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                         SetPageError(page);
  84                 } else {
  85                         SetPageUptodate(page);
  86                 }
  87                 unlock_page(page);
  88         }
  89         if (bio->bi_private)
  90                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
  91         bio_put(bio);
  92 }
  93
  94 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
  95
  96 static void decrypt_work(struct work_struct *work)
  97 {
  98         struct bio_post_read_ctx *ctx =
  99                 container_of(work, struct bio_post_read_ctx, work);
 100
 101         fscrypt_decrypt_bio(ctx->bio);
 102
 103         bio_post_read_processing(ctx);
 104 }
 105
 106 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 107 {
 108         switch (++ctx->cur_step) {
 109         case STEP_DECRYPT:
 110                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 111                         INIT_WORK(&ctx->work, decrypt_work);
 112                         fscrypt_enqueue_decrypt_work(&ctx->work);
 113                         return;
 114                 }
 115                 ctx->cur_step++;
 116                 /* fall-through */
 117         default:
 118                 __read_end_io(ctx->bio);
 119         }
 120 }
 121
 122 static bool f2fs_bio_post_read_required(struct bio *bio)
 123 {
 124         return bio->bi_private && !bio->bi_status;
 125 }
 126
 127 static void f2fs_read_end_io(struct bio *bio)
 128 {
 129 #ifdef CONFIG_F2FS_FAULT_INJECTION
 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 #endif
 135
 136         if (f2fs_bio_post_read_required(bio)) {
 137                 struct bio_post_read_ctx *ctx = bio->bi_private;
 138
 139                 ctx->cur_step = STEP_INITIAL;
 140                 bio_post_read_processing(ctx);
 141                 return;
 142         }
 143
 144         __read_end_io(bio);
 145 }
 146
 147 static void f2fs_write_end_io(struct bio *bio)
 148 {
 149         struct f2fs_sb_info *sbi = bio->bi_private;
 150         struct bio_vec *bvec;
 151         int i;
 152
 153         bio_for_each_segment_all(bvec, bio, i) {
 154                 struct page *page = bvec->bv_page;
 155                 enum count_type type = WB_DATA_TYPE(page);
 156
 157                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 158                         set_page_private(page, (unsigned long)NULL);
 159                         ClearPagePrivate(page);
 160                         unlock_page(page);
 161                         mempool_free(page, sbi->write_io_dummy);
 162
 163                         if (unlikely(bio->bi_status))
 164                                 f2fs_stop_checkpoint(sbi, true);
 165                         continue;
 166                 }
 167
 168                 fscrypt_pullback_bio_page(&page, true);
 169
 170                 if (unlikely(bio->bi_status)) {
 171                         mapping_set_error(page->mapping, -EIO);
 172                         if (type == F2FS_WB_CP_DATA)
 173                                 f2fs_stop_checkpoint(sbi, true);
 174                 }
 175
 176                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 177                                         page->index != nid_of_node(page));
 178
 179                 dec_page_count(sbi, type);
 180                 clear_cold_data(page);
 181                 end_page_writeback(page);
 182         }
 183         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 184                                 wq_has_sleeper(&sbi->cp_wait))
 185                 wake_up(&sbi->cp_wait);
 186
 187         bio_put(bio);
 188 }
 189
 190 /*
 191  * Return true, if pre_bio's bdev is same as its target device.
 192  */
 193 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 194                                 block_t blk_addr, struct bio *bio)
 195 {
 196         struct block_device *bdev = sbi->sb->s_bdev;
 197         int i;
 198
 199         for (i = 0; i < sbi->s_ndevs; i++) {
 200                 if (FDEV(i).start_blk <= blk_addr &&
 201                                         FDEV(i).end_blk >= blk_addr) {
 202                         blk_addr -= FDEV(i).start_blk;
 203                         bdev = FDEV(i).bdev;
 204                         break;
 205                 }
 206         }
 207         if (bio) {
 208                 bio_set_dev(bio, bdev);
 209                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 210         }
 211         return bdev;
 212 }
 213
 214 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 215 {
 216         int i;
 217
 218         for (i = 0; i < sbi->s_ndevs; i++)
 219                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 220                         return i;
 221         return 0;
 222 }
 223
 224 static bool __same_bdev(struct f2fs_sb_info *sbi,
 225                                 block_t blk_addr, struct bio *bio)
 226 {
 227         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 228         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 229 }
 230
 231 /*
 232  * Low-level block read/write IO operations.
 233  */
 234 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 235                                 struct writeback_control *wbc,
 236                                 int npages, bool is_read,
 237                                 enum page_type type, enum temp_type temp)
 238 {
 239         struct bio *bio;
 240
 241         bio = f2fs_bio_alloc(sbi, npages, true);
 242
 243         f2fs_target_device(sbi, blk_addr, bio);
 244         if (is_read) {
 245                 bio->bi_end_io = f2fs_read_end_io;
 246                 bio->bi_private = NULL;
 247         } else {
 248                 bio->bi_end_io = f2fs_write_end_io;
 249                 bio->bi_private = sbi;
 250                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 251         }
 252         if (wbc)
 253                 wbc_init_bio(wbc, bio);
 254
 255         return bio;
 256 }
 257
 258 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 259                                 struct bio *bio, enum page_type type)
 260 {
 261         if (!is_read_io(bio_op(bio))) {
 262                 unsigned int start;
 263
 264                 if (type != DATA && type != NODE)
 265                         goto submit_io;
 266
 267                 if (test_opt(sbi, LFS) && current->plug)
 268                         blk_finish_plug(current->plug);
 269
 270                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 271                 start %= F2FS_IO_SIZE(sbi);
 272
 273                 if (start == 0)
 274                         goto submit_io;
 275
 276                 /* fill dummy pages */
 277                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 278                         struct page *page =
 279                                 mempool_alloc(sbi->write_io_dummy,
 280                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 281                         f2fs_bug_on(sbi, !page);
 282
 283                         SetPagePrivate(page);
 284                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 285                         lock_page(page);
 286                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 287                                 f2fs_bug_on(sbi, 1);
 288                 }
 289                 /*
 290                  * In the NODE case, we lose next block address chain. So, we
 291                  * need to do checkpoint in f2fs_sync_file.
 292                  */
 293                 if (type == NODE)
 294                         set_sbi_flag(sbi, SBI_NEED_CP);
 295         }
 296 submit_io:
 297         if (is_read_io(bio_op(bio)))
 298                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 299         else
 300                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 301         submit_bio(bio);
 302 }
 303
 304 static void __submit_merged_bio(struct f2fs_bio_info *io)
 305 {
 306         struct f2fs_io_info *fio = &io->fio;
 307
 308         if (!io->bio)
 309                 return;
 310
 311         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 312
 313         if (is_read_io(fio->op))
 314                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 315         else
 316                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 317
 318         __submit_bio(io->sbi, io->bio, fio->type);
 319         io->bio = NULL;
 320 }
 321
 322 static bool __has_merged_page(struct f2fs_bio_info *io,
 323                                 struct inode *inode, nid_t ino, pgoff_t idx)
 324 {
 325         struct bio_vec *bvec;
 326         struct page *target;
 327         int i;
 328
 329         if (!io->bio)
 330                 return false;
 331
 332         if (!inode && !ino)
 333                 return true;
 334
 335         bio_for_each_segment_all(bvec, io->bio, i) {
 336
 337                 if (bvec->bv_page->mapping)
 338                         target = bvec->bv_page;
 339                 else
 340                         target = fscrypt_control_page(bvec->bv_page);
 341
 342                 if (idx != target->index)
 343                         continue;
 344
 345                 if (inode && inode == target->mapping->host)
 346                         return true;
 347                 if (ino && ino == ino_of_node(target))
 348                         return true;
 349         }
 350
 351         return false;
 352 }
 353
 354 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 355                                 nid_t ino, pgoff_t idx, enum page_type type)
 356 {
 357         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 358         enum temp_type temp;
 359         struct f2fs_bio_info *io;
 360         bool ret = false;
 361
 362         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 363                 io = sbi->write_io[btype] + temp;
 364
 365                 down_read(&io->io_rwsem);
 366                 ret = __has_merged_page(io, inode, ino, idx);
 367                 up_read(&io->io_rwsem);
 368
 369                 /* TODO: use HOT temp only for meta pages now. */
 370                 if (ret || btype == META)
 371                         break;
 372         }
 373         return ret;
 374 }
 375
 376 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 377                                 enum page_type type, enum temp_type temp)
 378 {
 379         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 380         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 381
 382         down_write(&io->io_rwsem);
 383
 384         /* change META to META_FLUSH in the checkpoint procedure */
 385         if (type >= META_FLUSH) {
 386                 io->fio.type = META_FLUSH;
 387                 io->fio.op = REQ_OP_WRITE;
 388                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 389                 if (!test_opt(sbi, NOBARRIER))
 390                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 391         }
 392         __submit_merged_bio(io);
 393         up_write(&io->io_rwsem);
 394 }
 395
 396 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 397                                 struct inode *inode, nid_t ino, pgoff_t idx,
 398                                 enum page_type type, bool force)
 399 {
 400         enum temp_type temp;
 401
 402         if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 403                 return;
 404
 405         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 406
 407                 __f2fs_submit_merged_write(sbi, type, temp);
 408
 409                 /* TODO: use HOT temp only for meta pages now. */
 410                 if (type >= META)
 411                         break;
 412         }
 413 }
 414
 415 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 416 {
 417         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 418 }
 419
 420 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 421                                 struct inode *inode, nid_t ino, pgoff_t idx,
 422                                 enum page_type type)
 423 {
 424         __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 425 }
 426
 427 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 428 {
 429         f2fs_submit_merged_write(sbi, DATA);
 430         f2fs_submit_merged_write(sbi, NODE);
 431         f2fs_submit_merged_write(sbi, META);
 432 }
 433
 434 /*
 435  * Fill the locked page with data located in the block address.
 436  * A caller needs to unlock the page on failure.
 437  */
 438 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 439 {
 440         struct bio *bio;
 441         struct page *page = fio->encrypted_page ?
 442                         fio->encrypted_page : fio->page;
 443
 444         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 445                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 446                 return -EFAULT;
 447
 448         trace_f2fs_submit_page_bio(page, fio);
 449         f2fs_trace_ios(fio, 0);
 450
 451         /* Allocate a new bio */
 452         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 453                                 1, is_read_io(fio->op), fio->type, fio->temp);
 454
 455         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 456                 bio_put(bio);
 457                 return -EFAULT;
 458         }
 459         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 460
 461         __submit_bio(fio->sbi, bio, fio->type);
 462
 463         if (!is_read_io(fio->op))
 464                 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 465         return 0;
 466 }
 467
 468 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 469 {
 470         struct f2fs_sb_info *sbi = fio->sbi;
 471         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 472         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 473         struct page *bio_page;
 474
 475         f2fs_bug_on(sbi, is_read_io(fio->op));
 476
 477         down_write(&io->io_rwsem);
 478 next:
 479         if (fio->in_list) {
 480                 spin_lock(&io->io_lock);
 481                 if (list_empty(&io->io_list)) {
 482                         spin_unlock(&io->io_lock);
 483                         goto out;
 484                 }
 485                 fio = list_first_entry(&io->io_list,
 486                                                 struct f2fs_io_info, list);
 487                 list_del(&fio->list);
 488                 spin_unlock(&io->io_lock);
 489         }
 490
 491         if (__is_valid_data_blkaddr(fio->old_blkaddr))
 492                 verify_block_addr(fio, fio->old_blkaddr);
 493         verify_block_addr(fio, fio->new_blkaddr);
 494
 495         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 496
 497         /* set submitted = true as a return value */
 498         fio->submitted = true;
 499
 500         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 501
 502         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 503             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 504                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 505                 __submit_merged_bio(io);
 506 alloc_new:
 507         if (io->bio == NULL) {
 508                 if ((fio->type == DATA || fio->type == NODE) &&
 509                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 510                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 511                         fio->retry = true;
 512                         goto skip;
 513                 }
 514                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 515                                                 BIO_MAX_PAGES, false,
 516                                                 fio->type, fio->temp);
 517                 io->fio = *fio;
 518         }
 519
 520         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 521                 __submit_merged_bio(io);
 522                 goto alloc_new;
 523         }
 524
 525         if (fio->io_wbc)
 526                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 527
 528         io->last_block_in_bio = fio->new_blkaddr;
 529         f2fs_trace_ios(fio, 0);
 530
 531         trace_f2fs_submit_page_write(fio->page, fio);
 532 skip:
 533         if (fio->in_list)
 534                 goto next;
 535 out:
 536         up_write(&io->io_rwsem);
 537 }
 538
 539 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 540                                         unsigned nr_pages, unsigned op_flag)
 541 {
 542         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 543         struct bio *bio;
 544         struct bio_post_read_ctx *ctx;
 545         unsigned int post_read_steps = 0;
 546
 547         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 548         if (!bio)
 549                 return ERR_PTR(-ENOMEM);
 550         f2fs_target_device(sbi, blkaddr, bio);
 551         bio->bi_end_io = f2fs_read_end_io;
 552         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 553
 554         if (f2fs_encrypted_file(inode))
 555                 post_read_steps |= 1 << STEP_DECRYPT;
 556         if (post_read_steps) {
 557                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 558                 if (!ctx) {
 559                         bio_put(bio);
 560                         return ERR_PTR(-ENOMEM);
 561                 }
 562                 ctx->bio = bio;
 563                 ctx->enabled_steps = post_read_steps;
 564                 bio->bi_private = ctx;
 565
 566                 /* wait the page to be moved by cleaning */
 567                 f2fs_wait_on_block_writeback(sbi, blkaddr);
 568         }
 569
 570         return bio;
 571 }
 572
 573 /* This can handle encryption stuffs */
 574 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 575                                                         block_t blkaddr)
 576 {
 577         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 578
 579         if (IS_ERR(bio))
 580                 return PTR_ERR(bio);
 581
 582         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 583                 bio_put(bio);
 584                 return -EFAULT;
 585         }
 586         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 587         return 0;
 588 }
 589
 590 static void __set_data_blkaddr(struct dnode_of_data *dn)
 591 {
 592         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 593         __le32 *addr_array;
 594         int base = 0;
 595
 596         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 597                 base = get_extra_isize(dn->inode);
 598
 599         /* Get physical address of data block */
 600         addr_array = blkaddr_in_node(rn);
 601         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 602 }
 603
 604 /*
 605  * Lock ordering for the change of data block address:
 606  * ->data_page
 607  *  ->node_page
 608  *    update block addresses in the node page
 609  */
 610 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 611 {
 612         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 613         __set_data_blkaddr(dn);
 614         if (set_page_dirty(dn->node_page))
 615                 dn->node_changed = true;
 616 }
 617
 618 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 619 {
 620         dn->data_blkaddr = blkaddr;
 621         f2fs_set_data_blkaddr(dn);
 622         f2fs_update_extent_cache(dn);
 623 }
 624
 625 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 626 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 627 {
 628         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 629         int err;
 630
 631         if (!count)
 632                 return 0;
 633
 634         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 635                 return -EPERM;
 636         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 637                 return err;
 638
 639         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 640                                                 dn->ofs_in_node, count);
 641
 642         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 643
 644         for (; count > 0; dn->ofs_in_node++) {
 645                 block_t blkaddr = datablock_addr(dn->inode,
 646                                         dn->node_page, dn->ofs_in_node);
 647                 if (blkaddr == NULL_ADDR) {
 648                         dn->data_blkaddr = NEW_ADDR;
 649                         __set_data_blkaddr(dn);
 650                         count--;
 651                 }
 652         }
 653
 654         if (set_page_dirty(dn->node_page))
 655                 dn->node_changed = true;
 656         return 0;
 657 }
 658
 659 /* Should keep dn->ofs_in_node unchanged */
 660 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 661 {
 662         unsigned int ofs_in_node = dn->ofs_in_node;
 663         int ret;
 664
 665         ret = f2fs_reserve_new_blocks(dn, 1);
 666         dn->ofs_in_node = ofs_in_node;
 667         return ret;
 668 }
 669
 670 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 671 {
 672         bool need_put = dn->inode_page ? false : true;
 673         int err;
 674
 675         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 676         if (err)
 677                 return err;
 678
 679         if (dn->data_blkaddr == NULL_ADDR)
 680                 err = f2fs_reserve_new_block(dn);
 681         if (err || need_put)
 682                 f2fs_put_dnode(dn);
 683         return err;
 684 }
 685
 686 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 687 {
 688         struct extent_info ei  = {0,0,0};
 689         struct inode *inode = dn->inode;
 690
 691         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 692                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 693                 return 0;
 694         }
 695
 696         return f2fs_reserve_block(dn, index);
 697 }
 698
 699 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 700                                                 int op_flags, bool for_write)
 701 {
 702         struct address_space *mapping = inode->i_mapping;
 703         struct dnode_of_data dn;
 704         struct page *page;
 705         struct extent_info ei = {0,0,0};
 706         int err;
 707
 708         page = f2fs_grab_cache_page(mapping, index, for_write);
 709         if (!page)
 710                 return ERR_PTR(-ENOMEM);
 711
 712         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 713                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 714                 goto got_it;
 715         }
 716
 717         set_new_dnode(&dn, inode, NULL, NULL, 0);
 718         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 719         if (err)
 720                 goto put_err;
 721         f2fs_put_dnode(&dn);
 722
 723         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 724                 err = -ENOENT;
 725                 goto put_err;
 726         }
 727 got_it:
 728         if (PageUptodate(page)) {
 729                 unlock_page(page);
 730                 return page;
 731         }
 732
 733         /*
 734          * A new dentry page is allocated but not able to be written, since its
 735          * new inode page couldn't be allocated due to -ENOSPC.
 736          * In such the case, its blkaddr can be remained as NEW_ADDR.
 737          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 738          * f2fs_init_inode_metadata.
 739          */
 740         if (dn.data_blkaddr == NEW_ADDR) {
 741                 zero_user_segment(page, 0, PAGE_SIZE);
 742                 if (!PageUptodate(page))
 743                         SetPageUptodate(page);
 744                 unlock_page(page);
 745                 return page;
 746         }
 747
 748         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 749         if (err)
 750                 goto put_err;
 751         return page;
 752
 753 put_err:
 754         f2fs_put_page(page, 1);
 755         return ERR_PTR(err);
 756 }
 757
 758 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 759 {
 760         struct address_space *mapping = inode->i_mapping;
 761         struct page *page;
 762
 763         page = find_get_page(mapping, index);
 764         if (page && PageUptodate(page))
 765                 return page;
 766         f2fs_put_page(page, 0);
 767
 768         page = f2fs_get_read_data_page(inode, index, 0, false);
 769         if (IS_ERR(page))
 770                 return page;
 771
 772         if (PageUptodate(page))
 773                 return page;
 774
 775         wait_on_page_locked(page);
 776         if (unlikely(!PageUptodate(page))) {
 777                 f2fs_put_page(page, 0);
 778                 return ERR_PTR(-EIO);
 779         }
 780         return page;
 781 }
 782
 783 /*
 784  * If it tries to access a hole, return an error.
 785  * Because, the callers, functions in dir.c and GC, should be able to know
 786  * whether this page exists or not.
 787  */
 788 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 789                                                         bool for_write)
 790 {
 791         struct address_space *mapping = inode->i_mapping;
 792         struct page *page;
 793 repeat:
 794         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 795         if (IS_ERR(page))
 796                 return page;
 797
 798         /* wait for read completion */
 799         lock_page(page);
 800         if (unlikely(page->mapping != mapping)) {
 801                 f2fs_put_page(page, 1);
 802                 goto repeat;
 803         }
 804         if (unlikely(!PageUptodate(page))) {
 805                 f2fs_put_page(page, 1);
 806                 return ERR_PTR(-EIO);
 807         }
 808         return page;
 809 }
 810
 811 /*
 812  * Caller ensures that this data page is never allocated.
 813  * A new zero-filled data page is allocated in the page cache.
 814  *
 815  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 816  * f2fs_unlock_op().
 817  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 818  * ipage should be released by this function.
 819  */
 820 struct page *f2fs_get_new_data_page(struct inode *inode,
 821                 struct page *ipage, pgoff_t index, bool new_i_size)
 822 {
 823         struct address_space *mapping = inode->i_mapping;
 824         struct page *page;
 825         struct dnode_of_data dn;
 826         int err;
 827
 828         page = f2fs_grab_cache_page(mapping, index, true);
 829         if (!page) {
 830                 /*
 831                  * before exiting, we should make sure ipage will be released
 832                  * if any error occur.
 833                  */
 834                 f2fs_put_page(ipage, 1);
 835                 return ERR_PTR(-ENOMEM);
 836         }
 837
 838         set_new_dnode(&dn, inode, ipage, NULL, 0);
 839         err = f2fs_reserve_block(&dn, index);
 840         if (err) {
 841                 f2fs_put_page(page, 1);
 842                 return ERR_PTR(err);
 843         }
 844         if (!ipage)
 845                 f2fs_put_dnode(&dn);
 846
 847         if (PageUptodate(page))
 848                 goto got_it;
 849
 850         if (dn.data_blkaddr == NEW_ADDR) {
 851                 zero_user_segment(page, 0, PAGE_SIZE);
 852                 if (!PageUptodate(page))
 853                         SetPageUptodate(page);
 854         } else {
 855                 f2fs_put_page(page, 1);
 856
 857                 /* if ipage exists, blkaddr should be NEW_ADDR */
 858                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 859                 page = f2fs_get_lock_data_page(inode, index, true);
 860                 if (IS_ERR(page))
 861                         return page;
 862         }
 863 got_it:
 864         if (new_i_size && i_size_read(inode) <
 865                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 866                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 867         return page;
 868 }
 869
 870 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 871 {
 872         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 873         struct f2fs_summary sum;
 874         struct node_info ni;
 875         pgoff_t fofs;
 876         blkcnt_t count = 1;
 877         int err;
 878
 879         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 880                 return -EPERM;
 881
 882         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 883         if (err)
 884                 return err;
 885
 886         dn->data_blkaddr = datablock_addr(dn->inode,
 887                                 dn->node_page, dn->ofs_in_node);
 888         if (dn->data_blkaddr == NEW_ADDR)
 889                 goto alloc;
 890
 891         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 892                 return err;
 893
 894 alloc:
 895         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 896
 897         f2fs_allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 898                                         &sum, seg_type, NULL, false);
 899         f2fs_set_data_blkaddr(dn);
 900
 901         /* update i_size */
 902         fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 903                                                         dn->ofs_in_node;
 904         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 905                 f2fs_i_size_write(dn->inode,
 906                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 907         return 0;
 908 }
 909
 910 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 911 {
 912         struct inode *inode = file_inode(iocb->ki_filp);
 913         struct f2fs_map_blocks map;
 914         int flag;
 915         int err = 0;
 916         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 917
 918         /* convert inline data for Direct I/O*/
 919         if (direct_io) {
 920                 err = f2fs_convert_inline_inode(inode);
 921                 if (err)
 922                         return err;
 923         }
 924
 925         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 926                 return 0;
 927
 928         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 929         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 930         if (map.m_len > map.m_lblk)
 931                 map.m_len -= map.m_lblk;
 932         else
 933                 map.m_len = 0;
 934
 935         map.m_next_pgofs = NULL;
 936         map.m_next_extent = NULL;
 937         map.m_seg_type = NO_CHECK_TYPE;
 938
 939         if (direct_io) {
 940                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 941                 flag = f2fs_force_buffered_io(inode, WRITE) ?
 942                                         F2FS_GET_BLOCK_PRE_AIO :
 943                                         F2FS_GET_BLOCK_PRE_DIO;
 944                 goto map_blocks;
 945         }
 946         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 947                 err = f2fs_convert_inline_inode(inode);
 948                 if (err)
 949                         return err;
 950         }
 951         if (f2fs_has_inline_data(inode))
 952                 return err;
 953
 954         flag = F2FS_GET_BLOCK_PRE_AIO;
 955
 956 map_blocks:
 957         err = f2fs_map_blocks(inode, &map, 1, flag);
 958         if (map.m_len > 0 && err == -ENOSPC) {
 959                 if (!direct_io)
 960                         set_inode_flag(inode, FI_NO_PREALLOC);
 961                 err = 0;
 962         }
 963         return err;
 964 }
 965
 966 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 967 {
 968         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 969                 if (lock)
 970                         down_read(&sbi->node_change);
 971                 else
 972                         up_read(&sbi->node_change);
 973         } else {
 974                 if (lock)
 975                         f2fs_lock_op(sbi);
 976                 else
 977                         f2fs_unlock_op(sbi);
 978         }
 979 }
 980
 981 /*
 982  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 983  * f2fs_map_blocks structure.
 984  * If original data blocks are allocated, then give them to blockdev.
 985  * Otherwise,
 986  *     a. preallocate requested block addresses
 987  *     b. do not use extent cache for better performance
 988  *     c. give the block addresses to blockdev
 989  */
 990 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 991                                                 int create, int flag)
 992 {
 993         unsigned int maxblocks = map->m_len;
 994         struct dnode_of_data dn;
 995         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 996         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 997         pgoff_t pgofs, end_offset, end;
 998         int err = 0, ofs = 1;
 999         unsigned int ofs_in_node, last_ofs_in_node;
1000         blkcnt_t prealloc;
1001         struct extent_info ei = {0,0,0};
1002         block_t blkaddr;
1003         unsigned int start_pgofs;
1004
1005         if (!maxblocks)
1006                 return 0;
1007
1008         map->m_len = 0;
1009         map->m_flags = 0;
1010
1011         /* it only supports block size == page size */
1012         pgofs = (pgoff_t)map->m_lblk;
1013         end = pgofs + maxblocks;
1014
1015         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1016                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1017                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1018                 map->m_flags = F2FS_MAP_MAPPED;
1019                 if (map->m_next_extent)
1020                         *map->m_next_extent = pgofs + map->m_len;
1021                 goto out;
1022         }
1023
1024 next_dnode:
1025         if (create)
1026                 __do_map_lock(sbi, flag, true);
1027
1028         /* When reading holes, we need its node page */
1029         set_new_dnode(&dn, inode, NULL, NULL, 0);
1030         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1031         if (err) {
1032                 if (flag == F2FS_GET_BLOCK_BMAP)
1033                         map->m_pblk = 0;
1034                 if (err == -ENOENT) {
1035                         err = 0;
1036                         if (map->m_next_pgofs)
1037                                 *map->m_next_pgofs =
1038                                         f2fs_get_next_page_offset(&dn, pgofs);
1039                         if (map->m_next_extent)
1040                                 *map->m_next_extent =
1041                                         f2fs_get_next_page_offset(&dn, pgofs);
1042                 }
1043                 goto unlock_out;
1044         }
1045
1046         start_pgofs = pgofs;
1047         prealloc = 0;
1048         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1049         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1050
1051 next_block:
1052         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1053
1054         if (__is_valid_data_blkaddr(blkaddr) &&
1055                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1056                 err = -EFAULT;
1057                 goto sync_out;
1058         }
1059
1060         if (!is_valid_data_blkaddr(sbi, blkaddr)) {
1061                 if (create) {
1062                         if (unlikely(f2fs_cp_error(sbi))) {
1063                                 err = -EIO;
1064                                 goto sync_out;
1065                         }
1066                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1067                                 if (blkaddr == NULL_ADDR) {
1068                                         prealloc++;
1069                                         last_ofs_in_node = dn.ofs_in_node;
1070                                 }
1071                         } else {
1072                                 err = __allocate_data_block(&dn,
1073                                                         map->m_seg_type);
1074                                 if (!err)
1075                                         set_inode_flag(inode, FI_APPEND_WRITE);
1076                         }
1077                         if (err)
1078                                 goto sync_out;
1079                         map->m_flags |= F2FS_MAP_NEW;
1080                         blkaddr = dn.data_blkaddr;
1081                 } else {
1082                         if (flag == F2FS_GET_BLOCK_BMAP) {
1083                                 map->m_pblk = 0;
1084                                 goto sync_out;
1085                         }
1086                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1087                                 goto sync_out;
1088                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1089                                                 blkaddr == NULL_ADDR) {
1090                                 if (map->m_next_pgofs)
1091                                         *map->m_next_pgofs = pgofs + 1;
1092                                 goto sync_out;
1093                         }
1094                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1095                                 /* for defragment case */
1096                                 if (map->m_next_pgofs)
1097                                         *map->m_next_pgofs = pgofs + 1;
1098                                 goto sync_out;
1099                         }
1100                 }
1101         }
1102
1103         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1104                 goto skip;
1105
1106         if (map->m_len == 0) {
1107                 /* preallocated unwritten block should be mapped for fiemap. */
1108                 if (blkaddr == NEW_ADDR)
1109                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1110                 map->m_flags |= F2FS_MAP_MAPPED;
1111
1112                 map->m_pblk = blkaddr;
1113                 map->m_len = 1;
1114         } else if ((map->m_pblk != NEW_ADDR &&
1115                         blkaddr == (map->m_pblk + ofs)) ||
1116                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1117                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1118                 ofs++;
1119                 map->m_len++;
1120         } else {
1121                 goto sync_out;
1122         }
1123
1124 skip:
1125         dn.ofs_in_node++;
1126         pgofs++;
1127
1128         /* preallocate blocks in batch for one dnode page */
1129         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1130                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1131
1132                 dn.ofs_in_node = ofs_in_node;
1133                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1134                 if (err)
1135                         goto sync_out;
1136
1137                 map->m_len += dn.ofs_in_node - ofs_in_node;
1138                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1139                         err = -ENOSPC;
1140                         goto sync_out;
1141                 }
1142                 dn.ofs_in_node = end_offset;
1143         }
1144
1145         if (pgofs >= end)
1146                 goto sync_out;
1147         else if (dn.ofs_in_node < end_offset)
1148                 goto next_block;
1149
1150         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1151                 if (map->m_flags & F2FS_MAP_MAPPED) {
1152                         unsigned int ofs = start_pgofs - map->m_lblk;
1153
1154                         f2fs_update_extent_cache_range(&dn,
1155                                 start_pgofs, map->m_pblk + ofs,
1156                                 map->m_len - ofs);
1157                 }
1158         }
1159
1160         f2fs_put_dnode(&dn);
1161
1162         if (create) {
1163                 __do_map_lock(sbi, flag, false);
1164                 f2fs_balance_fs(sbi, dn.node_changed);
1165         }
1166         goto next_dnode;
1167
1168 sync_out:
1169         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1170                 if (map->m_flags & F2FS_MAP_MAPPED) {
1171                         unsigned int ofs = start_pgofs - map->m_lblk;
1172
1173                         f2fs_update_extent_cache_range(&dn,
1174                                 start_pgofs, map->m_pblk + ofs,
1175                                 map->m_len - ofs);
1176                 }
1177                 if (map->m_next_extent)
1178                         *map->m_next_extent = pgofs + 1;
1179         }
1180         f2fs_put_dnode(&dn);
1181 unlock_out:
1182         if (create) {
1183                 __do_map_lock(sbi, flag, false);
1184                 f2fs_balance_fs(sbi, dn.node_changed);
1185         }
1186 out:
1187         trace_f2fs_map_blocks(inode, map, err);
1188         return err;
1189 }
1190
1191 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1192 {
1193         struct f2fs_map_blocks map;
1194         block_t last_lblk;
1195         int err;
1196
1197         if (pos + len > i_size_read(inode))
1198                 return false;
1199
1200         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1201         map.m_next_pgofs = NULL;
1202         map.m_next_extent = NULL;
1203         map.m_seg_type = NO_CHECK_TYPE;
1204         last_lblk = F2FS_BLK_ALIGN(pos + len);
1205
1206         while (map.m_lblk < last_lblk) {
1207                 map.m_len = last_lblk - map.m_lblk;
1208                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1209                 if (err || map.m_len == 0)
1210                         return false;
1211                 map.m_lblk += map.m_len;
1212         }
1213         return true;
1214 }
1215
1216 static int __get_data_block(struct inode *inode, sector_t iblock,
1217                         struct buffer_head *bh, int create, int flag,
1218                         pgoff_t *next_pgofs, int seg_type)
1219 {
1220         struct f2fs_map_blocks map;
1221         int err;
1222
1223         map.m_lblk = iblock;
1224         map.m_len = bh->b_size >> inode->i_blkbits;
1225         map.m_next_pgofs = next_pgofs;
1226         map.m_next_extent = NULL;
1227         map.m_seg_type = seg_type;
1228
1229         err = f2fs_map_blocks(inode, &map, create, flag);
1230         if (!err) {
1231                 map_bh(bh, inode->i_sb, map.m_pblk);
1232                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1233                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1234         }
1235         return err;
1236 }
1237
1238 static int get_data_block(struct inode *inode, sector_t iblock,
1239                         struct buffer_head *bh_result, int create, int flag,
1240                         pgoff_t *next_pgofs)
1241 {
1242         return __get_data_block(inode, iblock, bh_result, create,
1243                                                         flag, next_pgofs,
1244                                                         NO_CHECK_TYPE);
1245 }
1246
1247 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1248                         struct buffer_head *bh_result, int create)
1249 {
1250         return __get_data_block(inode, iblock, bh_result, create,
1251                                                 F2FS_GET_BLOCK_DEFAULT, NULL,
1252                                                 f2fs_rw_hint_to_seg_type(
1253                                                         inode->i_write_hint));
1254 }
1255
1256 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1257                         struct buffer_head *bh_result, int create)
1258 {
1259         /* Block number less than F2FS MAX BLOCKS */
1260         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1261                 return -EFBIG;
1262
1263         return __get_data_block(inode, iblock, bh_result, create,
1264                                                 F2FS_GET_BLOCK_BMAP, NULL,
1265                                                 NO_CHECK_TYPE);
1266 }
1267
1268 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1269 {
1270         return (offset >> inode->i_blkbits);
1271 }
1272
1273 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1274 {
1275         return (blk << inode->i_blkbits);
1276 }
1277
1278 static int f2fs_xattr_fiemap(struct inode *inode,
1279                                 struct fiemap_extent_info *fieinfo)
1280 {
1281         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1282         struct page *page;
1283         struct node_info ni;
1284         __u64 phys = 0, len;
1285         __u32 flags;
1286         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1287         int err = 0;
1288
1289         if (f2fs_has_inline_xattr(inode)) {
1290                 int offset;
1291
1292                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1293                                                 inode->i_ino, false);
1294                 if (!page)
1295                         return -ENOMEM;
1296
1297                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1298                 if (err) {
1299                         f2fs_put_page(page, 1);
1300                         return err;
1301                 }
1302
1303                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1304                 offset = offsetof(struct f2fs_inode, i_addr) +
1305                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1306                                         get_inline_xattr_addrs(inode));
1307
1308                 phys += offset;
1309                 len = inline_xattr_size(inode);
1310
1311                 f2fs_put_page(page, 1);
1312
1313                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1314
1315                 if (!xnid)
1316                         flags |= FIEMAP_EXTENT_LAST;
1317
1318                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1319                 if (err || err == 1)
1320                         return err;
1321         }
1322
1323         if (xnid) {
1324                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1325                 if (!page)
1326                         return -ENOMEM;
1327
1328                 err = f2fs_get_node_info(sbi, xnid, &ni);
1329                 if (err) {
1330                         f2fs_put_page(page, 1);
1331                         return err;
1332                 }
1333
1334                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1335                 len = inode->i_sb->s_blocksize;
1336
1337                 f2fs_put_page(page, 1);
1338
1339                 flags = FIEMAP_EXTENT_LAST;
1340         }
1341
1342         if (phys)
1343                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1344
1345         return (err < 0 ? err : 0);
1346 }
1347
1348 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1349                 u64 start, u64 len)
1350 {
1351         struct buffer_head map_bh;
1352         sector_t start_blk, last_blk;
1353         pgoff_t next_pgofs;
1354         u64 logical = 0, phys = 0, size = 0;
1355         u32 flags = 0;
1356         int ret = 0;
1357
1358         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1359                 ret = f2fs_precache_extents(inode);
1360                 if (ret)
1361                         return ret;
1362         }
1363
1364         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1365         if (ret)
1366                 return ret;
1367
1368         inode_lock(inode);
1369
1370         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1371                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1372                 goto out;
1373         }
1374
1375         if (f2fs_has_inline_data(inode)) {
1376                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1377                 if (ret != -EAGAIN)
1378                         goto out;
1379         }
1380
1381         if (logical_to_blk(inode, len) == 0)
1382                 len = blk_to_logical(inode, 1);
1383
1384         start_blk = logical_to_blk(inode, start);
1385         last_blk = logical_to_blk(inode, start + len - 1);
1386
1387 next:
1388         memset(&map_bh, 0, sizeof(struct buffer_head));
1389         map_bh.b_size = len;
1390
1391         ret = get_data_block(inode, start_blk, &map_bh, 0,
1392                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1393         if (ret)
1394                 goto out;
1395
1396         /* HOLE */
1397         if (!buffer_mapped(&map_bh)) {
1398                 start_blk = next_pgofs;
1399
1400                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1401                                         F2FS_I_SB(inode)->max_file_blocks))
1402                         goto prep_next;
1403
1404                 flags |= FIEMAP_EXTENT_LAST;
1405         }
1406
1407         if (size) {
1408                 if (f2fs_encrypted_inode(inode))
1409                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1410
1411                 ret = fiemap_fill_next_extent(fieinfo, logical,
1412                                 phys, size, flags);
1413         }
1414
1415         if (start_blk > last_blk || ret)
1416                 goto out;
1417
1418         logical = blk_to_logical(inode, start_blk);
1419         phys = blk_to_logical(inode, map_bh.b_blocknr);
1420         size = map_bh.b_size;
1421         flags = 0;
1422         if (buffer_unwritten(&map_bh))
1423                 flags = FIEMAP_EXTENT_UNWRITTEN;
1424
1425         start_blk += logical_to_blk(inode, size);
1426
1427 prep_next:
1428         cond_resched();
1429         if (fatal_signal_pending(current))
1430                 ret = -EINTR;
1431         else
1432                 goto next;
1433 out:
1434         if (ret == 1)
1435                 ret = 0;
1436
1437         inode_unlock(inode);
1438         return ret;
1439 }
1440
1441 /*
1442  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1443  * Major change was from block_size == page_size in f2fs by default.
1444  *
1445  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1446  * this function ever deviates from doing just read-ahead, it should either
1447  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1448  * from read-ahead.
1449  */
1450 static int f2fs_mpage_readpages(struct address_space *mapping,
1451                         struct list_head *pages, struct page *page,
1452                         unsigned nr_pages, bool is_readahead)
1453 {
1454         struct bio *bio = NULL;
1455         sector_t last_block_in_bio = 0;
1456         struct inode *inode = mapping->host;
1457         const unsigned blkbits = inode->i_blkbits;
1458         const unsigned blocksize = 1 << blkbits;
1459         sector_t block_in_file;
1460         sector_t last_block;
1461         sector_t last_block_in_file;
1462         sector_t block_nr;
1463         struct f2fs_map_blocks map;
1464
1465         map.m_pblk = 0;
1466         map.m_lblk = 0;
1467         map.m_len = 0;
1468         map.m_flags = 0;
1469         map.m_next_pgofs = NULL;
1470         map.m_next_extent = NULL;
1471         map.m_seg_type = NO_CHECK_TYPE;
1472
1473         for (; nr_pages; nr_pages--) {
1474                 if (pages) {
1475                         page = list_last_entry(pages, struct page, lru);
1476
1477                         prefetchw(&page->flags);
1478                         list_del(&page->lru);
1479                         if (add_to_page_cache_lru(page, mapping,
1480                                                   page->index,
1481                                                   readahead_gfp_mask(mapping)))
1482                                 goto next_page;
1483                 }
1484
1485                 block_in_file = (sector_t)page->index;
1486                 last_block = block_in_file + nr_pages;
1487                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1488                                                                 blkbits;
1489                 if (last_block > last_block_in_file)
1490                         last_block = last_block_in_file;
1491
1492                 /*
1493                  * Map blocks using the previous result first.
1494                  */
1495                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1496                                 block_in_file > map.m_lblk &&
1497                                 block_in_file < (map.m_lblk + map.m_len))
1498                         goto got_it;
1499
1500                 /*
1501                  * Then do more f2fs_map_blocks() calls until we are
1502                  * done with this page.
1503                  */
1504                 map.m_flags = 0;
1505
1506                 if (block_in_file < last_block) {
1507                         map.m_lblk = block_in_file;
1508                         map.m_len = last_block - block_in_file;
1509
1510                         if (f2fs_map_blocks(inode, &map, 0,
1511                                                 F2FS_GET_BLOCK_DEFAULT))
1512                                 goto set_error_page;
1513                 }
1514 got_it:
1515                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1516                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1517                         SetPageMappedToDisk(page);
1518
1519                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1520                                 SetPageUptodate(page);
1521                                 goto confused;
1522                         }
1523
1524                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1525                                                                 DATA_GENERIC))
1526                                 goto set_error_page;
1527                 } else {
1528                         zero_user_segment(page, 0, PAGE_SIZE);
1529                         if (!PageUptodate(page))
1530                                 SetPageUptodate(page);
1531                         unlock_page(page);
1532                         goto next_page;
1533                 }
1534
1535                 /*
1536                  * This page will go to BIO.  Do we need to send this
1537                  * BIO off first?
1538                  */
1539                 if (bio && (last_block_in_bio != block_nr - 1 ||
1540                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1541 submit_and_realloc:
1542                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1543                         bio = NULL;
1544                 }
1545                 if (bio == NULL) {
1546                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1547                                         is_readahead ? REQ_RAHEAD : 0);
1548                         if (IS_ERR(bio)) {
1549                                 bio = NULL;
1550                                 goto set_error_page;
1551                         }
1552                 }
1553
1554                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1555                         goto submit_and_realloc;
1556
1557                 last_block_in_bio = block_nr;
1558                 goto next_page;
1559 set_error_page:
1560                 SetPageError(page);
1561                 zero_user_segment(page, 0, PAGE_SIZE);
1562                 unlock_page(page);
1563                 goto next_page;
1564 confused:
1565                 if (bio) {
1566                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1567                         bio = NULL;
1568                 }
1569                 unlock_page(page);
1570 next_page:
1571                 if (pages)
1572                         put_page(page);
1573         }
1574         BUG_ON(pages && !list_empty(pages));
1575         if (bio)
1576                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1577         return 0;
1578 }
1579
1580 static int f2fs_read_data_page(struct file *file, struct page *page)
1581 {
1582         struct inode *inode = page->mapping->host;
1583         int ret = -EAGAIN;
1584
1585         trace_f2fs_readpage(page, DATA);
1586
1587         /* If the file has inline data, try to read it directly */
1588         if (f2fs_has_inline_data(inode))
1589                 ret = f2fs_read_inline_data(inode, page);
1590         if (ret == -EAGAIN)
1591                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1592         return ret;
1593 }
1594
1595 static int f2fs_read_data_pages(struct file *file,
1596                         struct address_space *mapping,
1597                         struct list_head *pages, unsigned nr_pages)
1598 {
1599         struct inode *inode = mapping->host;
1600         struct page *page = list_last_entry(pages, struct page, lru);
1601
1602         trace_f2fs_readpages(inode, page, nr_pages);
1603
1604         /* If the file has inline data, skip readpages */
1605         if (f2fs_has_inline_data(inode))
1606                 return 0;
1607
1608         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1609 }
1610
1611 static int encrypt_one_page(struct f2fs_io_info *fio)
1612 {
1613         struct inode *inode = fio->page->mapping->host;
1614         gfp_t gfp_flags = GFP_NOFS;
1615
1616         if (!f2fs_encrypted_file(inode))
1617                 return 0;
1618
1619         /* wait for GCed page writeback via META_MAPPING */
1620         f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1621
1622 retry_encrypt:
1623         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1624                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1625         if (!IS_ERR(fio->encrypted_page))
1626                 return 0;
1627
1628         /* flush pending IOs and wait for a while in the ENOMEM case */
1629         if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1630                 f2fs_flush_merged_writes(fio->sbi);
1631                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1632                 gfp_flags |= __GFP_NOFAIL;
1633                 goto retry_encrypt;
1634         }
1635         return PTR_ERR(fio->encrypted_page);
1636 }
1637
1638 static inline bool check_inplace_update_policy(struct inode *inode,
1639                                 struct f2fs_io_info *fio)
1640 {
1641         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1642         unsigned int policy = SM_I(sbi)->ipu_policy;
1643
1644         if (policy & (0x1 << F2FS_IPU_FORCE))
1645                 return true;
1646         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1647                 return true;
1648         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1649                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1650                 return true;
1651         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1652                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1653                 return true;
1654
1655         /*
1656          * IPU for rewrite async pages
1657          */
1658         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1659                         fio && fio->op == REQ_OP_WRITE &&
1660                         !(fio->op_flags & REQ_SYNC) &&
1661                         !f2fs_encrypted_inode(inode))
1662                 return true;
1663
1664         /* this is only set during fdatasync */
1665         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1666                         is_inode_flag_set(inode, FI_NEED_IPU))
1667                 return true;
1668
1669         return false;
1670 }
1671
1672 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1673 {
1674         if (f2fs_is_pinned_file(inode))
1675                 return true;
1676
1677         /* if this is cold file, we should overwrite to avoid fragmentation */
1678         if (file_is_cold(inode))
1679                 return true;
1680
1681         return check_inplace_update_policy(inode, fio);
1682 }
1683
1684 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1685 {
1686         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1687
1688         if (test_opt(sbi, LFS))
1689                 return true;
1690         if (S_ISDIR(inode->i_mode))
1691                 return true;
1692         if (f2fs_is_atomic_file(inode))
1693                 return true;
1694         if (fio) {
1695                 if (is_cold_data(fio->page))
1696                         return true;
1697                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1698                         return true;
1699         }
1700         return false;
1701 }
1702
1703 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1704 {
1705         struct inode *inode = fio->page->mapping->host;
1706
1707         if (f2fs_should_update_outplace(inode, fio))
1708                 return false;
1709
1710         return f2fs_should_update_inplace(inode, fio);
1711 }
1712
1713 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1714 {
1715         struct page *page = fio->page;
1716         struct inode *inode = page->mapping->host;
1717         struct dnode_of_data dn;
1718         struct extent_info ei = {0,0,0};
1719         struct node_info ni;
1720         bool ipu_force = false;
1721         int err = 0;
1722
1723         set_new_dnode(&dn, inode, NULL, NULL, 0);
1724         if (need_inplace_update(fio) &&
1725                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1726                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1727
1728                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1729                                                         DATA_GENERIC))
1730                         return -EFAULT;
1731
1732                 ipu_force = true;
1733                 fio->need_lock = LOCK_DONE;
1734                 goto got_it;
1735         }
1736
1737         /* Deadlock due to between page->lock and f2fs_lock_op */
1738         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1739                 return -EAGAIN;
1740
1741         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1742         if (err)
1743                 goto out;
1744
1745         fio->old_blkaddr = dn.data_blkaddr;
1746
1747         /* This page is already truncated */
1748         if (fio->old_blkaddr == NULL_ADDR) {
1749                 ClearPageUptodate(page);
1750                 goto out_writepage;
1751         }
1752 got_it:
1753         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1754                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1755                                                         DATA_GENERIC)) {
1756                 err = -EFAULT;
1757                 goto out_writepage;
1758         }
1759         /*
1760          * If current allocation needs SSR,
1761          * it had better in-place writes for updated data.
1762          */
1763         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1764                                         need_inplace_update(fio))) {
1765                 err = encrypt_one_page(fio);
1766                 if (err)
1767                         goto out_writepage;
1768
1769                 set_page_writeback(page);
1770                 ClearPageError(page);
1771                 f2fs_put_dnode(&dn);
1772                 if (fio->need_lock == LOCK_REQ)
1773                         f2fs_unlock_op(fio->sbi);
1774                 err = f2fs_inplace_write_data(fio);
1775                 trace_f2fs_do_write_data_page(fio->page, IPU);
1776                 set_inode_flag(inode, FI_UPDATE_WRITE);
1777                 return err;
1778         }
1779
1780         if (fio->need_lock == LOCK_RETRY) {
1781                 if (!f2fs_trylock_op(fio->sbi)) {
1782                         err = -EAGAIN;
1783                         goto out_writepage;
1784                 }
1785                 fio->need_lock = LOCK_REQ;
1786         }
1787
1788         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1789         if (err)
1790                 goto out_writepage;
1791
1792         fio->version = ni.version;
1793
1794         err = encrypt_one_page(fio);
1795         if (err)
1796                 goto out_writepage;
1797
1798         set_page_writeback(page);
1799         ClearPageError(page);
1800
1801         /* LFS mode write path */
1802         f2fs_outplace_write_data(&dn, fio);
1803         trace_f2fs_do_write_data_page(page, OPU);
1804         set_inode_flag(inode, FI_APPEND_WRITE);
1805         if (page->index == 0)
1806                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1807 out_writepage:
1808         f2fs_put_dnode(&dn);
1809 out:
1810         if (fio->need_lock == LOCK_REQ)
1811                 f2fs_unlock_op(fio->sbi);
1812         return err;
1813 }
1814
1815 static int __write_data_page(struct page *page, bool *submitted,
1816                                 struct writeback_control *wbc,
1817                                 enum iostat_type io_type)
1818 {
1819         struct inode *inode = page->mapping->host;
1820         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1821         loff_t i_size = i_size_read(inode);
1822         const pgoff_t end_index = ((unsigned long long) i_size)
1823                                                         >> PAGE_SHIFT;
1824         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1825         unsigned offset = 0;
1826         bool need_balance_fs = false;
1827         int err = 0;
1828         struct f2fs_io_info fio = {
1829                 .sbi = sbi,
1830                 .ino = inode->i_ino,
1831                 .type = DATA,
1832                 .op = REQ_OP_WRITE,
1833                 .op_flags = wbc_to_write_flags(wbc),
1834                 .old_blkaddr = NULL_ADDR,
1835                 .page = page,
1836                 .encrypted_page = NULL,
1837                 .submitted = false,
1838                 .need_lock = LOCK_RETRY,
1839                 .io_type = io_type,
1840                 .io_wbc = wbc,
1841         };
1842
1843         trace_f2fs_writepage(page, DATA);
1844
1845         /* we should bypass data pages to proceed the kworkder jobs */
1846         if (unlikely(f2fs_cp_error(sbi))) {
1847                 mapping_set_error(page->mapping, -EIO);
1848                 /*
1849                  * don't drop any dirty dentry pages for keeping lastest
1850                  * directory structure.
1851                  */
1852                 if (S_ISDIR(inode->i_mode))
1853                         goto redirty_out;
1854                 goto out;
1855         }
1856
1857         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1858                 goto redirty_out;
1859
1860         if (page->index < end_index)
1861                 goto write;
1862
1863         /*
1864          * If the offset is out-of-range of file size,
1865          * this page does not have to be written to disk.
1866          */
1867         offset = i_size & (PAGE_SIZE - 1);
1868         if ((page->index >= end_index + 1) || !offset)
1869                 goto out;
1870
1871         zero_user_segment(page, offset, PAGE_SIZE);
1872 write:
1873         if (f2fs_is_drop_cache(inode))
1874                 goto out;
1875         /* we should not write 0'th page having journal header */
1876         if (f2fs_is_volatile_file(inode) && (!page->index ||
1877                         (!wbc->for_reclaim &&
1878                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1879                 goto redirty_out;
1880
1881         /* Dentry blocks are controlled by checkpoint */
1882         if (S_ISDIR(inode->i_mode)) {
1883                 fio.need_lock = LOCK_DONE;
1884                 err = f2fs_do_write_data_page(&fio);
1885                 goto done;
1886         }
1887
1888         if (!wbc->for_reclaim)
1889                 need_balance_fs = true;
1890         else if (has_not_enough_free_secs(sbi, 0, 0))
1891                 goto redirty_out;
1892         else
1893                 set_inode_flag(inode, FI_HOT_DATA);
1894
1895         err = -EAGAIN;
1896         if (f2fs_has_inline_data(inode)) {
1897                 err = f2fs_write_inline_data(inode, page);
1898                 if (!err)
1899                         goto out;
1900         }
1901
1902         if (err == -EAGAIN) {
1903                 err = f2fs_do_write_data_page(&fio);
1904                 if (err == -EAGAIN) {
1905                         fio.need_lock = LOCK_REQ;
1906                         err = f2fs_do_write_data_page(&fio);
1907                 }
1908         }
1909
1910         if (err) {
1911                 file_set_keep_isize(inode);
1912         } else {
1913                 down_write(&F2FS_I(inode)->i_sem);
1914                 if (F2FS_I(inode)->last_disk_size < psize)
1915                         F2FS_I(inode)->last_disk_size = psize;
1916                 up_write(&F2FS_I(inode)->i_sem);
1917         }
1918
1919 done:
1920         if (err && err != -ENOENT)
1921                 goto redirty_out;
1922
1923 out:
1924         inode_dec_dirty_pages(inode);
1925         if (err)
1926                 ClearPageUptodate(page);
1927
1928         if (wbc->for_reclaim) {
1929                 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1930                 clear_inode_flag(inode, FI_HOT_DATA);
1931                 f2fs_remove_dirty_inode(inode);
1932                 submitted = NULL;
1933         }
1934
1935         unlock_page(page);
1936         if (!S_ISDIR(inode->i_mode))
1937                 f2fs_balance_fs(sbi, need_balance_fs);
1938
1939         if (unlikely(f2fs_cp_error(sbi))) {
1940                 f2fs_submit_merged_write(sbi, DATA);
1941                 submitted = NULL;
1942         }
1943
1944         if (submitted)
1945                 *submitted = fio.submitted;
1946
1947         return 0;
1948
1949 redirty_out:
1950         redirty_page_for_writepage(wbc, page);
1951         /*
1952          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
1953          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
1954          * file_write_and_wait_range() will see EIO error, which is critical
1955          * to return value of fsync() followed by atomic_write failure to user.
1956          */
1957         if (!err || wbc->for_reclaim)
1958                 return AOP_WRITEPAGE_ACTIVATE;
1959         unlock_page(page);
1960         return err;
1961 }
1962
1963 static int f2fs_write_data_page(struct page *page,
1964                                         struct writeback_control *wbc)
1965 {
1966         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1967 }
1968
1969 /*
1970  * This function was copied from write_cche_pages from mm/page-writeback.c.
1971  * The major change is making write step of cold data page separately from
1972  * warm/hot data page.
1973  */
1974 static int f2fs_write_cache_pages(struct address_space *mapping,
1975                                         struct writeback_control *wbc,
1976                                         enum iostat_type io_type)
1977 {
1978         int ret = 0;
1979         int done = 0;
1980         struct pagevec pvec;
1981         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1982         int nr_pages;
1983         pgoff_t uninitialized_var(writeback_index);
1984         pgoff_t index;
1985         pgoff_t end;            /* Inclusive */
1986         pgoff_t done_index;
1987         pgoff_t last_idx = ULONG_MAX;
1988         int cycled;
1989         int range_whole = 0;
1990         int tag;
1991
1992         pagevec_init(&pvec);
1993
1994         if (get_dirty_pages(mapping->host) <=
1995                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1996                 set_inode_flag(mapping->host, FI_HOT_DATA);
1997         else
1998                 clear_inode_flag(mapping->host, FI_HOT_DATA);
1999
2000         if (wbc->range_cyclic) {
2001                 writeback_index = mapping->writeback_index; /* prev offset */
2002                 index = writeback_index;
2003                 if (index == 0)
2004                         cycled = 1;
2005                 else
2006                         cycled = 0;
2007                 end = -1;
2008         } else {
2009                 index = wbc->range_start >> PAGE_SHIFT;
2010                 end = wbc->range_end >> PAGE_SHIFT;
2011                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2012                         range_whole = 1;
2013                 cycled = 1; /* ignore range_cyclic tests */
2014         }
2015         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2016                 tag = PAGECACHE_TAG_TOWRITE;
2017         else
2018                 tag = PAGECACHE_TAG_DIRTY;
2019 retry:
2020         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2021                 tag_pages_for_writeback(mapping, index, end);
2022         done_index = index;
2023         while (!done && (index <= end)) {
2024                 int i;
2025
2026                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2027                                 tag);
2028                 if (nr_pages == 0)
2029                         break;
2030
2031                 for (i = 0; i < nr_pages; i++) {
2032                         struct page *page = pvec.pages[i];
2033                         bool submitted = false;
2034
2035                         /* give a priority to WB_SYNC threads */
2036                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2037                                         wbc->sync_mode == WB_SYNC_NONE) {
2038                                 done = 1;
2039                                 break;
2040                         }
2041
2042                         done_index = page->index;
2043 retry_write:
2044                         lock_page(page);
2045
2046                         if (unlikely(page->mapping != mapping)) {
2047 continue_unlock:
2048                                 unlock_page(page);
2049                                 continue;
2050                         }
2051
2052                         if (!PageDirty(page)) {
2053                                 /* someone wrote it for us */
2054                                 goto continue_unlock;
2055                         }
2056
2057                         if (PageWriteback(page)) {
2058                                 if (wbc->sync_mode != WB_SYNC_NONE)
2059                                         f2fs_wait_on_page_writeback(page,
2060                                                                 DATA, true);
2061                                 else
2062                                         goto continue_unlock;
2063                         }
2064
2065                         BUG_ON(PageWriteback(page));
2066                         if (!clear_page_dirty_for_io(page))
2067                                 goto continue_unlock;
2068
2069                         ret = __write_data_page(page, &submitted, wbc, io_type);
2070                         if (unlikely(ret)) {
2071                                 /*
2072                                  * keep nr_to_write, since vfs uses this to
2073                                  * get # of written pages.
2074                                  */
2075                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2076                                         unlock_page(page);
2077                                         ret = 0;
2078                                         continue;
2079                                 } else if (ret == -EAGAIN) {
2080                                         ret = 0;
2081                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2082                                                 cond_resched();
2083                                                 congestion_wait(BLK_RW_ASYNC,
2084                                                                         HZ/50);
2085                                                 goto retry_write;
2086                                         }
2087                                         continue;
2088                                 }
2089                                 done_index = page->index + 1;
2090                                 done = 1;
2091                                 break;
2092                         } else if (submitted) {
2093                                 last_idx = page->index;
2094                         }
2095
2096                         if (--wbc->nr_to_write <= 0 &&
2097                                         wbc->sync_mode == WB_SYNC_NONE) {
2098                                 done = 1;
2099                                 break;
2100                         }
2101                 }
2102                 pagevec_release(&pvec);
2103                 cond_resched();
2104         }
2105
2106         if (!cycled && !done) {
2107                 cycled = 1;
2108                 index = 0;
2109                 end = writeback_index - 1;
2110                 goto retry;
2111         }
2112         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2113                 mapping->writeback_index = done_index;
2114
2115         if (last_idx != ULONG_MAX)
2116                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2117                                                 0, last_idx, DATA);
2118
2119         return ret;
2120 }
2121
2122 static int __f2fs_write_data_pages(struct address_space *mapping,
2123                                                 struct writeback_control *wbc,
2124                                                 enum iostat_type io_type)
2125 {
2126         struct inode *inode = mapping->host;
2127         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2128         struct blk_plug plug;
2129         int ret;
2130
2131         /* deal with chardevs and other special file */
2132         if (!mapping->a_ops->writepage)
2133                 return 0;
2134
2135         /* skip writing if there is no dirty page in this inode */
2136         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2137                 return 0;
2138
2139         /* during POR, we don't need to trigger writepage at all. */
2140         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2141                 goto skip_write;
2142
2143         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2144                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2145                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2146                 goto skip_write;
2147
2148         /* skip writing during file defragment */
2149         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2150                 goto skip_write;
2151
2152         trace_f2fs_writepages(mapping->host, wbc, DATA);
2153
2154         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2155         if (wbc->sync_mode == WB_SYNC_ALL)
2156                 atomic_inc(&sbi->wb_sync_req[DATA]);
2157         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2158                 goto skip_write;
2159
2160         blk_start_plug(&plug);
2161         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2162         blk_finish_plug(&plug);
2163
2164         if (wbc->sync_mode == WB_SYNC_ALL)
2165                 atomic_dec(&sbi->wb_sync_req[DATA]);
2166         /*
2167          * if some pages were truncated, we cannot guarantee its mapping->host
2168          * to detect pending bios.
2169          */
2170
2171         f2fs_remove_dirty_inode(inode);
2172         return ret;
2173
2174 skip_write:
2175         wbc->pages_skipped += get_dirty_pages(inode);
2176         trace_f2fs_writepages(mapping->host, wbc, DATA);
2177         return 0;
2178 }
2179
2180 static int f2fs_write_data_pages(struct address_space *mapping,
2181                             struct writeback_control *wbc)
2182 {
2183         struct inode *inode = mapping->host;
2184
2185         return __f2fs_write_data_pages(mapping, wbc,
2186                         F2FS_I(inode)->cp_task == current ?
2187                         FS_CP_DATA_IO : FS_DATA_IO);
2188 }
2189
2190 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2191 {
2192         struct inode *inode = mapping->host;
2193         loff_t i_size = i_size_read(inode);
2194
2195         if (to > i_size) {
2196                 down_write(&F2FS_I(inode)->i_mmap_sem);
2197                 truncate_pagecache(inode, i_size);
2198                 f2fs_truncate_blocks(inode, i_size, true);
2199                 up_write(&F2FS_I(inode)->i_mmap_sem);
2200         }
2201 }
2202
2203 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2204                         struct page *page, loff_t pos, unsigned len,
2205                         block_t *blk_addr, bool *node_changed)
2206 {
2207         struct inode *inode = page->mapping->host;
2208         pgoff_t index = page->index;
2209         struct dnode_of_data dn;
2210         struct page *ipage;
2211         bool locked = false;
2212         struct extent_info ei = {0,0,0};
2213         int err = 0;
2214
2215         /*
2216          * we already allocated all the blocks, so we don't need to get
2217          * the block addresses when there is no need to fill the page.
2218          */
2219         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2220                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2221                 return 0;
2222
2223         if (f2fs_has_inline_data(inode) ||
2224                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2225                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
2226                 locked = true;
2227         }
2228 restart:
2229         /* check inline_data */
2230         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2231         if (IS_ERR(ipage)) {
2232                 err = PTR_ERR(ipage);
2233                 goto unlock_out;
2234         }
2235
2236         set_new_dnode(&dn, inode, ipage, ipage, 0);
2237
2238         if (f2fs_has_inline_data(inode)) {
2239                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2240                         f2fs_do_read_inline_data(page, ipage);
2241                         set_inode_flag(inode, FI_DATA_EXIST);
2242                         if (inode->i_nlink)
2243                                 set_inline_node(ipage);
2244                 } else {
2245                         err = f2fs_convert_inline_page(&dn, page);
2246                         if (err)
2247                                 goto out;
2248                         if (dn.data_blkaddr == NULL_ADDR)
2249                                 err = f2fs_get_block(&dn, index);
2250                 }
2251         } else if (locked) {
2252                 err = f2fs_get_block(&dn, index);
2253         } else {
2254                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2255                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2256                 } else {
2257                         /* hole case */
2258                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2259                         if (err || dn.data_blkaddr == NULL_ADDR) {
2260                                 f2fs_put_dnode(&dn);
2261                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2262                                                                 true);
2263                                 locked = true;
2264                                 goto restart;
2265                         }
2266                 }
2267         }
2268
2269         /* convert_inline_page can make node_changed */
2270         *blk_addr = dn.data_blkaddr;
2271         *node_changed = dn.node_changed;
2272 out:
2273         f2fs_put_dnode(&dn);
2274 unlock_out:
2275         if (locked)
2276                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
2277         return err;
2278 }
2279
2280 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2281                 loff_t pos, unsigned len, unsigned flags,
2282                 struct page **pagep, void **fsdata)
2283 {
2284         struct inode *inode = mapping->host;
2285         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2286         struct page *page = NULL;
2287         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2288         bool need_balance = false, drop_atomic = false;
2289         block_t blkaddr = NULL_ADDR;
2290         int err = 0;
2291
2292         trace_f2fs_write_begin(inode, pos, len, flags);
2293
2294         if ((f2fs_is_atomic_file(inode) &&
2295                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2296                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2297                 err = -ENOMEM;
2298                 drop_atomic = true;
2299                 goto fail;
2300         }
2301
2302         /*
2303          * We should check this at this moment to avoid deadlock on inode page
2304          * and #0 page. The locking rule for inline_data conversion should be:
2305          * lock_page(page #0) -> lock_page(inode_page)
2306          */
2307         if (index != 0) {
2308                 err = f2fs_convert_inline_inode(inode);
2309                 if (err)
2310                         goto fail;
2311         }
2312 repeat:
2313         /*
2314          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2315          * wait_for_stable_page. Will wait that below with our IO control.
2316          */
2317         page = f2fs_pagecache_get_page(mapping, index,
2318                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2319         if (!page) {
2320                 err = -ENOMEM;
2321                 goto fail;
2322         }
2323
2324         *pagep = page;
2325
2326         err = prepare_write_begin(sbi, page, pos, len,
2327                                         &blkaddr, &need_balance);
2328         if (err)
2329                 goto fail;
2330
2331         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2332                 unlock_page(page);
2333                 f2fs_balance_fs(sbi, true);
2334                 lock_page(page);
2335                 if (page->mapping != mapping) {
2336                         /* The page got truncated from under us */
2337                         f2fs_put_page(page, 1);
2338                         goto repeat;
2339                 }
2340         }
2341
2342         f2fs_wait_on_page_writeback(page, DATA, false);
2343
2344         /* wait for GCed page writeback via META_MAPPING */
2345         if (f2fs_post_read_required(inode))
2346                 f2fs_wait_on_block_writeback(sbi, blkaddr);
2347
2348         if (len == PAGE_SIZE || PageUptodate(page))
2349                 return 0;
2350
2351         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2352                 zero_user_segment(page, len, PAGE_SIZE);
2353                 return 0;
2354         }
2355
2356         if (blkaddr == NEW_ADDR) {
2357                 zero_user_segment(page, 0, PAGE_SIZE);
2358                 SetPageUptodate(page);
2359         } else {
2360                 err = f2fs_submit_page_read(inode, page, blkaddr);
2361                 if (err)
2362                         goto fail;
2363
2364                 lock_page(page);
2365                 if (unlikely(page->mapping != mapping)) {
2366                         f2fs_put_page(page, 1);
2367                         goto repeat;
2368                 }
2369                 if (unlikely(!PageUptodate(page))) {
2370                         err = -EIO;
2371                         goto fail;
2372                 }
2373         }
2374         return 0;
2375
2376 fail:
2377         f2fs_put_page(page, 1);
2378         f2fs_write_failed(mapping, pos + len);
2379         if (drop_atomic)
2380                 f2fs_drop_inmem_pages_all(sbi, false);
2381         return err;
2382 }
2383
2384 static int f2fs_write_end(struct file *file,
2385                         struct address_space *mapping,
2386                         loff_t pos, unsigned len, unsigned copied,
2387                         struct page *page, void *fsdata)
2388 {
2389         struct inode *inode = page->mapping->host;
2390
2391         trace_f2fs_write_end(inode, pos, len, copied);
2392
2393         /*
2394          * This should be come from len == PAGE_SIZE, and we expect copied
2395          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2396          * let generic_perform_write() try to copy data again through copied=0.
2397          */
2398         if (!PageUptodate(page)) {
2399                 if (unlikely(copied != len))
2400                         copied = 0;
2401                 else
2402                         SetPageUptodate(page);
2403         }
2404         if (!copied)
2405                 goto unlock_out;
2406
2407         set_page_dirty(page);
2408
2409         if (pos + copied > i_size_read(inode))
2410                 f2fs_i_size_write(inode, pos + copied);
2411 unlock_out:
2412         f2fs_put_page(page, 1);
2413         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2414         return copied;
2415 }
2416
2417 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2418                            loff_t offset)
2419 {
2420         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2421         unsigned blkbits = i_blkbits;
2422         unsigned blocksize_mask = (1 << blkbits) - 1;
2423         unsigned long align = offset | iov_iter_alignment(iter);
2424         struct block_device *bdev = inode->i_sb->s_bdev;
2425
2426         if (align & blocksize_mask) {
2427                 if (bdev)
2428                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2429                 blocksize_mask = (1 << blkbits) - 1;
2430                 if (align & blocksize_mask)
2431                         return -EINVAL;
2432                 return 1;
2433         }
2434         return 0;
2435 }
2436
2437 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2438 {
2439         struct address_space *mapping = iocb->ki_filp->f_mapping;
2440         struct inode *inode = mapping->host;
2441         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2442         size_t count = iov_iter_count(iter);
2443         loff_t offset = iocb->ki_pos;
2444         int rw = iov_iter_rw(iter);
2445         int err;
2446         enum rw_hint hint = iocb->ki_hint;
2447         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2448
2449         err = check_direct_IO(inode, iter, offset);
2450         if (err)
2451                 return err < 0 ? err : 0;
2452
2453         if (f2fs_force_buffered_io(inode, rw))
2454                 return 0;
2455
2456         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2457
2458         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2459                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2460
2461         if (!down_read_trylock(&F2FS_I(inode)->i_gc_rwsem[rw])) {
2462                 if (iocb->ki_flags & IOCB_NOWAIT) {
2463                         iocb->ki_hint = hint;
2464                         err = -EAGAIN;
2465                         goto out;
2466                 }
2467                 down_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2468         }
2469
2470         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2471         up_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2472
2473         if (rw == WRITE) {
2474                 if (whint_mode == WHINT_MODE_OFF)
2475                         iocb->ki_hint = hint;
2476                 if (err > 0) {
2477                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2478                                                                         err);
2479                         set_inode_flag(inode, FI_UPDATE_WRITE);
2480                 } else if (err < 0) {
2481                         f2fs_write_failed(mapping, offset + count);
2482                 }
2483         }
2484
2485 out:
2486         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2487
2488         return err;
2489 }
2490
2491 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2492                                                         unsigned int length)
2493 {
2494         struct inode *inode = page->mapping->host;
2495         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2496
2497         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2498                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2499                 return;
2500
2501         if (PageDirty(page)) {
2502                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2503                         dec_page_count(sbi, F2FS_DIRTY_META);
2504                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2505                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2506                 } else {
2507                         inode_dec_dirty_pages(inode);
2508                         f2fs_remove_dirty_inode(inode);
2509                 }
2510         }
2511
2512         /* This is atomic written page, keep Private */
2513         if (IS_ATOMIC_WRITTEN_PAGE(page))
2514                 return f2fs_drop_inmem_page(inode, page);
2515
2516         set_page_private(page, 0);
2517         ClearPagePrivate(page);
2518 }
2519
2520 int f2fs_release_page(struct page *page, gfp_t wait)
2521 {
2522         /* If this is dirty page, keep PagePrivate */
2523         if (PageDirty(page))
2524                 return 0;
2525
2526         /* This is atomic written page, keep Private */
2527         if (IS_ATOMIC_WRITTEN_PAGE(page))
2528                 return 0;
2529
2530         set_page_private(page, 0);
2531         ClearPagePrivate(page);
2532         return 1;
2533 }
2534
2535 static int f2fs_set_data_page_dirty(struct page *page)
2536 {
2537         struct address_space *mapping = page->mapping;
2538         struct inode *inode = mapping->host;
2539
2540         trace_f2fs_set_page_dirty(page, DATA);
2541
2542         if (!PageUptodate(page))
2543                 SetPageUptodate(page);
2544
2545         /* don't remain PG_checked flag which was set during GC */
2546         if (is_cold_data(page))
2547                 clear_cold_data(page);
2548
2549         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2550                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2551                         f2fs_register_inmem_page(inode, page);
2552                         return 1;
2553                 }
2554                 /*
2555                  * Previously, this page has been registered, we just
2556                  * return here.
2557                  */
2558                 return 0;
2559         }
2560
2561         if (!PageDirty(page)) {
2562                 __set_page_dirty_nobuffers(page);
2563                 f2fs_update_dirty_page(inode, page);
2564                 return 1;
2565         }
2566         return 0;
2567 }
2568
2569 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2570 {
2571         struct inode *inode = mapping->host;
2572
2573         if (f2fs_has_inline_data(inode))
2574                 return 0;
2575
2576         /* make sure allocating whole blocks */
2577         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2578                 filemap_write_and_wait(mapping);
2579
2580         return generic_block_bmap(mapping, block, get_data_block_bmap);
2581 }
2582
2583 #ifdef CONFIG_MIGRATION
2584 #include <linux/migrate.h>
2585
2586 int f2fs_migrate_page(struct address_space *mapping,
2587                 struct page *newpage, struct page *page, enum migrate_mode mode)
2588 {
2589         int rc, extra_count;
2590         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2591         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2592
2593         BUG_ON(PageWriteback(page));
2594
2595         /* migrating an atomic written page is safe with the inmem_lock hold */
2596         if (atomic_written) {
2597                 if (mode != MIGRATE_SYNC)
2598                         return -EBUSY;
2599                 if (!mutex_trylock(&fi->inmem_lock))
2600                         return -EAGAIN;
2601         }
2602
2603         /*
2604          * A reference is expected if PagePrivate set when move mapping,
2605          * however F2FS breaks this for maintaining dirty page counts when
2606          * truncating pages. So here adjusting the 'extra_count' make it work.
2607          */
2608         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2609         rc = migrate_page_move_mapping(mapping, newpage,
2610                                 page, NULL, mode, extra_count);
2611         if (rc != MIGRATEPAGE_SUCCESS) {
2612                 if (atomic_written)
2613                         mutex_unlock(&fi->inmem_lock);
2614                 return rc;
2615         }
2616
2617         if (atomic_written) {
2618                 struct inmem_pages *cur;
2619                 list_for_each_entry(cur, &fi->inmem_pages, list)
2620                         if (cur->page == page) {
2621                                 cur->page = newpage;
2622                                 break;
2623                         }
2624                 mutex_unlock(&fi->inmem_lock);
2625                 put_page(page);
2626                 get_page(newpage);
2627         }
2628
2629         if (PagePrivate(page))
2630                 SetPagePrivate(newpage);
2631         set_page_private(newpage, page_private(page));
2632
2633         if (mode != MIGRATE_SYNC_NO_COPY)
2634                 migrate_page_copy(newpage, page);
2635         else
2636                 migrate_page_states(newpage, page);
2637
2638         return MIGRATEPAGE_SUCCESS;
2639 }
2640 #endif
2641
2642 const struct address_space_operations f2fs_dblock_aops = {
2643         .readpage       = f2fs_read_data_page,
2644         .readpages      = f2fs_read_data_pages,
2645         .writepage      = f2fs_write_data_page,
2646         .writepages     = f2fs_write_data_pages,
2647         .write_begin    = f2fs_write_begin,
2648         .write_end      = f2fs_write_end,
2649         .set_page_dirty = f2fs_set_data_page_dirty,
2650         .invalidatepage = f2fs_invalidate_page,
2651         .releasepage    = f2fs_release_page,
2652         .direct_IO      = f2fs_direct_IO,
2653         .bmap           = f2fs_bmap,
2654 #ifdef CONFIG_MIGRATION
2655         .migratepage    = f2fs_migrate_page,
2656 #endif
2657 };
2658
2659 void f2fs_clear_radix_tree_dirty_tag(struct page *page)
2660 {
2661         struct address_space *mapping = page_mapping(page);
2662         unsigned long flags;
2663
2664         xa_lock_irqsave(&mapping->i_pages, flags);
2665         radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2666                                                 PAGECACHE_TAG_DIRTY);
2667         xa_unlock_irqrestore(&mapping->i_pages, flags);
2668 }
2669
2670 int __init f2fs_init_post_read_processing(void)
2671 {
2672         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2673         if (!bio_post_read_ctx_cache)
2674                 goto fail;
2675         bio_post_read_ctx_pool =
2676                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2677                                          bio_post_read_ctx_cache);
2678         if (!bio_post_read_ctx_pool)
2679                 goto fail_free_cache;
2680         return 0;
2681
2682 fail_free_cache:
2683         kmem_cache_destroy(bio_post_read_ctx_cache);
2684 fail:
2685         return -ENOMEM;
2686 }
2687
2688 void __exit f2fs_destroy_post_read_processing(void)
2689 {
2690         mempool_destroy(bio_post_read_ctx_pool);
2691         kmem_cache_destroy(bio_post_read_ctx_cache);
2692 }