fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/sched/mm.h>
  12 #include <linux/mpage.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/blk-crypto.h>
  18 #include <linux/swap.h>
  19 #include <linux/prefetch.h>
  20 #include <linux/uio.h>
  21 #include <linux/sched/signal.h>
  22 #include <linux/fiemap.h>
  23 #include <linux/iomap.h>
  24
  25 #include "f2fs.h"
  26 #include "node.h"
  27 #include "segment.h"
  28 #include "iostat.h"
  29 #include <trace/events/f2fs.h>
  30
  31 #define NUM_PREALLOC_POST_READ_CTXS     128
  32
  33 static struct kmem_cache *bio_post_read_ctx_cache;
  34 static struct kmem_cache *bio_entry_slab;
  35 static mempool_t *bio_post_read_ctx_pool;
  36 static struct bio_set f2fs_bioset;
  37
  38 #define F2FS_BIO_POOL_SIZE      NR_CURSEG_TYPE
  39
  40 int __init f2fs_init_bioset(void)
  41 {
  42         if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  43                                         0, BIOSET_NEED_BVECS))
  44                 return -ENOMEM;
  45         return 0;
  46 }
  47
  48 void f2fs_destroy_bioset(void)
  49 {
  50         bioset_exit(&f2fs_bioset);
  51 }
  52
  53 static bool __is_cp_guaranteed(struct page *page)
  54 {
  55         struct address_space *mapping = page->mapping;
  56         struct inode *inode;
  57         struct f2fs_sb_info *sbi;
  58
  59         if (!mapping)
  60                 return false;
  61
  62         inode = mapping->host;
  63         sbi = F2FS_I_SB(inode);
  64
  65         if (inode->i_ino == F2FS_META_INO(sbi) ||
  66                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  67                         S_ISDIR(inode->i_mode))
  68                 return true;
  69
  70         if (f2fs_is_compressed_page(page))
  71                 return false;
  72         if ((S_ISREG(inode->i_mode) &&
  73                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  74                         page_private_gcing(page))
  75                 return true;
  76         return false;
  77 }
  78
  79 static enum count_type __read_io_type(struct page *page)
  80 {
  81         struct address_space *mapping = page_file_mapping(page);
  82
  83         if (mapping) {
  84                 struct inode *inode = mapping->host;
  85                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  86
  87                 if (inode->i_ino == F2FS_META_INO(sbi))
  88                         return F2FS_RD_META;
  89
  90                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  91                         return F2FS_RD_NODE;
  92         }
  93         return F2FS_RD_DATA;
  94 }
  95
  96 /* postprocessing steps for read bios */
  97 enum bio_post_read_step {
  98 #ifdef CONFIG_FS_ENCRYPTION
  99         STEP_DECRYPT    = 1 << 0,
 100 #else
 101         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
 102 #endif
 103 #ifdef CONFIG_F2FS_FS_COMPRESSION
 104         STEP_DECOMPRESS = 1 << 1,
 105 #else
 106         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 107 #endif
 108 #ifdef CONFIG_FS_VERITY
 109         STEP_VERITY     = 1 << 2,
 110 #else
 111         STEP_VERITY     = 0,    /* compile out the verity-related code */
 112 #endif
 113 };
 114
 115 struct bio_post_read_ctx {
 116         struct bio *bio;
 117         struct f2fs_sb_info *sbi;
 118         struct work_struct work;
 119         unsigned int enabled_steps;
 120         block_t fs_blkaddr;
 121 };
 122
 123 static void f2fs_finish_read_bio(struct bio *bio)
 124 {
 125         struct bio_vec *bv;
 126         struct bvec_iter_all iter_all;
 127
 128         /*
 129          * Update and unlock the bio's pagecache pages, and put the
 130          * decompression context for any compressed pages.
 131          */
 132         bio_for_each_segment_all(bv, bio, iter_all) {
 133                 struct page *page = bv->bv_page;
 134
 135                 if (f2fs_is_compressed_page(page)) {
 136                         if (bio->bi_status)
 137                                 f2fs_end_read_compressed_page(page, true, 0);
 138                         f2fs_put_page_dic(page);
 139                         continue;
 140                 }
 141
 142                 /* PG_error was set if decryption or verity failed. */
 143                 if (bio->bi_status || PageError(page)) {
 144                         ClearPageUptodate(page);
 145                         /* will re-read again later */
 146                         ClearPageError(page);
 147                 } else {
 148                         SetPageUptodate(page);
 149                 }
 150                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 151                 unlock_page(page);
 152         }
 153
 154         if (bio->bi_private)
 155                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 156         bio_put(bio);
 157 }
 158
 159 static void f2fs_verify_bio(struct work_struct *work)
 160 {
 161         struct bio_post_read_ctx *ctx =
 162                 container_of(work, struct bio_post_read_ctx, work);
 163         struct bio *bio = ctx->bio;
 164         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 165
 166         /*
 167          * fsverity_verify_bio() may call readpages() again, and while verity
 168          * will be disabled for this, decryption and/or decompression may still
 169          * be needed, resulting in another bio_post_read_ctx being allocated.
 170          * So to prevent deadlocks we need to release the current ctx to the
 171          * mempool first.  This assumes that verity is the last post-read step.
 172          */
 173         mempool_free(ctx, bio_post_read_ctx_pool);
 174         bio->bi_private = NULL;
 175
 176         /*
 177          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 178          * as those were handled separately by f2fs_end_read_compressed_page().
 179          */
 180         if (may_have_compressed_pages) {
 181                 struct bio_vec *bv;
 182                 struct bvec_iter_all iter_all;
 183
 184                 bio_for_each_segment_all(bv, bio, iter_all) {
 185                         struct page *page = bv->bv_page;
 186
 187                         if (!f2fs_is_compressed_page(page) &&
 188                             !PageError(page) && !fsverity_verify_page(page))
 189                                 SetPageError(page);
 190                 }
 191         } else {
 192                 fsverity_verify_bio(bio);
 193         }
 194
 195         f2fs_finish_read_bio(bio);
 196 }
 197
 198 /*
 199  * If the bio's data needs to be verified with fs-verity, then enqueue the
 200  * verity work for the bio.  Otherwise finish the bio now.
 201  *
 202  * Note that to avoid deadlocks, the verity work can't be done on the
 203  * decryption/decompression workqueue.  This is because verifying the data pages
 204  * can involve reading verity metadata pages from the file, and these verity
 205  * metadata pages may be encrypted and/or compressed.
 206  */
 207 static void f2fs_verify_and_finish_bio(struct bio *bio)
 208 {
 209         struct bio_post_read_ctx *ctx = bio->bi_private;
 210
 211         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 212                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 213                 fsverity_enqueue_verify_work(&ctx->work);
 214         } else {
 215                 f2fs_finish_read_bio(bio);
 216         }
 217 }
 218
 219 /*
 220  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 221  * remaining page was read by @ctx->bio.
 222  *
 223  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 224  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 225  * that the bio includes at least one compressed page.  The actual decompression
 226  * is done on a per-cluster basis, not a per-bio basis.
 227  */
 228 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
 229 {
 230         struct bio_vec *bv;
 231         struct bvec_iter_all iter_all;
 232         bool all_compressed = true;
 233         block_t blkaddr = ctx->fs_blkaddr;
 234
 235         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 236                 struct page *page = bv->bv_page;
 237
 238                 /* PG_error was set if decryption failed. */
 239                 if (f2fs_is_compressed_page(page))
 240                         f2fs_end_read_compressed_page(page, PageError(page),
 241                                                 blkaddr);
 242                 else
 243                         all_compressed = false;
 244
 245                 blkaddr++;
 246         }
 247
 248         /*
 249          * Optimization: if all the bio's pages are compressed, then scheduling
 250          * the per-bio verity work is unnecessary, as verity will be fully
 251          * handled at the compression cluster level.
 252          */
 253         if (all_compressed)
 254                 ctx->enabled_steps &= ~STEP_VERITY;
 255 }
 256
 257 static void f2fs_post_read_work(struct work_struct *work)
 258 {
 259         struct bio_post_read_ctx *ctx =
 260                 container_of(work, struct bio_post_read_ctx, work);
 261
 262         if (ctx->enabled_steps & STEP_DECRYPT)
 263                 fscrypt_decrypt_bio(ctx->bio);
 264
 265         if (ctx->enabled_steps & STEP_DECOMPRESS)
 266                 f2fs_handle_step_decompress(ctx);
 267
 268         f2fs_verify_and_finish_bio(ctx->bio);
 269 }
 270
 271 static void f2fs_read_end_io(struct bio *bio)
 272 {
 273         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 274         struct bio_post_read_ctx *ctx;
 275
 276         iostat_update_and_unbind_ctx(bio, 0);
 277         ctx = bio->bi_private;
 278
 279         if (time_to_inject(sbi, FAULT_READ_IO)) {
 280                 f2fs_show_injection_info(sbi, FAULT_READ_IO);
 281                 bio->bi_status = BLK_STS_IOERR;
 282         }
 283
 284         if (bio->bi_status) {
 285                 f2fs_finish_read_bio(bio);
 286                 return;
 287         }
 288
 289         if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
 290                 INIT_WORK(&ctx->work, f2fs_post_read_work);
 291                 queue_work(ctx->sbi->post_read_wq, &ctx->work);
 292         } else {
 293                 f2fs_verify_and_finish_bio(bio);
 294         }
 295 }
 296
 297 static void f2fs_write_end_io(struct bio *bio)
 298 {
 299         struct f2fs_sb_info *sbi;
 300         struct bio_vec *bvec;
 301         struct bvec_iter_all iter_all;
 302
 303         iostat_update_and_unbind_ctx(bio, 1);
 304         sbi = bio->bi_private;
 305
 306         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 307                 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
 308                 bio->bi_status = BLK_STS_IOERR;
 309         }
 310
 311         bio_for_each_segment_all(bvec, bio, iter_all) {
 312                 struct page *page = bvec->bv_page;
 313                 enum count_type type = WB_DATA_TYPE(page);
 314
 315                 if (page_private_dummy(page)) {
 316                         clear_page_private_dummy(page);
 317                         unlock_page(page);
 318                         mempool_free(page, sbi->write_io_dummy);
 319
 320                         if (unlikely(bio->bi_status))
 321                                 f2fs_stop_checkpoint(sbi, true);
 322                         continue;
 323                 }
 324
 325                 fscrypt_finalize_bounce_page(&page);
 326
 327 #ifdef CONFIG_F2FS_FS_COMPRESSION
 328                 if (f2fs_is_compressed_page(page)) {
 329                         f2fs_compress_write_end_io(bio, page);
 330                         continue;
 331                 }
 332 #endif
 333
 334                 if (unlikely(bio->bi_status)) {
 335                         mapping_set_error(page->mapping, -EIO);
 336                         if (type == F2FS_WB_CP_DATA)
 337                                 f2fs_stop_checkpoint(sbi, true);
 338                 }
 339
 340                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 341                                         page->index != nid_of_node(page));
 342
 343                 dec_page_count(sbi, type);
 344                 if (f2fs_in_warm_node_list(sbi, page))
 345                         f2fs_del_fsync_node_entry(sbi, page);
 346                 clear_page_private_gcing(page);
 347                 end_page_writeback(page);
 348         }
 349         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 350                                 wq_has_sleeper(&sbi->cp_wait))
 351                 wake_up(&sbi->cp_wait);
 352
 353         bio_put(bio);
 354 }
 355
 356 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 357                 block_t blk_addr, sector_t *sector)
 358 {
 359         struct block_device *bdev = sbi->sb->s_bdev;
 360         int i;
 361
 362         if (f2fs_is_multi_device(sbi)) {
 363                 for (i = 0; i < sbi->s_ndevs; i++) {
 364                         if (FDEV(i).start_blk <= blk_addr &&
 365                             FDEV(i).end_blk >= blk_addr) {
 366                                 blk_addr -= FDEV(i).start_blk;
 367                                 bdev = FDEV(i).bdev;
 368                                 break;
 369                         }
 370                 }
 371         }
 372
 373         if (sector)
 374                 *sector = SECTOR_FROM_BLOCK(blk_addr);
 375         return bdev;
 376 }
 377
 378 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 379 {
 380         int i;
 381
 382         if (!f2fs_is_multi_device(sbi))
 383                 return 0;
 384
 385         for (i = 0; i < sbi->s_ndevs; i++)
 386                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 387                         return i;
 388         return 0;
 389 }
 390
 391 static void __attach_io_flag(struct f2fs_io_info *fio, unsigned int io_flag)
 392 {
 393         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 394         unsigned int fua_flag = io_flag & temp_mask;
 395         unsigned int meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 396
 397         /*
 398          * data/node io flag bits per temp:
 399          *      REQ_META     |      REQ_FUA      |
 400          *    5 |    4 |   3 |    2 |    1 |   0 |
 401          * Cold | Warm | Hot | Cold | Warm | Hot |
 402          */
 403         if ((1 << fio->temp) & meta_flag)
 404                 fio->op_flags |= REQ_META;
 405         if ((1 << fio->temp) & fua_flag)
 406                 fio->op_flags |= REQ_FUA;
 407 }
 408
 409 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 410 {
 411         struct f2fs_sb_info *sbi = fio->sbi;
 412         struct block_device *bdev;
 413         sector_t sector;
 414         struct bio *bio;
 415
 416         if (fio->type == DATA)
 417                 __attach_io_flag(fio, sbi->data_io_flag);
 418         else if (fio->type == NODE)
 419                 __attach_io_flag(fio, sbi->node_io_flag);
 420
 421         bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 422         bio = bio_alloc_bioset(bdev, npages, fio->op | fio->op_flags, GFP_NOIO,
 423                                &f2fs_bioset);
 424         bio->bi_iter.bi_sector = sector;
 425         if (is_read_io(fio->op)) {
 426                 bio->bi_end_io = f2fs_read_end_io;
 427                 bio->bi_private = NULL;
 428         } else {
 429                 bio->bi_end_io = f2fs_write_end_io;
 430                 bio->bi_private = sbi;
 431         }
 432         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 433
 434         if (fio->io_wbc)
 435                 wbc_init_bio(fio->io_wbc, bio);
 436
 437         return bio;
 438 }
 439
 440 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 441                                   pgoff_t first_idx,
 442                                   const struct f2fs_io_info *fio,
 443                                   gfp_t gfp_mask)
 444 {
 445         /*
 446          * The f2fs garbage collector sets ->encrypted_page when it wants to
 447          * read/write raw data without encryption.
 448          */
 449         if (!fio || !fio->encrypted_page)
 450                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 451 }
 452
 453 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 454                                      pgoff_t next_idx,
 455                                      const struct f2fs_io_info *fio)
 456 {
 457         /*
 458          * The f2fs garbage collector sets ->encrypted_page when it wants to
 459          * read/write raw data without encryption.
 460          */
 461         if (fio && fio->encrypted_page)
 462                 return !bio_has_crypt_ctx(bio);
 463
 464         return fscrypt_mergeable_bio(bio, inode, next_idx);
 465 }
 466
 467 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 468                                 struct bio *bio, enum page_type type)
 469 {
 470         if (!is_read_io(bio_op(bio))) {
 471                 unsigned int start;
 472
 473                 if (type != DATA && type != NODE)
 474                         goto submit_io;
 475
 476                 if (f2fs_lfs_mode(sbi) && current->plug)
 477                         blk_finish_plug(current->plug);
 478
 479                 if (!F2FS_IO_ALIGNED(sbi))
 480                         goto submit_io;
 481
 482                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 483                 start %= F2FS_IO_SIZE(sbi);
 484
 485                 if (start == 0)
 486                         goto submit_io;
 487
 488                 /* fill dummy pages */
 489                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 490                         struct page *page =
 491                                 mempool_alloc(sbi->write_io_dummy,
 492                                               GFP_NOIO | __GFP_NOFAIL);
 493                         f2fs_bug_on(sbi, !page);
 494
 495                         lock_page(page);
 496
 497                         zero_user_segment(page, 0, PAGE_SIZE);
 498                         set_page_private_dummy(page);
 499
 500                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 501                                 f2fs_bug_on(sbi, 1);
 502                 }
 503                 /*
 504                  * In the NODE case, we lose next block address chain. So, we
 505                  * need to do checkpoint in f2fs_sync_file.
 506                  */
 507                 if (type == NODE)
 508                         set_sbi_flag(sbi, SBI_NEED_CP);
 509         }
 510 submit_io:
 511         if (is_read_io(bio_op(bio)))
 512                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 513         else
 514                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 515
 516         iostat_update_submit_ctx(bio, type);
 517         submit_bio(bio);
 518 }
 519
 520 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 521                                 struct bio *bio, enum page_type type)
 522 {
 523         __submit_bio(sbi, bio, type);
 524 }
 525
 526 static void __submit_merged_bio(struct f2fs_bio_info *io)
 527 {
 528         struct f2fs_io_info *fio = &io->fio;
 529
 530         if (!io->bio)
 531                 return;
 532
 533         if (is_read_io(fio->op))
 534                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 535         else
 536                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 537
 538         __submit_bio(io->sbi, io->bio, fio->type);
 539         io->bio = NULL;
 540 }
 541
 542 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 543                                                 struct page *page, nid_t ino)
 544 {
 545         struct bio_vec *bvec;
 546         struct bvec_iter_all iter_all;
 547
 548         if (!bio)
 549                 return false;
 550
 551         if (!inode && !page && !ino)
 552                 return true;
 553
 554         bio_for_each_segment_all(bvec, bio, iter_all) {
 555                 struct page *target = bvec->bv_page;
 556
 557                 if (fscrypt_is_bounce_page(target)) {
 558                         target = fscrypt_pagecache_page(target);
 559                         if (IS_ERR(target))
 560                                 continue;
 561                 }
 562                 if (f2fs_is_compressed_page(target)) {
 563                         target = f2fs_compress_control_page(target);
 564                         if (IS_ERR(target))
 565                                 continue;
 566                 }
 567
 568                 if (inode && inode == target->mapping->host)
 569                         return true;
 570                 if (page && page == target)
 571                         return true;
 572                 if (ino && ino == ino_of_node(target))
 573                         return true;
 574         }
 575
 576         return false;
 577 }
 578
 579 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 580                                 enum page_type type, enum temp_type temp)
 581 {
 582         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 583         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 584
 585         f2fs_down_write(&io->io_rwsem);
 586
 587         /* change META to META_FLUSH in the checkpoint procedure */
 588         if (type >= META_FLUSH) {
 589                 io->fio.type = META_FLUSH;
 590                 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 591                 if (!test_opt(sbi, NOBARRIER))
 592                         io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 593         }
 594         __submit_merged_bio(io);
 595         f2fs_up_write(&io->io_rwsem);
 596 }
 597
 598 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 599                                 struct inode *inode, struct page *page,
 600                                 nid_t ino, enum page_type type, bool force)
 601 {
 602         enum temp_type temp;
 603         bool ret = true;
 604
 605         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 606                 if (!force)     {
 607                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 608                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 609
 610                         f2fs_down_read(&io->io_rwsem);
 611                         ret = __has_merged_page(io->bio, inode, page, ino);
 612                         f2fs_up_read(&io->io_rwsem);
 613                 }
 614                 if (ret)
 615                         __f2fs_submit_merged_write(sbi, type, temp);
 616
 617                 /* TODO: use HOT temp only for meta pages now. */
 618                 if (type >= META)
 619                         break;
 620         }
 621 }
 622
 623 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 624 {
 625         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 626 }
 627
 628 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 629                                 struct inode *inode, struct page *page,
 630                                 nid_t ino, enum page_type type)
 631 {
 632         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 633 }
 634
 635 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 636 {
 637         f2fs_submit_merged_write(sbi, DATA);
 638         f2fs_submit_merged_write(sbi, NODE);
 639         f2fs_submit_merged_write(sbi, META);
 640 }
 641
 642 /*
 643  * Fill the locked page with data located in the block address.
 644  * A caller needs to unlock the page on failure.
 645  */
 646 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 647 {
 648         struct bio *bio;
 649         struct page *page = fio->encrypted_page ?
 650                         fio->encrypted_page : fio->page;
 651
 652         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 653                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 654                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 655                 return -EFSCORRUPTED;
 656
 657         trace_f2fs_submit_page_bio(page, fio);
 658
 659         /* Allocate a new bio */
 660         bio = __bio_alloc(fio, 1);
 661
 662         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 663                                fio->page->index, fio, GFP_NOIO);
 664
 665         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 666                 bio_put(bio);
 667                 return -EFAULT;
 668         }
 669
 670         if (fio->io_wbc && !is_read_io(fio->op))
 671                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 672
 673         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 674                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 675
 676         __submit_bio(fio->sbi, bio, fio->type);
 677         return 0;
 678 }
 679
 680 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 681                                 block_t last_blkaddr, block_t cur_blkaddr)
 682 {
 683         if (unlikely(sbi->max_io_bytes &&
 684                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 685                 return false;
 686         if (last_blkaddr + 1 != cur_blkaddr)
 687                 return false;
 688         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 689 }
 690
 691 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 692                                                 struct f2fs_io_info *fio)
 693 {
 694         if (io->fio.op != fio->op)
 695                 return false;
 696         return io->fio.op_flags == fio->op_flags;
 697 }
 698
 699 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 700                                         struct f2fs_bio_info *io,
 701                                         struct f2fs_io_info *fio,
 702                                         block_t last_blkaddr,
 703                                         block_t cur_blkaddr)
 704 {
 705         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 706                 unsigned int filled_blocks =
 707                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 708                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 709                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 710
 711                 /* IOs in bio is aligned and left space of vectors is not enough */
 712                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 713                         return false;
 714         }
 715         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 716                 return false;
 717         return io_type_is_mergeable(io, fio);
 718 }
 719
 720 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 721                                 struct page *page, enum temp_type temp)
 722 {
 723         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 724         struct bio_entry *be;
 725
 726         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 727         be->bio = bio;
 728         bio_get(bio);
 729
 730         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 731                 f2fs_bug_on(sbi, 1);
 732
 733         f2fs_down_write(&io->bio_list_lock);
 734         list_add_tail(&be->list, &io->bio_list);
 735         f2fs_up_write(&io->bio_list_lock);
 736 }
 737
 738 static void del_bio_entry(struct bio_entry *be)
 739 {
 740         list_del(&be->list);
 741         kmem_cache_free(bio_entry_slab, be);
 742 }
 743
 744 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 745                                                         struct page *page)
 746 {
 747         struct f2fs_sb_info *sbi = fio->sbi;
 748         enum temp_type temp;
 749         bool found = false;
 750         int ret = -EAGAIN;
 751
 752         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 753                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 754                 struct list_head *head = &io->bio_list;
 755                 struct bio_entry *be;
 756
 757                 f2fs_down_write(&io->bio_list_lock);
 758                 list_for_each_entry(be, head, list) {
 759                         if (be->bio != *bio)
 760                                 continue;
 761
 762                         found = true;
 763
 764                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 765                                                             *fio->last_block,
 766                                                             fio->new_blkaddr));
 767                         if (f2fs_crypt_mergeable_bio(*bio,
 768                                         fio->page->mapping->host,
 769                                         fio->page->index, fio) &&
 770                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 771                                         PAGE_SIZE) {
 772                                 ret = 0;
 773                                 break;
 774                         }
 775
 776                         /* page can't be merged into bio; submit the bio */
 777                         del_bio_entry(be);
 778                         __submit_bio(sbi, *bio, DATA);
 779                         break;
 780                 }
 781                 f2fs_up_write(&io->bio_list_lock);
 782         }
 783
 784         if (ret) {
 785                 bio_put(*bio);
 786                 *bio = NULL;
 787         }
 788
 789         return ret;
 790 }
 791
 792 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 793                                         struct bio **bio, struct page *page)
 794 {
 795         enum temp_type temp;
 796         bool found = false;
 797         struct bio *target = bio ? *bio : NULL;
 798
 799         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 800                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 801                 struct list_head *head = &io->bio_list;
 802                 struct bio_entry *be;
 803
 804                 if (list_empty(head))
 805                         continue;
 806
 807                 f2fs_down_read(&io->bio_list_lock);
 808                 list_for_each_entry(be, head, list) {
 809                         if (target)
 810                                 found = (target == be->bio);
 811                         else
 812                                 found = __has_merged_page(be->bio, NULL,
 813                                                                 page, 0);
 814                         if (found)
 815                                 break;
 816                 }
 817                 f2fs_up_read(&io->bio_list_lock);
 818
 819                 if (!found)
 820                         continue;
 821
 822                 found = false;
 823
 824                 f2fs_down_write(&io->bio_list_lock);
 825                 list_for_each_entry(be, head, list) {
 826                         if (target)
 827                                 found = (target == be->bio);
 828                         else
 829                                 found = __has_merged_page(be->bio, NULL,
 830                                                                 page, 0);
 831                         if (found) {
 832                                 target = be->bio;
 833                                 del_bio_entry(be);
 834                                 break;
 835                         }
 836                 }
 837                 f2fs_up_write(&io->bio_list_lock);
 838         }
 839
 840         if (found)
 841                 __submit_bio(sbi, target, DATA);
 842         if (bio && *bio) {
 843                 bio_put(*bio);
 844                 *bio = NULL;
 845         }
 846 }
 847
 848 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 849 {
 850         struct bio *bio = *fio->bio;
 851         struct page *page = fio->encrypted_page ?
 852                         fio->encrypted_page : fio->page;
 853
 854         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 855                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 856                 return -EFSCORRUPTED;
 857
 858         trace_f2fs_submit_page_bio(page, fio);
 859
 860         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 861                                                 fio->new_blkaddr))
 862                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 863 alloc_new:
 864         if (!bio) {
 865                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 866                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 867                                        fio->page->index, fio, GFP_NOIO);
 868
 869                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 870         } else {
 871                 if (add_ipu_page(fio, &bio, page))
 872                         goto alloc_new;
 873         }
 874
 875         if (fio->io_wbc)
 876                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 877
 878         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 879
 880         *fio->last_block = fio->new_blkaddr;
 881         *fio->bio = bio;
 882
 883         return 0;
 884 }
 885
 886 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 887 {
 888         struct f2fs_sb_info *sbi = fio->sbi;
 889         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 890         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 891         struct page *bio_page;
 892
 893         f2fs_bug_on(sbi, is_read_io(fio->op));
 894
 895         f2fs_down_write(&io->io_rwsem);
 896 next:
 897         if (fio->in_list) {
 898                 spin_lock(&io->io_lock);
 899                 if (list_empty(&io->io_list)) {
 900                         spin_unlock(&io->io_lock);
 901                         goto out;
 902                 }
 903                 fio = list_first_entry(&io->io_list,
 904                                                 struct f2fs_io_info, list);
 905                 list_del(&fio->list);
 906                 spin_unlock(&io->io_lock);
 907         }
 908
 909         verify_fio_blkaddr(fio);
 910
 911         if (fio->encrypted_page)
 912                 bio_page = fio->encrypted_page;
 913         else if (fio->compressed_page)
 914                 bio_page = fio->compressed_page;
 915         else
 916                 bio_page = fio->page;
 917
 918         /* set submitted = true as a return value */
 919         fio->submitted = true;
 920
 921         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 922
 923         if (io->bio &&
 924             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 925                               fio->new_blkaddr) ||
 926              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 927                                        bio_page->index, fio)))
 928                 __submit_merged_bio(io);
 929 alloc_new:
 930         if (io->bio == NULL) {
 931                 if (F2FS_IO_ALIGNED(sbi) &&
 932                                 (fio->type == DATA || fio->type == NODE) &&
 933                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 934                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 935                         fio->retry = true;
 936                         goto skip;
 937                 }
 938                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
 939                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 940                                        bio_page->index, fio, GFP_NOIO);
 941                 io->fio = *fio;
 942         }
 943
 944         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 945                 __submit_merged_bio(io);
 946                 goto alloc_new;
 947         }
 948
 949         if (fio->io_wbc)
 950                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
 951
 952         io->last_block_in_bio = fio->new_blkaddr;
 953
 954         trace_f2fs_submit_page_write(fio->page, fio);
 955 skip:
 956         if (fio->in_list)
 957                 goto next;
 958 out:
 959         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 960                                 !f2fs_is_checkpoint_ready(sbi))
 961                 __submit_merged_bio(io);
 962         f2fs_up_write(&io->io_rwsem);
 963 }
 964
 965 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 966                                       unsigned nr_pages, unsigned op_flag,
 967                                       pgoff_t first_idx, bool for_write)
 968 {
 969         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 970         struct bio *bio;
 971         struct bio_post_read_ctx *ctx = NULL;
 972         unsigned int post_read_steps = 0;
 973         sector_t sector;
 974         struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
 975
 976         bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
 977                                REQ_OP_READ | op_flag,
 978                                for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
 979         if (!bio)
 980                 return ERR_PTR(-ENOMEM);
 981         bio->bi_iter.bi_sector = sector;
 982         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
 983         bio->bi_end_io = f2fs_read_end_io;
 984
 985         if (fscrypt_inode_uses_fs_layer_crypto(inode))
 986                 post_read_steps |= STEP_DECRYPT;
 987
 988         if (f2fs_need_verity(inode, first_idx))
 989                 post_read_steps |= STEP_VERITY;
 990
 991         /*
 992          * STEP_DECOMPRESS is handled specially, since a compressed file might
 993          * contain both compressed and uncompressed clusters.  We'll allocate a
 994          * bio_post_read_ctx if the file is compressed, but the caller is
 995          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
 996          */
 997
 998         if (post_read_steps || f2fs_compressed_file(inode)) {
 999                 /* Due to the mempool, this never fails. */
1000                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1001                 ctx->bio = bio;
1002                 ctx->sbi = sbi;
1003                 ctx->enabled_steps = post_read_steps;
1004                 ctx->fs_blkaddr = blkaddr;
1005                 bio->bi_private = ctx;
1006         }
1007         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1008
1009         return bio;
1010 }
1011
1012 /* This can handle encryption stuffs */
1013 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1014                                  block_t blkaddr, int op_flags, bool for_write)
1015 {
1016         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1017         struct bio *bio;
1018
1019         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1020                                         page->index, for_write);
1021         if (IS_ERR(bio))
1022                 return PTR_ERR(bio);
1023
1024         /* wait for GCed page writeback via META_MAPPING */
1025         f2fs_wait_on_block_writeback(inode, blkaddr);
1026
1027         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1028                 bio_put(bio);
1029                 return -EFAULT;
1030         }
1031         ClearPageError(page);
1032         inc_page_count(sbi, F2FS_RD_DATA);
1033         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1034         __submit_bio(sbi, bio, DATA);
1035         return 0;
1036 }
1037
1038 static void __set_data_blkaddr(struct dnode_of_data *dn)
1039 {
1040         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1041         __le32 *addr_array;
1042         int base = 0;
1043
1044         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1045                 base = get_extra_isize(dn->inode);
1046
1047         /* Get physical address of data block */
1048         addr_array = blkaddr_in_node(rn);
1049         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1050 }
1051
1052 /*
1053  * Lock ordering for the change of data block address:
1054  * ->data_page
1055  *  ->node_page
1056  *    update block addresses in the node page
1057  */
1058 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1059 {
1060         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1061         __set_data_blkaddr(dn);
1062         if (set_page_dirty(dn->node_page))
1063                 dn->node_changed = true;
1064 }
1065
1066 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1067 {
1068         dn->data_blkaddr = blkaddr;
1069         f2fs_set_data_blkaddr(dn);
1070         f2fs_update_extent_cache(dn);
1071 }
1072
1073 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1074 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1075 {
1076         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1077         int err;
1078
1079         if (!count)
1080                 return 0;
1081
1082         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1083                 return -EPERM;
1084         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1085                 return err;
1086
1087         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1088                                                 dn->ofs_in_node, count);
1089
1090         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1091
1092         for (; count > 0; dn->ofs_in_node++) {
1093                 block_t blkaddr = f2fs_data_blkaddr(dn);
1094
1095                 if (blkaddr == NULL_ADDR) {
1096                         dn->data_blkaddr = NEW_ADDR;
1097                         __set_data_blkaddr(dn);
1098                         count--;
1099                 }
1100         }
1101
1102         if (set_page_dirty(dn->node_page))
1103                 dn->node_changed = true;
1104         return 0;
1105 }
1106
1107 /* Should keep dn->ofs_in_node unchanged */
1108 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1109 {
1110         unsigned int ofs_in_node = dn->ofs_in_node;
1111         int ret;
1112
1113         ret = f2fs_reserve_new_blocks(dn, 1);
1114         dn->ofs_in_node = ofs_in_node;
1115         return ret;
1116 }
1117
1118 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1119 {
1120         bool need_put = dn->inode_page ? false : true;
1121         int err;
1122
1123         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1124         if (err)
1125                 return err;
1126
1127         if (dn->data_blkaddr == NULL_ADDR)
1128                 err = f2fs_reserve_new_block(dn);
1129         if (err || need_put)
1130                 f2fs_put_dnode(dn);
1131         return err;
1132 }
1133
1134 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1135 {
1136         struct extent_info ei = {0, };
1137         struct inode *inode = dn->inode;
1138
1139         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1140                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1141                 return 0;
1142         }
1143
1144         return f2fs_reserve_block(dn, index);
1145 }
1146
1147 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1148                                                 int op_flags, bool for_write)
1149 {
1150         struct address_space *mapping = inode->i_mapping;
1151         struct dnode_of_data dn;
1152         struct page *page;
1153         struct extent_info ei = {0, };
1154         int err;
1155
1156         page = f2fs_grab_cache_page(mapping, index, for_write);
1157         if (!page)
1158                 return ERR_PTR(-ENOMEM);
1159
1160         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1161                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1162                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1163                                                 DATA_GENERIC_ENHANCE_READ)) {
1164                         err = -EFSCORRUPTED;
1165                         goto put_err;
1166                 }
1167                 goto got_it;
1168         }
1169
1170         set_new_dnode(&dn, inode, NULL, NULL, 0);
1171         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1172         if (err)
1173                 goto put_err;
1174         f2fs_put_dnode(&dn);
1175
1176         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1177                 err = -ENOENT;
1178                 goto put_err;
1179         }
1180         if (dn.data_blkaddr != NEW_ADDR &&
1181                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1182                                                 dn.data_blkaddr,
1183                                                 DATA_GENERIC_ENHANCE)) {
1184                 err = -EFSCORRUPTED;
1185                 goto put_err;
1186         }
1187 got_it:
1188         if (PageUptodate(page)) {
1189                 unlock_page(page);
1190                 return page;
1191         }
1192
1193         /*
1194          * A new dentry page is allocated but not able to be written, since its
1195          * new inode page couldn't be allocated due to -ENOSPC.
1196          * In such the case, its blkaddr can be remained as NEW_ADDR.
1197          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1198          * f2fs_init_inode_metadata.
1199          */
1200         if (dn.data_blkaddr == NEW_ADDR) {
1201                 zero_user_segment(page, 0, PAGE_SIZE);
1202                 if (!PageUptodate(page))
1203                         SetPageUptodate(page);
1204                 unlock_page(page);
1205                 return page;
1206         }
1207
1208         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1209                                                 op_flags, for_write);
1210         if (err)
1211                 goto put_err;
1212         return page;
1213
1214 put_err:
1215         f2fs_put_page(page, 1);
1216         return ERR_PTR(err);
1217 }
1218
1219 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1220 {
1221         struct address_space *mapping = inode->i_mapping;
1222         struct page *page;
1223
1224         page = find_get_page(mapping, index);
1225         if (page && PageUptodate(page))
1226                 return page;
1227         f2fs_put_page(page, 0);
1228
1229         page = f2fs_get_read_data_page(inode, index, 0, false);
1230         if (IS_ERR(page))
1231                 return page;
1232
1233         if (PageUptodate(page))
1234                 return page;
1235
1236         wait_on_page_locked(page);
1237         if (unlikely(!PageUptodate(page))) {
1238                 f2fs_put_page(page, 0);
1239                 return ERR_PTR(-EIO);
1240         }
1241         return page;
1242 }
1243
1244 /*
1245  * If it tries to access a hole, return an error.
1246  * Because, the callers, functions in dir.c and GC, should be able to know
1247  * whether this page exists or not.
1248  */
1249 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1250                                                         bool for_write)
1251 {
1252         struct address_space *mapping = inode->i_mapping;
1253         struct page *page;
1254 repeat:
1255         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1256         if (IS_ERR(page))
1257                 return page;
1258
1259         /* wait for read completion */
1260         lock_page(page);
1261         if (unlikely(page->mapping != mapping)) {
1262                 f2fs_put_page(page, 1);
1263                 goto repeat;
1264         }
1265         if (unlikely(!PageUptodate(page))) {
1266                 f2fs_put_page(page, 1);
1267                 return ERR_PTR(-EIO);
1268         }
1269         return page;
1270 }
1271
1272 /*
1273  * Caller ensures that this data page is never allocated.
1274  * A new zero-filled data page is allocated in the page cache.
1275  *
1276  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1277  * f2fs_unlock_op().
1278  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1279  * ipage should be released by this function.
1280  */
1281 struct page *f2fs_get_new_data_page(struct inode *inode,
1282                 struct page *ipage, pgoff_t index, bool new_i_size)
1283 {
1284         struct address_space *mapping = inode->i_mapping;
1285         struct page *page;
1286         struct dnode_of_data dn;
1287         int err;
1288
1289         page = f2fs_grab_cache_page(mapping, index, true);
1290         if (!page) {
1291                 /*
1292                  * before exiting, we should make sure ipage will be released
1293                  * if any error occur.
1294                  */
1295                 f2fs_put_page(ipage, 1);
1296                 return ERR_PTR(-ENOMEM);
1297         }
1298
1299         set_new_dnode(&dn, inode, ipage, NULL, 0);
1300         err = f2fs_reserve_block(&dn, index);
1301         if (err) {
1302                 f2fs_put_page(page, 1);
1303                 return ERR_PTR(err);
1304         }
1305         if (!ipage)
1306                 f2fs_put_dnode(&dn);
1307
1308         if (PageUptodate(page))
1309                 goto got_it;
1310
1311         if (dn.data_blkaddr == NEW_ADDR) {
1312                 zero_user_segment(page, 0, PAGE_SIZE);
1313                 if (!PageUptodate(page))
1314                         SetPageUptodate(page);
1315         } else {
1316                 f2fs_put_page(page, 1);
1317
1318                 /* if ipage exists, blkaddr should be NEW_ADDR */
1319                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1320                 page = f2fs_get_lock_data_page(inode, index, true);
1321                 if (IS_ERR(page))
1322                         return page;
1323         }
1324 got_it:
1325         if (new_i_size && i_size_read(inode) <
1326                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1327                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1328         return page;
1329 }
1330
1331 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1332 {
1333         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1334         struct f2fs_summary sum;
1335         struct node_info ni;
1336         block_t old_blkaddr;
1337         blkcnt_t count = 1;
1338         int err;
1339
1340         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1341                 return -EPERM;
1342
1343         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1344         if (err)
1345                 return err;
1346
1347         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1348         if (dn->data_blkaddr != NULL_ADDR)
1349                 goto alloc;
1350
1351         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1352                 return err;
1353
1354 alloc:
1355         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1356         old_blkaddr = dn->data_blkaddr;
1357         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1358                                 &sum, seg_type, NULL);
1359         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1360                 invalidate_mapping_pages(META_MAPPING(sbi),
1361                                         old_blkaddr, old_blkaddr);
1362                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1363         }
1364         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1365         return 0;
1366 }
1367
1368 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1369 {
1370         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1371                 if (lock)
1372                         f2fs_down_read(&sbi->node_change);
1373                 else
1374                         f2fs_up_read(&sbi->node_change);
1375         } else {
1376                 if (lock)
1377                         f2fs_lock_op(sbi);
1378                 else
1379                         f2fs_unlock_op(sbi);
1380         }
1381 }
1382
1383 /*
1384  * f2fs_map_blocks() tries to find or build mapping relationship which
1385  * maps continuous logical blocks to physical blocks, and return such
1386  * info via f2fs_map_blocks structure.
1387  */
1388 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1389                                                 int create, int flag)
1390 {
1391         unsigned int maxblocks = map->m_len;
1392         struct dnode_of_data dn;
1393         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1394         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1395         pgoff_t pgofs, end_offset, end;
1396         int err = 0, ofs = 1;
1397         unsigned int ofs_in_node, last_ofs_in_node;
1398         blkcnt_t prealloc;
1399         struct extent_info ei = {0, };
1400         block_t blkaddr;
1401         unsigned int start_pgofs;
1402         int bidx = 0;
1403
1404         if (!maxblocks)
1405                 return 0;
1406
1407         map->m_bdev = inode->i_sb->s_bdev;
1408         map->m_multidev_dio =
1409                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1410
1411         map->m_len = 0;
1412         map->m_flags = 0;
1413
1414         /* it only supports block size == page size */
1415         pgofs = (pgoff_t)map->m_lblk;
1416         end = pgofs + maxblocks;
1417
1418         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1419                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1420                                                         map->m_may_create)
1421                         goto next_dnode;
1422
1423                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1424                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1425                 map->m_flags = F2FS_MAP_MAPPED;
1426                 if (map->m_next_extent)
1427                         *map->m_next_extent = pgofs + map->m_len;
1428
1429                 /* for hardware encryption, but to avoid potential issue in future */
1430                 if (flag == F2FS_GET_BLOCK_DIO)
1431                         f2fs_wait_on_block_writeback_range(inode,
1432                                                 map->m_pblk, map->m_len);
1433
1434                 if (map->m_multidev_dio) {
1435                         block_t blk_addr = map->m_pblk;
1436
1437                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1438
1439                         map->m_bdev = FDEV(bidx).bdev;
1440                         map->m_pblk -= FDEV(bidx).start_blk;
1441                         map->m_len = min(map->m_len,
1442                                 FDEV(bidx).end_blk + 1 - map->m_pblk);
1443
1444                         if (map->m_may_create)
1445                                 f2fs_update_device_state(sbi, inode->i_ino,
1446                                                         blk_addr, map->m_len);
1447                 }
1448                 goto out;
1449         }
1450
1451 next_dnode:
1452         if (map->m_may_create)
1453                 f2fs_do_map_lock(sbi, flag, true);
1454
1455         /* When reading holes, we need its node page */
1456         set_new_dnode(&dn, inode, NULL, NULL, 0);
1457         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1458         if (err) {
1459                 if (flag == F2FS_GET_BLOCK_BMAP)
1460                         map->m_pblk = 0;
1461
1462                 if (err == -ENOENT) {
1463                         /*
1464                          * There is one exceptional case that read_node_page()
1465                          * may return -ENOENT due to filesystem has been
1466                          * shutdown or cp_error, so force to convert error
1467                          * number to EIO for such case.
1468                          */
1469                         if (map->m_may_create &&
1470                                 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1471                                 f2fs_cp_error(sbi))) {
1472                                 err = -EIO;
1473                                 goto unlock_out;
1474                         }
1475
1476                         err = 0;
1477                         if (map->m_next_pgofs)
1478                                 *map->m_next_pgofs =
1479                                         f2fs_get_next_page_offset(&dn, pgofs);
1480                         if (map->m_next_extent)
1481                                 *map->m_next_extent =
1482                                         f2fs_get_next_page_offset(&dn, pgofs);
1483                 }
1484                 goto unlock_out;
1485         }
1486
1487         start_pgofs = pgofs;
1488         prealloc = 0;
1489         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1490         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1491
1492 next_block:
1493         blkaddr = f2fs_data_blkaddr(&dn);
1494
1495         if (__is_valid_data_blkaddr(blkaddr) &&
1496                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1497                 err = -EFSCORRUPTED;
1498                 goto sync_out;
1499         }
1500
1501         if (__is_valid_data_blkaddr(blkaddr)) {
1502                 /* use out-place-update for driect IO under LFS mode */
1503                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1504                                                         map->m_may_create) {
1505                         err = __allocate_data_block(&dn, map->m_seg_type);
1506                         if (err)
1507                                 goto sync_out;
1508                         blkaddr = dn.data_blkaddr;
1509                         set_inode_flag(inode, FI_APPEND_WRITE);
1510                 }
1511         } else {
1512                 if (create) {
1513                         if (unlikely(f2fs_cp_error(sbi))) {
1514                                 err = -EIO;
1515                                 goto sync_out;
1516                         }
1517                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1518                                 if (blkaddr == NULL_ADDR) {
1519                                         prealloc++;
1520                                         last_ofs_in_node = dn.ofs_in_node;
1521                                 }
1522                         } else {
1523                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1524                                         flag != F2FS_GET_BLOCK_DIO);
1525                                 err = __allocate_data_block(&dn,
1526                                                         map->m_seg_type);
1527                                 if (!err) {
1528                                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1529                                                 file_need_truncate(inode);
1530                                         set_inode_flag(inode, FI_APPEND_WRITE);
1531                                 }
1532                         }
1533                         if (err)
1534                                 goto sync_out;
1535                         map->m_flags |= F2FS_MAP_NEW;
1536                         blkaddr = dn.data_blkaddr;
1537                 } else {
1538                         if (f2fs_compressed_file(inode) &&
1539                                         f2fs_sanity_check_cluster(&dn) &&
1540                                         (flag != F2FS_GET_BLOCK_FIEMAP ||
1541                                         IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1542                                 err = -EFSCORRUPTED;
1543                                 goto sync_out;
1544                         }
1545                         if (flag == F2FS_GET_BLOCK_BMAP) {
1546                                 map->m_pblk = 0;
1547                                 goto sync_out;
1548                         }
1549                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1550                                 goto sync_out;
1551                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1552                                                 blkaddr == NULL_ADDR) {
1553                                 if (map->m_next_pgofs)
1554                                         *map->m_next_pgofs = pgofs + 1;
1555                                 goto sync_out;
1556                         }
1557                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1558                                 /* for defragment case */
1559                                 if (map->m_next_pgofs)
1560                                         *map->m_next_pgofs = pgofs + 1;
1561                                 goto sync_out;
1562                         }
1563                 }
1564         }
1565
1566         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1567                 goto skip;
1568
1569         if (map->m_multidev_dio)
1570                 bidx = f2fs_target_device_index(sbi, blkaddr);
1571
1572         if (map->m_len == 0) {
1573                 /* preallocated unwritten block should be mapped for fiemap. */
1574                 if (blkaddr == NEW_ADDR)
1575                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1576                 map->m_flags |= F2FS_MAP_MAPPED;
1577
1578                 map->m_pblk = blkaddr;
1579                 map->m_len = 1;
1580
1581                 if (map->m_multidev_dio)
1582                         map->m_bdev = FDEV(bidx).bdev;
1583         } else if ((map->m_pblk != NEW_ADDR &&
1584                         blkaddr == (map->m_pblk + ofs)) ||
1585                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1586                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1587                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1588                         goto sync_out;
1589                 ofs++;
1590                 map->m_len++;
1591         } else {
1592                 goto sync_out;
1593         }
1594
1595 skip:
1596         dn.ofs_in_node++;
1597         pgofs++;
1598
1599         /* preallocate blocks in batch for one dnode page */
1600         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1601                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1602
1603                 dn.ofs_in_node = ofs_in_node;
1604                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1605                 if (err)
1606                         goto sync_out;
1607
1608                 map->m_len += dn.ofs_in_node - ofs_in_node;
1609                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1610                         err = -ENOSPC;
1611                         goto sync_out;
1612                 }
1613                 dn.ofs_in_node = end_offset;
1614         }
1615
1616         if (pgofs >= end)
1617                 goto sync_out;
1618         else if (dn.ofs_in_node < end_offset)
1619                 goto next_block;
1620
1621         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1622                 if (map->m_flags & F2FS_MAP_MAPPED) {
1623                         unsigned int ofs = start_pgofs - map->m_lblk;
1624
1625                         f2fs_update_extent_cache_range(&dn,
1626                                 start_pgofs, map->m_pblk + ofs,
1627                                 map->m_len - ofs);
1628                 }
1629         }
1630
1631         f2fs_put_dnode(&dn);
1632
1633         if (map->m_may_create) {
1634                 f2fs_do_map_lock(sbi, flag, false);
1635                 f2fs_balance_fs(sbi, dn.node_changed);
1636         }
1637         goto next_dnode;
1638
1639 sync_out:
1640
1641         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1642                 /*
1643                  * for hardware encryption, but to avoid potential issue
1644                  * in future
1645                  */
1646                 f2fs_wait_on_block_writeback_range(inode,
1647                                                 map->m_pblk, map->m_len);
1648                 invalidate_mapping_pages(META_MAPPING(sbi),
1649                                                 map->m_pblk, map->m_pblk);
1650
1651                 if (map->m_multidev_dio) {
1652                         block_t blk_addr = map->m_pblk;
1653
1654                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1655
1656                         map->m_bdev = FDEV(bidx).bdev;
1657                         map->m_pblk -= FDEV(bidx).start_blk;
1658
1659                         if (map->m_may_create)
1660                                 f2fs_update_device_state(sbi, inode->i_ino,
1661                                                         blk_addr, map->m_len);
1662
1663                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1664                                                 FDEV(bidx).end_blk + 1);
1665                 }
1666         }
1667
1668         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1669                 if (map->m_flags & F2FS_MAP_MAPPED) {
1670                         unsigned int ofs = start_pgofs - map->m_lblk;
1671
1672                         f2fs_update_extent_cache_range(&dn,
1673                                 start_pgofs, map->m_pblk + ofs,
1674                                 map->m_len - ofs);
1675                 }
1676                 if (map->m_next_extent)
1677                         *map->m_next_extent = pgofs + 1;
1678         }
1679         f2fs_put_dnode(&dn);
1680 unlock_out:
1681         if (map->m_may_create) {
1682                 f2fs_do_map_lock(sbi, flag, false);
1683                 f2fs_balance_fs(sbi, dn.node_changed);
1684         }
1685 out:
1686         trace_f2fs_map_blocks(inode, map, create, flag, err);
1687         return err;
1688 }
1689
1690 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1691 {
1692         struct f2fs_map_blocks map;
1693         block_t last_lblk;
1694         int err;
1695
1696         if (pos + len > i_size_read(inode))
1697                 return false;
1698
1699         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1700         map.m_next_pgofs = NULL;
1701         map.m_next_extent = NULL;
1702         map.m_seg_type = NO_CHECK_TYPE;
1703         map.m_may_create = false;
1704         last_lblk = F2FS_BLK_ALIGN(pos + len);
1705
1706         while (map.m_lblk < last_lblk) {
1707                 map.m_len = last_lblk - map.m_lblk;
1708                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1709                 if (err || map.m_len == 0)
1710                         return false;
1711                 map.m_lblk += map.m_len;
1712         }
1713         return true;
1714 }
1715
1716 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1717 {
1718         return (bytes >> inode->i_blkbits);
1719 }
1720
1721 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1722 {
1723         return (blks << inode->i_blkbits);
1724 }
1725
1726 static int f2fs_xattr_fiemap(struct inode *inode,
1727                                 struct fiemap_extent_info *fieinfo)
1728 {
1729         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1730         struct page *page;
1731         struct node_info ni;
1732         __u64 phys = 0, len;
1733         __u32 flags;
1734         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1735         int err = 0;
1736
1737         if (f2fs_has_inline_xattr(inode)) {
1738                 int offset;
1739
1740                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1741                                                 inode->i_ino, false);
1742                 if (!page)
1743                         return -ENOMEM;
1744
1745                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1746                 if (err) {
1747                         f2fs_put_page(page, 1);
1748                         return err;
1749                 }
1750
1751                 phys = blks_to_bytes(inode, ni.blk_addr);
1752                 offset = offsetof(struct f2fs_inode, i_addr) +
1753                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1754                                         get_inline_xattr_addrs(inode));
1755
1756                 phys += offset;
1757                 len = inline_xattr_size(inode);
1758
1759                 f2fs_put_page(page, 1);
1760
1761                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1762
1763                 if (!xnid)
1764                         flags |= FIEMAP_EXTENT_LAST;
1765
1766                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1767                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1768                 if (err || err == 1)
1769                         return err;
1770         }
1771
1772         if (xnid) {
1773                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1774                 if (!page)
1775                         return -ENOMEM;
1776
1777                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1778                 if (err) {
1779                         f2fs_put_page(page, 1);
1780                         return err;
1781                 }
1782
1783                 phys = blks_to_bytes(inode, ni.blk_addr);
1784                 len = inode->i_sb->s_blocksize;
1785
1786                 f2fs_put_page(page, 1);
1787
1788                 flags = FIEMAP_EXTENT_LAST;
1789         }
1790
1791         if (phys) {
1792                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1793                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1794         }
1795
1796         return (err < 0 ? err : 0);
1797 }
1798
1799 static loff_t max_inode_blocks(struct inode *inode)
1800 {
1801         loff_t result = ADDRS_PER_INODE(inode);
1802         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1803
1804         /* two direct node blocks */
1805         result += (leaf_count * 2);
1806
1807         /* two indirect node blocks */
1808         leaf_count *= NIDS_PER_BLOCK;
1809         result += (leaf_count * 2);
1810
1811         /* one double indirect node block */
1812         leaf_count *= NIDS_PER_BLOCK;
1813         result += leaf_count;
1814
1815         return result;
1816 }
1817
1818 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1819                 u64 start, u64 len)
1820 {
1821         struct f2fs_map_blocks map;
1822         sector_t start_blk, last_blk;
1823         pgoff_t next_pgofs;
1824         u64 logical = 0, phys = 0, size = 0;
1825         u32 flags = 0;
1826         int ret = 0;
1827         bool compr_cluster = false, compr_appended;
1828         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1829         unsigned int count_in_cluster = 0;
1830         loff_t maxbytes;
1831
1832         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1833                 ret = f2fs_precache_extents(inode);
1834                 if (ret)
1835                         return ret;
1836         }
1837
1838         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1839         if (ret)
1840                 return ret;
1841
1842         inode_lock(inode);
1843
1844         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1845         if (start > maxbytes) {
1846                 ret = -EFBIG;
1847                 goto out;
1848         }
1849
1850         if (len > maxbytes || (maxbytes - len) < start)
1851                 len = maxbytes - start;
1852
1853         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1854                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1855                 goto out;
1856         }
1857
1858         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1859                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1860                 if (ret != -EAGAIN)
1861                         goto out;
1862         }
1863
1864         if (bytes_to_blks(inode, len) == 0)
1865                 len = blks_to_bytes(inode, 1);
1866
1867         start_blk = bytes_to_blks(inode, start);
1868         last_blk = bytes_to_blks(inode, start + len - 1);
1869
1870 next:
1871         memset(&map, 0, sizeof(map));
1872         map.m_lblk = start_blk;
1873         map.m_len = bytes_to_blks(inode, len);
1874         map.m_next_pgofs = &next_pgofs;
1875         map.m_seg_type = NO_CHECK_TYPE;
1876
1877         if (compr_cluster) {
1878                 map.m_lblk += 1;
1879                 map.m_len = cluster_size - count_in_cluster;
1880         }
1881
1882         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1883         if (ret)
1884                 goto out;
1885
1886         /* HOLE */
1887         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1888                 start_blk = next_pgofs;
1889
1890                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1891                                                 max_inode_blocks(inode)))
1892                         goto prep_next;
1893
1894                 flags |= FIEMAP_EXTENT_LAST;
1895         }
1896
1897         compr_appended = false;
1898         /* In a case of compressed cluster, append this to the last extent */
1899         if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1900                         !(map.m_flags & F2FS_MAP_FLAGS))) {
1901                 compr_appended = true;
1902                 goto skip_fill;
1903         }
1904
1905         if (size) {
1906                 flags |= FIEMAP_EXTENT_MERGED;
1907                 if (IS_ENCRYPTED(inode))
1908                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1909
1910                 ret = fiemap_fill_next_extent(fieinfo, logical,
1911                                 phys, size, flags);
1912                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1913                 if (ret)
1914                         goto out;
1915                 size = 0;
1916         }
1917
1918         if (start_blk > last_blk)
1919                 goto out;
1920
1921 skip_fill:
1922         if (map.m_pblk == COMPRESS_ADDR) {
1923                 compr_cluster = true;
1924                 count_in_cluster = 1;
1925         } else if (compr_appended) {
1926                 unsigned int appended_blks = cluster_size -
1927                                                 count_in_cluster + 1;
1928                 size += blks_to_bytes(inode, appended_blks);
1929                 start_blk += appended_blks;
1930                 compr_cluster = false;
1931         } else {
1932                 logical = blks_to_bytes(inode, start_blk);
1933                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
1934                         blks_to_bytes(inode, map.m_pblk) : 0;
1935                 size = blks_to_bytes(inode, map.m_len);
1936                 flags = 0;
1937
1938                 if (compr_cluster) {
1939                         flags = FIEMAP_EXTENT_ENCODED;
1940                         count_in_cluster += map.m_len;
1941                         if (count_in_cluster == cluster_size) {
1942                                 compr_cluster = false;
1943                                 size += blks_to_bytes(inode, 1);
1944                         }
1945                 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
1946                         flags = FIEMAP_EXTENT_UNWRITTEN;
1947                 }
1948
1949                 start_blk += bytes_to_blks(inode, size);
1950         }
1951
1952 prep_next:
1953         cond_resched();
1954         if (fatal_signal_pending(current))
1955                 ret = -EINTR;
1956         else
1957                 goto next;
1958 out:
1959         if (ret == 1)
1960                 ret = 0;
1961
1962         inode_unlock(inode);
1963         return ret;
1964 }
1965
1966 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1967 {
1968         if (IS_ENABLED(CONFIG_FS_VERITY) &&
1969             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1970                 return inode->i_sb->s_maxbytes;
1971
1972         return i_size_read(inode);
1973 }
1974
1975 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1976                                         unsigned nr_pages,
1977                                         struct f2fs_map_blocks *map,
1978                                         struct bio **bio_ret,
1979                                         sector_t *last_block_in_bio,
1980                                         bool is_readahead)
1981 {
1982         struct bio *bio = *bio_ret;
1983         const unsigned blocksize = blks_to_bytes(inode, 1);
1984         sector_t block_in_file;
1985         sector_t last_block;
1986         sector_t last_block_in_file;
1987         sector_t block_nr;
1988         int ret = 0;
1989
1990         block_in_file = (sector_t)page_index(page);
1991         last_block = block_in_file + nr_pages;
1992         last_block_in_file = bytes_to_blks(inode,
1993                         f2fs_readpage_limit(inode) + blocksize - 1);
1994         if (last_block > last_block_in_file)
1995                 last_block = last_block_in_file;
1996
1997         /* just zeroing out page which is beyond EOF */
1998         if (block_in_file >= last_block)
1999                 goto zero_out;
2000         /*
2001          * Map blocks using the previous result first.
2002          */
2003         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2004                         block_in_file > map->m_lblk &&
2005                         block_in_file < (map->m_lblk + map->m_len))
2006                 goto got_it;
2007
2008         /*
2009          * Then do more f2fs_map_blocks() calls until we are
2010          * done with this page.
2011          */
2012         map->m_lblk = block_in_file;
2013         map->m_len = last_block - block_in_file;
2014
2015         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2016         if (ret)
2017                 goto out;
2018 got_it:
2019         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2020                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2021                 SetPageMappedToDisk(page);
2022
2023                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2024                                                 DATA_GENERIC_ENHANCE_READ)) {
2025                         ret = -EFSCORRUPTED;
2026                         goto out;
2027                 }
2028         } else {
2029 zero_out:
2030                 zero_user_segment(page, 0, PAGE_SIZE);
2031                 if (f2fs_need_verity(inode, page->index) &&
2032                     !fsverity_verify_page(page)) {
2033                         ret = -EIO;
2034                         goto out;
2035                 }
2036                 if (!PageUptodate(page))
2037                         SetPageUptodate(page);
2038                 unlock_page(page);
2039                 goto out;
2040         }
2041
2042         /*
2043          * This page will go to BIO.  Do we need to send this
2044          * BIO off first?
2045          */
2046         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2047                                        *last_block_in_bio, block_nr) ||
2048                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2049 submit_and_realloc:
2050                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2051                 bio = NULL;
2052         }
2053         if (bio == NULL) {
2054                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2055                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2056                                 false);
2057                 if (IS_ERR(bio)) {
2058                         ret = PTR_ERR(bio);
2059                         bio = NULL;
2060                         goto out;
2061                 }
2062         }
2063
2064         /*
2065          * If the page is under writeback, we need to wait for
2066          * its completion to see the correct decrypted data.
2067          */
2068         f2fs_wait_on_block_writeback(inode, block_nr);
2069
2070         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2071                 goto submit_and_realloc;
2072
2073         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2074         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2075         ClearPageError(page);
2076         *last_block_in_bio = block_nr;
2077         goto out;
2078 out:
2079         *bio_ret = bio;
2080         return ret;
2081 }
2082
2083 #ifdef CONFIG_F2FS_FS_COMPRESSION
2084 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2085                                 unsigned nr_pages, sector_t *last_block_in_bio,
2086                                 bool is_readahead, bool for_write)
2087 {
2088         struct dnode_of_data dn;
2089         struct inode *inode = cc->inode;
2090         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2091         struct bio *bio = *bio_ret;
2092         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2093         sector_t last_block_in_file;
2094         const unsigned blocksize = blks_to_bytes(inode, 1);
2095         struct decompress_io_ctx *dic = NULL;
2096         struct extent_info ei = {0, };
2097         bool from_dnode = true;
2098         int i;
2099         int ret = 0;
2100
2101         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2102
2103         last_block_in_file = bytes_to_blks(inode,
2104                         f2fs_readpage_limit(inode) + blocksize - 1);
2105
2106         /* get rid of pages beyond EOF */
2107         for (i = 0; i < cc->cluster_size; i++) {
2108                 struct page *page = cc->rpages[i];
2109
2110                 if (!page)
2111                         continue;
2112                 if ((sector_t)page->index >= last_block_in_file) {
2113                         zero_user_segment(page, 0, PAGE_SIZE);
2114                         if (!PageUptodate(page))
2115                                 SetPageUptodate(page);
2116                 } else if (!PageUptodate(page)) {
2117                         continue;
2118                 }
2119                 unlock_page(page);
2120                 if (for_write)
2121                         put_page(page);
2122                 cc->rpages[i] = NULL;
2123                 cc->nr_rpages--;
2124         }
2125
2126         /* we are done since all pages are beyond EOF */
2127         if (f2fs_cluster_is_empty(cc))
2128                 goto out;
2129
2130         if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2131                 from_dnode = false;
2132
2133         if (!from_dnode)
2134                 goto skip_reading_dnode;
2135
2136         set_new_dnode(&dn, inode, NULL, NULL, 0);
2137         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2138         if (ret)
2139                 goto out;
2140
2141         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2142
2143 skip_reading_dnode:
2144         for (i = 1; i < cc->cluster_size; i++) {
2145                 block_t blkaddr;
2146
2147                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2148                                         dn.ofs_in_node + i) :
2149                                         ei.blk + i - 1;
2150
2151                 if (!__is_valid_data_blkaddr(blkaddr))
2152                         break;
2153
2154                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2155                         ret = -EFAULT;
2156                         goto out_put_dnode;
2157                 }
2158                 cc->nr_cpages++;
2159
2160                 if (!from_dnode && i >= ei.c_len)
2161                         break;
2162         }
2163
2164         /* nothing to decompress */
2165         if (cc->nr_cpages == 0) {
2166                 ret = 0;
2167                 goto out_put_dnode;
2168         }
2169
2170         dic = f2fs_alloc_dic(cc);
2171         if (IS_ERR(dic)) {
2172                 ret = PTR_ERR(dic);
2173                 goto out_put_dnode;
2174         }
2175
2176         for (i = 0; i < cc->nr_cpages; i++) {
2177                 struct page *page = dic->cpages[i];
2178                 block_t blkaddr;
2179                 struct bio_post_read_ctx *ctx;
2180
2181                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2182                                         dn.ofs_in_node + i + 1) :
2183                                         ei.blk + i;
2184
2185                 f2fs_wait_on_block_writeback(inode, blkaddr);
2186
2187                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2188                         if (atomic_dec_and_test(&dic->remaining_pages))
2189                                 f2fs_decompress_cluster(dic);
2190                         continue;
2191                 }
2192
2193                 if (bio && (!page_is_mergeable(sbi, bio,
2194                                         *last_block_in_bio, blkaddr) ||
2195                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2196 submit_and_realloc:
2197                         __submit_bio(sbi, bio, DATA);
2198                         bio = NULL;
2199                 }
2200
2201                 if (!bio) {
2202                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2203                                         is_readahead ? REQ_RAHEAD : 0,
2204                                         page->index, for_write);
2205                         if (IS_ERR(bio)) {
2206                                 ret = PTR_ERR(bio);
2207                                 f2fs_decompress_end_io(dic, ret);
2208                                 f2fs_put_dnode(&dn);
2209                                 *bio_ret = NULL;
2210                                 return ret;
2211                         }
2212                 }
2213
2214                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2215                         goto submit_and_realloc;
2216
2217                 ctx = get_post_read_ctx(bio);
2218                 ctx->enabled_steps |= STEP_DECOMPRESS;
2219                 refcount_inc(&dic->refcnt);
2220
2221                 inc_page_count(sbi, F2FS_RD_DATA);
2222                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2223                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2224                 ClearPageError(page);
2225                 *last_block_in_bio = blkaddr;
2226         }
2227
2228         if (from_dnode)
2229                 f2fs_put_dnode(&dn);
2230
2231         *bio_ret = bio;
2232         return 0;
2233
2234 out_put_dnode:
2235         if (from_dnode)
2236                 f2fs_put_dnode(&dn);
2237 out:
2238         for (i = 0; i < cc->cluster_size; i++) {
2239                 if (cc->rpages[i]) {
2240                         ClearPageUptodate(cc->rpages[i]);
2241                         ClearPageError(cc->rpages[i]);
2242                         unlock_page(cc->rpages[i]);
2243                 }
2244         }
2245         *bio_ret = bio;
2246         return ret;
2247 }
2248 #endif
2249
2250 /*
2251  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2252  * Major change was from block_size == page_size in f2fs by default.
2253  */
2254 static int f2fs_mpage_readpages(struct inode *inode,
2255                 struct readahead_control *rac, struct page *page)
2256 {
2257         struct bio *bio = NULL;
2258         sector_t last_block_in_bio = 0;
2259         struct f2fs_map_blocks map;
2260 #ifdef CONFIG_F2FS_FS_COMPRESSION
2261         struct compress_ctx cc = {
2262                 .inode = inode,
2263                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2264                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2265                 .cluster_idx = NULL_CLUSTER,
2266                 .rpages = NULL,
2267                 .cpages = NULL,
2268                 .nr_rpages = 0,
2269                 .nr_cpages = 0,
2270         };
2271         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2272 #endif
2273         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2274         unsigned max_nr_pages = nr_pages;
2275         int ret = 0;
2276
2277         map.m_pblk = 0;
2278         map.m_lblk = 0;
2279         map.m_len = 0;
2280         map.m_flags = 0;
2281         map.m_next_pgofs = NULL;
2282         map.m_next_extent = NULL;
2283         map.m_seg_type = NO_CHECK_TYPE;
2284         map.m_may_create = false;
2285
2286         for (; nr_pages; nr_pages--) {
2287                 if (rac) {
2288                         page = readahead_page(rac);
2289                         prefetchw(&page->flags);
2290                 }
2291
2292 #ifdef CONFIG_F2FS_FS_COMPRESSION
2293                 if (f2fs_compressed_file(inode)) {
2294                         /* there are remained comressed pages, submit them */
2295                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2296                                 ret = f2fs_read_multi_pages(&cc, &bio,
2297                                                         max_nr_pages,
2298                                                         &last_block_in_bio,
2299                                                         rac != NULL, false);
2300                                 f2fs_destroy_compress_ctx(&cc, false);
2301                                 if (ret)
2302                                         goto set_error_page;
2303                         }
2304                         if (cc.cluster_idx == NULL_CLUSTER) {
2305                                 if (nc_cluster_idx ==
2306                                         page->index >> cc.log_cluster_size) {
2307                                         goto read_single_page;
2308                                 }
2309
2310                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2311                                 if (ret < 0)
2312                                         goto set_error_page;
2313                                 else if (!ret) {
2314                                         nc_cluster_idx =
2315                                                 page->index >> cc.log_cluster_size;
2316                                         goto read_single_page;
2317                                 }
2318
2319                                 nc_cluster_idx = NULL_CLUSTER;
2320                         }
2321                         ret = f2fs_init_compress_ctx(&cc);
2322                         if (ret)
2323                                 goto set_error_page;
2324
2325                         f2fs_compress_ctx_add_page(&cc, page);
2326
2327                         goto next_page;
2328                 }
2329 read_single_page:
2330 #endif
2331
2332                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2333                                         &bio, &last_block_in_bio, rac);
2334                 if (ret) {
2335 #ifdef CONFIG_F2FS_FS_COMPRESSION
2336 set_error_page:
2337 #endif
2338                         SetPageError(page);
2339                         zero_user_segment(page, 0, PAGE_SIZE);
2340                         unlock_page(page);
2341                 }
2342 #ifdef CONFIG_F2FS_FS_COMPRESSION
2343 next_page:
2344 #endif
2345                 if (rac)
2346                         put_page(page);
2347
2348 #ifdef CONFIG_F2FS_FS_COMPRESSION
2349                 if (f2fs_compressed_file(inode)) {
2350                         /* last page */
2351                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2352                                 ret = f2fs_read_multi_pages(&cc, &bio,
2353                                                         max_nr_pages,
2354                                                         &last_block_in_bio,
2355                                                         rac != NULL, false);
2356                                 f2fs_destroy_compress_ctx(&cc, false);
2357                         }
2358                 }
2359 #endif
2360         }
2361         if (bio)
2362                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2363         return ret;
2364 }
2365
2366 static int f2fs_read_data_page(struct file *file, struct page *page)
2367 {
2368         struct inode *inode = page_file_mapping(page)->host;
2369         int ret = -EAGAIN;
2370
2371         trace_f2fs_readpage(page, DATA);
2372
2373         if (!f2fs_is_compress_backend_ready(inode)) {
2374                 unlock_page(page);
2375                 return -EOPNOTSUPP;
2376         }
2377
2378         /* If the file has inline data, try to read it directly */
2379         if (f2fs_has_inline_data(inode))
2380                 ret = f2fs_read_inline_data(inode, page);
2381         if (ret == -EAGAIN)
2382                 ret = f2fs_mpage_readpages(inode, NULL, page);
2383         return ret;
2384 }
2385
2386 static void f2fs_readahead(struct readahead_control *rac)
2387 {
2388         struct inode *inode = rac->mapping->host;
2389
2390         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2391
2392         if (!f2fs_is_compress_backend_ready(inode))
2393                 return;
2394
2395         /* If the file has inline data, skip readpages */
2396         if (f2fs_has_inline_data(inode))
2397                 return;
2398
2399         f2fs_mpage_readpages(inode, rac, NULL);
2400 }
2401
2402 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2403 {
2404         struct inode *inode = fio->page->mapping->host;
2405         struct page *mpage, *page;
2406         gfp_t gfp_flags = GFP_NOFS;
2407
2408         if (!f2fs_encrypted_file(inode))
2409                 return 0;
2410
2411         page = fio->compressed_page ? fio->compressed_page : fio->page;
2412
2413         /* wait for GCed page writeback via META_MAPPING */
2414         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2415
2416         if (fscrypt_inode_uses_inline_crypto(inode))
2417                 return 0;
2418
2419 retry_encrypt:
2420         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2421                                         PAGE_SIZE, 0, gfp_flags);
2422         if (IS_ERR(fio->encrypted_page)) {
2423                 /* flush pending IOs and wait for a while in the ENOMEM case */
2424                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2425                         f2fs_flush_merged_writes(fio->sbi);
2426                         memalloc_retry_wait(GFP_NOFS);
2427                         gfp_flags |= __GFP_NOFAIL;
2428                         goto retry_encrypt;
2429                 }
2430                 return PTR_ERR(fio->encrypted_page);
2431         }
2432
2433         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2434         if (mpage) {
2435                 if (PageUptodate(mpage))
2436                         memcpy(page_address(mpage),
2437                                 page_address(fio->encrypted_page), PAGE_SIZE);
2438                 f2fs_put_page(mpage, 1);
2439         }
2440         return 0;
2441 }
2442
2443 static inline bool check_inplace_update_policy(struct inode *inode,
2444                                 struct f2fs_io_info *fio)
2445 {
2446         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2447         unsigned int policy = SM_I(sbi)->ipu_policy;
2448
2449         if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
2450                         is_inode_flag_set(inode, FI_OPU_WRITE))
2451                 return false;
2452         if (policy & (0x1 << F2FS_IPU_FORCE))
2453                 return true;
2454         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2455                 return true;
2456         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2457                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2458                 return true;
2459         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2460                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2461                 return true;
2462
2463         /*
2464          * IPU for rewrite async pages
2465          */
2466         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2467                         fio && fio->op == REQ_OP_WRITE &&
2468                         !(fio->op_flags & REQ_SYNC) &&
2469                         !IS_ENCRYPTED(inode))
2470                 return true;
2471
2472         /* this is only set during fdatasync */
2473         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2474                         is_inode_flag_set(inode, FI_NEED_IPU))
2475                 return true;
2476
2477         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2478                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2479                 return true;
2480
2481         return false;
2482 }
2483
2484 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2485 {
2486         /* swap file is migrating in aligned write mode */
2487         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2488                 return false;
2489
2490         if (f2fs_is_pinned_file(inode))
2491                 return true;
2492
2493         /* if this is cold file, we should overwrite to avoid fragmentation */
2494         if (file_is_cold(inode))
2495                 return true;
2496
2497         return check_inplace_update_policy(inode, fio);
2498 }
2499
2500 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2501 {
2502         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2503
2504         /* The below cases were checked when setting it. */
2505         if (f2fs_is_pinned_file(inode))
2506                 return false;
2507         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2508                 return true;
2509         if (f2fs_lfs_mode(sbi))
2510                 return true;
2511         if (S_ISDIR(inode->i_mode))
2512                 return true;
2513         if (IS_NOQUOTA(inode))
2514                 return true;
2515         if (f2fs_is_atomic_file(inode))
2516                 return true;
2517
2518         /* swap file is migrating in aligned write mode */
2519         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2520                 return true;
2521
2522         if (is_inode_flag_set(inode, FI_OPU_WRITE))
2523                 return true;
2524
2525         if (fio) {
2526                 if (page_private_gcing(fio->page))
2527                         return true;
2528                 if (page_private_dummy(fio->page))
2529                         return true;
2530                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2531                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2532                         return true;
2533         }
2534         return false;
2535 }
2536
2537 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2538 {
2539         struct inode *inode = fio->page->mapping->host;
2540
2541         if (f2fs_should_update_outplace(inode, fio))
2542                 return false;
2543
2544         return f2fs_should_update_inplace(inode, fio);
2545 }
2546
2547 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2548 {
2549         struct page *page = fio->page;
2550         struct inode *inode = page->mapping->host;
2551         struct dnode_of_data dn;
2552         struct extent_info ei = {0, };
2553         struct node_info ni;
2554         bool ipu_force = false;
2555         int err = 0;
2556
2557         set_new_dnode(&dn, inode, NULL, NULL, 0);
2558         if (need_inplace_update(fio) &&
2559                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2560                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2561
2562                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2563                                                 DATA_GENERIC_ENHANCE))
2564                         return -EFSCORRUPTED;
2565
2566                 ipu_force = true;
2567                 fio->need_lock = LOCK_DONE;
2568                 goto got_it;
2569         }
2570
2571         /* Deadlock due to between page->lock and f2fs_lock_op */
2572         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2573                 return -EAGAIN;
2574
2575         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2576         if (err)
2577                 goto out;
2578
2579         fio->old_blkaddr = dn.data_blkaddr;
2580
2581         /* This page is already truncated */
2582         if (fio->old_blkaddr == NULL_ADDR) {
2583                 ClearPageUptodate(page);
2584                 clear_page_private_gcing(page);
2585                 goto out_writepage;
2586         }
2587 got_it:
2588         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2589                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2590                                                 DATA_GENERIC_ENHANCE)) {
2591                 err = -EFSCORRUPTED;
2592                 goto out_writepage;
2593         }
2594         /*
2595          * If current allocation needs SSR,
2596          * it had better in-place writes for updated data.
2597          */
2598         if (ipu_force ||
2599                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2600                                         need_inplace_update(fio))) {
2601                 err = f2fs_encrypt_one_page(fio);
2602                 if (err)
2603                         goto out_writepage;
2604
2605                 set_page_writeback(page);
2606                 ClearPageError(page);
2607                 f2fs_put_dnode(&dn);
2608                 if (fio->need_lock == LOCK_REQ)
2609                         f2fs_unlock_op(fio->sbi);
2610                 err = f2fs_inplace_write_data(fio);
2611                 if (err) {
2612                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2613                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2614                         if (PageWriteback(page))
2615                                 end_page_writeback(page);
2616                 } else {
2617                         set_inode_flag(inode, FI_UPDATE_WRITE);
2618                 }
2619                 trace_f2fs_do_write_data_page(fio->page, IPU);
2620                 return err;
2621         }
2622
2623         if (fio->need_lock == LOCK_RETRY) {
2624                 if (!f2fs_trylock_op(fio->sbi)) {
2625                         err = -EAGAIN;
2626                         goto out_writepage;
2627                 }
2628                 fio->need_lock = LOCK_REQ;
2629         }
2630
2631         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2632         if (err)
2633                 goto out_writepage;
2634
2635         fio->version = ni.version;
2636
2637         err = f2fs_encrypt_one_page(fio);
2638         if (err)
2639                 goto out_writepage;
2640
2641         set_page_writeback(page);
2642         ClearPageError(page);
2643
2644         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2645                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2646
2647         /* LFS mode write path */
2648         f2fs_outplace_write_data(&dn, fio);
2649         trace_f2fs_do_write_data_page(page, OPU);
2650         set_inode_flag(inode, FI_APPEND_WRITE);
2651         if (page->index == 0)
2652                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2653 out_writepage:
2654         f2fs_put_dnode(&dn);
2655 out:
2656         if (fio->need_lock == LOCK_REQ)
2657                 f2fs_unlock_op(fio->sbi);
2658         return err;
2659 }
2660
2661 int f2fs_write_single_data_page(struct page *page, int *submitted,
2662                                 struct bio **bio,
2663                                 sector_t *last_block,
2664                                 struct writeback_control *wbc,
2665                                 enum iostat_type io_type,
2666                                 int compr_blocks,
2667                                 bool allow_balance)
2668 {
2669         struct inode *inode = page->mapping->host;
2670         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2671         loff_t i_size = i_size_read(inode);
2672         const pgoff_t end_index = ((unsigned long long)i_size)
2673                                                         >> PAGE_SHIFT;
2674         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2675         unsigned offset = 0;
2676         bool need_balance_fs = false;
2677         int err = 0;
2678         struct f2fs_io_info fio = {
2679                 .sbi = sbi,
2680                 .ino = inode->i_ino,
2681                 .type = DATA,
2682                 .op = REQ_OP_WRITE,
2683                 .op_flags = wbc_to_write_flags(wbc),
2684                 .old_blkaddr = NULL_ADDR,
2685                 .page = page,
2686                 .encrypted_page = NULL,
2687                 .submitted = false,
2688                 .compr_blocks = compr_blocks,
2689                 .need_lock = LOCK_RETRY,
2690                 .io_type = io_type,
2691                 .io_wbc = wbc,
2692                 .bio = bio,
2693                 .last_block = last_block,
2694         };
2695
2696         trace_f2fs_writepage(page, DATA);
2697
2698         /* we should bypass data pages to proceed the kworkder jobs */
2699         if (unlikely(f2fs_cp_error(sbi))) {
2700                 mapping_set_error(page->mapping, -EIO);
2701                 /*
2702                  * don't drop any dirty dentry pages for keeping lastest
2703                  * directory structure.
2704                  */
2705                 if (S_ISDIR(inode->i_mode))
2706                         goto redirty_out;
2707                 goto out;
2708         }
2709
2710         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2711                 goto redirty_out;
2712
2713         if (page->index < end_index ||
2714                         f2fs_verity_in_progress(inode) ||
2715                         compr_blocks)
2716                 goto write;
2717
2718         /*
2719          * If the offset is out-of-range of file size,
2720          * this page does not have to be written to disk.
2721          */
2722         offset = i_size & (PAGE_SIZE - 1);
2723         if ((page->index >= end_index + 1) || !offset)
2724                 goto out;
2725
2726         zero_user_segment(page, offset, PAGE_SIZE);
2727 write:
2728         if (f2fs_is_drop_cache(inode))
2729                 goto out;
2730         /* we should not write 0'th page having journal header */
2731         if (f2fs_is_volatile_file(inode) && (!page->index ||
2732                         (!wbc->for_reclaim &&
2733                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2734                 goto redirty_out;
2735
2736         /* Dentry/quota blocks are controlled by checkpoint */
2737         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2738                 /*
2739                  * We need to wait for node_write to avoid block allocation during
2740                  * checkpoint. This can only happen to quota writes which can cause
2741                  * the below discard race condition.
2742                  */
2743                 if (IS_NOQUOTA(inode))
2744                         f2fs_down_read(&sbi->node_write);
2745
2746                 fio.need_lock = LOCK_DONE;
2747                 err = f2fs_do_write_data_page(&fio);
2748
2749                 if (IS_NOQUOTA(inode))
2750                         f2fs_up_read(&sbi->node_write);
2751
2752                 goto done;
2753         }
2754
2755         if (!wbc->for_reclaim)
2756                 need_balance_fs = true;
2757         else if (has_not_enough_free_secs(sbi, 0, 0))
2758                 goto redirty_out;
2759         else
2760                 set_inode_flag(inode, FI_HOT_DATA);
2761
2762         err = -EAGAIN;
2763         if (f2fs_has_inline_data(inode)) {
2764                 err = f2fs_write_inline_data(inode, page);
2765                 if (!err)
2766                         goto out;
2767         }
2768
2769         if (err == -EAGAIN) {
2770                 err = f2fs_do_write_data_page(&fio);
2771                 if (err == -EAGAIN) {
2772                         fio.need_lock = LOCK_REQ;
2773                         err = f2fs_do_write_data_page(&fio);
2774                 }
2775         }
2776
2777         if (err) {
2778                 file_set_keep_isize(inode);
2779         } else {
2780                 spin_lock(&F2FS_I(inode)->i_size_lock);
2781                 if (F2FS_I(inode)->last_disk_size < psize)
2782                         F2FS_I(inode)->last_disk_size = psize;
2783                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2784         }
2785
2786 done:
2787         if (err && err != -ENOENT)
2788                 goto redirty_out;
2789
2790 out:
2791         inode_dec_dirty_pages(inode);
2792         if (err) {
2793                 ClearPageUptodate(page);
2794                 clear_page_private_gcing(page);
2795         }
2796
2797         if (wbc->for_reclaim) {
2798                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2799                 clear_inode_flag(inode, FI_HOT_DATA);
2800                 f2fs_remove_dirty_inode(inode);
2801                 submitted = NULL;
2802         }
2803         unlock_page(page);
2804         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2805                         !F2FS_I(inode)->cp_task && allow_balance)
2806                 f2fs_balance_fs(sbi, need_balance_fs);
2807
2808         if (unlikely(f2fs_cp_error(sbi))) {
2809                 f2fs_submit_merged_write(sbi, DATA);
2810                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2811                 submitted = NULL;
2812         }
2813
2814         if (submitted)
2815                 *submitted = fio.submitted ? 1 : 0;
2816
2817         return 0;
2818
2819 redirty_out:
2820         redirty_page_for_writepage(wbc, page);
2821         /*
2822          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2823          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2824          * file_write_and_wait_range() will see EIO error, which is critical
2825          * to return value of fsync() followed by atomic_write failure to user.
2826          */
2827         if (!err || wbc->for_reclaim)
2828                 return AOP_WRITEPAGE_ACTIVATE;
2829         unlock_page(page);
2830         return err;
2831 }
2832
2833 static int f2fs_write_data_page(struct page *page,
2834                                         struct writeback_control *wbc)
2835 {
2836 #ifdef CONFIG_F2FS_FS_COMPRESSION
2837         struct inode *inode = page->mapping->host;
2838
2839         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2840                 goto out;
2841
2842         if (f2fs_compressed_file(inode)) {
2843                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2844                         redirty_page_for_writepage(wbc, page);
2845                         return AOP_WRITEPAGE_ACTIVATE;
2846                 }
2847         }
2848 out:
2849 #endif
2850
2851         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2852                                                 wbc, FS_DATA_IO, 0, true);
2853 }
2854
2855 /*
2856  * This function was copied from write_cche_pages from mm/page-writeback.c.
2857  * The major change is making write step of cold data page separately from
2858  * warm/hot data page.
2859  */
2860 static int f2fs_write_cache_pages(struct address_space *mapping,
2861                                         struct writeback_control *wbc,
2862                                         enum iostat_type io_type)
2863 {
2864         int ret = 0;
2865         int done = 0, retry = 0;
2866         struct pagevec pvec;
2867         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2868         struct bio *bio = NULL;
2869         sector_t last_block;
2870 #ifdef CONFIG_F2FS_FS_COMPRESSION
2871         struct inode *inode = mapping->host;
2872         struct compress_ctx cc = {
2873                 .inode = inode,
2874                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2875                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2876                 .cluster_idx = NULL_CLUSTER,
2877                 .rpages = NULL,
2878                 .nr_rpages = 0,
2879                 .cpages = NULL,
2880                 .valid_nr_cpages = 0,
2881                 .rbuf = NULL,
2882                 .cbuf = NULL,
2883                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2884                 .private = NULL,
2885         };
2886 #endif
2887         int nr_pages;
2888         pgoff_t index;
2889         pgoff_t end;            /* Inclusive */
2890         pgoff_t done_index;
2891         int range_whole = 0;
2892         xa_mark_t tag;
2893         int nwritten = 0;
2894         int submitted = 0;
2895         int i;
2896
2897         pagevec_init(&pvec);
2898
2899         if (get_dirty_pages(mapping->host) <=
2900                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2901                 set_inode_flag(mapping->host, FI_HOT_DATA);
2902         else
2903                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2904
2905         if (wbc->range_cyclic) {
2906                 index = mapping->writeback_index; /* prev offset */
2907                 end = -1;
2908         } else {
2909                 index = wbc->range_start >> PAGE_SHIFT;
2910                 end = wbc->range_end >> PAGE_SHIFT;
2911                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2912                         range_whole = 1;
2913         }
2914         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2915                 tag = PAGECACHE_TAG_TOWRITE;
2916         else
2917                 tag = PAGECACHE_TAG_DIRTY;
2918 retry:
2919         retry = 0;
2920         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2921                 tag_pages_for_writeback(mapping, index, end);
2922         done_index = index;
2923         while (!done && !retry && (index <= end)) {
2924                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2925                                 tag);
2926                 if (nr_pages == 0)
2927                         break;
2928
2929                 for (i = 0; i < nr_pages; i++) {
2930                         struct page *page = pvec.pages[i];
2931                         bool need_readd;
2932 readd:
2933                         need_readd = false;
2934 #ifdef CONFIG_F2FS_FS_COMPRESSION
2935                         if (f2fs_compressed_file(inode)) {
2936                                 void *fsdata = NULL;
2937                                 struct page *pagep;
2938                                 int ret2;
2939
2940                                 ret = f2fs_init_compress_ctx(&cc);
2941                                 if (ret) {
2942                                         done = 1;
2943                                         break;
2944                                 }
2945
2946                                 if (!f2fs_cluster_can_merge_page(&cc,
2947                                                                 page->index)) {
2948                                         ret = f2fs_write_multi_pages(&cc,
2949                                                 &submitted, wbc, io_type);
2950                                         if (!ret)
2951                                                 need_readd = true;
2952                                         goto result;
2953                                 }
2954
2955                                 if (unlikely(f2fs_cp_error(sbi)))
2956                                         goto lock_page;
2957
2958                                 if (!f2fs_cluster_is_empty(&cc))
2959                                         goto lock_page;
2960
2961                                 ret2 = f2fs_prepare_compress_overwrite(
2962                                                         inode, &pagep,
2963                                                         page->index, &fsdata);
2964                                 if (ret2 < 0) {
2965                                         ret = ret2;
2966                                         done = 1;
2967                                         break;
2968                                 } else if (ret2 &&
2969                                         (!f2fs_compress_write_end(inode,
2970                                                 fsdata, page->index, 1) ||
2971                                          !f2fs_all_cluster_page_loaded(&cc,
2972                                                 &pvec, i, nr_pages))) {
2973                                         retry = 1;
2974                                         break;
2975                                 }
2976                         }
2977 #endif
2978                         /* give a priority to WB_SYNC threads */
2979                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2980                                         wbc->sync_mode == WB_SYNC_NONE) {
2981                                 done = 1;
2982                                 break;
2983                         }
2984 #ifdef CONFIG_F2FS_FS_COMPRESSION
2985 lock_page:
2986 #endif
2987                         done_index = page->index;
2988 retry_write:
2989                         lock_page(page);
2990
2991                         if (unlikely(page->mapping != mapping)) {
2992 continue_unlock:
2993                                 unlock_page(page);
2994                                 continue;
2995                         }
2996
2997                         if (!PageDirty(page)) {
2998                                 /* someone wrote it for us */
2999                                 goto continue_unlock;
3000                         }
3001
3002                         if (PageWriteback(page)) {
3003                                 if (wbc->sync_mode != WB_SYNC_NONE)
3004                                         f2fs_wait_on_page_writeback(page,
3005                                                         DATA, true, true);
3006                                 else
3007                                         goto continue_unlock;
3008                         }
3009
3010                         if (!clear_page_dirty_for_io(page))
3011                                 goto continue_unlock;
3012
3013 #ifdef CONFIG_F2FS_FS_COMPRESSION
3014                         if (f2fs_compressed_file(inode)) {
3015                                 get_page(page);
3016                                 f2fs_compress_ctx_add_page(&cc, page);
3017                                 continue;
3018                         }
3019 #endif
3020                         ret = f2fs_write_single_data_page(page, &submitted,
3021                                         &bio, &last_block, wbc, io_type,
3022                                         0, true);
3023                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3024                                 unlock_page(page);
3025 #ifdef CONFIG_F2FS_FS_COMPRESSION
3026 result:
3027 #endif
3028                         nwritten += submitted;
3029                         wbc->nr_to_write -= submitted;
3030
3031                         if (unlikely(ret)) {
3032                                 /*
3033                                  * keep nr_to_write, since vfs uses this to
3034                                  * get # of written pages.
3035                                  */
3036                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3037                                         ret = 0;
3038                                         goto next;
3039                                 } else if (ret == -EAGAIN) {
3040                                         ret = 0;
3041                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3042                                                 f2fs_io_schedule_timeout(
3043                                                         DEFAULT_IO_TIMEOUT);
3044                                                 goto retry_write;
3045                                         }
3046                                         goto next;
3047                                 }
3048                                 done_index = page->index + 1;
3049                                 done = 1;
3050                                 break;
3051                         }
3052
3053                         if (wbc->nr_to_write <= 0 &&
3054                                         wbc->sync_mode == WB_SYNC_NONE) {
3055                                 done = 1;
3056                                 break;
3057                         }
3058 next:
3059                         if (need_readd)
3060                                 goto readd;
3061                 }
3062                 pagevec_release(&pvec);
3063                 cond_resched();
3064         }
3065 #ifdef CONFIG_F2FS_FS_COMPRESSION
3066         /* flush remained pages in compress cluster */
3067         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3068                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3069                 nwritten += submitted;
3070                 wbc->nr_to_write -= submitted;
3071                 if (ret) {
3072                         done = 1;
3073                         retry = 0;
3074                 }
3075         }
3076         if (f2fs_compressed_file(inode))
3077                 f2fs_destroy_compress_ctx(&cc, false);
3078 #endif
3079         if (retry) {
3080                 index = 0;
3081                 end = -1;
3082                 goto retry;
3083         }
3084         if (wbc->range_cyclic && !done)
3085                 done_index = 0;
3086         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3087                 mapping->writeback_index = done_index;
3088
3089         if (nwritten)
3090                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3091                                                                 NULL, 0, DATA);
3092         /* submit cached bio of IPU write */
3093         if (bio)
3094                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3095
3096         return ret;
3097 }
3098
3099 static inline bool __should_serialize_io(struct inode *inode,
3100                                         struct writeback_control *wbc)
3101 {
3102         /* to avoid deadlock in path of data flush */
3103         if (F2FS_I(inode)->cp_task)
3104                 return false;
3105
3106         if (!S_ISREG(inode->i_mode))
3107                 return false;
3108         if (IS_NOQUOTA(inode))
3109                 return false;
3110
3111         if (f2fs_need_compress_data(inode))
3112                 return true;
3113         if (wbc->sync_mode != WB_SYNC_ALL)
3114                 return true;
3115         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3116                 return true;
3117         return false;
3118 }
3119
3120 static int __f2fs_write_data_pages(struct address_space *mapping,
3121                                                 struct writeback_control *wbc,
3122                                                 enum iostat_type io_type)
3123 {
3124         struct inode *inode = mapping->host;
3125         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3126         struct blk_plug plug;
3127         int ret;
3128         bool locked = false;
3129
3130         /* deal with chardevs and other special file */
3131         if (!mapping->a_ops->writepage)
3132                 return 0;
3133
3134         /* skip writing if there is no dirty page in this inode */
3135         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3136                 return 0;
3137
3138         /* during POR, we don't need to trigger writepage at all. */
3139         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3140                 goto skip_write;
3141
3142         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3143                         wbc->sync_mode == WB_SYNC_NONE &&
3144                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3145                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3146                 goto skip_write;
3147
3148         /* skip writing in file defragment preparing stage */
3149         if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3150                 goto skip_write;
3151
3152         trace_f2fs_writepages(mapping->host, wbc, DATA);
3153
3154         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3155         if (wbc->sync_mode == WB_SYNC_ALL)
3156                 atomic_inc(&sbi->wb_sync_req[DATA]);
3157         else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3158                 /* to avoid potential deadlock */
3159                 if (current->plug)
3160                         blk_finish_plug(current->plug);
3161                 goto skip_write;
3162         }
3163
3164         if (__should_serialize_io(inode, wbc)) {
3165                 mutex_lock(&sbi->writepages);
3166                 locked = true;
3167         }
3168
3169         blk_start_plug(&plug);
3170         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3171         blk_finish_plug(&plug);
3172
3173         if (locked)
3174                 mutex_unlock(&sbi->writepages);
3175
3176         if (wbc->sync_mode == WB_SYNC_ALL)
3177                 atomic_dec(&sbi->wb_sync_req[DATA]);
3178         /*
3179          * if some pages were truncated, we cannot guarantee its mapping->host
3180          * to detect pending bios.
3181          */
3182
3183         f2fs_remove_dirty_inode(inode);
3184         return ret;
3185
3186 skip_write:
3187         wbc->pages_skipped += get_dirty_pages(inode);
3188         trace_f2fs_writepages(mapping->host, wbc, DATA);
3189         return 0;
3190 }
3191
3192 static int f2fs_write_data_pages(struct address_space *mapping,
3193                             struct writeback_control *wbc)
3194 {
3195         struct inode *inode = mapping->host;
3196
3197         return __f2fs_write_data_pages(mapping, wbc,
3198                         F2FS_I(inode)->cp_task == current ?
3199                         FS_CP_DATA_IO : FS_DATA_IO);
3200 }
3201
3202 void f2fs_write_failed(struct inode *inode, loff_t to)
3203 {
3204         loff_t i_size = i_size_read(inode);
3205
3206         if (IS_NOQUOTA(inode))
3207                 return;
3208
3209         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3210         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3211                 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3212                 filemap_invalidate_lock(inode->i_mapping);
3213
3214                 truncate_pagecache(inode, i_size);
3215                 f2fs_truncate_blocks(inode, i_size, true);
3216
3217                 filemap_invalidate_unlock(inode->i_mapping);
3218                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3219         }
3220 }
3221
3222 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3223                         struct page *page, loff_t pos, unsigned len,
3224                         block_t *blk_addr, bool *node_changed)
3225 {
3226         struct inode *inode = page->mapping->host;
3227         pgoff_t index = page->index;
3228         struct dnode_of_data dn;
3229         struct page *ipage;
3230         bool locked = false;
3231         struct extent_info ei = {0, };
3232         int err = 0;
3233         int flag;
3234
3235         /*
3236          * If a whole page is being written and we already preallocated all the
3237          * blocks, then there is no need to get a block address now.
3238          */
3239         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3240                 return 0;
3241
3242         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3243         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3244                 flag = F2FS_GET_BLOCK_DEFAULT;
3245         else
3246                 flag = F2FS_GET_BLOCK_PRE_AIO;
3247
3248         if (f2fs_has_inline_data(inode) ||
3249                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3250                 f2fs_do_map_lock(sbi, flag, true);
3251                 locked = true;
3252         }
3253
3254 restart:
3255         /* check inline_data */
3256         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3257         if (IS_ERR(ipage)) {
3258                 err = PTR_ERR(ipage);
3259                 goto unlock_out;
3260         }
3261
3262         set_new_dnode(&dn, inode, ipage, ipage, 0);
3263
3264         if (f2fs_has_inline_data(inode)) {
3265                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3266                         f2fs_do_read_inline_data(page, ipage);
3267                         set_inode_flag(inode, FI_DATA_EXIST);
3268                         if (inode->i_nlink)
3269                                 set_page_private_inline(ipage);
3270                 } else {
3271                         err = f2fs_convert_inline_page(&dn, page);
3272                         if (err)
3273                                 goto out;
3274                         if (dn.data_blkaddr == NULL_ADDR)
3275                                 err = f2fs_get_block(&dn, index);
3276                 }
3277         } else if (locked) {
3278                 err = f2fs_get_block(&dn, index);
3279         } else {
3280                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3281                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3282                 } else {
3283                         /* hole case */
3284                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3285                         if (err || dn.data_blkaddr == NULL_ADDR) {
3286                                 f2fs_put_dnode(&dn);
3287                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3288                                                                 true);
3289                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3290                                 locked = true;
3291                                 goto restart;
3292                         }
3293                 }
3294         }
3295
3296         /* convert_inline_page can make node_changed */
3297         *blk_addr = dn.data_blkaddr;
3298         *node_changed = dn.node_changed;
3299 out:
3300         f2fs_put_dnode(&dn);
3301 unlock_out:
3302         if (locked)
3303                 f2fs_do_map_lock(sbi, flag, false);
3304         return err;
3305 }
3306
3307 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3308                 loff_t pos, unsigned len, unsigned flags,
3309                 struct page **pagep, void **fsdata)
3310 {
3311         struct inode *inode = mapping->host;
3312         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3313         struct page *page = NULL;
3314         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3315         bool need_balance = false, drop_atomic = false;
3316         block_t blkaddr = NULL_ADDR;
3317         int err = 0;
3318
3319         trace_f2fs_write_begin(inode, pos, len, flags);
3320
3321         if (!f2fs_is_checkpoint_ready(sbi)) {
3322                 err = -ENOSPC;
3323                 goto fail;
3324         }
3325
3326         if ((f2fs_is_atomic_file(inode) &&
3327                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3328                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3329                 err = -ENOMEM;
3330                 drop_atomic = true;
3331                 goto fail;
3332         }
3333
3334         /*
3335          * We should check this at this moment to avoid deadlock on inode page
3336          * and #0 page. The locking rule for inline_data conversion should be:
3337          * lock_page(page #0) -> lock_page(inode_page)
3338          */
3339         if (index != 0) {
3340                 err = f2fs_convert_inline_inode(inode);
3341                 if (err)
3342                         goto fail;
3343         }
3344
3345 #ifdef CONFIG_F2FS_FS_COMPRESSION
3346         if (f2fs_compressed_file(inode)) {
3347                 int ret;
3348
3349                 *fsdata = NULL;
3350
3351                 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3352                         goto repeat;
3353
3354                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3355                                                         index, fsdata);
3356                 if (ret < 0) {
3357                         err = ret;
3358                         goto fail;
3359                 } else if (ret) {
3360                         return 0;
3361                 }
3362         }
3363 #endif
3364
3365 repeat:
3366         /*
3367          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3368          * wait_for_stable_page. Will wait that below with our IO control.
3369          */
3370         page = f2fs_pagecache_get_page(mapping, index,
3371                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3372         if (!page) {
3373                 err = -ENOMEM;
3374                 goto fail;
3375         }
3376
3377         /* TODO: cluster can be compressed due to race with .writepage */
3378
3379         *pagep = page;
3380
3381         err = prepare_write_begin(sbi, page, pos, len,
3382                                         &blkaddr, &need_balance);
3383         if (err)
3384                 goto fail;
3385
3386         if (need_balance && !IS_NOQUOTA(inode) &&
3387                         has_not_enough_free_secs(sbi, 0, 0)) {
3388                 unlock_page(page);
3389                 f2fs_balance_fs(sbi, true);
3390                 lock_page(page);
3391                 if (page->mapping != mapping) {
3392                         /* The page got truncated from under us */
3393                         f2fs_put_page(page, 1);
3394                         goto repeat;
3395                 }
3396         }
3397
3398         f2fs_wait_on_page_writeback(page, DATA, false, true);
3399
3400         if (len == PAGE_SIZE || PageUptodate(page))
3401                 return 0;
3402
3403         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3404             !f2fs_verity_in_progress(inode)) {
3405                 zero_user_segment(page, len, PAGE_SIZE);
3406                 return 0;
3407         }
3408
3409         if (blkaddr == NEW_ADDR) {
3410                 zero_user_segment(page, 0, PAGE_SIZE);
3411                 SetPageUptodate(page);
3412         } else {
3413                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3414                                 DATA_GENERIC_ENHANCE_READ)) {
3415                         err = -EFSCORRUPTED;
3416                         goto fail;
3417                 }
3418                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3419                 if (err)
3420                         goto fail;
3421
3422                 lock_page(page);
3423                 if (unlikely(page->mapping != mapping)) {
3424                         f2fs_put_page(page, 1);
3425                         goto repeat;
3426                 }
3427                 if (unlikely(!PageUptodate(page))) {
3428                         err = -EIO;
3429                         goto fail;
3430                 }
3431         }
3432         return 0;
3433
3434 fail:
3435         f2fs_put_page(page, 1);
3436         f2fs_write_failed(inode, pos + len);
3437         if (drop_atomic)
3438                 f2fs_drop_inmem_pages_all(sbi, false);
3439         return err;
3440 }
3441
3442 static int f2fs_write_end(struct file *file,
3443                         struct address_space *mapping,
3444                         loff_t pos, unsigned len, unsigned copied,
3445                         struct page *page, void *fsdata)
3446 {
3447         struct inode *inode = page->mapping->host;
3448
3449         trace_f2fs_write_end(inode, pos, len, copied);
3450
3451         /*
3452          * This should be come from len == PAGE_SIZE, and we expect copied
3453          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3454          * let generic_perform_write() try to copy data again through copied=0.
3455          */
3456         if (!PageUptodate(page)) {
3457                 if (unlikely(copied != len))
3458                         copied = 0;
3459                 else
3460                         SetPageUptodate(page);
3461         }
3462
3463 #ifdef CONFIG_F2FS_FS_COMPRESSION
3464         /* overwrite compressed file */
3465         if (f2fs_compressed_file(inode) && fsdata) {
3466                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3467                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3468
3469                 if (pos + copied > i_size_read(inode) &&
3470                                 !f2fs_verity_in_progress(inode))
3471                         f2fs_i_size_write(inode, pos + copied);
3472                 return copied;
3473         }
3474 #endif
3475
3476         if (!copied)
3477                 goto unlock_out;
3478
3479         set_page_dirty(page);
3480
3481         if (pos + copied > i_size_read(inode) &&
3482             !f2fs_verity_in_progress(inode))
3483                 f2fs_i_size_write(inode, pos + copied);
3484 unlock_out:
3485         f2fs_put_page(page, 1);
3486         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3487         return copied;
3488 }
3489
3490 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3491 {
3492         struct inode *inode = folio->mapping->host;
3493         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3494
3495         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3496                                 (offset || length != folio_size(folio)))
3497                 return;
3498
3499         if (folio_test_dirty(folio)) {
3500                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3501                         dec_page_count(sbi, F2FS_DIRTY_META);
3502                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3503                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3504                 } else {
3505                         inode_dec_dirty_pages(inode);
3506                         f2fs_remove_dirty_inode(inode);
3507                 }
3508         }
3509
3510         clear_page_private_gcing(&folio->page);
3511
3512         if (test_opt(sbi, COMPRESS_CACHE) &&
3513                         inode->i_ino == F2FS_COMPRESS_INO(sbi))
3514                 clear_page_private_data(&folio->page);
3515
3516         if (page_private_atomic(&folio->page))
3517                 return f2fs_drop_inmem_page(inode, &folio->page);
3518
3519         folio_detach_private(folio);
3520 }
3521
3522 int f2fs_release_page(struct page *page, gfp_t wait)
3523 {
3524         /* If this is dirty page, keep PagePrivate */
3525         if (PageDirty(page))
3526                 return 0;
3527
3528         /* This is atomic written page, keep Private */
3529         if (page_private_atomic(page))
3530                 return 0;
3531
3532         if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3533                 struct inode *inode = page->mapping->host;
3534
3535                 if (inode->i_ino == F2FS_COMPRESS_INO(F2FS_I_SB(inode)))
3536                         clear_page_private_data(page);
3537         }
3538
3539         clear_page_private_gcing(page);
3540
3541         detach_page_private(page);
3542         set_page_private(page, 0);
3543         return 1;
3544 }
3545
3546 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3547                 struct folio *folio)
3548 {
3549         struct inode *inode = mapping->host;
3550
3551         trace_f2fs_set_page_dirty(&folio->page, DATA);
3552
3553         if (!folio_test_uptodate(folio))
3554                 folio_mark_uptodate(folio);
3555         BUG_ON(folio_test_swapcache(folio));
3556
3557         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3558                 if (!page_private_atomic(&folio->page)) {
3559                         f2fs_register_inmem_page(inode, &folio->page);
3560                         return true;
3561                 }
3562                 /*
3563                  * Previously, this page has been registered, we just
3564                  * return here.
3565                  */
3566                 return false;
3567         }
3568
3569         if (!folio_test_dirty(folio)) {
3570                 filemap_dirty_folio(mapping, folio);
3571                 f2fs_update_dirty_folio(inode, folio);
3572                 return true;
3573         }
3574         return true;
3575 }
3576
3577
3578 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3579 {
3580 #ifdef CONFIG_F2FS_FS_COMPRESSION
3581         struct dnode_of_data dn;
3582         sector_t start_idx, blknr = 0;
3583         int ret;
3584
3585         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3586
3587         set_new_dnode(&dn, inode, NULL, NULL, 0);
3588         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3589         if (ret)
3590                 return 0;
3591
3592         if (dn.data_blkaddr != COMPRESS_ADDR) {
3593                 dn.ofs_in_node += block - start_idx;
3594                 blknr = f2fs_data_blkaddr(&dn);
3595                 if (!__is_valid_data_blkaddr(blknr))
3596                         blknr = 0;
3597         }
3598
3599         f2fs_put_dnode(&dn);
3600         return blknr;
3601 #else
3602         return 0;
3603 #endif
3604 }
3605
3606
3607 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3608 {
3609         struct inode *inode = mapping->host;
3610         sector_t blknr = 0;
3611
3612         if (f2fs_has_inline_data(inode))
3613                 goto out;
3614
3615         /* make sure allocating whole blocks */
3616         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3617                 filemap_write_and_wait(mapping);
3618
3619         /* Block number less than F2FS MAX BLOCKS */
3620         if (unlikely(block >= max_file_blocks(inode)))
3621                 goto out;
3622
3623         if (f2fs_compressed_file(inode)) {
3624                 blknr = f2fs_bmap_compress(inode, block);
3625         } else {
3626                 struct f2fs_map_blocks map;
3627
3628                 memset(&map, 0, sizeof(map));
3629                 map.m_lblk = block;
3630                 map.m_len = 1;
3631                 map.m_next_pgofs = NULL;
3632                 map.m_seg_type = NO_CHECK_TYPE;
3633
3634                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3635                         blknr = map.m_pblk;
3636         }
3637 out:
3638         trace_f2fs_bmap(inode, block, blknr);
3639         return blknr;
3640 }
3641
3642 #ifdef CONFIG_MIGRATION
3643 #include <linux/migrate.h>
3644
3645 int f2fs_migrate_page(struct address_space *mapping,
3646                 struct page *newpage, struct page *page, enum migrate_mode mode)
3647 {
3648         int rc, extra_count;
3649         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3650         bool atomic_written = page_private_atomic(page);
3651
3652         BUG_ON(PageWriteback(page));
3653
3654         /* migrating an atomic written page is safe with the inmem_lock hold */
3655         if (atomic_written) {
3656                 if (mode != MIGRATE_SYNC)
3657                         return -EBUSY;
3658                 if (!mutex_trylock(&fi->inmem_lock))
3659                         return -EAGAIN;
3660         }
3661
3662         /* one extra reference was held for atomic_write page */
3663         extra_count = atomic_written ? 1 : 0;
3664         rc = migrate_page_move_mapping(mapping, newpage,
3665                                 page, extra_count);
3666         if (rc != MIGRATEPAGE_SUCCESS) {
3667                 if (atomic_written)
3668                         mutex_unlock(&fi->inmem_lock);
3669                 return rc;
3670         }
3671
3672         if (atomic_written) {
3673                 struct inmem_pages *cur;
3674
3675                 list_for_each_entry(cur, &fi->inmem_pages, list)
3676                         if (cur->page == page) {
3677                                 cur->page = newpage;
3678                                 break;
3679                         }
3680                 mutex_unlock(&fi->inmem_lock);
3681                 put_page(page);
3682                 get_page(newpage);
3683         }
3684
3685         /* guarantee to start from no stale private field */
3686         set_page_private(newpage, 0);
3687         if (PagePrivate(page)) {
3688                 set_page_private(newpage, page_private(page));
3689                 SetPagePrivate(newpage);
3690                 get_page(newpage);
3691
3692                 set_page_private(page, 0);
3693                 ClearPagePrivate(page);
3694                 put_page(page);
3695         }
3696
3697         if (mode != MIGRATE_SYNC_NO_COPY)
3698                 migrate_page_copy(newpage, page);
3699         else
3700                 migrate_page_states(newpage, page);
3701
3702         return MIGRATEPAGE_SUCCESS;
3703 }
3704 #endif
3705
3706 #ifdef CONFIG_SWAP
3707 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3708                                                         unsigned int blkcnt)
3709 {
3710         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3711         unsigned int blkofs;
3712         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3713         unsigned int secidx = start_blk / blk_per_sec;
3714         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3715         int ret = 0;
3716
3717         f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3718         filemap_invalidate_lock(inode->i_mapping);
3719
3720         set_inode_flag(inode, FI_ALIGNED_WRITE);
3721         set_inode_flag(inode, FI_OPU_WRITE);
3722
3723         for (; secidx < end_sec; secidx++) {
3724                 f2fs_down_write(&sbi->pin_sem);
3725
3726                 f2fs_lock_op(sbi);
3727                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3728                 f2fs_unlock_op(sbi);
3729
3730                 set_inode_flag(inode, FI_SKIP_WRITES);
3731
3732                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3733                         struct page *page;
3734                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3735
3736                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3737                         if (IS_ERR(page)) {
3738                                 f2fs_up_write(&sbi->pin_sem);
3739                                 ret = PTR_ERR(page);
3740                                 goto done;
3741                         }
3742
3743                         set_page_dirty(page);
3744                         f2fs_put_page(page, 1);
3745                 }
3746
3747                 clear_inode_flag(inode, FI_SKIP_WRITES);
3748
3749                 ret = filemap_fdatawrite(inode->i_mapping);
3750
3751                 f2fs_up_write(&sbi->pin_sem);
3752
3753                 if (ret)
3754                         break;
3755         }
3756
3757 done:
3758         clear_inode_flag(inode, FI_SKIP_WRITES);
3759         clear_inode_flag(inode, FI_OPU_WRITE);
3760         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3761
3762         filemap_invalidate_unlock(inode->i_mapping);
3763         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3764
3765         return ret;
3766 }
3767
3768 static int check_swap_activate(struct swap_info_struct *sis,
3769                                 struct file *swap_file, sector_t *span)
3770 {
3771         struct address_space *mapping = swap_file->f_mapping;
3772         struct inode *inode = mapping->host;
3773         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3774         sector_t cur_lblock;
3775         sector_t last_lblock;
3776         sector_t pblock;
3777         sector_t lowest_pblock = -1;
3778         sector_t highest_pblock = 0;
3779         int nr_extents = 0;
3780         unsigned long nr_pblocks;
3781         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3782         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3783         unsigned int not_aligned = 0;
3784         int ret = 0;
3785
3786         /*
3787          * Map all the blocks into the extent list.  This code doesn't try
3788          * to be very smart.
3789          */
3790         cur_lblock = 0;
3791         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3792
3793         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3794                 struct f2fs_map_blocks map;
3795 retry:
3796                 cond_resched();
3797
3798                 memset(&map, 0, sizeof(map));
3799                 map.m_lblk = cur_lblock;
3800                 map.m_len = last_lblock - cur_lblock;
3801                 map.m_next_pgofs = NULL;
3802                 map.m_next_extent = NULL;
3803                 map.m_seg_type = NO_CHECK_TYPE;
3804                 map.m_may_create = false;
3805
3806                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3807                 if (ret)
3808                         goto out;
3809
3810                 /* hole */
3811                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3812                         f2fs_err(sbi, "Swapfile has holes");
3813                         ret = -EINVAL;
3814                         goto out;
3815                 }
3816
3817                 pblock = map.m_pblk;
3818                 nr_pblocks = map.m_len;
3819
3820                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3821                                 nr_pblocks & sec_blks_mask) {
3822                         not_aligned++;
3823
3824                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3825                         if (cur_lblock + nr_pblocks > sis->max)
3826                                 nr_pblocks -= blks_per_sec;
3827
3828                         if (!nr_pblocks) {
3829                                 /* this extent is last one */
3830                                 nr_pblocks = map.m_len;
3831                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3832                                 goto next;
3833                         }
3834
3835                         ret = f2fs_migrate_blocks(inode, cur_lblock,
3836                                                         nr_pblocks);
3837                         if (ret)
3838                                 goto out;
3839                         goto retry;
3840                 }
3841 next:
3842                 if (cur_lblock + nr_pblocks >= sis->max)
3843                         nr_pblocks = sis->max - cur_lblock;
3844
3845                 if (cur_lblock) {       /* exclude the header page */
3846                         if (pblock < lowest_pblock)
3847                                 lowest_pblock = pblock;
3848                         if (pblock + nr_pblocks - 1 > highest_pblock)
3849                                 highest_pblock = pblock + nr_pblocks - 1;
3850                 }
3851
3852                 /*
3853                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3854                  */
3855                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3856                 if (ret < 0)
3857                         goto out;
3858                 nr_extents += ret;
3859                 cur_lblock += nr_pblocks;
3860         }
3861         ret = nr_extents;
3862         *span = 1 + highest_pblock - lowest_pblock;
3863         if (cur_lblock == 0)
3864                 cur_lblock = 1; /* force Empty message */
3865         sis->max = cur_lblock;
3866         sis->pages = cur_lblock - 1;
3867         sis->highest_bit = cur_lblock - 1;
3868 out:
3869         if (not_aligned)
3870                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3871                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
3872         return ret;
3873 }
3874
3875 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3876                                 sector_t *span)
3877 {
3878         struct inode *inode = file_inode(file);
3879         int ret;
3880
3881         if (!S_ISREG(inode->i_mode))
3882                 return -EINVAL;
3883
3884         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3885                 return -EROFS;
3886
3887         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3888                 f2fs_err(F2FS_I_SB(inode),
3889                         "Swapfile not supported in LFS mode");
3890                 return -EINVAL;
3891         }
3892
3893         ret = f2fs_convert_inline_inode(inode);
3894         if (ret)
3895                 return ret;
3896
3897         if (!f2fs_disable_compressed_file(inode))
3898                 return -EINVAL;
3899
3900         f2fs_precache_extents(inode);
3901
3902         ret = check_swap_activate(sis, file, span);
3903         if (ret < 0)
3904                 return ret;
3905
3906         set_inode_flag(inode, FI_PIN_FILE);
3907         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3908         return ret;
3909 }
3910
3911 static void f2fs_swap_deactivate(struct file *file)
3912 {
3913         struct inode *inode = file_inode(file);
3914
3915         clear_inode_flag(inode, FI_PIN_FILE);
3916 }
3917 #else
3918 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3919                                 sector_t *span)
3920 {
3921         return -EOPNOTSUPP;
3922 }
3923
3924 static void f2fs_swap_deactivate(struct file *file)
3925 {
3926 }
3927 #endif
3928
3929 const struct address_space_operations f2fs_dblock_aops = {
3930         .readpage       = f2fs_read_data_page,
3931         .readahead      = f2fs_readahead,
3932         .writepage      = f2fs_write_data_page,
3933         .writepages     = f2fs_write_data_pages,
3934         .write_begin    = f2fs_write_begin,
3935         .write_end      = f2fs_write_end,
3936         .dirty_folio    = f2fs_dirty_data_folio,
3937         .invalidate_folio = f2fs_invalidate_folio,
3938         .releasepage    = f2fs_release_page,
3939         .direct_IO      = noop_direct_IO,
3940         .bmap           = f2fs_bmap,
3941         .swap_activate  = f2fs_swap_activate,
3942         .swap_deactivate = f2fs_swap_deactivate,
3943 #ifdef CONFIG_MIGRATION
3944         .migratepage    = f2fs_migrate_page,
3945 #endif
3946 };
3947
3948 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3949 {
3950         struct address_space *mapping = page_mapping(page);
3951         unsigned long flags;
3952
3953         xa_lock_irqsave(&mapping->i_pages, flags);
3954         __xa_clear_mark(&mapping->i_pages, page_index(page),
3955                                                 PAGECACHE_TAG_DIRTY);
3956         xa_unlock_irqrestore(&mapping->i_pages, flags);
3957 }
3958
3959 int __init f2fs_init_post_read_processing(void)
3960 {
3961         bio_post_read_ctx_cache =
3962                 kmem_cache_create("f2fs_bio_post_read_ctx",
3963                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
3964         if (!bio_post_read_ctx_cache)
3965                 goto fail;
3966         bio_post_read_ctx_pool =
3967                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3968                                          bio_post_read_ctx_cache);
3969         if (!bio_post_read_ctx_pool)
3970                 goto fail_free_cache;
3971         return 0;
3972
3973 fail_free_cache:
3974         kmem_cache_destroy(bio_post_read_ctx_cache);
3975 fail:
3976         return -ENOMEM;
3977 }
3978
3979 void f2fs_destroy_post_read_processing(void)
3980 {
3981         mempool_destroy(bio_post_read_ctx_pool);
3982         kmem_cache_destroy(bio_post_read_ctx_cache);
3983 }
3984
3985 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
3986 {
3987         if (!f2fs_sb_has_encrypt(sbi) &&
3988                 !f2fs_sb_has_verity(sbi) &&
3989                 !f2fs_sb_has_compression(sbi))
3990                 return 0;
3991
3992         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
3993                                                  WQ_UNBOUND | WQ_HIGHPRI,
3994                                                  num_online_cpus());
3995         if (!sbi->post_read_wq)
3996                 return -ENOMEM;
3997         return 0;
3998 }
3999
4000 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4001 {
4002         if (sbi->post_read_wq)
4003                 destroy_workqueue(sbi->post_read_wq);
4004 }
4005
4006 int __init f2fs_init_bio_entry_cache(void)
4007 {
4008         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4009                         sizeof(struct bio_entry));
4010         if (!bio_entry_slab)
4011                 return -ENOMEM;
4012         return 0;
4013 }
4014
4015 void f2fs_destroy_bio_entry_cache(void)
4016 {
4017         kmem_cache_destroy(bio_entry_slab);
4018 }
4019
4020 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4021                             unsigned int flags, struct iomap *iomap,
4022                             struct iomap *srcmap)
4023 {
4024         struct f2fs_map_blocks map = {};
4025         pgoff_t next_pgofs = 0;
4026         int err;
4027
4028         map.m_lblk = bytes_to_blks(inode, offset);
4029         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4030         map.m_next_pgofs = &next_pgofs;
4031         map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4032         if (flags & IOMAP_WRITE)
4033                 map.m_may_create = true;
4034
4035         err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4036                               F2FS_GET_BLOCK_DIO);
4037         if (err)
4038                 return err;
4039
4040         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4041
4042         /*
4043          * When inline encryption is enabled, sometimes I/O to an encrypted file
4044          * has to be broken up to guarantee DUN contiguity.  Handle this by
4045          * limiting the length of the mapping returned.
4046          */
4047         map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4048
4049         if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4050                 iomap->length = blks_to_bytes(inode, map.m_len);
4051                 if (map.m_flags & F2FS_MAP_MAPPED) {
4052                         iomap->type = IOMAP_MAPPED;
4053                         iomap->flags |= IOMAP_F_MERGED;
4054                 } else {
4055                         iomap->type = IOMAP_UNWRITTEN;
4056                 }
4057                 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4058                         return -EINVAL;
4059
4060                 iomap->bdev = map.m_bdev;
4061                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4062         } else {
4063                 iomap->length = blks_to_bytes(inode, next_pgofs) -
4064                                 iomap->offset;
4065                 iomap->type = IOMAP_HOLE;
4066                 iomap->addr = IOMAP_NULL_ADDR;
4067         }
4068
4069         if (map.m_flags & F2FS_MAP_NEW)
4070                 iomap->flags |= IOMAP_F_NEW;
4071         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4072             offset + length > i_size_read(inode))
4073                 iomap->flags |= IOMAP_F_DIRTY;
4074
4075         return 0;
4076 }
4077
4078 const struct iomap_ops f2fs_iomap_ops = {
4079         .iomap_begin    = f2fs_iomap_begin,
4080 };