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, struct bio *bio)
 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         if (bio) {
 373                 bio_set_dev(bio, bdev);
 374                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 375         }
 376         return bdev;
 377 }
 378
 379 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 380 {
 381         int i;
 382
 383         if (!f2fs_is_multi_device(sbi))
 384                 return 0;
 385
 386         for (i = 0; i < sbi->s_ndevs; i++)
 387                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 388                         return i;
 389         return 0;
 390 }
 391
 392 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 393 {
 394         struct f2fs_sb_info *sbi = fio->sbi;
 395         struct bio *bio;
 396
 397         bio = bio_alloc_bioset(GFP_NOIO, npages, &f2fs_bioset);
 398
 399         f2fs_target_device(sbi, fio->new_blkaddr, bio);
 400         if (is_read_io(fio->op)) {
 401                 bio->bi_end_io = f2fs_read_end_io;
 402                 bio->bi_private = NULL;
 403         } else {
 404                 bio->bi_end_io = f2fs_write_end_io;
 405                 bio->bi_private = sbi;
 406                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 407                                                 fio->type, fio->temp);
 408         }
 409         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 410
 411         if (fio->io_wbc)
 412                 wbc_init_bio(fio->io_wbc, bio);
 413
 414         return bio;
 415 }
 416
 417 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 418                                   pgoff_t first_idx,
 419                                   const struct f2fs_io_info *fio,
 420                                   gfp_t gfp_mask)
 421 {
 422         /*
 423          * The f2fs garbage collector sets ->encrypted_page when it wants to
 424          * read/write raw data without encryption.
 425          */
 426         if (!fio || !fio->encrypted_page)
 427                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 428 }
 429
 430 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 431                                      pgoff_t next_idx,
 432                                      const struct f2fs_io_info *fio)
 433 {
 434         /*
 435          * The f2fs garbage collector sets ->encrypted_page when it wants to
 436          * read/write raw data without encryption.
 437          */
 438         if (fio && fio->encrypted_page)
 439                 return !bio_has_crypt_ctx(bio);
 440
 441         return fscrypt_mergeable_bio(bio, inode, next_idx);
 442 }
 443
 444 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 445                                 struct bio *bio, enum page_type type)
 446 {
 447         if (!is_read_io(bio_op(bio))) {
 448                 unsigned int start;
 449
 450                 if (type != DATA && type != NODE)
 451                         goto submit_io;
 452
 453                 if (f2fs_lfs_mode(sbi) && current->plug)
 454                         blk_finish_plug(current->plug);
 455
 456                 if (!F2FS_IO_ALIGNED(sbi))
 457                         goto submit_io;
 458
 459                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 460                 start %= F2FS_IO_SIZE(sbi);
 461
 462                 if (start == 0)
 463                         goto submit_io;
 464
 465                 /* fill dummy pages */
 466                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 467                         struct page *page =
 468                                 mempool_alloc(sbi->write_io_dummy,
 469                                               GFP_NOIO | __GFP_NOFAIL);
 470                         f2fs_bug_on(sbi, !page);
 471
 472                         lock_page(page);
 473
 474                         zero_user_segment(page, 0, PAGE_SIZE);
 475                         set_page_private_dummy(page);
 476
 477                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 478                                 f2fs_bug_on(sbi, 1);
 479                 }
 480                 /*
 481                  * In the NODE case, we lose next block address chain. So, we
 482                  * need to do checkpoint in f2fs_sync_file.
 483                  */
 484                 if (type == NODE)
 485                         set_sbi_flag(sbi, SBI_NEED_CP);
 486         }
 487 submit_io:
 488         if (is_read_io(bio_op(bio)))
 489                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 490         else
 491                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 492
 493         iostat_update_submit_ctx(bio, type);
 494         submit_bio(bio);
 495 }
 496
 497 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 498                                 struct bio *bio, enum page_type type)
 499 {
 500         __submit_bio(sbi, bio, type);
 501 }
 502
 503 static void __attach_io_flag(struct f2fs_io_info *fio)
 504 {
 505         struct f2fs_sb_info *sbi = fio->sbi;
 506         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 507         unsigned int io_flag, fua_flag, meta_flag;
 508
 509         if (fio->type == DATA)
 510                 io_flag = sbi->data_io_flag;
 511         else if (fio->type == NODE)
 512                 io_flag = sbi->node_io_flag;
 513         else
 514                 return;
 515
 516         fua_flag = io_flag & temp_mask;
 517         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 518
 519         /*
 520          * data/node io flag bits per temp:
 521          *      REQ_META     |      REQ_FUA      |
 522          *    5 |    4 |   3 |    2 |    1 |   0 |
 523          * Cold | Warm | Hot | Cold | Warm | Hot |
 524          */
 525         if ((1 << fio->temp) & meta_flag)
 526                 fio->op_flags |= REQ_META;
 527         if ((1 << fio->temp) & fua_flag)
 528                 fio->op_flags |= REQ_FUA;
 529 }
 530
 531 static void __submit_merged_bio(struct f2fs_bio_info *io)
 532 {
 533         struct f2fs_io_info *fio = &io->fio;
 534
 535         if (!io->bio)
 536                 return;
 537
 538         __attach_io_flag(fio);
 539         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 540
 541         if (is_read_io(fio->op))
 542                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 543         else
 544                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 545
 546         __submit_bio(io->sbi, io->bio, fio->type);
 547         io->bio = NULL;
 548 }
 549
 550 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 551                                                 struct page *page, nid_t ino)
 552 {
 553         struct bio_vec *bvec;
 554         struct bvec_iter_all iter_all;
 555
 556         if (!bio)
 557                 return false;
 558
 559         if (!inode && !page && !ino)
 560                 return true;
 561
 562         bio_for_each_segment_all(bvec, bio, iter_all) {
 563                 struct page *target = bvec->bv_page;
 564
 565                 if (fscrypt_is_bounce_page(target)) {
 566                         target = fscrypt_pagecache_page(target);
 567                         if (IS_ERR(target))
 568                                 continue;
 569                 }
 570                 if (f2fs_is_compressed_page(target)) {
 571                         target = f2fs_compress_control_page(target);
 572                         if (IS_ERR(target))
 573                                 continue;
 574                 }
 575
 576                 if (inode && inode == target->mapping->host)
 577                         return true;
 578                 if (page && page == target)
 579                         return true;
 580                 if (ino && ino == ino_of_node(target))
 581                         return true;
 582         }
 583
 584         return false;
 585 }
 586
 587 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 588                                 enum page_type type, enum temp_type temp)
 589 {
 590         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 591         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 592
 593         down_write(&io->io_rwsem);
 594
 595         /* change META to META_FLUSH in the checkpoint procedure */
 596         if (type >= META_FLUSH) {
 597                 io->fio.type = META_FLUSH;
 598                 io->fio.op = REQ_OP_WRITE;
 599                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 600                 if (!test_opt(sbi, NOBARRIER))
 601                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 602         }
 603         __submit_merged_bio(io);
 604         up_write(&io->io_rwsem);
 605 }
 606
 607 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 608                                 struct inode *inode, struct page *page,
 609                                 nid_t ino, enum page_type type, bool force)
 610 {
 611         enum temp_type temp;
 612         bool ret = true;
 613
 614         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 615                 if (!force)     {
 616                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 617                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 618
 619                         down_read(&io->io_rwsem);
 620                         ret = __has_merged_page(io->bio, inode, page, ino);
 621                         up_read(&io->io_rwsem);
 622                 }
 623                 if (ret)
 624                         __f2fs_submit_merged_write(sbi, type, temp);
 625
 626                 /* TODO: use HOT temp only for meta pages now. */
 627                 if (type >= META)
 628                         break;
 629         }
 630 }
 631
 632 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 633 {
 634         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 635 }
 636
 637 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 638                                 struct inode *inode, struct page *page,
 639                                 nid_t ino, enum page_type type)
 640 {
 641         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 642 }
 643
 644 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 645 {
 646         f2fs_submit_merged_write(sbi, DATA);
 647         f2fs_submit_merged_write(sbi, NODE);
 648         f2fs_submit_merged_write(sbi, META);
 649 }
 650
 651 /*
 652  * Fill the locked page with data located in the block address.
 653  * A caller needs to unlock the page on failure.
 654  */
 655 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 656 {
 657         struct bio *bio;
 658         struct page *page = fio->encrypted_page ?
 659                         fio->encrypted_page : fio->page;
 660
 661         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 662                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 663                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 664                 return -EFSCORRUPTED;
 665
 666         trace_f2fs_submit_page_bio(page, fio);
 667
 668         /* Allocate a new bio */
 669         bio = __bio_alloc(fio, 1);
 670
 671         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 672                                fio->page->index, fio, GFP_NOIO);
 673
 674         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 675                 bio_put(bio);
 676                 return -EFAULT;
 677         }
 678
 679         if (fio->io_wbc && !is_read_io(fio->op))
 680                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 681
 682         __attach_io_flag(fio);
 683         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 684
 685         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 686                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 687
 688         __submit_bio(fio->sbi, bio, fio->type);
 689         return 0;
 690 }
 691
 692 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 693                                 block_t last_blkaddr, block_t cur_blkaddr)
 694 {
 695         if (unlikely(sbi->max_io_bytes &&
 696                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 697                 return false;
 698         if (last_blkaddr + 1 != cur_blkaddr)
 699                 return false;
 700         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 701 }
 702
 703 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 704                                                 struct f2fs_io_info *fio)
 705 {
 706         if (io->fio.op != fio->op)
 707                 return false;
 708         return io->fio.op_flags == fio->op_flags;
 709 }
 710
 711 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 712                                         struct f2fs_bio_info *io,
 713                                         struct f2fs_io_info *fio,
 714                                         block_t last_blkaddr,
 715                                         block_t cur_blkaddr)
 716 {
 717         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 718                 unsigned int filled_blocks =
 719                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 720                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 721                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 722
 723                 /* IOs in bio is aligned and left space of vectors is not enough */
 724                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 725                         return false;
 726         }
 727         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 728                 return false;
 729         return io_type_is_mergeable(io, fio);
 730 }
 731
 732 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 733                                 struct page *page, enum temp_type temp)
 734 {
 735         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 736         struct bio_entry *be;
 737
 738         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 739         be->bio = bio;
 740         bio_get(bio);
 741
 742         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 743                 f2fs_bug_on(sbi, 1);
 744
 745         down_write(&io->bio_list_lock);
 746         list_add_tail(&be->list, &io->bio_list);
 747         up_write(&io->bio_list_lock);
 748 }
 749
 750 static void del_bio_entry(struct bio_entry *be)
 751 {
 752         list_del(&be->list);
 753         kmem_cache_free(bio_entry_slab, be);
 754 }
 755
 756 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 757                                                         struct page *page)
 758 {
 759         struct f2fs_sb_info *sbi = fio->sbi;
 760         enum temp_type temp;
 761         bool found = false;
 762         int ret = -EAGAIN;
 763
 764         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 765                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 766                 struct list_head *head = &io->bio_list;
 767                 struct bio_entry *be;
 768
 769                 down_write(&io->bio_list_lock);
 770                 list_for_each_entry(be, head, list) {
 771                         if (be->bio != *bio)
 772                                 continue;
 773
 774                         found = true;
 775
 776                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 777                                                             *fio->last_block,
 778                                                             fio->new_blkaddr));
 779                         if (f2fs_crypt_mergeable_bio(*bio,
 780                                         fio->page->mapping->host,
 781                                         fio->page->index, fio) &&
 782                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 783                                         PAGE_SIZE) {
 784                                 ret = 0;
 785                                 break;
 786                         }
 787
 788                         /* page can't be merged into bio; submit the bio */
 789                         del_bio_entry(be);
 790                         __submit_bio(sbi, *bio, DATA);
 791                         break;
 792                 }
 793                 up_write(&io->bio_list_lock);
 794         }
 795
 796         if (ret) {
 797                 bio_put(*bio);
 798                 *bio = NULL;
 799         }
 800
 801         return ret;
 802 }
 803
 804 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 805                                         struct bio **bio, struct page *page)
 806 {
 807         enum temp_type temp;
 808         bool found = false;
 809         struct bio *target = bio ? *bio : NULL;
 810
 811         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 812                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 813                 struct list_head *head = &io->bio_list;
 814                 struct bio_entry *be;
 815
 816                 if (list_empty(head))
 817                         continue;
 818
 819                 down_read(&io->bio_list_lock);
 820                 list_for_each_entry(be, head, list) {
 821                         if (target)
 822                                 found = (target == be->bio);
 823                         else
 824                                 found = __has_merged_page(be->bio, NULL,
 825                                                                 page, 0);
 826                         if (found)
 827                                 break;
 828                 }
 829                 up_read(&io->bio_list_lock);
 830
 831                 if (!found)
 832                         continue;
 833
 834                 found = false;
 835
 836                 down_write(&io->bio_list_lock);
 837                 list_for_each_entry(be, head, list) {
 838                         if (target)
 839                                 found = (target == be->bio);
 840                         else
 841                                 found = __has_merged_page(be->bio, NULL,
 842                                                                 page, 0);
 843                         if (found) {
 844                                 target = be->bio;
 845                                 del_bio_entry(be);
 846                                 break;
 847                         }
 848                 }
 849                 up_write(&io->bio_list_lock);
 850         }
 851
 852         if (found)
 853                 __submit_bio(sbi, target, DATA);
 854         if (bio && *bio) {
 855                 bio_put(*bio);
 856                 *bio = NULL;
 857         }
 858 }
 859
 860 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 861 {
 862         struct bio *bio = *fio->bio;
 863         struct page *page = fio->encrypted_page ?
 864                         fio->encrypted_page : fio->page;
 865
 866         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 867                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 868                 return -EFSCORRUPTED;
 869
 870         trace_f2fs_submit_page_bio(page, fio);
 871
 872         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 873                                                 fio->new_blkaddr))
 874                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 875 alloc_new:
 876         if (!bio) {
 877                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 878                 __attach_io_flag(fio);
 879                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 880                                        fio->page->index, fio, GFP_NOIO);
 881                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 882
 883                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 884         } else {
 885                 if (add_ipu_page(fio, &bio, page))
 886                         goto alloc_new;
 887         }
 888
 889         if (fio->io_wbc)
 890                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 891
 892         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 893
 894         *fio->last_block = fio->new_blkaddr;
 895         *fio->bio = bio;
 896
 897         return 0;
 898 }
 899
 900 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 901 {
 902         struct f2fs_sb_info *sbi = fio->sbi;
 903         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 904         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 905         struct page *bio_page;
 906
 907         f2fs_bug_on(sbi, is_read_io(fio->op));
 908
 909         down_write(&io->io_rwsem);
 910 next:
 911         if (fio->in_list) {
 912                 spin_lock(&io->io_lock);
 913                 if (list_empty(&io->io_list)) {
 914                         spin_unlock(&io->io_lock);
 915                         goto out;
 916                 }
 917                 fio = list_first_entry(&io->io_list,
 918                                                 struct f2fs_io_info, list);
 919                 list_del(&fio->list);
 920                 spin_unlock(&io->io_lock);
 921         }
 922
 923         verify_fio_blkaddr(fio);
 924
 925         if (fio->encrypted_page)
 926                 bio_page = fio->encrypted_page;
 927         else if (fio->compressed_page)
 928                 bio_page = fio->compressed_page;
 929         else
 930                 bio_page = fio->page;
 931
 932         /* set submitted = true as a return value */
 933         fio->submitted = true;
 934
 935         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 936
 937         if (io->bio &&
 938             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 939                               fio->new_blkaddr) ||
 940              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 941                                        bio_page->index, fio)))
 942                 __submit_merged_bio(io);
 943 alloc_new:
 944         if (io->bio == NULL) {
 945                 if (F2FS_IO_ALIGNED(sbi) &&
 946                                 (fio->type == DATA || fio->type == NODE) &&
 947                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 948                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 949                         fio->retry = true;
 950                         goto skip;
 951                 }
 952                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
 953                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 954                                        bio_page->index, fio, GFP_NOIO);
 955                 io->fio = *fio;
 956         }
 957
 958         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 959                 __submit_merged_bio(io);
 960                 goto alloc_new;
 961         }
 962
 963         if (fio->io_wbc)
 964                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
 965
 966         io->last_block_in_bio = fio->new_blkaddr;
 967
 968         trace_f2fs_submit_page_write(fio->page, fio);
 969 skip:
 970         if (fio->in_list)
 971                 goto next;
 972 out:
 973         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 974                                 !f2fs_is_checkpoint_ready(sbi))
 975                 __submit_merged_bio(io);
 976         up_write(&io->io_rwsem);
 977 }
 978
 979 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 980                                       unsigned nr_pages, unsigned op_flag,
 981                                       pgoff_t first_idx, bool for_write)
 982 {
 983         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 984         struct bio *bio;
 985         struct bio_post_read_ctx *ctx = NULL;
 986         unsigned int post_read_steps = 0;
 987
 988         bio = bio_alloc_bioset(for_write ? GFP_NOIO : GFP_KERNEL,
 989                                bio_max_segs(nr_pages), &f2fs_bioset);
 990         if (!bio)
 991                 return ERR_PTR(-ENOMEM);
 992
 993         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
 994
 995         f2fs_target_device(sbi, blkaddr, bio);
 996         bio->bi_end_io = f2fs_read_end_io;
 997         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 998
 999         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1000                 post_read_steps |= STEP_DECRYPT;
1001
1002         if (f2fs_need_verity(inode, first_idx))
1003                 post_read_steps |= STEP_VERITY;
1004
1005         /*
1006          * STEP_DECOMPRESS is handled specially, since a compressed file might
1007          * contain both compressed and uncompressed clusters.  We'll allocate a
1008          * bio_post_read_ctx if the file is compressed, but the caller is
1009          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1010          */
1011
1012         if (post_read_steps || f2fs_compressed_file(inode)) {
1013                 /* Due to the mempool, this never fails. */
1014                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1015                 ctx->bio = bio;
1016                 ctx->sbi = sbi;
1017                 ctx->enabled_steps = post_read_steps;
1018                 ctx->fs_blkaddr = blkaddr;
1019                 bio->bi_private = ctx;
1020         }
1021         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1022
1023         return bio;
1024 }
1025
1026 /* This can handle encryption stuffs */
1027 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1028                                  block_t blkaddr, int op_flags, bool for_write)
1029 {
1030         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1031         struct bio *bio;
1032
1033         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1034                                         page->index, for_write);
1035         if (IS_ERR(bio))
1036                 return PTR_ERR(bio);
1037
1038         /* wait for GCed page writeback via META_MAPPING */
1039         f2fs_wait_on_block_writeback(inode, blkaddr);
1040
1041         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1042                 bio_put(bio);
1043                 return -EFAULT;
1044         }
1045         ClearPageError(page);
1046         inc_page_count(sbi, F2FS_RD_DATA);
1047         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1048         __submit_bio(sbi, bio, DATA);
1049         return 0;
1050 }
1051
1052 static void __set_data_blkaddr(struct dnode_of_data *dn)
1053 {
1054         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1055         __le32 *addr_array;
1056         int base = 0;
1057
1058         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1059                 base = get_extra_isize(dn->inode);
1060
1061         /* Get physical address of data block */
1062         addr_array = blkaddr_in_node(rn);
1063         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1064 }
1065
1066 /*
1067  * Lock ordering for the change of data block address:
1068  * ->data_page
1069  *  ->node_page
1070  *    update block addresses in the node page
1071  */
1072 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1073 {
1074         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1075         __set_data_blkaddr(dn);
1076         if (set_page_dirty(dn->node_page))
1077                 dn->node_changed = true;
1078 }
1079
1080 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1081 {
1082         dn->data_blkaddr = blkaddr;
1083         f2fs_set_data_blkaddr(dn);
1084         f2fs_update_extent_cache(dn);
1085 }
1086
1087 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1088 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1089 {
1090         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1091         int err;
1092
1093         if (!count)
1094                 return 0;
1095
1096         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1097                 return -EPERM;
1098         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1099                 return err;
1100
1101         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1102                                                 dn->ofs_in_node, count);
1103
1104         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1105
1106         for (; count > 0; dn->ofs_in_node++) {
1107                 block_t blkaddr = f2fs_data_blkaddr(dn);
1108
1109                 if (blkaddr == NULL_ADDR) {
1110                         dn->data_blkaddr = NEW_ADDR;
1111                         __set_data_blkaddr(dn);
1112                         count--;
1113                 }
1114         }
1115
1116         if (set_page_dirty(dn->node_page))
1117                 dn->node_changed = true;
1118         return 0;
1119 }
1120
1121 /* Should keep dn->ofs_in_node unchanged */
1122 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1123 {
1124         unsigned int ofs_in_node = dn->ofs_in_node;
1125         int ret;
1126
1127         ret = f2fs_reserve_new_blocks(dn, 1);
1128         dn->ofs_in_node = ofs_in_node;
1129         return ret;
1130 }
1131
1132 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1133 {
1134         bool need_put = dn->inode_page ? false : true;
1135         int err;
1136
1137         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1138         if (err)
1139                 return err;
1140
1141         if (dn->data_blkaddr == NULL_ADDR)
1142                 err = f2fs_reserve_new_block(dn);
1143         if (err || need_put)
1144                 f2fs_put_dnode(dn);
1145         return err;
1146 }
1147
1148 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1149 {
1150         struct extent_info ei = {0, };
1151         struct inode *inode = dn->inode;
1152
1153         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1154                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1155                 return 0;
1156         }
1157
1158         return f2fs_reserve_block(dn, index);
1159 }
1160
1161 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1162                                                 int op_flags, bool for_write)
1163 {
1164         struct address_space *mapping = inode->i_mapping;
1165         struct dnode_of_data dn;
1166         struct page *page;
1167         struct extent_info ei = {0, };
1168         int err;
1169
1170         page = f2fs_grab_cache_page(mapping, index, for_write);
1171         if (!page)
1172                 return ERR_PTR(-ENOMEM);
1173
1174         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1175                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1176                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1177                                                 DATA_GENERIC_ENHANCE_READ)) {
1178                         err = -EFSCORRUPTED;
1179                         goto put_err;
1180                 }
1181                 goto got_it;
1182         }
1183
1184         set_new_dnode(&dn, inode, NULL, NULL, 0);
1185         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1186         if (err)
1187                 goto put_err;
1188         f2fs_put_dnode(&dn);
1189
1190         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1191                 err = -ENOENT;
1192                 goto put_err;
1193         }
1194         if (dn.data_blkaddr != NEW_ADDR &&
1195                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1196                                                 dn.data_blkaddr,
1197                                                 DATA_GENERIC_ENHANCE)) {
1198                 err = -EFSCORRUPTED;
1199                 goto put_err;
1200         }
1201 got_it:
1202         if (PageUptodate(page)) {
1203                 unlock_page(page);
1204                 return page;
1205         }
1206
1207         /*
1208          * A new dentry page is allocated but not able to be written, since its
1209          * new inode page couldn't be allocated due to -ENOSPC.
1210          * In such the case, its blkaddr can be remained as NEW_ADDR.
1211          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1212          * f2fs_init_inode_metadata.
1213          */
1214         if (dn.data_blkaddr == NEW_ADDR) {
1215                 zero_user_segment(page, 0, PAGE_SIZE);
1216                 if (!PageUptodate(page))
1217                         SetPageUptodate(page);
1218                 unlock_page(page);
1219                 return page;
1220         }
1221
1222         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1223                                                 op_flags, for_write);
1224         if (err)
1225                 goto put_err;
1226         return page;
1227
1228 put_err:
1229         f2fs_put_page(page, 1);
1230         return ERR_PTR(err);
1231 }
1232
1233 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1234 {
1235         struct address_space *mapping = inode->i_mapping;
1236         struct page *page;
1237
1238         page = find_get_page(mapping, index);
1239         if (page && PageUptodate(page))
1240                 return page;
1241         f2fs_put_page(page, 0);
1242
1243         page = f2fs_get_read_data_page(inode, index, 0, false);
1244         if (IS_ERR(page))
1245                 return page;
1246
1247         if (PageUptodate(page))
1248                 return page;
1249
1250         wait_on_page_locked(page);
1251         if (unlikely(!PageUptodate(page))) {
1252                 f2fs_put_page(page, 0);
1253                 return ERR_PTR(-EIO);
1254         }
1255         return page;
1256 }
1257
1258 /*
1259  * If it tries to access a hole, return an error.
1260  * Because, the callers, functions in dir.c and GC, should be able to know
1261  * whether this page exists or not.
1262  */
1263 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1264                                                         bool for_write)
1265 {
1266         struct address_space *mapping = inode->i_mapping;
1267         struct page *page;
1268 repeat:
1269         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1270         if (IS_ERR(page))
1271                 return page;
1272
1273         /* wait for read completion */
1274         lock_page(page);
1275         if (unlikely(page->mapping != mapping)) {
1276                 f2fs_put_page(page, 1);
1277                 goto repeat;
1278         }
1279         if (unlikely(!PageUptodate(page))) {
1280                 f2fs_put_page(page, 1);
1281                 return ERR_PTR(-EIO);
1282         }
1283         return page;
1284 }
1285
1286 /*
1287  * Caller ensures that this data page is never allocated.
1288  * A new zero-filled data page is allocated in the page cache.
1289  *
1290  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1291  * f2fs_unlock_op().
1292  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1293  * ipage should be released by this function.
1294  */
1295 struct page *f2fs_get_new_data_page(struct inode *inode,
1296                 struct page *ipage, pgoff_t index, bool new_i_size)
1297 {
1298         struct address_space *mapping = inode->i_mapping;
1299         struct page *page;
1300         struct dnode_of_data dn;
1301         int err;
1302
1303         page = f2fs_grab_cache_page(mapping, index, true);
1304         if (!page) {
1305                 /*
1306                  * before exiting, we should make sure ipage will be released
1307                  * if any error occur.
1308                  */
1309                 f2fs_put_page(ipage, 1);
1310                 return ERR_PTR(-ENOMEM);
1311         }
1312
1313         set_new_dnode(&dn, inode, ipage, NULL, 0);
1314         err = f2fs_reserve_block(&dn, index);
1315         if (err) {
1316                 f2fs_put_page(page, 1);
1317                 return ERR_PTR(err);
1318         }
1319         if (!ipage)
1320                 f2fs_put_dnode(&dn);
1321
1322         if (PageUptodate(page))
1323                 goto got_it;
1324
1325         if (dn.data_blkaddr == NEW_ADDR) {
1326                 zero_user_segment(page, 0, PAGE_SIZE);
1327                 if (!PageUptodate(page))
1328                         SetPageUptodate(page);
1329         } else {
1330                 f2fs_put_page(page, 1);
1331
1332                 /* if ipage exists, blkaddr should be NEW_ADDR */
1333                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1334                 page = f2fs_get_lock_data_page(inode, index, true);
1335                 if (IS_ERR(page))
1336                         return page;
1337         }
1338 got_it:
1339         if (new_i_size && i_size_read(inode) <
1340                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1341                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1342         return page;
1343 }
1344
1345 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1346 {
1347         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1348         struct f2fs_summary sum;
1349         struct node_info ni;
1350         block_t old_blkaddr;
1351         blkcnt_t count = 1;
1352         int err;
1353
1354         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1355                 return -EPERM;
1356
1357         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1358         if (err)
1359                 return err;
1360
1361         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1362         if (dn->data_blkaddr != NULL_ADDR)
1363                 goto alloc;
1364
1365         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1366                 return err;
1367
1368 alloc:
1369         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1370         old_blkaddr = dn->data_blkaddr;
1371         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1372                                 &sum, seg_type, NULL);
1373         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1374                 invalidate_mapping_pages(META_MAPPING(sbi),
1375                                         old_blkaddr, old_blkaddr);
1376                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1377         }
1378         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1379         return 0;
1380 }
1381
1382 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1383 {
1384         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1385                 if (lock)
1386                         down_read(&sbi->node_change);
1387                 else
1388                         up_read(&sbi->node_change);
1389         } else {
1390                 if (lock)
1391                         f2fs_lock_op(sbi);
1392                 else
1393                         f2fs_unlock_op(sbi);
1394         }
1395 }
1396
1397 /*
1398  * f2fs_map_blocks() tries to find or build mapping relationship which
1399  * maps continuous logical blocks to physical blocks, and return such
1400  * info via f2fs_map_blocks structure.
1401  */
1402 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1403                                                 int create, int flag)
1404 {
1405         unsigned int maxblocks = map->m_len;
1406         struct dnode_of_data dn;
1407         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1408         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1409         pgoff_t pgofs, end_offset, end;
1410         int err = 0, ofs = 1;
1411         unsigned int ofs_in_node, last_ofs_in_node;
1412         blkcnt_t prealloc;
1413         struct extent_info ei = {0, };
1414         block_t blkaddr;
1415         unsigned int start_pgofs;
1416         int bidx = 0;
1417
1418         if (!maxblocks)
1419                 return 0;
1420
1421         map->m_bdev = inode->i_sb->s_bdev;
1422         map->m_multidev_dio =
1423                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1424
1425         map->m_len = 0;
1426         map->m_flags = 0;
1427
1428         /* it only supports block size == page size */
1429         pgofs = (pgoff_t)map->m_lblk;
1430         end = pgofs + maxblocks;
1431
1432         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1433                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1434                                                         map->m_may_create)
1435                         goto next_dnode;
1436
1437                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1438                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1439                 map->m_flags = F2FS_MAP_MAPPED;
1440                 if (map->m_next_extent)
1441                         *map->m_next_extent = pgofs + map->m_len;
1442
1443                 /* for hardware encryption, but to avoid potential issue in future */
1444                 if (flag == F2FS_GET_BLOCK_DIO)
1445                         f2fs_wait_on_block_writeback_range(inode,
1446                                                 map->m_pblk, map->m_len);
1447
1448                 if (map->m_multidev_dio) {
1449                         block_t blk_addr = map->m_pblk;
1450
1451                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1452
1453                         map->m_bdev = FDEV(bidx).bdev;
1454                         map->m_pblk -= FDEV(bidx).start_blk;
1455                         map->m_len = min(map->m_len,
1456                                 FDEV(bidx).end_blk + 1 - map->m_pblk);
1457
1458                         if (map->m_may_create)
1459                                 f2fs_update_device_state(sbi, inode->i_ino,
1460                                                         blk_addr, map->m_len);
1461                 }
1462                 goto out;
1463         }
1464
1465 next_dnode:
1466         if (map->m_may_create)
1467                 f2fs_do_map_lock(sbi, flag, true);
1468
1469         /* When reading holes, we need its node page */
1470         set_new_dnode(&dn, inode, NULL, NULL, 0);
1471         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1472         if (err) {
1473                 if (flag == F2FS_GET_BLOCK_BMAP)
1474                         map->m_pblk = 0;
1475
1476                 if (err == -ENOENT) {
1477                         /*
1478                          * There is one exceptional case that read_node_page()
1479                          * may return -ENOENT due to filesystem has been
1480                          * shutdown or cp_error, so force to convert error
1481                          * number to EIO for such case.
1482                          */
1483                         if (map->m_may_create &&
1484                                 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1485                                 f2fs_cp_error(sbi))) {
1486                                 err = -EIO;
1487                                 goto unlock_out;
1488                         }
1489
1490                         err = 0;
1491                         if (map->m_next_pgofs)
1492                                 *map->m_next_pgofs =
1493                                         f2fs_get_next_page_offset(&dn, pgofs);
1494                         if (map->m_next_extent)
1495                                 *map->m_next_extent =
1496                                         f2fs_get_next_page_offset(&dn, pgofs);
1497                 }
1498                 goto unlock_out;
1499         }
1500
1501         start_pgofs = pgofs;
1502         prealloc = 0;
1503         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1504         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1505
1506 next_block:
1507         blkaddr = f2fs_data_blkaddr(&dn);
1508
1509         if (__is_valid_data_blkaddr(blkaddr) &&
1510                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1511                 err = -EFSCORRUPTED;
1512                 goto sync_out;
1513         }
1514
1515         if (__is_valid_data_blkaddr(blkaddr)) {
1516                 /* use out-place-update for driect IO under LFS mode */
1517                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1518                                                         map->m_may_create) {
1519                         err = __allocate_data_block(&dn, map->m_seg_type);
1520                         if (err)
1521                                 goto sync_out;
1522                         blkaddr = dn.data_blkaddr;
1523                         set_inode_flag(inode, FI_APPEND_WRITE);
1524                 }
1525         } else {
1526                 if (create) {
1527                         if (unlikely(f2fs_cp_error(sbi))) {
1528                                 err = -EIO;
1529                                 goto sync_out;
1530                         }
1531                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1532                                 if (blkaddr == NULL_ADDR) {
1533                                         prealloc++;
1534                                         last_ofs_in_node = dn.ofs_in_node;
1535                                 }
1536                         } else {
1537                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1538                                         flag != F2FS_GET_BLOCK_DIO);
1539                                 err = __allocate_data_block(&dn,
1540                                                         map->m_seg_type);
1541                                 if (!err) {
1542                                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1543                                                 file_need_truncate(inode);
1544                                         set_inode_flag(inode, FI_APPEND_WRITE);
1545                                 }
1546                         }
1547                         if (err)
1548                                 goto sync_out;
1549                         map->m_flags |= F2FS_MAP_NEW;
1550                         blkaddr = dn.data_blkaddr;
1551                 } else {
1552                         if (f2fs_compressed_file(inode) &&
1553                                         f2fs_sanity_check_cluster(&dn) &&
1554                                         (flag != F2FS_GET_BLOCK_FIEMAP ||
1555                                         IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1556                                 err = -EFSCORRUPTED;
1557                                 goto sync_out;
1558                         }
1559                         if (flag == F2FS_GET_BLOCK_BMAP) {
1560                                 map->m_pblk = 0;
1561                                 goto sync_out;
1562                         }
1563                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1564                                 goto sync_out;
1565                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1566                                                 blkaddr == NULL_ADDR) {
1567                                 if (map->m_next_pgofs)
1568                                         *map->m_next_pgofs = pgofs + 1;
1569                                 goto sync_out;
1570                         }
1571                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1572                                 /* for defragment case */
1573                                 if (map->m_next_pgofs)
1574                                         *map->m_next_pgofs = pgofs + 1;
1575                                 goto sync_out;
1576                         }
1577                 }
1578         }
1579
1580         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1581                 goto skip;
1582
1583         if (map->m_multidev_dio)
1584                 bidx = f2fs_target_device_index(sbi, blkaddr);
1585
1586         if (map->m_len == 0) {
1587                 /* preallocated unwritten block should be mapped for fiemap. */
1588                 if (blkaddr == NEW_ADDR)
1589                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1590                 map->m_flags |= F2FS_MAP_MAPPED;
1591
1592                 map->m_pblk = blkaddr;
1593                 map->m_len = 1;
1594
1595                 if (map->m_multidev_dio)
1596                         map->m_bdev = FDEV(bidx).bdev;
1597         } else if ((map->m_pblk != NEW_ADDR &&
1598                         blkaddr == (map->m_pblk + ofs)) ||
1599                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1600                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1601                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1602                         goto sync_out;
1603                 ofs++;
1604                 map->m_len++;
1605         } else {
1606                 goto sync_out;
1607         }
1608
1609 skip:
1610         dn.ofs_in_node++;
1611         pgofs++;
1612
1613         /* preallocate blocks in batch for one dnode page */
1614         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1615                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1616
1617                 dn.ofs_in_node = ofs_in_node;
1618                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1619                 if (err)
1620                         goto sync_out;
1621
1622                 map->m_len += dn.ofs_in_node - ofs_in_node;
1623                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1624                         err = -ENOSPC;
1625                         goto sync_out;
1626                 }
1627                 dn.ofs_in_node = end_offset;
1628         }
1629
1630         if (pgofs >= end)
1631                 goto sync_out;
1632         else if (dn.ofs_in_node < end_offset)
1633                 goto next_block;
1634
1635         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1636                 if (map->m_flags & F2FS_MAP_MAPPED) {
1637                         unsigned int ofs = start_pgofs - map->m_lblk;
1638
1639                         f2fs_update_extent_cache_range(&dn,
1640                                 start_pgofs, map->m_pblk + ofs,
1641                                 map->m_len - ofs);
1642                 }
1643         }
1644
1645         f2fs_put_dnode(&dn);
1646
1647         if (map->m_may_create) {
1648                 f2fs_do_map_lock(sbi, flag, false);
1649                 f2fs_balance_fs(sbi, dn.node_changed);
1650         }
1651         goto next_dnode;
1652
1653 sync_out:
1654
1655         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1656                 /*
1657                  * for hardware encryption, but to avoid potential issue
1658                  * in future
1659                  */
1660                 f2fs_wait_on_block_writeback_range(inode,
1661                                                 map->m_pblk, map->m_len);
1662                 invalidate_mapping_pages(META_MAPPING(sbi),
1663                                                 map->m_pblk, map->m_pblk);
1664
1665                 if (map->m_multidev_dio) {
1666                         block_t blk_addr = map->m_pblk;
1667
1668                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1669
1670                         map->m_bdev = FDEV(bidx).bdev;
1671                         map->m_pblk -= FDEV(bidx).start_blk;
1672
1673                         if (map->m_may_create)
1674                                 f2fs_update_device_state(sbi, inode->i_ino,
1675                                                         blk_addr, map->m_len);
1676
1677                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1678                                                 FDEV(bidx).end_blk + 1);
1679                 }
1680         }
1681
1682         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1683                 if (map->m_flags & F2FS_MAP_MAPPED) {
1684                         unsigned int ofs = start_pgofs - map->m_lblk;
1685
1686                         f2fs_update_extent_cache_range(&dn,
1687                                 start_pgofs, map->m_pblk + ofs,
1688                                 map->m_len - ofs);
1689                 }
1690                 if (map->m_next_extent)
1691                         *map->m_next_extent = pgofs + 1;
1692         }
1693         f2fs_put_dnode(&dn);
1694 unlock_out:
1695         if (map->m_may_create) {
1696                 f2fs_do_map_lock(sbi, flag, false);
1697                 f2fs_balance_fs(sbi, dn.node_changed);
1698         }
1699 out:
1700         trace_f2fs_map_blocks(inode, map, create, flag, err);
1701         return err;
1702 }
1703
1704 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1705 {
1706         struct f2fs_map_blocks map;
1707         block_t last_lblk;
1708         int err;
1709
1710         if (pos + len > i_size_read(inode))
1711                 return false;
1712
1713         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1714         map.m_next_pgofs = NULL;
1715         map.m_next_extent = NULL;
1716         map.m_seg_type = NO_CHECK_TYPE;
1717         map.m_may_create = false;
1718         last_lblk = F2FS_BLK_ALIGN(pos + len);
1719
1720         while (map.m_lblk < last_lblk) {
1721                 map.m_len = last_lblk - map.m_lblk;
1722                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1723                 if (err || map.m_len == 0)
1724                         return false;
1725                 map.m_lblk += map.m_len;
1726         }
1727         return true;
1728 }
1729
1730 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1731 {
1732         return (bytes >> inode->i_blkbits);
1733 }
1734
1735 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1736 {
1737         return (blks << inode->i_blkbits);
1738 }
1739
1740 static int f2fs_xattr_fiemap(struct inode *inode,
1741                                 struct fiemap_extent_info *fieinfo)
1742 {
1743         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1744         struct page *page;
1745         struct node_info ni;
1746         __u64 phys = 0, len;
1747         __u32 flags;
1748         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1749         int err = 0;
1750
1751         if (f2fs_has_inline_xattr(inode)) {
1752                 int offset;
1753
1754                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1755                                                 inode->i_ino, false);
1756                 if (!page)
1757                         return -ENOMEM;
1758
1759                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1760                 if (err) {
1761                         f2fs_put_page(page, 1);
1762                         return err;
1763                 }
1764
1765                 phys = blks_to_bytes(inode, ni.blk_addr);
1766                 offset = offsetof(struct f2fs_inode, i_addr) +
1767                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1768                                         get_inline_xattr_addrs(inode));
1769
1770                 phys += offset;
1771                 len = inline_xattr_size(inode);
1772
1773                 f2fs_put_page(page, 1);
1774
1775                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1776
1777                 if (!xnid)
1778                         flags |= FIEMAP_EXTENT_LAST;
1779
1780                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1781                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1782                 if (err || err == 1)
1783                         return err;
1784         }
1785
1786         if (xnid) {
1787                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1788                 if (!page)
1789                         return -ENOMEM;
1790
1791                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1792                 if (err) {
1793                         f2fs_put_page(page, 1);
1794                         return err;
1795                 }
1796
1797                 phys = blks_to_bytes(inode, ni.blk_addr);
1798                 len = inode->i_sb->s_blocksize;
1799
1800                 f2fs_put_page(page, 1);
1801
1802                 flags = FIEMAP_EXTENT_LAST;
1803         }
1804
1805         if (phys) {
1806                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1807                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1808         }
1809
1810         return (err < 0 ? err : 0);
1811 }
1812
1813 static loff_t max_inode_blocks(struct inode *inode)
1814 {
1815         loff_t result = ADDRS_PER_INODE(inode);
1816         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1817
1818         /* two direct node blocks */
1819         result += (leaf_count * 2);
1820
1821         /* two indirect node blocks */
1822         leaf_count *= NIDS_PER_BLOCK;
1823         result += (leaf_count * 2);
1824
1825         /* one double indirect node block */
1826         leaf_count *= NIDS_PER_BLOCK;
1827         result += leaf_count;
1828
1829         return result;
1830 }
1831
1832 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1833                 u64 start, u64 len)
1834 {
1835         struct f2fs_map_blocks map;
1836         sector_t start_blk, last_blk;
1837         pgoff_t next_pgofs;
1838         u64 logical = 0, phys = 0, size = 0;
1839         u32 flags = 0;
1840         int ret = 0;
1841         bool compr_cluster = false, compr_appended;
1842         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1843         unsigned int count_in_cluster = 0;
1844         loff_t maxbytes;
1845
1846         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1847                 ret = f2fs_precache_extents(inode);
1848                 if (ret)
1849                         return ret;
1850         }
1851
1852         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1853         if (ret)
1854                 return ret;
1855
1856         inode_lock(inode);
1857
1858         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1859         if (start > maxbytes) {
1860                 ret = -EFBIG;
1861                 goto out;
1862         }
1863
1864         if (len > maxbytes || (maxbytes - len) < start)
1865                 len = maxbytes - start;
1866
1867         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1868                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1869                 goto out;
1870         }
1871
1872         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1873                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1874                 if (ret != -EAGAIN)
1875                         goto out;
1876         }
1877
1878         if (bytes_to_blks(inode, len) == 0)
1879                 len = blks_to_bytes(inode, 1);
1880
1881         start_blk = bytes_to_blks(inode, start);
1882         last_blk = bytes_to_blks(inode, start + len - 1);
1883
1884 next:
1885         memset(&map, 0, sizeof(map));
1886         map.m_lblk = start_blk;
1887         map.m_len = bytes_to_blks(inode, len);
1888         map.m_next_pgofs = &next_pgofs;
1889         map.m_seg_type = NO_CHECK_TYPE;
1890
1891         if (compr_cluster) {
1892                 map.m_lblk += 1;
1893                 map.m_len = cluster_size - count_in_cluster;
1894         }
1895
1896         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1897         if (ret)
1898                 goto out;
1899
1900         /* HOLE */
1901         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1902                 start_blk = next_pgofs;
1903
1904                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1905                                                 max_inode_blocks(inode)))
1906                         goto prep_next;
1907
1908                 flags |= FIEMAP_EXTENT_LAST;
1909         }
1910
1911         compr_appended = false;
1912         /* In a case of compressed cluster, append this to the last extent */
1913         if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1914                         !(map.m_flags & F2FS_MAP_FLAGS))) {
1915                 compr_appended = true;
1916                 goto skip_fill;
1917         }
1918
1919         if (size) {
1920                 flags |= FIEMAP_EXTENT_MERGED;
1921                 if (IS_ENCRYPTED(inode))
1922                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1923
1924                 ret = fiemap_fill_next_extent(fieinfo, logical,
1925                                 phys, size, flags);
1926                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1927                 if (ret)
1928                         goto out;
1929                 size = 0;
1930         }
1931
1932         if (start_blk > last_blk)
1933                 goto out;
1934
1935 skip_fill:
1936         if (map.m_pblk == COMPRESS_ADDR) {
1937                 compr_cluster = true;
1938                 count_in_cluster = 1;
1939         } else if (compr_appended) {
1940                 unsigned int appended_blks = cluster_size -
1941                                                 count_in_cluster + 1;
1942                 size += blks_to_bytes(inode, appended_blks);
1943                 start_blk += appended_blks;
1944                 compr_cluster = false;
1945         } else {
1946                 logical = blks_to_bytes(inode, start_blk);
1947                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
1948                         blks_to_bytes(inode, map.m_pblk) : 0;
1949                 size = blks_to_bytes(inode, map.m_len);
1950                 flags = 0;
1951
1952                 if (compr_cluster) {
1953                         flags = FIEMAP_EXTENT_ENCODED;
1954                         count_in_cluster += map.m_len;
1955                         if (count_in_cluster == cluster_size) {
1956                                 compr_cluster = false;
1957                                 size += blks_to_bytes(inode, 1);
1958                         }
1959                 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
1960                         flags = FIEMAP_EXTENT_UNWRITTEN;
1961                 }
1962
1963                 start_blk += bytes_to_blks(inode, size);
1964         }
1965
1966 prep_next:
1967         cond_resched();
1968         if (fatal_signal_pending(current))
1969                 ret = -EINTR;
1970         else
1971                 goto next;
1972 out:
1973         if (ret == 1)
1974                 ret = 0;
1975
1976         inode_unlock(inode);
1977         return ret;
1978 }
1979
1980 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1981 {
1982         if (IS_ENABLED(CONFIG_FS_VERITY) &&
1983             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1984                 return inode->i_sb->s_maxbytes;
1985
1986         return i_size_read(inode);
1987 }
1988
1989 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1990                                         unsigned nr_pages,
1991                                         struct f2fs_map_blocks *map,
1992                                         struct bio **bio_ret,
1993                                         sector_t *last_block_in_bio,
1994                                         bool is_readahead)
1995 {
1996         struct bio *bio = *bio_ret;
1997         const unsigned blocksize = blks_to_bytes(inode, 1);
1998         sector_t block_in_file;
1999         sector_t last_block;
2000         sector_t last_block_in_file;
2001         sector_t block_nr;
2002         int ret = 0;
2003
2004         block_in_file = (sector_t)page_index(page);
2005         last_block = block_in_file + nr_pages;
2006         last_block_in_file = bytes_to_blks(inode,
2007                         f2fs_readpage_limit(inode) + blocksize - 1);
2008         if (last_block > last_block_in_file)
2009                 last_block = last_block_in_file;
2010
2011         /* just zeroing out page which is beyond EOF */
2012         if (block_in_file >= last_block)
2013                 goto zero_out;
2014         /*
2015          * Map blocks using the previous result first.
2016          */
2017         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2018                         block_in_file > map->m_lblk &&
2019                         block_in_file < (map->m_lblk + map->m_len))
2020                 goto got_it;
2021
2022         /*
2023          * Then do more f2fs_map_blocks() calls until we are
2024          * done with this page.
2025          */
2026         map->m_lblk = block_in_file;
2027         map->m_len = last_block - block_in_file;
2028
2029         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2030         if (ret)
2031                 goto out;
2032 got_it:
2033         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2034                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2035                 SetPageMappedToDisk(page);
2036
2037                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2038                                                 DATA_GENERIC_ENHANCE_READ)) {
2039                         ret = -EFSCORRUPTED;
2040                         goto out;
2041                 }
2042         } else {
2043 zero_out:
2044                 zero_user_segment(page, 0, PAGE_SIZE);
2045                 if (f2fs_need_verity(inode, page->index) &&
2046                     !fsverity_verify_page(page)) {
2047                         ret = -EIO;
2048                         goto out;
2049                 }
2050                 if (!PageUptodate(page))
2051                         SetPageUptodate(page);
2052                 unlock_page(page);
2053                 goto out;
2054         }
2055
2056         /*
2057          * This page will go to BIO.  Do we need to send this
2058          * BIO off first?
2059          */
2060         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2061                                        *last_block_in_bio, block_nr) ||
2062                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2063 submit_and_realloc:
2064                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2065                 bio = NULL;
2066         }
2067         if (bio == NULL) {
2068                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2069                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2070                                 false);
2071                 if (IS_ERR(bio)) {
2072                         ret = PTR_ERR(bio);
2073                         bio = NULL;
2074                         goto out;
2075                 }
2076         }
2077
2078         /*
2079          * If the page is under writeback, we need to wait for
2080          * its completion to see the correct decrypted data.
2081          */
2082         f2fs_wait_on_block_writeback(inode, block_nr);
2083
2084         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2085                 goto submit_and_realloc;
2086
2087         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2088         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2089         ClearPageError(page);
2090         *last_block_in_bio = block_nr;
2091         goto out;
2092 out:
2093         *bio_ret = bio;
2094         return ret;
2095 }
2096
2097 #ifdef CONFIG_F2FS_FS_COMPRESSION
2098 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2099                                 unsigned nr_pages, sector_t *last_block_in_bio,
2100                                 bool is_readahead, bool for_write)
2101 {
2102         struct dnode_of_data dn;
2103         struct inode *inode = cc->inode;
2104         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2105         struct bio *bio = *bio_ret;
2106         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2107         sector_t last_block_in_file;
2108         const unsigned blocksize = blks_to_bytes(inode, 1);
2109         struct decompress_io_ctx *dic = NULL;
2110         struct extent_info ei = {0, };
2111         bool from_dnode = true;
2112         int i;
2113         int ret = 0;
2114
2115         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2116
2117         last_block_in_file = bytes_to_blks(inode,
2118                         f2fs_readpage_limit(inode) + blocksize - 1);
2119
2120         /* get rid of pages beyond EOF */
2121         for (i = 0; i < cc->cluster_size; i++) {
2122                 struct page *page = cc->rpages[i];
2123
2124                 if (!page)
2125                         continue;
2126                 if ((sector_t)page->index >= last_block_in_file) {
2127                         zero_user_segment(page, 0, PAGE_SIZE);
2128                         if (!PageUptodate(page))
2129                                 SetPageUptodate(page);
2130                 } else if (!PageUptodate(page)) {
2131                         continue;
2132                 }
2133                 unlock_page(page);
2134                 if (for_write)
2135                         put_page(page);
2136                 cc->rpages[i] = NULL;
2137                 cc->nr_rpages--;
2138         }
2139
2140         /* we are done since all pages are beyond EOF */
2141         if (f2fs_cluster_is_empty(cc))
2142                 goto out;
2143
2144         if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2145                 from_dnode = false;
2146
2147         if (!from_dnode)
2148                 goto skip_reading_dnode;
2149
2150         set_new_dnode(&dn, inode, NULL, NULL, 0);
2151         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2152         if (ret)
2153                 goto out;
2154
2155         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2156
2157 skip_reading_dnode:
2158         for (i = 1; i < cc->cluster_size; i++) {
2159                 block_t blkaddr;
2160
2161                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2162                                         dn.ofs_in_node + i) :
2163                                         ei.blk + i - 1;
2164
2165                 if (!__is_valid_data_blkaddr(blkaddr))
2166                         break;
2167
2168                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2169                         ret = -EFAULT;
2170                         goto out_put_dnode;
2171                 }
2172                 cc->nr_cpages++;
2173
2174                 if (!from_dnode && i >= ei.c_len)
2175                         break;
2176         }
2177
2178         /* nothing to decompress */
2179         if (cc->nr_cpages == 0) {
2180                 ret = 0;
2181                 goto out_put_dnode;
2182         }
2183
2184         dic = f2fs_alloc_dic(cc);
2185         if (IS_ERR(dic)) {
2186                 ret = PTR_ERR(dic);
2187                 goto out_put_dnode;
2188         }
2189
2190         for (i = 0; i < cc->nr_cpages; i++) {
2191                 struct page *page = dic->cpages[i];
2192                 block_t blkaddr;
2193                 struct bio_post_read_ctx *ctx;
2194
2195                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2196                                         dn.ofs_in_node + i + 1) :
2197                                         ei.blk + i;
2198
2199                 f2fs_wait_on_block_writeback(inode, blkaddr);
2200
2201                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2202                         if (atomic_dec_and_test(&dic->remaining_pages))
2203                                 f2fs_decompress_cluster(dic);
2204                         continue;
2205                 }
2206
2207                 if (bio && (!page_is_mergeable(sbi, bio,
2208                                         *last_block_in_bio, blkaddr) ||
2209                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2210 submit_and_realloc:
2211                         __submit_bio(sbi, bio, DATA);
2212                         bio = NULL;
2213                 }
2214
2215                 if (!bio) {
2216                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2217                                         is_readahead ? REQ_RAHEAD : 0,
2218                                         page->index, for_write);
2219                         if (IS_ERR(bio)) {
2220                                 ret = PTR_ERR(bio);
2221                                 f2fs_decompress_end_io(dic, ret);
2222                                 f2fs_put_dnode(&dn);
2223                                 *bio_ret = NULL;
2224                                 return ret;
2225                         }
2226                 }
2227
2228                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2229                         goto submit_and_realloc;
2230
2231                 ctx = get_post_read_ctx(bio);
2232                 ctx->enabled_steps |= STEP_DECOMPRESS;
2233                 refcount_inc(&dic->refcnt);
2234
2235                 inc_page_count(sbi, F2FS_RD_DATA);
2236                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2237                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2238                 ClearPageError(page);
2239                 *last_block_in_bio = blkaddr;
2240         }
2241
2242         if (from_dnode)
2243                 f2fs_put_dnode(&dn);
2244
2245         *bio_ret = bio;
2246         return 0;
2247
2248 out_put_dnode:
2249         if (from_dnode)
2250                 f2fs_put_dnode(&dn);
2251 out:
2252         for (i = 0; i < cc->cluster_size; i++) {
2253                 if (cc->rpages[i]) {
2254                         ClearPageUptodate(cc->rpages[i]);
2255                         ClearPageError(cc->rpages[i]);
2256                         unlock_page(cc->rpages[i]);
2257                 }
2258         }
2259         *bio_ret = bio;
2260         return ret;
2261 }
2262 #endif
2263
2264 /*
2265  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2266  * Major change was from block_size == page_size in f2fs by default.
2267  */
2268 static int f2fs_mpage_readpages(struct inode *inode,
2269                 struct readahead_control *rac, struct page *page)
2270 {
2271         struct bio *bio = NULL;
2272         sector_t last_block_in_bio = 0;
2273         struct f2fs_map_blocks map;
2274 #ifdef CONFIG_F2FS_FS_COMPRESSION
2275         struct compress_ctx cc = {
2276                 .inode = inode,
2277                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2278                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2279                 .cluster_idx = NULL_CLUSTER,
2280                 .rpages = NULL,
2281                 .cpages = NULL,
2282                 .nr_rpages = 0,
2283                 .nr_cpages = 0,
2284         };
2285         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2286 #endif
2287         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2288         unsigned max_nr_pages = nr_pages;
2289         int ret = 0;
2290
2291         map.m_pblk = 0;
2292         map.m_lblk = 0;
2293         map.m_len = 0;
2294         map.m_flags = 0;
2295         map.m_next_pgofs = NULL;
2296         map.m_next_extent = NULL;
2297         map.m_seg_type = NO_CHECK_TYPE;
2298         map.m_may_create = false;
2299
2300         for (; nr_pages; nr_pages--) {
2301                 if (rac) {
2302                         page = readahead_page(rac);
2303                         prefetchw(&page->flags);
2304                 }
2305
2306 #ifdef CONFIG_F2FS_FS_COMPRESSION
2307                 if (f2fs_compressed_file(inode)) {
2308                         /* there are remained comressed pages, submit them */
2309                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2310                                 ret = f2fs_read_multi_pages(&cc, &bio,
2311                                                         max_nr_pages,
2312                                                         &last_block_in_bio,
2313                                                         rac != NULL, false);
2314                                 f2fs_destroy_compress_ctx(&cc, false);
2315                                 if (ret)
2316                                         goto set_error_page;
2317                         }
2318                         if (cc.cluster_idx == NULL_CLUSTER) {
2319                                 if (nc_cluster_idx ==
2320                                         page->index >> cc.log_cluster_size) {
2321                                         goto read_single_page;
2322                                 }
2323
2324                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2325                                 if (ret < 0)
2326                                         goto set_error_page;
2327                                 else if (!ret) {
2328                                         nc_cluster_idx =
2329                                                 page->index >> cc.log_cluster_size;
2330                                         goto read_single_page;
2331                                 }
2332
2333                                 nc_cluster_idx = NULL_CLUSTER;
2334                         }
2335                         ret = f2fs_init_compress_ctx(&cc);
2336                         if (ret)
2337                                 goto set_error_page;
2338
2339                         f2fs_compress_ctx_add_page(&cc, page);
2340
2341                         goto next_page;
2342                 }
2343 read_single_page:
2344 #endif
2345
2346                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2347                                         &bio, &last_block_in_bio, rac);
2348                 if (ret) {
2349 #ifdef CONFIG_F2FS_FS_COMPRESSION
2350 set_error_page:
2351 #endif
2352                         SetPageError(page);
2353                         zero_user_segment(page, 0, PAGE_SIZE);
2354                         unlock_page(page);
2355                 }
2356 #ifdef CONFIG_F2FS_FS_COMPRESSION
2357 next_page:
2358 #endif
2359                 if (rac)
2360                         put_page(page);
2361
2362 #ifdef CONFIG_F2FS_FS_COMPRESSION
2363                 if (f2fs_compressed_file(inode)) {
2364                         /* last page */
2365                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2366                                 ret = f2fs_read_multi_pages(&cc, &bio,
2367                                                         max_nr_pages,
2368                                                         &last_block_in_bio,
2369                                                         rac != NULL, false);
2370                                 f2fs_destroy_compress_ctx(&cc, false);
2371                         }
2372                 }
2373 #endif
2374         }
2375         if (bio)
2376                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2377         return ret;
2378 }
2379
2380 static int f2fs_read_data_page(struct file *file, struct page *page)
2381 {
2382         struct inode *inode = page_file_mapping(page)->host;
2383         int ret = -EAGAIN;
2384
2385         trace_f2fs_readpage(page, DATA);
2386
2387         if (!f2fs_is_compress_backend_ready(inode)) {
2388                 unlock_page(page);
2389                 return -EOPNOTSUPP;
2390         }
2391
2392         /* If the file has inline data, try to read it directly */
2393         if (f2fs_has_inline_data(inode))
2394                 ret = f2fs_read_inline_data(inode, page);
2395         if (ret == -EAGAIN)
2396                 ret = f2fs_mpage_readpages(inode, NULL, page);
2397         return ret;
2398 }
2399
2400 static void f2fs_readahead(struct readahead_control *rac)
2401 {
2402         struct inode *inode = rac->mapping->host;
2403
2404         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2405
2406         if (!f2fs_is_compress_backend_ready(inode))
2407                 return;
2408
2409         /* If the file has inline data, skip readpages */
2410         if (f2fs_has_inline_data(inode))
2411                 return;
2412
2413         f2fs_mpage_readpages(inode, rac, NULL);
2414 }
2415
2416 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2417 {
2418         struct inode *inode = fio->page->mapping->host;
2419         struct page *mpage, *page;
2420         gfp_t gfp_flags = GFP_NOFS;
2421
2422         if (!f2fs_encrypted_file(inode))
2423                 return 0;
2424
2425         page = fio->compressed_page ? fio->compressed_page : fio->page;
2426
2427         /* wait for GCed page writeback via META_MAPPING */
2428         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2429
2430         if (fscrypt_inode_uses_inline_crypto(inode))
2431                 return 0;
2432
2433 retry_encrypt:
2434         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2435                                         PAGE_SIZE, 0, gfp_flags);
2436         if (IS_ERR(fio->encrypted_page)) {
2437                 /* flush pending IOs and wait for a while in the ENOMEM case */
2438                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2439                         f2fs_flush_merged_writes(fio->sbi);
2440                         memalloc_retry_wait(GFP_NOFS);
2441                         gfp_flags |= __GFP_NOFAIL;
2442                         goto retry_encrypt;
2443                 }
2444                 return PTR_ERR(fio->encrypted_page);
2445         }
2446
2447         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2448         if (mpage) {
2449                 if (PageUptodate(mpage))
2450                         memcpy(page_address(mpage),
2451                                 page_address(fio->encrypted_page), PAGE_SIZE);
2452                 f2fs_put_page(mpage, 1);
2453         }
2454         return 0;
2455 }
2456
2457 static inline bool check_inplace_update_policy(struct inode *inode,
2458                                 struct f2fs_io_info *fio)
2459 {
2460         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2461         unsigned int policy = SM_I(sbi)->ipu_policy;
2462
2463         if (policy & (0x1 << F2FS_IPU_FORCE))
2464                 return true;
2465         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2466                 return true;
2467         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2468                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2469                 return true;
2470         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2471                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2472                 return true;
2473
2474         /*
2475          * IPU for rewrite async pages
2476          */
2477         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2478                         fio && fio->op == REQ_OP_WRITE &&
2479                         !(fio->op_flags & REQ_SYNC) &&
2480                         !IS_ENCRYPTED(inode))
2481                 return true;
2482
2483         /* this is only set during fdatasync */
2484         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2485                         is_inode_flag_set(inode, FI_NEED_IPU))
2486                 return true;
2487
2488         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2489                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2490                 return true;
2491
2492         return false;
2493 }
2494
2495 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2496 {
2497         /* swap file is migrating in aligned write mode */
2498         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2499                 return false;
2500
2501         if (f2fs_is_pinned_file(inode))
2502                 return true;
2503
2504         /* if this is cold file, we should overwrite to avoid fragmentation */
2505         if (file_is_cold(inode))
2506                 return true;
2507
2508         return check_inplace_update_policy(inode, fio);
2509 }
2510
2511 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2512 {
2513         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2514
2515         /* The below cases were checked when setting it. */
2516         if (f2fs_is_pinned_file(inode))
2517                 return false;
2518         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2519                 return true;
2520         if (f2fs_lfs_mode(sbi))
2521                 return true;
2522         if (S_ISDIR(inode->i_mode))
2523                 return true;
2524         if (IS_NOQUOTA(inode))
2525                 return true;
2526         if (f2fs_is_atomic_file(inode))
2527                 return true;
2528
2529         /* swap file is migrating in aligned write mode */
2530         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2531                 return true;
2532
2533         if (fio) {
2534                 if (page_private_gcing(fio->page))
2535                         return true;
2536                 if (page_private_dummy(fio->page))
2537                         return true;
2538                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2539                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2540                         return true;
2541         }
2542         return false;
2543 }
2544
2545 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2546 {
2547         struct inode *inode = fio->page->mapping->host;
2548
2549         if (f2fs_should_update_outplace(inode, fio))
2550                 return false;
2551
2552         return f2fs_should_update_inplace(inode, fio);
2553 }
2554
2555 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2556 {
2557         struct page *page = fio->page;
2558         struct inode *inode = page->mapping->host;
2559         struct dnode_of_data dn;
2560         struct extent_info ei = {0, };
2561         struct node_info ni;
2562         bool ipu_force = false;
2563         int err = 0;
2564
2565         set_new_dnode(&dn, inode, NULL, NULL, 0);
2566         if (need_inplace_update(fio) &&
2567                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2568                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2569
2570                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2571                                                 DATA_GENERIC_ENHANCE))
2572                         return -EFSCORRUPTED;
2573
2574                 ipu_force = true;
2575                 fio->need_lock = LOCK_DONE;
2576                 goto got_it;
2577         }
2578
2579         /* Deadlock due to between page->lock and f2fs_lock_op */
2580         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2581                 return -EAGAIN;
2582
2583         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2584         if (err)
2585                 goto out;
2586
2587         fio->old_blkaddr = dn.data_blkaddr;
2588
2589         /* This page is already truncated */
2590         if (fio->old_blkaddr == NULL_ADDR) {
2591                 ClearPageUptodate(page);
2592                 clear_page_private_gcing(page);
2593                 goto out_writepage;
2594         }
2595 got_it:
2596         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2597                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2598                                                 DATA_GENERIC_ENHANCE)) {
2599                 err = -EFSCORRUPTED;
2600                 goto out_writepage;
2601         }
2602         /*
2603          * If current allocation needs SSR,
2604          * it had better in-place writes for updated data.
2605          */
2606         if (ipu_force ||
2607                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2608                                         need_inplace_update(fio))) {
2609                 err = f2fs_encrypt_one_page(fio);
2610                 if (err)
2611                         goto out_writepage;
2612
2613                 set_page_writeback(page);
2614                 ClearPageError(page);
2615                 f2fs_put_dnode(&dn);
2616                 if (fio->need_lock == LOCK_REQ)
2617                         f2fs_unlock_op(fio->sbi);
2618                 err = f2fs_inplace_write_data(fio);
2619                 if (err) {
2620                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2621                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2622                         if (PageWriteback(page))
2623                                 end_page_writeback(page);
2624                 } else {
2625                         set_inode_flag(inode, FI_UPDATE_WRITE);
2626                 }
2627                 trace_f2fs_do_write_data_page(fio->page, IPU);
2628                 return err;
2629         }
2630
2631         if (fio->need_lock == LOCK_RETRY) {
2632                 if (!f2fs_trylock_op(fio->sbi)) {
2633                         err = -EAGAIN;
2634                         goto out_writepage;
2635                 }
2636                 fio->need_lock = LOCK_REQ;
2637         }
2638
2639         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2640         if (err)
2641                 goto out_writepage;
2642
2643         fio->version = ni.version;
2644
2645         err = f2fs_encrypt_one_page(fio);
2646         if (err)
2647                 goto out_writepage;
2648
2649         set_page_writeback(page);
2650         ClearPageError(page);
2651
2652         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2653                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2654
2655         /* LFS mode write path */
2656         f2fs_outplace_write_data(&dn, fio);
2657         trace_f2fs_do_write_data_page(page, OPU);
2658         set_inode_flag(inode, FI_APPEND_WRITE);
2659         if (page->index == 0)
2660                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2661 out_writepage:
2662         f2fs_put_dnode(&dn);
2663 out:
2664         if (fio->need_lock == LOCK_REQ)
2665                 f2fs_unlock_op(fio->sbi);
2666         return err;
2667 }
2668
2669 int f2fs_write_single_data_page(struct page *page, int *submitted,
2670                                 struct bio **bio,
2671                                 sector_t *last_block,
2672                                 struct writeback_control *wbc,
2673                                 enum iostat_type io_type,
2674                                 int compr_blocks,
2675                                 bool allow_balance)
2676 {
2677         struct inode *inode = page->mapping->host;
2678         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2679         loff_t i_size = i_size_read(inode);
2680         const pgoff_t end_index = ((unsigned long long)i_size)
2681                                                         >> PAGE_SHIFT;
2682         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2683         unsigned offset = 0;
2684         bool need_balance_fs = false;
2685         int err = 0;
2686         struct f2fs_io_info fio = {
2687                 .sbi = sbi,
2688                 .ino = inode->i_ino,
2689                 .type = DATA,
2690                 .op = REQ_OP_WRITE,
2691                 .op_flags = wbc_to_write_flags(wbc),
2692                 .old_blkaddr = NULL_ADDR,
2693                 .page = page,
2694                 .encrypted_page = NULL,
2695                 .submitted = false,
2696                 .compr_blocks = compr_blocks,
2697                 .need_lock = LOCK_RETRY,
2698                 .io_type = io_type,
2699                 .io_wbc = wbc,
2700                 .bio = bio,
2701                 .last_block = last_block,
2702         };
2703
2704         trace_f2fs_writepage(page, DATA);
2705
2706         /* we should bypass data pages to proceed the kworkder jobs */
2707         if (unlikely(f2fs_cp_error(sbi))) {
2708                 mapping_set_error(page->mapping, -EIO);
2709                 /*
2710                  * don't drop any dirty dentry pages for keeping lastest
2711                  * directory structure.
2712                  */
2713                 if (S_ISDIR(inode->i_mode))
2714                         goto redirty_out;
2715                 goto out;
2716         }
2717
2718         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2719                 goto redirty_out;
2720
2721         if (page->index < end_index ||
2722                         f2fs_verity_in_progress(inode) ||
2723                         compr_blocks)
2724                 goto write;
2725
2726         /*
2727          * If the offset is out-of-range of file size,
2728          * this page does not have to be written to disk.
2729          */
2730         offset = i_size & (PAGE_SIZE - 1);
2731         if ((page->index >= end_index + 1) || !offset)
2732                 goto out;
2733
2734         zero_user_segment(page, offset, PAGE_SIZE);
2735 write:
2736         if (f2fs_is_drop_cache(inode))
2737                 goto out;
2738         /* we should not write 0'th page having journal header */
2739         if (f2fs_is_volatile_file(inode) && (!page->index ||
2740                         (!wbc->for_reclaim &&
2741                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2742                 goto redirty_out;
2743
2744         /* Dentry/quota blocks are controlled by checkpoint */
2745         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2746                 /*
2747                  * We need to wait for node_write to avoid block allocation during
2748                  * checkpoint. This can only happen to quota writes which can cause
2749                  * the below discard race condition.
2750                  */
2751                 if (IS_NOQUOTA(inode))
2752                         down_read(&sbi->node_write);
2753
2754                 fio.need_lock = LOCK_DONE;
2755                 err = f2fs_do_write_data_page(&fio);
2756
2757                 if (IS_NOQUOTA(inode))
2758                         up_read(&sbi->node_write);
2759
2760                 goto done;
2761         }
2762
2763         if (!wbc->for_reclaim)
2764                 need_balance_fs = true;
2765         else if (has_not_enough_free_secs(sbi, 0, 0))
2766                 goto redirty_out;
2767         else
2768                 set_inode_flag(inode, FI_HOT_DATA);
2769
2770         err = -EAGAIN;
2771         if (f2fs_has_inline_data(inode)) {
2772                 err = f2fs_write_inline_data(inode, page);
2773                 if (!err)
2774                         goto out;
2775         }
2776
2777         if (err == -EAGAIN) {
2778                 err = f2fs_do_write_data_page(&fio);
2779                 if (err == -EAGAIN) {
2780                         fio.need_lock = LOCK_REQ;
2781                         err = f2fs_do_write_data_page(&fio);
2782                 }
2783         }
2784
2785         if (err) {
2786                 file_set_keep_isize(inode);
2787         } else {
2788                 spin_lock(&F2FS_I(inode)->i_size_lock);
2789                 if (F2FS_I(inode)->last_disk_size < psize)
2790                         F2FS_I(inode)->last_disk_size = psize;
2791                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2792         }
2793
2794 done:
2795         if (err && err != -ENOENT)
2796                 goto redirty_out;
2797
2798 out:
2799         inode_dec_dirty_pages(inode);
2800         if (err) {
2801                 ClearPageUptodate(page);
2802                 clear_page_private_gcing(page);
2803         }
2804
2805         if (wbc->for_reclaim) {
2806                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2807                 clear_inode_flag(inode, FI_HOT_DATA);
2808                 f2fs_remove_dirty_inode(inode);
2809                 submitted = NULL;
2810         }
2811         unlock_page(page);
2812         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2813                         !F2FS_I(inode)->cp_task && allow_balance)
2814                 f2fs_balance_fs(sbi, need_balance_fs);
2815
2816         if (unlikely(f2fs_cp_error(sbi))) {
2817                 f2fs_submit_merged_write(sbi, DATA);
2818                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2819                 submitted = NULL;
2820         }
2821
2822         if (submitted)
2823                 *submitted = fio.submitted ? 1 : 0;
2824
2825         return 0;
2826
2827 redirty_out:
2828         redirty_page_for_writepage(wbc, page);
2829         /*
2830          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2831          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2832          * file_write_and_wait_range() will see EIO error, which is critical
2833          * to return value of fsync() followed by atomic_write failure to user.
2834          */
2835         if (!err || wbc->for_reclaim)
2836                 return AOP_WRITEPAGE_ACTIVATE;
2837         unlock_page(page);
2838         return err;
2839 }
2840
2841 static int f2fs_write_data_page(struct page *page,
2842                                         struct writeback_control *wbc)
2843 {
2844 #ifdef CONFIG_F2FS_FS_COMPRESSION
2845         struct inode *inode = page->mapping->host;
2846
2847         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2848                 goto out;
2849
2850         if (f2fs_compressed_file(inode)) {
2851                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2852                         redirty_page_for_writepage(wbc, page);
2853                         return AOP_WRITEPAGE_ACTIVATE;
2854                 }
2855         }
2856 out:
2857 #endif
2858
2859         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2860                                                 wbc, FS_DATA_IO, 0, true);
2861 }
2862
2863 /*
2864  * This function was copied from write_cche_pages from mm/page-writeback.c.
2865  * The major change is making write step of cold data page separately from
2866  * warm/hot data page.
2867  */
2868 static int f2fs_write_cache_pages(struct address_space *mapping,
2869                                         struct writeback_control *wbc,
2870                                         enum iostat_type io_type)
2871 {
2872         int ret = 0;
2873         int done = 0, retry = 0;
2874         struct pagevec pvec;
2875         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2876         struct bio *bio = NULL;
2877         sector_t last_block;
2878 #ifdef CONFIG_F2FS_FS_COMPRESSION
2879         struct inode *inode = mapping->host;
2880         struct compress_ctx cc = {
2881                 .inode = inode,
2882                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2883                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2884                 .cluster_idx = NULL_CLUSTER,
2885                 .rpages = NULL,
2886                 .nr_rpages = 0,
2887                 .cpages = NULL,
2888                 .valid_nr_cpages = 0,
2889                 .rbuf = NULL,
2890                 .cbuf = NULL,
2891                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2892                 .private = NULL,
2893         };
2894 #endif
2895         int nr_pages;
2896         pgoff_t index;
2897         pgoff_t end;            /* Inclusive */
2898         pgoff_t done_index;
2899         int range_whole = 0;
2900         xa_mark_t tag;
2901         int nwritten = 0;
2902         int submitted = 0;
2903         int i;
2904
2905         pagevec_init(&pvec);
2906
2907         if (get_dirty_pages(mapping->host) <=
2908                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2909                 set_inode_flag(mapping->host, FI_HOT_DATA);
2910         else
2911                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2912
2913         if (wbc->range_cyclic) {
2914                 index = mapping->writeback_index; /* prev offset */
2915                 end = -1;
2916         } else {
2917                 index = wbc->range_start >> PAGE_SHIFT;
2918                 end = wbc->range_end >> PAGE_SHIFT;
2919                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2920                         range_whole = 1;
2921         }
2922         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2923                 tag = PAGECACHE_TAG_TOWRITE;
2924         else
2925                 tag = PAGECACHE_TAG_DIRTY;
2926 retry:
2927         retry = 0;
2928         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2929                 tag_pages_for_writeback(mapping, index, end);
2930         done_index = index;
2931         while (!done && !retry && (index <= end)) {
2932                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2933                                 tag);
2934                 if (nr_pages == 0)
2935                         break;
2936
2937                 for (i = 0; i < nr_pages; i++) {
2938                         struct page *page = pvec.pages[i];
2939                         bool need_readd;
2940 readd:
2941                         need_readd = false;
2942 #ifdef CONFIG_F2FS_FS_COMPRESSION
2943                         if (f2fs_compressed_file(inode)) {
2944                                 void *fsdata = NULL;
2945                                 struct page *pagep;
2946                                 int ret2;
2947
2948                                 ret = f2fs_init_compress_ctx(&cc);
2949                                 if (ret) {
2950                                         done = 1;
2951                                         break;
2952                                 }
2953
2954                                 if (!f2fs_cluster_can_merge_page(&cc,
2955                                                                 page->index)) {
2956                                         ret = f2fs_write_multi_pages(&cc,
2957                                                 &submitted, wbc, io_type);
2958                                         if (!ret)
2959                                                 need_readd = true;
2960                                         goto result;
2961                                 }
2962
2963                                 if (unlikely(f2fs_cp_error(sbi)))
2964                                         goto lock_page;
2965
2966                                 if (!f2fs_cluster_is_empty(&cc))
2967                                         goto lock_page;
2968
2969                                 ret2 = f2fs_prepare_compress_overwrite(
2970                                                         inode, &pagep,
2971                                                         page->index, &fsdata);
2972                                 if (ret2 < 0) {
2973                                         ret = ret2;
2974                                         done = 1;
2975                                         break;
2976                                 } else if (ret2 &&
2977                                         (!f2fs_compress_write_end(inode,
2978                                                 fsdata, page->index, 1) ||
2979                                          !f2fs_all_cluster_page_loaded(&cc,
2980                                                 &pvec, i, nr_pages))) {
2981                                         retry = 1;
2982                                         break;
2983                                 }
2984                         }
2985 #endif
2986                         /* give a priority to WB_SYNC threads */
2987                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2988                                         wbc->sync_mode == WB_SYNC_NONE) {
2989                                 done = 1;
2990                                 break;
2991                         }
2992 #ifdef CONFIG_F2FS_FS_COMPRESSION
2993 lock_page:
2994 #endif
2995                         done_index = page->index;
2996 retry_write:
2997                         lock_page(page);
2998
2999                         if (unlikely(page->mapping != mapping)) {
3000 continue_unlock:
3001                                 unlock_page(page);
3002                                 continue;
3003                         }
3004
3005                         if (!PageDirty(page)) {
3006                                 /* someone wrote it for us */
3007                                 goto continue_unlock;
3008                         }
3009
3010                         if (PageWriteback(page)) {
3011                                 if (wbc->sync_mode != WB_SYNC_NONE)
3012                                         f2fs_wait_on_page_writeback(page,
3013                                                         DATA, true, true);
3014                                 else
3015                                         goto continue_unlock;
3016                         }
3017
3018                         if (!clear_page_dirty_for_io(page))
3019                                 goto continue_unlock;
3020
3021 #ifdef CONFIG_F2FS_FS_COMPRESSION
3022                         if (f2fs_compressed_file(inode)) {
3023                                 get_page(page);
3024                                 f2fs_compress_ctx_add_page(&cc, page);
3025                                 continue;
3026                         }
3027 #endif
3028                         ret = f2fs_write_single_data_page(page, &submitted,
3029                                         &bio, &last_block, wbc, io_type,
3030                                         0, true);
3031                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3032                                 unlock_page(page);
3033 #ifdef CONFIG_F2FS_FS_COMPRESSION
3034 result:
3035 #endif
3036                         nwritten += submitted;
3037                         wbc->nr_to_write -= submitted;
3038
3039                         if (unlikely(ret)) {
3040                                 /*
3041                                  * keep nr_to_write, since vfs uses this to
3042                                  * get # of written pages.
3043                                  */
3044                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3045                                         ret = 0;
3046                                         goto next;
3047                                 } else if (ret == -EAGAIN) {
3048                                         ret = 0;
3049                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3050                                                 cond_resched();
3051                                                 congestion_wait(BLK_RW_ASYNC,
3052                                                         DEFAULT_IO_TIMEOUT);
3053                                                 goto retry_write;
3054                                         }
3055                                         goto next;
3056                                 }
3057                                 done_index = page->index + 1;
3058                                 done = 1;
3059                                 break;
3060                         }
3061
3062                         if (wbc->nr_to_write <= 0 &&
3063                                         wbc->sync_mode == WB_SYNC_NONE) {
3064                                 done = 1;
3065                                 break;
3066                         }
3067 next:
3068                         if (need_readd)
3069                                 goto readd;
3070                 }
3071                 pagevec_release(&pvec);
3072                 cond_resched();
3073         }
3074 #ifdef CONFIG_F2FS_FS_COMPRESSION
3075         /* flush remained pages in compress cluster */
3076         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3077                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3078                 nwritten += submitted;
3079                 wbc->nr_to_write -= submitted;
3080                 if (ret) {
3081                         done = 1;
3082                         retry = 0;
3083                 }
3084         }
3085         if (f2fs_compressed_file(inode))
3086                 f2fs_destroy_compress_ctx(&cc, false);
3087 #endif
3088         if (retry) {
3089                 index = 0;
3090                 end = -1;
3091                 goto retry;
3092         }
3093         if (wbc->range_cyclic && !done)
3094                 done_index = 0;
3095         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3096                 mapping->writeback_index = done_index;
3097
3098         if (nwritten)
3099                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3100                                                                 NULL, 0, DATA);
3101         /* submit cached bio of IPU write */
3102         if (bio)
3103                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3104
3105         return ret;
3106 }
3107
3108 static inline bool __should_serialize_io(struct inode *inode,
3109                                         struct writeback_control *wbc)
3110 {
3111         /* to avoid deadlock in path of data flush */
3112         if (F2FS_I(inode)->cp_task)
3113                 return false;
3114
3115         if (!S_ISREG(inode->i_mode))
3116                 return false;
3117         if (IS_NOQUOTA(inode))
3118                 return false;
3119
3120         if (f2fs_need_compress_data(inode))
3121                 return true;
3122         if (wbc->sync_mode != WB_SYNC_ALL)
3123                 return true;
3124         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3125                 return true;
3126         return false;
3127 }
3128
3129 static int __f2fs_write_data_pages(struct address_space *mapping,
3130                                                 struct writeback_control *wbc,
3131                                                 enum iostat_type io_type)
3132 {
3133         struct inode *inode = mapping->host;
3134         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3135         struct blk_plug plug;
3136         int ret;
3137         bool locked = false;
3138
3139         /* deal with chardevs and other special file */
3140         if (!mapping->a_ops->writepage)
3141                 return 0;
3142
3143         /* skip writing if there is no dirty page in this inode */
3144         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3145                 return 0;
3146
3147         /* during POR, we don't need to trigger writepage at all. */
3148         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3149                 goto skip_write;
3150
3151         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3152                         wbc->sync_mode == WB_SYNC_NONE &&
3153                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3154                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3155                 goto skip_write;
3156
3157         /* skip writing during file defragment */
3158         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3159                 goto skip_write;
3160
3161         trace_f2fs_writepages(mapping->host, wbc, DATA);
3162
3163         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3164         if (wbc->sync_mode == WB_SYNC_ALL)
3165                 atomic_inc(&sbi->wb_sync_req[DATA]);
3166         else if (atomic_read(&sbi->wb_sync_req[DATA]))
3167                 goto skip_write;
3168
3169         if (__should_serialize_io(inode, wbc)) {
3170                 mutex_lock(&sbi->writepages);
3171                 locked = true;
3172         }
3173
3174         blk_start_plug(&plug);
3175         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3176         blk_finish_plug(&plug);
3177
3178         if (locked)
3179                 mutex_unlock(&sbi->writepages);
3180
3181         if (wbc->sync_mode == WB_SYNC_ALL)
3182                 atomic_dec(&sbi->wb_sync_req[DATA]);
3183         /*
3184          * if some pages were truncated, we cannot guarantee its mapping->host
3185          * to detect pending bios.
3186          */
3187
3188         f2fs_remove_dirty_inode(inode);
3189         return ret;
3190
3191 skip_write:
3192         wbc->pages_skipped += get_dirty_pages(inode);
3193         trace_f2fs_writepages(mapping->host, wbc, DATA);
3194         return 0;
3195 }
3196
3197 static int f2fs_write_data_pages(struct address_space *mapping,
3198                             struct writeback_control *wbc)
3199 {
3200         struct inode *inode = mapping->host;
3201
3202         return __f2fs_write_data_pages(mapping, wbc,
3203                         F2FS_I(inode)->cp_task == current ?
3204                         FS_CP_DATA_IO : FS_DATA_IO);
3205 }
3206
3207 void f2fs_write_failed(struct inode *inode, loff_t to)
3208 {
3209         loff_t i_size = i_size_read(inode);
3210
3211         if (IS_NOQUOTA(inode))
3212                 return;
3213
3214         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3215         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3216                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3217                 filemap_invalidate_lock(inode->i_mapping);
3218
3219                 truncate_pagecache(inode, i_size);
3220                 f2fs_truncate_blocks(inode, i_size, true);
3221
3222                 filemap_invalidate_unlock(inode->i_mapping);
3223                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3224         }
3225 }
3226
3227 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3228                         struct page *page, loff_t pos, unsigned len,
3229                         block_t *blk_addr, bool *node_changed)
3230 {
3231         struct inode *inode = page->mapping->host;
3232         pgoff_t index = page->index;
3233         struct dnode_of_data dn;
3234         struct page *ipage;
3235         bool locked = false;
3236         struct extent_info ei = {0, };
3237         int err = 0;
3238         int flag;
3239
3240         /*
3241          * If a whole page is being written and we already preallocated all the
3242          * blocks, then there is no need to get a block address now.
3243          */
3244         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3245                 return 0;
3246
3247         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3248         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3249                 flag = F2FS_GET_BLOCK_DEFAULT;
3250         else
3251                 flag = F2FS_GET_BLOCK_PRE_AIO;
3252
3253         if (f2fs_has_inline_data(inode) ||
3254                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3255                 f2fs_do_map_lock(sbi, flag, true);
3256                 locked = true;
3257         }
3258
3259 restart:
3260         /* check inline_data */
3261         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3262         if (IS_ERR(ipage)) {
3263                 err = PTR_ERR(ipage);
3264                 goto unlock_out;
3265         }
3266
3267         set_new_dnode(&dn, inode, ipage, ipage, 0);
3268
3269         if (f2fs_has_inline_data(inode)) {
3270                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3271                         f2fs_do_read_inline_data(page, ipage);
3272                         set_inode_flag(inode, FI_DATA_EXIST);
3273                         if (inode->i_nlink)
3274                                 set_page_private_inline(ipage);
3275                 } else {
3276                         err = f2fs_convert_inline_page(&dn, page);
3277                         if (err)
3278                                 goto out;
3279                         if (dn.data_blkaddr == NULL_ADDR)
3280                                 err = f2fs_get_block(&dn, index);
3281                 }
3282         } else if (locked) {
3283                 err = f2fs_get_block(&dn, index);
3284         } else {
3285                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3286                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3287                 } else {
3288                         /* hole case */
3289                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3290                         if (err || dn.data_blkaddr == NULL_ADDR) {
3291                                 f2fs_put_dnode(&dn);
3292                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3293                                                                 true);
3294                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3295                                 locked = true;
3296                                 goto restart;
3297                         }
3298                 }
3299         }
3300
3301         /* convert_inline_page can make node_changed */
3302         *blk_addr = dn.data_blkaddr;
3303         *node_changed = dn.node_changed;
3304 out:
3305         f2fs_put_dnode(&dn);
3306 unlock_out:
3307         if (locked)
3308                 f2fs_do_map_lock(sbi, flag, false);
3309         return err;
3310 }
3311
3312 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3313                 loff_t pos, unsigned len, unsigned flags,
3314                 struct page **pagep, void **fsdata)
3315 {
3316         struct inode *inode = mapping->host;
3317         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3318         struct page *page = NULL;
3319         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3320         bool need_balance = false, drop_atomic = false;
3321         block_t blkaddr = NULL_ADDR;
3322         int err = 0;
3323
3324         trace_f2fs_write_begin(inode, pos, len, flags);
3325
3326         if (!f2fs_is_checkpoint_ready(sbi)) {
3327                 err = -ENOSPC;
3328                 goto fail;
3329         }
3330
3331         if ((f2fs_is_atomic_file(inode) &&
3332                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3333                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3334                 err = -ENOMEM;
3335                 drop_atomic = true;
3336                 goto fail;
3337         }
3338
3339         /*
3340          * We should check this at this moment to avoid deadlock on inode page
3341          * and #0 page. The locking rule for inline_data conversion should be:
3342          * lock_page(page #0) -> lock_page(inode_page)
3343          */
3344         if (index != 0) {
3345                 err = f2fs_convert_inline_inode(inode);
3346                 if (err)
3347                         goto fail;
3348         }
3349
3350 #ifdef CONFIG_F2FS_FS_COMPRESSION
3351         if (f2fs_compressed_file(inode)) {
3352                 int ret;
3353
3354                 *fsdata = NULL;
3355
3356                 if (len == PAGE_SIZE)
3357                         goto repeat;
3358
3359                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3360                                                         index, fsdata);
3361                 if (ret < 0) {
3362                         err = ret;
3363                         goto fail;
3364                 } else if (ret) {
3365                         return 0;
3366                 }
3367         }
3368 #endif
3369
3370 repeat:
3371         /*
3372          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3373          * wait_for_stable_page. Will wait that below with our IO control.
3374          */
3375         page = f2fs_pagecache_get_page(mapping, index,
3376                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3377         if (!page) {
3378                 err = -ENOMEM;
3379                 goto fail;
3380         }
3381
3382         /* TODO: cluster can be compressed due to race with .writepage */
3383
3384         *pagep = page;
3385
3386         err = prepare_write_begin(sbi, page, pos, len,
3387                                         &blkaddr, &need_balance);
3388         if (err)
3389                 goto fail;
3390
3391         if (need_balance && !IS_NOQUOTA(inode) &&
3392                         has_not_enough_free_secs(sbi, 0, 0)) {
3393                 unlock_page(page);
3394                 f2fs_balance_fs(sbi, true);
3395                 lock_page(page);
3396                 if (page->mapping != mapping) {
3397                         /* The page got truncated from under us */
3398                         f2fs_put_page(page, 1);
3399                         goto repeat;
3400                 }
3401         }
3402
3403         f2fs_wait_on_page_writeback(page, DATA, false, true);
3404
3405         if (len == PAGE_SIZE || PageUptodate(page))
3406                 return 0;
3407
3408         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3409             !f2fs_verity_in_progress(inode)) {
3410                 zero_user_segment(page, len, PAGE_SIZE);
3411                 return 0;
3412         }
3413
3414         if (blkaddr == NEW_ADDR) {
3415                 zero_user_segment(page, 0, PAGE_SIZE);
3416                 SetPageUptodate(page);
3417         } else {
3418                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3419                                 DATA_GENERIC_ENHANCE_READ)) {
3420                         err = -EFSCORRUPTED;
3421                         goto fail;
3422                 }
3423                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3424                 if (err)
3425                         goto fail;
3426
3427                 lock_page(page);
3428                 if (unlikely(page->mapping != mapping)) {
3429                         f2fs_put_page(page, 1);
3430                         goto repeat;
3431                 }
3432                 if (unlikely(!PageUptodate(page))) {
3433                         err = -EIO;
3434                         goto fail;
3435                 }
3436         }
3437         return 0;
3438
3439 fail:
3440         f2fs_put_page(page, 1);
3441         f2fs_write_failed(inode, pos + len);
3442         if (drop_atomic)
3443                 f2fs_drop_inmem_pages_all(sbi, false);
3444         return err;
3445 }
3446
3447 static int f2fs_write_end(struct file *file,
3448                         struct address_space *mapping,
3449                         loff_t pos, unsigned len, unsigned copied,
3450                         struct page *page, void *fsdata)
3451 {
3452         struct inode *inode = page->mapping->host;
3453
3454         trace_f2fs_write_end(inode, pos, len, copied);
3455
3456         /*
3457          * This should be come from len == PAGE_SIZE, and we expect copied
3458          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3459          * let generic_perform_write() try to copy data again through copied=0.
3460          */
3461         if (!PageUptodate(page)) {
3462                 if (unlikely(copied != len))
3463                         copied = 0;
3464                 else
3465                         SetPageUptodate(page);
3466         }
3467
3468 #ifdef CONFIG_F2FS_FS_COMPRESSION
3469         /* overwrite compressed file */
3470         if (f2fs_compressed_file(inode) && fsdata) {
3471                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3472                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3473
3474                 if (pos + copied > i_size_read(inode) &&
3475                                 !f2fs_verity_in_progress(inode))
3476                         f2fs_i_size_write(inode, pos + copied);
3477                 return copied;
3478         }
3479 #endif
3480
3481         if (!copied)
3482                 goto unlock_out;
3483
3484         set_page_dirty(page);
3485
3486         if (pos + copied > i_size_read(inode) &&
3487             !f2fs_verity_in_progress(inode))
3488                 f2fs_i_size_write(inode, pos + copied);
3489 unlock_out:
3490         f2fs_put_page(page, 1);
3491         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3492         return copied;
3493 }
3494
3495 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3496                                                         unsigned int length)
3497 {
3498         struct inode *inode = page->mapping->host;
3499         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3500
3501         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3502                 (offset % PAGE_SIZE || length != PAGE_SIZE))
3503                 return;
3504
3505         if (PageDirty(page)) {
3506                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3507                         dec_page_count(sbi, F2FS_DIRTY_META);
3508                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3509                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3510                 } else {
3511                         inode_dec_dirty_pages(inode);
3512                         f2fs_remove_dirty_inode(inode);
3513                 }
3514         }
3515
3516         clear_page_private_gcing(page);
3517
3518         if (test_opt(sbi, COMPRESS_CACHE) &&
3519                         inode->i_ino == F2FS_COMPRESS_INO(sbi))
3520                 clear_page_private_data(page);
3521
3522         if (page_private_atomic(page))
3523                 return f2fs_drop_inmem_page(inode, page);
3524
3525         detach_page_private(page);
3526         set_page_private(page, 0);
3527 }
3528
3529 int f2fs_release_page(struct page *page, gfp_t wait)
3530 {
3531         /* If this is dirty page, keep PagePrivate */
3532         if (PageDirty(page))
3533                 return 0;
3534
3535         /* This is atomic written page, keep Private */
3536         if (page_private_atomic(page))
3537                 return 0;
3538
3539         if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3540                 struct inode *inode = page->mapping->host;
3541
3542                 if (inode->i_ino == F2FS_COMPRESS_INO(F2FS_I_SB(inode)))
3543                         clear_page_private_data(page);
3544         }
3545
3546         clear_page_private_gcing(page);
3547
3548         detach_page_private(page);
3549         set_page_private(page, 0);
3550         return 1;
3551 }
3552
3553 static int f2fs_set_data_page_dirty(struct page *page)
3554 {
3555         struct inode *inode = page_file_mapping(page)->host;
3556
3557         trace_f2fs_set_page_dirty(page, DATA);
3558
3559         if (!PageUptodate(page))
3560                 SetPageUptodate(page);
3561         if (PageSwapCache(page))
3562                 return __set_page_dirty_nobuffers(page);
3563
3564         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3565                 if (!page_private_atomic(page)) {
3566                         f2fs_register_inmem_page(inode, page);
3567                         return 1;
3568                 }
3569                 /*
3570                  * Previously, this page has been registered, we just
3571                  * return here.
3572                  */
3573                 return 0;
3574         }
3575
3576         if (!PageDirty(page)) {
3577                 __set_page_dirty_nobuffers(page);
3578                 f2fs_update_dirty_page(inode, page);
3579                 return 1;
3580         }
3581         return 0;
3582 }
3583
3584
3585 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3586 {
3587 #ifdef CONFIG_F2FS_FS_COMPRESSION
3588         struct dnode_of_data dn;
3589         sector_t start_idx, blknr = 0;
3590         int ret;
3591
3592         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3593
3594         set_new_dnode(&dn, inode, NULL, NULL, 0);
3595         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3596         if (ret)
3597                 return 0;
3598
3599         if (dn.data_blkaddr != COMPRESS_ADDR) {
3600                 dn.ofs_in_node += block - start_idx;
3601                 blknr = f2fs_data_blkaddr(&dn);
3602                 if (!__is_valid_data_blkaddr(blknr))
3603                         blknr = 0;
3604         }
3605
3606         f2fs_put_dnode(&dn);
3607         return blknr;
3608 #else
3609         return 0;
3610 #endif
3611 }
3612
3613
3614 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3615 {
3616         struct inode *inode = mapping->host;
3617         sector_t blknr = 0;
3618
3619         if (f2fs_has_inline_data(inode))
3620                 goto out;
3621
3622         /* make sure allocating whole blocks */
3623         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3624                 filemap_write_and_wait(mapping);
3625
3626         /* Block number less than F2FS MAX BLOCKS */
3627         if (unlikely(block >= max_file_blocks(inode)))
3628                 goto out;
3629
3630         if (f2fs_compressed_file(inode)) {
3631                 blknr = f2fs_bmap_compress(inode, block);
3632         } else {
3633                 struct f2fs_map_blocks map;
3634
3635                 memset(&map, 0, sizeof(map));
3636                 map.m_lblk = block;
3637                 map.m_len = 1;
3638                 map.m_next_pgofs = NULL;
3639                 map.m_seg_type = NO_CHECK_TYPE;
3640
3641                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3642                         blknr = map.m_pblk;
3643         }
3644 out:
3645         trace_f2fs_bmap(inode, block, blknr);
3646         return blknr;
3647 }
3648
3649 #ifdef CONFIG_MIGRATION
3650 #include <linux/migrate.h>
3651
3652 int f2fs_migrate_page(struct address_space *mapping,
3653                 struct page *newpage, struct page *page, enum migrate_mode mode)
3654 {
3655         int rc, extra_count;
3656         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3657         bool atomic_written = page_private_atomic(page);
3658
3659         BUG_ON(PageWriteback(page));
3660
3661         /* migrating an atomic written page is safe with the inmem_lock hold */
3662         if (atomic_written) {
3663                 if (mode != MIGRATE_SYNC)
3664                         return -EBUSY;
3665                 if (!mutex_trylock(&fi->inmem_lock))
3666                         return -EAGAIN;
3667         }
3668
3669         /* one extra reference was held for atomic_write page */
3670         extra_count = atomic_written ? 1 : 0;
3671         rc = migrate_page_move_mapping(mapping, newpage,
3672                                 page, extra_count);
3673         if (rc != MIGRATEPAGE_SUCCESS) {
3674                 if (atomic_written)
3675                         mutex_unlock(&fi->inmem_lock);
3676                 return rc;
3677         }
3678
3679         if (atomic_written) {
3680                 struct inmem_pages *cur;
3681
3682                 list_for_each_entry(cur, &fi->inmem_pages, list)
3683                         if (cur->page == page) {
3684                                 cur->page = newpage;
3685                                 break;
3686                         }
3687                 mutex_unlock(&fi->inmem_lock);
3688                 put_page(page);
3689                 get_page(newpage);
3690         }
3691
3692         /* guarantee to start from no stale private field */
3693         set_page_private(newpage, 0);
3694         if (PagePrivate(page)) {
3695                 set_page_private(newpage, page_private(page));
3696                 SetPagePrivate(newpage);
3697                 get_page(newpage);
3698
3699                 set_page_private(page, 0);
3700                 ClearPagePrivate(page);
3701                 put_page(page);
3702         }
3703
3704         if (mode != MIGRATE_SYNC_NO_COPY)
3705                 migrate_page_copy(newpage, page);
3706         else
3707                 migrate_page_states(newpage, page);
3708
3709         return MIGRATEPAGE_SUCCESS;
3710 }
3711 #endif
3712
3713 #ifdef CONFIG_SWAP
3714 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3715                                                         unsigned int blkcnt)
3716 {
3717         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3718         unsigned int blkofs;
3719         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3720         unsigned int secidx = start_blk / blk_per_sec;
3721         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3722         int ret = 0;
3723
3724         down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3725         filemap_invalidate_lock(inode->i_mapping);
3726
3727         set_inode_flag(inode, FI_ALIGNED_WRITE);
3728
3729         for (; secidx < end_sec; secidx++) {
3730                 down_write(&sbi->pin_sem);
3731
3732                 f2fs_lock_op(sbi);
3733                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3734                 f2fs_unlock_op(sbi);
3735
3736                 set_inode_flag(inode, FI_DO_DEFRAG);
3737
3738                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3739                         struct page *page;
3740                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3741
3742                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3743                         if (IS_ERR(page)) {
3744                                 up_write(&sbi->pin_sem);
3745                                 ret = PTR_ERR(page);
3746                                 goto done;
3747                         }
3748
3749                         set_page_dirty(page);
3750                         f2fs_put_page(page, 1);
3751                 }
3752
3753                 clear_inode_flag(inode, FI_DO_DEFRAG);
3754
3755                 ret = filemap_fdatawrite(inode->i_mapping);
3756
3757                 up_write(&sbi->pin_sem);
3758
3759                 if (ret)
3760                         break;
3761         }
3762
3763 done:
3764         clear_inode_flag(inode, FI_DO_DEFRAG);
3765         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3766
3767         filemap_invalidate_unlock(inode->i_mapping);
3768         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3769
3770         return ret;
3771 }
3772
3773 static int check_swap_activate(struct swap_info_struct *sis,
3774                                 struct file *swap_file, sector_t *span)
3775 {
3776         struct address_space *mapping = swap_file->f_mapping;
3777         struct inode *inode = mapping->host;
3778         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3779         sector_t cur_lblock;
3780         sector_t last_lblock;
3781         sector_t pblock;
3782         sector_t lowest_pblock = -1;
3783         sector_t highest_pblock = 0;
3784         int nr_extents = 0;
3785         unsigned long nr_pblocks;
3786         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3787         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3788         unsigned int not_aligned = 0;
3789         int ret = 0;
3790
3791         /*
3792          * Map all the blocks into the extent list.  This code doesn't try
3793          * to be very smart.
3794          */
3795         cur_lblock = 0;
3796         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3797
3798         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3799                 struct f2fs_map_blocks map;
3800 retry:
3801                 cond_resched();
3802
3803                 memset(&map, 0, sizeof(map));
3804                 map.m_lblk = cur_lblock;
3805                 map.m_len = last_lblock - cur_lblock;
3806                 map.m_next_pgofs = NULL;
3807                 map.m_next_extent = NULL;
3808                 map.m_seg_type = NO_CHECK_TYPE;
3809                 map.m_may_create = false;
3810
3811                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3812                 if (ret)
3813                         goto out;
3814
3815                 /* hole */
3816                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3817                         f2fs_err(sbi, "Swapfile has holes");
3818                         ret = -EINVAL;
3819                         goto out;
3820                 }
3821
3822                 pblock = map.m_pblk;
3823                 nr_pblocks = map.m_len;
3824
3825                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3826                                 nr_pblocks & sec_blks_mask) {
3827                         not_aligned++;
3828
3829                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3830                         if (cur_lblock + nr_pblocks > sis->max)
3831                                 nr_pblocks -= blks_per_sec;
3832
3833                         if (!nr_pblocks) {
3834                                 /* this extent is last one */
3835                                 nr_pblocks = map.m_len;
3836                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3837                                 goto next;
3838                         }
3839
3840                         ret = f2fs_migrate_blocks(inode, cur_lblock,
3841                                                         nr_pblocks);
3842                         if (ret)
3843                                 goto out;
3844                         goto retry;
3845                 }
3846 next:
3847                 if (cur_lblock + nr_pblocks >= sis->max)
3848                         nr_pblocks = sis->max - cur_lblock;
3849
3850                 if (cur_lblock) {       /* exclude the header page */
3851                         if (pblock < lowest_pblock)
3852                                 lowest_pblock = pblock;
3853                         if (pblock + nr_pblocks - 1 > highest_pblock)
3854                                 highest_pblock = pblock + nr_pblocks - 1;
3855                 }
3856
3857                 /*
3858                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3859                  */
3860                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3861                 if (ret < 0)
3862                         goto out;
3863                 nr_extents += ret;
3864                 cur_lblock += nr_pblocks;
3865         }
3866         ret = nr_extents;
3867         *span = 1 + highest_pblock - lowest_pblock;
3868         if (cur_lblock == 0)
3869                 cur_lblock = 1; /* force Empty message */
3870         sis->max = cur_lblock;
3871         sis->pages = cur_lblock - 1;
3872         sis->highest_bit = cur_lblock - 1;
3873 out:
3874         if (not_aligned)
3875                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3876                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
3877         return ret;
3878 }
3879
3880 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3881                                 sector_t *span)
3882 {
3883         struct inode *inode = file_inode(file);
3884         int ret;
3885
3886         if (!S_ISREG(inode->i_mode))
3887                 return -EINVAL;
3888
3889         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3890                 return -EROFS;
3891
3892         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3893                 f2fs_err(F2FS_I_SB(inode),
3894                         "Swapfile not supported in LFS mode");
3895                 return -EINVAL;
3896         }
3897
3898         ret = f2fs_convert_inline_inode(inode);
3899         if (ret)
3900                 return ret;
3901
3902         if (!f2fs_disable_compressed_file(inode))
3903                 return -EINVAL;
3904
3905         f2fs_precache_extents(inode);
3906
3907         ret = check_swap_activate(sis, file, span);
3908         if (ret < 0)
3909                 return ret;
3910
3911         set_inode_flag(inode, FI_PIN_FILE);
3912         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3913         return ret;
3914 }
3915
3916 static void f2fs_swap_deactivate(struct file *file)
3917 {
3918         struct inode *inode = file_inode(file);
3919
3920         clear_inode_flag(inode, FI_PIN_FILE);
3921 }
3922 #else
3923 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3924                                 sector_t *span)
3925 {
3926         return -EOPNOTSUPP;
3927 }
3928
3929 static void f2fs_swap_deactivate(struct file *file)
3930 {
3931 }
3932 #endif
3933
3934 const struct address_space_operations f2fs_dblock_aops = {
3935         .readpage       = f2fs_read_data_page,
3936         .readahead      = f2fs_readahead,
3937         .writepage      = f2fs_write_data_page,
3938         .writepages     = f2fs_write_data_pages,
3939         .write_begin    = f2fs_write_begin,
3940         .write_end      = f2fs_write_end,
3941         .set_page_dirty = f2fs_set_data_page_dirty,
3942         .invalidatepage = f2fs_invalidate_page,
3943         .releasepage    = f2fs_release_page,
3944         .direct_IO      = noop_direct_IO,
3945         .bmap           = f2fs_bmap,
3946         .swap_activate  = f2fs_swap_activate,
3947         .swap_deactivate = f2fs_swap_deactivate,
3948 #ifdef CONFIG_MIGRATION
3949         .migratepage    = f2fs_migrate_page,
3950 #endif
3951 };
3952
3953 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3954 {
3955         struct address_space *mapping = page_mapping(page);
3956         unsigned long flags;
3957
3958         xa_lock_irqsave(&mapping->i_pages, flags);
3959         __xa_clear_mark(&mapping->i_pages, page_index(page),
3960                                                 PAGECACHE_TAG_DIRTY);
3961         xa_unlock_irqrestore(&mapping->i_pages, flags);
3962 }
3963
3964 int __init f2fs_init_post_read_processing(void)
3965 {
3966         bio_post_read_ctx_cache =
3967                 kmem_cache_create("f2fs_bio_post_read_ctx",
3968                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
3969         if (!bio_post_read_ctx_cache)
3970                 goto fail;
3971         bio_post_read_ctx_pool =
3972                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3973                                          bio_post_read_ctx_cache);
3974         if (!bio_post_read_ctx_pool)
3975                 goto fail_free_cache;
3976         return 0;
3977
3978 fail_free_cache:
3979         kmem_cache_destroy(bio_post_read_ctx_cache);
3980 fail:
3981         return -ENOMEM;
3982 }
3983
3984 void f2fs_destroy_post_read_processing(void)
3985 {
3986         mempool_destroy(bio_post_read_ctx_pool);
3987         kmem_cache_destroy(bio_post_read_ctx_cache);
3988 }
3989
3990 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
3991 {
3992         if (!f2fs_sb_has_encrypt(sbi) &&
3993                 !f2fs_sb_has_verity(sbi) &&
3994                 !f2fs_sb_has_compression(sbi))
3995                 return 0;
3996
3997         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
3998                                                  WQ_UNBOUND | WQ_HIGHPRI,
3999                                                  num_online_cpus());
4000         if (!sbi->post_read_wq)
4001                 return -ENOMEM;
4002         return 0;
4003 }
4004
4005 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4006 {
4007         if (sbi->post_read_wq)
4008                 destroy_workqueue(sbi->post_read_wq);
4009 }
4010
4011 int __init f2fs_init_bio_entry_cache(void)
4012 {
4013         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4014                         sizeof(struct bio_entry));
4015         if (!bio_entry_slab)
4016                 return -ENOMEM;
4017         return 0;
4018 }
4019
4020 void f2fs_destroy_bio_entry_cache(void)
4021 {
4022         kmem_cache_destroy(bio_entry_slab);
4023 }
4024
4025 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4026                             unsigned int flags, struct iomap *iomap,
4027                             struct iomap *srcmap)
4028 {
4029         struct f2fs_map_blocks map = {};
4030         pgoff_t next_pgofs = 0;
4031         int err;
4032
4033         map.m_lblk = bytes_to_blks(inode, offset);
4034         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4035         map.m_next_pgofs = &next_pgofs;
4036         map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4037         if (flags & IOMAP_WRITE)
4038                 map.m_may_create = true;
4039
4040         err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4041                               F2FS_GET_BLOCK_DIO);
4042         if (err)
4043                 return err;
4044
4045         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4046
4047         if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4048                 iomap->length = blks_to_bytes(inode, map.m_len);
4049                 if (map.m_flags & F2FS_MAP_MAPPED) {
4050                         iomap->type = IOMAP_MAPPED;
4051                         iomap->flags |= IOMAP_F_MERGED;
4052                 } else {
4053                         iomap->type = IOMAP_UNWRITTEN;
4054                 }
4055                 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4056                         return -EINVAL;
4057
4058                 iomap->bdev = map.m_bdev;
4059                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4060         } else {
4061                 iomap->length = blks_to_bytes(inode, next_pgofs) -
4062                                 iomap->offset;
4063                 iomap->type = IOMAP_HOLE;
4064                 iomap->addr = IOMAP_NULL_ADDR;
4065         }
4066
4067         if (map.m_flags & F2FS_MAP_NEW)
4068                 iomap->flags |= IOMAP_F_NEW;
4069         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4070             offset + length > i_size_read(inode))
4071                 iomap->flags |= IOMAP_F_DIRTY;
4072
4073         return 0;
4074 }
4075
4076 const struct iomap_ops f2fs_iomap_ops = {
4077         .iomap_begin    = f2fs_iomap_begin,
4078 };