2fs_wait_on_page_writeback(page, DATA, false, true);
3683
3684         if (len == PAGE_SIZE || PageUptodate(page))
3685                 return 0;
3686
3687         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3688             !f2fs_verity_in_progress(inode)) {
3689                 zero_user_segment(page, len, PAGE_SIZE);
3690                 return 0;
3691         }
3692
3693         if (blkaddr == NEW_ADDR) {
3694                 zero_user_segment(page, 0, PAGE_SIZE);
3695                 SetPageUptodate(page);
3696         } else {
3697                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3698                                 DATA_GENERIC_ENHANCE_READ)) {
3699                         err = -EFSCORRUPTED;
3700                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
3701                         goto fail;
3702                 }
3703                 err = f2fs_submit_page_read(use_cow ?
3704                                 F2FS_I(inode)->cow_inode : inode, page,
3705                                 blkaddr, 0, true);
3706                 if (err)
3707                         goto fail;
3708
3709                 lock_page(page);
3710                 if (unlikely(page->mapping != mapping)) {
3711                         f2fs_put_page(page, 1);
3712                         goto repeat;
3713                 }
3714                 if (unlikely(!PageUptodate(page))) {
3715                         err = -EIO;
3716                         goto fail;
3717                 }
3718         }
3719         return 0;
3720
3721 fail:
3722         f2fs_put_page(page, 1);
3723         f2fs_write_failed(inode, pos + len);
3724         return err;
3725 }
3726
3727 static int f2fs_write_end(struct file *file,
3728                         struct address_space *mapping,
3729                         loff_t pos, unsigned len, unsigned copied,
3730                         struct page *page, void *fsdata)
3731 {
3732         struct inode *inode = page->mapping->host;
3733
3734         trace_f2fs_write_end(inode, pos, len, copied);
3735
3736         /*
3737          * This should be come from len == PAGE_SIZE, and we expect copied
3738          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3739          * let generic_perform_write() try to copy data again through copied=0.
3740          */
3741         if (!PageUptodate(page)) {
3742                 if (unlikely(copied != len))
3743                         copied = 0;
3744                 else
3745                         SetPageUptodate(page);
3746         }
3747
3748 #ifdef CONFIG_F2FS_FS_COMPRESSION
3749         /* overwrite compressed file */
3750         if (f2fs_compressed_file(inode) && fsdata) {
3751                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3752                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3753
3754                 if (pos + copied > i_size_read(inode) &&
3755                                 !f2fs_verity_in_progress(inode))
3756                         f2fs_i_size_write(inode, pos + copied);
3757                 return copied;
3758         }
3759 #endif
3760
3761         if (!copied)
3762                 goto unlock_out;
3763
3764         set_page_dirty(page);
3765
3766         if (pos + copied > i_size_read(inode) &&
3767             !f2fs_verity_in_progress(inode)) {
3768                 f2fs_i_size_write(inode, pos + copied);
3769                 if (f2fs_is_atomic_file(inode))
3770                         f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3771                                         pos + copied);
3772         }
3773 unlock_out:
3774         f2fs_put_page(page, 1);
3775         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3776         return copied;
3777 }
3778
3779 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3780 {
3781         struct inode *inode = folio->mapping->host;
3782         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3783
3784         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3785                                 (offset || length != folio_size(folio)))
3786                 return;
3787
3788         if (folio_test_dirty(folio)) {
3789                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3790                         dec_page_count(sbi, F2FS_DIRTY_META);
3791                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3792                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3793                 } else {
3794                         inode_dec_dirty_pages(inode);
3795                         f2fs_remove_dirty_inode(inode);
3796                 }
3797         }
3798         clear_page_private_all(&folio->page);
3799 }
3800
3801 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3802 {
3803         /* If this is dirty folio, keep private data */
3804         if (folio_test_dirty(folio))
3805                 return false;
3806
3807         clear_page_private_all(&folio->page);
3808         return true;
3809 }
3810
3811 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3812                 struct folio *folio)
3813 {
3814         struct inode *inode = mapping->host;
3815
3816         trace_f2fs_set_page_dirty(&folio->page, DATA);
3817
3818         if (!folio_test_uptodate(folio))
3819                 folio_mark_uptodate(folio);
3820         BUG_ON(folio_test_swapcache(folio));
3821
3822         if (filemap_dirty_folio(mapping, folio)) {
3823                 f2fs_update_dirty_folio(inode, folio);
3824                 return true;
3825         }
3826         return false;
3827 }
3828
3829
3830 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3831 {
3832 #ifdef CONFIG_F2FS_FS_COMPRESSION
3833         struct dnode_of_data dn;
3834         sector_t start_idx, blknr = 0;
3835         int ret;
3836
3837         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3838
3839         set_new_dnode(&dn, inode, NULL, NULL, 0);
3840         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3841         if (ret)
3842                 return 0;
3843
3844         if (dn.data_blkaddr != COMPRESS_ADDR) {
3845                 dn.ofs_in_node += block - start_idx;
3846                 blknr = f2fs_data_blkaddr(&dn);
3847                 if (!__is_valid_data_blkaddr(blknr))
3848                         blknr = 0;
3849         }
3850
3851         f2fs_put_dnode(&dn);
3852         return blknr;
3853 #else
3854         return 0;
3855 #endif
3856 }
3857
3858
3859 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3860 {
3861         struct inode *inode = mapping->host;
3862         sector_t blknr = 0;
3863
3864         if (f2fs_has_inline_data(inode))
3865                 goto out;
3866
3867         /* make sure allocating whole blocks */
3868         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3869                 filemap_write_and_wait(mapping);
3870
3871         /* Block number less than F2FS MAX BLOCKS */
3872         if (unlikely(block >= max_file_blocks(inode)))
3873                 goto out;
3874
3875         if (f2fs_compressed_file(inode)) {
3876                 blknr = f2fs_bmap_compress(inode, block);
3877         } else {
3878                 struct f2fs_map_blocks map;
3879
3880                 memset(&map, 0, sizeof(map));
3881                 map.m_lblk = block;
3882                 map.m_len = 1;
3883                 map.m_next_pgofs = NULL;
3884                 map.m_seg_type = NO_CHECK_TYPE;
3885
3886                 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3887                         blknr = map.m_pblk;
3888         }
3889 out:
3890         trace_f2fs_bmap(inode, block, blknr);
3891         return blknr;
3892 }
3893
3894 #ifdef CONFIG_SWAP
3895 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3896                                                         unsigned int blkcnt)
3897 {
3898         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3899         unsigned int blkofs;
3900         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3901         unsigned int secidx = start_blk / blk_per_sec;
3902         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3903         int ret = 0;
3904
3905         f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3906         filemap_invalidate_lock(inode->i_mapping);
3907
3908         set_inode_flag(inode, FI_ALIGNED_WRITE);
3909         set_inode_flag(inode, FI_OPU_WRITE);
3910
3911         for (; secidx < end_sec; secidx++) {
3912                 f2fs_down_write(&sbi->pin_sem);
3913
3914                 f2fs_lock_op(sbi);
3915                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3916                 f2fs_unlock_op(sbi);
3917
3918                 set_inode_flag(inode, FI_SKIP_WRITES);
3919
3920                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3921                         struct page *page;
3922                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3923
3924                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3925                         if (IS_ERR(page)) {
3926                                 f2fs_up_write(&sbi->pin_sem);
3927                                 ret = PTR_ERR(page);
3928                                 goto done;
3929                         }
3930
3931                         set_page_dirty(page);
3932                         f2fs_put_page(page, 1);
3933                 }
3934
3935                 clear_inode_flag(inode, FI_SKIP_WRITES);
3936
3937                 ret = filemap_fdatawrite(inode->i_mapping);
3938
3939                 f2fs_up_write(&sbi->pin_sem);
3940
3941                 if (ret)
3942                         break;
3943         }
3944
3945 done:
3946         clear_inode_flag(inode, FI_SKIP_WRITES);
3947         clear_inode_flag(inode, FI_OPU_WRITE);
3948         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3949
3950         filemap_invalidate_unlock(inode->i_mapping);
3951         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3952
3953         return ret;
3954 }
3955
3956 static int check_swap_activate(struct swap_info_struct *sis,
3957                                 struct file *swap_file, sector_t *span)
3958 {
3959         struct address_space *mapping = swap_file->f_mapping;
3960         struct inode *inode = mapping->host;
3961         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3962         sector_t cur_lblock;
3963         sector_t last_lblock;
3964         sector_t pblock;
3965         sector_t lowest_pblock = -1;
3966         sector_t highest_pblock = 0;
3967         int nr_extents = 0;
3968         unsigned long nr_pblocks;
3969         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3970         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3971         unsigned int not_aligned = 0;
3972         int ret = 0;
3973
3974         /*
3975          * Map all the blocks into the extent list.  This code doesn't try
3976          * to be very smart.
3977          */
3978         cur_lblock = 0;
3979         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3980
3981         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3982                 struct f2fs_map_blocks map;
3983 retry:
3984                 cond_resched();
3985
3986                 memset(&map, 0, sizeof(map));
3987                 map.m_lblk = cur_lblock;
3988                 map.m_len = last_lblock - cur_lblock;
3989                 map.m_next_pgofs = NULL;
3990                 map.m_next_extent = NULL;
3991                 map.m_seg_type = NO_CHECK_TYPE;
3992                 map.m_may_create = false;
3993
3994                 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
3995                 if (ret)
3996                         goto out;
3997
3998                 /* hole */
3999                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
4000                         f2fs_err(sbi, "Swapfile has holes");
4001                         ret = -EINVAL;
4002                         goto out;
4003                 }
4004
4005                 pblock = map.m_pblk;
4006                 nr_pblocks = map.m_len;
4007
4008                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
4009                                 nr_pblocks & sec_blks_mask) {
4010                         not_aligned++;
4011
4012                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
4013                         if (cur_lblock + nr_pblocks > sis->max)
4014                                 nr_pblocks -= blks_per_sec;
4015
4016                         if (!nr_pblocks) {
4017                                 /* this extent is last one */
4018                                 nr_pblocks = map.m_len;
4019                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
4020                                 goto next;
4021                         }
4022
4023                         ret = f2fs_migrate_blocks(inode, cur_lblock,
4024                                                         nr_pblocks);
4025                         if (ret)
4026                                 goto out;
4027                         goto retry;
4028                 }
4029 next:
4030                 if (cur_lblock + nr_pblocks >= sis->max)
4031                         nr_pblocks = sis->max - cur_lblock;
4032
4033                 if (cur_lblock) {       /* exclude the header page */
4034                         if (pblock < lowest_pblock)
4035                                 lowest_pblock = pblock;
4036                         if (pblock + nr_pblocks - 1 > highest_pblock)
4037                                 highest_pblock = pblock + nr_pblocks - 1;
4038                 }
4039
4040                 /*
4041                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4042                  */
4043                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4044                 if (ret < 0)
4045                         goto out;
4046                 nr_extents += ret;
4047                 cur_lblock += nr_pblocks;
4048         }
4049         ret = nr_extents;
4050         *span = 1 + highest_pblock - lowest_pblock;
4051         if (cur_lblock == 0)
4052                 cur_lblock = 1; /* force Empty message */
4053         sis->max = cur_lblock;
4054         sis->pages = cur_lblock - 1;
4055         sis->highest_bit = cur_lblock - 1;
4056 out:
4057         if (not_aligned)
4058                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
4059                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
4060         return ret;
4061 }
4062
4063 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4064                                 sector_t *span)
4065 {
4066         struct inode *inode = file_inode(file);
4067         int ret;
4068
4069         if (!S_ISREG(inode->i_mode))
4070                 return -EINVAL;
4071
4072         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4073                 return -EROFS;
4074
4075         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4076                 f2fs_err(F2FS_I_SB(inode),
4077                         "Swapfile not supported in LFS mode");
4078                 return -EINVAL;
4079         }
4080
4081         ret = f2fs_convert_inline_inode(inode);
4082         if (ret)
4083                 return ret;
4084
4085         if (!f2fs_disable_compressed_file(inode))
4086                 return -EINVAL;
4087
4088         f2fs_precache_extents(inode);
4089
4090         ret = check_swap_activate(sis, file, span);
4091         if (ret < 0)
4092                 return ret;
4093
4094         stat_inc_swapfile_inode(inode);
4095         set_inode_flag(inode, FI_PIN_FILE);
4096         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4097         return ret;
4098 }
4099
4100 static void f2fs_swap_deactivate(struct file *file)
4101 {
4102         struct inode *inode = file_inode(file);
4103
4104         stat_dec_swapfile_inode(inode);
4105         clear_inode_flag(inode, FI_PIN_FILE);
4106 }
4107 #else
4108 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4109                                 sector_t *span)
4110 {
4111         return -EOPNOTSUPP;
4112 }
4113
4114 static void f2fs_swap_deactivate(struct file *file)
4115 {
4116 }
4117 #endif
4118
4119 const struct address_space_operations f2fs_dblock_aops = {
4120         .read_folio     = f2fs_read_data_folio,
4121         .readahead      = f2fs_readahead,
4122         .writepage      = f2fs_write_data_page,
4123         .writepages     = f2fs_write_data_pages,
4124         .write_begin    = f2fs_write_begin,
4125         .write_end      = f2fs_write_end,
4126         .dirty_folio    = f2fs_dirty_data_folio,
4127         .migrate_folio  = filemap_migrate_folio,
4128         .invalidate_folio = f2fs_invalidate_folio,
4129         .release_folio  = f2fs_release_folio,
4130         .bmap           = f2fs_bmap,
4131         .swap_activate  = f2fs_swap_activate,
4132         .swap_deactivate = f2fs_swap_deactivate,
4133 };
4134
4135 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4136 {
4137         struct address_space *mapping = page_mapping(page);
4138         unsigned long flags;
4139
4140         xa_lock_irqsave(&mapping->i_pages, flags);
4141         __xa_clear_mark(&mapping->i_pages, page_index(page),
4142                                                 PAGECACHE_TAG_DIRTY);
4143         xa_unlock_irqrestore(&mapping->i_pages, flags);
4144 }
4145
4146 int __init f2fs_init_post_read_processing(void)
4147 {
4148         bio_post_read_ctx_cache =
4149                 kmem_cache_create("f2fs_bio_post_read_ctx",
4150                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4151         if (!bio_post_read_ctx_cache)
4152                 goto fail;
4153         bio_post_read_ctx_pool =
4154                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4155                                          bio_post_read_ctx_cache);
4156         if (!bio_post_read_ctx_pool)
4157                 goto fail_free_cache;
4158         return 0;
4159
4160 fail_free_cache:
4161         kmem_cache_destroy(bio_post_read_ctx_cache);
4162 fail:
4163         return -ENOMEM;
4164 }
4165
4166 void f2fs_destroy_post_read_processing(void)
4167 {
4168         mempool_destroy(bio_post_read_ctx_pool);
4169         kmem_cache_destroy(bio_post_read_ctx_cache);
4170 }
4171
4172 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4173 {
4174         if (!f2fs_sb_has_encrypt(sbi) &&
4175                 !f2fs_sb_has_verity(sbi) &&
4176                 !f2fs_sb_has_compression(sbi))
4177                 return 0;
4178
4179         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4180                                                  WQ_UNBOUND | WQ_HIGHPRI,
4181                                                  num_online_cpus());
4182         return sbi->post_read_wq ? 0 : -ENOMEM;
4183 }
4184
4185 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4186 {
4187         if (sbi->post_read_wq)
4188                 destroy_workqueue(sbi->post_read_wq);
4189 }
4190
4191 int __init f2fs_init_bio_entry_cache(void)
4192 {
4193         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4194                         sizeof(struct bio_entry));
4195         return bio_entry_slab ? 0 : -ENOMEM;
4196 }
4197
4198 void f2fs_destroy_bio_entry_cache(void)
4199 {
4200         kmem_cache_destroy(bio_entry_slab);
4201 }
4202
4203 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4204                             unsigned int flags, struct iomap *iomap,
4205                             struct iomap *srcmap)
4206 {
4207         struct f2fs_map_blocks map = {};
4208         pgoff_t next_pgofs = 0;
4209         int err;
4210
4211         map.m_lblk = bytes_to_blks(inode, offset);
4212         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4213         map.m_next_pgofs = &next_pgofs;
4214         map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4215         if (flags & IOMAP_WRITE)
4216                 map.m_may_create = true;
4217
4218         err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4219         if (err)
4220                 return err;
4221
4222         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4223
4224         /*
4225          * When inline encryption is enabled, sometimes I/O to an encrypted file
4226          * has to be broken up to guarantee DUN contiguity.  Handle this by
4227          * limiting the length of the mapping returned.
4228          */
4229         map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4230
4231         /*
4232          * We should never see delalloc or compressed extents here based on
4233          * prior flushing and checks.
4234          */
4235         if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4236                 return -EINVAL;
4237         if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4238                 return -EINVAL;
4239
4240         if (map.m_pblk != NULL_ADDR) {
4241                 iomap->length = blks_to_bytes(inode, map.m_len);
4242                 iomap->type = IOMAP_MAPPED;
4243                 iomap->flags |= IOMAP_F_MERGED;
4244                 iomap->bdev = map.m_bdev;
4245                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4246         } else {
4247                 if (flags & IOMAP_WRITE)
4248                         return -ENOTBLK;
4249                 iomap->length = blks_to_bytes(inode, next_pgofs) -
4250                                 iomap->offset;
4251                 iomap->type = IOMAP_HOLE;
4252                 iomap->addr = IOMAP_NULL_ADDR;
4253         }
4254
4255         if (map.m_flags & F2FS_MAP_NEW)
4256                 iomap->flags |= IOMAP_F_NEW;
4257         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4258             offset + length > i_size_read(inode))
4259                 iomap->flags |= IOMAP_F_DIRTY;
4260
4261         return 0;
4262 }
4263
4264 const struct iomap_ops f2fs_iomap_ops = {
4265         .iomap_begin    = f2fs_iomap_begin,
4266 };