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_pullback_bio_page(&page, true);
 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 bio *bio, 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 (!bio)
 358                 return false;
 359
 360         if (!inode && !page && !ino)
 361                 return true;
 362
 363         bio_for_each_segment_all(bvec, bio, iter_all) {
 364
 365                 if (bvec->bv_page->mapping)
 366                         target = bvec->bv_page;
 367                 else
 368                         target = fscrypt_control_page(bvec->bv_page);
 369
 370                 if (inode && inode == target->mapping->host)
 371                         return true;
 372                 if (page && page == target)
 373                         return true;
 374                 if (ino && ino == ino_of_node(target))
 375                         return true;
 376         }
 377
 378         return false;
 379 }
 380
 381 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 382                                 enum page_type type, enum temp_type temp)
 383 {
 384         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 385         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 386
 387         down_write(&io->io_rwsem);
 388
 389         /* change META to META_FLUSH in the checkpoint procedure */
 390         if (type >= META_FLUSH) {
 391                 io->fio.type = META_FLUSH;
 392                 io->fio.op = REQ_OP_WRITE;
 393                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 394                 if (!test_opt(sbi, NOBARRIER))
 395                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 396         }
 397         __submit_merged_bio(io);
 398         up_write(&io->io_rwsem);
 399 }
 400
 401 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 402                                 struct inode *inode, struct page *page,
 403                                 nid_t ino, enum page_type type, bool force)
 404 {
 405         enum temp_type temp;
 406         bool ret = true;
 407
 408         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 409                 if (!force)     {
 410                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 411                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 412
 413                         down_read(&io->io_rwsem);
 414                         ret = __has_merged_page(io->bio, inode, page, ino);
 415                         up_read(&io->io_rwsem);
 416                 }
 417                 if (ret)
 418                         __f2fs_submit_merged_write(sbi, type, temp);
 419
 420                 /* TODO: use HOT temp only for meta pages now. */
 421                 if (type >= META)
 422                         break;
 423         }
 424 }
 425
 426 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 427 {
 428         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 429 }
 430
 431 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 432                                 struct inode *inode, struct page *page,
 433                                 nid_t ino, enum page_type type)
 434 {
 435         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 436 }
 437
 438 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 439 {
 440         f2fs_submit_merged_write(sbi, DATA);
 441         f2fs_submit_merged_write(sbi, NODE);
 442         f2fs_submit_merged_write(sbi, META);
 443 }
 444
 445 /*
 446  * Fill the locked page with data located in the block address.
 447  * A caller needs to unlock the page on failure.
 448  */
 449 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 450 {
 451         struct bio *bio;
 452         struct page *page = fio->encrypted_page ?
 453                         fio->encrypted_page : fio->page;
 454
 455         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 456                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 457                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 458                 return -EFAULT;
 459
 460         trace_f2fs_submit_page_bio(page, fio);
 461         f2fs_trace_ios(fio, 0);
 462
 463         /* Allocate a new bio */
 464         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 465                                 1, is_read_io(fio->op), fio->type, fio->temp);
 466
 467         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 468                 bio_put(bio);
 469                 return -EFAULT;
 470         }
 471
 472         if (fio->io_wbc && !is_read_io(fio->op))
 473                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 474
 475         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 476
 477         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 478                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 479
 480         __submit_bio(fio->sbi, bio, fio->type);
 481         return 0;
 482 }
 483
 484 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 485 {
 486         struct bio *bio = *fio->bio;
 487         struct page *page = fio->encrypted_page ?
 488                         fio->encrypted_page : fio->page;
 489
 490         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 491                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 492                 return -EFAULT;
 493
 494         trace_f2fs_submit_page_bio(page, fio);
 495         f2fs_trace_ios(fio, 0);
 496
 497         if (bio && (*fio->last_block + 1 != fio->new_blkaddr ||
 498                         !__same_bdev(fio->sbi, fio->new_blkaddr, bio))) {
 499                 __submit_bio(fio->sbi, bio, fio->type);
 500                 bio = NULL;
 501         }
 502 alloc_new:
 503         if (!bio) {
 504                 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 505                                 BIO_MAX_PAGES, false, fio->type, fio->temp);
 506                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 507         }
 508
 509         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 510                 __submit_bio(fio->sbi, bio, fio->type);
 511                 bio = NULL;
 512                 goto alloc_new;
 513         }
 514
 515         if (fio->io_wbc)
 516                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 517
 518         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 519
 520         *fio->last_block = fio->new_blkaddr;
 521         *fio->bio = bio;
 522
 523         return 0;
 524 }
 525
 526 static void f2fs_submit_ipu_bio(struct f2fs_sb_info *sbi, struct bio **bio,
 527                                                         struct page *page)
 528 {
 529         if (!bio)
 530                 return;
 531
 532         if (!__has_merged_page(*bio, NULL, page, 0))
 533                 return;
 534
 535         __submit_bio(sbi, *bio, DATA);
 536         *bio = NULL;
 537 }
 538
 539 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 540 {
 541         struct f2fs_sb_info *sbi = fio->sbi;
 542         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 543         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 544         struct page *bio_page;
 545
 546         f2fs_bug_on(sbi, is_read_io(fio->op));
 547
 548         down_write(&io->io_rwsem);
 549 next:
 550         if (fio->in_list) {
 551                 spin_lock(&io->io_lock);
 552                 if (list_empty(&io->io_list)) {
 553                         spin_unlock(&io->io_lock);
 554                         goto out;
 555                 }
 556                 fio = list_first_entry(&io->io_list,
 557                                                 struct f2fs_io_info, list);
 558                 list_del(&fio->list);
 559                 spin_unlock(&io->io_lock);
 560         }
 561
 562         verify_fio_blkaddr(fio);
 563
 564         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 565
 566         /* set submitted = true as a return value */
 567         fio->submitted = true;
 568
 569         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 570
 571         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 572             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 573                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 574                 __submit_merged_bio(io);
 575 alloc_new:
 576         if (io->bio == NULL) {
 577                 if ((fio->type == DATA || fio->type == NODE) &&
 578                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 579                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 580                         fio->retry = true;
 581                         goto skip;
 582                 }
 583                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 584                                                 BIO_MAX_PAGES, false,
 585                                                 fio->type, fio->temp);
 586                 io->fio = *fio;
 587         }
 588
 589         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 590                 __submit_merged_bio(io);
 591                 goto alloc_new;
 592         }
 593
 594         if (fio->io_wbc)
 595                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 596
 597         io->last_block_in_bio = fio->new_blkaddr;
 598         f2fs_trace_ios(fio, 0);
 599
 600         trace_f2fs_submit_page_write(fio->page, fio);
 601 skip:
 602         if (fio->in_list)
 603                 goto next;
 604 out:
 605         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 606                                 f2fs_is_checkpoint_ready(sbi))
 607                 __submit_merged_bio(io);
 608         up_write(&io->io_rwsem);
 609 }
 610
 611 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 612                                         unsigned nr_pages, unsigned op_flag)
 613 {
 614         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 615         struct bio *bio;
 616         struct bio_post_read_ctx *ctx;
 617         unsigned int post_read_steps = 0;
 618
 619         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 620         if (!bio)
 621                 return ERR_PTR(-ENOMEM);
 622         f2fs_target_device(sbi, blkaddr, bio);
 623         bio->bi_end_io = f2fs_read_end_io;
 624         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 625
 626         if (f2fs_encrypted_file(inode))
 627                 post_read_steps |= 1 << STEP_DECRYPT;
 628         if (post_read_steps) {
 629                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 630                 if (!ctx) {
 631                         bio_put(bio);
 632                         return ERR_PTR(-ENOMEM);
 633                 }
 634                 ctx->bio = bio;
 635                 ctx->enabled_steps = post_read_steps;
 636                 bio->bi_private = ctx;
 637         }
 638
 639         return bio;
 640 }
 641
 642 /* This can handle encryption stuffs */
 643 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 644                                                         block_t blkaddr)
 645 {
 646         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 647         struct bio *bio;
 648
 649         bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 650         if (IS_ERR(bio))
 651                 return PTR_ERR(bio);
 652
 653         /* wait for GCed page writeback via META_MAPPING */
 654         f2fs_wait_on_block_writeback(inode, blkaddr);
 655
 656         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 657                 bio_put(bio);
 658                 return -EFAULT;
 659         }
 660         ClearPageError(page);
 661         inc_page_count(sbi, F2FS_RD_DATA);
 662         __submit_bio(sbi, bio, DATA);
 663         return 0;
 664 }
 665
 666 static void __set_data_blkaddr(struct dnode_of_data *dn)
 667 {
 668         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 669         __le32 *addr_array;
 670         int base = 0;
 671
 672         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 673                 base = get_extra_isize(dn->inode);
 674
 675         /* Get physical address of data block */
 676         addr_array = blkaddr_in_node(rn);
 677         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 678 }
 679
 680 /*
 681  * Lock ordering for the change of data block address:
 682  * ->data_page
 683  *  ->node_page
 684  *    update block addresses in the node page
 685  */
 686 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 687 {
 688         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 689         __set_data_blkaddr(dn);
 690         if (set_page_dirty(dn->node_page))
 691                 dn->node_changed = true;
 692 }
 693
 694 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 695 {
 696         dn->data_blkaddr = blkaddr;
 697         f2fs_set_data_blkaddr(dn);
 698         f2fs_update_extent_cache(dn);
 699 }
 700
 701 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 702 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 703 {
 704         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 705         int err;
 706
 707         if (!count)
 708                 return 0;
 709
 710         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 711                 return -EPERM;
 712         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 713                 return err;
 714
 715         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 716                                                 dn->ofs_in_node, count);
 717
 718         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 719
 720         for (; count > 0; dn->ofs_in_node++) {
 721                 block_t blkaddr = datablock_addr(dn->inode,
 722                                         dn->node_page, dn->ofs_in_node);
 723                 if (blkaddr == NULL_ADDR) {
 724                         dn->data_blkaddr = NEW_ADDR;
 725                         __set_data_blkaddr(dn);
 726                         count--;
 727                 }
 728         }
 729
 730         if (set_page_dirty(dn->node_page))
 731                 dn->node_changed = true;
 732         return 0;
 733 }
 734
 735 /* Should keep dn->ofs_in_node unchanged */
 736 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 737 {
 738         unsigned int ofs_in_node = dn->ofs_in_node;
 739         int ret;
 740
 741         ret = f2fs_reserve_new_blocks(dn, 1);
 742         dn->ofs_in_node = ofs_in_node;
 743         return ret;
 744 }
 745
 746 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 747 {
 748         bool need_put = dn->inode_page ? false : true;
 749         int err;
 750
 751         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 752         if (err)
 753                 return err;
 754
 755         if (dn->data_blkaddr == NULL_ADDR)
 756                 err = f2fs_reserve_new_block(dn);
 757         if (err || need_put)
 758                 f2fs_put_dnode(dn);
 759         return err;
 760 }
 761
 762 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 763 {
 764         struct extent_info ei  = {0,0,0};
 765         struct inode *inode = dn->inode;
 766
 767         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 768                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 769                 return 0;
 770         }
 771
 772         return f2fs_reserve_block(dn, index);
 773 }
 774
 775 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 776                                                 int op_flags, bool for_write)
 777 {
 778         struct address_space *mapping = inode->i_mapping;
 779         struct dnode_of_data dn;
 780         struct page *page;
 781         struct extent_info ei = {0,0,0};
 782         int err;
 783
 784         page = f2fs_grab_cache_page(mapping, index, for_write);
 785         if (!page)
 786                 return ERR_PTR(-ENOMEM);
 787
 788         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 789                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 790                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
 791                                                 DATA_GENERIC_ENHANCE_READ)) {
 792                         err = -EFAULT;
 793                         goto put_err;
 794                 }
 795                 goto got_it;
 796         }
 797
 798         set_new_dnode(&dn, inode, NULL, NULL, 0);
 799         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 800         if (err)
 801                 goto put_err;
 802         f2fs_put_dnode(&dn);
 803
 804         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 805                 err = -ENOENT;
 806                 goto put_err;
 807         }
 808         if (dn.data_blkaddr != NEW_ADDR &&
 809                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
 810                                                 dn.data_blkaddr,
 811                                                 DATA_GENERIC_ENHANCE)) {
 812                 err = -EFAULT;
 813                 goto put_err;
 814         }
 815 got_it:
 816         if (PageUptodate(page)) {
 817                 unlock_page(page);
 818                 return page;
 819         }
 820
 821         /*
 822          * A new dentry page is allocated but not able to be written, since its
 823          * new inode page couldn't be allocated due to -ENOSPC.
 824          * In such the case, its blkaddr can be remained as NEW_ADDR.
 825          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 826          * f2fs_init_inode_metadata.
 827          */
 828         if (dn.data_blkaddr == NEW_ADDR) {
 829                 zero_user_segment(page, 0, PAGE_SIZE);
 830                 if (!PageUptodate(page))
 831                         SetPageUptodate(page);
 832                 unlock_page(page);
 833                 return page;
 834         }
 835
 836         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 837         if (err)
 838                 goto put_err;
 839         return page;
 840
 841 put_err:
 842         f2fs_put_page(page, 1);
 843         return ERR_PTR(err);
 844 }
 845
 846 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 847 {
 848         struct address_space *mapping = inode->i_mapping;
 849         struct page *page;
 850
 851         page = find_get_page(mapping, index);
 852         if (page && PageUptodate(page))
 853                 return page;
 854         f2fs_put_page(page, 0);
 855
 856         page = f2fs_get_read_data_page(inode, index, 0, false);
 857         if (IS_ERR(page))
 858                 return page;
 859
 860         if (PageUptodate(page))
 861                 return page;
 862
 863         wait_on_page_locked(page);
 864         if (unlikely(!PageUptodate(page))) {
 865                 f2fs_put_page(page, 0);
 866                 return ERR_PTR(-EIO);
 867         }
 868         return page;
 869 }
 870
 871 /*
 872  * If it tries to access a hole, return an error.
 873  * Because, the callers, functions in dir.c and GC, should be able to know
 874  * whether this page exists or not.
 875  */
 876 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 877                                                         bool for_write)
 878 {
 879         struct address_space *mapping = inode->i_mapping;
 880         struct page *page;
 881 repeat:
 882         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 883         if (IS_ERR(page))
 884                 return page;
 885
 886         /* wait for read completion */
 887         lock_page(page);
 888         if (unlikely(page->mapping != mapping)) {
 889                 f2fs_put_page(page, 1);
 890                 goto repeat;
 891         }
 892         if (unlikely(!PageUptodate(page))) {
 893                 f2fs_put_page(page, 1);
 894                 return ERR_PTR(-EIO);
 895         }
 896         return page;
 897 }
 898
 899 /*
 900  * Caller ensures that this data page is never allocated.
 901  * A new zero-filled data page is allocated in the page cache.
 902  *
 903  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 904  * f2fs_unlock_op().
 905  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 906  * ipage should be released by this function.
 907  */
 908 struct page *f2fs_get_new_data_page(struct inode *inode,
 909                 struct page *ipage, pgoff_t index, bool new_i_size)
 910 {
 911         struct address_space *mapping = inode->i_mapping;
 912         struct page *page;
 913         struct dnode_of_data dn;
 914         int err;
 915
 916         page = f2fs_grab_cache_page(mapping, index, true);
 917         if (!page) {
 918                 /*
 919                  * before exiting, we should make sure ipage will be released
 920                  * if any error occur.
 921                  */
 922                 f2fs_put_page(ipage, 1);
 923                 return ERR_PTR(-ENOMEM);
 924         }
 925
 926         set_new_dnode(&dn, inode, ipage, NULL, 0);
 927         err = f2fs_reserve_block(&dn, index);
 928         if (err) {
 929                 f2fs_put_page(page, 1);
 930                 return ERR_PTR(err);
 931         }
 932         if (!ipage)
 933                 f2fs_put_dnode(&dn);
 934
 935         if (PageUptodate(page))
 936                 goto got_it;
 937
 938         if (dn.data_blkaddr == NEW_ADDR) {
 939                 zero_user_segment(page, 0, PAGE_SIZE);
 940                 if (!PageUptodate(page))
 941                         SetPageUptodate(page);
 942         } else {
 943                 f2fs_put_page(page, 1);
 944
 945                 /* if ipage exists, blkaddr should be NEW_ADDR */
 946                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 947                 page = f2fs_get_lock_data_page(inode, index, true);
 948                 if (IS_ERR(page))
 949                         return page;
 950         }
 951 got_it:
 952         if (new_i_size && i_size_read(inode) <
 953                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 954                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 955         return page;
 956 }
 957
 958 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 959 {
 960         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 961         struct f2fs_summary sum;
 962         struct node_info ni;
 963         block_t old_blkaddr;
 964         blkcnt_t count = 1;
 965         int err;
 966
 967         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 968                 return -EPERM;
 969
 970         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 971         if (err)
 972                 return err;
 973
 974         dn->data_blkaddr = datablock_addr(dn->inode,
 975                                 dn->node_page, dn->ofs_in_node);
 976         if (dn->data_blkaddr != NULL_ADDR)
 977                 goto alloc;
 978
 979         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 980                 return err;
 981
 982 alloc:
 983         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 984         old_blkaddr = dn->data_blkaddr;
 985         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 986                                         &sum, seg_type, NULL, false);
 987         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 988                 invalidate_mapping_pages(META_MAPPING(sbi),
 989                                         old_blkaddr, old_blkaddr);
 990         f2fs_set_data_blkaddr(dn);
 991
 992         /*
 993          * i_size will be updated by direct_IO. Otherwise, we'll get stale
 994          * data from unwritten block via dio_read.
 995          */
 996         return 0;
 997 }
 998
 999 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1000 {
1001         struct inode *inode = file_inode(iocb->ki_filp);
1002         struct f2fs_map_blocks map;
1003         int flag;
1004         int err = 0;
1005         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1006
1007         /* convert inline data for Direct I/O*/
1008         if (direct_io) {
1009                 err = f2fs_convert_inline_inode(inode);
1010                 if (err)
1011                         return err;
1012         }
1013
1014         if (direct_io && allow_outplace_dio(inode, iocb, from))
1015                 return 0;
1016
1017         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
1018                 return 0;
1019
1020         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1021         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1022         if (map.m_len > map.m_lblk)
1023                 map.m_len -= map.m_lblk;
1024         else
1025                 map.m_len = 0;
1026
1027         map.m_next_pgofs = NULL;
1028         map.m_next_extent = NULL;
1029         map.m_seg_type = NO_CHECK_TYPE;
1030         map.m_may_create = true;
1031
1032         if (direct_io) {
1033                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1034                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1035                                         F2FS_GET_BLOCK_PRE_AIO :
1036                                         F2FS_GET_BLOCK_PRE_DIO;
1037                 goto map_blocks;
1038         }
1039         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1040                 err = f2fs_convert_inline_inode(inode);
1041                 if (err)
1042                         return err;
1043         }
1044         if (f2fs_has_inline_data(inode))
1045                 return err;
1046
1047         flag = F2FS_GET_BLOCK_PRE_AIO;
1048
1049 map_blocks:
1050         err = f2fs_map_blocks(inode, &map, 1, flag);
1051         if (map.m_len > 0 && err == -ENOSPC) {
1052                 if (!direct_io)
1053                         set_inode_flag(inode, FI_NO_PREALLOC);
1054                 err = 0;
1055         }
1056         return err;
1057 }
1058
1059 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1060 {
1061         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1062                 if (lock)
1063                         down_read(&sbi->node_change);
1064                 else
1065                         up_read(&sbi->node_change);
1066         } else {
1067                 if (lock)
1068                         f2fs_lock_op(sbi);
1069                 else
1070                         f2fs_unlock_op(sbi);
1071         }
1072 }
1073
1074 /*
1075  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1076  * f2fs_map_blocks structure.
1077  * If original data blocks are allocated, then give them to blockdev.
1078  * Otherwise,
1079  *     a. preallocate requested block addresses
1080  *     b. do not use extent cache for better performance
1081  *     c. give the block addresses to blockdev
1082  */
1083 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1084                                                 int create, int flag)
1085 {
1086         unsigned int maxblocks = map->m_len;
1087         struct dnode_of_data dn;
1088         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1089         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1090         pgoff_t pgofs, end_offset, end;
1091         int err = 0, ofs = 1;
1092         unsigned int ofs_in_node, last_ofs_in_node;
1093         blkcnt_t prealloc;
1094         struct extent_info ei = {0,0,0};
1095         block_t blkaddr;
1096         unsigned int start_pgofs;
1097
1098         if (!maxblocks)
1099                 return 0;
1100
1101         map->m_len = 0;
1102         map->m_flags = 0;
1103
1104         /* it only supports block size == page size */
1105         pgofs = (pgoff_t)map->m_lblk;
1106         end = pgofs + maxblocks;
1107
1108         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1109                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1110                                                         map->m_may_create)
1111                         goto next_dnode;
1112
1113                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1114                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1115                 map->m_flags = F2FS_MAP_MAPPED;
1116                 if (map->m_next_extent)
1117                         *map->m_next_extent = pgofs + map->m_len;
1118
1119                 /* for hardware encryption, but to avoid potential issue in future */
1120                 if (flag == F2FS_GET_BLOCK_DIO)
1121                         f2fs_wait_on_block_writeback_range(inode,
1122                                                 map->m_pblk, map->m_len);
1123                 goto out;
1124         }
1125
1126 next_dnode:
1127         if (map->m_may_create)
1128                 __do_map_lock(sbi, flag, true);
1129
1130         /* When reading holes, we need its node page */
1131         set_new_dnode(&dn, inode, NULL, NULL, 0);
1132         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1133         if (err) {
1134                 if (flag == F2FS_GET_BLOCK_BMAP)
1135                         map->m_pblk = 0;
1136                 if (err == -ENOENT) {
1137                         err = 0;
1138                         if (map->m_next_pgofs)
1139                                 *map->m_next_pgofs =
1140                                         f2fs_get_next_page_offset(&dn, pgofs);
1141                         if (map->m_next_extent)
1142                                 *map->m_next_extent =
1143                                         f2fs_get_next_page_offset(&dn, pgofs);
1144                 }
1145                 goto unlock_out;
1146         }
1147
1148         start_pgofs = pgofs;
1149         prealloc = 0;
1150         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1151         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1152
1153 next_block:
1154         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1155
1156         if (__is_valid_data_blkaddr(blkaddr) &&
1157                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1158                 err = -EFAULT;
1159                 goto sync_out;
1160         }
1161
1162         if (__is_valid_data_blkaddr(blkaddr)) {
1163                 /* use out-place-update for driect IO under LFS mode */
1164                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1165                                                         map->m_may_create) {
1166                         err = __allocate_data_block(&dn, map->m_seg_type);
1167                         if (!err) {
1168                                 blkaddr = dn.data_blkaddr;
1169                                 set_inode_flag(inode, FI_APPEND_WRITE);
1170                         }
1171                 }
1172         } else {
1173                 if (create) {
1174                         if (unlikely(f2fs_cp_error(sbi))) {
1175                                 err = -EIO;
1176                                 goto sync_out;
1177                         }
1178                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1179                                 if (blkaddr == NULL_ADDR) {
1180                                         prealloc++;
1181                                         last_ofs_in_node = dn.ofs_in_node;
1182                                 }
1183                         } else {
1184                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1185                                         flag != F2FS_GET_BLOCK_DIO);
1186                                 err = __allocate_data_block(&dn,
1187                                                         map->m_seg_type);
1188                                 if (!err)
1189                                         set_inode_flag(inode, FI_APPEND_WRITE);
1190                         }
1191                         if (err)
1192                                 goto sync_out;
1193                         map->m_flags |= F2FS_MAP_NEW;
1194                         blkaddr = dn.data_blkaddr;
1195                 } else {
1196                         if (flag == F2FS_GET_BLOCK_BMAP) {
1197                                 map->m_pblk = 0;
1198                                 goto sync_out;
1199                         }
1200                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1201                                 goto sync_out;
1202                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1203                                                 blkaddr == NULL_ADDR) {
1204                                 if (map->m_next_pgofs)
1205                                         *map->m_next_pgofs = pgofs + 1;
1206                                 goto sync_out;
1207                         }
1208                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1209                                 /* for defragment case */
1210                                 if (map->m_next_pgofs)
1211                                         *map->m_next_pgofs = pgofs + 1;
1212                                 goto sync_out;
1213                         }
1214                 }
1215         }
1216
1217         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1218                 goto skip;
1219
1220         if (map->m_len == 0) {
1221                 /* preallocated unwritten block should be mapped for fiemap. */
1222                 if (blkaddr == NEW_ADDR)
1223                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1224                 map->m_flags |= F2FS_MAP_MAPPED;
1225
1226                 map->m_pblk = blkaddr;
1227                 map->m_len = 1;
1228         } else if ((map->m_pblk != NEW_ADDR &&
1229                         blkaddr == (map->m_pblk + ofs)) ||
1230                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1231                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1232                 ofs++;
1233                 map->m_len++;
1234         } else {
1235                 goto sync_out;
1236         }
1237
1238 skip:
1239         dn.ofs_in_node++;
1240         pgofs++;
1241
1242         /* preallocate blocks in batch for one dnode page */
1243         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1244                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1245
1246                 dn.ofs_in_node = ofs_in_node;
1247                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1248                 if (err)
1249                         goto sync_out;
1250
1251                 map->m_len += dn.ofs_in_node - ofs_in_node;
1252                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1253                         err = -ENOSPC;
1254                         goto sync_out;
1255                 }
1256                 dn.ofs_in_node = end_offset;
1257         }
1258
1259         if (pgofs >= end)
1260                 goto sync_out;
1261         else if (dn.ofs_in_node < end_offset)
1262                 goto next_block;
1263
1264         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1265                 if (map->m_flags & F2FS_MAP_MAPPED) {
1266                         unsigned int ofs = start_pgofs - map->m_lblk;
1267
1268                         f2fs_update_extent_cache_range(&dn,
1269                                 start_pgofs, map->m_pblk + ofs,
1270                                 map->m_len - ofs);
1271                 }
1272         }
1273
1274         f2fs_put_dnode(&dn);
1275
1276         if (map->m_may_create) {
1277                 __do_map_lock(sbi, flag, false);
1278                 f2fs_balance_fs(sbi, dn.node_changed);
1279         }
1280         goto next_dnode;
1281
1282 sync_out:
1283
1284         /* for hardware encryption, but to avoid potential issue in future */
1285         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1286                 f2fs_wait_on_block_writeback_range(inode,
1287                                                 map->m_pblk, map->m_len);
1288
1289         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1290                 if (map->m_flags & F2FS_MAP_MAPPED) {
1291                         unsigned int ofs = start_pgofs - map->m_lblk;
1292
1293                         f2fs_update_extent_cache_range(&dn,
1294                                 start_pgofs, map->m_pblk + ofs,
1295                                 map->m_len - ofs);
1296                 }
1297                 if (map->m_next_extent)
1298                         *map->m_next_extent = pgofs + 1;
1299         }
1300         f2fs_put_dnode(&dn);
1301 unlock_out:
1302         if (map->m_may_create) {
1303                 __do_map_lock(sbi, flag, false);
1304                 f2fs_balance_fs(sbi, dn.node_changed);
1305         }
1306 out:
1307         trace_f2fs_map_blocks(inode, map, err);
1308         return err;
1309 }
1310
1311 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1312 {
1313         struct f2fs_map_blocks map;
1314         block_t last_lblk;
1315         int err;
1316
1317         if (pos + len > i_size_read(inode))
1318                 return false;
1319
1320         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1321         map.m_next_pgofs = NULL;
1322         map.m_next_extent = NULL;
1323         map.m_seg_type = NO_CHECK_TYPE;
1324         map.m_may_create = false;
1325         last_lblk = F2FS_BLK_ALIGN(pos + len);
1326
1327         while (map.m_lblk < last_lblk) {
1328                 map.m_len = last_lblk - map.m_lblk;
1329                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1330                 if (err || map.m_len == 0)
1331                         return false;
1332                 map.m_lblk += map.m_len;
1333         }
1334         return true;
1335 }
1336
1337 static int __get_data_block(struct inode *inode, sector_t iblock,
1338                         struct buffer_head *bh, int create, int flag,
1339                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1340 {
1341         struct f2fs_map_blocks map;
1342         int err;
1343
1344         map.m_lblk = iblock;
1345         map.m_len = bh->b_size >> inode->i_blkbits;
1346         map.m_next_pgofs = next_pgofs;
1347         map.m_next_extent = NULL;
1348         map.m_seg_type = seg_type;
1349         map.m_may_create = may_write;
1350
1351         err = f2fs_map_blocks(inode, &map, create, flag);
1352         if (!err) {
1353                 map_bh(bh, inode->i_sb, map.m_pblk);
1354                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1355                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1356         }
1357         return err;
1358 }
1359
1360 static int get_data_block(struct inode *inode, sector_t iblock,
1361                         struct buffer_head *bh_result, int create, int flag,
1362                         pgoff_t *next_pgofs)
1363 {
1364         return __get_data_block(inode, iblock, bh_result, create,
1365                                                         flag, next_pgofs,
1366                                                         NO_CHECK_TYPE, create);
1367 }
1368
1369 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1370                         struct buffer_head *bh_result, int create)
1371 {
1372         return __get_data_block(inode, iblock, bh_result, create,
1373                                 F2FS_GET_BLOCK_DIO, NULL,
1374                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1375                                 true);
1376 }
1377
1378 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1379                         struct buffer_head *bh_result, int create)
1380 {
1381         return __get_data_block(inode, iblock, bh_result, create,
1382                                 F2FS_GET_BLOCK_DIO, NULL,
1383                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1384                                 false);
1385 }
1386
1387 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1388                         struct buffer_head *bh_result, int create)
1389 {
1390         /* Block number less than F2FS MAX BLOCKS */
1391         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1392                 return -EFBIG;
1393
1394         return __get_data_block(inode, iblock, bh_result, create,
1395                                                 F2FS_GET_BLOCK_BMAP, NULL,
1396                                                 NO_CHECK_TYPE, create);
1397 }
1398
1399 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1400 {
1401         return (offset >> inode->i_blkbits);
1402 }
1403
1404 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1405 {
1406         return (blk << inode->i_blkbits);
1407 }
1408
1409 static int f2fs_xattr_fiemap(struct inode *inode,
1410                                 struct fiemap_extent_info *fieinfo)
1411 {
1412         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1413         struct page *page;
1414         struct node_info ni;
1415         __u64 phys = 0, len;
1416         __u32 flags;
1417         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1418         int err = 0;
1419
1420         if (f2fs_has_inline_xattr(inode)) {
1421                 int offset;
1422
1423                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1424                                                 inode->i_ino, false);
1425                 if (!page)
1426                         return -ENOMEM;
1427
1428                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1429                 if (err) {
1430                         f2fs_put_page(page, 1);
1431                         return err;
1432                 }
1433
1434                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1435                 offset = offsetof(struct f2fs_inode, i_addr) +
1436                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1437                                         get_inline_xattr_addrs(inode));
1438
1439                 phys += offset;
1440                 len = inline_xattr_size(inode);
1441
1442                 f2fs_put_page(page, 1);
1443
1444                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1445
1446                 if (!xnid)
1447                         flags |= FIEMAP_EXTENT_LAST;
1448
1449                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1450                 if (err || err == 1)
1451                         return err;
1452         }
1453
1454         if (xnid) {
1455                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1456                 if (!page)
1457                         return -ENOMEM;
1458
1459                 err = f2fs_get_node_info(sbi, xnid, &ni);
1460                 if (err) {
1461                         f2fs_put_page(page, 1);
1462                         return err;
1463                 }
1464
1465                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1466                 len = inode->i_sb->s_blocksize;
1467
1468                 f2fs_put_page(page, 1);
1469
1470                 flags = FIEMAP_EXTENT_LAST;
1471         }
1472
1473         if (phys)
1474                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1475
1476         return (err < 0 ? err : 0);
1477 }
1478
1479 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1480                 u64 start, u64 len)
1481 {
1482         struct buffer_head map_bh;
1483         sector_t start_blk, last_blk;
1484         pgoff_t next_pgofs;
1485         u64 logical = 0, phys = 0, size = 0;
1486         u32 flags = 0;
1487         int ret = 0;
1488
1489         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1490                 ret = f2fs_precache_extents(inode);
1491                 if (ret)
1492                         return ret;
1493         }
1494
1495         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1496         if (ret)
1497                 return ret;
1498
1499         inode_lock(inode);
1500
1501         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1502                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1503                 goto out;
1504         }
1505
1506         if (f2fs_has_inline_data(inode)) {
1507                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1508                 if (ret != -EAGAIN)
1509                         goto out;
1510         }
1511
1512         if (logical_to_blk(inode, len) == 0)
1513                 len = blk_to_logical(inode, 1);
1514
1515         start_blk = logical_to_blk(inode, start);
1516         last_blk = logical_to_blk(inode, start + len - 1);
1517
1518 next:
1519         memset(&map_bh, 0, sizeof(struct buffer_head));
1520         map_bh.b_size = len;
1521
1522         ret = get_data_block(inode, start_blk, &map_bh, 0,
1523                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1524         if (ret)
1525                 goto out;
1526
1527         /* HOLE */
1528         if (!buffer_mapped(&map_bh)) {
1529                 start_blk = next_pgofs;
1530
1531                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1532                                         F2FS_I_SB(inode)->max_file_blocks))
1533                         goto prep_next;
1534
1535                 flags |= FIEMAP_EXTENT_LAST;
1536         }
1537
1538         if (size) {
1539                 if (IS_ENCRYPTED(inode))
1540                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1541
1542                 ret = fiemap_fill_next_extent(fieinfo, logical,
1543                                 phys, size, flags);
1544         }
1545
1546         if (start_blk > last_blk || ret)
1547                 goto out;
1548
1549         logical = blk_to_logical(inode, start_blk);
1550         phys = blk_to_logical(inode, map_bh.b_blocknr);
1551         size = map_bh.b_size;
1552         flags = 0;
1553         if (buffer_unwritten(&map_bh))
1554                 flags = FIEMAP_EXTENT_UNWRITTEN;
1555
1556         start_blk += logical_to_blk(inode, size);
1557
1558 prep_next:
1559         cond_resched();
1560         if (fatal_signal_pending(current))
1561                 ret = -EINTR;
1562         else
1563                 goto next;
1564 out:
1565         if (ret == 1)
1566                 ret = 0;
1567
1568         inode_unlock(inode);
1569         return ret;
1570 }
1571
1572 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1573                                         unsigned nr_pages,
1574                                         struct f2fs_map_blocks *map,
1575                                         struct bio **bio_ret,
1576                                         sector_t *last_block_in_bio,
1577                                         bool is_readahead)
1578 {
1579         struct bio *bio = *bio_ret;
1580         const unsigned blkbits = inode->i_blkbits;
1581         const unsigned blocksize = 1 << blkbits;
1582         sector_t block_in_file;
1583         sector_t last_block;
1584         sector_t last_block_in_file;
1585         sector_t block_nr;
1586         int ret = 0;
1587
1588         block_in_file = (sector_t)page->index;
1589         last_block = block_in_file + nr_pages;
1590         last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1591                                                         blkbits;
1592         if (last_block > last_block_in_file)
1593                 last_block = last_block_in_file;
1594
1595         /* just zeroing out page which is beyond EOF */
1596         if (block_in_file >= last_block)
1597                 goto zero_out;
1598         /*
1599          * Map blocks using the previous result first.
1600          */
1601         if ((map->m_flags & F2FS_MAP_MAPPED) &&
1602                         block_in_file > map->m_lblk &&
1603                         block_in_file < (map->m_lblk + map->m_len))
1604                 goto got_it;
1605
1606         /*
1607          * Then do more f2fs_map_blocks() calls until we are
1608          * done with this page.
1609          */
1610         map->m_lblk = block_in_file;
1611         map->m_len = last_block - block_in_file;
1612
1613         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1614         if (ret)
1615                 goto out;
1616 got_it:
1617         if ((map->m_flags & F2FS_MAP_MAPPED)) {
1618                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1619                 SetPageMappedToDisk(page);
1620
1621                 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1622                         SetPageUptodate(page);
1623                         goto confused;
1624                 }
1625
1626                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1627                                                 DATA_GENERIC_ENHANCE_READ)) {
1628                         ret = -EFAULT;
1629                         goto out;
1630                 }
1631         } else {
1632 zero_out:
1633                 zero_user_segment(page, 0, PAGE_SIZE);
1634                 if (!PageUptodate(page))
1635                         SetPageUptodate(page);
1636                 unlock_page(page);
1637                 goto out;
1638         }
1639
1640         /*
1641          * This page will go to BIO.  Do we need to send this
1642          * BIO off first?
1643          */
1644         if (bio && (*last_block_in_bio != block_nr - 1 ||
1645                 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1646 submit_and_realloc:
1647                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1648                 bio = NULL;
1649         }
1650         if (bio == NULL) {
1651                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1652                                 is_readahead ? REQ_RAHEAD : 0);
1653                 if (IS_ERR(bio)) {
1654                         ret = PTR_ERR(bio);
1655                         bio = NULL;
1656                         goto out;
1657                 }
1658         }
1659
1660         /*
1661          * If the page is under writeback, we need to wait for
1662          * its completion to see the correct decrypted data.
1663          */
1664         f2fs_wait_on_block_writeback(inode, block_nr);
1665
1666         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1667                 goto submit_and_realloc;
1668
1669         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1670         ClearPageError(page);
1671         *last_block_in_bio = block_nr;
1672         goto out;
1673 confused:
1674         if (bio) {
1675                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1676                 bio = NULL;
1677         }
1678         unlock_page(page);
1679 out:
1680         *bio_ret = bio;
1681         return ret;
1682 }
1683
1684 /*
1685  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1686  * Major change was from block_size == page_size in f2fs by default.
1687  *
1688  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1689  * this function ever deviates from doing just read-ahead, it should either
1690  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1691  * from read-ahead.
1692  */
1693 static int f2fs_mpage_readpages(struct address_space *mapping,
1694                         struct list_head *pages, struct page *page,
1695                         unsigned nr_pages, bool is_readahead)
1696 {
1697         struct bio *bio = NULL;
1698         sector_t last_block_in_bio = 0;
1699         struct inode *inode = mapping->host;
1700         struct f2fs_map_blocks map;
1701         int ret = 0;
1702
1703         map.m_pblk = 0;
1704         map.m_lblk = 0;
1705         map.m_len = 0;
1706         map.m_flags = 0;
1707         map.m_next_pgofs = NULL;
1708         map.m_next_extent = NULL;
1709         map.m_seg_type = NO_CHECK_TYPE;
1710         map.m_may_create = false;
1711
1712         for (; nr_pages; nr_pages--) {
1713                 if (pages) {
1714                         page = list_last_entry(pages, struct page, lru);
1715
1716                         prefetchw(&page->flags);
1717                         list_del(&page->lru);
1718                         if (add_to_page_cache_lru(page, mapping,
1719                                                   page->index,
1720                                                   readahead_gfp_mask(mapping)))
1721                                 goto next_page;
1722                 }
1723
1724                 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1725                                         &last_block_in_bio, is_readahead);
1726                 if (ret) {
1727                         SetPageError(page);
1728                         zero_user_segment(page, 0, PAGE_SIZE);
1729                         unlock_page(page);
1730                 }
1731 next_page:
1732                 if (pages)
1733                         put_page(page);
1734         }
1735         BUG_ON(pages && !list_empty(pages));
1736         if (bio)
1737                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1738         return pages ? 0 : ret;
1739 }
1740
1741 static int f2fs_read_data_page(struct file *file, struct page *page)
1742 {
1743         struct inode *inode = page->mapping->host;
1744         int ret = -EAGAIN;
1745
1746         trace_f2fs_readpage(page, DATA);
1747
1748         /* If the file has inline data, try to read it directly */
1749         if (f2fs_has_inline_data(inode))
1750                 ret = f2fs_read_inline_data(inode, page);
1751         if (ret == -EAGAIN)
1752                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1753         return ret;
1754 }
1755
1756 static int f2fs_read_data_pages(struct file *file,
1757                         struct address_space *mapping,
1758                         struct list_head *pages, unsigned nr_pages)
1759 {
1760         struct inode *inode = mapping->host;
1761         struct page *page = list_last_entry(pages, struct page, lru);
1762
1763         trace_f2fs_readpages(inode, page, nr_pages);
1764
1765         /* If the file has inline data, skip readpages */
1766         if (f2fs_has_inline_data(inode))
1767                 return 0;
1768
1769         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1770 }
1771
1772 static int encrypt_one_page(struct f2fs_io_info *fio)
1773 {
1774         struct inode *inode = fio->page->mapping->host;
1775         struct page *mpage;
1776         gfp_t gfp_flags = GFP_NOFS;
1777
1778         if (!f2fs_encrypted_file(inode))
1779                 return 0;
1780
1781         /* wait for GCed page writeback via META_MAPPING */
1782         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1783
1784 retry_encrypt:
1785         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1786                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1787         if (IS_ERR(fio->encrypted_page)) {
1788                 /* flush pending IOs and wait for a while in the ENOMEM case */
1789                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1790                         f2fs_flush_merged_writes(fio->sbi);
1791                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1792                         gfp_flags |= __GFP_NOFAIL;
1793                         goto retry_encrypt;
1794                 }
1795                 return PTR_ERR(fio->encrypted_page);
1796         }
1797
1798         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1799         if (mpage) {
1800                 if (PageUptodate(mpage))
1801                         memcpy(page_address(mpage),
1802                                 page_address(fio->encrypted_page), PAGE_SIZE);
1803                 f2fs_put_page(mpage, 1);
1804         }
1805         return 0;
1806 }
1807
1808 static inline bool check_inplace_update_policy(struct inode *inode,
1809                                 struct f2fs_io_info *fio)
1810 {
1811         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1812         unsigned int policy = SM_I(sbi)->ipu_policy;
1813
1814         if (policy & (0x1 << F2FS_IPU_FORCE))
1815                 return true;
1816         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1817                 return true;
1818         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1819                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1820                 return true;
1821         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1822                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1823                 return true;
1824
1825         /*
1826          * IPU for rewrite async pages
1827          */
1828         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1829                         fio && fio->op == REQ_OP_WRITE &&
1830                         !(fio->op_flags & REQ_SYNC) &&
1831                         !IS_ENCRYPTED(inode))
1832                 return true;
1833
1834         /* this is only set during fdatasync */
1835         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1836                         is_inode_flag_set(inode, FI_NEED_IPU))
1837                 return true;
1838
1839         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1840                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1841                 return true;
1842
1843         return false;
1844 }
1845
1846 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1847 {
1848         if (f2fs_is_pinned_file(inode))
1849                 return true;
1850
1851         /* if this is cold file, we should overwrite to avoid fragmentation */
1852         if (file_is_cold(inode))
1853                 return true;
1854
1855         return check_inplace_update_policy(inode, fio);
1856 }
1857
1858 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1859 {
1860         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1861
1862         if (test_opt(sbi, LFS))
1863                 return true;
1864         if (S_ISDIR(inode->i_mode))
1865                 return true;
1866         if (IS_NOQUOTA(inode))
1867                 return true;
1868         if (f2fs_is_atomic_file(inode))
1869                 return true;
1870         if (fio) {
1871                 if (is_cold_data(fio->page))
1872                         return true;
1873                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1874                         return true;
1875                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1876                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1877                         return true;
1878         }
1879         return false;
1880 }
1881
1882 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1883 {
1884         struct inode *inode = fio->page->mapping->host;
1885
1886         if (f2fs_should_update_outplace(inode, fio))
1887                 return false;
1888
1889         return f2fs_should_update_inplace(inode, fio);
1890 }
1891
1892 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1893 {
1894         struct page *page = fio->page;
1895         struct inode *inode = page->mapping->host;
1896         struct dnode_of_data dn;
1897         struct extent_info ei = {0,0,0};
1898         struct node_info ni;
1899         bool ipu_force = false;
1900         int err = 0;
1901
1902         set_new_dnode(&dn, inode, NULL, NULL, 0);
1903         if (need_inplace_update(fio) &&
1904                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1905                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1906
1907                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1908                                                 DATA_GENERIC_ENHANCE))
1909                         return -EFAULT;
1910
1911                 ipu_force = true;
1912                 fio->need_lock = LOCK_DONE;
1913                 goto got_it;
1914         }
1915
1916         /* Deadlock due to between page->lock and f2fs_lock_op */
1917         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1918                 return -EAGAIN;
1919
1920         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1921         if (err)
1922                 goto out;
1923
1924         fio->old_blkaddr = dn.data_blkaddr;
1925
1926         /* This page is already truncated */
1927         if (fio->old_blkaddr == NULL_ADDR) {
1928                 ClearPageUptodate(page);
1929                 clear_cold_data(page);
1930                 goto out_writepage;
1931         }
1932 got_it:
1933         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1934                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1935                                                 DATA_GENERIC_ENHANCE)) {
1936                 err = -EFAULT;
1937                 goto out_writepage;
1938         }
1939         /*
1940          * If current allocation needs SSR,
1941          * it had better in-place writes for updated data.
1942          */
1943         if (ipu_force ||
1944                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1945                                         need_inplace_update(fio))) {
1946                 err = encrypt_one_page(fio);
1947                 if (err)
1948                         goto out_writepage;
1949
1950                 set_page_writeback(page);
1951                 ClearPageError(page);
1952                 f2fs_put_dnode(&dn);
1953                 if (fio->need_lock == LOCK_REQ)
1954                         f2fs_unlock_op(fio->sbi);
1955                 err = f2fs_inplace_write_data(fio);
1956                 if (err) {
1957                         if (f2fs_encrypted_file(inode))
1958                                 fscrypt_pullback_bio_page(&fio->encrypted_page,
1959                                                                         true);
1960                         if (PageWriteback(page))
1961                                 end_page_writeback(page);
1962                 } else {
1963                         set_inode_flag(inode, FI_UPDATE_WRITE);
1964                 }
1965                 trace_f2fs_do_write_data_page(fio->page, IPU);
1966                 return err;
1967         }
1968
1969         if (fio->need_lock == LOCK_RETRY) {
1970                 if (!f2fs_trylock_op(fio->sbi)) {
1971                         err = -EAGAIN;
1972                         goto out_writepage;
1973                 }
1974                 fio->need_lock = LOCK_REQ;
1975         }
1976
1977         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1978         if (err)
1979                 goto out_writepage;
1980
1981         fio->version = ni.version;
1982
1983         err = encrypt_one_page(fio);
1984         if (err)
1985                 goto out_writepage;
1986
1987         set_page_writeback(page);
1988         ClearPageError(page);
1989
1990         /* LFS mode write path */
1991         f2fs_outplace_write_data(&dn, fio);
1992         trace_f2fs_do_write_data_page(page, OPU);
1993         set_inode_flag(inode, FI_APPEND_WRITE);
1994         if (page->index == 0)
1995                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1996 out_writepage:
1997         f2fs_put_dnode(&dn);
1998 out:
1999         if (fio->need_lock == LOCK_REQ)
2000                 f2fs_unlock_op(fio->sbi);
2001         return err;
2002 }
2003
2004 static int __write_data_page(struct page *page, bool *submitted,
2005                                 struct bio **bio,
2006                                 sector_t *last_block,
2007                                 struct writeback_control *wbc,
2008                                 enum iostat_type io_type)
2009 {
2010         struct inode *inode = page->mapping->host;
2011         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2012         loff_t i_size = i_size_read(inode);
2013         const pgoff_t end_index = ((unsigned long long) i_size)
2014                                                         >> PAGE_SHIFT;
2015         loff_t psize = (page->index + 1) << PAGE_SHIFT;
2016         unsigned offset = 0;
2017         bool need_balance_fs = false;
2018         int err = 0;
2019         struct f2fs_io_info fio = {
2020                 .sbi = sbi,
2021                 .ino = inode->i_ino,
2022                 .type = DATA,
2023                 .op = REQ_OP_WRITE,
2024                 .op_flags = wbc_to_write_flags(wbc),
2025                 .old_blkaddr = NULL_ADDR,
2026                 .page = page,
2027                 .encrypted_page = NULL,
2028                 .submitted = false,
2029                 .need_lock = LOCK_RETRY,
2030                 .io_type = io_type,
2031                 .io_wbc = wbc,
2032                 .bio = bio,
2033                 .last_block = last_block,
2034         };
2035
2036         trace_f2fs_writepage(page, DATA);
2037
2038         /* we should bypass data pages to proceed the kworkder jobs */
2039         if (unlikely(f2fs_cp_error(sbi))) {
2040                 mapping_set_error(page->mapping, -EIO);
2041                 /*
2042                  * don't drop any dirty dentry pages for keeping lastest
2043                  * directory structure.
2044                  */
2045                 if (S_ISDIR(inode->i_mode))
2046                         goto redirty_out;
2047                 goto out;
2048         }
2049
2050         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2051                 goto redirty_out;
2052
2053         if (page->index < end_index)
2054                 goto write;
2055
2056         /*
2057          * If the offset is out-of-range of file size,
2058          * this page does not have to be written to disk.
2059          */
2060         offset = i_size & (PAGE_SIZE - 1);
2061         if ((page->index >= end_index + 1) || !offset)
2062                 goto out;
2063
2064         zero_user_segment(page, offset, PAGE_SIZE);
2065 write:
2066         if (f2fs_is_drop_cache(inode))
2067                 goto out;
2068         /* we should not write 0'th page having journal header */
2069         if (f2fs_is_volatile_file(inode) && (!page->index ||
2070                         (!wbc->for_reclaim &&
2071                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2072                 goto redirty_out;
2073
2074         /* Dentry blocks are controlled by checkpoint */
2075         if (S_ISDIR(inode->i_mode)) {
2076                 fio.need_lock = LOCK_DONE;
2077                 err = f2fs_do_write_data_page(&fio);
2078                 goto done;
2079         }
2080
2081         if (!wbc->for_reclaim)
2082                 need_balance_fs = true;
2083         else if (has_not_enough_free_secs(sbi, 0, 0))
2084                 goto redirty_out;
2085         else
2086                 set_inode_flag(inode, FI_HOT_DATA);
2087
2088         err = -EAGAIN;
2089         if (f2fs_has_inline_data(inode)) {
2090                 err = f2fs_write_inline_data(inode, page);
2091                 if (!err)
2092                         goto out;
2093         }
2094
2095         if (err == -EAGAIN) {
2096                 err = f2fs_do_write_data_page(&fio);
2097                 if (err == -EAGAIN) {
2098                         fio.need_lock = LOCK_REQ;
2099                         err = f2fs_do_write_data_page(&fio);
2100                 }
2101         }
2102
2103         if (err) {
2104                 file_set_keep_isize(inode);
2105         } else {
2106                 down_write(&F2FS_I(inode)->i_sem);
2107                 if (F2FS_I(inode)->last_disk_size < psize)
2108                         F2FS_I(inode)->last_disk_size = psize;
2109                 up_write(&F2FS_I(inode)->i_sem);
2110         }
2111
2112 done:
2113         if (err && err != -ENOENT)
2114                 goto redirty_out;
2115
2116 out:
2117         inode_dec_dirty_pages(inode);
2118         if (err) {
2119                 ClearPageUptodate(page);
2120                 clear_cold_data(page);
2121         }
2122
2123         if (wbc->for_reclaim) {
2124                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2125                 clear_inode_flag(inode, FI_HOT_DATA);
2126                 f2fs_remove_dirty_inode(inode);
2127                 submitted = NULL;
2128         }
2129
2130         unlock_page(page);
2131         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2132                                         !F2FS_I(inode)->cp_task) {
2133                 f2fs_submit_ipu_bio(sbi, bio, page);
2134                 f2fs_balance_fs(sbi, need_balance_fs);
2135         }
2136
2137         if (unlikely(f2fs_cp_error(sbi))) {
2138                 f2fs_submit_ipu_bio(sbi, bio, page);
2139                 f2fs_submit_merged_write(sbi, DATA);
2140                 submitted = NULL;
2141         }
2142
2143         if (submitted)
2144                 *submitted = fio.submitted;
2145
2146         return 0;
2147
2148 redirty_out:
2149         redirty_page_for_writepage(wbc, page);
2150         /*
2151          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2152          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2153          * file_write_and_wait_range() will see EIO error, which is critical
2154          * to return value of fsync() followed by atomic_write failure to user.
2155          */
2156         if (!err || wbc->for_reclaim)
2157                 return AOP_WRITEPAGE_ACTIVATE;
2158         unlock_page(page);
2159         return err;
2160 }
2161
2162 static int f2fs_write_data_page(struct page *page,
2163                                         struct writeback_control *wbc)
2164 {
2165         return __write_data_page(page, NULL, NULL, NULL, wbc, FS_DATA_IO);
2166 }
2167
2168 /*
2169  * This function was copied from write_cche_pages from mm/page-writeback.c.
2170  * The major change is making write step of cold data page separately from
2171  * warm/hot data page.
2172  */
2173 static int f2fs_write_cache_pages(struct address_space *mapping,
2174                                         struct writeback_control *wbc,
2175                                         enum iostat_type io_type)
2176 {
2177         int ret = 0;
2178         int done = 0;
2179         struct pagevec pvec;
2180         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2181         struct bio *bio = NULL;
2182         sector_t last_block;
2183         int nr_pages;
2184         pgoff_t uninitialized_var(writeback_index);
2185         pgoff_t index;
2186         pgoff_t end;            /* Inclusive */
2187         pgoff_t done_index;
2188         int cycled;
2189         int range_whole = 0;
2190         xa_mark_t tag;
2191         int nwritten = 0;
2192
2193         pagevec_init(&pvec);
2194
2195         if (get_dirty_pages(mapping->host) <=
2196                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2197                 set_inode_flag(mapping->host, FI_HOT_DATA);
2198         else
2199                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2200
2201         if (wbc->range_cyclic) {
2202                 writeback_index = mapping->writeback_index; /* prev offset */
2203                 index = writeback_index;
2204                 if (index == 0)
2205                         cycled = 1;
2206                 else
2207                         cycled = 0;
2208                 end = -1;
2209         } else {
2210                 index = wbc->range_start >> PAGE_SHIFT;
2211                 end = wbc->range_end >> PAGE_SHIFT;
2212                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2213                         range_whole = 1;
2214                 cycled = 1; /* ignore range_cyclic tests */
2215         }
2216         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2217                 tag = PAGECACHE_TAG_TOWRITE;
2218         else
2219                 tag = PAGECACHE_TAG_DIRTY;
2220 retry:
2221         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2222                 tag_pages_for_writeback(mapping, index, end);
2223         done_index = index;
2224         while (!done && (index <= end)) {
2225                 int i;
2226
2227                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2228                                 tag);
2229                 if (nr_pages == 0)
2230                         break;
2231
2232                 for (i = 0; i < nr_pages; i++) {
2233                         struct page *page = pvec.pages[i];
2234                         bool submitted = false;
2235
2236                         /* give a priority to WB_SYNC threads */
2237                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2238                                         wbc->sync_mode == WB_SYNC_NONE) {
2239                                 done = 1;
2240                                 break;
2241                         }
2242
2243                         done_index = page->index;
2244 retry_write:
2245                         lock_page(page);
2246
2247                         if (unlikely(page->mapping != mapping)) {
2248 continue_unlock:
2249                                 unlock_page(page);
2250                                 continue;
2251                         }
2252
2253                         if (!PageDirty(page)) {
2254                                 /* someone wrote it for us */
2255                                 goto continue_unlock;
2256                         }
2257
2258                         if (PageWriteback(page)) {
2259                                 if (wbc->sync_mode != WB_SYNC_NONE) {
2260                                         f2fs_wait_on_page_writeback(page,
2261                                                         DATA, true, true);
2262                                         f2fs_submit_ipu_bio(sbi, &bio, page);
2263                                 } else {
2264                                         goto continue_unlock;
2265                                 }
2266                         }
2267
2268                         if (!clear_page_dirty_for_io(page))
2269                                 goto continue_unlock;
2270
2271                         ret = __write_data_page(page, &submitted, &bio,
2272                                         &last_block, wbc, io_type);
2273                         if (unlikely(ret)) {
2274                                 /*
2275                                  * keep nr_to_write, since vfs uses this to
2276                                  * get # of written pages.
2277                                  */
2278                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2279                                         unlock_page(page);
2280                                         ret = 0;
2281                                         continue;
2282                                 } else if (ret == -EAGAIN) {
2283                                         ret = 0;
2284                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2285                                                 cond_resched();
2286                                                 congestion_wait(BLK_RW_ASYNC,
2287                                                                         HZ/50);
2288                                                 goto retry_write;
2289                                         }
2290                                         continue;
2291                                 }
2292                                 done_index = page->index + 1;
2293                                 done = 1;
2294                                 break;
2295                         } else if (submitted) {
2296                                 nwritten++;
2297                         }
2298
2299                         if (--wbc->nr_to_write <= 0 &&
2300                                         wbc->sync_mode == WB_SYNC_NONE) {
2301                                 done = 1;
2302                                 break;
2303                         }
2304                 }
2305                 pagevec_release(&pvec);
2306                 cond_resched();
2307         }
2308
2309         if (!cycled && !done) {
2310                 cycled = 1;
2311                 index = 0;
2312                 end = writeback_index - 1;
2313                 goto retry;
2314         }
2315         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2316                 mapping->writeback_index = done_index;
2317
2318         if (nwritten)
2319                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2320                                                                 NULL, 0, DATA);
2321         /* submit cached bio of IPU write */
2322         if (bio)
2323                 __submit_bio(sbi, bio, DATA);
2324
2325         return ret;
2326 }
2327
2328 static inline bool __should_serialize_io(struct inode *inode,
2329                                         struct writeback_control *wbc)
2330 {
2331         if (!S_ISREG(inode->i_mode))
2332                 return false;
2333         if (IS_NOQUOTA(inode))
2334                 return false;
2335         /* to avoid deadlock in path of data flush */
2336         if (F2FS_I(inode)->cp_task)
2337                 return false;
2338         if (wbc->sync_mode != WB_SYNC_ALL)
2339                 return true;
2340         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2341                 return true;
2342         return false;
2343 }
2344
2345 static int __f2fs_write_data_pages(struct address_space *mapping,
2346                                                 struct writeback_control *wbc,
2347                                                 enum iostat_type io_type)
2348 {
2349         struct inode *inode = mapping->host;
2350         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2351         struct blk_plug plug;
2352         int ret;
2353         bool locked = false;
2354
2355         /* deal with chardevs and other special file */
2356         if (!mapping->a_ops->writepage)
2357                 return 0;
2358
2359         /* skip writing if there is no dirty page in this inode */
2360         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2361                 return 0;
2362
2363         /* during POR, we don't need to trigger writepage at all. */
2364         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2365                 goto skip_write;
2366
2367         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2368                         wbc->sync_mode == WB_SYNC_NONE &&
2369                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2370                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2371                 goto skip_write;
2372
2373         /* skip writing during file defragment */
2374         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2375                 goto skip_write;
2376
2377         trace_f2fs_writepages(mapping->host, wbc, DATA);
2378
2379         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2380         if (wbc->sync_mode == WB_SYNC_ALL)
2381                 atomic_inc(&sbi->wb_sync_req[DATA]);
2382         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2383                 goto skip_write;
2384
2385         if (__should_serialize_io(inode, wbc)) {
2386                 mutex_lock(&sbi->writepages);
2387                 locked = true;
2388         }
2389
2390         blk_start_plug(&plug);
2391         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2392         blk_finish_plug(&plug);
2393
2394         if (locked)
2395                 mutex_unlock(&sbi->writepages);
2396
2397         if (wbc->sync_mode == WB_SYNC_ALL)
2398                 atomic_dec(&sbi->wb_sync_req[DATA]);
2399         /*
2400          * if some pages were truncated, we cannot guarantee its mapping->host
2401          * to detect pending bios.
2402          */
2403
2404         f2fs_remove_dirty_inode(inode);
2405         return ret;
2406
2407 skip_write:
2408         wbc->pages_skipped += get_dirty_pages(inode);
2409         trace_f2fs_writepages(mapping->host, wbc, DATA);
2410         return 0;
2411 }
2412
2413 static int f2fs_write_data_pages(struct address_space *mapping,
2414                             struct writeback_control *wbc)
2415 {
2416         struct inode *inode = mapping->host;
2417
2418         return __f2fs_write_data_pages(mapping, wbc,
2419                         F2FS_I(inode)->cp_task == current ?
2420                         FS_CP_DATA_IO : FS_DATA_IO);
2421 }
2422
2423 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2424 {
2425         struct inode *inode = mapping->host;
2426         loff_t i_size = i_size_read(inode);
2427
2428         if (to > i_size) {
2429                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2430                 down_write(&F2FS_I(inode)->i_mmap_sem);
2431
2432                 truncate_pagecache(inode, i_size);
2433                 if (!IS_NOQUOTA(inode))
2434                         f2fs_truncate_blocks(inode, i_size, true);
2435
2436                 up_write(&F2FS_I(inode)->i_mmap_sem);
2437                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2438         }
2439 }
2440
2441 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2442                         struct page *page, loff_t pos, unsigned len,
2443                         block_t *blk_addr, bool *node_changed)
2444 {
2445         struct inode *inode = page->mapping->host;
2446         pgoff_t index = page->index;
2447         struct dnode_of_data dn;
2448         struct page *ipage;
2449         bool locked = false;
2450         struct extent_info ei = {0,0,0};
2451         int err = 0;
2452         int flag;
2453
2454         /*
2455          * we already allocated all the blocks, so we don't need to get
2456          * the block addresses when there is no need to fill the page.
2457          */
2458         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2459                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2460                 return 0;
2461
2462         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2463         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2464                 flag = F2FS_GET_BLOCK_DEFAULT;
2465         else
2466                 flag = F2FS_GET_BLOCK_PRE_AIO;
2467
2468         if (f2fs_has_inline_data(inode) ||
2469                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2470                 __do_map_lock(sbi, flag, true);
2471                 locked = true;
2472         }
2473 restart:
2474         /* check inline_data */
2475         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2476         if (IS_ERR(ipage)) {
2477                 err = PTR_ERR(ipage);
2478                 goto unlock_out;
2479         }
2480
2481         set_new_dnode(&dn, inode, ipage, ipage, 0);
2482
2483         if (f2fs_has_inline_data(inode)) {
2484                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2485                         f2fs_do_read_inline_data(page, ipage);
2486                         set_inode_flag(inode, FI_DATA_EXIST);
2487                         if (inode->i_nlink)
2488                                 set_inline_node(ipage);
2489                 } else {
2490                         err = f2fs_convert_inline_page(&dn, page);
2491                         if (err)
2492                                 goto out;
2493                         if (dn.data_blkaddr == NULL_ADDR)
2494                                 err = f2fs_get_block(&dn, index);
2495                 }
2496         } else if (locked) {
2497                 err = f2fs_get_block(&dn, index);
2498         } else {
2499                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2500                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2501                 } else {
2502                         /* hole case */
2503                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2504                         if (err || dn.data_blkaddr == NULL_ADDR) {
2505                                 f2fs_put_dnode(&dn);
2506                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2507                                                                 true);
2508                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2509                                 locked = true;
2510                                 goto restart;
2511                         }
2512                 }
2513         }
2514
2515         /* convert_inline_page can make node_changed */
2516         *blk_addr = dn.data_blkaddr;
2517         *node_changed = dn.node_changed;
2518 out:
2519         f2fs_put_dnode(&dn);
2520 unlock_out:
2521         if (locked)
2522                 __do_map_lock(sbi, flag, false);
2523         return err;
2524 }
2525
2526 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2527                 loff_t pos, unsigned len, unsigned flags,
2528                 struct page **pagep, void **fsdata)
2529 {
2530         struct inode *inode = mapping->host;
2531         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2532         struct page *page = NULL;
2533         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2534         bool need_balance = false, drop_atomic = false;
2535         block_t blkaddr = NULL_ADDR;
2536         int err = 0;
2537
2538         trace_f2fs_write_begin(inode, pos, len, flags);
2539
2540         err = f2fs_is_checkpoint_ready(sbi);
2541         if (err)
2542                 goto fail;
2543
2544         if ((f2fs_is_atomic_file(inode) &&
2545                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2546                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2547                 err = -ENOMEM;
2548                 drop_atomic = true;
2549                 goto fail;
2550         }
2551
2552         /*
2553          * We should check this at this moment to avoid deadlock on inode page
2554          * and #0 page. The locking rule for inline_data conversion should be:
2555          * lock_page(page #0) -> lock_page(inode_page)
2556          */
2557         if (index != 0) {
2558                 err = f2fs_convert_inline_inode(inode);
2559                 if (err)
2560                         goto fail;
2561         }
2562 repeat:
2563         /*
2564          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2565          * wait_for_stable_page. Will wait that below with our IO control.
2566          */
2567         page = f2fs_pagecache_get_page(mapping, index,
2568                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2569         if (!page) {
2570                 err = -ENOMEM;
2571                 goto fail;
2572         }
2573
2574         *pagep = page;
2575
2576         err = prepare_write_begin(sbi, page, pos, len,
2577                                         &blkaddr, &need_balance);
2578         if (err)
2579                 goto fail;
2580
2581         if (need_balance && !IS_NOQUOTA(inode) &&
2582                         has_not_enough_free_secs(sbi, 0, 0)) {
2583                 unlock_page(page);
2584                 f2fs_balance_fs(sbi, true);
2585                 lock_page(page);
2586                 if (page->mapping != mapping) {
2587                         /* The page got truncated from under us */
2588                         f2fs_put_page(page, 1);
2589                         goto repeat;
2590                 }
2591         }
2592
2593         f2fs_wait_on_page_writeback(page, DATA, false, true);
2594
2595         if (len == PAGE_SIZE || PageUptodate(page))
2596                 return 0;
2597
2598         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2599                 zero_user_segment(page, len, PAGE_SIZE);
2600                 return 0;
2601         }
2602
2603         if (blkaddr == NEW_ADDR) {
2604                 zero_user_segment(page, 0, PAGE_SIZE);
2605                 SetPageUptodate(page);
2606         } else {
2607                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2608                                 DATA_GENERIC_ENHANCE_READ)) {
2609                         err = -EFAULT;
2610                         goto fail;
2611                 }
2612                 err = f2fs_submit_page_read(inode, page, blkaddr);
2613                 if (err)
2614                         goto fail;
2615
2616                 lock_page(page);
2617                 if (unlikely(page->mapping != mapping)) {
2618                         f2fs_put_page(page, 1);
2619                         goto repeat;
2620                 }
2621                 if (unlikely(!PageUptodate(page))) {
2622                         err = -EIO;
2623                         goto fail;
2624                 }
2625         }
2626         return 0;
2627
2628 fail:
2629         f2fs_put_page(page, 1);
2630         f2fs_write_failed(mapping, pos + len);
2631         if (drop_atomic)
2632                 f2fs_drop_inmem_pages_all(sbi, false);
2633         return err;
2634 }
2635
2636 static int f2fs_write_end(struct file *file,
2637                         struct address_space *mapping,
2638                         loff_t pos, unsigned len, unsigned copied,
2639                         struct page *page, void *fsdata)
2640 {
2641         struct inode *inode = page->mapping->host;
2642
2643         trace_f2fs_write_end(inode, pos, len, copied);
2644
2645         /*
2646          * This should be come from len == PAGE_SIZE, and we expect copied
2647          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2648          * let generic_perform_write() try to copy data again through copied=0.
2649          */
2650         if (!PageUptodate(page)) {
2651                 if (unlikely(copied != len))
2652                         copied = 0;
2653                 else
2654                         SetPageUptodate(page);
2655         }
2656         if (!copied)
2657                 goto unlock_out;
2658
2659         set_page_dirty(page);
2660
2661         if (pos + copied > i_size_read(inode))
2662                 f2fs_i_size_write(inode, pos + copied);
2663 unlock_out:
2664         f2fs_put_page(page, 1);
2665         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2666         return copied;
2667 }
2668
2669 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2670                            loff_t offset)
2671 {
2672         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2673         unsigned blkbits = i_blkbits;
2674         unsigned blocksize_mask = (1 << blkbits) - 1;
2675         unsigned long align = offset | iov_iter_alignment(iter);
2676         struct block_device *bdev = inode->i_sb->s_bdev;
2677
2678         if (align & blocksize_mask) {
2679                 if (bdev)
2680                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2681                 blocksize_mask = (1 << blkbits) - 1;
2682                 if (align & blocksize_mask)
2683                         return -EINVAL;
2684                 return 1;
2685         }
2686         return 0;
2687 }
2688
2689 static void f2fs_dio_end_io(struct bio *bio)
2690 {
2691         struct f2fs_private_dio *dio = bio->bi_private;
2692
2693         dec_page_count(F2FS_I_SB(dio->inode),
2694                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2695
2696         bio->bi_private = dio->orig_private;
2697         bio->bi_end_io = dio->orig_end_io;
2698
2699         kvfree(dio);
2700
2701         bio_endio(bio);
2702 }
2703
2704 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2705                                                         loff_t file_offset)
2706 {
2707         struct f2fs_private_dio *dio;
2708         bool write = (bio_op(bio) == REQ_OP_WRITE);
2709
2710         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2711                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2712         if (!dio)
2713                 goto out;
2714
2715         dio->inode = inode;
2716         dio->orig_end_io = bio->bi_end_io;
2717         dio->orig_private = bio->bi_private;
2718         dio->write = write;
2719
2720         bio->bi_end_io = f2fs_dio_end_io;
2721         bio->bi_private = dio;
2722
2723         inc_page_count(F2FS_I_SB(inode),
2724                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2725
2726         submit_bio(bio);
2727         return;
2728 out:
2729         bio->bi_status = BLK_STS_IOERR;
2730         bio_endio(bio);
2731 }
2732
2733 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2734 {
2735         struct address_space *mapping = iocb->ki_filp->f_mapping;
2736         struct inode *inode = mapping->host;
2737         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2738         struct f2fs_inode_info *fi = F2FS_I(inode);
2739         size_t count = iov_iter_count(iter);
2740         loff_t offset = iocb->ki_pos;
2741         int rw = iov_iter_rw(iter);
2742         int err;
2743         enum rw_hint hint = iocb->ki_hint;
2744         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2745         bool do_opu;
2746
2747         err = check_direct_IO(inode, iter, offset);
2748         if (err)
2749                 return err < 0 ? err : 0;
2750
2751         if (f2fs_force_buffered_io(inode, iocb, iter))
2752                 return 0;
2753
2754         do_opu = allow_outplace_dio(inode, iocb, iter);
2755
2756         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2757
2758         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2759                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2760
2761         if (iocb->ki_flags & IOCB_NOWAIT) {
2762                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2763                         iocb->ki_hint = hint;
2764                         err = -EAGAIN;
2765                         goto out;
2766                 }
2767                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2768                         up_read(&fi->i_gc_rwsem[rw]);
2769                         iocb->ki_hint = hint;
2770                         err = -EAGAIN;
2771                         goto out;
2772                 }
2773         } else {
2774                 down_read(&fi->i_gc_rwsem[rw]);
2775                 if (do_opu)
2776                         down_read(&fi->i_gc_rwsem[READ]);
2777         }
2778
2779         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2780                         iter, rw == WRITE ? get_data_block_dio_write :
2781                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2782                         DIO_LOCKING | DIO_SKIP_HOLES);
2783
2784         if (do_opu)
2785                 up_read(&fi->i_gc_rwsem[READ]);
2786
2787         up_read(&fi->i_gc_rwsem[rw]);
2788
2789         if (rw == WRITE) {
2790                 if (whint_mode == WHINT_MODE_OFF)
2791                         iocb->ki_hint = hint;
2792                 if (err > 0) {
2793                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2794                                                                         err);
2795                         if (!do_opu)
2796                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2797                 } else if (err < 0) {
2798                         f2fs_write_failed(mapping, offset + count);
2799                 }
2800         }
2801
2802 out:
2803         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2804
2805         return err;
2806 }
2807
2808 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2809                                                         unsigned int length)
2810 {
2811         struct inode *inode = page->mapping->host;
2812         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2813
2814         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2815                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2816                 return;
2817
2818         if (PageDirty(page)) {
2819                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2820                         dec_page_count(sbi, F2FS_DIRTY_META);
2821                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2822                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2823                 } else {
2824                         inode_dec_dirty_pages(inode);
2825                         f2fs_remove_dirty_inode(inode);
2826                 }
2827         }
2828
2829         clear_cold_data(page);
2830
2831         if (IS_ATOMIC_WRITTEN_PAGE(page))
2832                 return f2fs_drop_inmem_page(inode, page);
2833
2834         f2fs_clear_page_private(page);
2835 }
2836
2837 int f2fs_release_page(struct page *page, gfp_t wait)
2838 {
2839         /* If this is dirty page, keep PagePrivate */
2840         if (PageDirty(page))
2841                 return 0;
2842
2843         /* This is atomic written page, keep Private */
2844         if (IS_ATOMIC_WRITTEN_PAGE(page))
2845                 return 0;
2846
2847         clear_cold_data(page);
2848         f2fs_clear_page_private(page);
2849         return 1;
2850 }
2851
2852 static int f2fs_set_data_page_dirty(struct page *page)
2853 {
2854         struct address_space *mapping = page->mapping;
2855         struct inode *inode = mapping->host;
2856
2857         trace_f2fs_set_page_dirty(page, DATA);
2858
2859         if (!PageUptodate(page))
2860                 SetPageUptodate(page);
2861
2862         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2863                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2864                         f2fs_register_inmem_page(inode, page);
2865                         return 1;
2866                 }
2867                 /*
2868                  * Previously, this page has been registered, we just
2869                  * return here.
2870                  */
2871                 return 0;
2872         }
2873
2874         if (!PageDirty(page)) {
2875                 __set_page_dirty_nobuffers(page);
2876                 f2fs_update_dirty_page(inode, page);
2877                 return 1;
2878         }
2879         return 0;
2880 }
2881
2882 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2883 {
2884         struct inode *inode = mapping->host;
2885
2886         if (f2fs_has_inline_data(inode))
2887                 return 0;
2888
2889         /* make sure allocating whole blocks */
2890         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2891                 filemap_write_and_wait(mapping);
2892
2893         return generic_block_bmap(mapping, block, get_data_block_bmap);
2894 }
2895
2896 #ifdef CONFIG_MIGRATION
2897 #include <linux/migrate.h>
2898
2899 int f2fs_migrate_page(struct address_space *mapping,
2900                 struct page *newpage, struct page *page, enum migrate_mode mode)
2901 {
2902         int rc, extra_count;
2903         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2904         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2905
2906         BUG_ON(PageWriteback(page));
2907
2908         /* migrating an atomic written page is safe with the inmem_lock hold */
2909         if (atomic_written) {
2910                 if (mode != MIGRATE_SYNC)
2911                         return -EBUSY;
2912                 if (!mutex_trylock(&fi->inmem_lock))
2913                         return -EAGAIN;
2914         }
2915
2916         /* one extra reference was held for atomic_write page */
2917         extra_count = atomic_written ? 1 : 0;
2918         rc = migrate_page_move_mapping(mapping, newpage,
2919                                 page, mode, extra_count);
2920         if (rc != MIGRATEPAGE_SUCCESS) {
2921                 if (atomic_written)
2922                         mutex_unlock(&fi->inmem_lock);
2923                 return rc;
2924         }
2925
2926         if (atomic_written) {
2927                 struct inmem_pages *cur;
2928                 list_for_each_entry(cur, &fi->inmem_pages, list)
2929                         if (cur->page == page) {
2930                                 cur->page = newpage;
2931                                 break;
2932                         }
2933                 mutex_unlock(&fi->inmem_lock);
2934                 put_page(page);
2935                 get_page(newpage);
2936         }
2937
2938         if (PagePrivate(page)) {
2939                 f2fs_set_page_private(newpage, page_private(page));
2940                 f2fs_clear_page_private(page);
2941         }
2942
2943         if (mode != MIGRATE_SYNC_NO_COPY)
2944                 migrate_page_copy(newpage, page);
2945         else
2946                 migrate_page_states(newpage, page);
2947
2948         return MIGRATEPAGE_SUCCESS;
2949 }
2950 #endif
2951
2952 const struct address_space_operations f2fs_dblock_aops = {
2953         .readpage       = f2fs_read_data_page,
2954         .readpages      = f2fs_read_data_pages,
2955         .writepage      = f2fs_write_data_page,
2956         .writepages     = f2fs_write_data_pages,
2957         .write_begin    = f2fs_write_begin,
2958         .write_end      = f2fs_write_end,
2959         .set_page_dirty = f2fs_set_data_page_dirty,
2960         .invalidatepage = f2fs_invalidate_page,
2961         .releasepage    = f2fs_release_page,
2962         .direct_IO      = f2fs_direct_IO,
2963         .bmap           = f2fs_bmap,
2964 #ifdef CONFIG_MIGRATION
2965         .migratepage    = f2fs_migrate_page,
2966 #endif
2967 };
2968
2969 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2970 {
2971         struct address_space *mapping = page_mapping(page);
2972         unsigned long flags;
2973
2974         xa_lock_irqsave(&mapping->i_pages, flags);
2975         __xa_clear_mark(&mapping->i_pages, page_index(page),
2976                                                 PAGECACHE_TAG_DIRTY);
2977         xa_unlock_irqrestore(&mapping->i_pages, flags);
2978 }
2979
2980 int __init f2fs_init_post_read_processing(void)
2981 {
2982         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2983         if (!bio_post_read_ctx_cache)
2984                 goto fail;
2985         bio_post_read_ctx_pool =
2986                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2987                                          bio_post_read_ctx_cache);
2988         if (!bio_post_read_ctx_pool)
2989                 goto fail_free_cache;
2990         return 0;
2991
2992 fail_free_cache:
2993         kmem_cache_destroy(bio_post_read_ctx_cache);
2994 fail:
2995         return -ENOMEM;
2996 }
2997
2998 void __exit f2fs_destroy_post_read_processing(void)
2999 {
3000         mempool_destroy(bio_post_read_ctx_pool);
3001         kmem_cache_destroy(bio_post_read_ctx_cache);
3002 }