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
 347         sync_inode_page(dn);
 348         return 0;
 349 }
 350
 351 /* Should keep dn->ofs_in_node unchanged */
 352 int reserve_new_block(struct dnode_of_data *dn)
 353 {
 354         unsigned int ofs_in_node = dn->ofs_in_node;
 355         int ret;
 356
 357         ret = reserve_new_blocks(dn, 1);
 358         dn->ofs_in_node = ofs_in_node;
 359         return ret;
 360 }
 361
 362 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 363 {
 364         bool need_put = dn->inode_page ? false : true;
 365         int err;
 366
 367         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 368         if (err)
 369                 return err;
 370
 371         if (dn->data_blkaddr == NULL_ADDR)
 372                 err = reserve_new_block(dn);
 373         if (err || need_put)
 374                 f2fs_put_dnode(dn);
 375         return err;
 376 }
 377
 378 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 379 {
 380         struct extent_info ei;
 381         struct inode *inode = dn->inode;
 382
 383         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 384                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 385                 return 0;
 386         }
 387
 388         return f2fs_reserve_block(dn, index);
 389 }
 390
 391 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 392                                                 int rw, bool for_write)
 393 {
 394         struct address_space *mapping = inode->i_mapping;
 395         struct dnode_of_data dn;
 396         struct page *page;
 397         struct extent_info ei;
 398         int err;
 399         struct f2fs_io_info fio = {
 400                 .sbi = F2FS_I_SB(inode),
 401                 .type = DATA,
 402                 .rw = rw,
 403                 .encrypted_page = NULL,
 404         };
 405
 406         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 407                 return read_mapping_page(mapping, index, NULL);
 408
 409         page = f2fs_grab_cache_page(mapping, index, for_write);
 410         if (!page)
 411                 return ERR_PTR(-ENOMEM);
 412
 413         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 414                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 415                 goto got_it;
 416         }
 417
 418         set_new_dnode(&dn, inode, NULL, NULL, 0);
 419         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 420         if (err)
 421                 goto put_err;
 422         f2fs_put_dnode(&dn);
 423
 424         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 425                 err = -ENOENT;
 426                 goto put_err;
 427         }
 428 got_it:
 429         if (PageUptodate(page)) {
 430                 unlock_page(page);
 431                 return page;
 432         }
 433
 434         /*
 435          * A new dentry page is allocated but not able to be written, since its
 436          * new inode page couldn't be allocated due to -ENOSPC.
 437          * In such the case, its blkaddr can be remained as NEW_ADDR.
 438          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 439          */
 440         if (dn.data_blkaddr == NEW_ADDR) {
 441                 zero_user_segment(page, 0, PAGE_SIZE);
 442                 SetPageUptodate(page);
 443                 unlock_page(page);
 444                 return page;
 445         }
 446
 447         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 448         fio.page = page;
 449         err = f2fs_submit_page_bio(&fio);
 450         if (err)
 451                 goto put_err;
 452         return page;
 453
 454 put_err:
 455         f2fs_put_page(page, 1);
 456         return ERR_PTR(err);
 457 }
 458
 459 struct page *find_data_page(struct inode *inode, pgoff_t index)
 460 {
 461         struct address_space *mapping = inode->i_mapping;
 462         struct page *page;
 463
 464         page = find_get_page(mapping, index);
 465         if (page && PageUptodate(page))
 466                 return page;
 467         f2fs_put_page(page, 0);
 468
 469         page = get_read_data_page(inode, index, READ_SYNC, false);
 470         if (IS_ERR(page))
 471                 return page;
 472
 473         if (PageUptodate(page))
 474                 return page;
 475
 476         wait_on_page_locked(page);
 477         if (unlikely(!PageUptodate(page))) {
 478                 f2fs_put_page(page, 0);
 479                 return ERR_PTR(-EIO);
 480         }
 481         return page;
 482 }
 483
 484 /*
 485  * If it tries to access a hole, return an error.
 486  * Because, the callers, functions in dir.c and GC, should be able to know
 487  * whether this page exists or not.
 488  */
 489 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 490                                                         bool for_write)
 491 {
 492         struct address_space *mapping = inode->i_mapping;
 493         struct page *page;
 494 repeat:
 495         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 496         if (IS_ERR(page))
 497                 return page;
 498
 499         /* wait for read completion */
 500         lock_page(page);
 501         if (unlikely(!PageUptodate(page))) {
 502                 f2fs_put_page(page, 1);
 503                 return ERR_PTR(-EIO);
 504         }
 505         if (unlikely(page->mapping != mapping)) {
 506                 f2fs_put_page(page, 1);
 507                 goto repeat;
 508         }
 509         return page;
 510 }
 511
 512 /*
 513  * Caller ensures that this data page is never allocated.
 514  * A new zero-filled data page is allocated in the page cache.
 515  *
 516  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 517  * f2fs_unlock_op().
 518  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 519  * ipage should be released by this function.
 520  */
 521 struct page *get_new_data_page(struct inode *inode,
 522                 struct page *ipage, pgoff_t index, bool new_i_size)
 523 {
 524         struct address_space *mapping = inode->i_mapping;
 525         struct page *page;
 526         struct dnode_of_data dn;
 527         int err;
 528
 529         page = f2fs_grab_cache_page(mapping, index, true);
 530         if (!page) {
 531                 /*
 532                  * before exiting, we should make sure ipage will be released
 533                  * if any error occur.
 534                  */
 535                 f2fs_put_page(ipage, 1);
 536                 return ERR_PTR(-ENOMEM);
 537         }
 538
 539         set_new_dnode(&dn, inode, ipage, NULL, 0);
 540         err = f2fs_reserve_block(&dn, index);
 541         if (err) {
 542                 f2fs_put_page(page, 1);
 543                 return ERR_PTR(err);
 544         }
 545         if (!ipage)
 546                 f2fs_put_dnode(&dn);
 547
 548         if (PageUptodate(page))
 549                 goto got_it;
 550
 551         if (dn.data_blkaddr == NEW_ADDR) {
 552                 zero_user_segment(page, 0, PAGE_SIZE);
 553                 SetPageUptodate(page);
 554         } else {
 555                 f2fs_put_page(page, 1);
 556
 557                 /* if ipage exists, blkaddr should be NEW_ADDR */
 558                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 559                 page = get_lock_data_page(inode, index, true);
 560                 if (IS_ERR(page))
 561                         return page;
 562         }
 563 got_it:
 564         if (new_i_size && i_size_read(inode) <
 565                                 ((loff_t)(index + 1) << PAGE_SHIFT)) {
 566                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 567                 /* Only the directory inode sets new_i_size */
 568                 set_inode_flag(inode, FI_UPDATE_DIR);
 569         }
 570         return page;
 571 }
 572
 573 static int __allocate_data_block(struct dnode_of_data *dn)
 574 {
 575         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 576         struct f2fs_summary sum;
 577         struct node_info ni;
 578         int seg = CURSEG_WARM_DATA;
 579         pgoff_t fofs;
 580         blkcnt_t count = 1;
 581
 582         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 583                 return -EPERM;
 584
 585         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 586         if (dn->data_blkaddr == NEW_ADDR)
 587                 goto alloc;
 588
 589         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 590                 return -ENOSPC;
 591
 592 alloc:
 593         get_node_info(sbi, dn->nid, &ni);
 594         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 595
 596         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 597                 seg = CURSEG_DIRECT_IO;
 598
 599         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 600                                                                 &sum, seg);
 601         set_data_blkaddr(dn);
 602
 603         /* update i_size */
 604         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 605                                                         dn->ofs_in_node;
 606         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 607                 f2fs_i_size_write(dn->inode,
 608                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 609         return 0;
 610 }
 611
 612 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 613 {
 614         struct inode *inode = file_inode(iocb->ki_filp);
 615         struct f2fs_map_blocks map;
 616         ssize_t ret = 0;
 617
 618         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 619         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 620         map.m_next_pgofs = NULL;
 621
 622         if (f2fs_encrypted_inode(inode))
 623                 return 0;
 624
 625         if (iocb->ki_flags & IOCB_DIRECT) {
 626                 ret = f2fs_convert_inline_inode(inode);
 627                 if (ret)
 628                         return ret;
 629                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 630         }
 631         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 632                 ret = f2fs_convert_inline_inode(inode);
 633                 if (ret)
 634                         return ret;
 635         }
 636         if (!f2fs_has_inline_data(inode))
 637                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 638         return ret;
 639 }
 640
 641 /*
 642  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 643  * f2fs_map_blocks structure.
 644  * If original data blocks are allocated, then give them to blockdev.
 645  * Otherwise,
 646  *     a. preallocate requested block addresses
 647  *     b. do not use extent cache for better performance
 648  *     c. give the block addresses to blockdev
 649  */
 650 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 651                                                 int create, int flag)
 652 {
 653         unsigned int maxblocks = map->m_len;
 654         struct dnode_of_data dn;
 655         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 656         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 657         pgoff_t pgofs, end_offset, end;
 658         int err = 0, ofs = 1;
 659         unsigned int ofs_in_node, last_ofs_in_node;
 660         blkcnt_t prealloc;
 661         struct extent_info ei;
 662         bool allocated = false;
 663         block_t blkaddr;
 664
 665         map->m_len = 0;
 666         map->m_flags = 0;
 667
 668         /* it only supports block size == page size */
 669         pgofs = (pgoff_t)map->m_lblk;
 670         end = pgofs + maxblocks;
 671
 672         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 673                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 674                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 675                 map->m_flags = F2FS_MAP_MAPPED;
 676                 goto out;
 677         }
 678
 679 next_dnode:
 680         if (create)
 681                 f2fs_lock_op(sbi);
 682
 683         /* When reading holes, we need its node page */
 684         set_new_dnode(&dn, inode, NULL, NULL, 0);
 685         err = get_dnode_of_data(&dn, pgofs, mode);
 686         if (err) {
 687                 if (flag == F2FS_GET_BLOCK_BMAP)
 688                         map->m_pblk = 0;
 689                 if (err == -ENOENT) {
 690                         err = 0;
 691                         if (map->m_next_pgofs)
 692                                 *map->m_next_pgofs =
 693                                         get_next_page_offset(&dn, pgofs);
 694                 }
 695                 goto unlock_out;
 696         }
 697
 698         prealloc = 0;
 699         ofs_in_node = dn.ofs_in_node;
 700         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 701
 702 next_block:
 703         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 704
 705         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 706                 if (create) {
 707                         if (unlikely(f2fs_cp_error(sbi))) {
 708                                 err = -EIO;
 709                                 goto sync_out;
 710                         }
 711                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 712                                 if (blkaddr == NULL_ADDR) {
 713                                         prealloc++;
 714                                         last_ofs_in_node = dn.ofs_in_node;
 715                                 }
 716                         } else {
 717                                 err = __allocate_data_block(&dn);
 718                                 if (!err) {
 719                                         set_inode_flag(inode, FI_APPEND_WRITE);
 720                                         allocated = true;
 721                                 }
 722                         }
 723                         if (err)
 724                                 goto sync_out;
 725                         map->m_flags = F2FS_MAP_NEW;
 726                         blkaddr = dn.data_blkaddr;
 727                 } else {
 728                         if (flag == F2FS_GET_BLOCK_BMAP) {
 729                                 map->m_pblk = 0;
 730                                 goto sync_out;
 731                         }
 732                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 733                                                 blkaddr == NULL_ADDR) {
 734                                 if (map->m_next_pgofs)
 735                                         *map->m_next_pgofs = pgofs + 1;
 736                         }
 737                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 738                                                 blkaddr != NEW_ADDR)
 739                                 goto sync_out;
 740                 }
 741         }
 742
 743         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 744                 goto skip;
 745
 746         if (map->m_len == 0) {
 747                 /* preallocated unwritten block should be mapped for fiemap. */
 748                 if (blkaddr == NEW_ADDR)
 749                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 750                 map->m_flags |= F2FS_MAP_MAPPED;
 751
 752                 map->m_pblk = blkaddr;
 753                 map->m_len = 1;
 754         } else if ((map->m_pblk != NEW_ADDR &&
 755                         blkaddr == (map->m_pblk + ofs)) ||
 756                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 757                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 758                 ofs++;
 759                 map->m_len++;
 760         } else {
 761                 goto sync_out;
 762         }
 763
 764 skip:
 765         dn.ofs_in_node++;
 766         pgofs++;
 767
 768         /* preallocate blocks in batch for one dnode page */
 769         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 770                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 771
 772                 dn.ofs_in_node = ofs_in_node;
 773                 err = reserve_new_blocks(&dn, prealloc);
 774                 if (err)
 775                         goto sync_out;
 776
 777                 map->m_len += dn.ofs_in_node - ofs_in_node;
 778                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 779                         err = -ENOSPC;
 780                         goto sync_out;
 781                 }
 782                 dn.ofs_in_node = end_offset;
 783         }
 784
 785         if (pgofs >= end)
 786                 goto sync_out;
 787         else if (dn.ofs_in_node < end_offset)
 788                 goto next_block;
 789
 790         if (allocated)
 791                 sync_inode_page(&dn);
 792         f2fs_put_dnode(&dn);
 793
 794         if (create) {
 795                 f2fs_unlock_op(sbi);
 796                 f2fs_balance_fs(sbi, allocated);
 797         }
 798         allocated = false;
 799         goto next_dnode;
 800
 801 sync_out:
 802         if (allocated)
 803                 sync_inode_page(&dn);
 804         f2fs_put_dnode(&dn);
 805 unlock_out:
 806         if (create) {
 807                 f2fs_unlock_op(sbi);
 808                 f2fs_balance_fs(sbi, allocated);
 809         }
 810 out:
 811         trace_f2fs_map_blocks(inode, map, err);
 812         return err;
 813 }
 814
 815 static int __get_data_block(struct inode *inode, sector_t iblock,
 816                         struct buffer_head *bh, int create, int flag,
 817                         pgoff_t *next_pgofs)
 818 {
 819         struct f2fs_map_blocks map;
 820         int ret;
 821
 822         map.m_lblk = iblock;
 823         map.m_len = bh->b_size >> inode->i_blkbits;
 824         map.m_next_pgofs = next_pgofs;
 825
 826         ret = f2fs_map_blocks(inode, &map, create, flag);
 827         if (!ret) {
 828                 map_bh(bh, inode->i_sb, map.m_pblk);
 829                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 830                 bh->b_size = map.m_len << inode->i_blkbits;
 831         }
 832         return ret;
 833 }
 834
 835 static int get_data_block(struct inode *inode, sector_t iblock,
 836                         struct buffer_head *bh_result, int create, int flag,
 837                         pgoff_t *next_pgofs)
 838 {
 839         return __get_data_block(inode, iblock, bh_result, create,
 840                                                         flag, next_pgofs);
 841 }
 842
 843 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 844                         struct buffer_head *bh_result, int create)
 845 {
 846         return __get_data_block(inode, iblock, bh_result, create,
 847                                                 F2FS_GET_BLOCK_DIO, NULL);
 848 }
 849
 850 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 851                         struct buffer_head *bh_result, int create)
 852 {
 853         /* Block number less than F2FS MAX BLOCKS */
 854         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 855                 return -EFBIG;
 856
 857         return __get_data_block(inode, iblock, bh_result, create,
 858                                                 F2FS_GET_BLOCK_BMAP, NULL);
 859 }
 860
 861 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 862 {
 863         return (offset >> inode->i_blkbits);
 864 }
 865
 866 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 867 {
 868         return (blk << inode->i_blkbits);
 869 }
 870
 871 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 872                 u64 start, u64 len)
 873 {
 874         struct buffer_head map_bh;
 875         sector_t start_blk, last_blk;
 876         pgoff_t next_pgofs;
 877         loff_t isize;
 878         u64 logical = 0, phys = 0, size = 0;
 879         u32 flags = 0;
 880         int ret = 0;
 881
 882         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 883         if (ret)
 884                 return ret;
 885
 886         if (f2fs_has_inline_data(inode)) {
 887                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 888                 if (ret != -EAGAIN)
 889                         return ret;
 890         }
 891
 892         inode_lock(inode);
 893
 894         isize = i_size_read(inode);
 895         if (start >= isize)
 896                 goto out;
 897
 898         if (start + len > isize)
 899                 len = isize - start;
 900
 901         if (logical_to_blk(inode, len) == 0)
 902                 len = blk_to_logical(inode, 1);
 903
 904         start_blk = logical_to_blk(inode, start);
 905         last_blk = logical_to_blk(inode, start + len - 1);
 906
 907 next:
 908         memset(&map_bh, 0, sizeof(struct buffer_head));
 909         map_bh.b_size = len;
 910
 911         ret = get_data_block(inode, start_blk, &map_bh, 0,
 912                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 913         if (ret)
 914                 goto out;
 915
 916         /* HOLE */
 917         if (!buffer_mapped(&map_bh)) {
 918                 start_blk = next_pgofs;
 919                 /* Go through holes util pass the EOF */
 920                 if (blk_to_logical(inode, start_blk) < isize)
 921                         goto prep_next;
 922                 /* Found a hole beyond isize means no more extents.
 923                  * Note that the premise is that filesystems don't
 924                  * punch holes beyond isize and keep size unchanged.
 925                  */
 926                 flags |= FIEMAP_EXTENT_LAST;
 927         }
 928
 929         if (size) {
 930                 if (f2fs_encrypted_inode(inode))
 931                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 932
 933                 ret = fiemap_fill_next_extent(fieinfo, logical,
 934                                 phys, size, flags);
 935         }
 936
 937         if (start_blk > last_blk || ret)
 938                 goto out;
 939
 940         logical = blk_to_logical(inode, start_blk);
 941         phys = blk_to_logical(inode, map_bh.b_blocknr);
 942         size = map_bh.b_size;
 943         flags = 0;
 944         if (buffer_unwritten(&map_bh))
 945                 flags = FIEMAP_EXTENT_UNWRITTEN;
 946
 947         start_blk += logical_to_blk(inode, size);
 948
 949 prep_next:
 950         cond_resched();
 951         if (fatal_signal_pending(current))
 952                 ret = -EINTR;
 953         else
 954                 goto next;
 955 out:
 956         if (ret == 1)
 957                 ret = 0;
 958
 959         inode_unlock(inode);
 960         return ret;
 961 }
 962
 963 /*
 964  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 965  * Major change was from block_size == page_size in f2fs by default.
 966  */
 967 static int f2fs_mpage_readpages(struct address_space *mapping,
 968                         struct list_head *pages, struct page *page,
 969                         unsigned nr_pages)
 970 {
 971         struct bio *bio = NULL;
 972         unsigned page_idx;
 973         sector_t last_block_in_bio = 0;
 974         struct inode *inode = mapping->host;
 975         const unsigned blkbits = inode->i_blkbits;
 976         const unsigned blocksize = 1 << blkbits;
 977         sector_t block_in_file;
 978         sector_t last_block;
 979         sector_t last_block_in_file;
 980         sector_t block_nr;
 981         struct block_device *bdev = inode->i_sb->s_bdev;
 982         struct f2fs_map_blocks map;
 983
 984         map.m_pblk = 0;
 985         map.m_lblk = 0;
 986         map.m_len = 0;
 987         map.m_flags = 0;
 988         map.m_next_pgofs = NULL;
 989
 990         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 991
 992                 prefetchw(&page->flags);
 993                 if (pages) {
 994                         page = list_entry(pages->prev, struct page, lru);
 995                         list_del(&page->lru);
 996                         if (add_to_page_cache_lru(page, mapping,
 997                                                   page->index, GFP_KERNEL))
 998                                 goto next_page;
 999                 }
1000
1001                 block_in_file = (sector_t)page->index;
1002                 last_block = block_in_file + nr_pages;
1003                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1004                                                                 blkbits;
1005                 if (last_block > last_block_in_file)
1006                         last_block = last_block_in_file;
1007
1008                 /*
1009                  * Map blocks using the previous result first.
1010                  */
1011                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1012                                 block_in_file > map.m_lblk &&
1013                                 block_in_file < (map.m_lblk + map.m_len))
1014                         goto got_it;
1015
1016                 /*
1017                  * Then do more f2fs_map_blocks() calls until we are
1018                  * done with this page.
1019                  */
1020                 map.m_flags = 0;
1021
1022                 if (block_in_file < last_block) {
1023                         map.m_lblk = block_in_file;
1024                         map.m_len = last_block - block_in_file;
1025
1026                         if (f2fs_map_blocks(inode, &map, 0,
1027                                                 F2FS_GET_BLOCK_READ))
1028                                 goto set_error_page;
1029                 }
1030 got_it:
1031                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1032                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1033                         SetPageMappedToDisk(page);
1034
1035                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1036                                 SetPageUptodate(page);
1037                                 goto confused;
1038                         }
1039                 } else {
1040                         zero_user_segment(page, 0, PAGE_SIZE);
1041                         SetPageUptodate(page);
1042                         unlock_page(page);
1043                         goto next_page;
1044                 }
1045
1046                 /*
1047                  * This page will go to BIO.  Do we need to send this
1048                  * BIO off first?
1049                  */
1050                 if (bio && (last_block_in_bio != block_nr - 1)) {
1051 submit_and_realloc:
1052                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1053                         bio = NULL;
1054                 }
1055                 if (bio == NULL) {
1056                         struct fscrypt_ctx *ctx = NULL;
1057
1058                         if (f2fs_encrypted_inode(inode) &&
1059                                         S_ISREG(inode->i_mode)) {
1060
1061                                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1062                                 if (IS_ERR(ctx))
1063                                         goto set_error_page;
1064
1065                                 /* wait the page to be moved by cleaning */
1066                                 f2fs_wait_on_encrypted_page_writeback(
1067                                                 F2FS_I_SB(inode), block_nr);
1068                         }
1069
1070                         bio = bio_alloc(GFP_KERNEL,
1071                                 min_t(int, nr_pages, BIO_MAX_PAGES));
1072                         if (!bio) {
1073                                 if (ctx)
1074                                         fscrypt_release_ctx(ctx);
1075                                 goto set_error_page;
1076                         }
1077                         bio->bi_bdev = bdev;
1078                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1079                         bio->bi_end_io = f2fs_read_end_io;
1080                         bio->bi_private = ctx;
1081                 }
1082
1083                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1084                         goto submit_and_realloc;
1085
1086                 last_block_in_bio = block_nr;
1087                 goto next_page;
1088 set_error_page:
1089                 SetPageError(page);
1090                 zero_user_segment(page, 0, PAGE_SIZE);
1091                 unlock_page(page);
1092                 goto next_page;
1093 confused:
1094                 if (bio) {
1095                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1096                         bio = NULL;
1097                 }
1098                 unlock_page(page);
1099 next_page:
1100                 if (pages)
1101                         put_page(page);
1102         }
1103         BUG_ON(pages && !list_empty(pages));
1104         if (bio)
1105                 __submit_bio(F2FS_I_SB(inode), READ, bio);
1106         return 0;
1107 }
1108
1109 static int f2fs_read_data_page(struct file *file, struct page *page)
1110 {
1111         struct inode *inode = page->mapping->host;
1112         int ret = -EAGAIN;
1113
1114         trace_f2fs_readpage(page, DATA);
1115
1116         /* If the file has inline data, try to read it directly */
1117         if (f2fs_has_inline_data(inode))
1118                 ret = f2fs_read_inline_data(inode, page);
1119         if (ret == -EAGAIN)
1120                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1121         return ret;
1122 }
1123
1124 static int f2fs_read_data_pages(struct file *file,
1125                         struct address_space *mapping,
1126                         struct list_head *pages, unsigned nr_pages)
1127 {
1128         struct inode *inode = file->f_mapping->host;
1129         struct page *page = list_entry(pages->prev, struct page, lru);
1130
1131         trace_f2fs_readpages(inode, page, nr_pages);
1132
1133         /* If the file has inline data, skip readpages */
1134         if (f2fs_has_inline_data(inode))
1135                 return 0;
1136
1137         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1138 }
1139
1140 int do_write_data_page(struct f2fs_io_info *fio)
1141 {
1142         struct page *page = fio->page;
1143         struct inode *inode = page->mapping->host;
1144         struct dnode_of_data dn;
1145         int err = 0;
1146
1147         set_new_dnode(&dn, inode, NULL, NULL, 0);
1148         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1149         if (err)
1150                 return err;
1151
1152         fio->old_blkaddr = dn.data_blkaddr;
1153
1154         /* This page is already truncated */
1155         if (fio->old_blkaddr == NULL_ADDR) {
1156                 ClearPageUptodate(page);
1157                 goto out_writepage;
1158         }
1159
1160         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1161                 gfp_t gfp_flags = GFP_NOFS;
1162
1163                 /* wait for GCed encrypted page writeback */
1164                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1165                                                         fio->old_blkaddr);
1166 retry_encrypt:
1167                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1168                                                                 gfp_flags);
1169                 if (IS_ERR(fio->encrypted_page)) {
1170                         err = PTR_ERR(fio->encrypted_page);
1171                         if (err == -ENOMEM) {
1172                                 /* flush pending ios and wait for a while */
1173                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1174                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1175                                 gfp_flags |= __GFP_NOFAIL;
1176                                 err = 0;
1177                                 goto retry_encrypt;
1178                         }
1179                         goto out_writepage;
1180                 }
1181         }
1182
1183         set_page_writeback(page);
1184
1185         /*
1186          * If current allocation needs SSR,
1187          * it had better in-place writes for updated data.
1188          */
1189         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1190                         !is_cold_data(page) &&
1191                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1192                         need_inplace_update(inode))) {
1193                 rewrite_data_page(fio);
1194                 set_inode_flag(inode, FI_UPDATE_WRITE);
1195                 trace_f2fs_do_write_data_page(page, IPU);
1196         } else {
1197                 write_data_page(&dn, fio);
1198                 trace_f2fs_do_write_data_page(page, OPU);
1199                 set_inode_flag(inode, FI_APPEND_WRITE);
1200                 if (page->index == 0)
1201                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1202         }
1203 out_writepage:
1204         f2fs_put_dnode(&dn);
1205         return err;
1206 }
1207
1208 static int f2fs_write_data_page(struct page *page,
1209                                         struct writeback_control *wbc)
1210 {
1211         struct inode *inode = page->mapping->host;
1212         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1213         loff_t i_size = i_size_read(inode);
1214         const pgoff_t end_index = ((unsigned long long) i_size)
1215                                                         >> PAGE_SHIFT;
1216         unsigned offset = 0;
1217         bool need_balance_fs = false;
1218         int err = 0;
1219         struct f2fs_io_info fio = {
1220                 .sbi = sbi,
1221                 .type = DATA,
1222                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1223                 .page = page,
1224                 .encrypted_page = NULL,
1225         };
1226
1227         trace_f2fs_writepage(page, DATA);
1228
1229         if (page->index < end_index)
1230                 goto write;
1231
1232         /*
1233          * If the offset is out-of-range of file size,
1234          * this page does not have to be written to disk.
1235          */
1236         offset = i_size & (PAGE_SIZE - 1);
1237         if ((page->index >= end_index + 1) || !offset)
1238                 goto out;
1239
1240         zero_user_segment(page, offset, PAGE_SIZE);
1241 write:
1242         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1243                 goto redirty_out;
1244         if (f2fs_is_drop_cache(inode))
1245                 goto out;
1246         /* we should not write 0'th page having journal header */
1247         if (f2fs_is_volatile_file(inode) && (!page->index ||
1248                         (!wbc->for_reclaim &&
1249                         available_free_memory(sbi, BASE_CHECK))))
1250                 goto redirty_out;
1251
1252         /* Dentry blocks are controlled by checkpoint */
1253         if (S_ISDIR(inode->i_mode)) {
1254                 if (unlikely(f2fs_cp_error(sbi)))
1255                         goto redirty_out;
1256                 err = do_write_data_page(&fio);
1257                 goto done;
1258         }
1259
1260         /* we should bypass data pages to proceed the kworkder jobs */
1261         if (unlikely(f2fs_cp_error(sbi))) {
1262                 SetPageError(page);
1263                 goto out;
1264         }
1265
1266         if (!wbc->for_reclaim)
1267                 need_balance_fs = true;
1268         else if (has_not_enough_free_secs(sbi, 0))
1269                 goto redirty_out;
1270
1271         err = -EAGAIN;
1272         f2fs_lock_op(sbi);
1273         if (f2fs_has_inline_data(inode))
1274                 err = f2fs_write_inline_data(inode, page);
1275         if (err == -EAGAIN)
1276                 err = do_write_data_page(&fio);
1277         f2fs_unlock_op(sbi);
1278 done:
1279         if (err && err != -ENOENT)
1280                 goto redirty_out;
1281
1282         clear_cold_data(page);
1283 out:
1284         inode_dec_dirty_pages(inode);
1285         if (err)
1286                 ClearPageUptodate(page);
1287
1288         if (wbc->for_reclaim) {
1289                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1290                 remove_dirty_inode(inode);
1291         }
1292
1293         unlock_page(page);
1294         f2fs_balance_fs(sbi, need_balance_fs);
1295
1296         if (unlikely(f2fs_cp_error(sbi)))
1297                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1298
1299         return 0;
1300
1301 redirty_out:
1302         redirty_page_for_writepage(wbc, page);
1303         return AOP_WRITEPAGE_ACTIVATE;
1304 }
1305
1306 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1307                         void *data)
1308 {
1309         struct address_space *mapping = data;
1310         int ret = mapping->a_ops->writepage(page, wbc);
1311         mapping_set_error(mapping, ret);
1312         return ret;
1313 }
1314
1315 /*
1316  * This function was copied from write_cche_pages from mm/page-writeback.c.
1317  * The major change is making write step of cold data page separately from
1318  * warm/hot data page.
1319  */
1320 static int f2fs_write_cache_pages(struct address_space *mapping,
1321                         struct writeback_control *wbc, writepage_t writepage,
1322                         void *data)
1323 {
1324         int ret = 0;
1325         int done = 0;
1326         struct pagevec pvec;
1327         int nr_pages;
1328         pgoff_t uninitialized_var(writeback_index);
1329         pgoff_t index;
1330         pgoff_t end;            /* Inclusive */
1331         pgoff_t done_index;
1332         int cycled;
1333         int range_whole = 0;
1334         int tag;
1335         int step = 0;
1336
1337         pagevec_init(&pvec, 0);
1338 next:
1339         if (wbc->range_cyclic) {
1340                 writeback_index = mapping->writeback_index; /* prev offset */
1341                 index = writeback_index;
1342                 if (index == 0)
1343                         cycled = 1;
1344                 else
1345                         cycled = 0;
1346                 end = -1;
1347         } else {
1348                 index = wbc->range_start >> PAGE_SHIFT;
1349                 end = wbc->range_end >> PAGE_SHIFT;
1350                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1351                         range_whole = 1;
1352                 cycled = 1; /* ignore range_cyclic tests */
1353         }
1354         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1355                 tag = PAGECACHE_TAG_TOWRITE;
1356         else
1357                 tag = PAGECACHE_TAG_DIRTY;
1358 retry:
1359         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1360                 tag_pages_for_writeback(mapping, index, end);
1361         done_index = index;
1362         while (!done && (index <= end)) {
1363                 int i;
1364
1365                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1366                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1367                 if (nr_pages == 0)
1368                         break;
1369
1370                 for (i = 0; i < nr_pages; i++) {
1371                         struct page *page = pvec.pages[i];
1372
1373                         if (page->index > end) {
1374                                 done = 1;
1375                                 break;
1376                         }
1377
1378                         done_index = page->index;
1379
1380                         lock_page(page);
1381
1382                         if (unlikely(page->mapping != mapping)) {
1383 continue_unlock:
1384                                 unlock_page(page);
1385                                 continue;
1386                         }
1387
1388                         if (!PageDirty(page)) {
1389                                 /* someone wrote it for us */
1390                                 goto continue_unlock;
1391                         }
1392
1393                         if (step == is_cold_data(page))
1394                                 goto continue_unlock;
1395
1396                         if (PageWriteback(page)) {
1397                                 if (wbc->sync_mode != WB_SYNC_NONE)
1398                                         f2fs_wait_on_page_writeback(page,
1399                                                                 DATA, true);
1400                                 else
1401                                         goto continue_unlock;
1402                         }
1403
1404                         BUG_ON(PageWriteback(page));
1405                         if (!clear_page_dirty_for_io(page))
1406                                 goto continue_unlock;
1407
1408                         ret = (*writepage)(page, wbc, data);
1409                         if (unlikely(ret)) {
1410                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1411                                         unlock_page(page);
1412                                         ret = 0;
1413                                 } else {
1414                                         done_index = page->index + 1;
1415                                         done = 1;
1416                                         break;
1417                                 }
1418                         }
1419
1420                         if (--wbc->nr_to_write <= 0 &&
1421                             wbc->sync_mode == WB_SYNC_NONE) {
1422                                 done = 1;
1423                                 break;
1424                         }
1425                 }
1426                 pagevec_release(&pvec);
1427                 cond_resched();
1428         }
1429
1430         if (step < 1) {
1431                 step++;
1432                 goto next;
1433         }
1434
1435         if (!cycled && !done) {
1436                 cycled = 1;
1437                 index = 0;
1438                 end = writeback_index - 1;
1439                 goto retry;
1440         }
1441         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1442                 mapping->writeback_index = done_index;
1443
1444         return ret;
1445 }
1446
1447 static int f2fs_write_data_pages(struct address_space *mapping,
1448                             struct writeback_control *wbc)
1449 {
1450         struct inode *inode = mapping->host;
1451         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1452         bool locked = false;
1453         int ret;
1454         long diff;
1455
1456         /* deal with chardevs and other special file */
1457         if (!mapping->a_ops->writepage)
1458                 return 0;
1459
1460         /* skip writing if there is no dirty page in this inode */
1461         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1462                 return 0;
1463
1464         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1465                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1466                         available_free_memory(sbi, DIRTY_DENTS))
1467                 goto skip_write;
1468
1469         /* skip writing during file defragment */
1470         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1471                 goto skip_write;
1472
1473         /* during POR, we don't need to trigger writepage at all. */
1474         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1475                 goto skip_write;
1476
1477         trace_f2fs_writepages(mapping->host, wbc, DATA);
1478
1479         diff = nr_pages_to_write(sbi, DATA, wbc);
1480
1481         if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1482                 mutex_lock(&sbi->writepages);
1483                 locked = true;
1484         }
1485         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1486         f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1487         if (locked)
1488                 mutex_unlock(&sbi->writepages);
1489
1490         remove_dirty_inode(inode);
1491
1492         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1493         return ret;
1494
1495 skip_write:
1496         wbc->pages_skipped += get_dirty_pages(inode);
1497         trace_f2fs_writepages(mapping->host, wbc, DATA);
1498         return 0;
1499 }
1500
1501 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1502 {
1503         struct inode *inode = mapping->host;
1504         loff_t i_size = i_size_read(inode);
1505
1506         if (to > i_size) {
1507                 truncate_pagecache(inode, i_size);
1508                 truncate_blocks(inode, i_size, true);
1509         }
1510 }
1511
1512 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1513                         struct page *page, loff_t pos, unsigned len,
1514                         block_t *blk_addr, bool *node_changed)
1515 {
1516         struct inode *inode = page->mapping->host;
1517         pgoff_t index = page->index;
1518         struct dnode_of_data dn;
1519         struct page *ipage;
1520         bool locked = false;
1521         struct extent_info ei;
1522         int err = 0;
1523
1524         /*
1525          * we already allocated all the blocks, so we don't need to get
1526          * the block addresses when there is no need to fill the page.
1527          */
1528         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1529                                         len == PAGE_SIZE)
1530                 return 0;
1531
1532         if (f2fs_has_inline_data(inode) ||
1533                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1534                 f2fs_lock_op(sbi);
1535                 locked = true;
1536         }
1537 restart:
1538         /* check inline_data */
1539         ipage = get_node_page(sbi, inode->i_ino);
1540         if (IS_ERR(ipage)) {
1541                 err = PTR_ERR(ipage);
1542                 goto unlock_out;
1543         }
1544
1545         set_new_dnode(&dn, inode, ipage, ipage, 0);
1546
1547         if (f2fs_has_inline_data(inode)) {
1548                 if (pos + len <= MAX_INLINE_DATA) {
1549                         read_inline_data(page, ipage);
1550                         set_inode_flag(inode, FI_DATA_EXIST);
1551                         if (inode->i_nlink)
1552                                 set_inline_node(ipage);
1553                 } else {
1554                         err = f2fs_convert_inline_page(&dn, page);
1555                         if (err)
1556                                 goto out;
1557                         if (dn.data_blkaddr == NULL_ADDR)
1558                                 err = f2fs_get_block(&dn, index);
1559                 }
1560         } else if (locked) {
1561                 err = f2fs_get_block(&dn, index);
1562         } else {
1563                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1564                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1565                 } else {
1566                         /* hole case */
1567                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1568                         if (err || dn.data_blkaddr == NULL_ADDR) {
1569                                 f2fs_put_dnode(&dn);
1570                                 f2fs_lock_op(sbi);
1571                                 locked = true;
1572                                 goto restart;
1573                         }
1574                 }
1575         }
1576
1577         /* convert_inline_page can make node_changed */
1578         *blk_addr = dn.data_blkaddr;
1579         *node_changed = dn.node_changed;
1580 out:
1581         f2fs_put_dnode(&dn);
1582 unlock_out:
1583         if (locked)
1584                 f2fs_unlock_op(sbi);
1585         return err;
1586 }
1587
1588 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1589                 loff_t pos, unsigned len, unsigned flags,
1590                 struct page **pagep, void **fsdata)
1591 {
1592         struct inode *inode = mapping->host;
1593         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1594         struct page *page = NULL;
1595         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1596         bool need_balance = false;
1597         block_t blkaddr = NULL_ADDR;
1598         int err = 0;
1599
1600         trace_f2fs_write_begin(inode, pos, len, flags);
1601
1602         /*
1603          * We should check this at this moment to avoid deadlock on inode page
1604          * and #0 page. The locking rule for inline_data conversion should be:
1605          * lock_page(page #0) -> lock_page(inode_page)
1606          */
1607         if (index != 0) {
1608                 err = f2fs_convert_inline_inode(inode);
1609                 if (err)
1610                         goto fail;
1611         }
1612 repeat:
1613         page = grab_cache_page_write_begin(mapping, index, flags);
1614         if (!page) {
1615                 err = -ENOMEM;
1616                 goto fail;
1617         }
1618
1619         *pagep = page;
1620
1621         err = prepare_write_begin(sbi, page, pos, len,
1622                                         &blkaddr, &need_balance);
1623         if (err)
1624                 goto fail;
1625
1626         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1627                 unlock_page(page);
1628                 f2fs_balance_fs(sbi, true);
1629                 lock_page(page);
1630                 if (page->mapping != mapping) {
1631                         /* The page got truncated from under us */
1632                         f2fs_put_page(page, 1);
1633                         goto repeat;
1634                 }
1635         }
1636
1637         f2fs_wait_on_page_writeback(page, DATA, false);
1638
1639         /* wait for GCed encrypted page writeback */
1640         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1641                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1642
1643         if (len == PAGE_SIZE)
1644                 goto out_update;
1645         if (PageUptodate(page))
1646                 goto out_clear;
1647
1648         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1649                 unsigned start = pos & (PAGE_SIZE - 1);
1650                 unsigned end = start + len;
1651
1652                 /* Reading beyond i_size is simple: memset to zero */
1653                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1654                 goto out_update;
1655         }
1656
1657         if (blkaddr == NEW_ADDR) {
1658                 zero_user_segment(page, 0, PAGE_SIZE);
1659         } else {
1660                 struct f2fs_io_info fio = {
1661                         .sbi = sbi,
1662                         .type = DATA,
1663                         .rw = READ_SYNC,
1664                         .old_blkaddr = blkaddr,
1665                         .new_blkaddr = blkaddr,
1666                         .page = page,
1667                         .encrypted_page = NULL,
1668                 };
1669                 err = f2fs_submit_page_bio(&fio);
1670                 if (err)
1671                         goto fail;
1672
1673                 lock_page(page);
1674                 if (unlikely(!PageUptodate(page))) {
1675                         err = -EIO;
1676                         goto fail;
1677                 }
1678                 if (unlikely(page->mapping != mapping)) {
1679                         f2fs_put_page(page, 1);
1680                         goto repeat;
1681                 }
1682
1683                 /* avoid symlink page */
1684                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1685                         err = fscrypt_decrypt_page(page);
1686                         if (err)
1687                                 goto fail;
1688                 }
1689         }
1690 out_update:
1691         SetPageUptodate(page);
1692 out_clear:
1693         clear_cold_data(page);
1694         return 0;
1695
1696 fail:
1697         f2fs_put_page(page, 1);
1698         f2fs_write_failed(mapping, pos + len);
1699         return err;
1700 }
1701
1702 static int f2fs_write_end(struct file *file,
1703                         struct address_space *mapping,
1704                         loff_t pos, unsigned len, unsigned copied,
1705                         struct page *page, void *fsdata)
1706 {
1707         struct inode *inode = page->mapping->host;
1708
1709         trace_f2fs_write_end(inode, pos, len, copied);
1710
1711         set_page_dirty(page);
1712
1713         if (pos + copied > i_size_read(inode))
1714                 f2fs_i_size_write(inode, pos + copied);
1715
1716         f2fs_put_page(page, 1);
1717         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1718         return copied;
1719 }
1720
1721 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1722                            loff_t offset)
1723 {
1724         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1725
1726         if (offset & blocksize_mask)
1727                 return -EINVAL;
1728
1729         if (iov_iter_alignment(iter) & blocksize_mask)
1730                 return -EINVAL;
1731
1732         return 0;
1733 }
1734
1735 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1736 {
1737         struct address_space *mapping = iocb->ki_filp->f_mapping;
1738         struct inode *inode = mapping->host;
1739         size_t count = iov_iter_count(iter);
1740         loff_t offset = iocb->ki_pos;
1741         int err;
1742
1743         err = check_direct_IO(inode, iter, offset);
1744         if (err)
1745                 return err;
1746
1747         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1748                 return 0;
1749
1750         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1751
1752         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1753         if (iov_iter_rw(iter) == WRITE) {
1754                 if (err > 0)
1755                         set_inode_flag(inode, FI_UPDATE_WRITE);
1756                 else if (err < 0)
1757                         f2fs_write_failed(mapping, offset + count);
1758         }
1759
1760         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1761
1762         return err;
1763 }
1764
1765 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1766                                                         unsigned int length)
1767 {
1768         struct inode *inode = page->mapping->host;
1769         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1770
1771         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1772                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1773                 return;
1774
1775         if (PageDirty(page)) {
1776                 if (inode->i_ino == F2FS_META_INO(sbi))
1777                         dec_page_count(sbi, F2FS_DIRTY_META);
1778                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1779                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1780                 else
1781                         inode_dec_dirty_pages(inode);
1782         }
1783
1784         /* This is atomic written page, keep Private */
1785         if (IS_ATOMIC_WRITTEN_PAGE(page))
1786                 return;
1787
1788         set_page_private(page, 0);
1789         ClearPagePrivate(page);
1790 }
1791
1792 int f2fs_release_page(struct page *page, gfp_t wait)
1793 {
1794         /* If this is dirty page, keep PagePrivate */
1795         if (PageDirty(page))
1796                 return 0;
1797
1798         /* This is atomic written page, keep Private */
1799         if (IS_ATOMIC_WRITTEN_PAGE(page))
1800                 return 0;
1801
1802         set_page_private(page, 0);
1803         ClearPagePrivate(page);
1804         return 1;
1805 }
1806
1807 static int f2fs_set_data_page_dirty(struct page *page)
1808 {
1809         struct address_space *mapping = page->mapping;
1810         struct inode *inode = mapping->host;
1811
1812         trace_f2fs_set_page_dirty(page, DATA);
1813
1814         SetPageUptodate(page);
1815
1816         if (f2fs_is_atomic_file(inode)) {
1817                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1818                         register_inmem_page(inode, page);
1819                         return 1;
1820                 }
1821                 /*
1822                  * Previously, this page has been registered, we just
1823                  * return here.
1824                  */
1825                 return 0;
1826         }
1827
1828         if (!PageDirty(page)) {
1829                 __set_page_dirty_nobuffers(page);
1830                 update_dirty_page(inode, page);
1831                 return 1;
1832         }
1833         return 0;
1834 }
1835
1836 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1837 {
1838         struct inode *inode = mapping->host;
1839
1840         if (f2fs_has_inline_data(inode))
1841                 return 0;
1842
1843         /* make sure allocating whole blocks */
1844         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1845                 filemap_write_and_wait(mapping);
1846
1847         return generic_block_bmap(mapping, block, get_data_block_bmap);
1848 }
1849
1850 const struct address_space_operations f2fs_dblock_aops = {
1851         .readpage       = f2fs_read_data_page,
1852         .readpages      = f2fs_read_data_pages,
1853         .writepage      = f2fs_write_data_page,
1854         .writepages     = f2fs_write_data_pages,
1855         .write_begin    = f2fs_write_begin,
1856         .write_end      = f2fs_write_end,
1857         .set_page_dirty = f2fs_set_data_page_dirty,
1858         .invalidatepage = f2fs_invalidate_page,
1859         .releasepage    = f2fs_release_page,
1860         .direct_IO      = f2fs_direct_IO,
1861         .bmap           = f2fs_bmap,
1862 };