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/prefetch.h>
  18 #include <linux/uio.h>
  19 #include <linux/cleancache.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include "trace.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define NUM_PREALLOC_POST_READ_CTXS     128
  29
  30 static struct kmem_cache *bio_post_read_ctx_cache;
  31 static mempool_t *bio_post_read_ctx_pool;
  32
  33 static bool __is_cp_guaranteed(struct page *page)
  34 {
  35         struct address_space *mapping = page->mapping;
  36         struct inode *inode;
  37         struct f2fs_sb_info *sbi;
  38
  39         if (!mapping)
  40                 return false;
  41
  42         inode = mapping->host;
  43         sbi = F2FS_I_SB(inode);
  44
  45         if (inode->i_ino == F2FS_META_INO(sbi) ||
  46                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47                         S_ISDIR(inode->i_mode) ||
  48                         (S_ISREG(inode->i_mode) &&
  49                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  50                         is_cold_data(page))
  51                 return true;
  52         return false;
  53 }
  54
  55 static enum count_type __read_io_type(struct page *page)
  56 {
  57         struct address_space *mapping = page->mapping;
  58
  59         if (mapping) {
  60                 struct inode *inode = mapping->host;
  61                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  62
  63                 if (inode->i_ino == F2FS_META_INO(sbi))
  64                         return F2FS_RD_META;
  65
  66                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  67                         return F2FS_RD_NODE;
  68         }
  69         return F2FS_RD_DATA;
  70 }
  71
  72 /* postprocessing steps for read bios */
  73 enum bio_post_read_step {
  74         STEP_INITIAL = 0,
  75         STEP_DECRYPT,
  76 };
  77
  78 struct bio_post_read_ctx {
  79         struct bio *bio;
  80         struct work_struct work;
  81         unsigned int cur_step;
  82         unsigned int enabled_steps;
  83 };
  84
  85 static void __read_end_io(struct bio *bio)
  86 {
  87         struct page *page;
  88         struct bio_vec *bv;
  89         struct bvec_iter_all iter_all;
  90
  91         bio_for_each_segment_all(bv, bio, iter_all) {
  92                 page = bv->bv_page;
  93
  94                 /* PG_error was set if any post_read step failed */
  95                 if (bio->bi_status || PageError(page)) {
  96                         ClearPageUptodate(page);
  97                         /* will re-read again later */
  98                         ClearPageError(page);
  99                 } else {
 100                         SetPageUptodate(page);
 101                 }
 102                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 103                 unlock_page(page);
 104         }
 105         if (bio->bi_private)
 106                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 107         bio_put(bio);
 108 }
 109
 110 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 111
 112 static void decrypt_work(struct work_struct *work)
 113 {
 114         struct bio_post_read_ctx *ctx =
 115                 container_of(work, struct bio_post_read_ctx, work);
 116
 117         fscrypt_decrypt_bio(ctx->bio);
 118
 119         bio_post_read_processing(ctx);
 120 }
 121
 122 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 123 {
 124         switch (++ctx->cur_step) {
 125         case STEP_DECRYPT:
 126                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 127                         INIT_WORK(&ctx->work, decrypt_work);
 128                         fscrypt_enqueue_decrypt_work(&ctx->work);
 129                         return;
 130                 }
 131                 ctx->cur_step++;
 132                 /* fall-through */
 133         default:
 134                 __read_end_io(ctx->bio);
 135         }
 136 }
 137
 138 static bool f2fs_bio_post_read_required(struct bio *bio)
 139 {
 140         return bio->bi_private && !bio->bi_status;
 141 }
 142
 143 static void f2fs_read_end_io(struct bio *bio)
 144 {
 145         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
 146                                                 FAULT_READ_IO)) {
 147                 f2fs_show_injection_info(FAULT_READ_IO);
 148                 bio->bi_status = BLK_STS_IOERR;
 149         }
 150
 151         if (f2fs_bio_post_read_required(bio)) {
 152                 struct bio_post_read_ctx *ctx = bio->bi_private;
 153
 154                 ctx->cur_step = STEP_INITIAL;
 155                 bio_post_read_processing(ctx);
 156                 return;
 157         }
 158
 159         __read_end_io(bio);
 160 }
 161
 162 static void f2fs_write_end_io(struct bio *bio)
 163 {
 164         struct f2fs_sb_info *sbi = bio->bi_private;
 165         struct bio_vec *bvec;
 166         struct bvec_iter_all iter_all;
 167
 168         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 169                 f2fs_show_injection_info(FAULT_WRITE_IO);
 170                 bio->bi_status = BLK_STS_IOERR;
 171         }
 172
 173         bio_for_each_segment_all(bvec, bio, iter_all) {
 174                 struct page *page = bvec->bv_page;
 175                 enum count_type type = WB_DATA_TYPE(page);
 176
 177                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 178                         set_page_private(page, (unsigned long)NULL);
 179                         ClearPagePrivate(page);
 180                         unlock_page(page);
 181                         mempool_free(page, sbi->write_io_dummy);
 182
 183                         if (unlikely(bio->bi_status))
 184                                 f2fs_stop_checkpoint(sbi, true);
 185                         continue;
 186                 }
 187
 188                 fscrypt_finalize_bounce_page(&page);
 189
 190                 if (unlikely(bio->bi_status)) {
 191                         mapping_set_error(page->mapping, -EIO);
 192                         if (type == F2FS_WB_CP_DATA)
 193                                 f2fs_stop_checkpoint(sbi, true);
 194                 }
 195
 196                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 197                                         page->index != nid_of_node(page));
 198
 199                 dec_page_count(sbi, type);
 200                 if (f2fs_in_warm_node_list(sbi, page))
 201                         f2fs_del_fsync_node_entry(sbi, page);
 202                 clear_cold_data(page);
 203                 end_page_writeback(page);
 204         }
 205         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 206                                 wq_has_sleeper(&sbi->cp_wait))
 207                 wake_up(&sbi->cp_wait);
 208
 209         bio_put(bio);
 210 }
 211
 212 /*
 213  * Return true, if pre_bio's bdev is same as its target device.
 214  */
 215 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 216                                 block_t blk_addr, struct bio *bio)
 217 {
 218         struct block_device *bdev = sbi->sb->s_bdev;
 219         int i;
 220
 221         if (f2fs_is_multi_device(sbi)) {
 222                 for (i = 0; i < sbi->s_ndevs; i++) {
 223                         if (FDEV(i).start_blk <= blk_addr &&
 224                             FDEV(i).end_blk >= blk_addr) {
 225                                 blk_addr -= FDEV(i).start_blk;
 226                                 bdev = FDEV(i).bdev;
 227                                 break;
 228                         }
 229                 }
 230         }
 231         if (bio) {
 232                 bio_set_dev(bio, bdev);
 233                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 234         }
 235         return bdev;
 236 }
 237
 238 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 239 {
 240         int i;
 241
 242         if (!f2fs_is_multi_device(sbi))
 243                 return 0;
 244
 245         for (i = 0; i < sbi->s_ndevs; i++)
 246                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 247                         return i;
 248         return 0;
 249 }
 250
 251 static bool __same_bdev(struct f2fs_sb_info *sbi,
 252                                 block_t blk_addr, struct bio *bio)
 253 {
 254         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 255         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 256 }
 257
 258 /*
 259  * Low-level block read/write IO operations.
 260  */
 261 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 262                                 struct writeback_control *wbc,
 263                                 int npages, bool is_read,
 264                                 enum page_type type, enum temp_type temp)
 265 {
 266         struct bio *bio;
 267
 268         bio = f2fs_bio_alloc(sbi, npages, true);
 269
 270         f2fs_target_device(sbi, blk_addr, bio);
 271         if (is_read) {
 272                 bio->bi_end_io = f2fs_read_end_io;
 273                 bio->bi_private = NULL;
 274         } else {
 275                 bio->bi_end_io = f2fs_write_end_io;
 276                 bio->bi_private = sbi;
 277                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 278         }
 279         if (wbc)
 280                 wbc_init_bio(wbc, bio);
 281
 282         return bio;
 283 }
 284
 285 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 286                                 struct bio *bio, enum page_type type)
 287 {
 288         if (!is_read_io(bio_op(bio))) {
 289                 unsigned int start;
 290
 291                 if (type != DATA && type != NODE)
 292                         goto submit_io;
 293
 294                 if (test_opt(sbi, LFS) && current->plug)
 295                         blk_finish_plug(current->plug);
 296
 297                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 298                 start %= F2FS_IO_SIZE(sbi);
 299
 300                 if (start == 0)
 301                         goto submit_io;
 302
 303                 /* fill dummy pages */
 304                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 305                         struct page *page =
 306                                 mempool_alloc(sbi->write_io_dummy,
 307                                               GFP_NOIO | __GFP_NOFAIL);
 308                         f2fs_bug_on(sbi, !page);
 309
 310                         zero_user_segment(page, 0, PAGE_SIZE);
 311                         SetPagePrivate(page);
 312                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 313                         lock_page(page);
 314                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 315                                 f2fs_bug_on(sbi, 1);
 316                 }
 317                 /*
 318                  * In the NODE case, we lose next block address chain. So, we
 319                  * need to do checkpoint in f2fs_sync_file.
 320                  */
 321                 if (type == NODE)
 322                         set_sbi_flag(sbi, SBI_NEED_CP);
 323         }
 324 submit_io:
 325         if (is_read_io(bio_op(bio)))
 326                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 327         else
 328                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 329         submit_bio(bio);
 330 }
 331
 332 static void __submit_merged_bio(struct f2fs_bio_info *io)
 333 {
 334         struct f2fs_io_info *fio = &io->fio;
 335
 336         if (!io->bio)
 337                 return;
 338
 339         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 340
 341         if (is_read_io(fio->op))
 342                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 343         else
 344                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 345
 346         __submit_bio(io->sbi, io->bio, fio->type);
 347         io->bio = NULL;
 348 }
 349
 350 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 351                                                 struct page *page, nid_t ino)
 352 {
 353         struct bio_vec *bvec;
 354         struct page *target;
 355         struct bvec_iter_all iter_all;
 356
 357         if (!io->bio)
 358                 return false;
 359
 360         if (!inode && !page && !ino)
 361                 return true;
 362
 363         bio_for_each_segment_all(bvec, io->bio, iter_all) {
 364
 365                 target = bvec->bv_page;
 366                 if (fscrypt_is_bounce_page(target))
 367                         target = fscrypt_pagecache_page(target);
 368
 369                 if (inode && inode == target->mapping->host)
 370                         return true;
 371                 if (page && page == target)
 372                         return true;
 373                 if (ino && ino == ino_of_node(target))
 374                         return true;
 375         }
 376
 377         return false;
 378 }
 379
 380 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 381                                 enum page_type type, enum temp_type temp)
 382 {
 383         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 384         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 385
 386         down_write(&io->io_rwsem);
 387
 388         /* change META to META_FLUSH in the checkpoint procedure */
 389         if (type >= META_FLUSH) {
 390                 io->fio.type = META_FLUSH;
 391                 io->fio.op = REQ_OP_WRITE;
 392                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 393                 if (!test_opt(sbi, NOBARRIER))
 394                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 395         }
 396         __submit_merged_bio(io);
 397         up_write(&io->io_rwsem);
 398 }
 399
 400 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 401                                 struct inode *inode, struct page *page,
 402                                 nid_t ino, enum page_type type, bool force)
 403 {
 404         enum temp_type temp;
 405         bool ret = true;
 406
 407         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 408                 if (!force)     {
 409                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 410                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 411
 412                         down_read(&io->io_rwsem);
 413                         ret = __has_merged_page(io, inode, page, ino);
 414                         up_read(&io->io_rwsem);
 415                 }
 416                 if (ret)
 417                         __f2fs_submit_merged_write(sbi, type, temp);
 418
 419                 /* TODO: use HOT temp only for meta pages now. */
 420                 if (type >= META)
 421                         break;
 422         }
 423 }
 424
 425 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 426 {
 427         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 428 }
 429
 430 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 431                                 struct inode *inode, struct page *page,
 432                                 nid_t ino, enum page_type type)
 433 {
 434         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 435 }
 436
 437 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 438 {
 439         f2fs_submit_merged_write(sbi, DATA);
 440         f2fs_submit_merged_write(sbi, NODE);
 441         f2fs_submit_merged_write(sbi, META);
 442 }
 443
 444 /*
 445  * Fill the locked page with data located in the block address.
 446  * A caller needs to unlock the page on failure.
 447  */
 448 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 449 {
 450         struct bio *bio;
 451         struct page *page = fio->encrypted_page ?
 452                         fio->encrypted_page : fio->page;
 453
 454         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 455                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 456                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 457                 return -EFAULT;
 458
 459         trace_f2fs_submit_page_bio(page, fio);
 460         f2fs_trace_ios(fio, 0);
 461
 462         /* Allocate a new bio */
 463         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 464                                 1, is_read_io(fio->op), fio->type, fio->temp);
 465
 466         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 467                 bio_put(bio);
 468                 return -EFAULT;
 469         }
 470
 471         if (fio->io_wbc && !is_read_io(fio->op))
 472                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 473
 474         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 475
 476         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 477                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 478
 479         __submit_bio(fio->sbi, bio, fio->type);
 480         return 0;
 481 }
 482
 483 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 484 {
 485         struct f2fs_sb_info *sbi = fio->sbi;
 486         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 487         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 488         struct page *bio_page;
 489
 490         f2fs_bug_on(sbi, is_read_io(fio->op));
 491
 492         down_write(&io->io_rwsem);
 493 next:
 494         if (fio->in_list) {
 495                 spin_lock(&io->io_lock);
 496                 if (list_empty(&io->io_list)) {
 497                         spin_unlock(&io->io_lock);
 498                         goto out;
 499                 }
 500                 fio = list_first_entry(&io->io_list,
 501                                                 struct f2fs_io_info, list);
 502                 list_del(&fio->list);
 503                 spin_unlock(&io->io_lock);
 504         }
 505
 506         verify_fio_blkaddr(fio);
 507
 508         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 509
 510         /* set submitted = true as a return value */
 511         fio->submitted = true;
 512
 513         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 514
 515         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 516             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 517                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 518                 __submit_merged_bio(io);
 519 alloc_new:
 520         if (io->bio == NULL) {
 521                 if ((fio->type == DATA || fio->type == NODE) &&
 522                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 523                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 524                         fio->retry = true;
 525                         goto skip;
 526                 }
 527                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 528                                                 BIO_MAX_PAGES, false,
 529                                                 fio->type, fio->temp);
 530                 io->fio = *fio;
 531         }
 532
 533         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 534                 __submit_merged_bio(io);
 535                 goto alloc_new;
 536         }
 537
 538         if (fio->io_wbc)
 539                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 540
 541         io->last_block_in_bio = fio->new_blkaddr;
 542         f2fs_trace_ios(fio, 0);
 543
 544         trace_f2fs_submit_page_write(fio->page, fio);
 545 skip:
 546         if (fio->in_list)
 547                 goto next;
 548 out:
 549         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 550                                 f2fs_is_checkpoint_ready(sbi))
 551                 __submit_merged_bio(io);
 552         up_write(&io->io_rwsem);
 553 }
 554
 555 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 556                                         unsigned nr_pages, unsigned op_flag)
 557 {
 558         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 559         struct bio *bio;
 560         struct bio_post_read_ctx *ctx;
 561         unsigned int post_read_steps = 0;
 562
 563         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 564         if (!bio)
 565                 return ERR_PTR(-ENOMEM);
 566         f2fs_target_device(sbi, blkaddr, bio);
 567         bio->bi_end_io = f2fs_read_end_io;
 568         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 569
 570         if (f2fs_encrypted_file(inode))
 571                 post_read_steps |= 1 << STEP_DECRYPT;
 572         if (post_read_steps) {
 573                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 574                 if (!ctx) {
 575                         bio_put(bio);
 576                         return ERR_PTR(-ENOMEM);
 577                 }
 578                 ctx->bio = bio;
 579                 ctx->enabled_steps = post_read_steps;
 580                 bio->bi_private = ctx;
 581         }
 582
 583         return bio;
 584 }
 585
 586 /* This can handle encryption stuffs */
 587 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 588                                                         block_t blkaddr)
 589 {
 590         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 591         struct bio *bio;
 592
 593         bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 594         if (IS_ERR(bio))
 595                 return PTR_ERR(bio);
 596
 597         /* wait for GCed page writeback via META_MAPPING */
 598         f2fs_wait_on_block_writeback(inode, blkaddr);
 599
 600         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 601                 bio_put(bio);
 602                 return -EFAULT;
 603         }
 604         ClearPageError(page);
 605         inc_page_count(sbi, F2FS_RD_DATA);
 606         __submit_bio(sbi, bio, DATA);
 607         return 0;
 608 }
 609
 610 static void __set_data_blkaddr(struct dnode_of_data *dn)
 611 {
 612         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 613         __le32 *addr_array;
 614         int base = 0;
 615
 616         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 617                 base = get_extra_isize(dn->inode);
 618
 619         /* Get physical address of data block */
 620         addr_array = blkaddr_in_node(rn);
 621         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 622 }
 623
 624 /*
 625  * Lock ordering for the change of data block address:
 626  * ->data_page
 627  *  ->node_page
 628  *    update block addresses in the node page
 629  */
 630 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 631 {
 632         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 633         __set_data_blkaddr(dn);
 634         if (set_page_dirty(dn->node_page))
 635                 dn->node_changed = true;
 636 }
 637
 638 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 639 {
 640         dn->data_blkaddr = blkaddr;
 641         f2fs_set_data_blkaddr(dn);
 642         f2fs_update_extent_cache(dn);
 643 }
 644
 645 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 646 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 647 {
 648         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 649         int err;
 650
 651         if (!count)
 652                 return 0;
 653
 654         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 655                 return -EPERM;
 656         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 657                 return err;
 658
 659         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 660                                                 dn->ofs_in_node, count);
 661
 662         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 663
 664         for (; count > 0; dn->ofs_in_node++) {
 665                 block_t blkaddr = datablock_addr(dn->inode,
 666                                         dn->node_page, dn->ofs_in_node);
 667                 if (blkaddr == NULL_ADDR) {
 668                         dn->data_blkaddr = NEW_ADDR;
 669                         __set_data_blkaddr(dn);
 670                         count--;
 671                 }
 672         }
 673
 674         if (set_page_dirty(dn->node_page))
 675                 dn->node_changed = true;
 676         return 0;
 677 }
 678
 679 /* Should keep dn->ofs_in_node unchanged */
 680 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 681 {
 682         unsigned int ofs_in_node = dn->ofs_in_node;
 683         int ret;
 684
 685         ret = f2fs_reserve_new_blocks(dn, 1);
 686         dn->ofs_in_node = ofs_in_node;
 687         return ret;
 688 }
 689
 690 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 691 {
 692         bool need_put = dn->inode_page ? false : true;
 693         int err;
 694
 695         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 696         if (err)
 697                 return err;
 698
 699         if (dn->data_blkaddr == NULL_ADDR)
 700                 err = f2fs_reserve_new_block(dn);
 701         if (err || need_put)
 702                 f2fs_put_dnode(dn);
 703         return err;
 704 }
 705
 706 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 707 {
 708         struct extent_info ei  = {0,0,0};
 709         struct inode *inode = dn->inode;
 710
 711         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 712                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 713                 return 0;
 714         }
 715
 716         return f2fs_reserve_block(dn, index);
 717 }
 718
 719 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 720                                                 int op_flags, bool for_write)
 721 {
 722         struct address_space *mapping = inode->i_mapping;
 723         struct dnode_of_data dn;
 724         struct page *page;
 725         struct extent_info ei = {0,0,0};
 726         int err;
 727
 728         page = f2fs_grab_cache_page(mapping, index, for_write);
 729         if (!page)
 730                 return ERR_PTR(-ENOMEM);
 731
 732         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 733                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 734                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
 735                                                 DATA_GENERIC_ENHANCE_READ)) {
 736                         err = -EFAULT;
 737                         goto put_err;
 738                 }
 739                 goto got_it;
 740         }
 741
 742         set_new_dnode(&dn, inode, NULL, NULL, 0);
 743         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 744         if (err)
 745                 goto put_err;
 746         f2fs_put_dnode(&dn);
 747
 748         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 749                 err = -ENOENT;
 750                 goto put_err;
 751         }
 752         if (dn.data_blkaddr != NEW_ADDR &&
 753                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
 754                                                 dn.data_blkaddr,
 755                                                 DATA_GENERIC_ENHANCE)) {
 756                 err = -EFAULT;
 757                 goto put_err;
 758         }
 759 got_it:
 760         if (PageUptodate(page)) {
 761                 unlock_page(page);
 762                 return page;
 763         }
 764
 765         /*
 766          * A new dentry page is allocated but not able to be written, since its
 767          * new inode page couldn't be allocated due to -ENOSPC.
 768          * In such the case, its blkaddr can be remained as NEW_ADDR.
 769          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 770          * f2fs_init_inode_metadata.
 771          */
 772         if (dn.data_blkaddr == NEW_ADDR) {
 773                 zero_user_segment(page, 0, PAGE_SIZE);
 774                 if (!PageUptodate(page))
 775                         SetPageUptodate(page);
 776                 unlock_page(page);
 777                 return page;
 778         }
 779
 780         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 781         if (err)
 782                 goto put_err;
 783         return page;
 784
 785 put_err:
 786         f2fs_put_page(page, 1);
 787         return ERR_PTR(err);
 788 }
 789
 790 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 791 {
 792         struct address_space *mapping = inode->i_mapping;
 793         struct page *page;
 794
 795         page = find_get_page(mapping, index);
 796         if (page && PageUptodate(page))
 797                 return page;
 798         f2fs_put_page(page, 0);
 799
 800         page = f2fs_get_read_data_page(inode, index, 0, false);
 801         if (IS_ERR(page))
 802                 return page;
 803
 804         if (PageUptodate(page))
 805                 return page;
 806
 807         wait_on_page_locked(page);
 808         if (unlikely(!PageUptodate(page))) {
 809                 f2fs_put_page(page, 0);
 810                 return ERR_PTR(-EIO);
 811         }
 812         return page;
 813 }
 814
 815 /*
 816  * If it tries to access a hole, return an error.
 817  * Because, the callers, functions in dir.c and GC, should be able to know
 818  * whether this page exists or not.
 819  */
 820 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 821                                                         bool for_write)
 822 {
 823         struct address_space *mapping = inode->i_mapping;
 824         struct page *page;
 825 repeat:
 826         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 827         if (IS_ERR(page))
 828                 return page;
 829
 830         /* wait for read completion */
 831         lock_page(page);
 832         if (unlikely(page->mapping != mapping)) {
 833                 f2fs_put_page(page, 1);
 834                 goto repeat;
 835         }
 836         if (unlikely(!PageUptodate(page))) {
 837                 f2fs_put_page(page, 1);
 838                 return ERR_PTR(-EIO);
 839         }
 840         return page;
 841 }
 842
 843 /*
 844  * Caller ensures that this data page is never allocated.
 845  * A new zero-filled data page is allocated in the page cache.
 846  *
 847  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 848  * f2fs_unlock_op().
 849  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 850  * ipage should be released by this function.
 851  */
 852 struct page *f2fs_get_new_data_page(struct inode *inode,
 853                 struct page *ipage, pgoff_t index, bool new_i_size)
 854 {
 855         struct address_space *mapping = inode->i_mapping;
 856         struct page *page;
 857         struct dnode_of_data dn;
 858         int err;
 859
 860         page = f2fs_grab_cache_page(mapping, index, true);
 861         if (!page) {
 862                 /*
 863                  * before exiting, we should make sure ipage will be released
 864                  * if any error occur.
 865                  */
 866                 f2fs_put_page(ipage, 1);
 867                 return ERR_PTR(-ENOMEM);
 868         }
 869
 870         set_new_dnode(&dn, inode, ipage, NULL, 0);
 871         err = f2fs_reserve_block(&dn, index);
 872         if (err) {
 873                 f2fs_put_page(page, 1);
 874                 return ERR_PTR(err);
 875         }
 876         if (!ipage)
 877                 f2fs_put_dnode(&dn);
 878
 879         if (PageUptodate(page))
 880                 goto got_it;
 881
 882         if (dn.data_blkaddr == NEW_ADDR) {
 883                 zero_user_segment(page, 0, PAGE_SIZE);
 884                 if (!PageUptodate(page))
 885                         SetPageUptodate(page);
 886         } else {
 887                 f2fs_put_page(page, 1);
 888
 889                 /* if ipage exists, blkaddr should be NEW_ADDR */
 890                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 891                 page = f2fs_get_lock_data_page(inode, index, true);
 892                 if (IS_ERR(page))
 893                         return page;
 894         }
 895 got_it:
 896         if (new_i_size && i_size_read(inode) <
 897                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 898                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 899         return page;
 900 }
 901
 902 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 903 {
 904         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 905         struct f2fs_summary sum;
 906         struct node_info ni;
 907         block_t old_blkaddr;
 908         blkcnt_t count = 1;
 909         int err;
 910
 911         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 912                 return -EPERM;
 913
 914         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 915         if (err)
 916                 return err;
 917
 918         dn->data_blkaddr = datablock_addr(dn->inode,
 919                                 dn->node_page, dn->ofs_in_node);
 920         if (dn->data_blkaddr != NULL_ADDR)
 921                 goto alloc;
 922
 923         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 924                 return err;
 925
 926 alloc:
 927         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 928         old_blkaddr = dn->data_blkaddr;
 929         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 930                                         &sum, seg_type, NULL, false);
 931         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 932                 invalidate_mapping_pages(META_MAPPING(sbi),
 933                                         old_blkaddr, old_blkaddr);
 934         f2fs_set_data_blkaddr(dn);
 935
 936         /*
 937          * i_size will be updated by direct_IO. Otherwise, we'll get stale
 938          * data from unwritten block via dio_read.
 939          */
 940         return 0;
 941 }
 942
 943 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 944 {
 945         struct inode *inode = file_inode(iocb->ki_filp);
 946         struct f2fs_map_blocks map;
 947         int flag;
 948         int err = 0;
 949         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 950
 951         /* convert inline data for Direct I/O*/
 952         if (direct_io) {
 953                 err = f2fs_convert_inline_inode(inode);
 954                 if (err)
 955                         return err;
 956         }
 957
 958         if (direct_io && allow_outplace_dio(inode, iocb, from))
 959                 return 0;
 960
 961         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 962                 return 0;
 963
 964         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 965         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 966         if (map.m_len > map.m_lblk)
 967                 map.m_len -= map.m_lblk;
 968         else
 969                 map.m_len = 0;
 970
 971         map.m_next_pgofs = NULL;
 972         map.m_next_extent = NULL;
 973         map.m_seg_type = NO_CHECK_TYPE;
 974         map.m_may_create = true;
 975
 976         if (direct_io) {
 977                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 978                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
 979                                         F2FS_GET_BLOCK_PRE_AIO :
 980                                         F2FS_GET_BLOCK_PRE_DIO;
 981                 goto map_blocks;
 982         }
 983         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 984                 err = f2fs_convert_inline_inode(inode);
 985                 if (err)
 986                         return err;
 987         }
 988         if (f2fs_has_inline_data(inode))
 989                 return err;
 990
 991         flag = F2FS_GET_BLOCK_PRE_AIO;
 992
 993 map_blocks:
 994         err = f2fs_map_blocks(inode, &map, 1, flag);
 995         if (map.m_len > 0 && err == -ENOSPC) {
 996                 if (!direct_io)
 997                         set_inode_flag(inode, FI_NO_PREALLOC);
 998                 err = 0;
 999         }
1000         return err;
1001 }
1002
1003 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1004 {
1005         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1006                 if (lock)
1007                         down_read(&sbi->node_change);
1008                 else
1009                         up_read(&sbi->node_change);
1010         } else {
1011                 if (lock)
1012                         f2fs_lock_op(sbi);
1013                 else
1014                         f2fs_unlock_op(sbi);
1015         }
1016 }
1017
1018 /*
1019  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1020  * f2fs_map_blocks structure.
1021  * If original data blocks are allocated, then give them to blockdev.
1022  * Otherwise,
1023  *     a. preallocate requested block addresses
1024  *     b. do not use extent cache for better performance
1025  *     c. give the block addresses to blockdev
1026  */
1027 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1028                                                 int create, int flag)
1029 {
1030         unsigned int maxblocks = map->m_len;
1031         struct dnode_of_data dn;
1032         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1033         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1034         pgoff_t pgofs, end_offset, end;
1035         int err = 0, ofs = 1;
1036         unsigned int ofs_in_node, last_ofs_in_node;
1037         blkcnt_t prealloc;
1038         struct extent_info ei = {0,0,0};
1039         block_t blkaddr;
1040         unsigned int start_pgofs;
1041
1042         if (!maxblocks)
1043                 return 0;
1044
1045         map->m_len = 0;
1046         map->m_flags = 0;
1047
1048         /* it only supports block size == page size */
1049         pgofs = (pgoff_t)map->m_lblk;
1050         end = pgofs + maxblocks;
1051
1052         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1053                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1054                                                         map->m_may_create)
1055                         goto next_dnode;
1056
1057                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1058                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1059                 map->m_flags = F2FS_MAP_MAPPED;
1060                 if (map->m_next_extent)
1061                         *map->m_next_extent = pgofs + map->m_len;
1062
1063                 /* for hardware encryption, but to avoid potential issue in future */
1064                 if (flag == F2FS_GET_BLOCK_DIO)
1065                         f2fs_wait_on_block_writeback_range(inode,
1066                                                 map->m_pblk, map->m_len);
1067                 goto out;
1068         }
1069
1070 next_dnode:
1071         if (map->m_may_create)
1072                 __do_map_lock(sbi, flag, true);
1073
1074         /* When reading holes, we need its node page */
1075         set_new_dnode(&dn, inode, NULL, NULL, 0);
1076         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1077         if (err) {
1078                 if (flag == F2FS_GET_BLOCK_BMAP)
1079                         map->m_pblk = 0;
1080                 if (err == -ENOENT) {
1081                         err = 0;
1082                         if (map->m_next_pgofs)
1083                                 *map->m_next_pgofs =
1084                                         f2fs_get_next_page_offset(&dn, pgofs);
1085                         if (map->m_next_extent)
1086                                 *map->m_next_extent =
1087                                         f2fs_get_next_page_offset(&dn, pgofs);
1088                 }
1089                 goto unlock_out;
1090         }
1091
1092         start_pgofs = pgofs;
1093         prealloc = 0;
1094         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1095         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1096
1097 next_block:
1098         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1099
1100         if (__is_valid_data_blkaddr(blkaddr) &&
1101                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1102                 err = -EFAULT;
1103                 goto sync_out;
1104         }
1105
1106         if (__is_valid_data_blkaddr(blkaddr)) {
1107                 /* use out-place-update for driect IO under LFS mode */
1108                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1109                                                         map->m_may_create) {
1110                         err = __allocate_data_block(&dn, map->m_seg_type);
1111                         if (!err) {
1112                                 blkaddr = dn.data_blkaddr;
1113                                 set_inode_flag(inode, FI_APPEND_WRITE);
1114                         }
1115                 }
1116         } else {
1117                 if (create) {
1118                         if (unlikely(f2fs_cp_error(sbi))) {
1119                                 err = -EIO;
1120                                 goto sync_out;
1121                         }
1122                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1123                                 if (blkaddr == NULL_ADDR) {
1124                                         prealloc++;
1125                                         last_ofs_in_node = dn.ofs_in_node;
1126                                 }
1127                         } else {
1128                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1129                                         flag != F2FS_GET_BLOCK_DIO);
1130                                 err = __allocate_data_block(&dn,
1131                                                         map->m_seg_type);
1132                                 if (!err)
1133                                         set_inode_flag(inode, FI_APPEND_WRITE);
1134                         }
1135                         if (err)
1136                                 goto sync_out;
1137                         map->m_flags |= F2FS_MAP_NEW;
1138                         blkaddr = dn.data_blkaddr;
1139                 } else {
1140                         if (flag == F2FS_GET_BLOCK_BMAP) {
1141                                 map->m_pblk = 0;
1142                                 goto sync_out;
1143                         }
1144                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1145                                 goto sync_out;
1146                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1147                                                 blkaddr == NULL_ADDR) {
1148                                 if (map->m_next_pgofs)
1149                                         *map->m_next_pgofs = pgofs + 1;
1150                                 goto sync_out;
1151                         }
1152                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1153                                 /* for defragment case */
1154                                 if (map->m_next_pgofs)
1155                                         *map->m_next_pgofs = pgofs + 1;
1156                                 goto sync_out;
1157                         }
1158                 }
1159         }
1160
1161         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1162                 goto skip;
1163
1164         if (map->m_len == 0) {
1165                 /* preallocated unwritten block should be mapped for fiemap. */
1166                 if (blkaddr == NEW_ADDR)
1167                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1168                 map->m_flags |= F2FS_MAP_MAPPED;
1169
1170                 map->m_pblk = blkaddr;
1171                 map->m_len = 1;
1172         } else if ((map->m_pblk != NEW_ADDR &&
1173                         blkaddr == (map->m_pblk + ofs)) ||
1174                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1175                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1176                 ofs++;
1177                 map->m_len++;
1178         } else {
1179                 goto sync_out;
1180         }
1181
1182 skip:
1183         dn.ofs_in_node++;
1184         pgofs++;
1185
1186         /* preallocate blocks in batch for one dnode page */
1187         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1188                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1189
1190                 dn.ofs_in_node = ofs_in_node;
1191                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1192                 if (err)
1193                         goto sync_out;
1194
1195                 map->m_len += dn.ofs_in_node - ofs_in_node;
1196                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1197                         err = -ENOSPC;
1198                         goto sync_out;
1199                 }
1200                 dn.ofs_in_node = end_offset;
1201         }
1202
1203         if (pgofs >= end)
1204                 goto sync_out;
1205         else if (dn.ofs_in_node < end_offset)
1206                 goto next_block;
1207
1208         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1209                 if (map->m_flags & F2FS_MAP_MAPPED) {
1210                         unsigned int ofs = start_pgofs - map->m_lblk;
1211
1212                         f2fs_update_extent_cache_range(&dn,
1213                                 start_pgofs, map->m_pblk + ofs,
1214                                 map->m_len - ofs);
1215                 }
1216         }
1217
1218         f2fs_put_dnode(&dn);
1219
1220         if (map->m_may_create) {
1221                 __do_map_lock(sbi, flag, false);
1222                 f2fs_balance_fs(sbi, dn.node_changed);
1223         }
1224         goto next_dnode;
1225
1226 sync_out:
1227
1228         /* for hardware encryption, but to avoid potential issue in future */
1229         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1230                 f2fs_wait_on_block_writeback_range(inode,
1231                                                 map->m_pblk, map->m_len);
1232
1233         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1234                 if (map->m_flags & F2FS_MAP_MAPPED) {
1235                         unsigned int ofs = start_pgofs - map->m_lblk;
1236
1237                         f2fs_update_extent_cache_range(&dn,
1238                                 start_pgofs, map->m_pblk + ofs,
1239                                 map->m_len - ofs);
1240                 }
1241                 if (map->m_next_extent)
1242                         *map->m_next_extent = pgofs + 1;
1243         }
1244         f2fs_put_dnode(&dn);
1245 unlock_out:
1246         if (map->m_may_create) {
1247                 __do_map_lock(sbi, flag, false);
1248                 f2fs_balance_fs(sbi, dn.node_changed);
1249         }
1250 out:
1251         trace_f2fs_map_blocks(inode, map, err);
1252         return err;
1253 }
1254
1255 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1256 {
1257         struct f2fs_map_blocks map;
1258         block_t last_lblk;
1259         int err;
1260
1261         if (pos + len > i_size_read(inode))
1262                 return false;
1263
1264         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1265         map.m_next_pgofs = NULL;
1266         map.m_next_extent = NULL;
1267         map.m_seg_type = NO_CHECK_TYPE;
1268         map.m_may_create = false;
1269         last_lblk = F2FS_BLK_ALIGN(pos + len);
1270
1271         while (map.m_lblk < last_lblk) {
1272                 map.m_len = last_lblk - map.m_lblk;
1273                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1274                 if (err || map.m_len == 0)
1275                         return false;
1276                 map.m_lblk += map.m_len;
1277         }
1278         return true;
1279 }
1280
1281 static int __get_data_block(struct inode *inode, sector_t iblock,
1282                         struct buffer_head *bh, int create, int flag,
1283                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1284 {
1285         struct f2fs_map_blocks map;
1286         int err;
1287
1288         map.m_lblk = iblock;
1289         map.m_len = bh->b_size >> inode->i_blkbits;
1290         map.m_next_pgofs = next_pgofs;
1291         map.m_next_extent = NULL;
1292         map.m_seg_type = seg_type;
1293         map.m_may_create = may_write;
1294
1295         err = f2fs_map_blocks(inode, &map, create, flag);
1296         if (!err) {
1297                 map_bh(bh, inode->i_sb, map.m_pblk);
1298                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1299                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1300         }
1301         return err;
1302 }
1303
1304 static int get_data_block(struct inode *inode, sector_t iblock,
1305                         struct buffer_head *bh_result, int create, int flag,
1306                         pgoff_t *next_pgofs)
1307 {
1308         return __get_data_block(inode, iblock, bh_result, create,
1309                                                         flag, next_pgofs,
1310                                                         NO_CHECK_TYPE, create);
1311 }
1312
1313 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1314                         struct buffer_head *bh_result, int create)
1315 {
1316         return __get_data_block(inode, iblock, bh_result, create,
1317                                 F2FS_GET_BLOCK_DIO, NULL,
1318                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1319                                 true);
1320 }
1321
1322 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1323                         struct buffer_head *bh_result, int create)
1324 {
1325         return __get_data_block(inode, iblock, bh_result, create,
1326                                 F2FS_GET_BLOCK_DIO, NULL,
1327                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1328                                 false);
1329 }
1330
1331 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1332                         struct buffer_head *bh_result, int create)
1333 {
1334         /* Block number less than F2FS MAX BLOCKS */
1335         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1336                 return -EFBIG;
1337
1338         return __get_data_block(inode, iblock, bh_result, create,
1339                                                 F2FS_GET_BLOCK_BMAP, NULL,
1340                                                 NO_CHECK_TYPE, create);
1341 }
1342
1343 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1344 {
1345         return (offset >> inode->i_blkbits);
1346 }
1347
1348 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1349 {
1350         return (blk << inode->i_blkbits);
1351 }
1352
1353 static int f2fs_xattr_fiemap(struct inode *inode,
1354                                 struct fiemap_extent_info *fieinfo)
1355 {
1356         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1357         struct page *page;
1358         struct node_info ni;
1359         __u64 phys = 0, len;
1360         __u32 flags;
1361         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1362         int err = 0;
1363
1364         if (f2fs_has_inline_xattr(inode)) {
1365                 int offset;
1366
1367                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1368                                                 inode->i_ino, false);
1369                 if (!page)
1370                         return -ENOMEM;
1371
1372                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1373                 if (err) {
1374                         f2fs_put_page(page, 1);
1375                         return err;
1376                 }
1377
1378                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1379                 offset = offsetof(struct f2fs_inode, i_addr) +
1380                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1381                                         get_inline_xattr_addrs(inode));
1382
1383                 phys += offset;
1384                 len = inline_xattr_size(inode);
1385
1386                 f2fs_put_page(page, 1);
1387
1388                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1389
1390                 if (!xnid)
1391                         flags |= FIEMAP_EXTENT_LAST;
1392
1393                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1394                 if (err || err == 1)
1395                         return err;
1396         }
1397
1398         if (xnid) {
1399                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1400                 if (!page)
1401                         return -ENOMEM;
1402
1403                 err = f2fs_get_node_info(sbi, xnid, &ni);
1404                 if (err) {
1405                         f2fs_put_page(page, 1);
1406                         return err;
1407                 }
1408
1409                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1410                 len = inode->i_sb->s_blocksize;
1411
1412                 f2fs_put_page(page, 1);
1413
1414                 flags = FIEMAP_EXTENT_LAST;
1415         }
1416
1417         if (phys)
1418                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1419
1420         return (err < 0 ? err : 0);
1421 }
1422
1423 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1424                 u64 start, u64 len)
1425 {
1426         struct buffer_head map_bh;
1427         sector_t start_blk, last_blk;
1428         pgoff_t next_pgofs;
1429         u64 logical = 0, phys = 0, size = 0;
1430         u32 flags = 0;
1431         int ret = 0;
1432
1433         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1434                 ret = f2fs_precache_extents(inode);
1435                 if (ret)
1436                         return ret;
1437         }
1438
1439         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1440         if (ret)
1441                 return ret;
1442
1443         inode_lock(inode);
1444
1445         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1446                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1447                 goto out;
1448         }
1449
1450         if (f2fs_has_inline_data(inode)) {
1451                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1452                 if (ret != -EAGAIN)
1453                         goto out;
1454         }
1455
1456         if (logical_to_blk(inode, len) == 0)
1457                 len = blk_to_logical(inode, 1);
1458
1459         start_blk = logical_to_blk(inode, start);
1460         last_blk = logical_to_blk(inode, start + len - 1);
1461
1462 next:
1463         memset(&map_bh, 0, sizeof(struct buffer_head));
1464         map_bh.b_size = len;
1465
1466         ret = get_data_block(inode, start_blk, &map_bh, 0,
1467                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1468         if (ret)
1469                 goto out;
1470
1471         /* HOLE */
1472         if (!buffer_mapped(&map_bh)) {
1473                 start_blk = next_pgofs;
1474
1475                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1476                                         F2FS_I_SB(inode)->max_file_blocks))
1477                         goto prep_next;
1478
1479                 flags |= FIEMAP_EXTENT_LAST;
1480         }
1481
1482         if (size) {
1483                 if (IS_ENCRYPTED(inode))
1484                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1485
1486                 ret = fiemap_fill_next_extent(fieinfo, logical,
1487                                 phys, size, flags);
1488         }
1489
1490         if (start_blk > last_blk || ret)
1491                 goto out;
1492
1493         logical = blk_to_logical(inode, start_blk);
1494         phys = blk_to_logical(inode, map_bh.b_blocknr);
1495         size = map_bh.b_size;
1496         flags = 0;
1497         if (buffer_unwritten(&map_bh))
1498                 flags = FIEMAP_EXTENT_UNWRITTEN;
1499
1500         start_blk += logical_to_blk(inode, size);
1501
1502 prep_next:
1503         cond_resched();
1504         if (fatal_signal_pending(current))
1505                 ret = -EINTR;
1506         else
1507                 goto next;
1508 out:
1509         if (ret == 1)
1510                 ret = 0;
1511
1512         inode_unlock(inode);
1513         return ret;
1514 }
1515
1516 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1517                                         unsigned nr_pages,
1518                                         struct f2fs_map_blocks *map,
1519                                         struct bio **bio_ret,
1520                                         sector_t *last_block_in_bio,
1521                                         bool is_readahead)
1522 {
1523         struct bio *bio = *bio_ret;
1524         const unsigned blkbits = inode->i_blkbits;
1525         const unsigned blocksize = 1 << blkbits;
1526         sector_t block_in_file;
1527         sector_t last_block;
1528         sector_t last_block_in_file;
1529         sector_t block_nr;
1530         int ret = 0;
1531
1532         block_in_file = (sector_t)page->index;
1533         last_block = block_in_file + nr_pages;
1534         last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1535                                                         blkbits;
1536         if (last_block > last_block_in_file)
1537                 last_block = last_block_in_file;
1538
1539         /* just zeroing out page which is beyond EOF */
1540         if (block_in_file >= last_block)
1541                 goto zero_out;
1542         /*
1543          * Map blocks using the previous result first.
1544          */
1545         if ((map->m_flags & F2FS_MAP_MAPPED) &&
1546                         block_in_file > map->m_lblk &&
1547                         block_in_file < (map->m_lblk + map->m_len))
1548                 goto got_it;
1549
1550         /*
1551          * Then do more f2fs_map_blocks() calls until we are
1552          * done with this page.
1553          */
1554         map->m_lblk = block_in_file;
1555         map->m_len = last_block - block_in_file;
1556
1557         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1558         if (ret)
1559                 goto out;
1560 got_it:
1561         if ((map->m_flags & F2FS_MAP_MAPPED)) {
1562                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1563                 SetPageMappedToDisk(page);
1564
1565                 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1566                         SetPageUptodate(page);
1567                         goto confused;
1568                 }
1569
1570                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1571                                                 DATA_GENERIC_ENHANCE_READ)) {
1572                         ret = -EFAULT;
1573                         goto out;
1574                 }
1575         } else {
1576 zero_out:
1577                 zero_user_segment(page, 0, PAGE_SIZE);
1578                 if (!PageUptodate(page))
1579                         SetPageUptodate(page);
1580                 unlock_page(page);
1581                 goto out;
1582         }
1583
1584         /*
1585          * This page will go to BIO.  Do we need to send this
1586          * BIO off first?
1587          */
1588         if (bio && (*last_block_in_bio != block_nr - 1 ||
1589                 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1590 submit_and_realloc:
1591                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1592                 bio = NULL;
1593         }
1594         if (bio == NULL) {
1595                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1596                                 is_readahead ? REQ_RAHEAD : 0);
1597                 if (IS_ERR(bio)) {
1598                         ret = PTR_ERR(bio);
1599                         bio = NULL;
1600                         goto out;
1601                 }
1602         }
1603
1604         /*
1605          * If the page is under writeback, we need to wait for
1606          * its completion to see the correct decrypted data.
1607          */
1608         f2fs_wait_on_block_writeback(inode, block_nr);
1609
1610         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1611                 goto submit_and_realloc;
1612
1613         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1614         ClearPageError(page);
1615         *last_block_in_bio = block_nr;
1616         goto out;
1617 confused:
1618         if (bio) {
1619                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1620                 bio = NULL;
1621         }
1622         unlock_page(page);
1623 out:
1624         *bio_ret = bio;
1625         return ret;
1626 }
1627
1628 /*
1629  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1630  * Major change was from block_size == page_size in f2fs by default.
1631  *
1632  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1633  * this function ever deviates from doing just read-ahead, it should either
1634  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1635  * from read-ahead.
1636  */
1637 static int f2fs_mpage_readpages(struct address_space *mapping,
1638                         struct list_head *pages, struct page *page,
1639                         unsigned nr_pages, bool is_readahead)
1640 {
1641         struct bio *bio = NULL;
1642         sector_t last_block_in_bio = 0;
1643         struct inode *inode = mapping->host;
1644         struct f2fs_map_blocks map;
1645         int ret = 0;
1646
1647         map.m_pblk = 0;
1648         map.m_lblk = 0;
1649         map.m_len = 0;
1650         map.m_flags = 0;
1651         map.m_next_pgofs = NULL;
1652         map.m_next_extent = NULL;
1653         map.m_seg_type = NO_CHECK_TYPE;
1654         map.m_may_create = false;
1655
1656         for (; nr_pages; nr_pages--) {
1657                 if (pages) {
1658                         page = list_last_entry(pages, struct page, lru);
1659
1660                         prefetchw(&page->flags);
1661                         list_del(&page->lru);
1662                         if (add_to_page_cache_lru(page, mapping,
1663                                                   page->index,
1664                                                   readahead_gfp_mask(mapping)))
1665                                 goto next_page;
1666                 }
1667
1668                 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1669                                         &last_block_in_bio, is_readahead);
1670                 if (ret) {
1671                         SetPageError(page);
1672                         zero_user_segment(page, 0, PAGE_SIZE);
1673                         unlock_page(page);
1674                 }
1675 next_page:
1676                 if (pages)
1677                         put_page(page);
1678         }
1679         BUG_ON(pages && !list_empty(pages));
1680         if (bio)
1681                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1682         return pages ? 0 : ret;
1683 }
1684
1685 static int f2fs_read_data_page(struct file *file, struct page *page)
1686 {
1687         struct inode *inode = page->mapping->host;
1688         int ret = -EAGAIN;
1689
1690         trace_f2fs_readpage(page, DATA);
1691
1692         /* If the file has inline data, try to read it directly */
1693         if (f2fs_has_inline_data(inode))
1694                 ret = f2fs_read_inline_data(inode, page);
1695         if (ret == -EAGAIN)
1696                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1697         return ret;
1698 }
1699
1700 static int f2fs_read_data_pages(struct file *file,
1701                         struct address_space *mapping,
1702                         struct list_head *pages, unsigned nr_pages)
1703 {
1704         struct inode *inode = mapping->host;
1705         struct page *page = list_last_entry(pages, struct page, lru);
1706
1707         trace_f2fs_readpages(inode, page, nr_pages);
1708
1709         /* If the file has inline data, skip readpages */
1710         if (f2fs_has_inline_data(inode))
1711                 return 0;
1712
1713         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1714 }
1715
1716 static int encrypt_one_page(struct f2fs_io_info *fio)
1717 {
1718         struct inode *inode = fio->page->mapping->host;
1719         struct page *mpage;
1720         gfp_t gfp_flags = GFP_NOFS;
1721
1722         if (!f2fs_encrypted_file(inode))
1723                 return 0;
1724
1725         /* wait for GCed page writeback via META_MAPPING */
1726         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1727
1728 retry_encrypt:
1729         fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(fio->page,
1730                                                                PAGE_SIZE, 0,
1731                                                                gfp_flags);
1732         if (IS_ERR(fio->encrypted_page)) {
1733                 /* flush pending IOs and wait for a while in the ENOMEM case */
1734                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1735                         f2fs_flush_merged_writes(fio->sbi);
1736                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1737                         gfp_flags |= __GFP_NOFAIL;
1738                         goto retry_encrypt;
1739                 }
1740                 return PTR_ERR(fio->encrypted_page);
1741         }
1742
1743         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1744         if (mpage) {
1745                 if (PageUptodate(mpage))
1746                         memcpy(page_address(mpage),
1747                                 page_address(fio->encrypted_page), PAGE_SIZE);
1748                 f2fs_put_page(mpage, 1);
1749         }
1750         return 0;
1751 }
1752
1753 static inline bool check_inplace_update_policy(struct inode *inode,
1754                                 struct f2fs_io_info *fio)
1755 {
1756         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1757         unsigned int policy = SM_I(sbi)->ipu_policy;
1758
1759         if (policy & (0x1 << F2FS_IPU_FORCE))
1760                 return true;
1761         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1762                 return true;
1763         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1764                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1765                 return true;
1766         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1767                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1768                 return true;
1769
1770         /*
1771          * IPU for rewrite async pages
1772          */
1773         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1774                         fio && fio->op == REQ_OP_WRITE &&
1775                         !(fio->op_flags & REQ_SYNC) &&
1776                         !IS_ENCRYPTED(inode))
1777                 return true;
1778
1779         /* this is only set during fdatasync */
1780         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1781                         is_inode_flag_set(inode, FI_NEED_IPU))
1782                 return true;
1783
1784         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1785                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1786                 return true;
1787
1788         return false;
1789 }
1790
1791 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1792 {
1793         if (f2fs_is_pinned_file(inode))
1794                 return true;
1795
1796         /* if this is cold file, we should overwrite to avoid fragmentation */
1797         if (file_is_cold(inode))
1798                 return true;
1799
1800         return check_inplace_update_policy(inode, fio);
1801 }
1802
1803 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1804 {
1805         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1806
1807         if (test_opt(sbi, LFS))
1808                 return true;
1809         if (S_ISDIR(inode->i_mode))
1810                 return true;
1811         if (IS_NOQUOTA(inode))
1812                 return true;
1813         if (f2fs_is_atomic_file(inode))
1814                 return true;
1815         if (fio) {
1816                 if (is_cold_data(fio->page))
1817                         return true;
1818                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1819                         return true;
1820                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1821                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1822                         return true;
1823         }
1824         return false;
1825 }
1826
1827 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1828 {
1829         struct inode *inode = fio->page->mapping->host;
1830
1831         if (f2fs_should_update_outplace(inode, fio))
1832                 return false;
1833
1834         return f2fs_should_update_inplace(inode, fio);
1835 }
1836
1837 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1838 {
1839         struct page *page = fio->page;
1840         struct inode *inode = page->mapping->host;
1841         struct dnode_of_data dn;
1842         struct extent_info ei = {0,0,0};
1843         struct node_info ni;
1844         bool ipu_force = false;
1845         int err = 0;
1846
1847         set_new_dnode(&dn, inode, NULL, NULL, 0);
1848         if (need_inplace_update(fio) &&
1849                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1850                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1851
1852                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1853                                                 DATA_GENERIC_ENHANCE))
1854                         return -EFAULT;
1855
1856                 ipu_force = true;
1857                 fio->need_lock = LOCK_DONE;
1858                 goto got_it;
1859         }
1860
1861         /* Deadlock due to between page->lock and f2fs_lock_op */
1862         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1863                 return -EAGAIN;
1864
1865         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1866         if (err)
1867                 goto out;
1868
1869         fio->old_blkaddr = dn.data_blkaddr;
1870
1871         /* This page is already truncated */
1872         if (fio->old_blkaddr == NULL_ADDR) {
1873                 ClearPageUptodate(page);
1874                 clear_cold_data(page);
1875                 goto out_writepage;
1876         }
1877 got_it:
1878         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1879                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1880                                                 DATA_GENERIC_ENHANCE)) {
1881                 err = -EFAULT;
1882                 goto out_writepage;
1883         }
1884         /*
1885          * If current allocation needs SSR,
1886          * it had better in-place writes for updated data.
1887          */
1888         if (ipu_force ||
1889                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1890                                         need_inplace_update(fio))) {
1891                 err = encrypt_one_page(fio);
1892                 if (err)
1893                         goto out_writepage;
1894
1895                 set_page_writeback(page);
1896                 ClearPageError(page);
1897                 f2fs_put_dnode(&dn);
1898                 if (fio->need_lock == LOCK_REQ)
1899                         f2fs_unlock_op(fio->sbi);
1900                 err = f2fs_inplace_write_data(fio);
1901                 if (err) {
1902                         if (f2fs_encrypted_file(inode))
1903                                 fscrypt_finalize_bounce_page(&fio->encrypted_page);
1904                         if (PageWriteback(page))
1905                                 end_page_writeback(page);
1906                 } else {
1907                         set_inode_flag(inode, FI_UPDATE_WRITE);
1908                 }
1909                 trace_f2fs_do_write_data_page(fio->page, IPU);
1910                 return err;
1911         }
1912
1913         if (fio->need_lock == LOCK_RETRY) {
1914                 if (!f2fs_trylock_op(fio->sbi)) {
1915                         err = -EAGAIN;
1916                         goto out_writepage;
1917                 }
1918                 fio->need_lock = LOCK_REQ;
1919         }
1920
1921         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1922         if (err)
1923                 goto out_writepage;
1924
1925         fio->version = ni.version;
1926
1927         err = encrypt_one_page(fio);
1928         if (err)
1929                 goto out_writepage;
1930
1931         set_page_writeback(page);
1932         ClearPageError(page);
1933
1934         /* LFS mode write path */
1935         f2fs_outplace_write_data(&dn, fio);
1936         trace_f2fs_do_write_data_page(page, OPU);
1937         set_inode_flag(inode, FI_APPEND_WRITE);
1938         if (page->index == 0)
1939                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1940 out_writepage:
1941         f2fs_put_dnode(&dn);
1942 out:
1943         if (fio->need_lock == LOCK_REQ)
1944                 f2fs_unlock_op(fio->sbi);
1945         return err;
1946 }
1947
1948 static int __write_data_page(struct page *page, bool *submitted,
1949                                 struct writeback_control *wbc,
1950                                 enum iostat_type io_type)
1951 {
1952         struct inode *inode = page->mapping->host;
1953         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1954         loff_t i_size = i_size_read(inode);
1955         const pgoff_t end_index = ((unsigned long long) i_size)
1956                                                         >> PAGE_SHIFT;
1957         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1958         unsigned offset = 0;
1959         bool need_balance_fs = false;
1960         int err = 0;
1961         struct f2fs_io_info fio = {
1962                 .sbi = sbi,
1963                 .ino = inode->i_ino,
1964                 .type = DATA,
1965                 .op = REQ_OP_WRITE,
1966                 .op_flags = wbc_to_write_flags(wbc),
1967                 .old_blkaddr = NULL_ADDR,
1968                 .page = page,
1969                 .encrypted_page = NULL,
1970                 .submitted = false,
1971                 .need_lock = LOCK_RETRY,
1972                 .io_type = io_type,
1973                 .io_wbc = wbc,
1974         };
1975
1976         trace_f2fs_writepage(page, DATA);
1977
1978         /* we should bypass data pages to proceed the kworkder jobs */
1979         if (unlikely(f2fs_cp_error(sbi))) {
1980                 mapping_set_error(page->mapping, -EIO);
1981                 /*
1982                  * don't drop any dirty dentry pages for keeping lastest
1983                  * directory structure.
1984                  */
1985                 if (S_ISDIR(inode->i_mode))
1986                         goto redirty_out;
1987                 goto out;
1988         }
1989
1990         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1991                 goto redirty_out;
1992
1993         if (page->index < end_index)
1994                 goto write;
1995
1996         /*
1997          * If the offset is out-of-range of file size,
1998          * this page does not have to be written to disk.
1999          */
2000         offset = i_size & (PAGE_SIZE - 1);
2001         if ((page->index >= end_index + 1) || !offset)
2002                 goto out;
2003
2004         zero_user_segment(page, offset, PAGE_SIZE);
2005 write:
2006         if (f2fs_is_drop_cache(inode))
2007                 goto out;
2008         /* we should not write 0'th page having journal header */
2009         if (f2fs_is_volatile_file(inode) && (!page->index ||
2010                         (!wbc->for_reclaim &&
2011                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2012                 goto redirty_out;
2013
2014         /* Dentry blocks are controlled by checkpoint */
2015         if (S_ISDIR(inode->i_mode)) {
2016                 fio.need_lock = LOCK_DONE;
2017                 err = f2fs_do_write_data_page(&fio);
2018                 goto done;
2019         }
2020
2021         if (!wbc->for_reclaim)
2022                 need_balance_fs = true;
2023         else if (has_not_enough_free_secs(sbi, 0, 0))
2024                 goto redirty_out;
2025         else
2026                 set_inode_flag(inode, FI_HOT_DATA);
2027
2028         err = -EAGAIN;
2029         if (f2fs_has_inline_data(inode)) {
2030                 err = f2fs_write_inline_data(inode, page);
2031                 if (!err)
2032                         goto out;
2033         }
2034
2035         if (err == -EAGAIN) {
2036                 err = f2fs_do_write_data_page(&fio);
2037                 if (err == -EAGAIN) {
2038                         fio.need_lock = LOCK_REQ;
2039                         err = f2fs_do_write_data_page(&fio);
2040                 }
2041         }
2042
2043         if (err) {
2044                 file_set_keep_isize(inode);
2045         } else {
2046                 down_write(&F2FS_I(inode)->i_sem);
2047                 if (F2FS_I(inode)->last_disk_size < psize)
2048                         F2FS_I(inode)->last_disk_size = psize;
2049                 up_write(&F2FS_I(inode)->i_sem);
2050         }
2051
2052 done:
2053         if (err && err != -ENOENT)
2054                 goto redirty_out;
2055
2056 out:
2057         inode_dec_dirty_pages(inode);
2058         if (err) {
2059                 ClearPageUptodate(page);
2060                 clear_cold_data(page);
2061         }
2062
2063         if (wbc->for_reclaim) {
2064                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2065                 clear_inode_flag(inode, FI_HOT_DATA);
2066                 f2fs_remove_dirty_inode(inode);
2067                 submitted = NULL;
2068         }
2069
2070         unlock_page(page);
2071         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2072                                         !F2FS_I(inode)->cp_task)
2073                 f2fs_balance_fs(sbi, need_balance_fs);
2074
2075         if (unlikely(f2fs_cp_error(sbi))) {
2076                 f2fs_submit_merged_write(sbi, DATA);
2077                 submitted = NULL;
2078         }
2079
2080         if (submitted)
2081                 *submitted = fio.submitted;
2082
2083         return 0;
2084
2085 redirty_out:
2086         redirty_page_for_writepage(wbc, page);
2087         /*
2088          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2089          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2090          * file_write_and_wait_range() will see EIO error, which is critical
2091          * to return value of fsync() followed by atomic_write failure to user.
2092          */
2093         if (!err || wbc->for_reclaim)
2094                 return AOP_WRITEPAGE_ACTIVATE;
2095         unlock_page(page);
2096         return err;
2097 }
2098
2099 static int f2fs_write_data_page(struct page *page,
2100                                         struct writeback_control *wbc)
2101 {
2102         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2103 }
2104
2105 /*
2106  * This function was copied from write_cche_pages from mm/page-writeback.c.
2107  * The major change is making write step of cold data page separately from
2108  * warm/hot data page.
2109  */
2110 static int f2fs_write_cache_pages(struct address_space *mapping,
2111                                         struct writeback_control *wbc,
2112                                         enum iostat_type io_type)
2113 {
2114         int ret = 0;
2115         int done = 0;
2116         struct pagevec pvec;
2117         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2118         int nr_pages;
2119         pgoff_t uninitialized_var(writeback_index);
2120         pgoff_t index;
2121         pgoff_t end;            /* Inclusive */
2122         pgoff_t done_index;
2123         int cycled;
2124         int range_whole = 0;
2125         xa_mark_t tag;
2126         int nwritten = 0;
2127
2128         pagevec_init(&pvec);
2129
2130         if (get_dirty_pages(mapping->host) <=
2131                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2132                 set_inode_flag(mapping->host, FI_HOT_DATA);
2133         else
2134                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2135
2136         if (wbc->range_cyclic) {
2137                 writeback_index = mapping->writeback_index; /* prev offset */
2138                 index = writeback_index;
2139                 if (index == 0)
2140                         cycled = 1;
2141                 else
2142                         cycled = 0;
2143                 end = -1;
2144         } else {
2145                 index = wbc->range_start >> PAGE_SHIFT;
2146                 end = wbc->range_end >> PAGE_SHIFT;
2147                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2148                         range_whole = 1;
2149                 cycled = 1; /* ignore range_cyclic tests */
2150         }
2151         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2152                 tag = PAGECACHE_TAG_TOWRITE;
2153         else
2154                 tag = PAGECACHE_TAG_DIRTY;
2155 retry:
2156         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2157                 tag_pages_for_writeback(mapping, index, end);
2158         done_index = index;
2159         while (!done && (index <= end)) {
2160                 int i;
2161
2162                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2163                                 tag);
2164                 if (nr_pages == 0)
2165                         break;
2166
2167                 for (i = 0; i < nr_pages; i++) {
2168                         struct page *page = pvec.pages[i];
2169                         bool submitted = false;
2170
2171                         /* give a priority to WB_SYNC threads */
2172                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2173                                         wbc->sync_mode == WB_SYNC_NONE) {
2174                                 done = 1;
2175                                 break;
2176                         }
2177
2178                         done_index = page->index;
2179 retry_write:
2180                         lock_page(page);
2181
2182                         if (unlikely(page->mapping != mapping)) {
2183 continue_unlock:
2184                                 unlock_page(page);
2185                                 continue;
2186                         }
2187
2188                         if (!PageDirty(page)) {
2189                                 /* someone wrote it for us */
2190                                 goto continue_unlock;
2191                         }
2192
2193                         if (PageWriteback(page)) {
2194                                 if (wbc->sync_mode != WB_SYNC_NONE)
2195                                         f2fs_wait_on_page_writeback(page,
2196                                                         DATA, true, true);
2197                                 else
2198                                         goto continue_unlock;
2199                         }
2200
2201                         if (!clear_page_dirty_for_io(page))
2202                                 goto continue_unlock;
2203
2204                         ret = __write_data_page(page, &submitted, wbc, io_type);
2205                         if (unlikely(ret)) {
2206                                 /*
2207                                  * keep nr_to_write, since vfs uses this to
2208                                  * get # of written pages.
2209                                  */
2210                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2211                                         unlock_page(page);
2212                                         ret = 0;
2213                                         continue;
2214                                 } else if (ret == -EAGAIN) {
2215                                         ret = 0;
2216                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2217                                                 cond_resched();
2218                                                 congestion_wait(BLK_RW_ASYNC,
2219                                                                         HZ/50);
2220                                                 goto retry_write;
2221                                         }
2222                                         continue;
2223                                 }
2224                                 done_index = page->index + 1;
2225                                 done = 1;
2226                                 break;
2227                         } else if (submitted) {
2228                                 nwritten++;
2229                         }
2230
2231                         if (--wbc->nr_to_write <= 0 &&
2232                                         wbc->sync_mode == WB_SYNC_NONE) {
2233                                 done = 1;
2234                                 break;
2235                         }
2236                 }
2237                 pagevec_release(&pvec);
2238                 cond_resched();
2239         }
2240
2241         if (!cycled && !done) {
2242                 cycled = 1;
2243                 index = 0;
2244                 end = writeback_index - 1;
2245                 goto retry;
2246         }
2247         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2248                 mapping->writeback_index = done_index;
2249
2250         if (nwritten)
2251                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2252                                                                 NULL, 0, DATA);
2253
2254         return ret;
2255 }
2256
2257 static inline bool __should_serialize_io(struct inode *inode,
2258                                         struct writeback_control *wbc)
2259 {
2260         if (!S_ISREG(inode->i_mode))
2261                 return false;
2262         if (IS_NOQUOTA(inode))
2263                 return false;
2264         if (wbc->sync_mode != WB_SYNC_ALL)
2265                 return true;
2266         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2267                 return true;
2268         return false;
2269 }
2270
2271 static int __f2fs_write_data_pages(struct address_space *mapping,
2272                                                 struct writeback_control *wbc,
2273                                                 enum iostat_type io_type)
2274 {
2275         struct inode *inode = mapping->host;
2276         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2277         struct blk_plug plug;
2278         int ret;
2279         bool locked = false;
2280
2281         /* deal with chardevs and other special file */
2282         if (!mapping->a_ops->writepage)
2283                 return 0;
2284
2285         /* skip writing if there is no dirty page in this inode */
2286         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2287                 return 0;
2288
2289         /* during POR, we don't need to trigger writepage at all. */
2290         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2291                 goto skip_write;
2292
2293         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2294                         wbc->sync_mode == WB_SYNC_NONE &&
2295                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2296                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2297                 goto skip_write;
2298
2299         /* skip writing during file defragment */
2300         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2301                 goto skip_write;
2302
2303         trace_f2fs_writepages(mapping->host, wbc, DATA);
2304
2305         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2306         if (wbc->sync_mode == WB_SYNC_ALL)
2307                 atomic_inc(&sbi->wb_sync_req[DATA]);
2308         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2309                 goto skip_write;
2310
2311         if (__should_serialize_io(inode, wbc)) {
2312                 mutex_lock(&sbi->writepages);
2313                 locked = true;
2314         }
2315
2316         blk_start_plug(&plug);
2317         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2318         blk_finish_plug(&plug);
2319
2320         if (locked)
2321                 mutex_unlock(&sbi->writepages);
2322
2323         if (wbc->sync_mode == WB_SYNC_ALL)
2324                 atomic_dec(&sbi->wb_sync_req[DATA]);
2325         /*
2326          * if some pages were truncated, we cannot guarantee its mapping->host
2327          * to detect pending bios.
2328          */
2329
2330         f2fs_remove_dirty_inode(inode);
2331         return ret;
2332
2333 skip_write:
2334         wbc->pages_skipped += get_dirty_pages(inode);
2335         trace_f2fs_writepages(mapping->host, wbc, DATA);
2336         return 0;
2337 }
2338
2339 static int f2fs_write_data_pages(struct address_space *mapping,
2340                             struct writeback_control *wbc)
2341 {
2342         struct inode *inode = mapping->host;
2343
2344         return __f2fs_write_data_pages(mapping, wbc,
2345                         F2FS_I(inode)->cp_task == current ?
2346                         FS_CP_DATA_IO : FS_DATA_IO);
2347 }
2348
2349 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2350 {
2351         struct inode *inode = mapping->host;
2352         loff_t i_size = i_size_read(inode);
2353
2354         if (to > i_size) {
2355                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2356                 down_write(&F2FS_I(inode)->i_mmap_sem);
2357
2358                 truncate_pagecache(inode, i_size);
2359                 if (!IS_NOQUOTA(inode))
2360                         f2fs_truncate_blocks(inode, i_size, true);
2361
2362                 up_write(&F2FS_I(inode)->i_mmap_sem);
2363                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2364         }
2365 }
2366
2367 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2368                         struct page *page, loff_t pos, unsigned len,
2369                         block_t *blk_addr, bool *node_changed)
2370 {
2371         struct inode *inode = page->mapping->host;
2372         pgoff_t index = page->index;
2373         struct dnode_of_data dn;
2374         struct page *ipage;
2375         bool locked = false;
2376         struct extent_info ei = {0,0,0};
2377         int err = 0;
2378         int flag;
2379
2380         /*
2381          * we already allocated all the blocks, so we don't need to get
2382          * the block addresses when there is no need to fill the page.
2383          */
2384         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2385                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2386                 return 0;
2387
2388         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2389         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2390                 flag = F2FS_GET_BLOCK_DEFAULT;
2391         else
2392                 flag = F2FS_GET_BLOCK_PRE_AIO;
2393
2394         if (f2fs_has_inline_data(inode) ||
2395                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2396                 __do_map_lock(sbi, flag, true);
2397                 locked = true;
2398         }
2399 restart:
2400         /* check inline_data */
2401         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2402         if (IS_ERR(ipage)) {
2403                 err = PTR_ERR(ipage);
2404                 goto unlock_out;
2405         }
2406
2407         set_new_dnode(&dn, inode, ipage, ipage, 0);
2408
2409         if (f2fs_has_inline_data(inode)) {
2410                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2411                         f2fs_do_read_inline_data(page, ipage);
2412                         set_inode_flag(inode, FI_DATA_EXIST);
2413                         if (inode->i_nlink)
2414                                 set_inline_node(ipage);
2415                 } else {
2416                         err = f2fs_convert_inline_page(&dn, page);
2417                         if (err)
2418                                 goto out;
2419                         if (dn.data_blkaddr == NULL_ADDR)
2420                                 err = f2fs_get_block(&dn, index);
2421                 }
2422         } else if (locked) {
2423                 err = f2fs_get_block(&dn, index);
2424         } else {
2425                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2426                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2427                 } else {
2428                         /* hole case */
2429                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2430                         if (err || dn.data_blkaddr == NULL_ADDR) {
2431                                 f2fs_put_dnode(&dn);
2432                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2433                                                                 true);
2434                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2435                                 locked = true;
2436                                 goto restart;
2437                         }
2438                 }
2439         }
2440
2441         /* convert_inline_page can make node_changed */
2442         *blk_addr = dn.data_blkaddr;
2443         *node_changed = dn.node_changed;
2444 out:
2445         f2fs_put_dnode(&dn);
2446 unlock_out:
2447         if (locked)
2448                 __do_map_lock(sbi, flag, false);
2449         return err;
2450 }
2451
2452 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2453                 loff_t pos, unsigned len, unsigned flags,
2454                 struct page **pagep, void **fsdata)
2455 {
2456         struct inode *inode = mapping->host;
2457         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2458         struct page *page = NULL;
2459         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2460         bool need_balance = false, drop_atomic = false;
2461         block_t blkaddr = NULL_ADDR;
2462         int err = 0;
2463
2464         trace_f2fs_write_begin(inode, pos, len, flags);
2465
2466         err = f2fs_is_checkpoint_ready(sbi);
2467         if (err)
2468                 goto fail;
2469
2470         if ((f2fs_is_atomic_file(inode) &&
2471                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2472                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2473                 err = -ENOMEM;
2474                 drop_atomic = true;
2475                 goto fail;
2476         }
2477
2478         /*
2479          * We should check this at this moment to avoid deadlock on inode page
2480          * and #0 page. The locking rule for inline_data conversion should be:
2481          * lock_page(page #0) -> lock_page(inode_page)
2482          */
2483         if (index != 0) {
2484                 err = f2fs_convert_inline_inode(inode);
2485                 if (err)
2486                         goto fail;
2487         }
2488 repeat:
2489         /*
2490          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2491          * wait_for_stable_page. Will wait that below with our IO control.
2492          */
2493         page = f2fs_pagecache_get_page(mapping, index,
2494                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2495         if (!page) {
2496                 err = -ENOMEM;
2497                 goto fail;
2498         }
2499
2500         *pagep = page;
2501
2502         err = prepare_write_begin(sbi, page, pos, len,
2503                                         &blkaddr, &need_balance);
2504         if (err)
2505                 goto fail;
2506
2507         if (need_balance && !IS_NOQUOTA(inode) &&
2508                         has_not_enough_free_secs(sbi, 0, 0)) {
2509                 unlock_page(page);
2510                 f2fs_balance_fs(sbi, true);
2511                 lock_page(page);
2512                 if (page->mapping != mapping) {
2513                         /* The page got truncated from under us */
2514                         f2fs_put_page(page, 1);
2515                         goto repeat;
2516                 }
2517         }
2518
2519         f2fs_wait_on_page_writeback(page, DATA, false, true);
2520
2521         if (len == PAGE_SIZE || PageUptodate(page))
2522                 return 0;
2523
2524         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2525                 zero_user_segment(page, len, PAGE_SIZE);
2526                 return 0;
2527         }
2528
2529         if (blkaddr == NEW_ADDR) {
2530                 zero_user_segment(page, 0, PAGE_SIZE);
2531                 SetPageUptodate(page);
2532         } else {
2533                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2534                                 DATA_GENERIC_ENHANCE_READ)) {
2535                         err = -EFAULT;
2536                         goto fail;
2537                 }
2538                 err = f2fs_submit_page_read(inode, page, blkaddr);
2539                 if (err)
2540                         goto fail;
2541
2542                 lock_page(page);
2543                 if (unlikely(page->mapping != mapping)) {
2544                         f2fs_put_page(page, 1);
2545                         goto repeat;
2546                 }
2547                 if (unlikely(!PageUptodate(page))) {
2548                         err = -EIO;
2549                         goto fail;
2550                 }
2551         }
2552         return 0;
2553
2554 fail:
2555         f2fs_put_page(page, 1);
2556         f2fs_write_failed(mapping, pos + len);
2557         if (drop_atomic)
2558                 f2fs_drop_inmem_pages_all(sbi, false);
2559         return err;
2560 }
2561
2562 static int f2fs_write_end(struct file *file,
2563                         struct address_space *mapping,
2564                         loff_t pos, unsigned len, unsigned copied,
2565                         struct page *page, void *fsdata)
2566 {
2567         struct inode *inode = page->mapping->host;
2568
2569         trace_f2fs_write_end(inode, pos, len, copied);
2570
2571         /*
2572          * This should be come from len == PAGE_SIZE, and we expect copied
2573          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2574          * let generic_perform_write() try to copy data again through copied=0.
2575          */
2576         if (!PageUptodate(page)) {
2577                 if (unlikely(copied != len))
2578                         copied = 0;
2579                 else
2580                         SetPageUptodate(page);
2581         }
2582         if (!copied)
2583                 goto unlock_out;
2584
2585         set_page_dirty(page);
2586
2587         if (pos + copied > i_size_read(inode))
2588                 f2fs_i_size_write(inode, pos + copied);
2589 unlock_out:
2590         f2fs_put_page(page, 1);
2591         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2592         return copied;
2593 }
2594
2595 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2596                            loff_t offset)
2597 {
2598         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2599         unsigned blkbits = i_blkbits;
2600         unsigned blocksize_mask = (1 << blkbits) - 1;
2601         unsigned long align = offset | iov_iter_alignment(iter);
2602         struct block_device *bdev = inode->i_sb->s_bdev;
2603
2604         if (align & blocksize_mask) {
2605                 if (bdev)
2606                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2607                 blocksize_mask = (1 << blkbits) - 1;
2608                 if (align & blocksize_mask)
2609                         return -EINVAL;
2610                 return 1;
2611         }
2612         return 0;
2613 }
2614
2615 static void f2fs_dio_end_io(struct bio *bio)
2616 {
2617         struct f2fs_private_dio *dio = bio->bi_private;
2618
2619         dec_page_count(F2FS_I_SB(dio->inode),
2620                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2621
2622         bio->bi_private = dio->orig_private;
2623         bio->bi_end_io = dio->orig_end_io;
2624
2625         kvfree(dio);
2626
2627         bio_endio(bio);
2628 }
2629
2630 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2631                                                         loff_t file_offset)
2632 {
2633         struct f2fs_private_dio *dio;
2634         bool write = (bio_op(bio) == REQ_OP_WRITE);
2635
2636         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2637                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2638         if (!dio)
2639                 goto out;
2640
2641         dio->inode = inode;
2642         dio->orig_end_io = bio->bi_end_io;
2643         dio->orig_private = bio->bi_private;
2644         dio->write = write;
2645
2646         bio->bi_end_io = f2fs_dio_end_io;
2647         bio->bi_private = dio;
2648
2649         inc_page_count(F2FS_I_SB(inode),
2650                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2651
2652         submit_bio(bio);
2653         return;
2654 out:
2655         bio->bi_status = BLK_STS_IOERR;
2656         bio_endio(bio);
2657 }
2658
2659 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2660 {
2661         struct address_space *mapping = iocb->ki_filp->f_mapping;
2662         struct inode *inode = mapping->host;
2663         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2664         struct f2fs_inode_info *fi = F2FS_I(inode);
2665         size_t count = iov_iter_count(iter);
2666         loff_t offset = iocb->ki_pos;
2667         int rw = iov_iter_rw(iter);
2668         int err;
2669         enum rw_hint hint = iocb->ki_hint;
2670         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2671         bool do_opu;
2672
2673         err = check_direct_IO(inode, iter, offset);
2674         if (err)
2675                 return err < 0 ? err : 0;
2676
2677         if (f2fs_force_buffered_io(inode, iocb, iter))
2678                 return 0;
2679
2680         do_opu = allow_outplace_dio(inode, iocb, iter);
2681
2682         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2683
2684         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2685                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2686
2687         if (iocb->ki_flags & IOCB_NOWAIT) {
2688                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2689                         iocb->ki_hint = hint;
2690                         err = -EAGAIN;
2691                         goto out;
2692                 }
2693                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2694                         up_read(&fi->i_gc_rwsem[rw]);
2695                         iocb->ki_hint = hint;
2696                         err = -EAGAIN;
2697                         goto out;
2698                 }
2699         } else {
2700                 down_read(&fi->i_gc_rwsem[rw]);
2701                 if (do_opu)
2702                         down_read(&fi->i_gc_rwsem[READ]);
2703         }
2704
2705         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2706                         iter, rw == WRITE ? get_data_block_dio_write :
2707                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2708                         DIO_LOCKING | DIO_SKIP_HOLES);
2709
2710         if (do_opu)
2711                 up_read(&fi->i_gc_rwsem[READ]);
2712
2713         up_read(&fi->i_gc_rwsem[rw]);
2714
2715         if (rw == WRITE) {
2716                 if (whint_mode == WHINT_MODE_OFF)
2717                         iocb->ki_hint = hint;
2718                 if (err > 0) {
2719                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2720                                                                         err);
2721                         if (!do_opu)
2722                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2723                 } else if (err < 0) {
2724                         f2fs_write_failed(mapping, offset + count);
2725                 }
2726         }
2727
2728 out:
2729         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2730
2731         return err;
2732 }
2733
2734 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2735                                                         unsigned int length)
2736 {
2737         struct inode *inode = page->mapping->host;
2738         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2739
2740         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2741                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2742                 return;
2743
2744         if (PageDirty(page)) {
2745                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2746                         dec_page_count(sbi, F2FS_DIRTY_META);
2747                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2748                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2749                 } else {
2750                         inode_dec_dirty_pages(inode);
2751                         f2fs_remove_dirty_inode(inode);
2752                 }
2753         }
2754
2755         clear_cold_data(page);
2756
2757         if (IS_ATOMIC_WRITTEN_PAGE(page))
2758                 return f2fs_drop_inmem_page(inode, page);
2759
2760         f2fs_clear_page_private(page);
2761 }
2762
2763 int f2fs_release_page(struct page *page, gfp_t wait)
2764 {
2765         /* If this is dirty page, keep PagePrivate */
2766         if (PageDirty(page))
2767                 return 0;
2768
2769         /* This is atomic written page, keep Private */
2770         if (IS_ATOMIC_WRITTEN_PAGE(page))
2771                 return 0;
2772
2773         clear_cold_data(page);
2774         f2fs_clear_page_private(page);
2775         return 1;
2776 }
2777
2778 static int f2fs_set_data_page_dirty(struct page *page)
2779 {
2780         struct address_space *mapping = page->mapping;
2781         struct inode *inode = mapping->host;
2782
2783         trace_f2fs_set_page_dirty(page, DATA);
2784
2785         if (!PageUptodate(page))
2786                 SetPageUptodate(page);
2787
2788         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2789                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2790                         f2fs_register_inmem_page(inode, page);
2791                         return 1;
2792                 }
2793                 /*
2794                  * Previously, this page has been registered, we just
2795                  * return here.
2796                  */
2797                 return 0;
2798         }
2799
2800         if (!PageDirty(page)) {
2801                 __set_page_dirty_nobuffers(page);
2802                 f2fs_update_dirty_page(inode, page);
2803                 return 1;
2804         }
2805         return 0;
2806 }
2807
2808 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2809 {
2810         struct inode *inode = mapping->host;
2811
2812         if (f2fs_has_inline_data(inode))
2813                 return 0;
2814
2815         /* make sure allocating whole blocks */
2816         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2817                 filemap_write_and_wait(mapping);
2818
2819         return generic_block_bmap(mapping, block, get_data_block_bmap);
2820 }
2821
2822 #ifdef CONFIG_MIGRATION
2823 #include <linux/migrate.h>
2824
2825 int f2fs_migrate_page(struct address_space *mapping,
2826                 struct page *newpage, struct page *page, enum migrate_mode mode)
2827 {
2828         int rc, extra_count;
2829         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2830         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2831
2832         BUG_ON(PageWriteback(page));
2833
2834         /* migrating an atomic written page is safe with the inmem_lock hold */
2835         if (atomic_written) {
2836                 if (mode != MIGRATE_SYNC)
2837                         return -EBUSY;
2838                 if (!mutex_trylock(&fi->inmem_lock))
2839                         return -EAGAIN;
2840         }
2841
2842         /* one extra reference was held for atomic_write page */
2843         extra_count = atomic_written ? 1 : 0;
2844         rc = migrate_page_move_mapping(mapping, newpage,
2845                                 page, mode, extra_count);
2846         if (rc != MIGRATEPAGE_SUCCESS) {
2847                 if (atomic_written)
2848                         mutex_unlock(&fi->inmem_lock);
2849                 return rc;
2850         }
2851
2852         if (atomic_written) {
2853                 struct inmem_pages *cur;
2854                 list_for_each_entry(cur, &fi->inmem_pages, list)
2855                         if (cur->page == page) {
2856                                 cur->page = newpage;
2857                                 break;
2858                         }
2859                 mutex_unlock(&fi->inmem_lock);
2860                 put_page(page);
2861                 get_page(newpage);
2862         }
2863
2864         if (PagePrivate(page)) {
2865                 f2fs_set_page_private(newpage, page_private(page));
2866                 f2fs_clear_page_private(page);
2867         }
2868
2869         if (mode != MIGRATE_SYNC_NO_COPY)
2870                 migrate_page_copy(newpage, page);
2871         else
2872                 migrate_page_states(newpage, page);
2873
2874         return MIGRATEPAGE_SUCCESS;
2875 }
2876 #endif
2877
2878 const struct address_space_operations f2fs_dblock_aops = {
2879         .readpage       = f2fs_read_data_page,
2880         .readpages      = f2fs_read_data_pages,
2881         .writepage      = f2fs_write_data_page,
2882         .writepages     = f2fs_write_data_pages,
2883         .write_begin    = f2fs_write_begin,
2884         .write_end      = f2fs_write_end,
2885         .set_page_dirty = f2fs_set_data_page_dirty,
2886         .invalidatepage = f2fs_invalidate_page,
2887         .releasepage    = f2fs_release_page,
2888         .direct_IO      = f2fs_direct_IO,
2889         .bmap           = f2fs_bmap,
2890 #ifdef CONFIG_MIGRATION
2891         .migratepage    = f2fs_migrate_page,
2892 #endif
2893 };
2894
2895 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2896 {
2897         struct address_space *mapping = page_mapping(page);
2898         unsigned long flags;
2899
2900         xa_lock_irqsave(&mapping->i_pages, flags);
2901         __xa_clear_mark(&mapping->i_pages, page_index(page),
2902                                                 PAGECACHE_TAG_DIRTY);
2903         xa_unlock_irqrestore(&mapping->i_pages, flags);
2904 }
2905
2906 int __init f2fs_init_post_read_processing(void)
2907 {
2908         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2909         if (!bio_post_read_ctx_cache)
2910                 goto fail;
2911         bio_post_read_ctx_pool =
2912                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2913                                          bio_post_read_ctx_cache);
2914         if (!bio_post_read_ctx_pool)
2915                 goto fail_free_cache;
2916         return 0;
2917
2918 fail_free_cache:
2919         kmem_cache_destroy(bio_post_read_ctx_cache);
2920 fail:
2921         return -ENOMEM;
2922 }
2923
2924 void __exit f2fs_destroy_post_read_processing(void)
2925 {
2926         mempool_destroy(bio_post_read_ctx_pool);
2927         kmem_cache_destroy(bio_post_read_ctx_cache);
2928 }