fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/blk-crypto.h>
  18 #include <linux/swap.h>
  19 #include <linux/prefetch.h>
  20 #include <linux/uio.h>
  21 #include <linux/cleancache.h>
  22 #include <linux/sched/signal.h>
  23 #include <linux/fiemap.h>
  24
  25 #include "f2fs.h"
  26 #include "node.h"
  27 #include "segment.h"
  28 #include <trace/events/f2fs.h>
  29
  30 #define NUM_PREALLOC_POST_READ_CTXS     128
  31
  32 static struct kmem_cache *bio_post_read_ctx_cache;
  33 static struct kmem_cache *bio_entry_slab;
  34 static mempool_t *bio_post_read_ctx_pool;
  35 static struct bio_set f2fs_bioset;
  36
  37 #define F2FS_BIO_POOL_SIZE      NR_CURSEG_TYPE
  38
  39 int __init f2fs_init_bioset(void)
  40 {
  41         if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  42                                         0, BIOSET_NEED_BVECS))
  43                 return -ENOMEM;
  44         return 0;
  45 }
  46
  47 void f2fs_destroy_bioset(void)
  48 {
  49         bioset_exit(&f2fs_bioset);
  50 }
  51
  52 static bool __is_cp_guaranteed(struct page *page)
  53 {
  54         struct address_space *mapping = page->mapping;
  55         struct inode *inode;
  56         struct f2fs_sb_info *sbi;
  57
  58         if (!mapping)
  59                 return false;
  60
  61         inode = mapping->host;
  62         sbi = F2FS_I_SB(inode);
  63
  64         if (inode->i_ino == F2FS_META_INO(sbi) ||
  65                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  66                         S_ISDIR(inode->i_mode))
  67                 return true;
  68
  69         if (f2fs_is_compressed_page(page))
  70                 return false;
  71         if ((S_ISREG(inode->i_mode) &&
  72                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  73                         page_private_gcing(page))
  74                 return true;
  75         return false;
  76 }
  77
  78 static enum count_type __read_io_type(struct page *page)
  79 {
  80         struct address_space *mapping = page_file_mapping(page);
  81
  82         if (mapping) {
  83                 struct inode *inode = mapping->host;
  84                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  85
  86                 if (inode->i_ino == F2FS_META_INO(sbi))
  87                         return F2FS_RD_META;
  88
  89                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  90                         return F2FS_RD_NODE;
  91         }
  92         return F2FS_RD_DATA;
  93 }
  94
  95 /* postprocessing steps for read bios */
  96 enum bio_post_read_step {
  97 #ifdef CONFIG_FS_ENCRYPTION
  98         STEP_DECRYPT    = 1 << 0,
  99 #else
 100         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
 101 #endif
 102 #ifdef CONFIG_F2FS_FS_COMPRESSION
 103         STEP_DECOMPRESS = 1 << 1,
 104 #else
 105         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 106 #endif
 107 #ifdef CONFIG_FS_VERITY
 108         STEP_VERITY     = 1 << 2,
 109 #else
 110         STEP_VERITY     = 0,    /* compile out the verity-related code */
 111 #endif
 112 };
 113
 114 struct bio_post_read_ctx {
 115         struct bio *bio;
 116         struct f2fs_sb_info *sbi;
 117         struct work_struct work;
 118         unsigned int enabled_steps;
 119 };
 120
 121 static void f2fs_finish_read_bio(struct bio *bio)
 122 {
 123         struct bio_vec *bv;
 124         struct bvec_iter_all iter_all;
 125
 126         /*
 127          * Update and unlock the bio's pagecache pages, and put the
 128          * decompression context for any compressed pages.
 129          */
 130         bio_for_each_segment_all(bv, bio, iter_all) {
 131                 struct page *page = bv->bv_page;
 132
 133                 if (f2fs_is_compressed_page(page)) {
 134                         if (bio->bi_status)
 135                                 f2fs_end_read_compressed_page(page, true, 0);
 136                         f2fs_put_page_dic(page);
 137                         continue;
 138                 }
 139
 140                 /* PG_error was set if decryption or verity failed. */
 141                 if (bio->bi_status || PageError(page)) {
 142                         ClearPageUptodate(page);
 143                         /* will re-read again later */
 144                         ClearPageError(page);
 145                 } else {
 146                         SetPageUptodate(page);
 147                 }
 148                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 149                 unlock_page(page);
 150         }
 151
 152         if (bio->bi_private)
 153                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 154         bio_put(bio);
 155 }
 156
 157 static void f2fs_verify_bio(struct work_struct *work)
 158 {
 159         struct bio_post_read_ctx *ctx =
 160                 container_of(work, struct bio_post_read_ctx, work);
 161         struct bio *bio = ctx->bio;
 162         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 163
 164         /*
 165          * fsverity_verify_bio() may call readpages() again, and while verity
 166          * will be disabled for this, decryption and/or decompression may still
 167          * be needed, resulting in another bio_post_read_ctx being allocated.
 168          * So to prevent deadlocks we need to release the current ctx to the
 169          * mempool first.  This assumes that verity is the last post-read step.
 170          */
 171         mempool_free(ctx, bio_post_read_ctx_pool);
 172         bio->bi_private = NULL;
 173
 174         /*
 175          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 176          * as those were handled separately by f2fs_end_read_compressed_page().
 177          */
 178         if (may_have_compressed_pages) {
 179                 struct bio_vec *bv;
 180                 struct bvec_iter_all iter_all;
 181
 182                 bio_for_each_segment_all(bv, bio, iter_all) {
 183                         struct page *page = bv->bv_page;
 184
 185                         if (!f2fs_is_compressed_page(page) &&
 186                             !PageError(page) && !fsverity_verify_page(page))
 187                                 SetPageError(page);
 188                 }
 189         } else {
 190                 fsverity_verify_bio(bio);
 191         }
 192
 193         f2fs_finish_read_bio(bio);
 194 }
 195
 196 /*
 197  * If the bio's data needs to be verified with fs-verity, then enqueue the
 198  * verity work for the bio.  Otherwise finish the bio now.
 199  *
 200  * Note that to avoid deadlocks, the verity work can't be done on the
 201  * decryption/decompression workqueue.  This is because verifying the data pages
 202  * can involve reading verity metadata pages from the file, and these verity
 203  * metadata pages may be encrypted and/or compressed.
 204  */
 205 static void f2fs_verify_and_finish_bio(struct bio *bio)
 206 {
 207         struct bio_post_read_ctx *ctx = bio->bi_private;
 208
 209         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 210                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 211                 fsverity_enqueue_verify_work(&ctx->work);
 212         } else {
 213                 f2fs_finish_read_bio(bio);
 214         }
 215 }
 216
 217 /*
 218  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 219  * remaining page was read by @ctx->bio.
 220  *
 221  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 222  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 223  * that the bio includes at least one compressed page.  The actual decompression
 224  * is done on a per-cluster basis, not a per-bio basis.
 225  */
 226 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
 227 {
 228         struct bio_vec *bv;
 229         struct bvec_iter_all iter_all;
 230         bool all_compressed = true;
 231         block_t blkaddr = SECTOR_TO_BLOCK(ctx->bio->bi_iter.bi_sector);
 232
 233         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 234                 struct page *page = bv->bv_page;
 235
 236                 /* PG_error was set if decryption failed. */
 237                 if (f2fs_is_compressed_page(page))
 238                         f2fs_end_read_compressed_page(page, PageError(page),
 239                                                 blkaddr);
 240                 else
 241                         all_compressed = false;
 242
 243                 blkaddr++;
 244         }
 245
 246         /*
 247          * Optimization: if all the bio's pages are compressed, then scheduling
 248          * the per-bio verity work is unnecessary, as verity will be fully
 249          * handled at the compression cluster level.
 250          */
 251         if (all_compressed)
 252                 ctx->enabled_steps &= ~STEP_VERITY;
 253 }
 254
 255 static void f2fs_post_read_work(struct work_struct *work)
 256 {
 257         struct bio_post_read_ctx *ctx =
 258                 container_of(work, struct bio_post_read_ctx, work);
 259
 260         if (ctx->enabled_steps & STEP_DECRYPT)
 261                 fscrypt_decrypt_bio(ctx->bio);
 262
 263         if (ctx->enabled_steps & STEP_DECOMPRESS)
 264                 f2fs_handle_step_decompress(ctx);
 265
 266         f2fs_verify_and_finish_bio(ctx->bio);
 267 }
 268
 269 static void f2fs_read_end_io(struct bio *bio)
 270 {
 271         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 272         struct bio_post_read_ctx *ctx = bio->bi_private;
 273
 274         if (time_to_inject(sbi, FAULT_READ_IO)) {
 275                 f2fs_show_injection_info(sbi, FAULT_READ_IO);
 276                 bio->bi_status = BLK_STS_IOERR;
 277         }
 278
 279         if (bio->bi_status) {
 280                 f2fs_finish_read_bio(bio);
 281                 return;
 282         }
 283
 284         if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
 285                 INIT_WORK(&ctx->work, f2fs_post_read_work);
 286                 queue_work(ctx->sbi->post_read_wq, &ctx->work);
 287         } else {
 288                 f2fs_verify_and_finish_bio(bio);
 289         }
 290 }
 291
 292 static void f2fs_write_end_io(struct bio *bio)
 293 {
 294         struct f2fs_sb_info *sbi = bio->bi_private;
 295         struct bio_vec *bvec;
 296         struct bvec_iter_all iter_all;
 297
 298         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 299                 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
 300                 bio->bi_status = BLK_STS_IOERR;
 301         }
 302
 303         bio_for_each_segment_all(bvec, bio, iter_all) {
 304                 struct page *page = bvec->bv_page;
 305                 enum count_type type = WB_DATA_TYPE(page);
 306
 307                 if (page_private_dummy(page)) {
 308                         clear_page_private_dummy(page);
 309                         unlock_page(page);
 310                         mempool_free(page, sbi->write_io_dummy);
 311
 312                         if (unlikely(bio->bi_status))
 313                                 f2fs_stop_checkpoint(sbi, true);
 314                         continue;
 315                 }
 316
 317                 fscrypt_finalize_bounce_page(&page);
 318
 319 #ifdef CONFIG_F2FS_FS_COMPRESSION
 320                 if (f2fs_is_compressed_page(page)) {
 321                         f2fs_compress_write_end_io(bio, page);
 322                         continue;
 323                 }
 324 #endif
 325
 326                 if (unlikely(bio->bi_status)) {
 327                         mapping_set_error(page->mapping, -EIO);
 328                         if (type == F2FS_WB_CP_DATA)
 329                                 f2fs_stop_checkpoint(sbi, true);
 330                 }
 331
 332                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 333                                         page->index != nid_of_node(page));
 334
 335                 dec_page_count(sbi, type);
 336                 if (f2fs_in_warm_node_list(sbi, page))
 337                         f2fs_del_fsync_node_entry(sbi, page);
 338                 clear_page_private_gcing(page);
 339                 end_page_writeback(page);
 340         }
 341         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 342                                 wq_has_sleeper(&sbi->cp_wait))
 343                 wake_up(&sbi->cp_wait);
 344
 345         bio_put(bio);
 346 }
 347
 348 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 349                                 block_t blk_addr, struct bio *bio)
 350 {
 351         struct block_device *bdev = sbi->sb->s_bdev;
 352         int i;
 353
 354         if (f2fs_is_multi_device(sbi)) {
 355                 for (i = 0; i < sbi->s_ndevs; i++) {
 356                         if (FDEV(i).start_blk <= blk_addr &&
 357                             FDEV(i).end_blk >= blk_addr) {
 358                                 blk_addr -= FDEV(i).start_blk;
 359                                 bdev = FDEV(i).bdev;
 360                                 break;
 361                         }
 362                 }
 363         }
 364         if (bio) {
 365                 bio_set_dev(bio, bdev);
 366                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 367         }
 368         return bdev;
 369 }
 370
 371 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 372 {
 373         int i;
 374
 375         if (!f2fs_is_multi_device(sbi))
 376                 return 0;
 377
 378         for (i = 0; i < sbi->s_ndevs; i++)
 379                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 380                         return i;
 381         return 0;
 382 }
 383
 384 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 385 {
 386         struct f2fs_sb_info *sbi = fio->sbi;
 387         struct bio *bio;
 388
 389         bio = bio_alloc_bioset(GFP_NOIO, npages, &f2fs_bioset);
 390
 391         f2fs_target_device(sbi, fio->new_blkaddr, bio);
 392         if (is_read_io(fio->op)) {
 393                 bio->bi_end_io = f2fs_read_end_io;
 394                 bio->bi_private = NULL;
 395         } else {
 396                 bio->bi_end_io = f2fs_write_end_io;
 397                 bio->bi_private = sbi;
 398                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 399                                                 fio->type, fio->temp);
 400         }
 401         if (fio->io_wbc)
 402                 wbc_init_bio(fio->io_wbc, bio);
 403
 404         return bio;
 405 }
 406
 407 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 408                                   pgoff_t first_idx,
 409                                   const struct f2fs_io_info *fio,
 410                                   gfp_t gfp_mask)
 411 {
 412         /*
 413          * The f2fs garbage collector sets ->encrypted_page when it wants to
 414          * read/write raw data without encryption.
 415          */
 416         if (!fio || !fio->encrypted_page)
 417                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 418 }
 419
 420 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 421                                      pgoff_t next_idx,
 422                                      const struct f2fs_io_info *fio)
 423 {
 424         /*
 425          * The f2fs garbage collector sets ->encrypted_page when it wants to
 426          * read/write raw data without encryption.
 427          */
 428         if (fio && fio->encrypted_page)
 429                 return !bio_has_crypt_ctx(bio);
 430
 431         return fscrypt_mergeable_bio(bio, inode, next_idx);
 432 }
 433
 434 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 435                                 struct bio *bio, enum page_type type)
 436 {
 437         if (!is_read_io(bio_op(bio))) {
 438                 unsigned int start;
 439
 440                 if (type != DATA && type != NODE)
 441                         goto submit_io;
 442
 443                 if (f2fs_lfs_mode(sbi) && current->plug)
 444                         blk_finish_plug(current->plug);
 445
 446                 if (!F2FS_IO_ALIGNED(sbi))
 447                         goto submit_io;
 448
 449                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 450                 start %= F2FS_IO_SIZE(sbi);
 451
 452                 if (start == 0)
 453                         goto submit_io;
 454
 455                 /* fill dummy pages */
 456                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 457                         struct page *page =
 458                                 mempool_alloc(sbi->write_io_dummy,
 459                                               GFP_NOIO | __GFP_NOFAIL);
 460                         f2fs_bug_on(sbi, !page);
 461
 462                         lock_page(page);
 463
 464                         zero_user_segment(page, 0, PAGE_SIZE);
 465                         set_page_private_dummy(page);
 466
 467                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 468                                 f2fs_bug_on(sbi, 1);
 469                 }
 470                 /*
 471                  * In the NODE case, we lose next block address chain. So, we
 472                  * need to do checkpoint in f2fs_sync_file.
 473                  */
 474                 if (type == NODE)
 475                         set_sbi_flag(sbi, SBI_NEED_CP);
 476         }
 477 submit_io:
 478         if (is_read_io(bio_op(bio)))
 479                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 480         else
 481                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 482         submit_bio(bio);
 483 }
 484
 485 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 486                                 struct bio *bio, enum page_type type)
 487 {
 488         __submit_bio(sbi, bio, type);
 489 }
 490
 491 static void __attach_io_flag(struct f2fs_io_info *fio)
 492 {
 493         struct f2fs_sb_info *sbi = fio->sbi;
 494         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 495         unsigned int io_flag, fua_flag, meta_flag;
 496
 497         if (fio->type == DATA)
 498                 io_flag = sbi->data_io_flag;
 499         else if (fio->type == NODE)
 500                 io_flag = sbi->node_io_flag;
 501         else
 502                 return;
 503
 504         fua_flag = io_flag & temp_mask;
 505         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 506
 507         /*
 508          * data/node io flag bits per temp:
 509          *      REQ_META     |      REQ_FUA      |
 510          *    5 |    4 |   3 |    2 |    1 |   0 |
 511          * Cold | Warm | Hot | Cold | Warm | Hot |
 512          */
 513         if ((1 << fio->temp) & meta_flag)
 514                 fio->op_flags |= REQ_META;
 515         if ((1 << fio->temp) & fua_flag)
 516                 fio->op_flags |= REQ_FUA;
 517 }
 518
 519 static void __submit_merged_bio(struct f2fs_bio_info *io)
 520 {
 521         struct f2fs_io_info *fio = &io->fio;
 522
 523         if (!io->bio)
 524                 return;
 525
 526         __attach_io_flag(fio);
 527         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 528
 529         if (is_read_io(fio->op))
 530                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 531         else
 532                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 533
 534         __submit_bio(io->sbi, io->bio, fio->type);
 535         io->bio = NULL;
 536 }
 537
 538 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 539                                                 struct page *page, nid_t ino)
 540 {
 541         struct bio_vec *bvec;
 542         struct bvec_iter_all iter_all;
 543
 544         if (!bio)
 545                 return false;
 546
 547         if (!inode && !page && !ino)
 548                 return true;
 549
 550         bio_for_each_segment_all(bvec, bio, iter_all) {
 551                 struct page *target = bvec->bv_page;
 552
 553                 if (fscrypt_is_bounce_page(target)) {
 554                         target = fscrypt_pagecache_page(target);
 555                         if (IS_ERR(target))
 556                                 continue;
 557                 }
 558                 if (f2fs_is_compressed_page(target)) {
 559                         target = f2fs_compress_control_page(target);
 560                         if (IS_ERR(target))
 561                                 continue;
 562                 }
 563
 564                 if (inode && inode == target->mapping->host)
 565                         return true;
 566                 if (page && page == target)
 567                         return true;
 568                 if (ino && ino == ino_of_node(target))
 569                         return true;
 570         }
 571
 572         return false;
 573 }
 574
 575 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 576                                 enum page_type type, enum temp_type temp)
 577 {
 578         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 579         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 580
 581         down_write(&io->io_rwsem);
 582
 583         /* change META to META_FLUSH in the checkpoint procedure */
 584         if (type >= META_FLUSH) {
 585                 io->fio.type = META_FLUSH;
 586                 io->fio.op = REQ_OP_WRITE;
 587                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 588                 if (!test_opt(sbi, NOBARRIER))
 589                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 590         }
 591         __submit_merged_bio(io);
 592         up_write(&io->io_rwsem);
 593 }
 594
 595 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 596                                 struct inode *inode, struct page *page,
 597                                 nid_t ino, enum page_type type, bool force)
 598 {
 599         enum temp_type temp;
 600         bool ret = true;
 601
 602         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 603                 if (!force)     {
 604                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 605                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 606
 607                         down_read(&io->io_rwsem);
 608                         ret = __has_merged_page(io->bio, inode, page, ino);
 609                         up_read(&io->io_rwsem);
 610                 }
 611                 if (ret)
 612                         __f2fs_submit_merged_write(sbi, type, temp);
 613
 614                 /* TODO: use HOT temp only for meta pages now. */
 615                 if (type >= META)
 616                         break;
 617         }
 618 }
 619
 620 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 621 {
 622         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 623 }
 624
 625 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 626                                 struct inode *inode, struct page *page,
 627                                 nid_t ino, enum page_type type)
 628 {
 629         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 630 }
 631
 632 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 633 {
 634         f2fs_submit_merged_write(sbi, DATA);
 635         f2fs_submit_merged_write(sbi, NODE);
 636         f2fs_submit_merged_write(sbi, META);
 637 }
 638
 639 /*
 640  * Fill the locked page with data located in the block address.
 641  * A caller needs to unlock the page on failure.
 642  */
 643 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 644 {
 645         struct bio *bio;
 646         struct page *page = fio->encrypted_page ?
 647                         fio->encrypted_page : fio->page;
 648
 649         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 650                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 651                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 652                 return -EFSCORRUPTED;
 653
 654         trace_f2fs_submit_page_bio(page, fio);
 655
 656         /* Allocate a new bio */
 657         bio = __bio_alloc(fio, 1);
 658
 659         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 660                                fio->page->index, fio, GFP_NOIO);
 661
 662         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 663                 bio_put(bio);
 664                 return -EFAULT;
 665         }
 666
 667         if (fio->io_wbc && !is_read_io(fio->op))
 668                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 669
 670         __attach_io_flag(fio);
 671         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 672
 673         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 674                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 675
 676         __submit_bio(fio->sbi, bio, fio->type);
 677         return 0;
 678 }
 679
 680 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 681                                 block_t last_blkaddr, block_t cur_blkaddr)
 682 {
 683         if (unlikely(sbi->max_io_bytes &&
 684                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 685                 return false;
 686         if (last_blkaddr + 1 != cur_blkaddr)
 687                 return false;
 688         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 689 }
 690
 691 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 692                                                 struct f2fs_io_info *fio)
 693 {
 694         if (io->fio.op != fio->op)
 695                 return false;
 696         return io->fio.op_flags == fio->op_flags;
 697 }
 698
 699 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 700                                         struct f2fs_bio_info *io,
 701                                         struct f2fs_io_info *fio,
 702                                         block_t last_blkaddr,
 703                                         block_t cur_blkaddr)
 704 {
 705         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 706                 unsigned int filled_blocks =
 707                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 708                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 709                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 710
 711                 /* IOs in bio is aligned and left space of vectors is not enough */
 712                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 713                         return false;
 714         }
 715         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 716                 return false;
 717         return io_type_is_mergeable(io, fio);
 718 }
 719
 720 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 721                                 struct page *page, enum temp_type temp)
 722 {
 723         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 724         struct bio_entry *be;
 725
 726         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 727         be->bio = bio;
 728         bio_get(bio);
 729
 730         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 731                 f2fs_bug_on(sbi, 1);
 732
 733         down_write(&io->bio_list_lock);
 734         list_add_tail(&be->list, &io->bio_list);
 735         up_write(&io->bio_list_lock);
 736 }
 737
 738 static void del_bio_entry(struct bio_entry *be)
 739 {
 740         list_del(&be->list);
 741         kmem_cache_free(bio_entry_slab, be);
 742 }
 743
 744 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 745                                                         struct page *page)
 746 {
 747         struct f2fs_sb_info *sbi = fio->sbi;
 748         enum temp_type temp;
 749         bool found = false;
 750         int ret = -EAGAIN;
 751
 752         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 753                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 754                 struct list_head *head = &io->bio_list;
 755                 struct bio_entry *be;
 756
 757                 down_write(&io->bio_list_lock);
 758                 list_for_each_entry(be, head, list) {
 759                         if (be->bio != *bio)
 760                                 continue;
 761
 762                         found = true;
 763
 764                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 765                                                             *fio->last_block,
 766                                                             fio->new_blkaddr));
 767                         if (f2fs_crypt_mergeable_bio(*bio,
 768                                         fio->page->mapping->host,
 769                                         fio->page->index, fio) &&
 770                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 771                                         PAGE_SIZE) {
 772                                 ret = 0;
 773                                 break;
 774                         }
 775
 776                         /* page can't be merged into bio; submit the bio */
 777                         del_bio_entry(be);
 778                         __submit_bio(sbi, *bio, DATA);
 779                         break;
 780                 }
 781                 up_write(&io->bio_list_lock);
 782         }
 783
 784         if (ret) {
 785                 bio_put(*bio);
 786                 *bio = NULL;
 787         }
 788
 789         return ret;
 790 }
 791
 792 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 793                                         struct bio **bio, struct page *page)
 794 {
 795         enum temp_type temp;
 796         bool found = false;
 797         struct bio *target = bio ? *bio : NULL;
 798
 799         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 800                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 801                 struct list_head *head = &io->bio_list;
 802                 struct bio_entry *be;
 803
 804                 if (list_empty(head))
 805                         continue;
 806
 807                 down_read(&io->bio_list_lock);
 808                 list_for_each_entry(be, head, list) {
 809                         if (target)
 810                                 found = (target == be->bio);
 811                         else
 812                                 found = __has_merged_page(be->bio, NULL,
 813                                                                 page, 0);
 814                         if (found)
 815                                 break;
 816                 }
 817                 up_read(&io->bio_list_lock);
 818
 819                 if (!found)
 820                         continue;
 821
 822                 found = false;
 823
 824                 down_write(&io->bio_list_lock);
 825                 list_for_each_entry(be, head, list) {
 826                         if (target)
 827                                 found = (target == be->bio);
 828                         else
 829                                 found = __has_merged_page(be->bio, NULL,
 830                                                                 page, 0);
 831                         if (found) {
 832                                 target = be->bio;
 833                                 del_bio_entry(be);
 834                                 break;
 835                         }
 836                 }
 837                 up_write(&io->bio_list_lock);
 838         }
 839
 840         if (found)
 841                 __submit_bio(sbi, target, DATA);
 842         if (bio && *bio) {
 843                 bio_put(*bio);
 844                 *bio = NULL;
 845         }
 846 }
 847
 848 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 849 {
 850         struct bio *bio = *fio->bio;
 851         struct page *page = fio->encrypted_page ?
 852                         fio->encrypted_page : fio->page;
 853
 854         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 855                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 856                 return -EFSCORRUPTED;
 857
 858         trace_f2fs_submit_page_bio(page, fio);
 859
 860         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 861                                                 fio->new_blkaddr))
 862                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 863 alloc_new:
 864         if (!bio) {
 865                 bio = __bio_alloc(fio, BIO_MAX_VECS);
 866                 __attach_io_flag(fio);
 867                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 868                                        fio->page->index, fio, GFP_NOIO);
 869                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 870
 871                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 872         } else {
 873                 if (add_ipu_page(fio, &bio, page))
 874                         goto alloc_new;
 875         }
 876
 877         if (fio->io_wbc)
 878                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 879
 880         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 881
 882         *fio->last_block = fio->new_blkaddr;
 883         *fio->bio = bio;
 884
 885         return 0;
 886 }
 887
 888 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 889 {
 890         struct f2fs_sb_info *sbi = fio->sbi;
 891         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 892         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 893         struct page *bio_page;
 894
 895         f2fs_bug_on(sbi, is_read_io(fio->op));
 896
 897         down_write(&io->io_rwsem);
 898 next:
 899         if (fio->in_list) {
 900                 spin_lock(&io->io_lock);
 901                 if (list_empty(&io->io_list)) {
 902                         spin_unlock(&io->io_lock);
 903                         goto out;
 904                 }
 905                 fio = list_first_entry(&io->io_list,
 906                                                 struct f2fs_io_info, list);
 907                 list_del(&fio->list);
 908                 spin_unlock(&io->io_lock);
 909         }
 910
 911         verify_fio_blkaddr(fio);
 912
 913         if (fio->encrypted_page)
 914                 bio_page = fio->encrypted_page;
 915         else if (fio->compressed_page)
 916                 bio_page = fio->compressed_page;
 917         else
 918                 bio_page = fio->page;
 919
 920         /* set submitted = true as a return value */
 921         fio->submitted = true;
 922
 923         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 924
 925         if (io->bio &&
 926             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 927                               fio->new_blkaddr) ||
 928              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 929                                        bio_page->index, fio)))
 930                 __submit_merged_bio(io);
 931 alloc_new:
 932         if (io->bio == NULL) {
 933                 if (F2FS_IO_ALIGNED(sbi) &&
 934                                 (fio->type == DATA || fio->type == NODE) &&
 935                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 936                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 937                         fio->retry = true;
 938                         goto skip;
 939                 }
 940                 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
 941                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 942                                        bio_page->index, fio, GFP_NOIO);
 943                 io->fio = *fio;
 944         }
 945
 946         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 947                 __submit_merged_bio(io);
 948                 goto alloc_new;
 949         }
 950
 951         if (fio->io_wbc)
 952                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
 953
 954         io->last_block_in_bio = fio->new_blkaddr;
 955
 956         trace_f2fs_submit_page_write(fio->page, fio);
 957 skip:
 958         if (fio->in_list)
 959                 goto next;
 960 out:
 961         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 962                                 !f2fs_is_checkpoint_ready(sbi))
 963                 __submit_merged_bio(io);
 964         up_write(&io->io_rwsem);
 965 }
 966
 967 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 968                                       unsigned nr_pages, unsigned op_flag,
 969                                       pgoff_t first_idx, bool for_write)
 970 {
 971         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 972         struct bio *bio;
 973         struct bio_post_read_ctx *ctx;
 974         unsigned int post_read_steps = 0;
 975
 976         bio = bio_alloc_bioset(for_write ? GFP_NOIO : GFP_KERNEL,
 977                                bio_max_segs(nr_pages), &f2fs_bioset);
 978         if (!bio)
 979                 return ERR_PTR(-ENOMEM);
 980
 981         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
 982
 983         f2fs_target_device(sbi, blkaddr, bio);
 984         bio->bi_end_io = f2fs_read_end_io;
 985         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 986
 987         if (fscrypt_inode_uses_fs_layer_crypto(inode))
 988                 post_read_steps |= STEP_DECRYPT;
 989
 990         if (f2fs_need_verity(inode, first_idx))
 991                 post_read_steps |= STEP_VERITY;
 992
 993         /*
 994          * STEP_DECOMPRESS is handled specially, since a compressed file might
 995          * contain both compressed and uncompressed clusters.  We'll allocate a
 996          * bio_post_read_ctx if the file is compressed, but the caller is
 997          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
 998          */
 999
1000         if (post_read_steps || f2fs_compressed_file(inode)) {
1001                 /* Due to the mempool, this never fails. */
1002                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1003                 ctx->bio = bio;
1004                 ctx->sbi = sbi;
1005                 ctx->enabled_steps = post_read_steps;
1006                 bio->bi_private = ctx;
1007         }
1008
1009         return bio;
1010 }
1011
1012 /* This can handle encryption stuffs */
1013 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1014                                  block_t blkaddr, int op_flags, bool for_write)
1015 {
1016         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1017         struct bio *bio;
1018
1019         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1020                                         page->index, for_write);
1021         if (IS_ERR(bio))
1022                 return PTR_ERR(bio);
1023
1024         /* wait for GCed page writeback via META_MAPPING */
1025         f2fs_wait_on_block_writeback(inode, blkaddr);
1026
1027         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1028                 bio_put(bio);
1029                 return -EFAULT;
1030         }
1031         ClearPageError(page);
1032         inc_page_count(sbi, F2FS_RD_DATA);
1033         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1034         __submit_bio(sbi, bio, DATA);
1035         return 0;
1036 }
1037
1038 static void __set_data_blkaddr(struct dnode_of_data *dn)
1039 {
1040         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1041         __le32 *addr_array;
1042         int base = 0;
1043
1044         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1045                 base = get_extra_isize(dn->inode);
1046
1047         /* Get physical address of data block */
1048         addr_array = blkaddr_in_node(rn);
1049         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1050 }
1051
1052 /*
1053  * Lock ordering for the change of data block address:
1054  * ->data_page
1055  *  ->node_page
1056  *    update block addresses in the node page
1057  */
1058 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1059 {
1060         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1061         __set_data_blkaddr(dn);
1062         if (set_page_dirty(dn->node_page))
1063                 dn->node_changed = true;
1064 }
1065
1066 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1067 {
1068         dn->data_blkaddr = blkaddr;
1069         f2fs_set_data_blkaddr(dn);
1070         f2fs_update_extent_cache(dn);
1071 }
1072
1073 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1074 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1075 {
1076         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1077         int err;
1078
1079         if (!count)
1080                 return 0;
1081
1082         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1083                 return -EPERM;
1084         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1085                 return err;
1086
1087         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1088                                                 dn->ofs_in_node, count);
1089
1090         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1091
1092         for (; count > 0; dn->ofs_in_node++) {
1093                 block_t blkaddr = f2fs_data_blkaddr(dn);
1094
1095                 if (blkaddr == NULL_ADDR) {
1096                         dn->data_blkaddr = NEW_ADDR;
1097                         __set_data_blkaddr(dn);
1098                         count--;
1099                 }
1100         }
1101
1102         if (set_page_dirty(dn->node_page))
1103                 dn->node_changed = true;
1104         return 0;
1105 }
1106
1107 /* Should keep dn->ofs_in_node unchanged */
1108 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1109 {
1110         unsigned int ofs_in_node = dn->ofs_in_node;
1111         int ret;
1112
1113         ret = f2fs_reserve_new_blocks(dn, 1);
1114         dn->ofs_in_node = ofs_in_node;
1115         return ret;
1116 }
1117
1118 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1119 {
1120         bool need_put = dn->inode_page ? false : true;
1121         int err;
1122
1123         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1124         if (err)
1125                 return err;
1126
1127         if (dn->data_blkaddr == NULL_ADDR)
1128                 err = f2fs_reserve_new_block(dn);
1129         if (err || need_put)
1130                 f2fs_put_dnode(dn);
1131         return err;
1132 }
1133
1134 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1135 {
1136         struct extent_info ei = {0, 0, 0};
1137         struct inode *inode = dn->inode;
1138
1139         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1140                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1141                 return 0;
1142         }
1143
1144         return f2fs_reserve_block(dn, index);
1145 }
1146
1147 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1148                                                 int op_flags, bool for_write)
1149 {
1150         struct address_space *mapping = inode->i_mapping;
1151         struct dnode_of_data dn;
1152         struct page *page;
1153         struct extent_info ei = {0,0,0};
1154         int err;
1155
1156         page = f2fs_grab_cache_page(mapping, index, for_write);
1157         if (!page)
1158                 return ERR_PTR(-ENOMEM);
1159
1160         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1161                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1162                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1163                                                 DATA_GENERIC_ENHANCE_READ)) {
1164                         err = -EFSCORRUPTED;
1165                         goto put_err;
1166                 }
1167                 goto got_it;
1168         }
1169
1170         set_new_dnode(&dn, inode, NULL, NULL, 0);
1171         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1172         if (err)
1173                 goto put_err;
1174         f2fs_put_dnode(&dn);
1175
1176         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1177                 err = -ENOENT;
1178                 goto put_err;
1179         }
1180         if (dn.data_blkaddr != NEW_ADDR &&
1181                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1182                                                 dn.data_blkaddr,
1183                                                 DATA_GENERIC_ENHANCE)) {
1184                 err = -EFSCORRUPTED;
1185                 goto put_err;
1186         }
1187 got_it:
1188         if (PageUptodate(page)) {
1189                 unlock_page(page);
1190                 return page;
1191         }
1192
1193         /*
1194          * A new dentry page is allocated but not able to be written, since its
1195          * new inode page couldn't be allocated due to -ENOSPC.
1196          * In such the case, its blkaddr can be remained as NEW_ADDR.
1197          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1198          * f2fs_init_inode_metadata.
1199          */
1200         if (dn.data_blkaddr == NEW_ADDR) {
1201                 zero_user_segment(page, 0, PAGE_SIZE);
1202                 if (!PageUptodate(page))
1203                         SetPageUptodate(page);
1204                 unlock_page(page);
1205                 return page;
1206         }
1207
1208         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1209                                                 op_flags, for_write);
1210         if (err)
1211                 goto put_err;
1212         return page;
1213
1214 put_err:
1215         f2fs_put_page(page, 1);
1216         return ERR_PTR(err);
1217 }
1218
1219 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1220 {
1221         struct address_space *mapping = inode->i_mapping;
1222         struct page *page;
1223
1224         page = find_get_page(mapping, index);
1225         if (page && PageUptodate(page))
1226                 return page;
1227         f2fs_put_page(page, 0);
1228
1229         page = f2fs_get_read_data_page(inode, index, 0, false);
1230         if (IS_ERR(page))
1231                 return page;
1232
1233         if (PageUptodate(page))
1234                 return page;
1235
1236         wait_on_page_locked(page);
1237         if (unlikely(!PageUptodate(page))) {
1238                 f2fs_put_page(page, 0);
1239                 return ERR_PTR(-EIO);
1240         }
1241         return page;
1242 }
1243
1244 /*
1245  * If it tries to access a hole, return an error.
1246  * Because, the callers, functions in dir.c and GC, should be able to know
1247  * whether this page exists or not.
1248  */
1249 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1250                                                         bool for_write)
1251 {
1252         struct address_space *mapping = inode->i_mapping;
1253         struct page *page;
1254 repeat:
1255         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1256         if (IS_ERR(page))
1257                 return page;
1258
1259         /* wait for read completion */
1260         lock_page(page);
1261         if (unlikely(page->mapping != mapping)) {
1262                 f2fs_put_page(page, 1);
1263                 goto repeat;
1264         }
1265         if (unlikely(!PageUptodate(page))) {
1266                 f2fs_put_page(page, 1);
1267                 return ERR_PTR(-EIO);
1268         }
1269         return page;
1270 }
1271
1272 /*
1273  * Caller ensures that this data page is never allocated.
1274  * A new zero-filled data page is allocated in the page cache.
1275  *
1276  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1277  * f2fs_unlock_op().
1278  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1279  * ipage should be released by this function.
1280  */
1281 struct page *f2fs_get_new_data_page(struct inode *inode,
1282                 struct page *ipage, pgoff_t index, bool new_i_size)
1283 {
1284         struct address_space *mapping = inode->i_mapping;
1285         struct page *page;
1286         struct dnode_of_data dn;
1287         int err;
1288
1289         page = f2fs_grab_cache_page(mapping, index, true);
1290         if (!page) {
1291                 /*
1292                  * before exiting, we should make sure ipage will be released
1293                  * if any error occur.
1294                  */
1295                 f2fs_put_page(ipage, 1);
1296                 return ERR_PTR(-ENOMEM);
1297         }
1298
1299         set_new_dnode(&dn, inode, ipage, NULL, 0);
1300         err = f2fs_reserve_block(&dn, index);
1301         if (err) {
1302                 f2fs_put_page(page, 1);
1303                 return ERR_PTR(err);
1304         }
1305         if (!ipage)
1306                 f2fs_put_dnode(&dn);
1307
1308         if (PageUptodate(page))
1309                 goto got_it;
1310
1311         if (dn.data_blkaddr == NEW_ADDR) {
1312                 zero_user_segment(page, 0, PAGE_SIZE);
1313                 if (!PageUptodate(page))
1314                         SetPageUptodate(page);
1315         } else {
1316                 f2fs_put_page(page, 1);
1317
1318                 /* if ipage exists, blkaddr should be NEW_ADDR */
1319                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1320                 page = f2fs_get_lock_data_page(inode, index, true);
1321                 if (IS_ERR(page))
1322                         return page;
1323         }
1324 got_it:
1325         if (new_i_size && i_size_read(inode) <
1326                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1327                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1328         return page;
1329 }
1330
1331 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1332 {
1333         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1334         struct f2fs_summary sum;
1335         struct node_info ni;
1336         block_t old_blkaddr;
1337         blkcnt_t count = 1;
1338         int err;
1339
1340         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1341                 return -EPERM;
1342
1343         err = f2fs_get_node_info(sbi, dn->nid, &ni);
1344         if (err)
1345                 return err;
1346
1347         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1348         if (dn->data_blkaddr != NULL_ADDR)
1349                 goto alloc;
1350
1351         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1352                 return err;
1353
1354 alloc:
1355         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1356         old_blkaddr = dn->data_blkaddr;
1357         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1358                                 &sum, seg_type, NULL);
1359         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1360                 invalidate_mapping_pages(META_MAPPING(sbi),
1361                                         old_blkaddr, old_blkaddr);
1362                 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1363         }
1364         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1365
1366         /*
1367          * i_size will be updated by direct_IO. Otherwise, we'll get stale
1368          * data from unwritten block via dio_read.
1369          */
1370         return 0;
1371 }
1372
1373 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1374 {
1375         struct inode *inode = file_inode(iocb->ki_filp);
1376         struct f2fs_map_blocks map;
1377         int flag;
1378         int err = 0;
1379         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1380
1381         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1382         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1383         if (map.m_len > map.m_lblk)
1384                 map.m_len -= map.m_lblk;
1385         else
1386                 map.m_len = 0;
1387
1388         map.m_next_pgofs = NULL;
1389         map.m_next_extent = NULL;
1390         map.m_seg_type = NO_CHECK_TYPE;
1391         map.m_may_create = true;
1392
1393         if (direct_io) {
1394                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1395                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1396                                         F2FS_GET_BLOCK_PRE_AIO :
1397                                         F2FS_GET_BLOCK_PRE_DIO;
1398                 goto map_blocks;
1399         }
1400         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1401                 err = f2fs_convert_inline_inode(inode);
1402                 if (err)
1403                         return err;
1404         }
1405         if (f2fs_has_inline_data(inode))
1406                 return err;
1407
1408         flag = F2FS_GET_BLOCK_PRE_AIO;
1409
1410 map_blocks:
1411         err = f2fs_map_blocks(inode, &map, 1, flag);
1412         if (map.m_len > 0 && err == -ENOSPC) {
1413                 if (!direct_io)
1414                         set_inode_flag(inode, FI_NO_PREALLOC);
1415                 err = 0;
1416         }
1417         return err;
1418 }
1419
1420 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1421 {
1422         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1423                 if (lock)
1424                         down_read(&sbi->node_change);
1425                 else
1426                         up_read(&sbi->node_change);
1427         } else {
1428                 if (lock)
1429                         f2fs_lock_op(sbi);
1430                 else
1431                         f2fs_unlock_op(sbi);
1432         }
1433 }
1434
1435 /*
1436  * f2fs_map_blocks() tries to find or build mapping relationship which
1437  * maps continuous logical blocks to physical blocks, and return such
1438  * info via f2fs_map_blocks structure.
1439  */
1440 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1441                                                 int create, int flag)
1442 {
1443         unsigned int maxblocks = map->m_len;
1444         struct dnode_of_data dn;
1445         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1446         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1447         pgoff_t pgofs, end_offset, end;
1448         int err = 0, ofs = 1;
1449         unsigned int ofs_in_node, last_ofs_in_node;
1450         blkcnt_t prealloc;
1451         struct extent_info ei = {0,0,0};
1452         block_t blkaddr;
1453         unsigned int start_pgofs;
1454
1455         if (!maxblocks)
1456                 return 0;
1457
1458         map->m_len = 0;
1459         map->m_flags = 0;
1460
1461         /* it only supports block size == page size */
1462         pgofs = (pgoff_t)map->m_lblk;
1463         end = pgofs + maxblocks;
1464
1465         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1466                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1467                                                         map->m_may_create)
1468                         goto next_dnode;
1469
1470                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1471                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1472                 map->m_flags = F2FS_MAP_MAPPED;
1473                 if (map->m_next_extent)
1474                         *map->m_next_extent = pgofs + map->m_len;
1475
1476                 /* for hardware encryption, but to avoid potential issue in future */
1477                 if (flag == F2FS_GET_BLOCK_DIO)
1478                         f2fs_wait_on_block_writeback_range(inode,
1479                                                 map->m_pblk, map->m_len);
1480                 goto out;
1481         }
1482
1483 next_dnode:
1484         if (map->m_may_create)
1485                 f2fs_do_map_lock(sbi, flag, true);
1486
1487         /* When reading holes, we need its node page */
1488         set_new_dnode(&dn, inode, NULL, NULL, 0);
1489         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1490         if (err) {
1491                 if (flag == F2FS_GET_BLOCK_BMAP)
1492                         map->m_pblk = 0;
1493                 if (err == -ENOENT) {
1494                         err = 0;
1495                         if (map->m_next_pgofs)
1496                                 *map->m_next_pgofs =
1497                                         f2fs_get_next_page_offset(&dn, pgofs);
1498                         if (map->m_next_extent)
1499                                 *map->m_next_extent =
1500                                         f2fs_get_next_page_offset(&dn, pgofs);
1501                 }
1502                 goto unlock_out;
1503         }
1504
1505         start_pgofs = pgofs;
1506         prealloc = 0;
1507         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1508         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1509
1510 next_block:
1511         blkaddr = f2fs_data_blkaddr(&dn);
1512
1513         if (__is_valid_data_blkaddr(blkaddr) &&
1514                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1515                 err = -EFSCORRUPTED;
1516                 goto sync_out;
1517         }
1518
1519         if (__is_valid_data_blkaddr(blkaddr)) {
1520                 /* use out-place-update for driect IO under LFS mode */
1521                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1522                                                         map->m_may_create) {
1523                         err = __allocate_data_block(&dn, map->m_seg_type);
1524                         if (err)
1525                                 goto sync_out;
1526                         blkaddr = dn.data_blkaddr;
1527                         set_inode_flag(inode, FI_APPEND_WRITE);
1528                 }
1529         } else {
1530                 if (create) {
1531                         if (unlikely(f2fs_cp_error(sbi))) {
1532                                 err = -EIO;
1533                                 goto sync_out;
1534                         }
1535                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1536                                 if (blkaddr == NULL_ADDR) {
1537                                         prealloc++;
1538                                         last_ofs_in_node = dn.ofs_in_node;
1539                                 }
1540                         } else {
1541                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1542                                         flag != F2FS_GET_BLOCK_DIO);
1543                                 err = __allocate_data_block(&dn,
1544                                                         map->m_seg_type);
1545                                 if (!err)
1546                                         set_inode_flag(inode, FI_APPEND_WRITE);
1547                         }
1548                         if (err)
1549                                 goto sync_out;
1550                         map->m_flags |= F2FS_MAP_NEW;
1551                         blkaddr = dn.data_blkaddr;
1552                 } else {
1553                         if (flag == F2FS_GET_BLOCK_BMAP) {
1554                                 map->m_pblk = 0;
1555                                 goto sync_out;
1556                         }
1557                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1558                                 goto sync_out;
1559                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1560                                                 blkaddr == NULL_ADDR) {
1561                                 if (map->m_next_pgofs)
1562                                         *map->m_next_pgofs = pgofs + 1;
1563                                 goto sync_out;
1564                         }
1565                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1566                                 /* for defragment case */
1567                                 if (map->m_next_pgofs)
1568                                         *map->m_next_pgofs = pgofs + 1;
1569                                 goto sync_out;
1570                         }
1571                 }
1572         }
1573
1574         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1575                 goto skip;
1576
1577         if (map->m_len == 0) {
1578                 /* preallocated unwritten block should be mapped for fiemap. */
1579                 if (blkaddr == NEW_ADDR)
1580                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1581                 map->m_flags |= F2FS_MAP_MAPPED;
1582
1583                 map->m_pblk = blkaddr;
1584                 map->m_len = 1;
1585         } else if ((map->m_pblk != NEW_ADDR &&
1586                         blkaddr == (map->m_pblk + ofs)) ||
1587                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1588                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1589                 ofs++;
1590                 map->m_len++;
1591         } else {
1592                 goto sync_out;
1593         }
1594
1595 skip:
1596         dn.ofs_in_node++;
1597         pgofs++;
1598
1599         /* preallocate blocks in batch for one dnode page */
1600         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1601                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1602
1603                 dn.ofs_in_node = ofs_in_node;
1604                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1605                 if (err)
1606                         goto sync_out;
1607
1608                 map->m_len += dn.ofs_in_node - ofs_in_node;
1609                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1610                         err = -ENOSPC;
1611                         goto sync_out;
1612                 }
1613                 dn.ofs_in_node = end_offset;
1614         }
1615
1616         if (pgofs >= end)
1617                 goto sync_out;
1618         else if (dn.ofs_in_node < end_offset)
1619                 goto next_block;
1620
1621         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1622                 if (map->m_flags & F2FS_MAP_MAPPED) {
1623                         unsigned int ofs = start_pgofs - map->m_lblk;
1624
1625                         f2fs_update_extent_cache_range(&dn,
1626                                 start_pgofs, map->m_pblk + ofs,
1627                                 map->m_len - ofs);
1628                 }
1629         }
1630
1631         f2fs_put_dnode(&dn);
1632
1633         if (map->m_may_create) {
1634                 f2fs_do_map_lock(sbi, flag, false);
1635                 f2fs_balance_fs(sbi, dn.node_changed);
1636         }
1637         goto next_dnode;
1638
1639 sync_out:
1640
1641         /* for hardware encryption, but to avoid potential issue in future */
1642         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1643                 f2fs_wait_on_block_writeback_range(inode,
1644                                                 map->m_pblk, map->m_len);
1645
1646         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1647                 if (map->m_flags & F2FS_MAP_MAPPED) {
1648                         unsigned int ofs = start_pgofs - map->m_lblk;
1649
1650                         f2fs_update_extent_cache_range(&dn,
1651                                 start_pgofs, map->m_pblk + ofs,
1652                                 map->m_len - ofs);
1653                 }
1654                 if (map->m_next_extent)
1655                         *map->m_next_extent = pgofs + 1;
1656         }
1657         f2fs_put_dnode(&dn);
1658 unlock_out:
1659         if (map->m_may_create) {
1660                 f2fs_do_map_lock(sbi, flag, false);
1661                 f2fs_balance_fs(sbi, dn.node_changed);
1662         }
1663 out:
1664         trace_f2fs_map_blocks(inode, map, err);
1665         return err;
1666 }
1667
1668 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1669 {
1670         struct f2fs_map_blocks map;
1671         block_t last_lblk;
1672         int err;
1673
1674         if (pos + len > i_size_read(inode))
1675                 return false;
1676
1677         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1678         map.m_next_pgofs = NULL;
1679         map.m_next_extent = NULL;
1680         map.m_seg_type = NO_CHECK_TYPE;
1681         map.m_may_create = false;
1682         last_lblk = F2FS_BLK_ALIGN(pos + len);
1683
1684         while (map.m_lblk < last_lblk) {
1685                 map.m_len = last_lblk - map.m_lblk;
1686                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1687                 if (err || map.m_len == 0)
1688                         return false;
1689                 map.m_lblk += map.m_len;
1690         }
1691         return true;
1692 }
1693
1694 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1695 {
1696         return (bytes >> inode->i_blkbits);
1697 }
1698
1699 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1700 {
1701         return (blks << inode->i_blkbits);
1702 }
1703
1704 static int __get_data_block(struct inode *inode, sector_t iblock,
1705                         struct buffer_head *bh, int create, int flag,
1706                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1707 {
1708         struct f2fs_map_blocks map;
1709         int err;
1710
1711         map.m_lblk = iblock;
1712         map.m_len = bytes_to_blks(inode, bh->b_size);
1713         map.m_next_pgofs = next_pgofs;
1714         map.m_next_extent = NULL;
1715         map.m_seg_type = seg_type;
1716         map.m_may_create = may_write;
1717
1718         err = f2fs_map_blocks(inode, &map, create, flag);
1719         if (!err) {
1720                 map_bh(bh, inode->i_sb, map.m_pblk);
1721                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1722                 bh->b_size = blks_to_bytes(inode, map.m_len);
1723         }
1724         return err;
1725 }
1726
1727 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1728                         struct buffer_head *bh_result, int create)
1729 {
1730         return __get_data_block(inode, iblock, bh_result, create,
1731                                 F2FS_GET_BLOCK_DIO, NULL,
1732                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1733                                 true);
1734 }
1735
1736 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1737                         struct buffer_head *bh_result, int create)
1738 {
1739         return __get_data_block(inode, iblock, bh_result, create,
1740                                 F2FS_GET_BLOCK_DIO, NULL,
1741                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1742                                 false);
1743 }
1744
1745 static int f2fs_xattr_fiemap(struct inode *inode,
1746                                 struct fiemap_extent_info *fieinfo)
1747 {
1748         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1749         struct page *page;
1750         struct node_info ni;
1751         __u64 phys = 0, len;
1752         __u32 flags;
1753         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1754         int err = 0;
1755
1756         if (f2fs_has_inline_xattr(inode)) {
1757                 int offset;
1758
1759                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1760                                                 inode->i_ino, false);
1761                 if (!page)
1762                         return -ENOMEM;
1763
1764                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1765                 if (err) {
1766                         f2fs_put_page(page, 1);
1767                         return err;
1768                 }
1769
1770                 phys = blks_to_bytes(inode, ni.blk_addr);
1771                 offset = offsetof(struct f2fs_inode, i_addr) +
1772                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1773                                         get_inline_xattr_addrs(inode));
1774
1775                 phys += offset;
1776                 len = inline_xattr_size(inode);
1777
1778                 f2fs_put_page(page, 1);
1779
1780                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1781
1782                 if (!xnid)
1783                         flags |= FIEMAP_EXTENT_LAST;
1784
1785                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1786                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1787                 if (err || err == 1)
1788                         return err;
1789         }
1790
1791         if (xnid) {
1792                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1793                 if (!page)
1794                         return -ENOMEM;
1795
1796                 err = f2fs_get_node_info(sbi, xnid, &ni);
1797                 if (err) {
1798                         f2fs_put_page(page, 1);
1799                         return err;
1800                 }
1801
1802                 phys = blks_to_bytes(inode, ni.blk_addr);
1803                 len = inode->i_sb->s_blocksize;
1804
1805                 f2fs_put_page(page, 1);
1806
1807                 flags = FIEMAP_EXTENT_LAST;
1808         }
1809
1810         if (phys) {
1811                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1812                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1813         }
1814
1815         return (err < 0 ? err : 0);
1816 }
1817
1818 static loff_t max_inode_blocks(struct inode *inode)
1819 {
1820         loff_t result = ADDRS_PER_INODE(inode);
1821         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1822
1823         /* two direct node blocks */
1824         result += (leaf_count * 2);
1825
1826         /* two indirect node blocks */
1827         leaf_count *= NIDS_PER_BLOCK;
1828         result += (leaf_count * 2);
1829
1830         /* one double indirect node block */
1831         leaf_count *= NIDS_PER_BLOCK;
1832         result += leaf_count;
1833
1834         return result;
1835 }
1836
1837 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1838                 u64 start, u64 len)
1839 {
1840         struct f2fs_map_blocks map;
1841         sector_t start_blk, last_blk;
1842         pgoff_t next_pgofs;
1843         u64 logical = 0, phys = 0, size = 0;
1844         u32 flags = 0;
1845         int ret = 0;
1846         bool compr_cluster = false;
1847         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1848         loff_t maxbytes;
1849
1850         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1851                 ret = f2fs_precache_extents(inode);
1852                 if (ret)
1853                         return ret;
1854         }
1855
1856         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1857         if (ret)
1858                 return ret;
1859
1860         inode_lock(inode);
1861
1862         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1863         if (start > maxbytes) {
1864                 ret = -EFBIG;
1865                 goto out;
1866         }
1867
1868         if (len > maxbytes || (maxbytes - len) < start)
1869                 len = maxbytes - start;
1870
1871         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1872                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1873                 goto out;
1874         }
1875
1876         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1877                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1878                 if (ret != -EAGAIN)
1879                         goto out;
1880         }
1881
1882         if (bytes_to_blks(inode, len) == 0)
1883                 len = blks_to_bytes(inode, 1);
1884
1885         start_blk = bytes_to_blks(inode, start);
1886         last_blk = bytes_to_blks(inode, start + len - 1);
1887
1888 next:
1889         memset(&map, 0, sizeof(map));
1890         map.m_lblk = start_blk;
1891         map.m_len = bytes_to_blks(inode, len);
1892         map.m_next_pgofs = &next_pgofs;
1893         map.m_seg_type = NO_CHECK_TYPE;
1894
1895         if (compr_cluster)
1896                 map.m_len = cluster_size - 1;
1897
1898         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1899         if (ret)
1900                 goto out;
1901
1902         /* HOLE */
1903         if (!(map.m_flags & F2FS_MAP_FLAGS)) {
1904                 start_blk = next_pgofs;
1905
1906                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1907                                                 max_inode_blocks(inode)))
1908                         goto prep_next;
1909
1910                 flags |= FIEMAP_EXTENT_LAST;
1911         }
1912
1913         if (size) {
1914                 flags |= FIEMAP_EXTENT_MERGED;
1915                 if (IS_ENCRYPTED(inode))
1916                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1917
1918                 ret = fiemap_fill_next_extent(fieinfo, logical,
1919                                 phys, size, flags);
1920                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1921                 if (ret)
1922                         goto out;
1923                 size = 0;
1924         }
1925
1926         if (start_blk > last_blk)
1927                 goto out;
1928
1929         if (compr_cluster) {
1930                 compr_cluster = false;
1931
1932
1933                 logical = blks_to_bytes(inode, start_blk - 1);
1934                 phys = blks_to_bytes(inode, map.m_pblk);
1935                 size = blks_to_bytes(inode, cluster_size);
1936
1937                 flags |= FIEMAP_EXTENT_ENCODED;
1938
1939                 start_blk += cluster_size - 1;
1940
1941                 if (start_blk > last_blk)
1942                         goto out;
1943
1944                 goto prep_next;
1945         }
1946
1947         if (map.m_pblk == COMPRESS_ADDR) {
1948                 compr_cluster = true;
1949                 start_blk++;
1950                 goto prep_next;
1951         }
1952
1953         logical = blks_to_bytes(inode, start_blk);
1954         phys = blks_to_bytes(inode, map.m_pblk);
1955         size = blks_to_bytes(inode, map.m_len);
1956         flags = 0;
1957         if (map.m_flags & F2FS_MAP_UNWRITTEN)
1958                 flags = FIEMAP_EXTENT_UNWRITTEN;
1959
1960         start_blk += bytes_to_blks(inode, size);
1961
1962 prep_next:
1963         cond_resched();
1964         if (fatal_signal_pending(current))
1965                 ret = -EINTR;
1966         else
1967                 goto next;
1968 out:
1969         if (ret == 1)
1970                 ret = 0;
1971
1972         inode_unlock(inode);
1973         return ret;
1974 }
1975
1976 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1977 {
1978         if (IS_ENABLED(CONFIG_FS_VERITY) &&
1979             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1980                 return inode->i_sb->s_maxbytes;
1981
1982         return i_size_read(inode);
1983 }
1984
1985 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1986                                         unsigned nr_pages,
1987                                         struct f2fs_map_blocks *map,
1988                                         struct bio **bio_ret,
1989                                         sector_t *last_block_in_bio,
1990                                         bool is_readahead)
1991 {
1992         struct bio *bio = *bio_ret;
1993         const unsigned blocksize = blks_to_bytes(inode, 1);
1994         sector_t block_in_file;
1995         sector_t last_block;
1996         sector_t last_block_in_file;
1997         sector_t block_nr;
1998         int ret = 0;
1999
2000         block_in_file = (sector_t)page_index(page);
2001         last_block = block_in_file + nr_pages;
2002         last_block_in_file = bytes_to_blks(inode,
2003                         f2fs_readpage_limit(inode) + blocksize - 1);
2004         if (last_block > last_block_in_file)
2005                 last_block = last_block_in_file;
2006
2007         /* just zeroing out page which is beyond EOF */
2008         if (block_in_file >= last_block)
2009                 goto zero_out;
2010         /*
2011          * Map blocks using the previous result first.
2012          */
2013         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2014                         block_in_file > map->m_lblk &&
2015                         block_in_file < (map->m_lblk + map->m_len))
2016                 goto got_it;
2017
2018         /*
2019          * Then do more f2fs_map_blocks() calls until we are
2020          * done with this page.
2021          */
2022         map->m_lblk = block_in_file;
2023         map->m_len = last_block - block_in_file;
2024
2025         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2026         if (ret)
2027                 goto out;
2028 got_it:
2029         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2030                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2031                 SetPageMappedToDisk(page);
2032
2033                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
2034                                         !cleancache_get_page(page))) {
2035                         SetPageUptodate(page);
2036                         goto confused;
2037                 }
2038
2039                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2040                                                 DATA_GENERIC_ENHANCE_READ)) {
2041                         ret = -EFSCORRUPTED;
2042                         goto out;
2043                 }
2044         } else {
2045 zero_out:
2046                 zero_user_segment(page, 0, PAGE_SIZE);
2047                 if (f2fs_need_verity(inode, page->index) &&
2048                     !fsverity_verify_page(page)) {
2049                         ret = -EIO;
2050                         goto out;
2051                 }
2052                 if (!PageUptodate(page))
2053                         SetPageUptodate(page);
2054                 unlock_page(page);
2055                 goto out;
2056         }
2057
2058         /*
2059          * This page will go to BIO.  Do we need to send this
2060          * BIO off first?
2061          */
2062         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2063                                        *last_block_in_bio, block_nr) ||
2064                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2065 submit_and_realloc:
2066                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2067                 bio = NULL;
2068         }
2069         if (bio == NULL) {
2070                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2071                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2072                                 false);
2073                 if (IS_ERR(bio)) {
2074                         ret = PTR_ERR(bio);
2075                         bio = NULL;
2076                         goto out;
2077                 }
2078         }
2079
2080         /*
2081          * If the page is under writeback, we need to wait for
2082          * its completion to see the correct decrypted data.
2083          */
2084         f2fs_wait_on_block_writeback(inode, block_nr);
2085
2086         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2087                 goto submit_and_realloc;
2088
2089         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2090         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2091         ClearPageError(page);
2092         *last_block_in_bio = block_nr;
2093         goto out;
2094 confused:
2095         if (bio) {
2096                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2097                 bio = NULL;
2098         }
2099         unlock_page(page);
2100 out:
2101         *bio_ret = bio;
2102         return ret;
2103 }
2104
2105 #ifdef CONFIG_F2FS_FS_COMPRESSION
2106 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2107                                 unsigned nr_pages, sector_t *last_block_in_bio,
2108                                 bool is_readahead, bool for_write)
2109 {
2110         struct dnode_of_data dn;
2111         struct inode *inode = cc->inode;
2112         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2113         struct bio *bio = *bio_ret;
2114         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2115         sector_t last_block_in_file;
2116         const unsigned blocksize = blks_to_bytes(inode, 1);
2117         struct decompress_io_ctx *dic = NULL;
2118         int i;
2119         int ret = 0;
2120
2121         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2122
2123         last_block_in_file = bytes_to_blks(inode,
2124                         f2fs_readpage_limit(inode) + blocksize - 1);
2125
2126         /* get rid of pages beyond EOF */
2127         for (i = 0; i < cc->cluster_size; i++) {
2128                 struct page *page = cc->rpages[i];
2129
2130                 if (!page)
2131                         continue;
2132                 if ((sector_t)page->index >= last_block_in_file) {
2133                         zero_user_segment(page, 0, PAGE_SIZE);
2134                         if (!PageUptodate(page))
2135                                 SetPageUptodate(page);
2136                 } else if (!PageUptodate(page)) {
2137                         continue;
2138                 }
2139                 unlock_page(page);
2140                 cc->rpages[i] = NULL;
2141                 cc->nr_rpages--;
2142         }
2143
2144         /* we are done since all pages are beyond EOF */
2145         if (f2fs_cluster_is_empty(cc))
2146                 goto out;
2147
2148         set_new_dnode(&dn, inode, NULL, NULL, 0);
2149         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2150         if (ret)
2151                 goto out;
2152
2153         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2154
2155         for (i = 1; i < cc->cluster_size; i++) {
2156                 block_t blkaddr;
2157
2158                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2159                                                 dn.ofs_in_node + i);
2160
2161                 if (!__is_valid_data_blkaddr(blkaddr))
2162                         break;
2163
2164                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2165                         ret = -EFAULT;
2166                         goto out_put_dnode;
2167                 }
2168                 cc->nr_cpages++;
2169         }
2170
2171         /* nothing to decompress */
2172         if (cc->nr_cpages == 0) {
2173                 ret = 0;
2174                 goto out_put_dnode;
2175         }
2176
2177         dic = f2fs_alloc_dic(cc);
2178         if (IS_ERR(dic)) {
2179                 ret = PTR_ERR(dic);
2180                 goto out_put_dnode;
2181         }
2182
2183         for (i = 0; i < cc->nr_cpages; i++) {
2184                 struct page *page = dic->cpages[i];
2185                 block_t blkaddr;
2186                 struct bio_post_read_ctx *ctx;
2187
2188                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2189                                                 dn.ofs_in_node + i + 1);
2190
2191                 f2fs_wait_on_block_writeback(inode, blkaddr);
2192
2193                 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2194                         if (atomic_dec_and_test(&dic->remaining_pages))
2195                                 f2fs_decompress_cluster(dic);
2196                         continue;
2197                 }
2198
2199                 if (bio && (!page_is_mergeable(sbi, bio,
2200                                         *last_block_in_bio, blkaddr) ||
2201                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2202 submit_and_realloc:
2203                         __submit_bio(sbi, bio, DATA);
2204                         bio = NULL;
2205                 }
2206
2207                 if (!bio) {
2208                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2209                                         is_readahead ? REQ_RAHEAD : 0,
2210                                         page->index, for_write);
2211                         if (IS_ERR(bio)) {
2212                                 ret = PTR_ERR(bio);
2213                                 f2fs_decompress_end_io(dic, ret);
2214                                 f2fs_put_dnode(&dn);
2215                                 *bio_ret = NULL;
2216                                 return ret;
2217                         }
2218                 }
2219
2220                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2221                         goto submit_and_realloc;
2222
2223                 ctx = bio->bi_private;
2224                 ctx->enabled_steps |= STEP_DECOMPRESS;
2225                 refcount_inc(&dic->refcnt);
2226
2227                 inc_page_count(sbi, F2FS_RD_DATA);
2228                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2229                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2230                 ClearPageError(page);
2231                 *last_block_in_bio = blkaddr;
2232         }
2233
2234         f2fs_put_dnode(&dn);
2235
2236         *bio_ret = bio;
2237         return 0;
2238
2239 out_put_dnode:
2240         f2fs_put_dnode(&dn);
2241 out:
2242         for (i = 0; i < cc->cluster_size; i++) {
2243                 if (cc->rpages[i]) {
2244                         ClearPageUptodate(cc->rpages[i]);
2245                         ClearPageError(cc->rpages[i]);
2246                         unlock_page(cc->rpages[i]);
2247                 }
2248         }
2249         *bio_ret = bio;
2250         return ret;
2251 }
2252 #endif
2253
2254 /*
2255  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2256  * Major change was from block_size == page_size in f2fs by default.
2257  */
2258 static int f2fs_mpage_readpages(struct inode *inode,
2259                 struct readahead_control *rac, struct page *page)
2260 {
2261         struct bio *bio = NULL;
2262         sector_t last_block_in_bio = 0;
2263         struct f2fs_map_blocks map;
2264 #ifdef CONFIG_F2FS_FS_COMPRESSION
2265         struct compress_ctx cc = {
2266                 .inode = inode,
2267                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2268                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2269                 .cluster_idx = NULL_CLUSTER,
2270                 .rpages = NULL,
2271                 .cpages = NULL,
2272                 .nr_rpages = 0,
2273                 .nr_cpages = 0,
2274         };
2275 #endif
2276         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2277         unsigned max_nr_pages = nr_pages;
2278         int ret = 0;
2279
2280         map.m_pblk = 0;
2281         map.m_lblk = 0;
2282         map.m_len = 0;
2283         map.m_flags = 0;
2284         map.m_next_pgofs = NULL;
2285         map.m_next_extent = NULL;
2286         map.m_seg_type = NO_CHECK_TYPE;
2287         map.m_may_create = false;
2288
2289         for (; nr_pages; nr_pages--) {
2290                 if (rac) {
2291                         page = readahead_page(rac);
2292                         prefetchw(&page->flags);
2293                 }
2294
2295 #ifdef CONFIG_F2FS_FS_COMPRESSION
2296                 if (f2fs_compressed_file(inode)) {
2297                         /* there are remained comressed pages, submit them */
2298                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2299                                 ret = f2fs_read_multi_pages(&cc, &bio,
2300                                                         max_nr_pages,
2301                                                         &last_block_in_bio,
2302                                                         rac != NULL, false);
2303                                 f2fs_destroy_compress_ctx(&cc, false);
2304                                 if (ret)
2305                                         goto set_error_page;
2306                         }
2307                         ret = f2fs_is_compressed_cluster(inode, page->index);
2308                         if (ret < 0)
2309                                 goto set_error_page;
2310                         else if (!ret)
2311                                 goto read_single_page;
2312
2313                         ret = f2fs_init_compress_ctx(&cc);
2314                         if (ret)
2315                                 goto set_error_page;
2316
2317                         f2fs_compress_ctx_add_page(&cc, page);
2318
2319                         goto next_page;
2320                 }
2321 read_single_page:
2322 #endif
2323
2324                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2325                                         &bio, &last_block_in_bio, rac);
2326                 if (ret) {
2327 #ifdef CONFIG_F2FS_FS_COMPRESSION
2328 set_error_page:
2329 #endif
2330                         SetPageError(page);
2331                         zero_user_segment(page, 0, PAGE_SIZE);
2332                         unlock_page(page);
2333                 }
2334 #ifdef CONFIG_F2FS_FS_COMPRESSION
2335 next_page:
2336 #endif
2337                 if (rac)
2338                         put_page(page);
2339
2340 #ifdef CONFIG_F2FS_FS_COMPRESSION
2341                 if (f2fs_compressed_file(inode)) {
2342                         /* last page */
2343                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2344                                 ret = f2fs_read_multi_pages(&cc, &bio,
2345                                                         max_nr_pages,
2346                                                         &last_block_in_bio,
2347                                                         rac != NULL, false);
2348                                 f2fs_destroy_compress_ctx(&cc, false);
2349                         }
2350                 }
2351 #endif
2352         }
2353         if (bio)
2354                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2355         return ret;
2356 }
2357
2358 static int f2fs_read_data_page(struct file *file, struct page *page)
2359 {
2360         struct inode *inode = page_file_mapping(page)->host;
2361         int ret = -EAGAIN;
2362
2363         trace_f2fs_readpage(page, DATA);
2364
2365         if (!f2fs_is_compress_backend_ready(inode)) {
2366                 unlock_page(page);
2367                 return -EOPNOTSUPP;
2368         }
2369
2370         /* If the file has inline data, try to read it directly */
2371         if (f2fs_has_inline_data(inode))
2372                 ret = f2fs_read_inline_data(inode, page);
2373         if (ret == -EAGAIN)
2374                 ret = f2fs_mpage_readpages(inode, NULL, page);
2375         return ret;
2376 }
2377
2378 static void f2fs_readahead(struct readahead_control *rac)
2379 {
2380         struct inode *inode = rac->mapping->host;
2381
2382         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2383
2384         if (!f2fs_is_compress_backend_ready(inode))
2385                 return;
2386
2387         /* If the file has inline data, skip readpages */
2388         if (f2fs_has_inline_data(inode))
2389                 return;
2390
2391         f2fs_mpage_readpages(inode, rac, NULL);
2392 }
2393
2394 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2395 {
2396         struct inode *inode = fio->page->mapping->host;
2397         struct page *mpage, *page;
2398         gfp_t gfp_flags = GFP_NOFS;
2399
2400         if (!f2fs_encrypted_file(inode))
2401                 return 0;
2402
2403         page = fio->compressed_page ? fio->compressed_page : fio->page;
2404
2405         /* wait for GCed page writeback via META_MAPPING */
2406         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2407
2408         if (fscrypt_inode_uses_inline_crypto(inode))
2409                 return 0;
2410
2411 retry_encrypt:
2412         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2413                                         PAGE_SIZE, 0, gfp_flags);
2414         if (IS_ERR(fio->encrypted_page)) {
2415                 /* flush pending IOs and wait for a while in the ENOMEM case */
2416                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2417                         f2fs_flush_merged_writes(fio->sbi);
2418                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2419                         gfp_flags |= __GFP_NOFAIL;
2420                         goto retry_encrypt;
2421                 }
2422                 return PTR_ERR(fio->encrypted_page);
2423         }
2424
2425         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2426         if (mpage) {
2427                 if (PageUptodate(mpage))
2428                         memcpy(page_address(mpage),
2429                                 page_address(fio->encrypted_page), PAGE_SIZE);
2430                 f2fs_put_page(mpage, 1);
2431         }
2432         return 0;
2433 }
2434
2435 static inline bool check_inplace_update_policy(struct inode *inode,
2436                                 struct f2fs_io_info *fio)
2437 {
2438         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2439         unsigned int policy = SM_I(sbi)->ipu_policy;
2440
2441         if (policy & (0x1 << F2FS_IPU_FORCE))
2442                 return true;
2443         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2444                 return true;
2445         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2446                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2447                 return true;
2448         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2449                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2450                 return true;
2451
2452         /*
2453          * IPU for rewrite async pages
2454          */
2455         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2456                         fio && fio->op == REQ_OP_WRITE &&
2457                         !(fio->op_flags & REQ_SYNC) &&
2458                         !IS_ENCRYPTED(inode))
2459                 return true;
2460
2461         /* this is only set during fdatasync */
2462         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2463                         is_inode_flag_set(inode, FI_NEED_IPU))
2464                 return true;
2465
2466         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2467                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2468                 return true;
2469
2470         return false;
2471 }
2472
2473 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2474 {
2475         /* swap file is migrating in aligned write mode */
2476         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2477                 return false;
2478
2479         if (f2fs_is_pinned_file(inode))
2480                 return true;
2481
2482         /* if this is cold file, we should overwrite to avoid fragmentation */
2483         if (file_is_cold(inode))
2484                 return true;
2485
2486         return check_inplace_update_policy(inode, fio);
2487 }
2488
2489 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2490 {
2491         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2492
2493         if (f2fs_lfs_mode(sbi))
2494                 return true;
2495         if (S_ISDIR(inode->i_mode))
2496                 return true;
2497         if (IS_NOQUOTA(inode))
2498                 return true;
2499         if (f2fs_is_atomic_file(inode))
2500                 return true;
2501
2502         /* swap file is migrating in aligned write mode */
2503         if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2504                 return true;
2505
2506         if (fio) {
2507                 if (page_private_gcing(fio->page))
2508                         return true;
2509                 if (page_private_dummy(fio->page))
2510                         return true;
2511                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2512                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2513                         return true;
2514         }
2515         return false;
2516 }
2517
2518 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2519 {
2520         struct inode *inode = fio->page->mapping->host;
2521
2522         if (f2fs_should_update_outplace(inode, fio))
2523                 return false;
2524
2525         return f2fs_should_update_inplace(inode, fio);
2526 }
2527
2528 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2529 {
2530         struct page *page = fio->page;
2531         struct inode *inode = page->mapping->host;
2532         struct dnode_of_data dn;
2533         struct extent_info ei = {0,0,0};
2534         struct node_info ni;
2535         bool ipu_force = false;
2536         int err = 0;
2537
2538         set_new_dnode(&dn, inode, NULL, NULL, 0);
2539         if (need_inplace_update(fio) &&
2540                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2541                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2542
2543                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2544                                                 DATA_GENERIC_ENHANCE))
2545                         return -EFSCORRUPTED;
2546
2547                 ipu_force = true;
2548                 fio->need_lock = LOCK_DONE;
2549                 goto got_it;
2550         }
2551
2552         /* Deadlock due to between page->lock and f2fs_lock_op */
2553         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2554                 return -EAGAIN;
2555
2556         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2557         if (err)
2558                 goto out;
2559
2560         fio->old_blkaddr = dn.data_blkaddr;
2561
2562         /* This page is already truncated */
2563         if (fio->old_blkaddr == NULL_ADDR) {
2564                 ClearPageUptodate(page);
2565                 clear_page_private_gcing(page);
2566                 goto out_writepage;
2567         }
2568 got_it:
2569         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2570                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2571                                                 DATA_GENERIC_ENHANCE)) {
2572                 err = -EFSCORRUPTED;
2573                 goto out_writepage;
2574         }
2575         /*
2576          * If current allocation needs SSR,
2577          * it had better in-place writes for updated data.
2578          */
2579         if (ipu_force ||
2580                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2581                                         need_inplace_update(fio))) {
2582                 err = f2fs_encrypt_one_page(fio);
2583                 if (err)
2584                         goto out_writepage;
2585
2586                 set_page_writeback(page);
2587                 ClearPageError(page);
2588                 f2fs_put_dnode(&dn);
2589                 if (fio->need_lock == LOCK_REQ)
2590                         f2fs_unlock_op(fio->sbi);
2591                 err = f2fs_inplace_write_data(fio);
2592                 if (err) {
2593                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2594                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2595                         if (PageWriteback(page))
2596                                 end_page_writeback(page);
2597                 } else {
2598                         set_inode_flag(inode, FI_UPDATE_WRITE);
2599                 }
2600                 trace_f2fs_do_write_data_page(fio->page, IPU);
2601                 return err;
2602         }
2603
2604         if (fio->need_lock == LOCK_RETRY) {
2605                 if (!f2fs_trylock_op(fio->sbi)) {
2606                         err = -EAGAIN;
2607                         goto out_writepage;
2608                 }
2609                 fio->need_lock = LOCK_REQ;
2610         }
2611
2612         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2613         if (err)
2614                 goto out_writepage;
2615
2616         fio->version = ni.version;
2617
2618         err = f2fs_encrypt_one_page(fio);
2619         if (err)
2620                 goto out_writepage;
2621
2622         set_page_writeback(page);
2623         ClearPageError(page);
2624
2625         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2626                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2627
2628         /* LFS mode write path */
2629         f2fs_outplace_write_data(&dn, fio);
2630         trace_f2fs_do_write_data_page(page, OPU);
2631         set_inode_flag(inode, FI_APPEND_WRITE);
2632         if (page->index == 0)
2633                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2634 out_writepage:
2635         f2fs_put_dnode(&dn);
2636 out:
2637         if (fio->need_lock == LOCK_REQ)
2638                 f2fs_unlock_op(fio->sbi);
2639         return err;
2640 }
2641
2642 int f2fs_write_single_data_page(struct page *page, int *submitted,
2643                                 struct bio **bio,
2644                                 sector_t *last_block,
2645                                 struct writeback_control *wbc,
2646                                 enum iostat_type io_type,
2647                                 int compr_blocks,
2648                                 bool allow_balance)
2649 {
2650         struct inode *inode = page->mapping->host;
2651         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2652         loff_t i_size = i_size_read(inode);
2653         const pgoff_t end_index = ((unsigned long long)i_size)
2654                                                         >> PAGE_SHIFT;
2655         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2656         unsigned offset = 0;
2657         bool need_balance_fs = false;
2658         int err = 0;
2659         struct f2fs_io_info fio = {
2660                 .sbi = sbi,
2661                 .ino = inode->i_ino,
2662                 .type = DATA,
2663                 .op = REQ_OP_WRITE,
2664                 .op_flags = wbc_to_write_flags(wbc),
2665                 .old_blkaddr = NULL_ADDR,
2666                 .page = page,
2667                 .encrypted_page = NULL,
2668                 .submitted = false,
2669                 .compr_blocks = compr_blocks,
2670                 .need_lock = LOCK_RETRY,
2671                 .io_type = io_type,
2672                 .io_wbc = wbc,
2673                 .bio = bio,
2674                 .last_block = last_block,
2675         };
2676
2677         trace_f2fs_writepage(page, DATA);
2678
2679         /* we should bypass data pages to proceed the kworkder jobs */
2680         if (unlikely(f2fs_cp_error(sbi))) {
2681                 mapping_set_error(page->mapping, -EIO);
2682                 /*
2683                  * don't drop any dirty dentry pages for keeping lastest
2684                  * directory structure.
2685                  */
2686                 if (S_ISDIR(inode->i_mode))
2687                         goto redirty_out;
2688                 goto out;
2689         }
2690
2691         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2692                 goto redirty_out;
2693
2694         if (page->index < end_index ||
2695                         f2fs_verity_in_progress(inode) ||
2696                         compr_blocks)
2697                 goto write;
2698
2699         /*
2700          * If the offset is out-of-range of file size,
2701          * this page does not have to be written to disk.
2702          */
2703         offset = i_size & (PAGE_SIZE - 1);
2704         if ((page->index >= end_index + 1) || !offset)
2705                 goto out;
2706
2707         zero_user_segment(page, offset, PAGE_SIZE);
2708 write:
2709         if (f2fs_is_drop_cache(inode))
2710                 goto out;
2711         /* we should not write 0'th page having journal header */
2712         if (f2fs_is_volatile_file(inode) && (!page->index ||
2713                         (!wbc->for_reclaim &&
2714                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2715                 goto redirty_out;
2716
2717         /* Dentry/quota blocks are controlled by checkpoint */
2718         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2719                 /*
2720                  * We need to wait for node_write to avoid block allocation during
2721                  * checkpoint. This can only happen to quota writes which can cause
2722                  * the below discard race condition.
2723                  */
2724                 if (IS_NOQUOTA(inode))
2725                         down_read(&sbi->node_write);
2726
2727                 fio.need_lock = LOCK_DONE;
2728                 err = f2fs_do_write_data_page(&fio);
2729
2730                 if (IS_NOQUOTA(inode))
2731                         up_read(&sbi->node_write);
2732
2733                 goto done;
2734         }
2735
2736         if (!wbc->for_reclaim)
2737                 need_balance_fs = true;
2738         else if (has_not_enough_free_secs(sbi, 0, 0))
2739                 goto redirty_out;
2740         else
2741                 set_inode_flag(inode, FI_HOT_DATA);
2742
2743         err = -EAGAIN;
2744         if (f2fs_has_inline_data(inode)) {
2745                 err = f2fs_write_inline_data(inode, page);
2746                 if (!err)
2747                         goto out;
2748         }
2749
2750         if (err == -EAGAIN) {
2751                 err = f2fs_do_write_data_page(&fio);
2752                 if (err == -EAGAIN) {
2753                         fio.need_lock = LOCK_REQ;
2754                         err = f2fs_do_write_data_page(&fio);
2755                 }
2756         }
2757
2758         if (err) {
2759                 file_set_keep_isize(inode);
2760         } else {
2761                 spin_lock(&F2FS_I(inode)->i_size_lock);
2762                 if (F2FS_I(inode)->last_disk_size < psize)
2763                         F2FS_I(inode)->last_disk_size = psize;
2764                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2765         }
2766
2767 done:
2768         if (err && err != -ENOENT)
2769                 goto redirty_out;
2770
2771 out:
2772         inode_dec_dirty_pages(inode);
2773         if (err) {
2774                 ClearPageUptodate(page);
2775                 clear_page_private_gcing(page);
2776         }
2777
2778         if (wbc->for_reclaim) {
2779                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2780                 clear_inode_flag(inode, FI_HOT_DATA);
2781                 f2fs_remove_dirty_inode(inode);
2782                 submitted = NULL;
2783         }
2784         unlock_page(page);
2785         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2786                         !F2FS_I(inode)->cp_task && allow_balance)
2787                 f2fs_balance_fs(sbi, need_balance_fs);
2788
2789         if (unlikely(f2fs_cp_error(sbi))) {
2790                 f2fs_submit_merged_write(sbi, DATA);
2791                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2792                 submitted = NULL;
2793         }
2794
2795         if (submitted)
2796                 *submitted = fio.submitted ? 1 : 0;
2797
2798         return 0;
2799
2800 redirty_out:
2801         redirty_page_for_writepage(wbc, page);
2802         /*
2803          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2804          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2805          * file_write_and_wait_range() will see EIO error, which is critical
2806          * to return value of fsync() followed by atomic_write failure to user.
2807          */
2808         if (!err || wbc->for_reclaim)
2809                 return AOP_WRITEPAGE_ACTIVATE;
2810         unlock_page(page);
2811         return err;
2812 }
2813
2814 static int f2fs_write_data_page(struct page *page,
2815                                         struct writeback_control *wbc)
2816 {
2817 #ifdef CONFIG_F2FS_FS_COMPRESSION
2818         struct inode *inode = page->mapping->host;
2819
2820         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2821                 goto out;
2822
2823         if (f2fs_compressed_file(inode)) {
2824                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2825                         redirty_page_for_writepage(wbc, page);
2826                         return AOP_WRITEPAGE_ACTIVATE;
2827                 }
2828         }
2829 out:
2830 #endif
2831
2832         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2833                                                 wbc, FS_DATA_IO, 0, true);
2834 }
2835
2836 /*
2837  * This function was copied from write_cche_pages from mm/page-writeback.c.
2838  * The major change is making write step of cold data page separately from
2839  * warm/hot data page.
2840  */
2841 static int f2fs_write_cache_pages(struct address_space *mapping,
2842                                         struct writeback_control *wbc,
2843                                         enum iostat_type io_type)
2844 {
2845         int ret = 0;
2846         int done = 0, retry = 0;
2847         struct pagevec pvec;
2848         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2849         struct bio *bio = NULL;
2850         sector_t last_block;
2851 #ifdef CONFIG_F2FS_FS_COMPRESSION
2852         struct inode *inode = mapping->host;
2853         struct compress_ctx cc = {
2854                 .inode = inode,
2855                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2856                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2857                 .cluster_idx = NULL_CLUSTER,
2858                 .rpages = NULL,
2859                 .nr_rpages = 0,
2860                 .cpages = NULL,
2861                 .rbuf = NULL,
2862                 .cbuf = NULL,
2863                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2864                 .private = NULL,
2865         };
2866 #endif
2867         int nr_pages;
2868         pgoff_t index;
2869         pgoff_t end;            /* Inclusive */
2870         pgoff_t done_index;
2871         int range_whole = 0;
2872         xa_mark_t tag;
2873         int nwritten = 0;
2874         int submitted = 0;
2875         int i;
2876
2877         pagevec_init(&pvec);
2878
2879         if (get_dirty_pages(mapping->host) <=
2880                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2881                 set_inode_flag(mapping->host, FI_HOT_DATA);
2882         else
2883                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2884
2885         if (wbc->range_cyclic) {
2886                 index = mapping->writeback_index; /* prev offset */
2887                 end = -1;
2888         } else {
2889                 index = wbc->range_start >> PAGE_SHIFT;
2890                 end = wbc->range_end >> PAGE_SHIFT;
2891                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2892                         range_whole = 1;
2893         }
2894         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2895                 tag = PAGECACHE_TAG_TOWRITE;
2896         else
2897                 tag = PAGECACHE_TAG_DIRTY;
2898 retry:
2899         retry = 0;
2900         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2901                 tag_pages_for_writeback(mapping, index, end);
2902         done_index = index;
2903         while (!done && !retry && (index <= end)) {
2904                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2905                                 tag);
2906                 if (nr_pages == 0)
2907                         break;
2908
2909                 for (i = 0; i < nr_pages; i++) {
2910                         struct page *page = pvec.pages[i];
2911                         bool need_readd;
2912 readd:
2913                         need_readd = false;
2914 #ifdef CONFIG_F2FS_FS_COMPRESSION
2915                         if (f2fs_compressed_file(inode)) {
2916                                 ret = f2fs_init_compress_ctx(&cc);
2917                                 if (ret) {
2918                                         done = 1;
2919                                         break;
2920                                 }
2921
2922                                 if (!f2fs_cluster_can_merge_page(&cc,
2923                                                                 page->index)) {
2924                                         ret = f2fs_write_multi_pages(&cc,
2925                                                 &submitted, wbc, io_type);
2926                                         if (!ret)
2927                                                 need_readd = true;
2928                                         goto result;
2929                                 }
2930
2931                                 if (unlikely(f2fs_cp_error(sbi)))
2932                                         goto lock_page;
2933
2934                                 if (f2fs_cluster_is_empty(&cc)) {
2935                                         void *fsdata = NULL;
2936                                         struct page *pagep;
2937                                         int ret2;
2938
2939                                         ret2 = f2fs_prepare_compress_overwrite(
2940                                                         inode, &pagep,
2941                                                         page->index, &fsdata);
2942                                         if (ret2 < 0) {
2943                                                 ret = ret2;
2944                                                 done = 1;
2945                                                 break;
2946                                         } else if (ret2 &&
2947                                                 !f2fs_compress_write_end(inode,
2948                                                                 fsdata, page->index,
2949                                                                 1)) {
2950                                                 retry = 1;
2951                                                 break;
2952                                         }
2953                                 } else {
2954                                         goto lock_page;
2955                                 }
2956                         }
2957 #endif
2958                         /* give a priority to WB_SYNC threads */
2959                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2960                                         wbc->sync_mode == WB_SYNC_NONE) {
2961                                 done = 1;
2962                                 break;
2963                         }
2964 #ifdef CONFIG_F2FS_FS_COMPRESSION
2965 lock_page:
2966 #endif
2967                         done_index = page->index;
2968 retry_write:
2969                         lock_page(page);
2970
2971                         if (unlikely(page->mapping != mapping)) {
2972 continue_unlock:
2973                                 unlock_page(page);
2974                                 continue;
2975                         }
2976
2977                         if (!PageDirty(page)) {
2978                                 /* someone wrote it for us */
2979                                 goto continue_unlock;
2980                         }
2981
2982                         if (PageWriteback(page)) {
2983                                 if (wbc->sync_mode != WB_SYNC_NONE)
2984                                         f2fs_wait_on_page_writeback(page,
2985                                                         DATA, true, true);
2986                                 else
2987                                         goto continue_unlock;
2988                         }
2989
2990                         if (!clear_page_dirty_for_io(page))
2991                                 goto continue_unlock;
2992
2993 #ifdef CONFIG_F2FS_FS_COMPRESSION
2994                         if (f2fs_compressed_file(inode)) {
2995                                 get_page(page);
2996                                 f2fs_compress_ctx_add_page(&cc, page);
2997                                 continue;
2998                         }
2999 #endif
3000                         ret = f2fs_write_single_data_page(page, &submitted,
3001                                         &bio, &last_block, wbc, io_type,
3002                                         0, true);
3003                         if (ret == AOP_WRITEPAGE_ACTIVATE)
3004                                 unlock_page(page);
3005 #ifdef CONFIG_F2FS_FS_COMPRESSION
3006 result:
3007 #endif
3008                         nwritten += submitted;
3009                         wbc->nr_to_write -= submitted;
3010
3011                         if (unlikely(ret)) {
3012                                 /*
3013                                  * keep nr_to_write, since vfs uses this to
3014                                  * get # of written pages.
3015                                  */
3016                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3017                                         ret = 0;
3018                                         goto next;
3019                                 } else if (ret == -EAGAIN) {
3020                                         ret = 0;
3021                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3022                                                 cond_resched();
3023                                                 congestion_wait(BLK_RW_ASYNC,
3024                                                         DEFAULT_IO_TIMEOUT);
3025                                                 goto retry_write;
3026                                         }
3027                                         goto next;
3028                                 }
3029                                 done_index = page->index + 1;
3030                                 done = 1;
3031                                 break;
3032                         }
3033
3034                         if (wbc->nr_to_write <= 0 &&
3035                                         wbc->sync_mode == WB_SYNC_NONE) {
3036                                 done = 1;
3037                                 break;
3038                         }
3039 next:
3040                         if (need_readd)
3041                                 goto readd;
3042                 }
3043                 pagevec_release(&pvec);
3044                 cond_resched();
3045         }
3046 #ifdef CONFIG_F2FS_FS_COMPRESSION
3047         /* flush remained pages in compress cluster */
3048         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3049                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3050                 nwritten += submitted;
3051                 wbc->nr_to_write -= submitted;
3052                 if (ret) {
3053                         done = 1;
3054                         retry = 0;
3055                 }
3056         }
3057         if (f2fs_compressed_file(inode))
3058                 f2fs_destroy_compress_ctx(&cc, false);
3059 #endif
3060         if (retry) {
3061                 index = 0;
3062                 end = -1;
3063                 goto retry;
3064         }
3065         if (wbc->range_cyclic && !done)
3066                 done_index = 0;
3067         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3068                 mapping->writeback_index = done_index;
3069
3070         if (nwritten)
3071                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3072                                                                 NULL, 0, DATA);
3073         /* submit cached bio of IPU write */
3074         if (bio)
3075                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3076
3077         return ret;
3078 }
3079
3080 static inline bool __should_serialize_io(struct inode *inode,
3081                                         struct writeback_control *wbc)
3082 {
3083         /* to avoid deadlock in path of data flush */
3084         if (F2FS_I(inode)->cp_task)
3085                 return false;
3086
3087         if (!S_ISREG(inode->i_mode))
3088                 return false;
3089         if (IS_NOQUOTA(inode))
3090                 return false;
3091
3092         if (f2fs_need_compress_data(inode))
3093                 return true;
3094         if (wbc->sync_mode != WB_SYNC_ALL)
3095                 return true;
3096         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3097                 return true;
3098         return false;
3099 }
3100
3101 static int __f2fs_write_data_pages(struct address_space *mapping,
3102                                                 struct writeback_control *wbc,
3103                                                 enum iostat_type io_type)
3104 {
3105         struct inode *inode = mapping->host;
3106         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3107         struct blk_plug plug;
3108         int ret;
3109         bool locked = false;
3110
3111         /* deal with chardevs and other special file */
3112         if (!mapping->a_ops->writepage)
3113                 return 0;
3114
3115         /* skip writing if there is no dirty page in this inode */
3116         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3117                 return 0;
3118
3119         /* during POR, we don't need to trigger writepage at all. */
3120         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3121                 goto skip_write;
3122
3123         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3124                         wbc->sync_mode == WB_SYNC_NONE &&
3125                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3126                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3127                 goto skip_write;
3128
3129         /* skip writing during file defragment */
3130         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3131                 goto skip_write;
3132
3133         trace_f2fs_writepages(mapping->host, wbc, DATA);
3134
3135         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3136         if (wbc->sync_mode == WB_SYNC_ALL)
3137                 atomic_inc(&sbi->wb_sync_req[DATA]);
3138         else if (atomic_read(&sbi->wb_sync_req[DATA]))
3139                 goto skip_write;
3140
3141         if (__should_serialize_io(inode, wbc)) {
3142                 mutex_lock(&sbi->writepages);
3143                 locked = true;
3144         }
3145
3146         blk_start_plug(&plug);
3147         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3148         blk_finish_plug(&plug);
3149
3150         if (locked)
3151                 mutex_unlock(&sbi->writepages);
3152
3153         if (wbc->sync_mode == WB_SYNC_ALL)
3154                 atomic_dec(&sbi->wb_sync_req[DATA]);
3155         /*
3156          * if some pages were truncated, we cannot guarantee its mapping->host
3157          * to detect pending bios.
3158          */
3159
3160         f2fs_remove_dirty_inode(inode);
3161         return ret;
3162
3163 skip_write:
3164         wbc->pages_skipped += get_dirty_pages(inode);
3165         trace_f2fs_writepages(mapping->host, wbc, DATA);
3166         return 0;
3167 }
3168
3169 static int f2fs_write_data_pages(struct address_space *mapping,
3170                             struct writeback_control *wbc)
3171 {
3172         struct inode *inode = mapping->host;
3173
3174         return __f2fs_write_data_pages(mapping, wbc,
3175                         F2FS_I(inode)->cp_task == current ?
3176                         FS_CP_DATA_IO : FS_DATA_IO);
3177 }
3178
3179 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
3180 {
3181         struct inode *inode = mapping->host;
3182         loff_t i_size = i_size_read(inode);
3183
3184         if (IS_NOQUOTA(inode))
3185                 return;
3186
3187         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3188         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3189                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3190                 filemap_invalidate_lock(mapping);
3191
3192                 truncate_pagecache(inode, i_size);
3193                 f2fs_truncate_blocks(inode, i_size, true);
3194
3195                 filemap_invalidate_unlock(mapping);
3196                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3197         }
3198 }
3199
3200 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3201                         struct page *page, loff_t pos, unsigned len,
3202                         block_t *blk_addr, bool *node_changed)
3203 {
3204         struct inode *inode = page->mapping->host;
3205         pgoff_t index = page->index;
3206         struct dnode_of_data dn;
3207         struct page *ipage;
3208         bool locked = false;
3209         struct extent_info ei = {0,0,0};
3210         int err = 0;
3211         int flag;
3212
3213         /*
3214          * we already allocated all the blocks, so we don't need to get
3215          * the block addresses when there is no need to fill the page.
3216          */
3217         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3218             !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3219             !f2fs_verity_in_progress(inode))
3220                 return 0;
3221
3222         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3223         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3224                 flag = F2FS_GET_BLOCK_DEFAULT;
3225         else
3226                 flag = F2FS_GET_BLOCK_PRE_AIO;
3227
3228         if (f2fs_has_inline_data(inode) ||
3229                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3230                 f2fs_do_map_lock(sbi, flag, true);
3231                 locked = true;
3232         }
3233
3234 restart:
3235         /* check inline_data */
3236         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3237         if (IS_ERR(ipage)) {
3238                 err = PTR_ERR(ipage);
3239                 goto unlock_out;
3240         }
3241
3242         set_new_dnode(&dn, inode, ipage, ipage, 0);
3243
3244         if (f2fs_has_inline_data(inode)) {
3245                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3246                         f2fs_do_read_inline_data(page, ipage);
3247                         set_inode_flag(inode, FI_DATA_EXIST);
3248                         if (inode->i_nlink)
3249                                 set_page_private_inline(ipage);
3250                 } else {
3251                         err = f2fs_convert_inline_page(&dn, page);
3252                         if (err)
3253                                 goto out;
3254                         if (dn.data_blkaddr == NULL_ADDR)
3255                                 err = f2fs_get_block(&dn, index);
3256                 }
3257         } else if (locked) {
3258                 err = f2fs_get_block(&dn, index);
3259         } else {
3260                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3261                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3262                 } else {
3263                         /* hole case */
3264                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3265                         if (err || dn.data_blkaddr == NULL_ADDR) {
3266                                 f2fs_put_dnode(&dn);
3267                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3268                                                                 true);
3269                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3270                                 locked = true;
3271                                 goto restart;
3272                         }
3273                 }
3274         }
3275
3276         /* convert_inline_page can make node_changed */
3277         *blk_addr = dn.data_blkaddr;
3278         *node_changed = dn.node_changed;
3279 out:
3280         f2fs_put_dnode(&dn);
3281 unlock_out:
3282         if (locked)
3283                 f2fs_do_map_lock(sbi, flag, false);
3284         return err;
3285 }
3286
3287 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3288                 loff_t pos, unsigned len, unsigned flags,
3289                 struct page **pagep, void **fsdata)
3290 {
3291         struct inode *inode = mapping->host;
3292         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3293         struct page *page = NULL;
3294         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3295         bool need_balance = false, drop_atomic = false;
3296         block_t blkaddr = NULL_ADDR;
3297         int err = 0;
3298
3299         trace_f2fs_write_begin(inode, pos, len, flags);
3300
3301         if (!f2fs_is_checkpoint_ready(sbi)) {
3302                 err = -ENOSPC;
3303                 goto fail;
3304         }
3305
3306         if ((f2fs_is_atomic_file(inode) &&
3307                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3308                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3309                 err = -ENOMEM;
3310                 drop_atomic = true;
3311                 goto fail;
3312         }
3313
3314         /*
3315          * We should check this at this moment to avoid deadlock on inode page
3316          * and #0 page. The locking rule for inline_data conversion should be:
3317          * lock_page(page #0) -> lock_page(inode_page)
3318          */
3319         if (index != 0) {
3320                 err = f2fs_convert_inline_inode(inode);
3321                 if (err)
3322                         goto fail;
3323         }
3324
3325 #ifdef CONFIG_F2FS_FS_COMPRESSION
3326         if (f2fs_compressed_file(inode)) {
3327                 int ret;
3328
3329                 *fsdata = NULL;
3330
3331                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3332                                                         index, fsdata);
3333                 if (ret < 0) {
3334                         err = ret;
3335                         goto fail;
3336                 } else if (ret) {
3337                         return 0;
3338                 }
3339         }
3340 #endif
3341
3342 repeat:
3343         /*
3344          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3345          * wait_for_stable_page. Will wait that below with our IO control.
3346          */
3347         page = f2fs_pagecache_get_page(mapping, index,
3348                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3349         if (!page) {
3350                 err = -ENOMEM;
3351                 goto fail;
3352         }
3353
3354         /* TODO: cluster can be compressed due to race with .writepage */
3355
3356         *pagep = page;
3357
3358         err = prepare_write_begin(sbi, page, pos, len,
3359                                         &blkaddr, &need_balance);
3360         if (err)
3361                 goto fail;
3362
3363         if (need_balance && !IS_NOQUOTA(inode) &&
3364                         has_not_enough_free_secs(sbi, 0, 0)) {
3365                 unlock_page(page);
3366                 f2fs_balance_fs(sbi, true);
3367                 lock_page(page);
3368                 if (page->mapping != mapping) {
3369                         /* The page got truncated from under us */
3370                         f2fs_put_page(page, 1);
3371                         goto repeat;
3372                 }
3373         }
3374
3375         f2fs_wait_on_page_writeback(page, DATA, false, true);
3376
3377         if (len == PAGE_SIZE || PageUptodate(page))
3378                 return 0;
3379
3380         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3381             !f2fs_verity_in_progress(inode)) {
3382                 zero_user_segment(page, len, PAGE_SIZE);
3383                 return 0;
3384         }
3385
3386         if (blkaddr == NEW_ADDR) {
3387                 zero_user_segment(page, 0, PAGE_SIZE);
3388                 SetPageUptodate(page);
3389         } else {
3390                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3391                                 DATA_GENERIC_ENHANCE_READ)) {
3392                         err = -EFSCORRUPTED;
3393                         goto fail;
3394                 }
3395                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3396                 if (err)
3397                         goto fail;
3398
3399                 lock_page(page);
3400                 if (unlikely(page->mapping != mapping)) {
3401                         f2fs_put_page(page, 1);
3402                         goto repeat;
3403                 }
3404                 if (unlikely(!PageUptodate(page))) {
3405                         err = -EIO;
3406                         goto fail;
3407                 }
3408         }
3409         return 0;
3410
3411 fail:
3412         f2fs_put_page(page, 1);
3413         f2fs_write_failed(mapping, pos + len);
3414         if (drop_atomic)
3415                 f2fs_drop_inmem_pages_all(sbi, false);
3416         return err;
3417 }
3418
3419 static int f2fs_write_end(struct file *file,
3420                         struct address_space *mapping,
3421                         loff_t pos, unsigned len, unsigned copied,
3422                         struct page *page, void *fsdata)
3423 {
3424         struct inode *inode = page->mapping->host;
3425
3426         trace_f2fs_write_end(inode, pos, len, copied);
3427
3428         /*
3429          * This should be come from len == PAGE_SIZE, and we expect copied
3430          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3431          * let generic_perform_write() try to copy data again through copied=0.
3432          */
3433         if (!PageUptodate(page)) {
3434                 if (unlikely(copied != len))
3435                         copied = 0;
3436                 else
3437                         SetPageUptodate(page);
3438         }
3439
3440 #ifdef CONFIG_F2FS_FS_COMPRESSION
3441         /* overwrite compressed file */
3442         if (f2fs_compressed_file(inode) && fsdata) {
3443                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3444                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3445
3446                 if (pos + copied > i_size_read(inode) &&
3447                                 !f2fs_verity_in_progress(inode))
3448                         f2fs_i_size_write(inode, pos + copied);
3449                 return copied;
3450         }
3451 #endif
3452
3453         if (!copied)
3454                 goto unlock_out;
3455
3456         set_page_dirty(page);
3457
3458         if (pos + copied > i_size_read(inode) &&
3459             !f2fs_verity_in_progress(inode))
3460                 f2fs_i_size_write(inode, pos + copied);
3461 unlock_out:
3462         f2fs_put_page(page, 1);
3463         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3464         return copied;
3465 }
3466
3467 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3468                            loff_t offset)
3469 {
3470         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3471         unsigned blkbits = i_blkbits;
3472         unsigned blocksize_mask = (1 << blkbits) - 1;
3473         unsigned long align = offset | iov_iter_alignment(iter);
3474         struct block_device *bdev = inode->i_sb->s_bdev;
3475
3476         if (iov_iter_rw(iter) == READ && offset >= i_size_read(inode))
3477                 return 1;
3478
3479         if (align & blocksize_mask) {
3480                 if (bdev)
3481                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
3482                 blocksize_mask = (1 << blkbits) - 1;
3483                 if (align & blocksize_mask)
3484                         return -EINVAL;
3485                 return 1;
3486         }
3487         return 0;
3488 }
3489
3490 static void f2fs_dio_end_io(struct bio *bio)
3491 {
3492         struct f2fs_private_dio *dio = bio->bi_private;
3493
3494         dec_page_count(F2FS_I_SB(dio->inode),
3495                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3496
3497         bio->bi_private = dio->orig_private;
3498         bio->bi_end_io = dio->orig_end_io;
3499
3500         kfree(dio);
3501
3502         bio_endio(bio);
3503 }
3504
3505 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3506                                                         loff_t file_offset)
3507 {
3508         struct f2fs_private_dio *dio;
3509         bool write = (bio_op(bio) == REQ_OP_WRITE);
3510
3511         dio = f2fs_kzalloc(F2FS_I_SB(inode),
3512                         sizeof(struct f2fs_private_dio), GFP_NOFS);
3513         if (!dio)
3514                 goto out;
3515
3516         dio->inode = inode;
3517         dio->orig_end_io = bio->bi_end_io;
3518         dio->orig_private = bio->bi_private;
3519         dio->write = write;
3520
3521         bio->bi_end_io = f2fs_dio_end_io;
3522         bio->bi_private = dio;
3523
3524         inc_page_count(F2FS_I_SB(inode),
3525                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3526
3527         submit_bio(bio);
3528         return;
3529 out:
3530         bio->bi_status = BLK_STS_IOERR;
3531         bio_endio(bio);
3532 }
3533
3534 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3535 {
3536         struct address_space *mapping = iocb->ki_filp->f_mapping;
3537         struct inode *inode = mapping->host;
3538         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3539         struct f2fs_inode_info *fi = F2FS_I(inode);
3540         size_t count = iov_iter_count(iter);
3541         loff_t offset = iocb->ki_pos;
3542         int rw = iov_iter_rw(iter);
3543         int err;
3544         enum rw_hint hint = iocb->ki_hint;
3545         int whint_mode = F2FS_OPTION(sbi).whint_mode;
3546         bool do_opu;
3547
3548         err = check_direct_IO(inode, iter, offset);
3549         if (err)
3550                 return err < 0 ? err : 0;
3551
3552         if (f2fs_force_buffered_io(inode, iocb, iter))
3553                 return 0;
3554
3555         do_opu = allow_outplace_dio(inode, iocb, iter);
3556
3557         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3558
3559         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3560                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
3561
3562         if (iocb->ki_flags & IOCB_NOWAIT) {
3563                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3564                         iocb->ki_hint = hint;
3565                         err = -EAGAIN;
3566                         goto out;
3567                 }
3568                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3569                         up_read(&fi->i_gc_rwsem[rw]);
3570                         iocb->ki_hint = hint;
3571                         err = -EAGAIN;
3572                         goto out;
3573                 }
3574         } else {
3575                 down_read(&fi->i_gc_rwsem[rw]);
3576                 if (do_opu)
3577                         down_read(&fi->i_gc_rwsem[READ]);
3578         }
3579
3580         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3581                         iter, rw == WRITE ? get_data_block_dio_write :
3582                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
3583                         rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3584                         DIO_SKIP_HOLES);
3585
3586         if (do_opu)
3587                 up_read(&fi->i_gc_rwsem[READ]);
3588
3589         up_read(&fi->i_gc_rwsem[rw]);
3590
3591         if (rw == WRITE) {
3592                 if (whint_mode == WHINT_MODE_OFF)
3593                         iocb->ki_hint = hint;
3594                 if (err > 0) {
3595                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3596                                                                         err);
3597                         if (!do_opu)
3598                                 set_inode_flag(inode, FI_UPDATE_WRITE);
3599                 } else if (err == -EIOCBQUEUED) {
3600                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3601                                                 count - iov_iter_count(iter));
3602                 } else if (err < 0) {
3603                         f2fs_write_failed(mapping, offset + count);
3604                 }
3605         } else {
3606                 if (err > 0)
3607                         f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3608                 else if (err == -EIOCBQUEUED)
3609                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_READ_IO,
3610                                                 count - iov_iter_count(iter));
3611         }
3612
3613 out:
3614         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3615
3616         return err;
3617 }
3618
3619 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3620                                                         unsigned int length)
3621 {
3622         struct inode *inode = page->mapping->host;
3623         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3624
3625         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3626                 (offset % PAGE_SIZE || length != PAGE_SIZE))
3627                 return;
3628
3629         if (PageDirty(page)) {
3630                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3631                         dec_page_count(sbi, F2FS_DIRTY_META);
3632                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3633                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3634                 } else {
3635                         inode_dec_dirty_pages(inode);
3636                         f2fs_remove_dirty_inode(inode);
3637                 }
3638         }
3639
3640         clear_page_private_gcing(page);
3641
3642         if (test_opt(sbi, COMPRESS_CACHE)) {
3643                 if (f2fs_compressed_file(inode))
3644                         f2fs_invalidate_compress_pages(sbi, inode->i_ino);
3645                 if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3646                         clear_page_private_data(page);
3647         }
3648
3649         if (page_private_atomic(page))
3650                 return f2fs_drop_inmem_page(inode, page);
3651
3652         detach_page_private(page);
3653         set_page_private(page, 0);
3654 }
3655
3656 int f2fs_release_page(struct page *page, gfp_t wait)
3657 {
3658         /* If this is dirty page, keep PagePrivate */
3659         if (PageDirty(page))
3660                 return 0;
3661
3662         /* This is atomic written page, keep Private */
3663         if (page_private_atomic(page))
3664                 return 0;
3665
3666         if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3667                 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3668                 struct inode *inode = page->mapping->host;
3669
3670                 if (f2fs_compressed_file(inode))
3671                         f2fs_invalidate_compress_pages(sbi, inode->i_ino);
3672                 if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3673                         clear_page_private_data(page);
3674         }
3675
3676         clear_page_private_gcing(page);
3677
3678         detach_page_private(page);
3679         set_page_private(page, 0);
3680         return 1;
3681 }
3682
3683 static int f2fs_set_data_page_dirty(struct page *page)
3684 {
3685         struct inode *inode = page_file_mapping(page)->host;
3686
3687         trace_f2fs_set_page_dirty(page, DATA);
3688
3689         if (!PageUptodate(page))
3690                 SetPageUptodate(page);
3691         if (PageSwapCache(page))
3692                 return __set_page_dirty_nobuffers(page);
3693
3694         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3695                 if (!page_private_atomic(page)) {
3696                         f2fs_register_inmem_page(inode, page);
3697                         return 1;
3698                 }
3699                 /*
3700                  * Previously, this page has been registered, we just
3701                  * return here.
3702                  */
3703                 return 0;
3704         }
3705
3706         if (!PageDirty(page)) {
3707                 __set_page_dirty_nobuffers(page);
3708                 f2fs_update_dirty_page(inode, page);
3709                 return 1;
3710         }
3711         return 0;
3712 }
3713
3714
3715 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3716 {
3717 #ifdef CONFIG_F2FS_FS_COMPRESSION
3718         struct dnode_of_data dn;
3719         sector_t start_idx, blknr = 0;
3720         int ret;
3721
3722         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3723
3724         set_new_dnode(&dn, inode, NULL, NULL, 0);
3725         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3726         if (ret)
3727                 return 0;
3728
3729         if (dn.data_blkaddr != COMPRESS_ADDR) {
3730                 dn.ofs_in_node += block - start_idx;
3731                 blknr = f2fs_data_blkaddr(&dn);
3732                 if (!__is_valid_data_blkaddr(blknr))
3733                         blknr = 0;
3734         }
3735
3736         f2fs_put_dnode(&dn);
3737         return blknr;
3738 #else
3739         return 0;
3740 #endif
3741 }
3742
3743
3744 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3745 {
3746         struct inode *inode = mapping->host;
3747         sector_t blknr = 0;
3748
3749         if (f2fs_has_inline_data(inode))
3750                 goto out;
3751
3752         /* make sure allocating whole blocks */
3753         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3754                 filemap_write_and_wait(mapping);
3755
3756         /* Block number less than F2FS MAX BLOCKS */
3757         if (unlikely(block >= max_file_blocks(inode)))
3758                 goto out;
3759
3760         if (f2fs_compressed_file(inode)) {
3761                 blknr = f2fs_bmap_compress(inode, block);
3762         } else {
3763                 struct f2fs_map_blocks map;
3764
3765                 memset(&map, 0, sizeof(map));
3766                 map.m_lblk = block;
3767                 map.m_len = 1;
3768                 map.m_next_pgofs = NULL;
3769                 map.m_seg_type = NO_CHECK_TYPE;
3770
3771                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3772                         blknr = map.m_pblk;
3773         }
3774 out:
3775         trace_f2fs_bmap(inode, block, blknr);
3776         return blknr;
3777 }
3778
3779 #ifdef CONFIG_MIGRATION
3780 #include <linux/migrate.h>
3781
3782 int f2fs_migrate_page(struct address_space *mapping,
3783                 struct page *newpage, struct page *page, enum migrate_mode mode)
3784 {
3785         int rc, extra_count;
3786         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3787         bool atomic_written = page_private_atomic(page);
3788
3789         BUG_ON(PageWriteback(page));
3790
3791         /* migrating an atomic written page is safe with the inmem_lock hold */
3792         if (atomic_written) {
3793                 if (mode != MIGRATE_SYNC)
3794                         return -EBUSY;
3795                 if (!mutex_trylock(&fi->inmem_lock))
3796                         return -EAGAIN;
3797         }
3798
3799         /* one extra reference was held for atomic_write page */
3800         extra_count = atomic_written ? 1 : 0;
3801         rc = migrate_page_move_mapping(mapping, newpage,
3802                                 page, extra_count);
3803         if (rc != MIGRATEPAGE_SUCCESS) {
3804                 if (atomic_written)
3805                         mutex_unlock(&fi->inmem_lock);
3806                 return rc;
3807         }
3808
3809         if (atomic_written) {
3810                 struct inmem_pages *cur;
3811
3812                 list_for_each_entry(cur, &fi->inmem_pages, list)
3813                         if (cur->page == page) {
3814                                 cur->page = newpage;
3815                                 break;
3816                         }
3817                 mutex_unlock(&fi->inmem_lock);
3818                 put_page(page);
3819                 get_page(newpage);
3820         }
3821
3822         /* guarantee to start from no stale private field */
3823         set_page_private(newpage, 0);
3824         if (PagePrivate(page)) {
3825                 set_page_private(newpage, page_private(page));
3826                 SetPagePrivate(newpage);
3827                 get_page(newpage);
3828
3829                 set_page_private(page, 0);
3830                 ClearPagePrivate(page);
3831                 put_page(page);
3832         }
3833
3834         if (mode != MIGRATE_SYNC_NO_COPY)
3835                 migrate_page_copy(newpage, page);
3836         else
3837                 migrate_page_states(newpage, page);
3838
3839         return MIGRATEPAGE_SUCCESS;
3840 }
3841 #endif
3842
3843 #ifdef CONFIG_SWAP
3844 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3845                                                         unsigned int blkcnt)
3846 {
3847         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3848         unsigned int blkofs;
3849         unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3850         unsigned int secidx = start_blk / blk_per_sec;
3851         unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3852         int ret = 0;
3853
3854         down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3855         filemap_invalidate_lock(inode->i_mapping);
3856
3857         set_inode_flag(inode, FI_ALIGNED_WRITE);
3858
3859         for (; secidx < end_sec; secidx++) {
3860                 down_write(&sbi->pin_sem);
3861
3862                 f2fs_lock_op(sbi);
3863                 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3864                 f2fs_unlock_op(sbi);
3865
3866                 set_inode_flag(inode, FI_DO_DEFRAG);
3867
3868                 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3869                         struct page *page;
3870                         unsigned int blkidx = secidx * blk_per_sec + blkofs;
3871
3872                         page = f2fs_get_lock_data_page(inode, blkidx, true);
3873                         if (IS_ERR(page)) {
3874                                 up_write(&sbi->pin_sem);
3875                                 ret = PTR_ERR(page);
3876                                 goto done;
3877                         }
3878
3879                         set_page_dirty(page);
3880                         f2fs_put_page(page, 1);
3881                 }
3882
3883                 clear_inode_flag(inode, FI_DO_DEFRAG);
3884
3885                 ret = filemap_fdatawrite(inode->i_mapping);
3886
3887                 up_write(&sbi->pin_sem);
3888
3889                 if (ret)
3890                         break;
3891         }
3892
3893 done:
3894         clear_inode_flag(inode, FI_DO_DEFRAG);
3895         clear_inode_flag(inode, FI_ALIGNED_WRITE);
3896
3897         filemap_invalidate_unlock(inode->i_mapping);
3898         up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3899
3900         return ret;
3901 }
3902
3903 static int check_swap_activate(struct swap_info_struct *sis,
3904                                 struct file *swap_file, sector_t *span)
3905 {
3906         struct address_space *mapping = swap_file->f_mapping;
3907         struct inode *inode = mapping->host;
3908         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3909         sector_t cur_lblock;
3910         sector_t last_lblock;
3911         sector_t pblock;
3912         sector_t lowest_pblock = -1;
3913         sector_t highest_pblock = 0;
3914         int nr_extents = 0;
3915         unsigned long nr_pblocks;
3916         unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3917         unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3918         unsigned int not_aligned = 0;
3919         int ret = 0;
3920
3921         /*
3922          * Map all the blocks into the extent list.  This code doesn't try
3923          * to be very smart.
3924          */
3925         cur_lblock = 0;
3926         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3927
3928         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3929                 struct f2fs_map_blocks map;
3930 retry:
3931                 cond_resched();
3932
3933                 memset(&map, 0, sizeof(map));
3934                 map.m_lblk = cur_lblock;
3935                 map.m_len = last_lblock - cur_lblock;
3936                 map.m_next_pgofs = NULL;
3937                 map.m_next_extent = NULL;
3938                 map.m_seg_type = NO_CHECK_TYPE;
3939                 map.m_may_create = false;
3940
3941                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3942                 if (ret)
3943                         goto out;
3944
3945                 /* hole */
3946                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3947                         f2fs_err(sbi, "Swapfile has holes");
3948                         ret = -EINVAL;
3949                         goto out;
3950                 }
3951
3952                 pblock = map.m_pblk;
3953                 nr_pblocks = map.m_len;
3954
3955                 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3956                                 nr_pblocks & sec_blks_mask) {
3957                         not_aligned++;
3958
3959                         nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3960                         if (cur_lblock + nr_pblocks > sis->max)
3961                                 nr_pblocks -= blks_per_sec;
3962
3963                         if (!nr_pblocks) {
3964                                 /* this extent is last one */
3965                                 nr_pblocks = map.m_len;
3966                                 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3967                                 goto next;
3968                         }
3969
3970                         ret = f2fs_migrate_blocks(inode, cur_lblock,
3971                                                         nr_pblocks);
3972                         if (ret)
3973                                 goto out;
3974                         goto retry;
3975                 }
3976 next:
3977                 if (cur_lblock + nr_pblocks >= sis->max)
3978                         nr_pblocks = sis->max - cur_lblock;
3979
3980                 if (cur_lblock) {       /* exclude the header page */
3981                         if (pblock < lowest_pblock)
3982                                 lowest_pblock = pblock;
3983                         if (pblock + nr_pblocks - 1 > highest_pblock)
3984                                 highest_pblock = pblock + nr_pblocks - 1;
3985                 }
3986
3987                 /*
3988                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3989                  */
3990                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3991                 if (ret < 0)
3992                         goto out;
3993                 nr_extents += ret;
3994                 cur_lblock += nr_pblocks;
3995         }
3996         ret = nr_extents;
3997         *span = 1 + highest_pblock - lowest_pblock;
3998         if (cur_lblock == 0)
3999                 cur_lblock = 1; /* force Empty message */
4000         sis->max = cur_lblock;
4001         sis->pages = cur_lblock - 1;
4002         sis->highest_bit = cur_lblock - 1;
4003 out:
4004         if (not_aligned)
4005                 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
4006                           not_aligned, blks_per_sec * F2FS_BLKSIZE);
4007         return ret;
4008 }
4009
4010 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4011                                 sector_t *span)
4012 {
4013         struct inode *inode = file_inode(file);
4014         int ret;
4015
4016         if (!S_ISREG(inode->i_mode))
4017                 return -EINVAL;
4018
4019         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4020                 return -EROFS;
4021
4022         if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4023                 f2fs_err(F2FS_I_SB(inode),
4024                         "Swapfile not supported in LFS mode");
4025                 return -EINVAL;
4026         }
4027
4028         ret = f2fs_convert_inline_inode(inode);
4029         if (ret)
4030                 return ret;
4031
4032         if (!f2fs_disable_compressed_file(inode))
4033                 return -EINVAL;
4034
4035         f2fs_precache_extents(inode);
4036
4037         ret = check_swap_activate(sis, file, span);
4038         if (ret < 0)
4039                 return ret;
4040
4041         set_inode_flag(inode, FI_PIN_FILE);
4042         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4043         return ret;
4044 }
4045
4046 static void f2fs_swap_deactivate(struct file *file)
4047 {
4048         struct inode *inode = file_inode(file);
4049
4050         clear_inode_flag(inode, FI_PIN_FILE);
4051 }
4052 #else
4053 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4054                                 sector_t *span)
4055 {
4056         return -EOPNOTSUPP;
4057 }
4058
4059 static void f2fs_swap_deactivate(struct file *file)
4060 {
4061 }
4062 #endif
4063
4064 const struct address_space_operations f2fs_dblock_aops = {
4065         .readpage       = f2fs_read_data_page,
4066         .readahead      = f2fs_readahead,
4067         .writepage      = f2fs_write_data_page,
4068         .writepages     = f2fs_write_data_pages,
4069         .write_begin    = f2fs_write_begin,
4070         .write_end      = f2fs_write_end,
4071         .set_page_dirty = f2fs_set_data_page_dirty,
4072         .invalidatepage = f2fs_invalidate_page,
4073         .releasepage    = f2fs_release_page,
4074         .direct_IO      = f2fs_direct_IO,
4075         .bmap           = f2fs_bmap,
4076         .swap_activate  = f2fs_swap_activate,
4077         .swap_deactivate = f2fs_swap_deactivate,
4078 #ifdef CONFIG_MIGRATION
4079         .migratepage    = f2fs_migrate_page,
4080 #endif
4081 };
4082
4083 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4084 {
4085         struct address_space *mapping = page_mapping(page);
4086         unsigned long flags;
4087
4088         xa_lock_irqsave(&mapping->i_pages, flags);
4089         __xa_clear_mark(&mapping->i_pages, page_index(page),
4090                                                 PAGECACHE_TAG_DIRTY);
4091         xa_unlock_irqrestore(&mapping->i_pages, flags);
4092 }
4093
4094 int __init f2fs_init_post_read_processing(void)
4095 {
4096         bio_post_read_ctx_cache =
4097                 kmem_cache_create("f2fs_bio_post_read_ctx",
4098                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4099         if (!bio_post_read_ctx_cache)
4100                 goto fail;
4101         bio_post_read_ctx_pool =
4102                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4103                                          bio_post_read_ctx_cache);
4104         if (!bio_post_read_ctx_pool)
4105                 goto fail_free_cache;
4106         return 0;
4107
4108 fail_free_cache:
4109         kmem_cache_destroy(bio_post_read_ctx_cache);
4110 fail:
4111         return -ENOMEM;
4112 }
4113
4114 void f2fs_destroy_post_read_processing(void)
4115 {
4116         mempool_destroy(bio_post_read_ctx_pool);
4117         kmem_cache_destroy(bio_post_read_ctx_cache);
4118 }
4119
4120 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4121 {
4122         if (!f2fs_sb_has_encrypt(sbi) &&
4123                 !f2fs_sb_has_verity(sbi) &&
4124                 !f2fs_sb_has_compression(sbi))
4125                 return 0;
4126
4127         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4128                                                  WQ_UNBOUND | WQ_HIGHPRI,
4129                                                  num_online_cpus());
4130         if (!sbi->post_read_wq)
4131                 return -ENOMEM;
4132         return 0;
4133 }
4134
4135 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4136 {
4137         if (sbi->post_read_wq)
4138                 destroy_workqueue(sbi->post_read_wq);
4139 }
4140
4141 int __init f2fs_init_bio_entry_cache(void)
4142 {
4143         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4144                         sizeof(struct bio_entry));
4145         if (!bio_entry_slab)
4146                 return -ENOMEM;
4147         return 0;
4148 }
4149
4150 void f2fs_destroy_bio_entry_cache(void)
4151 {
4152         kmem_cache_destroy(bio_entry_slab);
4153 }