fs/f2fs/data.c

   1 /*
   2  * fs/f2fs/data.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/mpage.h>
  15 #include <linux/writeback.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21 #include <linux/uio.h>
  22 #include <linux/cleancache.h>
  23
  24 #include "f2fs.h"
  25 #include "node.h"
  26 #include "segment.h"
  27 #include "trace.h"
  28 #include <trace/events/f2fs.h>
  29
  30 static void f2fs_read_end_io(struct bio *bio)
  31 {
  32         struct bio_vec *bvec;
  33         int i;
  34
  35         if (f2fs_bio_encrypted(bio)) {
  36                 if (bio->bi_error) {
  37                         fscrypt_release_ctx(bio->bi_private);
  38                 } else {
  39                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  40                         return;
  41                 }
  42         }
  43
  44         bio_for_each_segment_all(bvec, bio, i) {
  45                 struct page *page = bvec->bv_page;
  46
  47                 if (!bio->bi_error) {
  48                         SetPageUptodate(page);
  49                 } else {
  50                         ClearPageUptodate(page);
  51                         SetPageError(page);
  52                 }
  53                 unlock_page(page);
  54         }
  55         bio_put(bio);
  56 }
  57
  58 static void f2fs_write_end_io(struct bio *bio)
  59 {
  60         struct f2fs_sb_info *sbi = bio->bi_private;
  61         struct bio_vec *bvec;
  62         int i;
  63
  64         bio_for_each_segment_all(bvec, bio, i) {
  65                 struct page *page = bvec->bv_page;
  66
  67                 fscrypt_pullback_bio_page(&page, true);
  68
  69                 if (unlikely(bio->bi_error)) {
  70                         set_bit(AS_EIO, &page->mapping->flags);
  71                         f2fs_stop_checkpoint(sbi);
  72                 }
  73                 end_page_writeback(page);
  74                 dec_page_count(sbi, F2FS_WRITEBACK);
  75         }
  76
  77         if (!get_pages(sbi, F2FS_WRITEBACK) && wq_has_sleeper(&sbi->cp_wait))
  78                 wake_up(&sbi->cp_wait);
  79
  80         bio_put(bio);
  81 }
  82
  83 /*
  84  * Low-level block read/write IO operations.
  85  */
  86 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  87                                 int npages, bool is_read)
  88 {
  89         struct bio *bio;
  90
  91         bio = f2fs_bio_alloc(npages);
  92
  93         bio->bi_bdev = sbi->sb->s_bdev;
  94         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  95         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  96         bio->bi_private = is_read ? NULL : sbi;
  97
  98         return bio;
  99 }
 100
 101 static void __submit_merged_bio(struct f2fs_bio_info *io)
 102 {
 103         struct f2fs_io_info *fio = &io->fio;
 104
 105         if (!io->bio)
 106                 return;
 107
 108         if (is_read_io(fio->rw))
 109                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 110         else
 111                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 112
 113         submit_bio(fio->rw, io->bio);
 114         io->bio = NULL;
 115 }
 116
 117 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 118                                                 struct page *page, nid_t ino)
 119 {
 120         struct bio_vec *bvec;
 121         struct page *target;
 122         int i;
 123
 124         if (!io->bio)
 125                 return false;
 126
 127         if (!inode && !page && !ino)
 128                 return true;
 129
 130         bio_for_each_segment_all(bvec, io->bio, i) {
 131
 132                 if (bvec->bv_page->mapping)
 133                         target = bvec->bv_page;
 134                 else
 135                         target = fscrypt_control_page(bvec->bv_page);
 136
 137                 if (inode && inode == target->mapping->host)
 138                         return true;
 139                 if (page && page == target)
 140                         return true;
 141                 if (ino && ino == ino_of_node(target))
 142                         return true;
 143         }
 144
 145         return false;
 146 }
 147
 148 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 149                                                 struct page *page, nid_t ino,
 150                                                 enum page_type type)
 151 {
 152         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 153         struct f2fs_bio_info *io = &sbi->write_io[btype];
 154         bool ret;
 155
 156         down_read(&io->io_rwsem);
 157         ret = __has_merged_page(io, inode, page, ino);
 158         up_read(&io->io_rwsem);
 159         return ret;
 160 }
 161
 162 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 163                                 struct inode *inode, struct page *page,
 164                                 nid_t ino, enum page_type type, int rw)
 165 {
 166         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 167         struct f2fs_bio_info *io;
 168
 169         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 170
 171         down_write(&io->io_rwsem);
 172
 173         if (!__has_merged_page(io, inode, page, ino))
 174                 goto out;
 175
 176         /* change META to META_FLUSH in the checkpoint procedure */
 177         if (type >= META_FLUSH) {
 178                 io->fio.type = META_FLUSH;
 179                 if (test_opt(sbi, NOBARRIER))
 180                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 181                 else
 182                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 183         }
 184         __submit_merged_bio(io);
 185 out:
 186         up_write(&io->io_rwsem);
 187 }
 188
 189 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 190                                                                         int rw)
 191 {
 192         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 193 }
 194
 195 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 196                                 struct inode *inode, struct page *page,
 197                                 nid_t ino, enum page_type type, int rw)
 198 {
 199         if (has_merged_page(sbi, inode, page, ino, type))
 200                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 201 }
 202
 203 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 204 {
 205         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 206         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 207         f2fs_submit_merged_bio(sbi, META, WRITE);
 208 }
 209
 210 /*
 211  * Fill the locked page with data located in the block address.
 212  * Return unlocked page.
 213  */
 214 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 215 {
 216         struct bio *bio;
 217         struct page *page = fio->encrypted_page ?
 218                         fio->encrypted_page : fio->page;
 219
 220         trace_f2fs_submit_page_bio(page, fio);
 221         f2fs_trace_ios(fio, 0);
 222
 223         /* Allocate a new bio */
 224         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
 225
 226         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 227                 bio_put(bio);
 228                 return -EFAULT;
 229         }
 230
 231         submit_bio(fio->rw, bio);
 232         return 0;
 233 }
 234
 235 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 236 {
 237         struct f2fs_sb_info *sbi = fio->sbi;
 238         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 239         struct f2fs_bio_info *io;
 240         bool is_read = is_read_io(fio->rw);
 241         struct page *bio_page;
 242
 243         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 244
 245         if (fio->old_blkaddr != NEW_ADDR)
 246                 verify_block_addr(sbi, fio->old_blkaddr);
 247         verify_block_addr(sbi, fio->new_blkaddr);
 248
 249         down_write(&io->io_rwsem);
 250
 251         if (!is_read)
 252                 inc_page_count(sbi, F2FS_WRITEBACK);
 253
 254         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 255                                                 io->fio.rw != fio->rw))
 256                 __submit_merged_bio(io);
 257 alloc_new:
 258         if (io->bio == NULL) {
 259                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 260
 261                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 262                                                 bio_blocks, is_read);
 263                 io->fio = *fio;
 264         }
 265
 266         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 267
 268         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 269                                                         PAGE_SIZE) {
 270                 __submit_merged_bio(io);
 271                 goto alloc_new;
 272         }
 273
 274         io->last_block_in_bio = fio->new_blkaddr;
 275         f2fs_trace_ios(fio, 0);
 276
 277         up_write(&io->io_rwsem);
 278         trace_f2fs_submit_page_mbio(fio->page, fio);
 279 }
 280
 281 /*
 282  * Lock ordering for the change of data block address:
 283  * ->data_page
 284  *  ->node_page
 285  *    update block addresses in the node page
 286  */
 287 void set_data_blkaddr(struct dnode_of_data *dn)
 288 {
 289         struct f2fs_node *rn;
 290         __le32 *addr_array;
 291         struct page *node_page = dn->node_page;
 292         unsigned int ofs_in_node = dn->ofs_in_node;
 293
 294         f2fs_wait_on_page_writeback(node_page, NODE, true);
 295
 296         rn = F2FS_NODE(node_page);
 297
 298         /* Get physical address of data block */
 299         addr_array = blkaddr_in_node(rn);
 300         addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 301         if (set_page_dirty(node_page))
 302                 dn->node_changed = true;
 303 }
 304
 305 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 306 {
 307         dn->data_blkaddr = blkaddr;
 308         set_data_blkaddr(dn);
 309         f2fs_update_extent_cache(dn);
 310 }
 311
 312 int reserve_new_block(struct dnode_of_data *dn)
 313 {
 314         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 315
 316         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 317                 return -EPERM;
 318         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 319                 return -ENOSPC;
 320
 321         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 322
 323         dn->data_blkaddr = NEW_ADDR;
 324         set_data_blkaddr(dn);
 325         mark_inode_dirty(dn->inode);
 326         sync_inode_page(dn);
 327         return 0;
 328 }
 329
 330 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 331 {
 332         bool need_put = dn->inode_page ? false : true;
 333         int err;
 334
 335         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 336         if (err)
 337                 return err;
 338
 339         if (dn->data_blkaddr == NULL_ADDR)
 340                 err = reserve_new_block(dn);
 341         if (err || need_put)
 342                 f2fs_put_dnode(dn);
 343         return err;
 344 }
 345
 346 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 347 {
 348         struct extent_info ei;
 349         struct inode *inode = dn->inode;
 350
 351         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 352                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 353                 return 0;
 354         }
 355
 356         return f2fs_reserve_block(dn, index);
 357 }
 358
 359 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 360                                                 int rw, bool for_write)
 361 {
 362         struct address_space *mapping = inode->i_mapping;
 363         struct dnode_of_data dn;
 364         struct page *page;
 365         struct extent_info ei;
 366         int err;
 367         struct f2fs_io_info fio = {
 368                 .sbi = F2FS_I_SB(inode),
 369                 .type = DATA,
 370                 .rw = rw,
 371                 .encrypted_page = NULL,
 372         };
 373
 374         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 375                 return read_mapping_page(mapping, index, NULL);
 376
 377         page = f2fs_grab_cache_page(mapping, index, for_write);
 378         if (!page)
 379                 return ERR_PTR(-ENOMEM);
 380
 381         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 382                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 383                 goto got_it;
 384         }
 385
 386         set_new_dnode(&dn, inode, NULL, NULL, 0);
 387         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 388         if (err)
 389                 goto put_err;
 390         f2fs_put_dnode(&dn);
 391
 392         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 393                 err = -ENOENT;
 394                 goto put_err;
 395         }
 396 got_it:
 397         if (PageUptodate(page)) {
 398                 unlock_page(page);
 399                 return page;
 400         }
 401
 402         /*
 403          * A new dentry page is allocated but not able to be written, since its
 404          * new inode page couldn't be allocated due to -ENOSPC.
 405          * In such the case, its blkaddr can be remained as NEW_ADDR.
 406          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 407          */
 408         if (dn.data_blkaddr == NEW_ADDR) {
 409                 zero_user_segment(page, 0, PAGE_SIZE);
 410                 SetPageUptodate(page);
 411                 unlock_page(page);
 412                 return page;
 413         }
 414
 415         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 416         fio.page = page;
 417         err = f2fs_submit_page_bio(&fio);
 418         if (err)
 419                 goto put_err;
 420         return page;
 421
 422 put_err:
 423         f2fs_put_page(page, 1);
 424         return ERR_PTR(err);
 425 }
 426
 427 struct page *find_data_page(struct inode *inode, pgoff_t index)
 428 {
 429         struct address_space *mapping = inode->i_mapping;
 430         struct page *page;
 431
 432         page = find_get_page(mapping, index);
 433         if (page && PageUptodate(page))
 434                 return page;
 435         f2fs_put_page(page, 0);
 436
 437         page = get_read_data_page(inode, index, READ_SYNC, false);
 438         if (IS_ERR(page))
 439                 return page;
 440
 441         if (PageUptodate(page))
 442                 return page;
 443
 444         wait_on_page_locked(page);
 445         if (unlikely(!PageUptodate(page))) {
 446                 f2fs_put_page(page, 0);
 447                 return ERR_PTR(-EIO);
 448         }
 449         return page;
 450 }
 451
 452 /*
 453  * If it tries to access a hole, return an error.
 454  * Because, the callers, functions in dir.c and GC, should be able to know
 455  * whether this page exists or not.
 456  */
 457 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 458                                                         bool for_write)
 459 {
 460         struct address_space *mapping = inode->i_mapping;
 461         struct page *page;
 462 repeat:
 463         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 464         if (IS_ERR(page))
 465                 return page;
 466
 467         /* wait for read completion */
 468         lock_page(page);
 469         if (unlikely(!PageUptodate(page))) {
 470                 f2fs_put_page(page, 1);
 471                 return ERR_PTR(-EIO);
 472         }
 473         if (unlikely(page->mapping != mapping)) {
 474                 f2fs_put_page(page, 1);
 475                 goto repeat;
 476         }
 477         return page;
 478 }
 479
 480 /*
 481  * Caller ensures that this data page is never allocated.
 482  * A new zero-filled data page is allocated in the page cache.
 483  *
 484  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 485  * f2fs_unlock_op().
 486  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 487  * ipage should be released by this function.
 488  */
 489 struct page *get_new_data_page(struct inode *inode,
 490                 struct page *ipage, pgoff_t index, bool new_i_size)
 491 {
 492         struct address_space *mapping = inode->i_mapping;
 493         struct page *page;
 494         struct dnode_of_data dn;
 495         int err;
 496
 497         page = f2fs_grab_cache_page(mapping, index, true);
 498         if (!page) {
 499                 /*
 500                  * before exiting, we should make sure ipage will be released
 501                  * if any error occur.
 502                  */
 503                 f2fs_put_page(ipage, 1);
 504                 return ERR_PTR(-ENOMEM);
 505         }
 506
 507         set_new_dnode(&dn, inode, ipage, NULL, 0);
 508         err = f2fs_reserve_block(&dn, index);
 509         if (err) {
 510                 f2fs_put_page(page, 1);
 511                 return ERR_PTR(err);
 512         }
 513         if (!ipage)
 514                 f2fs_put_dnode(&dn);
 515
 516         if (PageUptodate(page))
 517                 goto got_it;
 518
 519         if (dn.data_blkaddr == NEW_ADDR) {
 520                 zero_user_segment(page, 0, PAGE_SIZE);
 521                 SetPageUptodate(page);
 522         } else {
 523                 f2fs_put_page(page, 1);
 524
 525                 /* if ipage exists, blkaddr should be NEW_ADDR */
 526                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 527                 page = get_lock_data_page(inode, index, true);
 528                 if (IS_ERR(page))
 529                         return page;
 530         }
 531 got_it:
 532         if (new_i_size && i_size_read(inode) <
 533                                 ((loff_t)(index + 1) << PAGE_SHIFT)) {
 534                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 535                 /* Only the directory inode sets new_i_size */
 536                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 537         }
 538         return page;
 539 }
 540
 541 static int __allocate_data_block(struct dnode_of_data *dn)
 542 {
 543         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 544         struct f2fs_summary sum;
 545         struct node_info ni;
 546         int seg = CURSEG_WARM_DATA;
 547         pgoff_t fofs;
 548
 549         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 550                 return -EPERM;
 551
 552         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 553         if (dn->data_blkaddr == NEW_ADDR)
 554                 goto alloc;
 555
 556         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 557                 return -ENOSPC;
 558
 559 alloc:
 560         get_node_info(sbi, dn->nid, &ni);
 561         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 562
 563         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 564                 seg = CURSEG_DIRECT_IO;
 565
 566         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 567                                                                 &sum, seg);
 568         set_data_blkaddr(dn);
 569
 570         /* update i_size */
 571         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 572                                                         dn->ofs_in_node;
 573         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 574                 i_size_write(dn->inode,
 575                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 576         return 0;
 577 }
 578
 579 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 580 {
 581         struct inode *inode = file_inode(iocb->ki_filp);
 582         struct f2fs_map_blocks map;
 583         ssize_t ret = 0;
 584
 585         map.m_lblk = F2FS_BYTES_TO_BLK(iocb->ki_pos);
 586         map.m_len = F2FS_BLK_ALIGN(iov_iter_count(from));
 587         map.m_next_pgofs = NULL;
 588
 589         if (f2fs_encrypted_inode(inode))
 590                 return 0;
 591
 592         if (iocb->ki_flags & IOCB_DIRECT) {
 593                 ret = f2fs_convert_inline_inode(inode);
 594                 if (ret)
 595                         return ret;
 596                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 597         }
 598         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 599                 ret = f2fs_convert_inline_inode(inode);
 600                 if (ret)
 601                         return ret;
 602         }
 603         if (!f2fs_has_inline_data(inode))
 604                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 605         return ret;
 606 }
 607
 608 /*
 609  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 610  * f2fs_map_blocks structure.
 611  * If original data blocks are allocated, then give them to blockdev.
 612  * Otherwise,
 613  *     a. preallocate requested block addresses
 614  *     b. do not use extent cache for better performance
 615  *     c. give the block addresses to blockdev
 616  */
 617 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 618                                                 int create, int flag)
 619 {
 620         unsigned int maxblocks = map->m_len;
 621         struct dnode_of_data dn;
 622         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 623         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 624         pgoff_t pgofs, end_offset;
 625         int err = 0, ofs = 1;
 626         struct extent_info ei;
 627         bool allocated = false;
 628         block_t blkaddr;
 629
 630         map->m_len = 0;
 631         map->m_flags = 0;
 632
 633         /* it only supports block size == page size */
 634         pgofs = (pgoff_t)map->m_lblk;
 635
 636         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 637                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 638                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 639                 map->m_flags = F2FS_MAP_MAPPED;
 640                 goto out;
 641         }
 642
 643 next_dnode:
 644         if (create)
 645                 f2fs_lock_op(sbi);
 646
 647         /* When reading holes, we need its node page */
 648         set_new_dnode(&dn, inode, NULL, NULL, 0);
 649         err = get_dnode_of_data(&dn, pgofs, mode);
 650         if (err) {
 651                 if (err == -ENOENT) {
 652                         err = 0;
 653                         if (map->m_next_pgofs)
 654                                 *map->m_next_pgofs =
 655                                         get_next_page_offset(&dn, pgofs);
 656                 }
 657                 goto unlock_out;
 658         }
 659
 660         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 661
 662 next_block:
 663         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 664
 665         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 666                 if (create) {
 667                         if (unlikely(f2fs_cp_error(sbi))) {
 668                                 err = -EIO;
 669                                 goto sync_out;
 670                         }
 671                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 672                                 if (blkaddr == NULL_ADDR)
 673                                         err = reserve_new_block(&dn);
 674                         } else {
 675                                 err = __allocate_data_block(&dn);
 676                         }
 677                         if (err)
 678                                 goto sync_out;
 679                         allocated = true;
 680                         map->m_flags = F2FS_MAP_NEW;
 681                         blkaddr = dn.data_blkaddr;
 682                 } else {
 683                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 684                                                 blkaddr == NULL_ADDR) {
 685                                 if (map->m_next_pgofs)
 686                                         *map->m_next_pgofs = pgofs + 1;
 687                         }
 688                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 689                                                 blkaddr != NEW_ADDR) {
 690                                 if (flag == F2FS_GET_BLOCK_BMAP)
 691                                         err = -ENOENT;
 692                                 goto sync_out;
 693                         }
 694                 }
 695         }
 696
 697         if (map->m_len == 0) {
 698                 /* preallocated unwritten block should be mapped for fiemap. */
 699                 if (blkaddr == NEW_ADDR)
 700                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 701                 map->m_flags |= F2FS_MAP_MAPPED;
 702
 703                 map->m_pblk = blkaddr;
 704                 map->m_len = 1;
 705         } else if ((map->m_pblk != NEW_ADDR &&
 706                         blkaddr == (map->m_pblk + ofs)) ||
 707                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 708                         flag == F2FS_GET_BLOCK_PRE_DIO ||
 709                         flag == F2FS_GET_BLOCK_PRE_AIO) {
 710                 ofs++;
 711                 map->m_len++;
 712         } else {
 713                 goto sync_out;
 714         }
 715
 716         dn.ofs_in_node++;
 717         pgofs++;
 718
 719         if (map->m_len < maxblocks) {
 720                 if (dn.ofs_in_node < end_offset)
 721                         goto next_block;
 722
 723                 if (allocated)
 724                         sync_inode_page(&dn);
 725                 f2fs_put_dnode(&dn);
 726
 727                 if (create) {
 728                         f2fs_unlock_op(sbi);
 729                         f2fs_balance_fs(sbi, allocated);
 730                 }
 731                 allocated = false;
 732                 goto next_dnode;
 733         }
 734
 735 sync_out:
 736         if (allocated)
 737                 sync_inode_page(&dn);
 738         f2fs_put_dnode(&dn);
 739 unlock_out:
 740         if (create) {
 741                 f2fs_unlock_op(sbi);
 742                 f2fs_balance_fs(sbi, allocated);
 743         }
 744 out:
 745         trace_f2fs_map_blocks(inode, map, err);
 746         return err;
 747 }
 748
 749 static int __get_data_block(struct inode *inode, sector_t iblock,
 750                         struct buffer_head *bh, int create, int flag,
 751                         pgoff_t *next_pgofs)
 752 {
 753         struct f2fs_map_blocks map;
 754         int ret;
 755
 756         map.m_lblk = iblock;
 757         map.m_len = bh->b_size >> inode->i_blkbits;
 758         map.m_next_pgofs = next_pgofs;
 759
 760         ret = f2fs_map_blocks(inode, &map, create, flag);
 761         if (!ret) {
 762                 map_bh(bh, inode->i_sb, map.m_pblk);
 763                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 764                 bh->b_size = map.m_len << inode->i_blkbits;
 765         }
 766         return ret;
 767 }
 768
 769 static int get_data_block(struct inode *inode, sector_t iblock,
 770                         struct buffer_head *bh_result, int create, int flag,
 771                         pgoff_t *next_pgofs)
 772 {
 773         return __get_data_block(inode, iblock, bh_result, create,
 774                                                         flag, next_pgofs);
 775 }
 776
 777 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 778                         struct buffer_head *bh_result, int create)
 779 {
 780         return __get_data_block(inode, iblock, bh_result, create,
 781                                                 F2FS_GET_BLOCK_DIO, NULL);
 782 }
 783
 784 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 785                         struct buffer_head *bh_result, int create)
 786 {
 787         /* Block number less than F2FS MAX BLOCKS */
 788         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 789                 return -EFBIG;
 790
 791         return __get_data_block(inode, iblock, bh_result, create,
 792                                                 F2FS_GET_BLOCK_BMAP, NULL);
 793 }
 794
 795 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 796 {
 797         return (offset >> inode->i_blkbits);
 798 }
 799
 800 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 801 {
 802         return (blk << inode->i_blkbits);
 803 }
 804
 805 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 806                 u64 start, u64 len)
 807 {
 808         struct buffer_head map_bh;
 809         sector_t start_blk, last_blk;
 810         pgoff_t next_pgofs;
 811         loff_t isize;
 812         u64 logical = 0, phys = 0, size = 0;
 813         u32 flags = 0;
 814         int ret = 0;
 815
 816         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 817         if (ret)
 818                 return ret;
 819
 820         if (f2fs_has_inline_data(inode)) {
 821                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 822                 if (ret != -EAGAIN)
 823                         return ret;
 824         }
 825
 826         inode_lock(inode);
 827
 828         isize = i_size_read(inode);
 829         if (start >= isize)
 830                 goto out;
 831
 832         if (start + len > isize)
 833                 len = isize - start;
 834
 835         if (logical_to_blk(inode, len) == 0)
 836                 len = blk_to_logical(inode, 1);
 837
 838         start_blk = logical_to_blk(inode, start);
 839         last_blk = logical_to_blk(inode, start + len - 1);
 840
 841 next:
 842         memset(&map_bh, 0, sizeof(struct buffer_head));
 843         map_bh.b_size = len;
 844
 845         ret = get_data_block(inode, start_blk, &map_bh, 0,
 846                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 847         if (ret)
 848                 goto out;
 849
 850         /* HOLE */
 851         if (!buffer_mapped(&map_bh)) {
 852                 start_blk = next_pgofs;
 853                 /* Go through holes util pass the EOF */
 854                 if (blk_to_logical(inode, start_blk) < isize)
 855                         goto prep_next;
 856                 /* Found a hole beyond isize means no more extents.
 857                  * Note that the premise is that filesystems don't
 858                  * punch holes beyond isize and keep size unchanged.
 859                  */
 860                 flags |= FIEMAP_EXTENT_LAST;
 861         }
 862
 863         if (size) {
 864                 if (f2fs_encrypted_inode(inode))
 865                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 866
 867                 ret = fiemap_fill_next_extent(fieinfo, logical,
 868                                 phys, size, flags);
 869         }
 870
 871         if (start_blk > last_blk || ret)
 872                 goto out;
 873
 874         logical = blk_to_logical(inode, start_blk);
 875         phys = blk_to_logical(inode, map_bh.b_blocknr);
 876         size = map_bh.b_size;
 877         flags = 0;
 878         if (buffer_unwritten(&map_bh))
 879                 flags = FIEMAP_EXTENT_UNWRITTEN;
 880
 881         start_blk += logical_to_blk(inode, size);
 882
 883 prep_next:
 884         cond_resched();
 885         if (fatal_signal_pending(current))
 886                 ret = -EINTR;
 887         else
 888                 goto next;
 889 out:
 890         if (ret == 1)
 891                 ret = 0;
 892
 893         inode_unlock(inode);
 894         return ret;
 895 }
 896
 897 /*
 898  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 899  * Major change was from block_size == page_size in f2fs by default.
 900  */
 901 static int f2fs_mpage_readpages(struct address_space *mapping,
 902                         struct list_head *pages, struct page *page,
 903                         unsigned nr_pages)
 904 {
 905         struct bio *bio = NULL;
 906         unsigned page_idx;
 907         sector_t last_block_in_bio = 0;
 908         struct inode *inode = mapping->host;
 909         const unsigned blkbits = inode->i_blkbits;
 910         const unsigned blocksize = 1 << blkbits;
 911         sector_t block_in_file;
 912         sector_t last_block;
 913         sector_t last_block_in_file;
 914         sector_t block_nr;
 915         struct block_device *bdev = inode->i_sb->s_bdev;
 916         struct f2fs_map_blocks map;
 917
 918         map.m_pblk = 0;
 919         map.m_lblk = 0;
 920         map.m_len = 0;
 921         map.m_flags = 0;
 922         map.m_next_pgofs = NULL;
 923
 924         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 925
 926                 prefetchw(&page->flags);
 927                 if (pages) {
 928                         page = list_entry(pages->prev, struct page, lru);
 929                         list_del(&page->lru);
 930                         if (add_to_page_cache_lru(page, mapping,
 931                                                   page->index, GFP_KERNEL))
 932                                 goto next_page;
 933                 }
 934
 935                 block_in_file = (sector_t)page->index;
 936                 last_block = block_in_file + nr_pages;
 937                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 938                                                                 blkbits;
 939                 if (last_block > last_block_in_file)
 940                         last_block = last_block_in_file;
 941
 942                 /*
 943                  * Map blocks using the previous result first.
 944                  */
 945                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 946                                 block_in_file > map.m_lblk &&
 947                                 block_in_file < (map.m_lblk + map.m_len))
 948                         goto got_it;
 949
 950                 /*
 951                  * Then do more f2fs_map_blocks() calls until we are
 952                  * done with this page.
 953                  */
 954                 map.m_flags = 0;
 955
 956                 if (block_in_file < last_block) {
 957                         map.m_lblk = block_in_file;
 958                         map.m_len = last_block - block_in_file;
 959
 960                         if (f2fs_map_blocks(inode, &map, 0,
 961                                                 F2FS_GET_BLOCK_READ))
 962                                 goto set_error_page;
 963                 }
 964 got_it:
 965                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 966                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 967                         SetPageMappedToDisk(page);
 968
 969                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 970                                 SetPageUptodate(page);
 971                                 goto confused;
 972                         }
 973                 } else {
 974                         zero_user_segment(page, 0, PAGE_SIZE);
 975                         SetPageUptodate(page);
 976                         unlock_page(page);
 977                         goto next_page;
 978                 }
 979
 980                 /*
 981                  * This page will go to BIO.  Do we need to send this
 982                  * BIO off first?
 983                  */
 984                 if (bio && (last_block_in_bio != block_nr - 1)) {
 985 submit_and_realloc:
 986                         submit_bio(READ, bio);
 987                         bio = NULL;
 988                 }
 989                 if (bio == NULL) {
 990                         struct fscrypt_ctx *ctx = NULL;
 991
 992                         if (f2fs_encrypted_inode(inode) &&
 993                                         S_ISREG(inode->i_mode)) {
 994
 995                                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 996                                 if (IS_ERR(ctx))
 997                                         goto set_error_page;
 998
 999                                 /* wait the page to be moved by cleaning */
1000                                 f2fs_wait_on_encrypted_page_writeback(
1001                                                 F2FS_I_SB(inode), block_nr);
1002                         }
1003
1004                         bio = bio_alloc(GFP_KERNEL,
1005                                 min_t(int, nr_pages, BIO_MAX_PAGES));
1006                         if (!bio) {
1007                                 if (ctx)
1008                                         fscrypt_release_ctx(ctx);
1009                                 goto set_error_page;
1010                         }
1011                         bio->bi_bdev = bdev;
1012                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1013                         bio->bi_end_io = f2fs_read_end_io;
1014                         bio->bi_private = ctx;
1015                 }
1016
1017                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1018                         goto submit_and_realloc;
1019
1020                 last_block_in_bio = block_nr;
1021                 goto next_page;
1022 set_error_page:
1023                 SetPageError(page);
1024                 zero_user_segment(page, 0, PAGE_SIZE);
1025                 unlock_page(page);
1026                 goto next_page;
1027 confused:
1028                 if (bio) {
1029                         submit_bio(READ, bio);
1030                         bio = NULL;
1031                 }
1032                 unlock_page(page);
1033 next_page:
1034                 if (pages)
1035                         put_page(page);
1036         }
1037         BUG_ON(pages && !list_empty(pages));
1038         if (bio)
1039                 submit_bio(READ, bio);
1040         return 0;
1041 }
1042
1043 static int f2fs_read_data_page(struct file *file, struct page *page)
1044 {
1045         struct inode *inode = page->mapping->host;
1046         int ret = -EAGAIN;
1047
1048         trace_f2fs_readpage(page, DATA);
1049
1050         /* If the file has inline data, try to read it directly */
1051         if (f2fs_has_inline_data(inode))
1052                 ret = f2fs_read_inline_data(inode, page);
1053         if (ret == -EAGAIN)
1054                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1055         return ret;
1056 }
1057
1058 static int f2fs_read_data_pages(struct file *file,
1059                         struct address_space *mapping,
1060                         struct list_head *pages, unsigned nr_pages)
1061 {
1062         struct inode *inode = file->f_mapping->host;
1063         struct page *page = list_entry(pages->prev, struct page, lru);
1064
1065         trace_f2fs_readpages(inode, page, nr_pages);
1066
1067         /* If the file has inline data, skip readpages */
1068         if (f2fs_has_inline_data(inode))
1069                 return 0;
1070
1071         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1072 }
1073
1074 int do_write_data_page(struct f2fs_io_info *fio)
1075 {
1076         struct page *page = fio->page;
1077         struct inode *inode = page->mapping->host;
1078         struct dnode_of_data dn;
1079         int err = 0;
1080
1081         set_new_dnode(&dn, inode, NULL, NULL, 0);
1082         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1083         if (err)
1084                 return err;
1085
1086         fio->old_blkaddr = dn.data_blkaddr;
1087
1088         /* This page is already truncated */
1089         if (fio->old_blkaddr == NULL_ADDR) {
1090                 ClearPageUptodate(page);
1091                 goto out_writepage;
1092         }
1093
1094         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1095                 gfp_t gfp_flags = GFP_NOFS;
1096
1097                 /* wait for GCed encrypted page writeback */
1098                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1099                                                         fio->old_blkaddr);
1100 retry_encrypt:
1101                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1102                                                                 gfp_flags);
1103                 if (IS_ERR(fio->encrypted_page)) {
1104                         err = PTR_ERR(fio->encrypted_page);
1105                         if (err == -ENOMEM) {
1106                                 /* flush pending ios and wait for a while */
1107                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1108                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1109                                 gfp_flags |= __GFP_NOFAIL;
1110                                 err = 0;
1111                                 goto retry_encrypt;
1112                         }
1113                         goto out_writepage;
1114                 }
1115         }
1116
1117         set_page_writeback(page);
1118
1119         /*
1120          * If current allocation needs SSR,
1121          * it had better in-place writes for updated data.
1122          */
1123         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1124                         !is_cold_data(page) &&
1125                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1126                         need_inplace_update(inode))) {
1127                 rewrite_data_page(fio);
1128                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1129                 trace_f2fs_do_write_data_page(page, IPU);
1130         } else {
1131                 write_data_page(&dn, fio);
1132                 trace_f2fs_do_write_data_page(page, OPU);
1133                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1134                 if (page->index == 0)
1135                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1136         }
1137 out_writepage:
1138         f2fs_put_dnode(&dn);
1139         return err;
1140 }
1141
1142 static int f2fs_write_data_page(struct page *page,
1143                                         struct writeback_control *wbc)
1144 {
1145         struct inode *inode = page->mapping->host;
1146         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1147         loff_t i_size = i_size_read(inode);
1148         const pgoff_t end_index = ((unsigned long long) i_size)
1149                                                         >> PAGE_SHIFT;
1150         unsigned offset = 0;
1151         bool need_balance_fs = false;
1152         int err = 0;
1153         struct f2fs_io_info fio = {
1154                 .sbi = sbi,
1155                 .type = DATA,
1156                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1157                 .page = page,
1158                 .encrypted_page = NULL,
1159         };
1160
1161         trace_f2fs_writepage(page, DATA);
1162
1163         if (page->index < end_index)
1164                 goto write;
1165
1166         /*
1167          * If the offset is out-of-range of file size,
1168          * this page does not have to be written to disk.
1169          */
1170         offset = i_size & (PAGE_SIZE - 1);
1171         if ((page->index >= end_index + 1) || !offset)
1172                 goto out;
1173
1174         zero_user_segment(page, offset, PAGE_SIZE);
1175 write:
1176         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1177                 goto redirty_out;
1178         if (f2fs_is_drop_cache(inode))
1179                 goto out;
1180         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1181                         available_free_memory(sbi, BASE_CHECK))
1182                 goto redirty_out;
1183
1184         /* Dentry blocks are controlled by checkpoint */
1185         if (S_ISDIR(inode->i_mode)) {
1186                 if (unlikely(f2fs_cp_error(sbi)))
1187                         goto redirty_out;
1188                 err = do_write_data_page(&fio);
1189                 goto done;
1190         }
1191
1192         /* we should bypass data pages to proceed the kworkder jobs */
1193         if (unlikely(f2fs_cp_error(sbi))) {
1194                 SetPageError(page);
1195                 goto out;
1196         }
1197
1198         if (!wbc->for_reclaim)
1199                 need_balance_fs = true;
1200         else if (has_not_enough_free_secs(sbi, 0))
1201                 goto redirty_out;
1202
1203         err = -EAGAIN;
1204         f2fs_lock_op(sbi);
1205         if (f2fs_has_inline_data(inode))
1206                 err = f2fs_write_inline_data(inode, page);
1207         if (err == -EAGAIN)
1208                 err = do_write_data_page(&fio);
1209         f2fs_unlock_op(sbi);
1210 done:
1211         if (err && err != -ENOENT)
1212                 goto redirty_out;
1213
1214         clear_cold_data(page);
1215 out:
1216         inode_dec_dirty_pages(inode);
1217         if (err)
1218                 ClearPageUptodate(page);
1219
1220         if (wbc->for_reclaim) {
1221                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1222                 remove_dirty_inode(inode);
1223         }
1224
1225         unlock_page(page);
1226         f2fs_balance_fs(sbi, need_balance_fs);
1227
1228         if (unlikely(f2fs_cp_error(sbi)))
1229                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1230
1231         return 0;
1232
1233 redirty_out:
1234         redirty_page_for_writepage(wbc, page);
1235         return AOP_WRITEPAGE_ACTIVATE;
1236 }
1237
1238 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1239                         void *data)
1240 {
1241         struct address_space *mapping = data;
1242         int ret = mapping->a_ops->writepage(page, wbc);
1243         mapping_set_error(mapping, ret);
1244         return ret;
1245 }
1246
1247 /*
1248  * This function was copied from write_cche_pages from mm/page-writeback.c.
1249  * The major change is making write step of cold data page separately from
1250  * warm/hot data page.
1251  */
1252 static int f2fs_write_cache_pages(struct address_space *mapping,
1253                         struct writeback_control *wbc, writepage_t writepage,
1254                         void *data)
1255 {
1256         int ret = 0;
1257         int done = 0;
1258         struct pagevec pvec;
1259         int nr_pages;
1260         pgoff_t uninitialized_var(writeback_index);
1261         pgoff_t index;
1262         pgoff_t end;            /* Inclusive */
1263         pgoff_t done_index;
1264         int cycled;
1265         int range_whole = 0;
1266         int tag;
1267         int step = 0;
1268
1269         pagevec_init(&pvec, 0);
1270 next:
1271         if (wbc->range_cyclic) {
1272                 writeback_index = mapping->writeback_index; /* prev offset */
1273                 index = writeback_index;
1274                 if (index == 0)
1275                         cycled = 1;
1276                 else
1277                         cycled = 0;
1278                 end = -1;
1279         } else {
1280                 index = wbc->range_start >> PAGE_SHIFT;
1281                 end = wbc->range_end >> PAGE_SHIFT;
1282                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1283                         range_whole = 1;
1284                 cycled = 1; /* ignore range_cyclic tests */
1285         }
1286         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1287                 tag = PAGECACHE_TAG_TOWRITE;
1288         else
1289                 tag = PAGECACHE_TAG_DIRTY;
1290 retry:
1291         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1292                 tag_pages_for_writeback(mapping, index, end);
1293         done_index = index;
1294         while (!done && (index <= end)) {
1295                 int i;
1296
1297                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1298                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1299                 if (nr_pages == 0)
1300                         break;
1301
1302                 for (i = 0; i < nr_pages; i++) {
1303                         struct page *page = pvec.pages[i];
1304
1305                         if (page->index > end) {
1306                                 done = 1;
1307                                 break;
1308                         }
1309
1310                         done_index = page->index;
1311
1312                         lock_page(page);
1313
1314                         if (unlikely(page->mapping != mapping)) {
1315 continue_unlock:
1316                                 unlock_page(page);
1317                                 continue;
1318                         }
1319
1320                         if (!PageDirty(page)) {
1321                                 /* someone wrote it for us */
1322                                 goto continue_unlock;
1323                         }
1324
1325                         if (step == is_cold_data(page))
1326                                 goto continue_unlock;
1327
1328                         if (PageWriteback(page)) {
1329                                 if (wbc->sync_mode != WB_SYNC_NONE)
1330                                         f2fs_wait_on_page_writeback(page,
1331                                                                 DATA, true);
1332                                 else
1333                                         goto continue_unlock;
1334                         }
1335
1336                         BUG_ON(PageWriteback(page));
1337                         if (!clear_page_dirty_for_io(page))
1338                                 goto continue_unlock;
1339
1340                         ret = (*writepage)(page, wbc, data);
1341                         if (unlikely(ret)) {
1342                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1343                                         unlock_page(page);
1344                                         ret = 0;
1345                                 } else {
1346                                         done_index = page->index + 1;
1347                                         done = 1;
1348                                         break;
1349                                 }
1350                         }
1351
1352                         if (--wbc->nr_to_write <= 0 &&
1353                             wbc->sync_mode == WB_SYNC_NONE) {
1354                                 done = 1;
1355                                 break;
1356                         }
1357                 }
1358                 pagevec_release(&pvec);
1359                 cond_resched();
1360         }
1361
1362         if (step < 1) {
1363                 step++;
1364                 goto next;
1365         }
1366
1367         if (!cycled && !done) {
1368                 cycled = 1;
1369                 index = 0;
1370                 end = writeback_index - 1;
1371                 goto retry;
1372         }
1373         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1374                 mapping->writeback_index = done_index;
1375
1376         return ret;
1377 }
1378
1379 static int f2fs_write_data_pages(struct address_space *mapping,
1380                             struct writeback_control *wbc)
1381 {
1382         struct inode *inode = mapping->host;
1383         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1384         bool locked = false;
1385         int ret;
1386         long diff;
1387
1388         /* deal with chardevs and other special file */
1389         if (!mapping->a_ops->writepage)
1390                 return 0;
1391
1392         /* skip writing if there is no dirty page in this inode */
1393         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1394                 return 0;
1395
1396         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1397                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1398                         available_free_memory(sbi, DIRTY_DENTS))
1399                 goto skip_write;
1400
1401         /* skip writing during file defragment */
1402         if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1403                 goto skip_write;
1404
1405         /* during POR, we don't need to trigger writepage at all. */
1406         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1407                 goto skip_write;
1408
1409         trace_f2fs_writepages(mapping->host, wbc, DATA);
1410
1411         diff = nr_pages_to_write(sbi, DATA, wbc);
1412
1413         if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1414                 mutex_lock(&sbi->writepages);
1415                 locked = true;
1416         }
1417         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1418         f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1419         if (locked)
1420                 mutex_unlock(&sbi->writepages);
1421
1422         remove_dirty_inode(inode);
1423
1424         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1425         return ret;
1426
1427 skip_write:
1428         wbc->pages_skipped += get_dirty_pages(inode);
1429         trace_f2fs_writepages(mapping->host, wbc, DATA);
1430         return 0;
1431 }
1432
1433 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1434 {
1435         struct inode *inode = mapping->host;
1436         loff_t i_size = i_size_read(inode);
1437
1438         if (to > i_size) {
1439                 truncate_pagecache(inode, i_size);
1440                 truncate_blocks(inode, i_size, true);
1441         }
1442 }
1443
1444 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1445                         struct page *page, loff_t pos, unsigned len,
1446                         block_t *blk_addr, bool *node_changed)
1447 {
1448         struct inode *inode = page->mapping->host;
1449         pgoff_t index = page->index;
1450         struct dnode_of_data dn;
1451         struct page *ipage;
1452         bool locked = false;
1453         struct extent_info ei;
1454         int err = 0;
1455
1456         /*
1457          * we already allocated all the blocks, so we don't need to get
1458          * the block addresses when there is no need to fill the page.
1459          */
1460         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1461                                         len == PAGE_SIZE)
1462                 return 0;
1463
1464         if (f2fs_has_inline_data(inode) ||
1465                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1466                 f2fs_lock_op(sbi);
1467                 locked = true;
1468         }
1469 restart:
1470         /* check inline_data */
1471         ipage = get_node_page(sbi, inode->i_ino);
1472         if (IS_ERR(ipage)) {
1473                 err = PTR_ERR(ipage);
1474                 goto unlock_out;
1475         }
1476
1477         set_new_dnode(&dn, inode, ipage, ipage, 0);
1478
1479         if (f2fs_has_inline_data(inode)) {
1480                 if (pos + len <= MAX_INLINE_DATA) {
1481                         read_inline_data(page, ipage);
1482                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1483                         set_inline_node(ipage);
1484                 } else {
1485                         err = f2fs_convert_inline_page(&dn, page);
1486                         if (err)
1487                                 goto out;
1488                         if (dn.data_blkaddr == NULL_ADDR)
1489                                 err = f2fs_get_block(&dn, index);
1490                 }
1491         } else if (locked) {
1492                 err = f2fs_get_block(&dn, index);
1493         } else {
1494                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1495                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1496                 } else {
1497                         /* hole case */
1498                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1499                         if (err || (!err && dn.data_blkaddr == NULL_ADDR)) {
1500                                 f2fs_put_dnode(&dn);
1501                                 f2fs_lock_op(sbi);
1502                                 locked = true;
1503                                 goto restart;
1504                         }
1505                 }
1506         }
1507
1508         /* convert_inline_page can make node_changed */
1509         *blk_addr = dn.data_blkaddr;
1510         *node_changed = dn.node_changed;
1511 out:
1512         f2fs_put_dnode(&dn);
1513 unlock_out:
1514         if (locked)
1515                 f2fs_unlock_op(sbi);
1516         return err;
1517 }
1518
1519 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1520                 loff_t pos, unsigned len, unsigned flags,
1521                 struct page **pagep, void **fsdata)
1522 {
1523         struct inode *inode = mapping->host;
1524         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1525         struct page *page = NULL;
1526         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1527         bool need_balance = false;
1528         block_t blkaddr = NULL_ADDR;
1529         int err = 0;
1530
1531         trace_f2fs_write_begin(inode, pos, len, flags);
1532
1533         /*
1534          * We should check this at this moment to avoid deadlock on inode page
1535          * and #0 page. The locking rule for inline_data conversion should be:
1536          * lock_page(page #0) -> lock_page(inode_page)
1537          */
1538         if (index != 0) {
1539                 err = f2fs_convert_inline_inode(inode);
1540                 if (err)
1541                         goto fail;
1542         }
1543 repeat:
1544         page = grab_cache_page_write_begin(mapping, index, flags);
1545         if (!page) {
1546                 err = -ENOMEM;
1547                 goto fail;
1548         }
1549
1550         *pagep = page;
1551
1552         err = prepare_write_begin(sbi, page, pos, len,
1553                                         &blkaddr, &need_balance);
1554         if (err)
1555                 goto fail;
1556
1557         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1558                 unlock_page(page);
1559                 f2fs_balance_fs(sbi, true);
1560                 lock_page(page);
1561                 if (page->mapping != mapping) {
1562                         /* The page got truncated from under us */
1563                         f2fs_put_page(page, 1);
1564                         goto repeat;
1565                 }
1566         }
1567
1568         f2fs_wait_on_page_writeback(page, DATA, false);
1569
1570         /* wait for GCed encrypted page writeback */
1571         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1572                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1573
1574         if (len == PAGE_SIZE)
1575                 goto out_update;
1576         if (PageUptodate(page))
1577                 goto out_clear;
1578
1579         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1580                 unsigned start = pos & (PAGE_SIZE - 1);
1581                 unsigned end = start + len;
1582
1583                 /* Reading beyond i_size is simple: memset to zero */
1584                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1585                 goto out_update;
1586         }
1587
1588         if (blkaddr == NEW_ADDR) {
1589                 zero_user_segment(page, 0, PAGE_SIZE);
1590         } else {
1591                 struct f2fs_io_info fio = {
1592                         .sbi = sbi,
1593                         .type = DATA,
1594                         .rw = READ_SYNC,
1595                         .old_blkaddr = blkaddr,
1596                         .new_blkaddr = blkaddr,
1597                         .page = page,
1598                         .encrypted_page = NULL,
1599                 };
1600                 err = f2fs_submit_page_bio(&fio);
1601                 if (err)
1602                         goto fail;
1603
1604                 lock_page(page);
1605                 if (unlikely(!PageUptodate(page))) {
1606                         err = -EIO;
1607                         goto fail;
1608                 }
1609                 if (unlikely(page->mapping != mapping)) {
1610                         f2fs_put_page(page, 1);
1611                         goto repeat;
1612                 }
1613
1614                 /* avoid symlink page */
1615                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1616                         err = fscrypt_decrypt_page(page);
1617                         if (err)
1618                                 goto fail;
1619                 }
1620         }
1621 out_update:
1622         SetPageUptodate(page);
1623 out_clear:
1624         clear_cold_data(page);
1625         return 0;
1626
1627 fail:
1628         f2fs_put_page(page, 1);
1629         f2fs_write_failed(mapping, pos + len);
1630         return err;
1631 }
1632
1633 static int f2fs_write_end(struct file *file,
1634                         struct address_space *mapping,
1635                         loff_t pos, unsigned len, unsigned copied,
1636                         struct page *page, void *fsdata)
1637 {
1638         struct inode *inode = page->mapping->host;
1639
1640         trace_f2fs_write_end(inode, pos, len, copied);
1641
1642         set_page_dirty(page);
1643
1644         if (pos + copied > i_size_read(inode)) {
1645                 i_size_write(inode, pos + copied);
1646                 mark_inode_dirty(inode);
1647         }
1648
1649         f2fs_put_page(page, 1);
1650         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1651         return copied;
1652 }
1653
1654 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1655                            loff_t offset)
1656 {
1657         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1658
1659         if (offset & blocksize_mask)
1660                 return -EINVAL;
1661
1662         if (iov_iter_alignment(iter) & blocksize_mask)
1663                 return -EINVAL;
1664
1665         return 0;
1666 }
1667
1668 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1669 {
1670         struct address_space *mapping = iocb->ki_filp->f_mapping;
1671         struct inode *inode = mapping->host;
1672         size_t count = iov_iter_count(iter);
1673         loff_t offset = iocb->ki_pos;
1674         int err;
1675
1676         err = check_direct_IO(inode, iter, offset);
1677         if (err)
1678                 return err;
1679
1680         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1681                 return 0;
1682
1683         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1684
1685         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1686         if (err < 0 && iov_iter_rw(iter) == WRITE)
1687                 f2fs_write_failed(mapping, offset + count);
1688
1689         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1690
1691         return err;
1692 }
1693
1694 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1695                                                         unsigned int length)
1696 {
1697         struct inode *inode = page->mapping->host;
1698         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1699
1700         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1701                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1702                 return;
1703
1704         if (PageDirty(page)) {
1705                 if (inode->i_ino == F2FS_META_INO(sbi))
1706                         dec_page_count(sbi, F2FS_DIRTY_META);
1707                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1708                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1709                 else
1710                         inode_dec_dirty_pages(inode);
1711         }
1712
1713         /* This is atomic written page, keep Private */
1714         if (IS_ATOMIC_WRITTEN_PAGE(page))
1715                 return;
1716
1717         ClearPagePrivate(page);
1718 }
1719
1720 int f2fs_release_page(struct page *page, gfp_t wait)
1721 {
1722         /* If this is dirty page, keep PagePrivate */
1723         if (PageDirty(page))
1724                 return 0;
1725
1726         /* This is atomic written page, keep Private */
1727         if (IS_ATOMIC_WRITTEN_PAGE(page))
1728                 return 0;
1729
1730         ClearPagePrivate(page);
1731         return 1;
1732 }
1733
1734 static int f2fs_set_data_page_dirty(struct page *page)
1735 {
1736         struct address_space *mapping = page->mapping;
1737         struct inode *inode = mapping->host;
1738
1739         trace_f2fs_set_page_dirty(page, DATA);
1740
1741         SetPageUptodate(page);
1742
1743         if (f2fs_is_atomic_file(inode)) {
1744                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1745                         register_inmem_page(inode, page);
1746                         return 1;
1747                 }
1748                 /*
1749                  * Previously, this page has been registered, we just
1750                  * return here.
1751                  */
1752                 return 0;
1753         }
1754
1755         if (!PageDirty(page)) {
1756                 __set_page_dirty_nobuffers(page);
1757                 update_dirty_page(inode, page);
1758                 return 1;
1759         }
1760         return 0;
1761 }
1762
1763 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1764 {
1765         struct inode *inode = mapping->host;
1766
1767         if (f2fs_has_inline_data(inode))
1768                 return 0;
1769
1770         /* make sure allocating whole blocks */
1771         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1772                 filemap_write_and_wait(mapping);
1773
1774         return generic_block_bmap(mapping, block, get_data_block_bmap);
1775 }
1776
1777 const struct address_space_operations f2fs_dblock_aops = {
1778         .readpage       = f2fs_read_data_page,
1779         .readpages      = f2fs_read_data_pages,
1780         .writepage      = f2fs_write_data_page,
1781         .writepages     = f2fs_write_data_pages,
1782         .write_begin    = f2fs_write_begin,
1783         .write_end      = f2fs_write_end,
1784         .set_page_dirty = f2fs_set_data_page_dirty,
1785         .invalidatepage = f2fs_invalidate_page,
1786         .releasepage    = f2fs_release_page,
1787         .direct_IO      = f2fs_direct_IO,
1788         .bmap           = f2fs_bmap,
1789 };