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->verity_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 (unlikely(sbi->max_io_bytes &&
 740                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 741                 return false;
 742         if (last_blkaddr + 1 != cur_blkaddr)
 743                 return false;
 744         return __same_bdev(sbi, cur_blkaddr, bio);
 745 }
 746
 747 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 748                                                 struct f2fs_io_info *fio)
 749 {
 750         if (io->fio.op != fio->op)
 751                 return false;
 752         return io->fio.op_flags == fio->op_flags;
 753 }
 754
 755 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 756                                         struct f2fs_bio_info *io,
 757                                         struct f2fs_io_info *fio,
 758                                         block_t last_blkaddr,
 759                                         block_t cur_blkaddr)
 760 {
 761         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 762                 unsigned int filled_blocks =
 763                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 764                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 765                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 766
 767                 /* IOs in bio is aligned and left space of vectors is not enough */
 768                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 769                         return false;
 770         }
 771         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 772                 return false;
 773         return io_type_is_mergeable(io, fio);
 774 }
 775
 776 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 777                                 struct page *page, enum temp_type temp)
 778 {
 779         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 780         struct bio_entry *be;
 781
 782         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 783         be->bio = bio;
 784         bio_get(bio);
 785
 786         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 787                 f2fs_bug_on(sbi, 1);
 788
 789         down_write(&io->bio_list_lock);
 790         list_add_tail(&be->list, &io->bio_list);
 791         up_write(&io->bio_list_lock);
 792 }
 793
 794 static void del_bio_entry(struct bio_entry *be)
 795 {
 796         list_del(&be->list);
 797         kmem_cache_free(bio_entry_slab, be);
 798 }
 799
 800 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 801                                                         struct page *page)
 802 {
 803         struct f2fs_sb_info *sbi = fio->sbi;
 804         enum temp_type temp;
 805         bool found = false;
 806         int ret = -EAGAIN;
 807
 808         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 809                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 810                 struct list_head *head = &io->bio_list;
 811                 struct bio_entry *be;
 812
 813                 down_write(&io->bio_list_lock);
 814                 list_for_each_entry(be, head, list) {
 815                         if (be->bio != *bio)
 816                                 continue;
 817
 818                         found = true;
 819
 820                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 821                                                             *fio->last_block,
 822                                                             fio->new_blkaddr));
 823                         if (f2fs_crypt_mergeable_bio(*bio,
 824                                         fio->page->mapping->host,
 825                                         fio->page->index, fio) &&
 826                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 827                                         PAGE_SIZE) {
 828                                 ret = 0;
 829                                 break;
 830                         }
 831
 832                         /* page can't be merged into bio; submit the bio */
 833                         del_bio_entry(be);
 834                         __submit_bio(sbi, *bio, DATA);
 835                         break;
 836                 }
 837                 up_write(&io->bio_list_lock);
 838         }
 839
 840         if (ret) {
 841                 bio_put(*bio);
 842                 *bio = NULL;
 843         }
 844
 845         return ret;
 846 }
 847
 848 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 849                                         struct bio **bio, struct page *page)
 850 {
 851         enum temp_type temp;
 852         bool found = false;
 853         struct bio *target = bio ? *bio : NULL;
 854
 855         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 856                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 857                 struct list_head *head = &io->bio_list;
 858                 struct bio_entry *be;
 859
 860                 if (list_empty(head))
 861                         continue;
 862
 863                 down_read(&io->bio_list_lock);
 864                 list_for_each_entry(be, head, list) {
 865                         if (target)
 866                                 found = (target == be->bio);
 867                         else
 868                                 found = __has_merged_page(be->bio, NULL,
 869                                                                 page, 0);
 870                         if (found)
 871                                 break;
 872                 }
 873                 up_read(&io->bio_list_lock);
 874
 875                 if (!found)
 876                         continue;
 877
 878                 found = false;
 879
 880                 down_write(&io->bio_list_lock);
 881                 list_for_each_entry(be, head, list) {
 882                         if (target)
 883                                 found = (target == be->bio);
 884                         else
 885                                 found = __has_merged_page(be->bio, NULL,
 886                                                                 page, 0);
 887                         if (found) {
 888                                 target = be->bio;
 889                                 del_bio_entry(be);
 890                                 break;
 891                         }
 892                 }
 893                 up_write(&io->bio_list_lock);
 894         }
 895
 896         if (found)
 897                 __submit_bio(sbi, target, DATA);
 898         if (bio && *bio) {
 899                 bio_put(*bio);
 900                 *bio = NULL;
 901         }
 902 }
 903
 904 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 905 {
 906         struct bio *bio = *fio->bio;
 907         struct page *page = fio->encrypted_page ?
 908                         fio->encrypted_page : fio->page;
 909
 910         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 911                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 912                 return -EFSCORRUPTED;
 913
 914         trace_f2fs_submit_page_bio(page, fio);
 915         f2fs_trace_ios(fio, 0);
 916
 917         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 918                                                 fio->new_blkaddr))
 919                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 920 alloc_new:
 921         if (!bio) {
 922                 bio = __bio_alloc(fio, BIO_MAX_PAGES);
 923                 __attach_io_flag(fio);
 924                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 925                                        fio->page->index, fio, GFP_NOIO);
 926                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 927
 928                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 929         } else {
 930                 if (add_ipu_page(fio, &bio, page))
 931                         goto alloc_new;
 932         }
 933
 934         if (fio->io_wbc)
 935                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 936
 937         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 938
 939         *fio->last_block = fio->new_blkaddr;
 940         *fio->bio = bio;
 941
 942         return 0;
 943 }
 944
 945 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 946 {
 947         struct f2fs_sb_info *sbi = fio->sbi;
 948         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 949         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 950         struct page *bio_page;
 951
 952         f2fs_bug_on(sbi, is_read_io(fio->op));
 953
 954         down_write(&io->io_rwsem);
 955 next:
 956         if (fio->in_list) {
 957                 spin_lock(&io->io_lock);
 958                 if (list_empty(&io->io_list)) {
 959                         spin_unlock(&io->io_lock);
 960                         goto out;
 961                 }
 962                 fio = list_first_entry(&io->io_list,
 963                                                 struct f2fs_io_info, list);
 964                 list_del(&fio->list);
 965                 spin_unlock(&io->io_lock);
 966         }
 967
 968         verify_fio_blkaddr(fio);
 969
 970         if (fio->encrypted_page)
 971                 bio_page = fio->encrypted_page;
 972         else if (fio->compressed_page)
 973                 bio_page = fio->compressed_page;
 974         else
 975                 bio_page = fio->page;
 976
 977         /* set submitted = true as a return value */
 978         fio->submitted = true;
 979
 980         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 981
 982         if (io->bio &&
 983             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 984                               fio->new_blkaddr) ||
 985              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 986                                        bio_page->index, fio)))
 987                 __submit_merged_bio(io);
 988 alloc_new:
 989         if (io->bio == NULL) {
 990                 if (F2FS_IO_ALIGNED(sbi) &&
 991                                 (fio->type == DATA || fio->type == NODE) &&
 992                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 993                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 994                         fio->retry = true;
 995                         goto skip;
 996                 }
 997                 io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
 998                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 999                                        bio_page->index, fio, GFP_NOIO);
1000                 io->fio = *fio;
1001         }
1002
1003         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1004                 __submit_merged_bio(io);
1005                 goto alloc_new;
1006         }
1007
1008         if (fio->io_wbc)
1009                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
1010
1011         io->last_block_in_bio = fio->new_blkaddr;
1012         f2fs_trace_ios(fio, 0);
1013
1014         trace_f2fs_submit_page_write(fio->page, fio);
1015 skip:
1016         if (fio->in_list)
1017                 goto next;
1018 out:
1019         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1020                                 !f2fs_is_checkpoint_ready(sbi))
1021                 __submit_merged_bio(io);
1022         up_write(&io->io_rwsem);
1023 }
1024
1025 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
1026 {
1027         return fsverity_active(inode) &&
1028                idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
1029 }
1030
1031 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1032                                       unsigned nr_pages, unsigned op_flag,
1033                                       pgoff_t first_idx, bool for_write,
1034                                       bool for_verity)
1035 {
1036         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1037         struct bio *bio;
1038         struct bio_post_read_ctx *ctx;
1039         unsigned int post_read_steps = 0;
1040
1041         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES),
1042                                                                 for_write);
1043         if (!bio)
1044                 return ERR_PTR(-ENOMEM);
1045
1046         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1047
1048         f2fs_target_device(sbi, blkaddr, bio);
1049         bio->bi_end_io = f2fs_read_end_io;
1050         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
1051
1052         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1053                 post_read_steps |= 1 << STEP_DECRYPT;
1054         if (f2fs_compressed_file(inode))
1055                 post_read_steps |= 1 << STEP_DECOMPRESS_NOWQ;
1056         if (for_verity && f2fs_need_verity(inode, first_idx))
1057                 post_read_steps |= 1 << STEP_VERITY;
1058
1059         if (post_read_steps) {
1060                 /* Due to the mempool, this never fails. */
1061                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1062                 ctx->bio = bio;
1063                 ctx->sbi = sbi;
1064                 ctx->enabled_steps = post_read_steps;
1065                 bio->bi_private = ctx;
1066         }
1067
1068         return bio;
1069 }
1070
1071 static void f2fs_release_read_bio(struct bio *bio)
1072 {
1073         if (bio->bi_private)
1074                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1075         bio_put(bio);
1076 }
1077
1078 /* This can handle encryption stuffs */
1079 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1080                                  block_t blkaddr, int op_flags, bool for_write)
1081 {
1082         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1083         struct bio *bio;
1084
1085         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1086                                         page->index, for_write, true);
1087         if (IS_ERR(bio))
1088                 return PTR_ERR(bio);
1089
1090         /* wait for GCed page writeback via META_MAPPING */
1091         f2fs_wait_on_block_writeback(inode, blkaddr);
1092
1093         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1094                 bio_put(bio);
1095                 return -EFAULT;
1096         }
1097         ClearPageError(page);
1098         inc_page_count(sbi, F2FS_RD_DATA);
1099         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1100         __submit_bio(sbi, bio, DATA);
1101         return 0;
1102 }
1103
1104 static void __set_data_blkaddr(struct dnode_of_data *dn)
1105 {
1106         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1107         __le32 *addr_array;
1108         int base = 0;
1109
1110         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1111                 base = get_extra_isize(dn->inode);
1112
1113         /* Get physical address of data block */
1114         addr_array = blkaddr_in_node(rn);
1115         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1116 }
1117
1118 /*
1119  * Lock ordering for the change of data block address:
1120  * ->data_page
1121  *  ->node_page
1122  *    update block addresses in the node page
1123  */
1124 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1125 {
1126         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1127         __set_data_blkaddr(dn);
1128         if (set_page_dirty(dn->node_page))
1129                 dn->node_changed = true;
1130 }
1131
1132 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1133 {
1134         dn->data_blkaddr = blkaddr;
1135         f2fs_set_data_blkaddr(dn);
1136         f2fs_update_extent_cache(dn);
1137 }
1138
1139 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1140 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1141 {
1142         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1143         int err;
1144
1145         if (!count)
1146                 return 0;
1147
1148         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1149                 return -EPERM;
1150         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1151                 return err;
1152
1153         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1154                                                 dn->ofs_in_node, count);
1155
1156         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1157
1158         for (; count > 0; dn->ofs_in_node++) {
1159                 block_t blkaddr = f2fs_data_blkaddr(dn);
1160                 if (blkaddr == NULL_ADDR) {
1161                         dn->data_blkaddr = NEW_ADDR;
1162                         __set_data_blkaddr(dn);
1163                         count--;
1164                 }
1165         }
1166
1167         if (set_page_dirty(dn->node_page))
1168                 dn->node_changed = true;
1169         return 0;
1170 }
1171
1172 /* Should keep dn->ofs_in_node unchanged */
1173 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1174 {
1175         unsigned int ofs_in_node = dn->ofs_in_node;
1176         int ret;
1177
1178         ret = f2fs_reserve_new_blocks(dn, 1);
1179         dn->ofs_in_node = ofs_in_node;
1180         return ret;
1181 }
1182
1183 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1184 {
1185         bool need_put = dn->inode_page ? false : true;
1186         int err;
1187
1188         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1189         if (err)
1190                 return err;
1191
1192         if (dn->data_blkaddr == NULL_ADDR)
1193                 err = f2fs_reserve_new_block(dn);
1194         if (err || need_put)
1195                 f2fs_put_dnode(dn);
1196         return err;
1197 }
1198
1199 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1200 {
1201         struct extent_info ei = {0, 0, 0};
1202         struct inode *inode = dn->inode;
1203
1204         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1205                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1206                 return 0;
1207         }
1208
1209         return f2fs_reserve_block(dn, index);
1210 }
1211
1212 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1213                                                 int op_flags, bool for_write)
1214 {
1215         struct address_space *mapping = inode->i_mapping;
1216         struct dnode_of_data dn;
1217         struct page *page;
1218         struct extent_info ei = {0,0,0};
1219         int err;
1220
1221         page = f2fs_grab_cache_page(mapping, index, for_write);
1222         if (!page)
1223                 return ERR_PTR(-ENOMEM);
1224
1225         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1226                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1227                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1228                                                 DATA_GENERIC_ENHANCE_READ)) {
1229                         err = -EFSCORRUPTED;
1230                         goto put_err;
1231                 }
1232                 goto got_it;
1233         }
1234
1235         set_new_dnode(&dn, inode, NULL, NULL, 0);
1236         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1237         if (err)
1238                 goto put_err;
1239         f2fs_put_dnode(&dn);
1240
1241         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1242                 err = -ENOENT;
1243                 goto put_err;
1244         }
1245         if (dn.data_blkaddr != NEW_ADDR &&
1246                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1247                                                 dn.data_blkaddr,
1248                                                 DATA_GENERIC_ENHANCE)) {
1249                 err = -EFSCORRUPTED;
1250                 goto put_err;
1251         }
1252 got_it:
1253         if (PageUptodate(page)) {
1254                 unlock_page(page);
1255                 return page;
1256         }
1257
1258         /*
1259          * A new dentry page is allocated but not able to be written, since its
1260          * new inode page couldn't be allocated due to -ENOSPC.
1261          * In such the case, its blkaddr can be remained as NEW_ADDR.
1262          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1263          * f2fs_init_inode_metadata.
1264          */
1265         if (dn.data_blkaddr == NEW_ADDR) {
1266                 zero_user_segment(page, 0, PAGE_SIZE);
1267                 if (!PageUptodate(page))
1268                         SetPageUptodate(page);
1269                 unlock_page(page);
1270                 return page;
1271         }
1272
1273         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1274                                                 op_flags, for_write);
1275         if (err)
1276                 goto put_err;
1277         return page;
1278
1279 put_err:
1280         f2fs_put_page(page, 1);
1281         return ERR_PTR(err);
1282 }
1283
1284 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1285 {
1286         struct address_space *mapping = inode->i_mapping;
1287         struct page *page;
1288
1289         page = find_get_page(mapping, index);
1290         if (page && PageUptodate(page))
1291                 return page;
1292         f2fs_put_page(page, 0);
1293
1294         page = f2fs_get_read_data_page(inode, index, 0, false);
1295         if (IS_ERR(page))
1296                 return page;
1297
1298         if (PageUptodate(page))
1299                 return page;
1300
1301         wait_on_page_locked(page);
1302         if (unlikely(!PageUptodate(page))) {
1303                 f2fs_put_page(page, 0);
1304                 return ERR_PTR(-EIO);
1305         }
1306         return page;
1307 }
1308
1309 /*
1310  * If it tries to access a hole, return an error.
1311  * Because, the callers, functions in dir.c and GC, should be able to know
1312  * whether this page exists or not.
1313  */
1314 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1315                                                         bool for_write)
1316 {
1317         struct address_space *mapping = inode->i_mapping;
1318         struct page *page;
1319 repeat:
1320         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1321         if (IS_ERR(page))
1322                 return page;
1323
1324         /* wait for read completion */
1325         lock_page(page);
1326         if (unlikely(page->mapping != mapping)) {
1327                 f2fs_put_page(page, 1);
1328                 goto repeat;
1329         }
1330         if (unlikely(!PageUptodate(page))) {
1331                 f2fs_put_page(page, 1);
1332                 return ERR_PTR(-EIO);
1333         }
1334         return page;
1335 }
1336
1337 /*
1338  * Caller ensures that this data page is never allocated.
1339  * A new zero-filled data page is allocated in the page cache.
1340  *
1341  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1342  * f2fs_unlock_op().
1343  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1344  * ipage should be released by this function.
1345  */
1346 struct page *f2fs_get_new_data_page(struct inode *inode,
1347                 struct page *ipage, pgoff_t index, bool new_i_size)
1348 {
1349         struct address_space *mapping = inode->i_mapping;
1350         struct page *page;
1351         struct dnode_of_data dn;
1352         int err;
1353
1354         page = f2fs_grab_cache_page(mapping, index, true);
1355         if (!page) {
1356                 /*
1357                  * before exiting, we should make sure ipage will be released
1358                  * if any error occur.
1359                  */
1360                 f2fs_put_page(ipage, 1);
1361                 return ERR_PTR(-ENOMEM);
1362         }
1363
1364         set_new_dnode(&dn, inode, ipage, NULL, 0);
1365         err = f2fs_reserve_block(&dn, index);
1366         if (err) {
1367                 f2fs_put_page(page, 1);
1368                 return ERR_PTR(err);
1369         }
1370         if (!ipage)
1371                 f2fs_put_dnode(&dn);
1372
1373         if (PageUptodate(page))
1374                 goto got_it;
1375
1376         if (dn.data_blkaddr == NEW_ADDR) {
1377                 zero_user_segment(page, 0, PAGE_SIZE);
1378                 if (!PageUptodate(page))
1379                         SetPageUptodate(page);
1380         } else {
1381                 f2fs_put_page(page, 1);
1382
1383                 /* if ipage exists, blkaddr should be NEW_ADDR */
1384                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1385                 page = f2fs_get_lock_data_page(inode, index, true);
1386                 if (IS_ERR(page))
1387                         return page;
1388         }
1389 got_it:
1390         if (new_i_size && i_size_read(inode) <
1391                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1392                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1393         return page;
1394 }
1395
1396 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1397 {
1398         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1399         struct f2fs_summary sum;
1400         struct node_info ni;
1401         block_t old_blkaddr;
1402         blkcnt_t count = 1;
1403         int err;
1404
1405         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1406                 return -EPERM;
1407
1408         err = f2fs_get_node_info(sbi, dn->nid, &ni);
1409         if (err)
1410                 return err;
1411
1412         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1413         if (dn->data_blkaddr != NULL_ADDR)
1414                 goto alloc;
1415
1416         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1417                 return err;
1418
1419 alloc:
1420         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1421         old_blkaddr = dn->data_blkaddr;
1422         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1423                                 &sum, seg_type, NULL);
1424         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1425                 invalidate_mapping_pages(META_MAPPING(sbi),
1426                                         old_blkaddr, old_blkaddr);
1427         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1428
1429         /*
1430          * i_size will be updated by direct_IO. Otherwise, we'll get stale
1431          * data from unwritten block via dio_read.
1432          */
1433         return 0;
1434 }
1435
1436 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1437 {
1438         struct inode *inode = file_inode(iocb->ki_filp);
1439         struct f2fs_map_blocks map;
1440         int flag;
1441         int err = 0;
1442         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1443
1444         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1445         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1446         if (map.m_len > map.m_lblk)
1447                 map.m_len -= map.m_lblk;
1448         else
1449                 map.m_len = 0;
1450
1451         map.m_next_pgofs = NULL;
1452         map.m_next_extent = NULL;
1453         map.m_seg_type = NO_CHECK_TYPE;
1454         map.m_may_create = true;
1455
1456         if (direct_io) {
1457                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1458                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1459                                         F2FS_GET_BLOCK_PRE_AIO :
1460                                         F2FS_GET_BLOCK_PRE_DIO;
1461                 goto map_blocks;
1462         }
1463         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1464                 err = f2fs_convert_inline_inode(inode);
1465                 if (err)
1466                         return err;
1467         }
1468         if (f2fs_has_inline_data(inode))
1469                 return err;
1470
1471         flag = F2FS_GET_BLOCK_PRE_AIO;
1472
1473 map_blocks:
1474         err = f2fs_map_blocks(inode, &map, 1, flag);
1475         if (map.m_len > 0 && err == -ENOSPC) {
1476                 if (!direct_io)
1477                         set_inode_flag(inode, FI_NO_PREALLOC);
1478                 err = 0;
1479         }
1480         return err;
1481 }
1482
1483 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1484 {
1485         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1486                 if (lock)
1487                         down_read(&sbi->node_change);
1488                 else
1489                         up_read(&sbi->node_change);
1490         } else {
1491                 if (lock)
1492                         f2fs_lock_op(sbi);
1493                 else
1494                         f2fs_unlock_op(sbi);
1495         }
1496 }
1497
1498 /*
1499  * f2fs_map_blocks() tries to find or build mapping relationship which
1500  * maps continuous logical blocks to physical blocks, and return such
1501  * info via f2fs_map_blocks structure.
1502  */
1503 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1504                                                 int create, int flag)
1505 {
1506         unsigned int maxblocks = map->m_len;
1507         struct dnode_of_data dn;
1508         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1509         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1510         pgoff_t pgofs, end_offset, end;
1511         int err = 0, ofs = 1;
1512         unsigned int ofs_in_node, last_ofs_in_node;
1513         blkcnt_t prealloc;
1514         struct extent_info ei = {0,0,0};
1515         block_t blkaddr;
1516         unsigned int start_pgofs;
1517
1518         if (!maxblocks)
1519                 return 0;
1520
1521         map->m_len = 0;
1522         map->m_flags = 0;
1523
1524         /* it only supports block size == page size */
1525         pgofs = (pgoff_t)map->m_lblk;
1526         end = pgofs + maxblocks;
1527
1528         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1529                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1530                                                         map->m_may_create)
1531                         goto next_dnode;
1532
1533                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1534                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1535                 map->m_flags = F2FS_MAP_MAPPED;
1536                 if (map->m_next_extent)
1537                         *map->m_next_extent = pgofs + map->m_len;
1538
1539                 /* for hardware encryption, but to avoid potential issue in future */
1540                 if (flag == F2FS_GET_BLOCK_DIO)
1541                         f2fs_wait_on_block_writeback_range(inode,
1542                                                 map->m_pblk, map->m_len);
1543                 goto out;
1544         }
1545
1546 next_dnode:
1547         if (map->m_may_create)
1548                 f2fs_do_map_lock(sbi, flag, true);
1549
1550         /* When reading holes, we need its node page */
1551         set_new_dnode(&dn, inode, NULL, NULL, 0);
1552         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1553         if (err) {
1554                 if (flag == F2FS_GET_BLOCK_BMAP)
1555                         map->m_pblk = 0;
1556                 if (err == -ENOENT) {
1557                         err = 0;
1558                         if (map->m_next_pgofs)
1559                                 *map->m_next_pgofs =
1560                                         f2fs_get_next_page_offset(&dn, pgofs);
1561                         if (map->m_next_extent)
1562                                 *map->m_next_extent =
1563                                         f2fs_get_next_page_offset(&dn, pgofs);
1564                 }
1565                 goto unlock_out;
1566         }
1567
1568         start_pgofs = pgofs;
1569         prealloc = 0;
1570         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1571         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1572
1573 next_block:
1574         blkaddr = f2fs_data_blkaddr(&dn);
1575
1576         if (__is_valid_data_blkaddr(blkaddr) &&
1577                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1578                 err = -EFSCORRUPTED;
1579                 goto sync_out;
1580         }
1581
1582         if (__is_valid_data_blkaddr(blkaddr)) {
1583                 /* use out-place-update for driect IO under LFS mode */
1584                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1585                                                         map->m_may_create) {
1586                         err = __allocate_data_block(&dn, map->m_seg_type);
1587                         if (err)
1588                                 goto sync_out;
1589                         blkaddr = dn.data_blkaddr;
1590                         set_inode_flag(inode, FI_APPEND_WRITE);
1591                 }
1592         } else {
1593                 if (create) {
1594                         if (unlikely(f2fs_cp_error(sbi))) {
1595                                 err = -EIO;
1596                                 goto sync_out;
1597                         }
1598                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1599                                 if (blkaddr == NULL_ADDR) {
1600                                         prealloc++;
1601                                         last_ofs_in_node = dn.ofs_in_node;
1602                                 }
1603                         } else {
1604                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1605                                         flag != F2FS_GET_BLOCK_DIO);
1606                                 err = __allocate_data_block(&dn,
1607                                                         map->m_seg_type);
1608                                 if (!err)
1609                                         set_inode_flag(inode, FI_APPEND_WRITE);
1610                         }
1611                         if (err)
1612                                 goto sync_out;
1613                         map->m_flags |= F2FS_MAP_NEW;
1614                         blkaddr = dn.data_blkaddr;
1615                 } else {
1616                         if (flag == F2FS_GET_BLOCK_BMAP) {
1617                                 map->m_pblk = 0;
1618                                 goto sync_out;
1619                         }
1620                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1621                                 goto sync_out;
1622                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1623                                                 blkaddr == NULL_ADDR) {
1624                                 if (map->m_next_pgofs)
1625                                         *map->m_next_pgofs = pgofs + 1;
1626                                 goto sync_out;
1627                         }
1628                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1629                                 /* for defragment case */
1630                                 if (map->m_next_pgofs)
1631                                         *map->m_next_pgofs = pgofs + 1;
1632                                 goto sync_out;
1633                         }
1634                 }
1635         }
1636
1637         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1638                 goto skip;
1639
1640         if (map->m_len == 0) {
1641                 /* preallocated unwritten block should be mapped for fiemap. */
1642                 if (blkaddr == NEW_ADDR)
1643                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1644                 map->m_flags |= F2FS_MAP_MAPPED;
1645
1646                 map->m_pblk = blkaddr;
1647                 map->m_len = 1;
1648         } else if ((map->m_pblk != NEW_ADDR &&
1649                         blkaddr == (map->m_pblk + ofs)) ||
1650                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1651                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1652                 ofs++;
1653                 map->m_len++;
1654         } else {
1655                 goto sync_out;
1656         }
1657
1658 skip:
1659         dn.ofs_in_node++;
1660         pgofs++;
1661
1662         /* preallocate blocks in batch for one dnode page */
1663         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1664                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1665
1666                 dn.ofs_in_node = ofs_in_node;
1667                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1668                 if (err)
1669                         goto sync_out;
1670
1671                 map->m_len += dn.ofs_in_node - ofs_in_node;
1672                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1673                         err = -ENOSPC;
1674                         goto sync_out;
1675                 }
1676                 dn.ofs_in_node = end_offset;
1677         }
1678
1679         if (pgofs >= end)
1680                 goto sync_out;
1681         else if (dn.ofs_in_node < end_offset)
1682                 goto next_block;
1683
1684         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1685                 if (map->m_flags & F2FS_MAP_MAPPED) {
1686                         unsigned int ofs = start_pgofs - map->m_lblk;
1687
1688                         f2fs_update_extent_cache_range(&dn,
1689                                 start_pgofs, map->m_pblk + ofs,
1690                                 map->m_len - ofs);
1691                 }
1692         }
1693
1694         f2fs_put_dnode(&dn);
1695
1696         if (map->m_may_create) {
1697                 f2fs_do_map_lock(sbi, flag, false);
1698                 f2fs_balance_fs(sbi, dn.node_changed);
1699         }
1700         goto next_dnode;
1701
1702 sync_out:
1703
1704         /* for hardware encryption, but to avoid potential issue in future */
1705         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1706                 f2fs_wait_on_block_writeback_range(inode,
1707                                                 map->m_pblk, map->m_len);
1708
1709         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1710                 if (map->m_flags & F2FS_MAP_MAPPED) {
1711                         unsigned int ofs = start_pgofs - map->m_lblk;
1712
1713                         f2fs_update_extent_cache_range(&dn,
1714                                 start_pgofs, map->m_pblk + ofs,
1715                                 map->m_len - ofs);
1716                 }
1717                 if (map->m_next_extent)
1718                         *map->m_next_extent = pgofs + 1;
1719         }
1720         f2fs_put_dnode(&dn);
1721 unlock_out:
1722         if (map->m_may_create) {
1723                 f2fs_do_map_lock(sbi, flag, false);
1724                 f2fs_balance_fs(sbi, dn.node_changed);
1725         }
1726 out:
1727         trace_f2fs_map_blocks(inode, map, err);
1728         return err;
1729 }
1730
1731 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1732 {
1733         struct f2fs_map_blocks map;
1734         block_t last_lblk;
1735         int err;
1736
1737         if (pos + len > i_size_read(inode))
1738                 return false;
1739
1740         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1741         map.m_next_pgofs = NULL;
1742         map.m_next_extent = NULL;
1743         map.m_seg_type = NO_CHECK_TYPE;
1744         map.m_may_create = false;
1745         last_lblk = F2FS_BLK_ALIGN(pos + len);
1746
1747         while (map.m_lblk < last_lblk) {
1748                 map.m_len = last_lblk - map.m_lblk;
1749                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1750                 if (err || map.m_len == 0)
1751                         return false;
1752                 map.m_lblk += map.m_len;
1753         }
1754         return true;
1755 }
1756
1757 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1758 {
1759         return (bytes >> inode->i_blkbits);
1760 }
1761
1762 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1763 {
1764         return (blks << inode->i_blkbits);
1765 }
1766
1767 static int __get_data_block(struct inode *inode, sector_t iblock,
1768                         struct buffer_head *bh, int create, int flag,
1769                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1770 {
1771         struct f2fs_map_blocks map;
1772         int err;
1773
1774         map.m_lblk = iblock;
1775         map.m_len = bytes_to_blks(inode, bh->b_size);
1776         map.m_next_pgofs = next_pgofs;
1777         map.m_next_extent = NULL;
1778         map.m_seg_type = seg_type;
1779         map.m_may_create = may_write;
1780
1781         err = f2fs_map_blocks(inode, &map, create, flag);
1782         if (!err) {
1783                 map_bh(bh, inode->i_sb, map.m_pblk);
1784                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1785                 bh->b_size = blks_to_bytes(inode, map.m_len);
1786         }
1787         return err;
1788 }
1789
1790 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1791                         struct buffer_head *bh_result, int create)
1792 {
1793         return __get_data_block(inode, iblock, bh_result, create,
1794                                 F2FS_GET_BLOCK_DIO, NULL,
1795                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1796                                 IS_SWAPFILE(inode) ? false : true);
1797 }
1798
1799 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1800                         struct buffer_head *bh_result, int create)
1801 {
1802         return __get_data_block(inode, iblock, bh_result, create,
1803                                 F2FS_GET_BLOCK_DIO, NULL,
1804                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1805                                 false);
1806 }
1807
1808 static int f2fs_xattr_fiemap(struct inode *inode,
1809                                 struct fiemap_extent_info *fieinfo)
1810 {
1811         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1812         struct page *page;
1813         struct node_info ni;
1814         __u64 phys = 0, len;
1815         __u32 flags;
1816         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1817         int err = 0;
1818
1819         if (f2fs_has_inline_xattr(inode)) {
1820                 int offset;
1821
1822                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1823                                                 inode->i_ino, false);
1824                 if (!page)
1825                         return -ENOMEM;
1826
1827                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1828                 if (err) {
1829                         f2fs_put_page(page, 1);
1830                         return err;
1831                 }
1832
1833                 phys = blks_to_bytes(inode, ni.blk_addr);
1834                 offset = offsetof(struct f2fs_inode, i_addr) +
1835                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1836                                         get_inline_xattr_addrs(inode));
1837
1838                 phys += offset;
1839                 len = inline_xattr_size(inode);
1840
1841                 f2fs_put_page(page, 1);
1842
1843                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1844
1845                 if (!xnid)
1846                         flags |= FIEMAP_EXTENT_LAST;
1847
1848                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1849                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1850                 if (err || err == 1)
1851                         return err;
1852         }
1853
1854         if (xnid) {
1855                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1856                 if (!page)
1857                         return -ENOMEM;
1858
1859                 err = f2fs_get_node_info(sbi, xnid, &ni);
1860                 if (err) {
1861                         f2fs_put_page(page, 1);
1862                         return err;
1863                 }
1864
1865                 phys = blks_to_bytes(inode, ni.blk_addr);
1866                 len = inode->i_sb->s_blocksize;
1867
1868                 f2fs_put_page(page, 1);
1869
1870                 flags = FIEMAP_EXTENT_LAST;
1871         }
1872
1873         if (phys) {
1874                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1875                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1876         }
1877
1878         return (err < 0 ? err : 0);
1879 }
1880
1881 static loff_t max_inode_blocks(struct inode *inode)
1882 {
1883         loff_t result = ADDRS_PER_INODE(inode);
1884         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1885
1886         /* two direct node blocks */
1887         result += (leaf_count * 2);
1888
1889         /* two indirect node blocks */
1890         leaf_count *= NIDS_PER_BLOCK;
1891         result += (leaf_count * 2);
1892
1893         /* one double indirect node block */
1894         leaf_count *= NIDS_PER_BLOCK;
1895         result += leaf_count;
1896
1897         return result;
1898 }
1899
1900 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1901                 u64 start, u64 len)
1902 {
1903         struct f2fs_map_blocks map;
1904         sector_t start_blk, last_blk;
1905         pgoff_t next_pgofs;
1906         u64 logical = 0, phys = 0, size = 0;
1907         u32 flags = 0;
1908         int ret = 0;
1909         bool compr_cluster = false;
1910         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1911
1912         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1913                 ret = f2fs_precache_extents(inode);
1914                 if (ret)
1915                         return ret;
1916         }
1917
1918         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1919         if (ret)
1920                 return ret;
1921
1922         inode_lock(inode);
1923
1924         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1925                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1926                 goto out;
1927         }
1928
1929         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1930                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1931                 if (ret != -EAGAIN)
1932                         goto out;
1933         }
1934
1935         if (bytes_to_blks(inode, len) == 0)
1936                 len = blks_to_bytes(inode, 1);
1937
1938         start_blk = bytes_to_blks(inode, start);
1939         last_blk = bytes_to_blks(inode, start + len - 1);
1940
1941 next:
1942         memset(&map, 0, sizeof(map));
1943         map.m_lblk = start_blk;
1944         map.m_len = bytes_to_blks(inode, len);
1945         map.m_next_pgofs = &next_pgofs;
1946         map.m_seg_type = NO_CHECK_TYPE;
1947
1948         if (compr_cluster)
1949                 map.m_len = cluster_size - 1;
1950
1951         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1952         if (ret)
1953                 goto out;
1954
1955         /* HOLE */
1956         if (!(map.m_flags & F2FS_MAP_FLAGS)) {
1957                 start_blk = next_pgofs;
1958
1959                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1960                                                 max_inode_blocks(inode)))
1961                         goto prep_next;
1962
1963                 flags |= FIEMAP_EXTENT_LAST;
1964         }
1965
1966         if (size) {
1967                 if (IS_ENCRYPTED(inode))
1968                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1969
1970                 ret = fiemap_fill_next_extent(fieinfo, logical,
1971                                 phys, size, flags);
1972                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1973                 if (ret)
1974                         goto out;
1975                 size = 0;
1976         }
1977
1978         if (start_blk > last_blk)
1979                 goto out;
1980
1981         if (compr_cluster) {
1982                 compr_cluster = false;
1983
1984
1985                 logical = blks_to_bytes(inode, start_blk - 1);
1986                 phys = blks_to_bytes(inode, map.m_pblk);
1987                 size = blks_to_bytes(inode, cluster_size);
1988
1989                 flags |= FIEMAP_EXTENT_ENCODED;
1990
1991                 start_blk += cluster_size - 1;
1992
1993                 if (start_blk > last_blk)
1994                         goto out;
1995
1996                 goto prep_next;
1997         }
1998
1999         if (map.m_pblk == COMPRESS_ADDR) {
2000                 compr_cluster = true;
2001                 start_blk++;
2002                 goto prep_next;
2003         }
2004
2005         logical = blks_to_bytes(inode, start_blk);
2006         phys = blks_to_bytes(inode, map.m_pblk);
2007         size = blks_to_bytes(inode, map.m_len);
2008         flags = 0;
2009         if (map.m_flags & F2FS_MAP_UNWRITTEN)
2010                 flags = FIEMAP_EXTENT_UNWRITTEN;
2011
2012         start_blk += bytes_to_blks(inode, size);
2013
2014 prep_next:
2015         cond_resched();
2016         if (fatal_signal_pending(current))
2017                 ret = -EINTR;
2018         else
2019                 goto next;
2020 out:
2021         if (ret == 1)
2022                 ret = 0;
2023
2024         inode_unlock(inode);
2025         return ret;
2026 }
2027
2028 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2029 {
2030         if (IS_ENABLED(CONFIG_FS_VERITY) &&
2031             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2032                 return inode->i_sb->s_maxbytes;
2033
2034         return i_size_read(inode);
2035 }
2036
2037 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2038                                         unsigned nr_pages,
2039                                         struct f2fs_map_blocks *map,
2040                                         struct bio **bio_ret,
2041                                         sector_t *last_block_in_bio,
2042                                         bool is_readahead)
2043 {
2044         struct bio *bio = *bio_ret;
2045         const unsigned blocksize = blks_to_bytes(inode, 1);
2046         sector_t block_in_file;
2047         sector_t last_block;
2048         sector_t last_block_in_file;
2049         sector_t block_nr;
2050         int ret = 0;
2051
2052         block_in_file = (sector_t)page_index(page);
2053         last_block = block_in_file + nr_pages;
2054         last_block_in_file = bytes_to_blks(inode,
2055                         f2fs_readpage_limit(inode) + blocksize - 1);
2056         if (last_block > last_block_in_file)
2057                 last_block = last_block_in_file;
2058
2059         /* just zeroing out page which is beyond EOF */
2060         if (block_in_file >= last_block)
2061                 goto zero_out;
2062         /*
2063          * Map blocks using the previous result first.
2064          */
2065         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2066                         block_in_file > map->m_lblk &&
2067                         block_in_file < (map->m_lblk + map->m_len))
2068                 goto got_it;
2069
2070         /*
2071          * Then do more f2fs_map_blocks() calls until we are
2072          * done with this page.
2073          */
2074         map->m_lblk = block_in_file;
2075         map->m_len = last_block - block_in_file;
2076
2077         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2078         if (ret)
2079                 goto out;
2080 got_it:
2081         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2082                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2083                 SetPageMappedToDisk(page);
2084
2085                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
2086                                         !cleancache_get_page(page))) {
2087                         SetPageUptodate(page);
2088                         goto confused;
2089                 }
2090
2091                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2092                                                 DATA_GENERIC_ENHANCE_READ)) {
2093                         ret = -EFSCORRUPTED;
2094                         goto out;
2095                 }
2096         } else {
2097 zero_out:
2098                 zero_user_segment(page, 0, PAGE_SIZE);
2099                 if (f2fs_need_verity(inode, page->index) &&
2100                     !fsverity_verify_page(page)) {
2101                         ret = -EIO;
2102                         goto out;
2103                 }
2104                 if (!PageUptodate(page))
2105                         SetPageUptodate(page);
2106                 unlock_page(page);
2107                 goto out;
2108         }
2109
2110         /*
2111          * This page will go to BIO.  Do we need to send this
2112          * BIO off first?
2113          */
2114         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2115                                        *last_block_in_bio, block_nr) ||
2116                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2117 submit_and_realloc:
2118                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2119                 bio = NULL;
2120         }
2121         if (bio == NULL) {
2122                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2123                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2124                                 false, true);
2125                 if (IS_ERR(bio)) {
2126                         ret = PTR_ERR(bio);
2127                         bio = NULL;
2128                         goto out;
2129                 }
2130         }
2131
2132         /*
2133          * If the page is under writeback, we need to wait for
2134          * its completion to see the correct decrypted data.
2135          */
2136         f2fs_wait_on_block_writeback(inode, block_nr);
2137
2138         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2139                 goto submit_and_realloc;
2140
2141         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2142         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2143         ClearPageError(page);
2144         *last_block_in_bio = block_nr;
2145         goto out;
2146 confused:
2147         if (bio) {
2148                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2149                 bio = NULL;
2150         }
2151         unlock_page(page);
2152 out:
2153         *bio_ret = bio;
2154         return ret;
2155 }
2156
2157 #ifdef CONFIG_F2FS_FS_COMPRESSION
2158 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2159                                 unsigned nr_pages, sector_t *last_block_in_bio,
2160                                 bool is_readahead, bool for_write)
2161 {
2162         struct dnode_of_data dn;
2163         struct inode *inode = cc->inode;
2164         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2165         struct bio *bio = *bio_ret;
2166         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2167         sector_t last_block_in_file;
2168         const unsigned blocksize = blks_to_bytes(inode, 1);
2169         struct decompress_io_ctx *dic = NULL;
2170         struct bio_post_read_ctx *ctx;
2171         bool for_verity = false;
2172         int i;
2173         int ret = 0;
2174
2175         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2176
2177         last_block_in_file = bytes_to_blks(inode,
2178                         f2fs_readpage_limit(inode) + blocksize - 1);
2179
2180         /* get rid of pages beyond EOF */
2181         for (i = 0; i < cc->cluster_size; i++) {
2182                 struct page *page = cc->rpages[i];
2183
2184                 if (!page)
2185                         continue;
2186                 if ((sector_t)page->index >= last_block_in_file) {
2187                         zero_user_segment(page, 0, PAGE_SIZE);
2188                         if (!PageUptodate(page))
2189                                 SetPageUptodate(page);
2190                 } else if (!PageUptodate(page)) {
2191                         continue;
2192                 }
2193                 unlock_page(page);
2194                 cc->rpages[i] = NULL;
2195                 cc->nr_rpages--;
2196         }
2197
2198         /* we are done since all pages are beyond EOF */
2199         if (f2fs_cluster_is_empty(cc))
2200                 goto out;
2201
2202         set_new_dnode(&dn, inode, NULL, NULL, 0);
2203         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2204         if (ret)
2205                 goto out;
2206
2207         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2208
2209         for (i = 1; i < cc->cluster_size; i++) {
2210                 block_t blkaddr;
2211
2212                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2213                                                 dn.ofs_in_node + i);
2214
2215                 if (!__is_valid_data_blkaddr(blkaddr))
2216                         break;
2217
2218                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2219                         ret = -EFAULT;
2220                         goto out_put_dnode;
2221                 }
2222                 cc->nr_cpages++;
2223         }
2224
2225         /* nothing to decompress */
2226         if (cc->nr_cpages == 0) {
2227                 ret = 0;
2228                 goto out_put_dnode;
2229         }
2230
2231         dic = f2fs_alloc_dic(cc);
2232         if (IS_ERR(dic)) {
2233                 ret = PTR_ERR(dic);
2234                 goto out_put_dnode;
2235         }
2236
2237         /*
2238          * It's possible to enable fsverity on the fly when handling a cluster,
2239          * which requires complicated error handling. Instead of adding more
2240          * complexity, let's give a rule where end_io post-processes fsverity
2241          * per cluster. In order to do that, we need to submit bio, if previous
2242          * bio sets a different post-process policy.
2243          */
2244         if (fsverity_active(cc->inode)) {
2245                 atomic_set(&dic->verity_pages, cc->nr_cpages);
2246                 for_verity = true;
2247
2248                 if (bio) {
2249                         ctx = bio->bi_private;
2250                         if (!(ctx->enabled_steps & (1 << STEP_VERITY))) {
2251                                 __submit_bio(sbi, bio, DATA);
2252                                 bio = NULL;
2253                         }
2254                 }
2255         }
2256
2257         for (i = 0; i < dic->nr_cpages; i++) {
2258                 struct page *page = dic->cpages[i];
2259                 block_t blkaddr;
2260
2261                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2262                                                 dn.ofs_in_node + i + 1);
2263
2264                 if (bio && (!page_is_mergeable(sbi, bio,
2265                                         *last_block_in_bio, blkaddr) ||
2266                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2267 submit_and_realloc:
2268                         __submit_bio(sbi, bio, DATA);
2269                         bio = NULL;
2270                 }
2271
2272                 if (!bio) {
2273                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2274                                         is_readahead ? REQ_RAHEAD : 0,
2275                                         page->index, for_write, for_verity);
2276                         if (IS_ERR(bio)) {
2277                                 unsigned int remained = dic->nr_cpages - i;
2278                                 bool release = false;
2279
2280                                 ret = PTR_ERR(bio);
2281                                 dic->failed = true;
2282
2283                                 if (for_verity) {
2284                                         if (!atomic_sub_return(remained,
2285                                                 &dic->verity_pages))
2286                                                 release = true;
2287                                 } else {
2288                                         if (!atomic_sub_return(remained,
2289                                                 &dic->pending_pages))
2290                                                 release = true;
2291                                 }
2292
2293                                 if (release) {
2294                                         f2fs_decompress_end_io(dic->rpages,
2295                                                 cc->cluster_size, true,
2296                                                 false);
2297                                         f2fs_free_dic(dic);
2298                                 }
2299
2300                                 f2fs_put_dnode(&dn);
2301                                 *bio_ret = NULL;
2302                                 return ret;
2303                         }
2304                 }
2305
2306                 f2fs_wait_on_block_writeback(inode, blkaddr);
2307
2308                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2309                         goto submit_and_realloc;
2310
2311                 /* tag STEP_DECOMPRESS to handle IO in wq */
2312                 ctx = bio->bi_private;
2313                 if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS)))
2314                         ctx->enabled_steps |= 1 << STEP_DECOMPRESS;
2315
2316                 inc_page_count(sbi, F2FS_RD_DATA);
2317                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2318                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2319                 ClearPageError(page);
2320                 *last_block_in_bio = blkaddr;
2321         }
2322
2323         f2fs_put_dnode(&dn);
2324
2325         *bio_ret = bio;
2326         return 0;
2327
2328 out_put_dnode:
2329         f2fs_put_dnode(&dn);
2330 out:
2331         f2fs_decompress_end_io(cc->rpages, cc->cluster_size, true, false);
2332         *bio_ret = bio;
2333         return ret;
2334 }
2335 #endif
2336
2337 /*
2338  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2339  * Major change was from block_size == page_size in f2fs by default.
2340  *
2341  * Note that the aops->readpages() function is ONLY used for read-ahead. If
2342  * this function ever deviates from doing just read-ahead, it should either
2343  * use ->readpage() or do the necessary surgery to decouple ->readpages()
2344  * from read-ahead.
2345  */
2346 static int f2fs_mpage_readpages(struct inode *inode,
2347                 struct readahead_control *rac, struct page *page)
2348 {
2349         struct bio *bio = NULL;
2350         sector_t last_block_in_bio = 0;
2351         struct f2fs_map_blocks map;
2352 #ifdef CONFIG_F2FS_FS_COMPRESSION
2353         struct compress_ctx cc = {
2354                 .inode = inode,
2355                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2356                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2357                 .cluster_idx = NULL_CLUSTER,
2358                 .rpages = NULL,
2359                 .cpages = NULL,
2360                 .nr_rpages = 0,
2361                 .nr_cpages = 0,
2362         };
2363 #endif
2364         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2365         unsigned max_nr_pages = nr_pages;
2366         int ret = 0;
2367         bool drop_ra = false;
2368
2369         map.m_pblk = 0;
2370         map.m_lblk = 0;
2371         map.m_len = 0;
2372         map.m_flags = 0;
2373         map.m_next_pgofs = NULL;
2374         map.m_next_extent = NULL;
2375         map.m_seg_type = NO_CHECK_TYPE;
2376         map.m_may_create = false;
2377
2378         /*
2379          * Two readahead threads for same address range can cause race condition
2380          * which fragments sequential read IOs. So let's avoid each other.
2381          */
2382         if (rac && readahead_count(rac)) {
2383                 if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
2384                         drop_ra = true;
2385                 else
2386                         WRITE_ONCE(F2FS_I(inode)->ra_offset,
2387                                                 readahead_index(rac));
2388         }
2389
2390         for (; nr_pages; nr_pages--) {
2391                 if (rac) {
2392                         page = readahead_page(rac);
2393                         prefetchw(&page->flags);
2394                         if (drop_ra) {
2395                                 f2fs_put_page(page, 1);
2396                                 continue;
2397                         }
2398                 }
2399
2400 #ifdef CONFIG_F2FS_FS_COMPRESSION
2401                 if (f2fs_compressed_file(inode)) {
2402                         /* there are remained comressed pages, submit them */
2403                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2404                                 ret = f2fs_read_multi_pages(&cc, &bio,
2405                                                         max_nr_pages,
2406                                                         &last_block_in_bio,
2407                                                         rac != NULL, false);
2408                                 f2fs_destroy_compress_ctx(&cc);
2409                                 if (ret)
2410                                         goto set_error_page;
2411                         }
2412                         ret = f2fs_is_compressed_cluster(inode, page->index);
2413                         if (ret < 0)
2414                                 goto set_error_page;
2415                         else if (!ret)
2416                                 goto read_single_page;
2417
2418                         ret = f2fs_init_compress_ctx(&cc);
2419                         if (ret)
2420                                 goto set_error_page;
2421
2422                         f2fs_compress_ctx_add_page(&cc, page);
2423
2424                         goto next_page;
2425                 }
2426 read_single_page:
2427 #endif
2428
2429                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2430                                         &bio, &last_block_in_bio, rac);
2431                 if (ret) {
2432 #ifdef CONFIG_F2FS_FS_COMPRESSION
2433 set_error_page:
2434 #endif
2435                         SetPageError(page);
2436                         zero_user_segment(page, 0, PAGE_SIZE);
2437                         unlock_page(page);
2438                 }
2439 #ifdef CONFIG_F2FS_FS_COMPRESSION
2440 next_page:
2441 #endif
2442                 if (rac)
2443                         put_page(page);
2444
2445 #ifdef CONFIG_F2FS_FS_COMPRESSION
2446                 if (f2fs_compressed_file(inode)) {
2447                         /* last page */
2448                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2449                                 ret = f2fs_read_multi_pages(&cc, &bio,
2450                                                         max_nr_pages,
2451                                                         &last_block_in_bio,
2452                                                         rac != NULL, false);
2453                                 f2fs_destroy_compress_ctx(&cc);
2454                         }
2455                 }
2456 #endif
2457         }
2458         if (bio)
2459                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2460
2461         if (rac && readahead_count(rac) && !drop_ra)
2462                 WRITE_ONCE(F2FS_I(inode)->ra_offset, -1);
2463         return ret;
2464 }
2465
2466 static int f2fs_read_data_page(struct file *file, struct page *page)
2467 {
2468         struct inode *inode = page_file_mapping(page)->host;
2469         int ret = -EAGAIN;
2470
2471         trace_f2fs_readpage(page, DATA);
2472
2473         if (!f2fs_is_compress_backend_ready(inode)) {
2474                 unlock_page(page);
2475                 return -EOPNOTSUPP;
2476         }
2477
2478         /* If the file has inline data, try to read it directly */
2479         if (f2fs_has_inline_data(inode))
2480                 ret = f2fs_read_inline_data(inode, page);
2481         if (ret == -EAGAIN)
2482                 ret = f2fs_mpage_readpages(inode, NULL, page);
2483         return ret;
2484 }
2485
2486 static void f2fs_readahead(struct readahead_control *rac)
2487 {
2488         struct inode *inode = rac->mapping->host;
2489
2490         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2491
2492         if (!f2fs_is_compress_backend_ready(inode))
2493                 return;
2494
2495         /* If the file has inline data, skip readpages */
2496         if (f2fs_has_inline_data(inode))
2497                 return;
2498
2499         f2fs_mpage_readpages(inode, rac, NULL);
2500 }
2501
2502 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2503 {
2504         struct inode *inode = fio->page->mapping->host;
2505         struct page *mpage, *page;
2506         gfp_t gfp_flags = GFP_NOFS;
2507
2508         if (!f2fs_encrypted_file(inode))
2509                 return 0;
2510
2511         page = fio->compressed_page ? fio->compressed_page : fio->page;
2512
2513         /* wait for GCed page writeback via META_MAPPING */
2514         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2515
2516         if (fscrypt_inode_uses_inline_crypto(inode))
2517                 return 0;
2518
2519 retry_encrypt:
2520         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2521                                         PAGE_SIZE, 0, gfp_flags);
2522         if (IS_ERR(fio->encrypted_page)) {
2523                 /* flush pending IOs and wait for a while in the ENOMEM case */
2524                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2525                         f2fs_flush_merged_writes(fio->sbi);
2526                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2527                         gfp_flags |= __GFP_NOFAIL;
2528                         goto retry_encrypt;
2529                 }
2530                 return PTR_ERR(fio->encrypted_page);
2531         }
2532
2533         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2534         if (mpage) {
2535                 if (PageUptodate(mpage))
2536                         memcpy(page_address(mpage),
2537                                 page_address(fio->encrypted_page), PAGE_SIZE);
2538                 f2fs_put_page(mpage, 1);
2539         }
2540         return 0;
2541 }
2542
2543 static inline bool check_inplace_update_policy(struct inode *inode,
2544                                 struct f2fs_io_info *fio)
2545 {
2546         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2547         unsigned int policy = SM_I(sbi)->ipu_policy;
2548
2549         if (policy & (0x1 << F2FS_IPU_FORCE))
2550                 return true;
2551         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2552                 return true;
2553         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2554                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2555                 return true;
2556         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2557                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2558                 return true;
2559
2560         /*
2561          * IPU for rewrite async pages
2562          */
2563         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2564                         fio && fio->op == REQ_OP_WRITE &&
2565                         !(fio->op_flags & REQ_SYNC) &&
2566                         !IS_ENCRYPTED(inode))
2567                 return true;
2568
2569         /* this is only set during fdatasync */
2570         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2571                         is_inode_flag_set(inode, FI_NEED_IPU))
2572                 return true;
2573
2574         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2575                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2576                 return true;
2577
2578         return false;
2579 }
2580
2581 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2582 {
2583         if (f2fs_is_pinned_file(inode))
2584                 return true;
2585
2586         /* if this is cold file, we should overwrite to avoid fragmentation */
2587         if (file_is_cold(inode))
2588                 return true;
2589
2590         return check_inplace_update_policy(inode, fio);
2591 }
2592
2593 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2594 {
2595         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2596
2597         if (f2fs_lfs_mode(sbi))
2598                 return true;
2599         if (S_ISDIR(inode->i_mode))
2600                 return true;
2601         if (IS_NOQUOTA(inode))
2602                 return true;
2603         if (f2fs_is_atomic_file(inode))
2604                 return true;
2605         if (fio) {
2606                 if (is_cold_data(fio->page))
2607                         return true;
2608                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
2609                         return true;
2610                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2611                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2612                         return true;
2613         }
2614         return false;
2615 }
2616
2617 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2618 {
2619         struct inode *inode = fio->page->mapping->host;
2620
2621         if (f2fs_should_update_outplace(inode, fio))
2622                 return false;
2623
2624         return f2fs_should_update_inplace(inode, fio);
2625 }
2626
2627 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2628 {
2629         struct page *page = fio->page;
2630         struct inode *inode = page->mapping->host;
2631         struct dnode_of_data dn;
2632         struct extent_info ei = {0,0,0};
2633         struct node_info ni;
2634         bool ipu_force = false;
2635         int err = 0;
2636
2637         set_new_dnode(&dn, inode, NULL, NULL, 0);
2638         if (need_inplace_update(fio) &&
2639                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2640                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2641
2642                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2643                                                 DATA_GENERIC_ENHANCE))
2644                         return -EFSCORRUPTED;
2645
2646                 ipu_force = true;
2647                 fio->need_lock = LOCK_DONE;
2648                 goto got_it;
2649         }
2650
2651         /* Deadlock due to between page->lock and f2fs_lock_op */
2652         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2653                 return -EAGAIN;
2654
2655         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2656         if (err)
2657                 goto out;
2658
2659         fio->old_blkaddr = dn.data_blkaddr;
2660
2661         /* This page is already truncated */
2662         if (fio->old_blkaddr == NULL_ADDR) {
2663                 ClearPageUptodate(page);
2664                 clear_cold_data(page);
2665                 goto out_writepage;
2666         }
2667 got_it:
2668         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2669                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2670                                                 DATA_GENERIC_ENHANCE)) {
2671                 err = -EFSCORRUPTED;
2672                 goto out_writepage;
2673         }
2674         /*
2675          * If current allocation needs SSR,
2676          * it had better in-place writes for updated data.
2677          */
2678         if (ipu_force ||
2679                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2680                                         need_inplace_update(fio))) {
2681                 err = f2fs_encrypt_one_page(fio);
2682                 if (err)
2683                         goto out_writepage;
2684
2685                 set_page_writeback(page);
2686                 ClearPageError(page);
2687                 f2fs_put_dnode(&dn);
2688                 if (fio->need_lock == LOCK_REQ)
2689                         f2fs_unlock_op(fio->sbi);
2690                 err = f2fs_inplace_write_data(fio);
2691                 if (err) {
2692                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2693                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2694                         if (PageWriteback(page))
2695                                 end_page_writeback(page);
2696                 } else {
2697                         set_inode_flag(inode, FI_UPDATE_WRITE);
2698                 }
2699                 trace_f2fs_do_write_data_page(fio->page, IPU);
2700                 return err;
2701         }
2702
2703         if (fio->need_lock == LOCK_RETRY) {
2704                 if (!f2fs_trylock_op(fio->sbi)) {
2705                         err = -EAGAIN;
2706                         goto out_writepage;
2707                 }
2708                 fio->need_lock = LOCK_REQ;
2709         }
2710
2711         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2712         if (err)
2713                 goto out_writepage;
2714
2715         fio->version = ni.version;
2716
2717         err = f2fs_encrypt_one_page(fio);
2718         if (err)
2719                 goto out_writepage;
2720
2721         set_page_writeback(page);
2722         ClearPageError(page);
2723
2724         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2725                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2726
2727         /* LFS mode write path */
2728         f2fs_outplace_write_data(&dn, fio);
2729         trace_f2fs_do_write_data_page(page, OPU);
2730         set_inode_flag(inode, FI_APPEND_WRITE);
2731         if (page->index == 0)
2732                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2733 out_writepage:
2734         f2fs_put_dnode(&dn);
2735 out:
2736         if (fio->need_lock == LOCK_REQ)
2737                 f2fs_unlock_op(fio->sbi);
2738         return err;
2739 }
2740
2741 int f2fs_write_single_data_page(struct page *page, int *submitted,
2742                                 struct bio **bio,
2743                                 sector_t *last_block,
2744                                 struct writeback_control *wbc,
2745                                 enum iostat_type io_type,
2746                                 int compr_blocks)
2747 {
2748         struct inode *inode = page->mapping->host;
2749         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2750         loff_t i_size = i_size_read(inode);
2751         const pgoff_t end_index = ((unsigned long long)i_size)
2752                                                         >> PAGE_SHIFT;
2753         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2754         unsigned offset = 0;
2755         bool need_balance_fs = false;
2756         int err = 0;
2757         struct f2fs_io_info fio = {
2758                 .sbi = sbi,
2759                 .ino = inode->i_ino,
2760                 .type = DATA,
2761                 .op = REQ_OP_WRITE,
2762                 .op_flags = wbc_to_write_flags(wbc),
2763                 .old_blkaddr = NULL_ADDR,
2764                 .page = page,
2765                 .encrypted_page = NULL,
2766                 .submitted = false,
2767                 .compr_blocks = compr_blocks,
2768                 .need_lock = LOCK_RETRY,
2769                 .io_type = io_type,
2770                 .io_wbc = wbc,
2771                 .bio = bio,
2772                 .last_block = last_block,
2773         };
2774
2775         trace_f2fs_writepage(page, DATA);
2776
2777         /* we should bypass data pages to proceed the kworkder jobs */
2778         if (unlikely(f2fs_cp_error(sbi))) {
2779                 mapping_set_error(page->mapping, -EIO);
2780                 /*
2781                  * don't drop any dirty dentry pages for keeping lastest
2782                  * directory structure.
2783                  */
2784                 if (S_ISDIR(inode->i_mode))
2785                         goto redirty_out;
2786                 goto out;
2787         }
2788
2789         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2790                 goto redirty_out;
2791
2792         if (page->index < end_index ||
2793                         f2fs_verity_in_progress(inode) ||
2794                         compr_blocks)
2795                 goto write;
2796
2797         /*
2798          * If the offset is out-of-range of file size,
2799          * this page does not have to be written to disk.
2800          */
2801         offset = i_size & (PAGE_SIZE - 1);
2802         if ((page->index >= end_index + 1) || !offset)
2803                 goto out;
2804
2805         zero_user_segment(page, offset, PAGE_SIZE);
2806 write:
2807         if (f2fs_is_drop_cache(inode))
2808                 goto out;
2809         /* we should not write 0'th page having journal header */
2810         if (f2fs_is_volatile_file(inode) && (!page->index ||
2811                         (!wbc->for_reclaim &&
2812                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2813                 goto redirty_out;
2814
2815         /* Dentry/quota blocks are controlled by checkpoint */
2816         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2817                 /*
2818                  * We need to wait for node_write to avoid block allocation during
2819                  * checkpoint. This can only happen to quota writes which can cause
2820                  * the below discard race condition.
2821                  */
2822                 if (IS_NOQUOTA(inode))
2823                         down_read(&sbi->node_write);
2824
2825                 fio.need_lock = LOCK_DONE;
2826                 err = f2fs_do_write_data_page(&fio);
2827
2828                 if (IS_NOQUOTA(inode))
2829                         up_read(&sbi->node_write);
2830
2831                 goto done;
2832         }
2833
2834         if (!wbc->for_reclaim)
2835                 need_balance_fs = true;
2836         else if (has_not_enough_free_secs(sbi, 0, 0))
2837                 goto redirty_out;
2838         else
2839                 set_inode_flag(inode, FI_HOT_DATA);
2840
2841         err = -EAGAIN;
2842         if (f2fs_has_inline_data(inode)) {
2843                 err = f2fs_write_inline_data(inode, page);
2844                 if (!err)
2845                         goto out;
2846         }
2847
2848         if (err == -EAGAIN) {
2849                 err = f2fs_do_write_data_page(&fio);
2850                 if (err == -EAGAIN) {
2851                         fio.need_lock = LOCK_REQ;
2852                         err = f2fs_do_write_data_page(&fio);
2853                 }
2854         }
2855
2856         if (err) {
2857                 file_set_keep_isize(inode);
2858         } else {
2859                 spin_lock(&F2FS_I(inode)->i_size_lock);
2860                 if (F2FS_I(inode)->last_disk_size < psize)
2861                         F2FS_I(inode)->last_disk_size = psize;
2862                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2863         }
2864
2865 done:
2866         if (err && err != -ENOENT)
2867                 goto redirty_out;
2868
2869 out:
2870         inode_dec_dirty_pages(inode);
2871         if (err) {
2872                 ClearPageUptodate(page);
2873                 clear_cold_data(page);
2874         }
2875
2876         if (wbc->for_reclaim) {
2877                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2878                 clear_inode_flag(inode, FI_HOT_DATA);
2879                 f2fs_remove_dirty_inode(inode);
2880                 submitted = NULL;
2881         }
2882         unlock_page(page);
2883         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2884                                         !F2FS_I(inode)->cp_task)
2885                 f2fs_balance_fs(sbi, need_balance_fs);
2886
2887         if (unlikely(f2fs_cp_error(sbi))) {
2888                 f2fs_submit_merged_write(sbi, DATA);
2889                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2890                 submitted = NULL;
2891         }
2892
2893         if (submitted)
2894                 *submitted = fio.submitted ? 1 : 0;
2895
2896         return 0;
2897
2898 redirty_out:
2899         redirty_page_for_writepage(wbc, page);
2900         /*
2901          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2902          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2903          * file_write_and_wait_range() will see EIO error, which is critical
2904          * to return value of fsync() followed by atomic_write failure to user.
2905          */
2906         if (!err || wbc->for_reclaim)
2907                 return AOP_WRITEPAGE_ACTIVATE;
2908         unlock_page(page);
2909         return err;
2910 }
2911
2912 static int f2fs_write_data_page(struct page *page,
2913                                         struct writeback_control *wbc)
2914 {
2915 #ifdef CONFIG_F2FS_FS_COMPRESSION
2916         struct inode *inode = page->mapping->host;
2917
2918         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2919                 goto out;
2920
2921         if (f2fs_compressed_file(inode)) {
2922                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2923                         redirty_page_for_writepage(wbc, page);
2924                         return AOP_WRITEPAGE_ACTIVATE;
2925                 }
2926         }
2927 out:
2928 #endif
2929
2930         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2931                                                 wbc, FS_DATA_IO, 0);
2932 }
2933
2934 /*
2935  * This function was copied from write_cche_pages from mm/page-writeback.c.
2936  * The major change is making write step of cold data page separately from
2937  * warm/hot data page.
2938  */
2939 static int f2fs_write_cache_pages(struct address_space *mapping,
2940                                         struct writeback_control *wbc,
2941                                         enum iostat_type io_type)
2942 {
2943         int ret = 0;
2944         int done = 0, retry = 0;
2945         struct pagevec pvec;
2946         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2947         struct bio *bio = NULL;
2948         sector_t last_block;
2949 #ifdef CONFIG_F2FS_FS_COMPRESSION
2950         struct inode *inode = mapping->host;
2951         struct compress_ctx cc = {
2952                 .inode = inode,
2953                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2954                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2955                 .cluster_idx = NULL_CLUSTER,
2956                 .rpages = NULL,
2957                 .nr_rpages = 0,
2958                 .cpages = NULL,
2959                 .rbuf = NULL,
2960                 .cbuf = NULL,
2961                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2962                 .private = NULL,
2963         };
2964 #endif
2965         int nr_pages;
2966         pgoff_t index;
2967         pgoff_t end;            /* Inclusive */
2968         pgoff_t done_index;
2969         int range_whole = 0;
2970         xa_mark_t tag;
2971         int nwritten = 0;
2972         int submitted = 0;
2973         int i;
2974
2975         pagevec_init(&pvec);
2976
2977         if (get_dirty_pages(mapping->host) <=
2978                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2979                 set_inode_flag(mapping->host, FI_HOT_DATA);
2980         else
2981                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2982
2983         if (wbc->range_cyclic) {
2984                 index = mapping->writeback_index; /* prev offset */
2985                 end = -1;
2986         } else {
2987                 index = wbc->range_start >> PAGE_SHIFT;
2988                 end = wbc->range_end >> PAGE_SHIFT;
2989                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2990                         range_whole = 1;
2991         }
2992         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2993                 tag = PAGECACHE_TAG_TOWRITE;
2994         else
2995                 tag = PAGECACHE_TAG_DIRTY;
2996 retry:
2997         retry = 0;
2998         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2999                 tag_pages_for_writeback(mapping, index, end);
3000         done_index = index;
3001         while (!done && !retry && (index <= end)) {
3002                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3003                                 tag);
3004                 if (nr_pages == 0)
3005                         break;
3006
3007                 for (i = 0; i < nr_pages; i++) {
3008                         struct page *page = pvec.pages[i];
3009                         bool need_readd;
3010 readd:
3011                         need_readd = false;
3012 #ifdef CONFIG_F2FS_FS_COMPRESSION
3013                         if (f2fs_compressed_file(inode)) {
3014                                 ret = f2fs_init_compress_ctx(&cc);
3015                                 if (ret) {
3016                                         done = 1;
3017                                         break;
3018                                 }
3019
3020                                 if (!f2fs_cluster_can_merge_page(&cc,
3021                                                                 page->index)) {
3022                                         ret = f2fs_write_multi_pages(&cc,
3023                                                 &submitted, wbc, io_type);
3024                                         if (!ret)
3025                                                 need_readd = true;
3026                                         goto result;
3027                                 }
3028
3029                                 if (unlikely(f2fs_cp_error(sbi)))
3030                                         goto lock_page;
3031
3032                                 if (f2fs_cluster_is_empty(&cc)) {
3033                                         void *fsdata = NULL;
3034                                         struct page *pagep;
3035                                         int ret2;
3036
3037                                         ret2 = f2fs_prepare_compress_overwrite(
3038                                                         inode, &pagep,
3039                                                         page->index, &fsdata);
3040                                         if (ret2 < 0) {
3041                                                 ret = ret2;
3042                                                 done = 1;
3043                                                 break;
3044                                         } else if (ret2 &&
3045                                                 !f2fs_compress_write_end(inode,
3046                                                                 fsdata, page->index,
3047                                                                 1)) {
3048                                                 retry = 1;
3049                                                 break;
3050                                         }
3051                                 } else {
3052                                         goto lock_page;
3053                                 }
3054                         }
3055 #endif
3056                         /* give a priority to WB_SYNC threads */
3057                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3058                                         wbc->sync_mode == WB_SYNC_NONE) {
3059                                 done = 1;
3060                                 break;
3061                         }
3062 #ifdef CONFIG_F2FS_FS_COMPRESSION
3063 lock_page:
3064 #endif
3065                         done_index = page->index;
3066 retry_write:
3067                         lock_page(page);
3068
3069                         if (unlikely(page->mapping != mapping)) {
3070 continue_unlock:
3071                                 unlock_page(page);
3072                                 continue;
3073                         }
3074
3075                         if (!PageDirty(page)) {
3076                                 /* someone wrote it for us */
3077                                 goto continue_unlock;
3078                         }
3079
3080                         if (PageWriteback(page)) {
3081                                 if (wbc->sync_mode != WB_SYNC_NONE)
3082                                         f2fs_wait_on_page_writeback(page,
3083                                                         DATA, true, true);
3084                                 else
3085                                         goto continue_unlock;
3086                         }
3087
3088                         if (!clear_page_dirty_for_io(page))
3089                                 goto continue_unlock;
3090
3091 #ifdef CONFIG_F2FS_FS_COMPRESSION
3092                         if (f2fs_compressed_file(inode)) {
3093                                 get_page(page);
3094                                 f2fs_compress_ctx_add_page(&cc, page);
3095                                 continue;
3096                         }
3097 #endif
3098                         ret = f2fs_write_single_data_page(page, &submitted,
3099                                         &bio, &last_block, wbc, io_type, 0);
3100                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3101                                 unlock_page(page);
3102 #ifdef CONFIG_F2FS_FS_COMPRESSION
3103 result:
3104 #endif
3105                         nwritten += submitted;
3106                         wbc->nr_to_write -= submitted;
3107
3108                         if (unlikely(ret)) {
3109                                 /*
3110                                  * keep nr_to_write, since vfs uses this to
3111                                  * get # of written pages.
3112                                  */
3113                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3114                                         ret = 0;
3115                                         goto next;
3116                                 } else if (ret == -EAGAIN) {
3117                                         ret = 0;
3118                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3119                                                 cond_resched();
3120                                                 congestion_wait(BLK_RW_ASYNC,
3121                                                         DEFAULT_IO_TIMEOUT);
3122                                                 goto retry_write;
3123                                         }
3124                                         goto next;
3125                                 }
3126                                 done_index = page->index + 1;
3127                                 done = 1;
3128                                 break;
3129                         }
3130
3131                         if (wbc->nr_to_write <= 0 &&
3132                                         wbc->sync_mode == WB_SYNC_NONE) {
3133                                 done = 1;
3134                                 break;
3135                         }
3136 next:
3137                         if (need_readd)
3138                                 goto readd;
3139                 }
3140                 pagevec_release(&pvec);
3141                 cond_resched();
3142         }
3143 #ifdef CONFIG_F2FS_FS_COMPRESSION
3144         /* flush remained pages in compress cluster */
3145         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3146                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3147                 nwritten += submitted;
3148                 wbc->nr_to_write -= submitted;
3149                 if (ret) {
3150                         done = 1;
3151                         retry = 0;
3152                 }
3153         }
3154         if (f2fs_compressed_file(inode))
3155                 f2fs_destroy_compress_ctx(&cc);
3156 #endif
3157         if (retry) {
3158                 index = 0;
3159                 end = -1;
3160                 goto retry;
3161         }
3162         if (wbc->range_cyclic && !done)
3163                 done_index = 0;
3164         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3165                 mapping->writeback_index = done_index;
3166
3167         if (nwritten)
3168                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3169                                                                 NULL, 0, DATA);
3170         /* submit cached bio of IPU write */
3171         if (bio)
3172                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3173
3174         return ret;
3175 }
3176
3177 static inline bool __should_serialize_io(struct inode *inode,
3178                                         struct writeback_control *wbc)
3179 {
3180         /* to avoid deadlock in path of data flush */
3181         if (F2FS_I(inode)->cp_task)
3182                 return false;
3183
3184         if (!S_ISREG(inode->i_mode))
3185                 return false;
3186         if (IS_NOQUOTA(inode))
3187                 return false;
3188
3189         if (f2fs_need_compress_data(inode))
3190                 return true;
3191         if (wbc->sync_mode != WB_SYNC_ALL)
3192                 return true;
3193         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3194                 return true;
3195         return false;
3196 }
3197
3198 static int __f2fs_write_data_pages(struct address_space *mapping,
3199                                                 struct writeback_control *wbc,
3200                                                 enum iostat_type io_type)
3201 {
3202         struct inode *inode = mapping->host;
3203         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3204         struct blk_plug plug;
3205         int ret;
3206         bool locked = false;
3207
3208         /* deal with chardevs and other special file */
3209         if (!mapping->a_ops->writepage)
3210                 return 0;
3211
3212         /* skip writing if there is no dirty page in this inode */
3213         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3214                 return 0;
3215
3216         /* during POR, we don't need to trigger writepage at all. */
3217         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3218                 goto skip_write;
3219
3220         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3221                         wbc->sync_mode == WB_SYNC_NONE &&
3222                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3223                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3224                 goto skip_write;
3225
3226         /* skip writing during file defragment */
3227         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3228                 goto skip_write;
3229
3230         trace_f2fs_writepages(mapping->host, wbc, DATA);
3231
3232         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3233         if (wbc->sync_mode == WB_SYNC_ALL)
3234                 atomic_inc(&sbi->wb_sync_req[DATA]);
3235         else if (atomic_read(&sbi->wb_sync_req[DATA]))
3236                 goto skip_write;
3237
3238         if (__should_serialize_io(inode, wbc)) {
3239                 mutex_lock(&sbi->writepages);
3240                 locked = true;
3241         }
3242
3243         blk_start_plug(&plug);
3244         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3245         blk_finish_plug(&plug);
3246
3247         if (locked)
3248                 mutex_unlock(&sbi->writepages);
3249
3250         if (wbc->sync_mode == WB_SYNC_ALL)
3251                 atomic_dec(&sbi->wb_sync_req[DATA]);
3252         /*
3253          * if some pages were truncated, we cannot guarantee its mapping->host
3254          * to detect pending bios.
3255          */
3256
3257         f2fs_remove_dirty_inode(inode);
3258         return ret;
3259
3260 skip_write:
3261         wbc->pages_skipped += get_dirty_pages(inode);
3262         trace_f2fs_writepages(mapping->host, wbc, DATA);
3263         return 0;
3264 }
3265
3266 static int f2fs_write_data_pages(struct address_space *mapping,
3267                             struct writeback_control *wbc)
3268 {
3269         struct inode *inode = mapping->host;
3270
3271         return __f2fs_write_data_pages(mapping, wbc,
3272                         F2FS_I(inode)->cp_task == current ?
3273                         FS_CP_DATA_IO : FS_DATA_IO);
3274 }
3275
3276 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
3277 {
3278         struct inode *inode = mapping->host;
3279         loff_t i_size = i_size_read(inode);
3280
3281         if (IS_NOQUOTA(inode))
3282                 return;
3283
3284         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3285         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3286                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3287                 down_write(&F2FS_I(inode)->i_mmap_sem);
3288
3289                 truncate_pagecache(inode, i_size);
3290                 f2fs_truncate_blocks(inode, i_size, true);
3291
3292                 up_write(&F2FS_I(inode)->i_mmap_sem);
3293                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3294         }
3295 }
3296
3297 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3298                         struct page *page, loff_t pos, unsigned len,
3299                         block_t *blk_addr, bool *node_changed)
3300 {
3301         struct inode *inode = page->mapping->host;
3302         pgoff_t index = page->index;
3303         struct dnode_of_data dn;
3304         struct page *ipage;
3305         bool locked = false;
3306         struct extent_info ei = {0,0,0};
3307         int err = 0;
3308         int flag;
3309
3310         /*
3311          * we already allocated all the blocks, so we don't need to get
3312          * the block addresses when there is no need to fill the page.
3313          */
3314         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3315             !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3316             !f2fs_verity_in_progress(inode))
3317                 return 0;
3318
3319         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3320         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3321                 flag = F2FS_GET_BLOCK_DEFAULT;
3322         else
3323                 flag = F2FS_GET_BLOCK_PRE_AIO;
3324
3325         if (f2fs_has_inline_data(inode) ||
3326                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3327                 f2fs_do_map_lock(sbi, flag, true);
3328                 locked = true;
3329         }
3330
3331 restart:
3332         /* check inline_data */
3333         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3334         if (IS_ERR(ipage)) {
3335                 err = PTR_ERR(ipage);
3336                 goto unlock_out;
3337         }
3338
3339         set_new_dnode(&dn, inode, ipage, ipage, 0);
3340
3341         if (f2fs_has_inline_data(inode)) {
3342                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3343                         f2fs_do_read_inline_data(page, ipage);
3344                         set_inode_flag(inode, FI_DATA_EXIST);
3345                         if (inode->i_nlink)
3346                                 set_inline_node(ipage);
3347                 } else {
3348                         err = f2fs_convert_inline_page(&dn, page);
3349                         if (err)
3350                                 goto out;
3351                         if (dn.data_blkaddr == NULL_ADDR)
3352                                 err = f2fs_get_block(&dn, index);
3353                 }
3354         } else if (locked) {
3355                 err = f2fs_get_block(&dn, index);
3356         } else {
3357                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3358                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3359                 } else {
3360                         /* hole case */
3361                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3362                         if (err || dn.data_blkaddr == NULL_ADDR) {
3363                                 f2fs_put_dnode(&dn);
3364                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3365                                                                 true);
3366                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3367                                 locked = true;
3368                                 goto restart;
3369                         }
3370                 }
3371         }
3372
3373         /* convert_inline_page can make node_changed */
3374         *blk_addr = dn.data_blkaddr;
3375         *node_changed = dn.node_changed;
3376 out:
3377         f2fs_put_dnode(&dn);
3378 unlock_out:
3379         if (locked)
3380                 f2fs_do_map_lock(sbi, flag, false);
3381         return err;
3382 }
3383
3384 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3385                 loff_t pos, unsigned len, unsigned flags,
3386                 struct page **pagep, void **fsdata)
3387 {
3388         struct inode *inode = mapping->host;
3389         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3390         struct page *page = NULL;
3391         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3392         bool need_balance = false, drop_atomic = false;
3393         block_t blkaddr = NULL_ADDR;
3394         int err = 0;
3395
3396         trace_f2fs_write_begin(inode, pos, len, flags);
3397
3398         if (!f2fs_is_checkpoint_ready(sbi)) {
3399                 err = -ENOSPC;
3400                 goto fail;
3401         }
3402
3403         if ((f2fs_is_atomic_file(inode) &&
3404                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3405                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3406                 err = -ENOMEM;
3407                 drop_atomic = true;
3408                 goto fail;
3409         }
3410
3411         /*
3412          * We should check this at this moment to avoid deadlock on inode page
3413          * and #0 page. The locking rule for inline_data conversion should be:
3414          * lock_page(page #0) -> lock_page(inode_page)
3415          */
3416         if (index != 0) {
3417                 err = f2fs_convert_inline_inode(inode);
3418                 if (err)
3419                         goto fail;
3420         }
3421
3422 #ifdef CONFIG_F2FS_FS_COMPRESSION
3423         if (f2fs_compressed_file(inode)) {
3424                 int ret;
3425
3426                 *fsdata = NULL;
3427
3428                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3429                                                         index, fsdata);
3430                 if (ret < 0) {
3431                         err = ret;
3432                         goto fail;
3433                 } else if (ret) {
3434                         return 0;
3435                 }
3436         }
3437 #endif
3438
3439 repeat:
3440         /*
3441          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3442          * wait_for_stable_page. Will wait that below with our IO control.
3443          */
3444         page = f2fs_pagecache_get_page(mapping, index,
3445                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3446         if (!page) {
3447                 err = -ENOMEM;
3448                 goto fail;
3449         }
3450
3451         /* TODO: cluster can be compressed due to race with .writepage */
3452
3453         *pagep = page;
3454
3455         err = prepare_write_begin(sbi, page, pos, len,
3456                                         &blkaddr, &need_balance);
3457         if (err)
3458                 goto fail;
3459
3460         if (need_balance && !IS_NOQUOTA(inode) &&
3461                         has_not_enough_free_secs(sbi, 0, 0)) {
3462                 unlock_page(page);
3463                 f2fs_balance_fs(sbi, true);
3464                 lock_page(page);
3465                 if (page->mapping != mapping) {
3466                         /* The page got truncated from under us */
3467                         f2fs_put_page(page, 1);
3468                         goto repeat;
3469                 }
3470         }
3471
3472         f2fs_wait_on_page_writeback(page, DATA, false, true);
3473
3474         if (len == PAGE_SIZE || PageUptodate(page))
3475                 return 0;
3476
3477         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3478             !f2fs_verity_in_progress(inode)) {
3479                 zero_user_segment(page, len, PAGE_SIZE);
3480                 return 0;
3481         }
3482
3483         if (blkaddr == NEW_ADDR) {
3484                 zero_user_segment(page, 0, PAGE_SIZE);
3485                 SetPageUptodate(page);
3486         } else {
3487                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3488                                 DATA_GENERIC_ENHANCE_READ)) {
3489                         err = -EFSCORRUPTED;
3490                         goto fail;
3491                 }
3492                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3493                 if (err)
3494                         goto fail;
3495
3496                 lock_page(page);
3497                 if (unlikely(page->mapping != mapping)) {
3498                         f2fs_put_page(page, 1);
3499                         goto repeat;
3500                 }
3501                 if (unlikely(!PageUptodate(page))) {
3502                         err = -EIO;
3503                         goto fail;
3504                 }
3505         }
3506         return 0;
3507
3508 fail:
3509         f2fs_put_page(page, 1);
3510         f2fs_write_failed(mapping, pos + len);
3511         if (drop_atomic)
3512                 f2fs_drop_inmem_pages_all(sbi, false);
3513         return err;
3514 }
3515
3516 static int f2fs_write_end(struct file *file,
3517                         struct address_space *mapping,
3518                         loff_t pos, unsigned len, unsigned copied,
3519                         struct page *page, void *fsdata)
3520 {
3521         struct inode *inode = page->mapping->host;
3522
3523         trace_f2fs_write_end(inode, pos, len, copied);
3524
3525         /*
3526          * This should be come from len == PAGE_SIZE, and we expect copied
3527          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3528          * let generic_perform_write() try to copy data again through copied=0.
3529          */
3530         if (!PageUptodate(page)) {
3531                 if (unlikely(copied != len))
3532                         copied = 0;
3533                 else
3534                         SetPageUptodate(page);
3535         }
3536
3537 #ifdef CONFIG_F2FS_FS_COMPRESSION
3538         /* overwrite compressed file */
3539         if (f2fs_compressed_file(inode) && fsdata) {
3540                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3541                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3542
3543                 if (pos + copied > i_size_read(inode) &&
3544                                 !f2fs_verity_in_progress(inode))
3545                         f2fs_i_size_write(inode, pos + copied);
3546                 return copied;
3547         }
3548 #endif
3549
3550         if (!copied)
3551                 goto unlock_out;
3552
3553         set_page_dirty(page);
3554
3555         if (pos + copied > i_size_read(inode) &&
3556             !f2fs_verity_in_progress(inode))
3557                 f2fs_i_size_write(inode, pos + copied);
3558 unlock_out:
3559         f2fs_put_page(page, 1);
3560         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3561         return copied;
3562 }
3563
3564 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3565                            loff_t offset)
3566 {
3567         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3568         unsigned blkbits = i_blkbits;
3569         unsigned blocksize_mask = (1 << blkbits) - 1;
3570         unsigned long align = offset | iov_iter_alignment(iter);
3571         struct block_device *bdev = inode->i_sb->s_bdev;
3572
3573         if (iov_iter_rw(iter) == READ && offset >= i_size_read(inode))
3574                 return 1;
3575
3576         if (align & blocksize_mask) {
3577                 if (bdev)
3578                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
3579                 blocksize_mask = (1 << blkbits) - 1;
3580                 if (align & blocksize_mask)
3581                         return -EINVAL;
3582                 return 1;
3583         }
3584         return 0;
3585 }
3586
3587 static void f2fs_dio_end_io(struct bio *bio)
3588 {
3589         struct f2fs_private_dio *dio = bio->bi_private;
3590
3591         dec_page_count(F2FS_I_SB(dio->inode),
3592                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3593
3594         bio->bi_private = dio->orig_private;
3595         bio->bi_end_io = dio->orig_end_io;
3596
3597         kfree(dio);
3598
3599         bio_endio(bio);
3600 }
3601
3602 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3603                                                         loff_t file_offset)
3604 {
3605         struct f2fs_private_dio *dio;
3606         bool write = (bio_op(bio) == REQ_OP_WRITE);
3607
3608         dio = f2fs_kzalloc(F2FS_I_SB(inode),
3609                         sizeof(struct f2fs_private_dio), GFP_NOFS);
3610         if (!dio)
3611                 goto out;
3612
3613         dio->inode = inode;
3614         dio->orig_end_io = bio->bi_end_io;
3615         dio->orig_private = bio->bi_private;
3616         dio->write = write;
3617
3618         bio->bi_end_io = f2fs_dio_end_io;
3619         bio->bi_private = dio;
3620
3621         inc_page_count(F2FS_I_SB(inode),
3622                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3623
3624         submit_bio(bio);
3625         return;
3626 out:
3627         bio->bi_status = BLK_STS_IOERR;
3628         bio_endio(bio);
3629 }
3630
3631 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3632 {
3633         struct address_space *mapping = iocb->ki_filp->f_mapping;
3634         struct inode *inode = mapping->host;
3635         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3636         struct f2fs_inode_info *fi = F2FS_I(inode);
3637         size_t count = iov_iter_count(iter);
3638         loff_t offset = iocb->ki_pos;
3639         int rw = iov_iter_rw(iter);
3640         int err;
3641         enum rw_hint hint = iocb->ki_hint;
3642         int whint_mode = F2FS_OPTION(sbi).whint_mode;
3643         bool do_opu;
3644
3645         err = check_direct_IO(inode, iter, offset);
3646         if (err)
3647                 return err < 0 ? err : 0;
3648
3649         if (f2fs_force_buffered_io(inode, iocb, iter))
3650                 return 0;
3651
3652         do_opu = allow_outplace_dio(inode, iocb, iter);
3653
3654         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3655
3656         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3657                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
3658
3659         if (iocb->ki_flags & IOCB_NOWAIT) {
3660                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3661                         iocb->ki_hint = hint;
3662                         err = -EAGAIN;
3663                         goto out;
3664                 }
3665                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3666                         up_read(&fi->i_gc_rwsem[rw]);
3667                         iocb->ki_hint = hint;
3668                         err = -EAGAIN;
3669                         goto out;
3670                 }
3671         } else {
3672                 down_read(&fi->i_gc_rwsem[rw]);
3673                 if (do_opu)
3674                         down_read(&fi->i_gc_rwsem[READ]);
3675         }
3676
3677         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3678                         iter, rw == WRITE ? get_data_block_dio_write :
3679                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
3680                         rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3681                         DIO_SKIP_HOLES);
3682
3683         if (do_opu)
3684                 up_read(&fi->i_gc_rwsem[READ]);
3685
3686         up_read(&fi->i_gc_rwsem[rw]);
3687
3688         if (rw == WRITE) {
3689                 if (whint_mode == WHINT_MODE_OFF)
3690                         iocb->ki_hint = hint;
3691                 if (err > 0) {
3692                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3693                                                                         err);
3694                         if (!do_opu)
3695                                 set_inode_flag(inode, FI_UPDATE_WRITE);
3696                 } else if (err == -EIOCBQUEUED) {
3697                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3698                                                 count - iov_iter_count(iter));
3699                 } else if (err < 0) {
3700                         f2fs_write_failed(mapping, offset + count);
3701                 }
3702         } else {
3703                 if (err > 0)
3704                         f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3705                 else if (err == -EIOCBQUEUED)
3706                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_READ_IO,
3707                                                 count - iov_iter_count(iter));
3708         }
3709
3710 out:
3711         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3712
3713         return err;
3714 }
3715
3716 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3717                                                         unsigned int length)
3718 {
3719         struct inode *inode = page->mapping->host;
3720         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3721
3722         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3723                 (offset % PAGE_SIZE || length != PAGE_SIZE))
3724                 return;
3725
3726         if (PageDirty(page)) {
3727                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3728                         dec_page_count(sbi, F2FS_DIRTY_META);
3729                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3730                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3731                 } else {
3732                         inode_dec_dirty_pages(inode);
3733                         f2fs_remove_dirty_inode(inode);
3734                 }
3735         }
3736
3737         clear_cold_data(page);
3738
3739         if (IS_ATOMIC_WRITTEN_PAGE(page))
3740                 return f2fs_drop_inmem_page(inode, page);
3741
3742         f2fs_clear_page_private(page);
3743 }
3744
3745 int f2fs_release_page(struct page *page, gfp_t wait)
3746 {
3747         /* If this is dirty page, keep PagePrivate */
3748         if (PageDirty(page))
3749                 return 0;
3750
3751         /* This is atomic written page, keep Private */
3752         if (IS_ATOMIC_WRITTEN_PAGE(page))
3753                 return 0;
3754
3755         clear_cold_data(page);
3756         f2fs_clear_page_private(page);
3757         return 1;
3758 }
3759
3760 static int f2fs_set_data_page_dirty(struct page *page)
3761 {
3762         struct inode *inode = page_file_mapping(page)->host;
3763
3764         trace_f2fs_set_page_dirty(page, DATA);
3765
3766         if (!PageUptodate(page))
3767                 SetPageUptodate(page);
3768         if (PageSwapCache(page))
3769                 return __set_page_dirty_nobuffers(page);
3770
3771         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3772                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
3773                         f2fs_register_inmem_page(inode, page);
3774                         return 1;
3775                 }
3776                 /*
3777                  * Previously, this page has been registered, we just
3778                  * return here.
3779                  */
3780                 return 0;
3781         }
3782
3783         if (!PageDirty(page)) {
3784                 __set_page_dirty_nobuffers(page);
3785                 f2fs_update_dirty_page(inode, page);
3786                 return 1;
3787         }
3788         return 0;
3789 }
3790
3791
3792 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3793 {
3794 #ifdef CONFIG_F2FS_FS_COMPRESSION
3795         struct dnode_of_data dn;
3796         sector_t start_idx, blknr = 0;
3797         int ret;
3798
3799         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3800
3801         set_new_dnode(&dn, inode, NULL, NULL, 0);
3802         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3803         if (ret)
3804                 return 0;
3805
3806         if (dn.data_blkaddr != COMPRESS_ADDR) {
3807                 dn.ofs_in_node += block - start_idx;
3808                 blknr = f2fs_data_blkaddr(&dn);
3809                 if (!__is_valid_data_blkaddr(blknr))
3810                         blknr = 0;
3811         }
3812
3813         f2fs_put_dnode(&dn);
3814         return blknr;
3815 #else
3816         return 0;
3817 #endif
3818 }
3819
3820
3821 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3822 {
3823         struct inode *inode = mapping->host;
3824         sector_t blknr = 0;
3825
3826         if (f2fs_has_inline_data(inode))
3827                 goto out;
3828
3829         /* make sure allocating whole blocks */
3830         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3831                 filemap_write_and_wait(mapping);
3832
3833         /* Block number less than F2FS MAX BLOCKS */
3834         if (unlikely(block >= F2FS_I_SB(inode)->max_file_blocks))
3835                 goto out;
3836
3837         if (f2fs_compressed_file(inode)) {
3838                 blknr = f2fs_bmap_compress(inode, block);
3839         } else {
3840                 struct f2fs_map_blocks map;
3841
3842                 memset(&map, 0, sizeof(map));
3843                 map.m_lblk = block;
3844                 map.m_len = 1;
3845                 map.m_next_pgofs = NULL;
3846                 map.m_seg_type = NO_CHECK_TYPE;
3847
3848                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3849                         blknr = map.m_pblk;
3850         }
3851 out:
3852         trace_f2fs_bmap(inode, block, blknr);
3853         return blknr;
3854 }
3855
3856 #ifdef CONFIG_MIGRATION
3857 #include <linux/migrate.h>
3858
3859 int f2fs_migrate_page(struct address_space *mapping,
3860                 struct page *newpage, struct page *page, enum migrate_mode mode)
3861 {
3862         int rc, extra_count;
3863         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3864         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
3865
3866         BUG_ON(PageWriteback(page));
3867
3868         /* migrating an atomic written page is safe with the inmem_lock hold */
3869         if (atomic_written) {
3870                 if (mode != MIGRATE_SYNC)
3871                         return -EBUSY;
3872                 if (!mutex_trylock(&fi->inmem_lock))
3873                         return -EAGAIN;
3874         }
3875
3876         /* one extra reference was held for atomic_write page */
3877         extra_count = atomic_written ? 1 : 0;
3878         rc = migrate_page_move_mapping(mapping, newpage,
3879                                 page, extra_count);
3880         if (rc != MIGRATEPAGE_SUCCESS) {
3881                 if (atomic_written)
3882                         mutex_unlock(&fi->inmem_lock);
3883                 return rc;
3884         }
3885
3886         if (atomic_written) {
3887                 struct inmem_pages *cur;
3888                 list_for_each_entry(cur, &fi->inmem_pages, list)
3889                         if (cur->page == page) {
3890                                 cur->page = newpage;
3891                                 break;
3892                         }
3893                 mutex_unlock(&fi->inmem_lock);
3894                 put_page(page);
3895                 get_page(newpage);
3896         }
3897
3898         if (PagePrivate(page)) {
3899                 f2fs_set_page_private(newpage, page_private(page));
3900                 f2fs_clear_page_private(page);
3901         }
3902
3903         if (mode != MIGRATE_SYNC_NO_COPY)
3904                 migrate_page_copy(newpage, page);
3905         else
3906                 migrate_page_states(newpage, page);
3907
3908         return MIGRATEPAGE_SUCCESS;
3909 }
3910 #endif
3911
3912 #ifdef CONFIG_SWAP
3913 static int check_swap_activate_fast(struct swap_info_struct *sis,
3914                                 struct file *swap_file, sector_t *span)
3915 {
3916         struct address_space *mapping = swap_file->f_mapping;
3917         struct inode *inode = mapping->host;
3918         sector_t cur_lblock;
3919         sector_t last_lblock;
3920         sector_t pblock;
3921         sector_t lowest_pblock = -1;
3922         sector_t highest_pblock = 0;
3923         int nr_extents = 0;
3924         unsigned long nr_pblocks;
3925         u64 len;
3926         int ret;
3927
3928         /*
3929          * Map all the blocks into the extent list.  This code doesn't try
3930          * to be very smart.
3931          */
3932         cur_lblock = 0;
3933         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3934         len = i_size_read(inode);
3935
3936         while (cur_lblock <= last_lblock && cur_lblock < sis->max) {
3937                 struct f2fs_map_blocks map;
3938                 pgoff_t next_pgofs;
3939
3940                 cond_resched();
3941
3942                 memset(&map, 0, sizeof(map));
3943                 map.m_lblk = cur_lblock;
3944                 map.m_len = bytes_to_blks(inode, len) - cur_lblock;
3945                 map.m_next_pgofs = &next_pgofs;
3946                 map.m_seg_type = NO_CHECK_TYPE;
3947
3948                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3949                 if (ret)
3950                         goto err_out;
3951
3952                 /* hole */
3953                 if (!(map.m_flags & F2FS_MAP_FLAGS))
3954                         goto err_out;
3955
3956                 pblock = map.m_pblk;
3957                 nr_pblocks = map.m_len;
3958
3959                 if (cur_lblock + nr_pblocks >= sis->max)
3960                         nr_pblocks = sis->max - cur_lblock;
3961
3962                 if (cur_lblock) {       /* exclude the header page */
3963                         if (pblock < lowest_pblock)
3964                                 lowest_pblock = pblock;
3965                         if (pblock + nr_pblocks - 1 > highest_pblock)
3966                                 highest_pblock = pblock + nr_pblocks - 1;
3967                 }
3968
3969                 /*
3970                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3971                  */
3972                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3973                 if (ret < 0)
3974                         goto out;
3975                 nr_extents += ret;
3976                 cur_lblock += nr_pblocks;
3977         }
3978         ret = nr_extents;
3979         *span = 1 + highest_pblock - lowest_pblock;
3980         if (cur_lblock == 0)
3981                 cur_lblock = 1; /* force Empty message */
3982         sis->max = cur_lblock;
3983         sis->pages = cur_lblock - 1;
3984         sis->highest_bit = cur_lblock - 1;
3985 out:
3986         return ret;
3987 err_out:
3988         pr_err("swapon: swapfile has holes\n");
3989         return -EINVAL;
3990 }
3991
3992 /* Copied from generic_swapfile_activate() to check any holes */
3993 static int check_swap_activate(struct swap_info_struct *sis,
3994                                 struct file *swap_file, sector_t *span)
3995 {
3996         struct address_space *mapping = swap_file->f_mapping;
3997         struct inode *inode = mapping->host;
3998         unsigned blocks_per_page;
3999         unsigned long page_no;
4000         sector_t probe_block;
4001         sector_t last_block;
4002         sector_t lowest_block = -1;
4003         sector_t highest_block = 0;
4004         int nr_extents = 0;
4005         int ret;
4006
4007         if (PAGE_SIZE == F2FS_BLKSIZE)
4008                 return check_swap_activate_fast(sis, swap_file, span);
4009
4010         blocks_per_page = bytes_to_blks(inode, PAGE_SIZE);
4011
4012         /*
4013          * Map all the blocks into the extent list.  This code doesn't try
4014          * to be very smart.
4015          */
4016         probe_block = 0;
4017         page_no = 0;
4018         last_block = bytes_to_blks(inode, i_size_read(inode));
4019         while ((probe_block + blocks_per_page) <= last_block &&
4020                         page_no < sis->max) {
4021                 unsigned block_in_page;
4022                 sector_t first_block;
4023                 sector_t block = 0;
4024                 int      err = 0;
4025
4026                 cond_resched();
4027
4028                 block = probe_block;
4029                 err = bmap(inode, &block);
4030                 if (err || !block)
4031                         goto bad_bmap;
4032                 first_block = block;
4033
4034                 /*
4035                  * It must be PAGE_SIZE aligned on-disk
4036                  */
4037                 if (first_block & (blocks_per_page - 1)) {
4038                         probe_block++;
4039                         goto reprobe;
4040                 }
4041
4042                 for (block_in_page = 1; block_in_page < blocks_per_page;
4043                                         block_in_page++) {
4044
4045                         block = probe_block + block_in_page;
4046                         err = bmap(inode, &block);
4047
4048                         if (err || !block)
4049                                 goto bad_bmap;
4050
4051                         if (block != first_block + block_in_page) {
4052                                 /* Discontiguity */
4053                                 probe_block++;
4054                                 goto reprobe;
4055                         }
4056                 }
4057
4058                 first_block >>= (PAGE_SHIFT - inode->i_blkbits);
4059                 if (page_no) {  /* exclude the header page */
4060                         if (first_block < lowest_block)
4061                                 lowest_block = first_block;
4062                         if (first_block > highest_block)
4063                                 highest_block = first_block;
4064                 }
4065
4066                 /*
4067                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4068                  */
4069                 ret = add_swap_extent(sis, page_no, 1, first_block);
4070                 if (ret < 0)
4071                         goto out;
4072                 nr_extents += ret;
4073                 page_no++;
4074                 probe_block += blocks_per_page;
4075 reprobe:
4076                 continue;
4077         }
4078         ret = nr_extents;
4079         *span = 1 + highest_block - lowest_block;
4080         if (page_no == 0)
4081                 page_no = 1;    /* force Empty message */
4082         sis->max = page_no;
4083         sis->pages = page_no - 1;
4084         sis->highest_bit = page_no - 1;
4085 out:
4086         return ret;
4087 bad_bmap:
4088         pr_err("swapon: swapfile has holes\n");
4089         return -EINVAL;
4090 }
4091
4092 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4093                                 sector_t *span)
4094 {
4095         struct inode *inode = file_inode(file);
4096         int ret;
4097
4098         if (!S_ISREG(inode->i_mode))
4099                 return -EINVAL;
4100
4101         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4102                 return -EROFS;
4103
4104         ret = f2fs_convert_inline_inode(inode);
4105         if (ret)
4106                 return ret;
4107
4108         if (!f2fs_disable_compressed_file(inode))
4109                 return -EINVAL;
4110
4111         ret = check_swap_activate(sis, file, span);
4112         if (ret < 0)
4113                 return ret;
4114
4115         set_inode_flag(inode, FI_PIN_FILE);
4116         f2fs_precache_extents(inode);
4117         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4118         return ret;
4119 }
4120
4121 static void f2fs_swap_deactivate(struct file *file)
4122 {
4123         struct inode *inode = file_inode(file);
4124
4125         clear_inode_flag(inode, FI_PIN_FILE);
4126 }
4127 #else
4128 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4129                                 sector_t *span)
4130 {
4131         return -EOPNOTSUPP;
4132 }
4133
4134 static void f2fs_swap_deactivate(struct file *file)
4135 {
4136 }
4137 #endif
4138
4139 const struct address_space_operations f2fs_dblock_aops = {
4140         .readpage       = f2fs_read_data_page,
4141         .readahead      = f2fs_readahead,
4142         .writepage      = f2fs_write_data_page,
4143         .writepages     = f2fs_write_data_pages,
4144         .write_begin    = f2fs_write_begin,
4145         .write_end      = f2fs_write_end,
4146         .set_page_dirty = f2fs_set_data_page_dirty,
4147         .invalidatepage = f2fs_invalidate_page,
4148         .releasepage    = f2fs_release_page,
4149         .direct_IO      = f2fs_direct_IO,
4150         .bmap           = f2fs_bmap,
4151         .swap_activate  = f2fs_swap_activate,
4152         .swap_deactivate = f2fs_swap_deactivate,
4153 #ifdef CONFIG_MIGRATION
4154         .migratepage    = f2fs_migrate_page,
4155 #endif
4156 };
4157
4158 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4159 {
4160         struct address_space *mapping = page_mapping(page);
4161         unsigned long flags;
4162
4163         xa_lock_irqsave(&mapping->i_pages, flags);
4164         __xa_clear_mark(&mapping->i_pages, page_index(page),
4165                                                 PAGECACHE_TAG_DIRTY);
4166         xa_unlock_irqrestore(&mapping->i_pages, flags);
4167 }
4168
4169 int __init f2fs_init_post_read_processing(void)
4170 {
4171         bio_post_read_ctx_cache =
4172                 kmem_cache_create("f2fs_bio_post_read_ctx",
4173                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4174         if (!bio_post_read_ctx_cache)
4175                 goto fail;
4176         bio_post_read_ctx_pool =
4177                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4178                                          bio_post_read_ctx_cache);
4179         if (!bio_post_read_ctx_pool)
4180                 goto fail_free_cache;
4181         return 0;
4182
4183 fail_free_cache:
4184         kmem_cache_destroy(bio_post_read_ctx_cache);
4185 fail:
4186         return -ENOMEM;
4187 }
4188
4189 void f2fs_destroy_post_read_processing(void)
4190 {
4191         mempool_destroy(bio_post_read_ctx_pool);
4192         kmem_cache_destroy(bio_post_read_ctx_cache);
4193 }
4194
4195 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4196 {
4197         if (!f2fs_sb_has_encrypt(sbi) &&
4198                 !f2fs_sb_has_verity(sbi) &&
4199                 !f2fs_sb_has_compression(sbi))
4200                 return 0;
4201
4202         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4203                                                  WQ_UNBOUND | WQ_HIGHPRI,
4204                                                  num_online_cpus());
4205         if (!sbi->post_read_wq)
4206                 return -ENOMEM;
4207         return 0;
4208 }
4209
4210 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4211 {
4212         if (sbi->post_read_wq)
4213                 destroy_workqueue(sbi->post_read_wq);
4214 }
4215
4216 int __init f2fs_init_bio_entry_cache(void)
4217 {
4218         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4219                         sizeof(struct bio_entry));
4220         if (!bio_entry_slab)
4221                 return -ENOMEM;
4222         return 0;
4223 }
4224
4225 void f2fs_destroy_bio_entry_cache(void)
4226 {
4227         kmem_cache_destroy(bio_entry_slab);
4228 }