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                         f2fs_release_crypto_ctx(bio->bi_private);
  38                 } else {
  39                         f2fs_end_io_crypto_work(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                 f2fs_restore_and_release_control_page(&page);
  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                         struct f2fs_crypto_ctx *ctx;
 136
 137                         /* encrypted page */
 138                         ctx = (struct f2fs_crypto_ctx *)page_private(
 139                                                                 bvec->bv_page);
 140                         target = ctx->w.control_page;
 141                 }
 142
 143                 if (inode && inode == target->mapping->host)
 144                         return true;
 145                 if (page && page == target)
 146                         return true;
 147                 if (ino && ino == ino_of_node(target))
 148                         return true;
 149         }
 150
 151         return false;
 152 }
 153
 154 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 155                                                 struct page *page, nid_t ino,
 156                                                 enum page_type type)
 157 {
 158         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 159         struct f2fs_bio_info *io = &sbi->write_io[btype];
 160         bool ret;
 161
 162         down_read(&io->io_rwsem);
 163         ret = __has_merged_page(io, inode, page, ino);
 164         up_read(&io->io_rwsem);
 165         return ret;
 166 }
 167
 168 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 169                                 struct inode *inode, struct page *page,
 170                                 nid_t ino, enum page_type type, int rw)
 171 {
 172         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 173         struct f2fs_bio_info *io;
 174
 175         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 176
 177         down_write(&io->io_rwsem);
 178
 179         if (!__has_merged_page(io, inode, page, ino))
 180                 goto out;
 181
 182         /* change META to META_FLUSH in the checkpoint procedure */
 183         if (type >= META_FLUSH) {
 184                 io->fio.type = META_FLUSH;
 185                 if (test_opt(sbi, NOBARRIER))
 186                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 187                 else
 188                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 189         }
 190         __submit_merged_bio(io);
 191 out:
 192         up_write(&io->io_rwsem);
 193 }
 194
 195 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 196                                                                         int rw)
 197 {
 198         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 199 }
 200
 201 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 202                                 struct inode *inode, struct page *page,
 203                                 nid_t ino, enum page_type type, int rw)
 204 {
 205         if (has_merged_page(sbi, inode, page, ino, type))
 206                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 207 }
 208
 209 /*
 210  * Fill the locked page with data located in the block address.
 211  * Return unlocked page.
 212  */
 213 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 214 {
 215         struct bio *bio;
 216         struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 217
 218         trace_f2fs_submit_page_bio(page, fio);
 219         f2fs_trace_ios(fio, 0);
 220
 221         /* Allocate a new bio */
 222         bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
 223
 224         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 225                 bio_put(bio);
 226                 return -EFAULT;
 227         }
 228
 229         submit_bio(fio->rw, bio);
 230         return 0;
 231 }
 232
 233 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 234 {
 235         struct f2fs_sb_info *sbi = fio->sbi;
 236         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 237         struct f2fs_bio_info *io;
 238         bool is_read = is_read_io(fio->rw);
 239         struct page *bio_page;
 240
 241         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 242
 243         verify_block_addr(sbi, fio->blk_addr);
 244
 245         down_write(&io->io_rwsem);
 246
 247         if (!is_read)
 248                 inc_page_count(sbi, F2FS_WRITEBACK);
 249
 250         if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
 251                                                 io->fio.rw != fio->rw))
 252                 __submit_merged_bio(io);
 253 alloc_new:
 254         if (io->bio == NULL) {
 255                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 256
 257                 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
 258                 io->fio = *fio;
 259         }
 260
 261         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 262
 263         if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
 264                                                         PAGE_CACHE_SIZE) {
 265                 __submit_merged_bio(io);
 266                 goto alloc_new;
 267         }
 268
 269         io->last_block_in_bio = fio->blk_addr;
 270         f2fs_trace_ios(fio, 0);
 271
 272         up_write(&io->io_rwsem);
 273         trace_f2fs_submit_page_mbio(fio->page, fio);
 274 }
 275
 276 /*
 277  * Lock ordering for the change of data block address:
 278  * ->data_page
 279  *  ->node_page
 280  *    update block addresses in the node page
 281  */
 282 void set_data_blkaddr(struct dnode_of_data *dn)
 283 {
 284         struct f2fs_node *rn;
 285         __le32 *addr_array;
 286         struct page *node_page = dn->node_page;
 287         unsigned int ofs_in_node = dn->ofs_in_node;
 288
 289         f2fs_wait_on_page_writeback(node_page, NODE, true);
 290
 291         rn = F2FS_NODE(node_page);
 292
 293         /* Get physical address of data block */
 294         addr_array = blkaddr_in_node(rn);
 295         addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 296         if (set_page_dirty(node_page))
 297                 dn->node_changed = true;
 298 }
 299
 300 int reserve_new_block(struct dnode_of_data *dn)
 301 {
 302         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 303
 304         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 305                 return -EPERM;
 306         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 307                 return -ENOSPC;
 308
 309         trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
 310
 311         dn->data_blkaddr = NEW_ADDR;
 312         set_data_blkaddr(dn);
 313         mark_inode_dirty(dn->inode);
 314         sync_inode_page(dn);
 315         return 0;
 316 }
 317
 318 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 319 {
 320         bool need_put = dn->inode_page ? false : true;
 321         int err;
 322
 323         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 324         if (err)
 325                 return err;
 326
 327         if (dn->data_blkaddr == NULL_ADDR)
 328                 err = reserve_new_block(dn);
 329         if (err || need_put)
 330                 f2fs_put_dnode(dn);
 331         return err;
 332 }
 333
 334 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 335 {
 336         struct extent_info ei;
 337         struct inode *inode = dn->inode;
 338
 339         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 340                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 341                 return 0;
 342         }
 343
 344         return f2fs_reserve_block(dn, index);
 345 }
 346
 347 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 348                                                 int rw, bool for_write)
 349 {
 350         struct address_space *mapping = inode->i_mapping;
 351         struct dnode_of_data dn;
 352         struct page *page;
 353         struct extent_info ei;
 354         int err;
 355         struct f2fs_io_info fio = {
 356                 .sbi = F2FS_I_SB(inode),
 357                 .type = DATA,
 358                 .rw = rw,
 359                 .encrypted_page = NULL,
 360         };
 361
 362         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 363                 return read_mapping_page(mapping, index, NULL);
 364
 365         page = f2fs_grab_cache_page(mapping, index, for_write);
 366         if (!page)
 367                 return ERR_PTR(-ENOMEM);
 368
 369         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 370                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 371                 goto got_it;
 372         }
 373
 374         set_new_dnode(&dn, inode, NULL, NULL, 0);
 375         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 376         if (err)
 377                 goto put_err;
 378         f2fs_put_dnode(&dn);
 379
 380         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 381                 err = -ENOENT;
 382                 goto put_err;
 383         }
 384 got_it:
 385         if (PageUptodate(page)) {
 386                 unlock_page(page);
 387                 return page;
 388         }
 389
 390         /*
 391          * A new dentry page is allocated but not able to be written, since its
 392          * new inode page couldn't be allocated due to -ENOSPC.
 393          * In such the case, its blkaddr can be remained as NEW_ADDR.
 394          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 395          */
 396         if (dn.data_blkaddr == NEW_ADDR) {
 397                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 398                 SetPageUptodate(page);
 399                 unlock_page(page);
 400                 return page;
 401         }
 402
 403         fio.blk_addr = dn.data_blkaddr;
 404         fio.page = page;
 405         err = f2fs_submit_page_bio(&fio);
 406         if (err)
 407                 goto put_err;
 408         return page;
 409
 410 put_err:
 411         f2fs_put_page(page, 1);
 412         return ERR_PTR(err);
 413 }
 414
 415 struct page *find_data_page(struct inode *inode, pgoff_t index)
 416 {
 417         struct address_space *mapping = inode->i_mapping;
 418         struct page *page;
 419
 420         page = find_get_page(mapping, index);
 421         if (page && PageUptodate(page))
 422                 return page;
 423         f2fs_put_page(page, 0);
 424
 425         page = get_read_data_page(inode, index, READ_SYNC, false);
 426         if (IS_ERR(page))
 427                 return page;
 428
 429         if (PageUptodate(page))
 430                 return page;
 431
 432         wait_on_page_locked(page);
 433         if (unlikely(!PageUptodate(page))) {
 434                 f2fs_put_page(page, 0);
 435                 return ERR_PTR(-EIO);
 436         }
 437         return page;
 438 }
 439
 440 /*
 441  * If it tries to access a hole, return an error.
 442  * Because, the callers, functions in dir.c and GC, should be able to know
 443  * whether this page exists or not.
 444  */
 445 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 446                                                         bool for_write)
 447 {
 448         struct address_space *mapping = inode->i_mapping;
 449         struct page *page;
 450 repeat:
 451         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 452         if (IS_ERR(page))
 453                 return page;
 454
 455         /* wait for read completion */
 456         lock_page(page);
 457         if (unlikely(!PageUptodate(page))) {
 458                 f2fs_put_page(page, 1);
 459                 return ERR_PTR(-EIO);
 460         }
 461         if (unlikely(page->mapping != mapping)) {
 462                 f2fs_put_page(page, 1);
 463                 goto repeat;
 464         }
 465         return page;
 466 }
 467
 468 /*
 469  * Caller ensures that this data page is never allocated.
 470  * A new zero-filled data page is allocated in the page cache.
 471  *
 472  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 473  * f2fs_unlock_op().
 474  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 475  * ipage should be released by this function.
 476  */
 477 struct page *get_new_data_page(struct inode *inode,
 478                 struct page *ipage, pgoff_t index, bool new_i_size)
 479 {
 480         struct address_space *mapping = inode->i_mapping;
 481         struct page *page;
 482         struct dnode_of_data dn;
 483         int err;
 484
 485         page = f2fs_grab_cache_page(mapping, index, true);
 486         if (!page) {
 487                 /*
 488                  * before exiting, we should make sure ipage will be released
 489                  * if any error occur.
 490                  */
 491                 f2fs_put_page(ipage, 1);
 492                 return ERR_PTR(-ENOMEM);
 493         }
 494
 495         set_new_dnode(&dn, inode, ipage, NULL, 0);
 496         err = f2fs_reserve_block(&dn, index);
 497         if (err) {
 498                 f2fs_put_page(page, 1);
 499                 return ERR_PTR(err);
 500         }
 501         if (!ipage)
 502                 f2fs_put_dnode(&dn);
 503
 504         if (PageUptodate(page))
 505                 goto got_it;
 506
 507         if (dn.data_blkaddr == NEW_ADDR) {
 508                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 509                 SetPageUptodate(page);
 510         } else {
 511                 f2fs_put_page(page, 1);
 512
 513                 /* if ipage exists, blkaddr should be NEW_ADDR */
 514                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 515                 page = get_lock_data_page(inode, index, true);
 516                 if (IS_ERR(page))
 517                         return page;
 518         }
 519 got_it:
 520         if (new_i_size && i_size_read(inode) <
 521                                 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
 522                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
 523                 /* Only the directory inode sets new_i_size */
 524                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 525         }
 526         return page;
 527 }
 528
 529 static int __allocate_data_block(struct dnode_of_data *dn)
 530 {
 531         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 532         struct f2fs_summary sum;
 533         struct node_info ni;
 534         int seg = CURSEG_WARM_DATA;
 535         pgoff_t fofs;
 536
 537         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 538                 return -EPERM;
 539
 540         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 541         if (dn->data_blkaddr == NEW_ADDR)
 542                 goto alloc;
 543
 544         if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
 545                 return -ENOSPC;
 546
 547 alloc:
 548         get_node_info(sbi, dn->nid, &ni);
 549         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 550
 551         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 552                 seg = CURSEG_DIRECT_IO;
 553
 554         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 555                                                                 &sum, seg);
 556         set_data_blkaddr(dn);
 557
 558         /* update i_size */
 559         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 560                                                         dn->ofs_in_node;
 561         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
 562                 i_size_write(dn->inode,
 563                                 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
 564         return 0;
 565 }
 566
 567 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 568 {
 569         struct inode *inode = file_inode(iocb->ki_filp);
 570         struct f2fs_map_blocks map;
 571         ssize_t ret = 0;
 572
 573         map.m_lblk = F2FS_BYTES_TO_BLK(iocb->ki_pos);
 574         map.m_len = F2FS_BLK_ALIGN(iov_iter_count(from));
 575         map.m_next_pgofs = NULL;
 576
 577         if (f2fs_encrypted_inode(inode))
 578                 return 0;
 579
 580         if (iocb->ki_flags & IOCB_DIRECT) {
 581                 ret = f2fs_convert_inline_inode(inode);
 582                 if (ret)
 583                         return ret;
 584                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 585         }
 586         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 587                 ret = f2fs_convert_inline_inode(inode);
 588                 if (ret)
 589                         return ret;
 590         }
 591         if (!f2fs_has_inline_data(inode))
 592                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 593         return ret;
 594 }
 595
 596 /*
 597  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 598  * f2fs_map_blocks structure.
 599  * If original data blocks are allocated, then give them to blockdev.
 600  * Otherwise,
 601  *     a. preallocate requested block addresses
 602  *     b. do not use extent cache for better performance
 603  *     c. give the block addresses to blockdev
 604  */
 605 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 606                                                 int create, int flag)
 607 {
 608         unsigned int maxblocks = map->m_len;
 609         struct dnode_of_data dn;
 610         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 611         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 612         pgoff_t pgofs, end_offset;
 613         int err = 0, ofs = 1;
 614         struct extent_info ei;
 615         bool allocated = false;
 616         block_t blkaddr;
 617
 618         map->m_len = 0;
 619         map->m_flags = 0;
 620
 621         /* it only supports block size == page size */
 622         pgofs = (pgoff_t)map->m_lblk;
 623
 624         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 625                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 626                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 627                 map->m_flags = F2FS_MAP_MAPPED;
 628                 goto out;
 629         }
 630
 631 next_dnode:
 632         if (create)
 633                 f2fs_lock_op(sbi);
 634
 635         /* When reading holes, we need its node page */
 636         set_new_dnode(&dn, inode, NULL, NULL, 0);
 637         err = get_dnode_of_data(&dn, pgofs, mode);
 638         if (err) {
 639                 if (err == -ENOENT) {
 640                         err = 0;
 641                         if (map->m_next_pgofs)
 642                                 *map->m_next_pgofs =
 643                                         get_next_page_offset(&dn, pgofs);
 644                 }
 645                 goto unlock_out;
 646         }
 647
 648         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 649
 650 next_block:
 651         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 652
 653         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 654                 if (create) {
 655                         if (unlikely(f2fs_cp_error(sbi))) {
 656                                 err = -EIO;
 657                                 goto sync_out;
 658                         }
 659                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 660                                 if (blkaddr == NULL_ADDR)
 661                                         err = reserve_new_block(&dn);
 662                         } else {
 663                                 err = __allocate_data_block(&dn);
 664                         }
 665                         if (err)
 666                                 goto sync_out;
 667                         allocated = true;
 668                         map->m_flags = F2FS_MAP_NEW;
 669                         blkaddr = dn.data_blkaddr;
 670                 } else {
 671                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 672                                                 blkaddr == NULL_ADDR) {
 673                                 if (map->m_next_pgofs)
 674                                         *map->m_next_pgofs = pgofs + 1;
 675                         }
 676                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 677                                                 blkaddr != NEW_ADDR) {
 678                                 if (flag == F2FS_GET_BLOCK_BMAP)
 679                                         err = -ENOENT;
 680                                 goto sync_out;
 681                         }
 682                 }
 683         }
 684
 685         if (map->m_len == 0) {
 686                 /* preallocated unwritten block should be mapped for fiemap. */
 687                 if (blkaddr == NEW_ADDR)
 688                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 689                 map->m_flags |= F2FS_MAP_MAPPED;
 690
 691                 map->m_pblk = blkaddr;
 692                 map->m_len = 1;
 693         } else if ((map->m_pblk != NEW_ADDR &&
 694                         blkaddr == (map->m_pblk + ofs)) ||
 695                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 696                         flag == F2FS_GET_BLOCK_PRE_DIO ||
 697                         flag == F2FS_GET_BLOCK_PRE_AIO) {
 698                 ofs++;
 699                 map->m_len++;
 700         } else {
 701                 goto sync_out;
 702         }
 703
 704         dn.ofs_in_node++;
 705         pgofs++;
 706
 707         if (map->m_len < maxblocks) {
 708                 if (dn.ofs_in_node < end_offset)
 709                         goto next_block;
 710
 711                 if (allocated)
 712                         sync_inode_page(&dn);
 713                 f2fs_put_dnode(&dn);
 714
 715                 if (create) {
 716                         f2fs_unlock_op(sbi);
 717                         f2fs_balance_fs(sbi, allocated);
 718                 }
 719                 allocated = false;
 720                 goto next_dnode;
 721         }
 722
 723 sync_out:
 724         if (allocated)
 725                 sync_inode_page(&dn);
 726         f2fs_put_dnode(&dn);
 727 unlock_out:
 728         if (create) {
 729                 f2fs_unlock_op(sbi);
 730                 f2fs_balance_fs(sbi, allocated);
 731         }
 732 out:
 733         trace_f2fs_map_blocks(inode, map, err);
 734         return err;
 735 }
 736
 737 static int __get_data_block(struct inode *inode, sector_t iblock,
 738                         struct buffer_head *bh, int create, int flag,
 739                         pgoff_t *next_pgofs)
 740 {
 741         struct f2fs_map_blocks map;
 742         int ret;
 743
 744         map.m_lblk = iblock;
 745         map.m_len = bh->b_size >> inode->i_blkbits;
 746         map.m_next_pgofs = next_pgofs;
 747
 748         ret = f2fs_map_blocks(inode, &map, create, flag);
 749         if (!ret) {
 750                 map_bh(bh, inode->i_sb, map.m_pblk);
 751                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 752                 bh->b_size = map.m_len << inode->i_blkbits;
 753         }
 754         return ret;
 755 }
 756
 757 static int get_data_block(struct inode *inode, sector_t iblock,
 758                         struct buffer_head *bh_result, int create, int flag,
 759                         pgoff_t *next_pgofs)
 760 {
 761         return __get_data_block(inode, iblock, bh_result, create,
 762                                                         flag, next_pgofs);
 763 }
 764
 765 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 766                         struct buffer_head *bh_result, int create)
 767 {
 768         return __get_data_block(inode, iblock, bh_result, create,
 769                                                 F2FS_GET_BLOCK_DIO, NULL);
 770 }
 771
 772 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 773                         struct buffer_head *bh_result, int create)
 774 {
 775         /* Block number less than F2FS MAX BLOCKS */
 776         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 777                 return -EFBIG;
 778
 779         return __get_data_block(inode, iblock, bh_result, create,
 780                                                 F2FS_GET_BLOCK_BMAP, NULL);
 781 }
 782
 783 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 784 {
 785         return (offset >> inode->i_blkbits);
 786 }
 787
 788 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 789 {
 790         return (blk << inode->i_blkbits);
 791 }
 792
 793 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 794                 u64 start, u64 len)
 795 {
 796         struct buffer_head map_bh;
 797         sector_t start_blk, last_blk;
 798         pgoff_t next_pgofs;
 799         loff_t isize;
 800         u64 logical = 0, phys = 0, size = 0;
 801         u32 flags = 0;
 802         int ret = 0;
 803
 804         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 805         if (ret)
 806                 return ret;
 807
 808         if (f2fs_has_inline_data(inode)) {
 809                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 810                 if (ret != -EAGAIN)
 811                         return ret;
 812         }
 813
 814         inode_lock(inode);
 815
 816         isize = i_size_read(inode);
 817         if (start >= isize)
 818                 goto out;
 819
 820         if (start + len > isize)
 821                 len = isize - start;
 822
 823         if (logical_to_blk(inode, len) == 0)
 824                 len = blk_to_logical(inode, 1);
 825
 826         start_blk = logical_to_blk(inode, start);
 827         last_blk = logical_to_blk(inode, start + len - 1);
 828
 829 next:
 830         memset(&map_bh, 0, sizeof(struct buffer_head));
 831         map_bh.b_size = len;
 832
 833         ret = get_data_block(inode, start_blk, &map_bh, 0,
 834                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 835         if (ret)
 836                 goto out;
 837
 838         /* HOLE */
 839         if (!buffer_mapped(&map_bh)) {
 840                 start_blk = next_pgofs;
 841                 /* Go through holes util pass the EOF */
 842                 if (blk_to_logical(inode, start_blk) < isize)
 843                         goto prep_next;
 844                 /* Found a hole beyond isize means no more extents.
 845                  * Note that the premise is that filesystems don't
 846                  * punch holes beyond isize and keep size unchanged.
 847                  */
 848                 flags |= FIEMAP_EXTENT_LAST;
 849         }
 850
 851         if (size) {
 852                 if (f2fs_encrypted_inode(inode))
 853                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 854
 855                 ret = fiemap_fill_next_extent(fieinfo, logical,
 856                                 phys, size, flags);
 857         }
 858
 859         if (start_blk > last_blk || ret)
 860                 goto out;
 861
 862         logical = blk_to_logical(inode, start_blk);
 863         phys = blk_to_logical(inode, map_bh.b_blocknr);
 864         size = map_bh.b_size;
 865         flags = 0;
 866         if (buffer_unwritten(&map_bh))
 867                 flags = FIEMAP_EXTENT_UNWRITTEN;
 868
 869         start_blk += logical_to_blk(inode, size);
 870
 871 prep_next:
 872         cond_resched();
 873         if (fatal_signal_pending(current))
 874                 ret = -EINTR;
 875         else
 876                 goto next;
 877 out:
 878         if (ret == 1)
 879                 ret = 0;
 880
 881         inode_unlock(inode);
 882         return ret;
 883 }
 884
 885 /*
 886  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 887  * Major change was from block_size == page_size in f2fs by default.
 888  */
 889 static int f2fs_mpage_readpages(struct address_space *mapping,
 890                         struct list_head *pages, struct page *page,
 891                         unsigned nr_pages)
 892 {
 893         struct bio *bio = NULL;
 894         unsigned page_idx;
 895         sector_t last_block_in_bio = 0;
 896         struct inode *inode = mapping->host;
 897         const unsigned blkbits = inode->i_blkbits;
 898         const unsigned blocksize = 1 << blkbits;
 899         sector_t block_in_file;
 900         sector_t last_block;
 901         sector_t last_block_in_file;
 902         sector_t block_nr;
 903         struct block_device *bdev = inode->i_sb->s_bdev;
 904         struct f2fs_map_blocks map;
 905
 906         map.m_pblk = 0;
 907         map.m_lblk = 0;
 908         map.m_len = 0;
 909         map.m_flags = 0;
 910         map.m_next_pgofs = NULL;
 911
 912         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 913
 914                 prefetchw(&page->flags);
 915                 if (pages) {
 916                         page = list_entry(pages->prev, struct page, lru);
 917                         list_del(&page->lru);
 918                         if (add_to_page_cache_lru(page, mapping,
 919                                                   page->index, GFP_KERNEL))
 920                                 goto next_page;
 921                 }
 922
 923                 block_in_file = (sector_t)page->index;
 924                 last_block = block_in_file + nr_pages;
 925                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 926                                                                 blkbits;
 927                 if (last_block > last_block_in_file)
 928                         last_block = last_block_in_file;
 929
 930                 /*
 931                  * Map blocks using the previous result first.
 932                  */
 933                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
 934                                 block_in_file > map.m_lblk &&
 935                                 block_in_file < (map.m_lblk + map.m_len))
 936                         goto got_it;
 937
 938                 /*
 939                  * Then do more f2fs_map_blocks() calls until we are
 940                  * done with this page.
 941                  */
 942                 map.m_flags = 0;
 943
 944                 if (block_in_file < last_block) {
 945                         map.m_lblk = block_in_file;
 946                         map.m_len = last_block - block_in_file;
 947
 948                         if (f2fs_map_blocks(inode, &map, 0,
 949                                                 F2FS_GET_BLOCK_READ))
 950                                 goto set_error_page;
 951                 }
 952 got_it:
 953                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
 954                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
 955                         SetPageMappedToDisk(page);
 956
 957                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
 958                                 SetPageUptodate(page);
 959                                 goto confused;
 960                         }
 961                 } else {
 962                         zero_user_segment(page, 0, PAGE_CACHE_SIZE);
 963                         SetPageUptodate(page);
 964                         unlock_page(page);
 965                         goto next_page;
 966                 }
 967
 968                 /*
 969                  * This page will go to BIO.  Do we need to send this
 970                  * BIO off first?
 971                  */
 972                 if (bio && (last_block_in_bio != block_nr - 1)) {
 973 submit_and_realloc:
 974                         submit_bio(READ, bio);
 975                         bio = NULL;
 976                 }
 977                 if (bio == NULL) {
 978                         struct f2fs_crypto_ctx *ctx = NULL;
 979
 980                         if (f2fs_encrypted_inode(inode) &&
 981                                         S_ISREG(inode->i_mode)) {
 982
 983                                 ctx = f2fs_get_crypto_ctx(inode);
 984                                 if (IS_ERR(ctx))
 985                                         goto set_error_page;
 986
 987                                 /* wait the page to be moved by cleaning */
 988                                 f2fs_wait_on_encrypted_page_writeback(
 989                                                 F2FS_I_SB(inode), block_nr);
 990                         }
 991
 992                         bio = bio_alloc(GFP_KERNEL,
 993                                 min_t(int, nr_pages, BIO_MAX_PAGES));
 994                         if (!bio) {
 995                                 if (ctx)
 996                                         f2fs_release_crypto_ctx(ctx);
 997                                 goto set_error_page;
 998                         }
 999                         bio->bi_bdev = bdev;
1000                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1001                         bio->bi_end_io = f2fs_read_end_io;
1002                         bio->bi_private = ctx;
1003                 }
1004
1005                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1006                         goto submit_and_realloc;
1007
1008                 last_block_in_bio = block_nr;
1009                 goto next_page;
1010 set_error_page:
1011                 SetPageError(page);
1012                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1013                 unlock_page(page);
1014                 goto next_page;
1015 confused:
1016                 if (bio) {
1017                         submit_bio(READ, bio);
1018                         bio = NULL;
1019                 }
1020                 unlock_page(page);
1021 next_page:
1022                 if (pages)
1023                         page_cache_release(page);
1024         }
1025         BUG_ON(pages && !list_empty(pages));
1026         if (bio)
1027                 submit_bio(READ, bio);
1028         return 0;
1029 }
1030
1031 static int f2fs_read_data_page(struct file *file, struct page *page)
1032 {
1033         struct inode *inode = page->mapping->host;
1034         int ret = -EAGAIN;
1035
1036         trace_f2fs_readpage(page, DATA);
1037
1038         /* If the file has inline data, try to read it directly */
1039         if (f2fs_has_inline_data(inode))
1040                 ret = f2fs_read_inline_data(inode, page);
1041         if (ret == -EAGAIN)
1042                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1043         return ret;
1044 }
1045
1046 static int f2fs_read_data_pages(struct file *file,
1047                         struct address_space *mapping,
1048                         struct list_head *pages, unsigned nr_pages)
1049 {
1050         struct inode *inode = file->f_mapping->host;
1051         struct page *page = list_entry(pages->prev, struct page, lru);
1052
1053         trace_f2fs_readpages(inode, page, nr_pages);
1054
1055         /* If the file has inline data, skip readpages */
1056         if (f2fs_has_inline_data(inode))
1057                 return 0;
1058
1059         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1060 }
1061
1062 int do_write_data_page(struct f2fs_io_info *fio)
1063 {
1064         struct page *page = fio->page;
1065         struct inode *inode = page->mapping->host;
1066         struct dnode_of_data dn;
1067         int err = 0;
1068
1069         set_new_dnode(&dn, inode, NULL, NULL, 0);
1070         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1071         if (err)
1072                 return err;
1073
1074         fio->blk_addr = dn.data_blkaddr;
1075
1076         /* This page is already truncated */
1077         if (fio->blk_addr == NULL_ADDR) {
1078                 ClearPageUptodate(page);
1079                 goto out_writepage;
1080         }
1081
1082         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1083
1084                 /* wait for GCed encrypted page writeback */
1085                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1086                                                         fio->blk_addr);
1087
1088                 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1089                 if (IS_ERR(fio->encrypted_page)) {
1090                         err = PTR_ERR(fio->encrypted_page);
1091                         goto out_writepage;
1092                 }
1093         }
1094
1095         set_page_writeback(page);
1096
1097         /*
1098          * If current allocation needs SSR,
1099          * it had better in-place writes for updated data.
1100          */
1101         if (unlikely(fio->blk_addr != NEW_ADDR &&
1102                         !is_cold_data(page) &&
1103                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1104                         need_inplace_update(inode))) {
1105                 rewrite_data_page(fio);
1106                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1107                 trace_f2fs_do_write_data_page(page, IPU);
1108         } else {
1109                 write_data_page(&dn, fio);
1110                 set_data_blkaddr(&dn);
1111                 f2fs_update_extent_cache(&dn);
1112                 trace_f2fs_do_write_data_page(page, OPU);
1113                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1114                 if (page->index == 0)
1115                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1116         }
1117 out_writepage:
1118         f2fs_put_dnode(&dn);
1119         return err;
1120 }
1121
1122 static int f2fs_write_data_page(struct page *page,
1123                                         struct writeback_control *wbc)
1124 {
1125         struct inode *inode = page->mapping->host;
1126         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1127         loff_t i_size = i_size_read(inode);
1128         const pgoff_t end_index = ((unsigned long long) i_size)
1129                                                         >> PAGE_CACHE_SHIFT;
1130         unsigned offset = 0;
1131         bool need_balance_fs = false;
1132         int err = 0;
1133         struct f2fs_io_info fio = {
1134                 .sbi = sbi,
1135                 .type = DATA,
1136                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1137                 .page = page,
1138                 .encrypted_page = NULL,
1139         };
1140
1141         trace_f2fs_writepage(page, DATA);
1142
1143         if (page->index < end_index)
1144                 goto write;
1145
1146         /*
1147          * If the offset is out-of-range of file size,
1148          * this page does not have to be written to disk.
1149          */
1150         offset = i_size & (PAGE_CACHE_SIZE - 1);
1151         if ((page->index >= end_index + 1) || !offset)
1152                 goto out;
1153
1154         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1155 write:
1156         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1157                 goto redirty_out;
1158         if (f2fs_is_drop_cache(inode))
1159                 goto out;
1160         if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1161                         available_free_memory(sbi, BASE_CHECK))
1162                 goto redirty_out;
1163
1164         /* Dentry blocks are controlled by checkpoint */
1165         if (S_ISDIR(inode->i_mode)) {
1166                 if (unlikely(f2fs_cp_error(sbi)))
1167                         goto redirty_out;
1168                 err = do_write_data_page(&fio);
1169                 goto done;
1170         }
1171
1172         /* we should bypass data pages to proceed the kworkder jobs */
1173         if (unlikely(f2fs_cp_error(sbi))) {
1174                 SetPageError(page);
1175                 goto out;
1176         }
1177
1178         if (!wbc->for_reclaim)
1179                 need_balance_fs = true;
1180         else if (has_not_enough_free_secs(sbi, 0))
1181                 goto redirty_out;
1182
1183         err = -EAGAIN;
1184         f2fs_lock_op(sbi);
1185         if (f2fs_has_inline_data(inode))
1186                 err = f2fs_write_inline_data(inode, page);
1187         if (err == -EAGAIN)
1188                 err = do_write_data_page(&fio);
1189         f2fs_unlock_op(sbi);
1190 done:
1191         if (err && err != -ENOENT)
1192                 goto redirty_out;
1193
1194         clear_cold_data(page);
1195 out:
1196         inode_dec_dirty_pages(inode);
1197         if (err)
1198                 ClearPageUptodate(page);
1199
1200         if (wbc->for_reclaim) {
1201                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1202                 remove_dirty_inode(inode);
1203         }
1204
1205         unlock_page(page);
1206         f2fs_balance_fs(sbi, need_balance_fs);
1207
1208         if (unlikely(f2fs_cp_error(sbi)))
1209                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1210
1211         return 0;
1212
1213 redirty_out:
1214         redirty_page_for_writepage(wbc, page);
1215         return AOP_WRITEPAGE_ACTIVATE;
1216 }
1217
1218 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1219                         void *data)
1220 {
1221         struct address_space *mapping = data;
1222         int ret = mapping->a_ops->writepage(page, wbc);
1223         mapping_set_error(mapping, ret);
1224         return ret;
1225 }
1226
1227 /*
1228  * This function was copied from write_cche_pages from mm/page-writeback.c.
1229  * The major change is making write step of cold data page separately from
1230  * warm/hot data page.
1231  */
1232 static int f2fs_write_cache_pages(struct address_space *mapping,
1233                         struct writeback_control *wbc, writepage_t writepage,
1234                         void *data)
1235 {
1236         int ret = 0;
1237         int done = 0;
1238         struct pagevec pvec;
1239         int nr_pages;
1240         pgoff_t uninitialized_var(writeback_index);
1241         pgoff_t index;
1242         pgoff_t end;            /* Inclusive */
1243         pgoff_t done_index;
1244         int cycled;
1245         int range_whole = 0;
1246         int tag;
1247         int step = 0;
1248
1249         pagevec_init(&pvec, 0);
1250 next:
1251         if (wbc->range_cyclic) {
1252                 writeback_index = mapping->writeback_index; /* prev offset */
1253                 index = writeback_index;
1254                 if (index == 0)
1255                         cycled = 1;
1256                 else
1257                         cycled = 0;
1258                 end = -1;
1259         } else {
1260                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1261                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1262                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1263                         range_whole = 1;
1264                 cycled = 1; /* ignore range_cyclic tests */
1265         }
1266         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1267                 tag = PAGECACHE_TAG_TOWRITE;
1268         else
1269                 tag = PAGECACHE_TAG_DIRTY;
1270 retry:
1271         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1272                 tag_pages_for_writeback(mapping, index, end);
1273         done_index = index;
1274         while (!done && (index <= end)) {
1275                 int i;
1276
1277                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1278                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1279                 if (nr_pages == 0)
1280                         break;
1281
1282                 for (i = 0; i < nr_pages; i++) {
1283                         struct page *page = pvec.pages[i];
1284
1285                         if (page->index > end) {
1286                                 done = 1;
1287                                 break;
1288                         }
1289
1290                         done_index = page->index;
1291
1292                         lock_page(page);
1293
1294                         if (unlikely(page->mapping != mapping)) {
1295 continue_unlock:
1296                                 unlock_page(page);
1297                                 continue;
1298                         }
1299
1300                         if (!PageDirty(page)) {
1301                                 /* someone wrote it for us */
1302                                 goto continue_unlock;
1303                         }
1304
1305                         if (step == is_cold_data(page))
1306                                 goto continue_unlock;
1307
1308                         if (PageWriteback(page)) {
1309                                 if (wbc->sync_mode != WB_SYNC_NONE)
1310                                         f2fs_wait_on_page_writeback(page,
1311                                                                 DATA, true);
1312                                 else
1313                                         goto continue_unlock;
1314                         }
1315
1316                         BUG_ON(PageWriteback(page));
1317                         if (!clear_page_dirty_for_io(page))
1318                                 goto continue_unlock;
1319
1320                         ret = (*writepage)(page, wbc, data);
1321                         if (unlikely(ret)) {
1322                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1323                                         unlock_page(page);
1324                                         ret = 0;
1325                                 } else {
1326                                         done_index = page->index + 1;
1327                                         done = 1;
1328                                         break;
1329                                 }
1330                         }
1331
1332                         if (--wbc->nr_to_write <= 0 &&
1333                             wbc->sync_mode == WB_SYNC_NONE) {
1334                                 done = 1;
1335                                 break;
1336                         }
1337                 }
1338                 pagevec_release(&pvec);
1339                 cond_resched();
1340         }
1341
1342         if (step < 1) {
1343                 step++;
1344                 goto next;
1345         }
1346
1347         if (!cycled && !done) {
1348                 cycled = 1;
1349                 index = 0;
1350                 end = writeback_index - 1;
1351                 goto retry;
1352         }
1353         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1354                 mapping->writeback_index = done_index;
1355
1356         return ret;
1357 }
1358
1359 static int f2fs_write_data_pages(struct address_space *mapping,
1360                             struct writeback_control *wbc)
1361 {
1362         struct inode *inode = mapping->host;
1363         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1364         bool locked = false;
1365         int ret;
1366         long diff;
1367
1368         /* deal with chardevs and other special file */
1369         if (!mapping->a_ops->writepage)
1370                 return 0;
1371
1372         /* skip writing if there is no dirty page in this inode */
1373         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1374                 return 0;
1375
1376         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1377                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1378                         available_free_memory(sbi, DIRTY_DENTS))
1379                 goto skip_write;
1380
1381         /* skip writing during file defragment */
1382         if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1383                 goto skip_write;
1384
1385         /* during POR, we don't need to trigger writepage at all. */
1386         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1387                 goto skip_write;
1388
1389         trace_f2fs_writepages(mapping->host, wbc, DATA);
1390
1391         diff = nr_pages_to_write(sbi, DATA, wbc);
1392
1393         if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1394                 mutex_lock(&sbi->writepages);
1395                 locked = true;
1396         }
1397         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1398         f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1399         if (locked)
1400                 mutex_unlock(&sbi->writepages);
1401
1402         remove_dirty_inode(inode);
1403
1404         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1405         return ret;
1406
1407 skip_write:
1408         wbc->pages_skipped += get_dirty_pages(inode);
1409         trace_f2fs_writepages(mapping->host, wbc, DATA);
1410         return 0;
1411 }
1412
1413 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1414 {
1415         struct inode *inode = mapping->host;
1416         loff_t i_size = i_size_read(inode);
1417
1418         if (to > i_size) {
1419                 truncate_pagecache(inode, i_size);
1420                 truncate_blocks(inode, i_size, true);
1421         }
1422 }
1423
1424 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1425                         struct page *page, loff_t pos, unsigned len,
1426                         block_t *blk_addr, bool *node_changed)
1427 {
1428         struct inode *inode = page->mapping->host;
1429         pgoff_t index = page->index;
1430         struct dnode_of_data dn;
1431         struct page *ipage;
1432         bool locked = false;
1433         struct extent_info ei;
1434         int err = 0;
1435
1436         /*
1437          * we already allocated all the blocks, so we don't need to get
1438          * the block addresses when there is no need to fill the page.
1439          */
1440         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1441                                         len == PAGE_CACHE_SIZE)
1442                 return 0;
1443
1444         if (f2fs_has_inline_data(inode) ||
1445                         (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1446                 f2fs_lock_op(sbi);
1447                 locked = true;
1448         }
1449 restart:
1450         /* check inline_data */
1451         ipage = get_node_page(sbi, inode->i_ino);
1452         if (IS_ERR(ipage)) {
1453                 err = PTR_ERR(ipage);
1454                 goto unlock_out;
1455         }
1456
1457         set_new_dnode(&dn, inode, ipage, ipage, 0);
1458
1459         if (f2fs_has_inline_data(inode)) {
1460                 if (pos + len <= MAX_INLINE_DATA) {
1461                         read_inline_data(page, ipage);
1462                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1463                         set_inline_node(ipage);
1464                 } else {
1465                         err = f2fs_convert_inline_page(&dn, page);
1466                         if (err)
1467                                 goto out;
1468                         if (dn.data_blkaddr == NULL_ADDR)
1469                                 err = f2fs_get_block(&dn, index);
1470                 }
1471         } else if (locked) {
1472                 err = f2fs_get_block(&dn, index);
1473         } else {
1474                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1475                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1476                 } else {
1477                         /* hole case */
1478                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1479                         if (err || (!err && dn.data_blkaddr == NULL_ADDR)) {
1480                                 f2fs_put_dnode(&dn);
1481                                 f2fs_lock_op(sbi);
1482                                 locked = true;
1483                                 goto restart;
1484                         }
1485                 }
1486         }
1487
1488         /* convert_inline_page can make node_changed */
1489         *blk_addr = dn.data_blkaddr;
1490         *node_changed = dn.node_changed;
1491 out:
1492         f2fs_put_dnode(&dn);
1493 unlock_out:
1494         if (locked)
1495                 f2fs_unlock_op(sbi);
1496         return err;
1497 }
1498
1499 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1500                 loff_t pos, unsigned len, unsigned flags,
1501                 struct page **pagep, void **fsdata)
1502 {
1503         struct inode *inode = mapping->host;
1504         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1505         struct page *page = NULL;
1506         pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1507         bool need_balance = false;
1508         block_t blkaddr = NULL_ADDR;
1509         int err = 0;
1510
1511         trace_f2fs_write_begin(inode, pos, len, flags);
1512
1513         /*
1514          * We should check this at this moment to avoid deadlock on inode page
1515          * and #0 page. The locking rule for inline_data conversion should be:
1516          * lock_page(page #0) -> lock_page(inode_page)
1517          */
1518         if (index != 0) {
1519                 err = f2fs_convert_inline_inode(inode);
1520                 if (err)
1521                         goto fail;
1522         }
1523 repeat:
1524         page = grab_cache_page_write_begin(mapping, index, flags);
1525         if (!page) {
1526                 err = -ENOMEM;
1527                 goto fail;
1528         }
1529
1530         *pagep = page;
1531
1532         err = prepare_write_begin(sbi, page, pos, len,
1533                                         &blkaddr, &need_balance);
1534         if (err)
1535                 goto fail;
1536
1537         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1538                 unlock_page(page);
1539                 f2fs_balance_fs(sbi, true);
1540                 lock_page(page);
1541                 if (page->mapping != mapping) {
1542                         /* The page got truncated from under us */
1543                         f2fs_put_page(page, 1);
1544                         goto repeat;
1545                 }
1546         }
1547
1548         f2fs_wait_on_page_writeback(page, DATA, false);
1549
1550         /* wait for GCed encrypted page writeback */
1551         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1552                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1553
1554         if (len == PAGE_CACHE_SIZE)
1555                 goto out_update;
1556         if (PageUptodate(page))
1557                 goto out_clear;
1558
1559         if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1560                 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1561                 unsigned end = start + len;
1562
1563                 /* Reading beyond i_size is simple: memset to zero */
1564                 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1565                 goto out_update;
1566         }
1567
1568         if (blkaddr == NEW_ADDR) {
1569                 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1570         } else {
1571                 struct f2fs_io_info fio = {
1572                         .sbi = sbi,
1573                         .type = DATA,
1574                         .rw = READ_SYNC,
1575                         .blk_addr = blkaddr,
1576                         .page = page,
1577                         .encrypted_page = NULL,
1578                 };
1579                 err = f2fs_submit_page_bio(&fio);
1580                 if (err)
1581                         goto fail;
1582
1583                 lock_page(page);
1584                 if (unlikely(!PageUptodate(page))) {
1585                         err = -EIO;
1586                         goto fail;
1587                 }
1588                 if (unlikely(page->mapping != mapping)) {
1589                         f2fs_put_page(page, 1);
1590                         goto repeat;
1591                 }
1592
1593                 /* avoid symlink page */
1594                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1595                         err = f2fs_decrypt_one(inode, page);
1596                         if (err)
1597                                 goto fail;
1598                 }
1599         }
1600 out_update:
1601         SetPageUptodate(page);
1602 out_clear:
1603         clear_cold_data(page);
1604         return 0;
1605
1606 fail:
1607         f2fs_put_page(page, 1);
1608         f2fs_write_failed(mapping, pos + len);
1609         return err;
1610 }
1611
1612 static int f2fs_write_end(struct file *file,
1613                         struct address_space *mapping,
1614                         loff_t pos, unsigned len, unsigned copied,
1615                         struct page *page, void *fsdata)
1616 {
1617         struct inode *inode = page->mapping->host;
1618
1619         trace_f2fs_write_end(inode, pos, len, copied);
1620
1621         set_page_dirty(page);
1622
1623         if (pos + copied > i_size_read(inode)) {
1624                 i_size_write(inode, pos + copied);
1625                 mark_inode_dirty(inode);
1626         }
1627
1628         f2fs_put_page(page, 1);
1629         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1630         return copied;
1631 }
1632
1633 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1634                            loff_t offset)
1635 {
1636         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1637
1638         if (offset & blocksize_mask)
1639                 return -EINVAL;
1640
1641         if (iov_iter_alignment(iter) & blocksize_mask)
1642                 return -EINVAL;
1643
1644         return 0;
1645 }
1646
1647 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1648                               loff_t offset)
1649 {
1650         struct address_space *mapping = iocb->ki_filp->f_mapping;
1651         struct inode *inode = mapping->host;
1652         size_t count = iov_iter_count(iter);
1653         int err;
1654
1655         err = check_direct_IO(inode, iter, offset);
1656         if (err)
1657                 return err;
1658
1659         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1660                 return 0;
1661
1662         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1663
1664         err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1665         if (err < 0 && iov_iter_rw(iter) == WRITE)
1666                 f2fs_write_failed(mapping, offset + count);
1667
1668         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1669
1670         return err;
1671 }
1672
1673 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1674                                                         unsigned int length)
1675 {
1676         struct inode *inode = page->mapping->host;
1677         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1678
1679         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1680                 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1681                 return;
1682
1683         if (PageDirty(page)) {
1684                 if (inode->i_ino == F2FS_META_INO(sbi))
1685                         dec_page_count(sbi, F2FS_DIRTY_META);
1686                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1687                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1688                 else
1689                         inode_dec_dirty_pages(inode);
1690         }
1691
1692         /* This is atomic written page, keep Private */
1693         if (IS_ATOMIC_WRITTEN_PAGE(page))
1694                 return;
1695
1696         ClearPagePrivate(page);
1697 }
1698
1699 int f2fs_release_page(struct page *page, gfp_t wait)
1700 {
1701         /* If this is dirty page, keep PagePrivate */
1702         if (PageDirty(page))
1703                 return 0;
1704
1705         /* This is atomic written page, keep Private */
1706         if (IS_ATOMIC_WRITTEN_PAGE(page))
1707                 return 0;
1708
1709         ClearPagePrivate(page);
1710         return 1;
1711 }
1712
1713 static int f2fs_set_data_page_dirty(struct page *page)
1714 {
1715         struct address_space *mapping = page->mapping;
1716         struct inode *inode = mapping->host;
1717
1718         trace_f2fs_set_page_dirty(page, DATA);
1719
1720         SetPageUptodate(page);
1721
1722         if (f2fs_is_atomic_file(inode)) {
1723                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1724                         register_inmem_page(inode, page);
1725                         return 1;
1726                 }
1727                 /*
1728                  * Previously, this page has been registered, we just
1729                  * return here.
1730                  */
1731                 return 0;
1732         }
1733
1734         if (!PageDirty(page)) {
1735                 __set_page_dirty_nobuffers(page);
1736                 update_dirty_page(inode, page);
1737                 return 1;
1738         }
1739         return 0;
1740 }
1741
1742 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1743 {
1744         struct inode *inode = mapping->host;
1745
1746         if (f2fs_has_inline_data(inode))
1747                 return 0;
1748
1749         /* make sure allocating whole blocks */
1750         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1751                 filemap_write_and_wait(mapping);
1752
1753         return generic_block_bmap(mapping, block, get_data_block_bmap);
1754 }
1755
1756 const struct address_space_operations f2fs_dblock_aops = {
1757         .readpage       = f2fs_read_data_page,
1758         .readpages      = f2fs_read_data_pages,
1759         .writepage      = f2fs_write_data_page,
1760         .writepages     = f2fs_write_data_pages,
1761         .write_begin    = f2fs_write_begin,
1762         .write_end      = f2fs_write_end,
1763         .set_page_dirty = f2fs_set_data_page_dirty,
1764         .invalidatepage = f2fs_invalidate_page,
1765         .releasepage    = f2fs_release_page,
1766         .direct_IO      = f2fs_direct_IO,
1767         .bmap           = f2fs_bmap,
1768 };