fs/f2fs/data.c

   1 /*
   2  * fs/f2fs/data.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/mpage.h>
  15 #include <linux/writeback.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21 #include <linux/uio.h>
  22 #include <linux/cleancache.h>
  23 #include <linux/sched/signal.h>
  24
  25 #include "f2fs.h"
  26 #include "node.h"
  27 #include "segment.h"
  28 #include "trace.h"
  29 #include <trace/events/f2fs.h>
  30
  31 #define NUM_PREALLOC_POST_READ_CTXS     128
  32
  33 static struct kmem_cache *bio_post_read_ctx_cache;
  34 static mempool_t *bio_post_read_ctx_pool;
  35
  36 static bool __is_cp_guaranteed(struct page *page)
  37 {
  38         struct address_space *mapping = page->mapping;
  39         struct inode *inode;
  40         struct f2fs_sb_info *sbi;
  41
  42         if (!mapping)
  43                 return false;
  44
  45         inode = mapping->host;
  46         sbi = F2FS_I_SB(inode);
  47
  48         if (inode->i_ino == F2FS_META_INO(sbi) ||
  49                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  50                         S_ISDIR(inode->i_mode) ||
  51                         (S_ISREG(inode->i_mode) &&
  52                         is_inode_flag_set(inode, FI_ATOMIC_FILE)) ||
  53                         is_cold_data(page))
  54                 return true;
  55         return false;
  56 }
  57
  58 /* postprocessing steps for read bios */
  59 enum bio_post_read_step {
  60         STEP_INITIAL = 0,
  61         STEP_DECRYPT,
  62 };
  63
  64 struct bio_post_read_ctx {
  65         struct bio *bio;
  66         struct work_struct work;
  67         unsigned int cur_step;
  68         unsigned int enabled_steps;
  69 };
  70
  71 static void __read_end_io(struct bio *bio)
  72 {
  73         struct page *page;
  74         struct bio_vec *bv;
  75         int i;
  76
  77         bio_for_each_segment_all(bv, bio, i) {
  78                 page = bv->bv_page;
  79
  80                 /* PG_error was set if any post_read step failed */
  81                 if (bio->bi_status || PageError(page)) {
  82                         ClearPageUptodate(page);
  83                         /* will re-read again later */
  84                         ClearPageError(page);
  85                 } else {
  86                         SetPageUptodate(page);
  87                 }
  88                 unlock_page(page);
  89         }
  90         if (bio->bi_private)
  91                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
  92         bio_put(bio);
  93 }
  94
  95 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
  96
  97 static void decrypt_work(struct work_struct *work)
  98 {
  99         struct bio_post_read_ctx *ctx =
 100                 container_of(work, struct bio_post_read_ctx, work);
 101
 102         fscrypt_decrypt_bio(ctx->bio);
 103
 104         bio_post_read_processing(ctx);
 105 }
 106
 107 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 108 {
 109         switch (++ctx->cur_step) {
 110         case STEP_DECRYPT:
 111                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 112                         INIT_WORK(&ctx->work, decrypt_work);
 113                         fscrypt_enqueue_decrypt_work(&ctx->work);
 114                         return;
 115                 }
 116                 ctx->cur_step++;
 117                 /* fall-through */
 118         default:
 119                 __read_end_io(ctx->bio);
 120         }
 121 }
 122
 123 static bool f2fs_bio_post_read_required(struct bio *bio)
 124 {
 125         return bio->bi_private && !bio->bi_status;
 126 }
 127
 128 static void f2fs_read_end_io(struct bio *bio)
 129 {
 130         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)), FAULT_IO)) {
 131                 f2fs_show_injection_info(FAULT_IO);
 132                 bio->bi_status = BLK_STS_IOERR;
 133         }
 134
 135         if (f2fs_bio_post_read_required(bio)) {
 136                 struct bio_post_read_ctx *ctx = bio->bi_private;
 137
 138                 ctx->cur_step = STEP_INITIAL;
 139                 bio_post_read_processing(ctx);
 140                 return;
 141         }
 142
 143         __read_end_io(bio);
 144 }
 145
 146 static void f2fs_write_end_io(struct bio *bio)
 147 {
 148         struct f2fs_sb_info *sbi = bio->bi_private;
 149         struct bio_vec *bvec;
 150         int i;
 151
 152         bio_for_each_segment_all(bvec, bio, i) {
 153                 struct page *page = bvec->bv_page;
 154                 enum count_type type = WB_DATA_TYPE(page);
 155
 156                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 157                         set_page_private(page, (unsigned long)NULL);
 158                         ClearPagePrivate(page);
 159                         unlock_page(page);
 160                         mempool_free(page, sbi->write_io_dummy);
 161
 162                         if (unlikely(bio->bi_status))
 163                                 f2fs_stop_checkpoint(sbi, true);
 164                         continue;
 165                 }
 166
 167                 fscrypt_pullback_bio_page(&page, true);
 168
 169                 if (unlikely(bio->bi_status)) {
 170                         mapping_set_error(page->mapping, -EIO);
 171                         if (type == F2FS_WB_CP_DATA)
 172                                 f2fs_stop_checkpoint(sbi, true);
 173                 }
 174
 175                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 176                                         page->index != nid_of_node(page));
 177
 178                 dec_page_count(sbi, type);
 179                 if (f2fs_in_warm_node_list(sbi, page))
 180                         f2fs_del_fsync_node_entry(sbi, page);
 181                 clear_cold_data(page);
 182                 end_page_writeback(page);
 183         }
 184         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 185                                 wq_has_sleeper(&sbi->cp_wait))
 186                 wake_up(&sbi->cp_wait);
 187
 188         bio_put(bio);
 189 }
 190
 191 /*
 192  * Return true, if pre_bio's bdev is same as its target device.
 193  */
 194 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 195                                 block_t blk_addr, struct bio *bio)
 196 {
 197         struct block_device *bdev = sbi->sb->s_bdev;
 198         int i;
 199
 200         if (f2fs_is_multi_device(sbi)) {
 201                 for (i = 0; i < sbi->s_ndevs; i++) {
 202                         if (FDEV(i).start_blk <= blk_addr &&
 203                             FDEV(i).end_blk >= blk_addr) {
 204                                 blk_addr -= FDEV(i).start_blk;
 205                                 bdev = FDEV(i).bdev;
 206                                 break;
 207                         }
 208                 }
 209         }
 210         if (bio) {
 211                 bio_set_dev(bio, bdev);
 212                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 213         }
 214         return bdev;
 215 }
 216
 217 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 218 {
 219         int i;
 220
 221         if (!f2fs_is_multi_device(sbi))
 222                 return 0;
 223
 224         for (i = 0; i < sbi->s_ndevs; i++)
 225                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 226                         return i;
 227         return 0;
 228 }
 229
 230 static bool __same_bdev(struct f2fs_sb_info *sbi,
 231                                 block_t blk_addr, struct bio *bio)
 232 {
 233         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 234         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 235 }
 236
 237 /*
 238  * Low-level block read/write IO operations.
 239  */
 240 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 241                                 struct writeback_control *wbc,
 242                                 int npages, bool is_read,
 243                                 enum page_type type, enum temp_type temp)
 244 {
 245         struct bio *bio;
 246
 247         bio = f2fs_bio_alloc(sbi, npages, true);
 248
 249         f2fs_target_device(sbi, blk_addr, bio);
 250         if (is_read) {
 251                 bio->bi_end_io = f2fs_read_end_io;
 252                 bio->bi_private = NULL;
 253         } else {
 254                 bio->bi_end_io = f2fs_write_end_io;
 255                 bio->bi_private = sbi;
 256                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 257         }
 258         if (wbc)
 259                 wbc_init_bio(wbc, bio);
 260
 261         return bio;
 262 }
 263
 264 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 265                                 struct bio *bio, enum page_type type)
 266 {
 267         if (!is_read_io(bio_op(bio))) {
 268                 unsigned int start;
 269
 270                 if (type != DATA && type != NODE)
 271                         goto submit_io;
 272
 273                 if (test_opt(sbi, LFS) && current->plug)
 274                         blk_finish_plug(current->plug);
 275
 276                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 277                 start %= F2FS_IO_SIZE(sbi);
 278
 279                 if (start == 0)
 280                         goto submit_io;
 281
 282                 /* fill dummy pages */
 283                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 284                         struct page *page =
 285                                 mempool_alloc(sbi->write_io_dummy,
 286                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 287                         f2fs_bug_on(sbi, !page);
 288
 289                         SetPagePrivate(page);
 290                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 291                         lock_page(page);
 292                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 293                                 f2fs_bug_on(sbi, 1);
 294                 }
 295                 /*
 296                  * In the NODE case, we lose next block address chain. So, we
 297                  * need to do checkpoint in f2fs_sync_file.
 298                  */
 299                 if (type == NODE)
 300                         set_sbi_flag(sbi, SBI_NEED_CP);
 301         }
 302 submit_io:
 303         if (is_read_io(bio_op(bio)))
 304                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 305         else
 306                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 307         submit_bio(bio);
 308 }
 309
 310 static void __submit_merged_bio(struct f2fs_bio_info *io)
 311 {
 312         struct f2fs_io_info *fio = &io->fio;
 313
 314         if (!io->bio)
 315                 return;
 316
 317         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 318
 319         if (is_read_io(fio->op))
 320                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 321         else
 322                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 323
 324         __submit_bio(io->sbi, io->bio, fio->type);
 325         io->bio = NULL;
 326 }
 327
 328 static bool __has_merged_page(struct f2fs_bio_info *io,
 329                                 struct inode *inode, nid_t ino, pgoff_t idx)
 330 {
 331         struct bio_vec *bvec;
 332         struct page *target;
 333         int i;
 334
 335         if (!io->bio)
 336                 return false;
 337
 338         if (!inode && !ino)
 339                 return true;
 340
 341         bio_for_each_segment_all(bvec, io->bio, i) {
 342
 343                 if (bvec->bv_page->mapping)
 344                         target = bvec->bv_page;
 345                 else
 346                         target = fscrypt_control_page(bvec->bv_page);
 347
 348                 if (idx != target->index)
 349                         continue;
 350
 351                 if (inode && inode == target->mapping->host)
 352                         return true;
 353                 if (ino && ino == ino_of_node(target))
 354                         return true;
 355         }
 356
 357         return false;
 358 }
 359
 360 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 361                                 nid_t ino, pgoff_t idx, enum page_type type)
 362 {
 363         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 364         enum temp_type temp;
 365         struct f2fs_bio_info *io;
 366         bool ret = false;
 367
 368         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 369                 io = sbi->write_io[btype] + temp;
 370
 371                 down_read(&io->io_rwsem);
 372                 ret = __has_merged_page(io, inode, ino, idx);
 373                 up_read(&io->io_rwsem);
 374
 375                 /* TODO: use HOT temp only for meta pages now. */
 376                 if (ret || btype == META)
 377                         break;
 378         }
 379         return ret;
 380 }
 381
 382 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 383                                 enum page_type type, enum temp_type temp)
 384 {
 385         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 386         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 387
 388         down_write(&io->io_rwsem);
 389
 390         /* change META to META_FLUSH in the checkpoint procedure */
 391         if (type >= META_FLUSH) {
 392                 io->fio.type = META_FLUSH;
 393                 io->fio.op = REQ_OP_WRITE;
 394                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 395                 if (!test_opt(sbi, NOBARRIER))
 396                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 397         }
 398         __submit_merged_bio(io);
 399         up_write(&io->io_rwsem);
 400 }
 401
 402 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 403                                 struct inode *inode, nid_t ino, pgoff_t idx,
 404                                 enum page_type type, bool force)
 405 {
 406         enum temp_type temp;
 407
 408         if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 409                 return;
 410
 411         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 412
 413                 __f2fs_submit_merged_write(sbi, type, temp);
 414
 415                 /* TODO: use HOT temp only for meta pages now. */
 416                 if (type >= META)
 417                         break;
 418         }
 419 }
 420
 421 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 422 {
 423         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 424 }
 425
 426 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 427                                 struct inode *inode, nid_t ino, pgoff_t idx,
 428                                 enum page_type type)
 429 {
 430         __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 431 }
 432
 433 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 434 {
 435         f2fs_submit_merged_write(sbi, DATA);
 436         f2fs_submit_merged_write(sbi, NODE);
 437         f2fs_submit_merged_write(sbi, META);
 438 }
 439
 440 /*
 441  * Fill the locked page with data located in the block address.
 442  * A caller needs to unlock the page on failure.
 443  */
 444 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 445 {
 446         struct bio *bio;
 447         struct page *page = fio->encrypted_page ?
 448                         fio->encrypted_page : fio->page;
 449
 450         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 451                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 452                 return -EFSCORRUPTED;
 453
 454         trace_f2fs_submit_page_bio(page, fio);
 455         f2fs_trace_ios(fio, 0);
 456
 457         /* Allocate a new bio */
 458         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 459                                 1, is_read_io(fio->op), fio->type, fio->temp);
 460
 461         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 462                 bio_put(bio);
 463                 return -EFAULT;
 464         }
 465
 466         if (fio->io_wbc && !is_read_io(fio->op))
 467                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 468
 469         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 470
 471         if (!is_read_io(fio->op))
 472                 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 473
 474         __submit_bio(fio->sbi, bio, fio->type);
 475         return 0;
 476 }
 477
 478 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 479 {
 480         struct f2fs_sb_info *sbi = fio->sbi;
 481         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 482         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 483         struct page *bio_page;
 484
 485         f2fs_bug_on(sbi, is_read_io(fio->op));
 486
 487         down_write(&io->io_rwsem);
 488 next:
 489         if (fio->in_list) {
 490                 spin_lock(&io->io_lock);
 491                 if (list_empty(&io->io_list)) {
 492                         spin_unlock(&io->io_lock);
 493                         goto out;
 494                 }
 495                 fio = list_first_entry(&io->io_list,
 496                                                 struct f2fs_io_info, list);
 497                 list_del(&fio->list);
 498                 spin_unlock(&io->io_lock);
 499         }
 500
 501         if (__is_valid_data_blkaddr(fio->old_blkaddr))
 502                 verify_block_addr(fio, fio->old_blkaddr);
 503         verify_block_addr(fio, fio->new_blkaddr);
 504
 505         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 506
 507         /* set submitted = true as a return value */
 508         fio->submitted = true;
 509
 510         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 511
 512         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 513             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 514                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 515                 __submit_merged_bio(io);
 516 alloc_new:
 517         if (io->bio == NULL) {
 518                 if ((fio->type == DATA || fio->type == NODE) &&
 519                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 520                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 521                         fio->retry = true;
 522                         goto skip;
 523                 }
 524                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 525                                                 BIO_MAX_PAGES, false,
 526                                                 fio->type, fio->temp);
 527                 io->fio = *fio;
 528         }
 529
 530         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 531                 __submit_merged_bio(io);
 532                 goto alloc_new;
 533         }
 534
 535         if (fio->io_wbc)
 536                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 537
 538         io->last_block_in_bio = fio->new_blkaddr;
 539         f2fs_trace_ios(fio, 0);
 540
 541         trace_f2fs_submit_page_write(fio->page, fio);
 542 skip:
 543         if (fio->in_list)
 544                 goto next;
 545 out:
 546         up_write(&io->io_rwsem);
 547 }
 548
 549 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 550                                         unsigned nr_pages, unsigned op_flag)
 551 {
 552         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 553         struct bio *bio;
 554         struct bio_post_read_ctx *ctx;
 555         unsigned int post_read_steps = 0;
 556
 557         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
 558                 return ERR_PTR(-EFAULT);
 559
 560         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 561         if (!bio)
 562                 return ERR_PTR(-ENOMEM);
 563         f2fs_target_device(sbi, blkaddr, bio);
 564         bio->bi_end_io = f2fs_read_end_io;
 565         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 566
 567         if (f2fs_encrypted_file(inode))
 568                 post_read_steps |= 1 << STEP_DECRYPT;
 569         if (post_read_steps) {
 570                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 571                 if (!ctx) {
 572                         bio_put(bio);
 573                         return ERR_PTR(-ENOMEM);
 574                 }
 575                 ctx->bio = bio;
 576                 ctx->enabled_steps = post_read_steps;
 577                 bio->bi_private = ctx;
 578
 579                 /* wait the page to be moved by cleaning */
 580                 f2fs_wait_on_block_writeback(sbi, blkaddr);
 581         }
 582
 583         return bio;
 584 }
 585
 586 /* This can handle encryption stuffs */
 587 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 588                                                         block_t blkaddr)
 589 {
 590         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 591
 592         if (IS_ERR(bio))
 593                 return PTR_ERR(bio);
 594
 595         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 596                 bio_put(bio);
 597                 return -EFAULT;
 598         }
 599         ClearPageError(page);
 600         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 601         return 0;
 602 }
 603
 604 static void __set_data_blkaddr(struct dnode_of_data *dn)
 605 {
 606         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 607         __le32 *addr_array;
 608         int base = 0;
 609
 610         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 611                 base = get_extra_isize(dn->inode);
 612
 613         /* Get physical address of data block */
 614         addr_array = blkaddr_in_node(rn);
 615         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 616 }
 617
 618 /*
 619  * Lock ordering for the change of data block address:
 620  * ->data_page
 621  *  ->node_page
 622  *    update block addresses in the node page
 623  */
 624 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 625 {
 626         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 627         __set_data_blkaddr(dn);
 628         if (set_page_dirty(dn->node_page))
 629                 dn->node_changed = true;
 630 }
 631
 632 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 633 {
 634         dn->data_blkaddr = blkaddr;
 635         f2fs_set_data_blkaddr(dn);
 636         f2fs_update_extent_cache(dn);
 637 }
 638
 639 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 640 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 641 {
 642         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 643         int err;
 644
 645         if (!count)
 646                 return 0;
 647
 648         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 649                 return -EPERM;
 650         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 651                 return err;
 652
 653         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 654                                                 dn->ofs_in_node, count);
 655
 656         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 657
 658         for (; count > 0; dn->ofs_in_node++) {
 659                 block_t blkaddr = datablock_addr(dn->inode,
 660                                         dn->node_page, dn->ofs_in_node);
 661                 if (blkaddr == NULL_ADDR) {
 662                         dn->data_blkaddr = NEW_ADDR;
 663                         __set_data_blkaddr(dn);
 664                         count--;
 665                 }
 666         }
 667
 668         if (set_page_dirty(dn->node_page))
 669                 dn->node_changed = true;
 670         return 0;
 671 }
 672
 673 /* Should keep dn->ofs_in_node unchanged */
 674 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 675 {
 676         unsigned int ofs_in_node = dn->ofs_in_node;
 677         int ret;
 678
 679         ret = f2fs_reserve_new_blocks(dn, 1);
 680         dn->ofs_in_node = ofs_in_node;
 681         return ret;
 682 }
 683
 684 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 685 {
 686         bool need_put = dn->inode_page ? false : true;
 687         int err;
 688
 689         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 690         if (err)
 691                 return err;
 692
 693         if (dn->data_blkaddr == NULL_ADDR)
 694                 err = f2fs_reserve_new_block(dn);
 695         if (err || need_put)
 696                 f2fs_put_dnode(dn);
 697         return err;
 698 }
 699
 700 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 701 {
 702         struct extent_info ei  = {0,0,0};
 703         struct inode *inode = dn->inode;
 704
 705         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 706                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 707                 return 0;
 708         }
 709
 710         return f2fs_reserve_block(dn, index);
 711 }
 712
 713 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 714                                                 int op_flags, bool for_write)
 715 {
 716         struct address_space *mapping = inode->i_mapping;
 717         struct dnode_of_data dn;
 718         struct page *page;
 719         struct extent_info ei = {0,0,0};
 720         int err;
 721
 722         page = f2fs_grab_cache_page(mapping, index, for_write);
 723         if (!page)
 724                 return ERR_PTR(-ENOMEM);
 725
 726         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 727                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 728                 goto got_it;
 729         }
 730
 731         set_new_dnode(&dn, inode, NULL, NULL, 0);
 732         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 733         if (err)
 734                 goto put_err;
 735         f2fs_put_dnode(&dn);
 736
 737         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 738                 err = -ENOENT;
 739                 goto put_err;
 740         }
 741 got_it:
 742         if (PageUptodate(page)) {
 743                 unlock_page(page);
 744                 return page;
 745         }
 746
 747         /*
 748          * A new dentry page is allocated but not able to be written, since its
 749          * new inode page couldn't be allocated due to -ENOSPC.
 750          * In such the case, its blkaddr can be remained as NEW_ADDR.
 751          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 752          * f2fs_init_inode_metadata.
 753          */
 754         if (dn.data_blkaddr == NEW_ADDR) {
 755                 zero_user_segment(page, 0, PAGE_SIZE);
 756                 if (!PageUptodate(page))
 757                         SetPageUptodate(page);
 758                 unlock_page(page);
 759                 return page;
 760         }
 761
 762         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 763         if (err)
 764                 goto put_err;
 765         return page;
 766
 767 put_err:
 768         f2fs_put_page(page, 1);
 769         return ERR_PTR(err);
 770 }
 771
 772 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 773 {
 774         struct address_space *mapping = inode->i_mapping;
 775         struct page *page;
 776
 777         page = find_get_page(mapping, index);
 778         if (page && PageUptodate(page))
 779                 return page;
 780         f2fs_put_page(page, 0);
 781
 782         page = f2fs_get_read_data_page(inode, index, 0, false);
 783         if (IS_ERR(page))
 784                 return page;
 785
 786         if (PageUptodate(page))
 787                 return page;
 788
 789         wait_on_page_locked(page);
 790         if (unlikely(!PageUptodate(page))) {
 791                 f2fs_put_page(page, 0);
 792                 return ERR_PTR(-EIO);
 793         }
 794         return page;
 795 }
 796
 797 /*
 798  * If it tries to access a hole, return an error.
 799  * Because, the callers, functions in dir.c and GC, should be able to know
 800  * whether this page exists or not.
 801  */
 802 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 803                                                         bool for_write)
 804 {
 805         struct address_space *mapping = inode->i_mapping;
 806         struct page *page;
 807 repeat:
 808         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 809         if (IS_ERR(page))
 810                 return page;
 811
 812         /* wait for read completion */
 813         lock_page(page);
 814         if (unlikely(page->mapping != mapping)) {
 815                 f2fs_put_page(page, 1);
 816                 goto repeat;
 817         }
 818         if (unlikely(!PageUptodate(page))) {
 819                 f2fs_put_page(page, 1);
 820                 return ERR_PTR(-EIO);
 821         }
 822         return page;
 823 }
 824
 825 /*
 826  * Caller ensures that this data page is never allocated.
 827  * A new zero-filled data page is allocated in the page cache.
 828  *
 829  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 830  * f2fs_unlock_op().
 831  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 832  * ipage should be released by this function.
 833  */
 834 struct page *f2fs_get_new_data_page(struct inode *inode,
 835                 struct page *ipage, pgoff_t index, bool new_i_size)
 836 {
 837         struct address_space *mapping = inode->i_mapping;
 838         struct page *page;
 839         struct dnode_of_data dn;
 840         int err;
 841
 842         page = f2fs_grab_cache_page(mapping, index, true);
 843         if (!page) {
 844                 /*
 845                  * before exiting, we should make sure ipage will be released
 846                  * if any error occur.
 847                  */
 848                 f2fs_put_page(ipage, 1);
 849                 return ERR_PTR(-ENOMEM);
 850         }
 851
 852         set_new_dnode(&dn, inode, ipage, NULL, 0);
 853         err = f2fs_reserve_block(&dn, index);
 854         if (err) {
 855                 f2fs_put_page(page, 1);
 856                 return ERR_PTR(err);
 857         }
 858         if (!ipage)
 859                 f2fs_put_dnode(&dn);
 860
 861         if (PageUptodate(page))
 862                 goto got_it;
 863
 864         if (dn.data_blkaddr == NEW_ADDR) {
 865                 zero_user_segment(page, 0, PAGE_SIZE);
 866                 if (!PageUptodate(page))
 867                         SetPageUptodate(page);
 868         } else {
 869                 f2fs_put_page(page, 1);
 870
 871                 /* if ipage exists, blkaddr should be NEW_ADDR */
 872                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 873                 page = f2fs_get_lock_data_page(inode, index, true);
 874                 if (IS_ERR(page))
 875                         return page;
 876         }
 877 got_it:
 878         if (new_i_size && i_size_read(inode) <
 879                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 880                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 881         return page;
 882 }
 883
 884 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 885 {
 886         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 887         struct f2fs_summary sum;
 888         struct node_info ni;
 889         block_t old_blkaddr;
 890         pgoff_t fofs;
 891         blkcnt_t count = 1;
 892         int err;
 893
 894         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 895                 return -EPERM;
 896
 897         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 898         if (err)
 899                 return err;
 900
 901         dn->data_blkaddr = datablock_addr(dn->inode,
 902                                 dn->node_page, dn->ofs_in_node);
 903         if (dn->data_blkaddr == NEW_ADDR)
 904                 goto alloc;
 905
 906         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 907                 return err;
 908
 909 alloc:
 910         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 911         old_blkaddr = dn->data_blkaddr;
 912         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 913                                         &sum, seg_type, NULL, false);
 914         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 915                 invalidate_mapping_pages(META_MAPPING(sbi),
 916                                         old_blkaddr, old_blkaddr);
 917         f2fs_set_data_blkaddr(dn);
 918
 919         /* update i_size */
 920         fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 921                                                         dn->ofs_in_node;
 922         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 923                 f2fs_i_size_write(dn->inode,
 924                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 925         return 0;
 926 }
 927
 928 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 929 {
 930         struct inode *inode = file_inode(iocb->ki_filp);
 931         struct f2fs_map_blocks map;
 932         int flag;
 933         int err = 0;
 934         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 935
 936         /* convert inline data for Direct I/O*/
 937         if (direct_io) {
 938                 err = f2fs_convert_inline_inode(inode);
 939                 if (err)
 940                         return err;
 941         }
 942
 943         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 944                 return 0;
 945
 946         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 947         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 948         if (map.m_len > map.m_lblk)
 949                 map.m_len -= map.m_lblk;
 950         else
 951                 map.m_len = 0;
 952
 953         map.m_next_pgofs = NULL;
 954         map.m_next_extent = NULL;
 955         map.m_seg_type = NO_CHECK_TYPE;
 956
 957         if (direct_io) {
 958                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 959                 flag = f2fs_force_buffered_io(inode, WRITE) ?
 960                                         F2FS_GET_BLOCK_PRE_AIO :
 961                                         F2FS_GET_BLOCK_PRE_DIO;
 962                 goto map_blocks;
 963         }
 964         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 965                 err = f2fs_convert_inline_inode(inode);
 966                 if (err)
 967                         return err;
 968         }
 969         if (f2fs_has_inline_data(inode))
 970                 return err;
 971
 972         flag = F2FS_GET_BLOCK_PRE_AIO;
 973
 974 map_blocks:
 975         err = f2fs_map_blocks(inode, &map, 1, flag);
 976         if (map.m_len > 0 && err == -ENOSPC) {
 977                 if (!direct_io)
 978                         set_inode_flag(inode, FI_NO_PREALLOC);
 979                 err = 0;
 980         }
 981         return err;
 982 }
 983
 984 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 985 {
 986         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 987                 if (lock)
 988                         down_read(&sbi->node_change);
 989                 else
 990                         up_read(&sbi->node_change);
 991         } else {
 992                 if (lock)
 993                         f2fs_lock_op(sbi);
 994                 else
 995                         f2fs_unlock_op(sbi);
 996         }
 997 }
 998
 999 /*
1000  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1001  * f2fs_map_blocks structure.
1002  * If original data blocks are allocated, then give them to blockdev.
1003  * Otherwise,
1004  *     a. preallocate requested block addresses
1005  *     b. do not use extent cache for better performance
1006  *     c. give the block addresses to blockdev
1007  */
1008 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1009                                                 int create, int flag)
1010 {
1011         unsigned int maxblocks = map->m_len;
1012         struct dnode_of_data dn;
1013         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1014         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
1015         pgoff_t pgofs, end_offset, end;
1016         int err = 0, ofs = 1;
1017         unsigned int ofs_in_node, last_ofs_in_node;
1018         blkcnt_t prealloc;
1019         struct extent_info ei = {0,0,0};
1020         block_t blkaddr;
1021         unsigned int start_pgofs;
1022
1023         if (!maxblocks)
1024                 return 0;
1025
1026         map->m_len = 0;
1027         map->m_flags = 0;
1028
1029         /* it only supports block size == page size */
1030         pgofs = (pgoff_t)map->m_lblk;
1031         end = pgofs + maxblocks;
1032
1033         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1034                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1035                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1036                 map->m_flags = F2FS_MAP_MAPPED;
1037                 if (map->m_next_extent)
1038                         *map->m_next_extent = pgofs + map->m_len;
1039                 goto out;
1040         }
1041
1042 next_dnode:
1043         if (create)
1044                 __do_map_lock(sbi, flag, true);
1045
1046         /* When reading holes, we need its node page */
1047         set_new_dnode(&dn, inode, NULL, NULL, 0);
1048         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1049         if (err) {
1050                 if (flag == F2FS_GET_BLOCK_BMAP)
1051                         map->m_pblk = 0;
1052                 if (err == -ENOENT) {
1053                         err = 0;
1054                         if (map->m_next_pgofs)
1055                                 *map->m_next_pgofs =
1056                                         f2fs_get_next_page_offset(&dn, pgofs);
1057                         if (map->m_next_extent)
1058                                 *map->m_next_extent =
1059                                         f2fs_get_next_page_offset(&dn, pgofs);
1060                 }
1061                 goto unlock_out;
1062         }
1063
1064         start_pgofs = pgofs;
1065         prealloc = 0;
1066         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1067         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1068
1069 next_block:
1070         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1071
1072         if (__is_valid_data_blkaddr(blkaddr) &&
1073                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1074                 err = -EFSCORRUPTED;
1075                 goto sync_out;
1076         }
1077
1078         if (!is_valid_data_blkaddr(sbi, blkaddr)) {
1079                 if (create) {
1080                         if (unlikely(f2fs_cp_error(sbi))) {
1081                                 err = -EIO;
1082                                 goto sync_out;
1083                         }
1084                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1085                                 if (blkaddr == NULL_ADDR) {
1086                                         prealloc++;
1087                                         last_ofs_in_node = dn.ofs_in_node;
1088                                 }
1089                         } else {
1090                                 err = __allocate_data_block(&dn,
1091                                                         map->m_seg_type);
1092                                 if (!err)
1093                                         set_inode_flag(inode, FI_APPEND_WRITE);
1094                         }
1095                         if (err)
1096                                 goto sync_out;
1097                         map->m_flags |= F2FS_MAP_NEW;
1098                         blkaddr = dn.data_blkaddr;
1099                 } else {
1100                         if (flag == F2FS_GET_BLOCK_BMAP) {
1101                                 map->m_pblk = 0;
1102                                 goto sync_out;
1103                         }
1104                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1105                                 goto sync_out;
1106                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1107                                                 blkaddr == NULL_ADDR) {
1108                                 if (map->m_next_pgofs)
1109                                         *map->m_next_pgofs = pgofs + 1;
1110                                 goto sync_out;
1111                         }
1112                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1113                                 /* for defragment case */
1114                                 if (map->m_next_pgofs)
1115                                         *map->m_next_pgofs = pgofs + 1;
1116                                 goto sync_out;
1117                         }
1118                 }
1119         }
1120
1121         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1122                 goto skip;
1123
1124         if (map->m_len == 0) {
1125                 /* preallocated unwritten block should be mapped for fiemap. */
1126                 if (blkaddr == NEW_ADDR)
1127                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1128                 map->m_flags |= F2FS_MAP_MAPPED;
1129
1130                 map->m_pblk = blkaddr;
1131                 map->m_len = 1;
1132         } else if ((map->m_pblk != NEW_ADDR &&
1133                         blkaddr == (map->m_pblk + ofs)) ||
1134                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1135                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1136                 ofs++;
1137                 map->m_len++;
1138         } else {
1139                 goto sync_out;
1140         }
1141
1142 skip:
1143         dn.ofs_in_node++;
1144         pgofs++;
1145
1146         /* preallocate blocks in batch for one dnode page */
1147         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1148                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1149
1150                 dn.ofs_in_node = ofs_in_node;
1151                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1152                 if (err)
1153                         goto sync_out;
1154
1155                 map->m_len += dn.ofs_in_node - ofs_in_node;
1156                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1157                         err = -ENOSPC;
1158                         goto sync_out;
1159                 }
1160                 dn.ofs_in_node = end_offset;
1161         }
1162
1163         if (pgofs >= end)
1164                 goto sync_out;
1165         else if (dn.ofs_in_node < end_offset)
1166                 goto next_block;
1167
1168         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1169                 if (map->m_flags & F2FS_MAP_MAPPED) {
1170                         unsigned int ofs = start_pgofs - map->m_lblk;
1171
1172                         f2fs_update_extent_cache_range(&dn,
1173                                 start_pgofs, map->m_pblk + ofs,
1174                                 map->m_len - ofs);
1175                 }
1176         }
1177
1178         f2fs_put_dnode(&dn);
1179
1180         if (create) {
1181                 __do_map_lock(sbi, flag, false);
1182                 f2fs_balance_fs(sbi, dn.node_changed);
1183         }
1184         goto next_dnode;
1185
1186 sync_out:
1187         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1188                 if (map->m_flags & F2FS_MAP_MAPPED) {
1189                         unsigned int ofs = start_pgofs - map->m_lblk;
1190
1191                         f2fs_update_extent_cache_range(&dn,
1192                                 start_pgofs, map->m_pblk + ofs,
1193                                 map->m_len - ofs);
1194                 }
1195                 if (map->m_next_extent)
1196                         *map->m_next_extent = pgofs + 1;
1197         }
1198         f2fs_put_dnode(&dn);
1199 unlock_out:
1200         if (create) {
1201                 __do_map_lock(sbi, flag, false);
1202                 f2fs_balance_fs(sbi, dn.node_changed);
1203         }
1204 out:
1205         trace_f2fs_map_blocks(inode, map, err);
1206         return err;
1207 }
1208
1209 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1210 {
1211         struct f2fs_map_blocks map;
1212         block_t last_lblk;
1213         int err;
1214
1215         if (pos + len > i_size_read(inode))
1216                 return false;
1217
1218         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1219         map.m_next_pgofs = NULL;
1220         map.m_next_extent = NULL;
1221         map.m_seg_type = NO_CHECK_TYPE;
1222         last_lblk = F2FS_BLK_ALIGN(pos + len);
1223
1224         while (map.m_lblk < last_lblk) {
1225                 map.m_len = last_lblk - map.m_lblk;
1226                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1227                 if (err || map.m_len == 0)
1228                         return false;
1229                 map.m_lblk += map.m_len;
1230         }
1231         return true;
1232 }
1233
1234 static int __get_data_block(struct inode *inode, sector_t iblock,
1235                         struct buffer_head *bh, int create, int flag,
1236                         pgoff_t *next_pgofs, int seg_type)
1237 {
1238         struct f2fs_map_blocks map;
1239         int err;
1240
1241         map.m_lblk = iblock;
1242         map.m_len = bh->b_size >> inode->i_blkbits;
1243         map.m_next_pgofs = next_pgofs;
1244         map.m_next_extent = NULL;
1245         map.m_seg_type = seg_type;
1246
1247         err = f2fs_map_blocks(inode, &map, create, flag);
1248         if (!err) {
1249                 map_bh(bh, inode->i_sb, map.m_pblk);
1250                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1251                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1252         }
1253         return err;
1254 }
1255
1256 static int get_data_block(struct inode *inode, sector_t iblock,
1257                         struct buffer_head *bh_result, int create, int flag,
1258                         pgoff_t *next_pgofs)
1259 {
1260         return __get_data_block(inode, iblock, bh_result, create,
1261                                                         flag, next_pgofs,
1262                                                         NO_CHECK_TYPE);
1263 }
1264
1265 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1266                         struct buffer_head *bh_result, int create)
1267 {
1268         return __get_data_block(inode, iblock, bh_result, create,
1269                                                 F2FS_GET_BLOCK_DEFAULT, NULL,
1270                                                 f2fs_rw_hint_to_seg_type(
1271                                                         inode->i_write_hint));
1272 }
1273
1274 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1275                         struct buffer_head *bh_result, int create)
1276 {
1277         /* Block number less than F2FS MAX BLOCKS */
1278         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1279                 return -EFBIG;
1280
1281         return __get_data_block(inode, iblock, bh_result, create,
1282                                                 F2FS_GET_BLOCK_BMAP, NULL,
1283                                                 NO_CHECK_TYPE);
1284 }
1285
1286 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1287 {
1288         return (offset >> inode->i_blkbits);
1289 }
1290
1291 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1292 {
1293         return (blk << inode->i_blkbits);
1294 }
1295
1296 static int f2fs_xattr_fiemap(struct inode *inode,
1297                                 struct fiemap_extent_info *fieinfo)
1298 {
1299         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1300         struct page *page;
1301         struct node_info ni;
1302         __u64 phys = 0, len;
1303         __u32 flags;
1304         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1305         int err = 0;
1306
1307         if (f2fs_has_inline_xattr(inode)) {
1308                 int offset;
1309
1310                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1311                                                 inode->i_ino, false);
1312                 if (!page)
1313                         return -ENOMEM;
1314
1315                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1316                 if (err) {
1317                         f2fs_put_page(page, 1);
1318                         return err;
1319                 }
1320
1321                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1322                 offset = offsetof(struct f2fs_inode, i_addr) +
1323                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1324                                         get_inline_xattr_addrs(inode));
1325
1326                 phys += offset;
1327                 len = inline_xattr_size(inode);
1328
1329                 f2fs_put_page(page, 1);
1330
1331                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1332
1333                 if (!xnid)
1334                         flags |= FIEMAP_EXTENT_LAST;
1335
1336                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1337                 if (err || err == 1)
1338                         return err;
1339         }
1340
1341         if (xnid) {
1342                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1343                 if (!page)
1344                         return -ENOMEM;
1345
1346                 err = f2fs_get_node_info(sbi, xnid, &ni);
1347                 if (err) {
1348                         f2fs_put_page(page, 1);
1349                         return err;
1350                 }
1351
1352                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1353                 len = inode->i_sb->s_blocksize;
1354
1355                 f2fs_put_page(page, 1);
1356
1357                 flags = FIEMAP_EXTENT_LAST;
1358         }
1359
1360         if (phys)
1361                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1362
1363         return (err < 0 ? err : 0);
1364 }
1365
1366 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1367                 u64 start, u64 len)
1368 {
1369         struct buffer_head map_bh;
1370         sector_t start_blk, last_blk;
1371         pgoff_t next_pgofs;
1372         u64 logical = 0, phys = 0, size = 0;
1373         u32 flags = 0;
1374         int ret = 0;
1375
1376         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1377                 ret = f2fs_precache_extents(inode);
1378                 if (ret)
1379                         return ret;
1380         }
1381
1382         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1383         if (ret)
1384                 return ret;
1385
1386         inode_lock(inode);
1387
1388         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1389                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1390                 goto out;
1391         }
1392
1393         if (f2fs_has_inline_data(inode)) {
1394                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1395                 if (ret != -EAGAIN)
1396                         goto out;
1397         }
1398
1399         if (logical_to_blk(inode, len) == 0)
1400                 len = blk_to_logical(inode, 1);
1401
1402         start_blk = logical_to_blk(inode, start);
1403         last_blk = logical_to_blk(inode, start + len - 1);
1404
1405 next:
1406         memset(&map_bh, 0, sizeof(struct buffer_head));
1407         map_bh.b_size = len;
1408
1409         ret = get_data_block(inode, start_blk, &map_bh, 0,
1410                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1411         if (ret)
1412                 goto out;
1413
1414         /* HOLE */
1415         if (!buffer_mapped(&map_bh)) {
1416                 start_blk = next_pgofs;
1417
1418                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1419                                         F2FS_I_SB(inode)->max_file_blocks))
1420                         goto prep_next;
1421
1422                 flags |= FIEMAP_EXTENT_LAST;
1423         }
1424
1425         if (size) {
1426                 if (f2fs_encrypted_inode(inode))
1427                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1428
1429                 ret = fiemap_fill_next_extent(fieinfo, logical,
1430                                 phys, size, flags);
1431         }
1432
1433         if (start_blk > last_blk || ret)
1434                 goto out;
1435
1436         logical = blk_to_logical(inode, start_blk);
1437         phys = blk_to_logical(inode, map_bh.b_blocknr);
1438         size = map_bh.b_size;
1439         flags = 0;
1440         if (buffer_unwritten(&map_bh))
1441                 flags = FIEMAP_EXTENT_UNWRITTEN;
1442
1443         start_blk += logical_to_blk(inode, size);
1444
1445 prep_next:
1446         cond_resched();
1447         if (fatal_signal_pending(current))
1448                 ret = -EINTR;
1449         else
1450                 goto next;
1451 out:
1452         if (ret == 1)
1453                 ret = 0;
1454
1455         inode_unlock(inode);
1456         return ret;
1457 }
1458
1459 /*
1460  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1461  * Major change was from block_size == page_size in f2fs by default.
1462  *
1463  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1464  * this function ever deviates from doing just read-ahead, it should either
1465  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1466  * from read-ahead.
1467  */
1468 static int f2fs_mpage_readpages(struct address_space *mapping,
1469                         struct list_head *pages, struct page *page,
1470                         unsigned nr_pages, bool is_readahead)
1471 {
1472         struct bio *bio = NULL;
1473         sector_t last_block_in_bio = 0;
1474         struct inode *inode = mapping->host;
1475         const unsigned blkbits = inode->i_blkbits;
1476         const unsigned blocksize = 1 << blkbits;
1477         sector_t block_in_file;
1478         sector_t last_block;
1479         sector_t last_block_in_file;
1480         sector_t block_nr;
1481         struct f2fs_map_blocks map;
1482
1483         map.m_pblk = 0;
1484         map.m_lblk = 0;
1485         map.m_len = 0;
1486         map.m_flags = 0;
1487         map.m_next_pgofs = NULL;
1488         map.m_next_extent = NULL;
1489         map.m_seg_type = NO_CHECK_TYPE;
1490
1491         for (; nr_pages; nr_pages--) {
1492                 if (pages) {
1493                         page = list_last_entry(pages, struct page, lru);
1494
1495                         prefetchw(&page->flags);
1496                         list_del(&page->lru);
1497                         if (add_to_page_cache_lru(page, mapping,
1498                                                   page->index,
1499                                                   readahead_gfp_mask(mapping)))
1500                                 goto next_page;
1501                 }
1502
1503                 block_in_file = (sector_t)page->index;
1504                 last_block = block_in_file + nr_pages;
1505                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1506                                                                 blkbits;
1507                 if (last_block > last_block_in_file)
1508                         last_block = last_block_in_file;
1509
1510                 /*
1511                  * Map blocks using the previous result first.
1512                  */
1513                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1514                                 block_in_file > map.m_lblk &&
1515                                 block_in_file < (map.m_lblk + map.m_len))
1516                         goto got_it;
1517
1518                 /*
1519                  * Then do more f2fs_map_blocks() calls until we are
1520                  * done with this page.
1521                  */
1522                 map.m_flags = 0;
1523
1524                 if (block_in_file < last_block) {
1525                         map.m_lblk = block_in_file;
1526                         map.m_len = last_block - block_in_file;
1527
1528                         if (f2fs_map_blocks(inode, &map, 0,
1529                                                 F2FS_GET_BLOCK_DEFAULT))
1530                                 goto set_error_page;
1531                 }
1532 got_it:
1533                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1534                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1535                         SetPageMappedToDisk(page);
1536
1537                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1538                                 SetPageUptodate(page);
1539                                 goto confused;
1540                         }
1541
1542                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1543                                                                 DATA_GENERIC))
1544                                 goto set_error_page;
1545                 } else {
1546                         zero_user_segment(page, 0, PAGE_SIZE);
1547                         if (!PageUptodate(page))
1548                                 SetPageUptodate(page);
1549                         unlock_page(page);
1550                         goto next_page;
1551                 }
1552
1553                 /*
1554                  * This page will go to BIO.  Do we need to send this
1555                  * BIO off first?
1556                  */
1557                 if (bio && (last_block_in_bio != block_nr - 1 ||
1558                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1559 submit_and_realloc:
1560                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1561                         bio = NULL;
1562                 }
1563                 if (bio == NULL) {
1564                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1565                                         is_readahead ? REQ_RAHEAD : 0);
1566                         if (IS_ERR(bio)) {
1567                                 bio = NULL;
1568                                 goto set_error_page;
1569                         }
1570                 }
1571
1572                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1573                         goto submit_and_realloc;
1574
1575                 ClearPageError(page);
1576                 last_block_in_bio = block_nr;
1577                 goto next_page;
1578 set_error_page:
1579                 SetPageError(page);
1580                 zero_user_segment(page, 0, PAGE_SIZE);
1581                 unlock_page(page);
1582                 goto next_page;
1583 confused:
1584                 if (bio) {
1585                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1586                         bio = NULL;
1587                 }
1588                 unlock_page(page);
1589 next_page:
1590                 if (pages)
1591                         put_page(page);
1592         }
1593         BUG_ON(pages && !list_empty(pages));
1594         if (bio)
1595                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1596         return 0;
1597 }
1598
1599 static int f2fs_read_data_page(struct file *file, struct page *page)
1600 {
1601         struct inode *inode = page->mapping->host;
1602         int ret = -EAGAIN;
1603
1604         trace_f2fs_readpage(page, DATA);
1605
1606         /* If the file has inline data, try to read it directly */
1607         if (f2fs_has_inline_data(inode))
1608                 ret = f2fs_read_inline_data(inode, page);
1609         if (ret == -EAGAIN)
1610                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1611         return ret;
1612 }
1613
1614 static int f2fs_read_data_pages(struct file *file,
1615                         struct address_space *mapping,
1616                         struct list_head *pages, unsigned nr_pages)
1617 {
1618         struct inode *inode = mapping->host;
1619         struct page *page = list_last_entry(pages, struct page, lru);
1620
1621         trace_f2fs_readpages(inode, page, nr_pages);
1622
1623         /* If the file has inline data, skip readpages */
1624         if (f2fs_has_inline_data(inode))
1625                 return 0;
1626
1627         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1628 }
1629
1630 static int encrypt_one_page(struct f2fs_io_info *fio)
1631 {
1632         struct inode *inode = fio->page->mapping->host;
1633         struct page *mpage;
1634         gfp_t gfp_flags = GFP_NOFS;
1635
1636         if (!f2fs_encrypted_file(inode))
1637                 return 0;
1638
1639         /* wait for GCed page writeback via META_MAPPING */
1640         f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1641
1642 retry_encrypt:
1643         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1644                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1645         if (IS_ERR(fio->encrypted_page)) {
1646                 /* flush pending IOs and wait for a while in the ENOMEM case */
1647                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1648                         f2fs_flush_merged_writes(fio->sbi);
1649                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1650                         gfp_flags |= __GFP_NOFAIL;
1651                         goto retry_encrypt;
1652                 }
1653                 return PTR_ERR(fio->encrypted_page);
1654         }
1655
1656         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1657         if (mpage) {
1658                 if (PageUptodate(mpage))
1659                         memcpy(page_address(mpage),
1660                                 page_address(fio->encrypted_page), PAGE_SIZE);
1661                 f2fs_put_page(mpage, 1);
1662         }
1663         return 0;
1664 }
1665
1666 static inline bool check_inplace_update_policy(struct inode *inode,
1667                                 struct f2fs_io_info *fio)
1668 {
1669         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1670         unsigned int policy = SM_I(sbi)->ipu_policy;
1671
1672         if (policy & (0x1 << F2FS_IPU_FORCE))
1673                 return true;
1674         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1675                 return true;
1676         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1677                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1678                 return true;
1679         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1680                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1681                 return true;
1682
1683         /*
1684          * IPU for rewrite async pages
1685          */
1686         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1687                         fio && fio->op == REQ_OP_WRITE &&
1688                         !(fio->op_flags & REQ_SYNC) &&
1689                         !f2fs_encrypted_inode(inode))
1690                 return true;
1691
1692         /* this is only set during fdatasync */
1693         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1694                         is_inode_flag_set(inode, FI_NEED_IPU))
1695                 return true;
1696
1697         return false;
1698 }
1699
1700 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1701 {
1702         if (f2fs_is_pinned_file(inode))
1703                 return true;
1704
1705         /* if this is cold file, we should overwrite to avoid fragmentation */
1706         if (file_is_cold(inode))
1707                 return true;
1708
1709         return check_inplace_update_policy(inode, fio);
1710 }
1711
1712 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1713 {
1714         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1715
1716         if (test_opt(sbi, LFS))
1717                 return true;
1718         if (S_ISDIR(inode->i_mode))
1719                 return true;
1720         if (f2fs_is_atomic_file(inode))
1721                 return true;
1722         if (fio) {
1723                 if (is_cold_data(fio->page))
1724                         return true;
1725                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1726                         return true;
1727         }
1728         return false;
1729 }
1730
1731 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1732 {
1733         struct inode *inode = fio->page->mapping->host;
1734
1735         if (f2fs_should_update_outplace(inode, fio))
1736                 return false;
1737
1738         return f2fs_should_update_inplace(inode, fio);
1739 }
1740
1741 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1742 {
1743         struct page *page = fio->page;
1744         struct inode *inode = page->mapping->host;
1745         struct dnode_of_data dn;
1746         struct extent_info ei = {0,0,0};
1747         struct node_info ni;
1748         bool ipu_force = false;
1749         int err = 0;
1750
1751         set_new_dnode(&dn, inode, NULL, NULL, 0);
1752         if (need_inplace_update(fio) &&
1753                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1754                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1755
1756                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1757                                                         DATA_GENERIC))
1758                         return -EFSCORRUPTED;
1759
1760                 ipu_force = true;
1761                 fio->need_lock = LOCK_DONE;
1762                 goto got_it;
1763         }
1764
1765         /* Deadlock due to between page->lock and f2fs_lock_op */
1766         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1767                 return -EAGAIN;
1768
1769         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1770         if (err)
1771                 goto out;
1772
1773         fio->old_blkaddr = dn.data_blkaddr;
1774
1775         /* This page is already truncated */
1776         if (fio->old_blkaddr == NULL_ADDR) {
1777                 ClearPageUptodate(page);
1778                 goto out_writepage;
1779         }
1780 got_it:
1781         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1782                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1783                                                         DATA_GENERIC)) {
1784                 err = -EFSCORRUPTED;
1785                 goto out_writepage;
1786         }
1787         /*
1788          * If current allocation needs SSR,
1789          * it had better in-place writes for updated data.
1790          */
1791         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1792                                         need_inplace_update(fio))) {
1793                 err = encrypt_one_page(fio);
1794                 if (err)
1795                         goto out_writepage;
1796
1797                 set_page_writeback(page);
1798                 ClearPageError(page);
1799                 f2fs_put_dnode(&dn);
1800                 if (fio->need_lock == LOCK_REQ)
1801                         f2fs_unlock_op(fio->sbi);
1802                 err = f2fs_inplace_write_data(fio);
1803                 trace_f2fs_do_write_data_page(fio->page, IPU);
1804                 set_inode_flag(inode, FI_UPDATE_WRITE);
1805                 return err;
1806         }
1807
1808         if (fio->need_lock == LOCK_RETRY) {
1809                 if (!f2fs_trylock_op(fio->sbi)) {
1810                         err = -EAGAIN;
1811                         goto out_writepage;
1812                 }
1813                 fio->need_lock = LOCK_REQ;
1814         }
1815
1816         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1817         if (err)
1818                 goto out_writepage;
1819
1820         fio->version = ni.version;
1821
1822         err = encrypt_one_page(fio);
1823         if (err)
1824                 goto out_writepage;
1825
1826         set_page_writeback(page);
1827         ClearPageError(page);
1828
1829         /* LFS mode write path */
1830         f2fs_outplace_write_data(&dn, fio);
1831         trace_f2fs_do_write_data_page(page, OPU);
1832         set_inode_flag(inode, FI_APPEND_WRITE);
1833         if (page->index == 0)
1834                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1835 out_writepage:
1836         f2fs_put_dnode(&dn);
1837 out:
1838         if (fio->need_lock == LOCK_REQ)
1839                 f2fs_unlock_op(fio->sbi);
1840         return err;
1841 }
1842
1843 static int __write_data_page(struct page *page, bool *submitted,
1844                                 struct writeback_control *wbc,
1845                                 enum iostat_type io_type)
1846 {
1847         struct inode *inode = page->mapping->host;
1848         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1849         loff_t i_size = i_size_read(inode);
1850         const pgoff_t end_index = ((unsigned long long) i_size)
1851                                                         >> PAGE_SHIFT;
1852         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1853         unsigned offset = 0;
1854         bool need_balance_fs = false;
1855         int err = 0;
1856         struct f2fs_io_info fio = {
1857                 .sbi = sbi,
1858                 .ino = inode->i_ino,
1859                 .type = DATA,
1860                 .op = REQ_OP_WRITE,
1861                 .op_flags = wbc_to_write_flags(wbc),
1862                 .old_blkaddr = NULL_ADDR,
1863                 .page = page,
1864                 .encrypted_page = NULL,
1865                 .submitted = false,
1866                 .need_lock = LOCK_RETRY,
1867                 .io_type = io_type,
1868                 .io_wbc = wbc,
1869         };
1870
1871         trace_f2fs_writepage(page, DATA);
1872
1873         /* we should bypass data pages to proceed the kworkder jobs */
1874         if (unlikely(f2fs_cp_error(sbi))) {
1875                 mapping_set_error(page->mapping, -EIO);
1876                 /*
1877                  * don't drop any dirty dentry pages for keeping lastest
1878                  * directory structure.
1879                  */
1880                 if (S_ISDIR(inode->i_mode))
1881                         goto redirty_out;
1882                 goto out;
1883         }
1884
1885         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1886                 goto redirty_out;
1887
1888         if (page->index < end_index)
1889                 goto write;
1890
1891         /*
1892          * If the offset is out-of-range of file size,
1893          * this page does not have to be written to disk.
1894          */
1895         offset = i_size & (PAGE_SIZE - 1);
1896         if ((page->index >= end_index + 1) || !offset)
1897                 goto out;
1898
1899         zero_user_segment(page, offset, PAGE_SIZE);
1900 write:
1901         if (f2fs_is_drop_cache(inode))
1902                 goto out;
1903         /* we should not write 0'th page having journal header */
1904         if (f2fs_is_volatile_file(inode) && (!page->index ||
1905                         (!wbc->for_reclaim &&
1906                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1907                 goto redirty_out;
1908
1909         /* Dentry blocks are controlled by checkpoint */
1910         if (S_ISDIR(inode->i_mode)) {
1911                 fio.need_lock = LOCK_DONE;
1912                 err = f2fs_do_write_data_page(&fio);
1913                 goto done;
1914         }
1915
1916         if (!wbc->for_reclaim)
1917                 need_balance_fs = true;
1918         else if (has_not_enough_free_secs(sbi, 0, 0))
1919                 goto redirty_out;
1920         else
1921                 set_inode_flag(inode, FI_HOT_DATA);
1922
1923         err = -EAGAIN;
1924         if (f2fs_has_inline_data(inode)) {
1925                 err = f2fs_write_inline_data(inode, page);
1926                 if (!err)
1927                         goto out;
1928         }
1929
1930         if (err == -EAGAIN) {
1931                 err = f2fs_do_write_data_page(&fio);
1932                 if (err == -EAGAIN) {
1933                         fio.need_lock = LOCK_REQ;
1934                         err = f2fs_do_write_data_page(&fio);
1935                 }
1936         }
1937
1938         if (err) {
1939                 file_set_keep_isize(inode);
1940         } else {
1941                 down_write(&F2FS_I(inode)->i_sem);
1942                 if (F2FS_I(inode)->last_disk_size < psize)
1943                         F2FS_I(inode)->last_disk_size = psize;
1944                 up_write(&F2FS_I(inode)->i_sem);
1945         }
1946
1947 done:
1948         if (err && err != -ENOENT)
1949                 goto redirty_out;
1950
1951 out:
1952         inode_dec_dirty_pages(inode);
1953         if (err)
1954                 ClearPageUptodate(page);
1955
1956         if (wbc->for_reclaim) {
1957                 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1958                 clear_inode_flag(inode, FI_HOT_DATA);
1959                 f2fs_remove_dirty_inode(inode);
1960                 submitted = NULL;
1961         }
1962
1963         unlock_page(page);
1964         if (!S_ISDIR(inode->i_mode))
1965                 f2fs_balance_fs(sbi, need_balance_fs);
1966
1967         if (unlikely(f2fs_cp_error(sbi))) {
1968                 f2fs_submit_merged_write(sbi, DATA);
1969                 submitted = NULL;
1970         }
1971
1972         if (submitted)
1973                 *submitted = fio.submitted;
1974
1975         return 0;
1976
1977 redirty_out:
1978         redirty_page_for_writepage(wbc, page);
1979         /*
1980          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
1981          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
1982          * file_write_and_wait_range() will see EIO error, which is critical
1983          * to return value of fsync() followed by atomic_write failure to user.
1984          */
1985         if (!err || wbc->for_reclaim)
1986                 return AOP_WRITEPAGE_ACTIVATE;
1987         unlock_page(page);
1988         return err;
1989 }
1990
1991 static int f2fs_write_data_page(struct page *page,
1992                                         struct writeback_control *wbc)
1993 {
1994         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1995 }
1996
1997 /*
1998  * This function was copied from write_cche_pages from mm/page-writeback.c.
1999  * The major change is making write step of cold data page separately from
2000  * warm/hot data page.
2001  */
2002 static int f2fs_write_cache_pages(struct address_space *mapping,
2003                                         struct writeback_control *wbc,
2004                                         enum iostat_type io_type)
2005 {
2006         int ret = 0;
2007         int done = 0;
2008         struct pagevec pvec;
2009         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2010         int nr_pages;
2011         pgoff_t uninitialized_var(writeback_index);
2012         pgoff_t index;
2013         pgoff_t end;            /* Inclusive */
2014         pgoff_t done_index;
2015         pgoff_t last_idx = ULONG_MAX;
2016         int cycled;
2017         int range_whole = 0;
2018         int tag;
2019
2020         pagevec_init(&pvec);
2021
2022         if (get_dirty_pages(mapping->host) <=
2023                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2024                 set_inode_flag(mapping->host, FI_HOT_DATA);
2025         else
2026                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2027
2028         if (wbc->range_cyclic) {
2029                 writeback_index = mapping->writeback_index; /* prev offset */
2030                 index = writeback_index;
2031                 if (index == 0)
2032                         cycled = 1;
2033                 else
2034                         cycled = 0;
2035                 end = -1;
2036         } else {
2037                 index = wbc->range_start >> PAGE_SHIFT;
2038                 end = wbc->range_end >> PAGE_SHIFT;
2039                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2040                         range_whole = 1;
2041                 cycled = 1; /* ignore range_cyclic tests */
2042         }
2043         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2044                 tag = PAGECACHE_TAG_TOWRITE;
2045         else
2046                 tag = PAGECACHE_TAG_DIRTY;
2047 retry:
2048         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2049                 tag_pages_for_writeback(mapping, index, end);
2050         done_index = index;
2051         while (!done && (index <= end)) {
2052                 int i;
2053
2054                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2055                                 tag);
2056                 if (nr_pages == 0)
2057                         break;
2058
2059                 for (i = 0; i < nr_pages; i++) {
2060                         struct page *page = pvec.pages[i];
2061                         bool submitted = false;
2062
2063                         /* give a priority to WB_SYNC threads */
2064                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2065                                         wbc->sync_mode == WB_SYNC_NONE) {
2066                                 done = 1;
2067                                 break;
2068                         }
2069
2070                         done_index = page->index;
2071 retry_write:
2072                         lock_page(page);
2073
2074                         if (unlikely(page->mapping != mapping)) {
2075 continue_unlock:
2076                                 unlock_page(page);
2077                                 continue;
2078                         }
2079
2080                         if (!PageDirty(page)) {
2081                                 /* someone wrote it for us */
2082                                 goto continue_unlock;
2083                         }
2084
2085                         if (PageWriteback(page)) {
2086                                 if (wbc->sync_mode != WB_SYNC_NONE)
2087                                         f2fs_wait_on_page_writeback(page,
2088                                                                 DATA, true);
2089                                 else
2090                                         goto continue_unlock;
2091                         }
2092
2093                         BUG_ON(PageWriteback(page));
2094                         if (!clear_page_dirty_for_io(page))
2095                                 goto continue_unlock;
2096
2097                         ret = __write_data_page(page, &submitted, wbc, io_type);
2098                         if (unlikely(ret)) {
2099                                 /*
2100                                  * keep nr_to_write, since vfs uses this to
2101                                  * get # of written pages.
2102                                  */
2103                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2104                                         unlock_page(page);
2105                                         ret = 0;
2106                                         continue;
2107                                 } else if (ret == -EAGAIN) {
2108                                         ret = 0;
2109                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2110                                                 cond_resched();
2111                                                 congestion_wait(BLK_RW_ASYNC,
2112                                                                         HZ/50);
2113                                                 goto retry_write;
2114                                         }
2115                                         continue;
2116                                 }
2117                                 done_index = page->index + 1;
2118                                 done = 1;
2119                                 break;
2120                         } else if (submitted) {
2121                                 last_idx = page->index;
2122                         }
2123
2124                         if (--wbc->nr_to_write <= 0 &&
2125                                         wbc->sync_mode == WB_SYNC_NONE) {
2126                                 done = 1;
2127                                 break;
2128                         }
2129                 }
2130                 pagevec_release(&pvec);
2131                 cond_resched();
2132         }
2133
2134         if (!cycled && !done) {
2135                 cycled = 1;
2136                 index = 0;
2137                 end = writeback_index - 1;
2138                 goto retry;
2139         }
2140         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2141                 mapping->writeback_index = done_index;
2142
2143         if (last_idx != ULONG_MAX)
2144                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2145                                                 0, last_idx, DATA);
2146
2147         return ret;
2148 }
2149
2150 static inline bool __should_serialize_io(struct inode *inode,
2151                                         struct writeback_control *wbc)
2152 {
2153         if (!S_ISREG(inode->i_mode))
2154                 return false;
2155         if (wbc->sync_mode != WB_SYNC_ALL)
2156                 return true;
2157         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2158                 return true;
2159         return false;
2160 }
2161
2162 static int __f2fs_write_data_pages(struct address_space *mapping,
2163                                                 struct writeback_control *wbc,
2164                                                 enum iostat_type io_type)
2165 {
2166         struct inode *inode = mapping->host;
2167         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2168         struct blk_plug plug;
2169         int ret;
2170         bool locked = false;
2171
2172         /* deal with chardevs and other special file */
2173         if (!mapping->a_ops->writepage)
2174                 return 0;
2175
2176         /* skip writing if there is no dirty page in this inode */
2177         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2178                 return 0;
2179
2180         /* during POR, we don't need to trigger writepage at all. */
2181         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2182                 goto skip_write;
2183
2184         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2185                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2186                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2187                 goto skip_write;
2188
2189         /* skip writing during file defragment */
2190         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2191                 goto skip_write;
2192
2193         trace_f2fs_writepages(mapping->host, wbc, DATA);
2194
2195         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2196         if (wbc->sync_mode == WB_SYNC_ALL)
2197                 atomic_inc(&sbi->wb_sync_req[DATA]);
2198         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2199                 goto skip_write;
2200
2201         if (__should_serialize_io(inode, wbc)) {
2202                 mutex_lock(&sbi->writepages);
2203                 locked = true;
2204         }
2205
2206         blk_start_plug(&plug);
2207         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2208         blk_finish_plug(&plug);
2209
2210         if (locked)
2211                 mutex_unlock(&sbi->writepages);
2212
2213         if (wbc->sync_mode == WB_SYNC_ALL)
2214                 atomic_dec(&sbi->wb_sync_req[DATA]);
2215         /*
2216          * if some pages were truncated, we cannot guarantee its mapping->host
2217          * to detect pending bios.
2218          */
2219
2220         f2fs_remove_dirty_inode(inode);
2221         return ret;
2222
2223 skip_write:
2224         wbc->pages_skipped += get_dirty_pages(inode);
2225         trace_f2fs_writepages(mapping->host, wbc, DATA);
2226         return 0;
2227 }
2228
2229 static int f2fs_write_data_pages(struct address_space *mapping,
2230                             struct writeback_control *wbc)
2231 {
2232         struct inode *inode = mapping->host;
2233
2234         return __f2fs_write_data_pages(mapping, wbc,
2235                         F2FS_I(inode)->cp_task == current ?
2236                         FS_CP_DATA_IO : FS_DATA_IO);
2237 }
2238
2239 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2240 {
2241         struct inode *inode = mapping->host;
2242         loff_t i_size = i_size_read(inode);
2243
2244         if (to > i_size) {
2245                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2246                 down_write(&F2FS_I(inode)->i_mmap_sem);
2247
2248                 truncate_pagecache(inode, i_size);
2249                 f2fs_truncate_blocks(inode, i_size, true);
2250
2251                 up_write(&F2FS_I(inode)->i_mmap_sem);
2252                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2253         }
2254 }
2255
2256 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2257                         struct page *page, loff_t pos, unsigned len,
2258                         block_t *blk_addr, bool *node_changed)
2259 {
2260         struct inode *inode = page->mapping->host;
2261         pgoff_t index = page->index;
2262         struct dnode_of_data dn;
2263         struct page *ipage;
2264         bool locked = false;
2265         struct extent_info ei = {0,0,0};
2266         int err = 0;
2267         int flag;
2268
2269         /*
2270          * we already allocated all the blocks, so we don't need to get
2271          * the block addresses when there is no need to fill the page.
2272          */
2273         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2274                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2275                 return 0;
2276
2277         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2278         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2279                 flag = F2FS_GET_BLOCK_DEFAULT;
2280         else
2281                 flag = F2FS_GET_BLOCK_PRE_AIO;
2282
2283         if (f2fs_has_inline_data(inode) ||
2284                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2285                 __do_map_lock(sbi, flag, true);
2286                 locked = true;
2287         }
2288 restart:
2289         /* check inline_data */
2290         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2291         if (IS_ERR(ipage)) {
2292                 err = PTR_ERR(ipage);
2293                 goto unlock_out;
2294         }
2295
2296         set_new_dnode(&dn, inode, ipage, ipage, 0);
2297
2298         if (f2fs_has_inline_data(inode)) {
2299                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2300                         f2fs_do_read_inline_data(page, ipage);
2301                         set_inode_flag(inode, FI_DATA_EXIST);
2302                         if (inode->i_nlink)
2303                                 set_inline_node(ipage);
2304                 } else {
2305                         err = f2fs_convert_inline_page(&dn, page);
2306                         if (err)
2307                                 goto out;
2308                         if (dn.data_blkaddr == NULL_ADDR)
2309                                 err = f2fs_get_block(&dn, index);
2310                 }
2311         } else if (locked) {
2312                 err = f2fs_get_block(&dn, index);
2313         } else {
2314                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2315                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2316                 } else {
2317                         /* hole case */
2318                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2319                         if (err || dn.data_blkaddr == NULL_ADDR) {
2320                                 f2fs_put_dnode(&dn);
2321                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2322                                                                 true);
2323                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2324                                 locked = true;
2325                                 goto restart;
2326                         }
2327                 }
2328         }
2329
2330         /* convert_inline_page can make node_changed */
2331         *blk_addr = dn.data_blkaddr;
2332         *node_changed = dn.node_changed;
2333 out:
2334         f2fs_put_dnode(&dn);
2335 unlock_out:
2336         if (locked)
2337                 __do_map_lock(sbi, flag, false);
2338         return err;
2339 }
2340
2341 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2342                 loff_t pos, unsigned len, unsigned flags,
2343                 struct page **pagep, void **fsdata)
2344 {
2345         struct inode *inode = mapping->host;
2346         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2347         struct page *page = NULL;
2348         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2349         bool need_balance = false, drop_atomic = false;
2350         block_t blkaddr = NULL_ADDR;
2351         int err = 0;
2352
2353         trace_f2fs_write_begin(inode, pos, len, flags);
2354
2355         if ((f2fs_is_atomic_file(inode) &&
2356                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2357                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2358                 err = -ENOMEM;
2359                 drop_atomic = true;
2360                 goto fail;
2361         }
2362
2363         /*
2364          * We should check this at this moment to avoid deadlock on inode page
2365          * and #0 page. The locking rule for inline_data conversion should be:
2366          * lock_page(page #0) -> lock_page(inode_page)
2367          */
2368         if (index != 0) {
2369                 err = f2fs_convert_inline_inode(inode);
2370                 if (err)
2371                         goto fail;
2372         }
2373 repeat:
2374         /*
2375          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2376          * wait_for_stable_page. Will wait that below with our IO control.
2377          */
2378         page = f2fs_pagecache_get_page(mapping, index,
2379                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2380         if (!page) {
2381                 err = -ENOMEM;
2382                 goto fail;
2383         }
2384
2385         *pagep = page;
2386
2387         err = prepare_write_begin(sbi, page, pos, len,
2388                                         &blkaddr, &need_balance);
2389         if (err)
2390                 goto fail;
2391
2392         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2393                 unlock_page(page);
2394                 f2fs_balance_fs(sbi, true);
2395                 lock_page(page);
2396                 if (page->mapping != mapping) {
2397                         /* The page got truncated from under us */
2398                         f2fs_put_page(page, 1);
2399                         goto repeat;
2400                 }
2401         }
2402
2403         f2fs_wait_on_page_writeback(page, DATA, false);
2404
2405         /* wait for GCed page writeback via META_MAPPING */
2406         if (f2fs_post_read_required(inode))
2407                 f2fs_wait_on_block_writeback(sbi, blkaddr);
2408
2409         if (len == PAGE_SIZE || PageUptodate(page))
2410                 return 0;
2411
2412         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2413                 zero_user_segment(page, len, PAGE_SIZE);
2414                 return 0;
2415         }
2416
2417         if (blkaddr == NEW_ADDR) {
2418                 zero_user_segment(page, 0, PAGE_SIZE);
2419                 SetPageUptodate(page);
2420         } else {
2421                 err = f2fs_submit_page_read(inode, page, blkaddr);
2422                 if (err)
2423                         goto fail;
2424
2425                 lock_page(page);
2426                 if (unlikely(page->mapping != mapping)) {
2427                         f2fs_put_page(page, 1);
2428                         goto repeat;
2429                 }
2430                 if (unlikely(!PageUptodate(page))) {
2431                         err = -EIO;
2432                         goto fail;
2433                 }
2434         }
2435         return 0;
2436
2437 fail:
2438         f2fs_put_page(page, 1);
2439         f2fs_write_failed(mapping, pos + len);
2440         if (drop_atomic)
2441                 f2fs_drop_inmem_pages_all(sbi, false);
2442         return err;
2443 }
2444
2445 static int f2fs_write_end(struct file *file,
2446                         struct address_space *mapping,
2447                         loff_t pos, unsigned len, unsigned copied,
2448                         struct page *page, void *fsdata)
2449 {
2450         struct inode *inode = page->mapping->host;
2451
2452         trace_f2fs_write_end(inode, pos, len, copied);
2453
2454         /*
2455          * This should be come from len == PAGE_SIZE, and we expect copied
2456          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2457          * let generic_perform_write() try to copy data again through copied=0.
2458          */
2459         if (!PageUptodate(page)) {
2460                 if (unlikely(copied != len))
2461                         copied = 0;
2462                 else
2463                         SetPageUptodate(page);
2464         }
2465         if (!copied)
2466                 goto unlock_out;
2467
2468         set_page_dirty(page);
2469
2470         if (pos + copied > i_size_read(inode))
2471                 f2fs_i_size_write(inode, pos + copied);
2472 unlock_out:
2473         f2fs_put_page(page, 1);
2474         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2475         return copied;
2476 }
2477
2478 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2479                            loff_t offset)
2480 {
2481         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2482         unsigned blkbits = i_blkbits;
2483         unsigned blocksize_mask = (1 << blkbits) - 1;
2484         unsigned long align = offset | iov_iter_alignment(iter);
2485         struct block_device *bdev = inode->i_sb->s_bdev;
2486
2487         if (align & blocksize_mask) {
2488                 if (bdev)
2489                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2490                 blocksize_mask = (1 << blkbits) - 1;
2491                 if (align & blocksize_mask)
2492                         return -EINVAL;
2493                 return 1;
2494         }
2495         return 0;
2496 }
2497
2498 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2499 {
2500         struct address_space *mapping = iocb->ki_filp->f_mapping;
2501         struct inode *inode = mapping->host;
2502         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2503         size_t count = iov_iter_count(iter);
2504         loff_t offset = iocb->ki_pos;
2505         int rw = iov_iter_rw(iter);
2506         int err;
2507         enum rw_hint hint = iocb->ki_hint;
2508         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2509
2510         err = check_direct_IO(inode, iter, offset);
2511         if (err)
2512                 return err < 0 ? err : 0;
2513
2514         if (f2fs_force_buffered_io(inode, rw))
2515                 return 0;
2516
2517         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2518
2519         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2520                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2521
2522         if (!down_read_trylock(&F2FS_I(inode)->i_gc_rwsem[rw])) {
2523                 if (iocb->ki_flags & IOCB_NOWAIT) {
2524                         iocb->ki_hint = hint;
2525                         err = -EAGAIN;
2526                         goto out;
2527                 }
2528                 down_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2529         }
2530
2531         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2532         up_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2533
2534         if (rw == WRITE) {
2535                 if (whint_mode == WHINT_MODE_OFF)
2536                         iocb->ki_hint = hint;
2537                 if (err > 0) {
2538                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2539                                                                         err);
2540                         set_inode_flag(inode, FI_UPDATE_WRITE);
2541                 } else if (err < 0) {
2542                         f2fs_write_failed(mapping, offset + count);
2543                 }
2544         }
2545
2546 out:
2547         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2548
2549         return err;
2550 }
2551
2552 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2553                                                         unsigned int length)
2554 {
2555         struct inode *inode = page->mapping->host;
2556         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2557
2558         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2559                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2560                 return;
2561
2562         if (PageDirty(page)) {
2563                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2564                         dec_page_count(sbi, F2FS_DIRTY_META);
2565                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2566                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2567                 } else {
2568                         inode_dec_dirty_pages(inode);
2569                         f2fs_remove_dirty_inode(inode);
2570                 }
2571         }
2572
2573         /* This is atomic written page, keep Private */
2574         if (IS_ATOMIC_WRITTEN_PAGE(page))
2575                 return f2fs_drop_inmem_page(inode, page);
2576
2577         set_page_private(page, 0);
2578         ClearPagePrivate(page);
2579 }
2580
2581 int f2fs_release_page(struct page *page, gfp_t wait)
2582 {
2583         /* If this is dirty page, keep PagePrivate */
2584         if (PageDirty(page))
2585                 return 0;
2586
2587         /* This is atomic written page, keep Private */
2588         if (IS_ATOMIC_WRITTEN_PAGE(page))
2589                 return 0;
2590
2591         set_page_private(page, 0);
2592         ClearPagePrivate(page);
2593         return 1;
2594 }
2595
2596 static int f2fs_set_data_page_dirty(struct page *page)
2597 {
2598         struct address_space *mapping = page->mapping;
2599         struct inode *inode = mapping->host;
2600
2601         trace_f2fs_set_page_dirty(page, DATA);
2602
2603         if (!PageUptodate(page))
2604                 SetPageUptodate(page);
2605
2606         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2607                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2608                         f2fs_register_inmem_page(inode, page);
2609                         return 1;
2610                 }
2611                 /*
2612                  * Previously, this page has been registered, we just
2613                  * return here.
2614                  */
2615                 return 0;
2616         }
2617
2618         if (!PageDirty(page)) {
2619                 __set_page_dirty_nobuffers(page);
2620                 f2fs_update_dirty_page(inode, page);
2621                 return 1;
2622         }
2623         return 0;
2624 }
2625
2626 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2627 {
2628         struct inode *inode = mapping->host;
2629
2630         if (f2fs_has_inline_data(inode))
2631                 return 0;
2632
2633         /* make sure allocating whole blocks */
2634         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2635                 filemap_write_and_wait(mapping);
2636
2637         return generic_block_bmap(mapping, block, get_data_block_bmap);
2638 }
2639
2640 #ifdef CONFIG_MIGRATION
2641 #include <linux/migrate.h>
2642
2643 int f2fs_migrate_page(struct address_space *mapping,
2644                 struct page *newpage, struct page *page, enum migrate_mode mode)
2645 {
2646         int rc, extra_count;
2647         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2648         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2649
2650         BUG_ON(PageWriteback(page));
2651
2652         /* migrating an atomic written page is safe with the inmem_lock hold */
2653         if (atomic_written) {
2654                 if (mode != MIGRATE_SYNC)
2655                         return -EBUSY;
2656                 if (!mutex_trylock(&fi->inmem_lock))
2657                         return -EAGAIN;
2658         }
2659
2660         /*
2661          * A reference is expected if PagePrivate set when move mapping,
2662          * however F2FS breaks this for maintaining dirty page counts when
2663          * truncating pages. So here adjusting the 'extra_count' make it work.
2664          */
2665         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2666         rc = migrate_page_move_mapping(mapping, newpage,
2667                                 page, NULL, mode, extra_count);
2668         if (rc != MIGRATEPAGE_SUCCESS) {
2669                 if (atomic_written)
2670                         mutex_unlock(&fi->inmem_lock);
2671                 return rc;
2672         }
2673
2674         if (atomic_written) {
2675                 struct inmem_pages *cur;
2676                 list_for_each_entry(cur, &fi->inmem_pages, list)
2677                         if (cur->page == page) {
2678                                 cur->page = newpage;
2679                                 break;
2680                         }
2681                 mutex_unlock(&fi->inmem_lock);
2682                 put_page(page);
2683                 get_page(newpage);
2684         }
2685
2686         if (PagePrivate(page))
2687                 SetPagePrivate(newpage);
2688         set_page_private(newpage, page_private(page));
2689
2690         if (mode != MIGRATE_SYNC_NO_COPY)
2691                 migrate_page_copy(newpage, page);
2692         else
2693                 migrate_page_states(newpage, page);
2694
2695         return MIGRATEPAGE_SUCCESS;
2696 }
2697 #endif
2698
2699 const struct address_space_operations f2fs_dblock_aops = {
2700         .readpage       = f2fs_read_data_page,
2701         .readpages      = f2fs_read_data_pages,
2702         .writepage      = f2fs_write_data_page,
2703         .writepages     = f2fs_write_data_pages,
2704         .write_begin    = f2fs_write_begin,
2705         .write_end      = f2fs_write_end,
2706         .set_page_dirty = f2fs_set_data_page_dirty,
2707         .invalidatepage = f2fs_invalidate_page,
2708         .releasepage    = f2fs_release_page,
2709         .direct_IO      = f2fs_direct_IO,
2710         .bmap           = f2fs_bmap,
2711 #ifdef CONFIG_MIGRATION
2712         .migratepage    = f2fs_migrate_page,
2713 #endif
2714 };
2715
2716 void f2fs_clear_radix_tree_dirty_tag(struct page *page)
2717 {
2718         struct address_space *mapping = page_mapping(page);
2719         unsigned long flags;
2720
2721         xa_lock_irqsave(&mapping->i_pages, flags);
2722         radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2723                                                 PAGECACHE_TAG_DIRTY);
2724         xa_unlock_irqrestore(&mapping->i_pages, flags);
2725 }
2726
2727 int __init f2fs_init_post_read_processing(void)
2728 {
2729         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2730         if (!bio_post_read_ctx_cache)
2731                 goto fail;
2732         bio_post_read_ctx_pool =
2733                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2734                                          bio_post_read_ctx_cache);
2735         if (!bio_post_read_ctx_pool)
2736                 goto fail_free_cache;
2737         return 0;
2738
2739 fail_free_cache:
2740         kmem_cache_destroy(bio_post_read_ctx_cache);
2741 fail:
2742         return -ENOMEM;
2743 }
2744
2745 void __exit f2fs_destroy_post_read_processing(void)
2746 {
2747         mempool_destroy(bio_post_read_ctx_pool);
2748         kmem_cache_destroy(bio_post_read_ctx_cache);
2749 }