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, true);
  72                 }
  73                 end_page_writeback(page);
  74         }
  75         if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
  76                                 wq_has_sleeper(&sbi->cp_wait))
  77                 wake_up(&sbi->cp_wait);
  78
  79         bio_put(bio);
  80 }
  81
  82 /*
  83  * Low-level block read/write IO operations.
  84  */
  85 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  86                                 int npages, bool is_read)
  87 {
  88         struct bio *bio;
  89
  90         bio = f2fs_bio_alloc(npages);
  91
  92         bio->bi_bdev = sbi->sb->s_bdev;
  93         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  94         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  95         bio->bi_private = is_read ? NULL : sbi;
  96
  97         return bio;
  98 }
  99
 100 static inline void __submit_bio(struct f2fs_sb_info *sbi, int rw,
 101                                                 struct bio *bio)
 102 {
 103         if (!is_read_io(rw))
 104                 atomic_inc(&sbi->nr_wb_bios);
 105         submit_bio(rw, bio);
 106 }
 107
 108 static void __submit_merged_bio(struct f2fs_bio_info *io)
 109 {
 110         struct f2fs_io_info *fio = &io->fio;
 111
 112         if (!io->bio)
 113                 return;
 114
 115         if (is_read_io(fio->rw))
 116                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 117         else
 118                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 119
 120         __submit_bio(io->sbi, fio->rw, io->bio);
 121         io->bio = NULL;
 122 }
 123
 124 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 125                                                 struct page *page, nid_t ino)
 126 {
 127         struct bio_vec *bvec;
 128         struct page *target;
 129         int i;
 130
 131         if (!io->bio)
 132                 return false;
 133
 134         if (!inode && !page && !ino)
 135                 return true;
 136
 137         bio_for_each_segment_all(bvec, io->bio, i) {
 138
 139                 if (bvec->bv_page->mapping)
 140                         target = bvec->bv_page;
 141                 else
 142                         target = fscrypt_control_page(bvec->bv_page);
 143
 144                 if (inode && inode == target->mapping->host)
 145                         return true;
 146                 if (page && page == target)
 147                         return true;
 148                 if (ino && ino == ino_of_node(target))
 149                         return true;
 150         }
 151
 152         return false;
 153 }
 154
 155 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 156                                                 struct page *page, nid_t ino,
 157                                                 enum page_type type)
 158 {
 159         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 160         struct f2fs_bio_info *io = &sbi->write_io[btype];
 161         bool ret;
 162
 163         down_read(&io->io_rwsem);
 164         ret = __has_merged_page(io, inode, page, ino);
 165         up_read(&io->io_rwsem);
 166         return ret;
 167 }
 168
 169 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 170                                 struct inode *inode, struct page *page,
 171                                 nid_t ino, enum page_type type, int rw)
 172 {
 173         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 174         struct f2fs_bio_info *io;
 175
 176         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 177
 178         down_write(&io->io_rwsem);
 179
 180         if (!__has_merged_page(io, inode, page, ino))
 181                 goto out;
 182
 183         /* change META to META_FLUSH in the checkpoint procedure */
 184         if (type >= META_FLUSH) {
 185                 io->fio.type = META_FLUSH;
 186                 if (test_opt(sbi, NOBARRIER))
 187                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 188                 else
 189                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 190         }
 191         __submit_merged_bio(io);
 192 out:
 193         up_write(&io->io_rwsem);
 194 }
 195
 196 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 197                                                                         int rw)
 198 {
 199         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 200 }
 201
 202 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 203                                 struct inode *inode, struct page *page,
 204                                 nid_t ino, enum page_type type, int rw)
 205 {
 206         if (has_merged_page(sbi, inode, page, ino, type))
 207                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 208 }
 209
 210 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 211 {
 212         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 213         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 214         f2fs_submit_merged_bio(sbi, META, WRITE);
 215 }
 216
 217 /*
 218  * Fill the locked page with data located in the block address.
 219  * Return unlocked page.
 220  */
 221 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 222 {
 223         struct bio *bio;
 224         struct page *page = fio->encrypted_page ?
 225                         fio->encrypted_page : fio->page;
 226
 227         trace_f2fs_submit_page_bio(page, fio);
 228         f2fs_trace_ios(fio, 0);
 229
 230         /* Allocate a new bio */
 231         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
 232
 233         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 234                 bio_put(bio);
 235                 return -EFAULT;
 236         }
 237
 238         __submit_bio(fio->sbi, fio->rw, bio);
 239         return 0;
 240 }
 241
 242 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 243 {
 244         struct f2fs_sb_info *sbi = fio->sbi;
 245         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 246         struct f2fs_bio_info *io;
 247         bool is_read = is_read_io(fio->rw);
 248         struct page *bio_page;
 249
 250         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 251
 252         if (fio->old_blkaddr != NEW_ADDR)
 253                 verify_block_addr(sbi, fio->old_blkaddr);
 254         verify_block_addr(sbi, fio->new_blkaddr);
 255
 256         down_write(&io->io_rwsem);
 257
 258         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 259                                                 io->fio.rw != fio->rw))
 260                 __submit_merged_bio(io);
 261 alloc_new:
 262         if (io->bio == NULL) {
 263                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 264
 265                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 266                                                 bio_blocks, is_read);
 267                 io->fio = *fio;
 268         }
 269
 270         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 271
 272         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 273                                                         PAGE_SIZE) {
 274                 __submit_merged_bio(io);
 275                 goto alloc_new;
 276         }
 277
 278         io->last_block_in_bio = fio->new_blkaddr;
 279         f2fs_trace_ios(fio, 0);
 280
 281         up_write(&io->io_rwsem);
 282         trace_f2fs_submit_page_mbio(fio->page, fio);
 283 }
 284
 285 static void __set_data_blkaddr(struct dnode_of_data *dn)
 286 {
 287         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 288         __le32 *addr_array;
 289
 290         /* Get physical address of data block */
 291         addr_array = blkaddr_in_node(rn);
 292         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 293 }
 294
 295 /*
 296  * Lock ordering for the change of data block address:
 297  * ->data_page
 298  *  ->node_page
 299  *    update block addresses in the node page
 300  */
 301 void set_data_blkaddr(struct dnode_of_data *dn)
 302 {
 303         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 304         __set_data_blkaddr(dn);
 305         if (set_page_dirty(dn->node_page))
 306                 dn->node_changed = true;
 307 }
 308
 309 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 310 {
 311         dn->data_blkaddr = blkaddr;
 312         set_data_blkaddr(dn);
 313         f2fs_update_extent_cache(dn);
 314 }
 315
 316 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 317 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 318 {
 319         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 320
 321         if (!count)
 322                 return 0;
 323
 324         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 325                 return -EPERM;
 326         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 327                 return -ENOSPC;
 328
 329         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 330                                                 dn->ofs_in_node, count);
 331
 332         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 333
 334         for (; count > 0; dn->ofs_in_node++) {
 335                 block_t blkaddr =
 336                         datablock_addr(dn->node_page, dn->ofs_in_node);
 337                 if (blkaddr == NULL_ADDR) {
 338                         dn->data_blkaddr = NEW_ADDR;
 339                         __set_data_blkaddr(dn);
 340                         count--;
 341                 }
 342         }
 343
 344         if (set_page_dirty(dn->node_page))
 345                 dn->node_changed = true;
 346         return 0;
 347 }
 348
 349 /* Should keep dn->ofs_in_node unchanged */
 350 int reserve_new_block(struct dnode_of_data *dn)
 351 {
 352         unsigned int ofs_in_node = dn->ofs_in_node;
 353         int ret;
 354
 355         ret = reserve_new_blocks(dn, 1);
 356         dn->ofs_in_node = ofs_in_node;
 357         return ret;
 358 }
 359
 360 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 361 {
 362         bool need_put = dn->inode_page ? false : true;
 363         int err;
 364
 365         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 366         if (err)
 367                 return err;
 368
 369         if (dn->data_blkaddr == NULL_ADDR)
 370                 err = reserve_new_block(dn);
 371         if (err || need_put)
 372                 f2fs_put_dnode(dn);
 373         return err;
 374 }
 375
 376 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 377 {
 378         struct extent_info ei;
 379         struct inode *inode = dn->inode;
 380
 381         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 382                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 383                 return 0;
 384         }
 385
 386         return f2fs_reserve_block(dn, index);
 387 }
 388
 389 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 390                                                 int rw, bool for_write)
 391 {
 392         struct address_space *mapping = inode->i_mapping;
 393         struct dnode_of_data dn;
 394         struct page *page;
 395         struct extent_info ei;
 396         int err;
 397         struct f2fs_io_info fio = {
 398                 .sbi = F2FS_I_SB(inode),
 399                 .type = DATA,
 400                 .rw = rw,
 401                 .encrypted_page = NULL,
 402         };
 403
 404         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 405                 return read_mapping_page(mapping, index, NULL);
 406
 407         page = f2fs_grab_cache_page(mapping, index, for_write);
 408         if (!page)
 409                 return ERR_PTR(-ENOMEM);
 410
 411         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 412                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 413                 goto got_it;
 414         }
 415
 416         set_new_dnode(&dn, inode, NULL, NULL, 0);
 417         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 418         if (err)
 419                 goto put_err;
 420         f2fs_put_dnode(&dn);
 421
 422         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 423                 err = -ENOENT;
 424                 goto put_err;
 425         }
 426 got_it:
 427         if (PageUptodate(page)) {
 428                 unlock_page(page);
 429                 return page;
 430         }
 431
 432         /*
 433          * A new dentry page is allocated but not able to be written, since its
 434          * new inode page couldn't be allocated due to -ENOSPC.
 435          * In such the case, its blkaddr can be remained as NEW_ADDR.
 436          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 437          */
 438         if (dn.data_blkaddr == NEW_ADDR) {
 439                 zero_user_segment(page, 0, PAGE_SIZE);
 440                 SetPageUptodate(page);
 441                 unlock_page(page);
 442                 return page;
 443         }
 444
 445         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 446         fio.page = page;
 447         err = f2fs_submit_page_bio(&fio);
 448         if (err)
 449                 goto put_err;
 450         return page;
 451
 452 put_err:
 453         f2fs_put_page(page, 1);
 454         return ERR_PTR(err);
 455 }
 456
 457 struct page *find_data_page(struct inode *inode, pgoff_t index)
 458 {
 459         struct address_space *mapping = inode->i_mapping;
 460         struct page *page;
 461
 462         page = find_get_page(mapping, index);
 463         if (page && PageUptodate(page))
 464                 return page;
 465         f2fs_put_page(page, 0);
 466
 467         page = get_read_data_page(inode, index, READ_SYNC, false);
 468         if (IS_ERR(page))
 469                 return page;
 470
 471         if (PageUptodate(page))
 472                 return page;
 473
 474         wait_on_page_locked(page);
 475         if (unlikely(!PageUptodate(page))) {
 476                 f2fs_put_page(page, 0);
 477                 return ERR_PTR(-EIO);
 478         }
 479         return page;
 480 }
 481
 482 /*
 483  * If it tries to access a hole, return an error.
 484  * Because, the callers, functions in dir.c and GC, should be able to know
 485  * whether this page exists or not.
 486  */
 487 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 488                                                         bool for_write)
 489 {
 490         struct address_space *mapping = inode->i_mapping;
 491         struct page *page;
 492 repeat:
 493         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 494         if (IS_ERR(page))
 495                 return page;
 496
 497         /* wait for read completion */
 498         lock_page(page);
 499         if (unlikely(!PageUptodate(page))) {
 500                 f2fs_put_page(page, 1);
 501                 return ERR_PTR(-EIO);
 502         }
 503         if (unlikely(page->mapping != mapping)) {
 504                 f2fs_put_page(page, 1);
 505                 goto repeat;
 506         }
 507         return page;
 508 }
 509
 510 /*
 511  * Caller ensures that this data page is never allocated.
 512  * A new zero-filled data page is allocated in the page cache.
 513  *
 514  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 515  * f2fs_unlock_op().
 516  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 517  * ipage should be released by this function.
 518  */
 519 struct page *get_new_data_page(struct inode *inode,
 520                 struct page *ipage, pgoff_t index, bool new_i_size)
 521 {
 522         struct address_space *mapping = inode->i_mapping;
 523         struct page *page;
 524         struct dnode_of_data dn;
 525         int err;
 526
 527         page = f2fs_grab_cache_page(mapping, index, true);
 528         if (!page) {
 529                 /*
 530                  * before exiting, we should make sure ipage will be released
 531                  * if any error occur.
 532                  */
 533                 f2fs_put_page(ipage, 1);
 534                 return ERR_PTR(-ENOMEM);
 535         }
 536
 537         set_new_dnode(&dn, inode, ipage, NULL, 0);
 538         err = f2fs_reserve_block(&dn, index);
 539         if (err) {
 540                 f2fs_put_page(page, 1);
 541                 return ERR_PTR(err);
 542         }
 543         if (!ipage)
 544                 f2fs_put_dnode(&dn);
 545
 546         if (PageUptodate(page))
 547                 goto got_it;
 548
 549         if (dn.data_blkaddr == NEW_ADDR) {
 550                 zero_user_segment(page, 0, PAGE_SIZE);
 551                 SetPageUptodate(page);
 552         } else {
 553                 f2fs_put_page(page, 1);
 554
 555                 /* if ipage exists, blkaddr should be NEW_ADDR */
 556                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 557                 page = get_lock_data_page(inode, index, true);
 558                 if (IS_ERR(page))
 559                         return page;
 560         }
 561 got_it:
 562         if (new_i_size && i_size_read(inode) <
 563                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 564                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 565         return page;
 566 }
 567
 568 static int __allocate_data_block(struct dnode_of_data *dn)
 569 {
 570         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 571         struct f2fs_summary sum;
 572         struct node_info ni;
 573         int seg = CURSEG_WARM_DATA;
 574         pgoff_t fofs;
 575         blkcnt_t count = 1;
 576
 577         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 578                 return -EPERM;
 579
 580         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 581         if (dn->data_blkaddr == NEW_ADDR)
 582                 goto alloc;
 583
 584         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 585                 return -ENOSPC;
 586
 587 alloc:
 588         get_node_info(sbi, dn->nid, &ni);
 589         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 590
 591         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 592                 seg = CURSEG_DIRECT_IO;
 593
 594         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 595                                                                 &sum, seg);
 596         set_data_blkaddr(dn);
 597
 598         /* update i_size */
 599         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 600                                                         dn->ofs_in_node;
 601         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 602                 f2fs_i_size_write(dn->inode,
 603                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 604         return 0;
 605 }
 606
 607 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 608 {
 609         struct inode *inode = file_inode(iocb->ki_filp);
 610         struct f2fs_map_blocks map;
 611         ssize_t ret = 0;
 612
 613         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 614         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 615         map.m_next_pgofs = NULL;
 616
 617         if (f2fs_encrypted_inode(inode))
 618                 return 0;
 619
 620         if (iocb->ki_flags & IOCB_DIRECT) {
 621                 ret = f2fs_convert_inline_inode(inode);
 622                 if (ret)
 623                         return ret;
 624                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 625         }
 626         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 627                 ret = f2fs_convert_inline_inode(inode);
 628                 if (ret)
 629                         return ret;
 630         }
 631         if (!f2fs_has_inline_data(inode))
 632                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 633         return ret;
 634 }
 635
 636 /*
 637  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 638  * f2fs_map_blocks structure.
 639  * If original data blocks are allocated, then give them to blockdev.
 640  * Otherwise,
 641  *     a. preallocate requested block addresses
 642  *     b. do not use extent cache for better performance
 643  *     c. give the block addresses to blockdev
 644  */
 645 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 646                                                 int create, int flag)
 647 {
 648         unsigned int maxblocks = map->m_len;
 649         struct dnode_of_data dn;
 650         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 651         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 652         pgoff_t pgofs, end_offset, end;
 653         int err = 0, ofs = 1;
 654         unsigned int ofs_in_node, last_ofs_in_node;
 655         blkcnt_t prealloc;
 656         struct extent_info ei;
 657         bool allocated = false;
 658         block_t blkaddr;
 659
 660         map->m_len = 0;
 661         map->m_flags = 0;
 662
 663         /* it only supports block size == page size */
 664         pgofs = (pgoff_t)map->m_lblk;
 665         end = pgofs + maxblocks;
 666
 667         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 668                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 669                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 670                 map->m_flags = F2FS_MAP_MAPPED;
 671                 goto out;
 672         }
 673
 674 next_dnode:
 675         if (create)
 676                 f2fs_lock_op(sbi);
 677
 678         /* When reading holes, we need its node page */
 679         set_new_dnode(&dn, inode, NULL, NULL, 0);
 680         err = get_dnode_of_data(&dn, pgofs, mode);
 681         if (err) {
 682                 if (flag == F2FS_GET_BLOCK_BMAP)
 683                         map->m_pblk = 0;
 684                 if (err == -ENOENT) {
 685                         err = 0;
 686                         if (map->m_next_pgofs)
 687                                 *map->m_next_pgofs =
 688                                         get_next_page_offset(&dn, pgofs);
 689                 }
 690                 goto unlock_out;
 691         }
 692
 693         prealloc = 0;
 694         ofs_in_node = dn.ofs_in_node;
 695         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 696
 697 next_block:
 698         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 699
 700         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 701                 if (create) {
 702                         if (unlikely(f2fs_cp_error(sbi))) {
 703                                 err = -EIO;
 704                                 goto sync_out;
 705                         }
 706                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 707                                 if (blkaddr == NULL_ADDR) {
 708                                         prealloc++;
 709                                         last_ofs_in_node = dn.ofs_in_node;
 710                                 }
 711                         } else {
 712                                 err = __allocate_data_block(&dn);
 713                                 if (!err) {
 714                                         set_inode_flag(inode, FI_APPEND_WRITE);
 715                                         allocated = true;
 716                                 }
 717                         }
 718                         if (err)
 719                                 goto sync_out;
 720                         map->m_flags = F2FS_MAP_NEW;
 721                         blkaddr = dn.data_blkaddr;
 722                 } else {
 723                         if (flag == F2FS_GET_BLOCK_BMAP) {
 724                                 map->m_pblk = 0;
 725                                 goto sync_out;
 726                         }
 727                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 728                                                 blkaddr == NULL_ADDR) {
 729                                 if (map->m_next_pgofs)
 730                                         *map->m_next_pgofs = pgofs + 1;
 731                         }
 732                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 733                                                 blkaddr != NEW_ADDR)
 734                                 goto sync_out;
 735                 }
 736         }
 737
 738         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 739                 goto skip;
 740
 741         if (map->m_len == 0) {
 742                 /* preallocated unwritten block should be mapped for fiemap. */
 743                 if (blkaddr == NEW_ADDR)
 744                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 745                 map->m_flags |= F2FS_MAP_MAPPED;
 746
 747                 map->m_pblk = blkaddr;
 748                 map->m_len = 1;
 749         } else if ((map->m_pblk != NEW_ADDR &&
 750                         blkaddr == (map->m_pblk + ofs)) ||
 751                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 752                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 753                 ofs++;
 754                 map->m_len++;
 755         } else {
 756                 goto sync_out;
 757         }
 758
 759 skip:
 760         dn.ofs_in_node++;
 761         pgofs++;
 762
 763         /* preallocate blocks in batch for one dnode page */
 764         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 765                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 766
 767                 dn.ofs_in_node = ofs_in_node;
 768                 err = reserve_new_blocks(&dn, prealloc);
 769                 if (err)
 770                         goto sync_out;
 771
 772                 map->m_len += dn.ofs_in_node - ofs_in_node;
 773                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 774                         err = -ENOSPC;
 775                         goto sync_out;
 776                 }
 777                 dn.ofs_in_node = end_offset;
 778         }
 779
 780         if (pgofs >= end)
 781                 goto sync_out;
 782         else if (dn.ofs_in_node < end_offset)
 783                 goto next_block;
 784
 785         f2fs_put_dnode(&dn);
 786
 787         if (create) {
 788                 f2fs_unlock_op(sbi);
 789                 f2fs_balance_fs(sbi, allocated);
 790         }
 791         allocated = false;
 792         goto next_dnode;
 793
 794 sync_out:
 795         f2fs_put_dnode(&dn);
 796 unlock_out:
 797         if (create) {
 798                 f2fs_unlock_op(sbi);
 799                 f2fs_balance_fs(sbi, allocated);
 800         }
 801 out:
 802         trace_f2fs_map_blocks(inode, map, err);
 803         return err;
 804 }
 805
 806 static int __get_data_block(struct inode *inode, sector_t iblock,
 807                         struct buffer_head *bh, int create, int flag,
 808                         pgoff_t *next_pgofs)
 809 {
 810         struct f2fs_map_blocks map;
 811         int ret;
 812
 813         map.m_lblk = iblock;
 814         map.m_len = bh->b_size >> inode->i_blkbits;
 815         map.m_next_pgofs = next_pgofs;
 816
 817         ret = f2fs_map_blocks(inode, &map, create, flag);
 818         if (!ret) {
 819                 map_bh(bh, inode->i_sb, map.m_pblk);
 820                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 821                 bh->b_size = map.m_len << inode->i_blkbits;
 822         }
 823         return ret;
 824 }
 825
 826 static int get_data_block(struct inode *inode, sector_t iblock,
 827                         struct buffer_head *bh_result, int create, int flag,
 828                         pgoff_t *next_pgofs)
 829 {
 830         return __get_data_block(inode, iblock, bh_result, create,
 831                                                         flag, next_pgofs);
 832 }
 833
 834 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 835                         struct buffer_head *bh_result, int create)
 836 {
 837         return __get_data_block(inode, iblock, bh_result, create,
 838                                                 F2FS_GET_BLOCK_DIO, NULL);
 839 }
 840
 841 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 842                         struct buffer_head *bh_result, int create)
 843 {
 844         /* Block number less than F2FS MAX BLOCKS */
 845         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 846                 return -EFBIG;
 847
 848         return __get_data_block(inode, iblock, bh_result, create,
 849                                                 F2FS_GET_BLOCK_BMAP, NULL);
 850 }
 851
 852 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 853 {
 854         return (offset >> inode->i_blkbits);
 855 }
 856
 857 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 858 {
 859         return (blk << inode->i_blkbits);
 860 }
 861
 862 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 863                 u64 start, u64 len)
 864 {
 865         struct buffer_head map_bh;
 866         sector_t start_blk, last_blk;
 867         pgoff_t next_pgofs;
 868         loff_t isize;
 869         u64 logical = 0, phys = 0, size = 0;
 870         u32 flags = 0;
 871         int ret = 0;
 872
 873         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 874         if (ret)
 875                 return ret;
 876
 877         if (f2fs_has_inline_data(inode)) {
 878                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 879                 if (ret != -EAGAIN)
 880                         return ret;
 881         }
 882
 883         inode_lock(inode);
 884
 885         isize = i_size_read(inode);
 886         if (start >= isize)
 887                 goto out;
 888
 889         if (start + len > isize)
 890                 len = isize - start;
 891
 892         if (logical_to_blk(inode, len) == 0)
 893                 len = blk_to_logical(inode, 1);
 894
 895         start_blk = logical_to_blk(inode, start);
 896         last_blk = logical_to_blk(inode, start + len - 1);
 897
 898 next:
 899         memset(&map_bh, 0, sizeof(struct buffer_head));
 900         map_bh.b_size = len;
 901
 902         ret = get_data_block(inode, start_blk, &map_bh, 0,
 903                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 904         if (ret)
 905                 goto out;
 906
 907         /* HOLE */
 908         if (!buffer_mapped(&map_bh)) {
 909                 start_blk = next_pgofs;
 910                 /* Go through holes util pass the EOF */
 911                 if (blk_to_logical(inode, start_blk) < isize)
 912                         goto prep_next;
 913                 /* Found a hole beyond isize means no more extents.
 914                  * Note that the premise is that filesystems don't
 915                  * punch holes beyond isize and keep size unchanged.
 916                  */
 917                 flags |= FIEMAP_EXTENT_LAST;
 918         }
 919
 920         if (size) {
 921                 if (f2fs_encrypted_inode(inode))
 922                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 923
 924                 ret = fiemap_fill_next_extent(fieinfo, logical,
 925                                 phys, size, flags);
 926         }
 927
 928         if (start_blk > last_blk || ret)
 929                 goto out;
 930
 931         logical = blk_to_logical(inode, start_blk);
 932         phys = blk_to_logical(inode, map_bh.b_blocknr);
 933         size = map_bh.b_size;
 934         flags = 0;
 935         if (buffer_unwritten(&map_bh))
 936                 flags = FIEMAP_EXTENT_UNWRITTEN;
 937
 938         start_blk += logical_to_blk(inode, size);
 939
 940 prep_next:
 941         cond_resched();
 942         if (fatal_signal_pending(current))
 943                 ret = -EINTR;
 944         else
 945                 goto next;
 946 out:
 947         if (ret == 1)
 948                 ret = 0;
 949
 950         inode_unlock(inode);
 951         return ret;
 952 }
 953
 954 /*
 955  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 956  * Major change was from block_size == page_size in f2fs by default.
 957  */
 958 static int f2fs_mpage_readpages(struct address_space *mapping,
 959                         struct list_head *pages, struct page *page,
 960                         unsigned nr_pages)
 961 {
 962         struct bio *bio = NULL;
 963         unsigned page_idx;
 964         sector_t last_block_in_bio = 0;
 965         struct inode *inode = mapping->host;
 966         const unsigned blkbits = inode->i_blkbits;
 967         const unsigned blocksize = 1 << blkbits;
 968         sector_t block_in_file;
 969         sector_t last_block;
 970         sector_t last_block_in_file;
 971         sector_t block_nr;
 972         struct block_device *bdev = inode->i_sb->s_bdev;
 973         struct f2fs_map_blocks map;
 974
 975         map.m_pblk = 0;
 976         map.m_lblk = 0;
 977         map.m_len = 0;
 978         map.m_flags = 0;
 979         map.m_next_pgofs = NULL;
 980
 981         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 982
 983                 prefetchw(&page->flags);
 984                 if (pages) {
 985                         page = list_entry(pages->prev, struct page, lru);
 986                         list_del(&page->lru);
 987                         if (add_to_page_cache_lru(page, mapping,
 988                                                   page->index, GFP_KERNEL))
 989                                 goto next_page;
 990                 }
 991
 992                 block_in_file = (sector_t)page->index;
 993                 last_block = block_in_file + nr_pages;
 994                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 995                                                                 blkbits;
 996                 if (last_block > last_block_in_file)
 997                         last_block = last_block_in_file;
 998
 999                 /*
1000                  * Map blocks using the previous result first.
1001                  */
1002                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1003                                 block_in_file > map.m_lblk &&
1004                                 block_in_file < (map.m_lblk + map.m_len))
1005                         goto got_it;
1006
1007                 /*
1008                  * Then do more f2fs_map_blocks() calls until we are
1009                  * done with this page.
1010                  */
1011                 map.m_flags = 0;
1012
1013                 if (block_in_file < last_block) {
1014                         map.m_lblk = block_in_file;
1015                         map.m_len = last_block - block_in_file;
1016
1017                         if (f2fs_map_blocks(inode, &map, 0,
1018                                                 F2FS_GET_BLOCK_READ))
1019                                 goto set_error_page;
1020                 }
1021 got_it:
1022                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1023                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1024                         SetPageMappedToDisk(page);
1025
1026                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1027                                 SetPageUptodate(page);
1028                                 goto confused;
1029                         }
1030                 } else {
1031                         zero_user_segment(page, 0, PAGE_SIZE);
1032                         SetPageUptodate(page);
1033                         unlock_page(page);
1034                         goto next_page;
1035                 }
1036
1037                 /*
1038                  * This page will go to BIO.  Do we need to send this
1039                  * BIO off first?
1040                  */
1041                 if (bio && (last_block_in_bio != block_nr - 1)) {
1042 submit_and_realloc:
1043                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1044                         bio = NULL;
1045                 }
1046                 if (bio == NULL) {
1047                         struct fscrypt_ctx *ctx = NULL;
1048
1049                         if (f2fs_encrypted_inode(inode) &&
1050                                         S_ISREG(inode->i_mode)) {
1051
1052                                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1053                                 if (IS_ERR(ctx))
1054                                         goto set_error_page;
1055
1056                                 /* wait the page to be moved by cleaning */
1057                                 f2fs_wait_on_encrypted_page_writeback(
1058                                                 F2FS_I_SB(inode), block_nr);
1059                         }
1060
1061                         bio = bio_alloc(GFP_KERNEL,
1062                                 min_t(int, nr_pages, BIO_MAX_PAGES));
1063                         if (!bio) {
1064                                 if (ctx)
1065                                         fscrypt_release_ctx(ctx);
1066                                 goto set_error_page;
1067                         }
1068                         bio->bi_bdev = bdev;
1069                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1070                         bio->bi_end_io = f2fs_read_end_io;
1071                         bio->bi_private = ctx;
1072                 }
1073
1074                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1075                         goto submit_and_realloc;
1076
1077                 last_block_in_bio = block_nr;
1078                 goto next_page;
1079 set_error_page:
1080                 SetPageError(page);
1081                 zero_user_segment(page, 0, PAGE_SIZE);
1082                 unlock_page(page);
1083                 goto next_page;
1084 confused:
1085                 if (bio) {
1086                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1087                         bio = NULL;
1088                 }
1089                 unlock_page(page);
1090 next_page:
1091                 if (pages)
1092                         put_page(page);
1093         }
1094         BUG_ON(pages && !list_empty(pages));
1095         if (bio)
1096                 __submit_bio(F2FS_I_SB(inode), READ, bio);
1097         return 0;
1098 }
1099
1100 static int f2fs_read_data_page(struct file *file, struct page *page)
1101 {
1102         struct inode *inode = page->mapping->host;
1103         int ret = -EAGAIN;
1104
1105         trace_f2fs_readpage(page, DATA);
1106
1107         /* If the file has inline data, try to read it directly */
1108         if (f2fs_has_inline_data(inode))
1109                 ret = f2fs_read_inline_data(inode, page);
1110         if (ret == -EAGAIN)
1111                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1112         return ret;
1113 }
1114
1115 static int f2fs_read_data_pages(struct file *file,
1116                         struct address_space *mapping,
1117                         struct list_head *pages, unsigned nr_pages)
1118 {
1119         struct inode *inode = file->f_mapping->host;
1120         struct page *page = list_entry(pages->prev, struct page, lru);
1121
1122         trace_f2fs_readpages(inode, page, nr_pages);
1123
1124         /* If the file has inline data, skip readpages */
1125         if (f2fs_has_inline_data(inode))
1126                 return 0;
1127
1128         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1129 }
1130
1131 int do_write_data_page(struct f2fs_io_info *fio)
1132 {
1133         struct page *page = fio->page;
1134         struct inode *inode = page->mapping->host;
1135         struct dnode_of_data dn;
1136         int err = 0;
1137
1138         set_new_dnode(&dn, inode, NULL, NULL, 0);
1139         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1140         if (err)
1141                 return err;
1142
1143         fio->old_blkaddr = dn.data_blkaddr;
1144
1145         /* This page is already truncated */
1146         if (fio->old_blkaddr == NULL_ADDR) {
1147                 ClearPageUptodate(page);
1148                 goto out_writepage;
1149         }
1150
1151         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1152                 gfp_t gfp_flags = GFP_NOFS;
1153
1154                 /* wait for GCed encrypted page writeback */
1155                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1156                                                         fio->old_blkaddr);
1157 retry_encrypt:
1158                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1159                                                                 gfp_flags);
1160                 if (IS_ERR(fio->encrypted_page)) {
1161                         err = PTR_ERR(fio->encrypted_page);
1162                         if (err == -ENOMEM) {
1163                                 /* flush pending ios and wait for a while */
1164                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1165                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1166                                 gfp_flags |= __GFP_NOFAIL;
1167                                 err = 0;
1168                                 goto retry_encrypt;
1169                         }
1170                         goto out_writepage;
1171                 }
1172         }
1173
1174         set_page_writeback(page);
1175
1176         /*
1177          * If current allocation needs SSR,
1178          * it had better in-place writes for updated data.
1179          */
1180         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1181                         !is_cold_data(page) &&
1182                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1183                         need_inplace_update(inode))) {
1184                 rewrite_data_page(fio);
1185                 set_inode_flag(inode, FI_UPDATE_WRITE);
1186                 trace_f2fs_do_write_data_page(page, IPU);
1187         } else {
1188                 write_data_page(&dn, fio);
1189                 trace_f2fs_do_write_data_page(page, OPU);
1190                 set_inode_flag(inode, FI_APPEND_WRITE);
1191                 if (page->index == 0)
1192                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1193         }
1194 out_writepage:
1195         f2fs_put_dnode(&dn);
1196         return err;
1197 }
1198
1199 static int f2fs_write_data_page(struct page *page,
1200                                         struct writeback_control *wbc)
1201 {
1202         struct inode *inode = page->mapping->host;
1203         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1204         loff_t i_size = i_size_read(inode);
1205         const pgoff_t end_index = ((unsigned long long) i_size)
1206                                                         >> PAGE_SHIFT;
1207         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1208         unsigned offset = 0;
1209         bool need_balance_fs = false;
1210         int err = 0;
1211         struct f2fs_io_info fio = {
1212                 .sbi = sbi,
1213                 .type = DATA,
1214                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1215                 .page = page,
1216                 .encrypted_page = NULL,
1217         };
1218
1219         trace_f2fs_writepage(page, DATA);
1220
1221         if (page->index < end_index)
1222                 goto write;
1223
1224         /*
1225          * If the offset is out-of-range of file size,
1226          * this page does not have to be written to disk.
1227          */
1228         offset = i_size & (PAGE_SIZE - 1);
1229         if ((page->index >= end_index + 1) || !offset)
1230                 goto out;
1231
1232         zero_user_segment(page, offset, PAGE_SIZE);
1233 write:
1234         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1235                 goto redirty_out;
1236         if (f2fs_is_drop_cache(inode))
1237                 goto out;
1238         /* we should not write 0'th page having journal header */
1239         if (f2fs_is_volatile_file(inode) && (!page->index ||
1240                         (!wbc->for_reclaim &&
1241                         available_free_memory(sbi, BASE_CHECK))))
1242                 goto redirty_out;
1243
1244         /* Dentry blocks are controlled by checkpoint */
1245         if (S_ISDIR(inode->i_mode)) {
1246                 if (unlikely(f2fs_cp_error(sbi)))
1247                         goto redirty_out;
1248                 err = do_write_data_page(&fio);
1249                 goto done;
1250         }
1251
1252         /* we should bypass data pages to proceed the kworkder jobs */
1253         if (unlikely(f2fs_cp_error(sbi))) {
1254                 SetPageError(page);
1255                 goto out;
1256         }
1257
1258         if (!wbc->for_reclaim)
1259                 need_balance_fs = true;
1260         else if (has_not_enough_free_secs(sbi, 0))
1261                 goto redirty_out;
1262
1263         err = -EAGAIN;
1264         f2fs_lock_op(sbi);
1265         if (f2fs_has_inline_data(inode))
1266                 err = f2fs_write_inline_data(inode, page);
1267         if (err == -EAGAIN)
1268                 err = do_write_data_page(&fio);
1269         if (F2FS_I(inode)->last_disk_size < psize)
1270                 F2FS_I(inode)->last_disk_size = psize;
1271         f2fs_unlock_op(sbi);
1272 done:
1273         if (err && err != -ENOENT)
1274                 goto redirty_out;
1275
1276         clear_cold_data(page);
1277 out:
1278         inode_dec_dirty_pages(inode);
1279         if (err)
1280                 ClearPageUptodate(page);
1281
1282         if (wbc->for_reclaim) {
1283                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1284                 remove_dirty_inode(inode);
1285         }
1286
1287         unlock_page(page);
1288         f2fs_balance_fs(sbi, need_balance_fs);
1289
1290         if (unlikely(f2fs_cp_error(sbi)))
1291                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1292
1293         return 0;
1294
1295 redirty_out:
1296         redirty_page_for_writepage(wbc, page);
1297         return AOP_WRITEPAGE_ACTIVATE;
1298 }
1299
1300 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1301                         void *data)
1302 {
1303         struct address_space *mapping = data;
1304         int ret = mapping->a_ops->writepage(page, wbc);
1305         mapping_set_error(mapping, ret);
1306         return ret;
1307 }
1308
1309 /*
1310  * This function was copied from write_cche_pages from mm/page-writeback.c.
1311  * The major change is making write step of cold data page separately from
1312  * warm/hot data page.
1313  */
1314 static int f2fs_write_cache_pages(struct address_space *mapping,
1315                         struct writeback_control *wbc, writepage_t writepage,
1316                         void *data)
1317 {
1318         int ret = 0;
1319         int done = 0;
1320         struct pagevec pvec;
1321         int nr_pages;
1322         pgoff_t uninitialized_var(writeback_index);
1323         pgoff_t index;
1324         pgoff_t end;            /* Inclusive */
1325         pgoff_t done_index;
1326         int cycled;
1327         int range_whole = 0;
1328         int tag;
1329         int step = 0;
1330
1331         pagevec_init(&pvec, 0);
1332 next:
1333         if (wbc->range_cyclic) {
1334                 writeback_index = mapping->writeback_index; /* prev offset */
1335                 index = writeback_index;
1336                 if (index == 0)
1337                         cycled = 1;
1338                 else
1339                         cycled = 0;
1340                 end = -1;
1341         } else {
1342                 index = wbc->range_start >> PAGE_SHIFT;
1343                 end = wbc->range_end >> PAGE_SHIFT;
1344                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1345                         range_whole = 1;
1346                 cycled = 1; /* ignore range_cyclic tests */
1347         }
1348         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1349                 tag = PAGECACHE_TAG_TOWRITE;
1350         else
1351                 tag = PAGECACHE_TAG_DIRTY;
1352 retry:
1353         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1354                 tag_pages_for_writeback(mapping, index, end);
1355         done_index = index;
1356         while (!done && (index <= end)) {
1357                 int i;
1358
1359                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1360                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1361                 if (nr_pages == 0)
1362                         break;
1363
1364                 for (i = 0; i < nr_pages; i++) {
1365                         struct page *page = pvec.pages[i];
1366
1367                         if (page->index > end) {
1368                                 done = 1;
1369                                 break;
1370                         }
1371
1372                         done_index = page->index;
1373
1374                         lock_page(page);
1375
1376                         if (unlikely(page->mapping != mapping)) {
1377 continue_unlock:
1378                                 unlock_page(page);
1379                                 continue;
1380                         }
1381
1382                         if (!PageDirty(page)) {
1383                                 /* someone wrote it for us */
1384                                 goto continue_unlock;
1385                         }
1386
1387                         if (step == is_cold_data(page))
1388                                 goto continue_unlock;
1389
1390                         if (PageWriteback(page)) {
1391                                 if (wbc->sync_mode != WB_SYNC_NONE)
1392                                         f2fs_wait_on_page_writeback(page,
1393                                                                 DATA, true);
1394                                 else
1395                                         goto continue_unlock;
1396                         }
1397
1398                         BUG_ON(PageWriteback(page));
1399                         if (!clear_page_dirty_for_io(page))
1400                                 goto continue_unlock;
1401
1402                         ret = (*writepage)(page, wbc, data);
1403                         if (unlikely(ret)) {
1404                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1405                                         unlock_page(page);
1406                                         ret = 0;
1407                                 } else {
1408                                         done_index = page->index + 1;
1409                                         done = 1;
1410                                         break;
1411                                 }
1412                         }
1413
1414                         if (--wbc->nr_to_write <= 0 &&
1415                             wbc->sync_mode == WB_SYNC_NONE) {
1416                                 done = 1;
1417                                 break;
1418                         }
1419                 }
1420                 pagevec_release(&pvec);
1421                 cond_resched();
1422         }
1423
1424         if (step < 1) {
1425                 step++;
1426                 goto next;
1427         }
1428
1429         if (!cycled && !done) {
1430                 cycled = 1;
1431                 index = 0;
1432                 end = writeback_index - 1;
1433                 goto retry;
1434         }
1435         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1436                 mapping->writeback_index = done_index;
1437
1438         return ret;
1439 }
1440
1441 static int f2fs_write_data_pages(struct address_space *mapping,
1442                             struct writeback_control *wbc)
1443 {
1444         struct inode *inode = mapping->host;
1445         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1446         int ret;
1447         long diff;
1448
1449         /* deal with chardevs and other special file */
1450         if (!mapping->a_ops->writepage)
1451                 return 0;
1452
1453         /* skip writing if there is no dirty page in this inode */
1454         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1455                 return 0;
1456
1457         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1458                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1459                         available_free_memory(sbi, DIRTY_DENTS))
1460                 goto skip_write;
1461
1462         /* skip writing during file defragment */
1463         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1464                 goto skip_write;
1465
1466         /* during POR, we don't need to trigger writepage at all. */
1467         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1468                 goto skip_write;
1469
1470         trace_f2fs_writepages(mapping->host, wbc, DATA);
1471
1472         diff = nr_pages_to_write(sbi, DATA, wbc);
1473
1474         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1475         f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1476
1477         remove_dirty_inode(inode);
1478
1479         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1480         return ret;
1481
1482 skip_write:
1483         wbc->pages_skipped += get_dirty_pages(inode);
1484         trace_f2fs_writepages(mapping->host, wbc, DATA);
1485         return 0;
1486 }
1487
1488 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1489 {
1490         struct inode *inode = mapping->host;
1491         loff_t i_size = i_size_read(inode);
1492
1493         if (to > i_size) {
1494                 truncate_pagecache(inode, i_size);
1495                 truncate_blocks(inode, i_size, true);
1496         }
1497 }
1498
1499 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1500                         struct page *page, loff_t pos, unsigned len,
1501                         block_t *blk_addr, bool *node_changed)
1502 {
1503         struct inode *inode = page->mapping->host;
1504         pgoff_t index = page->index;
1505         struct dnode_of_data dn;
1506         struct page *ipage;
1507         bool locked = false;
1508         struct extent_info ei;
1509         int err = 0;
1510
1511         /*
1512          * we already allocated all the blocks, so we don't need to get
1513          * the block addresses when there is no need to fill the page.
1514          */
1515         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1516                                         len == PAGE_SIZE)
1517                 return 0;
1518
1519         if (f2fs_has_inline_data(inode) ||
1520                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1521                 f2fs_lock_op(sbi);
1522                 locked = true;
1523         }
1524 restart:
1525         /* check inline_data */
1526         ipage = get_node_page(sbi, inode->i_ino);
1527         if (IS_ERR(ipage)) {
1528                 err = PTR_ERR(ipage);
1529                 goto unlock_out;
1530         }
1531
1532         set_new_dnode(&dn, inode, ipage, ipage, 0);
1533
1534         if (f2fs_has_inline_data(inode)) {
1535                 if (pos + len <= MAX_INLINE_DATA) {
1536                         read_inline_data(page, ipage);
1537                         set_inode_flag(inode, FI_DATA_EXIST);
1538                         if (inode->i_nlink)
1539                                 set_inline_node(ipage);
1540                 } else {
1541                         err = f2fs_convert_inline_page(&dn, page);
1542                         if (err)
1543                                 goto out;
1544                         if (dn.data_blkaddr == NULL_ADDR)
1545                                 err = f2fs_get_block(&dn, index);
1546                 }
1547         } else if (locked) {
1548                 err = f2fs_get_block(&dn, index);
1549         } else {
1550                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1551                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1552                 } else {
1553                         /* hole case */
1554                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1555                         if (err || dn.data_blkaddr == NULL_ADDR) {
1556                                 f2fs_put_dnode(&dn);
1557                                 f2fs_lock_op(sbi);
1558                                 locked = true;
1559                                 goto restart;
1560                         }
1561                 }
1562         }
1563
1564         /* convert_inline_page can make node_changed */
1565         *blk_addr = dn.data_blkaddr;
1566         *node_changed = dn.node_changed;
1567 out:
1568         f2fs_put_dnode(&dn);
1569 unlock_out:
1570         if (locked)
1571                 f2fs_unlock_op(sbi);
1572         return err;
1573 }
1574
1575 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1576                 loff_t pos, unsigned len, unsigned flags,
1577                 struct page **pagep, void **fsdata)
1578 {
1579         struct inode *inode = mapping->host;
1580         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1581         struct page *page = NULL;
1582         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1583         bool need_balance = false;
1584         block_t blkaddr = NULL_ADDR;
1585         int err = 0;
1586
1587         trace_f2fs_write_begin(inode, pos, len, flags);
1588
1589         /*
1590          * We should check this at this moment to avoid deadlock on inode page
1591          * and #0 page. The locking rule for inline_data conversion should be:
1592          * lock_page(page #0) -> lock_page(inode_page)
1593          */
1594         if (index != 0) {
1595                 err = f2fs_convert_inline_inode(inode);
1596                 if (err)
1597                         goto fail;
1598         }
1599 repeat:
1600         page = grab_cache_page_write_begin(mapping, index, flags);
1601         if (!page) {
1602                 err = -ENOMEM;
1603                 goto fail;
1604         }
1605
1606         *pagep = page;
1607
1608         err = prepare_write_begin(sbi, page, pos, len,
1609                                         &blkaddr, &need_balance);
1610         if (err)
1611                 goto fail;
1612
1613         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1614                 unlock_page(page);
1615                 f2fs_balance_fs(sbi, true);
1616                 lock_page(page);
1617                 if (page->mapping != mapping) {
1618                         /* The page got truncated from under us */
1619                         f2fs_put_page(page, 1);
1620                         goto repeat;
1621                 }
1622         }
1623
1624         f2fs_wait_on_page_writeback(page, DATA, false);
1625
1626         /* wait for GCed encrypted page writeback */
1627         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1628                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1629
1630         if (len == PAGE_SIZE)
1631                 goto out_update;
1632         if (PageUptodate(page))
1633                 goto out_clear;
1634
1635         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1636                 unsigned start = pos & (PAGE_SIZE - 1);
1637                 unsigned end = start + len;
1638
1639                 /* Reading beyond i_size is simple: memset to zero */
1640                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1641                 goto out_update;
1642         }
1643
1644         if (blkaddr == NEW_ADDR) {
1645                 zero_user_segment(page, 0, PAGE_SIZE);
1646         } else {
1647                 struct f2fs_io_info fio = {
1648                         .sbi = sbi,
1649                         .type = DATA,
1650                         .rw = READ_SYNC,
1651                         .old_blkaddr = blkaddr,
1652                         .new_blkaddr = blkaddr,
1653                         .page = page,
1654                         .encrypted_page = NULL,
1655                 };
1656                 err = f2fs_submit_page_bio(&fio);
1657                 if (err)
1658                         goto fail;
1659
1660                 lock_page(page);
1661                 if (unlikely(!PageUptodate(page))) {
1662                         err = -EIO;
1663                         goto fail;
1664                 }
1665                 if (unlikely(page->mapping != mapping)) {
1666                         f2fs_put_page(page, 1);
1667                         goto repeat;
1668                 }
1669
1670                 /* avoid symlink page */
1671                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1672                         err = fscrypt_decrypt_page(page);
1673                         if (err)
1674                                 goto fail;
1675                 }
1676         }
1677 out_update:
1678         SetPageUptodate(page);
1679 out_clear:
1680         clear_cold_data(page);
1681         return 0;
1682
1683 fail:
1684         f2fs_put_page(page, 1);
1685         f2fs_write_failed(mapping, pos + len);
1686         return err;
1687 }
1688
1689 static int f2fs_write_end(struct file *file,
1690                         struct address_space *mapping,
1691                         loff_t pos, unsigned len, unsigned copied,
1692                         struct page *page, void *fsdata)
1693 {
1694         struct inode *inode = page->mapping->host;
1695
1696         trace_f2fs_write_end(inode, pos, len, copied);
1697
1698         set_page_dirty(page);
1699
1700         if (pos + copied > i_size_read(inode))
1701                 f2fs_i_size_write(inode, pos + copied);
1702
1703         f2fs_put_page(page, 1);
1704         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1705         return copied;
1706 }
1707
1708 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1709                            loff_t offset)
1710 {
1711         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1712
1713         if (offset & blocksize_mask)
1714                 return -EINVAL;
1715
1716         if (iov_iter_alignment(iter) & blocksize_mask)
1717                 return -EINVAL;
1718
1719         return 0;
1720 }
1721
1722 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1723 {
1724         struct address_space *mapping = iocb->ki_filp->f_mapping;
1725         struct inode *inode = mapping->host;
1726         size_t count = iov_iter_count(iter);
1727         loff_t offset = iocb->ki_pos;
1728         int err;
1729
1730         err = check_direct_IO(inode, iter, offset);
1731         if (err)
1732                 return err;
1733
1734         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1735                 return 0;
1736
1737         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1738
1739         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1740         if (iov_iter_rw(iter) == WRITE) {
1741                 if (err > 0)
1742                         set_inode_flag(inode, FI_UPDATE_WRITE);
1743                 else if (err < 0)
1744                         f2fs_write_failed(mapping, offset + count);
1745         }
1746
1747         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1748
1749         return err;
1750 }
1751
1752 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1753                                                         unsigned int length)
1754 {
1755         struct inode *inode = page->mapping->host;
1756         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1757
1758         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1759                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1760                 return;
1761
1762         if (PageDirty(page)) {
1763                 if (inode->i_ino == F2FS_META_INO(sbi))
1764                         dec_page_count(sbi, F2FS_DIRTY_META);
1765                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1766                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1767                 else
1768                         inode_dec_dirty_pages(inode);
1769         }
1770
1771         /* This is atomic written page, keep Private */
1772         if (IS_ATOMIC_WRITTEN_PAGE(page))
1773                 return;
1774
1775         set_page_private(page, 0);
1776         ClearPagePrivate(page);
1777 }
1778
1779 int f2fs_release_page(struct page *page, gfp_t wait)
1780 {
1781         /* If this is dirty page, keep PagePrivate */
1782         if (PageDirty(page))
1783                 return 0;
1784
1785         /* This is atomic written page, keep Private */
1786         if (IS_ATOMIC_WRITTEN_PAGE(page))
1787                 return 0;
1788
1789         set_page_private(page, 0);
1790         ClearPagePrivate(page);
1791         return 1;
1792 }
1793
1794 static int f2fs_set_data_page_dirty(struct page *page)
1795 {
1796         struct address_space *mapping = page->mapping;
1797         struct inode *inode = mapping->host;
1798
1799         trace_f2fs_set_page_dirty(page, DATA);
1800
1801         SetPageUptodate(page);
1802
1803         if (f2fs_is_atomic_file(inode)) {
1804                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1805                         register_inmem_page(inode, page);
1806                         return 1;
1807                 }
1808                 /*
1809                  * Previously, this page has been registered, we just
1810                  * return here.
1811                  */
1812                 return 0;
1813         }
1814
1815         if (!PageDirty(page)) {
1816                 __set_page_dirty_nobuffers(page);
1817                 update_dirty_page(inode, page);
1818                 return 1;
1819         }
1820         return 0;
1821 }
1822
1823 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1824 {
1825         struct inode *inode = mapping->host;
1826
1827         if (f2fs_has_inline_data(inode))
1828                 return 0;
1829
1830         /* make sure allocating whole blocks */
1831         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1832                 filemap_write_and_wait(mapping);
1833
1834         return generic_block_bmap(mapping, block, get_data_block_bmap);
1835 }
1836
1837 const struct address_space_operations f2fs_dblock_aops = {
1838         .readpage       = f2fs_read_data_page,
1839         .readpages      = f2fs_read_data_pages,
1840         .writepage      = f2fs_write_data_page,
1841         .writepages     = f2fs_write_data_pages,
1842         .write_begin    = f2fs_write_begin,
1843         .write_end      = f2fs_write_end,
1844         .set_page_dirty = f2fs_set_data_page_dirty,
1845         .invalidatepage = f2fs_invalidate_page,
1846         .releasepage    = f2fs_release_page,
1847         .direct_IO      = f2fs_direct_IO,
1848         .bmap           = f2fs_bmap,
1849 };