fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/prefetch.h>
  18 #include <linux/uio.h>
  19 #include <linux/cleancache.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include "trace.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define NUM_PREALLOC_POST_READ_CTXS     128
  29
  30 static struct kmem_cache *bio_post_read_ctx_cache;
  31 static mempool_t *bio_post_read_ctx_pool;
  32
  33 static bool __is_cp_guaranteed(struct page *page)
  34 {
  35         struct address_space *mapping = page->mapping;
  36         struct inode *inode;
  37         struct f2fs_sb_info *sbi;
  38
  39         if (!mapping)
  40                 return false;
  41
  42         inode = mapping->host;
  43         sbi = F2FS_I_SB(inode);
  44
  45         if (inode->i_ino == F2FS_META_INO(sbi) ||
  46                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47                         S_ISDIR(inode->i_mode) ||
  48                         (S_ISREG(inode->i_mode) &&
  49                         is_inode_flag_set(inode, FI_ATOMIC_FILE)) ||
  50                         is_cold_data(page))
  51                 return true;
  52         return false;
  53 }
  54
  55 /* postprocessing steps for read bios */
  56 enum bio_post_read_step {
  57         STEP_INITIAL = 0,
  58         STEP_DECRYPT,
  59 };
  60
  61 struct bio_post_read_ctx {
  62         struct bio *bio;
  63         struct work_struct work;
  64         unsigned int cur_step;
  65         unsigned int enabled_steps;
  66 };
  67
  68 static void __read_end_io(struct bio *bio)
  69 {
  70         struct page *page;
  71         struct bio_vec *bv;
  72         int i;
  73
  74         bio_for_each_segment_all(bv, bio, i) {
  75                 page = bv->bv_page;
  76
  77                 /* PG_error was set if any post_read step failed */
  78                 if (bio->bi_status || PageError(page)) {
  79                         ClearPageUptodate(page);
  80                         SetPageError(page);
  81                 } else {
  82                         SetPageUptodate(page);
  83                 }
  84                 unlock_page(page);
  85         }
  86         if (bio->bi_private)
  87                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
  88         bio_put(bio);
  89 }
  90
  91 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
  92
  93 static void decrypt_work(struct work_struct *work)
  94 {
  95         struct bio_post_read_ctx *ctx =
  96                 container_of(work, struct bio_post_read_ctx, work);
  97
  98         fscrypt_decrypt_bio(ctx->bio);
  99
 100         bio_post_read_processing(ctx);
 101 }
 102
 103 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 104 {
 105         switch (++ctx->cur_step) {
 106         case STEP_DECRYPT:
 107                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 108                         INIT_WORK(&ctx->work, decrypt_work);
 109                         fscrypt_enqueue_decrypt_work(&ctx->work);
 110                         return;
 111                 }
 112                 ctx->cur_step++;
 113                 /* fall-through */
 114         default:
 115                 __read_end_io(ctx->bio);
 116         }
 117 }
 118
 119 static bool f2fs_bio_post_read_required(struct bio *bio)
 120 {
 121         return bio->bi_private && !bio->bi_status;
 122 }
 123
 124 static void f2fs_read_end_io(struct bio *bio)
 125 {
 126         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
 127                                                 FAULT_READ_IO)) {
 128                 f2fs_show_injection_info(FAULT_READ_IO);
 129                 bio->bi_status = BLK_STS_IOERR;
 130         }
 131
 132         if (f2fs_bio_post_read_required(bio)) {
 133                 struct bio_post_read_ctx *ctx = bio->bi_private;
 134
 135                 ctx->cur_step = STEP_INITIAL;
 136                 bio_post_read_processing(ctx);
 137                 return;
 138         }
 139
 140         __read_end_io(bio);
 141 }
 142
 143 static void f2fs_write_end_io(struct bio *bio)
 144 {
 145         struct f2fs_sb_info *sbi = bio->bi_private;
 146         struct bio_vec *bvec;
 147         int i;
 148
 149         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 150                 f2fs_show_injection_info(FAULT_WRITE_IO);
 151                 bio->bi_status = BLK_STS_IOERR;
 152         }
 153
 154         bio_for_each_segment_all(bvec, bio, i) {
 155                 struct page *page = bvec->bv_page;
 156                 enum count_type type = WB_DATA_TYPE(page);
 157
 158                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 159                         set_page_private(page, (unsigned long)NULL);
 160                         ClearPagePrivate(page);
 161                         unlock_page(page);
 162                         mempool_free(page, sbi->write_io_dummy);
 163
 164                         if (unlikely(bio->bi_status))
 165                                 f2fs_stop_checkpoint(sbi, true);
 166                         continue;
 167                 }
 168
 169                 fscrypt_pullback_bio_page(&page, true);
 170
 171                 if (unlikely(bio->bi_status)) {
 172                         mapping_set_error(page->mapping, -EIO);
 173                         if (type == F2FS_WB_CP_DATA)
 174                                 f2fs_stop_checkpoint(sbi, true);
 175                 }
 176
 177                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 178                                         page->index != nid_of_node(page));
 179
 180                 dec_page_count(sbi, type);
 181                 if (f2fs_in_warm_node_list(sbi, page))
 182                         f2fs_del_fsync_node_entry(sbi, page);
 183                 clear_cold_data(page);
 184                 end_page_writeback(page);
 185         }
 186         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 187                                 wq_has_sleeper(&sbi->cp_wait))
 188                 wake_up(&sbi->cp_wait);
 189
 190         bio_put(bio);
 191 }
 192
 193 /*
 194  * Return true, if pre_bio's bdev is same as its target device.
 195  */
 196 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 197                                 block_t blk_addr, struct bio *bio)
 198 {
 199         struct block_device *bdev = sbi->sb->s_bdev;
 200         int i;
 201
 202         for (i = 0; i < sbi->s_ndevs; i++) {
 203                 if (FDEV(i).start_blk <= blk_addr &&
 204                                         FDEV(i).end_blk >= blk_addr) {
 205                         blk_addr -= FDEV(i).start_blk;
 206                         bdev = FDEV(i).bdev;
 207                         break;
 208                 }
 209         }
 210         if (bio) {
 211                 bio_set_dev(bio, bdev);
 212                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 213         }
 214         return bdev;
 215 }
 216
 217 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 218 {
 219         int i;
 220
 221         for (i = 0; i < sbi->s_ndevs; i++)
 222                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 223                         return i;
 224         return 0;
 225 }
 226
 227 static bool __same_bdev(struct f2fs_sb_info *sbi,
 228                                 block_t blk_addr, struct bio *bio)
 229 {
 230         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 231         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 232 }
 233
 234 /*
 235  * Low-level block read/write IO operations.
 236  */
 237 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 238                                 struct writeback_control *wbc,
 239                                 int npages, bool is_read,
 240                                 enum page_type type, enum temp_type temp)
 241 {
 242         struct bio *bio;
 243
 244         bio = f2fs_bio_alloc(sbi, npages, true);
 245
 246         f2fs_target_device(sbi, blk_addr, bio);
 247         if (is_read) {
 248                 bio->bi_end_io = f2fs_read_end_io;
 249                 bio->bi_private = NULL;
 250         } else {
 251                 bio->bi_end_io = f2fs_write_end_io;
 252                 bio->bi_private = sbi;
 253                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 254         }
 255         if (wbc)
 256                 wbc_init_bio(wbc, bio);
 257
 258         return bio;
 259 }
 260
 261 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 262                                 struct bio *bio, enum page_type type)
 263 {
 264         if (!is_read_io(bio_op(bio))) {
 265                 unsigned int start;
 266
 267                 if (type != DATA && type != NODE)
 268                         goto submit_io;
 269
 270                 if (test_opt(sbi, LFS) && current->plug)
 271                         blk_finish_plug(current->plug);
 272
 273                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 274                 start %= F2FS_IO_SIZE(sbi);
 275
 276                 if (start == 0)
 277                         goto submit_io;
 278
 279                 /* fill dummy pages */
 280                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 281                         struct page *page =
 282                                 mempool_alloc(sbi->write_io_dummy,
 283                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 284                         f2fs_bug_on(sbi, !page);
 285
 286                         SetPagePrivate(page);
 287                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 288                         lock_page(page);
 289                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 290                                 f2fs_bug_on(sbi, 1);
 291                 }
 292                 /*
 293                  * In the NODE case, we lose next block address chain. So, we
 294                  * need to do checkpoint in f2fs_sync_file.
 295                  */
 296                 if (type == NODE)
 297                         set_sbi_flag(sbi, SBI_NEED_CP);
 298         }
 299 submit_io:
 300         if (is_read_io(bio_op(bio)))
 301                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 302         else
 303                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 304         submit_bio(bio);
 305 }
 306
 307 static void __submit_merged_bio(struct f2fs_bio_info *io)
 308 {
 309         struct f2fs_io_info *fio = &io->fio;
 310
 311         if (!io->bio)
 312                 return;
 313
 314         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 315
 316         if (is_read_io(fio->op))
 317                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 318         else
 319                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 320
 321         __submit_bio(io->sbi, io->bio, fio->type);
 322         io->bio = NULL;
 323 }
 324
 325 static bool __has_merged_page(struct f2fs_bio_info *io,
 326                                 struct inode *inode, nid_t ino, pgoff_t idx)
 327 {
 328         struct bio_vec *bvec;
 329         struct page *target;
 330         int i;
 331
 332         if (!io->bio)
 333                 return false;
 334
 335         if (!inode && !ino)
 336                 return true;
 337
 338         bio_for_each_segment_all(bvec, io->bio, i) {
 339
 340                 if (bvec->bv_page->mapping)
 341                         target = bvec->bv_page;
 342                 else
 343                         target = fscrypt_control_page(bvec->bv_page);
 344
 345                 if (idx != target->index)
 346                         continue;
 347
 348                 if (inode && inode == target->mapping->host)
 349                         return true;
 350                 if (ino && ino == ino_of_node(target))
 351                         return true;
 352         }
 353
 354         return false;
 355 }
 356
 357 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 358                                 nid_t ino, pgoff_t idx, enum page_type type)
 359 {
 360         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 361         enum temp_type temp;
 362         struct f2fs_bio_info *io;
 363         bool ret = false;
 364
 365         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 366                 io = sbi->write_io[btype] + temp;
 367
 368                 down_read(&io->io_rwsem);
 369                 ret = __has_merged_page(io, inode, ino, idx);
 370                 up_read(&io->io_rwsem);
 371
 372                 /* TODO: use HOT temp only for meta pages now. */
 373                 if (ret || btype == META)
 374                         break;
 375         }
 376         return ret;
 377 }
 378
 379 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 380                                 enum page_type type, enum temp_type temp)
 381 {
 382         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 383         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 384
 385         down_write(&io->io_rwsem);
 386
 387         /* change META to META_FLUSH in the checkpoint procedure */
 388         if (type >= META_FLUSH) {
 389                 io->fio.type = META_FLUSH;
 390                 io->fio.op = REQ_OP_WRITE;
 391                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 392                 if (!test_opt(sbi, NOBARRIER))
 393                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 394         }
 395         __submit_merged_bio(io);
 396         up_write(&io->io_rwsem);
 397 }
 398
 399 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 400                                 struct inode *inode, nid_t ino, pgoff_t idx,
 401                                 enum page_type type, bool force)
 402 {
 403         enum temp_type temp;
 404
 405         if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 406                 return;
 407
 408         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 409
 410                 __f2fs_submit_merged_write(sbi, type, temp);
 411
 412                 /* TODO: use HOT temp only for meta pages now. */
 413                 if (type >= META)
 414                         break;
 415         }
 416 }
 417
 418 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 419 {
 420         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 421 }
 422
 423 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 424                                 struct inode *inode, nid_t ino, pgoff_t idx,
 425                                 enum page_type type)
 426 {
 427         __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 428 }
 429
 430 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 431 {
 432         f2fs_submit_merged_write(sbi, DATA);
 433         f2fs_submit_merged_write(sbi, NODE);
 434         f2fs_submit_merged_write(sbi, META);
 435 }
 436
 437 /*
 438  * Fill the locked page with data located in the block address.
 439  * A caller needs to unlock the page on failure.
 440  */
 441 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 442 {
 443         struct bio *bio;
 444         struct page *page = fio->encrypted_page ?
 445                         fio->encrypted_page : fio->page;
 446
 447         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 448                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 449                 return -EFAULT;
 450
 451         trace_f2fs_submit_page_bio(page, fio);
 452         f2fs_trace_ios(fio, 0);
 453
 454         /* Allocate a new bio */
 455         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 456                                 1, is_read_io(fio->op), fio->type, fio->temp);
 457
 458         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 459                 bio_put(bio);
 460                 return -EFAULT;
 461         }
 462         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 463
 464         __submit_bio(fio->sbi, bio, fio->type);
 465
 466         if (!is_read_io(fio->op))
 467                 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 468         return 0;
 469 }
 470
 471 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 472 {
 473         struct f2fs_sb_info *sbi = fio->sbi;
 474         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 475         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 476         struct page *bio_page;
 477
 478         f2fs_bug_on(sbi, is_read_io(fio->op));
 479
 480         down_write(&io->io_rwsem);
 481 next:
 482         if (fio->in_list) {
 483                 spin_lock(&io->io_lock);
 484                 if (list_empty(&io->io_list)) {
 485                         spin_unlock(&io->io_lock);
 486                         goto out;
 487                 }
 488                 fio = list_first_entry(&io->io_list,
 489                                                 struct f2fs_io_info, list);
 490                 list_del(&fio->list);
 491                 spin_unlock(&io->io_lock);
 492         }
 493
 494         if (__is_valid_data_blkaddr(fio->old_blkaddr))
 495                 verify_block_addr(fio, fio->old_blkaddr);
 496         verify_block_addr(fio, fio->new_blkaddr);
 497
 498         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 499
 500         /* set submitted = true as a return value */
 501         fio->submitted = true;
 502
 503         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 504
 505         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 506             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 507                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 508                 __submit_merged_bio(io);
 509 alloc_new:
 510         if (io->bio == NULL) {
 511                 if ((fio->type == DATA || fio->type == NODE) &&
 512                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 513                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 514                         fio->retry = true;
 515                         goto skip;
 516                 }
 517                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 518                                                 BIO_MAX_PAGES, false,
 519                                                 fio->type, fio->temp);
 520                 io->fio = *fio;
 521         }
 522
 523         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 524                 __submit_merged_bio(io);
 525                 goto alloc_new;
 526         }
 527
 528         if (fio->io_wbc)
 529                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 530
 531         io->last_block_in_bio = fio->new_blkaddr;
 532         f2fs_trace_ios(fio, 0);
 533
 534         trace_f2fs_submit_page_write(fio->page, fio);
 535 skip:
 536         if (fio->in_list)
 537                 goto next;
 538 out:
 539         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN))
 540                 __submit_merged_bio(io);
 541         up_write(&io->io_rwsem);
 542 }
 543
 544 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 545                                         unsigned nr_pages, unsigned op_flag)
 546 {
 547         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 548         struct bio *bio;
 549         struct bio_post_read_ctx *ctx;
 550         unsigned int post_read_steps = 0;
 551
 552         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
 553                 return ERR_PTR(-EFAULT);
 554
 555         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 556         if (!bio)
 557                 return ERR_PTR(-ENOMEM);
 558         f2fs_target_device(sbi, blkaddr, bio);
 559         bio->bi_end_io = f2fs_read_end_io;
 560         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 561
 562         if (f2fs_encrypted_file(inode))
 563                 post_read_steps |= 1 << STEP_DECRYPT;
 564         if (post_read_steps) {
 565                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 566                 if (!ctx) {
 567                         bio_put(bio);
 568                         return ERR_PTR(-ENOMEM);
 569                 }
 570                 ctx->bio = bio;
 571                 ctx->enabled_steps = post_read_steps;
 572                 bio->bi_private = ctx;
 573         }
 574
 575         return bio;
 576 }
 577
 578 /* This can handle encryption stuffs */
 579 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 580                                                         block_t blkaddr)
 581 {
 582         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 583
 584         if (IS_ERR(bio))
 585                 return PTR_ERR(bio);
 586
 587         /* wait for GCed page writeback via META_MAPPING */
 588         f2fs_wait_on_block_writeback(inode, blkaddr);
 589
 590         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 591                 bio_put(bio);
 592                 return -EFAULT;
 593         }
 594         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 595         return 0;
 596 }
 597
 598 static void __set_data_blkaddr(struct dnode_of_data *dn)
 599 {
 600         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 601         __le32 *addr_array;
 602         int base = 0;
 603
 604         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 605                 base = get_extra_isize(dn->inode);
 606
 607         /* Get physical address of data block */
 608         addr_array = blkaddr_in_node(rn);
 609         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 610 }
 611
 612 /*
 613  * Lock ordering for the change of data block address:
 614  * ->data_page
 615  *  ->node_page
 616  *    update block addresses in the node page
 617  */
 618 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 619 {
 620         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 621         __set_data_blkaddr(dn);
 622         if (set_page_dirty(dn->node_page))
 623                 dn->node_changed = true;
 624 }
 625
 626 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 627 {
 628         dn->data_blkaddr = blkaddr;
 629         f2fs_set_data_blkaddr(dn);
 630         f2fs_update_extent_cache(dn);
 631 }
 632
 633 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 634 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 635 {
 636         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 637         int err;
 638
 639         if (!count)
 640                 return 0;
 641
 642         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 643                 return -EPERM;
 644         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 645                 return err;
 646
 647         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 648                                                 dn->ofs_in_node, count);
 649
 650         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 651
 652         for (; count > 0; dn->ofs_in_node++) {
 653                 block_t blkaddr = datablock_addr(dn->inode,
 654                                         dn->node_page, dn->ofs_in_node);
 655                 if (blkaddr == NULL_ADDR) {
 656                         dn->data_blkaddr = NEW_ADDR;
 657                         __set_data_blkaddr(dn);
 658                         count--;
 659                 }
 660         }
 661
 662         if (set_page_dirty(dn->node_page))
 663                 dn->node_changed = true;
 664         return 0;
 665 }
 666
 667 /* Should keep dn->ofs_in_node unchanged */
 668 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 669 {
 670         unsigned int ofs_in_node = dn->ofs_in_node;
 671         int ret;
 672
 673         ret = f2fs_reserve_new_blocks(dn, 1);
 674         dn->ofs_in_node = ofs_in_node;
 675         return ret;
 676 }
 677
 678 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 679 {
 680         bool need_put = dn->inode_page ? false : true;
 681         int err;
 682
 683         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 684         if (err)
 685                 return err;
 686
 687         if (dn->data_blkaddr == NULL_ADDR)
 688                 err = f2fs_reserve_new_block(dn);
 689         if (err || need_put)
 690                 f2fs_put_dnode(dn);
 691         return err;
 692 }
 693
 694 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 695 {
 696         struct extent_info ei  = {0,0,0};
 697         struct inode *inode = dn->inode;
 698
 699         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 700                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 701                 return 0;
 702         }
 703
 704         return f2fs_reserve_block(dn, index);
 705 }
 706
 707 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 708                                                 int op_flags, bool for_write)
 709 {
 710         struct address_space *mapping = inode->i_mapping;
 711         struct dnode_of_data dn;
 712         struct page *page;
 713         struct extent_info ei = {0,0,0};
 714         int err;
 715
 716         page = f2fs_grab_cache_page(mapping, index, for_write);
 717         if (!page)
 718                 return ERR_PTR(-ENOMEM);
 719
 720         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 721                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 722                 goto got_it;
 723         }
 724
 725         set_new_dnode(&dn, inode, NULL, NULL, 0);
 726         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 727         if (err)
 728                 goto put_err;
 729         f2fs_put_dnode(&dn);
 730
 731         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 732                 err = -ENOENT;
 733                 goto put_err;
 734         }
 735 got_it:
 736         if (PageUptodate(page)) {
 737                 unlock_page(page);
 738                 return page;
 739         }
 740
 741         /*
 742          * A new dentry page is allocated but not able to be written, since its
 743          * new inode page couldn't be allocated due to -ENOSPC.
 744          * In such the case, its blkaddr can be remained as NEW_ADDR.
 745          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 746          * f2fs_init_inode_metadata.
 747          */
 748         if (dn.data_blkaddr == NEW_ADDR) {
 749                 zero_user_segment(page, 0, PAGE_SIZE);
 750                 if (!PageUptodate(page))
 751                         SetPageUptodate(page);
 752                 unlock_page(page);
 753                 return page;
 754         }
 755
 756         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 757         if (err)
 758                 goto put_err;
 759         return page;
 760
 761 put_err:
 762         f2fs_put_page(page, 1);
 763         return ERR_PTR(err);
 764 }
 765
 766 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 767 {
 768         struct address_space *mapping = inode->i_mapping;
 769         struct page *page;
 770
 771         page = find_get_page(mapping, index);
 772         if (page && PageUptodate(page))
 773                 return page;
 774         f2fs_put_page(page, 0);
 775
 776         page = f2fs_get_read_data_page(inode, index, 0, false);
 777         if (IS_ERR(page))
 778                 return page;
 779
 780         if (PageUptodate(page))
 781                 return page;
 782
 783         wait_on_page_locked(page);
 784         if (unlikely(!PageUptodate(page))) {
 785                 f2fs_put_page(page, 0);
 786                 return ERR_PTR(-EIO);
 787         }
 788         return page;
 789 }
 790
 791 /*
 792  * If it tries to access a hole, return an error.
 793  * Because, the callers, functions in dir.c and GC, should be able to know
 794  * whether this page exists or not.
 795  */
 796 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 797                                                         bool for_write)
 798 {
 799         struct address_space *mapping = inode->i_mapping;
 800         struct page *page;
 801 repeat:
 802         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 803         if (IS_ERR(page))
 804                 return page;
 805
 806         /* wait for read completion */
 807         lock_page(page);
 808         if (unlikely(page->mapping != mapping)) {
 809                 f2fs_put_page(page, 1);
 810                 goto repeat;
 811         }
 812         if (unlikely(!PageUptodate(page))) {
 813                 f2fs_put_page(page, 1);
 814                 return ERR_PTR(-EIO);
 815         }
 816         return page;
 817 }
 818
 819 /*
 820  * Caller ensures that this data page is never allocated.
 821  * A new zero-filled data page is allocated in the page cache.
 822  *
 823  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 824  * f2fs_unlock_op().
 825  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 826  * ipage should be released by this function.
 827  */
 828 struct page *f2fs_get_new_data_page(struct inode *inode,
 829                 struct page *ipage, pgoff_t index, bool new_i_size)
 830 {
 831         struct address_space *mapping = inode->i_mapping;
 832         struct page *page;
 833         struct dnode_of_data dn;
 834         int err;
 835
 836         page = f2fs_grab_cache_page(mapping, index, true);
 837         if (!page) {
 838                 /*
 839                  * before exiting, we should make sure ipage will be released
 840                  * if any error occur.
 841                  */
 842                 f2fs_put_page(ipage, 1);
 843                 return ERR_PTR(-ENOMEM);
 844         }
 845
 846         set_new_dnode(&dn, inode, ipage, NULL, 0);
 847         err = f2fs_reserve_block(&dn, index);
 848         if (err) {
 849                 f2fs_put_page(page, 1);
 850                 return ERR_PTR(err);
 851         }
 852         if (!ipage)
 853                 f2fs_put_dnode(&dn);
 854
 855         if (PageUptodate(page))
 856                 goto got_it;
 857
 858         if (dn.data_blkaddr == NEW_ADDR) {
 859                 zero_user_segment(page, 0, PAGE_SIZE);
 860                 if (!PageUptodate(page))
 861                         SetPageUptodate(page);
 862         } else {
 863                 f2fs_put_page(page, 1);
 864
 865                 /* if ipage exists, blkaddr should be NEW_ADDR */
 866                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 867                 page = f2fs_get_lock_data_page(inode, index, true);
 868                 if (IS_ERR(page))
 869                         return page;
 870         }
 871 got_it:
 872         if (new_i_size && i_size_read(inode) <
 873                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 874                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 875         return page;
 876 }
 877
 878 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 879 {
 880         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 881         struct f2fs_summary sum;
 882         struct node_info ni;
 883         block_t old_blkaddr;
 884         pgoff_t fofs;
 885         blkcnt_t count = 1;
 886         int err;
 887
 888         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 889                 return -EPERM;
 890
 891         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 892         if (err)
 893                 return err;
 894
 895         dn->data_blkaddr = datablock_addr(dn->inode,
 896                                 dn->node_page, dn->ofs_in_node);
 897         if (dn->data_blkaddr == NEW_ADDR)
 898                 goto alloc;
 899
 900         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 901                 return err;
 902
 903 alloc:
 904         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 905         old_blkaddr = dn->data_blkaddr;
 906         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 907                                         &sum, seg_type, NULL, false);
 908         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 909                 invalidate_mapping_pages(META_MAPPING(sbi),
 910                                         old_blkaddr, old_blkaddr);
 911         f2fs_set_data_blkaddr(dn);
 912
 913         /* update i_size */
 914         fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 915                                                         dn->ofs_in_node;
 916         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 917                 f2fs_i_size_write(dn->inode,
 918                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 919         return 0;
 920 }
 921
 922 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 923 {
 924         struct inode *inode = file_inode(iocb->ki_filp);
 925         struct f2fs_map_blocks map;
 926         int flag;
 927         int err = 0;
 928         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 929
 930         /* convert inline data for Direct I/O*/
 931         if (direct_io) {
 932                 err = f2fs_convert_inline_inode(inode);
 933                 if (err)
 934                         return err;
 935         }
 936
 937         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 938                 return 0;
 939
 940         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 941         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 942         if (map.m_len > map.m_lblk)
 943                 map.m_len -= map.m_lblk;
 944         else
 945                 map.m_len = 0;
 946
 947         map.m_next_pgofs = NULL;
 948         map.m_next_extent = NULL;
 949         map.m_seg_type = NO_CHECK_TYPE;
 950
 951         if (direct_io) {
 952                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 953                 flag = f2fs_force_buffered_io(inode, WRITE) ?
 954                                         F2FS_GET_BLOCK_PRE_AIO :
 955                                         F2FS_GET_BLOCK_PRE_DIO;
 956                 goto map_blocks;
 957         }
 958         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 959                 err = f2fs_convert_inline_inode(inode);
 960                 if (err)
 961                         return err;
 962         }
 963         if (f2fs_has_inline_data(inode))
 964                 return err;
 965
 966         flag = F2FS_GET_BLOCK_PRE_AIO;
 967
 968 map_blocks:
 969         err = f2fs_map_blocks(inode, &map, 1, flag);
 970         if (map.m_len > 0 && err == -ENOSPC) {
 971                 if (!direct_io)
 972                         set_inode_flag(inode, FI_NO_PREALLOC);
 973                 err = 0;
 974         }
 975         return err;
 976 }
 977
 978 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 979 {
 980         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 981                 if (lock)
 982                         down_read(&sbi->node_change);
 983                 else
 984                         up_read(&sbi->node_change);
 985         } else {
 986                 if (lock)
 987                         f2fs_lock_op(sbi);
 988                 else
 989                         f2fs_unlock_op(sbi);
 990         }
 991 }
 992
 993 /*
 994  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 995  * f2fs_map_blocks structure.
 996  * If original data blocks are allocated, then give them to blockdev.
 997  * Otherwise,
 998  *     a. preallocate requested block addresses
 999  *     b. do not use extent cache for better performance
1000  *     c. give the block addresses to blockdev
1001  */
1002 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1003                                                 int create, int flag)
1004 {
1005         unsigned int maxblocks = map->m_len;
1006         struct dnode_of_data dn;
1007         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1008         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
1009         pgoff_t pgofs, end_offset, end;
1010         int err = 0, ofs = 1;
1011         unsigned int ofs_in_node, last_ofs_in_node;
1012         blkcnt_t prealloc;
1013         struct extent_info ei = {0,0,0};
1014         block_t blkaddr;
1015         unsigned int start_pgofs;
1016
1017         if (!maxblocks)
1018                 return 0;
1019
1020         map->m_len = 0;
1021         map->m_flags = 0;
1022
1023         /* it only supports block size == page size */
1024         pgofs = (pgoff_t)map->m_lblk;
1025         end = pgofs + maxblocks;
1026
1027         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1028                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1029                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1030                 map->m_flags = F2FS_MAP_MAPPED;
1031                 if (map->m_next_extent)
1032                         *map->m_next_extent = pgofs + map->m_len;
1033                 goto out;
1034         }
1035
1036 next_dnode:
1037         if (create)
1038                 __do_map_lock(sbi, flag, true);
1039
1040         /* When reading holes, we need its node page */
1041         set_new_dnode(&dn, inode, NULL, NULL, 0);
1042         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1043         if (err) {
1044                 if (flag == F2FS_GET_BLOCK_BMAP)
1045                         map->m_pblk = 0;
1046                 if (err == -ENOENT) {
1047                         err = 0;
1048                         if (map->m_next_pgofs)
1049                                 *map->m_next_pgofs =
1050                                         f2fs_get_next_page_offset(&dn, pgofs);
1051                         if (map->m_next_extent)
1052                                 *map->m_next_extent =
1053                                         f2fs_get_next_page_offset(&dn, pgofs);
1054                 }
1055                 goto unlock_out;
1056         }
1057
1058         start_pgofs = pgofs;
1059         prealloc = 0;
1060         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1061         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1062
1063 next_block:
1064         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1065
1066         if (__is_valid_data_blkaddr(blkaddr) &&
1067                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1068                 err = -EFAULT;
1069                 goto sync_out;
1070         }
1071
1072         if (!is_valid_data_blkaddr(sbi, blkaddr)) {
1073                 if (create) {
1074                         if (unlikely(f2fs_cp_error(sbi))) {
1075                                 err = -EIO;
1076                                 goto sync_out;
1077                         }
1078                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1079                                 if (blkaddr == NULL_ADDR) {
1080                                         prealloc++;
1081                                         last_ofs_in_node = dn.ofs_in_node;
1082                                 }
1083                         } else {
1084                                 err = __allocate_data_block(&dn,
1085                                                         map->m_seg_type);
1086                                 if (!err)
1087                                         set_inode_flag(inode, FI_APPEND_WRITE);
1088                         }
1089                         if (err)
1090                                 goto sync_out;
1091                         map->m_flags |= F2FS_MAP_NEW;
1092                         blkaddr = dn.data_blkaddr;
1093                 } else {
1094                         if (flag == F2FS_GET_BLOCK_BMAP) {
1095                                 map->m_pblk = 0;
1096                                 goto sync_out;
1097                         }
1098                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1099                                 goto sync_out;
1100                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1101                                                 blkaddr == NULL_ADDR) {
1102                                 if (map->m_next_pgofs)
1103                                         *map->m_next_pgofs = pgofs + 1;
1104                                 goto sync_out;
1105                         }
1106                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1107                                 /* for defragment case */
1108                                 if (map->m_next_pgofs)
1109                                         *map->m_next_pgofs = pgofs + 1;
1110                                 goto sync_out;
1111                         }
1112                 }
1113         }
1114
1115         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1116                 goto skip;
1117
1118         if (map->m_len == 0) {
1119                 /* preallocated unwritten block should be mapped for fiemap. */
1120                 if (blkaddr == NEW_ADDR)
1121                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1122                 map->m_flags |= F2FS_MAP_MAPPED;
1123
1124                 map->m_pblk = blkaddr;
1125                 map->m_len = 1;
1126         } else if ((map->m_pblk != NEW_ADDR &&
1127                         blkaddr == (map->m_pblk + ofs)) ||
1128                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1129                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1130                 ofs++;
1131                 map->m_len++;
1132         } else {
1133                 goto sync_out;
1134         }
1135
1136 skip:
1137         dn.ofs_in_node++;
1138         pgofs++;
1139
1140         /* preallocate blocks in batch for one dnode page */
1141         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1142                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1143
1144                 dn.ofs_in_node = ofs_in_node;
1145                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1146                 if (err)
1147                         goto sync_out;
1148
1149                 map->m_len += dn.ofs_in_node - ofs_in_node;
1150                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1151                         err = -ENOSPC;
1152                         goto sync_out;
1153                 }
1154                 dn.ofs_in_node = end_offset;
1155         }
1156
1157         if (pgofs >= end)
1158                 goto sync_out;
1159         else if (dn.ofs_in_node < end_offset)
1160                 goto next_block;
1161
1162         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1163                 if (map->m_flags & F2FS_MAP_MAPPED) {
1164                         unsigned int ofs = start_pgofs - map->m_lblk;
1165
1166                         f2fs_update_extent_cache_range(&dn,
1167                                 start_pgofs, map->m_pblk + ofs,
1168                                 map->m_len - ofs);
1169                 }
1170         }
1171
1172         f2fs_put_dnode(&dn);
1173
1174         if (create) {
1175                 __do_map_lock(sbi, flag, false);
1176                 f2fs_balance_fs(sbi, dn.node_changed);
1177         }
1178         goto next_dnode;
1179
1180 sync_out:
1181         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1182                 if (map->m_flags & F2FS_MAP_MAPPED) {
1183                         unsigned int ofs = start_pgofs - map->m_lblk;
1184
1185                         f2fs_update_extent_cache_range(&dn,
1186                                 start_pgofs, map->m_pblk + ofs,
1187                                 map->m_len - ofs);
1188                 }
1189                 if (map->m_next_extent)
1190                         *map->m_next_extent = pgofs + 1;
1191         }
1192         f2fs_put_dnode(&dn);
1193 unlock_out:
1194         if (create) {
1195                 __do_map_lock(sbi, flag, false);
1196                 f2fs_balance_fs(sbi, dn.node_changed);
1197         }
1198 out:
1199         trace_f2fs_map_blocks(inode, map, err);
1200         return err;
1201 }
1202
1203 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1204 {
1205         struct f2fs_map_blocks map;
1206         block_t last_lblk;
1207         int err;
1208
1209         if (pos + len > i_size_read(inode))
1210                 return false;
1211
1212         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1213         map.m_next_pgofs = NULL;
1214         map.m_next_extent = NULL;
1215         map.m_seg_type = NO_CHECK_TYPE;
1216         last_lblk = F2FS_BLK_ALIGN(pos + len);
1217
1218         while (map.m_lblk < last_lblk) {
1219                 map.m_len = last_lblk - map.m_lblk;
1220                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1221                 if (err || map.m_len == 0)
1222                         return false;
1223                 map.m_lblk += map.m_len;
1224         }
1225         return true;
1226 }
1227
1228 static int __get_data_block(struct inode *inode, sector_t iblock,
1229                         struct buffer_head *bh, int create, int flag,
1230                         pgoff_t *next_pgofs, int seg_type)
1231 {
1232         struct f2fs_map_blocks map;
1233         int err;
1234
1235         map.m_lblk = iblock;
1236         map.m_len = bh->b_size >> inode->i_blkbits;
1237         map.m_next_pgofs = next_pgofs;
1238         map.m_next_extent = NULL;
1239         map.m_seg_type = seg_type;
1240
1241         err = f2fs_map_blocks(inode, &map, create, flag);
1242         if (!err) {
1243                 map_bh(bh, inode->i_sb, map.m_pblk);
1244                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1245                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1246         }
1247         return err;
1248 }
1249
1250 static int get_data_block(struct inode *inode, sector_t iblock,
1251                         struct buffer_head *bh_result, int create, int flag,
1252                         pgoff_t *next_pgofs)
1253 {
1254         return __get_data_block(inode, iblock, bh_result, create,
1255                                                         flag, next_pgofs,
1256                                                         NO_CHECK_TYPE);
1257 }
1258
1259 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1260                         struct buffer_head *bh_result, int create)
1261 {
1262         return __get_data_block(inode, iblock, bh_result, create,
1263                                                 F2FS_GET_BLOCK_DEFAULT, NULL,
1264                                                 f2fs_rw_hint_to_seg_type(
1265                                                         inode->i_write_hint));
1266 }
1267
1268 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1269                         struct buffer_head *bh_result, int create)
1270 {
1271         /* Block number less than F2FS MAX BLOCKS */
1272         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1273                 return -EFBIG;
1274
1275         return __get_data_block(inode, iblock, bh_result, create,
1276                                                 F2FS_GET_BLOCK_BMAP, NULL,
1277                                                 NO_CHECK_TYPE);
1278 }
1279
1280 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1281 {
1282         return (offset >> inode->i_blkbits);
1283 }
1284
1285 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1286 {
1287         return (blk << inode->i_blkbits);
1288 }
1289
1290 static int f2fs_xattr_fiemap(struct inode *inode,
1291                                 struct fiemap_extent_info *fieinfo)
1292 {
1293         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1294         struct page *page;
1295         struct node_info ni;
1296         __u64 phys = 0, len;
1297         __u32 flags;
1298         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1299         int err = 0;
1300
1301         if (f2fs_has_inline_xattr(inode)) {
1302                 int offset;
1303
1304                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1305                                                 inode->i_ino, false);
1306                 if (!page)
1307                         return -ENOMEM;
1308
1309                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1310                 if (err) {
1311                         f2fs_put_page(page, 1);
1312                         return err;
1313                 }
1314
1315                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1316                 offset = offsetof(struct f2fs_inode, i_addr) +
1317                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1318                                         get_inline_xattr_addrs(inode));
1319
1320                 phys += offset;
1321                 len = inline_xattr_size(inode);
1322
1323                 f2fs_put_page(page, 1);
1324
1325                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1326
1327                 if (!xnid)
1328                         flags |= FIEMAP_EXTENT_LAST;
1329
1330                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1331                 if (err || err == 1)
1332                         return err;
1333         }
1334
1335         if (xnid) {
1336                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1337                 if (!page)
1338                         return -ENOMEM;
1339
1340                 err = f2fs_get_node_info(sbi, xnid, &ni);
1341                 if (err) {
1342                         f2fs_put_page(page, 1);
1343                         return err;
1344                 }
1345
1346                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1347                 len = inode->i_sb->s_blocksize;
1348
1349                 f2fs_put_page(page, 1);
1350
1351                 flags = FIEMAP_EXTENT_LAST;
1352         }
1353
1354         if (phys)
1355                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1356
1357         return (err < 0 ? err : 0);
1358 }
1359
1360 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1361                 u64 start, u64 len)
1362 {
1363         struct buffer_head map_bh;
1364         sector_t start_blk, last_blk;
1365         pgoff_t next_pgofs;
1366         u64 logical = 0, phys = 0, size = 0;
1367         u32 flags = 0;
1368         int ret = 0;
1369
1370         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1371                 ret = f2fs_precache_extents(inode);
1372                 if (ret)
1373                         return ret;
1374         }
1375
1376         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1377         if (ret)
1378                 return ret;
1379
1380         inode_lock(inode);
1381
1382         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1383                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1384                 goto out;
1385         }
1386
1387         if (f2fs_has_inline_data(inode)) {
1388                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1389                 if (ret != -EAGAIN)
1390                         goto out;
1391         }
1392
1393         if (logical_to_blk(inode, len) == 0)
1394                 len = blk_to_logical(inode, 1);
1395
1396         start_blk = logical_to_blk(inode, start);
1397         last_blk = logical_to_blk(inode, start + len - 1);
1398
1399 next:
1400         memset(&map_bh, 0, sizeof(struct buffer_head));
1401         map_bh.b_size = len;
1402
1403         ret = get_data_block(inode, start_blk, &map_bh, 0,
1404                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1405         if (ret)
1406                 goto out;
1407
1408         /* HOLE */
1409         if (!buffer_mapped(&map_bh)) {
1410                 start_blk = next_pgofs;
1411
1412                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1413                                         F2FS_I_SB(inode)->max_file_blocks))
1414                         goto prep_next;
1415
1416                 flags |= FIEMAP_EXTENT_LAST;
1417         }
1418
1419         if (size) {
1420                 if (f2fs_encrypted_inode(inode))
1421                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1422
1423                 ret = fiemap_fill_next_extent(fieinfo, logical,
1424                                 phys, size, flags);
1425         }
1426
1427         if (start_blk > last_blk || ret)
1428                 goto out;
1429
1430         logical = blk_to_logical(inode, start_blk);
1431         phys = blk_to_logical(inode, map_bh.b_blocknr);
1432         size = map_bh.b_size;
1433         flags = 0;
1434         if (buffer_unwritten(&map_bh))
1435                 flags = FIEMAP_EXTENT_UNWRITTEN;
1436
1437         start_blk += logical_to_blk(inode, size);
1438
1439 prep_next:
1440         cond_resched();
1441         if (fatal_signal_pending(current))
1442                 ret = -EINTR;
1443         else
1444                 goto next;
1445 out:
1446         if (ret == 1)
1447                 ret = 0;
1448
1449         inode_unlock(inode);
1450         return ret;
1451 }
1452
1453 /*
1454  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1455  * Major change was from block_size == page_size in f2fs by default.
1456  *
1457  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1458  * this function ever deviates from doing just read-ahead, it should either
1459  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1460  * from read-ahead.
1461  */
1462 static int f2fs_mpage_readpages(struct address_space *mapping,
1463                         struct list_head *pages, struct page *page,
1464                         unsigned nr_pages, bool is_readahead)
1465 {
1466         struct bio *bio = NULL;
1467         sector_t last_block_in_bio = 0;
1468         struct inode *inode = mapping->host;
1469         const unsigned blkbits = inode->i_blkbits;
1470         const unsigned blocksize = 1 << blkbits;
1471         sector_t block_in_file;
1472         sector_t last_block;
1473         sector_t last_block_in_file;
1474         sector_t block_nr;
1475         struct f2fs_map_blocks map;
1476
1477         map.m_pblk = 0;
1478         map.m_lblk = 0;
1479         map.m_len = 0;
1480         map.m_flags = 0;
1481         map.m_next_pgofs = NULL;
1482         map.m_next_extent = NULL;
1483         map.m_seg_type = NO_CHECK_TYPE;
1484
1485         for (; nr_pages; nr_pages--) {
1486                 if (pages) {
1487                         page = list_last_entry(pages, struct page, lru);
1488
1489                         prefetchw(&page->flags);
1490                         list_del(&page->lru);
1491                         if (add_to_page_cache_lru(page, mapping,
1492                                                   page->index,
1493                                                   readahead_gfp_mask(mapping)))
1494                                 goto next_page;
1495                 }
1496
1497                 block_in_file = (sector_t)page->index;
1498                 last_block = block_in_file + nr_pages;
1499                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1500                                                                 blkbits;
1501                 if (last_block > last_block_in_file)
1502                         last_block = last_block_in_file;
1503
1504                 /*
1505                  * Map blocks using the previous result first.
1506                  */
1507                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1508                                 block_in_file > map.m_lblk &&
1509                                 block_in_file < (map.m_lblk + map.m_len))
1510                         goto got_it;
1511
1512                 /*
1513                  * Then do more f2fs_map_blocks() calls until we are
1514                  * done with this page.
1515                  */
1516                 map.m_flags = 0;
1517
1518                 if (block_in_file < last_block) {
1519                         map.m_lblk = block_in_file;
1520                         map.m_len = last_block - block_in_file;
1521
1522                         if (f2fs_map_blocks(inode, &map, 0,
1523                                                 F2FS_GET_BLOCK_DEFAULT))
1524                                 goto set_error_page;
1525                 }
1526 got_it:
1527                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1528                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1529                         SetPageMappedToDisk(page);
1530
1531                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1532                                 SetPageUptodate(page);
1533                                 goto confused;
1534                         }
1535
1536                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1537                                                                 DATA_GENERIC))
1538                                 goto set_error_page;
1539                 } else {
1540                         zero_user_segment(page, 0, PAGE_SIZE);
1541                         if (!PageUptodate(page))
1542                                 SetPageUptodate(page);
1543                         unlock_page(page);
1544                         goto next_page;
1545                 }
1546
1547                 /*
1548                  * This page will go to BIO.  Do we need to send this
1549                  * BIO off first?
1550                  */
1551                 if (bio && (last_block_in_bio != block_nr - 1 ||
1552                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1553 submit_and_realloc:
1554                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1555                         bio = NULL;
1556                 }
1557                 if (bio == NULL) {
1558                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1559                                         is_readahead ? REQ_RAHEAD : 0);
1560                         if (IS_ERR(bio)) {
1561                                 bio = NULL;
1562                                 goto set_error_page;
1563                         }
1564                 }
1565
1566                 /*
1567                  * If the page is under writeback, we need to wait for
1568                  * its completion to see the correct decrypted data.
1569                  */
1570                 f2fs_wait_on_block_writeback(inode, block_nr);
1571
1572                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1573                         goto submit_and_realloc;
1574
1575                 last_block_in_bio = block_nr;
1576                 goto next_page;
1577 set_error_page:
1578                 SetPageError(page);
1579                 zero_user_segment(page, 0, PAGE_SIZE);
1580                 unlock_page(page);
1581                 goto next_page;
1582 confused:
1583                 if (bio) {
1584                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1585                         bio = NULL;
1586                 }
1587                 unlock_page(page);
1588 next_page:
1589                 if (pages)
1590                         put_page(page);
1591         }
1592         BUG_ON(pages && !list_empty(pages));
1593         if (bio)
1594                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1595         return 0;
1596 }
1597
1598 static int f2fs_read_data_page(struct file *file, struct page *page)
1599 {
1600         struct inode *inode = page->mapping->host;
1601         int ret = -EAGAIN;
1602
1603         trace_f2fs_readpage(page, DATA);
1604
1605         /* If the file has inline data, try to read it directly */
1606         if (f2fs_has_inline_data(inode))
1607                 ret = f2fs_read_inline_data(inode, page);
1608         if (ret == -EAGAIN)
1609                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1610         return ret;
1611 }
1612
1613 static int f2fs_read_data_pages(struct file *file,
1614                         struct address_space *mapping,
1615                         struct list_head *pages, unsigned nr_pages)
1616 {
1617         struct inode *inode = mapping->host;
1618         struct page *page = list_last_entry(pages, struct page, lru);
1619
1620         trace_f2fs_readpages(inode, page, nr_pages);
1621
1622         /* If the file has inline data, skip readpages */
1623         if (f2fs_has_inline_data(inode))
1624                 return 0;
1625
1626         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1627 }
1628
1629 static int encrypt_one_page(struct f2fs_io_info *fio)
1630 {
1631         struct inode *inode = fio->page->mapping->host;
1632         struct page *mpage;
1633         gfp_t gfp_flags = GFP_NOFS;
1634
1635         if (!f2fs_encrypted_file(inode))
1636                 return 0;
1637
1638         /* wait for GCed page writeback via META_MAPPING */
1639         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1640
1641 retry_encrypt:
1642         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1643                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1644         if (IS_ERR(fio->encrypted_page)) {
1645                 /* flush pending IOs and wait for a while in the ENOMEM case */
1646                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1647                         f2fs_flush_merged_writes(fio->sbi);
1648                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1649                         gfp_flags |= __GFP_NOFAIL;
1650                         goto retry_encrypt;
1651                 }
1652                 return PTR_ERR(fio->encrypted_page);
1653         }
1654
1655         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1656         if (mpage) {
1657                 if (PageUptodate(mpage))
1658                         memcpy(page_address(mpage),
1659                                 page_address(fio->encrypted_page), PAGE_SIZE);
1660                 f2fs_put_page(mpage, 1);
1661         }
1662         return 0;
1663 }
1664
1665 static inline bool check_inplace_update_policy(struct inode *inode,
1666                                 struct f2fs_io_info *fio)
1667 {
1668         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1669         unsigned int policy = SM_I(sbi)->ipu_policy;
1670
1671         if (policy & (0x1 << F2FS_IPU_FORCE))
1672                 return true;
1673         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1674                 return true;
1675         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1676                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1677                 return true;
1678         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1679                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1680                 return true;
1681
1682         /*
1683          * IPU for rewrite async pages
1684          */
1685         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1686                         fio && fio->op == REQ_OP_WRITE &&
1687                         !(fio->op_flags & REQ_SYNC) &&
1688                         !f2fs_encrypted_inode(inode))
1689                 return true;
1690
1691         /* this is only set during fdatasync */
1692         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1693                         is_inode_flag_set(inode, FI_NEED_IPU))
1694                 return true;
1695
1696         return false;
1697 }
1698
1699 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1700 {
1701         if (f2fs_is_pinned_file(inode))
1702                 return true;
1703
1704         /* if this is cold file, we should overwrite to avoid fragmentation */
1705         if (file_is_cold(inode))
1706                 return true;
1707
1708         return check_inplace_update_policy(inode, fio);
1709 }
1710
1711 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1712 {
1713         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1714
1715         if (test_opt(sbi, LFS))
1716                 return true;
1717         if (S_ISDIR(inode->i_mode))
1718                 return true;
1719         if (f2fs_is_atomic_file(inode))
1720                 return true;
1721         if (fio) {
1722                 if (is_cold_data(fio->page))
1723                         return true;
1724                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1725                         return true;
1726         }
1727         return false;
1728 }
1729
1730 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1731 {
1732         struct inode *inode = fio->page->mapping->host;
1733
1734         if (f2fs_should_update_outplace(inode, fio))
1735                 return false;
1736
1737         return f2fs_should_update_inplace(inode, fio);
1738 }
1739
1740 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1741 {
1742         struct page *page = fio->page;
1743         struct inode *inode = page->mapping->host;
1744         struct dnode_of_data dn;
1745         struct extent_info ei = {0,0,0};
1746         struct node_info ni;
1747         bool ipu_force = false;
1748         int err = 0;
1749
1750         set_new_dnode(&dn, inode, NULL, NULL, 0);
1751         if (need_inplace_update(fio) &&
1752                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1753                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1754
1755                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1756                                                         DATA_GENERIC))
1757                         return -EFAULT;
1758
1759                 ipu_force = true;
1760                 fio->need_lock = LOCK_DONE;
1761                 goto got_it;
1762         }
1763
1764         /* Deadlock due to between page->lock and f2fs_lock_op */
1765         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1766                 return -EAGAIN;
1767
1768         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1769         if (err)
1770                 goto out;
1771
1772         fio->old_blkaddr = dn.data_blkaddr;
1773
1774         /* This page is already truncated */
1775         if (fio->old_blkaddr == NULL_ADDR) {
1776                 ClearPageUptodate(page);
1777                 goto out_writepage;
1778         }
1779 got_it:
1780         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1781                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1782                                                         DATA_GENERIC)) {
1783                 err = -EFAULT;
1784                 goto out_writepage;
1785         }
1786         /*
1787          * If current allocation needs SSR,
1788          * it had better in-place writes for updated data.
1789          */
1790         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1791                                         need_inplace_update(fio))) {
1792                 err = encrypt_one_page(fio);
1793                 if (err)
1794                         goto out_writepage;
1795
1796                 set_page_writeback(page);
1797                 ClearPageError(page);
1798                 f2fs_put_dnode(&dn);
1799                 if (fio->need_lock == LOCK_REQ)
1800                         f2fs_unlock_op(fio->sbi);
1801                 err = f2fs_inplace_write_data(fio);
1802                 trace_f2fs_do_write_data_page(fio->page, IPU);
1803                 set_inode_flag(inode, FI_UPDATE_WRITE);
1804                 return err;
1805         }
1806
1807         if (fio->need_lock == LOCK_RETRY) {
1808                 if (!f2fs_trylock_op(fio->sbi)) {
1809                         err = -EAGAIN;
1810                         goto out_writepage;
1811                 }
1812                 fio->need_lock = LOCK_REQ;
1813         }
1814
1815         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1816         if (err)
1817                 goto out_writepage;
1818
1819         fio->version = ni.version;
1820
1821         err = encrypt_one_page(fio);
1822         if (err)
1823                 goto out_writepage;
1824
1825         set_page_writeback(page);
1826         ClearPageError(page);
1827
1828         /* LFS mode write path */
1829         f2fs_outplace_write_data(&dn, fio);
1830         trace_f2fs_do_write_data_page(page, OPU);
1831         set_inode_flag(inode, FI_APPEND_WRITE);
1832         if (page->index == 0)
1833                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1834 out_writepage:
1835         f2fs_put_dnode(&dn);
1836 out:
1837         if (fio->need_lock == LOCK_REQ)
1838                 f2fs_unlock_op(fio->sbi);
1839         return err;
1840 }
1841
1842 static int __write_data_page(struct page *page, bool *submitted,
1843                                 struct writeback_control *wbc,
1844                                 enum iostat_type io_type)
1845 {
1846         struct inode *inode = page->mapping->host;
1847         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1848         loff_t i_size = i_size_read(inode);
1849         const pgoff_t end_index = ((unsigned long long) i_size)
1850                                                         >> PAGE_SHIFT;
1851         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1852         unsigned offset = 0;
1853         bool need_balance_fs = false;
1854         int err = 0;
1855         struct f2fs_io_info fio = {
1856                 .sbi = sbi,
1857                 .ino = inode->i_ino,
1858                 .type = DATA,
1859                 .op = REQ_OP_WRITE,
1860                 .op_flags = wbc_to_write_flags(wbc),
1861                 .old_blkaddr = NULL_ADDR,
1862                 .page = page,
1863                 .encrypted_page = NULL,
1864                 .submitted = false,
1865                 .need_lock = LOCK_RETRY,
1866                 .io_type = io_type,
1867                 .io_wbc = wbc,
1868         };
1869
1870         trace_f2fs_writepage(page, DATA);
1871
1872         /* we should bypass data pages to proceed the kworkder jobs */
1873         if (unlikely(f2fs_cp_error(sbi))) {
1874                 mapping_set_error(page->mapping, -EIO);
1875                 /*
1876                  * don't drop any dirty dentry pages for keeping lastest
1877                  * directory structure.
1878                  */
1879                 if (S_ISDIR(inode->i_mode))
1880                         goto redirty_out;
1881                 goto out;
1882         }
1883
1884         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1885                 goto redirty_out;
1886
1887         if (page->index < end_index)
1888                 goto write;
1889
1890         /*
1891          * If the offset is out-of-range of file size,
1892          * this page does not have to be written to disk.
1893          */
1894         offset = i_size & (PAGE_SIZE - 1);
1895         if ((page->index >= end_index + 1) || !offset)
1896                 goto out;
1897
1898         zero_user_segment(page, offset, PAGE_SIZE);
1899 write:
1900         if (f2fs_is_drop_cache(inode))
1901                 goto out;
1902         /* we should not write 0'th page having journal header */
1903         if (f2fs_is_volatile_file(inode) && (!page->index ||
1904                         (!wbc->for_reclaim &&
1905                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1906                 goto redirty_out;
1907
1908         /* Dentry blocks are controlled by checkpoint */
1909         if (S_ISDIR(inode->i_mode)) {
1910                 fio.need_lock = LOCK_DONE;
1911                 err = f2fs_do_write_data_page(&fio);
1912                 goto done;
1913         }
1914
1915         if (!wbc->for_reclaim)
1916                 need_balance_fs = true;
1917         else if (has_not_enough_free_secs(sbi, 0, 0))
1918                 goto redirty_out;
1919         else
1920                 set_inode_flag(inode, FI_HOT_DATA);
1921
1922         err = -EAGAIN;
1923         if (f2fs_has_inline_data(inode)) {
1924                 err = f2fs_write_inline_data(inode, page);
1925                 if (!err)
1926                         goto out;
1927         }
1928
1929         if (err == -EAGAIN) {
1930                 err = f2fs_do_write_data_page(&fio);
1931                 if (err == -EAGAIN) {
1932                         fio.need_lock = LOCK_REQ;
1933                         err = f2fs_do_write_data_page(&fio);
1934                 }
1935         }
1936
1937         if (err) {
1938                 file_set_keep_isize(inode);
1939         } else {
1940                 down_write(&F2FS_I(inode)->i_sem);
1941                 if (F2FS_I(inode)->last_disk_size < psize)
1942                         F2FS_I(inode)->last_disk_size = psize;
1943                 up_write(&F2FS_I(inode)->i_sem);
1944         }
1945
1946 done:
1947         if (err && err != -ENOENT)
1948                 goto redirty_out;
1949
1950 out:
1951         inode_dec_dirty_pages(inode);
1952         if (err)
1953                 ClearPageUptodate(page);
1954
1955         if (wbc->for_reclaim) {
1956                 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1957                 clear_inode_flag(inode, FI_HOT_DATA);
1958                 f2fs_remove_dirty_inode(inode);
1959                 submitted = NULL;
1960         }
1961
1962         unlock_page(page);
1963         if (!S_ISDIR(inode->i_mode))
1964                 f2fs_balance_fs(sbi, need_balance_fs);
1965
1966         if (unlikely(f2fs_cp_error(sbi))) {
1967                 f2fs_submit_merged_write(sbi, DATA);
1968                 submitted = NULL;
1969         }
1970
1971         if (submitted)
1972                 *submitted = fio.submitted;
1973
1974         return 0;
1975
1976 redirty_out:
1977         redirty_page_for_writepage(wbc, page);
1978         /*
1979          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
1980          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
1981          * file_write_and_wait_range() will see EIO error, which is critical
1982          * to return value of fsync() followed by atomic_write failure to user.
1983          */
1984         if (!err || wbc->for_reclaim)
1985                 return AOP_WRITEPAGE_ACTIVATE;
1986         unlock_page(page);
1987         return err;
1988 }
1989
1990 static int f2fs_write_data_page(struct page *page,
1991                                         struct writeback_control *wbc)
1992 {
1993         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1994 }
1995
1996 /*
1997  * This function was copied from write_cche_pages from mm/page-writeback.c.
1998  * The major change is making write step of cold data page separately from
1999  * warm/hot data page.
2000  */
2001 static int f2fs_write_cache_pages(struct address_space *mapping,
2002                                         struct writeback_control *wbc,
2003                                         enum iostat_type io_type)
2004 {
2005         int ret = 0;
2006         int done = 0;
2007         struct pagevec pvec;
2008         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2009         int nr_pages;
2010         pgoff_t uninitialized_var(writeback_index);
2011         pgoff_t index;
2012         pgoff_t end;            /* Inclusive */
2013         pgoff_t done_index;
2014         pgoff_t last_idx = ULONG_MAX;
2015         int cycled;
2016         int range_whole = 0;
2017         int tag;
2018
2019         pagevec_init(&pvec);
2020
2021         if (get_dirty_pages(mapping->host) <=
2022                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2023                 set_inode_flag(mapping->host, FI_HOT_DATA);
2024         else
2025                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2026
2027         if (wbc->range_cyclic) {
2028                 writeback_index = mapping->writeback_index; /* prev offset */
2029                 index = writeback_index;
2030                 if (index == 0)
2031                         cycled = 1;
2032                 else
2033                         cycled = 0;
2034                 end = -1;
2035         } else {
2036                 index = wbc->range_start >> PAGE_SHIFT;
2037                 end = wbc->range_end >> PAGE_SHIFT;
2038                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2039                         range_whole = 1;
2040                 cycled = 1; /* ignore range_cyclic tests */
2041         }
2042         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2043                 tag = PAGECACHE_TAG_TOWRITE;
2044         else
2045                 tag = PAGECACHE_TAG_DIRTY;
2046 retry:
2047         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2048                 tag_pages_for_writeback(mapping, index, end);
2049         done_index = index;
2050         while (!done && (index <= end)) {
2051                 int i;
2052
2053                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2054                                 tag);
2055                 if (nr_pages == 0)
2056                         break;
2057
2058                 for (i = 0; i < nr_pages; i++) {
2059                         struct page *page = pvec.pages[i];
2060                         bool submitted = false;
2061
2062                         /* give a priority to WB_SYNC threads */
2063                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2064                                         wbc->sync_mode == WB_SYNC_NONE) {
2065                                 done = 1;
2066                                 break;
2067                         }
2068
2069                         done_index = page->index;
2070 retry_write:
2071                         lock_page(page);
2072
2073                         if (unlikely(page->mapping != mapping)) {
2074 continue_unlock:
2075                                 unlock_page(page);
2076                                 continue;
2077                         }
2078
2079                         if (!PageDirty(page)) {
2080                                 /* someone wrote it for us */
2081                                 goto continue_unlock;
2082                         }
2083
2084                         if (PageWriteback(page)) {
2085                                 if (wbc->sync_mode != WB_SYNC_NONE)
2086                                         f2fs_wait_on_page_writeback(page,
2087                                                                 DATA, true);
2088                                 else
2089                                         goto continue_unlock;
2090                         }
2091
2092                         BUG_ON(PageWriteback(page));
2093                         if (!clear_page_dirty_for_io(page))
2094                                 goto continue_unlock;
2095
2096                         ret = __write_data_page(page, &submitted, wbc, io_type);
2097                         if (unlikely(ret)) {
2098                                 /*
2099                                  * keep nr_to_write, since vfs uses this to
2100                                  * get # of written pages.
2101                                  */
2102                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2103                                         unlock_page(page);
2104                                         ret = 0;
2105                                         continue;
2106                                 } else if (ret == -EAGAIN) {
2107                                         ret = 0;
2108                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2109                                                 cond_resched();
2110                                                 congestion_wait(BLK_RW_ASYNC,
2111                                                                         HZ/50);
2112                                                 goto retry_write;
2113                                         }
2114                                         continue;
2115                                 }
2116                                 done_index = page->index + 1;
2117                                 done = 1;
2118                                 break;
2119                         } else if (submitted) {
2120                                 last_idx = page->index;
2121                         }
2122
2123                         if (--wbc->nr_to_write <= 0 &&
2124                                         wbc->sync_mode == WB_SYNC_NONE) {
2125                                 done = 1;
2126                                 break;
2127                         }
2128                 }
2129                 pagevec_release(&pvec);
2130                 cond_resched();
2131         }
2132
2133         if (!cycled && !done) {
2134                 cycled = 1;
2135                 index = 0;
2136                 end = writeback_index - 1;
2137                 goto retry;
2138         }
2139         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2140                 mapping->writeback_index = done_index;
2141
2142         if (last_idx != ULONG_MAX)
2143                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2144                                                 0, last_idx, DATA);
2145
2146         return ret;
2147 }
2148
2149 static inline bool __should_serialize_io(struct inode *inode,
2150                                         struct writeback_control *wbc)
2151 {
2152         if (!S_ISREG(inode->i_mode))
2153                 return false;
2154         if (wbc->sync_mode != WB_SYNC_ALL)
2155                 return true;
2156         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2157                 return true;
2158         return false;
2159 }
2160
2161 static int __f2fs_write_data_pages(struct address_space *mapping,
2162                                                 struct writeback_control *wbc,
2163                                                 enum iostat_type io_type)
2164 {
2165         struct inode *inode = mapping->host;
2166         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2167         struct blk_plug plug;
2168         int ret;
2169         bool locked = false;
2170
2171         /* deal with chardevs and other special file */
2172         if (!mapping->a_ops->writepage)
2173                 return 0;
2174
2175         /* skip writing if there is no dirty page in this inode */
2176         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2177                 return 0;
2178
2179         /* during POR, we don't need to trigger writepage at all. */
2180         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2181                 goto skip_write;
2182
2183         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2184                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2185                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2186                 goto skip_write;
2187
2188         /* skip writing during file defragment */
2189         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2190                 goto skip_write;
2191
2192         trace_f2fs_writepages(mapping->host, wbc, DATA);
2193
2194         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2195         if (wbc->sync_mode == WB_SYNC_ALL)
2196                 atomic_inc(&sbi->wb_sync_req[DATA]);
2197         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2198                 goto skip_write;
2199
2200         if (__should_serialize_io(inode, wbc)) {
2201                 mutex_lock(&sbi->writepages);
2202                 locked = true;
2203         }
2204
2205         blk_start_plug(&plug);
2206         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2207         blk_finish_plug(&plug);
2208
2209         if (locked)
2210                 mutex_unlock(&sbi->writepages);
2211
2212         if (wbc->sync_mode == WB_SYNC_ALL)
2213                 atomic_dec(&sbi->wb_sync_req[DATA]);
2214         /*
2215          * if some pages were truncated, we cannot guarantee its mapping->host
2216          * to detect pending bios.
2217          */
2218
2219         f2fs_remove_dirty_inode(inode);
2220         return ret;
2221
2222 skip_write:
2223         wbc->pages_skipped += get_dirty_pages(inode);
2224         trace_f2fs_writepages(mapping->host, wbc, DATA);
2225         return 0;
2226 }
2227
2228 static int f2fs_write_data_pages(struct address_space *mapping,
2229                             struct writeback_control *wbc)
2230 {
2231         struct inode *inode = mapping->host;
2232
2233         return __f2fs_write_data_pages(mapping, wbc,
2234                         F2FS_I(inode)->cp_task == current ?
2235                         FS_CP_DATA_IO : FS_DATA_IO);
2236 }
2237
2238 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2239 {
2240         struct inode *inode = mapping->host;
2241         loff_t i_size = i_size_read(inode);
2242
2243         if (to > i_size) {
2244                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2245                 down_write(&F2FS_I(inode)->i_mmap_sem);
2246
2247                 truncate_pagecache(inode, i_size);
2248                 f2fs_truncate_blocks(inode, i_size, true);
2249
2250                 up_write(&F2FS_I(inode)->i_mmap_sem);
2251                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2252         }
2253 }
2254
2255 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2256                         struct page *page, loff_t pos, unsigned len,
2257                         block_t *blk_addr, bool *node_changed)
2258 {
2259         struct inode *inode = page->mapping->host;
2260         pgoff_t index = page->index;
2261         struct dnode_of_data dn;
2262         struct page *ipage;
2263         bool locked = false;
2264         struct extent_info ei = {0,0,0};
2265         int err = 0;
2266
2267         /*
2268          * we already allocated all the blocks, so we don't need to get
2269          * the block addresses when there is no need to fill the page.
2270          */
2271         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2272                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2273                 return 0;
2274
2275         if (f2fs_has_inline_data(inode) ||
2276                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2277                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
2278                 locked = true;
2279         }
2280 restart:
2281         /* check inline_data */
2282         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2283         if (IS_ERR(ipage)) {
2284                 err = PTR_ERR(ipage);
2285                 goto unlock_out;
2286         }
2287
2288         set_new_dnode(&dn, inode, ipage, ipage, 0);
2289
2290         if (f2fs_has_inline_data(inode)) {
2291                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2292                         f2fs_do_read_inline_data(page, ipage);
2293                         set_inode_flag(inode, FI_DATA_EXIST);
2294                         if (inode->i_nlink)
2295                                 set_inline_node(ipage);
2296                 } else {
2297                         err = f2fs_convert_inline_page(&dn, page);
2298                         if (err)
2299                                 goto out;
2300                         if (dn.data_blkaddr == NULL_ADDR)
2301                                 err = f2fs_get_block(&dn, index);
2302                 }
2303         } else if (locked) {
2304                 err = f2fs_get_block(&dn, index);
2305         } else {
2306                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2307                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2308                 } else {
2309                         /* hole case */
2310                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2311                         if (err || dn.data_blkaddr == NULL_ADDR) {
2312                                 f2fs_put_dnode(&dn);
2313                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2314                                                                 true);
2315                                 locked = true;
2316                                 goto restart;
2317                         }
2318                 }
2319         }
2320
2321         /* convert_inline_page can make node_changed */
2322         *blk_addr = dn.data_blkaddr;
2323         *node_changed = dn.node_changed;
2324 out:
2325         f2fs_put_dnode(&dn);
2326 unlock_out:
2327         if (locked)
2328                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
2329         return err;
2330 }
2331
2332 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2333                 loff_t pos, unsigned len, unsigned flags,
2334                 struct page **pagep, void **fsdata)
2335 {
2336         struct inode *inode = mapping->host;
2337         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2338         struct page *page = NULL;
2339         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2340         bool need_balance = false, drop_atomic = false;
2341         block_t blkaddr = NULL_ADDR;
2342         int err = 0;
2343
2344         trace_f2fs_write_begin(inode, pos, len, flags);
2345
2346         if ((f2fs_is_atomic_file(inode) &&
2347                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2348                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2349                 err = -ENOMEM;
2350                 drop_atomic = true;
2351                 goto fail;
2352         }
2353
2354         /*
2355          * We should check this at this moment to avoid deadlock on inode page
2356          * and #0 page. The locking rule for inline_data conversion should be:
2357          * lock_page(page #0) -> lock_page(inode_page)
2358          */
2359         if (index != 0) {
2360                 err = f2fs_convert_inline_inode(inode);
2361                 if (err)
2362                         goto fail;
2363         }
2364 repeat:
2365         /*
2366          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2367          * wait_for_stable_page. Will wait that below with our IO control.
2368          */
2369         page = f2fs_pagecache_get_page(mapping, index,
2370                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2371         if (!page) {
2372                 err = -ENOMEM;
2373                 goto fail;
2374         }
2375
2376         *pagep = page;
2377
2378         err = prepare_write_begin(sbi, page, pos, len,
2379                                         &blkaddr, &need_balance);
2380         if (err)
2381                 goto fail;
2382
2383         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2384                 unlock_page(page);
2385                 f2fs_balance_fs(sbi, true);
2386                 lock_page(page);
2387                 if (page->mapping != mapping) {
2388                         /* The page got truncated from under us */
2389                         f2fs_put_page(page, 1);
2390                         goto repeat;
2391                 }
2392         }
2393
2394         f2fs_wait_on_page_writeback(page, DATA, false);
2395
2396         if (len == PAGE_SIZE || PageUptodate(page))
2397                 return 0;
2398
2399         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2400                 zero_user_segment(page, len, PAGE_SIZE);
2401                 return 0;
2402         }
2403
2404         if (blkaddr == NEW_ADDR) {
2405                 zero_user_segment(page, 0, PAGE_SIZE);
2406                 SetPageUptodate(page);
2407         } else {
2408                 err = f2fs_submit_page_read(inode, page, blkaddr);
2409                 if (err)
2410                         goto fail;
2411
2412                 lock_page(page);
2413                 if (unlikely(page->mapping != mapping)) {
2414                         f2fs_put_page(page, 1);
2415                         goto repeat;
2416                 }
2417                 if (unlikely(!PageUptodate(page))) {
2418                         err = -EIO;
2419                         goto fail;
2420                 }
2421         }
2422         return 0;
2423
2424 fail:
2425         f2fs_put_page(page, 1);
2426         f2fs_write_failed(mapping, pos + len);
2427         if (drop_atomic)
2428                 f2fs_drop_inmem_pages_all(sbi, false);
2429         return err;
2430 }
2431
2432 static int f2fs_write_end(struct file *file,
2433                         struct address_space *mapping,
2434                         loff_t pos, unsigned len, unsigned copied,
2435                         struct page *page, void *fsdata)
2436 {
2437         struct inode *inode = page->mapping->host;
2438
2439         trace_f2fs_write_end(inode, pos, len, copied);
2440
2441         /*
2442          * This should be come from len == PAGE_SIZE, and we expect copied
2443          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2444          * let generic_perform_write() try to copy data again through copied=0.
2445          */
2446         if (!PageUptodate(page)) {
2447                 if (unlikely(copied != len))
2448                         copied = 0;
2449                 else
2450                         SetPageUptodate(page);
2451         }
2452         if (!copied)
2453                 goto unlock_out;
2454
2455         set_page_dirty(page);
2456
2457         if (pos + copied > i_size_read(inode))
2458                 f2fs_i_size_write(inode, pos + copied);
2459 unlock_out:
2460         f2fs_put_page(page, 1);
2461         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2462         return copied;
2463 }
2464
2465 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2466                            loff_t offset)
2467 {
2468         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2469         unsigned blkbits = i_blkbits;
2470         unsigned blocksize_mask = (1 << blkbits) - 1;
2471         unsigned long align = offset | iov_iter_alignment(iter);
2472         struct block_device *bdev = inode->i_sb->s_bdev;
2473
2474         if (align & blocksize_mask) {
2475                 if (bdev)
2476                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2477                 blocksize_mask = (1 << blkbits) - 1;
2478                 if (align & blocksize_mask)
2479                         return -EINVAL;
2480                 return 1;
2481         }
2482         return 0;
2483 }
2484
2485 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2486 {
2487         struct address_space *mapping = iocb->ki_filp->f_mapping;
2488         struct inode *inode = mapping->host;
2489         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2490         size_t count = iov_iter_count(iter);
2491         loff_t offset = iocb->ki_pos;
2492         int rw = iov_iter_rw(iter);
2493         int err;
2494         enum rw_hint hint = iocb->ki_hint;
2495         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2496
2497         err = check_direct_IO(inode, iter, offset);
2498         if (err)
2499                 return err < 0 ? err : 0;
2500
2501         if (f2fs_force_buffered_io(inode, rw))
2502                 return 0;
2503
2504         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2505
2506         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2507                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2508
2509         if (!down_read_trylock(&F2FS_I(inode)->i_gc_rwsem[rw])) {
2510                 if (iocb->ki_flags & IOCB_NOWAIT) {
2511                         iocb->ki_hint = hint;
2512                         err = -EAGAIN;
2513                         goto out;
2514                 }
2515                 down_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2516         }
2517
2518         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2519         up_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2520
2521         if (rw == WRITE) {
2522                 if (whint_mode == WHINT_MODE_OFF)
2523                         iocb->ki_hint = hint;
2524                 if (err > 0) {
2525                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2526                                                                         err);
2527                         set_inode_flag(inode, FI_UPDATE_WRITE);
2528                 } else if (err < 0) {
2529                         f2fs_write_failed(mapping, offset + count);
2530                 }
2531         }
2532
2533 out:
2534         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2535
2536         return err;
2537 }
2538
2539 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2540                                                         unsigned int length)
2541 {
2542         struct inode *inode = page->mapping->host;
2543         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2544
2545         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2546                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2547                 return;
2548
2549         if (PageDirty(page)) {
2550                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2551                         dec_page_count(sbi, F2FS_DIRTY_META);
2552                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2553                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2554                 } else {
2555                         inode_dec_dirty_pages(inode);
2556                         f2fs_remove_dirty_inode(inode);
2557                 }
2558         }
2559
2560         /* This is atomic written page, keep Private */
2561         if (IS_ATOMIC_WRITTEN_PAGE(page))
2562                 return f2fs_drop_inmem_page(inode, page);
2563
2564         set_page_private(page, 0);
2565         ClearPagePrivate(page);
2566 }
2567
2568 int f2fs_release_page(struct page *page, gfp_t wait)
2569 {
2570         /* If this is dirty page, keep PagePrivate */
2571         if (PageDirty(page))
2572                 return 0;
2573
2574         /* This is atomic written page, keep Private */
2575         if (IS_ATOMIC_WRITTEN_PAGE(page))
2576                 return 0;
2577
2578         set_page_private(page, 0);
2579         ClearPagePrivate(page);
2580         return 1;
2581 }
2582
2583 static int f2fs_set_data_page_dirty(struct page *page)
2584 {
2585         struct address_space *mapping = page->mapping;
2586         struct inode *inode = mapping->host;
2587
2588         trace_f2fs_set_page_dirty(page, DATA);
2589
2590         if (!PageUptodate(page))
2591                 SetPageUptodate(page);
2592
2593         /* don't remain PG_checked flag which was set during GC */
2594         if (is_cold_data(page))
2595                 clear_cold_data(page);
2596
2597         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2598                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2599                         f2fs_register_inmem_page(inode, page);
2600                         return 1;
2601                 }
2602                 /*
2603                  * Previously, this page has been registered, we just
2604                  * return here.
2605                  */
2606                 return 0;
2607         }
2608
2609         if (!PageDirty(page)) {
2610                 __set_page_dirty_nobuffers(page);
2611                 f2fs_update_dirty_page(inode, page);
2612                 return 1;
2613         }
2614         return 0;
2615 }
2616
2617 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2618 {
2619         struct inode *inode = mapping->host;
2620
2621         if (f2fs_has_inline_data(inode))
2622                 return 0;
2623
2624         /* make sure allocating whole blocks */
2625         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2626                 filemap_write_and_wait(mapping);
2627
2628         return generic_block_bmap(mapping, block, get_data_block_bmap);
2629 }
2630
2631 #ifdef CONFIG_MIGRATION
2632 #include <linux/migrate.h>
2633
2634 int f2fs_migrate_page(struct address_space *mapping,
2635                 struct page *newpage, struct page *page, enum migrate_mode mode)
2636 {
2637         int rc, extra_count;
2638         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2639         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2640
2641         BUG_ON(PageWriteback(page));
2642
2643         /* migrating an atomic written page is safe with the inmem_lock hold */
2644         if (atomic_written) {
2645                 if (mode != MIGRATE_SYNC)
2646                         return -EBUSY;
2647                 if (!mutex_trylock(&fi->inmem_lock))
2648                         return -EAGAIN;
2649         }
2650
2651         /*
2652          * A reference is expected if PagePrivate set when move mapping,
2653          * however F2FS breaks this for maintaining dirty page counts when
2654          * truncating pages. So here adjusting the 'extra_count' make it work.
2655          */
2656         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2657         rc = migrate_page_move_mapping(mapping, newpage,
2658                                 page, NULL, mode, extra_count);
2659         if (rc != MIGRATEPAGE_SUCCESS) {
2660                 if (atomic_written)
2661                         mutex_unlock(&fi->inmem_lock);
2662                 return rc;
2663         }
2664
2665         if (atomic_written) {
2666                 struct inmem_pages *cur;
2667                 list_for_each_entry(cur, &fi->inmem_pages, list)
2668                         if (cur->page == page) {
2669                                 cur->page = newpage;
2670                                 break;
2671                         }
2672                 mutex_unlock(&fi->inmem_lock);
2673                 put_page(page);
2674                 get_page(newpage);
2675         }
2676
2677         if (PagePrivate(page))
2678                 SetPagePrivate(newpage);
2679         set_page_private(newpage, page_private(page));
2680
2681         if (mode != MIGRATE_SYNC_NO_COPY)
2682                 migrate_page_copy(newpage, page);
2683         else
2684                 migrate_page_states(newpage, page);
2685
2686         return MIGRATEPAGE_SUCCESS;
2687 }
2688 #endif
2689
2690 const struct address_space_operations f2fs_dblock_aops = {
2691         .readpage       = f2fs_read_data_page,
2692         .readpages      = f2fs_read_data_pages,
2693         .writepage      = f2fs_write_data_page,
2694         .writepages     = f2fs_write_data_pages,
2695         .write_begin    = f2fs_write_begin,
2696         .write_end      = f2fs_write_end,
2697         .set_page_dirty = f2fs_set_data_page_dirty,
2698         .invalidatepage = f2fs_invalidate_page,
2699         .releasepage    = f2fs_release_page,
2700         .direct_IO      = f2fs_direct_IO,
2701         .bmap           = f2fs_bmap,
2702 #ifdef CONFIG_MIGRATION
2703         .migratepage    = f2fs_migrate_page,
2704 #endif
2705 };
2706
2707 void f2fs_clear_radix_tree_dirty_tag(struct page *page)
2708 {
2709         struct address_space *mapping = page_mapping(page);
2710         unsigned long flags;
2711
2712         xa_lock_irqsave(&mapping->i_pages, flags);
2713         radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2714                                                 PAGECACHE_TAG_DIRTY);
2715         xa_unlock_irqrestore(&mapping->i_pages, flags);
2716 }
2717
2718 int __init f2fs_init_post_read_processing(void)
2719 {
2720         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2721         if (!bio_post_read_ctx_cache)
2722                 goto fail;
2723         bio_post_read_ctx_pool =
2724                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2725                                          bio_post_read_ctx_cache);
2726         if (!bio_post_read_ctx_pool)
2727                 goto fail_free_cache;
2728         return 0;
2729
2730 fail_free_cache:
2731         kmem_cache_destroy(bio_post_read_ctx_cache);
2732 fail:
2733         return -ENOMEM;
2734 }
2735
2736 void __exit f2fs_destroy_post_read_processing(void)
2737 {
2738         mempool_destroy(bio_post_read_ctx_pool);
2739         kmem_cache_destroy(bio_post_read_ctx_cache);
2740 }