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/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.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/cleancache.h>
  22 #include <linux/sched/signal.h>
  23 #include <linux/fiemap.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);
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
1380         /*
1381          * i_size will be updated by direct_IO. Otherwise, we'll get stale
1382          * data from unwritten block via dio_read.
1383          */
1384         return 0;
1385 }
1386
1387 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1388 {
1389         struct inode *inode = file_inode(iocb->ki_filp);
1390         struct f2fs_map_blocks map;
1391         int flag;
1392         int err = 0;
1393         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1394
1395         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1396         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1397         if (map.m_len > map.m_lblk)
1398                 map.m_len -= map.m_lblk;
1399         else
1400                 map.m_len = 0;
1401
1402         map.m_next_pgofs = NULL;
1403         map.m_next_extent = NULL;
1404         map.m_seg_type = NO_CHECK_TYPE;
1405         map.m_may_create = true;
1406
1407         if (direct_io) {
1408                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1409                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1410                                         F2FS_GET_BLOCK_PRE_AIO :
1411                                         F2FS_GET_BLOCK_PRE_DIO;
1412                 goto map_blocks;
1413         }
1414         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1415                 err = f2fs_convert_inline_inode(inode);
1416                 if (err)
1417                         return err;
1418         }
1419         if (f2fs_has_inline_data(inode))
1420                 return err;
1421
1422         flag = F2FS_GET_BLOCK_PRE_AIO;
1423
1424 map_blocks:
1425         err = f2fs_map_blocks(inode, &map, 1, flag);
1426         if (map.m_len > 0 && err == -ENOSPC) {
1427                 if (!direct_io)
1428                         set_inode_flag(inode, FI_NO_PREALLOC);
1429                 err = 0;
1430         }
1431         return err;
1432 }
1433
1434 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1435 {
1436         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1437                 if (lock)
1438                         down_read(&sbi->node_change);
1439                 else
1440                         up_read(&sbi->node_change);
1441         } else {
1442                 if (lock)
1443                         f2fs_lock_op(sbi);
1444                 else
1445                         f2fs_unlock_op(sbi);
1446         }
1447 }
1448
1449 /*
1450  * f2fs_map_blocks() tries to find or build mapping relationship which
1451  * maps continuous logical blocks to physical blocks, and return such
1452  * info via f2fs_map_blocks structure.
1453  */
1454 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1455                                                 int create, int flag)
1456 {
1457         unsigned int maxblocks = map->m_len;
1458         struct dnode_of_data dn;
1459         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1460         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1461         pgoff_t pgofs, end_offset, end;
1462         int err = 0, ofs = 1;
1463         unsigned int ofs_in_node, last_ofs_in_node;
1464         blkcnt_t prealloc;
1465         struct extent_info ei = {0, };
1466         block_t blkaddr;
1467         unsigned int start_pgofs;
1468         int bidx = 0;
1469
1470         if (!maxblocks)
1471                 return 0;
1472
1473         map->m_bdev = inode->i_sb->s_bdev;
1474         map->m_multidev_dio =
1475                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1476
1477         map->m_len = 0;
1478         map->m_flags = 0;
1479
1480         /* it only supports block size == page size */
1481         pgofs = (pgoff_t)map->m_lblk;
1482         end = pgofs + maxblocks;
1483
1484         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1485                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1486                                                         map->m_may_create)
1487                         goto next_dnode;
1488
1489                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1490                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1491                 map->m_flags = F2FS_MAP_MAPPED;
1492                 if (map->m_next_extent)
1493                         *map->m_next_extent = pgofs + map->m_len;
1494
1495                 /* for hardware encryption, but to avoid potential issue in future */
1496                 if (flag == F2FS_GET_BLOCK_DIO)
1497                         f2fs_wait_on_block_writeback_range(inode,
1498                                                 map->m_pblk, map->m_len);
1499
1500                 if (map->m_multidev_dio) {
1501                         block_t blk_addr = map->m_pblk;
1502
1503                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1504
1505                         map->m_bdev = FDEV(bidx).bdev;
1506                         map->m_pblk -= FDEV(bidx).start_blk;
1507                         map->m_len = min(map->m_len,
1508                                 FDEV(bidx).end_blk + 1 - map->m_pblk);
1509
1510                         if (map->m_may_create)
1511                                 f2fs_update_device_state(sbi, inode->i_ino,
1512                                                         blk_addr, map->m_len);
1513                 }
1514                 goto out;
1515         }
1516
1517 next_dnode:
1518         if (map->m_may_create)
1519                 f2fs_do_map_lock(sbi, flag, true);
1520
1521         /* When reading holes, we need its node page */
1522         set_new_dnode(&dn, inode, NULL, NULL, 0);
1523         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1524         if (err) {
1525                 if (flag == F2FS_GET_BLOCK_BMAP)
1526                         map->m_pblk = 0;
1527
1528                 if (err == -ENOENT) {
1529                         /*
1530                          * There is one exceptional case that read_node_page()
1531                          * may return -ENOENT due to filesystem has been
1532                          * shutdown or cp_error, so force to convert error
1533                          * number to EIO for such case.
1534                          */
1535                         if (map->m_may_create &&
1536                                 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1537                                 f2fs_cp_error(sbi))) {
1538                                 err = -EIO;
1539                                 goto unlock_out;
1540                         }
1541
1542                         err = 0;
1543                         if (map->m_next_pgofs)
1544                                 *map->m_next_pgofs =
1545                                         f2fs_get_next_page_offset(&dn, pgofs);
1546                         if (map->m_next_extent)
1547                                 *map->m_next_extent =
1548                                         f2fs_get_next_page_offset(&dn, pgofs);
1549                 }
1550                 goto unlock_out;
1551         }
1552
1553         start_pgofs = pgofs;
1554         prealloc = 0;
1555         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1556         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1557
1558 next_block:
1559         blkaddr = f2fs_data_blkaddr(&dn);
1560
1561         if (__is_valid_data_blkaddr(blkaddr) &&
1562                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1563                 err = -EFSCORRUPTED;
1564                 goto sync_out;
1565         }
1566
1567         if (__is_valid_data_blkaddr(blkaddr)) {
1568                 /* use out-place-update for driect IO under LFS mode */
1569                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1570                                                         map->m_may_create) {
1571                         err = __allocate_data_block(&dn, map->m_seg_type);
1572                         if (err)
1573                                 goto sync_out;
1574                         blkaddr = dn.data_blkaddr;
1575                         set_inode_flag(inode, FI_APPEND_WRITE);
1576                 }
1577         } else {
1578                 if (create) {
1579                         if (unlikely(f2fs_cp_error(sbi))) {
1580                                 err = -EIO;
1581                                 goto sync_out;
1582                         }
1583                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1584                                 if (blkaddr == NULL_ADDR) {
1585                                         prealloc++;
1586                                         last_ofs_in_node = dn.ofs_in_node;
1587                                 }
1588                         } else {
1589                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1590                                         flag != F2FS_GET_BLOCK_DIO);
1591                                 err = __allocate_data_block(&dn,
1592                                                         map->m_seg_type);
1593                                 if (!err)
1594                                         set_inode_flag(inode, FI_APPEND_WRITE);
1595                         }
1596                         if (err)
1597                                 goto sync_out;
1598                         map->m_flags |= F2FS_MAP_NEW;
1599                         blkaddr = dn.data_blkaddr;
1600                 } else {
1601                         if (f2fs_compressed_file(inode) &&
1602                                         f2fs_sanity_check_cluster(&dn) &&
1603                                         (flag != F2FS_GET_BLOCK_FIEMAP ||
1604                                         IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1605                                 err = -EFSCORRUPTED;
1606                                 goto sync_out;
1607                         }
1608                         if (flag == F2FS_GET_BLOCK_BMAP) {
1609                                 map->m_pblk = 0;
1610                                 goto sync_out;
1611                         }
1612                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1613                                 goto sync_out;
1614                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1615                                                 blkaddr == NULL_ADDR) {
1616                                 if (map->m_next_pgofs)
1617                                         *map->m_next_pgofs = pgofs + 1;
1618                                 goto sync_out;
1619                         }
1620                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1621                                 /* for defragment case */
1622                                 if (map->m_next_pgofs)
1623                                         *map->m_next_pgofs = pgofs + 1;
1624                                 goto sync_out;
1625                         }
1626                 }
1627         }
1628
1629         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1630                 goto skip;
1631
1632         if (map->m_multidev_dio)
1633                 bidx = f2fs_target_device_index(sbi, blkaddr);
1634
1635         if (map->m_len == 0) {
1636                 /* preallocated unwritten block should be mapped for fiemap. */
1637                 if (blkaddr == NEW_ADDR)
1638                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1639                 map->m_flags |= F2FS_MAP_MAPPED;
1640
1641                 map->m_pblk = blkaddr;
1642                 map->m_len = 1;
1643
1644                 if (map->m_multidev_dio)
1645                         map->m_bdev = FDEV(bidx).bdev;
1646         } else if ((map->m_pblk != NEW_ADDR &&
1647                         blkaddr == (map->m_pblk + ofs)) ||
1648                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1649                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1650                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1651                         goto sync_out;
1652                 ofs++;
1653                 map->m_len++;
1654         } else {
1655                 goto sync_out;
1656         }
1657
1658 skip:
1659         dn.ofs_in_node++;
1660         pgofs++;
1661
1662         /* preallocate blocks in batch for one dnode page */
1663         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1664                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1665
1666                 dn.ofs_in_node = ofs_in_node;
1667                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1668                 if (err)
1669                         goto sync_out;
1670
1671                 map->m_len += dn.ofs_in_node - ofs_in_node;
1672                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1673                         err = -ENOSPC;
1674                         goto sync_out;
1675                 }
1676                 dn.ofs_in_node = end_offset;
1677         }
1678
1679         if (pgofs >= end)
1680                 goto sync_out;
1681         else if (dn.ofs_in_node < end_offset)
1682                 goto next_block;
1683
1684         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1685                 if (map->m_flags & F2FS_MAP_MAPPED) {
1686                         unsigned int ofs = start_pgofs - map->m_lblk;
1687
1688                         f2fs_update_extent_cache_range(&dn,
1689                                 start_pgofs, map->m_pblk + ofs,
1690                                 map->m_len - ofs);
1691                 }
1692         }
1693
1694         f2fs_put_dnode(&dn);
1695
1696         if (map->m_may_create) {
1697                 f2fs_do_map_lock(sbi, flag, false);
1698                 f2fs_balance_fs(sbi, dn.node_changed);
1699         }
1700         goto next_dnode;
1701
1702 sync_out:
1703
1704         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1705                 /*
1706                  * for hardware encryption, but to avoid potential issue
1707                  * in future
1708                  */
1709                 f2fs_wait_on_block_writeback_range(inode,
1710                                                 map->m_pblk, map->m_len);
1711                 invalidate_mapping_pages(META_MAPPING(sbi),
1712                                                 map->m_pblk, map->m_pblk);
1713
1714                 if (map->m_multidev_dio) {
1715                         block_t blk_addr = map->m_pblk;
1716
1717                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1718
1719                         map->m_bdev = FDEV(bidx).bdev;
1720                         map->m_pblk -= FDEV(bidx).start_blk;
1721
1722                         if (map->m_may_create)
1723                                 f2fs_update_device_state(sbi, inode->i_ino,
1724                                                         blk_addr, map->m_len);
1725
1726                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1727                                                 FDEV(bidx).end_blk + 1);
1728                 }
1729         }
1730
1731         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1732                 if (map->m_flags & F2FS_MAP_MAPPED) {
1733                         unsigned int ofs = start_pgofs - map->m_lblk;
1734
1735                         f2fs_update_extent_cache_range(&dn,
1736                                 start_pgofs, map->m_pblk + ofs,
1737                                 map->m_len - ofs);
1738                 }
1739                 if (map->m_next_extent)
1740                         *map->m_next_extent = pgofs + 1;
1741         }
1742         f2fs_put_dnode(&dn);
1743 unlock_out:
1744         if (map->m_may_create) {
1745                 f2fs_do_map_lock(sbi, flag, false);
1746                 f2fs_balance_fs(sbi, dn.node_changed);
1747         }
1748 out:
1749         trace_f2fs_map_blocks(inode, map, create, flag, err);
1750         return err;
1751 }
1752
1753 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1754 {
1755         struct f2fs_map_blocks map;
1756         block_t last_lblk;
1757         int err;
1758
1759         if (pos + len > i_size_read(inode))
1760                 return false;
1761
1762         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1763         map.m_next_pgofs = NULL;
1764         map.m_next_extent = NULL;
1765         map.m_seg_type = NO_CHECK_TYPE;
1766         map.m_may_create = false;
1767         last_lblk = F2FS_BLK_ALIGN(pos + len);
1768
1769         while (map.m_lblk < last_lblk) {
1770                 map.m_len = last_lblk - map.m_lblk;
1771                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1772                 if (err || map.m_len == 0)
1773                         return false;
1774                 map.m_lblk += map.m_len;
1775         }
1776         return true;
1777 }
1778
1779 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1780 {
1781         return (bytes >> inode->i_blkbits);
1782 }
1783
1784 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1785 {
1786         return (blks << inode->i_blkbits);
1787 }
1788
1789 static int __get_data_block(struct inode *inode, sector_t iblock,
1790                         struct buffer_head *bh, int create, int flag,
1791                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1792 {
1793         struct f2fs_map_blocks map;
1794         int err;
1795
1796         map.m_lblk = iblock;
1797         map.m_len = bytes_to_blks(inode, bh->b_size);
1798         map.m_next_pgofs = next_pgofs;
1799         map.m_next_extent = NULL;
1800         map.m_seg_type = seg_type;
1801         map.m_may_create = may_write;
1802
1803         err = f2fs_map_blocks(inode, &map, create, flag);
1804         if (!err) {
1805                 map_bh(bh, inode->i_sb, map.m_pblk);
1806                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1807                 bh->b_size = blks_to_bytes(inode, map.m_len);
1808
1809                 if (map.m_multidev_dio)
1810                         bh->b_bdev = map.m_bdev;
1811         }
1812         return err;
1813 }
1814
1815 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1816                         struct buffer_head *bh_result, int create)
1817 {
1818         return __get_data_block(inode, iblock, bh_result, create,
1819                                 F2FS_GET_BLOCK_DIO, NULL,
1820                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1821                                 true);
1822 }
1823
1824 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1825                         struct buffer_head *bh_result, int create)
1826 {
1827         return __get_data_block(inode, iblock, bh_result, create,
1828                                 F2FS_GET_BLOCK_DIO, NULL,
1829                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1830                                 false);
1831 }
1832
1833 static int f2fs_xattr_fiemap(struct inode *inode,
1834                                 struct fiemap_extent_info *fieinfo)
1835 {
1836         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1837         struct page *page;
1838         struct node_info ni;
1839         __u64 phys = 0, len;
1840         __u32 flags;
1841         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1842         int err = 0;
1843
1844         if (f2fs_has_inline_xattr(inode)) {
1845                 int offset;
1846
1847                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1848                                                 inode->i_ino, false);
1849                 if (!page)
1850                         return -ENOMEM;
1851
1852                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1853                 if (err) {
1854                         f2fs_put_page(page, 1);
1855                         return err;
1856                 }
1857
1858                 phys = blks_to_bytes(inode, ni.blk_addr);
1859                 offset = offsetof(struct f2fs_inode, i_addr) +
1860                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1861                                         get_inline_xattr_addrs(inode));
1862
1863                 phys += offset;
1864                 len = inline_xattr_size(inode);
1865
1866                 f2fs_put_page(page, 1);
1867
1868                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1869
1870                 if (!xnid)
1871                         flags |= FIEMAP_EXTENT_LAST;
1872
1873                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1874                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1875                 if (err || err == 1)
1876                         return err;
1877         }
1878
1879         if (xnid) {
1880                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1881                 if (!page)
1882                         return -ENOMEM;
1883
1884                 err = f2fs_get_node_info(sbi, xnid, &ni);
1885                 if (err) {
1886                         f2fs_put_page(page, 1);
1887                         return err;
1888                 }
1889
1890                 phys = blks_to_bytes(inode, ni.blk_addr);
1891                 len = inode->i_sb->s_blocksize;
1892
1893                 f2fs_put_page(page, 1);
1894
1895                 flags = FIEMAP_EXTENT_LAST;
1896         }
1897
1898         if (phys) {
1899                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1900                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1901         }
1902
1903         return (err < 0 ? err : 0);
1904 }
1905
1906 static loff_t max_inode_blocks(struct inode *inode)
1907 {
1908         loff_t result = ADDRS_PER_INODE(inode);
1909         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1910
1911         /* two direct node blocks */
1912         result += (leaf_count * 2);
1913
1914         /* two indirect node blocks */
1915         leaf_count *= NIDS_PER_BLOCK;
1916         result += (leaf_count * 2);
1917
1918         /* one double indirect node block */
1919         leaf_count *= NIDS_PER_BLOCK;
1920         result += leaf_count;
1921
1922         return result;
1923 }
1924
1925 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1926                 u64 start, u64 len)
1927 {
1928         struct f2fs_map_blocks map;
1929         sector_t start_blk, last_blk;
1930         pgoff_t next_pgofs;
1931         u64 logical = 0, phys = 0, size = 0;
1932         u32 flags = 0;
1933         int ret = 0;
1934         bool compr_cluster = false, compr_appended;
1935         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1936         unsigned int count_in_cluster = 0;
1937         loff_t maxbytes;
1938
1939         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1940                 ret = f2fs_precache_extents(inode);
1941                 if (ret)
1942                         return ret;
1943         }
1944
1945         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1946         if (ret)
1947                 return ret;
1948
1949         inode_lock(inode);
1950
1951         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1952         if (start > maxbytes) {
1953                 ret = -EFBIG;
1954                 goto out;
1955         }
1956
1957         if (len > maxbytes || (maxbytes - len) < start)
1958                 len = maxbytes - start;
1959
1960         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1961                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1962                 goto out;
1963         }
1964
1965         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1966                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1967                 if (ret != -EAGAIN)
1968                         goto out;
1969         }
1970
1971         if (bytes_to_blks(inode, len) == 0)
1972                 len = blks_to_bytes(inode, 1);
1973
1974         start_blk = bytes_to_blks(inode, start);
1975         last_blk = bytes_to_blks(inode, start + len - 1);
1976
1977 next:
1978         memset(&map, 0, sizeof(map));
1979         map.m_lblk = start_blk;
1980         map.m_len = bytes_to_blks(inode, len);
1981         map.m_next_pgofs = &next_pgofs;
1982         map.m_seg_type = NO_CHECK_TYPE;
1983
1984         if (compr_cluster) {
1985                 map.m_lblk += 1;
1986                 map.m_len = cluster_size - count_in_cluster;
1987         }
1988
1989         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1990         if (ret)
1991                 goto out;
1992
1993         /* HOLE */
1994         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1995                 start_blk = next_pgofs;
1996
1997                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1998                                                 max_inode_blocks(inode)))
1999                         goto prep_next;
2000
2001                 flags |= FIEMAP_EXTENT_LAST;
2002         }
2003
2004         compr_appended = false;
2005         /* In a case of compressed cluster, append this to the last extent */
2006         if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
2007                         !(map.m_flags & F2FS_MAP_FLAGS))) {
2008                 compr_appended = true;
2009                 goto skip_fill;
2010         }
2011
2012         if (size) {
2013                 flags |= FIEMAP_EXTENT_MERGED;
2014                 if (IS_ENCRYPTED(inode))
2015                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2016
2017                 ret = fiemap_fill_next_extent(fieinfo, logical,
2018                                 phys, size, flags);
2019                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2020                 if (ret)
2021                         goto out;
2022                 size = 0;
2023         }
2024
2025         if (start_blk > last_blk)
2026                 goto out;
2027
2028 skip_fill:
2029         if (map.m_pblk == COMPRESS_ADDR) {
2030                 compr_cluster = true;
2031                 count_in_cluster = 1;
2032         } else if (compr_appended) {
2033                 unsigned int appended_blks = cluster_size -
2034                                                 count_in_cluster + 1;
2035                 size += blks_to_bytes(inode, appended_blks);
2036                 start_blk += appended_blks;
2037                 compr_cluster = false;
2038         } else {
2039                 logical = blks_to_bytes(inode, start_blk);
2040                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2041                         blks_to_bytes(inode, map.m_pblk) : 0;
2042                 size = blks_to_bytes(inode, map.m_len);
2043                 flags = 0;
2044
2045                 if (compr_cluster) {
2046                         flags = FIEMAP_EXTENT_ENCODED;
2047                         count_in_cluster += map.m_len;
2048                         if (count_in_cluster == cluster_size) {
2049                                 compr_cluster = false;
2050                                 size += blks_to_bytes(inode, 1);
2051                         }
2052                 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
2053                         flags = FIEMAP_EXTENT_UNWRITTEN;
2054                 }
2055
2056                 start_blk += bytes_to_blks(inode, size);
2057         }
2058
2059 prep_next:
2060         cond_resched();
2061         if (fatal_signal_pending(current))
2062                 ret = -EINTR;
2063         else
2064                 goto next;
2065 out:
2066         if (ret == 1)
2067                 ret = 0;
2068
2069         inode_unlock(inode);
2070         return ret;
2071 }
2072
2073 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2074 {
2075         if (IS_ENABLED(CONFIG_FS_VERITY) &&
2076             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2077                 return inode->i_sb->s_maxbytes;
2078
2079         return i_size_read(inode);
2080 }
2081
2082 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2083                                         unsigned nr_pages,
2084                                         struct f2fs_map_blocks *map,
2085                                         struct bio **bio_ret,
2086                                         sector_t *last_block_in_bio,
2087                                         bool is_readahead)
2088 {
2089         struct bio *bio = *bio_ret;
2090         const unsigned blocksize = blks_to_bytes(inode, 1);
2091         sector_t block_in_file;
2092         sector_t last_block;
2093         sector_t last_block_in_file;
2094         sector_t block_nr;
2095         int ret = 0;
2096
2097         block_in_file = (sector_t)page_index(page);
2098         last_block = block_in_file + nr_pages;
2099         last_block_in_file = bytes_to_blks(inode,
2100                         f2fs_readpage_limit(inode) + blocksize - 1);
2101         if (last_block > last_block_in_file)
2102                 last_block = last_block_in_file;
2103
2104         /* just zeroing out page which is beyond EOF */
2105         if (block_in_file >= last_block)
2106                 goto zero_out;
2107         /*
2108          * Map blocks using the previous result first.
2109          */
2110         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2111                         block_in_file > map->m_lblk &&
2112                         block_in_file < (map->m_lblk + map->m_len))
2113                 goto got_it;
2114
2115         /*
2116          * Then do more f2fs_map_blocks() calls until we are
2117          * done with this page.
2118          */
2119         map->m_lblk = block_in_file;
2120         map->m_len = last_block - block_in_file;
2121
2122         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2123         if (ret)
2124                 goto out;
2125 got_it:
2126         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2127                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2128                 SetPageMappedToDisk(page);
2129
2130                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
2131                                         !cleancache_get_page(page))) {
2132                         SetPageUptodate(page);
2133                         goto confused;
2134                 }
2135
2136                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2137                                                 DATA_GENERIC_ENHANCE_READ)) {
2138                         ret = -EFSCORRUPTED;
2139                         goto out;
2140                 }
2141         } else {
2142 zero_out:
2143                 zero_user_segment(page, 0, PAGE_SIZE);
2144                 if (f2fs_need_verity(inode, page->index) &&
2145                     !fsverity_verify_page(page)) {
2146                         ret = -EIO;
2147                         goto out;
2148                 }
2149                 if (!PageUptodate(page))
2150                         SetPageUptodate(page);
2151                 unlock_page(page);
2152                 goto out;
2153         }
2154
2155         /*
2156          * This page will go to BIO.  Do we need to send this
2157          * BIO off first?
2158          */
2159         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2160                                        *last_block_in_bio, block_nr) ||
2161                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2162 submit_and_realloc:
2163                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2164                 bio = NULL;
2165         }
2166         if (bio == NULL) {
2167                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2168                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2169                                 false);
2170                 if (IS_ERR(bio)) {
2171                         ret = PTR_ERR(bio);
2172                         bio = NULL;
2173                         goto out;
2174                 }
2175         }
2176
2177         /*
2178          * If the page is under writeback, we need to wait for
2179          * its completion to see the correct decrypted data.
2180          */
2181         f2fs_wait_on_block_writeback(inode, block_nr);
2182
2183         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2184                 goto submit_and_realloc;
2185
2186         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2187         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2188         ClearPageError(page);
2189         *last_block_in_bio = block_nr;
2190         goto out;
2191 confused:
2192         if (bio) {
2193                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2194                 bio = NULL;
2195         }
2196         unlock_page(page);
2197 out:
2198         *bio_ret = bio;
2199         return ret;
2200 }
2201
2202 #ifdef CONFIG_F2FS_FS_COMPRESSION
2203 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2204                                 unsigned nr_pages, sector_t *last_block_in_bio,
2205                                 bool is_readahead, bool for_write)
2206 {
2207         struct dnode_of_data dn;
2208         struct inode *inode = cc->inode;
2209         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2210         struct bio *bio = *bio_ret;
2211         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2212         sector_t last_block_in_file;
2213         const unsigned blocksize = blks_to_bytes(inode, 1);
2214         struct decompress_io_ctx *dic = NULL;
2215         struct extent_info ei = {0, };
2216         bool from_dnode = true;
2217         int i;
2218         int ret = 0;
2219
2220         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2221
2222         last_block_in_file = bytes_to_blks(inode,
2223                         f2fs_readpage_limit(inode) + blocksize - 1);
2224
2225         /* get rid of pages beyond EOF */
2226         for (i = 0; i < cc->cluster_size; i++) {
2227                 struct page *page = cc->rpages[i];
2228
2229                 if (!page)
2230                         continue;
2231                 if ((sector_t)page->index >= last_block_in_file) {
2232                         zero_user_segment(page, 0, PAGE_SIZE);
2233                         if (!PageUptodate(page))
2234                                 SetPageUptodate(page);
2235                 } else if (!PageUptodate(page)) {
2236                         continue;
2237                 }
2238                 unlock_page(page);
2239                 if (for_write)
2240                         put_page(page);
2241                 cc->rpages[i] = NULL;
2242                 cc->nr_rpages--;
2243         }
2244
2245         /* we are done since all pages are beyond EOF */
2246         if (f2fs_cluster_is_empty(cc))
2247                 goto out;
2248
2249         if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2250                 from_dnode = false;
2251
2252         if (!from_dnode)
2253                 goto skip_reading_dnode;
2254
2255         set_new_dnode(&dn, inode, NULL, NULL, 0);
2256         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2257         if (ret)
2258                 goto out;
2259
2260         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2261
2262 skip_reading_dnode:
2263         for (i = 1; i < cc->cluster_size; i++) {
2264                 block_t blkaddr;
2265
2266                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2267                                         dn.ofs_in_node + i) :
2268                                         ei.blk + i - 1;
2269
2270                 if (!__is_valid_data_blkaddr(blkaddr))
2271                         break;
2272
2273                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2274                         ret = -EFAULT;
2275                         goto out_put_dnode;
2276                 }
2277                 cc->nr_cpages++;
2278
2279                 if (!from_dnode && i >= ei.c_len)
2280                         break;
2281         }
2282
2283         /* nothing to decompress */
2284         if (cc->nr_cpages == 0) {
2285                 ret = 0;
2286                 goto out_put_dnode;
2287         }
2288
2289         dic = f2fs_alloc_dic(cc);
2290         if (IS_ERR(dic)) {
2291                 ret = PTR_ERR(dic);
2292                 goto out_put_dnode;
2293         }
2294
2295         for (i = 0; i < cc->nr_cpages; i++) {
2296                 struct page *page = dic->cpages[i];
2297                 block_t blkaddr;
2298                 struct bio_post_read_ctx *ctx;
2299
2300                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2301                                         dn.ofs_in_node + i + 1) :
2302                                         ei.blk + i;
2303
2304                 f2fs_wait_on_block_writeback(inode, blkaddr);
2305
2306                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2307                         if (atomic_dec_and_test(&dic->remaining_pages))
2308                                 f2fs_decompress_cluster(dic);
2309                         continue;
2310                 }
2311
2312                 if (bio && (!page_is_mergeable(sbi, bio,
2313                                         *last_block_in_bio, blkaddr) ||
2314                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2315 submit_and_realloc:
2316                         __submit_bio(sbi, bio, DATA);
2317                         bio = NULL;
2318                 }
2319
2320                 if (!bio) {
2321                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2322                                         is_readahead ? REQ_RAHEAD : 0,
2323                                         page->index, for_write);
2324                         if (IS_ERR(bio)) {
2325                                 ret = PTR_ERR(bio);
2326                                 f2fs_decompress_end_io(dic, ret);
2327                                 f2fs_put_dnode(&dn);
2328                                 *bio_ret = NULL;
2329                                 return ret;
2330                         }
2331                 }
2332
2333                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2334                         goto submit_and_realloc;
2335
2336                 ctx = get_post_read_ctx(bio);
2337                 ctx->enabled_steps |= STEP_DECOMPRESS;
2338                 refcount_inc(&dic->refcnt);
2339
2340                 inc_page_count(sbi, F2FS_RD_DATA);
2341                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2342                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2343                 ClearPageError(page);
2344                 *last_block_in_bio = blkaddr;
2345         }
2346
2347         if (from_dnode)
2348                 f2fs_put_dnode(&dn);
2349
2350         *bio_ret = bio;
2351         return 0;
2352
2353 out_put_dnode:
2354         if (from_dnode)
2355                 f2fs_put_dnode(&dn);
2356 out:
2357         for (i = 0; i < cc->cluster_size; i++) {
2358                 if (cc->rpages[i]) {
2359                         ClearPageUptodate(cc->rpages[i]);
2360                         ClearPageError(cc->rpages[i]);
2361                         unlock_page(cc->rpages[i]);
2362                 }
2363         }
2364         *bio_ret = bio;
2365         return ret;
2366 }
2367 #endif
2368
2369 /*
2370  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2371  * Major change was from block_size == page_size in f2fs by default.
2372  */
2373 static int f2fs_mpage_readpages(struct inode *inode,
2374                 struct readahead_control *rac, struct page *page)
2375 {
2376         struct bio *bio = NULL;
2377         sector_t last_block_in_bio = 0;
2378         struct f2fs_map_blocks map;
2379 #ifdef CONFIG_F2FS_FS_COMPRESSION
2380         struct compress_ctx cc = {
2381                 .inode = inode,
2382                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2383                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2384                 .cluster_idx = NULL_CLUSTER,
2385                 .rpages = NULL,
2386                 .cpages = NULL,
2387                 .nr_rpages = 0,
2388                 .nr_cpages = 0,
2389         };
2390         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2391 #endif
2392         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2393         unsigned max_nr_pages = nr_pages;
2394         int ret = 0;
2395
2396         map.m_pblk = 0;
2397         map.m_lblk = 0;
2398         map.m_len = 0;
2399         map.m_flags = 0;
2400         map.m_next_pgofs = NULL;
2401         map.m_next_extent = NULL;
2402         map.m_seg_type = NO_CHECK_TYPE;
2403         map.m_may_create = false;
2404
2405         for (; nr_pages; nr_pages--) {
2406                 if (rac) {
2407                         page = readahead_page(rac);
2408                         prefetchw(&page->flags);
2409                 }
2410
2411 #ifdef CONFIG_F2FS_FS_COMPRESSION
2412                 if (f2fs_compressed_file(inode)) {
2413                         /* there are remained comressed pages, submit them */
2414                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2415                                 ret = f2fs_read_multi_pages(&cc, &bio,
2416                                                         max_nr_pages,
2417                                                         &last_block_in_bio,
2418                                                         rac != NULL, false);
2419                                 f2fs_destroy_compress_ctx(&cc, false);
2420                                 if (ret)
2421                                         goto set_error_page;
2422                         }
2423                         if (cc.cluster_idx == NULL_CLUSTER) {
2424                                 if (nc_cluster_idx ==
2425                                         page->index >> cc.log_cluster_size) {
2426                                         goto read_single_page;
2427                                 }
2428
2429                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2430                                 if (ret < 0)
2431                                         goto set_error_page;
2432                                 else if (!ret) {
2433                                         nc_cluster_idx =
2434                                                 page->index >> cc.log_cluster_size;
2435                                         goto read_single_page;
2436                                 }
2437
2438                                 nc_cluster_idx = NULL_CLUSTER;
2439                         }
2440                         ret = f2fs_init_compress_ctx(&cc);
2441                         if (ret)
2442                                 goto set_error_page;
2443
2444                         f2fs_compress_ctx_add_page(&cc, page);
2445
2446                         goto next_page;
2447                 }
2448 read_single_page:
2449 #endif
2450
2451                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2452                                         &bio, &last_block_in_bio, rac);
2453                 if (ret) {
2454 #ifdef CONFIG_F2FS_FS_COMPRESSION
2455 set_error_page:
2456 #endif
2457                         SetPageError(page);
2458                         zero_user_segment(page, 0, PAGE_SIZE);
2459                         unlock_page(page);
2460                 }
2461 #ifdef CONFIG_F2FS_FS_COMPRESSION
2462 next_page:
2463 #endif
2464                 if (rac)
2465                         put_page(page);
2466
2467 #ifdef CONFIG_F2FS_FS_COMPRESSION
2468                 if (f2fs_compressed_file(inode)) {
2469                         /* last page */
2470                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2471                                 ret = f2fs_read_multi_pages(&cc, &bio,
2472                                                         max_nr_pages,
2473                                                         &last_block_in_bio,
2474                                                         rac != NULL, false);
2475                                 f2fs_destroy_compress_ctx(&cc, false);
2476                         }
2477                 }
2478 #endif
2479         }
2480         if (bio)
2481                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2482         return ret;
2483 }
2484
2485 static int f2fs_read_data_page(struct file *file, struct page *page)
2486 {
2487         struct inode *inode = page_file_mapping(page)->host;
2488         int ret = -EAGAIN;
2489
2490         trace_f2fs_readpage(page, DATA);
2491
2492         if (!f2fs_is_compress_backend_ready(inode)) {
2493                 unlock_page(page);
2494                 return -EOPNOTSUPP;
2495         }
2496
2497         /* If the file has inline data, try to read it directly */
2498         if (f2fs_has_inline_data(inode))
2499                 ret = f2fs_read_inline_data(inode, page);
2500         if (ret == -EAGAIN)
2501                 ret = f2fs_mpage_readpages(inode, NULL, page);
2502         return ret;
2503 }
2504
2505 static void f2fs_readahead(struct readahead_control *rac)
2506 {
2507         struct inode *inode = rac->mapping->host;
2508
2509         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2510
2511         if (!f2fs_is_compress_backend_ready(inode))
2512                 return;
2513
2514         /* If the file has inline data, skip readpages */
2515         if (f2fs_has_inline_data(inode))
2516                 return;
2517
2518         f2fs_mpage_readpages(inode, rac, NULL);
2519 }
2520
2521 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2522 {
2523         struct inode *inode = fio->page->mapping->host;
2524         struct page *mpage, *page;
2525         gfp_t gfp_flags = GFP_NOFS;
2526
2527         if (!f2fs_encrypted_file(inode))
2528                 return 0;
2529
2530         page = fio->compressed_page ? fio->compressed_page : fio->page;
2531
2532         /* wait for GCed page writeback via META_MAPPING */
2533         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2534
2535         if (fscrypt_inode_uses_inline_crypto(inode))
2536                 return 0;
2537
2538 retry_encrypt:
2539         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2540                                         PAGE_SIZE, 0, gfp_flags);
2541         if (IS_ERR(fio->encrypted_page)) {
2542                 /* flush pending IOs and wait for a while in the ENOMEM case */
2543                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2544                         f2fs_flush_merged_writes(fio->sbi);
2545                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2546                         gfp_flags |= __GFP_NOFAIL;
2547                         goto retry_encrypt;
2548                 }
2549                 return PTR_ERR(fio->encrypted_page);
2550         }
2551
2552         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2553         if (mpage) {
2554                 if (PageUptodate(mpage))
2555                         memcpy(page_address(mpage),
2556                                 page_address(fio->encrypted_page), PAGE_SIZE);
2557                 f2fs_put_page(mpage, 1);
2558         }
2559         return 0;
2560 }
2561
2562 static inline bool check_inplace_update_policy(struct inode *inode,
2563                                 struct f2fs_io_info *fio)
2564 {
2565         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2566         unsigned int policy = SM_I(sbi)->ipu_policy;
2567
2568         if (policy & (0x1 << F2FS_IPU_FORCE))
2569                 return true;
2570         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2571                 return true;
2572         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2573                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2574                 return true;
2575         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2576                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2577                 return true;
2578
2579         /*
2580          * IPU for rewrite async pages
2581          */
2582         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2583                         fio && fio->op == REQ_OP_WRITE &&
2584                         !(fio->op_flags & REQ_SYNC) &&
2585                         !IS_ENCRYPTED(inode))
2586                 return true;
2587
2588         /* this is only set during fdatasync */
2589         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2590                         is_inode_flag_set(inode, FI_NEED_IPU))
2591                 return true;
2592
2593         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2594                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2595                 return true;
2596
2597         return false;
2598 }
2599
2600 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2601 {
2602         /* swap file is migrating in aligned write mode */
2603         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2604                 return false;
2605
2606         if (f2fs_is_pinned_file(inode))
2607                 return true;
2608
2609         /* if this is cold file, we should overwrite to avoid fragmentation */
2610         if (file_is_cold(inode))
2611                 return true;
2612
2613         return check_inplace_update_policy(inode, fio);
2614 }
2615
2616 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2617 {
2618         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2619
2620         if (f2fs_lfs_mode(sbi))
2621                 return true;
2622         if (S_ISDIR(inode->i_mode))
2623                 return true;
2624         if (IS_NOQUOTA(inode))
2625                 return true;
2626         if (f2fs_is_atomic_file(inode))
2627                 return true;
2628         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2629                 return true;
2630
2631         /* swap file is migrating in aligned write mode */
2632         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2633                 return true;
2634
2635         if (fio) {
2636                 if (page_private_gcing(fio->page))
2637                         return true;
2638                 if (page_private_dummy(fio->page))
2639                         return true;
2640                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2641                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2642                         return true;
2643         }
2644         return false;
2645 }
2646
2647 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2648 {
2649         struct inode *inode = fio->page->mapping->host;
2650
2651         if (f2fs_should_update_outplace(inode, fio))
2652                 return false;
2653
2654         return f2fs_should_update_inplace(inode, fio);
2655 }
2656
2657 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2658 {
2659         struct page *page = fio->page;
2660         struct inode *inode = page->mapping->host;
2661         struct dnode_of_data dn;
2662         struct extent_info ei = {0, };
2663         struct node_info ni;
2664         bool ipu_force = false;
2665         int err = 0;
2666
2667         set_new_dnode(&dn, inode, NULL, NULL, 0);
2668         if (need_inplace_update(fio) &&
2669                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2670                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2671
2672                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2673                                                 DATA_GENERIC_ENHANCE))
2674                         return -EFSCORRUPTED;
2675
2676                 ipu_force = true;
2677                 fio->need_lock = LOCK_DONE;
2678                 goto got_it;
2679         }
2680
2681         /* Deadlock due to between page->lock and f2fs_lock_op */
2682         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2683                 return -EAGAIN;
2684
2685         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2686         if (err)
2687                 goto out;
2688
2689         fio->old_blkaddr = dn.data_blkaddr;
2690
2691         /* This page is already truncated */
2692         if (fio->old_blkaddr == NULL_ADDR) {
2693                 ClearPageUptodate(page);
2694                 clear_page_private_gcing(page);
2695                 goto out_writepage;
2696         }
2697 got_it:
2698         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2699                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2700                                                 DATA_GENERIC_ENHANCE)) {
2701                 err = -EFSCORRUPTED;
2702                 goto out_writepage;
2703         }
2704         /*
2705          * If current allocation needs SSR,
2706          * it had better in-place writes for updated data.
2707          */
2708         if (ipu_force ||
2709                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2710                                         need_inplace_update(fio))) {
2711                 err = f2fs_encrypt_one_page(fio);
2712                 if (err)
2713                         goto out_writepage;
2714
2715                 set_page_writeback(page);
2716                 ClearPageError(page);
2717                 f2fs_put_dnode(&dn);
2718                 if (fio->need_lock == LOCK_REQ)
2719                         f2fs_unlock_op(fio->sbi);
2720                 err = f2fs_inplace_write_data(fio);
2721                 if (err) {
2722                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2723                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2724                         if (PageWriteback(page))
2725                                 end_page_writeback(page);
2726                 } else {
2727                         set_inode_flag(inode, FI_UPDATE_WRITE);
2728                 }
2729                 trace_f2fs_do_write_data_page(fio->page, IPU);
2730                 return err;
2731         }
2732
2733         if (fio->need_lock == LOCK_RETRY) {
2734                 if (!f2fs_trylock_op(fio->sbi)) {
2735                         err = -EAGAIN;
2736                         goto out_writepage;
2737                 }
2738                 fio->need_lock = LOCK_REQ;
2739         }
2740
2741         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2742         if (err)
2743                 goto out_writepage;
2744
2745         fio->version = ni.version;
2746
2747         err = f2fs_encrypt_one_page(fio);
2748         if (err)
2749                 goto out_writepage;
2750
2751         set_page_writeback(page);
2752         ClearPageError(page);
2753
2754         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2755                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2756
2757         /* LFS mode write path */
2758         f2fs_outplace_write_data(&dn, fio);
2759         trace_f2fs_do_write_data_page(page, OPU);
2760         set_inode_flag(inode, FI_APPEND_WRITE);
2761         if (page->index == 0)
2762                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2763 out_writepage:
2764         f2fs_put_dnode(&dn);
2765 out:
2766         if (fio->need_lock == LOCK_REQ)
2767                 f2fs_unlock_op(fio->sbi);
2768         return err;
2769 }
2770
2771 int f2fs_write_single_data_page(struct page *page, int *submitted,
2772                                 struct bio **bio,
2773                                 sector_t *last_block,
2774                                 struct writeback_control *wbc,
2775                                 enum iostat_type io_type,
2776                                 int compr_blocks,
2777                                 bool allow_balance)
2778 {
2779         struct inode *inode = page->mapping->host;
2780         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2781         loff_t i_size = i_size_read(inode);
2782         const pgoff_t end_index = ((unsigned long long)i_size)
2783                                                         >> PAGE_SHIFT;
2784         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2785         unsigned offset = 0;
2786         bool need_balance_fs = false;
2787         int err = 0;
2788         struct f2fs_io_info fio = {
2789                 .sbi = sbi,
2790                 .ino = inode->i_ino,
2791                 .type = DATA,
2792                 .op = REQ_OP_WRITE,
2793                 .op_flags = wbc_to_write_flags(wbc),
2794                 .old_blkaddr = NULL_ADDR,
2795                 .page = page,
2796                 .encrypted_page = NULL,
2797                 .submitted = false,
2798                 .compr_blocks = compr_blocks,
2799                 .need_lock = LOCK_RETRY,
2800                 .io_type = io_type,
2801                 .io_wbc = wbc,
2802                 .bio = bio,
2803                 .last_block = last_block,
2804         };
2805
2806         trace_f2fs_writepage(page, DATA);
2807
2808         /* we should bypass data pages to proceed the kworkder jobs */
2809         if (unlikely(f2fs_cp_error(sbi))) {
2810                 mapping_set_error(page->mapping, -EIO);
2811                 /*
2812                  * don't drop any dirty dentry pages for keeping lastest
2813                  * directory structure.
2814                  */
2815                 if (S_ISDIR(inode->i_mode))
2816                         goto redirty_out;
2817                 goto out;
2818         }
2819
2820         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2821                 goto redirty_out;
2822
2823         if (page->index < end_index ||
2824                         f2fs_verity_in_progress(inode) ||
2825                         compr_blocks)
2826                 goto write;
2827
2828         /*
2829          * If the offset is out-of-range of file size,
2830          * this page does not have to be written to disk.
2831          */
2832         offset = i_size & (PAGE_SIZE - 1);
2833         if ((page->index >= end_index + 1) || !offset)
2834                 goto out;
2835
2836         zero_user_segment(page, offset, PAGE_SIZE);
2837 write:
2838         if (f2fs_is_drop_cache(inode))
2839                 goto out;
2840         /* we should not write 0'th page having journal header */
2841         if (f2fs_is_volatile_file(inode) && (!page->index ||
2842                         (!wbc->for_reclaim &&
2843                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2844                 goto redirty_out;
2845
2846         /* Dentry/quota blocks are controlled by checkpoint */
2847         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2848                 /*
2849                  * We need to wait for node_write to avoid block allocation during
2850                  * checkpoint. This can only happen to quota writes which can cause
2851                  * the below discard race condition.
2852                  */
2853                 if (IS_NOQUOTA(inode))
2854                         down_read(&sbi->node_write);
2855
2856                 fio.need_lock = LOCK_DONE;
2857                 err = f2fs_do_write_data_page(&fio);
2858
2859                 if (IS_NOQUOTA(inode))
2860                         up_read(&sbi->node_write);
2861
2862                 goto done;
2863         }
2864
2865         if (!wbc->for_reclaim)
2866                 need_balance_fs = true;
2867         else if (has_not_enough_free_secs(sbi, 0, 0))
2868                 goto redirty_out;
2869         else
2870                 set_inode_flag(inode, FI_HOT_DATA);
2871
2872         err = -EAGAIN;
2873         if (f2fs_has_inline_data(inode)) {
2874                 err = f2fs_write_inline_data(inode, page);
2875                 if (!err)
2876                         goto out;
2877         }
2878
2879         if (err == -EAGAIN) {
2880                 err = f2fs_do_write_data_page(&fio);
2881                 if (err == -EAGAIN) {
2882                         fio.need_lock = LOCK_REQ;
2883                         err = f2fs_do_write_data_page(&fio);
2884                 }
2885         }
2886
2887         if (err) {
2888                 file_set_keep_isize(inode);
2889         } else {
2890                 spin_lock(&F2FS_I(inode)->i_size_lock);
2891                 if (F2FS_I(inode)->last_disk_size < psize)
2892                         F2FS_I(inode)->last_disk_size = psize;
2893                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2894         }
2895
2896 done:
2897         if (err && err != -ENOENT)
2898                 goto redirty_out;
2899
2900 out:
2901         inode_dec_dirty_pages(inode);
2902         if (err) {
2903                 ClearPageUptodate(page);
2904                 clear_page_private_gcing(page);
2905         }
2906
2907         if (wbc->for_reclaim) {
2908                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2909                 clear_inode_flag(inode, FI_HOT_DATA);
2910                 f2fs_remove_dirty_inode(inode);
2911                 submitted = NULL;
2912         }
2913         unlock_page(page);
2914         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2915                         !F2FS_I(inode)->cp_task && allow_balance)
2916                 f2fs_balance_fs(sbi, need_balance_fs);
2917
2918         if (unlikely(f2fs_cp_error(sbi))) {
2919                 f2fs_submit_merged_write(sbi, DATA);
2920                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2921                 submitted = NULL;
2922         }
2923
2924         if (submitted)
2925                 *submitted = fio.submitted ? 1 : 0;
2926
2927         return 0;
2928
2929 redirty_out:
2930         redirty_page_for_writepage(wbc, page);
2931         /*
2932          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2933          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2934          * file_write_and_wait_range() will see EIO error, which is critical
2935          * to return value of fsync() followed by atomic_write failure to user.
2936          */
2937         if (!err || wbc->for_reclaim)
2938                 return AOP_WRITEPAGE_ACTIVATE;
2939         unlock_page(page);
2940         return err;
2941 }
2942
2943 static int f2fs_write_data_page(struct page *page,
2944                                         struct writeback_control *wbc)
2945 {
2946 #ifdef CONFIG_F2FS_FS_COMPRESSION
2947         struct inode *inode = page->mapping->host;
2948
2949         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2950                 goto out;
2951
2952         if (f2fs_compressed_file(inode)) {
2953                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2954                         redirty_page_for_writepage(wbc, page);
2955                         return AOP_WRITEPAGE_ACTIVATE;
2956                 }
2957         }
2958 out:
2959 #endif
2960
2961         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2962                                                 wbc, FS_DATA_IO, 0, true);
2963 }
2964
2965 /*
2966  * This function was copied from write_cche_pages from mm/page-writeback.c.
2967  * The major change is making write step of cold data page separately from
2968  * warm/hot data page.
2969  */
2970 static int f2fs_write_cache_pages(struct address_space *mapping,
2971                                         struct writeback_control *wbc,
2972                                         enum iostat_type io_type)
2973 {
2974         int ret = 0;
2975         int done = 0, retry = 0;
2976         struct pagevec pvec;
2977         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2978         struct bio *bio = NULL;
2979         sector_t last_block;
2980 #ifdef CONFIG_F2FS_FS_COMPRESSION
2981         struct inode *inode = mapping->host;
2982         struct compress_ctx cc = {
2983                 .inode = inode,
2984                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2985                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2986                 .cluster_idx = NULL_CLUSTER,
2987                 .rpages = NULL,
2988                 .nr_rpages = 0,
2989                 .cpages = NULL,
2990                 .valid_nr_cpages = 0,
2991                 .rbuf = NULL,
2992                 .cbuf = NULL,
2993                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2994                 .private = NULL,
2995         };
2996 #endif
2997         int nr_pages;
2998         pgoff_t index;
2999         pgoff_t end;            /* Inclusive */
3000         pgoff_t done_index;
3001         int range_whole = 0;
3002         xa_mark_t tag;
3003         int nwritten = 0;
3004         int submitted = 0;
3005         int i;
3006
3007         pagevec_init(&pvec);
3008
3009         if (get_dirty_pages(mapping->host) <=
3010                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3011                 set_inode_flag(mapping->host, FI_HOT_DATA);
3012         else
3013                 clear_inode_flag(mapping->host, FI_HOT_DATA);
3014
3015         if (wbc->range_cyclic) {
3016                 index = mapping->writeback_index; /* prev offset */
3017                 end = -1;
3018         } else {
3019                 index = wbc->range_start >> PAGE_SHIFT;
3020                 end = wbc->range_end >> PAGE_SHIFT;
3021                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3022                         range_whole = 1;
3023         }
3024         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3025                 tag = PAGECACHE_TAG_TOWRITE;
3026         else
3027                 tag = PAGECACHE_TAG_DIRTY;
3028 retry:
3029         retry = 0;
3030         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3031                 tag_pages_for_writeback(mapping, index, end);
3032         done_index = index;
3033         while (!done && !retry && (index <= end)) {
3034                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3035                                 tag);
3036                 if (nr_pages == 0)
3037                         break;
3038
3039                 for (i = 0; i < nr_pages; i++) {
3040                         struct page *page = pvec.pages[i];
3041                         bool need_readd;
3042 readd:
3043                         need_readd = false;
3044 #ifdef CONFIG_F2FS_FS_COMPRESSION
3045                         if (f2fs_compressed_file(inode)) {
3046                                 void *fsdata = NULL;
3047                                 struct page *pagep;
3048                                 int ret2;
3049
3050                                 ret = f2fs_init_compress_ctx(&cc);
3051                                 if (ret) {
3052                                         done = 1;
3053                                         break;
3054                                 }
3055
3056                                 if (!f2fs_cluster_can_merge_page(&cc,
3057                                                                 page->index)) {
3058                                         ret = f2fs_write_multi_pages(&cc,
3059                                                 &submitted, wbc, io_type);
3060                                         if (!ret)
3061                                                 need_readd = true;
3062                                         goto result;
3063                                 }
3064
3065                                 if (unlikely(f2fs_cp_error(sbi)))
3066                                         goto lock_page;
3067
3068                                 if (!f2fs_cluster_is_empty(&cc))
3069                                         goto lock_page;
3070
3071                                 ret2 = f2fs_prepare_compress_overwrite(
3072                                                         inode, &pagep,
3073                                                         page->index, &fsdata);
3074                                 if (ret2 < 0) {
3075                                         ret = ret2;
3076                                         done = 1;
3077                                         break;
3078                                 } else if (ret2 &&
3079                                         (!f2fs_compress_write_end(inode,
3080                                                 fsdata, page->index, 1) ||
3081                                          !f2fs_all_cluster_page_loaded(&cc,
3082                                                 &pvec, i, nr_pages))) {
3083                                         retry = 1;
3084                                         break;
3085                                 }
3086                         }
3087 #endif
3088                         /* give a priority to WB_SYNC threads */
3089                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3090                                         wbc->sync_mode == WB_SYNC_NONE) {
3091                                 done = 1;
3092                                 break;
3093                         }
3094 #ifdef CONFIG_F2FS_FS_COMPRESSION
3095 lock_page:
3096 #endif
3097                         done_index = page->index;
3098 retry_write:
3099                         lock_page(page);
3100
3101                         if (unlikely(page->mapping != mapping)) {
3102 continue_unlock:
3103                                 unlock_page(page);
3104                                 continue;
3105                         }
3106
3107                         if (!PageDirty(page)) {
3108                                 /* someone wrote it for us */
3109                                 goto continue_unlock;
3110                         }
3111
3112                         if (PageWriteback(page)) {
3113                                 if (wbc->sync_mode != WB_SYNC_NONE)
3114                                         f2fs_wait_on_page_writeback(page,
3115                                                         DATA, true, true);
3116                                 else
3117                                         goto continue_unlock;
3118                         }
3119
3120                         if (!clear_page_dirty_for_io(page))
3121                                 goto continue_unlock;
3122
3123 #ifdef CONFIG_F2FS_FS_COMPRESSION
3124                         if (f2fs_compressed_file(inode)) {
3125                                 get_page(page);
3126                                 f2fs_compress_ctx_add_page(&cc, page);
3127                                 continue;
3128                         }
3129 #endif
3130                         ret = f2fs_write_single_data_page(page, &submitted,
3131                                         &bio, &last_block, wbc, io_type,
3132                                         0, true);
3133                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3134                                 unlock_page(page);
3135 #ifdef CONFIG_F2FS_FS_COMPRESSION
3136 result:
3137 #endif
3138                         nwritten += submitted;
3139                         wbc->nr_to_write -= submitted;
3140
3141                         if (unlikely(ret)) {
3142                                 /*
3143                                  * keep nr_to_write, since vfs uses this to
3144                                  * get # of written pages.
3145                                  */
3146                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3147                                         ret = 0;
3148                                         goto next;
3149                                 } else if (ret == -EAGAIN) {
3150                                         ret = 0;
3151                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3152                                                 cond_resched();
3153                                                 congestion_wait(BLK_RW_ASYNC,
3154                                                         DEFAULT_IO_TIMEOUT);
3155                                                 goto retry_write;
3156                                         }
3157                                         goto next;
3158                                 }
3159                                 done_index = page->index + 1;
3160                                 done = 1;
3161                                 break;
3162                         }
3163
3164                         if (wbc->nr_to_write <= 0 &&
3165                                         wbc->sync_mode == WB_SYNC_NONE) {
3166                                 done = 1;
3167                                 break;
3168                         }
3169 next:
3170                         if (need_readd)
3171                                 goto readd;
3172                 }
3173                 pagevec_release(&pvec);
3174                 cond_resched();
3175         }
3176 #ifdef CONFIG_F2FS_FS_COMPRESSION
3177         /* flush remained pages in compress cluster */
3178         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3179                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3180                 nwritten += submitted;
3181                 wbc->nr_to_write -= submitted;
3182                 if (ret) {
3183                         done = 1;
3184                         retry = 0;
3185                 }
3186         }
3187         if (f2fs_compressed_file(inode))
3188                 f2fs_destroy_compress_ctx(&cc, false);
3189 #endif
3190         if (retry) {
3191                 index = 0;
3192                 end = -1;
3193                 goto retry;
3194         }
3195         if (wbc->range_cyclic && !done)
3196                 done_index = 0;
3197         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3198                 mapping->writeback_index = done_index;
3199
3200         if (nwritten)
3201                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3202                                                                 NULL, 0, DATA);
3203         /* submit cached bio of IPU write */
3204         if (bio)
3205                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3206
3207         return ret;
3208 }
3209
3210 static inline bool __should_serialize_io(struct inode *inode,
3211                                         struct writeback_control *wbc)
3212 {
3213         /* to avoid deadlock in path of data flush */
3214         if (F2FS_I(inode)->cp_task)
3215                 return false;
3216
3217         if (!S_ISREG(inode->i_mode))
3218                 return false;
3219         if (IS_NOQUOTA(inode))
3220                 return false;
3221
3222         if (f2fs_need_compress_data(inode))
3223                 return true;
3224         if (wbc->sync_mode != WB_SYNC_ALL)
3225                 return true;
3226         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3227                 return true;
3228         return false;
3229 }
3230
3231 static int __f2fs_write_data_pages(struct address_space *mapping,
3232                                                 struct writeback_control *wbc,
3233                                                 enum iostat_type io_type)
3234 {
3235         struct inode *inode = mapping->host;
3236         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3237         struct blk_plug plug;
3238         int ret;
3239         bool locked = false;
3240
3241         /* deal with chardevs and other special file */
3242         if (!mapping->a_ops->writepage)
3243                 return 0;
3244
3245         /* skip writing if there is no dirty page in this inode */
3246         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3247                 return 0;
3248
3249         /* during POR, we don't need to trigger writepage at all. */
3250         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3251                 goto skip_write;
3252
3253         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3254                         wbc->sync_mode == WB_SYNC_NONE &&
3255                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3256                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3257                 goto skip_write;
3258
3259         /* skip writing during file defragment */
3260         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3261                 goto skip_write;
3262
3263         trace_f2fs_writepages(mapping->host, wbc, DATA);
3264
3265         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3266         if (wbc->sync_mode == WB_SYNC_ALL)
3267                 atomic_inc(&sbi->wb_sync_req[DATA]);
3268         else if (atomic_read(&sbi->wb_sync_req[DATA]))
3269                 goto skip_write;
3270
3271         if (__should_serialize_io(inode, wbc)) {
3272                 mutex_lock(&sbi->writepages);
3273                 locked = true;
3274         }
3275
3276         blk_start_plug(&plug);
3277         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3278         blk_finish_plug(&plug);
3279
3280         if (locked)
3281                 mutex_unlock(&sbi->writepages);
3282
3283         if (wbc->sync_mode == WB_SYNC_ALL)
3284                 atomic_dec(&sbi->wb_sync_req[DATA]);
3285         /*
3286          * if some pages were truncated, we cannot guarantee its mapping->host
3287          * to detect pending bios.
3288          */
3289
3290         f2fs_remove_dirty_inode(inode);
3291         return ret;
3292
3293 skip_write:
3294         wbc->pages_skipped += get_dirty_pages(inode);
3295         trace_f2fs_writepages(mapping->host, wbc, DATA);
3296         return 0;
3297 }
3298
3299 static int f2fs_write_data_pages(struct address_space *mapping,
3300                             struct writeback_control *wbc)
3301 {
3302         struct inode *inode = mapping->host;
3303
3304         return __f2fs_write_data_pages(mapping, wbc,
3305                         F2FS_I(inode)->cp_task == current ?
3306                         FS_CP_DATA_IO : FS_DATA_IO);
3307 }
3308
3309 static void f2fs_write_failed(struct inode *inode, loff_t to)
3310 {
3311         loff_t i_size = i_size_read(inode);
3312
3313         if (IS_NOQUOTA(inode))
3314                 return;
3315
3316         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3317         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3318                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3319                 filemap_invalidate_lock(inode->i_mapping);
3320
3321                 truncate_pagecache(inode, i_size);
3322                 f2fs_truncate_blocks(inode, i_size, true);
3323
3324                 filemap_invalidate_unlock(inode->i_mapping);
3325                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3326         }
3327 }
3328
3329 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3330                         struct page *page, loff_t pos, unsigned len,
3331                         block_t *blk_addr, bool *node_changed)
3332 {
3333         struct inode *inode = page->mapping->host;
3334         pgoff_t index = page->index;
3335         struct dnode_of_data dn;
3336         struct page *ipage;
3337         bool locked = false;
3338         struct extent_info ei = {0, };
3339         int err = 0;
3340         int flag;
3341
3342         /*
3343          * we already allocated all the blocks, so we don't need to get
3344          * the block addresses when there is no need to fill the page.
3345          */
3346         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3347             !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3348             !f2fs_verity_in_progress(inode))
3349                 return 0;
3350
3351         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3352         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3353                 flag = F2FS_GET_BLOCK_DEFAULT;
3354         else
3355                 flag = F2FS_GET_BLOCK_PRE_AIO;
3356
3357         if (f2fs_has_inline_data(inode) ||
3358                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3359                 f2fs_do_map_lock(sbi, flag, true);
3360                 locked = true;
3361         }
3362
3363 restart:
3364         /* check inline_data */
3365         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3366         if (IS_ERR(ipage)) {
3367                 err = PTR_ERR(ipage);
3368                 goto unlock_out;
3369         }
3370
3371         set_new_dnode(&dn, inode, ipage, ipage, 0);
3372
3373         if (f2fs_has_inline_data(inode)) {
3374                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3375                         f2fs_do_read_inline_data(page, ipage);
3376                         set_inode_flag(inode, FI_DATA_EXIST);
3377                         if (inode->i_nlink)
3378                                 set_page_private_inline(ipage);
3379                 } else {
3380                         err = f2fs_convert_inline_page(&dn, page);
3381                         if (err)
3382                                 goto out;
3383                         if (dn.data_blkaddr == NULL_ADDR)
3384                                 err = f2fs_get_block(&dn, index);
3385                 }
3386         } else if (locked) {
3387                 err = f2fs_get_block(&dn, index);
3388         } else {
3389                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3390                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3391                 } else {
3392                         /* hole case */
3393                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3394                         if (err || dn.data_blkaddr == NULL_ADDR) {
3395                                 f2fs_put_dnode(&dn);
3396                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3397                                                                 true);
3398                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3399                                 locked = true;
3400                                 goto restart;
3401                         }
3402                 }
3403         }
3404
3405         /* convert_inline_page can make node_changed */
3406         *blk_addr = dn.data_blkaddr;
3407         *node_changed = dn.node_changed;
3408 out:
3409         f2fs_put_dnode(&dn);
3410 unlock_out:
3411         if (locked)
3412                 f2fs_do_map_lock(sbi, flag, false);
3413         return err;
3414 }
3415
3416 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3417                 loff_t pos, unsigned len, unsigned flags,
3418                 struct page **pagep, void **fsdata)
3419 {
3420         struct inode *inode = mapping->host;
3421         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3422         struct page *page = NULL;
3423         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3424         bool need_balance = false, drop_atomic = false;
3425         block_t blkaddr = NULL_ADDR;
3426         int err = 0;
3427
3428         trace_f2fs_write_begin(inode, pos, len, flags);
3429
3430         if (!f2fs_is_checkpoint_ready(sbi)) {
3431                 err = -ENOSPC;
3432                 goto fail;
3433         }
3434
3435         if ((f2fs_is_atomic_file(inode) &&
3436                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3437                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3438                 err = -ENOMEM;
3439                 drop_atomic = true;
3440                 goto fail;
3441         }
3442
3443         /*
3444          * We should check this at this moment to avoid deadlock on inode page
3445          * and #0 page. The locking rule for inline_data conversion should be:
3446          * lock_page(page #0) -> lock_page(inode_page)
3447          */
3448         if (index != 0) {
3449                 err = f2fs_convert_inline_inode(inode);
3450                 if (err)
3451                         goto fail;
3452         }
3453
3454 #ifdef CONFIG_F2FS_FS_COMPRESSION
3455         if (f2fs_compressed_file(inode)) {
3456                 int ret;
3457
3458                 *fsdata = NULL;
3459
3460                 if (len == PAGE_SIZE)
3461                         goto repeat;
3462
3463                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3464                                                         index, fsdata);
3465                 if (ret < 0) {
3466                         err = ret;
3467                         goto fail;
3468                 } else if (ret) {
3469                         return 0;
3470                 }
3471         }
3472 #endif
3473
3474 repeat:
3475         /*
3476          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3477          * wait_for_stable_page. Will wait that below with our IO control.
3478          */
3479         page = f2fs_pagecache_get_page(mapping, index,
3480                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3481         if (!page) {
3482                 err = -ENOMEM;
3483                 goto fail;
3484         }
3485
3486         /* TODO: cluster can be compressed due to race with .writepage */
3487
3488         *pagep = page;
3489
3490         err = prepare_write_begin(sbi, page, pos, len,
3491                                         &blkaddr, &need_balance);
3492         if (err)
3493                 goto fail;
3494
3495         if (need_balance && !IS_NOQUOTA(inode) &&
3496                         has_not_enough_free_secs(sbi, 0, 0)) {
3497                 unlock_page(page);
3498                 f2fs_balance_fs(sbi, true);
3499                 lock_page(page);
3500                 if (page->mapping != mapping) {
3501                         /* The page got truncated from under us */
3502                         f2fs_put_page(page, 1);
3503                         goto repeat;
3504                 }
3505         }
3506
3507         f2fs_wait_on_page_writeback(page, DATA, false, true);
3508
3509         if (len == PAGE_SIZE || PageUptodate(page))
3510                 return 0;
3511
3512         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3513             !f2fs_verity_in_progress(inode)) {
3514                 zero_user_segment(page, len, PAGE_SIZE);
3515                 return 0;
3516         }
3517
3518         if (blkaddr == NEW_ADDR) {
3519                 zero_user_segment(page, 0, PAGE_SIZE);
3520                 SetPageUptodate(page);
3521         } else {
3522                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3523                                 DATA_GENERIC_ENHANCE_READ)) {
3524                         err = -EFSCORRUPTED;
3525                         goto fail;
3526                 }
3527                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3528                 if (err)
3529                         goto fail;
3530
3531                 lock_page(page);
3532                 if (unlikely(page->mapping != mapping)) {
3533                         f2fs_put_page(page, 1);
3534                         goto repeat;
3535                 }
3536                 if (unlikely(!PageUptodate(page))) {
3537                         err = -EIO;
3538                         goto fail;
3539                 }
3540         }
3541         return 0;
3542
3543 fail:
3544         f2fs_put_page(page, 1);
3545         f2fs_write_failed(inode, pos + len);
3546         if (drop_atomic)
3547                 f2fs_drop_inmem_pages_all(sbi, false);
3548         return err;
3549 }
3550
3551 static int f2fs_write_end(struct file *file,
3552                         struct address_space *mapping,
3553                         loff_t pos, unsigned len, unsigned copied,
3554                         struct page *page, void *fsdata)
3555 {
3556         struct inode *inode = page->mapping->host;
3557
3558         trace_f2fs_write_end(inode, pos, len, copied);
3559
3560         /*
3561          * This should be come from len == PAGE_SIZE, and we expect copied
3562          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3563          * let generic_perform_write() try to copy data again through copied=0.
3564          */
3565         if (!PageUptodate(page)) {
3566                 if (unlikely(copied != len))
3567                         copied = 0;
3568                 else
3569                         SetPageUptodate(page);
3570         }
3571
3572 #ifdef CONFIG_F2FS_FS_COMPRESSION
3573         /* overwrite compressed file */
3574         if (f2fs_compressed_file(inode) && fsdata) {
3575                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3576                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3577
3578                 if (pos + copied > i_size_read(inode) &&
3579                                 !f2fs_verity_in_progress(inode))
3580                         f2fs_i_size_write(inode, pos + copied);
3581                 return copied;
3582         }
3583 #endif
3584
3585         if (!copied)
3586                 goto unlock_out;
3587
3588         set_page_dirty(page);
3589
3590         if (pos + copied > i_size_read(inode) &&
3591             !f2fs_verity_in_progress(inode))
3592                 f2fs_i_size_write(inode, pos + copied);
3593 unlock_out:
3594         f2fs_put_page(page, 1);
3595         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3596         return copied;
3597 }
3598
3599 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3600                            loff_t offset)
3601 {
3602         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3603         unsigned blkbits = i_blkbits;
3604         unsigned blocksize_mask = (1 << blkbits) - 1;
3605         unsigned long align = offset | iov_iter_alignment(iter);
3606         struct block_device *bdev = inode->i_sb->s_bdev;
3607
3608         if (iov_iter_rw(iter) == READ && offset >= i_size_read(inode))
3609                 return 1;
3610
3611         if (align & blocksize_mask) {
3612                 if (bdev)
3613                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
3614                 blocksize_mask = (1 << blkbits) - 1;
3615                 if (align & blocksize_mask)
3616                         return -EINVAL;
3617                 return 1;
3618         }
3619         return 0;
3620 }
3621
3622 static void f2fs_dio_end_io(struct bio *bio)
3623 {
3624         struct f2fs_private_dio *dio = bio->bi_private;
3625
3626         dec_page_count(F2FS_I_SB(dio->inode),
3627                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3628
3629         bio->bi_private = dio->orig_private;
3630         bio->bi_end_io = dio->orig_end_io;
3631
3632         kfree(dio);
3633
3634         bio_endio(bio);
3635 }
3636
3637 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3638                                                         loff_t file_offset)
3639 {
3640         struct f2fs_private_dio *dio;
3641         bool write = (bio_op(bio) == REQ_OP_WRITE);
3642
3643         dio = f2fs_kzalloc(F2FS_I_SB(inode),
3644                         sizeof(struct f2fs_private_dio), GFP_NOFS);
3645         if (!dio)
3646                 goto out;
3647
3648         dio->inode = inode;
3649         dio->orig_end_io = bio->bi_end_io;
3650         dio->orig_private = bio->bi_private;
3651         dio->write = write;
3652
3653         bio->bi_end_io = f2fs_dio_end_io;
3654         bio->bi_private = dio;
3655
3656         inc_page_count(F2FS_I_SB(inode),
3657                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3658
3659         submit_bio(bio);
3660         return;
3661 out:
3662         bio->bi_status = BLK_STS_IOERR;
3663         bio_endio(bio);
3664 }
3665
3666 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3667 {
3668         struct address_space *mapping = iocb->ki_filp->f_mapping;
3669         struct inode *inode = mapping->host;
3670         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3671         struct f2fs_inode_info *fi = F2FS_I(inode);
3672         size_t count = iov_iter_count(iter);
3673         loff_t offset = iocb->ki_pos;
3674         int rw = iov_iter_rw(iter);
3675         int err;
3676         enum rw_hint hint = iocb->ki_hint;
3677         int whint_mode = F2FS_OPTION(sbi).whint_mode;
3678         bool do_opu;
3679
3680         err = check_direct_IO(inode, iter, offset);
3681         if (err)
3682                 return err < 0 ? err : 0;
3683
3684         if (f2fs_force_buffered_io(inode, iocb, iter))
3685                 return 0;
3686
3687         do_opu = rw == WRITE && f2fs_lfs_mode(sbi);
3688
3689         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3690
3691         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3692                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
3693
3694         if (iocb->ki_flags & IOCB_NOWAIT) {
3695                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3696                         iocb->ki_hint = hint;
3697                         err = -EAGAIN;
3698                         goto out;
3699                 }
3700                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3701                         up_read(&fi->i_gc_rwsem[rw]);
3702                         iocb->ki_hint = hint;
3703                         err = -EAGAIN;
3704                         goto out;
3705                 }
3706         } else {
3707                 down_read(&fi->i_gc_rwsem[rw]);
3708                 if (do_opu)
3709                         down_read(&fi->i_gc_rwsem[READ]);
3710         }
3711
3712         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3713                         iter, rw == WRITE ? get_data_block_dio_write :
3714                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
3715                         rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3716                         DIO_SKIP_HOLES);
3717
3718         if (do_opu)
3719                 up_read(&fi->i_gc_rwsem[READ]);
3720
3721         up_read(&fi->i_gc_rwsem[rw]);
3722
3723         if (rw == WRITE) {
3724                 if (whint_mode == WHINT_MODE_OFF)
3725                         iocb->ki_hint = hint;
3726                 if (err > 0) {
3727                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3728                                                                         err);
3729                         if (!do_opu)
3730                                 set_inode_flag(inode, FI_UPDATE_WRITE);
3731                 } else if (err == -EIOCBQUEUED) {
3732                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3733                                                 count - iov_iter_count(iter));
3734                 } else if (err < 0) {
3735                         f2fs_write_failed(inode, offset + count);
3736                 }
3737         } else {
3738                 if (err > 0)
3739                         f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3740                 else if (err == -EIOCBQUEUED)
3741                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_READ_IO,
3742                                                 count - iov_iter_count(iter));
3743         }
3744
3745 out:
3746         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3747
3748         return err;
3749 }
3750
3751 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3752                                                         unsigned int length)
3753 {
3754         struct inode *inode = page->mapping->host;
3755         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3756
3757         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3758                 (offset % PAGE_SIZE || length != PAGE_SIZE))
3759                 return;
3760
3761         if (PageDirty(page)) {
3762                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3763                         dec_page_count(sbi, F2FS_DIRTY_META);
3764                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3765                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3766                 } else {
3767                         inode_dec_dirty_pages(inode);
3768                         f2fs_remove_dirty_inode(inode);
3769                 }
3770         }
3771
3772         clear_page_private_gcing(page);
3773
3774         if (test_opt(sbi, COMPRESS_CACHE)) {
3775                 if (f2fs_compressed_file(inode))
3776                         f2fs_invalidate_compress_pages(sbi, inode->i_ino);
3777                 if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3778                         clear_page_private_data(page);
3779         }
3780
3781         if (page_private_atomic(page))
3782                 return f2fs_drop_inmem_page(inode, page);
3783
3784         detach_page_private(page);
3785         set_page_private(page, 0);
3786 }
3787
3788 int f2fs_release_page(struct page *page, gfp_t wait)
3789 {
3790         /* If this is dirty page, keep PagePrivate */
3791         if (PageDirty(page))
3792                 return 0;
3793
3794         /* This is atomic written page, keep Private */
3795         if (page_private_atomic(page))
3796                 return 0;
3797
3798         if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3799                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3800                 struct inode *inode = page->mapping->host;
3801
3802                 if (f2fs_compressed_file(inode))
3803                         f2fs_invalidate_compress_pages(sbi, inode->i_ino);
3804                 if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3805                         clear_page_private_data(page);
3806         }
3807
3808         clear_page_private_gcing(page);
3809
3810         detach_page_private(page);
3811         set_page_private(page, 0);
3812         return 1;
3813 }
3814
3815 static int f2fs_set_data_page_dirty(struct page *page)
3816 {
3817         struct inode *inode = page_file_mapping(page)->host;
3818
3819         trace_f2fs_set_page_dirty(page, DATA);
3820
3821         if (!PageUptodate(page))
3822                 SetPageUptodate(page);
3823         if (PageSwapCache(page))
3824                 return __set_page_dirty_nobuffers(page);
3825
3826         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3827                 if (!page_private_atomic(page)) {
3828                         f2fs_register_inmem_page(inode, page);
3829                         return 1;
3830                 }
3831                 /*
3832                  * Previously, this page has been registered, we just
3833                  * return here.
3834                  */
3835                 return 0;
3836         }
3837
3838         if (!PageDirty(page)) {
3839                 __set_page_dirty_nobuffers(page);
3840                 f2fs_update_dirty_page(inode, page);
3841                 return 1;
3842         }
3843         return 0;
3844 }
3845
3846
3847 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3848 {
3849 #ifdef CONFIG_F2FS_FS_COMPRESSION
3850         struct dnode_of_data dn;
3851         sector_t start_idx, blknr = 0;
3852         int ret;
3853
3854         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3855
3856         set_new_dnode(&dn, inode, NULL, NULL, 0);
3857         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3858         if (ret)
3859                 return 0;
3860
3861         if (dn.data_blkaddr != COMPRESS_ADDR) {
3862                 dn.ofs_in_node += block - start_idx;
3863                 blknr = f2fs_data_blkaddr(&dn);
3864                 if (!__is_valid_data_blkaddr(blknr))
3865                         blknr = 0;
3866         }
3867
3868         f2fs_put_dnode(&dn);
3869         return blknr;
3870 #else
3871         return 0;
3872 #endif
3873 }
3874
3875
3876 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3877 {
3878         struct inode *inode = mapping->host;
3879         sector_t blknr = 0;
3880
3881         if (f2fs_has_inline_data(inode))
3882                 goto out;
3883
3884         /* make sure allocating whole blocks */
3885         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3886                 filemap_write_and_wait(mapping);
3887
3888         /* Block number less than F2FS MAX BLOCKS */
3889         if (unlikely(block >= max_file_blocks(inode)))
3890                 goto out;
3891
3892         if (f2fs_compressed_file(inode)) {
3893                 blknr = f2fs_bmap_compress(inode, block);
3894         } else {
3895                 struct f2fs_map_blocks map;
3896
3897                 memset(&map, 0, sizeof(map));
3898                 map.m_lblk = block;
3899                 map.m_len = 1;
3900                 map.m_next_pgofs = NULL;
3901                 map.m_seg_type = NO_CHECK_TYPE;
3902
3903                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3904                         blknr = map.m_pblk;
3905         }
3906 out:
3907         trace_f2fs_bmap(inode, block, blknr);
3908         return blknr;
3909 }
3910
3911 #ifdef CONFIG_MIGRATION
3912 #include <linux/migrate.h>
3913
3914 int f2fs_migrate_page(struct address_space *mapping,
3915                 struct page *newpage, struct page *page, enum migrate_mode mode)
3916 {
3917         int rc, extra_count;
3918         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3919         bool atomic_written = page_private_atomic(page);
3920
3921         BUG_ON(PageWriteback(page));
3922
3923         /* migrating an atomic written page is safe with the inmem_lock hold */
3924         if (atomic_written) {
3925                 if (mode != MIGRATE_SYNC)
3926                         return -EBUSY;
3927                 if (!mutex_trylock(&fi->inmem_lock))
3928                         return -EAGAIN;
3929         }
3930
3931         /* one extra reference was held for atomic_write page */
3932         extra_count = atomic_written ? 1 : 0;
3933         rc = migrate_page_move_mapping(mapping, newpage,
3934                                 page, extra_count);
3935         if (rc != MIGRATEPAGE_SUCCESS) {
3936                 if (atomic_written)
3937                         mutex_unlock(&fi->inmem_lock);
3938                 return rc;
3939         }
3940
3941         if (atomic_written) {
3942                 struct inmem_pages *cur;
3943
3944                 list_for_each_entry(cur, &fi->inmem_pages, list)
3945                         if (cur->page == page) {
3946                                 cur->page = newpage;
3947                                 break;
3948                         }
3949                 mutex_unlock(&fi->inmem_lock);
3950                 put_page(page);
3951                 get_page(newpage);
3952         }
3953
3954         /* guarantee to start from no stale private field */
3955         set_page_private(newpage, 0);
3956         if (PagePrivate(page)) {
3957                 set_page_private(newpage, page_private(page));
3958                 SetPagePrivate(newpage);
3959                 get_page(newpage);
3960
3961                 set_page_private(page, 0);
3962                 ClearPagePrivate(page);
3963                 put_page(page);
3964         }
3965
3966         if (mode != MIGRATE_SYNC_NO_COPY)
3967                 migrate_page_copy(newpage, page);
3968         else
3969                 migrate_page_states(newpage, page);
3970
3971         return MIGRATEPAGE_SUCCESS;
3972 }
3973 #endif
3974
3975 #ifdef CONFIG_SWAP
3976 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3977                                                         unsigned int blkcnt)
3978 {
3979         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3980         unsigned int blkofs;
3981         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3982         unsigned int secidx = start_blk / blk_per_sec;
3983         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3984         int ret = 0;
3985
3986         down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3987         filemap_invalidate_lock(inode->i_mapping);
3988
3989         set_inode_flag(inode, FI_ALIGNED_WRITE);
3990
3991         for (; secidx < end_sec; secidx++) {
3992                 down_write(&sbi->pin_sem);
3993
3994                 f2fs_lock_op(sbi);
3995                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3996                 f2fs_unlock_op(sbi);
3997
3998                 set_inode_flag(inode, FI_DO_DEFRAG);
3999
4000                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
4001                         struct page *page;
4002                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
4003
4004                         page = f2fs_get_lock_data_page(inode, blkidx, true);
4005                         if (IS_ERR(page)) {
4006                                 up_write(&sbi->pin_sem);
4007                                 ret = PTR_ERR(page);
4008                                 goto done;
4009                         }
4010
4011                         set_page_dirty(page);
4012                         f2fs_put_page(page, 1);
4013                 }
4014
4015                 clear_inode_flag(inode, FI_DO_DEFRAG);
4016
4017                 ret = filemap_fdatawrite(inode->i_mapping);
4018
4019                 up_write(&sbi->pin_sem);
4020
4021                 if (ret)
4022                         break;
4023         }
4024
4025 done:
4026         clear_inode_flag(inode, FI_DO_DEFRAG);
4027         clear_inode_flag(inode, FI_ALIGNED_WRITE);
4028
4029         filemap_invalidate_unlock(inode->i_mapping);
4030         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
4031
4032         return ret;
4033 }
4034
4035 static int check_swap_activate(struct swap_info_struct *sis,
4036                                 struct file *swap_file, sector_t *span)
4037 {
4038         struct address_space *mapping = swap_file->f_mapping;
4039         struct inode *inode = mapping->host;
4040         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4041         sector_t cur_lblock;
4042         sector_t last_lblock;
4043         sector_t pblock;
4044         sector_t lowest_pblock = -1;
4045         sector_t highest_pblock = 0;
4046         int nr_extents = 0;
4047         unsigned long nr_pblocks;
4048         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
4049         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
4050         unsigned int not_aligned = 0;
4051         int ret = 0;
4052
4053         /*
4054          * Map all the blocks into the extent list.  This code doesn't try
4055          * to be very smart.
4056          */
4057         cur_lblock = 0;
4058         last_lblock = bytes_to_blks(inode, i_size_read(inode));
4059
4060         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
4061                 struct f2fs_map_blocks map;
4062 retry:
4063                 cond_resched();
4064
4065                 memset(&map, 0, sizeof(map));
4066                 map.m_lblk = cur_lblock;
4067                 map.m_len = last_lblock - cur_lblock;
4068                 map.m_next_pgofs = NULL;
4069                 map.m_next_extent = NULL;
4070                 map.m_seg_type = NO_CHECK_TYPE;
4071                 map.m_may_create = false;
4072
4073                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
4074                 if (ret)
4075                         goto out;
4076
4077                 /* hole */
4078                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
4079                         f2fs_err(sbi, "Swapfile has holes");
4080                         ret = -EINVAL;
4081                         goto out;
4082                 }
4083
4084                 pblock = map.m_pblk;
4085                 nr_pblocks = map.m_len;
4086
4087                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
4088                                 nr_pblocks & sec_blks_mask) {
4089                         not_aligned++;
4090
4091                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
4092                         if (cur_lblock + nr_pblocks > sis->max)
4093                                 nr_pblocks -= blks_per_sec;
4094
4095                         if (!nr_pblocks) {
4096                                 /* this extent is last one */
4097                                 nr_pblocks = map.m_len;
4098                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
4099                                 goto next;
4100                         }
4101
4102                         ret = f2fs_migrate_blocks(inode, cur_lblock,
4103                                                         nr_pblocks);
4104                         if (ret)
4105                                 goto out;
4106                         goto retry;
4107                 }
4108 next:
4109                 if (cur_lblock + nr_pblocks >= sis->max)
4110                         nr_pblocks = sis->max - cur_lblock;
4111
4112                 if (cur_lblock) {       /* exclude the header page */
4113                         if (pblock < lowest_pblock)
4114                                 lowest_pblock = pblock;
4115                         if (pblock + nr_pblocks - 1 > highest_pblock)
4116                                 highest_pblock = pblock + nr_pblocks - 1;
4117                 }
4118
4119                 /*
4120                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4121                  */
4122                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4123                 if (ret < 0)
4124                         goto out;
4125                 nr_extents += ret;
4126                 cur_lblock += nr_pblocks;
4127         }
4128         ret = nr_extents;
4129         *span = 1 + highest_pblock - lowest_pblock;
4130         if (cur_lblock == 0)
4131                 cur_lblock = 1; /* force Empty message */
4132         sis->max = cur_lblock;
4133         sis->pages = cur_lblock - 1;
4134         sis->highest_bit = cur_lblock - 1;
4135 out:
4136         if (not_aligned)
4137                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
4138                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
4139         return ret;
4140 }
4141
4142 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4143                                 sector_t *span)
4144 {
4145         struct inode *inode = file_inode(file);
4146         int ret;
4147
4148         if (!S_ISREG(inode->i_mode))
4149                 return -EINVAL;
4150
4151         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4152                 return -EROFS;
4153
4154         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4155                 f2fs_err(F2FS_I_SB(inode),
4156                         "Swapfile not supported in LFS mode");
4157                 return -EINVAL;
4158         }
4159
4160         ret = f2fs_convert_inline_inode(inode);
4161         if (ret)
4162                 return ret;
4163
4164         if (!f2fs_disable_compressed_file(inode))
4165                 return -EINVAL;
4166
4167         f2fs_precache_extents(inode);
4168
4169         ret = check_swap_activate(sis, file, span);
4170         if (ret < 0)
4171                 return ret;
4172
4173         set_inode_flag(inode, FI_PIN_FILE);
4174         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4175         return ret;
4176 }
4177
4178 static void f2fs_swap_deactivate(struct file *file)
4179 {
4180         struct inode *inode = file_inode(file);
4181
4182         clear_inode_flag(inode, FI_PIN_FILE);
4183 }
4184 #else
4185 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4186                                 sector_t *span)
4187 {
4188         return -EOPNOTSUPP;
4189 }
4190
4191 static void f2fs_swap_deactivate(struct file *file)
4192 {
4193 }
4194 #endif
4195
4196 const struct address_space_operations f2fs_dblock_aops = {
4197         .readpage       = f2fs_read_data_page,
4198         .readahead      = f2fs_readahead,
4199         .writepage      = f2fs_write_data_page,
4200         .writepages     = f2fs_write_data_pages,
4201         .write_begin    = f2fs_write_begin,
4202         .write_end      = f2fs_write_end,
4203         .set_page_dirty = f2fs_set_data_page_dirty,
4204         .invalidatepage = f2fs_invalidate_page,
4205         .releasepage    = f2fs_release_page,
4206         .direct_IO      = f2fs_direct_IO,
4207         .bmap           = f2fs_bmap,
4208         .swap_activate  = f2fs_swap_activate,
4209         .swap_deactivate = f2fs_swap_deactivate,
4210 #ifdef CONFIG_MIGRATION
4211         .migratepage    = f2fs_migrate_page,
4212 #endif
4213 };
4214
4215 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4216 {
4217         struct address_space *mapping = page_mapping(page);
4218         unsigned long flags;
4219
4220         xa_lock_irqsave(&mapping->i_pages, flags);
4221         __xa_clear_mark(&mapping->i_pages, page_index(page),
4222                                                 PAGECACHE_TAG_DIRTY);
4223         xa_unlock_irqrestore(&mapping->i_pages, flags);
4224 }
4225
4226 int __init f2fs_init_post_read_processing(void)
4227 {
4228         bio_post_read_ctx_cache =
4229                 kmem_cache_create("f2fs_bio_post_read_ctx",
4230                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4231         if (!bio_post_read_ctx_cache)
4232                 goto fail;
4233         bio_post_read_ctx_pool =
4234                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4235                                          bio_post_read_ctx_cache);
4236         if (!bio_post_read_ctx_pool)
4237                 goto fail_free_cache;
4238         return 0;
4239
4240 fail_free_cache:
4241         kmem_cache_destroy(bio_post_read_ctx_cache);
4242 fail:
4243         return -ENOMEM;
4244 }
4245
4246 void f2fs_destroy_post_read_processing(void)
4247 {
4248         mempool_destroy(bio_post_read_ctx_pool);
4249         kmem_cache_destroy(bio_post_read_ctx_cache);
4250 }
4251
4252 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4253 {
4254         if (!f2fs_sb_has_encrypt(sbi) &&
4255                 !f2fs_sb_has_verity(sbi) &&
4256                 !f2fs_sb_has_compression(sbi))
4257                 return 0;
4258
4259         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4260                                                  WQ_UNBOUND | WQ_HIGHPRI,
4261                                                  num_online_cpus());
4262         if (!sbi->post_read_wq)
4263                 return -ENOMEM;
4264         return 0;
4265 }
4266
4267 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4268 {
4269         if (sbi->post_read_wq)
4270                 destroy_workqueue(sbi->post_read_wq);
4271 }
4272
4273 int __init f2fs_init_bio_entry_cache(void)
4274 {
4275         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4276                         sizeof(struct bio_entry));
4277         if (!bio_entry_slab)
4278                 return -ENOMEM;
4279         return 0;
4280 }
4281
4282 void f2fs_destroy_bio_entry_cache(void)
4283 {
4284         kmem_cache_destroy(bio_entry_slab);
4285 }