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         return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  43                                         0, BIOSET_NEED_BVECS);
  44 }
  45
  46 void f2fs_destroy_bioset(void)
  47 {
  48         bioset_exit(&f2fs_bioset);
  49 }
  50
  51 bool f2fs_is_cp_guaranteed(struct page *page)
  52 {
  53         struct address_space *mapping = page->mapping;
  54         struct inode *inode;
  55         struct f2fs_sb_info *sbi;
  56
  57         if (!mapping)
  58                 return false;
  59
  60         inode = mapping->host;
  61         sbi = F2FS_I_SB(inode);
  62
  63         if (inode->i_ino == F2FS_META_INO(sbi) ||
  64                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  65                         S_ISDIR(inode->i_mode))
  66                 return true;
  67
  68         if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
  69                         page_private_gcing(page))
  70                 return true;
  71         return false;
  72 }
  73
  74 static enum count_type __read_io_type(struct page *page)
  75 {
  76         struct address_space *mapping = page_file_mapping(page);
  77
  78         if (mapping) {
  79                 struct inode *inode = mapping->host;
  80                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  81
  82                 if (inode->i_ino == F2FS_META_INO(sbi))
  83                         return F2FS_RD_META;
  84
  85                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  86                         return F2FS_RD_NODE;
  87         }
  88         return F2FS_RD_DATA;
  89 }
  90
  91 /* postprocessing steps for read bios */
  92 enum bio_post_read_step {
  93 #ifdef CONFIG_FS_ENCRYPTION
  94         STEP_DECRYPT    = BIT(0),
  95 #else
  96         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
  97 #endif
  98 #ifdef CONFIG_F2FS_FS_COMPRESSION
  99         STEP_DECOMPRESS = BIT(1),
 100 #else
 101         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 102 #endif
 103 #ifdef CONFIG_FS_VERITY
 104         STEP_VERITY     = BIT(2),
 105 #else
 106         STEP_VERITY     = 0,    /* compile out the verity-related code */
 107 #endif
 108 };
 109
 110 struct bio_post_read_ctx {
 111         struct bio *bio;
 112         struct f2fs_sb_info *sbi;
 113         struct work_struct work;
 114         unsigned int enabled_steps;
 115         /*
 116          * decompression_attempted keeps track of whether
 117          * f2fs_end_read_compressed_page() has been called on the pages in the
 118          * bio that belong to a compressed cluster yet.
 119          */
 120         bool decompression_attempted;
 121         block_t fs_blkaddr;
 122 };
 123
 124 /*
 125  * Update and unlock a bio's pages, and free the bio.
 126  *
 127  * This marks pages up-to-date only if there was no error in the bio (I/O error,
 128  * decryption error, or verity error), as indicated by bio->bi_status.
 129  *
 130  * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk)
 131  * aren't marked up-to-date here, as decompression is done on a per-compression-
 132  * cluster basis rather than a per-bio basis.  Instead, we only must do two
 133  * things for each compressed page here: call f2fs_end_read_compressed_page()
 134  * with failed=true if an error occurred before it would have normally gotten
 135  * called (i.e., I/O error or decryption error, but *not* verity error), and
 136  * release the bio's reference to the decompress_io_ctx of the page's cluster.
 137  */
 138 static void f2fs_finish_read_bio(struct bio *bio, bool in_task)
 139 {
 140         struct bio_vec *bv;
 141         struct bvec_iter_all iter_all;
 142         struct bio_post_read_ctx *ctx = bio->bi_private;
 143
 144         bio_for_each_segment_all(bv, bio, iter_all) {
 145                 struct page *page = bv->bv_page;
 146
 147                 if (f2fs_is_compressed_page(page)) {
 148                         if (ctx && !ctx->decompression_attempted)
 149                                 f2fs_end_read_compressed_page(page, true, 0,
 150                                                         in_task);
 151                         f2fs_put_page_dic(page, in_task);
 152                         continue;
 153                 }
 154
 155                 if (bio->bi_status)
 156                         ClearPageUptodate(page);
 157                 else
 158                         SetPageUptodate(page);
 159                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 160                 unlock_page(page);
 161         }
 162
 163         if (ctx)
 164                 mempool_free(ctx, bio_post_read_ctx_pool);
 165         bio_put(bio);
 166 }
 167
 168 static void f2fs_verify_bio(struct work_struct *work)
 169 {
 170         struct bio_post_read_ctx *ctx =
 171                 container_of(work, struct bio_post_read_ctx, work);
 172         struct bio *bio = ctx->bio;
 173         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 174
 175         /*
 176          * fsverity_verify_bio() may call readahead() again, and while verity
 177          * will be disabled for this, decryption and/or decompression may still
 178          * be needed, resulting in another bio_post_read_ctx being allocated.
 179          * So to prevent deadlocks we need to release the current ctx to the
 180          * mempool first.  This assumes that verity is the last post-read step.
 181          */
 182         mempool_free(ctx, bio_post_read_ctx_pool);
 183         bio->bi_private = NULL;
 184
 185         /*
 186          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 187          * as those were handled separately by f2fs_end_read_compressed_page().
 188          */
 189         if (may_have_compressed_pages) {
 190                 struct bio_vec *bv;
 191                 struct bvec_iter_all iter_all;
 192
 193                 bio_for_each_segment_all(bv, bio, iter_all) {
 194                         struct page *page = bv->bv_page;
 195
 196                         if (!f2fs_is_compressed_page(page) &&
 197                             !fsverity_verify_page(page)) {
 198                                 bio->bi_status = BLK_STS_IOERR;
 199                                 break;
 200                         }
 201                 }
 202         } else {
 203                 fsverity_verify_bio(bio);
 204         }
 205
 206         f2fs_finish_read_bio(bio, true);
 207 }
 208
 209 /*
 210  * If the bio's data needs to be verified with fs-verity, then enqueue the
 211  * verity work for the bio.  Otherwise finish the bio now.
 212  *
 213  * Note that to avoid deadlocks, the verity work can't be done on the
 214  * decryption/decompression workqueue.  This is because verifying the data pages
 215  * can involve reading verity metadata pages from the file, and these verity
 216  * metadata pages may be encrypted and/or compressed.
 217  */
 218 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task)
 219 {
 220         struct bio_post_read_ctx *ctx = bio->bi_private;
 221
 222         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 223                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 224                 fsverity_enqueue_verify_work(&ctx->work);
 225         } else {
 226                 f2fs_finish_read_bio(bio, in_task);
 227         }
 228 }
 229
 230 /*
 231  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 232  * remaining page was read by @ctx->bio.
 233  *
 234  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 235  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 236  * that the bio includes at least one compressed page.  The actual decompression
 237  * is done on a per-cluster basis, not a per-bio basis.
 238  */
 239 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx,
 240                 bool in_task)
 241 {
 242         struct bio_vec *bv;
 243         struct bvec_iter_all iter_all;
 244         bool all_compressed = true;
 245         block_t blkaddr = ctx->fs_blkaddr;
 246
 247         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 248                 struct page *page = bv->bv_page;
 249
 250                 if (f2fs_is_compressed_page(page))
 251                         f2fs_end_read_compressed_page(page, false, blkaddr,
 252                                                       in_task);
 253                 else
 254                         all_compressed = false;
 255
 256                 blkaddr++;
 257         }
 258
 259         ctx->decompression_attempted = true;
 260
 261         /*
 262          * Optimization: if all the bio's pages are compressed, then scheduling
 263          * the per-bio verity work is unnecessary, as verity will be fully
 264          * handled at the compression cluster level.
 265          */
 266         if (all_compressed)
 267                 ctx->enabled_steps &= ~STEP_VERITY;
 268 }
 269
 270 static void f2fs_post_read_work(struct work_struct *work)
 271 {
 272         struct bio_post_read_ctx *ctx =
 273                 container_of(work, struct bio_post_read_ctx, work);
 274         struct bio *bio = ctx->bio;
 275
 276         if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) {
 277                 f2fs_finish_read_bio(bio, true);
 278                 return;
 279         }
 280
 281         if (ctx->enabled_steps & STEP_DECOMPRESS)
 282                 f2fs_handle_step_decompress(ctx, true);
 283
 284         f2fs_verify_and_finish_bio(bio, true);
 285 }
 286
 287 static void f2fs_read_end_io(struct bio *bio)
 288 {
 289         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 290         struct bio_post_read_ctx *ctx;
 291         bool intask = in_task();
 292
 293         iostat_update_and_unbind_ctx(bio);
 294         ctx = bio->bi_private;
 295
 296         if (time_to_inject(sbi, FAULT_READ_IO))
 297                 bio->bi_status = BLK_STS_IOERR;
 298
 299         if (bio->bi_status) {
 300                 f2fs_finish_read_bio(bio, intask);
 301                 return;
 302         }
 303
 304         if (ctx) {
 305                 unsigned int enabled_steps = ctx->enabled_steps &
 306                                         (STEP_DECRYPT | STEP_DECOMPRESS);
 307
 308                 /*
 309                  * If we have only decompression step between decompression and
 310                  * decrypt, we don't need post processing for this.
 311                  */
 312                 if (enabled_steps == STEP_DECOMPRESS &&
 313                                 !f2fs_low_mem_mode(sbi)) {
 314                         f2fs_handle_step_decompress(ctx, intask);
 315                 } else if (enabled_steps) {
 316                         INIT_WORK(&ctx->work, f2fs_post_read_work);
 317                         queue_work(ctx->sbi->post_read_wq, &ctx->work);
 318                         return;
 319                 }
 320         }
 321
 322         f2fs_verify_and_finish_bio(bio, intask);
 323 }
 324
 325 static void f2fs_write_end_io(struct bio *bio)
 326 {
 327         struct f2fs_sb_info *sbi;
 328         struct bio_vec *bvec;
 329         struct bvec_iter_all iter_all;
 330
 331         iostat_update_and_unbind_ctx(bio);
 332         sbi = bio->bi_private;
 333
 334         if (time_to_inject(sbi, FAULT_WRITE_IO))
 335                 bio->bi_status = BLK_STS_IOERR;
 336
 337         bio_for_each_segment_all(bvec, bio, iter_all) {
 338                 struct page *page = bvec->bv_page;
 339                 enum count_type type = WB_DATA_TYPE(page, false);
 340
 341                 if (page_private_dummy(page)) {
 342                         clear_page_private_dummy(page);
 343                         unlock_page(page);
 344                         mempool_free(page, sbi->write_io_dummy);
 345
 346                         if (unlikely(bio->bi_status))
 347                                 f2fs_stop_checkpoint(sbi, true,
 348                                                 STOP_CP_REASON_WRITE_FAIL);
 349                         continue;
 350                 }
 351
 352                 fscrypt_finalize_bounce_page(&page);
 353
 354 #ifdef CONFIG_F2FS_FS_COMPRESSION
 355                 if (f2fs_is_compressed_page(page)) {
 356                         f2fs_compress_write_end_io(bio, page);
 357                         continue;
 358                 }
 359 #endif
 360
 361                 if (unlikely(bio->bi_status)) {
 362                         mapping_set_error(page->mapping, -EIO);
 363                         if (type == F2FS_WB_CP_DATA)
 364                                 f2fs_stop_checkpoint(sbi, true,
 365                                                 STOP_CP_REASON_WRITE_FAIL);
 366                 }
 367
 368                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 369                                         page->index != nid_of_node(page));
 370
 371                 dec_page_count(sbi, type);
 372                 if (f2fs_in_warm_node_list(sbi, page))
 373                         f2fs_del_fsync_node_entry(sbi, page);
 374                 clear_page_private_gcing(page);
 375                 end_page_writeback(page);
 376         }
 377         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 378                                 wq_has_sleeper(&sbi->cp_wait))
 379                 wake_up(&sbi->cp_wait);
 380
 381         bio_put(bio);
 382 }
 383
 384 #ifdef CONFIG_BLK_DEV_ZONED
 385 static void f2fs_zone_write_end_io(struct bio *bio)
 386 {
 387         struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
 388
 389         bio->bi_private = io->bi_private;
 390         complete(&io->zone_wait);
 391         f2fs_write_end_io(bio);
 392 }
 393 #endif
 394
 395 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 396                 block_t blk_addr, sector_t *sector)
 397 {
 398         struct block_device *bdev = sbi->sb->s_bdev;
 399         int i;
 400
 401         if (f2fs_is_multi_device(sbi)) {
 402                 for (i = 0; i < sbi->s_ndevs; i++) {
 403                         if (FDEV(i).start_blk <= blk_addr &&
 404                             FDEV(i).end_blk >= blk_addr) {
 405                                 blk_addr -= FDEV(i).start_blk;
 406                                 bdev = FDEV(i).bdev;
 407                                 break;
 408                         }
 409                 }
 410         }
 411
 412         if (sector)
 413                 *sector = SECTOR_FROM_BLOCK(blk_addr);
 414         return bdev;
 415 }
 416
 417 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 418 {
 419         int i;
 420
 421         if (!f2fs_is_multi_device(sbi))
 422                 return 0;
 423
 424         for (i = 0; i < sbi->s_ndevs; i++)
 425                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 426                         return i;
 427         return 0;
 428 }
 429
 430 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
 431 {
 432         unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0);
 433         unsigned int fua_flag, meta_flag, io_flag;
 434         blk_opf_t op_flags = 0;
 435
 436         if (fio->op != REQ_OP_WRITE)
 437                 return 0;
 438         if (fio->type == DATA)
 439                 io_flag = fio->sbi->data_io_flag;
 440         else if (fio->type == NODE)
 441                 io_flag = fio->sbi->node_io_flag;
 442         else
 443                 return 0;
 444
 445         fua_flag = io_flag & temp_mask;
 446         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 447
 448         /*
 449          * data/node io flag bits per temp:
 450          *      REQ_META     |      REQ_FUA      |
 451          *    5 |    4 |   3 |    2 |    1 |   0 |
 452          * Cold | Warm | Hot | Cold | Warm | Hot |
 453          */
 454         if (BIT(fio->temp) & meta_flag)
 455                 op_flags |= REQ_META;
 456         if (BIT(fio->temp) & fua_flag)
 457                 op_flags |= REQ_FUA;
 458         return op_flags;
 459 }
 460
 461 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 462 {
 463         struct f2fs_sb_info *sbi = fio->sbi;
 464         struct block_device *bdev;
 465         sector_t sector;
 466         struct bio *bio;
 467
 468         bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 469         bio = bio_alloc_bioset(bdev, npages,
 470                                 fio->op | fio->op_flags | f2fs_io_flags(fio),
 471                                 GFP_NOIO, &f2fs_bioset);
 472         bio->bi_iter.bi_sector = sector;
 473         if (is_read_io(fio->op)) {
 474                 bio->bi_end_io = f2fs_read_end_io;
 475                 bio->bi_private = NULL;
 476         } else {
 477                 bio->bi_end_io = f2fs_write_end_io;
 478                 bio->bi_private = sbi;
 479         }
 480         iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 481
 482         if (fio->io_wbc)
 483                 wbc_init_bio(fio->io_wbc, bio);
 484
 485         return bio;
 486 }
 487
 488 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 489                                   pgoff_t first_idx,
 490                                   const struct f2fs_io_info *fio,
 491                                   gfp_t gfp_mask)
 492 {
 493         /*
 494          * The f2fs garbage collector sets ->encrypted_page when it wants to
 495          * read/write raw data without encryption.
 496          */
 497         if (!fio || !fio->encrypted_page)
 498                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 499 }
 500
 501 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 502                                      pgoff_t next_idx,
 503                                      const struct f2fs_io_info *fio)
 504 {
 505         /*
 506          * The f2fs garbage collector sets ->encrypted_page when it wants to
 507          * read/write raw data without encryption.
 508          */
 509         if (fio && fio->encrypted_page)
 510                 return !bio_has_crypt_ctx(bio);
 511
 512         return fscrypt_mergeable_bio(bio, inode, next_idx);
 513 }
 514
 515 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 516                                  enum page_type type)
 517 {
 518         WARN_ON_ONCE(!is_read_io(bio_op(bio)));
 519         trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 520
 521         iostat_update_submit_ctx(bio, type);
 522         submit_bio(bio);
 523 }
 524
 525 static void f2fs_align_write_bio(struct f2fs_sb_info *sbi, struct bio *bio)
 526 {
 527         unsigned int start =
 528                 (bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS) % F2FS_IO_SIZE(sbi);
 529
 530         if (start == 0)
 531                 return;
 532
 533         /* fill dummy pages */
 534         for (; start < F2FS_IO_SIZE(sbi); start++) {
 535                 struct page *page =
 536                         mempool_alloc(sbi->write_io_dummy,
 537                                       GFP_NOIO | __GFP_NOFAIL);
 538                 f2fs_bug_on(sbi, !page);
 539
 540                 lock_page(page);
 541
 542                 zero_user_segment(page, 0, PAGE_SIZE);
 543                 set_page_private_dummy(page);
 544
 545                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 546                         f2fs_bug_on(sbi, 1);
 547         }
 548 }
 549
 550 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 551                                   enum page_type type)
 552 {
 553         WARN_ON_ONCE(is_read_io(bio_op(bio)));
 554
 555         if (type == DATA || type == NODE) {
 556                 if (f2fs_lfs_mode(sbi) && current->plug)
 557                         blk_finish_plug(current->plug);
 558
 559                 if (F2FS_IO_ALIGNED(sbi)) {
 560                         f2fs_align_write_bio(sbi, bio);
 561                         /*
 562                          * In the NODE case, we lose next block address chain.
 563                          * So, we need to do checkpoint in f2fs_sync_file.
 564                          */
 565                         if (type == NODE)
 566                                 set_sbi_flag(sbi, SBI_NEED_CP);
 567                 }
 568         }
 569
 570         trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 571         iostat_update_submit_ctx(bio, type);
 572         submit_bio(bio);
 573 }
 574
 575 static void __submit_merged_bio(struct f2fs_bio_info *io)
 576 {
 577         struct f2fs_io_info *fio = &io->fio;
 578
 579         if (!io->bio)
 580                 return;
 581
 582         if (is_read_io(fio->op)) {
 583                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 584                 f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
 585         } else {
 586                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 587                 f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
 588         }
 589         io->bio = NULL;
 590 }
 591
 592 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 593                                                 struct page *page, nid_t ino)
 594 {
 595         struct bio_vec *bvec;
 596         struct bvec_iter_all iter_all;
 597
 598         if (!bio)
 599                 return false;
 600
 601         if (!inode && !page && !ino)
 602                 return true;
 603
 604         bio_for_each_segment_all(bvec, bio, iter_all) {
 605                 struct page *target = bvec->bv_page;
 606
 607                 if (fscrypt_is_bounce_page(target)) {
 608                         target = fscrypt_pagecache_page(target);
 609                         if (IS_ERR(target))
 610                                 continue;
 611                 }
 612                 if (f2fs_is_compressed_page(target)) {
 613                         target = f2fs_compress_control_page(target);
 614                         if (IS_ERR(target))
 615                                 continue;
 616                 }
 617
 618                 if (inode && inode == target->mapping->host)
 619                         return true;
 620                 if (page && page == target)
 621                         return true;
 622                 if (ino && ino == ino_of_node(target))
 623                         return true;
 624         }
 625
 626         return false;
 627 }
 628
 629 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
 630 {
 631         int i;
 632
 633         for (i = 0; i < NR_PAGE_TYPE; i++) {
 634                 int n = (i == META) ? 1 : NR_TEMP_TYPE;
 635                 int j;
 636
 637                 sbi->write_io[i] = f2fs_kmalloc(sbi,
 638                                 array_size(n, sizeof(struct f2fs_bio_info)),
 639                                 GFP_KERNEL);
 640                 if (!sbi->write_io[i])
 641                         return -ENOMEM;
 642
 643                 for (j = HOT; j < n; j++) {
 644                         init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
 645                         sbi->write_io[i][j].sbi = sbi;
 646                         sbi->write_io[i][j].bio = NULL;
 647                         spin_lock_init(&sbi->write_io[i][j].io_lock);
 648                         INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
 649                         INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
 650                         init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
 651 #ifdef CONFIG_BLK_DEV_ZONED
 652                         init_completion(&sbi->write_io[i][j].zone_wait);
 653                         sbi->write_io[i][j].zone_pending_bio = NULL;
 654                         sbi->write_io[i][j].bi_private = NULL;
 655 #endif
 656                 }
 657         }
 658
 659         return 0;
 660 }
 661
 662 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 663                                 enum page_type type, enum temp_type temp)
 664 {
 665         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 666         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 667
 668         f2fs_down_write(&io->io_rwsem);
 669
 670         if (!io->bio)
 671                 goto unlock_out;
 672
 673         /* change META to META_FLUSH in the checkpoint procedure */
 674         if (type >= META_FLUSH) {
 675                 io->fio.type = META_FLUSH;
 676                 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 677                 if (!test_opt(sbi, NOBARRIER))
 678                         io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 679         }
 680         __submit_merged_bio(io);
 681 unlock_out:
 682         f2fs_up_write(&io->io_rwsem);
 683 }
 684
 685 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 686                                 struct inode *inode, struct page *page,
 687                                 nid_t ino, enum page_type type, bool force)
 688 {
 689         enum temp_type temp;
 690         bool ret = true;
 691
 692         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 693                 if (!force)     {
 694                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 695                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 696
 697                         f2fs_down_read(&io->io_rwsem);
 698                         ret = __has_merged_page(io->bio, inode, page, ino);
 699                         f2fs_up_read(&io->io_rwsem);
 700                 }
 701                 if (ret)
 702                         __f2fs_submit_merged_write(sbi, type, temp);
 703
 704                 /* TODO: use HOT temp only for meta pages now. */
 705                 if (type >= META)
 706                         break;
 707         }
 708 }
 709
 710 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 711 {
 712         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 713 }
 714
 715 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 716                                 struct inode *inode, struct page *page,
 717                                 nid_t ino, enum page_type type)
 718 {
 719         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 720 }
 721
 722 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 723 {
 724         f2fs_submit_merged_write(sbi, DATA);
 725         f2fs_submit_merged_write(sbi, NODE);
 726         f2fs_submit_merged_write(sbi, META);
 727 }
 728
 729 /*
 730  * Fill the locked page with data located in the block address.
 731  * A caller needs to unlock the page on failure.
 732  */
 733 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 734 {
 735         struct bio *bio;
 736         struct page *page = fio->encrypted_page ?
 737                         fio->encrypted_page : fio->page;
 738
 739         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 740                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 741                         META_GENERIC : DATA_GENERIC_ENHANCE))) {
 742                 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
 743                 return -EFSCORRUPTED;
 744         }
 745
 746         trace_f2fs_submit_page_bio(page, fio);
 747
 748         /* Allocate a new bio */
 749         bio = __bio_alloc(fio, 1);
 750
 751         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 752                                fio->page->index, fio, GFP_NOIO);
 753
 754         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 755                 bio_put(bio);
 756                 return -EFAULT;
 757         }
 758
 759         if (fio->io_wbc && !is_read_io(fio->op))
 760                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 761
 762         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 763                         __read_io_type(page) : WB_DATA_TYPE(fio->page, false));
 764
 765         if (is_read_io(bio_op(bio)))
 766                 f2fs_submit_read_bio(fio->sbi, bio, fio->type);
 767         else
 768                 f2fs_submit_write_bio(fio->sbi, bio, fio->type);
 769         return 0;
 770 }
 771
 772 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 773                                 block_t last_blkaddr, block_t cur_blkaddr)
 774 {
 775         if (unlikely(sbi->max_io_bytes &&
 776                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 777                 return false;
 778         if (last_blkaddr + 1 != cur_blkaddr)
 779                 return false;
 780         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 781 }
 782
 783 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 784                                                 struct f2fs_io_info *fio)
 785 {
 786         if (io->fio.op != fio->op)
 787                 return false;
 788         return io->fio.op_flags == fio->op_flags;
 789 }
 790
 791 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 792                                         struct f2fs_bio_info *io,
 793                                         struct f2fs_io_info *fio,
 794                                         block_t last_blkaddr,
 795                                         block_t cur_blkaddr)
 796 {
 797         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 798                 unsigned int filled_blocks =
 799                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 800                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 801                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 802
 803                 /* IOs in bio is aligned and left space of vectors is not enough */
 804                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 805                         return false;
 806         }
 807         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 808                 return false;
 809         return io_type_is_mergeable(io, fio);
 810 }
 811
 812 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 813                                 struct page *page, enum temp_type temp)
 814 {
 815         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 816         struct bio_entry *be;
 817
 818         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 819         be->bio = bio;
 820         bio_get(bio);
 821
 822         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 823                 f2fs_bug_on(sbi, 1);
 824
 825         f2fs_down_write(&io->bio_list_lock);
 826         list_add_tail(&be->list, &io->bio_list);
 827         f2fs_up_write(&io->bio_list_lock);
 828 }
 829
 830 static void del_bio_entry(struct bio_entry *be)
 831 {
 832         list_del(&be->list);
 833         kmem_cache_free(bio_entry_slab, be);
 834 }
 835
 836 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 837                                                         struct page *page)
 838 {
 839         struct f2fs_sb_info *sbi = fio->sbi;
 840         enum temp_type temp;
 841         bool found = false;
 842         int ret = -EAGAIN;
 843
 844         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 845                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 846                 struct list_head *head = &io->bio_list;
 847                 struct bio_entry *be;
 848
 849                 f2fs_down_write(&io->bio_list_lock);
 850                 list_for_each_entry(be, head, list) {
 851                         if (be->bio != *bio)
 852                                 continue;
 853
 854                         found = true;
 855
 856                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 857                                                             *fio->last_block,
 858                                                             fio->new_blkaddr));
 859                         if (f2fs_crypt_mergeable_bio(*bio,
 860                                         fio->page->mapping->host,
 861                                         fio->page->index, fio) &&
 862                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 863                                         PAGE_SIZE) {
 864                                 ret = 0;
 865                                 break;
 866                         }
 867
 868                         /* page can't be merged into bio; submit the bio */
 869                         del_bio_entry(be);
 870                         f2fs_submit_write_bio(sbi, *bio, DATA);
 871                         break;
 872                 }
 873                 f2fs_up_write(&io->bio_list_lock);
 874         }
 875
 876         if (ret) {
 877                 bio_put(*bio);
 878                 *bio = NULL;
 879         }
 880
 881         return ret;
 882 }
 883
 884 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 885                                         struct bio **bio, struct page *page)
 886 {
 887         enum temp_type temp;
 888         bool found = false;
 889         struct bio *target = bio ? *bio : NULL;
 890
 891         f2fs_bug_on(sbi, !target && !page);
 892
 893         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 894                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 895                 struct list_head *head = &io->bio_list;
 896                 struct bio_entry *be;
 897
 898                 if (list_empty(head))
 899                         continue;
 900
 901                 f2fs_down_read(&io->bio_list_lock);
 902                 list_for_each_entry(be, head, list) {
 903                         if (target)
 904                                 found = (target == be->bio);
 905                         else
 906                                 found = __has_merged_page(be->bio, NULL,
 907                                                                 page, 0);
 908                         if (found)
 909                                 break;
 910                 }
 911                 f2fs_up_read(&io->bio_list_lock);
 912
 913                 if (!found)
 914                         continue;
 915
 916                 found = false;
 917
 918                 f2fs_down_write(&io->bio_list_lock);
 919                 list_for_each_entry(be, head, list) {
 920                         if (target)
 921                                 found = (target == be->bio);
 922                         else
 923                                 found = __has_merged_page(be->bio, NULL,
 924                                                                 page, 0);
 925                         if (found) {
 926                                 target = be->bio;
 927                                 del_bio_entry(be);
 928                                 break;
 929                         }
 930                 }
 931                 f2fs_up_write(&io->bio_list_lock);
 932         }
 933
 934         if (found)
 935                 f2fs_submit_write_bio(sbi, target, DATA);
 936         if (bio && *bio) {
 937                 bio_put(*bio);
 938                 *bio = NULL;
 939         }
 940 }
 941
 942 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 943 {
 944         struct bio *bio = *fio->bio;
 945         struct page *page = fio->encrypted_page ?
 946                         fio->encrypted_page : fio->page;
 947
 948         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 949                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) {
 950                 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
 951                 return -EFSCORRUPTED;
 952         }
 953
 954         trace_f2fs_submit_page_bio(page, fio);
 955
 956         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 957                                                 fio->new_blkaddr))
 958                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 959 alloc_new:
 960         if (!bio) {
 961                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 962                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 963                                        fio->page->index, fio, GFP_NOIO);
 964
 965                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 966         } else {
 967                 if (add_ipu_page(fio, &bio, page))
 968                         goto alloc_new;
 969         }
 970
 971         if (fio->io_wbc)
 972                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 973
 974         inc_page_count(fio->sbi, WB_DATA_TYPE(page, false));
 975
 976         *fio->last_block = fio->new_blkaddr;
 977         *fio->bio = bio;
 978
 979         return 0;
 980 }
 981
 982 #ifdef CONFIG_BLK_DEV_ZONED
 983 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
 984 {
 985         int devi = 0;
 986
 987         if (f2fs_is_multi_device(sbi)) {
 988                 devi = f2fs_target_device_index(sbi, blkaddr);
 989                 if (blkaddr < FDEV(devi).start_blk ||
 990                     blkaddr > FDEV(devi).end_blk) {
 991                         f2fs_err(sbi, "Invalid block %x", blkaddr);
 992                         return false;
 993                 }
 994                 blkaddr -= FDEV(devi).start_blk;
 995         }
 996         return bdev_is_zoned(FDEV(devi).bdev) &&
 997                 f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
 998                 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
 999 }
1000 #endif
1001
1002 void f2fs_submit_page_write(struct f2fs_io_info *fio)
1003 {
1004         struct f2fs_sb_info *sbi = fio->sbi;
1005         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
1006         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
1007         struct page *bio_page;
1008         enum count_type type;
1009
1010         f2fs_bug_on(sbi, is_read_io(fio->op));
1011
1012         f2fs_down_write(&io->io_rwsem);
1013 next:
1014 #ifdef CONFIG_BLK_DEV_ZONED
1015         if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
1016                 wait_for_completion_io(&io->zone_wait);
1017                 bio_put(io->zone_pending_bio);
1018                 io->zone_pending_bio = NULL;
1019                 io->bi_private = NULL;
1020         }
1021 #endif
1022
1023         if (fio->in_list) {
1024                 spin_lock(&io->io_lock);
1025                 if (list_empty(&io->io_list)) {
1026                         spin_unlock(&io->io_lock);
1027                         goto out;
1028                 }
1029                 fio = list_first_entry(&io->io_list,
1030                                                 struct f2fs_io_info, list);
1031                 list_del(&fio->list);
1032                 spin_unlock(&io->io_lock);
1033         }
1034
1035         verify_fio_blkaddr(fio);
1036
1037         if (fio->encrypted_page)
1038                 bio_page = fio->encrypted_page;
1039         else if (fio->compressed_page)
1040                 bio_page = fio->compressed_page;
1041         else
1042                 bio_page = fio->page;
1043
1044         /* set submitted = true as a return value */
1045         fio->submitted = 1;
1046
1047         type = WB_DATA_TYPE(bio_page, fio->compressed_page);
1048         inc_page_count(sbi, type);
1049
1050         if (io->bio &&
1051             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
1052                               fio->new_blkaddr) ||
1053              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
1054                                        bio_page->index, fio)))
1055                 __submit_merged_bio(io);
1056 alloc_new:
1057         if (io->bio == NULL) {
1058                 if (F2FS_IO_ALIGNED(sbi) &&
1059                                 (fio->type == DATA || fio->type == NODE) &&
1060                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
1061                         dec_page_count(sbi, WB_DATA_TYPE(bio_page,
1062                                                 fio->compressed_page));
1063                         fio->retry = 1;
1064                         goto skip;
1065                 }
1066                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1067                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1068                                        bio_page->index, fio, GFP_NOIO);
1069                 io->fio = *fio;
1070         }
1071
1072         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1073                 __submit_merged_bio(io);
1074                 goto alloc_new;
1075         }
1076
1077         if (fio->io_wbc)
1078                 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1079
1080         io->last_block_in_bio = fio->new_blkaddr;
1081
1082         trace_f2fs_submit_page_write(fio->page, fio);
1083 #ifdef CONFIG_BLK_DEV_ZONED
1084         if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1085                         is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1086                 bio_get(io->bio);
1087                 reinit_completion(&io->zone_wait);
1088                 io->bi_private = io->bio->bi_private;
1089                 io->bio->bi_private = io;
1090                 io->bio->bi_end_io = f2fs_zone_write_end_io;
1091                 io->zone_pending_bio = io->bio;
1092                 __submit_merged_bio(io);
1093         }
1094 #endif
1095 skip:
1096         if (fio->in_list)
1097                 goto next;
1098 out:
1099         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1100                                 !f2fs_is_checkpoint_ready(sbi))
1101                 __submit_merged_bio(io);
1102         f2fs_up_write(&io->io_rwsem);
1103 }
1104
1105 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1106                                       unsigned nr_pages, blk_opf_t op_flag,
1107                                       pgoff_t first_idx, bool for_write)
1108 {
1109         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1110         struct bio *bio;
1111         struct bio_post_read_ctx *ctx = NULL;
1112         unsigned int post_read_steps = 0;
1113         sector_t sector;
1114         struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
1115
1116         bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1117                                REQ_OP_READ | op_flag,
1118                                for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1119         if (!bio)
1120                 return ERR_PTR(-ENOMEM);
1121         bio->bi_iter.bi_sector = sector;
1122         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1123         bio->bi_end_io = f2fs_read_end_io;
1124
1125         if (fscrypt_inode_uses_fs_layer_crypto(inode))
1126                 post_read_steps |= STEP_DECRYPT;
1127
1128         if (f2fs_need_verity(inode, first_idx))
1129                 post_read_steps |= STEP_VERITY;
1130
1131         /*
1132          * STEP_DECOMPRESS is handled specially, since a compressed file might
1133          * contain both compressed and uncompressed clusters.  We'll allocate a
1134          * bio_post_read_ctx if the file is compressed, but the caller is
1135          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1136          */
1137
1138         if (post_read_steps || f2fs_compressed_file(inode)) {
1139                 /* Due to the mempool, this never fails. */
1140                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1141                 ctx->bio = bio;
1142                 ctx->sbi = sbi;
1143                 ctx->enabled_steps = post_read_steps;
1144                 ctx->fs_blkaddr = blkaddr;
1145                 ctx->decompression_attempted = false;
1146                 bio->bi_private = ctx;
1147         }
1148         iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1149
1150         return bio;
1151 }
1152
1153 /* This can handle encryption stuffs */
1154 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1155                                  block_t blkaddr, blk_opf_t op_flags,
1156                                  bool for_write)
1157 {
1158         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1159         struct bio *bio;
1160
1161         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1162                                         page->index, for_write);
1163         if (IS_ERR(bio))
1164                 return PTR_ERR(bio);
1165
1166         /* wait for GCed page writeback via META_MAPPING */
1167         f2fs_wait_on_block_writeback(inode, blkaddr);
1168
1169         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1170                 iostat_update_and_unbind_ctx(bio);
1171                 if (bio->bi_private)
1172                         mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1173                 bio_put(bio);
1174                 return -EFAULT;
1175         }
1176         inc_page_count(sbi, F2FS_RD_DATA);
1177         f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1178         f2fs_submit_read_bio(sbi, bio, DATA);
1179         return 0;
1180 }
1181
1182 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1183 {
1184         __le32 *addr = get_dnode_addr(dn->inode, dn->node_page);
1185
1186         dn->data_blkaddr = blkaddr;
1187         addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1188 }
1189
1190 /*
1191  * Lock ordering for the change of data block address:
1192  * ->data_page
1193  *  ->node_page
1194  *    update block addresses in the node page
1195  */
1196 void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1197 {
1198         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1199         __set_data_blkaddr(dn, blkaddr);
1200         if (set_page_dirty(dn->node_page))
1201                 dn->node_changed = true;
1202 }
1203
1204 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1205 {
1206         f2fs_set_data_blkaddr(dn, blkaddr);
1207         f2fs_update_read_extent_cache(dn);
1208 }
1209
1210 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1211 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1212 {
1213         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1214         int err;
1215
1216         if (!count)
1217                 return 0;
1218
1219         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1220                 return -EPERM;
1221         err = inc_valid_block_count(sbi, dn->inode, &count, true);
1222         if (unlikely(err))
1223                 return err;
1224
1225         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1226                                                 dn->ofs_in_node, count);
1227
1228         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1229
1230         for (; count > 0; dn->ofs_in_node++) {
1231                 block_t blkaddr = f2fs_data_blkaddr(dn);
1232
1233                 if (blkaddr == NULL_ADDR) {
1234                         __set_data_blkaddr(dn, NEW_ADDR);
1235                         count--;
1236                 }
1237         }
1238
1239         if (set_page_dirty(dn->node_page))
1240                 dn->node_changed = true;
1241         return 0;
1242 }
1243
1244 /* Should keep dn->ofs_in_node unchanged */
1245 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1246 {
1247         unsigned int ofs_in_node = dn->ofs_in_node;
1248         int ret;
1249
1250         ret = f2fs_reserve_new_blocks(dn, 1);
1251         dn->ofs_in_node = ofs_in_node;
1252         return ret;
1253 }
1254
1255 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1256 {
1257         bool need_put = dn->inode_page ? false : true;
1258         int err;
1259
1260         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1261         if (err)
1262                 return err;
1263
1264         if (dn->data_blkaddr == NULL_ADDR)
1265                 err = f2fs_reserve_new_block(dn);
1266         if (err || need_put)
1267                 f2fs_put_dnode(dn);
1268         return err;
1269 }
1270
1271 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1272                                      blk_opf_t op_flags, bool for_write,
1273                                      pgoff_t *next_pgofs)
1274 {
1275         struct address_space *mapping = inode->i_mapping;
1276         struct dnode_of_data dn;
1277         struct page *page;
1278         int err;
1279
1280         page = f2fs_grab_cache_page(mapping, index, for_write);
1281         if (!page)
1282                 return ERR_PTR(-ENOMEM);
1283
1284         if (f2fs_lookup_read_extent_cache_block(inode, index,
1285                                                 &dn.data_blkaddr)) {
1286                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1287                                                 DATA_GENERIC_ENHANCE_READ)) {
1288                         err = -EFSCORRUPTED;
1289                         f2fs_handle_error(F2FS_I_SB(inode),
1290                                                 ERROR_INVALID_BLKADDR);
1291                         goto put_err;
1292                 }
1293                 goto got_it;
1294         }
1295
1296         set_new_dnode(&dn, inode, NULL, NULL, 0);
1297         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1298         if (err) {
1299                 if (err == -ENOENT && next_pgofs)
1300                         *next_pgofs = f2fs_get_next_page_offset(&dn, index);
1301                 goto put_err;
1302         }
1303         f2fs_put_dnode(&dn);
1304
1305         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1306                 err = -ENOENT;
1307                 if (next_pgofs)
1308                         *next_pgofs = index + 1;
1309                 goto put_err;
1310         }
1311         if (dn.data_blkaddr != NEW_ADDR &&
1312                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1313                                                 dn.data_blkaddr,
1314                                                 DATA_GENERIC_ENHANCE)) {
1315                 err = -EFSCORRUPTED;
1316                 f2fs_handle_error(F2FS_I_SB(inode),
1317                                         ERROR_INVALID_BLKADDR);
1318                 goto put_err;
1319         }
1320 got_it:
1321         if (PageUptodate(page)) {
1322                 unlock_page(page);
1323                 return page;
1324         }
1325
1326         /*
1327          * A new dentry page is allocated but not able to be written, since its
1328          * new inode page couldn't be allocated due to -ENOSPC.
1329          * In such the case, its blkaddr can be remained as NEW_ADDR.
1330          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1331          * f2fs_init_inode_metadata.
1332          */
1333         if (dn.data_blkaddr == NEW_ADDR) {
1334                 zero_user_segment(page, 0, PAGE_SIZE);
1335                 if (!PageUptodate(page))
1336                         SetPageUptodate(page);
1337                 unlock_page(page);
1338                 return page;
1339         }
1340
1341         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1342                                                 op_flags, for_write);
1343         if (err)
1344                 goto put_err;
1345         return page;
1346
1347 put_err:
1348         f2fs_put_page(page, 1);
1349         return ERR_PTR(err);
1350 }
1351
1352 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1353                                         pgoff_t *next_pgofs)
1354 {
1355         struct address_space *mapping = inode->i_mapping;
1356         struct page *page;
1357
1358         page = find_get_page(mapping, index);
1359         if (page && PageUptodate(page))
1360                 return page;
1361         f2fs_put_page(page, 0);
1362
1363         page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1364         if (IS_ERR(page))
1365                 return page;
1366
1367         if (PageUptodate(page))
1368                 return page;
1369
1370         wait_on_page_locked(page);
1371         if (unlikely(!PageUptodate(page))) {
1372                 f2fs_put_page(page, 0);
1373                 return ERR_PTR(-EIO);
1374         }
1375         return page;
1376 }
1377
1378 /*
1379  * If it tries to access a hole, return an error.
1380  * Because, the callers, functions in dir.c and GC, should be able to know
1381  * whether this page exists or not.
1382  */
1383 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1384                                                         bool for_write)
1385 {
1386         struct address_space *mapping = inode->i_mapping;
1387         struct page *page;
1388
1389         page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1390         if (IS_ERR(page))
1391                 return page;
1392
1393         /* wait for read completion */
1394         lock_page(page);
1395         if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1396                 f2fs_put_page(page, 1);
1397                 return ERR_PTR(-EIO);
1398         }
1399         return page;
1400 }
1401
1402 /*
1403  * Caller ensures that this data page is never allocated.
1404  * A new zero-filled data page is allocated in the page cache.
1405  *
1406  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1407  * f2fs_unlock_op().
1408  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1409  * ipage should be released by this function.
1410  */
1411 struct page *f2fs_get_new_data_page(struct inode *inode,
1412                 struct page *ipage, pgoff_t index, bool new_i_size)
1413 {
1414         struct address_space *mapping = inode->i_mapping;
1415         struct page *page;
1416         struct dnode_of_data dn;
1417         int err;
1418
1419         page = f2fs_grab_cache_page(mapping, index, true);
1420         if (!page) {
1421                 /*
1422                  * before exiting, we should make sure ipage will be released
1423                  * if any error occur.
1424                  */
1425                 f2fs_put_page(ipage, 1);
1426                 return ERR_PTR(-ENOMEM);
1427         }
1428
1429         set_new_dnode(&dn, inode, ipage, NULL, 0);
1430         err = f2fs_reserve_block(&dn, index);
1431         if (err) {
1432                 f2fs_put_page(page, 1);
1433                 return ERR_PTR(err);
1434         }
1435         if (!ipage)
1436                 f2fs_put_dnode(&dn);
1437
1438         if (PageUptodate(page))
1439                 goto got_it;
1440
1441         if (dn.data_blkaddr == NEW_ADDR) {
1442                 zero_user_segment(page, 0, PAGE_SIZE);
1443                 if (!PageUptodate(page))
1444                         SetPageUptodate(page);
1445         } else {
1446                 f2fs_put_page(page, 1);
1447
1448                 /* if ipage exists, blkaddr should be NEW_ADDR */
1449                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1450                 page = f2fs_get_lock_data_page(inode, index, true);
1451                 if (IS_ERR(page))
1452                         return page;
1453         }
1454 got_it:
1455         if (new_i_size && i_size_read(inode) <
1456                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1457                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1458         return page;
1459 }
1460
1461 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1462 {
1463         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1464         struct f2fs_summary sum;
1465         struct node_info ni;
1466         block_t old_blkaddr;
1467         blkcnt_t count = 1;
1468         int err;
1469
1470         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1471                 return -EPERM;
1472
1473         err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1474         if (err)
1475                 return err;
1476
1477         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1478         if (dn->data_blkaddr == NULL_ADDR) {
1479                 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1480                 if (unlikely(err))
1481                         return err;
1482         }
1483
1484         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1485         old_blkaddr = dn->data_blkaddr;
1486         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1487                                 &sum, seg_type, NULL);
1488         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1489                 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
1490
1491         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1492         return 0;
1493 }
1494
1495 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1496 {
1497         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1498                 f2fs_down_read(&sbi->node_change);
1499         else
1500                 f2fs_lock_op(sbi);
1501 }
1502
1503 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1504 {
1505         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1506                 f2fs_up_read(&sbi->node_change);
1507         else
1508                 f2fs_unlock_op(sbi);
1509 }
1510
1511 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1512 {
1513         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1514         int err = 0;
1515
1516         f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1517         if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1518                                                 &dn->data_blkaddr))
1519                 err = f2fs_reserve_block(dn, index);
1520         f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1521
1522         return err;
1523 }
1524
1525 static int f2fs_map_no_dnode(struct inode *inode,
1526                 struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1527                 pgoff_t pgoff)
1528 {
1529         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1530
1531         /*
1532          * There is one exceptional case that read_node_page() may return
1533          * -ENOENT due to filesystem has been shutdown or cp_error, return
1534          * -EIO in that case.
1535          */
1536         if (map->m_may_create &&
1537             (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1538                 return -EIO;
1539
1540         if (map->m_next_pgofs)
1541                 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1542         if (map->m_next_extent)
1543                 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1544         return 0;
1545 }
1546
1547 static bool f2fs_map_blocks_cached(struct inode *inode,
1548                 struct f2fs_map_blocks *map, int flag)
1549 {
1550         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1551         unsigned int maxblocks = map->m_len;
1552         pgoff_t pgoff = (pgoff_t)map->m_lblk;
1553         struct extent_info ei = {};
1554
1555         if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1556                 return false;
1557
1558         map->m_pblk = ei.blk + pgoff - ei.fofs;
1559         map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1560         map->m_flags = F2FS_MAP_MAPPED;
1561         if (map->m_next_extent)
1562                 *map->m_next_extent = pgoff + map->m_len;
1563
1564         /* for hardware encryption, but to avoid potential issue in future */
1565         if (flag == F2FS_GET_BLOCK_DIO)
1566                 f2fs_wait_on_block_writeback_range(inode,
1567                                         map->m_pblk, map->m_len);
1568
1569         if (f2fs_allow_multi_device_dio(sbi, flag)) {
1570                 int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1571                 struct f2fs_dev_info *dev = &sbi->devs[bidx];
1572
1573                 map->m_bdev = dev->bdev;
1574                 map->m_pblk -= dev->start_blk;
1575                 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1576         } else {
1577                 map->m_bdev = inode->i_sb->s_bdev;
1578         }
1579         return true;
1580 }
1581
1582 /*
1583  * f2fs_map_blocks() tries to find or build mapping relationship which
1584  * maps continuous logical blocks to physical blocks, and return such
1585  * info via f2fs_map_blocks structure.
1586  */
1587 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1588 {
1589         unsigned int maxblocks = map->m_len;
1590         struct dnode_of_data dn;
1591         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1592         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1593         pgoff_t pgofs, end_offset, end;
1594         int err = 0, ofs = 1;
1595         unsigned int ofs_in_node, last_ofs_in_node;
1596         blkcnt_t prealloc;
1597         block_t blkaddr;
1598         unsigned int start_pgofs;
1599         int bidx = 0;
1600         bool is_hole;
1601
1602         if (!maxblocks)
1603                 return 0;
1604
1605         if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1606                 goto out;
1607
1608         map->m_bdev = inode->i_sb->s_bdev;
1609         map->m_multidev_dio =
1610                 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1611
1612         map->m_len = 0;
1613         map->m_flags = 0;
1614
1615         /* it only supports block size == page size */
1616         pgofs = (pgoff_t)map->m_lblk;
1617         end = pgofs + maxblocks;
1618
1619 next_dnode:
1620         if (map->m_may_create)
1621                 f2fs_map_lock(sbi, flag);
1622
1623         /* When reading holes, we need its node page */
1624         set_new_dnode(&dn, inode, NULL, NULL, 0);
1625         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1626         if (err) {
1627                 if (flag == F2FS_GET_BLOCK_BMAP)
1628                         map->m_pblk = 0;
1629                 if (err == -ENOENT)
1630                         err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1631                 goto unlock_out;
1632         }
1633
1634         start_pgofs = pgofs;
1635         prealloc = 0;
1636         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1637         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1638
1639 next_block:
1640         blkaddr = f2fs_data_blkaddr(&dn);
1641         is_hole = !__is_valid_data_blkaddr(blkaddr);
1642         if (!is_hole &&
1643             !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1644                 err = -EFSCORRUPTED;
1645                 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1646                 goto sync_out;
1647         }
1648
1649         /* use out-place-update for direct IO under LFS mode */
1650         if (map->m_may_create &&
1651             (is_hole || (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO))) {
1652                 if (unlikely(f2fs_cp_error(sbi))) {
1653                         err = -EIO;
1654                         goto sync_out;
1655                 }
1656
1657                 switch (flag) {
1658                 case F2FS_GET_BLOCK_PRE_AIO:
1659                         if (blkaddr == NULL_ADDR) {
1660                                 prealloc++;
1661                                 last_ofs_in_node = dn.ofs_in_node;
1662                         }
1663                         break;
1664                 case F2FS_GET_BLOCK_PRE_DIO:
1665                 case F2FS_GET_BLOCK_DIO:
1666                         err = __allocate_data_block(&dn, map->m_seg_type);
1667                         if (err)
1668                                 goto sync_out;
1669                         if (flag == F2FS_GET_BLOCK_PRE_DIO)
1670                                 file_need_truncate(inode);
1671                         set_inode_flag(inode, FI_APPEND_WRITE);
1672                         break;
1673                 default:
1674                         WARN_ON_ONCE(1);
1675                         err = -EIO;
1676                         goto sync_out;
1677                 }
1678
1679                 blkaddr = dn.data_blkaddr;
1680                 if (is_hole)
1681                         map->m_flags |= F2FS_MAP_NEW;
1682         } else if (is_hole) {
1683                 if (f2fs_compressed_file(inode) &&
1684                     f2fs_sanity_check_cluster(&dn)) {
1685                         err = -EFSCORRUPTED;
1686                         f2fs_handle_error(sbi,
1687                                         ERROR_CORRUPTED_CLUSTER);
1688                         goto sync_out;
1689                 }
1690
1691                 switch (flag) {
1692                 case F2FS_GET_BLOCK_PRECACHE:
1693                         goto sync_out;
1694                 case F2FS_GET_BLOCK_BMAP:
1695                         map->m_pblk = 0;
1696                         goto sync_out;
1697                 case F2FS_GET_BLOCK_FIEMAP:
1698                         if (blkaddr == NULL_ADDR) {
1699                                 if (map->m_next_pgofs)
1700                                         *map->m_next_pgofs = pgofs + 1;
1701                                 goto sync_out;
1702                         }
1703                         break;
1704                 default:
1705                         /* for defragment case */
1706                         if (map->m_next_pgofs)
1707                                 *map->m_next_pgofs = pgofs + 1;
1708                         goto sync_out;
1709                 }
1710         }
1711
1712         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1713                 goto skip;
1714
1715         if (map->m_multidev_dio)
1716                 bidx = f2fs_target_device_index(sbi, blkaddr);
1717
1718         if (map->m_len == 0) {
1719                 /* reserved delalloc block should be mapped for fiemap. */
1720                 if (blkaddr == NEW_ADDR)
1721                         map->m_flags |= F2FS_MAP_DELALLOC;
1722                 map->m_flags |= F2FS_MAP_MAPPED;
1723
1724                 map->m_pblk = blkaddr;
1725                 map->m_len = 1;
1726
1727                 if (map->m_multidev_dio)
1728                         map->m_bdev = FDEV(bidx).bdev;
1729         } else if ((map->m_pblk != NEW_ADDR &&
1730                         blkaddr == (map->m_pblk + ofs)) ||
1731                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1732                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1733                 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1734                         goto sync_out;
1735                 ofs++;
1736                 map->m_len++;
1737         } else {
1738                 goto sync_out;
1739         }
1740
1741 skip:
1742         dn.ofs_in_node++;
1743         pgofs++;
1744
1745         /* preallocate blocks in batch for one dnode page */
1746         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1747                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1748
1749                 dn.ofs_in_node = ofs_in_node;
1750                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1751                 if (err)
1752                         goto sync_out;
1753
1754                 map->m_len += dn.ofs_in_node - ofs_in_node;
1755                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1756                         err = -ENOSPC;
1757                         goto sync_out;
1758                 }
1759                 dn.ofs_in_node = end_offset;
1760         }
1761
1762         if (pgofs >= end)
1763                 goto sync_out;
1764         else if (dn.ofs_in_node < end_offset)
1765                 goto next_block;
1766
1767         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1768                 if (map->m_flags & F2FS_MAP_MAPPED) {
1769                         unsigned int ofs = start_pgofs - map->m_lblk;
1770
1771                         f2fs_update_read_extent_cache_range(&dn,
1772                                 start_pgofs, map->m_pblk + ofs,
1773                                 map->m_len - ofs);
1774                 }
1775         }
1776
1777         f2fs_put_dnode(&dn);
1778
1779         if (map->m_may_create) {
1780                 f2fs_map_unlock(sbi, flag);
1781                 f2fs_balance_fs(sbi, dn.node_changed);
1782         }
1783         goto next_dnode;
1784
1785 sync_out:
1786
1787         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1788                 /*
1789                  * for hardware encryption, but to avoid potential issue
1790                  * in future
1791                  */
1792                 f2fs_wait_on_block_writeback_range(inode,
1793                                                 map->m_pblk, map->m_len);
1794
1795                 if (map->m_multidev_dio) {
1796                         block_t blk_addr = map->m_pblk;
1797
1798                         bidx = f2fs_target_device_index(sbi, map->m_pblk);
1799
1800                         map->m_bdev = FDEV(bidx).bdev;
1801                         map->m_pblk -= FDEV(bidx).start_blk;
1802
1803                         if (map->m_may_create)
1804                                 f2fs_update_device_state(sbi, inode->i_ino,
1805                                                         blk_addr, map->m_len);
1806
1807                         f2fs_bug_on(sbi, blk_addr + map->m_len >
1808                                                 FDEV(bidx).end_blk + 1);
1809                 }
1810         }
1811
1812         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1813                 if (map->m_flags & F2FS_MAP_MAPPED) {
1814                         unsigned int ofs = start_pgofs - map->m_lblk;
1815
1816                         f2fs_update_read_extent_cache_range(&dn,
1817                                 start_pgofs, map->m_pblk + ofs,
1818                                 map->m_len - ofs);
1819                 }
1820                 if (map->m_next_extent)
1821                         *map->m_next_extent = pgofs + 1;
1822         }
1823         f2fs_put_dnode(&dn);
1824 unlock_out:
1825         if (map->m_may_create) {
1826                 f2fs_map_unlock(sbi, flag);
1827                 f2fs_balance_fs(sbi, dn.node_changed);
1828         }
1829 out:
1830         trace_f2fs_map_blocks(inode, map, flag, err);
1831         return err;
1832 }
1833
1834 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1835 {
1836         struct f2fs_map_blocks map;
1837         block_t last_lblk;
1838         int err;
1839
1840         if (pos + len > i_size_read(inode))
1841                 return false;
1842
1843         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1844         map.m_next_pgofs = NULL;
1845         map.m_next_extent = NULL;
1846         map.m_seg_type = NO_CHECK_TYPE;
1847         map.m_may_create = false;
1848         last_lblk = F2FS_BLK_ALIGN(pos + len);
1849
1850         while (map.m_lblk < last_lblk) {
1851                 map.m_len = last_lblk - map.m_lblk;
1852                 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1853                 if (err || map.m_len == 0)
1854                         return false;
1855                 map.m_lblk += map.m_len;
1856         }
1857         return true;
1858 }
1859
1860 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1861 {
1862         return (bytes >> inode->i_blkbits);
1863 }
1864
1865 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1866 {
1867         return (blks << inode->i_blkbits);
1868 }
1869
1870 static int f2fs_xattr_fiemap(struct inode *inode,
1871                                 struct fiemap_extent_info *fieinfo)
1872 {
1873         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1874         struct page *page;
1875         struct node_info ni;
1876         __u64 phys = 0, len;
1877         __u32 flags;
1878         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1879         int err = 0;
1880
1881         if (f2fs_has_inline_xattr(inode)) {
1882                 int offset;
1883
1884                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1885                                                 inode->i_ino, false);
1886                 if (!page)
1887                         return -ENOMEM;
1888
1889                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1890                 if (err) {
1891                         f2fs_put_page(page, 1);
1892                         return err;
1893                 }
1894
1895                 phys = blks_to_bytes(inode, ni.blk_addr);
1896                 offset = offsetof(struct f2fs_inode, i_addr) +
1897                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1898                                         get_inline_xattr_addrs(inode));
1899
1900                 phys += offset;
1901                 len = inline_xattr_size(inode);
1902
1903                 f2fs_put_page(page, 1);
1904
1905                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1906
1907                 if (!xnid)
1908                         flags |= FIEMAP_EXTENT_LAST;
1909
1910                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1911                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1912                 if (err)
1913                         return err;
1914         }
1915
1916         if (xnid) {
1917                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1918                 if (!page)
1919                         return -ENOMEM;
1920
1921                 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1922                 if (err) {
1923                         f2fs_put_page(page, 1);
1924                         return err;
1925                 }
1926
1927                 phys = blks_to_bytes(inode, ni.blk_addr);
1928                 len = inode->i_sb->s_blocksize;
1929
1930                 f2fs_put_page(page, 1);
1931
1932                 flags = FIEMAP_EXTENT_LAST;
1933         }
1934
1935         if (phys) {
1936                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1937                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1938         }
1939
1940         return (err < 0 ? err : 0);
1941 }
1942
1943 static loff_t max_inode_blocks(struct inode *inode)
1944 {
1945         loff_t result = ADDRS_PER_INODE(inode);
1946         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1947
1948         /* two direct node blocks */
1949         result += (leaf_count * 2);
1950
1951         /* two indirect node blocks */
1952         leaf_count *= NIDS_PER_BLOCK;
1953         result += (leaf_count * 2);
1954
1955         /* one double indirect node block */
1956         leaf_count *= NIDS_PER_BLOCK;
1957         result += leaf_count;
1958
1959         return result;
1960 }
1961
1962 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1963                 u64 start, u64 len)
1964 {
1965         struct f2fs_map_blocks map;
1966         sector_t start_blk, last_blk;
1967         pgoff_t next_pgofs;
1968         u64 logical = 0, phys = 0, size = 0;
1969         u32 flags = 0;
1970         int ret = 0;
1971         bool compr_cluster = false, compr_appended;
1972         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1973         unsigned int count_in_cluster = 0;
1974         loff_t maxbytes;
1975
1976         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1977                 ret = f2fs_precache_extents(inode);
1978                 if (ret)
1979                         return ret;
1980         }
1981
1982         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1983         if (ret)
1984                 return ret;
1985
1986         inode_lock_shared(inode);
1987
1988         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1989         if (start > maxbytes) {
1990                 ret = -EFBIG;
1991                 goto out;
1992         }
1993
1994         if (len > maxbytes || (maxbytes - len) < start)
1995                 len = maxbytes - start;
1996
1997         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1998                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1999                 goto out;
2000         }
2001
2002         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
2003                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
2004                 if (ret != -EAGAIN)
2005                         goto out;
2006         }
2007
2008         if (bytes_to_blks(inode, len) == 0)
2009                 len = blks_to_bytes(inode, 1);
2010
2011         start_blk = bytes_to_blks(inode, start);
2012         last_blk = bytes_to_blks(inode, start + len - 1);
2013
2014 next:
2015         memset(&map, 0, sizeof(map));
2016         map.m_lblk = start_blk;
2017         map.m_len = bytes_to_blks(inode, len);
2018         map.m_next_pgofs = &next_pgofs;
2019         map.m_seg_type = NO_CHECK_TYPE;
2020
2021         if (compr_cluster) {
2022                 map.m_lblk += 1;
2023                 map.m_len = cluster_size - count_in_cluster;
2024         }
2025
2026         ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
2027         if (ret)
2028                 goto out;
2029
2030         /* HOLE */
2031         if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
2032                 start_blk = next_pgofs;
2033
2034                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
2035                                                 max_inode_blocks(inode)))
2036                         goto prep_next;
2037
2038                 flags |= FIEMAP_EXTENT_LAST;
2039         }
2040
2041         compr_appended = false;
2042         /* In a case of compressed cluster, append this to the last extent */
2043         if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
2044                         !(map.m_flags & F2FS_MAP_FLAGS))) {
2045                 compr_appended = true;
2046                 goto skip_fill;
2047         }
2048
2049         if (size) {
2050                 flags |= FIEMAP_EXTENT_MERGED;
2051                 if (IS_ENCRYPTED(inode))
2052                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2053
2054                 ret = fiemap_fill_next_extent(fieinfo, logical,
2055                                 phys, size, flags);
2056                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2057                 if (ret)
2058                         goto out;
2059                 size = 0;
2060         }
2061
2062         if (start_blk > last_blk)
2063                 goto out;
2064
2065 skip_fill:
2066         if (map.m_pblk == COMPRESS_ADDR) {
2067                 compr_cluster = true;
2068                 count_in_cluster = 1;
2069         } else if (compr_appended) {
2070                 unsigned int appended_blks = cluster_size -
2071                                                 count_in_cluster + 1;
2072                 size += blks_to_bytes(inode, appended_blks);
2073                 start_blk += appended_blks;
2074                 compr_cluster = false;
2075         } else {
2076                 logical = blks_to_bytes(inode, start_blk);
2077                 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2078                         blks_to_bytes(inode, map.m_pblk) : 0;
2079                 size = blks_to_bytes(inode, map.m_len);
2080                 flags = 0;
2081
2082                 if (compr_cluster) {
2083                         flags = FIEMAP_EXTENT_ENCODED;
2084                         count_in_cluster += map.m_len;
2085                         if (count_in_cluster == cluster_size) {
2086                                 compr_cluster = false;
2087                                 size += blks_to_bytes(inode, 1);
2088                         }
2089                 } else if (map.m_flags & F2FS_MAP_DELALLOC) {
2090                         flags = FIEMAP_EXTENT_UNWRITTEN;
2091                 }
2092
2093                 start_blk += bytes_to_blks(inode, size);
2094         }
2095
2096 prep_next:
2097         cond_resched();
2098         if (fatal_signal_pending(current))
2099                 ret = -EINTR;
2100         else
2101                 goto next;
2102 out:
2103         if (ret == 1)
2104                 ret = 0;
2105
2106         inode_unlock_shared(inode);
2107         return ret;
2108 }
2109
2110 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2111 {
2112         if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2113                 return inode->i_sb->s_maxbytes;
2114
2115         return i_size_read(inode);
2116 }
2117
2118 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2119                                         unsigned nr_pages,
2120                                         struct f2fs_map_blocks *map,
2121                                         struct bio **bio_ret,
2122                                         sector_t *last_block_in_bio,
2123                                         bool is_readahead)
2124 {
2125         struct bio *bio = *bio_ret;
2126         const unsigned blocksize = blks_to_bytes(inode, 1);
2127         sector_t block_in_file;
2128         sector_t last_block;
2129         sector_t last_block_in_file;
2130         sector_t block_nr;
2131         int ret = 0;
2132
2133         block_in_file = (sector_t)page_index(page);
2134         last_block = block_in_file + nr_pages;
2135         last_block_in_file = bytes_to_blks(inode,
2136                         f2fs_readpage_limit(inode) + blocksize - 1);
2137         if (last_block > last_block_in_file)
2138                 last_block = last_block_in_file;
2139
2140         /* just zeroing out page which is beyond EOF */
2141         if (block_in_file >= last_block)
2142                 goto zero_out;
2143         /*
2144          * Map blocks using the previous result first.
2145          */
2146         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2147                         block_in_file > map->m_lblk &&
2148                         block_in_file < (map->m_lblk + map->m_len))
2149                 goto got_it;
2150
2151         /*
2152          * Then do more f2fs_map_blocks() calls until we are
2153          * done with this page.
2154          */
2155         map->m_lblk = block_in_file;
2156         map->m_len = last_block - block_in_file;
2157
2158         ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2159         if (ret)
2160                 goto out;
2161 got_it:
2162         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2163                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2164                 SetPageMappedToDisk(page);
2165
2166                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2167                                                 DATA_GENERIC_ENHANCE_READ)) {
2168                         ret = -EFSCORRUPTED;
2169                         f2fs_handle_error(F2FS_I_SB(inode),
2170                                                 ERROR_INVALID_BLKADDR);
2171                         goto out;
2172                 }
2173         } else {
2174 zero_out:
2175                 zero_user_segment(page, 0, PAGE_SIZE);
2176                 if (f2fs_need_verity(inode, page->index) &&
2177                     !fsverity_verify_page(page)) {
2178                         ret = -EIO;
2179                         goto out;
2180                 }
2181                 if (!PageUptodate(page))
2182                         SetPageUptodate(page);
2183                 unlock_page(page);
2184                 goto out;
2185         }
2186
2187         /*
2188          * This page will go to BIO.  Do we need to send this
2189          * BIO off first?
2190          */
2191         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2192                                        *last_block_in_bio, block_nr) ||
2193                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2194 submit_and_realloc:
2195                 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2196                 bio = NULL;
2197         }
2198         if (bio == NULL) {
2199                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2200                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2201                                 false);
2202                 if (IS_ERR(bio)) {
2203                         ret = PTR_ERR(bio);
2204                         bio = NULL;
2205                         goto out;
2206                 }
2207         }
2208
2209         /*
2210          * If the page is under writeback, we need to wait for
2211          * its completion to see the correct decrypted data.
2212          */
2213         f2fs_wait_on_block_writeback(inode, block_nr);
2214
2215         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2216                 goto submit_and_realloc;
2217
2218         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2219         f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2220                                                         F2FS_BLKSIZE);
2221         *last_block_in_bio = block_nr;
2222 out:
2223         *bio_ret = bio;
2224         return ret;
2225 }
2226
2227 #ifdef CONFIG_F2FS_FS_COMPRESSION
2228 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2229                                 unsigned nr_pages, sector_t *last_block_in_bio,
2230                                 bool is_readahead, bool for_write)
2231 {
2232         struct dnode_of_data dn;
2233         struct inode *inode = cc->inode;
2234         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2235         struct bio *bio = *bio_ret;
2236         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2237         sector_t last_block_in_file;
2238         const unsigned blocksize = blks_to_bytes(inode, 1);
2239         struct decompress_io_ctx *dic = NULL;
2240         struct extent_info ei = {};
2241         bool from_dnode = true;
2242         int i;
2243         int ret = 0;
2244
2245         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2246
2247         last_block_in_file = bytes_to_blks(inode,
2248                         f2fs_readpage_limit(inode) + blocksize - 1);
2249
2250         /* get rid of pages beyond EOF */
2251         for (i = 0; i < cc->cluster_size; i++) {
2252                 struct page *page = cc->rpages[i];
2253
2254                 if (!page)
2255                         continue;
2256                 if ((sector_t)page->index >= last_block_in_file) {
2257                         zero_user_segment(page, 0, PAGE_SIZE);
2258                         if (!PageUptodate(page))
2259                                 SetPageUptodate(page);
2260                 } else if (!PageUptodate(page)) {
2261                         continue;
2262                 }
2263                 unlock_page(page);
2264                 if (for_write)
2265                         put_page(page);
2266                 cc->rpages[i] = NULL;
2267                 cc->nr_rpages--;
2268         }
2269
2270         /* we are done since all pages are beyond EOF */
2271         if (f2fs_cluster_is_empty(cc))
2272                 goto out;
2273
2274         if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2275                 from_dnode = false;
2276
2277         if (!from_dnode)
2278                 goto skip_reading_dnode;
2279
2280         set_new_dnode(&dn, inode, NULL, NULL, 0);
2281         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2282         if (ret)
2283                 goto out;
2284
2285         if (unlikely(f2fs_cp_error(sbi))) {
2286                 ret = -EIO;
2287                 goto out_put_dnode;
2288         }
2289         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2290
2291 skip_reading_dnode:
2292         for (i = 1; i < cc->cluster_size; i++) {
2293                 block_t blkaddr;
2294
2295                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2296                                         dn.ofs_in_node + i) :
2297                                         ei.blk + i - 1;
2298
2299                 if (!__is_valid_data_blkaddr(blkaddr))
2300                         break;
2301
2302                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2303                         ret = -EFAULT;
2304                         goto out_put_dnode;
2305                 }
2306                 cc->nr_cpages++;
2307
2308                 if (!from_dnode && i >= ei.c_len)
2309                         break;
2310         }
2311
2312         /* nothing to decompress */
2313         if (cc->nr_cpages == 0) {
2314                 ret = 0;
2315                 goto out_put_dnode;
2316         }
2317
2318         dic = f2fs_alloc_dic(cc);
2319         if (IS_ERR(dic)) {
2320                 ret = PTR_ERR(dic);
2321                 goto out_put_dnode;
2322         }
2323
2324         for (i = 0; i < cc->nr_cpages; i++) {
2325                 struct page *page = dic->cpages[i];
2326                 block_t blkaddr;
2327                 struct bio_post_read_ctx *ctx;
2328
2329                 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2330                                         dn.ofs_in_node + i + 1) :
2331                                         ei.blk + i;
2332
2333                 f2fs_wait_on_block_writeback(inode, blkaddr);
2334
2335                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2336                         if (atomic_dec_and_test(&dic->remaining_pages)) {
2337                                 f2fs_decompress_cluster(dic, true);
2338                                 break;
2339                         }
2340                         continue;
2341                 }
2342
2343                 if (bio && (!page_is_mergeable(sbi, bio,
2344                                         *last_block_in_bio, blkaddr) ||
2345                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2346 submit_and_realloc:
2347                         f2fs_submit_read_bio(sbi, bio, DATA);
2348                         bio = NULL;
2349                 }
2350
2351                 if (!bio) {
2352                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2353                                         is_readahead ? REQ_RAHEAD : 0,
2354                                         page->index, for_write);
2355                         if (IS_ERR(bio)) {
2356                                 ret = PTR_ERR(bio);
2357                                 f2fs_decompress_end_io(dic, ret, true);
2358                                 f2fs_put_dnode(&dn);
2359                                 *bio_ret = NULL;
2360                                 return ret;
2361                         }
2362                 }
2363
2364                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2365                         goto submit_and_realloc;
2366
2367                 ctx = get_post_read_ctx(bio);
2368                 ctx->enabled_steps |= STEP_DECOMPRESS;
2369                 refcount_inc(&dic->refcnt);
2370
2371                 inc_page_count(sbi, F2FS_RD_DATA);
2372                 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2373                 *last_block_in_bio = blkaddr;
2374         }
2375
2376         if (from_dnode)
2377                 f2fs_put_dnode(&dn);
2378
2379         *bio_ret = bio;
2380         return 0;
2381
2382 out_put_dnode:
2383         if (from_dnode)
2384                 f2fs_put_dnode(&dn);
2385 out:
2386         for (i = 0; i < cc->cluster_size; i++) {
2387                 if (cc->rpages[i]) {
2388                         ClearPageUptodate(cc->rpages[i]);
2389                         unlock_page(cc->rpages[i]);
2390                 }
2391         }
2392         *bio_ret = bio;
2393         return ret;
2394 }
2395 #endif
2396
2397 /*
2398  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2399  * Major change was from block_size == page_size in f2fs by default.
2400  */
2401 static int f2fs_mpage_readpages(struct inode *inode,
2402                 struct readahead_control *rac, struct page *page)
2403 {
2404         struct bio *bio = NULL;
2405         sector_t last_block_in_bio = 0;
2406         struct f2fs_map_blocks map;
2407 #ifdef CONFIG_F2FS_FS_COMPRESSION
2408         struct compress_ctx cc = {
2409                 .inode = inode,
2410                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2411                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2412                 .cluster_idx = NULL_CLUSTER,
2413                 .rpages = NULL,
2414                 .cpages = NULL,
2415                 .nr_rpages = 0,
2416                 .nr_cpages = 0,
2417         };
2418         pgoff_t nc_cluster_idx = NULL_CLUSTER;
2419 #endif
2420         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2421         unsigned max_nr_pages = nr_pages;
2422         int ret = 0;
2423
2424         map.m_pblk = 0;
2425         map.m_lblk = 0;
2426         map.m_len = 0;
2427         map.m_flags = 0;
2428         map.m_next_pgofs = NULL;
2429         map.m_next_extent = NULL;
2430         map.m_seg_type = NO_CHECK_TYPE;
2431         map.m_may_create = false;
2432
2433         for (; nr_pages; nr_pages--) {
2434                 if (rac) {
2435                         page = readahead_page(rac);
2436                         prefetchw(&page->flags);
2437                 }
2438
2439 #ifdef CONFIG_F2FS_FS_COMPRESSION
2440                 if (f2fs_compressed_file(inode)) {
2441                         /* there are remained compressed pages, submit them */
2442                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2443                                 ret = f2fs_read_multi_pages(&cc, &bio,
2444                                                         max_nr_pages,
2445                                                         &last_block_in_bio,
2446                                                         rac != NULL, false);
2447                                 f2fs_destroy_compress_ctx(&cc, false);
2448                                 if (ret)
2449                                         goto set_error_page;
2450                         }
2451                         if (cc.cluster_idx == NULL_CLUSTER) {
2452                                 if (nc_cluster_idx ==
2453                                         page->index >> cc.log_cluster_size) {
2454                                         goto read_single_page;
2455                                 }
2456
2457                                 ret = f2fs_is_compressed_cluster(inode, page->index);
2458                                 if (ret < 0)
2459                                         goto set_error_page;
2460                                 else if (!ret) {
2461                                         nc_cluster_idx =
2462                                                 page->index >> cc.log_cluster_size;
2463                                         goto read_single_page;
2464                                 }
2465
2466                                 nc_cluster_idx = NULL_CLUSTER;
2467                         }
2468                         ret = f2fs_init_compress_ctx(&cc);
2469                         if (ret)
2470                                 goto set_error_page;
2471
2472                         f2fs_compress_ctx_add_page(&cc, page);
2473
2474                         goto next_page;
2475                 }
2476 read_single_page:
2477 #endif
2478
2479                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2480                                         &bio, &last_block_in_bio, rac);
2481                 if (ret) {
2482 #ifdef CONFIG_F2FS_FS_COMPRESSION
2483 set_error_page:
2484 #endif
2485                         zero_user_segment(page, 0, PAGE_SIZE);
2486                         unlock_page(page);
2487                 }
2488 #ifdef CONFIG_F2FS_FS_COMPRESSION
2489 next_page:
2490 #endif
2491                 if (rac)
2492                         put_page(page);
2493
2494 #ifdef CONFIG_F2FS_FS_COMPRESSION
2495                 if (f2fs_compressed_file(inode)) {
2496                         /* last page */
2497                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2498                                 ret = f2fs_read_multi_pages(&cc, &bio,
2499                                                         max_nr_pages,
2500                                                         &last_block_in_bio,
2501                                                         rac != NULL, false);
2502                                 f2fs_destroy_compress_ctx(&cc, false);
2503                         }
2504                 }
2505 #endif
2506         }
2507         if (bio)
2508                 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2509         return ret;
2510 }
2511
2512 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2513 {
2514         struct page *page = &folio->page;
2515         struct inode *inode = page_file_mapping(page)->host;
2516         int ret = -EAGAIN;
2517
2518         trace_f2fs_readpage(page, DATA);
2519
2520         if (!f2fs_is_compress_backend_ready(inode)) {
2521                 unlock_page(page);
2522                 return -EOPNOTSUPP;
2523         }
2524
2525         /* If the file has inline data, try to read it directly */
2526         if (f2fs_has_inline_data(inode))
2527                 ret = f2fs_read_inline_data(inode, page);
2528         if (ret == -EAGAIN)
2529                 ret = f2fs_mpage_readpages(inode, NULL, page);
2530         return ret;
2531 }
2532
2533 static void f2fs_readahead(struct readahead_control *rac)
2534 {
2535         struct inode *inode = rac->mapping->host;
2536
2537         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2538
2539         if (!f2fs_is_compress_backend_ready(inode))
2540                 return;
2541
2542         /* If the file has inline data, skip readahead */
2543         if (f2fs_has_inline_data(inode))
2544                 return;
2545
2546         f2fs_mpage_readpages(inode, rac, NULL);
2547 }
2548
2549 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2550 {
2551         struct inode *inode = fio->page->mapping->host;
2552         struct page *mpage, *page;
2553         gfp_t gfp_flags = GFP_NOFS;
2554
2555         if (!f2fs_encrypted_file(inode))
2556                 return 0;
2557
2558         page = fio->compressed_page ? fio->compressed_page : fio->page;
2559
2560         if (fscrypt_inode_uses_inline_crypto(inode))
2561                 return 0;
2562
2563 retry_encrypt:
2564         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2565                                         PAGE_SIZE, 0, gfp_flags);
2566         if (IS_ERR(fio->encrypted_page)) {
2567                 /* flush pending IOs and wait for a while in the ENOMEM case */
2568                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2569                         f2fs_flush_merged_writes(fio->sbi);
2570                         memalloc_retry_wait(GFP_NOFS);
2571                         gfp_flags |= __GFP_NOFAIL;
2572                         goto retry_encrypt;
2573                 }
2574                 return PTR_ERR(fio->encrypted_page);
2575         }
2576
2577         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2578         if (mpage) {
2579                 if (PageUptodate(mpage))
2580                         memcpy(page_address(mpage),
2581                                 page_address(fio->encrypted_page), PAGE_SIZE);
2582                 f2fs_put_page(mpage, 1);
2583         }
2584         return 0;
2585 }
2586
2587 static inline bool check_inplace_update_policy(struct inode *inode,
2588                                 struct f2fs_io_info *fio)
2589 {
2590         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2591
2592         if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2593             is_inode_flag_set(inode, FI_OPU_WRITE))
2594                 return false;
2595         if (IS_F2FS_IPU_FORCE(sbi))
2596                 return true;
2597         if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2598                 return true;
2599         if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2600                 return true;
2601         if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2602             utilization(sbi) > SM_I(sbi)->min_ipu_util)
2603                 return true;
2604
2605         /*
2606          * IPU for rewrite async pages
2607          */
2608         if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2609             !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2610                 return true;
2611
2612         /* this is only set during fdatasync */
2613         if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2614                 return true;
2615
2616         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2617                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2618                 return true;
2619
2620         return false;
2621 }
2622
2623 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2624 {
2625         /* swap file is migrating in aligned write mode */
2626         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2627                 return false;
2628
2629         if (f2fs_is_pinned_file(inode))
2630                 return true;
2631
2632         /* if this is cold file, we should overwrite to avoid fragmentation */
2633         if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2634                 return true;
2635
2636         return check_inplace_update_policy(inode, fio);
2637 }
2638
2639 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2640 {
2641         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2642
2643         /* The below cases were checked when setting it. */
2644         if (f2fs_is_pinned_file(inode))
2645                 return false;
2646         if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2647                 return true;
2648         if (f2fs_lfs_mode(sbi))
2649                 return true;
2650         if (S_ISDIR(inode->i_mode))
2651                 return true;
2652         if (IS_NOQUOTA(inode))
2653                 return true;
2654         if (f2fs_is_atomic_file(inode))
2655                 return true;
2656         /* rewrite low ratio compress data w/ OPU mode to avoid fragmentation */
2657         if (f2fs_compressed_file(inode) &&
2658                 F2FS_OPTION(sbi).compress_mode == COMPR_MODE_USER &&
2659                 is_inode_flag_set(inode, FI_ENABLE_COMPRESS))
2660                 return true;
2661
2662         /* swap file is migrating in aligned write mode */
2663         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2664                 return true;
2665
2666         if (is_inode_flag_set(inode, FI_OPU_WRITE))
2667                 return true;
2668
2669         if (fio) {
2670                 if (page_private_gcing(fio->page))
2671                         return true;
2672                 if (page_private_dummy(fio->page))
2673                         return true;
2674                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2675                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2676                         return true;
2677         }
2678         return false;
2679 }
2680
2681 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2682 {
2683         struct inode *inode = fio->page->mapping->host;
2684
2685         if (f2fs_should_update_outplace(inode, fio))
2686                 return false;
2687
2688         return f2fs_should_update_inplace(inode, fio);
2689 }
2690
2691 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2692 {
2693         struct page *page = fio->page;
2694         struct inode *inode = page->mapping->host;
2695         struct dnode_of_data dn;
2696         struct node_info ni;
2697         bool ipu_force = false;
2698         int err = 0;
2699
2700         /* Use COW inode to make dnode_of_data for atomic write */
2701         if (f2fs_is_atomic_file(inode))
2702                 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2703         else
2704                 set_new_dnode(&dn, inode, NULL, NULL, 0);
2705
2706         if (need_inplace_update(fio) &&
2707             f2fs_lookup_read_extent_cache_block(inode, page->index,
2708                                                 &fio->old_blkaddr)) {
2709                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2710                                                 DATA_GENERIC_ENHANCE)) {
2711                         f2fs_handle_error(fio->sbi,
2712                                                 ERROR_INVALID_BLKADDR);
2713                         return -EFSCORRUPTED;
2714                 }
2715
2716                 ipu_force = true;
2717                 fio->need_lock = LOCK_DONE;
2718                 goto got_it;
2719         }
2720
2721         /* Deadlock due to between page->lock and f2fs_lock_op */
2722         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2723                 return -EAGAIN;
2724
2725         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2726         if (err)
2727                 goto out;
2728
2729         fio->old_blkaddr = dn.data_blkaddr;
2730
2731         /* This page is already truncated */
2732         if (fio->old_blkaddr == NULL_ADDR) {
2733                 ClearPageUptodate(page);
2734                 clear_page_private_gcing(page);
2735                 goto out_writepage;
2736         }
2737 got_it:
2738         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2739                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2740                                                 DATA_GENERIC_ENHANCE)) {
2741                 err = -EFSCORRUPTED;
2742                 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
2743                 goto out_writepage;
2744         }
2745
2746         /* wait for GCed page writeback via META_MAPPING */
2747         if (fio->post_read)
2748                 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2749
2750         /*
2751          * If current allocation needs SSR,
2752          * it had better in-place writes for updated data.
2753          */
2754         if (ipu_force ||
2755                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2756                                         need_inplace_update(fio))) {
2757                 err = f2fs_encrypt_one_page(fio);
2758                 if (err)
2759                         goto out_writepage;
2760
2761                 set_page_writeback(page);
2762                 f2fs_put_dnode(&dn);
2763                 if (fio->need_lock == LOCK_REQ)
2764                         f2fs_unlock_op(fio->sbi);
2765                 err = f2fs_inplace_write_data(fio);
2766                 if (err) {
2767                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2768                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2769                         if (PageWriteback(page))
2770                                 end_page_writeback(page);
2771                 } else {
2772                         set_inode_flag(inode, FI_UPDATE_WRITE);
2773                 }
2774                 trace_f2fs_do_write_data_page(fio->page, IPU);
2775                 return err;
2776         }
2777
2778         if (fio->need_lock == LOCK_RETRY) {
2779                 if (!f2fs_trylock_op(fio->sbi)) {
2780                         err = -EAGAIN;
2781                         goto out_writepage;
2782                 }
2783                 fio->need_lock = LOCK_REQ;
2784         }
2785
2786         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2787         if (err)
2788                 goto out_writepage;
2789
2790         fio->version = ni.version;
2791
2792         err = f2fs_encrypt_one_page(fio);
2793         if (err)
2794                 goto out_writepage;
2795
2796         set_page_writeback(page);
2797
2798         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2799                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2800
2801         /* LFS mode write path */
2802         f2fs_outplace_write_data(&dn, fio);
2803         trace_f2fs_do_write_data_page(page, OPU);
2804         set_inode_flag(inode, FI_APPEND_WRITE);
2805 out_writepage:
2806         f2fs_put_dnode(&dn);
2807 out:
2808         if (fio->need_lock == LOCK_REQ)
2809                 f2fs_unlock_op(fio->sbi);
2810         return err;
2811 }
2812
2813 int f2fs_write_single_data_page(struct page *page, int *submitted,
2814                                 struct bio **bio,
2815                                 sector_t *last_block,
2816                                 struct writeback_control *wbc,
2817                                 enum iostat_type io_type,
2818                                 int compr_blocks,
2819                                 bool allow_balance)
2820 {
2821         struct inode *inode = page->mapping->host;
2822         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2823         loff_t i_size = i_size_read(inode);
2824         const pgoff_t end_index = ((unsigned long long)i_size)
2825                                                         >> PAGE_SHIFT;
2826         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2827         unsigned offset = 0;
2828         bool need_balance_fs = false;
2829         bool quota_inode = IS_NOQUOTA(inode);
2830         int err = 0;
2831         struct f2fs_io_info fio = {
2832                 .sbi = sbi,
2833                 .ino = inode->i_ino,
2834                 .type = DATA,
2835                 .op = REQ_OP_WRITE,
2836                 .op_flags = wbc_to_write_flags(wbc),
2837                 .old_blkaddr = NULL_ADDR,
2838                 .page = page,
2839                 .encrypted_page = NULL,
2840                 .submitted = 0,
2841                 .compr_blocks = compr_blocks,
2842                 .need_lock = compr_blocks ? LOCK_DONE : LOCK_RETRY,
2843                 .post_read = f2fs_post_read_required(inode) ? 1 : 0,
2844                 .io_type = io_type,
2845                 .io_wbc = wbc,
2846                 .bio = bio,
2847                 .last_block = last_block,
2848         };
2849
2850         trace_f2fs_writepage(page, DATA);
2851
2852         /* we should bypass data pages to proceed the kworker jobs */
2853         if (unlikely(f2fs_cp_error(sbi))) {
2854                 mapping_set_error(page->mapping, -EIO);
2855                 /*
2856                  * don't drop any dirty dentry pages for keeping lastest
2857                  * directory structure.
2858                  */
2859                 if (S_ISDIR(inode->i_mode) &&
2860                                 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2861                         goto redirty_out;
2862
2863                 /* keep data pages in remount-ro mode */
2864                 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2865                         goto redirty_out;
2866                 goto out;
2867         }
2868
2869         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2870                 goto redirty_out;
2871
2872         if (page->index < end_index ||
2873                         f2fs_verity_in_progress(inode) ||
2874                         compr_blocks)
2875                 goto write;
2876
2877         /*
2878          * If the offset is out-of-range of file size,
2879          * this page does not have to be written to disk.
2880          */
2881         offset = i_size & (PAGE_SIZE - 1);
2882         if ((page->index >= end_index + 1) || !offset)
2883                 goto out;
2884
2885         zero_user_segment(page, offset, PAGE_SIZE);
2886 write:
2887         /* Dentry/quota blocks are controlled by checkpoint */
2888         if (S_ISDIR(inode->i_mode) || quota_inode) {
2889                 /*
2890                  * We need to wait for node_write to avoid block allocation during
2891                  * checkpoint. This can only happen to quota writes which can cause
2892                  * the below discard race condition.
2893                  */
2894                 if (quota_inode)
2895                         f2fs_down_read(&sbi->node_write);
2896
2897                 fio.need_lock = LOCK_DONE;
2898                 err = f2fs_do_write_data_page(&fio);
2899
2900                 if (quota_inode)
2901                         f2fs_up_read(&sbi->node_write);
2902
2903                 goto done;
2904         }
2905
2906         if (!wbc->for_reclaim)
2907                 need_balance_fs = true;
2908         else if (has_not_enough_free_secs(sbi, 0, 0))
2909                 goto redirty_out;
2910         else
2911                 set_inode_flag(inode, FI_HOT_DATA);
2912
2913         err = -EAGAIN;
2914         if (f2fs_has_inline_data(inode)) {
2915                 err = f2fs_write_inline_data(inode, page);
2916                 if (!err)
2917                         goto out;
2918         }
2919
2920         if (err == -EAGAIN) {
2921                 err = f2fs_do_write_data_page(&fio);
2922                 if (err == -EAGAIN) {
2923                         f2fs_bug_on(sbi, compr_blocks);
2924                         fio.need_lock = LOCK_REQ;
2925                         err = f2fs_do_write_data_page(&fio);
2926                 }
2927         }
2928
2929         if (err) {
2930                 file_set_keep_isize(inode);
2931         } else {
2932                 spin_lock(&F2FS_I(inode)->i_size_lock);
2933                 if (F2FS_I(inode)->last_disk_size < psize)
2934                         F2FS_I(inode)->last_disk_size = psize;
2935                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2936         }
2937
2938 done:
2939         if (err && err != -ENOENT)
2940                 goto redirty_out;
2941
2942 out:
2943         inode_dec_dirty_pages(inode);
2944         if (err) {
2945                 ClearPageUptodate(page);
2946                 clear_page_private_gcing(page);
2947         }
2948
2949         if (wbc->for_reclaim) {
2950                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2951                 clear_inode_flag(inode, FI_HOT_DATA);
2952                 f2fs_remove_dirty_inode(inode);
2953                 submitted = NULL;
2954         }
2955         unlock_page(page);
2956         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2957                         !F2FS_I(inode)->wb_task && allow_balance)
2958                 f2fs_balance_fs(sbi, need_balance_fs);
2959
2960         if (unlikely(f2fs_cp_error(sbi))) {
2961                 f2fs_submit_merged_write(sbi, DATA);
2962                 if (bio && *bio)
2963                         f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2964                 submitted = NULL;
2965         }
2966
2967         if (submitted)
2968                 *submitted = fio.submitted;
2969
2970         return 0;
2971
2972 redirty_out:
2973         redirty_page_for_writepage(wbc, page);
2974         /*
2975          * pageout() in MM translates EAGAIN, so calls handle_write_error()
2976          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2977          * file_write_and_wait_range() will see EIO error, which is critical
2978          * to return value of fsync() followed by atomic_write failure to user.
2979          */
2980         if (!err || wbc->for_reclaim)
2981                 return AOP_WRITEPAGE_ACTIVATE;
2982         unlock_page(page);
2983         return err;
2984 }
2985
2986 static int f2fs_write_data_page(struct page *page,
2987                                         struct writeback_control *wbc)
2988 {
2989 #ifdef CONFIG_F2FS_FS_COMPRESSION
2990         struct inode *inode = page->mapping->host;
2991
2992         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2993                 goto out;
2994
2995         if (f2fs_compressed_file(inode)) {
2996                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2997                         redirty_page_for_writepage(wbc, page);
2998                         return AOP_WRITEPAGE_ACTIVATE;
2999                 }
3000         }
3001 out:
3002 #endif
3003
3004         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
3005                                                 wbc, FS_DATA_IO, 0, true);
3006 }
3007
3008 /*
3009  * This function was copied from write_cache_pages from mm/page-writeback.c.
3010  * The major change is making write step of cold data page separately from
3011  * warm/hot data page.
3012  */
3013 static int f2fs_write_cache_pages(struct address_space *mapping,
3014                                         struct writeback_control *wbc,
3015                                         enum iostat_type io_type)
3016 {
3017         int ret = 0;
3018         int done = 0, retry = 0;
3019         struct page *pages_local[F2FS_ONSTACK_PAGES];
3020         struct page **pages = pages_local;
3021         struct folio_batch fbatch;
3022         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
3023         struct bio *bio = NULL;
3024         sector_t last_block;
3025 #ifdef CONFIG_F2FS_FS_COMPRESSION
3026         struct inode *inode = mapping->host;
3027         struct compress_ctx cc = {
3028                 .inode = inode,
3029                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
3030                 .cluster_size = F2FS_I(inode)->i_cluster_size,
3031                 .cluster_idx = NULL_CLUSTER,
3032                 .rpages = NULL,
3033                 .nr_rpages = 0,
3034                 .cpages = NULL,
3035                 .valid_nr_cpages = 0,
3036                 .rbuf = NULL,
3037                 .cbuf = NULL,
3038                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
3039                 .private = NULL,
3040         };
3041 #endif
3042         int nr_folios, p, idx;
3043         int nr_pages;
3044         unsigned int max_pages = F2FS_ONSTACK_PAGES;
3045         pgoff_t index;
3046         pgoff_t end;            /* Inclusive */
3047         pgoff_t done_index;
3048         int range_whole = 0;
3049         xa_mark_t tag;
3050         int nwritten = 0;
3051         int submitted = 0;
3052         int i;
3053
3054 #ifdef CONFIG_F2FS_FS_COMPRESSION
3055         if (f2fs_compressed_file(inode) &&
3056                 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) {
3057                 pages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
3058                                 cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL);
3059                 max_pages = 1 << cc.log_cluster_size;
3060         }
3061 #endif
3062
3063         folio_batch_init(&fbatch);
3064
3065         if (get_dirty_pages(mapping->host) <=
3066                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3067                 set_inode_flag(mapping->host, FI_HOT_DATA);
3068         else
3069                 clear_inode_flag(mapping->host, FI_HOT_DATA);
3070
3071         if (wbc->range_cyclic) {
3072                 index = mapping->writeback_index; /* prev offset */
3073                 end = -1;
3074         } else {
3075                 index = wbc->range_start >> PAGE_SHIFT;
3076                 end = wbc->range_end >> PAGE_SHIFT;
3077                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3078                         range_whole = 1;
3079         }
3080         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3081                 tag = PAGECACHE_TAG_TOWRITE;
3082         else
3083                 tag = PAGECACHE_TAG_DIRTY;
3084 retry:
3085         retry = 0;
3086         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3087                 tag_pages_for_writeback(mapping, index, end);
3088         done_index = index;
3089         while (!done && !retry && (index <= end)) {
3090                 nr_pages = 0;
3091 again:
3092                 nr_folios = filemap_get_folios_tag(mapping, &index, end,
3093                                 tag, &fbatch);
3094                 if (nr_folios == 0) {
3095                         if (nr_pages)
3096                                 goto write;
3097                         break;
3098                 }
3099
3100                 for (i = 0; i < nr_folios; i++) {
3101                         struct folio *folio = fbatch.folios[i];
3102
3103                         idx = 0;
3104                         p = folio_nr_pages(folio);
3105 add_more:
3106                         pages[nr_pages] = folio_page(folio, idx);
3107                         folio_get(folio);
3108                         if (++nr_pages == max_pages) {
3109                                 index = folio->index + idx + 1;
3110                                 folio_batch_release(&fbatch);
3111                                 goto write;
3112                         }
3113                         if (++idx < p)
3114                                 goto add_more;
3115                 }
3116                 folio_batch_release(&fbatch);
3117                 goto again;
3118 write:
3119                 for (i = 0; i < nr_pages; i++) {
3120                         struct page *page = pages[i];
3121                         struct folio *folio = page_folio(page);
3122                         bool need_readd;
3123 readd:
3124                         need_readd = false;
3125 #ifdef CONFIG_F2FS_FS_COMPRESSION
3126                         if (f2fs_compressed_file(inode)) {
3127                                 void *fsdata = NULL;
3128                                 struct page *pagep;
3129                                 int ret2;
3130
3131                                 ret = f2fs_init_compress_ctx(&cc);
3132                                 if (ret) {
3133                                         done = 1;
3134                                         break;
3135                                 }
3136
3137                                 if (!f2fs_cluster_can_merge_page(&cc,
3138                                                                 folio->index)) {
3139                                         ret = f2fs_write_multi_pages(&cc,
3140                                                 &submitted, wbc, io_type);
3141                                         if (!ret)
3142                                                 need_readd = true;
3143                                         goto result;
3144                                 }
3145
3146                                 if (unlikely(f2fs_cp_error(sbi)))
3147                                         goto lock_folio;
3148
3149                                 if (!f2fs_cluster_is_empty(&cc))
3150                                         goto lock_folio;
3151
3152                                 if (f2fs_all_cluster_page_ready(&cc,
3153                                         pages, i, nr_pages, true))
3154                                         goto lock_folio;
3155
3156                                 ret2 = f2fs_prepare_compress_overwrite(
3157                                                         inode, &pagep,
3158                                                         folio->index, &fsdata);
3159                                 if (ret2 < 0) {
3160                                         ret = ret2;
3161                                         done = 1;
3162                                         break;
3163                                 } else if (ret2 &&
3164                                         (!f2fs_compress_write_end(inode,
3165                                                 fsdata, folio->index, 1) ||
3166                                          !f2fs_all_cluster_page_ready(&cc,
3167                                                 pages, i, nr_pages,
3168                                                 false))) {
3169                                         retry = 1;
3170                                         break;
3171                                 }
3172                         }
3173 #endif
3174                         /* give a priority to WB_SYNC threads */
3175                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3176                                         wbc->sync_mode == WB_SYNC_NONE) {
3177                                 done = 1;
3178                                 break;
3179                         }
3180 #ifdef CONFIG_F2FS_FS_COMPRESSION
3181 lock_folio:
3182 #endif
3183                         done_index = folio->index;
3184 retry_write:
3185                         folio_lock(folio);
3186
3187                         if (unlikely(folio->mapping != mapping)) {
3188 continue_unlock:
3189                                 folio_unlock(folio);
3190                                 continue;
3191                         }
3192
3193                         if (!folio_test_dirty(folio)) {
3194                                 /* someone wrote it for us */
3195                                 goto continue_unlock;
3196                         }
3197
3198                         if (folio_test_writeback(folio)) {
3199                                 if (wbc->sync_mode == WB_SYNC_NONE)
3200                                         goto continue_unlock;
3201                                 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3202                         }
3203
3204                         if (!folio_clear_dirty_for_io(folio))
3205                                 goto continue_unlock;
3206
3207 #ifdef CONFIG_F2FS_FS_COMPRESSION
3208                         if (f2fs_compressed_file(inode)) {
3209                                 folio_get(folio);
3210                                 f2fs_compress_ctx_add_page(&cc, &folio->page);
3211                                 continue;
3212                         }
3213 #endif
3214                         ret = f2fs_write_single_data_page(&folio->page,
3215                                         &submitted, &bio, &last_block,
3216                                         wbc, io_type, 0, true);
3217                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3218                                 folio_unlock(folio);
3219 #ifdef CONFIG_F2FS_FS_COMPRESSION
3220 result:
3221 #endif
3222                         nwritten += submitted;
3223                         wbc->nr_to_write -= submitted;
3224
3225                         if (unlikely(ret)) {
3226                                 /*
3227                                  * keep nr_to_write, since vfs uses this to
3228                                  * get # of written pages.
3229                                  */
3230                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3231                                         ret = 0;
3232                                         goto next;
3233                                 } else if (ret == -EAGAIN) {
3234                                         ret = 0;
3235                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3236                                                 f2fs_io_schedule_timeout(
3237                                                         DEFAULT_IO_TIMEOUT);
3238                                                 goto retry_write;
3239                                         }
3240                                         goto next;
3241                                 }
3242                                 done_index = folio_next_index(folio);
3243                                 done = 1;
3244                                 break;
3245                         }
3246
3247                         if (wbc->nr_to_write <= 0 &&
3248                                         wbc->sync_mode == WB_SYNC_NONE) {
3249                                 done = 1;
3250                                 break;
3251                         }
3252 next:
3253                         if (need_readd)
3254                                 goto readd;
3255                 }
3256                 release_pages(pages, nr_pages);
3257                 cond_resched();
3258         }
3259 #ifdef CONFIG_F2FS_FS_COMPRESSION
3260         /* flush remained pages in compress cluster */
3261         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3262                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3263                 nwritten += submitted;
3264                 wbc->nr_to_write -= submitted;
3265                 if (ret) {
3266                         done = 1;
3267                         retry = 0;
3268                 }
3269         }
3270         if (f2fs_compressed_file(inode))
3271                 f2fs_destroy_compress_ctx(&cc, false);
3272 #endif
3273         if (retry) {
3274                 index = 0;
3275                 end = -1;
3276                 goto retry;
3277         }
3278         if (wbc->range_cyclic && !done)
3279                 done_index = 0;
3280         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3281                 mapping->writeback_index = done_index;
3282
3283         if (nwritten)
3284                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3285                                                                 NULL, 0, DATA);
3286         /* submit cached bio of IPU write */
3287         if (bio)
3288                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3289
3290 #ifdef CONFIG_F2FS_FS_COMPRESSION
3291         if (pages != pages_local)
3292                 kfree(pages);
3293 #endif
3294
3295         return ret;
3296 }
3297
3298 static inline bool __should_serialize_io(struct inode *inode,
3299                                         struct writeback_control *wbc)
3300 {
3301         /* to avoid deadlock in path of data flush */
3302         if (F2FS_I(inode)->wb_task)
3303                 return false;
3304
3305         if (!S_ISREG(inode->i_mode))
3306                 return false;
3307         if (IS_NOQUOTA(inode))
3308                 return false;
3309
3310         if (f2fs_need_compress_data(inode))
3311                 return true;
3312         if (wbc->sync_mode != WB_SYNC_ALL)
3313                 return true;
3314         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3315                 return true;
3316         return false;
3317 }
3318
3319 static int __f2fs_write_data_pages(struct address_space *mapping,
3320                                                 struct writeback_control *wbc,
3321                                                 enum iostat_type io_type)
3322 {
3323         struct inode *inode = mapping->host;
3324         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3325         struct blk_plug plug;
3326         int ret;
3327         bool locked = false;
3328
3329         /* deal with chardevs and other special file */
3330         if (!mapping->a_ops->writepage)
3331                 return 0;
3332
3333         /* skip writing if there is no dirty page in this inode */
3334         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3335                 return 0;
3336
3337         /* during POR, we don't need to trigger writepage at all. */
3338         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3339                 goto skip_write;
3340
3341         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3342                         wbc->sync_mode == WB_SYNC_NONE &&
3343                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3344                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3345                 goto skip_write;
3346
3347         /* skip writing in file defragment preparing stage */
3348         if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3349                 goto skip_write;
3350
3351         trace_f2fs_writepages(mapping->host, wbc, DATA);
3352
3353         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3354         if (wbc->sync_mode == WB_SYNC_ALL)
3355                 atomic_inc(&sbi->wb_sync_req[DATA]);
3356         else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3357                 /* to avoid potential deadlock */
3358                 if (current->plug)
3359                         blk_finish_plug(current->plug);
3360                 goto skip_write;
3361         }
3362
3363         if (__should_serialize_io(inode, wbc)) {
3364                 mutex_lock(&sbi->writepages);
3365                 locked = true;
3366         }
3367
3368         blk_start_plug(&plug);
3369         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3370         blk_finish_plug(&plug);
3371
3372         if (locked)
3373                 mutex_unlock(&sbi->writepages);
3374
3375         if (wbc->sync_mode == WB_SYNC_ALL)
3376                 atomic_dec(&sbi->wb_sync_req[DATA]);
3377         /*
3378          * if some pages were truncated, we cannot guarantee its mapping->host
3379          * to detect pending bios.
3380          */
3381
3382         f2fs_remove_dirty_inode(inode);
3383         return ret;
3384
3385 skip_write:
3386         wbc->pages_skipped += get_dirty_pages(inode);
3387         trace_f2fs_writepages(mapping->host, wbc, DATA);
3388         return 0;
3389 }
3390
3391 static int f2fs_write_data_pages(struct address_space *mapping,
3392                             struct writeback_control *wbc)
3393 {
3394         struct inode *inode = mapping->host;
3395
3396         return __f2fs_write_data_pages(mapping, wbc,
3397                         F2FS_I(inode)->cp_task == current ?
3398                         FS_CP_DATA_IO : FS_DATA_IO);
3399 }
3400
3401 void f2fs_write_failed(struct inode *inode, loff_t to)
3402 {
3403         loff_t i_size = i_size_read(inode);
3404
3405         if (IS_NOQUOTA(inode))
3406                 return;
3407
3408         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3409         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3410                 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3411                 filemap_invalidate_lock(inode->i_mapping);
3412
3413                 truncate_pagecache(inode, i_size);
3414                 f2fs_truncate_blocks(inode, i_size, true);
3415
3416                 filemap_invalidate_unlock(inode->i_mapping);
3417                 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3418         }
3419 }
3420
3421 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3422                         struct page *page, loff_t pos, unsigned len,
3423                         block_t *blk_addr, bool *node_changed)
3424 {
3425         struct inode *inode = page->mapping->host;
3426         pgoff_t index = page->index;
3427         struct dnode_of_data dn;
3428         struct page *ipage;
3429         bool locked = false;
3430         int flag = F2FS_GET_BLOCK_PRE_AIO;
3431         int err = 0;
3432
3433         /*
3434          * If a whole page is being written and we already preallocated all the
3435          * blocks, then there is no need to get a block address now.
3436          */
3437         if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3438                 return 0;
3439
3440         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3441         if (f2fs_has_inline_data(inode)) {
3442                 if (pos + len > MAX_INLINE_DATA(inode))
3443                         flag = F2FS_GET_BLOCK_DEFAULT;
3444                 f2fs_map_lock(sbi, flag);
3445                 locked = true;
3446         } else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3447                 f2fs_map_lock(sbi, flag);
3448                 locked = true;
3449         }
3450
3451 restart:
3452         /* check inline_data */
3453         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3454         if (IS_ERR(ipage)) {
3455                 err = PTR_ERR(ipage);
3456                 goto unlock_out;
3457         }
3458
3459         set_new_dnode(&dn, inode, ipage, ipage, 0);
3460
3461         if (f2fs_has_inline_data(inode)) {
3462                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3463                         f2fs_do_read_inline_data(page, ipage);
3464                         set_inode_flag(inode, FI_DATA_EXIST);
3465                         if (inode->i_nlink)
3466                                 set_page_private_inline(ipage);
3467                         goto out;
3468                 }
3469                 err = f2fs_convert_inline_page(&dn, page);
3470                 if (err || dn.data_blkaddr != NULL_ADDR)
3471                         goto out;
3472         }
3473
3474         if (!f2fs_lookup_read_extent_cache_block(inode, index,
3475                                                  &dn.data_blkaddr)) {
3476                 if (locked) {
3477                         err = f2fs_reserve_block(&dn, index);
3478                         goto out;
3479                 }
3480
3481                 /* hole case */
3482                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3483                 if (!err && dn.data_blkaddr != NULL_ADDR)
3484                         goto out;
3485                 f2fs_put_dnode(&dn);
3486                 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3487                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3488                 locked = true;
3489                 goto restart;
3490         }
3491 out:
3492         if (!err) {
3493                 /* convert_inline_page can make node_changed */
3494                 *blk_addr = dn.data_blkaddr;
3495                 *node_changed = dn.node_changed;
3496         }
3497         f2fs_put_dnode(&dn);
3498 unlock_out:
3499         if (locked)
3500                 f2fs_map_unlock(sbi, flag);
3501         return err;
3502 }
3503
3504 static int __find_data_block(struct inode *inode, pgoff_t index,
3505                                 block_t *blk_addr)
3506 {
3507         struct dnode_of_data dn;
3508         struct page *ipage;
3509         int err = 0;
3510
3511         ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3512         if (IS_ERR(ipage))
3513                 return PTR_ERR(ipage);
3514
3515         set_new_dnode(&dn, inode, ipage, ipage, 0);
3516
3517         if (!f2fs_lookup_read_extent_cache_block(inode, index,
3518                                                  &dn.data_blkaddr)) {
3519                 /* hole case */
3520                 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3521                 if (err) {
3522                         dn.data_blkaddr = NULL_ADDR;
3523                         err = 0;
3524                 }
3525         }
3526         *blk_addr = dn.data_blkaddr;
3527         f2fs_put_dnode(&dn);
3528         return err;
3529 }
3530
3531 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3532                                 block_t *blk_addr, bool *node_changed)
3533 {
3534         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3535         struct dnode_of_data dn;
3536         struct page *ipage;
3537         int err = 0;
3538
3539         f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3540
3541         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3542         if (IS_ERR(ipage)) {
3543                 err = PTR_ERR(ipage);
3544                 goto unlock_out;
3545         }
3546         set_new_dnode(&dn, inode, ipage, ipage, 0);
3547
3548         if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3549                                                 &dn.data_blkaddr))
3550                 err = f2fs_reserve_block(&dn, index);
3551
3552         *blk_addr = dn.data_blkaddr;
3553         *node_changed = dn.node_changed;
3554         f2fs_put_dnode(&dn);
3555
3556 unlock_out:
3557         f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3558         return err;
3559 }
3560
3561 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3562                         struct page *page, loff_t pos, unsigned int len,
3563                         block_t *blk_addr, bool *node_changed, bool *use_cow)
3564 {
3565         struct inode *inode = page->mapping->host;
3566         struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3567         pgoff_t index = page->index;
3568         int err = 0;
3569         block_t ori_blk_addr = NULL_ADDR;
3570
3571         /* If pos is beyond the end of file, reserve a new block in COW inode */
3572         if ((pos & PAGE_MASK) >= i_size_read(inode))
3573                 goto reserve_block;
3574
3575         /* Look for the block in COW inode first */
3576         err = __find_data_block(cow_inode, index, blk_addr);
3577         if (err) {
3578                 return err;
3579         } else if (*blk_addr != NULL_ADDR) {
3580                 *use_cow = true;
3581                 return 0;
3582         }
3583
3584         if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3585                 goto reserve_block;
3586
3587         /* Look for the block in the original inode */
3588         err = __find_data_block(inode, index, &ori_blk_addr);
3589         if (err)
3590                 return err;
3591
3592 reserve_block:
3593         /* Finally, we should reserve a new block in COW inode for the update */
3594         err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3595         if (err)
3596                 return err;
3597         inc_atomic_write_cnt(inode);
3598
3599         if (ori_blk_addr != NULL_ADDR)
3600                 *blk_addr = ori_blk_addr;
3601         return 0;
3602 }
3603
3604 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3605                 loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3606 {
3607         struct inode *inode = mapping->host;
3608         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3609         struct page *page = NULL;
3610         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3611         bool need_balance = false;
3612         bool use_cow = false;
3613         block_t blkaddr = NULL_ADDR;
3614         int err = 0;
3615
3616         trace_f2fs_write_begin(inode, pos, len);
3617
3618         if (!f2fs_is_checkpoint_ready(sbi)) {
3619                 err = -ENOSPC;
3620                 goto fail;
3621         }
3622
3623         /*
3624          * We should check this at this moment to avoid deadlock on inode page
3625          * and #0 page. The locking rule for inline_data conversion should be:
3626          * lock_page(page #0) -> lock_page(inode_page)
3627          */
3628         if (index != 0) {
3629                 err = f2fs_convert_inline_inode(inode);
3630                 if (err)
3631                         goto fail;
3632         }
3633
3634 #ifdef CONFIG_F2FS_FS_COMPRESSION
3635         if (f2fs_compressed_file(inode)) {
3636                 int ret;
3637
3638                 *fsdata = NULL;
3639
3640                 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3641                         goto repeat;
3642
3643                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3644                                                         index, fsdata);
3645                 if (ret < 0) {
3646                         err = ret;
3647                         goto fail;
3648                 } else if (ret) {
3649                         return 0;
3650                 }
3651         }
3652 #endif
3653
3654 repeat:
3655         /*
3656          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3657          * wait_for_stable_page. Will wait that below with our IO control.
3658          */
3659         page = f2fs_pagecache_get_page(mapping, index,
3660                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3661         if (!page) {
3662                 err = -ENOMEM;
3663                 goto fail;
3664         }
3665
3666         /* TODO: cluster can be compressed due to race with .writepage */
3667
3668         *pagep = page;
3669
3670         if (f2fs_is_atomic_file(inode))
3671                 err = prepare_atomic_write_begin(sbi, page, pos, len,
3672                                         &blkaddr, &need_balance, &use_cow);
3673         else
3674                 err = prepare_write_begin(sbi, page, pos, len,
3675                                         &blkaddr, &need_balance);
3676         if (err)
3677                 goto fail;
3678
3679         if (need_balance && !IS_NOQUOTA(inode) &&
3680                         has_not_enough_free_secs(sbi, 0, 0)) {
3681                 unlock_page(page);
3682                 f2fs_balance_fs(sbi, true);
3683                 lock_page(page);
3684                 if (page->mapping != mapping) {
3685                         /* The page got truncated from under us */
3686                         f2fs_put_page(page, 1);
3687                         goto repeat;
3688                 }
3689         }
3690
3691         f2fs_wait_on_page_writeback(page, DATA, false, true);
3692
3693         if (len == PAGE_SIZE || PageUptodate(page))
3694                 return 0;
3695
3696         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3697             !f2fs_verity_in_progress(inode)) {
3698                 zero_user_segment(page, len, PAGE_SIZE);
3699                 return 0;
3700         }
3701
3702         if (blkaddr == NEW_ADDR) {
3703                 zero_user_segment(page, 0, PAGE_SIZE);
3704                 SetPageUptodate(page);
3705         } else {
3706                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3707                                 DATA_GENERIC_ENHANCE_READ)) {
3708                         err = -EFSCORRUPTED;
3709                         f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
3710                         goto fail;
3711                 }
3712                 err = f2fs_submit_page_read(use_cow ?
3713                                 F2FS_I(inode)->cow_inode : inode, page,
3714                                 blkaddr, 0, true);
3715                 if (err)
3716                         goto fail;
3717
3718                 lock_page(page);
3719                 if (unlikely(page->mapping != mapping)) {
3720                         f2fs_put_page(page, 1);
3721                         goto repeat;
3722                 }
3723                 if (unlikely(!PageUptodate(page))) {
3724                         err = -EIO;
3725                         goto fail;
3726                 }
3727         }
3728         return 0;
3729
3730 fail:
3731         f2fs_put_page(page, 1);
3732         f2fs_write_failed(inode, pos + len);
3733         return err;
3734 }
3735
3736 static int f2fs_write_end(struct file *file,
3737                         struct address_space *mapping,
3738                         loff_t pos, unsigned len, unsigned copied,
3739                         struct page *page, void *fsdata)
3740 {
3741         struct inode *inode = page->mapping->host;
3742
3743         trace_f2fs_write_end(inode, pos, len, copied);
3744
3745         /*
3746          * This should be come from len == PAGE_SIZE, and we expect copied
3747          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3748          * let generic_perform_write() try to copy data again through copied=0.
3749          */
3750         if (!PageUptodate(page)) {
3751                 if (unlikely(copied != len))
3752                         copied = 0;
3753                 else
3754                         SetPageUptodate(page);
3755         }
3756
3757 #ifdef CONFIG_F2FS_FS_COMPRESSION
3758         /* overwrite compressed file */
3759         if (f2fs_compressed_file(inode) && fsdata) {
3760                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3761                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3762
3763                 if (pos + copied > i_size_read(inode) &&
3764                                 !f2fs_verity_in_progress(inode))
3765                         f2fs_i_size_write(inode, pos + copied);
3766                 return copied;
3767         }
3768 #endif
3769
3770         if (!copied)
3771                 goto unlock_out;
3772
3773         set_page_dirty(page);
3774
3775         if (pos + copied > i_size_read(inode) &&
3776             !f2fs_verity_in_progress(inode)) {
3777                 f2fs_i_size_write(inode, pos + copied);
3778                 if (f2fs_is_atomic_file(inode))
3779                         f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3780                                         pos + copied);
3781         }
3782 unlock_out:
3783         f2fs_put_page(page, 1);
3784         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3785         return copied;
3786 }
3787
3788 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3789 {
3790         struct inode *inode = folio->mapping->host;
3791         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3792
3793         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3794                                 (offset || length != folio_size(folio)))
3795                 return;
3796
3797         if (folio_test_dirty(folio)) {
3798                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3799                         dec_page_count(sbi, F2FS_DIRTY_META);
3800                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3801                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3802                 } else {
3803                         inode_dec_dirty_pages(inode);
3804                         f2fs_remove_dirty_inode(inode);
3805                 }
3806         }
3807         clear_page_private_all(&folio->page);
3808 }
3809
3810 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3811 {
3812         /* If this is dirty folio, keep private data */
3813         if (folio_test_dirty(folio))
3814                 return false;
3815
3816         clear_page_private_all(&folio->page);
3817         return true;
3818 }
3819
3820 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3821                 struct folio *folio)
3822 {
3823         struct inode *inode = mapping->host;
3824
3825         trace_f2fs_set_page_dirty(&folio->page, DATA);
3826
3827         if (!folio_test_uptodate(folio))
3828                 folio_mark_uptodate(folio);
3829         BUG_ON(folio_test_swapcache(folio));
3830
3831         if (filemap_dirty_folio(mapping, folio)) {
3832                 f2fs_update_dirty_folio(inode, folio);
3833                 return true;
3834         }
3835         return false;
3836 }
3837
3838
3839 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3840 {
3841 #ifdef CONFIG_F2FS_FS_COMPRESSION
3842         struct dnode_of_data dn;
3843         sector_t start_idx, blknr = 0;
3844         int ret;
3845
3846         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3847
3848         set_new_dnode(&dn, inode, NULL, NULL, 0);
3849         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3850         if (ret)
3851                 return 0;
3852
3853         if (dn.data_blkaddr != COMPRESS_ADDR) {
3854                 dn.ofs_in_node += block - start_idx;
3855                 blknr = f2fs_data_blkaddr(&dn);
3856                 if (!__is_valid_data_blkaddr(blknr))
3857                         blknr = 0;
3858         }
3859
3860         f2fs_put_dnode(&dn);
3861         return blknr;
3862 #else
3863         return 0;
3864 #endif
3865 }
3866
3867
3868 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3869 {
3870         struct inode *inode = mapping->host;
3871         sector_t blknr = 0;
3872
3873         if (f2fs_has_inline_data(inode))
3874                 goto out;
3875
3876         /* make sure allocating whole blocks */
3877         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3878                 filemap_write_and_wait(mapping);
3879
3880         /* Block number less than F2FS MAX BLOCKS */
3881         if (unlikely(block >= max_file_blocks(inode)))
3882                 goto out;
3883
3884         if (f2fs_compressed_file(inode)) {
3885                 blknr = f2fs_bmap_compress(inode, block);
3886         } else {
3887                 struct f2fs_map_blocks map;
3888
3889                 memset(&map, 0, sizeof(map));
3890                 map.m_lblk = block;
3891                 map.m_len = 1;
3892                 map.m_next_pgofs = NULL;
3893                 map.m_seg_type = NO_CHECK_TYPE;
3894
3895                 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3896                         blknr = map.m_pblk;
3897         }
3898 out:
3899         trace_f2fs_bmap(inode, block, blknr);
3900         return blknr;
3901 }
3902
3903 #ifdef CONFIG_SWAP
3904 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3905                                                         unsigned int blkcnt)
3906 {
3907         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3908         unsigned int blkofs;
3909         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3910         unsigned int secidx = start_blk / blk_per_sec;
3911         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3912         int ret = 0;
3913
3914         f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3915         filemap_invalidate_lock(inode->i_mapping);
3916
3917         set_inode_flag(inode, FI_ALIGNED_WRITE);
3918         set_inode_flag(inode, FI_OPU_WRITE);
3919
3920         for (; secidx < end_sec; secidx++) {
3921                 f2fs_down_write(&sbi->pin_sem);
3922
3923                 f2fs_lock_op(sbi);
3924                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3925                 f2fs_unlock_op(sbi);
3926
3927                 set_inode_flag(inode, FI_SKIP_WRITES);
3928
3929                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3930                         struct page *page;
3931                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3932
3933                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3934                         if (IS_ERR(page)) {
3935                                 f2fs_up_write(&sbi->pin_sem);
3936                                 ret = PTR_ERR(page);
3937                                 goto done;
3938                         }
3939
3940                         set_page_dirty(page);
3941                         f2fs_put_page(page, 1);
3942                 }
3943
3944                 clear_inode_flag(inode, FI_SKIP_WRITES);
3945
3946                 ret = filemap_fdatawrite(inode->i_mapping);
3947
3948                 f2fs_up_write(&sbi->pin_sem);
3949
3950                 if (ret)
3951                         break;
3952         }
3953
3954 done:
3955         clear_inode_flag(inode, FI_SKIP_WRITES);
3956         clear_inode_flag(inode, FI_OPU_WRITE);
3957         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3958
3959         filemap_invalidate_unlock(inode->i_mapping);
3960         f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3961
3962         return ret;
3963 }
3964
3965 static int check_swap_activate(struct swap_info_struct *sis,
3966                                 struct file *swap_file, sector_t *span)
3967 {
3968         struct address_space *mapping = swap_file->f_mapping;
3969         struct inode *inode = mapping->host;
3970         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3971         sector_t cur_lblock;
3972         sector_t last_lblock;
3973         sector_t pblock;
3974         sector_t lowest_pblock = -1;
3975         sector_t highest_pblock = 0;
3976         int nr_extents = 0;
3977         unsigned long nr_pblocks;
3978         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3979         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3980         unsigned int not_aligned = 0;
3981         int ret = 0;
3982
3983         /*
3984          * Map all the blocks into the extent list.  This code doesn't try
3985          * to be very smart.
3986          */
3987         cur_lblock = 0;
3988         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3989
3990         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3991                 struct f2fs_map_blocks map;
3992 retry:
3993                 cond_resched();
3994
3995                 memset(&map, 0, sizeof(map));
3996                 map.m_lblk = cur_lblock;
3997                 map.m_len = last_lblock - cur_lblock;
3998                 map.m_next_pgofs = NULL;
3999                 map.m_next_extent = NULL;
4000                 map.m_seg_type = NO_CHECK_TYPE;
4001                 map.m_may_create = false;
4002
4003                 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
4004                 if (ret)
4005                         goto out;
4006
4007                 /* hole */
4008                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
4009                         f2fs_err(sbi, "Swapfile has holes");
4010                         ret = -EINVAL;
4011                         goto out;
4012                 }
4013
4014                 pblock = map.m_pblk;
4015                 nr_pblocks = map.m_len;
4016
4017                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
4018                                 nr_pblocks & sec_blks_mask) {
4019                         not_aligned++;
4020
4021                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
4022                         if (cur_lblock + nr_pblocks > sis->max)
4023                                 nr_pblocks -= blks_per_sec;
4024
4025                         if (!nr_pblocks) {
4026                                 /* this extent is last one */
4027                                 nr_pblocks = map.m_len;
4028                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
4029                                 goto next;
4030                         }
4031
4032                         ret = f2fs_migrate_blocks(inode, cur_lblock,
4033                                                         nr_pblocks);
4034                         if (ret)
4035                                 goto out;
4036                         goto retry;
4037                 }
4038 next:
4039                 if (cur_lblock + nr_pblocks >= sis->max)
4040                         nr_pblocks = sis->max - cur_lblock;
4041
4042                 if (cur_lblock) {       /* exclude the header page */
4043                         if (pblock < lowest_pblock)
4044                                 lowest_pblock = pblock;
4045                         if (pblock + nr_pblocks - 1 > highest_pblock)
4046                                 highest_pblock = pblock + nr_pblocks - 1;
4047                 }
4048
4049                 /*
4050                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4051                  */
4052                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4053                 if (ret < 0)
4054                         goto out;
4055                 nr_extents += ret;
4056                 cur_lblock += nr_pblocks;
4057         }
4058         ret = nr_extents;
4059         *span = 1 + highest_pblock - lowest_pblock;
4060         if (cur_lblock == 0)
4061                 cur_lblock = 1; /* force Empty message */
4062         sis->max = cur_lblock;
4063         sis->pages = cur_lblock - 1;
4064         sis->highest_bit = cur_lblock - 1;
4065 out:
4066         if (not_aligned)
4067                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%lu * N)",
4068                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
4069         return ret;
4070 }
4071
4072 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4073                                 sector_t *span)
4074 {
4075         struct inode *inode = file_inode(file);
4076         int ret;
4077
4078         if (!S_ISREG(inode->i_mode))
4079                 return -EINVAL;
4080
4081         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4082                 return -EROFS;
4083
4084         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4085                 f2fs_err(F2FS_I_SB(inode),
4086                         "Swapfile not supported in LFS mode");
4087                 return -EINVAL;
4088         }
4089
4090         ret = f2fs_convert_inline_inode(inode);
4091         if (ret)
4092                 return ret;
4093
4094         if (!f2fs_disable_compressed_file(inode))
4095                 return -EINVAL;
4096
4097         f2fs_precache_extents(inode);
4098
4099         ret = check_swap_activate(sis, file, span);
4100         if (ret < 0)
4101                 return ret;
4102
4103         stat_inc_swapfile_inode(inode);
4104         set_inode_flag(inode, FI_PIN_FILE);
4105         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4106         return ret;
4107 }
4108
4109 static void f2fs_swap_deactivate(struct file *file)
4110 {
4111         struct inode *inode = file_inode(file);
4112
4113         stat_dec_swapfile_inode(inode);
4114         clear_inode_flag(inode, FI_PIN_FILE);
4115 }
4116 #else
4117 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4118                                 sector_t *span)
4119 {
4120         return -EOPNOTSUPP;
4121 }
4122
4123 static void f2fs_swap_deactivate(struct file *file)
4124 {
4125 }
4126 #endif
4127
4128 const struct address_space_operations f2fs_dblock_aops = {
4129         .read_folio     = f2fs_read_data_folio,
4130         .readahead      = f2fs_readahead,
4131         .writepage      = f2fs_write_data_page,
4132         .writepages     = f2fs_write_data_pages,
4133         .write_begin    = f2fs_write_begin,
4134         .write_end      = f2fs_write_end,
4135         .dirty_folio    = f2fs_dirty_data_folio,
4136         .migrate_folio  = filemap_migrate_folio,
4137         .invalidate_folio = f2fs_invalidate_folio,
4138         .release_folio  = f2fs_release_folio,
4139         .bmap           = f2fs_bmap,
4140         .swap_activate  = f2fs_swap_activate,
4141         .swap_deactivate = f2fs_swap_deactivate,
4142 };
4143
4144 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4145 {
4146         struct address_space *mapping = page_mapping(page);
4147         unsigned long flags;
4148
4149         xa_lock_irqsave(&mapping->i_pages, flags);
4150         __xa_clear_mark(&mapping->i_pages, page_index(page),
4151                                                 PAGECACHE_TAG_DIRTY);
4152         xa_unlock_irqrestore(&mapping->i_pages, flags);
4153 }
4154
4155 int __init f2fs_init_post_read_processing(void)
4156 {
4157         bio_post_read_ctx_cache =
4158                 kmem_cache_create("f2fs_bio_post_read_ctx",
4159                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4160         if (!bio_post_read_ctx_cache)
4161                 goto fail;
4162         bio_post_read_ctx_pool =
4163                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4164                                          bio_post_read_ctx_cache);
4165         if (!bio_post_read_ctx_pool)
4166                 goto fail_free_cache;
4167         return 0;
4168
4169 fail_free_cache:
4170         kmem_cache_destroy(bio_post_read_ctx_cache);
4171 fail:
4172         return -ENOMEM;
4173 }
4174
4175 void f2fs_destroy_post_read_processing(void)
4176 {
4177         mempool_destroy(bio_post_read_ctx_pool);
4178         kmem_cache_destroy(bio_post_read_ctx_cache);
4179 }
4180
4181 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4182 {
4183         if (!f2fs_sb_has_encrypt(sbi) &&
4184                 !f2fs_sb_has_verity(sbi) &&
4185                 !f2fs_sb_has_compression(sbi))
4186                 return 0;
4187
4188         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4189                                                  WQ_UNBOUND | WQ_HIGHPRI,
4190                                                  num_online_cpus());
4191         return sbi->post_read_wq ? 0 : -ENOMEM;
4192 }
4193
4194 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4195 {
4196         if (sbi->post_read_wq)
4197                 destroy_workqueue(sbi->post_read_wq);
4198 }
4199
4200 int __init f2fs_init_bio_entry_cache(void)
4201 {
4202         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4203                         sizeof(struct bio_entry));
4204         return bio_entry_slab ? 0 : -ENOMEM;
4205 }
4206
4207 void f2fs_destroy_bio_entry_cache(void)
4208 {
4209         kmem_cache_destroy(bio_entry_slab);
4210 }
4211
4212 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4213                             unsigned int flags, struct iomap *iomap,
4214                             struct iomap *srcmap)
4215 {
4216         struct f2fs_map_blocks map = {};
4217         pgoff_t next_pgofs = 0;
4218         int err;
4219
4220         map.m_lblk = bytes_to_blks(inode, offset);
4221         map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4222         map.m_next_pgofs = &next_pgofs;
4223         map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4224         if (flags & IOMAP_WRITE)
4225                 map.m_may_create = true;
4226
4227         err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4228         if (err)
4229                 return err;
4230
4231         iomap->offset = blks_to_bytes(inode, map.m_lblk);
4232
4233         /*
4234          * When inline encryption is enabled, sometimes I/O to an encrypted file
4235          * has to be broken up to guarantee DUN contiguity.  Handle this by
4236          * limiting the length of the mapping returned.
4237          */
4238         map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4239
4240         /*
4241          * We should never see delalloc or compressed extents here based on
4242          * prior flushing and checks.
4243          */
4244         if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4245                 return -EINVAL;
4246         if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4247                 return -EINVAL;
4248
4249         if (map.m_pblk != NULL_ADDR) {
4250                 iomap->length = blks_to_bytes(inode, map.m_len);
4251                 iomap->type = IOMAP_MAPPED;
4252                 iomap->flags |= IOMAP_F_MERGED;
4253                 iomap->bdev = map.m_bdev;
4254                 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4255         } else {
4256                 if (flags & IOMAP_WRITE)
4257                         return -ENOTBLK;
4258                 iomap->length = blks_to_bytes(inode, next_pgofs) -
4259                                 iomap->offset;
4260                 iomap->type = IOMAP_HOLE;
4261                 iomap->addr = IOMAP_NULL_ADDR;
4262         }
4263
4264         if (map.m_flags & F2FS_MAP_NEW)
4265                 iomap->flags |= IOMAP_F_NEW;
4266         if ((inode->i_state & I_DIRTY_DATASYNC) ||
4267             offset + length > i_size_read(inode))
4268                 iomap->flags |= IOMAP_F_DIRTY;
4269
4270         return 0;
4271 }
4272
4273 const struct iomap_ops f2fs_iomap_ops = {
4274         .iomap_begin    = f2fs_iomap_begin,
4275 };