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 "trace.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 inline struct bio *__f2fs_bio_alloc(gfp_t gfp_mask,
  54                                                 unsigned int nr_iovecs)
  55 {
  56         return bio_alloc_bioset(gfp_mask, nr_iovecs, &f2fs_bioset);
  57 }
  58
  59 struct bio *f2fs_bio_alloc(struct f2fs_sb_info *sbi, int npages, bool noio)
  60 {
  61         if (noio) {
  62                 /* No failure on bio allocation */
  63                 return __f2fs_bio_alloc(GFP_NOIO, npages);
  64         }
  65
  66         if (time_to_inject(sbi, FAULT_ALLOC_BIO)) {
  67                 f2fs_show_injection_info(sbi, FAULT_ALLOC_BIO);
  68                 return NULL;
  69         }
  70
  71         return __f2fs_bio_alloc(GFP_KERNEL, npages);
  72 }
  73
  74 static bool __is_cp_guaranteed(struct page *page)
  75 {
  76         struct address_space *mapping = page->mapping;
  77         struct inode *inode;
  78         struct f2fs_sb_info *sbi;
  79
  80         if (!mapping)
  81                 return false;
  82
  83         if (f2fs_is_compressed_page(page))
  84                 return false;
  85
  86         inode = mapping->host;
  87         sbi = F2FS_I_SB(inode);
  88
  89         if (inode->i_ino == F2FS_META_INO(sbi) ||
  90                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  91                         S_ISDIR(inode->i_mode) ||
  92                         (S_ISREG(inode->i_mode) &&
  93                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  94                         is_cold_data(page))
  95                 return true;
  96         return false;
  97 }
  98
  99 static enum count_type __read_io_type(struct page *page)
 100 {
 101         struct address_space *mapping = page_file_mapping(page);
 102
 103         if (mapping) {
 104                 struct inode *inode = mapping->host;
 105                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 106
 107                 if (inode->i_ino == F2FS_META_INO(sbi))
 108                         return F2FS_RD_META;
 109
 110                 if (inode->i_ino == F2FS_NODE_INO(sbi))
 111                         return F2FS_RD_NODE;
 112         }
 113         return F2FS_RD_DATA;
 114 }
 115
 116 /* postprocessing steps for read bios */
 117 enum bio_post_read_step {
 118         STEP_DECRYPT,
 119         STEP_DECOMPRESS_NOWQ,           /* handle normal cluster data inplace */
 120         STEP_DECOMPRESS,                /* handle compressed cluster data in workqueue */
 121         STEP_VERITY,
 122 };
 123
 124 struct bio_post_read_ctx {
 125         struct bio *bio;
 126         struct f2fs_sb_info *sbi;
 127         struct work_struct work;
 128         unsigned int enabled_steps;
 129 };
 130
 131 static void __read_end_io(struct bio *bio, bool compr, bool verity)
 132 {
 133         struct page *page;
 134         struct bio_vec *bv;
 135         struct bvec_iter_all iter_all;
 136
 137         bio_for_each_segment_all(bv, bio, iter_all) {
 138                 page = bv->bv_page;
 139
 140 #ifdef CONFIG_F2FS_FS_COMPRESSION
 141                 if (compr && f2fs_is_compressed_page(page)) {
 142                         f2fs_decompress_pages(bio, page, verity);
 143                         continue;
 144                 }
 145                 if (verity)
 146                         continue;
 147 #endif
 148
 149                 /* PG_error was set if any post_read step failed */
 150                 if (bio->bi_status || PageError(page)) {
 151                         ClearPageUptodate(page);
 152                         /* will re-read again later */
 153                         ClearPageError(page);
 154                 } else {
 155                         SetPageUptodate(page);
 156                 }
 157                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 158                 unlock_page(page);
 159         }
 160 }
 161
 162 static void f2fs_release_read_bio(struct bio *bio);
 163 static void __f2fs_read_end_io(struct bio *bio, bool compr, bool verity)
 164 {
 165         if (!compr)
 166                 __read_end_io(bio, false, verity);
 167         f2fs_release_read_bio(bio);
 168 }
 169
 170 static void f2fs_decompress_bio(struct bio *bio, bool verity)
 171 {
 172         __read_end_io(bio, true, verity);
 173 }
 174
 175 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 176
 177 static void f2fs_decrypt_work(struct bio_post_read_ctx *ctx)
 178 {
 179         fscrypt_decrypt_bio(ctx->bio);
 180 }
 181
 182 static void f2fs_decompress_work(struct bio_post_read_ctx *ctx)
 183 {
 184         f2fs_decompress_bio(ctx->bio, ctx->enabled_steps & (1 << STEP_VERITY));
 185 }
 186
 187 #ifdef CONFIG_F2FS_FS_COMPRESSION
 188 static void f2fs_verify_pages(struct page **rpages, unsigned int cluster_size)
 189 {
 190         f2fs_decompress_end_io(rpages, cluster_size, false, true);
 191 }
 192
 193 static void f2fs_verify_bio(struct bio *bio)
 194 {
 195         struct bio_vec *bv;
 196         struct bvec_iter_all iter_all;
 197
 198         bio_for_each_segment_all(bv, bio, iter_all) {
 199                 struct page *page = bv->bv_page;
 200                 struct decompress_io_ctx *dic;
 201
 202                 dic = (struct decompress_io_ctx *)page_private(page);
 203
 204                 if (dic) {
 205                         if (atomic_dec_return(&dic->pending_pages))
 206                                 continue;
 207                         f2fs_verify_pages(dic->rpages,
 208                                                 dic->cluster_size);
 209                         f2fs_free_dic(dic);
 210                         continue;
 211                 }
 212
 213                 if (bio->bi_status || PageError(page))
 214                         goto clear_uptodate;
 215
 216                 if (fsverity_verify_page(page)) {
 217                         SetPageUptodate(page);
 218                         goto unlock;
 219                 }
 220 clear_uptodate:
 221                 ClearPageUptodate(page);
 222                 ClearPageError(page);
 223 unlock:
 224                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 225                 unlock_page(page);
 226         }
 227 }
 228 #endif
 229
 230 static void f2fs_verity_work(struct work_struct *work)
 231 {
 232         struct bio_post_read_ctx *ctx =
 233                 container_of(work, struct bio_post_read_ctx, work);
 234         struct bio *bio = ctx->bio;
 235 #ifdef CONFIG_F2FS_FS_COMPRESSION
 236         unsigned int enabled_steps = ctx->enabled_steps;
 237 #endif
 238
 239         /*
 240          * fsverity_verify_bio() may call readpages() again, and while verity
 241          * will be disabled for this, decryption may still be needed, resulting
 242          * in another bio_post_read_ctx being allocated.  So to prevent
 243          * deadlocks we need to release the current ctx to the mempool first.
 244          * This assumes that verity is the last post-read step.
 245          */
 246         mempool_free(ctx, bio_post_read_ctx_pool);
 247         bio->bi_private = NULL;
 248
 249 #ifdef CONFIG_F2FS_FS_COMPRESSION
 250         /* previous step is decompression */
 251         if (enabled_steps & (1 << STEP_DECOMPRESS)) {
 252                 f2fs_verify_bio(bio);
 253                 f2fs_release_read_bio(bio);
 254                 return;
 255         }
 256 #endif
 257
 258         fsverity_verify_bio(bio);
 259         __f2fs_read_end_io(bio, false, false);
 260 }
 261
 262 static void f2fs_post_read_work(struct work_struct *work)
 263 {
 264         struct bio_post_read_ctx *ctx =
 265                 container_of(work, struct bio_post_read_ctx, work);
 266
 267         if (ctx->enabled_steps & (1 << STEP_DECRYPT))
 268                 f2fs_decrypt_work(ctx);
 269
 270         if (ctx->enabled_steps & (1 << STEP_DECOMPRESS))
 271                 f2fs_decompress_work(ctx);
 272
 273         if (ctx->enabled_steps & (1 << STEP_VERITY)) {
 274                 INIT_WORK(&ctx->work, f2fs_verity_work);
 275                 fsverity_enqueue_verify_work(&ctx->work);
 276                 return;
 277         }
 278
 279         __f2fs_read_end_io(ctx->bio,
 280                 ctx->enabled_steps & (1 << STEP_DECOMPRESS), false);
 281 }
 282
 283 static void f2fs_enqueue_post_read_work(struct f2fs_sb_info *sbi,
 284                                                 struct work_struct *work)
 285 {
 286         queue_work(sbi->post_read_wq, work);
 287 }
 288
 289 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 290 {
 291         /*
 292          * We use different work queues for decryption and for verity because
 293          * verity may require reading metadata pages that need decryption, and
 294          * we shouldn't recurse to the same workqueue.
 295          */
 296
 297         if (ctx->enabled_steps & (1 << STEP_DECRYPT) ||
 298                 ctx->enabled_steps & (1 << STEP_DECOMPRESS)) {
 299                 INIT_WORK(&ctx->work, f2fs_post_read_work);
 300                 f2fs_enqueue_post_read_work(ctx->sbi, &ctx->work);
 301                 return;
 302         }
 303
 304         if (ctx->enabled_steps & (1 << STEP_VERITY)) {
 305                 INIT_WORK(&ctx->work, f2fs_verity_work);
 306                 fsverity_enqueue_verify_work(&ctx->work);
 307                 return;
 308         }
 309
 310         __f2fs_read_end_io(ctx->bio, false, false);
 311 }
 312
 313 static bool f2fs_bio_post_read_required(struct bio *bio)
 314 {
 315         return bio->bi_private;
 316 }
 317
 318 static void f2fs_read_end_io(struct bio *bio)
 319 {
 320         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 321
 322         if (time_to_inject(sbi, FAULT_READ_IO)) {
 323                 f2fs_show_injection_info(sbi, FAULT_READ_IO);
 324                 bio->bi_status = BLK_STS_IOERR;
 325         }
 326
 327         if (f2fs_bio_post_read_required(bio)) {
 328                 struct bio_post_read_ctx *ctx = bio->bi_private;
 329
 330                 bio_post_read_processing(ctx);
 331                 return;
 332         }
 333
 334         __f2fs_read_end_io(bio, false, false);
 335 }
 336
 337 static void f2fs_write_end_io(struct bio *bio)
 338 {
 339         struct f2fs_sb_info *sbi = bio->bi_private;
 340         struct bio_vec *bvec;
 341         struct bvec_iter_all iter_all;
 342
 343         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 344                 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
 345                 bio->bi_status = BLK_STS_IOERR;
 346         }
 347
 348         bio_for_each_segment_all(bvec, bio, iter_all) {
 349                 struct page *page = bvec->bv_page;
 350                 enum count_type type = WB_DATA_TYPE(page);
 351
 352                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 353                         set_page_private(page, (unsigned long)NULL);
 354                         ClearPagePrivate(page);
 355                         unlock_page(page);
 356                         mempool_free(page, sbi->write_io_dummy);
 357
 358                         if (unlikely(bio->bi_status))
 359                                 f2fs_stop_checkpoint(sbi, true);
 360                         continue;
 361                 }
 362
 363                 fscrypt_finalize_bounce_page(&page);
 364
 365 #ifdef CONFIG_F2FS_FS_COMPRESSION
 366                 if (f2fs_is_compressed_page(page)) {
 367                         f2fs_compress_write_end_io(bio, page);
 368                         continue;
 369                 }
 370 #endif
 371
 372                 if (unlikely(bio->bi_status)) {
 373                         mapping_set_error(page->mapping, -EIO);
 374                         if (type == F2FS_WB_CP_DATA)
 375                                 f2fs_stop_checkpoint(sbi, true);
 376                 }
 377
 378                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 379                                         page->index != nid_of_node(page));
 380
 381                 dec_page_count(sbi, type);
 382                 if (f2fs_in_warm_node_list(sbi, page))
 383                         f2fs_del_fsync_node_entry(sbi, page);
 384                 clear_cold_data(page);
 385                 end_page_writeback(page);
 386         }
 387         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 388                                 wq_has_sleeper(&sbi->cp_wait))
 389                 wake_up(&sbi->cp_wait);
 390
 391         bio_put(bio);
 392 }
 393
 394 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 395                                 block_t blk_addr, struct bio *bio)
 396 {
 397         struct block_device *bdev = sbi->sb->s_bdev;
 398         int i;
 399
 400         if (f2fs_is_multi_device(sbi)) {
 401                 for (i = 0; i < sbi->s_ndevs; i++) {
 402                         if (FDEV(i).start_blk <= blk_addr &&
 403                             FDEV(i).end_blk >= blk_addr) {
 404                                 blk_addr -= FDEV(i).start_blk;
 405                                 bdev = FDEV(i).bdev;
 406                                 break;
 407                         }
 408                 }
 409         }
 410         if (bio) {
 411                 bio_set_dev(bio, bdev);
 412                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 413         }
 414         return bdev;
 415 }
 416
 417 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 418 {
 419         int i;
 420
 421         if (!f2fs_is_multi_device(sbi))
 422                 return 0;
 423
 424         for (i = 0; i < sbi->s_ndevs; i++)
 425                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 426                         return i;
 427         return 0;
 428 }
 429
 430 /*
 431  * Return true, if pre_bio's bdev is same as its target device.
 432  */
 433 static bool __same_bdev(struct f2fs_sb_info *sbi,
 434                                 block_t blk_addr, struct bio *bio)
 435 {
 436         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 437         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 438 }
 439
 440 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 441 {
 442         struct f2fs_sb_info *sbi = fio->sbi;
 443         struct bio *bio;
 444
 445         bio = f2fs_bio_alloc(sbi, npages, true);
 446
 447         f2fs_target_device(sbi, fio->new_blkaddr, bio);
 448         if (is_read_io(fio->op)) {
 449                 bio->bi_end_io = f2fs_read_end_io;
 450                 bio->bi_private = NULL;
 451         } else {
 452                 bio->bi_end_io = f2fs_write_end_io;
 453                 bio->bi_private = sbi;
 454                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 455                                                 fio->type, fio->temp);
 456         }
 457         if (fio->io_wbc)
 458                 wbc_init_bio(fio->io_wbc, bio);
 459
 460         return bio;
 461 }
 462
 463 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 464                                   pgoff_t first_idx,
 465                                   const struct f2fs_io_info *fio,
 466                                   gfp_t gfp_mask)
 467 {
 468         /*
 469          * The f2fs garbage collector sets ->encrypted_page when it wants to
 470          * read/write raw data without encryption.
 471          */
 472         if (!fio || !fio->encrypted_page)
 473                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 474 }
 475
 476 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 477                                      pgoff_t next_idx,
 478                                      const struct f2fs_io_info *fio)
 479 {
 480         /*
 481          * The f2fs garbage collector sets ->encrypted_page when it wants to
 482          * read/write raw data without encryption.
 483          */
 484         if (fio && fio->encrypted_page)
 485                 return !bio_has_crypt_ctx(bio);
 486
 487         return fscrypt_mergeable_bio(bio, inode, next_idx);
 488 }
 489
 490 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 491                                 struct bio *bio, enum page_type type)
 492 {
 493         if (!is_read_io(bio_op(bio))) {
 494                 unsigned int start;
 495
 496                 if (type != DATA && type != NODE)
 497                         goto submit_io;
 498
 499                 if (f2fs_lfs_mode(sbi) && current->plug)
 500                         blk_finish_plug(current->plug);
 501
 502                 if (F2FS_IO_ALIGNED(sbi))
 503                         goto submit_io;
 504
 505                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 506                 start %= F2FS_IO_SIZE(sbi);
 507
 508                 if (start == 0)
 509                         goto submit_io;
 510
 511                 /* fill dummy pages */
 512                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 513                         struct page *page =
 514                                 mempool_alloc(sbi->write_io_dummy,
 515                                               GFP_NOIO | __GFP_NOFAIL);
 516                         f2fs_bug_on(sbi, !page);
 517
 518                         zero_user_segment(page, 0, PAGE_SIZE);
 519                         SetPagePrivate(page);
 520                         set_page_private(page, DUMMY_WRITTEN_PAGE);
 521                         lock_page(page);
 522                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 523                                 f2fs_bug_on(sbi, 1);
 524                 }
 525                 /*
 526                  * In the NODE case, we lose next block address chain. So, we
 527                  * need to do checkpoint in f2fs_sync_file.
 528                  */
 529                 if (type == NODE)
 530                         set_sbi_flag(sbi, SBI_NEED_CP);
 531         }
 532 submit_io:
 533         if (is_read_io(bio_op(bio)))
 534                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 535         else
 536                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 537         submit_bio(bio);
 538 }
 539
 540 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 541                                 struct bio *bio, enum page_type type)
 542 {
 543         __submit_bio(sbi, bio, type);
 544 }
 545
 546 static void __attach_io_flag(struct f2fs_io_info *fio)
 547 {
 548         struct f2fs_sb_info *sbi = fio->sbi;
 549         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 550         unsigned int io_flag, fua_flag, meta_flag;
 551
 552         if (fio->type == DATA)
 553                 io_flag = sbi->data_io_flag;
 554         else if (fio->type == NODE)
 555                 io_flag = sbi->node_io_flag;
 556         else
 557                 return;
 558
 559         fua_flag = io_flag & temp_mask;
 560         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 561
 562         /*
 563          * data/node io flag bits per temp:
 564          *      REQ_META     |      REQ_FUA      |
 565          *    5 |    4 |   3 |    2 |    1 |   0 |
 566          * Cold | Warm | Hot | Cold | Warm | Hot |
 567          */
 568         if ((1 << fio->temp) & meta_flag)
 569                 fio->op_flags |= REQ_META;
 570         if ((1 << fio->temp) & fua_flag)
 571                 fio->op_flags |= REQ_FUA;
 572 }
 573
 574 static void __submit_merged_bio(struct f2fs_bio_info *io)
 575 {
 576         struct f2fs_io_info *fio = &io->fio;
 577
 578         if (!io->bio)
 579                 return;
 580
 581         __attach_io_flag(fio);
 582         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 583
 584         if (is_read_io(fio->op))
 585                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 586         else
 587                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 588
 589         __submit_bio(io->sbi, io->bio, fio->type);
 590         io->bio = NULL;
 591 }
 592
 593 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 594                                                 struct page *page, nid_t ino)
 595 {
 596         struct bio_vec *bvec;
 597         struct bvec_iter_all iter_all;
 598
 599         if (!bio)
 600                 return false;
 601
 602         if (!inode && !page && !ino)
 603                 return true;
 604
 605         bio_for_each_segment_all(bvec, bio, iter_all) {
 606                 struct page *target = bvec->bv_page;
 607
 608                 if (fscrypt_is_bounce_page(target)) {
 609                         target = fscrypt_pagecache_page(target);
 610                         if (IS_ERR(target))
 611                                 continue;
 612                 }
 613                 if (f2fs_is_compressed_page(target)) {
 614                         target = f2fs_compress_control_page(target);
 615                         if (IS_ERR(target))
 616                                 continue;
 617                 }
 618
 619                 if (inode && inode == target->mapping->host)
 620                         return true;
 621                 if (page && page == target)
 622                         return true;
 623                 if (ino && ino == ino_of_node(target))
 624                         return true;
 625         }
 626
 627         return false;
 628 }
 629
 630 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 631                                 enum page_type type, enum temp_type temp)
 632 {
 633         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 634         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 635
 636         down_write(&io->io_rwsem);
 637
 638         /* change META to META_FLUSH in the checkpoint procedure */
 639         if (type >= META_FLUSH) {
 640                 io->fio.type = META_FLUSH;
 641                 io->fio.op = REQ_OP_WRITE;
 642                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 643                 if (!test_opt(sbi, NOBARRIER))
 644                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 645         }
 646         __submit_merged_bio(io);
 647         up_write(&io->io_rwsem);
 648 }
 649
 650 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 651                                 struct inode *inode, struct page *page,
 652                                 nid_t ino, enum page_type type, bool force)
 653 {
 654         enum temp_type temp;
 655         bool ret = true;
 656
 657         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 658                 if (!force)     {
 659                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 660                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 661
 662                         down_read(&io->io_rwsem);
 663                         ret = __has_merged_page(io->bio, inode, page, ino);
 664                         up_read(&io->io_rwsem);
 665                 }
 666                 if (ret)
 667                         __f2fs_submit_merged_write(sbi, type, temp);
 668
 669                 /* TODO: use HOT temp only for meta pages now. */
 670                 if (type >= META)
 671                         break;
 672         }
 673 }
 674
 675 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 676 {
 677         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 678 }
 679
 680 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 681                                 struct inode *inode, struct page *page,
 682                                 nid_t ino, enum page_type type)
 683 {
 684         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 685 }
 686
 687 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 688 {
 689         f2fs_submit_merged_write(sbi, DATA);
 690         f2fs_submit_merged_write(sbi, NODE);
 691         f2fs_submit_merged_write(sbi, META);
 692 }
 693
 694 /*
 695  * Fill the locked page with data located in the block address.
 696  * A caller needs to unlock the page on failure.
 697  */
 698 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 699 {
 700         struct bio *bio;
 701         struct page *page = fio->encrypted_page ?
 702                         fio->encrypted_page : fio->page;
 703
 704         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 705                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 706                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 707                 return -EFSCORRUPTED;
 708
 709         trace_f2fs_submit_page_bio(page, fio);
 710         f2fs_trace_ios(fio, 0);
 711
 712         /* Allocate a new bio */
 713         bio = __bio_alloc(fio, 1);
 714
 715         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 716                                fio->page->index, fio, GFP_NOIO);
 717
 718         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 719                 bio_put(bio);
 720                 return -EFAULT;
 721         }
 722
 723         if (fio->io_wbc && !is_read_io(fio->op))
 724                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 725
 726         __attach_io_flag(fio);
 727         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 728
 729         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 730                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 731
 732         __submit_bio(fio->sbi, bio, fio->type);
 733         return 0;
 734 }
 735
 736 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 737                                 block_t last_blkaddr, block_t cur_blkaddr)
 738 {
 739         if (last_blkaddr + 1 != cur_blkaddr)
 740                 return false;
 741         return __same_bdev(sbi, cur_blkaddr, bio);
 742 }
 743
 744 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 745                                                 struct f2fs_io_info *fio)
 746 {
 747         if (io->fio.op != fio->op)
 748                 return false;
 749         return io->fio.op_flags == fio->op_flags;
 750 }
 751
 752 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 753                                         struct f2fs_bio_info *io,
 754                                         struct f2fs_io_info *fio,
 755                                         block_t last_blkaddr,
 756                                         block_t cur_blkaddr)
 757 {
 758         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 759                 unsigned int filled_blocks =
 760                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 761                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 762                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 763
 764                 /* IOs in bio is aligned and left space of vectors is not enough */
 765                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 766                         return false;
 767         }
 768         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 769                 return false;
 770         return io_type_is_mergeable(io, fio);
 771 }
 772
 773 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 774                                 struct page *page, enum temp_type temp)
 775 {
 776         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 777         struct bio_entry *be;
 778
 779         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 780         be->bio = bio;
 781         bio_get(bio);
 782
 783         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 784                 f2fs_bug_on(sbi, 1);
 785
 786         down_write(&io->bio_list_lock);
 787         list_add_tail(&be->list, &io->bio_list);
 788         up_write(&io->bio_list_lock);
 789 }
 790
 791 static void del_bio_entry(struct bio_entry *be)
 792 {
 793         list_del(&be->list);
 794         kmem_cache_free(bio_entry_slab, be);
 795 }
 796
 797 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 798                                                         struct page *page)
 799 {
 800         struct f2fs_sb_info *sbi = fio->sbi;
 801         enum temp_type temp;
 802         bool found = false;
 803         int ret = -EAGAIN;
 804
 805         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 806                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 807                 struct list_head *head = &io->bio_list;
 808                 struct bio_entry *be;
 809
 810                 down_write(&io->bio_list_lock);
 811                 list_for_each_entry(be, head, list) {
 812                         if (be->bio != *bio)
 813                                 continue;
 814
 815                         found = true;
 816
 817                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 818                                                             *fio->last_block,
 819                                                             fio->new_blkaddr));
 820                         if (f2fs_crypt_mergeable_bio(*bio,
 821                                         fio->page->mapping->host,
 822                                         fio->page->index, fio) &&
 823                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 824                                         PAGE_SIZE) {
 825                                 ret = 0;
 826                                 break;
 827                         }
 828
 829                         /* page can't be merged into bio; submit the bio */
 830                         del_bio_entry(be);
 831                         __submit_bio(sbi, *bio, DATA);
 832                         break;
 833                 }
 834                 up_write(&io->bio_list_lock);
 835         }
 836
 837         if (ret) {
 838                 bio_put(*bio);
 839                 *bio = NULL;
 840         }
 841
 842         return ret;
 843 }
 844
 845 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 846                                         struct bio **bio, struct page *page)
 847 {
 848         enum temp_type temp;
 849         bool found = false;
 850         struct bio *target = bio ? *bio : NULL;
 851
 852         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 853                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 854                 struct list_head *head = &io->bio_list;
 855                 struct bio_entry *be;
 856
 857                 if (list_empty(head))
 858                         continue;
 859
 860                 down_read(&io->bio_list_lock);
 861                 list_for_each_entry(be, head, list) {
 862                         if (target)
 863                                 found = (target == be->bio);
 864                         else
 865                                 found = __has_merged_page(be->bio, NULL,
 866                                                                 page, 0);
 867                         if (found)
 868                                 break;
 869                 }
 870                 up_read(&io->bio_list_lock);
 871
 872                 if (!found)
 873                         continue;
 874
 875                 found = false;
 876
 877                 down_write(&io->bio_list_lock);
 878                 list_for_each_entry(be, head, list) {
 879                         if (target)
 880                                 found = (target == be->bio);
 881                         else
 882                                 found = __has_merged_page(be->bio, NULL,
 883                                                                 page, 0);
 884                         if (found) {
 885                                 target = be->bio;
 886                                 del_bio_entry(be);
 887                                 break;
 888                         }
 889                 }
 890                 up_write(&io->bio_list_lock);
 891         }
 892
 893         if (found)
 894                 __submit_bio(sbi, target, DATA);
 895         if (bio && *bio) {
 896                 bio_put(*bio);
 897                 *bio = NULL;
 898         }
 899 }
 900
 901 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 902 {
 903         struct bio *bio = *fio->bio;
 904         struct page *page = fio->encrypted_page ?
 905                         fio->encrypted_page : fio->page;
 906
 907         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 908                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 909                 return -EFSCORRUPTED;
 910
 911         trace_f2fs_submit_page_bio(page, fio);
 912         f2fs_trace_ios(fio, 0);
 913
 914         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 915                                                 fio->new_blkaddr))
 916                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 917 alloc_new:
 918         if (!bio) {
 919                 bio = __bio_alloc(fio, BIO_MAX_PAGES);
 920                 __attach_io_flag(fio);
 921                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 922                                        fio->page->index, fio, GFP_NOIO);
 923                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 924
 925                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 926         } else {
 927                 if (add_ipu_page(fio, &bio, page))
 928                         goto alloc_new;
 929         }
 930
 931         if (fio->io_wbc)
 932                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 933
 934         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 935
 936         *fio->last_block = fio->new_blkaddr;
 937         *fio->bio = bio;
 938
 939         return 0;
 940 }
 941
 942 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 943 {
 944         struct f2fs_sb_info *sbi = fio->sbi;
 945         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 946         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 947         struct page *bio_page;
 948
 949         f2fs_bug_on(sbi, is_read_io(fio->op));
 950
 951         down_write(&io->io_rwsem);
 952 next:
 953         if (fio->in_list) {
 954                 spin_lock(&io->io_lock);
 955                 if (list_empty(&io->io_list)) {
 956                         spin_unlock(&io->io_lock);
 957                         goto out;
 958                 }
 959                 fio = list_first_entry(&io->io_list,
 960                                                 struct f2fs_io_info, list);
 961                 list_del(&fio->list);
 962                 spin_unlock(&io->io_lock);
 963         }
 964
 965         verify_fio_blkaddr(fio);
 966
 967         if (fio->encrypted_page)
 968                 bio_page = fio->encrypted_page;
 969         else if (fio->compressed_page)
 970                 bio_page = fio->compressed_page;
 971         else
 972                 bio_page = fio->page;
 973
 974         /* set submitted = true as a return value */
 975         fio->submitted = true;
 976
 977         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 978
 979         if (io->bio &&
 980             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 981                               fio->new_blkaddr) ||
 982              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 983                                        bio_page->index, fio)))
 984                 __submit_merged_bio(io);
 985 alloc_new:
 986         if (io->bio == NULL) {
 987                 if (F2FS_IO_ALIGNED(sbi) &&
 988                                 (fio->type == DATA || fio->type == NODE) &&
 989                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 990                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 991                         fio->retry = true;
 992                         goto skip;
 993                 }
 994                 io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
 995                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 996                                        bio_page->index, fio, GFP_NOIO);
 997                 io->fio = *fio;
 998         }
 999
1000         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1001                 __submit_merged_bio(io);
1002                 goto alloc_new;
1003         }
1004
1005         if (fio->io_wbc)
1006                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
1007
1008         io->last_block_in_bio = fio->new_blkaddr;
1009         f2fs_trace_ios(fio, 0);
1010
1011         trace_f2fs_submit_page_write(fio->page, fio);
1012 skip:
1013         if (fio->in_list)
1014                 goto next;
1015 out:
1016         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1017                                 !f2fs_is_checkpoint_ready(sbi))
1018                 __submit_merged_bio(io);
1019         up_write(&io->io_rwsem);
1020 }
1021
1022 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
1023 {
1024         return fsverity_active(inode) &&
1025                idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
1026 }
1027
1028 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1029                                       unsigned nr_pages, unsigned op_flag,
1030                                       pgoff_t first_idx, bool for_write)
1031 {
1032         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1033         struct bio *bio;
1034         struct bio_post_read_ctx *ctx;
1035         unsigned int post_read_steps = 0;
1036
1037         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES),
1038                                                                 for_write);
1039         if (!bio)
1040                 return ERR_PTR(-ENOMEM);
1041
1042         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1043
1044         f2fs_target_device(sbi, blkaddr, bio);
1045         bio->bi_end_io = f2fs_read_end_io;
1046         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
1047
1048         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1049                 post_read_steps |= 1 << STEP_DECRYPT;
1050         if (f2fs_compressed_file(inode))
1051                 post_read_steps |= 1 << STEP_DECOMPRESS_NOWQ;
1052         if (f2fs_need_verity(inode, first_idx))
1053                 post_read_steps |= 1 << STEP_VERITY;
1054
1055         if (post_read_steps) {
1056                 /* Due to the mempool, this never fails. */
1057                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1058                 ctx->bio = bio;
1059                 ctx->sbi = sbi;
1060                 ctx->enabled_steps = post_read_steps;
1061                 bio->bi_private = ctx;
1062         }
1063
1064         return bio;
1065 }
1066
1067 static void f2fs_release_read_bio(struct bio *bio)
1068 {
1069         if (bio->bi_private)
1070                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1071         bio_put(bio);
1072 }
1073
1074 /* This can handle encryption stuffs */
1075 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1076                                  block_t blkaddr, int op_flags, bool for_write)
1077 {
1078         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1079         struct bio *bio;
1080
1081         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1082                                         page->index, for_write);
1083         if (IS_ERR(bio))
1084                 return PTR_ERR(bio);
1085
1086         /* wait for GCed page writeback via META_MAPPING */
1087         f2fs_wait_on_block_writeback(inode, blkaddr);
1088
1089         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1090                 bio_put(bio);
1091                 return -EFAULT;
1092         }
1093         ClearPageError(page);
1094         inc_page_count(sbi, F2FS_RD_DATA);
1095         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1096         __submit_bio(sbi, bio, DATA);
1097         return 0;
1098 }
1099
1100 static void __set_data_blkaddr(struct dnode_of_data *dn)
1101 {
1102         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1103         __le32 *addr_array;
1104         int base = 0;
1105
1106         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1107                 base = get_extra_isize(dn->inode);
1108
1109         /* Get physical address of data block */
1110         addr_array = blkaddr_in_node(rn);
1111         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1112 }
1113
1114 /*
1115  * Lock ordering for the change of data block address:
1116  * ->data_page
1117  *  ->node_page
1118  *    update block addresses in the node page
1119  */
1120 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1121 {
1122         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1123         __set_data_blkaddr(dn);
1124         if (set_page_dirty(dn->node_page))
1125                 dn->node_changed = true;
1126 }
1127
1128 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1129 {
1130         dn->data_blkaddr = blkaddr;
1131         f2fs_set_data_blkaddr(dn);
1132         f2fs_update_extent_cache(dn);
1133 }
1134
1135 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1136 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1137 {
1138         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1139         int err;
1140
1141         if (!count)
1142                 return 0;
1143
1144         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1145                 return -EPERM;
1146         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1147                 return err;
1148
1149         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1150                                                 dn->ofs_in_node, count);
1151
1152         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1153
1154         for (; count > 0; dn->ofs_in_node++) {
1155                 block_t blkaddr = f2fs_data_blkaddr(dn);
1156                 if (blkaddr == NULL_ADDR) {
1157                         dn->data_blkaddr = NEW_ADDR;
1158                         __set_data_blkaddr(dn);
1159                         count--;
1160                 }
1161         }
1162
1163         if (set_page_dirty(dn->node_page))
1164                 dn->node_changed = true;
1165         return 0;
1166 }
1167
1168 /* Should keep dn->ofs_in_node unchanged */
1169 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1170 {
1171         unsigned int ofs_in_node = dn->ofs_in_node;
1172         int ret;
1173
1174         ret = f2fs_reserve_new_blocks(dn, 1);
1175         dn->ofs_in_node = ofs_in_node;
1176         return ret;
1177 }
1178
1179 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1180 {
1181         bool need_put = dn->inode_page ? false : true;
1182         int err;
1183
1184         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1185         if (err)
1186                 return err;
1187
1188         if (dn->data_blkaddr == NULL_ADDR)
1189                 err = f2fs_reserve_new_block(dn);
1190         if (err || need_put)
1191                 f2fs_put_dnode(dn);
1192         return err;
1193 }
1194
1195 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1196 {
1197         struct extent_info ei = {0, 0, 0};
1198         struct inode *inode = dn->inode;
1199
1200         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1201                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1202                 return 0;
1203         }
1204
1205         return f2fs_reserve_block(dn, index);
1206 }
1207
1208 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1209                                                 int op_flags, bool for_write)
1210 {
1211         struct address_space *mapping = inode->i_mapping;
1212         struct dnode_of_data dn;
1213         struct page *page;
1214         struct extent_info ei = {0,0,0};
1215         int err;
1216
1217         page = f2fs_grab_cache_page(mapping, index, for_write);
1218         if (!page)
1219                 return ERR_PTR(-ENOMEM);
1220
1221         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1222                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1223                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1224                                                 DATA_GENERIC_ENHANCE_READ)) {
1225                         err = -EFSCORRUPTED;
1226                         goto put_err;
1227                 }
1228                 goto got_it;
1229         }
1230
1231         set_new_dnode(&dn, inode, NULL, NULL, 0);
1232         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1233         if (err)
1234                 goto put_err;
1235         f2fs_put_dnode(&dn);
1236
1237         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1238                 err = -ENOENT;
1239                 goto put_err;
1240         }
1241         if (dn.data_blkaddr != NEW_ADDR &&
1242                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1243                                                 dn.data_blkaddr,
1244                                                 DATA_GENERIC_ENHANCE)) {
1245                 err = -EFSCORRUPTED;
1246                 goto put_err;
1247         }
1248 got_it:
1249         if (PageUptodate(page)) {
1250                 unlock_page(page);
1251                 return page;
1252         }
1253
1254         /*
1255          * A new dentry page is allocated but not able to be written, since its
1256          * new inode page couldn't be allocated due to -ENOSPC.
1257          * In such the case, its blkaddr can be remained as NEW_ADDR.
1258          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1259          * f2fs_init_inode_metadata.
1260          */
1261         if (dn.data_blkaddr == NEW_ADDR) {
1262                 zero_user_segment(page, 0, PAGE_SIZE);
1263                 if (!PageUptodate(page))
1264                         SetPageUptodate(page);
1265                 unlock_page(page);
1266                 return page;
1267         }
1268
1269         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1270                                                 op_flags, for_write);
1271         if (err)
1272                 goto put_err;
1273         return page;
1274
1275 put_err:
1276         f2fs_put_page(page, 1);
1277         return ERR_PTR(err);
1278 }
1279
1280 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1281 {
1282         struct address_space *mapping = inode->i_mapping;
1283         struct page *page;
1284
1285         page = find_get_page(mapping, index);
1286         if (page && PageUptodate(page))
1287                 return page;
1288         f2fs_put_page(page, 0);
1289
1290         page = f2fs_get_read_data_page(inode, index, 0, false);
1291         if (IS_ERR(page))
1292                 return page;
1293
1294         if (PageUptodate(page))
1295                 return page;
1296
1297         wait_on_page_locked(page);
1298         if (unlikely(!PageUptodate(page))) {
1299                 f2fs_put_page(page, 0);
1300                 return ERR_PTR(-EIO);
1301         }
1302         return page;
1303 }
1304
1305 /*
1306  * If it tries to access a hole, return an error.
1307  * Because, the callers, functions in dir.c and GC, should be able to know
1308  * whether this page exists or not.
1309  */
1310 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1311                                                         bool for_write)
1312 {
1313         struct address_space *mapping = inode->i_mapping;
1314         struct page *page;
1315 repeat:
1316         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1317         if (IS_ERR(page))
1318                 return page;
1319
1320         /* wait for read completion */
1321         lock_page(page);
1322         if (unlikely(page->mapping != mapping)) {
1323                 f2fs_put_page(page, 1);
1324                 goto repeat;
1325         }
1326         if (unlikely(!PageUptodate(page))) {
1327                 f2fs_put_page(page, 1);
1328                 return ERR_PTR(-EIO);
1329         }
1330         return page;
1331 }
1332
1333 /*
1334  * Caller ensures that this data page is never allocated.
1335  * A new zero-filled data page is allocated in the page cache.
1336  *
1337  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1338  * f2fs_unlock_op().
1339  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1340  * ipage should be released by this function.
1341  */
1342 struct page *f2fs_get_new_data_page(struct inode *inode,
1343                 struct page *ipage, pgoff_t index, bool new_i_size)
1344 {
1345         struct address_space *mapping = inode->i_mapping;
1346         struct page *page;
1347         struct dnode_of_data dn;
1348         int err;
1349
1350         page = f2fs_grab_cache_page(mapping, index, true);
1351         if (!page) {
1352                 /*
1353                  * before exiting, we should make sure ipage will be released
1354                  * if any error occur.
1355                  */
1356                 f2fs_put_page(ipage, 1);
1357                 return ERR_PTR(-ENOMEM);
1358         }
1359
1360         set_new_dnode(&dn, inode, ipage, NULL, 0);
1361         err = f2fs_reserve_block(&dn, index);
1362         if (err) {
1363                 f2fs_put_page(page, 1);
1364                 return ERR_PTR(err);
1365         }
1366         if (!ipage)
1367                 f2fs_put_dnode(&dn);
1368
1369         if (PageUptodate(page))
1370                 goto got_it;
1371
1372         if (dn.data_blkaddr == NEW_ADDR) {
1373                 zero_user_segment(page, 0, PAGE_SIZE);
1374                 if (!PageUptodate(page))
1375                         SetPageUptodate(page);
1376         } else {
1377                 f2fs_put_page(page, 1);
1378
1379                 /* if ipage exists, blkaddr should be NEW_ADDR */
1380                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1381                 page = f2fs_get_lock_data_page(inode, index, true);
1382                 if (IS_ERR(page))
1383                         return page;
1384         }
1385 got_it:
1386         if (new_i_size && i_size_read(inode) <
1387                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1388                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1389         return page;
1390 }
1391
1392 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1393 {
1394         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1395         struct f2fs_summary sum;
1396         struct node_info ni;
1397         block_t old_blkaddr;
1398         blkcnt_t count = 1;
1399         int err;
1400
1401         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1402                 return -EPERM;
1403
1404         err = f2fs_get_node_info(sbi, dn->nid, &ni);
1405         if (err)
1406                 return err;
1407
1408         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1409         if (dn->data_blkaddr != NULL_ADDR)
1410                 goto alloc;
1411
1412         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1413                 return err;
1414
1415 alloc:
1416         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1417         old_blkaddr = dn->data_blkaddr;
1418         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1419                                 &sum, seg_type, NULL);
1420         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1421                 invalidate_mapping_pages(META_MAPPING(sbi),
1422                                         old_blkaddr, old_blkaddr);
1423         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1424
1425         /*
1426          * i_size will be updated by direct_IO. Otherwise, we'll get stale
1427          * data from unwritten block via dio_read.
1428          */
1429         return 0;
1430 }
1431
1432 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1433 {
1434         struct inode *inode = file_inode(iocb->ki_filp);
1435         struct f2fs_map_blocks map;
1436         int flag;
1437         int err = 0;
1438         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1439
1440         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1441         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1442         if (map.m_len > map.m_lblk)
1443                 map.m_len -= map.m_lblk;
1444         else
1445                 map.m_len = 0;
1446
1447         map.m_next_pgofs = NULL;
1448         map.m_next_extent = NULL;
1449         map.m_seg_type = NO_CHECK_TYPE;
1450         map.m_may_create = true;
1451
1452         if (direct_io) {
1453                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1454                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1455                                         F2FS_GET_BLOCK_PRE_AIO :
1456                                         F2FS_GET_BLOCK_PRE_DIO;
1457                 goto map_blocks;
1458         }
1459         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1460                 err = f2fs_convert_inline_inode(inode);
1461                 if (err)
1462                         return err;
1463         }
1464         if (f2fs_has_inline_data(inode))
1465                 return err;
1466
1467         flag = F2FS_GET_BLOCK_PRE_AIO;
1468
1469 map_blocks:
1470         err = f2fs_map_blocks(inode, &map, 1, flag);
1471         if (map.m_len > 0 && err == -ENOSPC) {
1472                 if (!direct_io)
1473                         set_inode_flag(inode, FI_NO_PREALLOC);
1474                 err = 0;
1475         }
1476         return err;
1477 }
1478
1479 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1480 {
1481         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1482                 if (lock)
1483                         down_read(&sbi->node_change);
1484                 else
1485                         up_read(&sbi->node_change);
1486         } else {
1487                 if (lock)
1488                         f2fs_lock_op(sbi);
1489                 else
1490                         f2fs_unlock_op(sbi);
1491         }
1492 }
1493
1494 /*
1495  * f2fs_map_blocks() tries to find or build mapping relationship which
1496  * maps continuous logical blocks to physical blocks, and return such
1497  * info via f2fs_map_blocks structure.
1498  */
1499 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1500                                                 int create, int flag)
1501 {
1502         unsigned int maxblocks = map->m_len;
1503         struct dnode_of_data dn;
1504         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1505         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1506         pgoff_t pgofs, end_offset, end;
1507         int err = 0, ofs = 1;
1508         unsigned int ofs_in_node, last_ofs_in_node;
1509         blkcnt_t prealloc;
1510         struct extent_info ei = {0,0,0};
1511         block_t blkaddr;
1512         unsigned int start_pgofs;
1513
1514         if (!maxblocks)
1515                 return 0;
1516
1517         map->m_len = 0;
1518         map->m_flags = 0;
1519
1520         /* it only supports block size == page size */
1521         pgofs = (pgoff_t)map->m_lblk;
1522         end = pgofs + maxblocks;
1523
1524         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1525                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1526                                                         map->m_may_create)
1527                         goto next_dnode;
1528
1529                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1530                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1531                 map->m_flags = F2FS_MAP_MAPPED;
1532                 if (map->m_next_extent)
1533                         *map->m_next_extent = pgofs + map->m_len;
1534
1535                 /* for hardware encryption, but to avoid potential issue in future */
1536                 if (flag == F2FS_GET_BLOCK_DIO)
1537                         f2fs_wait_on_block_writeback_range(inode,
1538                                                 map->m_pblk, map->m_len);
1539                 goto out;
1540         }
1541
1542 next_dnode:
1543         if (map->m_may_create)
1544                 f2fs_do_map_lock(sbi, flag, true);
1545
1546         /* When reading holes, we need its node page */
1547         set_new_dnode(&dn, inode, NULL, NULL, 0);
1548         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1549         if (err) {
1550                 if (flag == F2FS_GET_BLOCK_BMAP)
1551                         map->m_pblk = 0;
1552                 if (err == -ENOENT) {
1553                         err = 0;
1554                         if (map->m_next_pgofs)
1555                                 *map->m_next_pgofs =
1556                                         f2fs_get_next_page_offset(&dn, pgofs);
1557                         if (map->m_next_extent)
1558                                 *map->m_next_extent =
1559                                         f2fs_get_next_page_offset(&dn, pgofs);
1560                 }
1561                 goto unlock_out;
1562         }
1563
1564         start_pgofs = pgofs;
1565         prealloc = 0;
1566         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1567         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1568
1569 next_block:
1570         blkaddr = f2fs_data_blkaddr(&dn);
1571
1572         if (__is_valid_data_blkaddr(blkaddr) &&
1573                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1574                 err = -EFSCORRUPTED;
1575                 goto sync_out;
1576         }
1577
1578         if (__is_valid_data_blkaddr(blkaddr)) {
1579                 /* use out-place-update for driect IO under LFS mode */
1580                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1581                                                         map->m_may_create) {
1582                         err = __allocate_data_block(&dn, map->m_seg_type);
1583                         if (err)
1584                                 goto sync_out;
1585                         blkaddr = dn.data_blkaddr;
1586                         set_inode_flag(inode, FI_APPEND_WRITE);
1587                 }
1588         } else {
1589                 if (create) {
1590                         if (unlikely(f2fs_cp_error(sbi))) {
1591                                 err = -EIO;
1592                                 goto sync_out;
1593                         }
1594                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1595                                 if (blkaddr == NULL_ADDR) {
1596                                         prealloc++;
1597                                         last_ofs_in_node = dn.ofs_in_node;
1598                                 }
1599                         } else {
1600                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1601                                         flag != F2FS_GET_BLOCK_DIO);
1602                                 err = __allocate_data_block(&dn,
1603                                                         map->m_seg_type);
1604                                 if (!err)
1605                                         set_inode_flag(inode, FI_APPEND_WRITE);
1606                         }
1607                         if (err)
1608                                 goto sync_out;
1609                         map->m_flags |= F2FS_MAP_NEW;
1610                         blkaddr = dn.data_blkaddr;
1611                 } else {
1612                         if (flag == F2FS_GET_BLOCK_BMAP) {
1613                                 map->m_pblk = 0;
1614                                 goto sync_out;
1615                         }
1616                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1617                                 goto sync_out;
1618                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1619                                                 blkaddr == NULL_ADDR) {
1620                                 if (map->m_next_pgofs)
1621                                         *map->m_next_pgofs = pgofs + 1;
1622                                 goto sync_out;
1623                         }
1624                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1625                                 /* for defragment case */
1626                                 if (map->m_next_pgofs)
1627                                         *map->m_next_pgofs = pgofs + 1;
1628                                 goto sync_out;
1629                         }
1630                 }
1631         }
1632
1633         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1634                 goto skip;
1635
1636         if (map->m_len == 0) {
1637                 /* preallocated unwritten block should be mapped for fiemap. */
1638                 if (blkaddr == NEW_ADDR)
1639                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1640                 map->m_flags |= F2FS_MAP_MAPPED;
1641
1642                 map->m_pblk = blkaddr;
1643                 map->m_len = 1;
1644         } else if ((map->m_pblk != NEW_ADDR &&
1645                         blkaddr == (map->m_pblk + ofs)) ||
1646                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1647                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1648                 ofs++;
1649                 map->m_len++;
1650         } else {
1651                 goto sync_out;
1652         }
1653
1654 skip:
1655         dn.ofs_in_node++;
1656         pgofs++;
1657
1658         /* preallocate blocks in batch for one dnode page */
1659         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1660                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1661
1662                 dn.ofs_in_node = ofs_in_node;
1663                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1664                 if (err)
1665                         goto sync_out;
1666
1667                 map->m_len += dn.ofs_in_node - ofs_in_node;
1668                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1669                         err = -ENOSPC;
1670                         goto sync_out;
1671                 }
1672                 dn.ofs_in_node = end_offset;
1673         }
1674
1675         if (pgofs >= end)
1676                 goto sync_out;
1677         else if (dn.ofs_in_node < end_offset)
1678                 goto next_block;
1679
1680         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1681                 if (map->m_flags & F2FS_MAP_MAPPED) {
1682                         unsigned int ofs = start_pgofs - map->m_lblk;
1683
1684                         f2fs_update_extent_cache_range(&dn,
1685                                 start_pgofs, map->m_pblk + ofs,
1686                                 map->m_len - ofs);
1687                 }
1688         }
1689
1690         f2fs_put_dnode(&dn);
1691
1692         if (map->m_may_create) {
1693                 f2fs_do_map_lock(sbi, flag, false);
1694                 f2fs_balance_fs(sbi, dn.node_changed);
1695         }
1696         goto next_dnode;
1697
1698 sync_out:
1699
1700         /* for hardware encryption, but to avoid potential issue in future */
1701         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1702                 f2fs_wait_on_block_writeback_range(inode,
1703                                                 map->m_pblk, map->m_len);
1704
1705         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1706                 if (map->m_flags & F2FS_MAP_MAPPED) {
1707                         unsigned int ofs = start_pgofs - map->m_lblk;
1708
1709                         f2fs_update_extent_cache_range(&dn,
1710                                 start_pgofs, map->m_pblk + ofs,
1711                                 map->m_len - ofs);
1712                 }
1713                 if (map->m_next_extent)
1714                         *map->m_next_extent = pgofs + 1;
1715         }
1716         f2fs_put_dnode(&dn);
1717 unlock_out:
1718         if (map->m_may_create) {
1719                 f2fs_do_map_lock(sbi, flag, false);
1720                 f2fs_balance_fs(sbi, dn.node_changed);
1721         }
1722 out:
1723         trace_f2fs_map_blocks(inode, map, err);
1724         return err;
1725 }
1726
1727 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1728 {
1729         struct f2fs_map_blocks map;
1730         block_t last_lblk;
1731         int err;
1732
1733         if (pos + len > i_size_read(inode))
1734                 return false;
1735
1736         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1737         map.m_next_pgofs = NULL;
1738         map.m_next_extent = NULL;
1739         map.m_seg_type = NO_CHECK_TYPE;
1740         map.m_may_create = false;
1741         last_lblk = F2FS_BLK_ALIGN(pos + len);
1742
1743         while (map.m_lblk < last_lblk) {
1744                 map.m_len = last_lblk - map.m_lblk;
1745                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1746                 if (err || map.m_len == 0)
1747                         return false;
1748                 map.m_lblk += map.m_len;
1749         }
1750         return true;
1751 }
1752
1753 static int __get_data_block(struct inode *inode, sector_t iblock,
1754                         struct buffer_head *bh, int create, int flag,
1755                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1756 {
1757         struct f2fs_map_blocks map;
1758         int err;
1759
1760         map.m_lblk = iblock;
1761         map.m_len = bh->b_size >> inode->i_blkbits;
1762         map.m_next_pgofs = next_pgofs;
1763         map.m_next_extent = NULL;
1764         map.m_seg_type = seg_type;
1765         map.m_may_create = may_write;
1766
1767         err = f2fs_map_blocks(inode, &map, create, flag);
1768         if (!err) {
1769                 map_bh(bh, inode->i_sb, map.m_pblk);
1770                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1771                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1772         }
1773         return err;
1774 }
1775
1776 static int get_data_block(struct inode *inode, sector_t iblock,
1777                         struct buffer_head *bh_result, int create, int flag,
1778                         pgoff_t *next_pgofs)
1779 {
1780         return __get_data_block(inode, iblock, bh_result, create,
1781                                                         flag, next_pgofs,
1782                                                         NO_CHECK_TYPE, create);
1783 }
1784
1785 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1786                         struct buffer_head *bh_result, int create)
1787 {
1788         return __get_data_block(inode, iblock, bh_result, create,
1789                                 F2FS_GET_BLOCK_DIO, NULL,
1790                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1791                                 IS_SWAPFILE(inode) ? false : true);
1792 }
1793
1794 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1795                         struct buffer_head *bh_result, int create)
1796 {
1797         return __get_data_block(inode, iblock, bh_result, create,
1798                                 F2FS_GET_BLOCK_DIO, NULL,
1799                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1800                                 false);
1801 }
1802
1803 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1804                         struct buffer_head *bh_result, int create)
1805 {
1806         return __get_data_block(inode, iblock, bh_result, create,
1807                                                 F2FS_GET_BLOCK_BMAP, NULL,
1808                                                 NO_CHECK_TYPE, create);
1809 }
1810
1811 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1812 {
1813         return (offset >> inode->i_blkbits);
1814 }
1815
1816 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1817 {
1818         return (blk << inode->i_blkbits);
1819 }
1820
1821 static int f2fs_xattr_fiemap(struct inode *inode,
1822                                 struct fiemap_extent_info *fieinfo)
1823 {
1824         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1825         struct page *page;
1826         struct node_info ni;
1827         __u64 phys = 0, len;
1828         __u32 flags;
1829         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1830         int err = 0;
1831
1832         if (f2fs_has_inline_xattr(inode)) {
1833                 int offset;
1834
1835                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1836                                                 inode->i_ino, false);
1837                 if (!page)
1838                         return -ENOMEM;
1839
1840                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1841                 if (err) {
1842                         f2fs_put_page(page, 1);
1843                         return err;
1844                 }
1845
1846                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1847                 offset = offsetof(struct f2fs_inode, i_addr) +
1848                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1849                                         get_inline_xattr_addrs(inode));
1850
1851                 phys += offset;
1852                 len = inline_xattr_size(inode);
1853
1854                 f2fs_put_page(page, 1);
1855
1856                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1857
1858                 if (!xnid)
1859                         flags |= FIEMAP_EXTENT_LAST;
1860
1861                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1862                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1863                 if (err || err == 1)
1864                         return err;
1865         }
1866
1867         if (xnid) {
1868                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1869                 if (!page)
1870                         return -ENOMEM;
1871
1872                 err = f2fs_get_node_info(sbi, xnid, &ni);
1873                 if (err) {
1874                         f2fs_put_page(page, 1);
1875                         return err;
1876                 }
1877
1878                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1879                 len = inode->i_sb->s_blocksize;
1880
1881                 f2fs_put_page(page, 1);
1882
1883                 flags = FIEMAP_EXTENT_LAST;
1884         }
1885
1886         if (phys) {
1887                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1888                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1889         }
1890
1891         return (err < 0 ? err : 0);
1892 }
1893
1894 static loff_t max_inode_blocks(struct inode *inode)
1895 {
1896         loff_t result = ADDRS_PER_INODE(inode);
1897         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1898
1899         /* two direct node blocks */
1900         result += (leaf_count * 2);
1901
1902         /* two indirect node blocks */
1903         leaf_count *= NIDS_PER_BLOCK;
1904         result += (leaf_count * 2);
1905
1906         /* one double indirect node block */
1907         leaf_count *= NIDS_PER_BLOCK;
1908         result += leaf_count;
1909
1910         return result;
1911 }
1912
1913 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1914                 u64 start, u64 len)
1915 {
1916         struct buffer_head map_bh;
1917         sector_t start_blk, last_blk;
1918         pgoff_t next_pgofs;
1919         u64 logical = 0, phys = 0, size = 0;
1920         u32 flags = 0;
1921         int ret = 0;
1922         bool compr_cluster = false;
1923         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1924
1925         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1926                 ret = f2fs_precache_extents(inode);
1927                 if (ret)
1928                         return ret;
1929         }
1930
1931         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1932         if (ret)
1933                 return ret;
1934
1935         inode_lock(inode);
1936
1937         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1938                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1939                 goto out;
1940         }
1941
1942         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1943                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1944                 if (ret != -EAGAIN)
1945                         goto out;
1946         }
1947
1948         if (logical_to_blk(inode, len) == 0)
1949                 len = blk_to_logical(inode, 1);
1950
1951         start_blk = logical_to_blk(inode, start);
1952         last_blk = logical_to_blk(inode, start + len - 1);
1953
1954 next:
1955         memset(&map_bh, 0, sizeof(struct buffer_head));
1956         map_bh.b_size = len;
1957
1958         if (compr_cluster)
1959                 map_bh.b_size = blk_to_logical(inode, cluster_size - 1);
1960
1961         ret = get_data_block(inode, start_blk, &map_bh, 0,
1962                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1963         if (ret)
1964                 goto out;
1965
1966         /* HOLE */
1967         if (!buffer_mapped(&map_bh)) {
1968                 start_blk = next_pgofs;
1969
1970                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1971                                                 max_inode_blocks(inode)))
1972                         goto prep_next;
1973
1974                 flags |= FIEMAP_EXTENT_LAST;
1975         }
1976
1977         if (size) {
1978                 if (IS_ENCRYPTED(inode))
1979                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1980
1981                 ret = fiemap_fill_next_extent(fieinfo, logical,
1982                                 phys, size, flags);
1983                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1984                 if (ret)
1985                         goto out;
1986                 size = 0;
1987         }
1988
1989         if (start_blk > last_blk)
1990                 goto out;
1991
1992         if (compr_cluster) {
1993                 compr_cluster = false;
1994
1995
1996                 logical = blk_to_logical(inode, start_blk - 1);
1997                 phys = blk_to_logical(inode, map_bh.b_blocknr);
1998                 size = blk_to_logical(inode, cluster_size);
1999
2000                 flags |= FIEMAP_EXTENT_ENCODED;
2001
2002                 start_blk += cluster_size - 1;
2003
2004                 if (start_blk > last_blk)
2005                         goto out;
2006
2007                 goto prep_next;
2008         }
2009
2010         if (map_bh.b_blocknr == COMPRESS_ADDR) {
2011                 compr_cluster = true;
2012                 start_blk++;
2013                 goto prep_next;
2014         }
2015
2016         logical = blk_to_logical(inode, start_blk);
2017         phys = blk_to_logical(inode, map_bh.b_blocknr);
2018         size = map_bh.b_size;
2019         flags = 0;
2020         if (buffer_unwritten(&map_bh))
2021                 flags = FIEMAP_EXTENT_UNWRITTEN;
2022
2023         start_blk += logical_to_blk(inode, size);
2024
2025 prep_next:
2026         cond_resched();
2027         if (fatal_signal_pending(current))
2028                 ret = -EINTR;
2029         else
2030                 goto next;
2031 out:
2032         if (ret == 1)
2033                 ret = 0;
2034
2035         inode_unlock(inode);
2036         return ret;
2037 }
2038
2039 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2040 {
2041         if (IS_ENABLED(CONFIG_FS_VERITY) &&
2042             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2043                 return inode->i_sb->s_maxbytes;
2044
2045         return i_size_read(inode);
2046 }
2047
2048 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2049                                         unsigned nr_pages,
2050                                         struct f2fs_map_blocks *map,
2051                                         struct bio **bio_ret,
2052                                         sector_t *last_block_in_bio,
2053                                         bool is_readahead)
2054 {
2055         struct bio *bio = *bio_ret;
2056         const unsigned blkbits = inode->i_blkbits;
2057         const unsigned blocksize = 1 << blkbits;
2058         sector_t block_in_file;
2059         sector_t last_block;
2060         sector_t last_block_in_file;
2061         sector_t block_nr;
2062         int ret = 0;
2063
2064         block_in_file = (sector_t)page_index(page);
2065         last_block = block_in_file + nr_pages;
2066         last_block_in_file = (f2fs_readpage_limit(inode) + blocksize - 1) >>
2067                                                         blkbits;
2068         if (last_block > last_block_in_file)
2069                 last_block = last_block_in_file;
2070
2071         /* just zeroing out page which is beyond EOF */
2072         if (block_in_file >= last_block)
2073                 goto zero_out;
2074         /*
2075          * Map blocks using the previous result first.
2076          */
2077         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2078                         block_in_file > map->m_lblk &&
2079                         block_in_file < (map->m_lblk + map->m_len))
2080                 goto got_it;
2081
2082         /*
2083          * Then do more f2fs_map_blocks() calls until we are
2084          * done with this page.
2085          */
2086         map->m_lblk = block_in_file;
2087         map->m_len = last_block - block_in_file;
2088
2089         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2090         if (ret)
2091                 goto out;
2092 got_it:
2093         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2094                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2095                 SetPageMappedToDisk(page);
2096
2097                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
2098                                         !cleancache_get_page(page))) {
2099                         SetPageUptodate(page);
2100                         goto confused;
2101                 }
2102
2103                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2104                                                 DATA_GENERIC_ENHANCE_READ)) {
2105                         ret = -EFSCORRUPTED;
2106                         goto out;
2107                 }
2108         } else {
2109 zero_out:
2110                 zero_user_segment(page, 0, PAGE_SIZE);
2111                 if (f2fs_need_verity(inode, page->index) &&
2112                     !fsverity_verify_page(page)) {
2113                         ret = -EIO;
2114                         goto out;
2115                 }
2116                 if (!PageUptodate(page))
2117                         SetPageUptodate(page);
2118                 unlock_page(page);
2119                 goto out;
2120         }
2121
2122         /*
2123          * This page will go to BIO.  Do we need to send this
2124          * BIO off first?
2125          */
2126         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2127                                        *last_block_in_bio, block_nr) ||
2128                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2129 submit_and_realloc:
2130                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2131                 bio = NULL;
2132         }
2133         if (bio == NULL) {
2134                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2135                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2136                                 false);
2137                 if (IS_ERR(bio)) {
2138                         ret = PTR_ERR(bio);
2139                         bio = NULL;
2140                         goto out;
2141                 }
2142         }
2143
2144         /*
2145          * If the page is under writeback, we need to wait for
2146          * its completion to see the correct decrypted data.
2147          */
2148         f2fs_wait_on_block_writeback(inode, block_nr);
2149
2150         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2151                 goto submit_and_realloc;
2152
2153         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2154         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2155         ClearPageError(page);
2156         *last_block_in_bio = block_nr;
2157         goto out;
2158 confused:
2159         if (bio) {
2160                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2161                 bio = NULL;
2162         }
2163         unlock_page(page);
2164 out:
2165         *bio_ret = bio;
2166         return ret;
2167 }
2168
2169 #ifdef CONFIG_F2FS_FS_COMPRESSION
2170 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2171                                 unsigned nr_pages, sector_t *last_block_in_bio,
2172                                 bool is_readahead, bool for_write)
2173 {
2174         struct dnode_of_data dn;
2175         struct inode *inode = cc->inode;
2176         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2177         struct bio *bio = *bio_ret;
2178         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2179         sector_t last_block_in_file;
2180         const unsigned blkbits = inode->i_blkbits;
2181         const unsigned blocksize = 1 << blkbits;
2182         struct decompress_io_ctx *dic = NULL;
2183         int i;
2184         int ret = 0;
2185
2186         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2187
2188         last_block_in_file = (f2fs_readpage_limit(inode) +
2189                                         blocksize - 1) >> blkbits;
2190
2191         /* get rid of pages beyond EOF */
2192         for (i = 0; i < cc->cluster_size; i++) {
2193                 struct page *page = cc->rpages[i];
2194
2195                 if (!page)
2196                         continue;
2197                 if ((sector_t)page->index >= last_block_in_file) {
2198                         zero_user_segment(page, 0, PAGE_SIZE);
2199                         if (!PageUptodate(page))
2200                                 SetPageUptodate(page);
2201                 } else if (!PageUptodate(page)) {
2202                         continue;
2203                 }
2204                 unlock_page(page);
2205                 cc->rpages[i] = NULL;
2206                 cc->nr_rpages--;
2207         }
2208
2209         /* we are done since all pages are beyond EOF */
2210         if (f2fs_cluster_is_empty(cc))
2211                 goto out;
2212
2213         set_new_dnode(&dn, inode, NULL, NULL, 0);
2214         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2215         if (ret)
2216                 goto out;
2217
2218         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2219
2220         for (i = 1; i < cc->cluster_size; i++) {
2221                 block_t blkaddr;
2222
2223                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2224                                                 dn.ofs_in_node + i);
2225
2226                 if (!__is_valid_data_blkaddr(blkaddr))
2227                         break;
2228
2229                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2230                         ret = -EFAULT;
2231                         goto out_put_dnode;
2232                 }
2233                 cc->nr_cpages++;
2234         }
2235
2236         /* nothing to decompress */
2237         if (cc->nr_cpages == 0) {
2238                 ret = 0;
2239                 goto out_put_dnode;
2240         }
2241
2242         dic = f2fs_alloc_dic(cc);
2243         if (IS_ERR(dic)) {
2244                 ret = PTR_ERR(dic);
2245                 goto out_put_dnode;
2246         }
2247
2248         for (i = 0; i < dic->nr_cpages; i++) {
2249                 struct page *page = dic->cpages[i];
2250                 block_t blkaddr;
2251                 struct bio_post_read_ctx *ctx;
2252
2253                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2254                                                 dn.ofs_in_node + i + 1);
2255
2256                 if (bio && (!page_is_mergeable(sbi, bio,
2257                                         *last_block_in_bio, blkaddr) ||
2258                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2259 submit_and_realloc:
2260                         __submit_bio(sbi, bio, DATA);
2261                         bio = NULL;
2262                 }
2263
2264                 if (!bio) {
2265                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2266                                         is_readahead ? REQ_RAHEAD : 0,
2267                                         page->index, for_write);
2268                         if (IS_ERR(bio)) {
2269                                 ret = PTR_ERR(bio);
2270                                 dic->failed = true;
2271                                 if (!atomic_sub_return(dic->nr_cpages - i,
2272                                                         &dic->pending_pages)) {
2273                                         f2fs_decompress_end_io(dic->rpages,
2274                                                         cc->cluster_size, true,
2275                                                         false);
2276                                         f2fs_free_dic(dic);
2277                                 }
2278                                 f2fs_put_dnode(&dn);
2279                                 *bio_ret = NULL;
2280                                 return ret;
2281                         }
2282                 }
2283
2284                 f2fs_wait_on_block_writeback(inode, blkaddr);
2285
2286                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2287                         goto submit_and_realloc;
2288
2289                 /* tag STEP_DECOMPRESS to handle IO in wq */
2290                 ctx = bio->bi_private;
2291                 if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS)))
2292                         ctx->enabled_steps |= 1 << STEP_DECOMPRESS;
2293
2294                 inc_page_count(sbi, F2FS_RD_DATA);
2295                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2296                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2297                 ClearPageError(page);
2298                 *last_block_in_bio = blkaddr;
2299         }
2300
2301         f2fs_put_dnode(&dn);
2302
2303         *bio_ret = bio;
2304         return 0;
2305
2306 out_put_dnode:
2307         f2fs_put_dnode(&dn);
2308 out:
2309         f2fs_decompress_end_io(cc->rpages, cc->cluster_size, true, false);
2310         *bio_ret = bio;
2311         return ret;
2312 }
2313 #endif
2314
2315 /*
2316  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2317  * Major change was from block_size == page_size in f2fs by default.
2318  *
2319  * Note that the aops->readpages() function is ONLY used for read-ahead. If
2320  * this function ever deviates from doing just read-ahead, it should either
2321  * use ->readpage() or do the necessary surgery to decouple ->readpages()
2322  * from read-ahead.
2323  */
2324 static int f2fs_mpage_readpages(struct inode *inode,
2325                 struct readahead_control *rac, struct page *page)
2326 {
2327         struct bio *bio = NULL;
2328         sector_t last_block_in_bio = 0;
2329         struct f2fs_map_blocks map;
2330 #ifdef CONFIG_F2FS_FS_COMPRESSION
2331         struct compress_ctx cc = {
2332                 .inode = inode,
2333                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2334                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2335                 .cluster_idx = NULL_CLUSTER,
2336                 .rpages = NULL,
2337                 .cpages = NULL,
2338                 .nr_rpages = 0,
2339                 .nr_cpages = 0,
2340         };
2341 #endif
2342         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2343         unsigned max_nr_pages = nr_pages;
2344         int ret = 0;
2345         bool drop_ra = false;
2346
2347         map.m_pblk = 0;
2348         map.m_lblk = 0;
2349         map.m_len = 0;
2350         map.m_flags = 0;
2351         map.m_next_pgofs = NULL;
2352         map.m_next_extent = NULL;
2353         map.m_seg_type = NO_CHECK_TYPE;
2354         map.m_may_create = false;
2355
2356         /*
2357          * Two readahead threads for same address range can cause race condition
2358          * which fragments sequential read IOs. So let's avoid each other.
2359          */
2360         if (rac && readahead_count(rac)) {
2361                 if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
2362                         drop_ra = true;
2363                 else
2364                         WRITE_ONCE(F2FS_I(inode)->ra_offset,
2365                                                 readahead_index(rac));
2366         }
2367
2368         for (; nr_pages; nr_pages--) {
2369                 if (rac) {
2370                         page = readahead_page(rac);
2371                         prefetchw(&page->flags);
2372                         if (drop_ra) {
2373                                 f2fs_put_page(page, 1);
2374                                 continue;
2375                         }
2376                 }
2377
2378 #ifdef CONFIG_F2FS_FS_COMPRESSION
2379                 if (f2fs_compressed_file(inode)) {
2380                         /* there are remained comressed pages, submit them */
2381                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2382                                 ret = f2fs_read_multi_pages(&cc, &bio,
2383                                                         max_nr_pages,
2384                                                         &last_block_in_bio,
2385                                                         rac != NULL, false);
2386                                 f2fs_destroy_compress_ctx(&cc);
2387                                 if (ret)
2388                                         goto set_error_page;
2389                         }
2390                         ret = f2fs_is_compressed_cluster(inode, page->index);
2391                         if (ret < 0)
2392                                 goto set_error_page;
2393                         else if (!ret)
2394                                 goto read_single_page;
2395
2396                         ret = f2fs_init_compress_ctx(&cc);
2397                         if (ret)
2398                                 goto set_error_page;
2399
2400                         f2fs_compress_ctx_add_page(&cc, page);
2401
2402                         goto next_page;
2403                 }
2404 read_single_page:
2405 #endif
2406
2407                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2408                                         &bio, &last_block_in_bio, rac);
2409                 if (ret) {
2410 #ifdef CONFIG_F2FS_FS_COMPRESSION
2411 set_error_page:
2412 #endif
2413                         SetPageError(page);
2414                         zero_user_segment(page, 0, PAGE_SIZE);
2415                         unlock_page(page);
2416                 }
2417 #ifdef CONFIG_F2FS_FS_COMPRESSION
2418 next_page:
2419 #endif
2420                 if (rac)
2421                         put_page(page);
2422
2423 #ifdef CONFIG_F2FS_FS_COMPRESSION
2424                 if (f2fs_compressed_file(inode)) {
2425                         /* last page */
2426                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2427                                 ret = f2fs_read_multi_pages(&cc, &bio,
2428                                                         max_nr_pages,
2429                                                         &last_block_in_bio,
2430                                                         rac != NULL, false);
2431                                 f2fs_destroy_compress_ctx(&cc);
2432                         }
2433                 }
2434 #endif
2435         }
2436         if (bio)
2437                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2438
2439         if (rac && readahead_count(rac) && !drop_ra)
2440                 WRITE_ONCE(F2FS_I(inode)->ra_offset, -1);
2441         return ret;
2442 }
2443
2444 static int f2fs_read_data_page(struct file *file, struct page *page)
2445 {
2446         struct inode *inode = page_file_mapping(page)->host;
2447         int ret = -EAGAIN;
2448
2449         trace_f2fs_readpage(page, DATA);
2450
2451         if (!f2fs_is_compress_backend_ready(inode)) {
2452                 unlock_page(page);
2453                 return -EOPNOTSUPP;
2454         }
2455
2456         /* If the file has inline data, try to read it directly */
2457         if (f2fs_has_inline_data(inode))
2458                 ret = f2fs_read_inline_data(inode, page);
2459         if (ret == -EAGAIN)
2460                 ret = f2fs_mpage_readpages(inode, NULL, page);
2461         return ret;
2462 }
2463
2464 static void f2fs_readahead(struct readahead_control *rac)
2465 {
2466         struct inode *inode = rac->mapping->host;
2467
2468         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2469
2470         if (!f2fs_is_compress_backend_ready(inode))
2471                 return;
2472
2473         /* If the file has inline data, skip readpages */
2474         if (f2fs_has_inline_data(inode))
2475                 return;
2476
2477         f2fs_mpage_readpages(inode, rac, NULL);
2478 }
2479
2480 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2481 {
2482         struct inode *inode = fio->page->mapping->host;
2483         struct page *mpage, *page;
2484         gfp_t gfp_flags = GFP_NOFS;
2485
2486         if (!f2fs_encrypted_file(inode))
2487                 return 0;
2488
2489         page = fio->compressed_page ? fio->compressed_page : fio->page;
2490
2491         /* wait for GCed page writeback via META_MAPPING */
2492         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2493
2494         if (fscrypt_inode_uses_inline_crypto(inode))
2495                 return 0;
2496
2497 retry_encrypt:
2498         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2499                                         PAGE_SIZE, 0, gfp_flags);
2500         if (IS_ERR(fio->encrypted_page)) {
2501                 /* flush pending IOs and wait for a while in the ENOMEM case */
2502                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2503                         f2fs_flush_merged_writes(fio->sbi);
2504                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2505                         gfp_flags |= __GFP_NOFAIL;
2506                         goto retry_encrypt;
2507                 }
2508                 return PTR_ERR(fio->encrypted_page);
2509         }
2510
2511         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2512         if (mpage) {
2513                 if (PageUptodate(mpage))
2514                         memcpy(page_address(mpage),
2515                                 page_address(fio->encrypted_page), PAGE_SIZE);
2516                 f2fs_put_page(mpage, 1);
2517         }
2518         return 0;
2519 }
2520
2521 static inline bool check_inplace_update_policy(struct inode *inode,
2522                                 struct f2fs_io_info *fio)
2523 {
2524         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2525         unsigned int policy = SM_I(sbi)->ipu_policy;
2526
2527         if (policy & (0x1 << F2FS_IPU_FORCE))
2528                 return true;
2529         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2530                 return true;
2531         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2532                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2533                 return true;
2534         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2535                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2536                 return true;
2537
2538         /*
2539          * IPU for rewrite async pages
2540          */
2541         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2542                         fio && fio->op == REQ_OP_WRITE &&
2543                         !(fio->op_flags & REQ_SYNC) &&
2544                         !IS_ENCRYPTED(inode))
2545                 return true;
2546
2547         /* this is only set during fdatasync */
2548         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2549                         is_inode_flag_set(inode, FI_NEED_IPU))
2550                 return true;
2551
2552         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2553                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2554                 return true;
2555
2556         return false;
2557 }
2558
2559 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2560 {
2561         if (f2fs_is_pinned_file(inode))
2562                 return true;
2563
2564         /* if this is cold file, we should overwrite to avoid fragmentation */
2565         if (file_is_cold(inode))
2566                 return true;
2567
2568         return check_inplace_update_policy(inode, fio);
2569 }
2570
2571 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2572 {
2573         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2574
2575         if (f2fs_lfs_mode(sbi))
2576                 return true;
2577         if (S_ISDIR(inode->i_mode))
2578                 return true;
2579         if (IS_NOQUOTA(inode))
2580                 return true;
2581         if (f2fs_is_atomic_file(inode))
2582                 return true;
2583         if (fio) {
2584                 if (is_cold_data(fio->page))
2585                         return true;
2586                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
2587                         return true;
2588                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2589                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2590                         return true;
2591         }
2592         return false;
2593 }
2594
2595 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2596 {
2597         struct inode *inode = fio->page->mapping->host;
2598
2599         if (f2fs_should_update_outplace(inode, fio))
2600                 return false;
2601
2602         return f2fs_should_update_inplace(inode, fio);
2603 }
2604
2605 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2606 {
2607         struct page *page = fio->page;
2608         struct inode *inode = page->mapping->host;
2609         struct dnode_of_data dn;
2610         struct extent_info ei = {0,0,0};
2611         struct node_info ni;
2612         bool ipu_force = false;
2613         int err = 0;
2614
2615         set_new_dnode(&dn, inode, NULL, NULL, 0);
2616         if (need_inplace_update(fio) &&
2617                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2618                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2619
2620                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2621                                                 DATA_GENERIC_ENHANCE))
2622                         return -EFSCORRUPTED;
2623
2624                 ipu_force = true;
2625                 fio->need_lock = LOCK_DONE;
2626                 goto got_it;
2627         }
2628
2629         /* Deadlock due to between page->lock and f2fs_lock_op */
2630         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2631                 return -EAGAIN;
2632
2633         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2634         if (err)
2635                 goto out;
2636
2637         fio->old_blkaddr = dn.data_blkaddr;
2638
2639         /* This page is already truncated */
2640         if (fio->old_blkaddr == NULL_ADDR) {
2641                 ClearPageUptodate(page);
2642                 clear_cold_data(page);
2643                 goto out_writepage;
2644         }
2645 got_it:
2646         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2647                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2648                                                 DATA_GENERIC_ENHANCE)) {
2649                 err = -EFSCORRUPTED;
2650                 goto out_writepage;
2651         }
2652         /*
2653          * If current allocation needs SSR,
2654          * it had better in-place writes for updated data.
2655          */
2656         if (ipu_force ||
2657                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2658                                         need_inplace_update(fio))) {
2659                 err = f2fs_encrypt_one_page(fio);
2660                 if (err)
2661                         goto out_writepage;
2662
2663                 set_page_writeback(page);
2664                 ClearPageError(page);
2665                 f2fs_put_dnode(&dn);
2666                 if (fio->need_lock == LOCK_REQ)
2667                         f2fs_unlock_op(fio->sbi);
2668                 err = f2fs_inplace_write_data(fio);
2669                 if (err) {
2670                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2671                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2672                         if (PageWriteback(page))
2673                                 end_page_writeback(page);
2674                 } else {
2675                         set_inode_flag(inode, FI_UPDATE_WRITE);
2676                 }
2677                 trace_f2fs_do_write_data_page(fio->page, IPU);
2678                 return err;
2679         }
2680
2681         if (fio->need_lock == LOCK_RETRY) {
2682                 if (!f2fs_trylock_op(fio->sbi)) {
2683                         err = -EAGAIN;
2684                         goto out_writepage;
2685                 }
2686                 fio->need_lock = LOCK_REQ;
2687         }
2688
2689         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2690         if (err)
2691                 goto out_writepage;
2692
2693         fio->version = ni.version;
2694
2695         err = f2fs_encrypt_one_page(fio);
2696         if (err)
2697                 goto out_writepage;
2698
2699         set_page_writeback(page);
2700         ClearPageError(page);
2701
2702         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2703                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2704
2705         /* LFS mode write path */
2706         f2fs_outplace_write_data(&dn, fio);
2707         trace_f2fs_do_write_data_page(page, OPU);
2708         set_inode_flag(inode, FI_APPEND_WRITE);
2709         if (page->index == 0)
2710                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2711 out_writepage:
2712         f2fs_put_dnode(&dn);
2713 out:
2714         if (fio->need_lock == LOCK_REQ)
2715                 f2fs_unlock_op(fio->sbi);
2716         return err;
2717 }
2718
2719 int f2fs_write_single_data_page(struct page *page, int *submitted,
2720                                 struct bio **bio,
2721                                 sector_t *last_block,
2722                                 struct writeback_control *wbc,
2723                                 enum iostat_type io_type,
2724                                 int compr_blocks)
2725 {
2726         struct inode *inode = page->mapping->host;
2727         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2728         loff_t i_size = i_size_read(inode);
2729         const pgoff_t end_index = ((unsigned long long)i_size)
2730                                                         >> PAGE_SHIFT;
2731         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2732         unsigned offset = 0;
2733         bool need_balance_fs = false;
2734         int err = 0;
2735         struct f2fs_io_info fio = {
2736                 .sbi = sbi,
2737                 .ino = inode->i_ino,
2738                 .type = DATA,
2739                 .op = REQ_OP_WRITE,
2740                 .op_flags = wbc_to_write_flags(wbc),
2741                 .old_blkaddr = NULL_ADDR,
2742                 .page = page,
2743                 .encrypted_page = NULL,
2744                 .submitted = false,
2745                 .compr_blocks = compr_blocks,
2746                 .need_lock = LOCK_RETRY,
2747                 .io_type = io_type,
2748                 .io_wbc = wbc,
2749                 .bio = bio,
2750                 .last_block = last_block,
2751         };
2752
2753         trace_f2fs_writepage(page, DATA);
2754
2755         /* we should bypass data pages to proceed the kworkder jobs */
2756         if (unlikely(f2fs_cp_error(sbi))) {
2757                 mapping_set_error(page->mapping, -EIO);
2758                 /*
2759                  * don't drop any dirty dentry pages for keeping lastest
2760                  * directory structure.
2761                  */
2762                 if (S_ISDIR(inode->i_mode))
2763                         goto redirty_out;
2764                 goto out;
2765         }
2766
2767         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2768                 goto redirty_out;
2769
2770         if (page->index < end_index ||
2771                         f2fs_verity_in_progress(inode) ||
2772                         compr_blocks)
2773                 goto write;
2774
2775         /*
2776          * If the offset is out-of-range of file size,
2777          * this page does not have to be written to disk.
2778          */
2779         offset = i_size & (PAGE_SIZE - 1);
2780         if ((page->index >= end_index + 1) || !offset)
2781                 goto out;
2782
2783         zero_user_segment(page, offset, PAGE_SIZE);
2784 write:
2785         if (f2fs_is_drop_cache(inode))
2786                 goto out;
2787         /* we should not write 0'th page having journal header */
2788         if (f2fs_is_volatile_file(inode) && (!page->index ||
2789                         (!wbc->for_reclaim &&
2790                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2791                 goto redirty_out;
2792
2793         /* Dentry/quota blocks are controlled by checkpoint */
2794         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2795                 /*
2796                  * We need to wait for node_write to avoid block allocation during
2797                  * checkpoint. This can only happen to quota writes which can cause
2798                  * the below discard race condition.
2799                  */
2800                 if (IS_NOQUOTA(inode))
2801                         down_read(&sbi->node_write);
2802
2803                 fio.need_lock = LOCK_DONE;
2804                 err = f2fs_do_write_data_page(&fio);
2805
2806                 if (IS_NOQUOTA(inode))
2807                         up_read(&sbi->node_write);
2808
2809                 goto done;
2810         }
2811
2812         if (!wbc->for_reclaim)
2813                 need_balance_fs = true;
2814         else if (has_not_enough_free_secs(sbi, 0, 0))
2815                 goto redirty_out;
2816         else
2817                 set_inode_flag(inode, FI_HOT_DATA);
2818
2819         err = -EAGAIN;
2820         if (f2fs_has_inline_data(inode)) {
2821                 err = f2fs_write_inline_data(inode, page);
2822                 if (!err)
2823                         goto out;
2824         }
2825
2826         if (err == -EAGAIN) {
2827                 err = f2fs_do_write_data_page(&fio);
2828                 if (err == -EAGAIN) {
2829                         fio.need_lock = LOCK_REQ;
2830                         err = f2fs_do_write_data_page(&fio);
2831                 }
2832         }
2833
2834         if (err) {
2835                 file_set_keep_isize(inode);
2836         } else {
2837                 spin_lock(&F2FS_I(inode)->i_size_lock);
2838                 if (F2FS_I(inode)->last_disk_size < psize)
2839                         F2FS_I(inode)->last_disk_size = psize;
2840                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2841         }
2842
2843 done:
2844         if (err && err != -ENOENT)
2845                 goto redirty_out;
2846
2847 out:
2848         inode_dec_dirty_pages(inode);
2849         if (err) {
2850                 ClearPageUptodate(page);
2851                 clear_cold_data(page);
2852         }
2853
2854         if (wbc->for_reclaim) {
2855                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2856                 clear_inode_flag(inode, FI_HOT_DATA);
2857                 f2fs_remove_dirty_inode(inode);
2858                 submitted = NULL;
2859         }
2860         unlock_page(page);
2861         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2862                                         !F2FS_I(inode)->cp_task)
2863                 f2fs_balance_fs(sbi, need_balance_fs);
2864
2865         if (unlikely(f2fs_cp_error(sbi))) {
2866                 f2fs_submit_merged_write(sbi, DATA);
2867                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2868                 submitted = NULL;
2869         }
2870
2871         if (submitted)
2872                 *submitted = fio.submitted ? 1 : 0;
2873
2874         return 0;
2875
2876 redirty_out:
2877         redirty_page_for_writepage(wbc, page);
2878         /*
2879          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2880          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2881          * file_write_and_wait_range() will see EIO error, which is critical
2882          * to return value of fsync() followed by atomic_write failure to user.
2883          */
2884         if (!err || wbc->for_reclaim)
2885                 return AOP_WRITEPAGE_ACTIVATE;
2886         unlock_page(page);
2887         return err;
2888 }
2889
2890 static int f2fs_write_data_page(struct page *page,
2891                                         struct writeback_control *wbc)
2892 {
2893 #ifdef CONFIG_F2FS_FS_COMPRESSION
2894         struct inode *inode = page->mapping->host;
2895
2896         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2897                 goto out;
2898
2899         if (f2fs_compressed_file(inode)) {
2900                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2901                         redirty_page_for_writepage(wbc, page);
2902                         return AOP_WRITEPAGE_ACTIVATE;
2903                 }
2904         }
2905 out:
2906 #endif
2907
2908         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2909                                                 wbc, FS_DATA_IO, 0);
2910 }
2911
2912 /*
2913  * This function was copied from write_cche_pages from mm/page-writeback.c.
2914  * The major change is making write step of cold data page separately from
2915  * warm/hot data page.
2916  */
2917 static int f2fs_write_cache_pages(struct address_space *mapping,
2918                                         struct writeback_control *wbc,
2919                                         enum iostat_type io_type)
2920 {
2921         int ret = 0;
2922         int done = 0, retry = 0;
2923         struct pagevec pvec;
2924         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2925         struct bio *bio = NULL;
2926         sector_t last_block;
2927 #ifdef CONFIG_F2FS_FS_COMPRESSION
2928         struct inode *inode = mapping->host;
2929         struct compress_ctx cc = {
2930                 .inode = inode,
2931                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2932                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2933                 .cluster_idx = NULL_CLUSTER,
2934                 .rpages = NULL,
2935                 .nr_rpages = 0,
2936                 .cpages = NULL,
2937                 .rbuf = NULL,
2938                 .cbuf = NULL,
2939                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2940                 .private = NULL,
2941         };
2942 #endif
2943         int nr_pages;
2944         pgoff_t index;
2945         pgoff_t end;            /* Inclusive */
2946         pgoff_t done_index;
2947         int range_whole = 0;
2948         xa_mark_t tag;
2949         int nwritten = 0;
2950         int submitted = 0;
2951         int i;
2952
2953         pagevec_init(&pvec);
2954
2955         if (get_dirty_pages(mapping->host) <=
2956                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2957                 set_inode_flag(mapping->host, FI_HOT_DATA);
2958         else
2959                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2960
2961         if (wbc->range_cyclic) {
2962                 index = mapping->writeback_index; /* prev offset */
2963                 end = -1;
2964         } else {
2965                 index = wbc->range_start >> PAGE_SHIFT;
2966                 end = wbc->range_end >> PAGE_SHIFT;
2967                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2968                         range_whole = 1;
2969         }
2970         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2971                 tag = PAGECACHE_TAG_TOWRITE;
2972         else
2973                 tag = PAGECACHE_TAG_DIRTY;
2974 retry:
2975         retry = 0;
2976         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2977                 tag_pages_for_writeback(mapping, index, end);
2978         done_index = index;
2979         while (!done && !retry && (index <= end)) {
2980                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2981                                 tag);
2982                 if (nr_pages == 0)
2983                         break;
2984
2985                 for (i = 0; i < nr_pages; i++) {
2986                         struct page *page = pvec.pages[i];
2987                         bool need_readd;
2988 readd:
2989                         need_readd = false;
2990 #ifdef CONFIG_F2FS_FS_COMPRESSION
2991                         if (f2fs_compressed_file(inode)) {
2992                                 ret = f2fs_init_compress_ctx(&cc);
2993                                 if (ret) {
2994                                         done = 1;
2995                                         break;
2996                                 }
2997
2998                                 if (!f2fs_cluster_can_merge_page(&cc,
2999                                                                 page->index)) {
3000                                         ret = f2fs_write_multi_pages(&cc,
3001                                                 &submitted, wbc, io_type);
3002                                         if (!ret)
3003                                                 need_readd = true;
3004                                         goto result;
3005                                 }
3006
3007                                 if (unlikely(f2fs_cp_error(sbi)))
3008                                         goto lock_page;
3009
3010                                 if (f2fs_cluster_is_empty(&cc)) {
3011                                         void *fsdata = NULL;
3012                                         struct page *pagep;
3013                                         int ret2;
3014
3015                                         ret2 = f2fs_prepare_compress_overwrite(
3016                                                         inode, &pagep,
3017                                                         page->index, &fsdata);
3018                                         if (ret2 < 0) {
3019                                                 ret = ret2;
3020                                                 done = 1;
3021                                                 break;
3022                                         } else if (ret2 &&
3023                                                 !f2fs_compress_write_end(inode,
3024                                                                 fsdata, page->index,
3025                                                                 1)) {
3026                                                 retry = 1;
3027                                                 break;
3028                                         }
3029                                 } else {
3030                                         goto lock_page;
3031                                 }
3032                         }
3033 #endif
3034                         /* give a priority to WB_SYNC threads */
3035                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3036                                         wbc->sync_mode == WB_SYNC_NONE) {
3037                                 done = 1;
3038                                 break;
3039                         }
3040 #ifdef CONFIG_F2FS_FS_COMPRESSION
3041 lock_page:
3042 #endif
3043                         done_index = page->index;
3044 retry_write:
3045                         lock_page(page);
3046
3047                         if (unlikely(page->mapping != mapping)) {
3048 continue_unlock:
3049                                 unlock_page(page);
3050                                 continue;
3051                         }
3052
3053                         if (!PageDirty(page)) {
3054                                 /* someone wrote it for us */
3055                                 goto continue_unlock;
3056                         }
3057
3058                         if (PageWriteback(page)) {
3059                                 if (wbc->sync_mode != WB_SYNC_NONE)
3060                                         f2fs_wait_on_page_writeback(page,
3061                                                         DATA, true, true);
3062                                 else
3063                                         goto continue_unlock;
3064                         }
3065
3066                         if (!clear_page_dirty_for_io(page))
3067                                 goto continue_unlock;
3068
3069 #ifdef CONFIG_F2FS_FS_COMPRESSION
3070                         if (f2fs_compressed_file(inode)) {
3071                                 get_page(page);
3072                                 f2fs_compress_ctx_add_page(&cc, page);
3073                                 continue;
3074                         }
3075 #endif
3076                         ret = f2fs_write_single_data_page(page, &submitted,
3077                                         &bio, &last_block, wbc, io_type, 0);
3078                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3079                                 unlock_page(page);
3080 #ifdef CONFIG_F2FS_FS_COMPRESSION
3081 result:
3082 #endif
3083                         nwritten += submitted;
3084                         wbc->nr_to_write -= submitted;
3085
3086                         if (unlikely(ret)) {
3087                                 /*
3088                                  * keep nr_to_write, since vfs uses this to
3089                                  * get # of written pages.
3090                                  */
3091                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3092                                         ret = 0;
3093                                         goto next;
3094                                 } else if (ret == -EAGAIN) {
3095                                         ret = 0;
3096                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3097                                                 cond_resched();
3098                                                 congestion_wait(BLK_RW_ASYNC,
3099                                                         DEFAULT_IO_TIMEOUT);
3100                                                 goto retry_write;
3101                                         }
3102                                         goto next;
3103                                 }
3104                                 done_index = page->index + 1;
3105                                 done = 1;
3106                                 break;
3107                         }
3108
3109                         if (wbc->nr_to_write <= 0 &&
3110                                         wbc->sync_mode == WB_SYNC_NONE) {
3111                                 done = 1;
3112                                 break;
3113                         }
3114 next:
3115                         if (need_readd)
3116                                 goto readd;
3117                 }
3118                 pagevec_release(&pvec);
3119                 cond_resched();
3120         }
3121 #ifdef CONFIG_F2FS_FS_COMPRESSION
3122         /* flush remained pages in compress cluster */
3123         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3124                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3125                 nwritten += submitted;
3126                 wbc->nr_to_write -= submitted;
3127                 if (ret) {
3128                         done = 1;
3129                         retry = 0;
3130                 }
3131         }
3132         if (f2fs_compressed_file(inode))
3133                 f2fs_destroy_compress_ctx(&cc);
3134 #endif
3135         if (retry) {
3136                 index = 0;
3137                 end = -1;
3138                 goto retry;
3139         }
3140         if (wbc->range_cyclic && !done)
3141                 done_index = 0;
3142         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3143                 mapping->writeback_index = done_index;
3144
3145         if (nwritten)
3146                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3147                                                                 NULL, 0, DATA);
3148         /* submit cached bio of IPU write */
3149         if (bio)
3150                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3151
3152         return ret;
3153 }
3154
3155 static inline bool __should_serialize_io(struct inode *inode,
3156                                         struct writeback_control *wbc)
3157 {
3158         /* to avoid deadlock in path of data flush */
3159         if (F2FS_I(inode)->cp_task)
3160                 return false;
3161
3162         if (!S_ISREG(inode->i_mode))
3163                 return false;
3164         if (IS_NOQUOTA(inode))
3165                 return false;
3166
3167         if (f2fs_compressed_file(inode))
3168                 return true;
3169         if (wbc->sync_mode != WB_SYNC_ALL)
3170                 return true;
3171         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3172                 return true;
3173         return false;
3174 }
3175
3176 static int __f2fs_write_data_pages(struct address_space *mapping,
3177                                                 struct writeback_control *wbc,
3178                                                 enum iostat_type io_type)
3179 {
3180         struct inode *inode = mapping->host;
3181         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3182         struct blk_plug plug;
3183         int ret;
3184         bool locked = false;
3185
3186         /* deal with chardevs and other special file */
3187         if (!mapping->a_ops->writepage)
3188                 return 0;
3189
3190         /* skip writing if there is no dirty page in this inode */
3191         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3192                 return 0;
3193
3194         /* during POR, we don't need to trigger writepage at all. */
3195         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3196                 goto skip_write;
3197
3198         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3199                         wbc->sync_mode == WB_SYNC_NONE &&
3200                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3201                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3202                 goto skip_write;
3203
3204         /* skip writing during file defragment */
3205         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3206                 goto skip_write;
3207
3208         trace_f2fs_writepages(mapping->host, wbc, DATA);
3209
3210         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3211         if (wbc->sync_mode == WB_SYNC_ALL)
3212                 atomic_inc(&sbi->wb_sync_req[DATA]);
3213         else if (atomic_read(&sbi->wb_sync_req[DATA]))
3214                 goto skip_write;
3215
3216         if (__should_serialize_io(inode, wbc)) {
3217                 mutex_lock(&sbi->writepages);
3218                 locked = true;
3219         }
3220
3221         blk_start_plug(&plug);
3222         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3223         blk_finish_plug(&plug);
3224
3225         if (locked)
3226                 mutex_unlock(&sbi->writepages);
3227
3228         if (wbc->sync_mode == WB_SYNC_ALL)
3229                 atomic_dec(&sbi->wb_sync_req[DATA]);
3230         /*
3231          * if some pages were truncated, we cannot guarantee its mapping->host
3232          * to detect pending bios.
3233          */
3234
3235         f2fs_remove_dirty_inode(inode);
3236         return ret;
3237
3238 skip_write:
3239         wbc->pages_skipped += get_dirty_pages(inode);
3240         trace_f2fs_writepages(mapping->host, wbc, DATA);
3241         return 0;
3242 }
3243
3244 static int f2fs_write_data_pages(struct address_space *mapping,
3245                             struct writeback_control *wbc)
3246 {
3247         struct inode *inode = mapping->host;
3248
3249         return __f2fs_write_data_pages(mapping, wbc,
3250                         F2FS_I(inode)->cp_task == current ?
3251                         FS_CP_DATA_IO : FS_DATA_IO);
3252 }
3253
3254 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
3255 {
3256         struct inode *inode = mapping->host;
3257         loff_t i_size = i_size_read(inode);
3258
3259         if (IS_NOQUOTA(inode))
3260                 return;
3261
3262         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3263         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3264                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3265                 down_write(&F2FS_I(inode)->i_mmap_sem);
3266
3267                 truncate_pagecache(inode, i_size);
3268                 f2fs_truncate_blocks(inode, i_size, true);
3269
3270                 up_write(&F2FS_I(inode)->i_mmap_sem);
3271                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3272         }
3273 }
3274
3275 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3276                         struct page *page, loff_t pos, unsigned len,
3277                         block_t *blk_addr, bool *node_changed)
3278 {
3279         struct inode *inode = page->mapping->host;
3280         pgoff_t index = page->index;
3281         struct dnode_of_data dn;
3282         struct page *ipage;
3283         bool locked = false;
3284         struct extent_info ei = {0,0,0};
3285         int err = 0;
3286         int flag;
3287
3288         /*
3289          * we already allocated all the blocks, so we don't need to get
3290          * the block addresses when there is no need to fill the page.
3291          */
3292         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3293             !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3294             !f2fs_verity_in_progress(inode))
3295                 return 0;
3296
3297         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3298         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3299                 flag = F2FS_GET_BLOCK_DEFAULT;
3300         else
3301                 flag = F2FS_GET_BLOCK_PRE_AIO;
3302
3303         if (f2fs_has_inline_data(inode) ||
3304                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3305                 f2fs_do_map_lock(sbi, flag, true);
3306                 locked = true;
3307         }
3308
3309 restart:
3310         /* check inline_data */
3311         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3312         if (IS_ERR(ipage)) {
3313                 err = PTR_ERR(ipage);
3314                 goto unlock_out;
3315         }
3316
3317         set_new_dnode(&dn, inode, ipage, ipage, 0);
3318
3319         if (f2fs_has_inline_data(inode)) {
3320                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3321                         f2fs_do_read_inline_data(page, ipage);
3322                         set_inode_flag(inode, FI_DATA_EXIST);
3323                         if (inode->i_nlink)
3324                                 set_inline_node(ipage);
3325                 } else {
3326                         err = f2fs_convert_inline_page(&dn, page);
3327                         if (err)
3328                                 goto out;
3329                         if (dn.data_blkaddr == NULL_ADDR)
3330                                 err = f2fs_get_block(&dn, index);
3331                 }
3332         } else if (locked) {
3333                 err = f2fs_get_block(&dn, index);
3334         } else {
3335                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3336                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3337                 } else {
3338                         /* hole case */
3339                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3340                         if (err || dn.data_blkaddr == NULL_ADDR) {
3341                                 f2fs_put_dnode(&dn);
3342                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3343                                                                 true);
3344                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3345                                 locked = true;
3346                                 goto restart;
3347                         }
3348                 }
3349         }
3350
3351         /* convert_inline_page can make node_changed */
3352         *blk_addr = dn.data_blkaddr;
3353         *node_changed = dn.node_changed;
3354 out:
3355         f2fs_put_dnode(&dn);
3356 unlock_out:
3357         if (locked)
3358                 f2fs_do_map_lock(sbi, flag, false);
3359         return err;
3360 }
3361
3362 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3363                 loff_t pos, unsigned len, unsigned flags,
3364                 struct page **pagep, void **fsdata)
3365 {
3366         struct inode *inode = mapping->host;
3367         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3368         struct page *page = NULL;
3369         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3370         bool need_balance = false, drop_atomic = false;
3371         block_t blkaddr = NULL_ADDR;
3372         int err = 0;
3373
3374         trace_f2fs_write_begin(inode, pos, len, flags);
3375
3376         if (!f2fs_is_checkpoint_ready(sbi)) {
3377                 err = -ENOSPC;
3378                 goto fail;
3379         }
3380
3381         if ((f2fs_is_atomic_file(inode) &&
3382                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3383                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3384                 err = -ENOMEM;
3385                 drop_atomic = true;
3386                 goto fail;
3387         }
3388
3389         /*
3390          * We should check this at this moment to avoid deadlock on inode page
3391          * and #0 page. The locking rule for inline_data conversion should be:
3392          * lock_page(page #0) -> lock_page(inode_page)
3393          */
3394         if (index != 0) {
3395                 err = f2fs_convert_inline_inode(inode);
3396                 if (err)
3397                         goto fail;
3398         }
3399
3400 #ifdef CONFIG_F2FS_FS_COMPRESSION
3401         if (f2fs_compressed_file(inode)) {
3402                 int ret;
3403
3404                 *fsdata = NULL;
3405
3406                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3407                                                         index, fsdata);
3408                 if (ret < 0) {
3409                         err = ret;
3410                         goto fail;
3411                 } else if (ret) {
3412                         return 0;
3413                 }
3414         }
3415 #endif
3416
3417 repeat:
3418         /*
3419          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3420          * wait_for_stable_page. Will wait that below with our IO control.
3421          */
3422         page = f2fs_pagecache_get_page(mapping, index,
3423                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3424         if (!page) {
3425                 err = -ENOMEM;
3426                 goto fail;
3427         }
3428
3429         /* TODO: cluster can be compressed due to race with .writepage */
3430
3431         *pagep = page;
3432
3433         err = prepare_write_begin(sbi, page, pos, len,
3434                                         &blkaddr, &need_balance);
3435         if (err)
3436                 goto fail;
3437
3438         if (need_balance && !IS_NOQUOTA(inode) &&
3439                         has_not_enough_free_secs(sbi, 0, 0)) {
3440                 unlock_page(page);
3441                 f2fs_balance_fs(sbi, true);
3442                 lock_page(page);
3443                 if (page->mapping != mapping) {
3444                         /* The page got truncated from under us */
3445                         f2fs_put_page(page, 1);
3446                         goto repeat;
3447                 }
3448         }
3449
3450         f2fs_wait_on_page_writeback(page, DATA, false, true);
3451
3452         if (len == PAGE_SIZE || PageUptodate(page))
3453                 return 0;
3454
3455         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3456             !f2fs_verity_in_progress(inode)) {
3457                 zero_user_segment(page, len, PAGE_SIZE);
3458                 return 0;
3459         }
3460
3461         if (blkaddr == NEW_ADDR) {
3462                 zero_user_segment(page, 0, PAGE_SIZE);
3463                 SetPageUptodate(page);
3464         } else {
3465                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3466                                 DATA_GENERIC_ENHANCE_READ)) {
3467                         err = -EFSCORRUPTED;
3468                         goto fail;
3469                 }
3470                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3471                 if (err)
3472                         goto fail;
3473
3474                 lock_page(page);
3475                 if (unlikely(page->mapping != mapping)) {
3476                         f2fs_put_page(page, 1);
3477                         goto repeat;
3478                 }
3479                 if (unlikely(!PageUptodate(page))) {
3480                         err = -EIO;
3481                         goto fail;
3482                 }
3483         }
3484         return 0;
3485
3486 fail:
3487         f2fs_put_page(page, 1);
3488         f2fs_write_failed(mapping, pos + len);
3489         if (drop_atomic)
3490                 f2fs_drop_inmem_pages_all(sbi, false);
3491         return err;
3492 }
3493
3494 static int f2fs_write_end(struct file *file,
3495                         struct address_space *mapping,
3496                         loff_t pos, unsigned len, unsigned copied,
3497                         struct page *page, void *fsdata)
3498 {
3499         struct inode *inode = page->mapping->host;
3500
3501         trace_f2fs_write_end(inode, pos, len, copied);
3502
3503         /*
3504          * This should be come from len == PAGE_SIZE, and we expect copied
3505          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3506          * let generic_perform_write() try to copy data again through copied=0.
3507          */
3508         if (!PageUptodate(page)) {
3509                 if (unlikely(copied != len))
3510                         copied = 0;
3511                 else
3512                         SetPageUptodate(page);
3513         }
3514
3515 #ifdef CONFIG_F2FS_FS_COMPRESSION
3516         /* overwrite compressed file */
3517         if (f2fs_compressed_file(inode) && fsdata) {
3518                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3519                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3520
3521                 if (pos + copied > i_size_read(inode) &&
3522                                 !f2fs_verity_in_progress(inode))
3523                         f2fs_i_size_write(inode, pos + copied);
3524                 return copied;
3525         }
3526 #endif
3527
3528         if (!copied)
3529                 goto unlock_out;
3530
3531         set_page_dirty(page);
3532
3533         if (pos + copied > i_size_read(inode) &&
3534             !f2fs_verity_in_progress(inode))
3535                 f2fs_i_size_write(inode, pos + copied);
3536 unlock_out:
3537         f2fs_put_page(page, 1);
3538         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3539         return copied;
3540 }
3541
3542 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3543                            loff_t offset)
3544 {
3545         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3546         unsigned blkbits = i_blkbits;
3547         unsigned blocksize_mask = (1 << blkbits) - 1;
3548         unsigned long align = offset | iov_iter_alignment(iter);
3549         struct block_device *bdev = inode->i_sb->s_bdev;
3550
3551         if (iov_iter_rw(iter) == READ && offset >= i_size_read(inode))
3552                 return 1;
3553
3554         if (align & blocksize_mask) {
3555                 if (bdev)
3556                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
3557                 blocksize_mask = (1 << blkbits) - 1;
3558                 if (align & blocksize_mask)
3559                         return -EINVAL;
3560                 return 1;
3561         }
3562         return 0;
3563 }
3564
3565 static void f2fs_dio_end_io(struct bio *bio)
3566 {
3567         struct f2fs_private_dio *dio = bio->bi_private;
3568
3569         dec_page_count(F2FS_I_SB(dio->inode),
3570                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3571
3572         bio->bi_private = dio->orig_private;
3573         bio->bi_end_io = dio->orig_end_io;
3574
3575         kfree(dio);
3576
3577         bio_endio(bio);
3578 }
3579
3580 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3581                                                         loff_t file_offset)
3582 {
3583         struct f2fs_private_dio *dio;
3584         bool write = (bio_op(bio) == REQ_OP_WRITE);
3585
3586         dio = f2fs_kzalloc(F2FS_I_SB(inode),
3587                         sizeof(struct f2fs_private_dio), GFP_NOFS);
3588         if (!dio)
3589                 goto out;
3590
3591         dio->inode = inode;
3592         dio->orig_end_io = bio->bi_end_io;
3593         dio->orig_private = bio->bi_private;
3594         dio->write = write;
3595
3596         bio->bi_end_io = f2fs_dio_end_io;
3597         bio->bi_private = dio;
3598
3599         inc_page_count(F2FS_I_SB(inode),
3600                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3601
3602         submit_bio(bio);
3603         return;
3604 out:
3605         bio->bi_status = BLK_STS_IOERR;
3606         bio_endio(bio);
3607 }
3608
3609 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3610 {
3611         struct address_space *mapping = iocb->ki_filp->f_mapping;
3612         struct inode *inode = mapping->host;
3613         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3614         struct f2fs_inode_info *fi = F2FS_I(inode);
3615         size_t count = iov_iter_count(iter);
3616         loff_t offset = iocb->ki_pos;
3617         int rw = iov_iter_rw(iter);
3618         int err;
3619         enum rw_hint hint = iocb->ki_hint;
3620         int whint_mode = F2FS_OPTION(sbi).whint_mode;
3621         bool do_opu;
3622
3623         err = check_direct_IO(inode, iter, offset);
3624         if (err)
3625                 return err < 0 ? err : 0;
3626
3627         if (f2fs_force_buffered_io(inode, iocb, iter))
3628                 return 0;
3629
3630         do_opu = allow_outplace_dio(inode, iocb, iter);
3631
3632         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3633
3634         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3635                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
3636
3637         if (iocb->ki_flags & IOCB_NOWAIT) {
3638                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3639                         iocb->ki_hint = hint;
3640                         err = -EAGAIN;
3641                         goto out;
3642                 }
3643                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3644                         up_read(&fi->i_gc_rwsem[rw]);
3645                         iocb->ki_hint = hint;
3646                         err = -EAGAIN;
3647                         goto out;
3648                 }
3649         } else {
3650                 down_read(&fi->i_gc_rwsem[rw]);
3651                 if (do_opu)
3652                         down_read(&fi->i_gc_rwsem[READ]);
3653         }
3654
3655         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3656                         iter, rw == WRITE ? get_data_block_dio_write :
3657                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
3658                         rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3659                         DIO_SKIP_HOLES);
3660
3661         if (do_opu)
3662                 up_read(&fi->i_gc_rwsem[READ]);
3663
3664         up_read(&fi->i_gc_rwsem[rw]);
3665
3666         if (rw == WRITE) {
3667                 if (whint_mode == WHINT_MODE_OFF)
3668                         iocb->ki_hint = hint;
3669                 if (err > 0) {
3670                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3671                                                                         err);
3672                         if (!do_opu)
3673                                 set_inode_flag(inode, FI_UPDATE_WRITE);
3674                 } else if (err == -EIOCBQUEUED) {
3675                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3676                                                 count - iov_iter_count(iter));
3677                 } else if (err < 0) {
3678                         f2fs_write_failed(mapping, offset + count);
3679                 }
3680         } else {
3681                 if (err > 0)
3682                         f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3683                 else if (err == -EIOCBQUEUED)
3684                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_READ_IO,
3685                                                 count - iov_iter_count(iter));
3686         }
3687
3688 out:
3689         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3690
3691         return err;
3692 }
3693
3694 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3695                                                         unsigned int length)
3696 {
3697         struct inode *inode = page->mapping->host;
3698         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3699
3700         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3701                 (offset % PAGE_SIZE || length != PAGE_SIZE))
3702                 return;
3703
3704         if (PageDirty(page)) {
3705                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3706                         dec_page_count(sbi, F2FS_DIRTY_META);
3707                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3708                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3709                 } else {
3710                         inode_dec_dirty_pages(inode);
3711                         f2fs_remove_dirty_inode(inode);
3712                 }
3713         }
3714
3715         clear_cold_data(page);
3716
3717         if (IS_ATOMIC_WRITTEN_PAGE(page))
3718                 return f2fs_drop_inmem_page(inode, page);
3719
3720         f2fs_clear_page_private(page);
3721 }
3722
3723 int f2fs_release_page(struct page *page, gfp_t wait)
3724 {
3725         /* If this is dirty page, keep PagePrivate */
3726         if (PageDirty(page))
3727                 return 0;
3728
3729         /* This is atomic written page, keep Private */
3730         if (IS_ATOMIC_WRITTEN_PAGE(page))
3731                 return 0;
3732
3733         clear_cold_data(page);
3734         f2fs_clear_page_private(page);
3735         return 1;
3736 }
3737
3738 static int f2fs_set_data_page_dirty(struct page *page)
3739 {
3740         struct inode *inode = page_file_mapping(page)->host;
3741
3742         trace_f2fs_set_page_dirty(page, DATA);
3743
3744         if (!PageUptodate(page))
3745                 SetPageUptodate(page);
3746         if (PageSwapCache(page))
3747                 return __set_page_dirty_nobuffers(page);
3748
3749         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3750                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
3751                         f2fs_register_inmem_page(inode, page);
3752                         return 1;
3753                 }
3754                 /*
3755                  * Previously, this page has been registered, we just
3756                  * return here.
3757                  */
3758                 return 0;
3759         }
3760
3761         if (!PageDirty(page)) {
3762                 __set_page_dirty_nobuffers(page);
3763                 f2fs_update_dirty_page(inode, page);
3764                 return 1;
3765         }
3766         return 0;
3767 }
3768
3769
3770 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3771 {
3772 #ifdef CONFIG_F2FS_FS_COMPRESSION
3773         struct dnode_of_data dn;
3774         sector_t start_idx, blknr = 0;
3775         int ret;
3776
3777         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3778
3779         set_new_dnode(&dn, inode, NULL, NULL, 0);
3780         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3781         if (ret)
3782                 return 0;
3783
3784         if (dn.data_blkaddr != COMPRESS_ADDR) {
3785                 dn.ofs_in_node += block - start_idx;
3786                 blknr = f2fs_data_blkaddr(&dn);
3787                 if (!__is_valid_data_blkaddr(blknr))
3788                         blknr = 0;
3789         }
3790
3791         f2fs_put_dnode(&dn);
3792         return blknr;
3793 #else
3794         return 0;
3795 #endif
3796 }
3797
3798
3799 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3800 {
3801         struct inode *inode = mapping->host;
3802         struct buffer_head tmp = {
3803                 .b_size = i_blocksize(inode),
3804         };
3805         sector_t blknr = 0;
3806
3807         if (f2fs_has_inline_data(inode))
3808                 goto out;
3809
3810         /* make sure allocating whole blocks */
3811         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3812                 filemap_write_and_wait(mapping);
3813
3814         /* Block number less than F2FS MAX BLOCKS */
3815         if (unlikely(block >= F2FS_I_SB(inode)->max_file_blocks))
3816                 goto out;
3817
3818         if (f2fs_compressed_file(inode)) {
3819                 blknr = f2fs_bmap_compress(inode, block);
3820         } else {
3821                 if (!get_data_block_bmap(inode, block, &tmp, 0))
3822                         blknr = tmp.b_blocknr;
3823         }
3824 out:
3825         trace_f2fs_bmap(inode, block, blknr);
3826         return blknr;
3827 }
3828
3829 #ifdef CONFIG_MIGRATION
3830 #include <linux/migrate.h>
3831
3832 int f2fs_migrate_page(struct address_space *mapping,
3833                 struct page *newpage, struct page *page, enum migrate_mode mode)
3834 {
3835         int rc, extra_count;
3836         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3837         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
3838
3839         BUG_ON(PageWriteback(page));
3840
3841         /* migrating an atomic written page is safe with the inmem_lock hold */
3842         if (atomic_written) {
3843                 if (mode != MIGRATE_SYNC)
3844                         return -EBUSY;
3845                 if (!mutex_trylock(&fi->inmem_lock))
3846                         return -EAGAIN;
3847         }
3848
3849         /* one extra reference was held for atomic_write page */
3850         extra_count = atomic_written ? 1 : 0;
3851         rc = migrate_page_move_mapping(mapping, newpage,
3852                                 page, extra_count);
3853         if (rc != MIGRATEPAGE_SUCCESS) {
3854                 if (atomic_written)
3855                         mutex_unlock(&fi->inmem_lock);
3856                 return rc;
3857         }
3858
3859         if (atomic_written) {
3860                 struct inmem_pages *cur;
3861                 list_for_each_entry(cur, &fi->inmem_pages, list)
3862                         if (cur->page == page) {
3863                                 cur->page = newpage;
3864                                 break;
3865                         }
3866                 mutex_unlock(&fi->inmem_lock);
3867                 put_page(page);
3868                 get_page(newpage);
3869         }
3870
3871         if (PagePrivate(page)) {
3872                 f2fs_set_page_private(newpage, page_private(page));
3873                 f2fs_clear_page_private(page);
3874         }
3875
3876         if (mode != MIGRATE_SYNC_NO_COPY)
3877                 migrate_page_copy(newpage, page);
3878         else
3879                 migrate_page_states(newpage, page);
3880
3881         return MIGRATEPAGE_SUCCESS;
3882 }
3883 #endif
3884
3885 #ifdef CONFIG_SWAP
3886 static int check_swap_activate_fast(struct swap_info_struct *sis,
3887                                 struct file *swap_file, sector_t *span)
3888 {
3889         struct address_space *mapping = swap_file->f_mapping;
3890         struct inode *inode = mapping->host;
3891         sector_t cur_lblock;
3892         sector_t last_lblock;
3893         sector_t pblock;
3894         sector_t lowest_pblock = -1;
3895         sector_t highest_pblock = 0;
3896         int nr_extents = 0;
3897         unsigned long nr_pblocks;
3898         unsigned long len;
3899         int ret;
3900
3901         /*
3902          * Map all the blocks into the extent list.  This code doesn't try
3903          * to be very smart.
3904          */
3905         cur_lblock = 0;
3906         last_lblock = logical_to_blk(inode, i_size_read(inode));
3907         len = i_size_read(inode);
3908
3909         while (cur_lblock <= last_lblock && cur_lblock < sis->max) {
3910                 struct buffer_head map_bh;
3911                 pgoff_t next_pgofs;
3912
3913                 cond_resched();
3914
3915                 memset(&map_bh, 0, sizeof(struct buffer_head));
3916                 map_bh.b_size = len - cur_lblock;
3917
3918                 ret = get_data_block(inode, cur_lblock, &map_bh, 0,
3919                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
3920                 if (ret)
3921                         goto err_out;
3922
3923                 /* hole */
3924                 if (!buffer_mapped(&map_bh))
3925                         goto err_out;
3926
3927                 pblock = map_bh.b_blocknr;
3928                 nr_pblocks = logical_to_blk(inode, map_bh.b_size);
3929
3930                 if (cur_lblock + nr_pblocks >= sis->max)
3931                         nr_pblocks = sis->max - cur_lblock;
3932
3933                 if (cur_lblock) {       /* exclude the header page */
3934                         if (pblock < lowest_pblock)
3935                                 lowest_pblock = pblock;
3936                         if (pblock + nr_pblocks - 1 > highest_pblock)
3937                                 highest_pblock = pblock + nr_pblocks - 1;
3938                 }
3939
3940                 /*
3941                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3942                  */
3943                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3944                 if (ret < 0)
3945                         goto out;
3946                 nr_extents += ret;
3947                 cur_lblock += nr_pblocks;
3948         }
3949         ret = nr_extents;
3950         *span = 1 + highest_pblock - lowest_pblock;
3951         if (cur_lblock == 0)
3952                 cur_lblock = 1; /* force Empty message */
3953         sis->max = cur_lblock;
3954         sis->pages = cur_lblock - 1;
3955         sis->highest_bit = cur_lblock - 1;
3956 out:
3957         return ret;
3958 err_out:
3959         pr_err("swapon: swapfile has holes\n");
3960         return -EINVAL;
3961 }
3962
3963 /* Copied from generic_swapfile_activate() to check any holes */
3964 static int check_swap_activate(struct swap_info_struct *sis,
3965                                 struct file *swap_file, sector_t *span)
3966 {
3967         struct address_space *mapping = swap_file->f_mapping;
3968         struct inode *inode = mapping->host;
3969         unsigned blocks_per_page;
3970         unsigned long page_no;
3971         unsigned blkbits;
3972         sector_t probe_block;
3973         sector_t last_block;
3974         sector_t lowest_block = -1;
3975         sector_t highest_block = 0;
3976         int nr_extents = 0;
3977         int ret;
3978
3979         if (PAGE_SIZE == F2FS_BLKSIZE)
3980                 return check_swap_activate_fast(sis, swap_file, span);
3981
3982         blkbits = inode->i_blkbits;
3983         blocks_per_page = PAGE_SIZE >> blkbits;
3984
3985         /*
3986          * Map all the blocks into the extent list.  This code doesn't try
3987          * to be very smart.
3988          */
3989         probe_block = 0;
3990         page_no = 0;
3991         last_block = i_size_read(inode) >> blkbits;
3992         while ((probe_block + blocks_per_page) <= last_block &&
3993                         page_no < sis->max) {
3994                 unsigned block_in_page;
3995                 sector_t first_block;
3996                 sector_t block = 0;
3997                 int      err = 0;
3998
3999                 cond_resched();
4000
4001                 block = probe_block;
4002                 err = bmap(inode, &block);
4003                 if (err || !block)
4004                         goto bad_bmap;
4005                 first_block = block;
4006
4007                 /*
4008                  * It must be PAGE_SIZE aligned on-disk
4009                  */
4010                 if (first_block & (blocks_per_page - 1)) {
4011                         probe_block++;
4012                         goto reprobe;
4013                 }
4014
4015                 for (block_in_page = 1; block_in_page < blocks_per_page;
4016                                         block_in_page++) {
4017
4018                         block = probe_block + block_in_page;
4019                         err = bmap(inode, &block);
4020
4021                         if (err || !block)
4022                                 goto bad_bmap;
4023
4024                         if (block != first_block + block_in_page) {
4025                                 /* Discontiguity */
4026                                 probe_block++;
4027                                 goto reprobe;
4028                         }
4029                 }
4030
4031                 first_block >>= (PAGE_SHIFT - blkbits);
4032                 if (page_no) {  /* exclude the header page */
4033                         if (first_block < lowest_block)
4034                                 lowest_block = first_block;
4035                         if (first_block > highest_block)
4036                                 highest_block = first_block;
4037                 }
4038
4039                 /*
4040                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4041                  */
4042                 ret = add_swap_extent(sis, page_no, 1, first_block);
4043                 if (ret < 0)
4044                         goto out;
4045                 nr_extents += ret;
4046                 page_no++;
4047                 probe_block += blocks_per_page;
4048 reprobe:
4049                 continue;
4050         }
4051         ret = nr_extents;
4052         *span = 1 + highest_block - lowest_block;
4053         if (page_no == 0)
4054                 page_no = 1;    /* force Empty message */
4055         sis->max = page_no;
4056         sis->pages = page_no - 1;
4057         sis->highest_bit = page_no - 1;
4058 out:
4059         return ret;
4060 bad_bmap:
4061         pr_err("swapon: swapfile has holes\n");
4062         return -EINVAL;
4063 }
4064
4065 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4066                                 sector_t *span)
4067 {
4068         struct inode *inode = file_inode(file);
4069         int ret;
4070
4071         if (!S_ISREG(inode->i_mode))
4072                 return -EINVAL;
4073
4074         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4075                 return -EROFS;
4076
4077         ret = f2fs_convert_inline_inode(inode);
4078         if (ret)
4079                 return ret;
4080
4081         if (!f2fs_disable_compressed_file(inode))
4082                 return -EINVAL;
4083
4084         ret = check_swap_activate(sis, file, span);
4085         if (ret < 0)
4086                 return ret;
4087
4088         set_inode_flag(inode, FI_PIN_FILE);
4089         f2fs_precache_extents(inode);
4090         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4091         return ret;
4092 }
4093
4094 static void f2fs_swap_deactivate(struct file *file)
4095 {
4096         struct inode *inode = file_inode(file);
4097
4098         clear_inode_flag(inode, FI_PIN_FILE);
4099 }
4100 #else
4101 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4102                                 sector_t *span)
4103 {
4104         return -EOPNOTSUPP;
4105 }
4106
4107 static void f2fs_swap_deactivate(struct file *file)
4108 {
4109 }
4110 #endif
4111
4112 const struct address_space_operations f2fs_dblock_aops = {
4113         .readpage       = f2fs_read_data_page,
4114         .readahead      = f2fs_readahead,
4115         .writepage      = f2fs_write_data_page,
4116         .writepages     = f2fs_write_data_pages,
4117         .write_begin    = f2fs_write_begin,
4118         .write_end      = f2fs_write_end,
4119         .set_page_dirty = f2fs_set_data_page_dirty,
4120         .invalidatepage = f2fs_invalidate_page,
4121         .releasepage    = f2fs_release_page,
4122         .direct_IO      = f2fs_direct_IO,
4123         .bmap           = f2fs_bmap,
4124         .swap_activate  = f2fs_swap_activate,
4125         .swap_deactivate = f2fs_swap_deactivate,
4126 #ifdef CONFIG_MIGRATION
4127         .migratepage    = f2fs_migrate_page,
4128 #endif
4129 };
4130
4131 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4132 {
4133         struct address_space *mapping = page_mapping(page);
4134         unsigned long flags;
4135
4136         xa_lock_irqsave(&mapping->i_pages, flags);
4137         __xa_clear_mark(&mapping->i_pages, page_index(page),
4138                                                 PAGECACHE_TAG_DIRTY);
4139         xa_unlock_irqrestore(&mapping->i_pages, flags);
4140 }
4141
4142 int __init f2fs_init_post_read_processing(void)
4143 {
4144         bio_post_read_ctx_cache =
4145                 kmem_cache_create("f2fs_bio_post_read_ctx",
4146                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4147         if (!bio_post_read_ctx_cache)
4148                 goto fail;
4149         bio_post_read_ctx_pool =
4150                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4151                                          bio_post_read_ctx_cache);
4152         if (!bio_post_read_ctx_pool)
4153                 goto fail_free_cache;
4154         return 0;
4155
4156 fail_free_cache:
4157         kmem_cache_destroy(bio_post_read_ctx_cache);
4158 fail:
4159         return -ENOMEM;
4160 }
4161
4162 void f2fs_destroy_post_read_processing(void)
4163 {
4164         mempool_destroy(bio_post_read_ctx_pool);
4165         kmem_cache_destroy(bio_post_read_ctx_cache);
4166 }
4167
4168 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4169 {
4170         if (!f2fs_sb_has_encrypt(sbi) &&
4171                 !f2fs_sb_has_verity(sbi) &&
4172                 !f2fs_sb_has_compression(sbi))
4173                 return 0;
4174
4175         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4176                                                  WQ_UNBOUND | WQ_HIGHPRI,
4177                                                  num_online_cpus());
4178         if (!sbi->post_read_wq)
4179                 return -ENOMEM;
4180         return 0;
4181 }
4182
4183 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4184 {
4185         if (sbi->post_read_wq)
4186                 destroy_workqueue(sbi->post_read_wq);
4187 }
4188
4189 int __init f2fs_init_bio_entry_cache(void)
4190 {
4191         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4192                         sizeof(struct bio_entry));
4193         if (!bio_entry_slab)
4194                 return -ENOMEM;
4195         return 0;
4196 }
4197
4198 void f2fs_destroy_bio_entry_cache(void)
4199 {
4200         kmem_cache_destroy(bio_entry_slab);
4201 }