fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/sched/mm.h>
  12 #include <linux/mpage.h>
  13 #include <linux/writeback.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/blk-crypto.h>
  18 #include <linux/swap.h>
  19 #include <linux/prefetch.h>
  20 #include <linux/uio.h>
  21 #include <linux/sched/signal.h>
  22 #include <linux/fiemap.h>
  23 #include <linux/iomap.h>
  24
  25 #include "f2fs.h"
  26 #include "node.h"
  27 #include "segment.h"
  28 #include "iostat.h"
  29 #include <trace/events/f2fs.h>
  30
  31 #define NUM_PREALLOC_POST_READ_CTXS     128
  32
  33 static struct kmem_cache *bio_post_read_ctx_cache;
  34 static struct kmem_cache *bio_entry_slab;
  35 static mempool_t *bio_post_read_ctx_pool;
  36 static struct bio_set f2fs_bioset;
  37
  38 #define F2FS_BIO_POOL_SIZE      NR_CURSEG_TYPE
  39
  40 int __init f2fs_init_bioset(void)
  41 {
  42         if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  43                                         0, BIOSET_NEED_BVECS))
  44                 return -ENOMEM;
  45         return 0;
  46 }
  47
  48 void f2fs_destroy_bioset(void)
  49 {
  50         bioset_exit(&f2fs_bioset);
  51 }
  52
  53 static bool __is_cp_guaranteed(struct page *page)
  54 {
  55         struct address_space *mapping = page->mapping;
  56         struct inode *inode;
  57         struct f2fs_sb_info *sbi;
  58
  59         if (!mapping)
  60                 return false;
  61
  62         inode = mapping->host;
  63         sbi = F2FS_I_SB(inode);
  64
  65         if (inode->i_ino == F2FS_META_INO(sbi) ||
  66                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  67                         S_ISDIR(inode->i_mode))
  68                 return true;
  69
  70         if (f2fs_is_compressed_page(page))
  71                 return false;
  72         if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
  73                         page_private_gcing(page))
  74                 return true;
  75         return false;
  76 }
  77
  78 static enum count_type __read_io_type(struct page *page)
  79 {
  80         struct address_space *mapping = page_file_mapping(page);
  81
  82         if (mapping) {
  83                 struct inode *inode = mapping->host;
  84                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  85
  86                 if (inode->i_ino == F2FS_META_INO(sbi))
  87                         return F2FS_RD_META;
  88
  89                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  90                         return F2FS_RD_NODE;
  91         }
  92         return F2FS_RD_DATA;
  93 }
  94
  95 /* postprocessing steps for read bios */
  96 enum bio_post_read_step {
  97 #ifdef CONFIG_FS_ENCRYPTION
  98         STEP_DECRYPT    = 1 << 0,
  99 #else
 100         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
 101 #endif
 102 #ifdef CONFIG_F2FS_FS_COMPRESSION
 103         STEP_DECOMPRESS = 1 << 1,
 104 #else
 105         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 106 #endif
 107 #ifdef CONFIG_FS_VERITY
 108         STEP_VERITY     = 1 << 2,
 109 #else
 110         STEP_VERITY     = 0,    /* compile out the verity-related code */
 111 #endif
 112 };
 113
 114 struct bio_post_read_ctx {
 115         struct bio *bio;
 116         struct f2fs_sb_info *sbi;
 117         struct work_struct work;
 118         unsigned int enabled_steps;
 119         block_t fs_blkaddr;
 120 };
 121
 122 static void f2fs_finish_read_bio(struct bio *bio)
 123 {
 124         struct bio_vec *bv;
 125         struct bvec_iter_all iter_all;
 126
 127         /*
 128          * Update and unlock the bio's pagecache pages, and put the
 129          * decompression context for any compressed pages.
 130          */
 131         bio_for_each_segment_all(bv, bio, iter_all) {
 132                 struct page *page = bv->bv_page;
 133
 134                 if (f2fs_is_compressed_page(page)) {
 135                         if (bio->bi_status)
 136                                 f2fs_end_read_compressed_page(page, true, 0);
 137                         f2fs_put_page_dic(page);
 138                         continue;
 139                 }
 140
 141                 /* PG_error was set if decryption or verity failed. */
 142                 if (bio->bi_status || PageError(page)) {
 143                         ClearPageUptodate(page);
 144                         /* will re-read again later */
 145                         ClearPageError(page);
 146                 } else {
 147                         SetPageUptodate(page);
 148                 }
 149                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 150                 unlock_page(page);
 151         }
 152
 153         if (bio->bi_private)
 154                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 155         bio_put(bio);
 156 }
 157
 158 static void f2fs_verify_bio(struct work_struct *work)
 159 {
 160         struct bio_post_read_ctx *ctx =
 161                 container_of(work, struct bio_post_read_ctx, work);
 162         struct bio *bio = ctx->bio;
 163         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 164
 165         /*
 166          * fsverity_verify_bio() may call readahead() again, and while verity
 167          * will be disabled for this, decryption and/or decompression may still
 168          * be needed, resulting in another bio_post_read_ctx being allocated.
 169          * So to prevent deadlocks we need to release the current ctx to the
 170          * mempool first.  This assumes that verity is the last post-read step.
 171          */
 172         mempool_free(ctx, bio_post_read_ctx_pool);
 173         bio->bi_private = NULL;
 174
 175         /*
 176          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 177          * as those were handled separately by f2fs_end_read_compressed_page().
 178          */
 179         if (may_have_compressed_pages) {
 180                 struct bio_vec *bv;
 181                 struct bvec_iter_all iter_all;
 182
 183                 bio_for_each_segment_all(bv, bio, iter_all) {
 184                         struct page *page = bv->bv_page;
 185
 186                         if (!f2fs_is_compressed_page(page) &&
 187                             !PageError(page) && !fsverity_verify_page(page))
 188                                 SetPageError(page);
 189                 }
 190         } else {
 191                 fsverity_verify_bio(bio);
 192         }
 193
 194         f2fs_finish_read_bio(bio);
 195 }
 196
 197 /*
 198  * If the bio's data needs to be verified with fs-verity, then enqueue the
 199  * verity work for the bio.  Otherwise finish the bio now.
 200  *
 201  * Note that to avoid deadlocks, the verity work can't be done on the
 202  * decryption/decompression workqueue.  This is because verifying the data pages
 203  * can involve reading verity metadata pages from the file, and these verity
 204  * metadata pages may be encrypted and/or compressed.
 205  */
 206 static void f2fs_verify_and_finish_bio(struct bio *bio)
 207 {
 208         struct bio_post_read_ctx *ctx = bio->bi_private;
 209
 210         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 211                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 212                 fsverity_enqueue_verify_work(&ctx->work);
 213         } else {
 214                 f2fs_finish_read_bio(bio);
 215         }
 216 }
 217
 218 /*
 219  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 220  * remaining page was read by @ctx->bio.
 221  *
 222  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 223  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 224  * that the bio includes at least one compressed page.  The actual decompression
 225  * is done on a per-cluster basis, not a per-bio basis.
 226  */
 227 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
 228 {
 229         struct bio_vec *bv;
 230         struct bvec_iter_all iter_all;
 231         bool all_compressed = true;
 232         block_t blkaddr = ctx->fs_blkaddr;
 233
 234         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 235                 struct page *page = bv->bv_page;
 236
 237                 /* PG_error was set if decryption failed. */
 238                 if (f2fs_is_compressed_page(page))
 239                         f2fs_end_read_compressed_page(page, PageError(page),
 240                                                 blkaddr);
 241                 else
 242                         all_compressed = false;
 243
 244                 blkaddr++;
 245         }
 246
 247         /*
 248          * Optimization: if all the bio's pages are compressed, then scheduling
 249          * the per-bio verity work is unnecessary, as verity will be fully
 250          * handled at the compression cluster level.
 251          */
 252         if (all_compressed)
 253                 ctx->enabled_steps &= ~STEP_VERITY;
 254 }
 255
 256 static void f2fs_post_read_work(struct work_struct *work)
 257 {
 258         struct bio_post_read_ctx *ctx =
 259                 container_of(work, struct bio_post_read_ctx, work);
 260
 261         if (ctx->enabled_steps & STEP_DECRYPT)
 262                 fscrypt_decrypt_bio(ctx->bio);
 263
 264         if (ctx->enabled_steps & STEP_DECOMPRESS)
 265                 f2fs_handle_step_decompress(ctx);
 266
 267         f2fs_verify_and_finish_bio(ctx->bio);
 268 }
 269
 270 static void f2fs_read_end_io(struct bio *bio)
 271 {
 272         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 273         struct bio_post_read_ctx *ctx;
 274
 275         iostat_update_and_unbind_ctx(bio, 0);
 276         ctx = bio->bi_private;
 277
 278         if (time_to_inject(sbi, FAULT_READ_IO)) {
 279                 f2fs_show_injection_info(sbi, FAULT_READ_IO);
 280                 bio->bi_status = BLK_STS_IOERR;
 281         }
 282
 283         if (bio->bi_status) {
 284                 f2fs_finish_read_bio(bio);
 285                 return;
 286         }
 287
 288         if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
 289                 INIT_WORK(&ctx->work, f2fs_post_read_work);
 290                 queue_work(ctx->sbi->post_read_wq, &ctx->work);
 291         } else {
 292                 f2fs_verify_and_finish_bio(bio);
 293         }
 294 }
 295
 296 static void f2fs_write_end_io(struct bio *bio)
 297 {
 298         struct f2fs_sb_info *sbi;
 299         struct bio_vec *bvec;
 300         struct bvec_iter_all iter_all;
 301
 302         iostat_update_and_unbind_ctx(bio, 1);
 303         sbi = bio->bi_private;
 304
 305         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 306                 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
 307                 bio->bi_status = BLK_STS_IOERR;
 308         }
 309
 310         bio_for_each_segment_all(bvec, bio, iter_all) {
 311                 struct page *page = bvec->bv_page;
 312                 enum count_type type = WB_DATA_TYPE(page);
 313
 314                 if (page_private_dummy(page)) {
 315                         clear_page_private_dummy(page);
 316                         unlock_page(page);
 317                         mempool_free(page, sbi->write_io_dummy);
 318
 319                         if (unlikely(bio->bi_status))
 320                                 f2fs_stop_checkpoint(sbi, true);
 321                         continue;
 322                 }
 323
 324                 fscrypt_finalize_bounce_page(&page);
 325
 326 #ifdef CONFIG_F2FS_FS_COMPRESSION
 327                 if (f2fs_is_compressed_page(page)) {
 328                         f2fs_compress_write_end_io(bio, page);
 329                         continue;
 330                 }
 331 #endif
 332
 333                 if (unlikely(bio->bi_status)) {
 334                         mapping_set_error(page->mapping, -EIO);
 335                         if (type == F2FS_WB_CP_DATA)
 336                                 f2fs_stop_checkpoint(sbi, true);
 337                 }
 338
 339                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 340                                         page->index != nid_of_node(page));
 341
 342                 dec_page_count(sbi, type);
 343                 if (f2fs_in_warm_node_list(sbi, page))
 344                         f2fs_del_fsync_node_entry(sbi, page);
 345                 clear_page_private_gcing(page);
 346                 end_page_writeback(page);
 347         }
 348         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 349                                 wq_has_sleeper(&sbi->cp_wait))
 350                 wake_up(&sbi->cp_wait);
 351
 352         bio_put(bio);
 353 }
 354
 355 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 356                 block_t blk_addr, sector_t *sector)
 357 {
 358         struct block_device *bdev = sbi->sb->s_bdev;
 359         int i;
 360
 361         if (f2fs_is_multi_device(sbi)) {
 362                 for (i = 0; i < sbi->s_ndevs; i++) {
 363                         if (FDEV(i).start_blk <= blk_addr &&
 364                             FDEV(i).end_blk >= blk_addr) {
 365                                 blk_addr -= FDEV(i).start_blk;
 366                                 bdev = FDEV(i).bdev;
 367                                 break;
 368                         }
 369                 }
 370         }
 371
 372         if (sector)
 373                 *sector = SECTOR_FROM_BLOCK(blk_addr);
 374         return bdev;
 375 }
 376
 377 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 378 {
 379         int i;
 380
 381         if (!f2fs_is_multi_device(sbi))
 382                 return 0;
 383
 384         for (i = 0; i < sbi->s_ndevs; i++)
 385                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 386                         return i;
 387         return 0;
 388 }
 389
 390 static unsigned int f2fs_io_flags(struct f2fs_io_info *fio)
 391 {
 392         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 393         unsigned int fua_flag, meta_flag, io_flag;
 394         unsigned int op_flags = 0;
 395
 396         if (fio->op != REQ_OP_WRITE)
 397                 return 0;
 398         if (fio->type == DATA)
 399                 io_flag = fio->sbi->data_io_flag;
 400         else if (fio->type == NODE)
 401                 io_flag = fio->sbi->node_io_flag;
 402         else
 403                 return 0;
 404
 405         fua_flag = io_flag & temp_mask;
 406         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 407
 408         /*
 409          * data/node io flag bits per temp:
 410          *      REQ_META     |      REQ_FUA      |
 411          *    5 |    4 |   3 |    2 |    1 |   0 |
 412          * Cold | Warm | Hot | Cold | Warm | Hot |
 413          */
 414         if ((1 << fio->temp) & meta_flag)
 415                 op_flags |= REQ_META;
 416         if ((1 << fio->temp) & fua_flag)
 417                 op_flags |= REQ_FUA;
 418         return op_flags;
 419 }
 420
 421 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 422 {
 423         struct f2fs_sb_info *sbi = fio->sbi;
 424         struct block_device *bdev;
 425         sector_t sector;
 426         struct bio *bio;
 427
 428         bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 429         bio = bio_alloc_bioset(bdev, npages,
 430                                 fio->op | fio->op_flags | f2fs_io_flags(fio),
 431                                 GFP_NOIO, &f2fs_bioset);
 432         bio->bi_iter.bi_sector = sector;
 433         if (is_read_io(fio->op)) {
 434                 bio->bi_end_io = f2fs_read_end_io;
 435                 bio->bi_private = NULL;
 436         } else {
 437                 bio->bi_end_io = f2fs_write_end_io;
 438                 bio->bi_private = sbi;
 439         }
 440         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 441
 442         if (fio->io_wbc)
 443                 wbc_init_bio(fio->io_wbc, bio);
 444
 445         return bio;
 446 }
 447
 448 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 449                                   pgoff_t first_idx,
 450                                   const struct f2fs_io_info *fio,
 451                                   gfp_t gfp_mask)
 452 {
 453         /*
 454          * The f2fs garbage collector sets ->encrypted_page when it wants to
 455          * read/write raw data without encryption.
 456          */
 457         if (!fio || !fio->encrypted_page)
 458                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 459 }
 460
 461 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 462                                      pgoff_t next_idx,
 463                                      const struct f2fs_io_info *fio)
 464 {
 465         /*
 466          * The f2fs garbage collector sets ->encrypted_page when it wants to
 467          * read/write raw data without encryption.
 468          */
 469         if (fio && fio->encrypted_page)
 470                 return !bio_has_crypt_ctx(bio);
 471
 472         return fscrypt_mergeable_bio(bio, inode, next_idx);
 473 }
 474
 475 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 476                                 struct bio *bio, enum page_type type)
 477 {
 478         if (!is_read_io(bio_op(bio))) {
 479                 unsigned int start;
 480
 481                 if (type != DATA && type != NODE)
 482                         goto submit_io;
 483
 484                 if (f2fs_lfs_mode(sbi) && current->plug)
 485                         blk_finish_plug(current->plug);
 486
 487                 if (!F2FS_IO_ALIGNED(sbi))
 488                         goto submit_io;
 489
 490                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 491                 start %= F2FS_IO_SIZE(sbi);
 492
 493                 if (start == 0)
 494                         goto submit_io;
 495
 496                 /* fill dummy pages */
 497                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 498                         struct page *page =
 499                                 mempool_alloc(sbi->write_io_dummy,
 500                                               GFP_NOIO | __GFP_NOFAIL);
 501                         f2fs_bug_on(sbi, !page);
 502
 503                         lock_page(page);
 504
 505                         zero_user_segment(page, 0, PAGE_SIZE);
 506                         set_page_private_dummy(page);
 507
 508                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 509                                 f2fs_bug_on(sbi, 1);
 510                 }
 511                 /*
 512                  * In the NODE case, we lose next block address chain. So, we
 513                  * need to do checkpoint in f2fs_sync_file.
 514                  */
 515                 if (type == NODE)
 516                         set_sbi_flag(sbi, SBI_NEED_CP);
 517         }
 518 submit_io:
 519         if (is_read_io(bio_op(bio)))
 520                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 521         else
 522                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 523
 524         iostat_update_submit_ctx(bio, type);
 525         submit_bio(bio);
 526 }
 527
 528 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 529                                 struct bio *bio, enum page_type type)
 530 {
 531         __submit_bio(sbi, bio, type);
 532 }
 533
 534 static void __submit_merged_bio(struct f2fs_bio_info *io)
 535 {
 536         struct f2fs_io_info *fio = &io->fio;
 537
 538         if (!io->bio)
 539                 return;
 540
 541         if (is_read_io(fio->op))
 542                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 543         else
 544                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 545
 546         __submit_bio(io->sbi, io->bio, fio->type);
 547         io->bio = NULL;
 548 }
 549
 550 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 551                                                 struct page *page, nid_t ino)
 552 {
 553         struct bio_vec *bvec;
 554         struct bvec_iter_all iter_all;
 555
 556         if (!bio)
 557                 return false;
 558
 559         if (!inode && !page && !ino)
 560                 return true;
 561
 562         bio_for_each_segment_all(bvec, bio, iter_all) {
 563                 struct page *target = bvec->bv_page;
 564
 565                 if (fscrypt_is_bounce_page(target)) {
 566                         target = fscrypt_pagecache_page(target);
 567                         if (IS_ERR(target))
 568                                 continue;
 569                 }
 570                 if (f2fs_is_compressed_page(target)) {
 571                         target = f2fs_compress_control_page(target);
 572                         if (IS_ERR(target))
 573                                 continue;
 574                 }
 575
 576                 if (inode && inode == target->mapping->host)
 577                         return true;
 578                 if (page && page == target)
 579                         return true;
 580                 if (ino && ino == ino_of_node(target))
 581                         return true;
 582         }
 583
 584         return false;
 585 }
 586
 587 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
 588 {
 589         int i;
 590
 591         for (i = 0; i < NR_PAGE_TYPE; i++) {
 592                 int n = (i == META) ? 1 : NR_TEMP_TYPE;
 593                 int j;
 594
 595                 sbi->write_io[i] = f2fs_kmalloc(sbi,
 596                                 array_size(n, sizeof(struct f2fs_bio_info)),
 597                                 GFP_KERNEL);
 598                 if (!sbi->write_io[i])
 599                         return -ENOMEM;
 600
 601                 for (j = HOT; j < n; j++) {
 602                         init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
 603                         sbi->write_io[i][j].sbi = sbi;
 604                         sbi->write_io[i][j].bio = NULL;
 605                         spin_lock_init(&sbi->write_io[i][j].io_lock);
 606                         INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
 607                         INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
 608                         init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
 609                 }
 610         }
 611
 612         return 0;
 613 }
 614
 615 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 616                                 enum page_type type, enum temp_type temp)
 617 {
 618         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 619         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 620
 621         f2fs_down_write(&io->io_rwsem);
 622
 623         /* change META to META_FLUSH in the checkpoint procedure */
 624         if (type >= META_FLUSH) {
 625                 io->fio.type = META_FLUSH;
 626                 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 627                 if (!test_opt(sbi, NOBARRIER))
 628                         io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 629         }
 630         __submit_merged_bio(io);
 631         f2fs_up_write(&io->io_rwsem);
 632 }
 633
 634 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 635                                 struct inode *inode, struct page *page,
 636                                 nid_t ino, enum page_type type, bool force)
 637 {
 638         enum temp_type temp;
 639         bool ret = true;
 640
 641         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 642                 if (!force)     {
 643                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 644                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 645
 646                         f2fs_down_read(&io->io_rwsem);
 647                         ret = __has_merged_page(io->bio, inode, page, ino);
 648                         f2fs_up_read(&io->io_rwsem);
 649                 }
 650                 if (ret)
 651                         __f2fs_submit_merged_write(sbi, type, temp);
 652
 653                 /* TODO: use HOT temp only for meta pages now. */
 654                 if (type >= META)
 655                         break;
 656         }
 657 }
 658
 659 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 660 {
 661         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 662 }
 663
 664 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 665                                 struct inode *inode, struct page *page,
 666                                 nid_t ino, enum page_type type)
 667 {
 668         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 669 }
 670
 671 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 672 {
 673         f2fs_submit_merged_write(sbi, DATA);
 674         f2fs_submit_merged_write(sbi, NODE);
 675         f2fs_submit_merged_write(sbi, META);
 676 }
 677
 678 /*
 679  * Fill the locked page with data located in the block address.
 680  * A caller needs to unlock the page on failure.
 681  */
 682 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 683 {
 684         struct bio *bio;
 685         struct page *page = fio->encrypted_page ?
 686                         fio->encrypted_page : fio->page;
 687
 688         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 689                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 690                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 691                 return -EFSCORRUPTED;
 692
 693         trace_f2fs_submit_page_bio(page, fio);
 694
 695         /* Allocate a new bio */
 696         bio = __bio_alloc(fio, 1);
 697
 698         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 699                                fio->page->index, fio, GFP_NOIO);
 700
 701         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 702                 bio_put(bio);
 703                 return -EFAULT;
 704         }
 705
 706         if (fio->io_wbc && !is_read_io(fio->op))
 707                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 708
 709         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 710                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 711
 712         __submit_bio(fio->sbi, bio, fio->type);
 713         return 0;
 714 }
 715
 716 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 717                                 block_t last_blkaddr, block_t cur_blkaddr)
 718 {
 719         if (unlikely(sbi->max_io_bytes &&
 720                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 721                 return false;
 722         if (last_blkaddr + 1 != cur_blkaddr)
 723                 return false;
 724         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 725 }
 726
 727 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 728                                                 struct f2fs_io_info *fio)
 729 {
 730         if (io->fio.op != fio->op)
 731                 return false;
 732         return io->fio.op_flags == fio->op_flags;
 733 }
 734
 735 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 736                                         struct f2fs_bio_info *io,
 737                                         struct f2fs_io_info *fio,
 738                                         block_t last_blkaddr,
 739                                         block_t cur_blkaddr)
 740 {
 741         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 742                 unsigned int filled_blocks =
 743                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 744                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 745                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 746
 747                 /* IOs in bio is aligned and left space of vectors is not enough */
 748                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 749                         return false;
 750         }
 751         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 752                 return false;
 753         return io_type_is_mergeable(io, fio);
 754 }
 755
 756 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 757                                 struct page *page, enum temp_type temp)
 758 {
 759         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 760         struct bio_entry *be;
 761
 762         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 763         be->bio = bio;
 764         bio_get(bio);
 765
 766         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 767                 f2fs_bug_on(sbi, 1);
 768
 769         f2fs_down_write(&io->bio_list_lock);
 770         list_add_tail(&be->list, &io->bio_list);
 771         f2fs_up_write(&io->bio_list_lock);
 772 }
 773
 774 static void del_bio_entry(struct bio_entry *be)
 775 {
 776         list_del(&be->list);
 777         kmem_cache_free(bio_entry_slab, be);
 778 }
 779
 780 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 781                                                         struct page *page)
 782 {
 783         struct f2fs_sb_info *sbi = fio->sbi;
 784         enum temp_type temp;
 785         bool found = false;
 786         int ret = -EAGAIN;
 787
 788         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 789                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 790                 struct list_head *head = &io->bio_list;
 791                 struct bio_entry *be;
 792
 793                 f2fs_down_write(&io->bio_list_lock);
 794                 list_for_each_entry(be, head, list) {
 795                         if (be->bio != *bio)
 796                                 continue;
 797
 798                         found = true;
 799
 800                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 801                                                             *fio->last_block,
 802                                                             fio->new_blkaddr));
 803                         if (f2fs_crypt_mergeable_bio(*bio,
 804                                         fio->page->mapping->host,
 805                                         fio->page->index, fio) &&
 806                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 807                                         PAGE_SIZE) {
 808                                 ret = 0;
 809                                 break;
 810                         }
 811
 812                         /* page can't be merged into bio; submit the bio */
 813                         del_bio_entry(be);
 814                         __submit_bio(sbi, *bio, DATA);
 815                         break;
 816                 }
 817                 f2fs_up_write(&io->bio_list_lock);
 818         }
 819
 820         if (ret) {
 821                 bio_put(*bio);
 822                 *bio = NULL;
 823         }
 824
 825         return ret;
 826 }
 827
 828 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 829                                         struct bio **bio, struct page *page)
 830 {
 831         enum temp_type temp;
 832         bool found = false;
 833         struct bio *target = bio ? *bio : NULL;
 834
 835         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 836                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 837                 struct list_head *head = &io->bio_list;
 838                 struct bio_entry *be;
 839
 840                 if (list_empty(head))
 841                         continue;
 842
 843                 f2fs_down_read(&io->bio_list_lock);
 844                 list_for_each_entry(be, head, list) {
 845                         if (target)
 846                                 found = (target == be->bio);
 847                         else
 848                                 found = __has_merged_page(be->bio, NULL,
 849                                                                 page, 0);
 850                         if (found)
 851                                 break;
 852                 }
 853                 f2fs_up_read(&io->bio_list_lock);
 854
 855                 if (!found)
 856                         continue;
 857
 858                 found = false;
 859
 860                 f2fs_down_write(&io->bio_list_lock);
 861                 list_for_each_entry(be, head, list) {
 862                         if (target)
 863                                 found = (target == be->bio);
 864                         else
 865                                 found = __has_merged_page(be->bio, NULL,
 866                                                                 page, 0);
 867                         if (found) {
 868                                 target = be->bio;
 869                                 del_bio_entry(be);
 870                                 break;
 871                         }
 872                 }
 873                 f2fs_up_write(&io->bio_list_lock);
 874         }
 875
 876         if (found)
 877                 __submit_bio(sbi, target, DATA);
 878         if (bio && *bio) {
 879                 bio_put(*bio);
 880                 *bio = NULL;
 881         }
 882 }
 883
 884 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 885 {
 886         struct bio *bio = *fio->bio;
 887         struct page *page = fio->encrypted_page ?
 888                         fio->encrypted_page : fio->page;
 889
 890         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 891                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 892                 return -EFSCORRUPTED;
 893
 894         trace_f2fs_submit_page_bio(page, fio);
 895
 896         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 897                                                 fio->new_blkaddr))
 898                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 899 alloc_new:
 900         if (!bio) {
 901                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 902                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 903                                        fio->page->index, fio, GFP_NOIO);
 904
 905                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 906         } else {
 907                 if (add_ipu_page(fio, &bio, page))
 908                         goto alloc_new;
 909         }
 910
 911         if (fio->io_wbc)
 912                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 913
 914         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 915
 916         *fio->last_block = fio->new_blkaddr;
 917         *fio->bio = bio;
 918
 919         return 0;
 920 }
 921
 922 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 923 {
 924         struct f2fs_sb_info *sbi = fio->sbi;
 925         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 926         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 927         struct page *bio_page;
 928
 929         f2fs_bug_on(sbi, is_read_io(fio->op));
 930
 931         f2fs_down_write(&io->io_rwsem);
 932 next:
 933         if (fio->in_list) {
 934                 spin_lock(&io->io_lock);
 935                 if (list_empty(&io->io_list)) {
 936                         spin_unlock(&io->io_lock);
 937                         goto out;
 938                 }
 939                 fio = list_first_entry(&io->io_list,
 940                                                 struct f2fs_io_info, list);
 941                 list_del(&fio->list);
 942                 spin_unlock(&io->io_lock);
 943         }
 944
 945         verify_fio_blkaddr(fio);
 946
 947         if (fio->encrypted_page)
 948                 bio_page = fio->encrypted_page;
 949         else if (fio->compressed_page)
 950                 bio_page = fio->compressed_page;
 951         else
 952                 bio_page = fio->page;
 953
 954         /* set submitted = true as a return value */
 955         fio->submitted = true;
 956
 957         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 958
 959         if (io->bio &&
 960             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 961                               fio->new_blkaddr) ||
 962              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 963                                        bio_page->index, fio)))
 964                 __submit_merged_bio(io);
 965 alloc_new:
 966         if (io->bio == NULL) {
 967                 if (F2FS_IO_ALIGNED(sbi) &&
 968                                 (fio->type == DATA || fio->type == NODE) &&
 969                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 970                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 971                         fio->retry = true;
 972                         goto skip;
 973                 }
 974                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
 975                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 976                                        bio_page->index, fio, GFP_NOIO);
 977                 io->fio = *fio;
 978         }
 979
 980         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 981                 __submit_merged_bio(io);
 982                 goto alloc_new;
 983         }
 984
 985         if (fio->io_wbc)
 986                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
 987
 988         io->last_block_in_bio = fio->new_blkaddr;
 989
 990         trace_f2fs_submit_page_write(fio->page, fio);
 991 skip:
 992         if (fio->in_list)
 993                 goto next;
 994 out:
 995         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 996                                 !f2fs_is_checkpoint_ready(sbi))
 997                 __submit_merged_bio(io);
 998         f2fs_up_write(&io->io_rwsem);
 999 }
1000
1001 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1002                                       unsigned nr_pages, unsigned op_flag,
1003                                       pgoff_t first_idx, bool for_write)
1004 {
1005         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1006         struct bio *bio;
1007         struct bio_post_read_ctx *ctx = NULL;
1008         unsigned int post_read_steps = 0;
1009         sector_t sector;
1010         struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
1011
1012         bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1013                                REQ_OP_READ | op_flag,
1014                                for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1015         if (!bio)
1016                 return ERR_PTR(-ENOMEM);
1017         bio->bi_iter.bi_sector = sector;
1018         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1019         bio->bi_end_io = f2fs_read_end_io;
1020
1021         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1022                 post_read_steps |= STEP_DECRYPT;
1023
1024         if (f2fs_need_verity(inode, first_idx))
1025                 post_read_steps |= STEP_VERITY;
1026
1027         /*
1028          * STEP_DECOMPRESS is handled specially, since a compressed file might
1029          * contain both compressed and uncompressed clusters.  We'll allocate a
1030          * bio_post_read_ctx if the file is compressed, but the caller is
1031          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1032          */
1033
1034         if (post_read_steps || f2fs_compressed_file(inode)) {
1035                 /* Due to the mempool, this never fails. */
1036                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1037                 ctx->bio = bio;
1038                 ctx->sbi = sbi;
1039                 ctx->enabled_steps = post_read_steps;
1040                 ctx->fs_blkaddr = blkaddr;
1041                 bio->bi_private = ctx;
1042         }
1043         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1044
1045         return bio;
1046 }
1047
1048 /* This can handle encryption stuffs */
1049 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1050                                  block_t blkaddr, int op_flags, bool for_write)
1051 {
1052         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1053         struct bio *bio;
1054
1055         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1056                                         page->index, for_write);
1057         if (IS_ERR(bio))
1058                 return PTR_ERR(bio);
1059
1060         /* wait for GCed page writeback via META_MAPPING */
1061         f2fs_wait_on_block_writeback(inode, blkaddr);
1062
1063         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1064                 bio_put(bio);
1065                 return -EFAULT;
1066         }
1067         ClearPageError(page);
1068         inc_page_count(sbi, F2FS_RD_DATA);
1069         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1070         __submit_bio(sbi, bio, DATA);
1071         return 0;
1072 }
1073
1074 static void __set_data_blkaddr(struct dnode_of_data *dn)
1075 {
1076         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1077         __le32 *addr_array;
1078         int base = 0;
1079
1080         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1081                 base = get_extra_isize(dn->inode);
1082
1083         /* Get physical address of data block */
1084         addr_array = blkaddr_in_node(rn);
1085         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1086 }
1087
1088 /*
1089  * Lock ordering for the change of data block address:
1090  * ->data_page
1091  *  ->node_page
1092  *    update block addresses in the node page
1093  */
1094 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1095 {
1096         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1097         __set_data_blkaddr(dn);
1098         if (set_page_dirty(dn->node_page))
1099                 dn->node_changed = true;
1100 }
1101
1102 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1103 {
1104         dn->data_blkaddr = blkaddr;
1105         f2fs_set_data_blkaddr(dn);
1106         f2fs_update_extent_cache(dn);
1107 }
1108
1109 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1110 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1111 {
1112         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1113         int err;
1114
1115         if (!count)
1116                 return 0;
1117
1118         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1119                 return -EPERM;
1120         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1121                 return err;
1122
1123         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1124                                                 dn->ofs_in_node, count);
1125
1126         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1127
1128         for (; count > 0; dn->ofs_in_node++) {
1129                 block_t blkaddr = f2fs_data_blkaddr(dn);
1130
1131                 if (blkaddr == NULL_ADDR) {
1132                         dn->data_blkaddr = NEW_ADDR;
1133                         __set_data_blkaddr(dn);
1134                         count--;
1135                 }
1136         }
1137
1138         if (set_page_dirty(dn->node_page))
1139                 dn->node_changed = true;
1140         return 0;
1141 }
1142
1143 /* Should keep dn->ofs_in_node unchanged */
1144 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1145 {
1146         unsigned int ofs_in_node = dn->ofs_in_node;
1147         int ret;
1148
1149         ret = f2fs_reserve_new_blocks(dn, 1);
1150         dn->ofs_in_node = ofs_in_node;
1151         return ret;
1152 }
1153
1154 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1155 {
1156         bool need_put = dn->inode_page ? false : true;
1157         int err;
1158
1159         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1160         if (err)
1161                 return err;
1162
1163         if (dn->data_blkaddr == NULL_ADDR)
1164                 err = f2fs_reserve_new_block(dn);
1165         if (err || need_put)
1166                 f2fs_put_dnode(dn);
1167         return err;
1168 }
1169
1170 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1171 {
1172         struct extent_info ei = {0, };
1173         struct inode *inode = dn->inode;
1174
1175         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1176                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1177                 return 0;
1178         }
1179
1180         return f2fs_reserve_block(dn, index);
1181 }
1182
1183 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1184                                                 int op_flags, bool for_write)
1185 {
1186         struct address_space *mapping = inode->i_mapping;
1187         struct dnode_of_data dn;
1188         struct page *page;
1189         struct extent_info ei = {0, };
1190         int err;
1191
1192         page = f2fs_grab_cache_page(mapping, index, for_write);
1193         if (!page)
1194                 return ERR_PTR(-ENOMEM);
1195
1196         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1197                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1198                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1199                                                 DATA_GENERIC_ENHANCE_READ)) {
1200                         err = -EFSCORRUPTED;
1201                         goto put_err;
1202                 }
1203                 goto got_it;
1204         }
1205
1206         set_new_dnode(&dn, inode, NULL, NULL, 0);
1207         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1208         if (err)
1209                 goto put_err;
1210         f2fs_put_dnode(&dn);
1211
1212         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1213                 err = -ENOENT;
1214                 goto put_err;
1215         }
1216         if (dn.data_blkaddr != NEW_ADDR &&
1217                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1218                                                 dn.data_blkaddr,
1219                                                 DATA_GENERIC_ENHANCE)) {
1220                 err = -EFSCORRUPTED;
1221                 goto put_err;
1222         }
1223 got_it:
1224         if (PageUptodate(page)) {
1225                 unlock_page(page);
1226                 return page;
1227         }
1228
1229         /*
1230          * A new dentry page is allocated but not able to be written, since its
1231          * new inode page couldn't be allocated due to -ENOSPC.
1232          * In such the case, its blkaddr can be remained as NEW_ADDR.
1233          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1234          * f2fs_init_inode_metadata.
1235          */
1236         if (dn.data_blkaddr == NEW_ADDR) {
1237                 zero_user_segment(page, 0, PAGE_SIZE);
1238                 if (!PageUptodate(page))
1239                         SetPageUptodate(page);
1240                 unlock_page(page);
1241                 return page;
1242         }
1243
1244         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1245                                                 op_flags, for_write);
1246         if (err)
1247                 goto put_err;
1248         return page;
1249
1250 put_err:
1251         f2fs_put_page(page, 1);
1252         return ERR_PTR(err);
1253 }
1254
1255 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1256 {
1257         struct address_space *mapping = inode->i_mapping;
1258         struct page *page;
1259
1260         page = find_get_page(mapping, index);
1261         if (page && PageUptodate(page))
1262                 return page;
1263         f2fs_put_page(page, 0);
1264
1265         page = f2fs_get_read_data_page(inode, index, 0, false);
1266         if (IS_ERR(page))
1267                 return page;
1268
1269         if (PageUptodate(page))
1270                 return page;
1271
1272         wait_on_page_locked(page);
1273         if (unlikely(!PageUptodate(page))) {
1274                 f2fs_put_page(page, 0);
1275                 return ERR_PTR(-EIO);
1276         }
1277         return page;
1278 }
1279
1280 /*
1281  * If it tries to access a hole, return an error.
1282  * Because, the callers, functions in dir.c and GC, should be able to know
1283  * whether this page exists or not.
1284  */
1285 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1286                                                         bool for_write)
1287 {
1288         struct address_space *mapping = inode->i_mapping;
1289         struct page *page;
1290 repeat:
1291         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1292         if (IS_ERR(page))
1293                 return page;
1294
1295         /* wait for read completion */
1296         lock_page(page);
1297         if (unlikely(page->mapping != mapping)) {
1298                 f2fs_put_page(page, 1);
1299                 goto repeat;
1300         }
1301         if (unlikely(!PageUptodate(page))) {
1302                 f2fs_put_page(page, 1);
1303                 return ERR_PTR(-EIO);
1304         }
1305         return page;
1306 }
1307
1308 /*
1309  * Caller ensures that this data page is never allocated.
1310  * A new zero-filled data page is allocated in the page cache.
1311  *
1312  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1313  * f2fs_unlock_op().
1314  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1315  * ipage should be released by this function.
1316  */
1317 struct page *f2fs_get_new_data_page(struct inode *inode,
1318                 struct page *ipage, pgoff_t index, bool new_i_size)
1319 {
1320         struct address_space *mapping = inode->i_mapping;
1321         struct page *page;
1322         struct dnode_of_data dn;
1323         int err;
1324
1325         page = f2fs_grab_cache_page(mapping, index, true);
1326         if (!page) {
1327                 /*
1328                  * before exiting, we should make sure ipage will be released
1329                  * if any error occur.
1330                  */
1331                 f2fs_put_page(ipage, 1);
1332                 return ERR_PTR(-ENOMEM);
1333         }
1334
1335         set_new_dnode(&dn, inode, ipage, NULL, 0);
1336         err = f2fs_reserve_block(&dn, index);
1337         if (err) {
1338                 f2fs_put_page(page, 1);
1339                 return ERR_PTR(err);
1340         }
1341         if (!ipage)
1342                 f2fs_put_dnode(&dn);
1343
1344         if (PageUptodate(page))
1345                 goto got_it;
1346
1347         if (dn.data_blkaddr == NEW_ADDR) {
1348                 zero_user_segment(page, 0, PAGE_SIZE);
1349                 if (!PageUptodate(page))
1350                         SetPageUptodate(page);
1351         } else {
1352                 f2fs_put_page(page, 1);
1353
1354                 /* if ipage exists, blkaddr should be NEW_ADDR */
1355                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1356                 page = f2fs_get_lock_data_page(inode, index, true);
1357                 if (IS_ERR(page))
1358                         return page;
1359         }
1360 got_it:
1361         if (new_i_size && i_size_read(inode) <
1362                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1363                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1364         return page;
1365 }
1366
1367 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1368 {
1369         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1370         struct f2fs_summary sum;
1371         struct node_info ni;
1372         block_t old_blkaddr;
1373         blkcnt_t count = 1;
1374         int err;
1375
1376         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1377                 return -EPERM;
1378
1379         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1380         if (err)
1381                 return err;
1382
1383         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1384         if (dn->data_blkaddr != NULL_ADDR)
1385                 goto alloc;
1386
1387         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1388                 return err;
1389
1390 alloc:
1391         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1392         old_blkaddr = dn->data_blkaddr;
1393         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1394                                 &sum, seg_type, NULL);
1395         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1396                 invalidate_mapping_pages(META_MAPPING(sbi),
1397                                         old_blkaddr, old_blkaddr);
1398                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1399         }
1400         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1401         return 0;
1402 }
1403
1404 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1405 {
1406         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1407                 if (lock)
1408                         f2fs_down_read(&sbi->node_change);
1409                 else
1410                         f2fs_up_read(&sbi->node_change);
1411         } else {
1412                 if (lock)
1413                         f2fs_lock_op(sbi);
1414                 else
1415                         f2fs_unlock_op(sbi);
1416         }
1417 }
1418
1419 /*
1420  * f2fs_map_blocks() tries to find or build mapping relationship which
1421  * maps continuous logical blocks to physical blocks, and return such
1422  * info via f2fs_map_blocks structure.
1423  */
1424 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1425                                                 int create, int flag)
1426 {
1427         unsigned int maxblocks = map->m_len;
1428         struct dnode_of_data dn;
1429         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1430         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1431         pgoff_t pgofs, end_offset, end;
1432         int err = 0, ofs = 1;
1433         unsigned int ofs_in_node, last_ofs_in_node;
1434         blkcnt_t prealloc;
1435         struct extent_info ei = {0, };
1436         block_t blkaddr;
1437         unsigned int start_pgofs;
1438         int bidx = 0;
1439
1440         if (!maxblocks)
1441                 return 0;
1442
1443         map->m_bdev = inode->i_sb->s_bdev;
1444         map->m_multidev_dio =
1445                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1446
1447         map->m_len = 0;
1448         map->m_flags = 0;
1449
1450         /* it only supports block size == page size */
1451         pgofs = (pgoff_t)map->m_lblk;
1452         end = pgofs + maxblocks;
1453
1454         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1455                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1456                                                         map->m_may_create)
1457                         goto next_dnode;
1458
1459                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1460                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1461                 map->m_flags = F2FS_MAP_MAPPED;
1462                 if (map->m_next_extent)
1463                         *map->m_next_extent = pgofs + map->m_len;
1464
1465                 /* for hardware encryption, but to avoid potential issue in future */
1466                 if (flag == F2FS_GET_BLOCK_DIO)
1467                         f2fs_wait_on_block_writeback_range(inode,
1468                                                 map->m_pblk, map->m_len);
1469
1470                 if (map->m_multidev_dio) {
1471                         block_t blk_addr = map->m_pblk;
1472
1473                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1474
1475                         map->m_bdev = FDEV(bidx).bdev;
1476                         map->m_pblk -= FDEV(bidx).start_blk;
1477                         map->m_len = min(map->m_len,
1478                                 FDEV(bidx).end_blk + 1 - map->m_pblk);
1479
1480                         if (map->m_may_create)
1481                                 f2fs_update_device_state(sbi, inode->i_ino,
1482                                                         blk_addr, map->m_len);
1483                 }
1484                 goto out;
1485         }
1486
1487 next_dnode:
1488         if (map->m_may_create)
1489                 f2fs_do_map_lock(sbi, flag, true);
1490
1491         /* When reading holes, we need its node page */
1492         set_new_dnode(&dn, inode, NULL, NULL, 0);
1493         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1494         if (err) {
1495                 if (flag == F2FS_GET_BLOCK_BMAP)
1496                         map->m_pblk = 0;
1497
1498                 if (err == -ENOENT) {
1499                         /*
1500                          * There is one exceptional case that read_node_page()
1501                          * may return -ENOENT due to filesystem has been
1502                          * shutdown or cp_error, so force to convert error
1503                          * number to EIO for such case.
1504                          */
1505                         if (map->m_may_create &&
1506                                 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1507                                 f2fs_cp_error(sbi))) {
1508                                 err = -EIO;
1509                                 goto unlock_out;
1510                         }
1511
1512                         err = 0;
1513                         if (map->m_next_pgofs)
1514                                 *map->m_next_pgofs =
1515                                         f2fs_get_next_page_offset(&dn, pgofs);
1516                         if (map->m_next_extent)
1517                                 *map->m_next_extent =
1518                                         f2fs_get_next_page_offset(&dn, pgofs);
1519                 }
1520                 goto unlock_out;
1521         }
1522
1523         start_pgofs = pgofs;
1524         prealloc = 0;
1525         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1526         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1527
1528 next_block:
1529         blkaddr = f2fs_data_blkaddr(&dn);
1530
1531         if (__is_valid_data_blkaddr(blkaddr) &&
1532                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1533                 err = -EFSCORRUPTED;
1534                 goto sync_out;
1535         }
1536
1537         if (__is_valid_data_blkaddr(blkaddr)) {
1538                 /* use out-place-update for driect IO under LFS mode */
1539                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1540                                                         map->m_may_create) {
1541                         err = __allocate_data_block(&dn, map->m_seg_type);
1542                         if (err)
1543                                 goto sync_out;
1544                         blkaddr = dn.data_blkaddr;
1545                         set_inode_flag(inode, FI_APPEND_WRITE);
1546                 }
1547         } else {
1548                 if (create) {
1549                         if (unlikely(f2fs_cp_error(sbi))) {
1550                                 err = -EIO;
1551                                 goto sync_out;
1552                         }
1553                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1554                                 if (blkaddr == NULL_ADDR) {
1555                                         prealloc++;
1556                                         last_ofs_in_node = dn.ofs_in_node;
1557                                 }
1558                         } else {
1559                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1560                                         flag != F2FS_GET_BLOCK_DIO);
1561                                 err = __allocate_data_block(&dn,
1562                                                         map->m_seg_type);
1563                                 if (!err) {
1564                                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1565                                                 file_need_truncate(inode);
1566                                         set_inode_flag(inode, FI_APPEND_WRITE);
1567                                 }
1568                         }
1569                         if (err)
1570                                 goto sync_out;
1571                         map->m_flags |= F2FS_MAP_NEW;
1572                         blkaddr = dn.data_blkaddr;
1573                 } else {
1574                         if (f2fs_compressed_file(inode) &&
1575                                         f2fs_sanity_check_cluster(&dn) &&
1576                                         (flag != F2FS_GET_BLOCK_FIEMAP ||
1577                                         IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1578                                 err = -EFSCORRUPTED;
1579                                 goto sync_out;
1580                         }
1581                         if (flag == F2FS_GET_BLOCK_BMAP) {
1582                                 map->m_pblk = 0;
1583                                 goto sync_out;
1584                         }
1585                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1586                                 goto sync_out;
1587                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1588                                                 blkaddr == NULL_ADDR) {
1589                                 if (map->m_next_pgofs)
1590                                         *map->m_next_pgofs = pgofs + 1;
1591                                 goto sync_out;
1592                         }
1593                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1594                                 /* for defragment case */
1595                                 if (map->m_next_pgofs)
1596                                         *map->m_next_pgofs = pgofs + 1;
1597                                 goto sync_out;
1598                         }
1599                 }
1600         }
1601
1602         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1603                 goto skip;
1604
1605         if (map->m_multidev_dio)
1606                 bidx = f2fs_target_device_index(sbi, blkaddr);
1607
1608         if (map->m_len == 0) {
1609                 /* preallocated unwritten block should be mapped for fiemap. */
1610                 if (blkaddr == NEW_ADDR)
1611                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1612                 map->m_flags |= F2FS_MAP_MAPPED;
1613
1614                 map->m_pblk = blkaddr;
1615                 map->m_len = 1;
1616
1617                 if (map->m_multidev_dio)
1618                         map->m_bdev = FDEV(bidx).bdev;
1619         } else if ((map->m_pblk != NEW_ADDR &&
1620                         blkaddr == (map->m_pblk + ofs)) ||
1621                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1622                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1623                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1624                         goto sync_out;
1625                 ofs++;
1626                 map->m_len++;
1627         } else {
1628                 goto sync_out;
1629         }
1630
1631 skip:
1632         dn.ofs_in_node++;
1633         pgofs++;
1634
1635         /* preallocate blocks in batch for one dnode page */
1636         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1637                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1638
1639                 dn.ofs_in_node = ofs_in_node;
1640                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1641                 if (err)
1642                         goto sync_out;
1643
1644                 map->m_len += dn.ofs_in_node - ofs_in_node;
1645                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1646                         err = -ENOSPC;
1647                         goto sync_out;
1648                 }
1649                 dn.ofs_in_node = end_offset;
1650         }
1651
1652         if (pgofs >= end)
1653                 goto sync_out;
1654         else if (dn.ofs_in_node < end_offset)
1655                 goto next_block;
1656
1657         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1658                 if (map->m_flags & F2FS_MAP_MAPPED) {
1659                         unsigned int ofs = start_pgofs - map->m_lblk;
1660
1661                         f2fs_update_extent_cache_range(&dn,
1662                                 start_pgofs, map->m_pblk + ofs,
1663                                 map->m_len - ofs);
1664                 }
1665         }
1666
1667         f2fs_put_dnode(&dn);
1668
1669         if (map->m_may_create) {
1670                 f2fs_do_map_lock(sbi, flag, false);
1671                 f2fs_balance_fs(sbi, dn.node_changed);
1672         }
1673         goto next_dnode;
1674
1675 sync_out:
1676
1677         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1678                 /*
1679                  * for hardware encryption, but to avoid potential issue
1680                  * in future
1681                  */
1682                 f2fs_wait_on_block_writeback_range(inode,
1683                                                 map->m_pblk, map->m_len);
1684                 invalidate_mapping_pages(META_MAPPING(sbi),
1685                                                 map->m_pblk, map->m_pblk);
1686
1687                 if (map->m_multidev_dio) {
1688                         block_t blk_addr = map->m_pblk;
1689
1690                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1691
1692                         map->m_bdev = FDEV(bidx).bdev;
1693                         map->m_pblk -= FDEV(bidx).start_blk;
1694
1695                         if (map->m_may_create)
1696                                 f2fs_update_device_state(sbi, inode->i_ino,
1697                                                         blk_addr, map->m_len);
1698
1699                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1700                                                 FDEV(bidx).end_blk + 1);
1701                 }
1702         }
1703
1704         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1705                 if (map->m_flags & F2FS_MAP_MAPPED) {
1706                         unsigned int ofs = start_pgofs - map->m_lblk;
1707
1708                         f2fs_update_extent_cache_range(&dn,
1709                                 start_pgofs, map->m_pblk + ofs,
1710                                 map->m_len - ofs);
1711                 }
1712                 if (map->m_next_extent)
1713                         *map->m_next_extent = pgofs + 1;
1714         }
1715         f2fs_put_dnode(&dn);
1716 unlock_out:
1717         if (map->m_may_create) {
1718                 f2fs_do_map_lock(sbi, flag, false);
1719                 f2fs_balance_fs(sbi, dn.node_changed);
1720         }
1721 out:
1722         trace_f2fs_map_blocks(inode, map, create, flag, err);
1723         return err;
1724 }
1725
1726 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1727 {
1728         struct f2fs_map_blocks map;
1729         block_t last_lblk;
1730         int err;
1731
1732         if (pos + len > i_size_read(inode))
1733                 return false;
1734
1735         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1736         map.m_next_pgofs = NULL;
1737         map.m_next_extent = NULL;
1738         map.m_seg_type = NO_CHECK_TYPE;
1739         map.m_may_create = false;
1740         last_lblk = F2FS_BLK_ALIGN(pos + len);
1741
1742         while (map.m_lblk < last_lblk) {
1743                 map.m_len = last_lblk - map.m_lblk;
1744                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1745                 if (err || map.m_len == 0)
1746                         return false;
1747                 map.m_lblk += map.m_len;
1748         }
1749         return true;
1750 }
1751
1752 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1753 {
1754         return (bytes >> inode->i_blkbits);
1755 }
1756
1757 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1758 {
1759         return (blks << inode->i_blkbits);
1760 }
1761
1762 static int f2fs_xattr_fiemap(struct inode *inode,
1763                                 struct fiemap_extent_info *fieinfo)
1764 {
1765         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1766         struct page *page;
1767         struct node_info ni;
1768         __u64 phys = 0, len;
1769         __u32 flags;
1770         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1771         int err = 0;
1772
1773         if (f2fs_has_inline_xattr(inode)) {
1774                 int offset;
1775
1776                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1777                                                 inode->i_ino, false);
1778                 if (!page)
1779                         return -ENOMEM;
1780
1781                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1782                 if (err) {
1783                         f2fs_put_page(page, 1);
1784                         return err;
1785                 }
1786
1787                 phys = blks_to_bytes(inode, ni.blk_addr);
1788                 offset = offsetof(struct f2fs_inode, i_addr) +
1789                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1790                                         get_inline_xattr_addrs(inode));
1791
1792                 phys += offset;
1793                 len = inline_xattr_size(inode);
1794
1795                 f2fs_put_page(page, 1);
1796
1797                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1798
1799                 if (!xnid)
1800                         flags |= FIEMAP_EXTENT_LAST;
1801
1802                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1803                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1804                 if (err || err == 1)
1805                         return err;
1806         }
1807
1808         if (xnid) {
1809                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1810                 if (!page)
1811                         return -ENOMEM;
1812
1813                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1814                 if (err) {
1815                         f2fs_put_page(page, 1);
1816                         return err;
1817                 }
1818
1819                 phys = blks_to_bytes(inode, ni.blk_addr);
1820                 len = inode->i_sb->s_blocksize;
1821
1822                 f2fs_put_page(page, 1);
1823
1824                 flags = FIEMAP_EXTENT_LAST;
1825         }
1826
1827         if (phys) {
1828                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1829                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1830         }
1831
1832         return (err < 0 ? err : 0);
1833 }
1834
1835 static loff_t max_inode_blocks(struct inode *inode)
1836 {
1837         loff_t result = ADDRS_PER_INODE(inode);
1838         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1839
1840         /* two direct node blocks */
1841         result += (leaf_count * 2);
1842
1843         /* two indirect node blocks */
1844         leaf_count *= NIDS_PER_BLOCK;
1845         result += (leaf_count * 2);
1846
1847         /* one double indirect node block */
1848         leaf_count *= NIDS_PER_BLOCK;
1849         result += leaf_count;
1850
1851         return result;
1852 }
1853
1854 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1855                 u64 start, u64 len)
1856 {
1857         struct f2fs_map_blocks map;
1858         sector_t start_blk, last_blk;
1859         pgoff_t next_pgofs;
1860         u64 logical = 0, phys = 0, size = 0;
1861         u32 flags = 0;
1862         int ret = 0;
1863         bool compr_cluster = false, compr_appended;
1864         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1865         unsigned int count_in_cluster = 0;
1866         loff_t maxbytes;
1867
1868         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1869                 ret = f2fs_precache_extents(inode);
1870                 if (ret)
1871                         return ret;
1872         }
1873
1874         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1875         if (ret)
1876                 return ret;
1877
1878         inode_lock(inode);
1879
1880         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1881         if (start > maxbytes) {
1882                 ret = -EFBIG;
1883                 goto out;
1884         }
1885
1886         if (len > maxbytes || (maxbytes - len) < start)
1887                 len = maxbytes - start;
1888
1889         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1890                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1891                 goto out;
1892         }
1893
1894         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1895                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1896                 if (ret != -EAGAIN)
1897                         goto out;
1898         }
1899
1900         if (bytes_to_blks(inode, len) == 0)
1901                 len = blks_to_bytes(inode, 1);
1902
1903         start_blk = bytes_to_blks(inode, start);
1904         last_blk = bytes_to_blks(inode, start + len - 1);
1905
1906 next:
1907         memset(&map, 0, sizeof(map));
1908         map.m_lblk = start_blk;
1909         map.m_len = bytes_to_blks(inode, len);
1910         map.m_next_pgofs = &next_pgofs;
1911         map.m_seg_type = NO_CHECK_TYPE;
1912
1913         if (compr_cluster) {
1914                 map.m_lblk += 1;
1915                 map.m_len = cluster_size - count_in_cluster;
1916         }
1917
1918         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1919         if (ret)
1920                 goto out;
1921
1922         /* HOLE */
1923         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1924                 start_blk = next_pgofs;
1925
1926                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1927                                                 max_inode_blocks(inode)))
1928                         goto prep_next;
1929
1930                 flags |= FIEMAP_EXTENT_LAST;
1931         }
1932
1933         compr_appended = false;
1934         /* In a case of compressed cluster, append this to the last extent */
1935         if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1936                         !(map.m_flags & F2FS_MAP_FLAGS))) {
1937                 compr_appended = true;
1938                 goto skip_fill;
1939         }
1940
1941         if (size) {
1942                 flags |= FIEMAP_EXTENT_MERGED;
1943                 if (IS_ENCRYPTED(inode))
1944                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1945
1946                 ret = fiemap_fill_next_extent(fieinfo, logical,
1947                                 phys, size, flags);
1948                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1949                 if (ret)
1950                         goto out;
1951                 size = 0;
1952         }
1953
1954         if (start_blk > last_blk)
1955                 goto out;
1956
1957 skip_fill:
1958         if (map.m_pblk == COMPRESS_ADDR) {
1959                 compr_cluster = true;
1960                 count_in_cluster = 1;
1961         } else if (compr_appended) {
1962                 unsigned int appended_blks = cluster_size -
1963                                                 count_in_cluster + 1;
1964                 size += blks_to_bytes(inode, appended_blks);
1965                 start_blk += appended_blks;
1966                 compr_cluster = false;
1967         } else {
1968                 logical = blks_to_bytes(inode, start_blk);
1969                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
1970                         blks_to_bytes(inode, map.m_pblk) : 0;
1971                 size = blks_to_bytes(inode, map.m_len);
1972                 flags = 0;
1973
1974                 if (compr_cluster) {
1975                         flags = FIEMAP_EXTENT_ENCODED;
1976                         count_in_cluster += map.m_len;
1977                         if (count_in_cluster == cluster_size) {
1978                                 compr_cluster = false;
1979                                 size += blks_to_bytes(inode, 1);
1980                         }
1981                 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
1982                         flags = FIEMAP_EXTENT_UNWRITTEN;
1983                 }
1984
1985                 start_blk += bytes_to_blks(inode, size);
1986         }
1987
1988 prep_next:
1989         cond_resched();
1990         if (fatal_signal_pending(current))
1991                 ret = -EINTR;
1992         else
1993                 goto next;
1994 out:
1995         if (ret == 1)
1996                 ret = 0;
1997
1998         inode_unlock(inode);
1999         return ret;
2000 }
2001
2002 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2003 {
2004         if (IS_ENABLED(CONFIG_FS_VERITY) &&
2005             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2006                 return inode->i_sb->s_maxbytes;
2007
2008         return i_size_read(inode);
2009 }
2010
2011 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2012                                         unsigned nr_pages,
2013                                         struct f2fs_map_blocks *map,
2014                                         struct bio **bio_ret,
2015                                         sector_t *last_block_in_bio,
2016                                         bool is_readahead)
2017 {
2018         struct bio *bio = *bio_ret;
2019         const unsigned blocksize = blks_to_bytes(inode, 1);
2020         sector_t block_in_file;
2021         sector_t last_block;
2022         sector_t last_block_in_file;
2023         sector_t block_nr;
2024         int ret = 0;
2025
2026         block_in_file = (sector_t)page_index(page);
2027         last_block = block_in_file + nr_pages;
2028         last_block_in_file = bytes_to_blks(inode,
2029                         f2fs_readpage_limit(inode) + blocksize - 1);
2030         if (last_block > last_block_in_file)
2031                 last_block = last_block_in_file;
2032
2033         /* just zeroing out page which is beyond EOF */
2034         if (block_in_file >= last_block)
2035                 goto zero_out;
2036         /*
2037          * Map blocks using the previous result first.
2038          */
2039         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2040                         block_in_file > map->m_lblk &&
2041                         block_in_file < (map->m_lblk + map->m_len))
2042                 goto got_it;
2043
2044         /*
2045          * Then do more f2fs_map_blocks() calls until we are
2046          * done with this page.
2047          */
2048         map->m_lblk = block_in_file;
2049         map->m_len = last_block - block_in_file;
2050
2051         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2052         if (ret)
2053                 goto out;
2054 got_it:
2055         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2056                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2057                 SetPageMappedToDisk(page);
2058
2059                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2060                                                 DATA_GENERIC_ENHANCE_READ)) {
2061                         ret = -EFSCORRUPTED;
2062                         goto out;
2063                 }
2064         } else {
2065 zero_out:
2066                 zero_user_segment(page, 0, PAGE_SIZE);
2067                 if (f2fs_need_verity(inode, page->index) &&
2068                     !fsverity_verify_page(page)) {
2069                         ret = -EIO;
2070                         goto out;
2071                 }
2072                 if (!PageUptodate(page))
2073                         SetPageUptodate(page);
2074                 unlock_page(page);
2075                 goto out;
2076         }
2077
2078         /*
2079          * This page will go to BIO.  Do we need to send this
2080          * BIO off first?
2081          */
2082         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2083                                        *last_block_in_bio, block_nr) ||
2084                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2085 submit_and_realloc:
2086                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2087                 bio = NULL;
2088         }
2089         if (bio == NULL) {
2090                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2091                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2092                                 false);
2093                 if (IS_ERR(bio)) {
2094                         ret = PTR_ERR(bio);
2095                         bio = NULL;
2096                         goto out;
2097                 }
2098         }
2099
2100         /*
2101          * If the page is under writeback, we need to wait for
2102          * its completion to see the correct decrypted data.
2103          */
2104         f2fs_wait_on_block_writeback(inode, block_nr);
2105
2106         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2107                 goto submit_and_realloc;
2108
2109         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2110         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2111         ClearPageError(page);
2112         *last_block_in_bio = block_nr;
2113         goto out;
2114 out:
2115         *bio_ret = bio;
2116         return ret;
2117 }
2118
2119 #ifdef CONFIG_F2FS_FS_COMPRESSION
2120 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2121                                 unsigned nr_pages, sector_t *last_block_in_bio,
2122                                 bool is_readahead, bool for_write)
2123 {
2124         struct dnode_of_data dn;
2125         struct inode *inode = cc->inode;
2126         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2127         struct bio *bio = *bio_ret;
2128         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2129         sector_t last_block_in_file;
2130         const unsigned blocksize = blks_to_bytes(inode, 1);
2131         struct decompress_io_ctx *dic = NULL;
2132         struct extent_info ei = {0, };
2133         bool from_dnode = true;
2134         int i;
2135         int ret = 0;
2136
2137         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2138
2139         last_block_in_file = bytes_to_blks(inode,
2140                         f2fs_readpage_limit(inode) + blocksize - 1);
2141
2142         /* get rid of pages beyond EOF */
2143         for (i = 0; i < cc->cluster_size; i++) {
2144                 struct page *page = cc->rpages[i];
2145
2146                 if (!page)
2147                         continue;
2148                 if ((sector_t)page->index >= last_block_in_file) {
2149                         zero_user_segment(page, 0, PAGE_SIZE);
2150                         if (!PageUptodate(page))
2151                                 SetPageUptodate(page);
2152                 } else if (!PageUptodate(page)) {
2153                         continue;
2154                 }
2155                 unlock_page(page);
2156                 if (for_write)
2157                         put_page(page);
2158                 cc->rpages[i] = NULL;
2159                 cc->nr_rpages--;
2160         }
2161
2162         /* we are done since all pages are beyond EOF */
2163         if (f2fs_cluster_is_empty(cc))
2164                 goto out;
2165
2166         if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2167                 from_dnode = false;
2168
2169         if (!from_dnode)
2170                 goto skip_reading_dnode;
2171
2172         set_new_dnode(&dn, inode, NULL, NULL, 0);
2173         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2174         if (ret)
2175                 goto out;
2176
2177         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2178
2179 skip_reading_dnode:
2180         for (i = 1; i < cc->cluster_size; i++) {
2181                 block_t blkaddr;
2182
2183                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2184                                         dn.ofs_in_node + i) :
2185                                         ei.blk + i - 1;
2186
2187                 if (!__is_valid_data_blkaddr(blkaddr))
2188                         break;
2189
2190                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2191                         ret = -EFAULT;
2192                         goto out_put_dnode;
2193                 }
2194                 cc->nr_cpages++;
2195
2196                 if (!from_dnode && i >= ei.c_len)
2197                         break;
2198         }
2199
2200         /* nothing to decompress */
2201         if (cc->nr_cpages == 0) {
2202                 ret = 0;
2203                 goto out_put_dnode;
2204         }
2205
2206         dic = f2fs_alloc_dic(cc);
2207         if (IS_ERR(dic)) {
2208                 ret = PTR_ERR(dic);
2209                 goto out_put_dnode;
2210         }
2211
2212         for (i = 0; i < cc->nr_cpages; i++) {
2213                 struct page *page = dic->cpages[i];
2214                 block_t blkaddr;
2215                 struct bio_post_read_ctx *ctx;
2216
2217                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2218                                         dn.ofs_in_node + i + 1) :
2219                                         ei.blk + i;
2220
2221                 f2fs_wait_on_block_writeback(inode, blkaddr);
2222
2223                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2224                         if (atomic_dec_and_test(&dic->remaining_pages))
2225                                 f2fs_decompress_cluster(dic);
2226                         continue;
2227                 }
2228
2229                 if (bio && (!page_is_mergeable(sbi, bio,
2230                                         *last_block_in_bio, blkaddr) ||
2231                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2232 submit_and_realloc:
2233                         __submit_bio(sbi, bio, DATA);
2234                         bio = NULL;
2235                 }
2236
2237                 if (!bio) {
2238                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2239                                         is_readahead ? REQ_RAHEAD : 0,
2240                                         page->index, for_write);
2241                         if (IS_ERR(bio)) {
2242                                 ret = PTR_ERR(bio);
2243                                 f2fs_decompress_end_io(dic, ret);
2244                                 f2fs_put_dnode(&dn);
2245                                 *bio_ret = NULL;
2246                                 return ret;
2247                         }
2248                 }
2249
2250                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2251                         goto submit_and_realloc;
2252
2253                 ctx = get_post_read_ctx(bio);
2254                 ctx->enabled_steps |= STEP_DECOMPRESS;
2255                 refcount_inc(&dic->refcnt);
2256
2257                 inc_page_count(sbi, F2FS_RD_DATA);
2258                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2259                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2260                 ClearPageError(page);
2261                 *last_block_in_bio = blkaddr;
2262         }
2263
2264         if (from_dnode)
2265                 f2fs_put_dnode(&dn);
2266
2267         *bio_ret = bio;
2268         return 0;
2269
2270 out_put_dnode:
2271         if (from_dnode)
2272                 f2fs_put_dnode(&dn);
2273 out:
2274         for (i = 0; i < cc->cluster_size; i++) {
2275                 if (cc->rpages[i]) {
2276                         ClearPageUptodate(cc->rpages[i]);
2277                         ClearPageError(cc->rpages[i]);
2278                         unlock_page(cc->rpages[i]);
2279                 }
2280         }
2281         *bio_ret = bio;
2282         return ret;
2283 }
2284 #endif
2285
2286 /*
2287  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2288  * Major change was from block_size == page_size in f2fs by default.
2289  */
2290 static int f2fs_mpage_readpages(struct inode *inode,
2291                 struct readahead_control *rac, struct page *page)
2292 {
2293         struct bio *bio = NULL;
2294         sector_t last_block_in_bio = 0;
2295         struct f2fs_map_blocks map;
2296 #ifdef CONFIG_F2FS_FS_COMPRESSION
2297         struct compress_ctx cc = {
2298                 .inode = inode,
2299                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2300                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2301                 .cluster_idx = NULL_CLUSTER,
2302                 .rpages = NULL,
2303                 .cpages = NULL,
2304                 .nr_rpages = 0,
2305                 .nr_cpages = 0,
2306         };
2307         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2308 #endif
2309         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2310         unsigned max_nr_pages = nr_pages;
2311         int ret = 0;
2312
2313         map.m_pblk = 0;
2314         map.m_lblk = 0;
2315         map.m_len = 0;
2316         map.m_flags = 0;
2317         map.m_next_pgofs = NULL;
2318         map.m_next_extent = NULL;
2319         map.m_seg_type = NO_CHECK_TYPE;
2320         map.m_may_create = false;
2321
2322         for (; nr_pages; nr_pages--) {
2323                 if (rac) {
2324                         page = readahead_page(rac);
2325                         prefetchw(&page->flags);
2326                 }
2327
2328 #ifdef CONFIG_F2FS_FS_COMPRESSION
2329                 if (f2fs_compressed_file(inode)) {
2330                         /* there are remained comressed pages, submit them */
2331                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2332                                 ret = f2fs_read_multi_pages(&cc, &bio,
2333                                                         max_nr_pages,
2334                                                         &last_block_in_bio,
2335                                                         rac != NULL, false);
2336                                 f2fs_destroy_compress_ctx(&cc, false);
2337                                 if (ret)
2338                                         goto set_error_page;
2339                         }
2340                         if (cc.cluster_idx == NULL_CLUSTER) {
2341                                 if (nc_cluster_idx ==
2342                                         page->index >> cc.log_cluster_size) {
2343                                         goto read_single_page;
2344                                 }
2345
2346                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2347                                 if (ret < 0)
2348                                         goto set_error_page;
2349                                 else if (!ret) {
2350                                         nc_cluster_idx =
2351                                                 page->index >> cc.log_cluster_size;
2352                                         goto read_single_page;
2353                                 }
2354
2355                                 nc_cluster_idx = NULL_CLUSTER;
2356                         }
2357                         ret = f2fs_init_compress_ctx(&cc);
2358                         if (ret)
2359                                 goto set_error_page;
2360
2361                         f2fs_compress_ctx_add_page(&cc, page);
2362
2363                         goto next_page;
2364                 }
2365 read_single_page:
2366 #endif
2367
2368                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2369                                         &bio, &last_block_in_bio, rac);
2370                 if (ret) {
2371 #ifdef CONFIG_F2FS_FS_COMPRESSION
2372 set_error_page:
2373 #endif
2374                         SetPageError(page);
2375                         zero_user_segment(page, 0, PAGE_SIZE);
2376                         unlock_page(page);
2377                 }
2378 #ifdef CONFIG_F2FS_FS_COMPRESSION
2379 next_page:
2380 #endif
2381                 if (rac)
2382                         put_page(page);
2383
2384 #ifdef CONFIG_F2FS_FS_COMPRESSION
2385                 if (f2fs_compressed_file(inode)) {
2386                         /* last page */
2387                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2388                                 ret = f2fs_read_multi_pages(&cc, &bio,
2389                                                         max_nr_pages,
2390                                                         &last_block_in_bio,
2391                                                         rac != NULL, false);
2392                                 f2fs_destroy_compress_ctx(&cc, false);
2393                         }
2394                 }
2395 #endif
2396         }
2397         if (bio)
2398                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2399         return ret;
2400 }
2401
2402 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2403 {
2404         struct page *page = &folio->page;
2405         struct inode *inode = page_file_mapping(page)->host;
2406         int ret = -EAGAIN;
2407
2408         trace_f2fs_readpage(page, DATA);
2409
2410         if (!f2fs_is_compress_backend_ready(inode)) {
2411                 unlock_page(page);
2412                 return -EOPNOTSUPP;
2413         }
2414
2415         /* If the file has inline data, try to read it directly */
2416         if (f2fs_has_inline_data(inode))
2417                 ret = f2fs_read_inline_data(inode, page);
2418         if (ret == -EAGAIN)
2419                 ret = f2fs_mpage_readpages(inode, NULL, page);
2420         return ret;
2421 }
2422
2423 static void f2fs_readahead(struct readahead_control *rac)
2424 {
2425         struct inode *inode = rac->mapping->host;
2426
2427         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2428
2429         if (!f2fs_is_compress_backend_ready(inode))
2430                 return;
2431
2432         /* If the file has inline data, skip readahead */
2433         if (f2fs_has_inline_data(inode))
2434                 return;
2435
2436         f2fs_mpage_readpages(inode, rac, NULL);
2437 }
2438
2439 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2440 {
2441         struct inode *inode = fio->page->mapping->host;
2442         struct page *mpage, *page;
2443         gfp_t gfp_flags = GFP_NOFS;
2444
2445         if (!f2fs_encrypted_file(inode))
2446                 return 0;
2447
2448         page = fio->compressed_page ? fio->compressed_page : fio->page;
2449
2450         /* wait for GCed page writeback via META_MAPPING */
2451         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2452
2453         if (fscrypt_inode_uses_inline_crypto(inode))
2454                 return 0;
2455
2456 retry_encrypt:
2457         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2458                                         PAGE_SIZE, 0, gfp_flags);
2459         if (IS_ERR(fio->encrypted_page)) {
2460                 /* flush pending IOs and wait for a while in the ENOMEM case */
2461                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2462                         f2fs_flush_merged_writes(fio->sbi);
2463                         memalloc_retry_wait(GFP_NOFS);
2464                         gfp_flags |= __GFP_NOFAIL;
2465                         goto retry_encrypt;
2466                 }
2467                 return PTR_ERR(fio->encrypted_page);
2468         }
2469
2470         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2471         if (mpage) {
2472                 if (PageUptodate(mpage))
2473                         memcpy(page_address(mpage),
2474                                 page_address(fio->encrypted_page), PAGE_SIZE);
2475                 f2fs_put_page(mpage, 1);
2476         }
2477         return 0;
2478 }
2479
2480 static inline bool check_inplace_update_policy(struct inode *inode,
2481                                 struct f2fs_io_info *fio)
2482 {
2483         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2484         unsigned int policy = SM_I(sbi)->ipu_policy;
2485
2486         if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
2487                         is_inode_flag_set(inode, FI_OPU_WRITE))
2488                 return false;
2489         if (policy & (0x1 << F2FS_IPU_FORCE))
2490                 return true;
2491         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2492                 return true;
2493         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2494                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2495                 return true;
2496         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2497                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2498                 return true;
2499
2500         /*
2501          * IPU for rewrite async pages
2502          */
2503         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2504                         fio && fio->op == REQ_OP_WRITE &&
2505                         !(fio->op_flags & REQ_SYNC) &&
2506                         !IS_ENCRYPTED(inode))
2507                 return true;
2508
2509         /* this is only set during fdatasync */
2510         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2511                         is_inode_flag_set(inode, FI_NEED_IPU))
2512                 return true;
2513
2514         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2515                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2516                 return true;
2517
2518         return false;
2519 }
2520
2521 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2522 {
2523         /* swap file is migrating in aligned write mode */
2524         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2525                 return false;
2526
2527         if (f2fs_is_pinned_file(inode))
2528                 return true;
2529
2530         /* if this is cold file, we should overwrite to avoid fragmentation */
2531         if (file_is_cold(inode))
2532                 return true;
2533
2534         return check_inplace_update_policy(inode, fio);
2535 }
2536
2537 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2538 {
2539         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2540
2541         /* The below cases were checked when setting it. */
2542         if (f2fs_is_pinned_file(inode))
2543                 return false;
2544         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2545                 return true;
2546         if (f2fs_lfs_mode(sbi))
2547                 return true;
2548         if (S_ISDIR(inode->i_mode))
2549                 return true;
2550         if (IS_NOQUOTA(inode))
2551                 return true;
2552         if (f2fs_is_atomic_file(inode))
2553                 return true;
2554
2555         /* swap file is migrating in aligned write mode */
2556         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2557                 return true;
2558
2559         if (is_inode_flag_set(inode, FI_OPU_WRITE))
2560                 return true;
2561
2562         if (fio) {
2563                 if (page_private_gcing(fio->page))
2564                         return true;
2565                 if (page_private_dummy(fio->page))
2566                         return true;
2567                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2568                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2569                         return true;
2570         }
2571         return false;
2572 }
2573
2574 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2575 {
2576         struct inode *inode = fio->page->mapping->host;
2577
2578         if (f2fs_should_update_outplace(inode, fio))
2579                 return false;
2580
2581         return f2fs_should_update_inplace(inode, fio);
2582 }
2583
2584 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2585 {
2586         struct page *page = fio->page;
2587         struct inode *inode = page->mapping->host;
2588         struct dnode_of_data dn;
2589         struct extent_info ei = {0, };
2590         struct node_info ni;
2591         bool ipu_force = false;
2592         int err = 0;
2593
2594         /* Use COW inode to make dnode_of_data for atomic write */
2595         if (f2fs_is_atomic_file(inode))
2596                 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2597         else
2598                 set_new_dnode(&dn, inode, NULL, NULL, 0);
2599
2600         if (need_inplace_update(fio) &&
2601                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2602                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2603
2604                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2605                                                 DATA_GENERIC_ENHANCE))
2606                         return -EFSCORRUPTED;
2607
2608                 ipu_force = true;
2609                 fio->need_lock = LOCK_DONE;
2610                 goto got_it;
2611         }
2612
2613         /* Deadlock due to between page->lock and f2fs_lock_op */
2614         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2615                 return -EAGAIN;
2616
2617         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2618         if (err)
2619                 goto out;
2620
2621         fio->old_blkaddr = dn.data_blkaddr;
2622
2623         /* This page is already truncated */
2624         if (fio->old_blkaddr == NULL_ADDR) {
2625                 ClearPageUptodate(page);
2626                 clear_page_private_gcing(page);
2627                 goto out_writepage;
2628         }
2629 got_it:
2630         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2631                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2632                                                 DATA_GENERIC_ENHANCE)) {
2633                 err = -EFSCORRUPTED;
2634                 goto out_writepage;
2635         }
2636
2637         /*
2638          * If current allocation needs SSR,
2639          * it had better in-place writes for updated data.
2640          */
2641         if (ipu_force ||
2642                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2643                                         need_inplace_update(fio))) {
2644                 err = f2fs_encrypt_one_page(fio);
2645                 if (err)
2646                         goto out_writepage;
2647
2648                 set_page_writeback(page);
2649                 ClearPageError(page);
2650                 f2fs_put_dnode(&dn);
2651                 if (fio->need_lock == LOCK_REQ)
2652                         f2fs_unlock_op(fio->sbi);
2653                 err = f2fs_inplace_write_data(fio);
2654                 if (err) {
2655                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2656                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2657                         if (PageWriteback(page))
2658                                 end_page_writeback(page);
2659                 } else {
2660                         set_inode_flag(inode, FI_UPDATE_WRITE);
2661                 }
2662                 trace_f2fs_do_write_data_page(fio->page, IPU);
2663                 return err;
2664         }
2665
2666         if (fio->need_lock == LOCK_RETRY) {
2667                 if (!f2fs_trylock_op(fio->sbi)) {
2668                         err = -EAGAIN;
2669                         goto out_writepage;
2670                 }
2671                 fio->need_lock = LOCK_REQ;
2672         }
2673
2674         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2675         if (err)
2676                 goto out_writepage;
2677
2678         fio->version = ni.version;
2679
2680         err = f2fs_encrypt_one_page(fio);
2681         if (err)
2682                 goto out_writepage;
2683
2684         set_page_writeback(page);
2685         ClearPageError(page);
2686
2687         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2688                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2689
2690         /* LFS mode write path */
2691         f2fs_outplace_write_data(&dn, fio);
2692         trace_f2fs_do_write_data_page(page, OPU);
2693         set_inode_flag(inode, FI_APPEND_WRITE);
2694         if (page->index == 0)
2695                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2696 out_writepage:
2697         f2fs_put_dnode(&dn);
2698 out:
2699         if (fio->need_lock == LOCK_REQ)
2700                 f2fs_unlock_op(fio->sbi);
2701         return err;
2702 }
2703
2704 int f2fs_write_single_data_page(struct page *page, int *submitted,
2705                                 struct bio **bio,
2706                                 sector_t *last_block,
2707                                 struct writeback_control *wbc,
2708                                 enum iostat_type io_type,
2709                                 int compr_blocks,
2710                                 bool allow_balance)
2711 {
2712         struct inode *inode = page->mapping->host;
2713         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2714         loff_t i_size = i_size_read(inode);
2715         const pgoff_t end_index = ((unsigned long long)i_size)
2716                                                         >> PAGE_SHIFT;
2717         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2718         unsigned offset = 0;
2719         bool need_balance_fs = false;
2720         int err = 0;
2721         struct f2fs_io_info fio = {
2722                 .sbi = sbi,
2723                 .ino = inode->i_ino,
2724                 .type = DATA,
2725                 .op = REQ_OP_WRITE,
2726                 .op_flags = wbc_to_write_flags(wbc),
2727                 .old_blkaddr = NULL_ADDR,
2728                 .page = page,
2729                 .encrypted_page = NULL,
2730                 .submitted = false,
2731                 .compr_blocks = compr_blocks,
2732                 .need_lock = LOCK_RETRY,
2733                 .io_type = io_type,
2734                 .io_wbc = wbc,
2735                 .bio = bio,
2736                 .last_block = last_block,
2737         };
2738
2739         trace_f2fs_writepage(page, DATA);
2740
2741         /* we should bypass data pages to proceed the kworkder jobs */
2742         if (unlikely(f2fs_cp_error(sbi))) {
2743                 mapping_set_error(page->mapping, -EIO);
2744                 /*
2745                  * don't drop any dirty dentry pages for keeping lastest
2746                  * directory structure.
2747                  */
2748                 if (S_ISDIR(inode->i_mode))
2749                         goto redirty_out;
2750                 goto out;
2751         }
2752
2753         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2754                 goto redirty_out;
2755
2756         if (page->index < end_index ||
2757                         f2fs_verity_in_progress(inode) ||
2758                         compr_blocks)
2759                 goto write;
2760
2761         /*
2762          * If the offset is out-of-range of file size,
2763          * this page does not have to be written to disk.
2764          */
2765         offset = i_size & (PAGE_SIZE - 1);
2766         if ((page->index >= end_index + 1) || !offset)
2767                 goto out;
2768
2769         zero_user_segment(page, offset, PAGE_SIZE);
2770 write:
2771         if (f2fs_is_drop_cache(inode))
2772                 goto out;
2773
2774         /* Dentry/quota blocks are controlled by checkpoint */
2775         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2776                 /*
2777                  * We need to wait for node_write to avoid block allocation during
2778                  * checkpoint. This can only happen to quota writes which can cause
2779                  * the below discard race condition.
2780                  */
2781                 if (IS_NOQUOTA(inode))
2782                         f2fs_down_read(&sbi->node_write);
2783
2784                 fio.need_lock = LOCK_DONE;
2785                 err = f2fs_do_write_data_page(&fio);
2786
2787                 if (IS_NOQUOTA(inode))
2788                         f2fs_up_read(&sbi->node_write);
2789
2790                 goto done;
2791         }
2792
2793         if (!wbc->for_reclaim)
2794                 need_balance_fs = true;
2795         else if (has_not_enough_free_secs(sbi, 0, 0))
2796                 goto redirty_out;
2797         else
2798                 set_inode_flag(inode, FI_HOT_DATA);
2799
2800         err = -EAGAIN;
2801         if (f2fs_has_inline_data(inode)) {
2802                 err = f2fs_write_inline_data(inode, page);
2803                 if (!err)
2804                         goto out;
2805         }
2806
2807         if (err == -EAGAIN) {
2808                 err = f2fs_do_write_data_page(&fio);
2809                 if (err == -EAGAIN) {
2810                         fio.need_lock = LOCK_REQ;
2811                         err = f2fs_do_write_data_page(&fio);
2812                 }
2813         }
2814
2815         if (err) {
2816                 file_set_keep_isize(inode);
2817         } else {
2818                 spin_lock(&F2FS_I(inode)->i_size_lock);
2819                 if (F2FS_I(inode)->last_disk_size < psize)
2820                         F2FS_I(inode)->last_disk_size = psize;
2821                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2822         }
2823
2824 done:
2825         if (err && err != -ENOENT)
2826                 goto redirty_out;
2827
2828 out:
2829         inode_dec_dirty_pages(inode);
2830         if (err) {
2831                 ClearPageUptodate(page);
2832                 clear_page_private_gcing(page);
2833         }
2834
2835         if (wbc->for_reclaim) {
2836                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2837                 clear_inode_flag(inode, FI_HOT_DATA);
2838                 f2fs_remove_dirty_inode(inode);
2839                 submitted = NULL;
2840         }
2841         unlock_page(page);
2842         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2843                         !F2FS_I(inode)->cp_task && allow_balance)
2844                 f2fs_balance_fs(sbi, need_balance_fs);
2845
2846         if (unlikely(f2fs_cp_error(sbi))) {
2847                 f2fs_submit_merged_write(sbi, DATA);
2848                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2849                 submitted = NULL;
2850         }
2851
2852         if (submitted)
2853                 *submitted = fio.submitted ? 1 : 0;
2854
2855         return 0;
2856
2857 redirty_out:
2858         redirty_page_for_writepage(wbc, page);
2859         /*
2860          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2861          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2862          * file_write_and_wait_range() will see EIO error, which is critical
2863          * to return value of fsync() followed by atomic_write failure to user.
2864          */
2865         if (!err || wbc->for_reclaim)
2866                 return AOP_WRITEPAGE_ACTIVATE;
2867         unlock_page(page);
2868         return err;
2869 }
2870
2871 static int f2fs_write_data_page(struct page *page,
2872                                         struct writeback_control *wbc)
2873 {
2874 #ifdef CONFIG_F2FS_FS_COMPRESSION
2875         struct inode *inode = page->mapping->host;
2876
2877         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2878                 goto out;
2879
2880         if (f2fs_compressed_file(inode)) {
2881                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2882                         redirty_page_for_writepage(wbc, page);
2883                         return AOP_WRITEPAGE_ACTIVATE;
2884                 }
2885         }
2886 out:
2887 #endif
2888
2889         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2890                                                 wbc, FS_DATA_IO, 0, true);
2891 }
2892
2893 /*
2894  * This function was copied from write_cche_pages from mm/page-writeback.c.
2895  * The major change is making write step of cold data page separately from
2896  * warm/hot data page.
2897  */
2898 static int f2fs_write_cache_pages(struct address_space *mapping,
2899                                         struct writeback_control *wbc,
2900                                         enum iostat_type io_type)
2901 {
2902         int ret = 0;
2903         int done = 0, retry = 0;
2904         struct pagevec pvec;
2905         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2906         struct bio *bio = NULL;
2907         sector_t last_block;
2908 #ifdef CONFIG_F2FS_FS_COMPRESSION
2909         struct inode *inode = mapping->host;
2910         struct compress_ctx cc = {
2911                 .inode = inode,
2912                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2913                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2914                 .cluster_idx = NULL_CLUSTER,
2915                 .rpages = NULL,
2916                 .nr_rpages = 0,
2917                 .cpages = NULL,
2918                 .valid_nr_cpages = 0,
2919                 .rbuf = NULL,
2920                 .cbuf = NULL,
2921                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2922                 .private = NULL,
2923         };
2924 #endif
2925         int nr_pages;
2926         pgoff_t index;
2927         pgoff_t end;            /* Inclusive */
2928         pgoff_t done_index;
2929         int range_whole = 0;
2930         xa_mark_t tag;
2931         int nwritten = 0;
2932         int submitted = 0;
2933         int i;
2934
2935         pagevec_init(&pvec);
2936
2937         if (get_dirty_pages(mapping->host) <=
2938                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2939                 set_inode_flag(mapping->host, FI_HOT_DATA);
2940         else
2941                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2942
2943         if (wbc->range_cyclic) {
2944                 index = mapping->writeback_index; /* prev offset */
2945                 end = -1;
2946         } else {
2947                 index = wbc->range_start >> PAGE_SHIFT;
2948                 end = wbc->range_end >> PAGE_SHIFT;
2949                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2950                         range_whole = 1;
2951         }
2952         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2953                 tag = PAGECACHE_TAG_TOWRITE;
2954         else
2955                 tag = PAGECACHE_TAG_DIRTY;
2956 retry:
2957         retry = 0;
2958         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2959                 tag_pages_for_writeback(mapping, index, end);
2960         done_index = index;
2961         while (!done && !retry && (index <= end)) {
2962                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2963                                 tag);
2964                 if (nr_pages == 0)
2965                         break;
2966
2967                 for (i = 0; i < nr_pages; i++) {
2968                         struct page *page = pvec.pages[i];
2969                         bool need_readd;
2970 readd:
2971                         need_readd = false;
2972 #ifdef CONFIG_F2FS_FS_COMPRESSION
2973                         if (f2fs_compressed_file(inode)) {
2974                                 void *fsdata = NULL;
2975                                 struct page *pagep;
2976                                 int ret2;
2977
2978                                 ret = f2fs_init_compress_ctx(&cc);
2979                                 if (ret) {
2980                                         done = 1;
2981                                         break;
2982                                 }
2983
2984                                 if (!f2fs_cluster_can_merge_page(&cc,
2985                                                                 page->index)) {
2986                                         ret = f2fs_write_multi_pages(&cc,
2987                                                 &submitted, wbc, io_type);
2988                                         if (!ret)
2989                                                 need_readd = true;
2990                                         goto result;
2991                                 }
2992
2993                                 if (unlikely(f2fs_cp_error(sbi)))
2994                                         goto lock_page;
2995
2996                                 if (!f2fs_cluster_is_empty(&cc))
2997                                         goto lock_page;
2998
2999                                 ret2 = f2fs_prepare_compress_overwrite(
3000                                                         inode, &pagep,
3001                                                         page->index, &fsdata);
3002                                 if (ret2 < 0) {
3003                                         ret = ret2;
3004                                         done = 1;
3005                                         break;
3006                                 } else if (ret2 &&
3007                                         (!f2fs_compress_write_end(inode,
3008                                                 fsdata, page->index, 1) ||
3009                                          !f2fs_all_cluster_page_loaded(&cc,
3010                                                 &pvec, i, nr_pages))) {
3011                                         retry = 1;
3012                                         break;
3013                                 }
3014                         }
3015 #endif
3016                         /* give a priority to WB_SYNC threads */
3017                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3018                                         wbc->sync_mode == WB_SYNC_NONE) {
3019                                 done = 1;
3020                                 break;
3021                         }
3022 #ifdef CONFIG_F2FS_FS_COMPRESSION
3023 lock_page:
3024 #endif
3025                         done_index = page->index;
3026 retry_write:
3027                         lock_page(page);
3028
3029                         if (unlikely(page->mapping != mapping)) {
3030 continue_unlock:
3031                                 unlock_page(page);
3032                                 continue;
3033                         }
3034
3035                         if (!PageDirty(page)) {
3036                                 /* someone wrote it for us */
3037                                 goto continue_unlock;
3038                         }
3039
3040                         if (PageWriteback(page)) {
3041                                 if (wbc->sync_mode != WB_SYNC_NONE)
3042                                         f2fs_wait_on_page_writeback(page,
3043                                                         DATA, true, true);
3044                                 else
3045                                         goto continue_unlock;
3046                         }
3047
3048                         if (!clear_page_dirty_for_io(page))
3049                                 goto continue_unlock;
3050
3051 #ifdef CONFIG_F2FS_FS_COMPRESSION
3052                         if (f2fs_compressed_file(inode)) {
3053                                 get_page(page);
3054                                 f2fs_compress_ctx_add_page(&cc, page);
3055                                 continue;
3056                         }
3057 #endif
3058                         ret = f2fs_write_single_data_page(page, &submitted,
3059                                         &bio, &last_block, wbc, io_type,
3060                                         0, true);
3061                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3062                                 unlock_page(page);
3063 #ifdef CONFIG_F2FS_FS_COMPRESSION
3064 result:
3065 #endif
3066                         nwritten += submitted;
3067                         wbc->nr_to_write -= submitted;
3068
3069                         if (unlikely(ret)) {
3070                                 /*
3071                                  * keep nr_to_write, since vfs uses this to
3072                                  * get # of written pages.
3073                                  */
3074                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3075                                         ret = 0;
3076                                         goto next;
3077                                 } else if (ret == -EAGAIN) {
3078                                         ret = 0;
3079                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3080                                                 f2fs_io_schedule_timeout(
3081                                                         DEFAULT_IO_TIMEOUT);
3082                                                 goto retry_write;
3083                                         }
3084                                         goto next;
3085                                 }
3086                                 done_index = page->index + 1;
3087                                 done = 1;
3088                                 break;
3089                         }
3090
3091                         if (wbc->nr_to_write <= 0 &&
3092                                         wbc->sync_mode == WB_SYNC_NONE) {
3093                                 done = 1;
3094                                 break;
3095                         }
3096 next:
3097                         if (need_readd)
3098                                 goto readd;
3099                 }
3100                 pagevec_release(&pvec);
3101                 cond_resched();
3102         }
3103 #ifdef CONFIG_F2FS_FS_COMPRESSION
3104         /* flush remained pages in compress cluster */
3105         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3106                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3107                 nwritten += submitted;
3108                 wbc->nr_to_write -= submitted;
3109                 if (ret) {
3110                         done = 1;
3111                         retry = 0;
3112                 }
3113         }
3114         if (f2fs_compressed_file(inode))
3115                 f2fs_destroy_compress_ctx(&cc, false);
3116 #endif
3117         if (retry) {
3118                 index = 0;
3119                 end = -1;
3120                 goto retry;
3121         }
3122         if (wbc->range_cyclic && !done)
3123                 done_index = 0;
3124         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3125                 mapping->writeback_index = done_index;
3126
3127         if (nwritten)
3128                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3129                                                                 NULL, 0, DATA);
3130         /* submit cached bio of IPU write */
3131         if (bio)
3132                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3133
3134         return ret;
3135 }
3136
3137 static inline bool __should_serialize_io(struct inode *inode,
3138                                         struct writeback_control *wbc)
3139 {
3140         /* to avoid deadlock in path of data flush */
3141         if (F2FS_I(inode)->cp_task)
3142                 return false;
3143
3144         if (!S_ISREG(inode->i_mode))
3145                 return false;
3146         if (IS_NOQUOTA(inode))
3147                 return false;
3148
3149         if (f2fs_need_compress_data(inode))
3150                 return true;
3151         if (wbc->sync_mode != WB_SYNC_ALL)
3152                 return true;
3153         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3154                 return true;
3155         return false;
3156 }
3157
3158 static int __f2fs_write_data_pages(struct address_space *mapping,
3159                                                 struct writeback_control *wbc,
3160                                                 enum iostat_type io_type)
3161 {
3162         struct inode *inode = mapping->host;
3163         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3164         struct blk_plug plug;
3165         int ret;
3166         bool locked = false;
3167
3168         /* deal with chardevs and other special file */
3169         if (!mapping->a_ops->writepage)
3170                 return 0;
3171
3172         /* skip writing if there is no dirty page in this inode */
3173         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3174                 return 0;
3175
3176         /* during POR, we don't need to trigger writepage at all. */
3177         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3178                 goto skip_write;
3179
3180         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3181                         wbc->sync_mode == WB_SYNC_NONE &&
3182                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3183                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3184                 goto skip_write;
3185
3186         /* skip writing in file defragment preparing stage */
3187         if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3188                 goto skip_write;
3189
3190         trace_f2fs_writepages(mapping->host, wbc, DATA);
3191
3192         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3193         if (wbc->sync_mode == WB_SYNC_ALL)
3194                 atomic_inc(&sbi->wb_sync_req[DATA]);
3195         else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3196                 /* to avoid potential deadlock */
3197                 if (current->plug)
3198                         blk_finish_plug(current->plug);
3199                 goto skip_write;
3200         }
3201
3202         if (__should_serialize_io(inode, wbc)) {
3203                 mutex_lock(&sbi->writepages);
3204                 locked = true;
3205         }
3206
3207         blk_start_plug(&plug);
3208         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3209         blk_finish_plug(&plug);
3210
3211         if (locked)
3212                 mutex_unlock(&sbi->writepages);
3213
3214         if (wbc->sync_mode == WB_SYNC_ALL)
3215                 atomic_dec(&sbi->wb_sync_req[DATA]);
3216         /*
3217          * if some pages were truncated, we cannot guarantee its mapping->host
3218          * to detect pending bios.
3219          */
3220
3221         f2fs_remove_dirty_inode(inode);
3222         return ret;
3223
3224 skip_write:
3225         wbc->pages_skipped += get_dirty_pages(inode);
3226         trace_f2fs_writepages(mapping->host, wbc, DATA);
3227         return 0;
3228 }
3229
3230 static int f2fs_write_data_pages(struct address_space *mapping,
3231                             struct writeback_control *wbc)
3232 {
3233         struct inode *inode = mapping->host;
3234
3235         return __f2fs_write_data_pages(mapping, wbc,
3236                         F2FS_I(inode)->cp_task == current ?
3237                         FS_CP_DATA_IO : FS_DATA_IO);
3238 }
3239
3240 void f2fs_write_failed(struct inode *inode, loff_t to)
3241 {
3242         loff_t i_size = i_size_read(inode);
3243
3244         if (IS_NOQUOTA(inode))
3245                 return;
3246
3247         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3248         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3249                 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3250                 filemap_invalidate_lock(inode->i_mapping);
3251
3252                 truncate_pagecache(inode, i_size);
3253                 f2fs_truncate_blocks(inode, i_size, true);
3254
3255                 filemap_invalidate_unlock(inode->i_mapping);
3256                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3257         }
3258 }
3259
3260 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3261                         struct page *page, loff_t pos, unsigned len,
3262                         block_t *blk_addr, bool *node_changed)
3263 {
3264         struct inode *inode = page->mapping->host;
3265         pgoff_t index = page->index;
3266         struct dnode_of_data dn;
3267         struct page *ipage;
3268         bool locked = false;
3269         struct extent_info ei = {0, };
3270         int err = 0;
3271         int flag;
3272
3273         /*
3274          * If a whole page is being written and we already preallocated all the
3275          * blocks, then there is no need to get a block address now.
3276          */
3277         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3278                 return 0;
3279
3280         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3281         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3282                 flag = F2FS_GET_BLOCK_DEFAULT;
3283         else
3284                 flag = F2FS_GET_BLOCK_PRE_AIO;
3285
3286         if (f2fs_has_inline_data(inode) ||
3287                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3288                 f2fs_do_map_lock(sbi, flag, true);
3289                 locked = true;
3290         }
3291
3292 restart:
3293         /* check inline_data */
3294         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3295         if (IS_ERR(ipage)) {
3296                 err = PTR_ERR(ipage);
3297                 goto unlock_out;
3298         }
3299
3300         set_new_dnode(&dn, inode, ipage, ipage, 0);
3301
3302         if (f2fs_has_inline_data(inode)) {
3303                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3304                         f2fs_do_read_inline_data(page, ipage);
3305                         set_inode_flag(inode, FI_DATA_EXIST);
3306                         if (inode->i_nlink)
3307                                 set_page_private_inline(ipage);
3308                 } else {
3309                         err = f2fs_convert_inline_page(&dn, page);
3310                         if (err)
3311                                 goto out;
3312                         if (dn.data_blkaddr == NULL_ADDR)
3313                                 err = f2fs_get_block(&dn, index);
3314                 }
3315         } else if (locked) {
3316                 err = f2fs_get_block(&dn, index);
3317         } else {
3318                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3319                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3320                 } else {
3321                         /* hole case */
3322                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3323                         if (err || dn.data_blkaddr == NULL_ADDR) {
3324                                 f2fs_put_dnode(&dn);
3325                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3326                                                                 true);
3327                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3328                                 locked = true;
3329                                 goto restart;
3330                         }
3331                 }
3332         }
3333
3334         /* convert_inline_page can make node_changed */
3335         *blk_addr = dn.data_blkaddr;
3336         *node_changed = dn.node_changed;
3337 out:
3338         f2fs_put_dnode(&dn);
3339 unlock_out:
3340         if (locked)
3341                 f2fs_do_map_lock(sbi, flag, false);
3342         return err;
3343 }
3344
3345 static int __find_data_block(struct inode *inode, pgoff_t index,
3346                                 block_t *blk_addr)
3347 {
3348         struct dnode_of_data dn;
3349         struct page *ipage;
3350         struct extent_info ei = {0, };
3351         int err = 0;
3352
3353         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3354         if (IS_ERR(ipage))
3355                 return PTR_ERR(ipage);
3356
3357         set_new_dnode(&dn, inode, ipage, ipage, 0);
3358
3359         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3360                 dn.data_blkaddr = ei.blk + index - ei.fofs;
3361         } else {
3362                 /* hole case */
3363                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3364                 if (err) {
3365                         dn.data_blkaddr = NULL_ADDR;
3366                         err = 0;
3367                 }
3368         }
3369         *blk_addr = dn.data_blkaddr;
3370         f2fs_put_dnode(&dn);
3371         return err;
3372 }
3373
3374 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3375                                 block_t *blk_addr, bool *node_changed)
3376 {
3377         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3378         struct dnode_of_data dn;
3379         struct page *ipage;
3380         int err = 0;
3381
3382         f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
3383
3384         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3385         if (IS_ERR(ipage)) {
3386                 err = PTR_ERR(ipage);
3387                 goto unlock_out;
3388         }
3389         set_new_dnode(&dn, inode, ipage, ipage, 0);
3390
3391         err = f2fs_get_block(&dn, index);
3392
3393         *blk_addr = dn.data_blkaddr;
3394         *node_changed = dn.node_changed;
3395         f2fs_put_dnode(&dn);
3396
3397 unlock_out:
3398         f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
3399         return err;
3400 }
3401
3402 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3403                         struct page *page, loff_t pos, unsigned int len,
3404                         block_t *blk_addr, bool *node_changed)
3405 {
3406         struct inode *inode = page->mapping->host;
3407         struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3408         pgoff_t index = page->index;
3409         int err = 0;
3410         block_t ori_blk_addr;
3411
3412         /* If pos is beyond the end of file, reserve a new block in COW inode */
3413         if ((pos & PAGE_MASK) >= i_size_read(inode))
3414                 return __reserve_data_block(cow_inode, index, blk_addr,
3415                                         node_changed);
3416
3417         /* Look for the block in COW inode first */
3418         err = __find_data_block(cow_inode, index, blk_addr);
3419         if (err)
3420                 return err;
3421         else if (*blk_addr != NULL_ADDR)
3422                 return 0;
3423
3424         /* Look for the block in the original inode */
3425         err = __find_data_block(inode, index, &ori_blk_addr);
3426         if (err)
3427                 return err;
3428
3429         /* Finally, we should reserve a new block in COW inode for the update */
3430         err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3431         if (err)
3432                 return err;
3433
3434         if (ori_blk_addr != NULL_ADDR)
3435                 *blk_addr = ori_blk_addr;
3436         return 0;
3437 }
3438
3439 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3440                 loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3441 {
3442         struct inode *inode = mapping->host;
3443         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3444         struct page *page = NULL;
3445         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3446         bool need_balance = false;
3447         block_t blkaddr = NULL_ADDR;
3448         int err = 0;
3449
3450         trace_f2fs_write_begin(inode, pos, len);
3451
3452         if (!f2fs_is_checkpoint_ready(sbi)) {
3453                 err = -ENOSPC;
3454                 goto fail;
3455         }
3456
3457         /*
3458          * We should check this at this moment to avoid deadlock on inode page
3459          * and #0 page. The locking rule for inline_data conversion should be:
3460          * lock_page(page #0) -> lock_page(inode_page)
3461          */
3462         if (index != 0) {
3463                 err = f2fs_convert_inline_inode(inode);
3464                 if (err)
3465                         goto fail;
3466         }
3467
3468 #ifdef CONFIG_F2FS_FS_COMPRESSION
3469         if (f2fs_compressed_file(inode)) {
3470                 int ret;
3471
3472                 *fsdata = NULL;
3473
3474                 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3475                         goto repeat;
3476
3477                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3478                                                         index, fsdata);
3479                 if (ret < 0) {
3480                         err = ret;
3481                         goto fail;
3482                 } else if (ret) {
3483                         return 0;
3484                 }
3485         }
3486 #endif
3487
3488 repeat:
3489         /*
3490          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3491          * wait_for_stable_page. Will wait that below with our IO control.
3492          */
3493         page = f2fs_pagecache_get_page(mapping, index,
3494                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3495         if (!page) {
3496                 err = -ENOMEM;
3497                 goto fail;
3498         }
3499
3500         /* TODO: cluster can be compressed due to race with .writepage */
3501
3502         *pagep = page;
3503
3504         if (f2fs_is_atomic_file(inode))
3505                 err = prepare_atomic_write_begin(sbi, page, pos, len,
3506                                         &blkaddr, &need_balance);
3507         else
3508                 err = prepare_write_begin(sbi, page, pos, len,
3509                                         &blkaddr, &need_balance);
3510         if (err)
3511                 goto fail;
3512
3513         if (need_balance && !IS_NOQUOTA(inode) &&
3514                         has_not_enough_free_secs(sbi, 0, 0)) {
3515                 unlock_page(page);
3516                 f2fs_balance_fs(sbi, true);
3517                 lock_page(page);
3518                 if (page->mapping != mapping) {
3519                         /* The page got truncated from under us */
3520                         f2fs_put_page(page, 1);
3521                         goto repeat;
3522                 }
3523         }
3524
3525         f2fs_wait_on_page_writeback(page, DATA, false, true);
3526
3527         if (len == PAGE_SIZE || PageUptodate(page))
3528                 return 0;
3529
3530         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3531             !f2fs_verity_in_progress(inode)) {
3532                 zero_user_segment(page, len, PAGE_SIZE);
3533                 return 0;
3534         }
3535
3536         if (blkaddr == NEW_ADDR) {
3537                 zero_user_segment(page, 0, PAGE_SIZE);
3538                 SetPageUptodate(page);
3539         } else {
3540                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3541                                 DATA_GENERIC_ENHANCE_READ)) {
3542                         err = -EFSCORRUPTED;
3543                         goto fail;
3544                 }
3545                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3546                 if (err)
3547                         goto fail;
3548
3549                 lock_page(page);
3550                 if (unlikely(page->mapping != mapping)) {
3551                         f2fs_put_page(page, 1);
3552                         goto repeat;
3553                 }
3554                 if (unlikely(!PageUptodate(page))) {
3555                         err = -EIO;
3556                         goto fail;
3557                 }
3558         }
3559         return 0;
3560
3561 fail:
3562         f2fs_put_page(page, 1);
3563         f2fs_write_failed(inode, pos + len);
3564         return err;
3565 }
3566
3567 static int f2fs_write_end(struct file *file,
3568                         struct address_space *mapping,
3569                         loff_t pos, unsigned len, unsigned copied,
3570                         struct page *page, void *fsdata)
3571 {
3572         struct inode *inode = page->mapping->host;
3573
3574         trace_f2fs_write_end(inode, pos, len, copied);
3575
3576         /*
3577          * This should be come from len == PAGE_SIZE, and we expect copied
3578          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3579          * let generic_perform_write() try to copy data again through copied=0.
3580          */
3581         if (!PageUptodate(page)) {
3582                 if (unlikely(copied != len))
3583                         copied = 0;
3584                 else
3585                         SetPageUptodate(page);
3586         }
3587
3588 #ifdef CONFIG_F2FS_FS_COMPRESSION
3589         /* overwrite compressed file */
3590         if (f2fs_compressed_file(inode) && fsdata) {
3591                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3592                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3593
3594                 if (pos + copied > i_size_read(inode) &&
3595                                 !f2fs_verity_in_progress(inode))
3596                         f2fs_i_size_write(inode, pos + copied);
3597                 return copied;
3598         }
3599 #endif
3600
3601         if (!copied)
3602                 goto unlock_out;
3603
3604         set_page_dirty(page);
3605
3606         if (pos + copied > i_size_read(inode) &&
3607             !f2fs_verity_in_progress(inode)) {
3608                 f2fs_i_size_write(inode, pos + copied);
3609                 if (f2fs_is_atomic_file(inode))
3610                         f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3611                                         pos + copied);
3612         }
3613 unlock_out:
3614         f2fs_put_page(page, 1);
3615         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3616         return copied;
3617 }
3618
3619 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3620 {
3621         struct inode *inode = folio->mapping->host;
3622         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3623
3624         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3625                                 (offset || length != folio_size(folio)))
3626                 return;
3627
3628         if (folio_test_dirty(folio)) {
3629                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3630                         dec_page_count(sbi, F2FS_DIRTY_META);
3631                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3632                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3633                 } else {
3634                         inode_dec_dirty_pages(inode);
3635                         f2fs_remove_dirty_inode(inode);
3636                 }
3637         }
3638
3639         clear_page_private_gcing(&folio->page);
3640
3641         if (test_opt(sbi, COMPRESS_CACHE) &&
3642                         inode->i_ino == F2FS_COMPRESS_INO(sbi))
3643                 clear_page_private_data(&folio->page);
3644
3645         folio_detach_private(folio);
3646 }
3647
3648 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3649 {
3650         struct f2fs_sb_info *sbi;
3651
3652         /* If this is dirty folio, keep private data */
3653         if (folio_test_dirty(folio))
3654                 return false;
3655
3656         sbi = F2FS_M_SB(folio->mapping);
3657         if (test_opt(sbi, COMPRESS_CACHE)) {
3658                 struct inode *inode = folio->mapping->host;
3659
3660                 if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3661                         clear_page_private_data(&folio->page);
3662         }
3663
3664         clear_page_private_gcing(&folio->page);
3665
3666         folio_detach_private(folio);
3667         return true;
3668 }
3669
3670 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3671                 struct folio *folio)
3672 {
3673         struct inode *inode = mapping->host;
3674
3675         trace_f2fs_set_page_dirty(&folio->page, DATA);
3676
3677         if (!folio_test_uptodate(folio))
3678                 folio_mark_uptodate(folio);
3679         BUG_ON(folio_test_swapcache(folio));
3680
3681         if (!folio_test_dirty(folio)) {
3682                 filemap_dirty_folio(mapping, folio);
3683                 f2fs_update_dirty_folio(inode, folio);
3684                 return true;
3685         }
3686         return false;
3687 }
3688
3689
3690 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3691 {
3692 #ifdef CONFIG_F2FS_FS_COMPRESSION
3693         struct dnode_of_data dn;
3694         sector_t start_idx, blknr = 0;
3695         int ret;
3696
3697         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3698
3699         set_new_dnode(&dn, inode, NULL, NULL, 0);
3700         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3701         if (ret)
3702                 return 0;
3703
3704         if (dn.data_blkaddr != COMPRESS_ADDR) {
3705                 dn.ofs_in_node += block - start_idx;
3706                 blknr = f2fs_data_blkaddr(&dn);
3707                 if (!__is_valid_data_blkaddr(blknr))
3708                         blknr = 0;
3709         }
3710
3711         f2fs_put_dnode(&dn);
3712         return blknr;
3713 #else
3714         return 0;
3715 #endif
3716 }
3717
3718
3719 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3720 {
3721         struct inode *inode = mapping->host;
3722         sector_t blknr = 0;
3723
3724         if (f2fs_has_inline_data(inode))
3725                 goto out;
3726
3727         /* make sure allocating whole blocks */
3728         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3729                 filemap_write_and_wait(mapping);
3730
3731         /* Block number less than F2FS MAX BLOCKS */
3732         if (unlikely(block >= max_file_blocks(inode)))
3733                 goto out;
3734
3735         if (f2fs_compressed_file(inode)) {
3736                 blknr = f2fs_bmap_compress(inode, block);
3737         } else {
3738                 struct f2fs_map_blocks map;
3739
3740                 memset(&map, 0, sizeof(map));
3741                 map.m_lblk = block;
3742                 map.m_len = 1;
3743                 map.m_next_pgofs = NULL;
3744                 map.m_seg_type = NO_CHECK_TYPE;
3745
3746                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3747                         blknr = map.m_pblk;
3748         }
3749 out:
3750         trace_f2fs_bmap(inode, block, blknr);
3751         return blknr;
3752 }
3753
3754 #ifdef CONFIG_MIGRATION
3755 #include <linux/migrate.h>
3756
3757 int f2fs_migrate_page(struct address_space *mapping,
3758                 struct page *newpage, struct page *page, enum migrate_mode mode)
3759 {
3760         int rc, extra_count = 0;
3761
3762         BUG_ON(PageWriteback(page));
3763
3764         rc = migrate_page_move_mapping(mapping, newpage,
3765                                 page, extra_count);
3766         if (rc != MIGRATEPAGE_SUCCESS)
3767                 return rc;
3768
3769         /* guarantee to start from no stale private field */
3770         set_page_private(newpage, 0);
3771         if (PagePrivate(page)) {
3772                 set_page_private(newpage, page_private(page));
3773                 SetPagePrivate(newpage);
3774                 get_page(newpage);
3775
3776                 set_page_private(page, 0);
3777                 ClearPagePrivate(page);
3778                 put_page(page);
3779         }
3780
3781         if (mode != MIGRATE_SYNC_NO_COPY)
3782                 migrate_page_copy(newpage, page);
3783         else
3784                 migrate_page_states(newpage, page);
3785
3786         return MIGRATEPAGE_SUCCESS;
3787 }
3788 #endif
3789
3790 #ifdef CONFIG_SWAP
3791 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3792                                                         unsigned int blkcnt)
3793 {
3794         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3795         unsigned int blkofs;
3796         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3797         unsigned int secidx = start_blk / blk_per_sec;
3798         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3799         int ret = 0;
3800
3801         f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3802         filemap_invalidate_lock(inode->i_mapping);
3803
3804         set_inode_flag(inode, FI_ALIGNED_WRITE);
3805         set_inode_flag(inode, FI_OPU_WRITE);
3806
3807         for (; secidx < end_sec; secidx++) {
3808                 f2fs_down_write(&sbi->pin_sem);
3809
3810                 f2fs_lock_op(sbi);
3811                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3812                 f2fs_unlock_op(sbi);
3813
3814                 set_inode_flag(inode, FI_SKIP_WRITES);
3815
3816                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3817                         struct page *page;
3818                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3819
3820                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3821                         if (IS_ERR(page)) {
3822                                 f2fs_up_write(&sbi->pin_sem);
3823                                 ret = PTR_ERR(page);
3824                                 goto done;
3825                         }
3826
3827                         set_page_dirty(page);
3828                         f2fs_put_page(page, 1);
3829                 }
3830
3831                 clear_inode_flag(inode, FI_SKIP_WRITES);
3832
3833                 ret = filemap_fdatawrite(inode->i_mapping);
3834
3835                 f2fs_up_write(&sbi->pin_sem);
3836
3837                 if (ret)
3838                         break;
3839         }
3840
3841 done:
3842         clear_inode_flag(inode, FI_SKIP_WRITES);
3843         clear_inode_flag(inode, FI_OPU_WRITE);
3844         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3845
3846         filemap_invalidate_unlock(inode->i_mapping);
3847         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3848
3849         return ret;
3850 }
3851
3852 static int check_swap_activate(struct swap_info_struct *sis,
3853                                 struct file *swap_file, sector_t *span)
3854 {
3855         struct address_space *mapping = swap_file->f_mapping;
3856         struct inode *inode = mapping->host;
3857         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3858         sector_t cur_lblock;
3859         sector_t last_lblock;
3860         sector_t pblock;
3861         sector_t lowest_pblock = -1;
3862         sector_t highest_pblock = 0;
3863         int nr_extents = 0;
3864         unsigned long nr_pblocks;
3865         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3866         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3867         unsigned int not_aligned = 0;
3868         int ret = 0;
3869
3870         /*
3871          * Map all the blocks into the extent list.  This code doesn't try
3872          * to be very smart.
3873          */
3874         cur_lblock = 0;
3875         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3876
3877         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3878                 struct f2fs_map_blocks map;
3879 retry:
3880                 cond_resched();
3881
3882                 memset(&map, 0, sizeof(map));
3883                 map.m_lblk = cur_lblock;
3884                 map.m_len = last_lblock - cur_lblock;
3885                 map.m_next_pgofs = NULL;
3886                 map.m_next_extent = NULL;
3887                 map.m_seg_type = NO_CHECK_TYPE;
3888                 map.m_may_create = false;
3889
3890                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3891                 if (ret)
3892                         goto out;
3893
3894                 /* hole */
3895                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3896                         f2fs_err(sbi, "Swapfile has holes");
3897                         ret = -EINVAL;
3898                         goto out;
3899                 }
3900
3901                 pblock = map.m_pblk;
3902                 nr_pblocks = map.m_len;
3903
3904                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3905                                 nr_pblocks & sec_blks_mask) {
3906                         not_aligned++;
3907
3908                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3909                         if (cur_lblock + nr_pblocks > sis->max)
3910                                 nr_pblocks -= blks_per_sec;
3911
3912                         if (!nr_pblocks) {
3913                                 /* this extent is last one */
3914                                 nr_pblocks = map.m_len;
3915                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3916                                 goto next;
3917                         }
3918
3919                         ret = f2fs_migrate_blocks(inode, cur_lblock,
3920                                                         nr_pblocks);
3921                         if (ret)
3922                                 goto out;
3923                         goto retry;
3924                 }
3925 next:
3926                 if (cur_lblock + nr_pblocks >= sis->max)
3927                         nr_pblocks = sis->max - cur_lblock;
3928
3929                 if (cur_lblock) {       /* exclude the header page */
3930                         if (pblock < lowest_pblock)
3931                                 lowest_pblock = pblock;
3932                         if (pblock + nr_pblocks - 1 > highest_pblock)
3933                                 highest_pblock = pblock + nr_pblocks - 1;
3934                 }
3935
3936                 /*
3937                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3938                  */
3939                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3940                 if (ret < 0)
3941                         goto out;
3942                 nr_extents += ret;
3943                 cur_lblock += nr_pblocks;
3944         }
3945         ret = nr_extents;
3946         *span = 1 + highest_pblock - lowest_pblock;
3947         if (cur_lblock == 0)
3948                 cur_lblock = 1; /* force Empty message */
3949         sis->max = cur_lblock;
3950         sis->pages = cur_lblock - 1;
3951         sis->highest_bit = cur_lblock - 1;
3952 out:
3953         if (not_aligned)
3954                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3955                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
3956         return ret;
3957 }
3958
3959 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3960                                 sector_t *span)
3961 {
3962         struct inode *inode = file_inode(file);
3963         int ret;
3964
3965         if (!S_ISREG(inode->i_mode))
3966                 return -EINVAL;
3967
3968         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3969                 return -EROFS;
3970
3971         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3972                 f2fs_err(F2FS_I_SB(inode),
3973                         "Swapfile not supported in LFS mode");
3974                 return -EINVAL;
3975         }
3976
3977         ret = f2fs_convert_inline_inode(inode);
3978         if (ret)
3979                 return ret;
3980
3981         if (!f2fs_disable_compressed_file(inode))
3982                 return -EINVAL;
3983
3984         f2fs_precache_extents(inode);
3985
3986         ret = check_swap_activate(sis, file, span);
3987         if (ret < 0)
3988                 return ret;
3989
3990         set_inode_flag(inode, FI_PIN_FILE);
3991         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3992         return ret;
3993 }
3994
3995 static void f2fs_swap_deactivate(struct file *file)
3996 {
3997         struct inode *inode = file_inode(file);
3998
3999         clear_inode_flag(inode, FI_PIN_FILE);
4000 }
4001 #else
4002 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4003                                 sector_t *span)
4004 {
4005         return -EOPNOTSUPP;
4006 }
4007
4008 static void f2fs_swap_deactivate(struct file *file)
4009 {
4010 }
4011 #endif
4012
4013 const struct address_space_operations f2fs_dblock_aops = {
4014         .read_folio     = f2fs_read_data_folio,
4015         .readahead      = f2fs_readahead,
4016         .writepage      = f2fs_write_data_page,
4017         .writepages     = f2fs_write_data_pages,
4018         .write_begin    = f2fs_write_begin,
4019         .write_end      = f2fs_write_end,
4020         .dirty_folio    = f2fs_dirty_data_folio,
4021         .invalidate_folio = f2fs_invalidate_folio,
4022         .release_folio  = f2fs_release_folio,
4023         .direct_IO      = noop_direct_IO,
4024         .bmap           = f2fs_bmap,
4025         .swap_activate  = f2fs_swap_activate,
4026         .swap_deactivate = f2fs_swap_deactivate,
4027 #ifdef CONFIG_MIGRATION
4028         .migratepage    = f2fs_migrate_page,
4029 #endif
4030 };
4031
4032 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4033 {
4034         struct address_space *mapping = page_mapping(page);
4035         unsigned long flags;
4036
4037         xa_lock_irqsave(&mapping->i_pages, flags);
4038         __xa_clear_mark(&mapping->i_pages, page_index(page),
4039                                                 PAGECACHE_TAG_DIRTY);
4040         xa_unlock_irqrestore(&mapping->i_pages, flags);
4041 }
4042
4043 int __init f2fs_init_post_read_processing(void)
4044 {
4045         bio_post_read_ctx_cache =
4046                 kmem_cache_create("f2fs_bio_post_read_ctx",
4047                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4048         if (!bio_post_read_ctx_cache)
4049                 goto fail;
4050         bio_post_read_ctx_pool =
4051                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4052                                          bio_post_read_ctx_cache);
4053         if (!bio_post_read_ctx_pool)
4054                 goto fail_free_cache;
4055         return 0;
4056
4057 fail_free_cache:
4058         kmem_cache_destroy(bio_post_read_ctx_cache);
4059 fail:
4060         return -ENOMEM;
4061 }
4062
4063 void f2fs_destroy_post_read_processing(void)
4064 {
4065         mempool_destroy(bio_post_read_ctx_pool);
4066         kmem_cache_destroy(bio_post_read_ctx_cache);
4067 }
4068
4069 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4070 {
4071         if (!f2fs_sb_has_encrypt(sbi) &&
4072                 !f2fs_sb_has_verity(sbi) &&
4073                 !f2fs_sb_has_compression(sbi))
4074                 return 0;
4075
4076         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4077                                                  WQ_UNBOUND | WQ_HIGHPRI,
4078                                                  num_online_cpus());
4079         if (!sbi->post_read_wq)
4080                 return -ENOMEM;
4081         return 0;
4082 }
4083
4084 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4085 {
4086         if (sbi->post_read_wq)
4087                 destroy_workqueue(sbi->post_read_wq);
4088 }
4089
4090 int __init f2fs_init_bio_entry_cache(void)
4091 {
4092         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4093                         sizeof(struct bio_entry));
4094         if (!bio_entry_slab)
4095                 return -ENOMEM;
4096         return 0;
4097 }
4098
4099 void f2fs_destroy_bio_entry_cache(void)
4100 {
4101         kmem_cache_destroy(bio_entry_slab);
4102 }
4103
4104 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4105                             unsigned int flags, struct iomap *iomap,
4106                             struct iomap *srcmap)
4107 {
4108         struct f2fs_map_blocks map = {};
4109         pgoff_t next_pgofs = 0;
4110         int err;
4111
4112         map.m_lblk = bytes_to_blks(inode, offset);
4113         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4114         map.m_next_pgofs = &next_pgofs;
4115         map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4116         if (flags & IOMAP_WRITE)
4117                 map.m_may_create = true;
4118
4119         err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4120                               F2FS_GET_BLOCK_DIO);
4121         if (err)
4122                 return err;
4123
4124         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4125
4126         /*
4127          * When inline encryption is enabled, sometimes I/O to an encrypted file
4128          * has to be broken up to guarantee DUN contiguity.  Handle this by
4129          * limiting the length of the mapping returned.
4130          */
4131         map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4132
4133         if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4134                 iomap->length = blks_to_bytes(inode, map.m_len);
4135                 if (map.m_flags & F2FS_MAP_MAPPED) {
4136                         iomap->type = IOMAP_MAPPED;
4137                         iomap->flags |= IOMAP_F_MERGED;
4138                 } else {
4139                         iomap->type = IOMAP_UNWRITTEN;
4140                 }
4141                 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4142                         return -EINVAL;
4143
4144                 iomap->bdev = map.m_bdev;
4145                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4146         } else {
4147                 iomap->length = blks_to_bytes(inode, next_pgofs) -
4148                                 iomap->offset;
4149                 iomap->type = IOMAP_HOLE;
4150                 iomap->addr = IOMAP_NULL_ADDR;
4151         }
4152
4153         if (map.m_flags & F2FS_MAP_NEW)
4154                 iomap->flags |= IOMAP_F_NEW;
4155         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4156             offset + length > i_size_read(inode))
4157                 iomap->flags |= IOMAP_F_DIRTY;
4158
4159         return 0;
4160 }
4161
4162 const struct iomap_ops f2fs_iomap_ops = {
4163         .iomap_begin    = f2fs_iomap_begin,
4164 };