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 bool f2fs_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 ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
  69                         page_private_gcing(page))
  70                 return true;
  71         return false;
  72 }
  73
  74 static enum count_type __read_io_type(struct page *page)
  75 {
  76         struct address_space *mapping = page_file_mapping(page);
  77
  78         if (mapping) {
  79                 struct inode *inode = mapping->host;
  80                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  81
  82                 if (inode->i_ino == F2FS_META_INO(sbi))
  83                         return F2FS_RD_META;
  84
  85                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  86                         return F2FS_RD_NODE;
  87         }
  88         return F2FS_RD_DATA;
  89 }
  90
  91 /* postprocessing steps for read bios */
  92 enum bio_post_read_step {
  93 #ifdef CONFIG_FS_ENCRYPTION
  94         STEP_DECRYPT    = BIT(0),
  95 #else
  96         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
  97 #endif
  98 #ifdef CONFIG_F2FS_FS_COMPRESSION
  99         STEP_DECOMPRESS = BIT(1),
 100 #else
 101         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 102 #endif
 103 #ifdef CONFIG_FS_VERITY
 104         STEP_VERITY     = BIT(2),
 105 #else
 106         STEP_VERITY     = 0,    /* compile out the verity-related code */
 107 #endif
 108 };
 109
 110 struct bio_post_read_ctx {
 111         struct bio *bio;
 112         struct f2fs_sb_info *sbi;
 113         struct work_struct work;
 114         unsigned int enabled_steps;
 115         /*
 116          * decompression_attempted keeps track of whether
 117          * f2fs_end_read_compressed_page() has been called on the pages in the
 118          * bio that belong to a compressed cluster yet.
 119          */
 120         bool decompression_attempted;
 121         block_t fs_blkaddr;
 122 };
 123
 124 /*
 125  * Update and unlock a bio's pages, and free the bio.
 126  *
 127  * This marks pages up-to-date only if there was no error in the bio (I/O error,
 128  * decryption error, or verity error), as indicated by bio->bi_status.
 129  *
 130  * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk)
 131  * aren't marked up-to-date here, as decompression is done on a per-compression-
 132  * cluster basis rather than a per-bio basis.  Instead, we only must do two
 133  * things for each compressed page here: call f2fs_end_read_compressed_page()
 134  * with failed=true if an error occurred before it would have normally gotten
 135  * called (i.e., I/O error or decryption error, but *not* verity error), and
 136  * release the bio's reference to the decompress_io_ctx of the page's cluster.
 137  */
 138 static void f2fs_finish_read_bio(struct bio *bio, bool in_task)
 139 {
 140         struct bio_vec *bv;
 141         struct bvec_iter_all iter_all;
 142         struct bio_post_read_ctx *ctx = bio->bi_private;
 143
 144         bio_for_each_segment_all(bv, bio, iter_all) {
 145                 struct page *page = bv->bv_page;
 146
 147                 if (f2fs_is_compressed_page(page)) {
 148                         if (ctx && !ctx->decompression_attempted)
 149                                 f2fs_end_read_compressed_page(page, true, 0,
 150                                                         in_task);
 151                         f2fs_put_page_dic(page, in_task);
 152                         continue;
 153                 }
 154
 155                 if (bio->bi_status)
 156                         ClearPageUptodate(page);
 157                 else
 158                         SetPageUptodate(page);
 159                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 160                 unlock_page(page);
 161         }
 162
 163         if (ctx)
 164                 mempool_free(ctx, bio_post_read_ctx_pool);
 165         bio_put(bio);
 166 }
 167
 168 static void f2fs_verify_bio(struct work_struct *work)
 169 {
 170         struct bio_post_read_ctx *ctx =
 171                 container_of(work, struct bio_post_read_ctx, work);
 172         struct bio *bio = ctx->bio;
 173         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 174
 175         /*
 176          * fsverity_verify_bio() may call readahead() again, and while verity
 177          * will be disabled for this, decryption and/or decompression may still
 178          * be needed, resulting in another bio_post_read_ctx being allocated.
 179          * So to prevent deadlocks we need to release the current ctx to the
 180          * mempool first.  This assumes that verity is the last post-read step.
 181          */
 182         mempool_free(ctx, bio_post_read_ctx_pool);
 183         bio->bi_private = NULL;
 184
 185         /*
 186          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 187          * as those were handled separately by f2fs_end_read_compressed_page().
 188          */
 189         if (may_have_compressed_pages) {
 190                 struct bio_vec *bv;
 191                 struct bvec_iter_all iter_all;
 192
 193                 bio_for_each_segment_all(bv, bio, iter_all) {
 194                         struct page *page = bv->bv_page;
 195
 196                         if (!f2fs_is_compressed_page(page) &&
 197                             !fsverity_verify_page(page)) {
 198                                 bio->bi_status = BLK_STS_IOERR;
 199                                 break;
 200                         }
 201                 }
 202         } else {
 203                 fsverity_verify_bio(bio);
 204         }
 205
 206         f2fs_finish_read_bio(bio, true);
 207 }
 208
 209 /*
 210  * If the bio's data needs to be verified with fs-verity, then enqueue the
 211  * verity work for the bio.  Otherwise finish the bio now.
 212  *
 213  * Note that to avoid deadlocks, the verity work can't be done on the
 214  * decryption/decompression workqueue.  This is because verifying the data pages
 215  * can involve reading verity metadata pages from the file, and these verity
 216  * metadata pages may be encrypted and/or compressed.
 217  */
 218 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task)
 219 {
 220         struct bio_post_read_ctx *ctx = bio->bi_private;
 221
 222         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 223                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 224                 fsverity_enqueue_verify_work(&ctx->work);
 225         } else {
 226                 f2fs_finish_read_bio(bio, in_task);
 227         }
 228 }
 229
 230 /*
 231  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 232  * remaining page was read by @ctx->bio.
 233  *
 234  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 235  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 236  * that the bio includes at least one compressed page.  The actual decompression
 237  * is done on a per-cluster basis, not a per-bio basis.
 238  */
 239 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx,
 240                 bool in_task)
 241 {
 242         struct bio_vec *bv;
 243         struct bvec_iter_all iter_all;
 244         bool all_compressed = true;
 245         block_t blkaddr = ctx->fs_blkaddr;
 246
 247         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 248                 struct page *page = bv->bv_page;
 249
 250                 if (f2fs_is_compressed_page(page))
 251                         f2fs_end_read_compressed_page(page, false, blkaddr,
 252                                                       in_task);
 253                 else
 254                         all_compressed = false;
 255
 256                 blkaddr++;
 257         }
 258
 259         ctx->decompression_attempted = true;
 260
 261         /*
 262          * Optimization: if all the bio's pages are compressed, then scheduling
 263          * the per-bio verity work is unnecessary, as verity will be fully
 264          * handled at the compression cluster level.
 265          */
 266         if (all_compressed)
 267                 ctx->enabled_steps &= ~STEP_VERITY;
 268 }
 269
 270 static void f2fs_post_read_work(struct work_struct *work)
 271 {
 272         struct bio_post_read_ctx *ctx =
 273                 container_of(work, struct bio_post_read_ctx, work);
 274         struct bio *bio = ctx->bio;
 275
 276         if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) {
 277                 f2fs_finish_read_bio(bio, true);
 278                 return;
 279         }
 280
 281         if (ctx->enabled_steps & STEP_DECOMPRESS)
 282                 f2fs_handle_step_decompress(ctx, true);
 283
 284         f2fs_verify_and_finish_bio(bio, true);
 285 }
 286
 287 static void f2fs_read_end_io(struct bio *bio)
 288 {
 289         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 290         struct bio_post_read_ctx *ctx;
 291         bool intask = in_task();
 292
 293         iostat_update_and_unbind_ctx(bio);
 294         ctx = bio->bi_private;
 295
 296         if (time_to_inject(sbi, FAULT_READ_IO))
 297                 bio->bi_status = BLK_STS_IOERR;
 298
 299         if (bio->bi_status) {
 300                 f2fs_finish_read_bio(bio, intask);
 301                 return;
 302         }
 303
 304         if (ctx) {
 305                 unsigned int enabled_steps = ctx->enabled_steps &
 306                                         (STEP_DECRYPT | STEP_DECOMPRESS);
 307
 308                 /*
 309                  * If we have only decompression step between decompression and
 310                  * decrypt, we don't need post processing for this.
 311                  */
 312                 if (enabled_steps == STEP_DECOMPRESS &&
 313                                 !f2fs_low_mem_mode(sbi)) {
 314                         f2fs_handle_step_decompress(ctx, intask);
 315                 } else if (enabled_steps) {
 316                         INIT_WORK(&ctx->work, f2fs_post_read_work);
 317                         queue_work(ctx->sbi->post_read_wq, &ctx->work);
 318                         return;
 319                 }
 320         }
 321
 322         f2fs_verify_and_finish_bio(bio, intask);
 323 }
 324
 325 static void f2fs_write_end_io(struct bio *bio)
 326 {
 327         struct f2fs_sb_info *sbi;
 328         struct bio_vec *bvec;
 329         struct bvec_iter_all iter_all;
 330
 331         iostat_update_and_unbind_ctx(bio);
 332         sbi = bio->bi_private;
 333
 334         if (time_to_inject(sbi, FAULT_WRITE_IO))
 335                 bio->bi_status = BLK_STS_IOERR;
 336
 337         bio_for_each_segment_all(bvec, bio, iter_all) {
 338                 struct page *page = bvec->bv_page;
 339                 enum count_type type = WB_DATA_TYPE(page, false);
 340
 341                 fscrypt_finalize_bounce_page(&page);
 342
 343 #ifdef CONFIG_F2FS_FS_COMPRESSION
 344                 if (f2fs_is_compressed_page(page)) {
 345                         f2fs_compress_write_end_io(bio, page);
 346                         continue;
 347                 }
 348 #endif
 349
 350                 if (unlikely(bio->bi_status)) {
 351                         mapping_set_error(page->mapping, -EIO);
 352                         if (type == F2FS_WB_CP_DATA)
 353                                 f2fs_stop_checkpoint(sbi, true,
 354                                                 STOP_CP_REASON_WRITE_FAIL);
 355                 }
 356
 357                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 358                                         page->index != nid_of_node(page));
 359
 360                 dec_page_count(sbi, type);
 361                 if (f2fs_in_warm_node_list(sbi, page))
 362                         f2fs_del_fsync_node_entry(sbi, page);
 363                 clear_page_private_gcing(page);
 364                 end_page_writeback(page);
 365         }
 366         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 367                                 wq_has_sleeper(&sbi->cp_wait))
 368                 wake_up(&sbi->cp_wait);
 369
 370         bio_put(bio);
 371 }
 372
 373 #ifdef CONFIG_BLK_DEV_ZONED
 374 static void f2fs_zone_write_end_io(struct bio *bio)
 375 {
 376         struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
 377
 378         bio->bi_private = io->bi_private;
 379         complete(&io->zone_wait);
 380         f2fs_write_end_io(bio);
 381 }
 382 #endif
 383
 384 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 385                 block_t blk_addr, sector_t *sector)
 386 {
 387         struct block_device *bdev = sbi->sb->s_bdev;
 388         int i;
 389
 390         if (f2fs_is_multi_device(sbi)) {
 391                 for (i = 0; i < sbi->s_ndevs; i++) {
 392                         if (FDEV(i).start_blk <= blk_addr &&
 393                             FDEV(i).end_blk >= blk_addr) {
 394                                 blk_addr -= FDEV(i).start_blk;
 395                                 bdev = FDEV(i).bdev;
 396                                 break;
 397                         }
 398                 }
 399         }
 400
 401         if (sector)
 402                 *sector = SECTOR_FROM_BLOCK(blk_addr);
 403         return bdev;
 404 }
 405
 406 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 407 {
 408         int i;
 409
 410         if (!f2fs_is_multi_device(sbi))
 411                 return 0;
 412
 413         for (i = 0; i < sbi->s_ndevs; i++)
 414                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 415                         return i;
 416         return 0;
 417 }
 418
 419 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
 420 {
 421         unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0);
 422         unsigned int fua_flag, meta_flag, io_flag;
 423         blk_opf_t op_flags = 0;
 424
 425         if (fio->op != REQ_OP_WRITE)
 426                 return 0;
 427         if (fio->type == DATA)
 428                 io_flag = fio->sbi->data_io_flag;
 429         else if (fio->type == NODE)
 430                 io_flag = fio->sbi->node_io_flag;
 431         else
 432                 return 0;
 433
 434         fua_flag = io_flag & temp_mask;
 435         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 436
 437         /*
 438          * data/node io flag bits per temp:
 439          *      REQ_META     |      REQ_FUA      |
 440          *    5 |    4 |   3 |    2 |    1 |   0 |
 441          * Cold | Warm | Hot | Cold | Warm | Hot |
 442          */
 443         if (BIT(fio->temp) & meta_flag)
 444                 op_flags |= REQ_META;
 445         if (BIT(fio->temp) & fua_flag)
 446                 op_flags |= REQ_FUA;
 447         return op_flags;
 448 }
 449
 450 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 451 {
 452         struct f2fs_sb_info *sbi = fio->sbi;
 453         struct block_device *bdev;
 454         sector_t sector;
 455         struct bio *bio;
 456
 457         bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 458         bio = bio_alloc_bioset(bdev, npages,
 459                                 fio->op | fio->op_flags | f2fs_io_flags(fio),
 460                                 GFP_NOIO, &f2fs_bioset);
 461         bio->bi_iter.bi_sector = sector;
 462         if (is_read_io(fio->op)) {
 463                 bio->bi_end_io = f2fs_read_end_io;
 464                 bio->bi_private = NULL;
 465         } else {
 466                 bio->bi_end_io = f2fs_write_end_io;
 467                 bio->bi_private = sbi;
 468                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 469                                                 fio->type, fio->temp);
 470         }
 471         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 472
 473         if (fio->io_wbc)
 474                 wbc_init_bio(fio->io_wbc, bio);
 475
 476         return bio;
 477 }
 478
 479 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 480                                   pgoff_t first_idx,
 481                                   const struct f2fs_io_info *fio,
 482                                   gfp_t gfp_mask)
 483 {
 484         /*
 485          * The f2fs garbage collector sets ->encrypted_page when it wants to
 486          * read/write raw data without encryption.
 487          */
 488         if (!fio || !fio->encrypted_page)
 489                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 490 }
 491
 492 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 493                                      pgoff_t next_idx,
 494                                      const struct f2fs_io_info *fio)
 495 {
 496         /*
 497          * The f2fs garbage collector sets ->encrypted_page when it wants to
 498          * read/write raw data without encryption.
 499          */
 500         if (fio && fio->encrypted_page)
 501                 return !bio_has_crypt_ctx(bio);
 502
 503         return fscrypt_mergeable_bio(bio, inode, next_idx);
 504 }
 505
 506 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 507                                  enum page_type type)
 508 {
 509         WARN_ON_ONCE(!is_read_io(bio_op(bio)));
 510         trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 511
 512         iostat_update_submit_ctx(bio, type);
 513         submit_bio(bio);
 514 }
 515
 516 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 517                                   enum page_type type)
 518 {
 519         WARN_ON_ONCE(is_read_io(bio_op(bio)));
 520
 521         if (f2fs_lfs_mode(sbi) && current->plug && PAGE_TYPE_ON_MAIN(type))
 522                 blk_finish_plug(current->plug);
 523
 524         trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 525         iostat_update_submit_ctx(bio, type);
 526         submit_bio(bio);
 527 }
 528
 529 static void __submit_merged_bio(struct f2fs_bio_info *io)
 530 {
 531         struct f2fs_io_info *fio = &io->fio;
 532
 533         if (!io->bio)
 534                 return;
 535
 536         if (is_read_io(fio->op)) {
 537                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 538                 f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
 539         } else {
 540                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 541                 f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
 542         }
 543         io->bio = NULL;
 544 }
 545
 546 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 547                                                 struct page *page, nid_t ino)
 548 {
 549         struct bio_vec *bvec;
 550         struct bvec_iter_all iter_all;
 551
 552         if (!bio)
 553                 return false;
 554
 555         if (!inode && !page && !ino)
 556                 return true;
 557
 558         bio_for_each_segment_all(bvec, bio, iter_all) {
 559                 struct page *target = bvec->bv_page;
 560
 561                 if (fscrypt_is_bounce_page(target)) {
 562                         target = fscrypt_pagecache_page(target);
 563                         if (IS_ERR(target))
 564                                 continue;
 565                 }
 566                 if (f2fs_is_compressed_page(target)) {
 567                         target = f2fs_compress_control_page(target);
 568                         if (IS_ERR(target))
 569                                 continue;
 570                 }
 571
 572                 if (inode && inode == target->mapping->host)
 573                         return true;
 574                 if (page && page == target)
 575                         return true;
 576                 if (ino && ino == ino_of_node(target))
 577                         return true;
 578         }
 579
 580         return false;
 581 }
 582
 583 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
 584 {
 585         int i;
 586
 587         for (i = 0; i < NR_PAGE_TYPE; i++) {
 588                 int n = (i == META) ? 1 : NR_TEMP_TYPE;
 589                 int j;
 590
 591                 sbi->write_io[i] = f2fs_kmalloc(sbi,
 592                                 array_size(n, sizeof(struct f2fs_bio_info)),
 593                                 GFP_KERNEL);
 594                 if (!sbi->write_io[i])
 595                         return -ENOMEM;
 596
 597                 for (j = HOT; j < n; j++) {
 598                         struct f2fs_bio_info *io = &sbi->write_io[i][j];
 599
 600                         init_f2fs_rwsem(&io->io_rwsem);
 601                         io->sbi = sbi;
 602                         io->bio = NULL;
 603                         io->last_block_in_bio = 0;
 604                         spin_lock_init(&io->io_lock);
 605                         INIT_LIST_HEAD(&io->io_list);
 606                         INIT_LIST_HEAD(&io->bio_list);
 607                         init_f2fs_rwsem(&io->bio_list_lock);
 608 #ifdef CONFIG_BLK_DEV_ZONED
 609                         init_completion(&io->zone_wait);
 610                         io->zone_pending_bio = NULL;
 611                         io->bi_private = NULL;
 612 #endif
 613                 }
 614         }
 615
 616         return 0;
 617 }
 618
 619 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 620                                 enum page_type type, enum temp_type temp)
 621 {
 622         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 623         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 624
 625         f2fs_down_write(&io->io_rwsem);
 626
 627         if (!io->bio)
 628                 goto unlock_out;
 629
 630         /* change META to META_FLUSH in the checkpoint procedure */
 631         if (type >= META_FLUSH) {
 632                 io->fio.type = META_FLUSH;
 633                 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 634                 if (!test_opt(sbi, NOBARRIER))
 635                         io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 636         }
 637         __submit_merged_bio(io);
 638 unlock_out:
 639         f2fs_up_write(&io->io_rwsem);
 640 }
 641
 642 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 643                                 struct inode *inode, struct page *page,
 644                                 nid_t ino, enum page_type type, bool force)
 645 {
 646         enum temp_type temp;
 647         bool ret = true;
 648
 649         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 650                 if (!force)     {
 651                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 652                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 653
 654                         f2fs_down_read(&io->io_rwsem);
 655                         ret = __has_merged_page(io->bio, inode, page, ino);
 656                         f2fs_up_read(&io->io_rwsem);
 657                 }
 658                 if (ret)
 659                         __f2fs_submit_merged_write(sbi, type, temp);
 660
 661                 /* TODO: use HOT temp only for meta pages now. */
 662                 if (type >= META)
 663                         break;
 664         }
 665 }
 666
 667 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 668 {
 669         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 670 }
 671
 672 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 673                                 struct inode *inode, struct page *page,
 674                                 nid_t ino, enum page_type type)
 675 {
 676         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 677 }
 678
 679 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 680 {
 681         f2fs_submit_merged_write(sbi, DATA);
 682         f2fs_submit_merged_write(sbi, NODE);
 683         f2fs_submit_merged_write(sbi, META);
 684 }
 685
 686 /*
 687  * Fill the locked page with data located in the block address.
 688  * A caller needs to unlock the page on failure.
 689  */
 690 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 691 {
 692         struct bio *bio;
 693         struct page *page = fio->encrypted_page ?
 694                         fio->encrypted_page : fio->page;
 695
 696         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 697                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 698                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 699                 return -EFSCORRUPTED;
 700
 701         trace_f2fs_submit_page_bio(page, fio);
 702
 703         /* Allocate a new bio */
 704         bio = __bio_alloc(fio, 1);
 705
 706         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 707                                fio->page->index, fio, GFP_NOIO);
 708
 709         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 710                 bio_put(bio);
 711                 return -EFAULT;
 712         }
 713
 714         if (fio->io_wbc && !is_read_io(fio->op))
 715                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 716
 717         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 718                         __read_io_type(page) : WB_DATA_TYPE(fio->page, false));
 719
 720         if (is_read_io(bio_op(bio)))
 721                 f2fs_submit_read_bio(fio->sbi, bio, fio->type);
 722         else
 723                 f2fs_submit_write_bio(fio->sbi, bio, fio->type);
 724         return 0;
 725 }
 726
 727 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 728                                 block_t last_blkaddr, block_t cur_blkaddr)
 729 {
 730         if (unlikely(sbi->max_io_bytes &&
 731                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 732                 return false;
 733         if (last_blkaddr + 1 != cur_blkaddr)
 734                 return false;
 735         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 736 }
 737
 738 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 739                                                 struct f2fs_io_info *fio)
 740 {
 741         if (io->fio.op != fio->op)
 742                 return false;
 743         return io->fio.op_flags == fio->op_flags;
 744 }
 745
 746 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 747                                         struct f2fs_bio_info *io,
 748                                         struct f2fs_io_info *fio,
 749                                         block_t last_blkaddr,
 750                                         block_t cur_blkaddr)
 751 {
 752         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 753                 return false;
 754         return io_type_is_mergeable(io, fio);
 755 }
 756
 757 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 758                                 struct page *page, enum temp_type temp)
 759 {
 760         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 761         struct bio_entry *be;
 762
 763         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 764         be->bio = bio;
 765         bio_get(bio);
 766
 767         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 768                 f2fs_bug_on(sbi, 1);
 769
 770         f2fs_down_write(&io->bio_list_lock);
 771         list_add_tail(&be->list, &io->bio_list);
 772         f2fs_up_write(&io->bio_list_lock);
 773 }
 774
 775 static void del_bio_entry(struct bio_entry *be)
 776 {
 777         list_del(&be->list);
 778         kmem_cache_free(bio_entry_slab, be);
 779 }
 780
 781 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 782                                                         struct page *page)
 783 {
 784         struct f2fs_sb_info *sbi = fio->sbi;
 785         enum temp_type temp;
 786         bool found = false;
 787         int ret = -EAGAIN;
 788
 789         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 790                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 791                 struct list_head *head = &io->bio_list;
 792                 struct bio_entry *be;
 793
 794                 f2fs_down_write(&io->bio_list_lock);
 795                 list_for_each_entry(be, head, list) {
 796                         if (be->bio != *bio)
 797                                 continue;
 798
 799                         found = true;
 800
 801                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 802                                                             *fio->last_block,
 803                                                             fio->new_blkaddr));
 804                         if (f2fs_crypt_mergeable_bio(*bio,
 805                                         fio->page->mapping->host,
 806                                         fio->page->index, fio) &&
 807                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 808                                         PAGE_SIZE) {
 809                                 ret = 0;
 810                                 break;
 811                         }
 812
 813                         /* page can't be merged into bio; submit the bio */
 814                         del_bio_entry(be);
 815                         f2fs_submit_write_bio(sbi, *bio, DATA);
 816                         break;
 817                 }
 818                 f2fs_up_write(&io->bio_list_lock);
 819         }
 820
 821         if (ret) {
 822                 bio_put(*bio);
 823                 *bio = NULL;
 824         }
 825
 826         return ret;
 827 }
 828
 829 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 830                                         struct bio **bio, struct page *page)
 831 {
 832         enum temp_type temp;
 833         bool found = false;
 834         struct bio *target = bio ? *bio : NULL;
 835
 836         f2fs_bug_on(sbi, !target && !page);
 837
 838         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 839                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 840                 struct list_head *head = &io->bio_list;
 841                 struct bio_entry *be;
 842
 843                 if (list_empty(head))
 844                         continue;
 845
 846                 f2fs_down_read(&io->bio_list_lock);
 847                 list_for_each_entry(be, head, list) {
 848                         if (target)
 849                                 found = (target == be->bio);
 850                         else
 851                                 found = __has_merged_page(be->bio, NULL,
 852                                                                 page, 0);
 853                         if (found)
 854                                 break;
 855                 }
 856                 f2fs_up_read(&io->bio_list_lock);
 857
 858                 if (!found)
 859                         continue;
 860
 861                 found = false;
 862
 863                 f2fs_down_write(&io->bio_list_lock);
 864                 list_for_each_entry(be, head, list) {
 865                         if (target)
 866                                 found = (target == be->bio);
 867                         else
 868                                 found = __has_merged_page(be->bio, NULL,
 869                                                                 page, 0);
 870                         if (found) {
 871                                 target = be->bio;
 872                                 del_bio_entry(be);
 873                                 break;
 874                         }
 875                 }
 876                 f2fs_up_write(&io->bio_list_lock);
 877         }
 878
 879         if (found)
 880                 f2fs_submit_write_bio(sbi, target, DATA);
 881         if (bio && *bio) {
 882                 bio_put(*bio);
 883                 *bio = NULL;
 884         }
 885 }
 886
 887 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 888 {
 889         struct bio *bio = *fio->bio;
 890         struct page *page = fio->encrypted_page ?
 891                         fio->encrypted_page : fio->page;
 892
 893         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 894                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 895                 return -EFSCORRUPTED;
 896
 897         trace_f2fs_submit_page_bio(page, fio);
 898
 899         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 900                                                 fio->new_blkaddr))
 901                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 902 alloc_new:
 903         if (!bio) {
 904                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 905                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 906                                        fio->page->index, fio, GFP_NOIO);
 907
 908                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 909         } else {
 910                 if (add_ipu_page(fio, &bio, page))
 911                         goto alloc_new;
 912         }
 913
 914         if (fio->io_wbc)
 915                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 916
 917         inc_page_count(fio->sbi, WB_DATA_TYPE(page, false));
 918
 919         *fio->last_block = fio->new_blkaddr;
 920         *fio->bio = bio;
 921
 922         return 0;
 923 }
 924
 925 #ifdef CONFIG_BLK_DEV_ZONED
 926 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
 927 {
 928         int devi = 0;
 929
 930         if (f2fs_is_multi_device(sbi)) {
 931                 devi = f2fs_target_device_index(sbi, blkaddr);
 932                 if (blkaddr < FDEV(devi).start_blk ||
 933                     blkaddr > FDEV(devi).end_blk) {
 934                         f2fs_err(sbi, "Invalid block %x", blkaddr);
 935                         return false;
 936                 }
 937                 blkaddr -= FDEV(devi).start_blk;
 938         }
 939         return bdev_is_zoned(FDEV(devi).bdev) &&
 940                 f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
 941                 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
 942 }
 943 #endif
 944
 945 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 946 {
 947         struct f2fs_sb_info *sbi = fio->sbi;
 948         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 949         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 950         struct page *bio_page;
 951         enum count_type type;
 952
 953         f2fs_bug_on(sbi, is_read_io(fio->op));
 954
 955         f2fs_down_write(&io->io_rwsem);
 956 next:
 957 #ifdef CONFIG_BLK_DEV_ZONED
 958         if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
 959                 wait_for_completion_io(&io->zone_wait);
 960                 bio_put(io->zone_pending_bio);
 961                 io->zone_pending_bio = NULL;
 962                 io->bi_private = NULL;
 963         }
 964 #endif
 965
 966         if (fio->in_list) {
 967                 spin_lock(&io->io_lock);
 968                 if (list_empty(&io->io_list)) {
 969                         spin_unlock(&io->io_lock);
 970                         goto out;
 971                 }
 972                 fio = list_first_entry(&io->io_list,
 973                                                 struct f2fs_io_info, list);
 974                 list_del(&fio->list);
 975                 spin_unlock(&io->io_lock);
 976         }
 977
 978         verify_fio_blkaddr(fio);
 979
 980         if (fio->encrypted_page)
 981                 bio_page = fio->encrypted_page;
 982         else if (fio->compressed_page)
 983                 bio_page = fio->compressed_page;
 984         else
 985                 bio_page = fio->page;
 986
 987         /* set submitted = true as a return value */
 988         fio->submitted = 1;
 989
 990         type = WB_DATA_TYPE(bio_page, fio->compressed_page);
 991         inc_page_count(sbi, type);
 992
 993         if (io->bio &&
 994             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 995                               fio->new_blkaddr) ||
 996              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 997                                        bio_page->index, fio)))
 998                 __submit_merged_bio(io);
 999 alloc_new:
1000         if (io->bio == NULL) {
1001                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1002                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1003                                        bio_page->index, fio, GFP_NOIO);
1004                 io->fio = *fio;
1005         }
1006
1007         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1008                 __submit_merged_bio(io);
1009                 goto alloc_new;
1010         }
1011
1012         if (fio->io_wbc)
1013                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1014
1015         io->last_block_in_bio = fio->new_blkaddr;
1016
1017         trace_f2fs_submit_page_write(fio->page, fio);
1018 #ifdef CONFIG_BLK_DEV_ZONED
1019         if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1020                         is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1021                 bio_get(io->bio);
1022                 reinit_completion(&io->zone_wait);
1023                 io->bi_private = io->bio->bi_private;
1024                 io->bio->bi_private = io;
1025                 io->bio->bi_end_io = f2fs_zone_write_end_io;
1026                 io->zone_pending_bio = io->bio;
1027                 __submit_merged_bio(io);
1028         }
1029 #endif
1030         if (fio->in_list)
1031                 goto next;
1032 out:
1033         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1034                                 !f2fs_is_checkpoint_ready(sbi))
1035                 __submit_merged_bio(io);
1036         f2fs_up_write(&io->io_rwsem);
1037 }
1038
1039 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1040                                       unsigned nr_pages, blk_opf_t op_flag,
1041                                       pgoff_t first_idx, bool for_write)
1042 {
1043         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1044         struct bio *bio;
1045         struct bio_post_read_ctx *ctx = NULL;
1046         unsigned int post_read_steps = 0;
1047         sector_t sector;
1048         struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
1049
1050         bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1051                                REQ_OP_READ | op_flag,
1052                                for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1053         if (!bio)
1054                 return ERR_PTR(-ENOMEM);
1055         bio->bi_iter.bi_sector = sector;
1056         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1057         bio->bi_end_io = f2fs_read_end_io;
1058
1059         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1060                 post_read_steps |= STEP_DECRYPT;
1061
1062         if (f2fs_need_verity(inode, first_idx))
1063                 post_read_steps |= STEP_VERITY;
1064
1065         /*
1066          * STEP_DECOMPRESS is handled specially, since a compressed file might
1067          * contain both compressed and uncompressed clusters.  We'll allocate a
1068          * bio_post_read_ctx if the file is compressed, but the caller is
1069          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1070          */
1071
1072         if (post_read_steps || f2fs_compressed_file(inode)) {
1073                 /* Due to the mempool, this never fails. */
1074                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1075                 ctx->bio = bio;
1076                 ctx->sbi = sbi;
1077                 ctx->enabled_steps = post_read_steps;
1078                 ctx->fs_blkaddr = blkaddr;
1079                 ctx->decompression_attempted = false;
1080                 bio->bi_private = ctx;
1081         }
1082         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1083
1084         return bio;
1085 }
1086
1087 /* This can handle encryption stuffs */
1088 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1089                                  block_t blkaddr, blk_opf_t op_flags,
1090                                  bool for_write)
1091 {
1092         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1093         struct bio *bio;
1094
1095         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1096                                         page->index, for_write);
1097         if (IS_ERR(bio))
1098                 return PTR_ERR(bio);
1099
1100         /* wait for GCed page writeback via META_MAPPING */
1101         f2fs_wait_on_block_writeback(inode, blkaddr);
1102
1103         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1104                 iostat_update_and_unbind_ctx(bio);
1105                 if (bio->bi_private)
1106                         mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1107                 bio_put(bio);
1108                 return -EFAULT;
1109         }
1110         inc_page_count(sbi, F2FS_RD_DATA);
1111         f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1112         f2fs_submit_read_bio(sbi, bio, DATA);
1113         return 0;
1114 }
1115
1116 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1117 {
1118         __le32 *addr = get_dnode_addr(dn->inode, dn->node_page);
1119
1120         dn->data_blkaddr = blkaddr;
1121         addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1122 }
1123
1124 /*
1125  * Lock ordering for the change of data block address:
1126  * ->data_page
1127  *  ->node_page
1128  *    update block addresses in the node page
1129  */
1130 void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1131 {
1132         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1133         __set_data_blkaddr(dn, blkaddr);
1134         if (set_page_dirty(dn->node_page))
1135                 dn->node_changed = true;
1136 }
1137
1138 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1139 {
1140         f2fs_set_data_blkaddr(dn, blkaddr);
1141         f2fs_update_read_extent_cache(dn);
1142 }
1143
1144 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1145 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1146 {
1147         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1148         int err;
1149
1150         if (!count)
1151                 return 0;
1152
1153         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1154                 return -EPERM;
1155         err = inc_valid_block_count(sbi, dn->inode, &count, true);
1156         if (unlikely(err))
1157                 return err;
1158
1159         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1160                                                 dn->ofs_in_node, count);
1161
1162         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1163
1164         for (; count > 0; dn->ofs_in_node++) {
1165                 block_t blkaddr = f2fs_data_blkaddr(dn);
1166
1167                 if (blkaddr == NULL_ADDR) {
1168                         __set_data_blkaddr(dn, NEW_ADDR);
1169                         count--;
1170                 }
1171         }
1172
1173         if (set_page_dirty(dn->node_page))
1174                 dn->node_changed = true;
1175         return 0;
1176 }
1177
1178 /* Should keep dn->ofs_in_node unchanged */
1179 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1180 {
1181         unsigned int ofs_in_node = dn->ofs_in_node;
1182         int ret;
1183
1184         ret = f2fs_reserve_new_blocks(dn, 1);
1185         dn->ofs_in_node = ofs_in_node;
1186         return ret;
1187 }
1188
1189 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1190 {
1191         bool need_put = dn->inode_page ? false : true;
1192         int err;
1193
1194         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1195         if (err)
1196                 return err;
1197
1198         if (dn->data_blkaddr == NULL_ADDR)
1199                 err = f2fs_reserve_new_block(dn);
1200         if (err || need_put)
1201                 f2fs_put_dnode(dn);
1202         return err;
1203 }
1204
1205 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1206                                      blk_opf_t op_flags, bool for_write,
1207                                      pgoff_t *next_pgofs)
1208 {
1209         struct address_space *mapping = inode->i_mapping;
1210         struct dnode_of_data dn;
1211         struct page *page;
1212         int err;
1213
1214         page = f2fs_grab_cache_page(mapping, index, for_write);
1215         if (!page)
1216                 return ERR_PTR(-ENOMEM);
1217
1218         if (f2fs_lookup_read_extent_cache_block(inode, index,
1219                                                 &dn.data_blkaddr)) {
1220                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1221                                                 DATA_GENERIC_ENHANCE_READ)) {
1222                         err = -EFSCORRUPTED;
1223                         goto put_err;
1224                 }
1225                 goto got_it;
1226         }
1227
1228         set_new_dnode(&dn, inode, NULL, NULL, 0);
1229         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1230         if (err) {
1231                 if (err == -ENOENT && next_pgofs)
1232                         *next_pgofs = f2fs_get_next_page_offset(&dn, index);
1233                 goto put_err;
1234         }
1235         f2fs_put_dnode(&dn);
1236
1237         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1238                 err = -ENOENT;
1239                 if (next_pgofs)
1240                         *next_pgofs = index + 1;
1241                 goto put_err;
1242         }
1243         if (dn.data_blkaddr != NEW_ADDR &&
1244                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1245                                                 dn.data_blkaddr,
1246                                                 DATA_GENERIC_ENHANCE)) {
1247                 err = -EFSCORRUPTED;
1248                 goto put_err;
1249         }
1250 got_it:
1251         if (PageUptodate(page)) {
1252                 unlock_page(page);
1253                 return page;
1254         }
1255
1256         /*
1257          * A new dentry page is allocated but not able to be written, since its
1258          * new inode page couldn't be allocated due to -ENOSPC.
1259          * In such the case, its blkaddr can be remained as NEW_ADDR.
1260          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1261          * f2fs_init_inode_metadata.
1262          */
1263         if (dn.data_blkaddr == NEW_ADDR) {
1264                 zero_user_segment(page, 0, PAGE_SIZE);
1265                 if (!PageUptodate(page))
1266                         SetPageUptodate(page);
1267                 unlock_page(page);
1268                 return page;
1269         }
1270
1271         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1272                                                 op_flags, for_write);
1273         if (err)
1274                 goto put_err;
1275         return page;
1276
1277 put_err:
1278         f2fs_put_page(page, 1);
1279         return ERR_PTR(err);
1280 }
1281
1282 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1283                                         pgoff_t *next_pgofs)
1284 {
1285         struct address_space *mapping = inode->i_mapping;
1286         struct page *page;
1287
1288         page = find_get_page(mapping, index);
1289         if (page && PageUptodate(page))
1290                 return page;
1291         f2fs_put_page(page, 0);
1292
1293         page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1294         if (IS_ERR(page))
1295                 return page;
1296
1297         if (PageUptodate(page))
1298                 return page;
1299
1300         wait_on_page_locked(page);
1301         if (unlikely(!PageUptodate(page))) {
1302                 f2fs_put_page(page, 0);
1303                 return ERR_PTR(-EIO);
1304         }
1305         return page;
1306 }
1307
1308 /*
1309  * If it tries to access a hole, return an error.
1310  * Because, the callers, functions in dir.c and GC, should be able to know
1311  * whether this page exists or not.
1312  */
1313 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1314                                                         bool for_write)
1315 {
1316         struct address_space *mapping = inode->i_mapping;
1317         struct page *page;
1318
1319         page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1320         if (IS_ERR(page))
1321                 return page;
1322
1323         /* wait for read completion */
1324         lock_page(page);
1325         if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1326                 f2fs_put_page(page, 1);
1327                 return ERR_PTR(-EIO);
1328         }
1329         return page;
1330 }
1331
1332 /*
1333  * Caller ensures that this data page is never allocated.
1334  * A new zero-filled data page is allocated in the page cache.
1335  *
1336  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1337  * f2fs_unlock_op().
1338  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1339  * ipage should be released by this function.
1340  */
1341 struct page *f2fs_get_new_data_page(struct inode *inode,
1342                 struct page *ipage, pgoff_t index, bool new_i_size)
1343 {
1344         struct address_space *mapping = inode->i_mapping;
1345         struct page *page;
1346         struct dnode_of_data dn;
1347         int err;
1348
1349         page = f2fs_grab_cache_page(mapping, index, true);
1350         if (!page) {
1351                 /*
1352                  * before exiting, we should make sure ipage will be released
1353                  * if any error occur.
1354                  */
1355                 f2fs_put_page(ipage, 1);
1356                 return ERR_PTR(-ENOMEM);
1357         }
1358
1359         set_new_dnode(&dn, inode, ipage, NULL, 0);
1360         err = f2fs_reserve_block(&dn, index);
1361         if (err) {
1362                 f2fs_put_page(page, 1);
1363                 return ERR_PTR(err);
1364         }
1365         if (!ipage)
1366                 f2fs_put_dnode(&dn);
1367
1368         if (PageUptodate(page))
1369                 goto got_it;
1370
1371         if (dn.data_blkaddr == NEW_ADDR) {
1372                 zero_user_segment(page, 0, PAGE_SIZE);
1373                 if (!PageUptodate(page))
1374                         SetPageUptodate(page);
1375         } else {
1376                 f2fs_put_page(page, 1);
1377
1378                 /* if ipage exists, blkaddr should be NEW_ADDR */
1379                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1380                 page = f2fs_get_lock_data_page(inode, index, true);
1381                 if (IS_ERR(page))
1382                         return page;
1383         }
1384 got_it:
1385         if (new_i_size && i_size_read(inode) <
1386                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1387                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1388         return page;
1389 }
1390
1391 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1392 {
1393         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1394         struct f2fs_summary sum;
1395         struct node_info ni;
1396         block_t old_blkaddr;
1397         blkcnt_t count = 1;
1398         int err;
1399
1400         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1401                 return -EPERM;
1402
1403         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1404         if (err)
1405                 return err;
1406
1407         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1408         if (dn->data_blkaddr == NULL_ADDR) {
1409                 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1410                 if (unlikely(err))
1411                         return err;
1412         }
1413
1414         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1415         old_blkaddr = dn->data_blkaddr;
1416         err = f2fs_allocate_data_block(sbi, NULL, old_blkaddr,
1417                                 &dn->data_blkaddr, &sum, seg_type, NULL);
1418         if (err)
1419                 return err;
1420
1421         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1422                 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
1423
1424         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1425         return 0;
1426 }
1427
1428 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1429 {
1430         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1431                 f2fs_down_read(&sbi->node_change);
1432         else
1433                 f2fs_lock_op(sbi);
1434 }
1435
1436 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1437 {
1438         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1439                 f2fs_up_read(&sbi->node_change);
1440         else
1441                 f2fs_unlock_op(sbi);
1442 }
1443
1444 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1445 {
1446         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1447         int err = 0;
1448
1449         f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1450         if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1451                                                 &dn->data_blkaddr))
1452                 err = f2fs_reserve_block(dn, index);
1453         f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1454
1455         return err;
1456 }
1457
1458 static int f2fs_map_no_dnode(struct inode *inode,
1459                 struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1460                 pgoff_t pgoff)
1461 {
1462         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1463
1464         /*
1465          * There is one exceptional case that read_node_page() may return
1466          * -ENOENT due to filesystem has been shutdown or cp_error, return
1467          * -EIO in that case.
1468          */
1469         if (map->m_may_create &&
1470             (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1471                 return -EIO;
1472
1473         if (map->m_next_pgofs)
1474                 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1475         if (map->m_next_extent)
1476                 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1477         return 0;
1478 }
1479
1480 static bool f2fs_map_blocks_cached(struct inode *inode,
1481                 struct f2fs_map_blocks *map, int flag)
1482 {
1483         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1484         unsigned int maxblocks = map->m_len;
1485         pgoff_t pgoff = (pgoff_t)map->m_lblk;
1486         struct extent_info ei = {};
1487
1488         if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1489                 return false;
1490
1491         map->m_pblk = ei.blk + pgoff - ei.fofs;
1492         map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1493         map->m_flags = F2FS_MAP_MAPPED;
1494         if (map->m_next_extent)
1495                 *map->m_next_extent = pgoff + map->m_len;
1496
1497         /* for hardware encryption, but to avoid potential issue in future */
1498         if (flag == F2FS_GET_BLOCK_DIO)
1499                 f2fs_wait_on_block_writeback_range(inode,
1500                                         map->m_pblk, map->m_len);
1501
1502         if (f2fs_allow_multi_device_dio(sbi, flag)) {
1503                 int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1504                 struct f2fs_dev_info *dev = &sbi->devs[bidx];
1505
1506                 map->m_bdev = dev->bdev;
1507                 map->m_pblk -= dev->start_blk;
1508                 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1509         } else {
1510                 map->m_bdev = inode->i_sb->s_bdev;
1511         }
1512         return true;
1513 }
1514
1515 static bool map_is_mergeable(struct f2fs_sb_info *sbi,
1516                                 struct f2fs_map_blocks *map,
1517                                 block_t blkaddr, int flag, int bidx,
1518                                 int ofs)
1519 {
1520         if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1521                 return false;
1522         if (map->m_pblk != NEW_ADDR && blkaddr == (map->m_pblk + ofs))
1523                 return true;
1524         if (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR)
1525                 return true;
1526         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1527                 return true;
1528         if (flag == F2FS_GET_BLOCK_DIO &&
1529                 map->m_pblk == NULL_ADDR && blkaddr == NULL_ADDR)
1530                 return true;
1531         return false;
1532 }
1533
1534 /*
1535  * f2fs_map_blocks() tries to find or build mapping relationship which
1536  * maps continuous logical blocks to physical blocks, and return such
1537  * info via f2fs_map_blocks structure.
1538  */
1539 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1540 {
1541         unsigned int maxblocks = map->m_len;
1542         struct dnode_of_data dn;
1543         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1544         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1545         pgoff_t pgofs, end_offset, end;
1546         int err = 0, ofs = 1;
1547         unsigned int ofs_in_node, last_ofs_in_node;
1548         blkcnt_t prealloc;
1549         block_t blkaddr;
1550         unsigned int start_pgofs;
1551         int bidx = 0;
1552         bool is_hole;
1553
1554         if (!maxblocks)
1555                 return 0;
1556
1557         if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1558                 goto out;
1559
1560         map->m_bdev = inode->i_sb->s_bdev;
1561         map->m_multidev_dio =
1562                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1563
1564         map->m_len = 0;
1565         map->m_flags = 0;
1566
1567         /* it only supports block size == page size */
1568         pgofs = (pgoff_t)map->m_lblk;
1569         end = pgofs + maxblocks;
1570
1571 next_dnode:
1572         if (map->m_may_create)
1573                 f2fs_map_lock(sbi, flag);
1574
1575         /* When reading holes, we need its node page */
1576         set_new_dnode(&dn, inode, NULL, NULL, 0);
1577         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1578         if (err) {
1579                 if (flag == F2FS_GET_BLOCK_BMAP)
1580                         map->m_pblk = 0;
1581                 if (err == -ENOENT)
1582                         err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1583                 goto unlock_out;
1584         }
1585
1586         start_pgofs = pgofs;
1587         prealloc = 0;
1588         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1589         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1590
1591 next_block:
1592         blkaddr = f2fs_data_blkaddr(&dn);
1593         is_hole = !__is_valid_data_blkaddr(blkaddr);
1594         if (!is_hole &&
1595             !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1596                 err = -EFSCORRUPTED;
1597                 goto sync_out;
1598         }
1599
1600         /* use out-place-update for direct IO under LFS mode */
1601         if (map->m_may_create && (is_hole ||
1602                 (flag == F2FS_GET_BLOCK_DIO && f2fs_lfs_mode(sbi) &&
1603                 !f2fs_is_pinned_file(inode)))) {
1604                 if (unlikely(f2fs_cp_error(sbi))) {
1605                         err = -EIO;
1606                         goto sync_out;
1607                 }
1608
1609                 switch (flag) {
1610                 case F2FS_GET_BLOCK_PRE_AIO:
1611                         if (blkaddr == NULL_ADDR) {
1612                                 prealloc++;
1613                                 last_ofs_in_node = dn.ofs_in_node;
1614                         }
1615                         break;
1616                 case F2FS_GET_BLOCK_PRE_DIO:
1617                 case F2FS_GET_BLOCK_DIO:
1618                         err = __allocate_data_block(&dn, map->m_seg_type);
1619                         if (err)
1620                                 goto sync_out;
1621                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1622                                 file_need_truncate(inode);
1623                         set_inode_flag(inode, FI_APPEND_WRITE);
1624                         break;
1625                 default:
1626                         WARN_ON_ONCE(1);
1627                         err = -EIO;
1628                         goto sync_out;
1629                 }
1630
1631                 blkaddr = dn.data_blkaddr;
1632                 if (is_hole)
1633                         map->m_flags |= F2FS_MAP_NEW;
1634         } else if (is_hole) {
1635                 if (f2fs_compressed_file(inode) &&
1636                     f2fs_sanity_check_cluster(&dn)) {
1637                         err = -EFSCORRUPTED;
1638                         f2fs_handle_error(sbi,
1639                                         ERROR_CORRUPTED_CLUSTER);
1640                         goto sync_out;
1641                 }
1642
1643                 switch (flag) {
1644                 case F2FS_GET_BLOCK_PRECACHE:
1645                         goto sync_out;
1646                 case F2FS_GET_BLOCK_BMAP:
1647                         map->m_pblk = 0;
1648                         goto sync_out;
1649                 case F2FS_GET_BLOCK_FIEMAP:
1650                         if (blkaddr == NULL_ADDR) {
1651                                 if (map->m_next_pgofs)
1652                                         *map->m_next_pgofs = pgofs + 1;
1653                                 goto sync_out;
1654                         }
1655                         break;
1656                 case F2FS_GET_BLOCK_DIO:
1657                         if (map->m_next_pgofs)
1658                                 *map->m_next_pgofs = pgofs + 1;
1659                         break;
1660                 default:
1661                         /* for defragment case */
1662                         if (map->m_next_pgofs)
1663                                 *map->m_next_pgofs = pgofs + 1;
1664                         goto sync_out;
1665                 }
1666         }
1667
1668         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1669                 goto skip;
1670
1671         if (map->m_multidev_dio)
1672                 bidx = f2fs_target_device_index(sbi, blkaddr);
1673
1674         if (map->m_len == 0) {
1675                 /* reserved delalloc block should be mapped for fiemap. */
1676                 if (blkaddr == NEW_ADDR)
1677                         map->m_flags |= F2FS_MAP_DELALLOC;
1678                 if (flag != F2FS_GET_BLOCK_DIO || !is_hole)
1679                         map->m_flags |= F2FS_MAP_MAPPED;
1680
1681                 map->m_pblk = blkaddr;
1682                 map->m_len = 1;
1683
1684                 if (map->m_multidev_dio)
1685                         map->m_bdev = FDEV(bidx).bdev;
1686         } else if (map_is_mergeable(sbi, map, blkaddr, flag, bidx, ofs)) {
1687                 ofs++;
1688                 map->m_len++;
1689         } else {
1690                 goto sync_out;
1691         }
1692
1693 skip:
1694         dn.ofs_in_node++;
1695         pgofs++;
1696
1697         /* preallocate blocks in batch for one dnode page */
1698         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1699                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1700
1701                 dn.ofs_in_node = ofs_in_node;
1702                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1703                 if (err)
1704                         goto sync_out;
1705
1706                 map->m_len += dn.ofs_in_node - ofs_in_node;
1707                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1708                         err = -ENOSPC;
1709                         goto sync_out;
1710                 }
1711                 dn.ofs_in_node = end_offset;
1712         }
1713
1714         if (pgofs >= end)
1715                 goto sync_out;
1716         else if (dn.ofs_in_node < end_offset)
1717                 goto next_block;
1718
1719         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1720                 if (map->m_flags & F2FS_MAP_MAPPED) {
1721                         unsigned int ofs = start_pgofs - map->m_lblk;
1722
1723                         f2fs_update_read_extent_cache_range(&dn,
1724                                 start_pgofs, map->m_pblk + ofs,
1725                                 map->m_len - ofs);
1726                 }
1727         }
1728
1729         f2fs_put_dnode(&dn);
1730
1731         if (map->m_may_create) {
1732                 f2fs_map_unlock(sbi, flag);
1733                 f2fs_balance_fs(sbi, dn.node_changed);
1734         }
1735         goto next_dnode;
1736
1737 sync_out:
1738
1739         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1740                 /*
1741                  * for hardware encryption, but to avoid potential issue
1742                  * in future
1743                  */
1744                 f2fs_wait_on_block_writeback_range(inode,
1745                                                 map->m_pblk, map->m_len);
1746
1747                 if (map->m_multidev_dio) {
1748                         block_t blk_addr = map->m_pblk;
1749
1750                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1751
1752                         map->m_bdev = FDEV(bidx).bdev;
1753                         map->m_pblk -= FDEV(bidx).start_blk;
1754
1755                         if (map->m_may_create)
1756                                 f2fs_update_device_state(sbi, inode->i_ino,
1757                                                         blk_addr, map->m_len);
1758
1759                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1760                                                 FDEV(bidx).end_blk + 1);
1761                 }
1762         }
1763
1764         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1765                 if (map->m_flags & F2FS_MAP_MAPPED) {
1766                         unsigned int ofs = start_pgofs - map->m_lblk;
1767
1768                         f2fs_update_read_extent_cache_range(&dn,
1769                                 start_pgofs, map->m_pblk + ofs,
1770                                 map->m_len - ofs);
1771                 }
1772                 if (map->m_next_extent)
1773                         *map->m_next_extent = pgofs + 1;
1774         }
1775         f2fs_put_dnode(&dn);
1776 unlock_out:
1777         if (map->m_may_create) {
1778                 f2fs_map_unlock(sbi, flag);
1779                 f2fs_balance_fs(sbi, dn.node_changed);
1780         }
1781 out:
1782         trace_f2fs_map_blocks(inode, map, flag, err);
1783         return err;
1784 }
1785
1786 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1787 {
1788         struct f2fs_map_blocks map;
1789         block_t last_lblk;
1790         int err;
1791
1792         if (pos + len > i_size_read(inode))
1793                 return false;
1794
1795         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1796         map.m_next_pgofs = NULL;
1797         map.m_next_extent = NULL;
1798         map.m_seg_type = NO_CHECK_TYPE;
1799         map.m_may_create = false;
1800         last_lblk = F2FS_BLK_ALIGN(pos + len);
1801
1802         while (map.m_lblk < last_lblk) {
1803                 map.m_len = last_lblk - map.m_lblk;
1804                 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1805                 if (err || map.m_len == 0)
1806                         return false;
1807                 map.m_lblk += map.m_len;
1808         }
1809         return true;
1810 }
1811
1812 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1813 {
1814         return (bytes >> inode->i_blkbits);
1815 }
1816
1817 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1818 {
1819         return (blks << inode->i_blkbits);
1820 }
1821
1822 static int f2fs_xattr_fiemap(struct inode *inode,
1823                                 struct fiemap_extent_info *fieinfo)
1824 {
1825         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1826         struct page *page;
1827         struct node_info ni;
1828         __u64 phys = 0, len;
1829         __u32 flags;
1830         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1831         int err = 0;
1832
1833         if (f2fs_has_inline_xattr(inode)) {
1834                 int offset;
1835
1836                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1837                                                 inode->i_ino, false);
1838                 if (!page)
1839                         return -ENOMEM;
1840
1841                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1842                 if (err) {
1843                         f2fs_put_page(page, 1);
1844                         return err;
1845                 }
1846
1847                 phys = blks_to_bytes(inode, ni.blk_addr);
1848                 offset = offsetof(struct f2fs_inode, i_addr) +
1849                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1850                                         get_inline_xattr_addrs(inode));
1851
1852                 phys += offset;
1853                 len = inline_xattr_size(inode);
1854
1855                 f2fs_put_page(page, 1);
1856
1857                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1858
1859                 if (!xnid)
1860                         flags |= FIEMAP_EXTENT_LAST;
1861
1862                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1863                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1864                 if (err)
1865                         return err;
1866         }
1867
1868         if (xnid) {
1869                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1870                 if (!page)
1871                         return -ENOMEM;
1872
1873                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1874                 if (err) {
1875                         f2fs_put_page(page, 1);
1876                         return err;
1877                 }
1878
1879                 phys = blks_to_bytes(inode, ni.blk_addr);
1880                 len = inode->i_sb->s_blocksize;
1881
1882                 f2fs_put_page(page, 1);
1883
1884                 flags = FIEMAP_EXTENT_LAST;
1885         }
1886
1887         if (phys) {
1888                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1889                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1890         }
1891
1892         return (err < 0 ? err : 0);
1893 }
1894
1895 static loff_t max_inode_blocks(struct inode *inode)
1896 {
1897         loff_t result = ADDRS_PER_INODE(inode);
1898         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1899
1900         /* two direct node blocks */
1901         result += (leaf_count * 2);
1902
1903         /* two indirect node blocks */
1904         leaf_count *= NIDS_PER_BLOCK;
1905         result += (leaf_count * 2);
1906
1907         /* one double indirect node block */
1908         leaf_count *= NIDS_PER_BLOCK;
1909         result += leaf_count;
1910
1911         return result;
1912 }
1913
1914 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1915                 u64 start, u64 len)
1916 {
1917         struct f2fs_map_blocks map;
1918         sector_t start_blk, last_blk;
1919         pgoff_t next_pgofs;
1920         u64 logical = 0, phys = 0, size = 0;
1921         u32 flags = 0;
1922         int ret = 0;
1923         bool compr_cluster = false, compr_appended;
1924         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1925         unsigned int count_in_cluster = 0;
1926         loff_t maxbytes;
1927
1928         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1929                 ret = f2fs_precache_extents(inode);
1930                 if (ret)
1931                         return ret;
1932         }
1933
1934         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1935         if (ret)
1936                 return ret;
1937
1938         inode_lock_shared(inode);
1939
1940         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1941         if (start > maxbytes) {
1942                 ret = -EFBIG;
1943                 goto out;
1944         }
1945
1946         if (len > maxbytes || (maxbytes - len) < start)
1947                 len = maxbytes - start;
1948
1949         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1950                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1951                 goto out;
1952         }
1953
1954         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1955                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1956                 if (ret != -EAGAIN)
1957                         goto out;
1958         }
1959
1960         if (bytes_to_blks(inode, len) == 0)
1961                 len = blks_to_bytes(inode, 1);
1962
1963         start_blk = bytes_to_blks(inode, start);
1964         last_blk = bytes_to_blks(inode, start + len - 1);
1965
1966 next:
1967         memset(&map, 0, sizeof(map));
1968         map.m_lblk = start_blk;
1969         map.m_len = bytes_to_blks(inode, len);
1970         map.m_next_pgofs = &next_pgofs;
1971         map.m_seg_type = NO_CHECK_TYPE;
1972
1973         if (compr_cluster) {
1974                 map.m_lblk += 1;
1975                 map.m_len = cluster_size - count_in_cluster;
1976         }
1977
1978         ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
1979         if (ret)
1980                 goto out;
1981
1982         /* HOLE */
1983         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1984                 start_blk = next_pgofs;
1985
1986                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1987                                                 max_inode_blocks(inode)))
1988                         goto prep_next;
1989
1990                 flags |= FIEMAP_EXTENT_LAST;
1991         }
1992
1993         compr_appended = false;
1994         /* In a case of compressed cluster, append this to the last extent */
1995         if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
1996                         !(map.m_flags & F2FS_MAP_FLAGS))) {
1997                 compr_appended = true;
1998                 goto skip_fill;
1999         }
2000
2001         if (size) {
2002                 flags |= FIEMAP_EXTENT_MERGED;
2003                 if (IS_ENCRYPTED(inode))
2004                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2005
2006                 ret = fiemap_fill_next_extent(fieinfo, logical,
2007                                 phys, size, flags);
2008                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2009                 if (ret)
2010                         goto out;
2011                 size = 0;
2012         }
2013
2014         if (start_blk > last_blk)
2015                 goto out;
2016
2017 skip_fill:
2018         if (map.m_pblk == COMPRESS_ADDR) {
2019                 compr_cluster = true;
2020                 count_in_cluster = 1;
2021         } else if (compr_appended) {
2022                 unsigned int appended_blks = cluster_size -
2023                                                 count_in_cluster + 1;
2024                 size += blks_to_bytes(inode, appended_blks);
2025                 start_blk += appended_blks;
2026                 compr_cluster = false;
2027         } else {
2028                 logical = blks_to_bytes(inode, start_blk);
2029                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2030                         blks_to_bytes(inode, map.m_pblk) : 0;
2031                 size = blks_to_bytes(inode, map.m_len);
2032                 flags = 0;
2033
2034                 if (compr_cluster) {
2035                         flags = FIEMAP_EXTENT_ENCODED;
2036                         count_in_cluster += map.m_len;
2037                         if (count_in_cluster == cluster_size) {
2038                                 compr_cluster = false;
2039                                 size += blks_to_bytes(inode, 1);
2040                         }
2041                 } else if (map.m_flags & F2FS_MAP_DELALLOC) {
2042                         flags = FIEMAP_EXTENT_UNWRITTEN;
2043                 }
2044
2045                 start_blk += bytes_to_blks(inode, size);
2046         }
2047
2048 prep_next:
2049         cond_resched();
2050         if (fatal_signal_pending(current))
2051                 ret = -EINTR;
2052         else
2053                 goto next;
2054 out:
2055         if (ret == 1)
2056                 ret = 0;
2057
2058         inode_unlock_shared(inode);
2059         return ret;
2060 }
2061
2062 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2063 {
2064         if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2065                 return inode->i_sb->s_maxbytes;
2066
2067         return i_size_read(inode);
2068 }
2069
2070 static int f2fs_read_single_page(struct inode *inode, struct folio *folio,
2071                                         unsigned nr_pages,
2072                                         struct f2fs_map_blocks *map,
2073                                         struct bio **bio_ret,
2074                                         sector_t *last_block_in_bio,
2075                                         bool is_readahead)
2076 {
2077         struct bio *bio = *bio_ret;
2078         const unsigned blocksize = blks_to_bytes(inode, 1);
2079         sector_t block_in_file;
2080         sector_t last_block;
2081         sector_t last_block_in_file;
2082         sector_t block_nr;
2083         pgoff_t index = folio_index(folio);
2084         int ret = 0;
2085
2086         block_in_file = (sector_t)index;
2087         last_block = block_in_file + nr_pages;
2088         last_block_in_file = bytes_to_blks(inode,
2089                         f2fs_readpage_limit(inode) + blocksize - 1);
2090         if (last_block > last_block_in_file)
2091                 last_block = last_block_in_file;
2092
2093         /* just zeroing out page which is beyond EOF */
2094         if (block_in_file >= last_block)
2095                 goto zero_out;
2096         /*
2097          * Map blocks using the previous result first.
2098          */
2099         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2100                         block_in_file > map->m_lblk &&
2101                         block_in_file < (map->m_lblk + map->m_len))
2102                 goto got_it;
2103
2104         /*
2105          * Then do more f2fs_map_blocks() calls until we are
2106          * done with this page.
2107          */
2108         map->m_lblk = block_in_file;
2109         map->m_len = last_block - block_in_file;
2110
2111         ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2112         if (ret)
2113                 goto out;
2114 got_it:
2115         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2116                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2117                 folio_set_mappedtodisk(folio);
2118
2119                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2120                                                 DATA_GENERIC_ENHANCE_READ)) {
2121                         ret = -EFSCORRUPTED;
2122                         goto out;
2123                 }
2124         } else {
2125 zero_out:
2126                 folio_zero_segment(folio, 0, folio_size(folio));
2127                 if (f2fs_need_verity(inode, index) &&
2128                     !fsverity_verify_folio(folio)) {
2129                         ret = -EIO;
2130                         goto out;
2131                 }
2132                 if (!folio_test_uptodate(folio))
2133                         folio_mark_uptodate(folio);
2134                 folio_unlock(folio);
2135                 goto out;
2136         }
2137
2138         /*
2139          * This page will go to BIO.  Do we need to send this
2140          * BIO off first?
2141          */
2142         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2143                                        *last_block_in_bio, block_nr) ||
2144                     !f2fs_crypt_mergeable_bio(bio, inode, index, NULL))) {
2145 submit_and_realloc:
2146                 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2147                 bio = NULL;
2148         }
2149         if (bio == NULL) {
2150                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2151                                 is_readahead ? REQ_RAHEAD : 0, index,
2152                                 false);
2153                 if (IS_ERR(bio)) {
2154                         ret = PTR_ERR(bio);
2155                         bio = NULL;
2156                         goto out;
2157                 }
2158         }
2159
2160         /*
2161          * If the page is under writeback, we need to wait for
2162          * its completion to see the correct decrypted data.
2163          */
2164         f2fs_wait_on_block_writeback(inode, block_nr);
2165
2166         if (!bio_add_folio(bio, folio, blocksize, 0))
2167                 goto submit_and_realloc;
2168
2169         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2170         f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2171                                                         F2FS_BLKSIZE);
2172         *last_block_in_bio = block_nr;
2173 out:
2174         *bio_ret = bio;
2175         return ret;
2176 }
2177
2178 #ifdef CONFIG_F2FS_FS_COMPRESSION
2179 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2180                                 unsigned nr_pages, sector_t *last_block_in_bio,
2181                                 bool is_readahead, bool for_write)
2182 {
2183         struct dnode_of_data dn;
2184         struct inode *inode = cc->inode;
2185         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2186         struct bio *bio = *bio_ret;
2187         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2188         sector_t last_block_in_file;
2189         const unsigned blocksize = blks_to_bytes(inode, 1);
2190         struct decompress_io_ctx *dic = NULL;
2191         struct extent_info ei = {};
2192         bool from_dnode = true;
2193         int i;
2194         int ret = 0;
2195
2196         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2197
2198         last_block_in_file = bytes_to_blks(inode,
2199                         f2fs_readpage_limit(inode) + blocksize - 1);
2200
2201         /* get rid of pages beyond EOF */
2202         for (i = 0; i < cc->cluster_size; i++) {
2203                 struct page *page = cc->rpages[i];
2204
2205                 if (!page)
2206                         continue;
2207                 if ((sector_t)page->index >= last_block_in_file) {
2208                         zero_user_segment(page, 0, PAGE_SIZE);
2209                         if (!PageUptodate(page))
2210                                 SetPageUptodate(page);
2211                 } else if (!PageUptodate(page)) {
2212                         continue;
2213                 }
2214                 unlock_page(page);
2215                 if (for_write)
2216                         put_page(page);
2217                 cc->rpages[i] = NULL;
2218                 cc->nr_rpages--;
2219         }
2220
2221         /* we are done since all pages are beyond EOF */
2222         if (f2fs_cluster_is_empty(cc))
2223                 goto out;
2224
2225         if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2226                 from_dnode = false;
2227
2228         if (!from_dnode)
2229                 goto skip_reading_dnode;
2230
2231         set_new_dnode(&dn, inode, NULL, NULL, 0);
2232         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2233         if (ret)
2234                 goto out;
2235
2236         if (unlikely(f2fs_cp_error(sbi))) {
2237                 ret = -EIO;
2238                 goto out_put_dnode;
2239         }
2240         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2241
2242 skip_reading_dnode:
2243         for (i = 1; i < cc->cluster_size; i++) {
2244                 block_t blkaddr;
2245
2246                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2247                                         dn.ofs_in_node + i) :
2248                                         ei.blk + i - 1;
2249
2250                 if (!__is_valid_data_blkaddr(blkaddr))
2251                         break;
2252
2253                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2254                         ret = -EFAULT;
2255                         goto out_put_dnode;
2256                 }
2257                 cc->nr_cpages++;
2258
2259                 if (!from_dnode && i >= ei.c_len)
2260                         break;
2261         }
2262
2263         /* nothing to decompress */
2264         if (cc->nr_cpages == 0) {
2265                 ret = 0;
2266                 goto out_put_dnode;
2267         }
2268
2269         dic = f2fs_alloc_dic(cc);
2270         if (IS_ERR(dic)) {
2271                 ret = PTR_ERR(dic);
2272                 goto out_put_dnode;
2273         }
2274
2275         for (i = 0; i < cc->nr_cpages; i++) {
2276                 struct page *page = dic->cpages[i];
2277                 block_t blkaddr;
2278                 struct bio_post_read_ctx *ctx;
2279
2280                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2281                                         dn.ofs_in_node + i + 1) :
2282                                         ei.blk + i;
2283
2284                 f2fs_wait_on_block_writeback(inode, blkaddr);
2285
2286                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2287                         if (atomic_dec_and_test(&dic->remaining_pages)) {
2288                                 f2fs_decompress_cluster(dic, true);
2289                                 break;
2290                         }
2291                         continue;
2292                 }
2293
2294                 if (bio && (!page_is_mergeable(sbi, bio,
2295                                         *last_block_in_bio, blkaddr) ||
2296                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2297 submit_and_realloc:
2298                         f2fs_submit_read_bio(sbi, bio, DATA);
2299                         bio = NULL;
2300                 }
2301
2302                 if (!bio) {
2303                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2304                                         is_readahead ? REQ_RAHEAD : 0,
2305                                         page->index, for_write);
2306                         if (IS_ERR(bio)) {
2307                                 ret = PTR_ERR(bio);
2308                                 f2fs_decompress_end_io(dic, ret, true);
2309                                 f2fs_put_dnode(&dn);
2310                                 *bio_ret = NULL;
2311                                 return ret;
2312                         }
2313                 }
2314
2315                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2316                         goto submit_and_realloc;
2317
2318                 ctx = get_post_read_ctx(bio);
2319                 ctx->enabled_steps |= STEP_DECOMPRESS;
2320                 refcount_inc(&dic->refcnt);
2321
2322                 inc_page_count(sbi, F2FS_RD_DATA);
2323                 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2324                 *last_block_in_bio = blkaddr;
2325         }
2326
2327         if (from_dnode)
2328                 f2fs_put_dnode(&dn);
2329
2330         *bio_ret = bio;
2331         return 0;
2332
2333 out_put_dnode:
2334         if (from_dnode)
2335                 f2fs_put_dnode(&dn);
2336 out:
2337         for (i = 0; i < cc->cluster_size; i++) {
2338                 if (cc->rpages[i]) {
2339                         ClearPageUptodate(cc->rpages[i]);
2340                         unlock_page(cc->rpages[i]);
2341                 }
2342         }
2343         *bio_ret = bio;
2344         return ret;
2345 }
2346 #endif
2347
2348 /*
2349  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2350  * Major change was from block_size == page_size in f2fs by default.
2351  */
2352 static int f2fs_mpage_readpages(struct inode *inode,
2353                 struct readahead_control *rac, struct folio *folio)
2354 {
2355         struct bio *bio = NULL;
2356         sector_t last_block_in_bio = 0;
2357         struct f2fs_map_blocks map;
2358 #ifdef CONFIG_F2FS_FS_COMPRESSION
2359         struct compress_ctx cc = {
2360                 .inode = inode,
2361                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2362                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2363                 .cluster_idx = NULL_CLUSTER,
2364                 .rpages = NULL,
2365                 .cpages = NULL,
2366                 .nr_rpages = 0,
2367                 .nr_cpages = 0,
2368         };
2369         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2370 #endif
2371         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2372         unsigned max_nr_pages = nr_pages;
2373         pgoff_t index;
2374         int ret = 0;
2375
2376         map.m_pblk = 0;
2377         map.m_lblk = 0;
2378         map.m_len = 0;
2379         map.m_flags = 0;
2380         map.m_next_pgofs = NULL;
2381         map.m_next_extent = NULL;
2382         map.m_seg_type = NO_CHECK_TYPE;
2383         map.m_may_create = false;
2384
2385         for (; nr_pages; nr_pages--) {
2386                 if (rac) {
2387                         folio = readahead_folio(rac);
2388                         prefetchw(&folio->flags);
2389                 }
2390
2391                 index = folio_index(folio);
2392
2393 #ifdef CONFIG_F2FS_FS_COMPRESSION
2394                 if (!f2fs_compressed_file(inode))
2395                         goto read_single_page;
2396
2397                 /* there are remained compressed pages, submit them */
2398                 if (!f2fs_cluster_can_merge_page(&cc, index)) {
2399                         ret = f2fs_read_multi_pages(&cc, &bio,
2400                                                 max_nr_pages,
2401                                                 &last_block_in_bio,
2402                                                 rac != NULL, false);
2403                         f2fs_destroy_compress_ctx(&cc, false);
2404                         if (ret)
2405                                 goto set_error_page;
2406                 }
2407                 if (cc.cluster_idx == NULL_CLUSTER) {
2408                         if (nc_cluster_idx == index >> cc.log_cluster_size)
2409                                 goto read_single_page;
2410
2411                         ret = f2fs_is_compressed_cluster(inode, index);
2412                         if (ret < 0)
2413                                 goto set_error_page;
2414                         else if (!ret) {
2415                                 nc_cluster_idx =
2416                                         index >> cc.log_cluster_size;
2417                                 goto read_single_page;
2418                         }
2419
2420                         nc_cluster_idx = NULL_CLUSTER;
2421                 }
2422                 ret = f2fs_init_compress_ctx(&cc);
2423                 if (ret)
2424                         goto set_error_page;
2425
2426                 f2fs_compress_ctx_add_page(&cc, &folio->page);
2427
2428                 goto next_page;
2429 read_single_page:
2430 #endif
2431
2432                 ret = f2fs_read_single_page(inode, folio, max_nr_pages, &map,
2433                                         &bio, &last_block_in_bio, rac);
2434                 if (ret) {
2435 #ifdef CONFIG_F2FS_FS_COMPRESSION
2436 set_error_page:
2437 #endif
2438                         folio_zero_segment(folio, 0, folio_size(folio));
2439                         folio_unlock(folio);
2440                 }
2441 #ifdef CONFIG_F2FS_FS_COMPRESSION
2442 next_page:
2443 #endif
2444
2445 #ifdef CONFIG_F2FS_FS_COMPRESSION
2446                 if (f2fs_compressed_file(inode)) {
2447                         /* last page */
2448                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2449                                 ret = f2fs_read_multi_pages(&cc, &bio,
2450                                                         max_nr_pages,
2451                                                         &last_block_in_bio,
2452                                                         rac != NULL, false);
2453                                 f2fs_destroy_compress_ctx(&cc, false);
2454                         }
2455                 }
2456 #endif
2457         }
2458         if (bio)
2459                 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2460         return ret;
2461 }
2462
2463 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2464 {
2465         struct inode *inode = folio_file_mapping(folio)->host;
2466         int ret = -EAGAIN;
2467
2468         trace_f2fs_readpage(folio, DATA);
2469
2470         if (!f2fs_is_compress_backend_ready(inode)) {
2471                 folio_unlock(folio);
2472                 return -EOPNOTSUPP;
2473         }
2474
2475         /* If the file has inline data, try to read it directly */
2476         if (f2fs_has_inline_data(inode))
2477                 ret = f2fs_read_inline_data(inode, folio);
2478         if (ret == -EAGAIN)
2479                 ret = f2fs_mpage_readpages(inode, NULL, folio);
2480         return ret;
2481 }
2482
2483 static void f2fs_readahead(struct readahead_control *rac)
2484 {
2485         struct inode *inode = rac->mapping->host;
2486
2487         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2488
2489         if (!f2fs_is_compress_backend_ready(inode))
2490                 return;
2491
2492         /* If the file has inline data, skip readahead */
2493         if (f2fs_has_inline_data(inode))
2494                 return;
2495
2496         f2fs_mpage_readpages(inode, rac, NULL);
2497 }
2498
2499 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2500 {
2501         struct inode *inode = fio->page->mapping->host;
2502         struct page *mpage, *page;
2503         gfp_t gfp_flags = GFP_NOFS;
2504
2505         if (!f2fs_encrypted_file(inode))
2506                 return 0;
2507
2508         page = fio->compressed_page ? fio->compressed_page : fio->page;
2509
2510         if (fscrypt_inode_uses_inline_crypto(inode))
2511                 return 0;
2512
2513 retry_encrypt:
2514         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2515                                         PAGE_SIZE, 0, gfp_flags);
2516         if (IS_ERR(fio->encrypted_page)) {
2517                 /* flush pending IOs and wait for a while in the ENOMEM case */
2518                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2519                         f2fs_flush_merged_writes(fio->sbi);
2520                         memalloc_retry_wait(GFP_NOFS);
2521                         gfp_flags |= __GFP_NOFAIL;
2522                         goto retry_encrypt;
2523                 }
2524                 return PTR_ERR(fio->encrypted_page);
2525         }
2526
2527         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2528         if (mpage) {
2529                 if (PageUptodate(mpage))
2530                         memcpy(page_address(mpage),
2531                                 page_address(fio->encrypted_page), PAGE_SIZE);
2532                 f2fs_put_page(mpage, 1);
2533         }
2534         return 0;
2535 }
2536
2537 static inline bool check_inplace_update_policy(struct inode *inode,
2538                                 struct f2fs_io_info *fio)
2539 {
2540         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2541
2542         if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2543             is_inode_flag_set(inode, FI_OPU_WRITE))
2544                 return false;
2545         if (IS_F2FS_IPU_FORCE(sbi))
2546                 return true;
2547         if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2548                 return true;
2549         if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2550                 return true;
2551         if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2552             utilization(sbi) > SM_I(sbi)->min_ipu_util)
2553                 return true;
2554
2555         /*
2556          * IPU for rewrite async pages
2557          */
2558         if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2559             !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2560                 return true;
2561
2562         /* this is only set during fdatasync */
2563         if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2564                 return true;
2565
2566         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2567                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2568                 return true;
2569
2570         return false;
2571 }
2572
2573 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2574 {
2575         /* swap file is migrating in aligned write mode */
2576         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2577                 return false;
2578
2579         if (f2fs_is_pinned_file(inode))
2580                 return true;
2581
2582         /* if this is cold file, we should overwrite to avoid fragmentation */
2583         if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2584                 return true;
2585
2586         return check_inplace_update_policy(inode, fio);
2587 }
2588
2589 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2590 {
2591         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2592
2593         /* The below cases were checked when setting it. */
2594         if (f2fs_is_pinned_file(inode))
2595                 return false;
2596         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2597                 return true;
2598         if (f2fs_lfs_mode(sbi))
2599                 return true;
2600         if (S_ISDIR(inode->i_mode))
2601                 return true;
2602         if (IS_NOQUOTA(inode))
2603                 return true;
2604         if (f2fs_is_atomic_file(inode))
2605                 return true;
2606         /* rewrite low ratio compress data w/ OPU mode to avoid fragmentation */
2607         if (f2fs_compressed_file(inode) &&
2608                 F2FS_OPTION(sbi).compress_mode == COMPR_MODE_USER &&
2609                 is_inode_flag_set(inode, FI_ENABLE_COMPRESS))
2610                 return true;
2611
2612         /* swap file is migrating in aligned write mode */
2613         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2614                 return true;
2615
2616         if (is_inode_flag_set(inode, FI_OPU_WRITE))
2617                 return true;
2618
2619         if (fio) {
2620                 if (page_private_gcing(fio->page))
2621                         return true;
2622                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2623                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2624                         return true;
2625         }
2626         return false;
2627 }
2628
2629 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2630 {
2631         struct inode *inode = fio->page->mapping->host;
2632
2633         if (f2fs_should_update_outplace(inode, fio))
2634                 return false;
2635
2636         return f2fs_should_update_inplace(inode, fio);
2637 }
2638
2639 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2640 {
2641         struct page *page = fio->page;
2642         struct inode *inode = page->mapping->host;
2643         struct dnode_of_data dn;
2644         struct node_info ni;
2645         bool ipu_force = false;
2646         int err = 0;
2647
2648         /* Use COW inode to make dnode_of_data for atomic write */
2649         if (f2fs_is_atomic_file(inode))
2650                 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2651         else
2652                 set_new_dnode(&dn, inode, NULL, NULL, 0);
2653
2654         if (need_inplace_update(fio) &&
2655             f2fs_lookup_read_extent_cache_block(inode, page->index,
2656                                                 &fio->old_blkaddr)) {
2657                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2658                                                 DATA_GENERIC_ENHANCE))
2659                         return -EFSCORRUPTED;
2660
2661                 ipu_force = true;
2662                 fio->need_lock = LOCK_DONE;
2663                 goto got_it;
2664         }
2665
2666         /* Deadlock due to between page->lock and f2fs_lock_op */
2667         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2668                 return -EAGAIN;
2669
2670         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2671         if (err)
2672                 goto out;
2673
2674         fio->old_blkaddr = dn.data_blkaddr;
2675
2676         /* This page is already truncated */
2677         if (fio->old_blkaddr == NULL_ADDR) {
2678                 ClearPageUptodate(page);
2679                 clear_page_private_gcing(page);
2680                 goto out_writepage;
2681         }
2682 got_it:
2683         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2684                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2685                                                 DATA_GENERIC_ENHANCE)) {
2686                 err = -EFSCORRUPTED;
2687                 goto out_writepage;
2688         }
2689
2690         /* wait for GCed page writeback via META_MAPPING */
2691         if (fio->post_read)
2692                 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2693
2694         /*
2695          * If current allocation needs SSR,
2696          * it had better in-place writes for updated data.
2697          */
2698         if (ipu_force ||
2699                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2700                                         need_inplace_update(fio))) {
2701                 err = f2fs_encrypt_one_page(fio);
2702                 if (err)
2703                         goto out_writepage;
2704
2705                 set_page_writeback(page);
2706                 f2fs_put_dnode(&dn);
2707                 if (fio->need_lock == LOCK_REQ)
2708                         f2fs_unlock_op(fio->sbi);
2709                 err = f2fs_inplace_write_data(fio);
2710                 if (err) {
2711                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2712                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2713                         end_page_writeback(page);
2714                 } else {
2715                         set_inode_flag(inode, FI_UPDATE_WRITE);
2716                 }
2717                 trace_f2fs_do_write_data_page(page_folio(page), IPU);
2718                 return err;
2719         }
2720
2721         if (fio->need_lock == LOCK_RETRY) {
2722                 if (!f2fs_trylock_op(fio->sbi)) {
2723                         err = -EAGAIN;
2724                         goto out_writepage;
2725                 }
2726                 fio->need_lock = LOCK_REQ;
2727         }
2728
2729         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2730         if (err)
2731                 goto out_writepage;
2732
2733         fio->version = ni.version;
2734
2735         err = f2fs_encrypt_one_page(fio);
2736         if (err)
2737                 goto out_writepage;
2738
2739         set_page_writeback(page);
2740
2741         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2742                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2743
2744         /* LFS mode write path */
2745         f2fs_outplace_write_data(&dn, fio);
2746         trace_f2fs_do_write_data_page(page_folio(page), OPU);
2747         set_inode_flag(inode, FI_APPEND_WRITE);
2748 out_writepage:
2749         f2fs_put_dnode(&dn);
2750 out:
2751         if (fio->need_lock == LOCK_REQ)
2752                 f2fs_unlock_op(fio->sbi);
2753         return err;
2754 }
2755
2756 int f2fs_write_single_data_page(struct page *page, int *submitted,
2757                                 struct bio **bio,
2758                                 sector_t *last_block,
2759                                 struct writeback_control *wbc,
2760                                 enum iostat_type io_type,
2761                                 int compr_blocks,
2762                                 bool allow_balance)
2763 {
2764         struct inode *inode = page->mapping->host;
2765         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2766         loff_t i_size = i_size_read(inode);
2767         const pgoff_t end_index = ((unsigned long long)i_size)
2768                                                         >> PAGE_SHIFT;
2769         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2770         unsigned offset = 0;
2771         bool need_balance_fs = false;
2772         bool quota_inode = IS_NOQUOTA(inode);
2773         int err = 0;
2774         struct f2fs_io_info fio = {
2775                 .sbi = sbi,
2776                 .ino = inode->i_ino,
2777                 .type = DATA,
2778                 .op = REQ_OP_WRITE,
2779                 .op_flags = wbc_to_write_flags(wbc),
2780                 .old_blkaddr = NULL_ADDR,
2781                 .page = page,
2782                 .encrypted_page = NULL,
2783                 .submitted = 0,
2784                 .compr_blocks = compr_blocks,
2785                 .need_lock = compr_blocks ? LOCK_DONE : LOCK_RETRY,
2786                 .post_read = f2fs_post_read_required(inode) ? 1 : 0,
2787                 .io_type = io_type,
2788                 .io_wbc = wbc,
2789                 .bio = bio,
2790                 .last_block = last_block,
2791         };
2792
2793         trace_f2fs_writepage(page_folio(page), DATA);
2794
2795         /* we should bypass data pages to proceed the kworker jobs */
2796         if (unlikely(f2fs_cp_error(sbi))) {
2797                 mapping_set_error(page->mapping, -EIO);
2798                 /*
2799                  * don't drop any dirty dentry pages for keeping lastest
2800                  * directory structure.
2801                  */
2802                 if (S_ISDIR(inode->i_mode) &&
2803                                 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2804                         goto redirty_out;
2805
2806                 /* keep data pages in remount-ro mode */
2807                 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2808                         goto redirty_out;
2809                 goto out;
2810         }
2811
2812         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2813                 goto redirty_out;
2814
2815         if (page->index < end_index ||
2816                         f2fs_verity_in_progress(inode) ||
2817                         compr_blocks)
2818                 goto write;
2819
2820         /*
2821          * If the offset is out-of-range of file size,
2822          * this page does not have to be written to disk.
2823          */
2824         offset = i_size & (PAGE_SIZE - 1);
2825         if ((page->index >= end_index + 1) || !offset)
2826                 goto out;
2827
2828         zero_user_segment(page, offset, PAGE_SIZE);
2829 write:
2830         /* Dentry/quota blocks are controlled by checkpoint */
2831         if (S_ISDIR(inode->i_mode) || quota_inode) {
2832                 /*
2833                  * We need to wait for node_write to avoid block allocation during
2834                  * checkpoint. This can only happen to quota writes which can cause
2835                  * the below discard race condition.
2836                  */
2837                 if (quota_inode)
2838                         f2fs_down_read(&sbi->node_write);
2839
2840                 fio.need_lock = LOCK_DONE;
2841                 err = f2fs_do_write_data_page(&fio);
2842
2843                 if (quota_inode)
2844                         f2fs_up_read(&sbi->node_write);
2845
2846                 goto done;
2847         }
2848
2849         if (!wbc->for_reclaim)
2850                 need_balance_fs = true;
2851         else if (has_not_enough_free_secs(sbi, 0, 0))
2852                 goto redirty_out;
2853         else
2854                 set_inode_flag(inode, FI_HOT_DATA);
2855
2856         err = -EAGAIN;
2857         if (f2fs_has_inline_data(inode)) {
2858                 err = f2fs_write_inline_data(inode, page);
2859                 if (!err)
2860                         goto out;
2861         }
2862
2863         if (err == -EAGAIN) {
2864                 err = f2fs_do_write_data_page(&fio);
2865                 if (err == -EAGAIN) {
2866                         f2fs_bug_on(sbi, compr_blocks);
2867                         fio.need_lock = LOCK_REQ;
2868                         err = f2fs_do_write_data_page(&fio);
2869                 }
2870         }
2871
2872         if (err) {
2873                 file_set_keep_isize(inode);
2874         } else {
2875                 spin_lock(&F2FS_I(inode)->i_size_lock);
2876                 if (F2FS_I(inode)->last_disk_size < psize)
2877                         F2FS_I(inode)->last_disk_size = psize;
2878                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2879         }
2880
2881 done:
2882         if (err && err != -ENOENT)
2883                 goto redirty_out;
2884
2885 out:
2886         inode_dec_dirty_pages(inode);
2887         if (err) {
2888                 ClearPageUptodate(page);
2889                 clear_page_private_gcing(page);
2890         }
2891
2892         if (wbc->for_reclaim) {
2893                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2894                 clear_inode_flag(inode, FI_HOT_DATA);
2895                 f2fs_remove_dirty_inode(inode);
2896                 submitted = NULL;
2897         }
2898         unlock_page(page);
2899         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2900                         !F2FS_I(inode)->wb_task && allow_balance)
2901                 f2fs_balance_fs(sbi, need_balance_fs);
2902
2903         if (unlikely(f2fs_cp_error(sbi))) {
2904                 f2fs_submit_merged_write(sbi, DATA);
2905                 if (bio && *bio)
2906                         f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2907                 submitted = NULL;
2908         }
2909
2910         if (submitted)
2911                 *submitted = fio.submitted;
2912
2913         return 0;
2914
2915 redirty_out:
2916         redirty_page_for_writepage(wbc, page);
2917         /*
2918          * pageout() in MM translates EAGAIN, so calls handle_write_error()
2919          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2920          * file_write_and_wait_range() will see EIO error, which is critical
2921          * to return value of fsync() followed by atomic_write failure to user.
2922          */
2923         if (!err || wbc->for_reclaim)
2924                 return AOP_WRITEPAGE_ACTIVATE;
2925         unlock_page(page);
2926         return err;
2927 }
2928
2929 static int f2fs_write_data_page(struct page *page,
2930                                         struct writeback_control *wbc)
2931 {
2932 #ifdef CONFIG_F2FS_FS_COMPRESSION
2933         struct inode *inode = page->mapping->host;
2934
2935         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2936                 goto out;
2937
2938         if (f2fs_compressed_file(inode)) {
2939                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2940                         redirty_page_for_writepage(wbc, page);
2941                         return AOP_WRITEPAGE_ACTIVATE;
2942                 }
2943         }
2944 out:
2945 #endif
2946
2947         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2948                                                 wbc, FS_DATA_IO, 0, true);
2949 }
2950
2951 /*
2952  * This function was copied from write_cache_pages from mm/page-writeback.c.
2953  * The major change is making write step of cold data page separately from
2954  * warm/hot data page.
2955  */
2956 static int f2fs_write_cache_pages(struct address_space *mapping,
2957                                         struct writeback_control *wbc,
2958                                         enum iostat_type io_type)
2959 {
2960         int ret = 0;
2961         int done = 0, retry = 0;
2962         struct page *pages_local[F2FS_ONSTACK_PAGES];
2963         struct page **pages = pages_local;
2964         struct folio_batch fbatch;
2965         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2966         struct bio *bio = NULL;
2967         sector_t last_block;
2968 #ifdef CONFIG_F2FS_FS_COMPRESSION
2969         struct inode *inode = mapping->host;
2970         struct compress_ctx cc = {
2971                 .inode = inode,
2972                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2973                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2974                 .cluster_idx = NULL_CLUSTER,
2975                 .rpages = NULL,
2976                 .nr_rpages = 0,
2977                 .cpages = NULL,
2978                 .valid_nr_cpages = 0,
2979                 .rbuf = NULL,
2980                 .cbuf = NULL,
2981                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2982                 .private = NULL,
2983         };
2984 #endif
2985         int nr_folios, p, idx;
2986         int nr_pages;
2987         unsigned int max_pages = F2FS_ONSTACK_PAGES;
2988         pgoff_t index;
2989         pgoff_t end;            /* Inclusive */
2990         pgoff_t done_index;
2991         int range_whole = 0;
2992         xa_mark_t tag;
2993         int nwritten = 0;
2994         int submitted = 0;
2995         int i;
2996
2997 #ifdef CONFIG_F2FS_FS_COMPRESSION
2998         if (f2fs_compressed_file(inode) &&
2999                 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) {
3000                 pages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
3001                                 cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL);
3002                 max_pages = 1 << cc.log_cluster_size;
3003         }
3004 #endif
3005
3006         folio_batch_init(&fbatch);
3007
3008         if (get_dirty_pages(mapping->host) <=
3009                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3010                 set_inode_flag(mapping->host, FI_HOT_DATA);
3011         else
3012                 clear_inode_flag(mapping->host, FI_HOT_DATA);
3013
3014         if (wbc->range_cyclic) {
3015                 index = mapping->writeback_index; /* prev offset */
3016                 end = -1;
3017         } else {
3018                 index = wbc->range_start >> PAGE_SHIFT;
3019                 end = wbc->range_end >> PAGE_SHIFT;
3020                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3021                         range_whole = 1;
3022         }
3023         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3024                 tag = PAGECACHE_TAG_TOWRITE;
3025         else
3026                 tag = PAGECACHE_TAG_DIRTY;
3027 retry:
3028         retry = 0;
3029         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3030                 tag_pages_for_writeback(mapping, index, end);
3031         done_index = index;
3032         while (!done && !retry && (index <= end)) {
3033                 nr_pages = 0;
3034 again:
3035                 nr_folios = filemap_get_folios_tag(mapping, &index, end,
3036                                 tag, &fbatch);
3037                 if (nr_folios == 0) {
3038                         if (nr_pages)
3039                                 goto write;
3040                         break;
3041                 }
3042
3043                 for (i = 0; i < nr_folios; i++) {
3044                         struct folio *folio = fbatch.folios[i];
3045
3046                         idx = 0;
3047                         p = folio_nr_pages(folio);
3048 add_more:
3049                         pages[nr_pages] = folio_page(folio, idx);
3050                         folio_get(folio);
3051                         if (++nr_pages == max_pages) {
3052                                 index = folio->index + idx + 1;
3053                                 folio_batch_release(&fbatch);
3054                                 goto write;
3055                         }
3056                         if (++idx < p)
3057                                 goto add_more;
3058                 }
3059                 folio_batch_release(&fbatch);
3060                 goto again;
3061 write:
3062                 for (i = 0; i < nr_pages; i++) {
3063                         struct page *page = pages[i];
3064                         struct folio *folio = page_folio(page);
3065                         bool need_readd;
3066 readd:
3067                         need_readd = false;
3068 #ifdef CONFIG_F2FS_FS_COMPRESSION
3069                         if (f2fs_compressed_file(inode)) {
3070                                 void *fsdata = NULL;
3071                                 struct page *pagep;
3072                                 int ret2;
3073
3074                                 ret = f2fs_init_compress_ctx(&cc);
3075                                 if (ret) {
3076                                         done = 1;
3077                                         break;
3078                                 }
3079
3080                                 if (!f2fs_cluster_can_merge_page(&cc,
3081                                                                 folio->index)) {
3082                                         ret = f2fs_write_multi_pages(&cc,
3083                                                 &submitted, wbc, io_type);
3084                                         if (!ret)
3085                                                 need_readd = true;
3086                                         goto result;
3087                                 }
3088
3089                                 if (unlikely(f2fs_cp_error(sbi)))
3090                                         goto lock_folio;
3091
3092                                 if (!f2fs_cluster_is_empty(&cc))
3093                                         goto lock_folio;
3094
3095                                 if (f2fs_all_cluster_page_ready(&cc,
3096                                         pages, i, nr_pages, true))
3097                                         goto lock_folio;
3098
3099                                 ret2 = f2fs_prepare_compress_overwrite(
3100                                                         inode, &pagep,
3101                                                         folio->index, &fsdata);
3102                                 if (ret2 < 0) {
3103                                         ret = ret2;
3104                                         done = 1;
3105                                         break;
3106                                 } else if (ret2 &&
3107                                         (!f2fs_compress_write_end(inode,
3108                                                 fsdata, folio->index, 1) ||
3109                                          !f2fs_all_cluster_page_ready(&cc,
3110                                                 pages, i, nr_pages,
3111                                                 false))) {
3112                                         retry = 1;
3113                                         break;
3114                                 }
3115                         }
3116 #endif
3117                         /* give a priority to WB_SYNC threads */
3118                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3119                                         wbc->sync_mode == WB_SYNC_NONE) {
3120                                 done = 1;
3121                                 break;
3122                         }
3123 #ifdef CONFIG_F2FS_FS_COMPRESSION
3124 lock_folio:
3125 #endif
3126                         done_index = folio->index;
3127 retry_write:
3128                         folio_lock(folio);
3129
3130                         if (unlikely(folio->mapping != mapping)) {
3131 continue_unlock:
3132                                 folio_unlock(folio);
3133                                 continue;
3134                         }
3135
3136                         if (!folio_test_dirty(folio)) {
3137                                 /* someone wrote it for us */
3138                                 goto continue_unlock;
3139                         }
3140
3141                         if (folio_test_writeback(folio)) {
3142                                 if (wbc->sync_mode == WB_SYNC_NONE)
3143                                         goto continue_unlock;
3144                                 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3145                         }
3146
3147                         if (!folio_clear_dirty_for_io(folio))
3148                                 goto continue_unlock;
3149
3150 #ifdef CONFIG_F2FS_FS_COMPRESSION
3151                         if (f2fs_compressed_file(inode)) {
3152                                 folio_get(folio);
3153                                 f2fs_compress_ctx_add_page(&cc, &folio->page);
3154                                 continue;
3155                         }
3156 #endif
3157                         ret = f2fs_write_single_data_page(&folio->page,
3158                                         &submitted, &bio, &last_block,
3159                                         wbc, io_type, 0, true);
3160                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3161                                 folio_unlock(folio);
3162 #ifdef CONFIG_F2FS_FS_COMPRESSION
3163 result:
3164 #endif
3165                         nwritten += submitted;
3166                         wbc->nr_to_write -= submitted;
3167
3168                         if (unlikely(ret)) {
3169                                 /*
3170                                  * keep nr_to_write, since vfs uses this to
3171                                  * get # of written pages.
3172                                  */
3173                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3174                                         ret = 0;
3175                                         goto next;
3176                                 } else if (ret == -EAGAIN) {
3177                                         ret = 0;
3178                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3179                                                 f2fs_io_schedule_timeout(
3180                                                         DEFAULT_IO_TIMEOUT);
3181                                                 goto retry_write;
3182                                         }
3183                                         goto next;
3184                                 }
3185                                 done_index = folio_next_index(folio);
3186                                 done = 1;
3187                                 break;
3188                         }
3189
3190                         if (wbc->nr_to_write <= 0 &&
3191                                         wbc->sync_mode == WB_SYNC_NONE) {
3192                                 done = 1;
3193                                 break;
3194                         }
3195 next:
3196                         if (need_readd)
3197                                 goto readd;
3198                 }
3199                 release_pages(pages, nr_pages);
3200                 cond_resched();
3201         }
3202 #ifdef CONFIG_F2FS_FS_COMPRESSION
3203         /* flush remained pages in compress cluster */
3204         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3205                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3206                 nwritten += submitted;
3207                 wbc->nr_to_write -= submitted;
3208                 if (ret) {
3209                         done = 1;
3210                         retry = 0;
3211                 }
3212         }
3213         if (f2fs_compressed_file(inode))
3214                 f2fs_destroy_compress_ctx(&cc, false);
3215 #endif
3216         if (retry) {
3217                 index = 0;
3218                 end = -1;
3219                 goto retry;
3220         }
3221         if (wbc->range_cyclic && !done)
3222                 done_index = 0;
3223         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3224                 mapping->writeback_index = done_index;
3225
3226         if (nwritten)
3227                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3228                                                                 NULL, 0, DATA);
3229         /* submit cached bio of IPU write */
3230         if (bio)
3231                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3232
3233 #ifdef CONFIG_F2FS_FS_COMPRESSION
3234         if (pages != pages_local)
3235                 kfree(pages);
3236 #endif
3237
3238         return ret;
3239 }
3240
3241 static inline bool __should_serialize_io(struct inode *inode,
3242                                         struct writeback_control *wbc)
3243 {
3244         /* to avoid deadlock in path of data flush */
3245         if (F2FS_I(inode)->wb_task)
3246                 return false;
3247
3248         if (!S_ISREG(inode->i_mode))
3249                 return false;
3250         if (IS_NOQUOTA(inode))
3251                 return false;
3252
3253         if (f2fs_need_compress_data(inode))
3254                 return true;
3255         if (wbc->sync_mode != WB_SYNC_ALL)
3256                 return true;
3257         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3258                 return true;
3259         return false;
3260 }
3261
3262 static int __f2fs_write_data_pages(struct address_space *mapping,
3263                                                 struct writeback_control *wbc,
3264                                                 enum iostat_type io_type)
3265 {
3266         struct inode *inode = mapping->host;
3267         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3268         struct blk_plug plug;
3269         int ret;
3270         bool locked = false;
3271
3272         /* deal with chardevs and other special file */
3273         if (!mapping->a_ops->writepage)
3274                 return 0;
3275
3276         /* skip writing if there is no dirty page in this inode */
3277         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3278                 return 0;
3279
3280         /* during POR, we don't need to trigger writepage at all. */
3281         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3282                 goto skip_write;
3283
3284         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3285                         wbc->sync_mode == WB_SYNC_NONE &&
3286                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3287                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3288                 goto skip_write;
3289
3290         /* skip writing in file defragment preparing stage */
3291         if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3292                 goto skip_write;
3293
3294         trace_f2fs_writepages(mapping->host, wbc, DATA);
3295
3296         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3297         if (wbc->sync_mode == WB_SYNC_ALL)
3298                 atomic_inc(&sbi->wb_sync_req[DATA]);
3299         else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3300                 /* to avoid potential deadlock */
3301                 if (current->plug)
3302                         blk_finish_plug(current->plug);
3303                 goto skip_write;
3304         }
3305
3306         if (__should_serialize_io(inode, wbc)) {
3307                 mutex_lock(&sbi->writepages);
3308                 locked = true;
3309         }
3310
3311         blk_start_plug(&plug);
3312         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3313         blk_finish_plug(&plug);
3314
3315         if (locked)
3316                 mutex_unlock(&sbi->writepages);
3317
3318         if (wbc->sync_mode == WB_SYNC_ALL)
3319                 atomic_dec(&sbi->wb_sync_req[DATA]);
3320         /*
3321          * if some pages were truncated, we cannot guarantee its mapping->host
3322          * to detect pending bios.
3323          */
3324
3325         f2fs_remove_dirty_inode(inode);
3326         return ret;
3327
3328 skip_write:
3329         wbc->pages_skipped += get_dirty_pages(inode);
3330         trace_f2fs_writepages(mapping->host, wbc, DATA);
3331         return 0;
3332 }
3333
3334 static int f2fs_write_data_pages(struct address_space *mapping,
3335                             struct writeback_control *wbc)
3336 {
3337         struct inode *inode = mapping->host;
3338
3339         return __f2fs_write_data_pages(mapping, wbc,
3340                         F2FS_I(inode)->cp_task == current ?
3341                         FS_CP_DATA_IO : FS_DATA_IO);
3342 }
3343
3344 void f2fs_write_failed(struct inode *inode, loff_t to)
3345 {
3346         loff_t i_size = i_size_read(inode);
3347
3348         if (IS_NOQUOTA(inode))
3349                 return;
3350
3351         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3352         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3353                 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3354                 filemap_invalidate_lock(inode->i_mapping);
3355
3356                 truncate_pagecache(inode, i_size);
3357                 f2fs_truncate_blocks(inode, i_size, true);
3358
3359                 filemap_invalidate_unlock(inode->i_mapping);
3360                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3361         }
3362 }
3363
3364 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3365                         struct page *page, loff_t pos, unsigned len,
3366                         block_t *blk_addr, bool *node_changed)
3367 {
3368         struct inode *inode = page->mapping->host;
3369         pgoff_t index = page->index;
3370         struct dnode_of_data dn;
3371         struct page *ipage;
3372         bool locked = false;
3373         int flag = F2FS_GET_BLOCK_PRE_AIO;
3374         int err = 0;
3375
3376         /*
3377          * If a whole page is being written and we already preallocated all the
3378          * blocks, then there is no need to get a block address now.
3379          */
3380         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3381                 return 0;
3382
3383         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3384         if (f2fs_has_inline_data(inode)) {
3385                 if (pos + len > MAX_INLINE_DATA(inode))
3386                         flag = F2FS_GET_BLOCK_DEFAULT;
3387                 f2fs_map_lock(sbi, flag);
3388                 locked = true;
3389         } else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3390                 f2fs_map_lock(sbi, flag);
3391                 locked = true;
3392         }
3393
3394 restart:
3395         /* check inline_data */
3396         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3397         if (IS_ERR(ipage)) {
3398                 err = PTR_ERR(ipage);
3399                 goto unlock_out;
3400         }
3401
3402         set_new_dnode(&dn, inode, ipage, ipage, 0);
3403
3404         if (f2fs_has_inline_data(inode)) {
3405                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3406                         f2fs_do_read_inline_data(page_folio(page), ipage);
3407                         set_inode_flag(inode, FI_DATA_EXIST);
3408                         if (inode->i_nlink)
3409                                 set_page_private_inline(ipage);
3410                         goto out;
3411                 }
3412                 err = f2fs_convert_inline_page(&dn, page);
3413                 if (err || dn.data_blkaddr != NULL_ADDR)
3414                         goto out;
3415         }
3416
3417         if (!f2fs_lookup_read_extent_cache_block(inode, index,
3418                                                  &dn.data_blkaddr)) {
3419                 if (locked) {
3420                         err = f2fs_reserve_block(&dn, index);
3421                         goto out;
3422                 }
3423
3424                 /* hole case */
3425                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3426                 if (!err && dn.data_blkaddr != NULL_ADDR)
3427                         goto out;
3428                 f2fs_put_dnode(&dn);
3429                 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3430                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3431                 locked = true;
3432                 goto restart;
3433         }
3434 out:
3435         if (!err) {
3436                 /* convert_inline_page can make node_changed */
3437                 *blk_addr = dn.data_blkaddr;
3438                 *node_changed = dn.node_changed;
3439         }
3440         f2fs_put_dnode(&dn);
3441 unlock_out:
3442         if (locked)
3443                 f2fs_map_unlock(sbi, flag);
3444         return err;
3445 }
3446
3447 static int __find_data_block(struct inode *inode, pgoff_t index,
3448                                 block_t *blk_addr)
3449 {
3450         struct dnode_of_data dn;
3451         struct page *ipage;
3452         int err = 0;
3453
3454         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3455         if (IS_ERR(ipage))
3456                 return PTR_ERR(ipage);
3457
3458         set_new_dnode(&dn, inode, ipage, ipage, 0);
3459
3460         if (!f2fs_lookup_read_extent_cache_block(inode, index,
3461                                                  &dn.data_blkaddr)) {
3462                 /* hole case */
3463                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3464                 if (err) {
3465                         dn.data_blkaddr = NULL_ADDR;
3466                         err = 0;
3467                 }
3468         }
3469         *blk_addr = dn.data_blkaddr;
3470         f2fs_put_dnode(&dn);
3471         return err;
3472 }
3473
3474 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3475                                 block_t *blk_addr, bool *node_changed)
3476 {
3477         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3478         struct dnode_of_data dn;
3479         struct page *ipage;
3480         int err = 0;
3481
3482         f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3483
3484         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3485         if (IS_ERR(ipage)) {
3486                 err = PTR_ERR(ipage);
3487                 goto unlock_out;
3488         }
3489         set_new_dnode(&dn, inode, ipage, ipage, 0);
3490
3491         if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3492                                                 &dn.data_blkaddr))
3493                 err = f2fs_reserve_block(&dn, index);
3494
3495         *blk_addr = dn.data_blkaddr;
3496         *node_changed = dn.node_changed;
3497         f2fs_put_dnode(&dn);
3498
3499 unlock_out:
3500         f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3501         return err;
3502 }
3503
3504 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3505                         struct page *page, loff_t pos, unsigned int len,
3506                         block_t *blk_addr, bool *node_changed, bool *use_cow)
3507 {
3508         struct inode *inode = page->mapping->host;
3509         struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3510         pgoff_t index = page->index;
3511         int err = 0;
3512         block_t ori_blk_addr = NULL_ADDR;
3513
3514         /* If pos is beyond the end of file, reserve a new block in COW inode */
3515         if ((pos & PAGE_MASK) >= i_size_read(inode))
3516                 goto reserve_block;
3517
3518         /* Look for the block in COW inode first */
3519         err = __find_data_block(cow_inode, index, blk_addr);
3520         if (err) {
3521                 return err;
3522         } else if (*blk_addr != NULL_ADDR) {
3523                 *use_cow = true;
3524                 return 0;
3525         }
3526
3527         if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3528                 goto reserve_block;
3529
3530         /* Look for the block in the original inode */
3531         err = __find_data_block(inode, index, &ori_blk_addr);
3532         if (err)
3533                 return err;
3534
3535 reserve_block:
3536         /* Finally, we should reserve a new block in COW inode for the update */
3537         err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3538         if (err)
3539                 return err;
3540         inc_atomic_write_cnt(inode);
3541
3542         if (ori_blk_addr != NULL_ADDR)
3543                 *blk_addr = ori_blk_addr;
3544         return 0;
3545 }
3546
3547 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3548                 loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3549 {
3550         struct inode *inode = mapping->host;
3551         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3552         struct page *page = NULL;
3553         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3554         bool need_balance = false;
3555         bool use_cow = false;
3556         block_t blkaddr = NULL_ADDR;
3557         int err = 0;
3558
3559         trace_f2fs_write_begin(inode, pos, len);
3560
3561         if (!f2fs_is_checkpoint_ready(sbi)) {
3562                 err = -ENOSPC;
3563                 goto fail;
3564         }
3565
3566         /*
3567          * We should check this at this moment to avoid deadlock on inode page
3568          * and #0 page. The locking rule for inline_data conversion should be:
3569          * lock_page(page #0) -> lock_page(inode_page)
3570          */
3571         if (index != 0) {
3572                 err = f2fs_convert_inline_inode(inode);
3573                 if (err)
3574                         goto fail;
3575         }
3576
3577 #ifdef CONFIG_F2FS_FS_COMPRESSION
3578         if (f2fs_compressed_file(inode)) {
3579                 int ret;
3580
3581                 *fsdata = NULL;
3582
3583                 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3584                         goto repeat;
3585
3586                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3587                                                         index, fsdata);
3588                 if (ret < 0) {
3589                         err = ret;
3590                         goto fail;
3591                 } else if (ret) {
3592                         return 0;
3593                 }
3594         }
3595 #endif
3596
3597 repeat:
3598         /*
3599          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3600          * wait_for_stable_page. Will wait that below with our IO control.
3601          */
3602         page = f2fs_pagecache_get_page(mapping, index,
3603                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3604         if (!page) {
3605                 err = -ENOMEM;
3606                 goto fail;
3607         }
3608
3609         /* TODO: cluster can be compressed due to race with .writepage */
3610
3611         *pagep = page;
3612
3613         if (f2fs_is_atomic_file(inode))
3614                 err = prepare_atomic_write_begin(sbi, page, pos, len,
3615                                         &blkaddr, &need_balance, &use_cow);
3616         else
3617                 err = prepare_write_begin(sbi, page, pos, len,
3618                                         &blkaddr, &need_balance);
3619         if (err)
3620                 goto fail;
3621
3622         if (need_balance && !IS_NOQUOTA(inode) &&
3623                         has_not_enough_free_secs(sbi, 0, 0)) {
3624                 unlock_page(page);
3625                 f2fs_balance_fs(sbi, true);
3626                 lock_page(page);
3627                 if (page->mapping != mapping) {
3628                         /* The page got truncated from under us */
3629                         f2fs_put_page(page, 1);
3630                         goto repeat;
3631                 }
3632         }
3633
3634         f2fs_wait_on_page_writeback(page, DATA, false, true);
3635
3636         if (len == PAGE_SIZE || PageUptodate(page))
3637                 return 0;
3638
3639         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3640             !f2fs_verity_in_progress(inode)) {
3641                 zero_user_segment(page, len, PAGE_SIZE);
3642                 return 0;
3643         }
3644
3645         if (blkaddr == NEW_ADDR) {
3646                 zero_user_segment(page, 0, PAGE_SIZE);
3647                 SetPageUptodate(page);
3648         } else {
3649                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3650                                 DATA_GENERIC_ENHANCE_READ)) {
3651                         err = -EFSCORRUPTED;
3652                         goto fail;
3653                 }
3654                 err = f2fs_submit_page_read(use_cow ?
3655                                 F2FS_I(inode)->cow_inode : inode, page,
3656                                 blkaddr, 0, true);
3657                 if (err)
3658                         goto fail;
3659
3660                 lock_page(page);
3661                 if (unlikely(page->mapping != mapping)) {
3662                         f2fs_put_page(page, 1);
3663                         goto repeat;
3664                 }
3665                 if (unlikely(!PageUptodate(page))) {
3666                         err = -EIO;
3667                         goto fail;
3668                 }
3669         }
3670         return 0;
3671
3672 fail:
3673         f2fs_put_page(page, 1);
3674         f2fs_write_failed(inode, pos + len);
3675         return err;
3676 }
3677
3678 static int f2fs_write_end(struct file *file,
3679                         struct address_space *mapping,
3680                         loff_t pos, unsigned len, unsigned copied,
3681                         struct page *page, void *fsdata)
3682 {
3683         struct inode *inode = page->mapping->host;
3684
3685         trace_f2fs_write_end(inode, pos, len, copied);
3686
3687         /*
3688          * This should be come from len == PAGE_SIZE, and we expect copied
3689          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3690          * let generic_perform_write() try to copy data again through copied=0.
3691          */
3692         if (!PageUptodate(page)) {
3693                 if (unlikely(copied != len))
3694                         copied = 0;
3695                 else
3696                         SetPageUptodate(page);
3697         }
3698
3699 #ifdef CONFIG_F2FS_FS_COMPRESSION
3700         /* overwrite compressed file */
3701         if (f2fs_compressed_file(inode) && fsdata) {
3702                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3703                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3704
3705                 if (pos + copied > i_size_read(inode) &&
3706                                 !f2fs_verity_in_progress(inode))
3707                         f2fs_i_size_write(inode, pos + copied);
3708                 return copied;
3709         }
3710 #endif
3711
3712         if (!copied)
3713                 goto unlock_out;
3714
3715         set_page_dirty(page);
3716
3717         if (pos + copied > i_size_read(inode) &&
3718             !f2fs_verity_in_progress(inode)) {
3719                 f2fs_i_size_write(inode, pos + copied);
3720                 if (f2fs_is_atomic_file(inode))
3721                         f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3722                                         pos + copied);
3723         }
3724 unlock_out:
3725         f2fs_put_page(page, 1);
3726         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3727         return copied;
3728 }
3729
3730 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3731 {
3732         struct inode *inode = folio->mapping->host;
3733         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3734
3735         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3736                                 (offset || length != folio_size(folio)))
3737                 return;
3738
3739         if (folio_test_dirty(folio)) {
3740                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3741                         dec_page_count(sbi, F2FS_DIRTY_META);
3742                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3743                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3744                 } else {
3745                         inode_dec_dirty_pages(inode);
3746                         f2fs_remove_dirty_inode(inode);
3747                 }
3748         }
3749         clear_page_private_all(&folio->page);
3750 }
3751
3752 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3753 {
3754         /* If this is dirty folio, keep private data */
3755         if (folio_test_dirty(folio))
3756                 return false;
3757
3758         clear_page_private_all(&folio->page);
3759         return true;
3760 }
3761
3762 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3763                 struct folio *folio)
3764 {
3765         struct inode *inode = mapping->host;
3766
3767         trace_f2fs_set_page_dirty(folio, DATA);
3768
3769         if (!folio_test_uptodate(folio))
3770                 folio_mark_uptodate(folio);
3771         BUG_ON(folio_test_swapcache(folio));
3772
3773         if (filemap_dirty_folio(mapping, folio)) {
3774                 f2fs_update_dirty_folio(inode, folio);
3775                 return true;
3776         }
3777         return false;
3778 }
3779
3780
3781 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3782 {
3783 #ifdef CONFIG_F2FS_FS_COMPRESSION
3784         struct dnode_of_data dn;
3785         sector_t start_idx, blknr = 0;
3786         int ret;
3787
3788         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3789
3790         set_new_dnode(&dn, inode, NULL, NULL, 0);
3791         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3792         if (ret)
3793                 return 0;
3794
3795         if (dn.data_blkaddr != COMPRESS_ADDR) {
3796                 dn.ofs_in_node += block - start_idx;
3797                 blknr = f2fs_data_blkaddr(&dn);
3798                 if (!__is_valid_data_blkaddr(blknr))
3799                         blknr = 0;
3800         }
3801
3802         f2fs_put_dnode(&dn);
3803         return blknr;
3804 #else
3805         return 0;
3806 #endif
3807 }
3808
3809
3810 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3811 {
3812         struct inode *inode = mapping->host;
3813         sector_t blknr = 0;
3814
3815         if (f2fs_has_inline_data(inode))
3816                 goto out;
3817
3818         /* make sure allocating whole blocks */
3819         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3820                 filemap_write_and_wait(mapping);
3821
3822         /* Block number less than F2FS MAX BLOCKS */
3823         if (unlikely(block >= max_file_blocks(inode)))
3824                 goto out;
3825
3826         if (f2fs_compressed_file(inode)) {
3827                 blknr = f2fs_bmap_compress(inode, block);
3828         } else {
3829                 struct f2fs_map_blocks map;
3830
3831                 memset(&map, 0, sizeof(map));
3832                 map.m_lblk = block;
3833                 map.m_len = 1;
3834                 map.m_next_pgofs = NULL;
3835                 map.m_seg_type = NO_CHECK_TYPE;
3836
3837                 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3838                         blknr = map.m_pblk;
3839         }
3840 out:
3841         trace_f2fs_bmap(inode, block, blknr);
3842         return blknr;
3843 }
3844
3845 #ifdef CONFIG_SWAP
3846 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3847                                                         unsigned int blkcnt)
3848 {
3849         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3850         unsigned int blkofs;
3851         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3852         unsigned int end_blk = start_blk + blkcnt - 1;
3853         unsigned int secidx = start_blk / blk_per_sec;
3854         unsigned int end_sec;
3855         int ret = 0;
3856
3857         if (!blkcnt)
3858                 return 0;
3859         end_sec = end_blk / blk_per_sec;
3860
3861         f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3862         filemap_invalidate_lock(inode->i_mapping);
3863
3864         set_inode_flag(inode, FI_ALIGNED_WRITE);
3865         set_inode_flag(inode, FI_OPU_WRITE);
3866
3867         for (; secidx <= end_sec; secidx++) {
3868                 unsigned int blkofs_end = secidx == end_sec ?
3869                                 end_blk % blk_per_sec : blk_per_sec - 1;
3870
3871                 f2fs_down_write(&sbi->pin_sem);
3872
3873                 ret = f2fs_allocate_pinning_section(sbi);
3874                 if (ret) {
3875                         f2fs_up_write(&sbi->pin_sem);
3876                         break;
3877                 }
3878
3879                 set_inode_flag(inode, FI_SKIP_WRITES);
3880
3881                 for (blkofs = 0; blkofs <= blkofs_end; blkofs++) {
3882                         struct page *page;
3883                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3884
3885                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3886                         if (IS_ERR(page)) {
3887                                 f2fs_up_write(&sbi->pin_sem);
3888                                 ret = PTR_ERR(page);
3889                                 goto done;
3890                         }
3891
3892                         set_page_dirty(page);
3893                         f2fs_put_page(page, 1);
3894                 }
3895
3896                 clear_inode_flag(inode, FI_SKIP_WRITES);
3897
3898                 ret = filemap_fdatawrite(inode->i_mapping);
3899
3900                 f2fs_up_write(&sbi->pin_sem);
3901
3902                 if (ret)
3903                         break;
3904         }
3905
3906 done:
3907         clear_inode_flag(inode, FI_SKIP_WRITES);
3908         clear_inode_flag(inode, FI_OPU_WRITE);
3909         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3910
3911         filemap_invalidate_unlock(inode->i_mapping);
3912         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3913
3914         return ret;
3915 }
3916
3917 static int check_swap_activate(struct swap_info_struct *sis,
3918                                 struct file *swap_file, sector_t *span)
3919 {
3920         struct address_space *mapping = swap_file->f_mapping;
3921         struct inode *inode = mapping->host;
3922         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3923         block_t cur_lblock;
3924         block_t last_lblock;
3925         block_t pblock;
3926         block_t lowest_pblock = -1;
3927         block_t highest_pblock = 0;
3928         int nr_extents = 0;
3929         unsigned int nr_pblocks;
3930         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3931         unsigned int not_aligned = 0;
3932         int ret = 0;
3933
3934         /*
3935          * Map all the blocks into the extent list.  This code doesn't try
3936          * to be very smart.
3937          */
3938         cur_lblock = 0;
3939         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3940
3941         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3942                 struct f2fs_map_blocks map;
3943 retry:
3944                 cond_resched();
3945
3946                 memset(&map, 0, sizeof(map));
3947                 map.m_lblk = cur_lblock;
3948                 map.m_len = last_lblock - cur_lblock;
3949                 map.m_next_pgofs = NULL;
3950                 map.m_next_extent = NULL;
3951                 map.m_seg_type = NO_CHECK_TYPE;
3952                 map.m_may_create = false;
3953
3954                 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
3955                 if (ret)
3956                         goto out;
3957
3958                 /* hole */
3959                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3960                         f2fs_err(sbi, "Swapfile has holes");
3961                         ret = -EINVAL;
3962                         goto out;
3963                 }
3964
3965                 pblock = map.m_pblk;
3966                 nr_pblocks = map.m_len;
3967
3968                 if ((pblock - SM_I(sbi)->main_blkaddr) % blks_per_sec ||
3969                                 nr_pblocks % blks_per_sec ||
3970                                 !f2fs_valid_pinned_area(sbi, pblock)) {
3971                         bool last_extent = false;
3972
3973                         not_aligned++;
3974
3975                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3976                         if (cur_lblock + nr_pblocks > sis->max)
3977                                 nr_pblocks -= blks_per_sec;
3978
3979                         /* this extent is last one */
3980                         if (!nr_pblocks) {
3981                                 nr_pblocks = last_lblock - cur_lblock;
3982                                 last_extent = true;
3983                         }
3984
3985                         ret = f2fs_migrate_blocks(inode, cur_lblock,
3986                                                         nr_pblocks);
3987                         if (ret) {
3988                                 if (ret == -ENOENT)
3989                                         ret = -EINVAL;
3990                                 goto out;
3991                         }
3992
3993                         if (!last_extent)
3994                                 goto retry;
3995                 }
3996
3997                 if (cur_lblock + nr_pblocks >= sis->max)
3998                         nr_pblocks = sis->max - cur_lblock;
3999
4000                 if (cur_lblock) {       /* exclude the header page */
4001                         if (pblock < lowest_pblock)
4002                                 lowest_pblock = pblock;
4003                         if (pblock + nr_pblocks - 1 > highest_pblock)
4004                                 highest_pblock = pblock + nr_pblocks - 1;
4005                 }
4006
4007                 /*
4008                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4009                  */
4010                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4011                 if (ret < 0)
4012                         goto out;
4013                 nr_extents += ret;
4014                 cur_lblock += nr_pblocks;
4015         }
4016         ret = nr_extents;
4017         *span = 1 + highest_pblock - lowest_pblock;
4018         if (cur_lblock == 0)
4019                 cur_lblock = 1; /* force Empty message */
4020         sis->max = cur_lblock;
4021         sis->pages = cur_lblock - 1;
4022         sis->highest_bit = cur_lblock - 1;
4023 out:
4024         if (not_aligned)
4025                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%lu * N)",
4026                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
4027         return ret;
4028 }
4029
4030 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4031                                 sector_t *span)
4032 {
4033         struct inode *inode = file_inode(file);
4034         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4035         int ret;
4036
4037         if (!S_ISREG(inode->i_mode))
4038                 return -EINVAL;
4039
4040         if (f2fs_readonly(sbi->sb))
4041                 return -EROFS;
4042
4043         if (f2fs_lfs_mode(sbi) && !f2fs_sb_has_blkzoned(sbi)) {
4044                 f2fs_err(sbi, "Swapfile not supported in LFS mode");
4045                 return -EINVAL;
4046         }
4047
4048         ret = f2fs_convert_inline_inode(inode);
4049         if (ret)
4050                 return ret;
4051
4052         if (!f2fs_disable_compressed_file(inode))
4053                 return -EINVAL;
4054
4055         ret = filemap_fdatawrite(inode->i_mapping);
4056         if (ret < 0)
4057                 return ret;
4058
4059         f2fs_precache_extents(inode);
4060
4061         ret = check_swap_activate(sis, file, span);
4062         if (ret < 0)
4063                 return ret;
4064
4065         stat_inc_swapfile_inode(inode);
4066         set_inode_flag(inode, FI_PIN_FILE);
4067         f2fs_update_time(sbi, REQ_TIME);
4068         return ret;
4069 }
4070
4071 static void f2fs_swap_deactivate(struct file *file)
4072 {
4073         struct inode *inode = file_inode(file);
4074
4075         stat_dec_swapfile_inode(inode);
4076         clear_inode_flag(inode, FI_PIN_FILE);
4077 }
4078 #else
4079 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4080                                 sector_t *span)
4081 {
4082         return -EOPNOTSUPP;
4083 }
4084
4085 static void f2fs_swap_deactivate(struct file *file)
4086 {
4087 }
4088 #endif
4089
4090 const struct address_space_operations f2fs_dblock_aops = {
4091         .read_folio     = f2fs_read_data_folio,
4092         .readahead      = f2fs_readahead,
4093         .writepage      = f2fs_write_data_page,
4094         .writepages     = f2fs_write_data_pages,
4095         .write_begin    = f2fs_write_begin,
4096         .write_end      = f2fs_write_end,
4097         .dirty_folio    = f2fs_dirty_data_folio,
4098         .migrate_folio  = filemap_migrate_folio,
4099         .invalidate_folio = f2fs_invalidate_folio,
4100         .release_folio  = f2fs_release_folio,
4101         .bmap           = f2fs_bmap,
4102         .swap_activate  = f2fs_swap_activate,
4103         .swap_deactivate = f2fs_swap_deactivate,
4104 };
4105
4106 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4107 {
4108         struct folio *folio = page_folio(page);
4109         struct address_space *mapping = folio->mapping;
4110         unsigned long flags;
4111
4112         xa_lock_irqsave(&mapping->i_pages, flags);
4113         __xa_clear_mark(&mapping->i_pages, folio->index,
4114                                                 PAGECACHE_TAG_DIRTY);
4115         xa_unlock_irqrestore(&mapping->i_pages, flags);
4116 }
4117
4118 int __init f2fs_init_post_read_processing(void)
4119 {
4120         bio_post_read_ctx_cache =
4121                 kmem_cache_create("f2fs_bio_post_read_ctx",
4122                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4123         if (!bio_post_read_ctx_cache)
4124                 goto fail;
4125         bio_post_read_ctx_pool =
4126                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4127                                          bio_post_read_ctx_cache);
4128         if (!bio_post_read_ctx_pool)
4129                 goto fail_free_cache;
4130         return 0;
4131
4132 fail_free_cache:
4133         kmem_cache_destroy(bio_post_read_ctx_cache);
4134 fail:
4135         return -ENOMEM;
4136 }
4137
4138 void f2fs_destroy_post_read_processing(void)
4139 {
4140         mempool_destroy(bio_post_read_ctx_pool);
4141         kmem_cache_destroy(bio_post_read_ctx_cache);
4142 }
4143
4144 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4145 {
4146         if (!f2fs_sb_has_encrypt(sbi) &&
4147                 !f2fs_sb_has_verity(sbi) &&
4148                 !f2fs_sb_has_compression(sbi))
4149                 return 0;
4150
4151         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4152                                                  WQ_UNBOUND | WQ_HIGHPRI,
4153                                                  num_online_cpus());
4154         return sbi->post_read_wq ? 0 : -ENOMEM;
4155 }
4156
4157 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4158 {
4159         if (sbi->post_read_wq)
4160                 destroy_workqueue(sbi->post_read_wq);
4161 }
4162
4163 int __init f2fs_init_bio_entry_cache(void)
4164 {
4165         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4166                         sizeof(struct bio_entry));
4167         return bio_entry_slab ? 0 : -ENOMEM;
4168 }
4169
4170 void f2fs_destroy_bio_entry_cache(void)
4171 {
4172         kmem_cache_destroy(bio_entry_slab);
4173 }
4174
4175 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4176                             unsigned int flags, struct iomap *iomap,
4177                             struct iomap *srcmap)
4178 {
4179         struct f2fs_map_blocks map = {};
4180         pgoff_t next_pgofs = 0;
4181         int err;
4182
4183         map.m_lblk = bytes_to_blks(inode, offset);
4184         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4185         map.m_next_pgofs = &next_pgofs;
4186         map.m_seg_type = f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
4187                                                 inode->i_write_hint);
4188         if (flags & IOMAP_WRITE)
4189                 map.m_may_create = true;
4190
4191         err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4192         if (err)
4193                 return err;
4194
4195         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4196
4197         /*
4198          * When inline encryption is enabled, sometimes I/O to an encrypted file
4199          * has to be broken up to guarantee DUN contiguity.  Handle this by
4200          * limiting the length of the mapping returned.
4201          */
4202         map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4203
4204         /*
4205          * We should never see delalloc or compressed extents here based on
4206          * prior flushing and checks.
4207          */
4208         if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4209                 return -EINVAL;
4210
4211         if (map.m_flags & F2FS_MAP_MAPPED) {
4212                 if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4213                         return -EINVAL;
4214
4215                 iomap->length = blks_to_bytes(inode, map.m_len);
4216                 iomap->type = IOMAP_MAPPED;
4217                 iomap->flags |= IOMAP_F_MERGED;
4218                 iomap->bdev = map.m_bdev;
4219                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4220         } else {
4221                 if (flags & IOMAP_WRITE)
4222                         return -ENOTBLK;
4223
4224                 if (map.m_pblk == NULL_ADDR) {
4225                         iomap->length = blks_to_bytes(inode, next_pgofs) -
4226                                                                 iomap->offset;
4227                         iomap->type = IOMAP_HOLE;
4228                 } else if (map.m_pblk == NEW_ADDR) {
4229                         iomap->length = blks_to_bytes(inode, map.m_len);
4230                         iomap->type = IOMAP_UNWRITTEN;
4231                 } else {
4232                         f2fs_bug_on(F2FS_I_SB(inode), 1);
4233                 }
4234                 iomap->addr = IOMAP_NULL_ADDR;
4235         }
4236
4237         if (map.m_flags & F2FS_MAP_NEW)
4238                 iomap->flags |= IOMAP_F_NEW;
4239         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4240             offset + length > i_size_read(inode))
4241                 iomap->flags |= IOMAP_F_DIRTY;
4242
4243         return 0;
4244 }
4245
4246 const struct iomap_ops f2fs_iomap_ops = {
4247         .iomap_begin    = f2fs_iomap_begin,
4248 };