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         if (f2fs_is_compressed_page(page))
  62                 return false;
  63
  64         inode = mapping->host;
  65         sbi = F2FS_I_SB(inode);
  66
  67         if (inode->i_ino == F2FS_META_INO(sbi) ||
  68                         inode->i_ino == F2FS_NODE_INO(sbi) ||
  69                         S_ISDIR(inode->i_mode) ||
  70                         (S_ISREG(inode->i_mode) &&
  71                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  72                         is_cold_data(page))
  73                 return true;
  74         return false;
  75 }
  76
  77 static enum count_type __read_io_type(struct page *page)
  78 {
  79         struct address_space *mapping = page_file_mapping(page);
  80
  81         if (mapping) {
  82                 struct inode *inode = mapping->host;
  83                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  84
  85                 if (inode->i_ino == F2FS_META_INO(sbi))
  86                         return F2FS_RD_META;
  87
  88                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  89                         return F2FS_RD_NODE;
  90         }
  91         return F2FS_RD_DATA;
  92 }
  93
  94 /* postprocessing steps for read bios */
  95 enum bio_post_read_step {
  96 #ifdef CONFIG_FS_ENCRYPTION
  97         STEP_DECRYPT    = 1 << 0,
  98 #else
  99         STEP_DECRYPT    = 0,    /* compile out the decryption-related code */
 100 #endif
 101 #ifdef CONFIG_F2FS_FS_COMPRESSION
 102         STEP_DECOMPRESS = 1 << 1,
 103 #else
 104         STEP_DECOMPRESS = 0,    /* compile out the decompression-related code */
 105 #endif
 106 #ifdef CONFIG_FS_VERITY
 107         STEP_VERITY     = 1 << 2,
 108 #else
 109         STEP_VERITY     = 0,    /* compile out the verity-related code */
 110 #endif
 111 };
 112
 113 struct bio_post_read_ctx {
 114         struct bio *bio;
 115         struct f2fs_sb_info *sbi;
 116         struct work_struct work;
 117         unsigned int enabled_steps;
 118 };
 119
 120 static void f2fs_finish_read_bio(struct bio *bio)
 121 {
 122         struct bio_vec *bv;
 123         struct bvec_iter_all iter_all;
 124
 125         /*
 126          * Update and unlock the bio's pagecache pages, and put the
 127          * decompression context for any compressed pages.
 128          */
 129         bio_for_each_segment_all(bv, bio, iter_all) {
 130                 struct page *page = bv->bv_page;
 131
 132                 if (f2fs_is_compressed_page(page)) {
 133                         if (bio->bi_status)
 134                                 f2fs_end_read_compressed_page(page, true);
 135                         f2fs_put_page_dic(page);
 136                         continue;
 137                 }
 138
 139                 /* PG_error was set if decryption or verity failed. */
 140                 if (bio->bi_status || PageError(page)) {
 141                         ClearPageUptodate(page);
 142                         /* will re-read again later */
 143                         ClearPageError(page);
 144                 } else {
 145                         SetPageUptodate(page);
 146                 }
 147                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 148                 unlock_page(page);
 149         }
 150
 151         if (bio->bi_private)
 152                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 153         bio_put(bio);
 154 }
 155
 156 static void f2fs_verify_bio(struct work_struct *work)
 157 {
 158         struct bio_post_read_ctx *ctx =
 159                 container_of(work, struct bio_post_read_ctx, work);
 160         struct bio *bio = ctx->bio;
 161         bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 162
 163         /*
 164          * fsverity_verify_bio() may call readpages() again, and while verity
 165          * will be disabled for this, decryption and/or decompression may still
 166          * be needed, resulting in another bio_post_read_ctx being allocated.
 167          * So to prevent deadlocks we need to release the current ctx to the
 168          * mempool first.  This assumes that verity is the last post-read step.
 169          */
 170         mempool_free(ctx, bio_post_read_ctx_pool);
 171         bio->bi_private = NULL;
 172
 173         /*
 174          * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 175          * as those were handled separately by f2fs_end_read_compressed_page().
 176          */
 177         if (may_have_compressed_pages) {
 178                 struct bio_vec *bv;
 179                 struct bvec_iter_all iter_all;
 180
 181                 bio_for_each_segment_all(bv, bio, iter_all) {
 182                         struct page *page = bv->bv_page;
 183
 184                         if (!f2fs_is_compressed_page(page) &&
 185                             !PageError(page) && !fsverity_verify_page(page))
 186                                 SetPageError(page);
 187                 }
 188         } else {
 189                 fsverity_verify_bio(bio);
 190         }
 191
 192         f2fs_finish_read_bio(bio);
 193 }
 194
 195 /*
 196  * If the bio's data needs to be verified with fs-verity, then enqueue the
 197  * verity work for the bio.  Otherwise finish the bio now.
 198  *
 199  * Note that to avoid deadlocks, the verity work can't be done on the
 200  * decryption/decompression workqueue.  This is because verifying the data pages
 201  * can involve reading verity metadata pages from the file, and these verity
 202  * metadata pages may be encrypted and/or compressed.
 203  */
 204 static void f2fs_verify_and_finish_bio(struct bio *bio)
 205 {
 206         struct bio_post_read_ctx *ctx = bio->bi_private;
 207
 208         if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 209                 INIT_WORK(&ctx->work, f2fs_verify_bio);
 210                 fsverity_enqueue_verify_work(&ctx->work);
 211         } else {
 212                 f2fs_finish_read_bio(bio);
 213         }
 214 }
 215
 216 /*
 217  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 218  * remaining page was read by @ctx->bio.
 219  *
 220  * Note that a bio may span clusters (even a mix of compressed and uncompressed
 221  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 222  * that the bio includes at least one compressed page.  The actual decompression
 223  * is done on a per-cluster basis, not a per-bio basis.
 224  */
 225 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
 226 {
 227         struct bio_vec *bv;
 228         struct bvec_iter_all iter_all;
 229         bool all_compressed = true;
 230
 231         bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 232                 struct page *page = bv->bv_page;
 233
 234                 /* PG_error was set if decryption failed. */
 235                 if (f2fs_is_compressed_page(page))
 236                         f2fs_end_read_compressed_page(page, PageError(page));
 237                 else
 238                         all_compressed = false;
 239         }
 240
 241         /*
 242          * Optimization: if all the bio's pages are compressed, then scheduling
 243          * the per-bio verity work is unnecessary, as verity will be fully
 244          * handled at the compression cluster level.
 245          */
 246         if (all_compressed)
 247                 ctx->enabled_steps &= ~STEP_VERITY;
 248 }
 249
 250 static void f2fs_post_read_work(struct work_struct *work)
 251 {
 252         struct bio_post_read_ctx *ctx =
 253                 container_of(work, struct bio_post_read_ctx, work);
 254
 255         if (ctx->enabled_steps & STEP_DECRYPT)
 256                 fscrypt_decrypt_bio(ctx->bio);
 257
 258         if (ctx->enabled_steps & STEP_DECOMPRESS)
 259                 f2fs_handle_step_decompress(ctx);
 260
 261         f2fs_verify_and_finish_bio(ctx->bio);
 262 }
 263
 264 static void f2fs_read_end_io(struct bio *bio)
 265 {
 266         struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 267         struct bio_post_read_ctx *ctx = bio->bi_private;
 268
 269         if (time_to_inject(sbi, FAULT_READ_IO)) {
 270                 f2fs_show_injection_info(sbi, FAULT_READ_IO);
 271                 bio->bi_status = BLK_STS_IOERR;
 272         }
 273
 274         if (bio->bi_status) {
 275                 f2fs_finish_read_bio(bio);
 276                 return;
 277         }
 278
 279         if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
 280                 INIT_WORK(&ctx->work, f2fs_post_read_work);
 281                 queue_work(ctx->sbi->post_read_wq, &ctx->work);
 282         } else {
 283                 f2fs_verify_and_finish_bio(bio);
 284         }
 285 }
 286
 287 static void f2fs_write_end_io(struct bio *bio)
 288 {
 289         struct f2fs_sb_info *sbi = bio->bi_private;
 290         struct bio_vec *bvec;
 291         struct bvec_iter_all iter_all;
 292
 293         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 294                 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
 295                 bio->bi_status = BLK_STS_IOERR;
 296         }
 297
 298         bio_for_each_segment_all(bvec, bio, iter_all) {
 299                 struct page *page = bvec->bv_page;
 300                 enum count_type type = WB_DATA_TYPE(page);
 301
 302                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 303                         set_page_private(page, (unsigned long)NULL);
 304                         ClearPagePrivate(page);
 305                         unlock_page(page);
 306                         mempool_free(page, sbi->write_io_dummy);
 307
 308                         if (unlikely(bio->bi_status))
 309                                 f2fs_stop_checkpoint(sbi, true);
 310                         continue;
 311                 }
 312
 313                 fscrypt_finalize_bounce_page(&page);
 314
 315 #ifdef CONFIG_F2FS_FS_COMPRESSION
 316                 if (f2fs_is_compressed_page(page)) {
 317                         f2fs_compress_write_end_io(bio, page);
 318                         continue;
 319                 }
 320 #endif
 321
 322                 if (unlikely(bio->bi_status)) {
 323                         mapping_set_error(page->mapping, -EIO);
 324                         if (type == F2FS_WB_CP_DATA)
 325                                 f2fs_stop_checkpoint(sbi, true);
 326                 }
 327
 328                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 329                                         page->index != nid_of_node(page));
 330
 331                 dec_page_count(sbi, type);
 332                 if (f2fs_in_warm_node_list(sbi, page))
 333                         f2fs_del_fsync_node_entry(sbi, page);
 334                 clear_cold_data(page);
 335                 end_page_writeback(page);
 336         }
 337         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 338                                 wq_has_sleeper(&sbi->cp_wait))
 339                 wake_up(&sbi->cp_wait);
 340
 341         bio_put(bio);
 342 }
 343
 344 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 345                                 block_t blk_addr, struct bio *bio)
 346 {
 347         struct block_device *bdev = sbi->sb->s_bdev;
 348         int i;
 349
 350         if (f2fs_is_multi_device(sbi)) {
 351                 for (i = 0; i < sbi->s_ndevs; i++) {
 352                         if (FDEV(i).start_blk <= blk_addr &&
 353                             FDEV(i).end_blk >= blk_addr) {
 354                                 blk_addr -= FDEV(i).start_blk;
 355                                 bdev = FDEV(i).bdev;
 356                                 break;
 357                         }
 358                 }
 359         }
 360         if (bio) {
 361                 bio_set_dev(bio, bdev);
 362                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 363         }
 364         return bdev;
 365 }
 366
 367 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 368 {
 369         int i;
 370
 371         if (!f2fs_is_multi_device(sbi))
 372                 return 0;
 373
 374         for (i = 0; i < sbi->s_ndevs; i++)
 375                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 376                         return i;
 377         return 0;
 378 }
 379
 380 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 381 {
 382         struct f2fs_sb_info *sbi = fio->sbi;
 383         struct bio *bio;
 384
 385         bio = bio_alloc_bioset(GFP_NOIO, npages, &f2fs_bioset);
 386
 387         f2fs_target_device(sbi, fio->new_blkaddr, bio);
 388         if (is_read_io(fio->op)) {
 389                 bio->bi_end_io = f2fs_read_end_io;
 390                 bio->bi_private = NULL;
 391         } else {
 392                 bio->bi_end_io = f2fs_write_end_io;
 393                 bio->bi_private = sbi;
 394                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
 395                                                 fio->type, fio->temp);
 396         }
 397         if (fio->io_wbc)
 398                 wbc_init_bio(fio->io_wbc, bio);
 399
 400         return bio;
 401 }
 402
 403 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 404                                   pgoff_t first_idx,
 405                                   const struct f2fs_io_info *fio,
 406                                   gfp_t gfp_mask)
 407 {
 408         /*
 409          * The f2fs garbage collector sets ->encrypted_page when it wants to
 410          * read/write raw data without encryption.
 411          */
 412         if (!fio || !fio->encrypted_page)
 413                 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 414 }
 415
 416 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 417                                      pgoff_t next_idx,
 418                                      const struct f2fs_io_info *fio)
 419 {
 420         /*
 421          * The f2fs garbage collector sets ->encrypted_page when it wants to
 422          * read/write raw data without encryption.
 423          */
 424         if (fio && fio->encrypted_page)
 425                 return !bio_has_crypt_ctx(bio);
 426
 427         return fscrypt_mergeable_bio(bio, inode, next_idx);
 428 }
 429
 430 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 431                                 struct bio *bio, enum page_type type)
 432 {
 433         if (!is_read_io(bio_op(bio))) {
 434                 unsigned int start;
 435
 436                 if (type != DATA && type != NODE)
 437                         goto submit_io;
 438
 439                 if (f2fs_lfs_mode(sbi) && current->plug)
 440                         blk_finish_plug(current->plug);
 441
 442                 if (!F2FS_IO_ALIGNED(sbi))
 443                         goto submit_io;
 444
 445                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 446                 start %= F2FS_IO_SIZE(sbi);
 447
 448                 if (start == 0)
 449                         goto submit_io;
 450
 451                 /* fill dummy pages */
 452                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 453                         struct page *page =
 454                                 mempool_alloc(sbi->write_io_dummy,
 455                                               GFP_NOIO | __GFP_NOFAIL);
 456                         f2fs_bug_on(sbi, !page);
 457
 458                         zero_user_segment(page, 0, PAGE_SIZE);
 459                         SetPagePrivate(page);
 460                         set_page_private(page, DUMMY_WRITTEN_PAGE);
 461                         lock_page(page);
 462                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 463                                 f2fs_bug_on(sbi, 1);
 464                 }
 465                 /*
 466                  * In the NODE case, we lose next block address chain. So, we
 467                  * need to do checkpoint in f2fs_sync_file.
 468                  */
 469                 if (type == NODE)
 470                         set_sbi_flag(sbi, SBI_NEED_CP);
 471         }
 472 submit_io:
 473         if (is_read_io(bio_op(bio)))
 474                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 475         else
 476                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 477         submit_bio(bio);
 478 }
 479
 480 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
 481                                 struct bio *bio, enum page_type type)
 482 {
 483         __submit_bio(sbi, bio, type);
 484 }
 485
 486 static void __attach_io_flag(struct f2fs_io_info *fio)
 487 {
 488         struct f2fs_sb_info *sbi = fio->sbi;
 489         unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
 490         unsigned int io_flag, fua_flag, meta_flag;
 491
 492         if (fio->type == DATA)
 493                 io_flag = sbi->data_io_flag;
 494         else if (fio->type == NODE)
 495                 io_flag = sbi->node_io_flag;
 496         else
 497                 return;
 498
 499         fua_flag = io_flag & temp_mask;
 500         meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 501
 502         /*
 503          * data/node io flag bits per temp:
 504          *      REQ_META     |      REQ_FUA      |
 505          *    5 |    4 |   3 |    2 |    1 |   0 |
 506          * Cold | Warm | Hot | Cold | Warm | Hot |
 507          */
 508         if ((1 << fio->temp) & meta_flag)
 509                 fio->op_flags |= REQ_META;
 510         if ((1 << fio->temp) & fua_flag)
 511                 fio->op_flags |= REQ_FUA;
 512 }
 513
 514 static void __submit_merged_bio(struct f2fs_bio_info *io)
 515 {
 516         struct f2fs_io_info *fio = &io->fio;
 517
 518         if (!io->bio)
 519                 return;
 520
 521         __attach_io_flag(fio);
 522         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 523
 524         if (is_read_io(fio->op))
 525                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 526         else
 527                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 528
 529         __submit_bio(io->sbi, io->bio, fio->type);
 530         io->bio = NULL;
 531 }
 532
 533 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 534                                                 struct page *page, nid_t ino)
 535 {
 536         struct bio_vec *bvec;
 537         struct bvec_iter_all iter_all;
 538
 539         if (!bio)
 540                 return false;
 541
 542         if (!inode && !page && !ino)
 543                 return true;
 544
 545         bio_for_each_segment_all(bvec, bio, iter_all) {
 546                 struct page *target = bvec->bv_page;
 547
 548                 if (fscrypt_is_bounce_page(target)) {
 549                         target = fscrypt_pagecache_page(target);
 550                         if (IS_ERR(target))
 551                                 continue;
 552                 }
 553                 if (f2fs_is_compressed_page(target)) {
 554                         target = f2fs_compress_control_page(target);
 555                         if (IS_ERR(target))
 556                                 continue;
 557                 }
 558
 559                 if (inode && inode == target->mapping->host)
 560                         return true;
 561                 if (page && page == target)
 562                         return true;
 563                 if (ino && ino == ino_of_node(target))
 564                         return true;
 565         }
 566
 567         return false;
 568 }
 569
 570 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 571                                 enum page_type type, enum temp_type temp)
 572 {
 573         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 574         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 575
 576         down_write(&io->io_rwsem);
 577
 578         /* change META to META_FLUSH in the checkpoint procedure */
 579         if (type >= META_FLUSH) {
 580                 io->fio.type = META_FLUSH;
 581                 io->fio.op = REQ_OP_WRITE;
 582                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 583                 if (!test_opt(sbi, NOBARRIER))
 584                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 585         }
 586         __submit_merged_bio(io);
 587         up_write(&io->io_rwsem);
 588 }
 589
 590 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 591                                 struct inode *inode, struct page *page,
 592                                 nid_t ino, enum page_type type, bool force)
 593 {
 594         enum temp_type temp;
 595         bool ret = true;
 596
 597         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 598                 if (!force)     {
 599                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 600                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 601
 602                         down_read(&io->io_rwsem);
 603                         ret = __has_merged_page(io->bio, inode, page, ino);
 604                         up_read(&io->io_rwsem);
 605                 }
 606                 if (ret)
 607                         __f2fs_submit_merged_write(sbi, type, temp);
 608
 609                 /* TODO: use HOT temp only for meta pages now. */
 610                 if (type >= META)
 611                         break;
 612         }
 613 }
 614
 615 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 616 {
 617         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 618 }
 619
 620 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 621                                 struct inode *inode, struct page *page,
 622                                 nid_t ino, enum page_type type)
 623 {
 624         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 625 }
 626
 627 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 628 {
 629         f2fs_submit_merged_write(sbi, DATA);
 630         f2fs_submit_merged_write(sbi, NODE);
 631         f2fs_submit_merged_write(sbi, META);
 632 }
 633
 634 /*
 635  * Fill the locked page with data located in the block address.
 636  * A caller needs to unlock the page on failure.
 637  */
 638 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 639 {
 640         struct bio *bio;
 641         struct page *page = fio->encrypted_page ?
 642                         fio->encrypted_page : fio->page;
 643
 644         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 645                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 646                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 647                 return -EFSCORRUPTED;
 648
 649         trace_f2fs_submit_page_bio(page, fio);
 650
 651         /* Allocate a new bio */
 652         bio = __bio_alloc(fio, 1);
 653
 654         f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 655                                fio->page->index, fio, GFP_NOIO);
 656
 657         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 658                 bio_put(bio);
 659                 return -EFAULT;
 660         }
 661
 662         if (fio->io_wbc && !is_read_io(fio->op))
 663                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 664
 665         __attach_io_flag(fio);
 666         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 667
 668         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 669                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 670
 671         __submit_bio(fio->sbi, bio, fio->type);
 672         return 0;
 673 }
 674
 675 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 676                                 block_t last_blkaddr, block_t cur_blkaddr)
 677 {
 678         if (unlikely(sbi->max_io_bytes &&
 679                         bio->bi_iter.bi_size >= sbi->max_io_bytes))
 680                 return false;
 681         if (last_blkaddr + 1 != cur_blkaddr)
 682                 return false;
 683         return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 684 }
 685
 686 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 687                                                 struct f2fs_io_info *fio)
 688 {
 689         if (io->fio.op != fio->op)
 690                 return false;
 691         return io->fio.op_flags == fio->op_flags;
 692 }
 693
 694 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 695                                         struct f2fs_bio_info *io,
 696                                         struct f2fs_io_info *fio,
 697                                         block_t last_blkaddr,
 698                                         block_t cur_blkaddr)
 699 {
 700         if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 701                 unsigned int filled_blocks =
 702                                 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 703                 unsigned int io_size = F2FS_IO_SIZE(sbi);
 704                 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 705
 706                 /* IOs in bio is aligned and left space of vectors is not enough */
 707                 if (!(filled_blocks % io_size) && left_vecs < io_size)
 708                         return false;
 709         }
 710         if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 711                 return false;
 712         return io_type_is_mergeable(io, fio);
 713 }
 714
 715 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 716                                 struct page *page, enum temp_type temp)
 717 {
 718         struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 719         struct bio_entry *be;
 720
 721         be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
 722         be->bio = bio;
 723         bio_get(bio);
 724
 725         if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 726                 f2fs_bug_on(sbi, 1);
 727
 728         down_write(&io->bio_list_lock);
 729         list_add_tail(&be->list, &io->bio_list);
 730         up_write(&io->bio_list_lock);
 731 }
 732
 733 static void del_bio_entry(struct bio_entry *be)
 734 {
 735         list_del(&be->list);
 736         kmem_cache_free(bio_entry_slab, be);
 737 }
 738
 739 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 740                                                         struct page *page)
 741 {
 742         struct f2fs_sb_info *sbi = fio->sbi;
 743         enum temp_type temp;
 744         bool found = false;
 745         int ret = -EAGAIN;
 746
 747         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 748                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 749                 struct list_head *head = &io->bio_list;
 750                 struct bio_entry *be;
 751
 752                 down_write(&io->bio_list_lock);
 753                 list_for_each_entry(be, head, list) {
 754                         if (be->bio != *bio)
 755                                 continue;
 756
 757                         found = true;
 758
 759                         f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 760                                                             *fio->last_block,
 761                                                             fio->new_blkaddr));
 762                         if (f2fs_crypt_mergeable_bio(*bio,
 763                                         fio->page->mapping->host,
 764                                         fio->page->index, fio) &&
 765                             bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 766                                         PAGE_SIZE) {
 767                                 ret = 0;
 768                                 break;
 769                         }
 770
 771                         /* page can't be merged into bio; submit the bio */
 772                         del_bio_entry(be);
 773                         __submit_bio(sbi, *bio, DATA);
 774                         break;
 775                 }
 776                 up_write(&io->bio_list_lock);
 777         }
 778
 779         if (ret) {
 780                 bio_put(*bio);
 781                 *bio = NULL;
 782         }
 783
 784         return ret;
 785 }
 786
 787 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 788                                         struct bio **bio, struct page *page)
 789 {
 790         enum temp_type temp;
 791         bool found = false;
 792         struct bio *target = bio ? *bio : NULL;
 793
 794         for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 795                 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 796                 struct list_head *head = &io->bio_list;
 797                 struct bio_entry *be;
 798
 799                 if (list_empty(head))
 800                         continue;
 801
 802                 down_read(&io->bio_list_lock);
 803                 list_for_each_entry(be, head, list) {
 804                         if (target)
 805                                 found = (target == be->bio);
 806                         else
 807                                 found = __has_merged_page(be->bio, NULL,
 808                                                                 page, 0);
 809                         if (found)
 810                                 break;
 811                 }
 812                 up_read(&io->bio_list_lock);
 813
 814                 if (!found)
 815                         continue;
 816
 817                 found = false;
 818
 819                 down_write(&io->bio_list_lock);
 820                 list_for_each_entry(be, head, list) {
 821                         if (target)
 822                                 found = (target == be->bio);
 823                         else
 824                                 found = __has_merged_page(be->bio, NULL,
 825                                                                 page, 0);
 826                         if (found) {
 827                                 target = be->bio;
 828                                 del_bio_entry(be);
 829                                 break;
 830                         }
 831                 }
 832                 up_write(&io->bio_list_lock);
 833         }
 834
 835         if (found)
 836                 __submit_bio(sbi, target, DATA);
 837         if (bio && *bio) {
 838                 bio_put(*bio);
 839                 *bio = NULL;
 840         }
 841 }
 842
 843 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 844 {
 845         struct bio *bio = *fio->bio;
 846         struct page *page = fio->encrypted_page ?
 847                         fio->encrypted_page : fio->page;
 848
 849         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 850                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 851                 return -EFSCORRUPTED;
 852
 853         trace_f2fs_submit_page_bio(page, fio);
 854
 855         if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 856                                                 fio->new_blkaddr))
 857                 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 858 alloc_new:
 859         if (!bio) {
 860                 bio = __bio_alloc(fio, BIO_MAX_PAGES);
 861                 __attach_io_flag(fio);
 862                 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 863                                        fio->page->index, fio, GFP_NOIO);
 864                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 865
 866                 add_bio_entry(fio->sbi, bio, page, fio->temp);
 867         } else {
 868                 if (add_ipu_page(fio, &bio, page))
 869                         goto alloc_new;
 870         }
 871
 872         if (fio->io_wbc)
 873                 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
 874
 875         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 876
 877         *fio->last_block = fio->new_blkaddr;
 878         *fio->bio = bio;
 879
 880         return 0;
 881 }
 882
 883 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 884 {
 885         struct f2fs_sb_info *sbi = fio->sbi;
 886         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 887         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 888         struct page *bio_page;
 889
 890         f2fs_bug_on(sbi, is_read_io(fio->op));
 891
 892         down_write(&io->io_rwsem);
 893 next:
 894         if (fio->in_list) {
 895                 spin_lock(&io->io_lock);
 896                 if (list_empty(&io->io_list)) {
 897                         spin_unlock(&io->io_lock);
 898                         goto out;
 899                 }
 900                 fio = list_first_entry(&io->io_list,
 901                                                 struct f2fs_io_info, list);
 902                 list_del(&fio->list);
 903                 spin_unlock(&io->io_lock);
 904         }
 905
 906         verify_fio_blkaddr(fio);
 907
 908         if (fio->encrypted_page)
 909                 bio_page = fio->encrypted_page;
 910         else if (fio->compressed_page)
 911                 bio_page = fio->compressed_page;
 912         else
 913                 bio_page = fio->page;
 914
 915         /* set submitted = true as a return value */
 916         fio->submitted = true;
 917
 918         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 919
 920         if (io->bio &&
 921             (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
 922                               fio->new_blkaddr) ||
 923              !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
 924                                        bio_page->index, fio)))
 925                 __submit_merged_bio(io);
 926 alloc_new:
 927         if (io->bio == NULL) {
 928                 if (F2FS_IO_ALIGNED(sbi) &&
 929                                 (fio->type == DATA || fio->type == NODE) &&
 930                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 931                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 932                         fio->retry = true;
 933                         goto skip;
 934                 }
 935                 io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
 936                 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
 937                                        bio_page->index, fio, GFP_NOIO);
 938                 io->fio = *fio;
 939         }
 940
 941         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 942                 __submit_merged_bio(io);
 943                 goto alloc_new;
 944         }
 945
 946         if (fio->io_wbc)
 947                 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
 948
 949         io->last_block_in_bio = fio->new_blkaddr;
 950
 951         trace_f2fs_submit_page_write(fio->page, fio);
 952 skip:
 953         if (fio->in_list)
 954                 goto next;
 955 out:
 956         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 957                                 !f2fs_is_checkpoint_ready(sbi))
 958                 __submit_merged_bio(io);
 959         up_write(&io->io_rwsem);
 960 }
 961
 962 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 963                                       unsigned nr_pages, unsigned op_flag,
 964                                       pgoff_t first_idx, bool for_write)
 965 {
 966         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 967         struct bio *bio;
 968         struct bio_post_read_ctx *ctx;
 969         unsigned int post_read_steps = 0;
 970
 971         bio = bio_alloc_bioset(for_write ? GFP_NOIO : GFP_KERNEL,
 972                                bio_max_segs(nr_pages), &f2fs_bioset);
 973         if (!bio)
 974                 return ERR_PTR(-ENOMEM);
 975
 976         f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
 977
 978         f2fs_target_device(sbi, blkaddr, bio);
 979         bio->bi_end_io = f2fs_read_end_io;
 980         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 981
 982         if (fscrypt_inode_uses_fs_layer_crypto(inode))
 983                 post_read_steps |= STEP_DECRYPT;
 984
 985         if (f2fs_need_verity(inode, first_idx))
 986                 post_read_steps |= STEP_VERITY;
 987
 988         /*
 989          * STEP_DECOMPRESS is handled specially, since a compressed file might
 990          * contain both compressed and uncompressed clusters.  We'll allocate a
 991          * bio_post_read_ctx if the file is compressed, but the caller is
 992          * responsible for enabling STEP_DECOMPRESS if it's actually needed.
 993          */
 994
 995         if (post_read_steps || f2fs_compressed_file(inode)) {
 996                 /* Due to the mempool, this never fails. */
 997                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 998                 ctx->bio = bio;
 999                 ctx->sbi = sbi;
1000                 ctx->enabled_steps = post_read_steps;
1001                 bio->bi_private = ctx;
1002         }
1003
1004         return bio;
1005 }
1006
1007 /* This can handle encryption stuffs */
1008 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1009                                  block_t blkaddr, int op_flags, bool for_write)
1010 {
1011         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1012         struct bio *bio;
1013
1014         bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1015                                         page->index, for_write);
1016         if (IS_ERR(bio))
1017                 return PTR_ERR(bio);
1018
1019         /* wait for GCed page writeback via META_MAPPING */
1020         f2fs_wait_on_block_writeback(inode, blkaddr);
1021
1022         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1023                 bio_put(bio);
1024                 return -EFAULT;
1025         }
1026         ClearPageError(page);
1027         inc_page_count(sbi, F2FS_RD_DATA);
1028         f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1029         __submit_bio(sbi, bio, DATA);
1030         return 0;
1031 }
1032
1033 static void __set_data_blkaddr(struct dnode_of_data *dn)
1034 {
1035         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1036         __le32 *addr_array;
1037         int base = 0;
1038
1039         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1040                 base = get_extra_isize(dn->inode);
1041
1042         /* Get physical address of data block */
1043         addr_array = blkaddr_in_node(rn);
1044         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1045 }
1046
1047 /*
1048  * Lock ordering for the change of data block address:
1049  * ->data_page
1050  *  ->node_page
1051  *    update block addresses in the node page
1052  */
1053 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1054 {
1055         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1056         __set_data_blkaddr(dn);
1057         if (set_page_dirty(dn->node_page))
1058                 dn->node_changed = true;
1059 }
1060
1061 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1062 {
1063         dn->data_blkaddr = blkaddr;
1064         f2fs_set_data_blkaddr(dn);
1065         f2fs_update_extent_cache(dn);
1066 }
1067
1068 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1069 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1070 {
1071         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1072         int err;
1073
1074         if (!count)
1075                 return 0;
1076
1077         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1078                 return -EPERM;
1079         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1080                 return err;
1081
1082         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1083                                                 dn->ofs_in_node, count);
1084
1085         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1086
1087         for (; count > 0; dn->ofs_in_node++) {
1088                 block_t blkaddr = f2fs_data_blkaddr(dn);
1089
1090                 if (blkaddr == NULL_ADDR) {
1091                         dn->data_blkaddr = NEW_ADDR;
1092                         __set_data_blkaddr(dn);
1093                         count--;
1094                 }
1095         }
1096
1097         if (set_page_dirty(dn->node_page))
1098                 dn->node_changed = true;
1099         return 0;
1100 }
1101
1102 /* Should keep dn->ofs_in_node unchanged */
1103 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1104 {
1105         unsigned int ofs_in_node = dn->ofs_in_node;
1106         int ret;
1107
1108         ret = f2fs_reserve_new_blocks(dn, 1);
1109         dn->ofs_in_node = ofs_in_node;
1110         return ret;
1111 }
1112
1113 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1114 {
1115         bool need_put = dn->inode_page ? false : true;
1116         int err;
1117
1118         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1119         if (err)
1120                 return err;
1121
1122         if (dn->data_blkaddr == NULL_ADDR)
1123                 err = f2fs_reserve_new_block(dn);
1124         if (err || need_put)
1125                 f2fs_put_dnode(dn);
1126         return err;
1127 }
1128
1129 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1130 {
1131         struct extent_info ei = {0, 0, 0};
1132         struct inode *inode = dn->inode;
1133
1134         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1135                 dn->data_blkaddr = ei.blk + index - ei.fofs;
1136                 return 0;
1137         }
1138
1139         return f2fs_reserve_block(dn, index);
1140 }
1141
1142 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1143                                                 int op_flags, bool for_write)
1144 {
1145         struct address_space *mapping = inode->i_mapping;
1146         struct dnode_of_data dn;
1147         struct page *page;
1148         struct extent_info ei = {0,0,0};
1149         int err;
1150
1151         page = f2fs_grab_cache_page(mapping, index, for_write);
1152         if (!page)
1153                 return ERR_PTR(-ENOMEM);
1154
1155         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1156                 dn.data_blkaddr = ei.blk + index - ei.fofs;
1157                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1158                                                 DATA_GENERIC_ENHANCE_READ)) {
1159                         err = -EFSCORRUPTED;
1160                         goto put_err;
1161                 }
1162                 goto got_it;
1163         }
1164
1165         set_new_dnode(&dn, inode, NULL, NULL, 0);
1166         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1167         if (err)
1168                 goto put_err;
1169         f2fs_put_dnode(&dn);
1170
1171         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1172                 err = -ENOENT;
1173                 goto put_err;
1174         }
1175         if (dn.data_blkaddr != NEW_ADDR &&
1176                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1177                                                 dn.data_blkaddr,
1178                                                 DATA_GENERIC_ENHANCE)) {
1179                 err = -EFSCORRUPTED;
1180                 goto put_err;
1181         }
1182 got_it:
1183         if (PageUptodate(page)) {
1184                 unlock_page(page);
1185                 return page;
1186         }
1187
1188         /*
1189          * A new dentry page is allocated but not able to be written, since its
1190          * new inode page couldn't be allocated due to -ENOSPC.
1191          * In such the case, its blkaddr can be remained as NEW_ADDR.
1192          * see, f2fs_add_link -> f2fs_get_new_data_page ->
1193          * f2fs_init_inode_metadata.
1194          */
1195         if (dn.data_blkaddr == NEW_ADDR) {
1196                 zero_user_segment(page, 0, PAGE_SIZE);
1197                 if (!PageUptodate(page))
1198                         SetPageUptodate(page);
1199                 unlock_page(page);
1200                 return page;
1201         }
1202
1203         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1204                                                 op_flags, for_write);
1205         if (err)
1206                 goto put_err;
1207         return page;
1208
1209 put_err:
1210         f2fs_put_page(page, 1);
1211         return ERR_PTR(err);
1212 }
1213
1214 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1215 {
1216         struct address_space *mapping = inode->i_mapping;
1217         struct page *page;
1218
1219         page = find_get_page(mapping, index);
1220         if (page && PageUptodate(page))
1221                 return page;
1222         f2fs_put_page(page, 0);
1223
1224         page = f2fs_get_read_data_page(inode, index, 0, false);
1225         if (IS_ERR(page))
1226                 return page;
1227
1228         if (PageUptodate(page))
1229                 return page;
1230
1231         wait_on_page_locked(page);
1232         if (unlikely(!PageUptodate(page))) {
1233                 f2fs_put_page(page, 0);
1234                 return ERR_PTR(-EIO);
1235         }
1236         return page;
1237 }
1238
1239 /*
1240  * If it tries to access a hole, return an error.
1241  * Because, the callers, functions in dir.c and GC, should be able to know
1242  * whether this page exists or not.
1243  */
1244 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1245                                                         bool for_write)
1246 {
1247         struct address_space *mapping = inode->i_mapping;
1248         struct page *page;
1249 repeat:
1250         page = f2fs_get_read_data_page(inode, index, 0, for_write);
1251         if (IS_ERR(page))
1252                 return page;
1253
1254         /* wait for read completion */
1255         lock_page(page);
1256         if (unlikely(page->mapping != mapping)) {
1257                 f2fs_put_page(page, 1);
1258                 goto repeat;
1259         }
1260         if (unlikely(!PageUptodate(page))) {
1261                 f2fs_put_page(page, 1);
1262                 return ERR_PTR(-EIO);
1263         }
1264         return page;
1265 }
1266
1267 /*
1268  * Caller ensures that this data page is never allocated.
1269  * A new zero-filled data page is allocated in the page cache.
1270  *
1271  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1272  * f2fs_unlock_op().
1273  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1274  * ipage should be released by this function.
1275  */
1276 struct page *f2fs_get_new_data_page(struct inode *inode,
1277                 struct page *ipage, pgoff_t index, bool new_i_size)
1278 {
1279         struct address_space *mapping = inode->i_mapping;
1280         struct page *page;
1281         struct dnode_of_data dn;
1282         int err;
1283
1284         page = f2fs_grab_cache_page(mapping, index, true);
1285         if (!page) {
1286                 /*
1287                  * before exiting, we should make sure ipage will be released
1288                  * if any error occur.
1289                  */
1290                 f2fs_put_page(ipage, 1);
1291                 return ERR_PTR(-ENOMEM);
1292         }
1293
1294         set_new_dnode(&dn, inode, ipage, NULL, 0);
1295         err = f2fs_reserve_block(&dn, index);
1296         if (err) {
1297                 f2fs_put_page(page, 1);
1298                 return ERR_PTR(err);
1299         }
1300         if (!ipage)
1301                 f2fs_put_dnode(&dn);
1302
1303         if (PageUptodate(page))
1304                 goto got_it;
1305
1306         if (dn.data_blkaddr == NEW_ADDR) {
1307                 zero_user_segment(page, 0, PAGE_SIZE);
1308                 if (!PageUptodate(page))
1309                         SetPageUptodate(page);
1310         } else {
1311                 f2fs_put_page(page, 1);
1312
1313                 /* if ipage exists, blkaddr should be NEW_ADDR */
1314                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1315                 page = f2fs_get_lock_data_page(inode, index, true);
1316                 if (IS_ERR(page))
1317                         return page;
1318         }
1319 got_it:
1320         if (new_i_size && i_size_read(inode) <
1321                                 ((loff_t)(index + 1) << PAGE_SHIFT))
1322                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1323         return page;
1324 }
1325
1326 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1327 {
1328         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1329         struct f2fs_summary sum;
1330         struct node_info ni;
1331         block_t old_blkaddr;
1332         blkcnt_t count = 1;
1333         int err;
1334
1335         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1336                 return -EPERM;
1337
1338         err = f2fs_get_node_info(sbi, dn->nid, &ni);
1339         if (err)
1340                 return err;
1341
1342         dn->data_blkaddr = f2fs_data_blkaddr(dn);
1343         if (dn->data_blkaddr != NULL_ADDR)
1344                 goto alloc;
1345
1346         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1347                 return err;
1348
1349 alloc:
1350         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1351         old_blkaddr = dn->data_blkaddr;
1352         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1353                                 &sum, seg_type, NULL);
1354         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1355                 invalidate_mapping_pages(META_MAPPING(sbi),
1356                                         old_blkaddr, old_blkaddr);
1357         f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1358
1359         /*
1360          * i_size will be updated by direct_IO. Otherwise, we'll get stale
1361          * data from unwritten block via dio_read.
1362          */
1363         return 0;
1364 }
1365
1366 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1367 {
1368         struct inode *inode = file_inode(iocb->ki_filp);
1369         struct f2fs_map_blocks map;
1370         int flag;
1371         int err = 0;
1372         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1373
1374         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1375         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1376         if (map.m_len > map.m_lblk)
1377                 map.m_len -= map.m_lblk;
1378         else
1379                 map.m_len = 0;
1380
1381         map.m_next_pgofs = NULL;
1382         map.m_next_extent = NULL;
1383         map.m_seg_type = NO_CHECK_TYPE;
1384         map.m_may_create = true;
1385
1386         if (direct_io) {
1387                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1388                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1389                                         F2FS_GET_BLOCK_PRE_AIO :
1390                                         F2FS_GET_BLOCK_PRE_DIO;
1391                 goto map_blocks;
1392         }
1393         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1394                 err = f2fs_convert_inline_inode(inode);
1395                 if (err)
1396                         return err;
1397         }
1398         if (f2fs_has_inline_data(inode))
1399                 return err;
1400
1401         flag = F2FS_GET_BLOCK_PRE_AIO;
1402
1403 map_blocks:
1404         err = f2fs_map_blocks(inode, &map, 1, flag);
1405         if (map.m_len > 0 && err == -ENOSPC) {
1406                 if (!direct_io)
1407                         set_inode_flag(inode, FI_NO_PREALLOC);
1408                 err = 0;
1409         }
1410         return err;
1411 }
1412
1413 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1414 {
1415         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1416                 if (lock)
1417                         down_read(&sbi->node_change);
1418                 else
1419                         up_read(&sbi->node_change);
1420         } else {
1421                 if (lock)
1422                         f2fs_lock_op(sbi);
1423                 else
1424                         f2fs_unlock_op(sbi);
1425         }
1426 }
1427
1428 /*
1429  * f2fs_map_blocks() tries to find or build mapping relationship which
1430  * maps continuous logical blocks to physical blocks, and return such
1431  * info via f2fs_map_blocks structure.
1432  */
1433 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1434                                                 int create, int flag)
1435 {
1436         unsigned int maxblocks = map->m_len;
1437         struct dnode_of_data dn;
1438         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1439         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1440         pgoff_t pgofs, end_offset, end;
1441         int err = 0, ofs = 1;
1442         unsigned int ofs_in_node, last_ofs_in_node;
1443         blkcnt_t prealloc;
1444         struct extent_info ei = {0,0,0};
1445         block_t blkaddr;
1446         unsigned int start_pgofs;
1447
1448         if (!maxblocks)
1449                 return 0;
1450
1451         map->m_len = 0;
1452         map->m_flags = 0;
1453
1454         /* it only supports block size == page size */
1455         pgofs = (pgoff_t)map->m_lblk;
1456         end = pgofs + maxblocks;
1457
1458         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1459                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1460                                                         map->m_may_create)
1461                         goto next_dnode;
1462
1463                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1464                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1465                 map->m_flags = F2FS_MAP_MAPPED;
1466                 if (map->m_next_extent)
1467                         *map->m_next_extent = pgofs + map->m_len;
1468
1469                 /* for hardware encryption, but to avoid potential issue in future */
1470                 if (flag == F2FS_GET_BLOCK_DIO)
1471                         f2fs_wait_on_block_writeback_range(inode,
1472                                                 map->m_pblk, map->m_len);
1473                 goto out;
1474         }
1475
1476 next_dnode:
1477         if (map->m_may_create)
1478                 f2fs_do_map_lock(sbi, flag, true);
1479
1480         /* When reading holes, we need its node page */
1481         set_new_dnode(&dn, inode, NULL, NULL, 0);
1482         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1483         if (err) {
1484                 if (flag == F2FS_GET_BLOCK_BMAP)
1485                         map->m_pblk = 0;
1486                 if (err == -ENOENT) {
1487                         err = 0;
1488                         if (map->m_next_pgofs)
1489                                 *map->m_next_pgofs =
1490                                         f2fs_get_next_page_offset(&dn, pgofs);
1491                         if (map->m_next_extent)
1492                                 *map->m_next_extent =
1493                                         f2fs_get_next_page_offset(&dn, pgofs);
1494                 }
1495                 goto unlock_out;
1496         }
1497
1498         start_pgofs = pgofs;
1499         prealloc = 0;
1500         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1501         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1502
1503 next_block:
1504         blkaddr = f2fs_data_blkaddr(&dn);
1505
1506         if (__is_valid_data_blkaddr(blkaddr) &&
1507                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1508                 err = -EFSCORRUPTED;
1509                 goto sync_out;
1510         }
1511
1512         if (__is_valid_data_blkaddr(blkaddr)) {
1513                 /* use out-place-update for driect IO under LFS mode */
1514                 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1515                                                         map->m_may_create) {
1516                         err = __allocate_data_block(&dn, map->m_seg_type);
1517                         if (err)
1518                                 goto sync_out;
1519                         blkaddr = dn.data_blkaddr;
1520                         set_inode_flag(inode, FI_APPEND_WRITE);
1521                 }
1522         } else {
1523                 if (create) {
1524                         if (unlikely(f2fs_cp_error(sbi))) {
1525                                 err = -EIO;
1526                                 goto sync_out;
1527                         }
1528                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1529                                 if (blkaddr == NULL_ADDR) {
1530                                         prealloc++;
1531                                         last_ofs_in_node = dn.ofs_in_node;
1532                                 }
1533                         } else {
1534                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1535                                         flag != F2FS_GET_BLOCK_DIO);
1536                                 err = __allocate_data_block(&dn,
1537                                                         map->m_seg_type);
1538                                 if (!err)
1539                                         set_inode_flag(inode, FI_APPEND_WRITE);
1540                         }
1541                         if (err)
1542                                 goto sync_out;
1543                         map->m_flags |= F2FS_MAP_NEW;
1544                         blkaddr = dn.data_blkaddr;
1545                 } else {
1546                         if (flag == F2FS_GET_BLOCK_BMAP) {
1547                                 map->m_pblk = 0;
1548                                 goto sync_out;
1549                         }
1550                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1551                                 goto sync_out;
1552                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1553                                                 blkaddr == NULL_ADDR) {
1554                                 if (map->m_next_pgofs)
1555                                         *map->m_next_pgofs = pgofs + 1;
1556                                 goto sync_out;
1557                         }
1558                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1559                                 /* for defragment case */
1560                                 if (map->m_next_pgofs)
1561                                         *map->m_next_pgofs = pgofs + 1;
1562                                 goto sync_out;
1563                         }
1564                 }
1565         }
1566
1567         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1568                 goto skip;
1569
1570         if (map->m_len == 0) {
1571                 /* preallocated unwritten block should be mapped for fiemap. */
1572                 if (blkaddr == NEW_ADDR)
1573                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1574                 map->m_flags |= F2FS_MAP_MAPPED;
1575
1576                 map->m_pblk = blkaddr;
1577                 map->m_len = 1;
1578         } else if ((map->m_pblk != NEW_ADDR &&
1579                         blkaddr == (map->m_pblk + ofs)) ||
1580                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1581                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1582                 ofs++;
1583                 map->m_len++;
1584         } else {
1585                 goto sync_out;
1586         }
1587
1588 skip:
1589         dn.ofs_in_node++;
1590         pgofs++;
1591
1592         /* preallocate blocks in batch for one dnode page */
1593         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1594                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1595
1596                 dn.ofs_in_node = ofs_in_node;
1597                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1598                 if (err)
1599                         goto sync_out;
1600
1601                 map->m_len += dn.ofs_in_node - ofs_in_node;
1602                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1603                         err = -ENOSPC;
1604                         goto sync_out;
1605                 }
1606                 dn.ofs_in_node = end_offset;
1607         }
1608
1609         if (pgofs >= end)
1610                 goto sync_out;
1611         else if (dn.ofs_in_node < end_offset)
1612                 goto next_block;
1613
1614         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1615                 if (map->m_flags & F2FS_MAP_MAPPED) {
1616                         unsigned int ofs = start_pgofs - map->m_lblk;
1617
1618                         f2fs_update_extent_cache_range(&dn,
1619                                 start_pgofs, map->m_pblk + ofs,
1620                                 map->m_len - ofs);
1621                 }
1622         }
1623
1624         f2fs_put_dnode(&dn);
1625
1626         if (map->m_may_create) {
1627                 f2fs_do_map_lock(sbi, flag, false);
1628                 f2fs_balance_fs(sbi, dn.node_changed);
1629         }
1630         goto next_dnode;
1631
1632 sync_out:
1633
1634         /* for hardware encryption, but to avoid potential issue in future */
1635         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1636                 f2fs_wait_on_block_writeback_range(inode,
1637                                                 map->m_pblk, map->m_len);
1638
1639         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1640                 if (map->m_flags & F2FS_MAP_MAPPED) {
1641                         unsigned int ofs = start_pgofs - map->m_lblk;
1642
1643                         f2fs_update_extent_cache_range(&dn,
1644                                 start_pgofs, map->m_pblk + ofs,
1645                                 map->m_len - ofs);
1646                 }
1647                 if (map->m_next_extent)
1648                         *map->m_next_extent = pgofs + 1;
1649         }
1650         f2fs_put_dnode(&dn);
1651 unlock_out:
1652         if (map->m_may_create) {
1653                 f2fs_do_map_lock(sbi, flag, false);
1654                 f2fs_balance_fs(sbi, dn.node_changed);
1655         }
1656 out:
1657         trace_f2fs_map_blocks(inode, map, err);
1658         return err;
1659 }
1660
1661 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1662 {
1663         struct f2fs_map_blocks map;
1664         block_t last_lblk;
1665         int err;
1666
1667         if (pos + len > i_size_read(inode))
1668                 return false;
1669
1670         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1671         map.m_next_pgofs = NULL;
1672         map.m_next_extent = NULL;
1673         map.m_seg_type = NO_CHECK_TYPE;
1674         map.m_may_create = false;
1675         last_lblk = F2FS_BLK_ALIGN(pos + len);
1676
1677         while (map.m_lblk < last_lblk) {
1678                 map.m_len = last_lblk - map.m_lblk;
1679                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1680                 if (err || map.m_len == 0)
1681                         return false;
1682                 map.m_lblk += map.m_len;
1683         }
1684         return true;
1685 }
1686
1687 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1688 {
1689         return (bytes >> inode->i_blkbits);
1690 }
1691
1692 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1693 {
1694         return (blks << inode->i_blkbits);
1695 }
1696
1697 static int __get_data_block(struct inode *inode, sector_t iblock,
1698                         struct buffer_head *bh, int create, int flag,
1699                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1700 {
1701         struct f2fs_map_blocks map;
1702         int err;
1703
1704         map.m_lblk = iblock;
1705         map.m_len = bytes_to_blks(inode, bh->b_size);
1706         map.m_next_pgofs = next_pgofs;
1707         map.m_next_extent = NULL;
1708         map.m_seg_type = seg_type;
1709         map.m_may_create = may_write;
1710
1711         err = f2fs_map_blocks(inode, &map, create, flag);
1712         if (!err) {
1713                 map_bh(bh, inode->i_sb, map.m_pblk);
1714                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1715                 bh->b_size = blks_to_bytes(inode, map.m_len);
1716         }
1717         return err;
1718 }
1719
1720 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1721                         struct buffer_head *bh_result, int create)
1722 {
1723         return __get_data_block(inode, iblock, bh_result, create,
1724                                 F2FS_GET_BLOCK_DIO, NULL,
1725                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1726                                 true);
1727 }
1728
1729 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1730                         struct buffer_head *bh_result, int create)
1731 {
1732         return __get_data_block(inode, iblock, bh_result, create,
1733                                 F2FS_GET_BLOCK_DIO, NULL,
1734                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1735                                 false);
1736 }
1737
1738 static int f2fs_xattr_fiemap(struct inode *inode,
1739                                 struct fiemap_extent_info *fieinfo)
1740 {
1741         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1742         struct page *page;
1743         struct node_info ni;
1744         __u64 phys = 0, len;
1745         __u32 flags;
1746         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1747         int err = 0;
1748
1749         if (f2fs_has_inline_xattr(inode)) {
1750                 int offset;
1751
1752                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1753                                                 inode->i_ino, false);
1754                 if (!page)
1755                         return -ENOMEM;
1756
1757                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1758                 if (err) {
1759                         f2fs_put_page(page, 1);
1760                         return err;
1761                 }
1762
1763                 phys = blks_to_bytes(inode, ni.blk_addr);
1764                 offset = offsetof(struct f2fs_inode, i_addr) +
1765                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1766                                         get_inline_xattr_addrs(inode));
1767
1768                 phys += offset;
1769                 len = inline_xattr_size(inode);
1770
1771                 f2fs_put_page(page, 1);
1772
1773                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1774
1775                 if (!xnid)
1776                         flags |= FIEMAP_EXTENT_LAST;
1777
1778                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1779                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1780                 if (err || err == 1)
1781                         return err;
1782         }
1783
1784         if (xnid) {
1785                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1786                 if (!page)
1787                         return -ENOMEM;
1788
1789                 err = f2fs_get_node_info(sbi, xnid, &ni);
1790                 if (err) {
1791                         f2fs_put_page(page, 1);
1792                         return err;
1793                 }
1794
1795                 phys = blks_to_bytes(inode, ni.blk_addr);
1796                 len = inode->i_sb->s_blocksize;
1797
1798                 f2fs_put_page(page, 1);
1799
1800                 flags = FIEMAP_EXTENT_LAST;
1801         }
1802
1803         if (phys) {
1804                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1805                 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1806         }
1807
1808         return (err < 0 ? err : 0);
1809 }
1810
1811 static loff_t max_inode_blocks(struct inode *inode)
1812 {
1813         loff_t result = ADDRS_PER_INODE(inode);
1814         loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1815
1816         /* two direct node blocks */
1817         result += (leaf_count * 2);
1818
1819         /* two indirect node blocks */
1820         leaf_count *= NIDS_PER_BLOCK;
1821         result += (leaf_count * 2);
1822
1823         /* one double indirect node block */
1824         leaf_count *= NIDS_PER_BLOCK;
1825         result += leaf_count;
1826
1827         return result;
1828 }
1829
1830 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1831                 u64 start, u64 len)
1832 {
1833         struct f2fs_map_blocks map;
1834         sector_t start_blk, last_blk;
1835         pgoff_t next_pgofs;
1836         u64 logical = 0, phys = 0, size = 0;
1837         u32 flags = 0;
1838         int ret = 0;
1839         bool compr_cluster = false;
1840         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1841         loff_t maxbytes;
1842
1843         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1844                 ret = f2fs_precache_extents(inode);
1845                 if (ret)
1846                         return ret;
1847         }
1848
1849         ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1850         if (ret)
1851                 return ret;
1852
1853         inode_lock(inode);
1854
1855         maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1856         if (start > maxbytes) {
1857                 ret = -EFBIG;
1858                 goto out;
1859         }
1860
1861         if (len > maxbytes || (maxbytes - len) < start)
1862                 len = maxbytes - start;
1863
1864         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1865                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1866                 goto out;
1867         }
1868
1869         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1870                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1871                 if (ret != -EAGAIN)
1872                         goto out;
1873         }
1874
1875         if (bytes_to_blks(inode, len) == 0)
1876                 len = blks_to_bytes(inode, 1);
1877
1878         start_blk = bytes_to_blks(inode, start);
1879         last_blk = bytes_to_blks(inode, start + len - 1);
1880
1881 next:
1882         memset(&map, 0, sizeof(map));
1883         map.m_lblk = start_blk;
1884         map.m_len = bytes_to_blks(inode, len);
1885         map.m_next_pgofs = &next_pgofs;
1886         map.m_seg_type = NO_CHECK_TYPE;
1887
1888         if (compr_cluster)
1889                 map.m_len = cluster_size - 1;
1890
1891         ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1892         if (ret)
1893                 goto out;
1894
1895         /* HOLE */
1896         if (!(map.m_flags & F2FS_MAP_FLAGS)) {
1897                 start_blk = next_pgofs;
1898
1899                 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1900                                                 max_inode_blocks(inode)))
1901                         goto prep_next;
1902
1903                 flags |= FIEMAP_EXTENT_LAST;
1904         }
1905
1906         if (size) {
1907                 flags |= FIEMAP_EXTENT_MERGED;
1908                 if (IS_ENCRYPTED(inode))
1909                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1910
1911                 ret = fiemap_fill_next_extent(fieinfo, logical,
1912                                 phys, size, flags);
1913                 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1914                 if (ret)
1915                         goto out;
1916                 size = 0;
1917         }
1918
1919         if (start_blk > last_blk)
1920                 goto out;
1921
1922         if (compr_cluster) {
1923                 compr_cluster = false;
1924
1925
1926                 logical = blks_to_bytes(inode, start_blk - 1);
1927                 phys = blks_to_bytes(inode, map.m_pblk);
1928                 size = blks_to_bytes(inode, cluster_size);
1929
1930                 flags |= FIEMAP_EXTENT_ENCODED;
1931
1932                 start_blk += cluster_size - 1;
1933
1934                 if (start_blk > last_blk)
1935                         goto out;
1936
1937                 goto prep_next;
1938         }
1939
1940         if (map.m_pblk == COMPRESS_ADDR) {
1941                 compr_cluster = true;
1942                 start_blk++;
1943                 goto prep_next;
1944         }
1945
1946         logical = blks_to_bytes(inode, start_blk);
1947         phys = blks_to_bytes(inode, map.m_pblk);
1948         size = blks_to_bytes(inode, map.m_len);
1949         flags = 0;
1950         if (map.m_flags & F2FS_MAP_UNWRITTEN)
1951                 flags = FIEMAP_EXTENT_UNWRITTEN;
1952
1953         start_blk += bytes_to_blks(inode, size);
1954
1955 prep_next:
1956         cond_resched();
1957         if (fatal_signal_pending(current))
1958                 ret = -EINTR;
1959         else
1960                 goto next;
1961 out:
1962         if (ret == 1)
1963                 ret = 0;
1964
1965         inode_unlock(inode);
1966         return ret;
1967 }
1968
1969 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1970 {
1971         if (IS_ENABLED(CONFIG_FS_VERITY) &&
1972             (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1973                 return inode->i_sb->s_maxbytes;
1974
1975         return i_size_read(inode);
1976 }
1977
1978 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1979                                         unsigned nr_pages,
1980                                         struct f2fs_map_blocks *map,
1981                                         struct bio **bio_ret,
1982                                         sector_t *last_block_in_bio,
1983                                         bool is_readahead)
1984 {
1985         struct bio *bio = *bio_ret;
1986         const unsigned blocksize = blks_to_bytes(inode, 1);
1987         sector_t block_in_file;
1988         sector_t last_block;
1989         sector_t last_block_in_file;
1990         sector_t block_nr;
1991         int ret = 0;
1992
1993         block_in_file = (sector_t)page_index(page);
1994         last_block = block_in_file + nr_pages;
1995         last_block_in_file = bytes_to_blks(inode,
1996                         f2fs_readpage_limit(inode) + blocksize - 1);
1997         if (last_block > last_block_in_file)
1998                 last_block = last_block_in_file;
1999
2000         /* just zeroing out page which is beyond EOF */
2001         if (block_in_file >= last_block)
2002                 goto zero_out;
2003         /*
2004          * Map blocks using the previous result first.
2005          */
2006         if ((map->m_flags & F2FS_MAP_MAPPED) &&
2007                         block_in_file > map->m_lblk &&
2008                         block_in_file < (map->m_lblk + map->m_len))
2009                 goto got_it;
2010
2011         /*
2012          * Then do more f2fs_map_blocks() calls until we are
2013          * done with this page.
2014          */
2015         map->m_lblk = block_in_file;
2016         map->m_len = last_block - block_in_file;
2017
2018         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2019         if (ret)
2020                 goto out;
2021 got_it:
2022         if ((map->m_flags & F2FS_MAP_MAPPED)) {
2023                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2024                 SetPageMappedToDisk(page);
2025
2026                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
2027                                         !cleancache_get_page(page))) {
2028                         SetPageUptodate(page);
2029                         goto confused;
2030                 }
2031
2032                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2033                                                 DATA_GENERIC_ENHANCE_READ)) {
2034                         ret = -EFSCORRUPTED;
2035                         goto out;
2036                 }
2037         } else {
2038 zero_out:
2039                 zero_user_segment(page, 0, PAGE_SIZE);
2040                 if (f2fs_need_verity(inode, page->index) &&
2041                     !fsverity_verify_page(page)) {
2042                         ret = -EIO;
2043                         goto out;
2044                 }
2045                 if (!PageUptodate(page))
2046                         SetPageUptodate(page);
2047                 unlock_page(page);
2048                 goto out;
2049         }
2050
2051         /*
2052          * This page will go to BIO.  Do we need to send this
2053          * BIO off first?
2054          */
2055         if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2056                                        *last_block_in_bio, block_nr) ||
2057                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2058 submit_and_realloc:
2059                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2060                 bio = NULL;
2061         }
2062         if (bio == NULL) {
2063                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2064                                 is_readahead ? REQ_RAHEAD : 0, page->index,
2065                                 false);
2066                 if (IS_ERR(bio)) {
2067                         ret = PTR_ERR(bio);
2068                         bio = NULL;
2069                         goto out;
2070                 }
2071         }
2072
2073         /*
2074          * If the page is under writeback, we need to wait for
2075          * its completion to see the correct decrypted data.
2076          */
2077         f2fs_wait_on_block_writeback(inode, block_nr);
2078
2079         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2080                 goto submit_and_realloc;
2081
2082         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2083         f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2084         ClearPageError(page);
2085         *last_block_in_bio = block_nr;
2086         goto out;
2087 confused:
2088         if (bio) {
2089                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2090                 bio = NULL;
2091         }
2092         unlock_page(page);
2093 out:
2094         *bio_ret = bio;
2095         return ret;
2096 }
2097
2098 #ifdef CONFIG_F2FS_FS_COMPRESSION
2099 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2100                                 unsigned nr_pages, sector_t *last_block_in_bio,
2101                                 bool is_readahead, bool for_write)
2102 {
2103         struct dnode_of_data dn;
2104         struct inode *inode = cc->inode;
2105         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2106         struct bio *bio = *bio_ret;
2107         unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2108         sector_t last_block_in_file;
2109         const unsigned blocksize = blks_to_bytes(inode, 1);
2110         struct decompress_io_ctx *dic = NULL;
2111         int i;
2112         int ret = 0;
2113
2114         f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2115
2116         last_block_in_file = bytes_to_blks(inode,
2117                         f2fs_readpage_limit(inode) + blocksize - 1);
2118
2119         /* get rid of pages beyond EOF */
2120         for (i = 0; i < cc->cluster_size; i++) {
2121                 struct page *page = cc->rpages[i];
2122
2123                 if (!page)
2124                         continue;
2125                 if ((sector_t)page->index >= last_block_in_file) {
2126                         zero_user_segment(page, 0, PAGE_SIZE);
2127                         if (!PageUptodate(page))
2128                                 SetPageUptodate(page);
2129                 } else if (!PageUptodate(page)) {
2130                         continue;
2131                 }
2132                 unlock_page(page);
2133                 cc->rpages[i] = NULL;
2134                 cc->nr_rpages--;
2135         }
2136
2137         /* we are done since all pages are beyond EOF */
2138         if (f2fs_cluster_is_empty(cc))
2139                 goto out;
2140
2141         set_new_dnode(&dn, inode, NULL, NULL, 0);
2142         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2143         if (ret)
2144                 goto out;
2145
2146         f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2147
2148         for (i = 1; i < cc->cluster_size; i++) {
2149                 block_t blkaddr;
2150
2151                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2152                                                 dn.ofs_in_node + i);
2153
2154                 if (!__is_valid_data_blkaddr(blkaddr))
2155                         break;
2156
2157                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2158                         ret = -EFAULT;
2159                         goto out_put_dnode;
2160                 }
2161                 cc->nr_cpages++;
2162         }
2163
2164         /* nothing to decompress */
2165         if (cc->nr_cpages == 0) {
2166                 ret = 0;
2167                 goto out_put_dnode;
2168         }
2169
2170         dic = f2fs_alloc_dic(cc);
2171         if (IS_ERR(dic)) {
2172                 ret = PTR_ERR(dic);
2173                 goto out_put_dnode;
2174         }
2175
2176         for (i = 0; i < dic->nr_cpages; i++) {
2177                 struct page *page = dic->cpages[i];
2178                 block_t blkaddr;
2179                 struct bio_post_read_ctx *ctx;
2180
2181                 blkaddr = data_blkaddr(dn.inode, dn.node_page,
2182                                                 dn.ofs_in_node + i + 1);
2183
2184                 if (bio && (!page_is_mergeable(sbi, bio,
2185                                         *last_block_in_bio, blkaddr) ||
2186                     !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2187 submit_and_realloc:
2188                         __submit_bio(sbi, bio, DATA);
2189                         bio = NULL;
2190                 }
2191
2192                 if (!bio) {
2193                         bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2194                                         is_readahead ? REQ_RAHEAD : 0,
2195                                         page->index, for_write);
2196                         if (IS_ERR(bio)) {
2197                                 ret = PTR_ERR(bio);
2198                                 f2fs_decompress_end_io(dic, ret);
2199                                 f2fs_put_dnode(&dn);
2200                                 *bio_ret = NULL;
2201                                 return ret;
2202                         }
2203                 }
2204
2205                 f2fs_wait_on_block_writeback(inode, blkaddr);
2206
2207                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2208                         goto submit_and_realloc;
2209
2210                 ctx = bio->bi_private;
2211                 ctx->enabled_steps |= STEP_DECOMPRESS;
2212                 refcount_inc(&dic->refcnt);
2213
2214                 inc_page_count(sbi, F2FS_RD_DATA);
2215                 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2216                 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2217                 ClearPageError(page);
2218                 *last_block_in_bio = blkaddr;
2219         }
2220
2221         f2fs_put_dnode(&dn);
2222
2223         *bio_ret = bio;
2224         return 0;
2225
2226 out_put_dnode:
2227         f2fs_put_dnode(&dn);
2228 out:
2229         for (i = 0; i < cc->cluster_size; i++) {
2230                 if (cc->rpages[i]) {
2231                         ClearPageUptodate(cc->rpages[i]);
2232                         ClearPageError(cc->rpages[i]);
2233                         unlock_page(cc->rpages[i]);
2234                 }
2235         }
2236         *bio_ret = bio;
2237         return ret;
2238 }
2239 #endif
2240
2241 /*
2242  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2243  * Major change was from block_size == page_size in f2fs by default.
2244  */
2245 static int f2fs_mpage_readpages(struct inode *inode,
2246                 struct readahead_control *rac, struct page *page)
2247 {
2248         struct bio *bio = NULL;
2249         sector_t last_block_in_bio = 0;
2250         struct f2fs_map_blocks map;
2251 #ifdef CONFIG_F2FS_FS_COMPRESSION
2252         struct compress_ctx cc = {
2253                 .inode = inode,
2254                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2255                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2256                 .cluster_idx = NULL_CLUSTER,
2257                 .rpages = NULL,
2258                 .cpages = NULL,
2259                 .nr_rpages = 0,
2260                 .nr_cpages = 0,
2261         };
2262 #endif
2263         unsigned nr_pages = rac ? readahead_count(rac) : 1;
2264         unsigned max_nr_pages = nr_pages;
2265         int ret = 0;
2266
2267         map.m_pblk = 0;
2268         map.m_lblk = 0;
2269         map.m_len = 0;
2270         map.m_flags = 0;
2271         map.m_next_pgofs = NULL;
2272         map.m_next_extent = NULL;
2273         map.m_seg_type = NO_CHECK_TYPE;
2274         map.m_may_create = false;
2275
2276         for (; nr_pages; nr_pages--) {
2277                 if (rac) {
2278                         page = readahead_page(rac);
2279                         prefetchw(&page->flags);
2280                 }
2281
2282 #ifdef CONFIG_F2FS_FS_COMPRESSION
2283                 if (f2fs_compressed_file(inode)) {
2284                         /* there are remained comressed pages, submit them */
2285                         if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2286                                 ret = f2fs_read_multi_pages(&cc, &bio,
2287                                                         max_nr_pages,
2288                                                         &last_block_in_bio,
2289                                                         rac != NULL, false);
2290                                 f2fs_destroy_compress_ctx(&cc);
2291                                 if (ret)
2292                                         goto set_error_page;
2293                         }
2294                         ret = f2fs_is_compressed_cluster(inode, page->index);
2295                         if (ret < 0)
2296                                 goto set_error_page;
2297                         else if (!ret)
2298                                 goto read_single_page;
2299
2300                         ret = f2fs_init_compress_ctx(&cc);
2301                         if (ret)
2302                                 goto set_error_page;
2303
2304                         f2fs_compress_ctx_add_page(&cc, page);
2305
2306                         goto next_page;
2307                 }
2308 read_single_page:
2309 #endif
2310
2311                 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2312                                         &bio, &last_block_in_bio, rac);
2313                 if (ret) {
2314 #ifdef CONFIG_F2FS_FS_COMPRESSION
2315 set_error_page:
2316 #endif
2317                         SetPageError(page);
2318                         zero_user_segment(page, 0, PAGE_SIZE);
2319                         unlock_page(page);
2320                 }
2321 #ifdef CONFIG_F2FS_FS_COMPRESSION
2322 next_page:
2323 #endif
2324                 if (rac)
2325                         put_page(page);
2326
2327 #ifdef CONFIG_F2FS_FS_COMPRESSION
2328                 if (f2fs_compressed_file(inode)) {
2329                         /* last page */
2330                         if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2331                                 ret = f2fs_read_multi_pages(&cc, &bio,
2332                                                         max_nr_pages,
2333                                                         &last_block_in_bio,
2334                                                         rac != NULL, false);
2335                                 f2fs_destroy_compress_ctx(&cc);
2336                         }
2337                 }
2338 #endif
2339         }
2340         if (bio)
2341                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2342         return ret;
2343 }
2344
2345 static int f2fs_read_data_page(struct file *file, struct page *page)
2346 {
2347         struct inode *inode = page_file_mapping(page)->host;
2348         int ret = -EAGAIN;
2349
2350         trace_f2fs_readpage(page, DATA);
2351
2352         if (!f2fs_is_compress_backend_ready(inode)) {
2353                 unlock_page(page);
2354                 return -EOPNOTSUPP;
2355         }
2356
2357         /* If the file has inline data, try to read it directly */
2358         if (f2fs_has_inline_data(inode))
2359                 ret = f2fs_read_inline_data(inode, page);
2360         if (ret == -EAGAIN)
2361                 ret = f2fs_mpage_readpages(inode, NULL, page);
2362         return ret;
2363 }
2364
2365 static void f2fs_readahead(struct readahead_control *rac)
2366 {
2367         struct inode *inode = rac->mapping->host;
2368
2369         trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2370
2371         if (!f2fs_is_compress_backend_ready(inode))
2372                 return;
2373
2374         /* If the file has inline data, skip readpages */
2375         if (f2fs_has_inline_data(inode))
2376                 return;
2377
2378         f2fs_mpage_readpages(inode, rac, NULL);
2379 }
2380
2381 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2382 {
2383         struct inode *inode = fio->page->mapping->host;
2384         struct page *mpage, *page;
2385         gfp_t gfp_flags = GFP_NOFS;
2386
2387         if (!f2fs_encrypted_file(inode))
2388                 return 0;
2389
2390         page = fio->compressed_page ? fio->compressed_page : fio->page;
2391
2392         /* wait for GCed page writeback via META_MAPPING */
2393         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2394
2395         if (fscrypt_inode_uses_inline_crypto(inode))
2396                 return 0;
2397
2398 retry_encrypt:
2399         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2400                                         PAGE_SIZE, 0, gfp_flags);
2401         if (IS_ERR(fio->encrypted_page)) {
2402                 /* flush pending IOs and wait for a while in the ENOMEM case */
2403                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2404                         f2fs_flush_merged_writes(fio->sbi);
2405                         congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2406                         gfp_flags |= __GFP_NOFAIL;
2407                         goto retry_encrypt;
2408                 }
2409                 return PTR_ERR(fio->encrypted_page);
2410         }
2411
2412         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2413         if (mpage) {
2414                 if (PageUptodate(mpage))
2415                         memcpy(page_address(mpage),
2416                                 page_address(fio->encrypted_page), PAGE_SIZE);
2417                 f2fs_put_page(mpage, 1);
2418         }
2419         return 0;
2420 }
2421
2422 static inline bool check_inplace_update_policy(struct inode *inode,
2423                                 struct f2fs_io_info *fio)
2424 {
2425         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2426         unsigned int policy = SM_I(sbi)->ipu_policy;
2427
2428         if (policy & (0x1 << F2FS_IPU_FORCE))
2429                 return true;
2430         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2431                 return true;
2432         if (policy & (0x1 << F2FS_IPU_UTIL) &&
2433                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2434                 return true;
2435         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2436                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
2437                 return true;
2438
2439         /*
2440          * IPU for rewrite async pages
2441          */
2442         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2443                         fio && fio->op == REQ_OP_WRITE &&
2444                         !(fio->op_flags & REQ_SYNC) &&
2445                         !IS_ENCRYPTED(inode))
2446                 return true;
2447
2448         /* this is only set during fdatasync */
2449         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2450                         is_inode_flag_set(inode, FI_NEED_IPU))
2451                 return true;
2452
2453         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2454                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2455                 return true;
2456
2457         return false;
2458 }
2459
2460 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2461 {
2462         if (f2fs_is_pinned_file(inode))
2463                 return true;
2464
2465         /* if this is cold file, we should overwrite to avoid fragmentation */
2466         if (file_is_cold(inode))
2467                 return true;
2468
2469         return check_inplace_update_policy(inode, fio);
2470 }
2471
2472 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2473 {
2474         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2475
2476         if (f2fs_lfs_mode(sbi))
2477                 return true;
2478         if (S_ISDIR(inode->i_mode))
2479                 return true;
2480         if (IS_NOQUOTA(inode))
2481                 return true;
2482         if (f2fs_is_atomic_file(inode))
2483                 return true;
2484         if (fio) {
2485                 if (is_cold_data(fio->page))
2486                         return true;
2487                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
2488                         return true;
2489                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2490                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2491                         return true;
2492         }
2493         return false;
2494 }
2495
2496 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2497 {
2498         struct inode *inode = fio->page->mapping->host;
2499
2500         if (f2fs_should_update_outplace(inode, fio))
2501                 return false;
2502
2503         return f2fs_should_update_inplace(inode, fio);
2504 }
2505
2506 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2507 {
2508         struct page *page = fio->page;
2509         struct inode *inode = page->mapping->host;
2510         struct dnode_of_data dn;
2511         struct extent_info ei = {0,0,0};
2512         struct node_info ni;
2513         bool ipu_force = false;
2514         int err = 0;
2515
2516         set_new_dnode(&dn, inode, NULL, NULL, 0);
2517         if (need_inplace_update(fio) &&
2518                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2519                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2520
2521                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2522                                                 DATA_GENERIC_ENHANCE))
2523                         return -EFSCORRUPTED;
2524
2525                 ipu_force = true;
2526                 fio->need_lock = LOCK_DONE;
2527                 goto got_it;
2528         }
2529
2530         /* Deadlock due to between page->lock and f2fs_lock_op */
2531         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2532                 return -EAGAIN;
2533
2534         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2535         if (err)
2536                 goto out;
2537
2538         fio->old_blkaddr = dn.data_blkaddr;
2539
2540         /* This page is already truncated */
2541         if (fio->old_blkaddr == NULL_ADDR) {
2542                 ClearPageUptodate(page);
2543                 clear_cold_data(page);
2544                 goto out_writepage;
2545         }
2546 got_it:
2547         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2548                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2549                                                 DATA_GENERIC_ENHANCE)) {
2550                 err = -EFSCORRUPTED;
2551                 goto out_writepage;
2552         }
2553         /*
2554          * If current allocation needs SSR,
2555          * it had better in-place writes for updated data.
2556          */
2557         if (ipu_force ||
2558                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2559                                         need_inplace_update(fio))) {
2560                 err = f2fs_encrypt_one_page(fio);
2561                 if (err)
2562                         goto out_writepage;
2563
2564                 set_page_writeback(page);
2565                 ClearPageError(page);
2566                 f2fs_put_dnode(&dn);
2567                 if (fio->need_lock == LOCK_REQ)
2568                         f2fs_unlock_op(fio->sbi);
2569                 err = f2fs_inplace_write_data(fio);
2570                 if (err) {
2571                         if (fscrypt_inode_uses_fs_layer_crypto(inode))
2572                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2573                         if (PageWriteback(page))
2574                                 end_page_writeback(page);
2575                 } else {
2576                         set_inode_flag(inode, FI_UPDATE_WRITE);
2577                 }
2578                 trace_f2fs_do_write_data_page(fio->page, IPU);
2579                 return err;
2580         }
2581
2582         if (fio->need_lock == LOCK_RETRY) {
2583                 if (!f2fs_trylock_op(fio->sbi)) {
2584                         err = -EAGAIN;
2585                         goto out_writepage;
2586                 }
2587                 fio->need_lock = LOCK_REQ;
2588         }
2589
2590         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2591         if (err)
2592                 goto out_writepage;
2593
2594         fio->version = ni.version;
2595
2596         err = f2fs_encrypt_one_page(fio);
2597         if (err)
2598                 goto out_writepage;
2599
2600         set_page_writeback(page);
2601         ClearPageError(page);
2602
2603         if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2604                 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2605
2606         /* LFS mode write path */
2607         f2fs_outplace_write_data(&dn, fio);
2608         trace_f2fs_do_write_data_page(page, OPU);
2609         set_inode_flag(inode, FI_APPEND_WRITE);
2610         if (page->index == 0)
2611                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2612 out_writepage:
2613         f2fs_put_dnode(&dn);
2614 out:
2615         if (fio->need_lock == LOCK_REQ)
2616                 f2fs_unlock_op(fio->sbi);
2617         return err;
2618 }
2619
2620 int f2fs_write_single_data_page(struct page *page, int *submitted,
2621                                 struct bio **bio,
2622                                 sector_t *last_block,
2623                                 struct writeback_control *wbc,
2624                                 enum iostat_type io_type,
2625                                 int compr_blocks,
2626                                 bool allow_balance)
2627 {
2628         struct inode *inode = page->mapping->host;
2629         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2630         loff_t i_size = i_size_read(inode);
2631         const pgoff_t end_index = ((unsigned long long)i_size)
2632                                                         >> PAGE_SHIFT;
2633         loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2634         unsigned offset = 0;
2635         bool need_balance_fs = false;
2636         int err = 0;
2637         struct f2fs_io_info fio = {
2638                 .sbi = sbi,
2639                 .ino = inode->i_ino,
2640                 .type = DATA,
2641                 .op = REQ_OP_WRITE,
2642                 .op_flags = wbc_to_write_flags(wbc),
2643                 .old_blkaddr = NULL_ADDR,
2644                 .page = page,
2645                 .encrypted_page = NULL,
2646                 .submitted = false,
2647                 .compr_blocks = compr_blocks,
2648                 .need_lock = LOCK_RETRY,
2649                 .io_type = io_type,
2650                 .io_wbc = wbc,
2651                 .bio = bio,
2652                 .last_block = last_block,
2653         };
2654
2655         trace_f2fs_writepage(page, DATA);
2656
2657         /* we should bypass data pages to proceed the kworkder jobs */
2658         if (unlikely(f2fs_cp_error(sbi))) {
2659                 mapping_set_error(page->mapping, -EIO);
2660                 /*
2661                  * don't drop any dirty dentry pages for keeping lastest
2662                  * directory structure.
2663                  */
2664                 if (S_ISDIR(inode->i_mode))
2665                         goto redirty_out;
2666                 goto out;
2667         }
2668
2669         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2670                 goto redirty_out;
2671
2672         if (page->index < end_index ||
2673                         f2fs_verity_in_progress(inode) ||
2674                         compr_blocks)
2675                 goto write;
2676
2677         /*
2678          * If the offset is out-of-range of file size,
2679          * this page does not have to be written to disk.
2680          */
2681         offset = i_size & (PAGE_SIZE - 1);
2682         if ((page->index >= end_index + 1) || !offset)
2683                 goto out;
2684
2685         zero_user_segment(page, offset, PAGE_SIZE);
2686 write:
2687         if (f2fs_is_drop_cache(inode))
2688                 goto out;
2689         /* we should not write 0'th page having journal header */
2690         if (f2fs_is_volatile_file(inode) && (!page->index ||
2691                         (!wbc->for_reclaim &&
2692                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2693                 goto redirty_out;
2694
2695         /* Dentry/quota blocks are controlled by checkpoint */
2696         if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2697                 /*
2698                  * We need to wait for node_write to avoid block allocation during
2699                  * checkpoint. This can only happen to quota writes which can cause
2700                  * the below discard race condition.
2701                  */
2702                 if (IS_NOQUOTA(inode))
2703                         down_read(&sbi->node_write);
2704
2705                 fio.need_lock = LOCK_DONE;
2706                 err = f2fs_do_write_data_page(&fio);
2707
2708                 if (IS_NOQUOTA(inode))
2709                         up_read(&sbi->node_write);
2710
2711                 goto done;
2712         }
2713
2714         if (!wbc->for_reclaim)
2715                 need_balance_fs = true;
2716         else if (has_not_enough_free_secs(sbi, 0, 0))
2717                 goto redirty_out;
2718         else
2719                 set_inode_flag(inode, FI_HOT_DATA);
2720
2721         err = -EAGAIN;
2722         if (f2fs_has_inline_data(inode)) {
2723                 err = f2fs_write_inline_data(inode, page);
2724                 if (!err)
2725                         goto out;
2726         }
2727
2728         if (err == -EAGAIN) {
2729                 err = f2fs_do_write_data_page(&fio);
2730                 if (err == -EAGAIN) {
2731                         fio.need_lock = LOCK_REQ;
2732                         err = f2fs_do_write_data_page(&fio);
2733                 }
2734         }
2735
2736         if (err) {
2737                 file_set_keep_isize(inode);
2738         } else {
2739                 spin_lock(&F2FS_I(inode)->i_size_lock);
2740                 if (F2FS_I(inode)->last_disk_size < psize)
2741                         F2FS_I(inode)->last_disk_size = psize;
2742                 spin_unlock(&F2FS_I(inode)->i_size_lock);
2743         }
2744
2745 done:
2746         if (err && err != -ENOENT)
2747                 goto redirty_out;
2748
2749 out:
2750         inode_dec_dirty_pages(inode);
2751         if (err) {
2752                 ClearPageUptodate(page);
2753                 clear_cold_data(page);
2754         }
2755
2756         if (wbc->for_reclaim) {
2757                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2758                 clear_inode_flag(inode, FI_HOT_DATA);
2759                 f2fs_remove_dirty_inode(inode);
2760                 submitted = NULL;
2761         }
2762         unlock_page(page);
2763         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2764                         !F2FS_I(inode)->cp_task && allow_balance)
2765                 f2fs_balance_fs(sbi, need_balance_fs);
2766
2767         if (unlikely(f2fs_cp_error(sbi))) {
2768                 f2fs_submit_merged_write(sbi, DATA);
2769                 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2770                 submitted = NULL;
2771         }
2772
2773         if (submitted)
2774                 *submitted = fio.submitted ? 1 : 0;
2775
2776         return 0;
2777
2778 redirty_out:
2779         redirty_page_for_writepage(wbc, page);
2780         /*
2781          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2782          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2783          * file_write_and_wait_range() will see EIO error, which is critical
2784          * to return value of fsync() followed by atomic_write failure to user.
2785          */
2786         if (!err || wbc->for_reclaim)
2787                 return AOP_WRITEPAGE_ACTIVATE;
2788         unlock_page(page);
2789         return err;
2790 }
2791
2792 static int f2fs_write_data_page(struct page *page,
2793                                         struct writeback_control *wbc)
2794 {
2795 #ifdef CONFIG_F2FS_FS_COMPRESSION
2796         struct inode *inode = page->mapping->host;
2797
2798         if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2799                 goto out;
2800
2801         if (f2fs_compressed_file(inode)) {
2802                 if (f2fs_is_compressed_cluster(inode, page->index)) {
2803                         redirty_page_for_writepage(wbc, page);
2804                         return AOP_WRITEPAGE_ACTIVATE;
2805                 }
2806         }
2807 out:
2808 #endif
2809
2810         return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2811                                                 wbc, FS_DATA_IO, 0, true);
2812 }
2813
2814 /*
2815  * This function was copied from write_cche_pages from mm/page-writeback.c.
2816  * The major change is making write step of cold data page separately from
2817  * warm/hot data page.
2818  */
2819 static int f2fs_write_cache_pages(struct address_space *mapping,
2820                                         struct writeback_control *wbc,
2821                                         enum iostat_type io_type)
2822 {
2823         int ret = 0;
2824         int done = 0, retry = 0;
2825         struct pagevec pvec;
2826         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2827         struct bio *bio = NULL;
2828         sector_t last_block;
2829 #ifdef CONFIG_F2FS_FS_COMPRESSION
2830         struct inode *inode = mapping->host;
2831         struct compress_ctx cc = {
2832                 .inode = inode,
2833                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2834                 .cluster_size = F2FS_I(inode)->i_cluster_size,
2835                 .cluster_idx = NULL_CLUSTER,
2836                 .rpages = NULL,
2837                 .nr_rpages = 0,
2838                 .cpages = NULL,
2839                 .rbuf = NULL,
2840                 .cbuf = NULL,
2841                 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2842                 .private = NULL,
2843         };
2844 #endif
2845         int nr_pages;
2846         pgoff_t index;
2847         pgoff_t end;            /* Inclusive */
2848         pgoff_t done_index;
2849         int range_whole = 0;
2850         xa_mark_t tag;
2851         int nwritten = 0;
2852         int submitted = 0;
2853         int i;
2854
2855         pagevec_init(&pvec);
2856
2857         if (get_dirty_pages(mapping->host) <=
2858                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2859                 set_inode_flag(mapping->host, FI_HOT_DATA);
2860         else
2861                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2862
2863         if (wbc->range_cyclic) {
2864                 index = mapping->writeback_index; /* prev offset */
2865                 end = -1;
2866         } else {
2867                 index = wbc->range_start >> PAGE_SHIFT;
2868                 end = wbc->range_end >> PAGE_SHIFT;
2869                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2870                         range_whole = 1;
2871         }
2872         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2873                 tag = PAGECACHE_TAG_TOWRITE;
2874         else
2875                 tag = PAGECACHE_TAG_DIRTY;
2876 retry:
2877         retry = 0;
2878         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2879                 tag_pages_for_writeback(mapping, index, end);
2880         done_index = index;
2881         while (!done && !retry && (index <= end)) {
2882                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2883                                 tag);
2884                 if (nr_pages == 0)
2885                         break;
2886
2887                 for (i = 0; i < nr_pages; i++) {
2888                         struct page *page = pvec.pages[i];
2889                         bool need_readd;
2890 readd:
2891                         need_readd = false;
2892 #ifdef CONFIG_F2FS_FS_COMPRESSION
2893                         if (f2fs_compressed_file(inode)) {
2894                                 ret = f2fs_init_compress_ctx(&cc);
2895                                 if (ret) {
2896                                         done = 1;
2897                                         break;
2898                                 }
2899
2900                                 if (!f2fs_cluster_can_merge_page(&cc,
2901                                                                 page->index)) {
2902                                         ret = f2fs_write_multi_pages(&cc,
2903                                                 &submitted, wbc, io_type);
2904                                         if (!ret)
2905                                                 need_readd = true;
2906                                         goto result;
2907                                 }
2908
2909                                 if (unlikely(f2fs_cp_error(sbi)))
2910                                         goto lock_page;
2911
2912                                 if (f2fs_cluster_is_empty(&cc)) {
2913                                         void *fsdata = NULL;
2914                                         struct page *pagep;
2915                                         int ret2;
2916
2917                                         ret2 = f2fs_prepare_compress_overwrite(
2918                                                         inode, &pagep,
2919                                                         page->index, &fsdata);
2920                                         if (ret2 < 0) {
2921                                                 ret = ret2;
2922                                                 done = 1;
2923                                                 break;
2924                                         } else if (ret2 &&
2925                                                 !f2fs_compress_write_end(inode,
2926                                                                 fsdata, page->index,
2927                                                                 1)) {
2928                                                 retry = 1;
2929                                                 break;
2930                                         }
2931                                 } else {
2932                                         goto lock_page;
2933                                 }
2934                         }
2935 #endif
2936                         /* give a priority to WB_SYNC threads */
2937                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2938                                         wbc->sync_mode == WB_SYNC_NONE) {
2939                                 done = 1;
2940                                 break;
2941                         }
2942 #ifdef CONFIG_F2FS_FS_COMPRESSION
2943 lock_page:
2944 #endif
2945                         done_index = page->index;
2946 retry_write:
2947                         lock_page(page);
2948
2949                         if (unlikely(page->mapping != mapping)) {
2950 continue_unlock:
2951                                 unlock_page(page);
2952                                 continue;
2953                         }
2954
2955                         if (!PageDirty(page)) {
2956                                 /* someone wrote it for us */
2957                                 goto continue_unlock;
2958                         }
2959
2960                         if (PageWriteback(page)) {
2961                                 if (wbc->sync_mode != WB_SYNC_NONE)
2962                                         f2fs_wait_on_page_writeback(page,
2963                                                         DATA, true, true);
2964                                 else
2965                                         goto continue_unlock;
2966                         }
2967
2968                         if (!clear_page_dirty_for_io(page))
2969                                 goto continue_unlock;
2970
2971 #ifdef CONFIG_F2FS_FS_COMPRESSION
2972                         if (f2fs_compressed_file(inode)) {
2973                                 get_page(page);
2974                                 f2fs_compress_ctx_add_page(&cc, page);
2975                                 continue;
2976                         }
2977 #endif
2978                         ret = f2fs_write_single_data_page(page, &submitted,
2979                                         &bio, &last_block, wbc, io_type,
2980                                         0, true);
2981                         if (ret == AOP_WRITEPAGE_ACTIVATE)
2982                                 unlock_page(page);
2983 #ifdef CONFIG_F2FS_FS_COMPRESSION
2984 result:
2985 #endif
2986                         nwritten += submitted;
2987                         wbc->nr_to_write -= submitted;
2988
2989                         if (unlikely(ret)) {
2990                                 /*
2991                                  * keep nr_to_write, since vfs uses this to
2992                                  * get # of written pages.
2993                                  */
2994                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2995                                         ret = 0;
2996                                         goto next;
2997                                 } else if (ret == -EAGAIN) {
2998                                         ret = 0;
2999                                         if (wbc->sync_mode == WB_SYNC_ALL) {
3000                                                 cond_resched();
3001                                                 congestion_wait(BLK_RW_ASYNC,
3002                                                         DEFAULT_IO_TIMEOUT);
3003                                                 goto retry_write;
3004                                         }
3005                                         goto next;
3006                                 }
3007                                 done_index = page->index + 1;
3008                                 done = 1;
3009                                 break;
3010                         }
3011
3012                         if (wbc->nr_to_write <= 0 &&
3013                                         wbc->sync_mode == WB_SYNC_NONE) {
3014                                 done = 1;
3015                                 break;
3016                         }
3017 next:
3018                         if (need_readd)
3019                                 goto readd;
3020                 }
3021                 pagevec_release(&pvec);
3022                 cond_resched();
3023         }
3024 #ifdef CONFIG_F2FS_FS_COMPRESSION
3025         /* flush remained pages in compress cluster */
3026         if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3027                 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3028                 nwritten += submitted;
3029                 wbc->nr_to_write -= submitted;
3030                 if (ret) {
3031                         done = 1;
3032                         retry = 0;
3033                 }
3034         }
3035         if (f2fs_compressed_file(inode))
3036                 f2fs_destroy_compress_ctx(&cc);
3037 #endif
3038         if (retry) {
3039                 index = 0;
3040                 end = -1;
3041                 goto retry;
3042         }
3043         if (wbc->range_cyclic && !done)
3044                 done_index = 0;
3045         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3046                 mapping->writeback_index = done_index;
3047
3048         if (nwritten)
3049                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3050                                                                 NULL, 0, DATA);
3051         /* submit cached bio of IPU write */
3052         if (bio)
3053                 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3054
3055         return ret;
3056 }
3057
3058 static inline bool __should_serialize_io(struct inode *inode,
3059                                         struct writeback_control *wbc)
3060 {
3061         /* to avoid deadlock in path of data flush */
3062         if (F2FS_I(inode)->cp_task)
3063                 return false;
3064
3065         if (!S_ISREG(inode->i_mode))
3066                 return false;
3067         if (IS_NOQUOTA(inode))
3068                 return false;
3069
3070         if (f2fs_need_compress_data(inode))
3071                 return true;
3072         if (wbc->sync_mode != WB_SYNC_ALL)
3073                 return true;
3074         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3075                 return true;
3076         return false;
3077 }
3078
3079 static int __f2fs_write_data_pages(struct address_space *mapping,
3080                                                 struct writeback_control *wbc,
3081                                                 enum iostat_type io_type)
3082 {
3083         struct inode *inode = mapping->host;
3084         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3085         struct blk_plug plug;
3086         int ret;
3087         bool locked = false;
3088
3089         /* deal with chardevs and other special file */
3090         if (!mapping->a_ops->writepage)
3091                 return 0;
3092
3093         /* skip writing if there is no dirty page in this inode */
3094         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3095                 return 0;
3096
3097         /* during POR, we don't need to trigger writepage at all. */
3098         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3099                 goto skip_write;
3100
3101         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3102                         wbc->sync_mode == WB_SYNC_NONE &&
3103                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3104                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
3105                 goto skip_write;
3106
3107         /* skip writing during file defragment */
3108         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3109                 goto skip_write;
3110
3111         trace_f2fs_writepages(mapping->host, wbc, DATA);
3112
3113         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3114         if (wbc->sync_mode == WB_SYNC_ALL)
3115                 atomic_inc(&sbi->wb_sync_req[DATA]);
3116         else if (atomic_read(&sbi->wb_sync_req[DATA]))
3117                 goto skip_write;
3118
3119         if (__should_serialize_io(inode, wbc)) {
3120                 mutex_lock(&sbi->writepages);
3121                 locked = true;
3122         }
3123
3124         blk_start_plug(&plug);
3125         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3126         blk_finish_plug(&plug);
3127
3128         if (locked)
3129                 mutex_unlock(&sbi->writepages);
3130
3131         if (wbc->sync_mode == WB_SYNC_ALL)
3132                 atomic_dec(&sbi->wb_sync_req[DATA]);
3133         /*
3134          * if some pages were truncated, we cannot guarantee its mapping->host
3135          * to detect pending bios.
3136          */
3137
3138         f2fs_remove_dirty_inode(inode);
3139         return ret;
3140
3141 skip_write:
3142         wbc->pages_skipped += get_dirty_pages(inode);
3143         trace_f2fs_writepages(mapping->host, wbc, DATA);
3144         return 0;
3145 }
3146
3147 static int f2fs_write_data_pages(struct address_space *mapping,
3148                             struct writeback_control *wbc)
3149 {
3150         struct inode *inode = mapping->host;
3151
3152         return __f2fs_write_data_pages(mapping, wbc,
3153                         F2FS_I(inode)->cp_task == current ?
3154                         FS_CP_DATA_IO : FS_DATA_IO);
3155 }
3156
3157 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
3158 {
3159         struct inode *inode = mapping->host;
3160         loff_t i_size = i_size_read(inode);
3161
3162         if (IS_NOQUOTA(inode))
3163                 return;
3164
3165         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3166         if (to > i_size && !f2fs_verity_in_progress(inode)) {
3167                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3168                 down_write(&F2FS_I(inode)->i_mmap_sem);
3169
3170                 truncate_pagecache(inode, i_size);
3171                 f2fs_truncate_blocks(inode, i_size, true);
3172
3173                 up_write(&F2FS_I(inode)->i_mmap_sem);
3174                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3175         }
3176 }
3177
3178 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3179                         struct page *page, loff_t pos, unsigned len,
3180                         block_t *blk_addr, bool *node_changed)
3181 {
3182         struct inode *inode = page->mapping->host;
3183         pgoff_t index = page->index;
3184         struct dnode_of_data dn;
3185         struct page *ipage;
3186         bool locked = false;
3187         struct extent_info ei = {0,0,0};
3188         int err = 0;
3189         int flag;
3190
3191         /*
3192          * we already allocated all the blocks, so we don't need to get
3193          * the block addresses when there is no need to fill the page.
3194          */
3195         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3196             !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3197             !f2fs_verity_in_progress(inode))
3198                 return 0;
3199
3200         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3201         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3202                 flag = F2FS_GET_BLOCK_DEFAULT;
3203         else
3204                 flag = F2FS_GET_BLOCK_PRE_AIO;
3205
3206         if (f2fs_has_inline_data(inode) ||
3207                         (pos & PAGE_MASK) >= i_size_read(inode)) {
3208                 f2fs_do_map_lock(sbi, flag, true);
3209                 locked = true;
3210         }
3211
3212 restart:
3213         /* check inline_data */
3214         ipage = f2fs_get_node_page(sbi, inode->i_ino);
3215         if (IS_ERR(ipage)) {
3216                 err = PTR_ERR(ipage);
3217                 goto unlock_out;
3218         }
3219
3220         set_new_dnode(&dn, inode, ipage, ipage, 0);
3221
3222         if (f2fs_has_inline_data(inode)) {
3223                 if (pos + len <= MAX_INLINE_DATA(inode)) {
3224                         f2fs_do_read_inline_data(page, ipage);
3225                         set_inode_flag(inode, FI_DATA_EXIST);
3226                         if (inode->i_nlink)
3227                                 set_inline_node(ipage);
3228                 } else {
3229                         err = f2fs_convert_inline_page(&dn, page);
3230                         if (err)
3231                                 goto out;
3232                         if (dn.data_blkaddr == NULL_ADDR)
3233                                 err = f2fs_get_block(&dn, index);
3234                 }
3235         } else if (locked) {
3236                 err = f2fs_get_block(&dn, index);
3237         } else {
3238                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3239                         dn.data_blkaddr = ei.blk + index - ei.fofs;
3240                 } else {
3241                         /* hole case */
3242                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3243                         if (err || dn.data_blkaddr == NULL_ADDR) {
3244                                 f2fs_put_dnode(&dn);
3245                                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3246                                                                 true);
3247                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3248                                 locked = true;
3249                                 goto restart;
3250                         }
3251                 }
3252         }
3253
3254         /* convert_inline_page can make node_changed */
3255         *blk_addr = dn.data_blkaddr;
3256         *node_changed = dn.node_changed;
3257 out:
3258         f2fs_put_dnode(&dn);
3259 unlock_out:
3260         if (locked)
3261                 f2fs_do_map_lock(sbi, flag, false);
3262         return err;
3263 }
3264
3265 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3266                 loff_t pos, unsigned len, unsigned flags,
3267                 struct page **pagep, void **fsdata)
3268 {
3269         struct inode *inode = mapping->host;
3270         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3271         struct page *page = NULL;
3272         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3273         bool need_balance = false, drop_atomic = false;
3274         block_t blkaddr = NULL_ADDR;
3275         int err = 0;
3276
3277         trace_f2fs_write_begin(inode, pos, len, flags);
3278
3279         if (!f2fs_is_checkpoint_ready(sbi)) {
3280                 err = -ENOSPC;
3281                 goto fail;
3282         }
3283
3284         if ((f2fs_is_atomic_file(inode) &&
3285                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3286                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3287                 err = -ENOMEM;
3288                 drop_atomic = true;
3289                 goto fail;
3290         }
3291
3292         /*
3293          * We should check this at this moment to avoid deadlock on inode page
3294          * and #0 page. The locking rule for inline_data conversion should be:
3295          * lock_page(page #0) -> lock_page(inode_page)
3296          */
3297         if (index != 0) {
3298                 err = f2fs_convert_inline_inode(inode);
3299                 if (err)
3300                         goto fail;
3301         }
3302
3303 #ifdef CONFIG_F2FS_FS_COMPRESSION
3304         if (f2fs_compressed_file(inode)) {
3305                 int ret;
3306
3307                 *fsdata = NULL;
3308
3309                 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3310                                                         index, fsdata);
3311                 if (ret < 0) {
3312                         err = ret;
3313                         goto fail;
3314                 } else if (ret) {
3315                         return 0;
3316                 }
3317         }
3318 #endif
3319
3320 repeat:
3321         /*
3322          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3323          * wait_for_stable_page. Will wait that below with our IO control.
3324          */
3325         page = f2fs_pagecache_get_page(mapping, index,
3326                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3327         if (!page) {
3328                 err = -ENOMEM;
3329                 goto fail;
3330         }
3331
3332         /* TODO: cluster can be compressed due to race with .writepage */
3333
3334         *pagep = page;
3335
3336         err = prepare_write_begin(sbi, page, pos, len,
3337                                         &blkaddr, &need_balance);
3338         if (err)
3339                 goto fail;
3340
3341         if (need_balance && !IS_NOQUOTA(inode) &&
3342                         has_not_enough_free_secs(sbi, 0, 0)) {
3343                 unlock_page(page);
3344                 f2fs_balance_fs(sbi, true);
3345                 lock_page(page);
3346                 if (page->mapping != mapping) {
3347                         /* The page got truncated from under us */
3348                         f2fs_put_page(page, 1);
3349                         goto repeat;
3350                 }
3351         }
3352
3353         f2fs_wait_on_page_writeback(page, DATA, false, true);
3354
3355         if (len == PAGE_SIZE || PageUptodate(page))
3356                 return 0;
3357
3358         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3359             !f2fs_verity_in_progress(inode)) {
3360                 zero_user_segment(page, len, PAGE_SIZE);
3361                 return 0;
3362         }
3363
3364         if (blkaddr == NEW_ADDR) {
3365                 zero_user_segment(page, 0, PAGE_SIZE);
3366                 SetPageUptodate(page);
3367         } else {
3368                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3369                                 DATA_GENERIC_ENHANCE_READ)) {
3370                         err = -EFSCORRUPTED;
3371                         goto fail;
3372                 }
3373                 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3374                 if (err)
3375                         goto fail;
3376
3377                 lock_page(page);
3378                 if (unlikely(page->mapping != mapping)) {
3379                         f2fs_put_page(page, 1);
3380                         goto repeat;
3381                 }
3382                 if (unlikely(!PageUptodate(page))) {
3383                         err = -EIO;
3384                         goto fail;
3385                 }
3386         }
3387         return 0;
3388
3389 fail:
3390         f2fs_put_page(page, 1);
3391         f2fs_write_failed(mapping, pos + len);
3392         if (drop_atomic)
3393                 f2fs_drop_inmem_pages_all(sbi, false);
3394         return err;
3395 }
3396
3397 static int f2fs_write_end(struct file *file,
3398                         struct address_space *mapping,
3399                         loff_t pos, unsigned len, unsigned copied,
3400                         struct page *page, void *fsdata)
3401 {
3402         struct inode *inode = page->mapping->host;
3403
3404         trace_f2fs_write_end(inode, pos, len, copied);
3405
3406         /*
3407          * This should be come from len == PAGE_SIZE, and we expect copied
3408          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3409          * let generic_perform_write() try to copy data again through copied=0.
3410          */
3411         if (!PageUptodate(page)) {
3412                 if (unlikely(copied != len))
3413                         copied = 0;
3414                 else
3415                         SetPageUptodate(page);
3416         }
3417
3418 #ifdef CONFIG_F2FS_FS_COMPRESSION
3419         /* overwrite compressed file */
3420         if (f2fs_compressed_file(inode) && fsdata) {
3421                 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3422                 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3423
3424                 if (pos + copied > i_size_read(inode) &&
3425                                 !f2fs_verity_in_progress(inode))
3426                         f2fs_i_size_write(inode, pos + copied);
3427                 return copied;
3428         }
3429 #endif
3430
3431         if (!copied)
3432                 goto unlock_out;
3433
3434         set_page_dirty(page);
3435
3436         if (pos + copied > i_size_read(inode) &&
3437             !f2fs_verity_in_progress(inode))
3438                 f2fs_i_size_write(inode, pos + copied);
3439 unlock_out:
3440         f2fs_put_page(page, 1);
3441         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3442         return copied;
3443 }
3444
3445 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3446                            loff_t offset)
3447 {
3448         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3449         unsigned blkbits = i_blkbits;
3450         unsigned blocksize_mask = (1 << blkbits) - 1;
3451         unsigned long align = offset | iov_iter_alignment(iter);
3452         struct block_device *bdev = inode->i_sb->s_bdev;
3453
3454         if (iov_iter_rw(iter) == READ && offset >= i_size_read(inode))
3455                 return 1;
3456
3457         if (align & blocksize_mask) {
3458                 if (bdev)
3459                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
3460                 blocksize_mask = (1 << blkbits) - 1;
3461                 if (align & blocksize_mask)
3462                         return -EINVAL;
3463                 return 1;
3464         }
3465         return 0;
3466 }
3467
3468 static void f2fs_dio_end_io(struct bio *bio)
3469 {
3470         struct f2fs_private_dio *dio = bio->bi_private;
3471
3472         dec_page_count(F2FS_I_SB(dio->inode),
3473                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3474
3475         bio->bi_private = dio->orig_private;
3476         bio->bi_end_io = dio->orig_end_io;
3477
3478         kfree(dio);
3479
3480         bio_endio(bio);
3481 }
3482
3483 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3484                                                         loff_t file_offset)
3485 {
3486         struct f2fs_private_dio *dio;
3487         bool write = (bio_op(bio) == REQ_OP_WRITE);
3488
3489         dio = f2fs_kzalloc(F2FS_I_SB(inode),
3490                         sizeof(struct f2fs_private_dio), GFP_NOFS);
3491         if (!dio)
3492                 goto out;
3493
3494         dio->inode = inode;
3495         dio->orig_end_io = bio->bi_end_io;
3496         dio->orig_private = bio->bi_private;
3497         dio->write = write;
3498
3499         bio->bi_end_io = f2fs_dio_end_io;
3500         bio->bi_private = dio;
3501
3502         inc_page_count(F2FS_I_SB(inode),
3503                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3504
3505         submit_bio(bio);
3506         return;
3507 out:
3508         bio->bi_status = BLK_STS_IOERR;
3509         bio_endio(bio);
3510 }
3511
3512 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3513 {
3514         struct address_space *mapping = iocb->ki_filp->f_mapping;
3515         struct inode *inode = mapping->host;
3516         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3517         struct f2fs_inode_info *fi = F2FS_I(inode);
3518         size_t count = iov_iter_count(iter);
3519         loff_t offset = iocb->ki_pos;
3520         int rw = iov_iter_rw(iter);
3521         int err;
3522         enum rw_hint hint = iocb->ki_hint;
3523         int whint_mode = F2FS_OPTION(sbi).whint_mode;
3524         bool do_opu;
3525
3526         err = check_direct_IO(inode, iter, offset);
3527         if (err)
3528                 return err < 0 ? err : 0;
3529
3530         if (f2fs_force_buffered_io(inode, iocb, iter))
3531                 return 0;
3532
3533         do_opu = allow_outplace_dio(inode, iocb, iter);
3534
3535         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3536
3537         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3538                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
3539
3540         if (iocb->ki_flags & IOCB_NOWAIT) {
3541                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3542                         iocb->ki_hint = hint;
3543                         err = -EAGAIN;
3544                         goto out;
3545                 }
3546                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3547                         up_read(&fi->i_gc_rwsem[rw]);
3548                         iocb->ki_hint = hint;
3549                         err = -EAGAIN;
3550                         goto out;
3551                 }
3552         } else {
3553                 down_read(&fi->i_gc_rwsem[rw]);
3554                 if (do_opu)
3555                         down_read(&fi->i_gc_rwsem[READ]);
3556         }
3557
3558         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3559                         iter, rw == WRITE ? get_data_block_dio_write :
3560                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
3561                         rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3562                         DIO_SKIP_HOLES);
3563
3564         if (do_opu)
3565                 up_read(&fi->i_gc_rwsem[READ]);
3566
3567         up_read(&fi->i_gc_rwsem[rw]);
3568
3569         if (rw == WRITE) {
3570                 if (whint_mode == WHINT_MODE_OFF)
3571                         iocb->ki_hint = hint;
3572                 if (err > 0) {
3573                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3574                                                                         err);
3575                         if (!do_opu)
3576                                 set_inode_flag(inode, FI_UPDATE_WRITE);
3577                 } else if (err == -EIOCBQUEUED) {
3578                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3579                                                 count - iov_iter_count(iter));
3580                 } else if (err < 0) {
3581                         f2fs_write_failed(mapping, offset + count);
3582                 }
3583         } else {
3584                 if (err > 0)
3585                         f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3586                 else if (err == -EIOCBQUEUED)
3587                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_READ_IO,
3588                                                 count - iov_iter_count(iter));
3589         }
3590
3591 out:
3592         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3593
3594         return err;
3595 }
3596
3597 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3598                                                         unsigned int length)
3599 {
3600         struct inode *inode = page->mapping->host;
3601         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3602
3603         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3604                 (offset % PAGE_SIZE || length != PAGE_SIZE))
3605                 return;
3606
3607         if (PageDirty(page)) {
3608                 if (inode->i_ino == F2FS_META_INO(sbi)) {
3609                         dec_page_count(sbi, F2FS_DIRTY_META);
3610                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3611                         dec_page_count(sbi, F2FS_DIRTY_NODES);
3612                 } else {
3613                         inode_dec_dirty_pages(inode);
3614                         f2fs_remove_dirty_inode(inode);
3615                 }
3616         }
3617
3618         clear_cold_data(page);
3619
3620         if (IS_ATOMIC_WRITTEN_PAGE(page))
3621                 return f2fs_drop_inmem_page(inode, page);
3622
3623         f2fs_clear_page_private(page);
3624 }
3625
3626 int f2fs_release_page(struct page *page, gfp_t wait)
3627 {
3628         /* If this is dirty page, keep PagePrivate */
3629         if (PageDirty(page))
3630                 return 0;
3631
3632         /* This is atomic written page, keep Private */
3633         if (IS_ATOMIC_WRITTEN_PAGE(page))
3634                 return 0;
3635
3636         clear_cold_data(page);
3637         f2fs_clear_page_private(page);
3638         return 1;
3639 }
3640
3641 static int f2fs_set_data_page_dirty(struct page *page)
3642 {
3643         struct inode *inode = page_file_mapping(page)->host;
3644
3645         trace_f2fs_set_page_dirty(page, DATA);
3646
3647         if (!PageUptodate(page))
3648                 SetPageUptodate(page);
3649         if (PageSwapCache(page))
3650                 return __set_page_dirty_nobuffers(page);
3651
3652         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3653                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
3654                         f2fs_register_inmem_page(inode, page);
3655                         return 1;
3656                 }
3657                 /*
3658                  * Previously, this page has been registered, we just
3659                  * return here.
3660                  */
3661                 return 0;
3662         }
3663
3664         if (!PageDirty(page)) {
3665                 __set_page_dirty_nobuffers(page);
3666                 f2fs_update_dirty_page(inode, page);
3667                 return 1;
3668         }
3669         return 0;
3670 }
3671
3672
3673 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3674 {
3675 #ifdef CONFIG_F2FS_FS_COMPRESSION
3676         struct dnode_of_data dn;
3677         sector_t start_idx, blknr = 0;
3678         int ret;
3679
3680         start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3681
3682         set_new_dnode(&dn, inode, NULL, NULL, 0);
3683         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3684         if (ret)
3685                 return 0;
3686
3687         if (dn.data_blkaddr != COMPRESS_ADDR) {
3688                 dn.ofs_in_node += block - start_idx;
3689                 blknr = f2fs_data_blkaddr(&dn);
3690                 if (!__is_valid_data_blkaddr(blknr))
3691                         blknr = 0;
3692         }
3693
3694         f2fs_put_dnode(&dn);
3695         return blknr;
3696 #else
3697         return 0;
3698 #endif
3699 }
3700
3701
3702 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3703 {
3704         struct inode *inode = mapping->host;
3705         sector_t blknr = 0;
3706
3707         if (f2fs_has_inline_data(inode))
3708                 goto out;
3709
3710         /* make sure allocating whole blocks */
3711         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3712                 filemap_write_and_wait(mapping);
3713
3714         /* Block number less than F2FS MAX BLOCKS */
3715         if (unlikely(block >= max_file_blocks(inode)))
3716                 goto out;
3717
3718         if (f2fs_compressed_file(inode)) {
3719                 blknr = f2fs_bmap_compress(inode, block);
3720         } else {
3721                 struct f2fs_map_blocks map;
3722
3723                 memset(&map, 0, sizeof(map));
3724                 map.m_lblk = block;
3725                 map.m_len = 1;
3726                 map.m_next_pgofs = NULL;
3727                 map.m_seg_type = NO_CHECK_TYPE;
3728
3729                 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3730                         blknr = map.m_pblk;
3731         }
3732 out:
3733         trace_f2fs_bmap(inode, block, blknr);
3734         return blknr;
3735 }
3736
3737 #ifdef CONFIG_MIGRATION
3738 #include <linux/migrate.h>
3739
3740 int f2fs_migrate_page(struct address_space *mapping,
3741                 struct page *newpage, struct page *page, enum migrate_mode mode)
3742 {
3743         int rc, extra_count;
3744         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3745         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
3746
3747         BUG_ON(PageWriteback(page));
3748
3749         /* migrating an atomic written page is safe with the inmem_lock hold */
3750         if (atomic_written) {
3751                 if (mode != MIGRATE_SYNC)
3752                         return -EBUSY;
3753                 if (!mutex_trylock(&fi->inmem_lock))
3754                         return -EAGAIN;
3755         }
3756
3757         /* one extra reference was held for atomic_write page */
3758         extra_count = atomic_written ? 1 : 0;
3759         rc = migrate_page_move_mapping(mapping, newpage,
3760                                 page, extra_count);
3761         if (rc != MIGRATEPAGE_SUCCESS) {
3762                 if (atomic_written)
3763                         mutex_unlock(&fi->inmem_lock);
3764                 return rc;
3765         }
3766
3767         if (atomic_written) {
3768                 struct inmem_pages *cur;
3769
3770                 list_for_each_entry(cur, &fi->inmem_pages, list)
3771                         if (cur->page == page) {
3772                                 cur->page = newpage;
3773                                 break;
3774                         }
3775                 mutex_unlock(&fi->inmem_lock);
3776                 put_page(page);
3777                 get_page(newpage);
3778         }
3779
3780         if (PagePrivate(page)) {
3781                 f2fs_set_page_private(newpage, page_private(page));
3782                 f2fs_clear_page_private(page);
3783         }
3784
3785         if (mode != MIGRATE_SYNC_NO_COPY)
3786                 migrate_page_copy(newpage, page);
3787         else
3788                 migrate_page_states(newpage, page);
3789
3790         return MIGRATEPAGE_SUCCESS;
3791 }
3792 #endif
3793
3794 #ifdef CONFIG_SWAP
3795 static int f2fs_is_file_aligned(struct inode *inode)
3796 {
3797         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3798         block_t main_blkaddr = SM_I(sbi)->main_blkaddr;
3799         block_t cur_lblock;
3800         block_t last_lblock;
3801         block_t pblock;
3802         unsigned long nr_pblocks;
3803         unsigned int blocks_per_sec = BLKS_PER_SEC(sbi);
3804         int ret = 0;
3805
3806         cur_lblock = 0;
3807         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3808
3809         while (cur_lblock < last_lblock) {
3810                 struct f2fs_map_blocks map;
3811
3812                 memset(&map, 0, sizeof(map));
3813                 map.m_lblk = cur_lblock;
3814                 map.m_len = last_lblock - cur_lblock;
3815                 map.m_next_pgofs = NULL;
3816                 map.m_next_extent = NULL;
3817                 map.m_seg_type = NO_CHECK_TYPE;
3818                 map.m_may_create = false;
3819
3820                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3821                 if (ret)
3822                         goto out;
3823
3824                 /* hole */
3825                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3826                         f2fs_err(sbi, "Swapfile has holes\n");
3827                         ret = -ENOENT;
3828                         goto out;
3829                 }
3830
3831                 pblock = map.m_pblk;
3832                 nr_pblocks = map.m_len;
3833
3834                 if ((pblock - main_blkaddr) & (blocks_per_sec - 1) ||
3835                         nr_pblocks & (blocks_per_sec - 1)) {
3836                         f2fs_err(sbi, "Swapfile does not align to section");
3837                         ret = -EINVAL;
3838                         goto out;
3839                 }
3840
3841                 cur_lblock += nr_pblocks;
3842         }
3843 out:
3844         return ret;
3845 }
3846
3847 static int check_swap_activate_fast(struct swap_info_struct *sis,
3848                                 struct file *swap_file, sector_t *span)
3849 {
3850         struct address_space *mapping = swap_file->f_mapping;
3851         struct inode *inode = mapping->host;
3852         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3853         sector_t cur_lblock;
3854         sector_t last_lblock;
3855         sector_t pblock;
3856         sector_t lowest_pblock = -1;
3857         sector_t highest_pblock = 0;
3858         int nr_extents = 0;
3859         unsigned long nr_pblocks;
3860         unsigned int blocks_per_sec = BLKS_PER_SEC(sbi);
3861         int ret = 0;
3862
3863         /*
3864          * Map all the blocks into the extent list.  This code doesn't try
3865          * to be very smart.
3866          */
3867         cur_lblock = 0;
3868         last_lblock = bytes_to_blks(inode, i_size_read(inode));
3869
3870         while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3871                 struct f2fs_map_blocks map;
3872
3873                 cond_resched();
3874
3875                 memset(&map, 0, sizeof(map));
3876                 map.m_lblk = cur_lblock;
3877                 map.m_len = last_lblock - cur_lblock;
3878                 map.m_next_pgofs = NULL;
3879                 map.m_next_extent = NULL;
3880                 map.m_seg_type = NO_CHECK_TYPE;
3881                 map.m_may_create = false;
3882
3883                 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3884                 if (ret)
3885                         goto out;
3886
3887                 /* hole */
3888                 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3889                         f2fs_err(sbi, "Swapfile has holes\n");
3890                         ret = -ENOENT;
3891                         goto out;
3892                 }
3893
3894                 pblock = map.m_pblk;
3895                 nr_pblocks = map.m_len;
3896
3897                 if ((pblock - SM_I(sbi)->main_blkaddr) & (blocks_per_sec - 1) ||
3898                                 nr_pblocks & (blocks_per_sec - 1)) {
3899                         f2fs_err(sbi, "Swapfile does not align to section");
3900                         ret = -EINVAL;
3901                         goto out;
3902                 }
3903
3904                 if (cur_lblock + nr_pblocks >= sis->max)
3905                         nr_pblocks = sis->max - cur_lblock;
3906
3907                 if (cur_lblock) {       /* exclude the header page */
3908                         if (pblock < lowest_pblock)
3909                                 lowest_pblock = pblock;
3910                         if (pblock + nr_pblocks - 1 > highest_pblock)
3911                                 highest_pblock = pblock + nr_pblocks - 1;
3912                 }
3913
3914                 /*
3915                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3916                  */
3917                 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3918                 if (ret < 0)
3919                         goto out;
3920                 nr_extents += ret;
3921                 cur_lblock += nr_pblocks;
3922         }
3923         ret = nr_extents;
3924         *span = 1 + highest_pblock - lowest_pblock;
3925         if (cur_lblock == 0)
3926                 cur_lblock = 1; /* force Empty message */
3927         sis->max = cur_lblock;
3928         sis->pages = cur_lblock - 1;
3929         sis->highest_bit = cur_lblock - 1;
3930 out:
3931         return ret;
3932 }
3933
3934 /* Copied from generic_swapfile_activate() to check any holes */
3935 static int check_swap_activate(struct swap_info_struct *sis,
3936                                 struct file *swap_file, sector_t *span)
3937 {
3938         struct address_space *mapping = swap_file->f_mapping;
3939         struct inode *inode = mapping->host;
3940         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3941         unsigned blocks_per_page;
3942         unsigned long page_no;
3943         sector_t probe_block;
3944         sector_t last_block;
3945         sector_t lowest_block = -1;
3946         sector_t highest_block = 0;
3947         int nr_extents = 0;
3948         int ret = 0;
3949
3950         if (PAGE_SIZE == F2FS_BLKSIZE)
3951                 return check_swap_activate_fast(sis, swap_file, span);
3952
3953         ret = f2fs_is_file_aligned(inode);
3954         if (ret)
3955                 goto out;
3956
3957         blocks_per_page = bytes_to_blks(inode, PAGE_SIZE);
3958
3959         /*
3960          * Map all the blocks into the extent list.  This code doesn't try
3961          * to be very smart.
3962          */
3963         probe_block = 0;
3964         page_no = 0;
3965         last_block = bytes_to_blks(inode, i_size_read(inode));
3966         while ((probe_block + blocks_per_page) <= last_block &&
3967                         page_no < sis->max) {
3968                 unsigned block_in_page;
3969                 sector_t first_block;
3970                 sector_t block = 0;
3971
3972                 cond_resched();
3973
3974                 block = probe_block;
3975                 ret = bmap(inode, &block);
3976                 if (ret)
3977                         goto out;
3978                 if (!block)
3979                         goto bad_bmap;
3980                 first_block = block;
3981
3982                 /*
3983                  * It must be PAGE_SIZE aligned on-disk
3984                  */
3985                 if (first_block & (blocks_per_page - 1)) {
3986                         probe_block++;
3987                         goto reprobe;
3988                 }
3989
3990                 for (block_in_page = 1; block_in_page < blocks_per_page;
3991                                         block_in_page++) {
3992
3993                         block = probe_block + block_in_page;
3994                         ret = bmap(inode, &block);
3995                         if (ret)
3996                                 goto out;
3997                         if (!block)
3998                                 goto bad_bmap;
3999
4000                         if (block != first_block + block_in_page) {
4001                                 /* Discontiguity */
4002                                 probe_block++;
4003                                 goto reprobe;
4004                         }
4005                 }
4006
4007                 first_block >>= (PAGE_SHIFT - inode->i_blkbits);
4008                 if (page_no) {  /* exclude the header page */
4009                         if (first_block < lowest_block)
4010                                 lowest_block = first_block;
4011                         if (first_block > highest_block)
4012                                 highest_block = first_block;
4013                 }
4014
4015                 /*
4016                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4017                  */
4018                 ret = add_swap_extent(sis, page_no, 1, first_block);
4019                 if (ret < 0)
4020                         goto out;
4021                 nr_extents += ret;
4022                 page_no++;
4023                 probe_block += blocks_per_page;
4024 reprobe:
4025                 continue;
4026         }
4027         ret = nr_extents;
4028         *span = 1 + highest_block - lowest_block;
4029         if (page_no == 0)
4030                 page_no = 1;    /* force Empty message */
4031         sis->max = page_no;
4032         sis->pages = page_no - 1;
4033         sis->highest_bit = page_no - 1;
4034 out:
4035         return ret;
4036 bad_bmap:
4037         f2fs_err(sbi, "Swapfile has holes\n");
4038         return -ENOENT;
4039 }
4040
4041 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4042                                 sector_t *span)
4043 {
4044         struct inode *inode = file_inode(file);
4045         int ret;
4046
4047         if (!S_ISREG(inode->i_mode))
4048                 return -EINVAL;
4049
4050         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4051                 return -EROFS;
4052
4053         ret = f2fs_convert_inline_inode(inode);
4054         if (ret)
4055                 return ret;
4056
4057         if (!f2fs_disable_compressed_file(inode))
4058                 return -EINVAL;
4059
4060         f2fs_precache_extents(inode);
4061
4062         ret = check_swap_activate(sis, file, span);
4063         if (ret < 0)
4064                 return ret;
4065
4066         set_inode_flag(inode, FI_PIN_FILE);
4067         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4068         return ret;
4069 }
4070
4071 static void f2fs_swap_deactivate(struct file *file)
4072 {
4073         struct inode *inode = file_inode(file);
4074
4075         clear_inode_flag(inode, FI_PIN_FILE);
4076 }
4077 #else
4078 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4079                                 sector_t *span)
4080 {
4081         return -EOPNOTSUPP;
4082 }
4083
4084 static void f2fs_swap_deactivate(struct file *file)
4085 {
4086 }
4087 #endif
4088
4089 const struct address_space_operations f2fs_dblock_aops = {
4090         .readpage       = f2fs_read_data_page,
4091         .readahead      = f2fs_readahead,
4092         .writepage      = f2fs_write_data_page,
4093         .writepages     = f2fs_write_data_pages,
4094         .write_begin    = f2fs_write_begin,
4095         .write_end      = f2fs_write_end,
4096         .set_page_dirty = f2fs_set_data_page_dirty,
4097         .invalidatepage = f2fs_invalidate_page,
4098         .releasepage    = f2fs_release_page,
4099         .direct_IO      = f2fs_direct_IO,
4100         .bmap           = f2fs_bmap,
4101         .swap_activate  = f2fs_swap_activate,
4102         .swap_deactivate = f2fs_swap_deactivate,
4103 #ifdef CONFIG_MIGRATION
4104         .migratepage    = f2fs_migrate_page,
4105 #endif
4106 };
4107
4108 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4109 {
4110         struct address_space *mapping = page_mapping(page);
4111         unsigned long flags;
4112
4113         xa_lock_irqsave(&mapping->i_pages, flags);
4114         __xa_clear_mark(&mapping->i_pages, page_index(page),
4115                                                 PAGECACHE_TAG_DIRTY);
4116         xa_unlock_irqrestore(&mapping->i_pages, flags);
4117 }
4118
4119 int __init f2fs_init_post_read_processing(void)
4120 {
4121         bio_post_read_ctx_cache =
4122                 kmem_cache_create("f2fs_bio_post_read_ctx",
4123                                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4124         if (!bio_post_read_ctx_cache)
4125                 goto fail;
4126         bio_post_read_ctx_pool =
4127                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4128                                          bio_post_read_ctx_cache);
4129         if (!bio_post_read_ctx_pool)
4130                 goto fail_free_cache;
4131         return 0;
4132
4133 fail_free_cache:
4134         kmem_cache_destroy(bio_post_read_ctx_cache);
4135 fail:
4136         return -ENOMEM;
4137 }
4138
4139 void f2fs_destroy_post_read_processing(void)
4140 {
4141         mempool_destroy(bio_post_read_ctx_pool);
4142         kmem_cache_destroy(bio_post_read_ctx_cache);
4143 }
4144
4145 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4146 {
4147         if (!f2fs_sb_has_encrypt(sbi) &&
4148                 !f2fs_sb_has_verity(sbi) &&
4149                 !f2fs_sb_has_compression(sbi))
4150                 return 0;
4151
4152         sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4153                                                  WQ_UNBOUND | WQ_HIGHPRI,
4154                                                  num_online_cpus());
4155         if (!sbi->post_read_wq)
4156                 return -ENOMEM;
4157         return 0;
4158 }
4159
4160 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4161 {
4162         if (sbi->post_read_wq)
4163                 destroy_workqueue(sbi->post_read_wq);
4164 }
4165
4166 int __init f2fs_init_bio_entry_cache(void)
4167 {
4168         bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4169                         sizeof(struct bio_entry));
4170         if (!bio_entry_slab)
4171                 return -ENOMEM;
4172         return 0;
4173 }
4174
4175 void f2fs_destroy_bio_entry_cache(void)
4176 {
4177         kmem_cache_destroy(bio_entry_slab);
4178 }