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(F2FS_I(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         mark_inode_dirty(dn->inode);
 348         sync_inode_page(dn);
 349         return 0;
 350 }
 351
 352 /* Should keep dn->ofs_in_node unchanged */
 353 int reserve_new_block(struct dnode_of_data *dn)
 354 {
 355         unsigned int ofs_in_node = dn->ofs_in_node;
 356         int ret;
 357
 358         ret = reserve_new_blocks(dn, 1);
 359         dn->ofs_in_node = ofs_in_node;
 360         return ret;
 361 }
 362
 363 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 364 {
 365         bool need_put = dn->inode_page ? false : true;
 366         int err;
 367
 368         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 369         if (err)
 370                 return err;
 371
 372         if (dn->data_blkaddr == NULL_ADDR)
 373                 err = reserve_new_block(dn);
 374         if (err || need_put)
 375                 f2fs_put_dnode(dn);
 376         return err;
 377 }
 378
 379 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 380 {
 381         struct extent_info ei;
 382         struct inode *inode = dn->inode;
 383
 384         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 385                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 386                 return 0;
 387         }
 388
 389         return f2fs_reserve_block(dn, index);
 390 }
 391
 392 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 393                                                 int rw, bool for_write)
 394 {
 395         struct address_space *mapping = inode->i_mapping;
 396         struct dnode_of_data dn;
 397         struct page *page;
 398         struct extent_info ei;
 399         int err;
 400         struct f2fs_io_info fio = {
 401                 .sbi = F2FS_I_SB(inode),
 402                 .type = DATA,
 403                 .rw = rw,
 404                 .encrypted_page = NULL,
 405         };
 406
 407         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 408                 return read_mapping_page(mapping, index, NULL);
 409
 410         page = f2fs_grab_cache_page(mapping, index, for_write);
 411         if (!page)
 412                 return ERR_PTR(-ENOMEM);
 413
 414         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 415                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 416                 goto got_it;
 417         }
 418
 419         set_new_dnode(&dn, inode, NULL, NULL, 0);
 420         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 421         if (err)
 422                 goto put_err;
 423         f2fs_put_dnode(&dn);
 424
 425         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 426                 err = -ENOENT;
 427                 goto put_err;
 428         }
 429 got_it:
 430         if (PageUptodate(page)) {
 431                 unlock_page(page);
 432                 return page;
 433         }
 434
 435         /*
 436          * A new dentry page is allocated but not able to be written, since its
 437          * new inode page couldn't be allocated due to -ENOSPC.
 438          * In such the case, its blkaddr can be remained as NEW_ADDR.
 439          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 440          */
 441         if (dn.data_blkaddr == NEW_ADDR) {
 442                 zero_user_segment(page, 0, PAGE_SIZE);
 443                 SetPageUptodate(page);
 444                 unlock_page(page);
 445                 return page;
 446         }
 447
 448         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 449         fio.page = page;
 450         err = f2fs_submit_page_bio(&fio);
 451         if (err)
 452                 goto put_err;
 453         return page;
 454
 455 put_err:
 456         f2fs_put_page(page, 1);
 457         return ERR_PTR(err);
 458 }
 459
 460 struct page *find_data_page(struct inode *inode, pgoff_t index)
 461 {
 462         struct address_space *mapping = inode->i_mapping;
 463         struct page *page;
 464
 465         page = find_get_page(mapping, index);
 466         if (page && PageUptodate(page))
 467                 return page;
 468         f2fs_put_page(page, 0);
 469
 470         page = get_read_data_page(inode, index, READ_SYNC, false);
 471         if (IS_ERR(page))
 472                 return page;
 473
 474         if (PageUptodate(page))
 475                 return page;
 476
 477         wait_on_page_locked(page);
 478         if (unlikely(!PageUptodate(page))) {
 479                 f2fs_put_page(page, 0);
 480                 return ERR_PTR(-EIO);
 481         }
 482         return page;
 483 }
 484
 485 /*
 486  * If it tries to access a hole, return an error.
 487  * Because, the callers, functions in dir.c and GC, should be able to know
 488  * whether this page exists or not.
 489  */
 490 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 491                                                         bool for_write)
 492 {
 493         struct address_space *mapping = inode->i_mapping;
 494         struct page *page;
 495 repeat:
 496         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 497         if (IS_ERR(page))
 498                 return page;
 499
 500         /* wait for read completion */
 501         lock_page(page);
 502         if (unlikely(!PageUptodate(page))) {
 503                 f2fs_put_page(page, 1);
 504                 return ERR_PTR(-EIO);
 505         }
 506         if (unlikely(page->mapping != mapping)) {
 507                 f2fs_put_page(page, 1);
 508                 goto repeat;
 509         }
 510         return page;
 511 }
 512
 513 /*
 514  * Caller ensures that this data page is never allocated.
 515  * A new zero-filled data page is allocated in the page cache.
 516  *
 517  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 518  * f2fs_unlock_op().
 519  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 520  * ipage should be released by this function.
 521  */
 522 struct page *get_new_data_page(struct inode *inode,
 523                 struct page *ipage, pgoff_t index, bool new_i_size)
 524 {
 525         struct address_space *mapping = inode->i_mapping;
 526         struct page *page;
 527         struct dnode_of_data dn;
 528         int err;
 529
 530         page = f2fs_grab_cache_page(mapping, index, true);
 531         if (!page) {
 532                 /*
 533                  * before exiting, we should make sure ipage will be released
 534                  * if any error occur.
 535                  */
 536                 f2fs_put_page(ipage, 1);
 537                 return ERR_PTR(-ENOMEM);
 538         }
 539
 540         set_new_dnode(&dn, inode, ipage, NULL, 0);
 541         err = f2fs_reserve_block(&dn, index);
 542         if (err) {
 543                 f2fs_put_page(page, 1);
 544                 return ERR_PTR(err);
 545         }
 546         if (!ipage)
 547                 f2fs_put_dnode(&dn);
 548
 549         if (PageUptodate(page))
 550                 goto got_it;
 551
 552         if (dn.data_blkaddr == NEW_ADDR) {
 553                 zero_user_segment(page, 0, PAGE_SIZE);
 554                 SetPageUptodate(page);
 555         } else {
 556                 f2fs_put_page(page, 1);
 557
 558                 /* if ipage exists, blkaddr should be NEW_ADDR */
 559                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 560                 page = get_lock_data_page(inode, index, true);
 561                 if (IS_ERR(page))
 562                         return page;
 563         }
 564 got_it:
 565         if (new_i_size && i_size_read(inode) <
 566                                 ((loff_t)(index + 1) << PAGE_SHIFT)) {
 567                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 568                 /* Only the directory inode sets new_i_size */
 569                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 570         }
 571         return page;
 572 }
 573
 574 static int __allocate_data_block(struct dnode_of_data *dn)
 575 {
 576         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 577         struct f2fs_summary sum;
 578         struct node_info ni;
 579         int seg = CURSEG_WARM_DATA;
 580         pgoff_t fofs;
 581         blkcnt_t count = 1;
 582
 583         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 584                 return -EPERM;
 585
 586         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 587         if (dn->data_blkaddr == NEW_ADDR)
 588                 goto alloc;
 589
 590         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 591                 return -ENOSPC;
 592
 593 alloc:
 594         get_node_info(sbi, dn->nid, &ni);
 595         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 596
 597         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 598                 seg = CURSEG_DIRECT_IO;
 599
 600         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 601                                                                 &sum, seg);
 602         set_data_blkaddr(dn);
 603
 604         /* update i_size */
 605         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 606                                                         dn->ofs_in_node;
 607         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 608                 i_size_write(dn->inode,
 609                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 610         return 0;
 611 }
 612
 613 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 614 {
 615         struct inode *inode = file_inode(iocb->ki_filp);
 616         struct f2fs_map_blocks map;
 617         ssize_t ret = 0;
 618
 619         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 620         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 621         map.m_next_pgofs = NULL;
 622
 623         if (f2fs_encrypted_inode(inode))
 624                 return 0;
 625
 626         if (iocb->ki_flags & IOCB_DIRECT) {
 627                 ret = f2fs_convert_inline_inode(inode);
 628                 if (ret)
 629                         return ret;
 630                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 631         }
 632         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 633                 ret = f2fs_convert_inline_inode(inode);
 634                 if (ret)
 635                         return ret;
 636         }
 637         if (!f2fs_has_inline_data(inode))
 638                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 639         return ret;
 640 }
 641
 642 /*
 643  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 644  * f2fs_map_blocks structure.
 645  * If original data blocks are allocated, then give them to blockdev.
 646  * Otherwise,
 647  *     a. preallocate requested block addresses
 648  *     b. do not use extent cache for better performance
 649  *     c. give the block addresses to blockdev
 650  */
 651 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 652                                                 int create, int flag)
 653 {
 654         unsigned int maxblocks = map->m_len;
 655         struct dnode_of_data dn;
 656         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 657         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 658         pgoff_t pgofs, end_offset, end;
 659         int err = 0, ofs = 1;
 660         unsigned int ofs_in_node, last_ofs_in_node;
 661         blkcnt_t prealloc;
 662         struct extent_info ei;
 663         bool allocated = false;
 664         block_t blkaddr;
 665
 666         map->m_len = 0;
 667         map->m_flags = 0;
 668
 669         /* it only supports block size == page size */
 670         pgofs = (pgoff_t)map->m_lblk;
 671         end = pgofs + maxblocks;
 672
 673         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 674                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 675                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 676                 map->m_flags = F2FS_MAP_MAPPED;
 677                 goto out;
 678         }
 679
 680 next_dnode:
 681         if (create)
 682                 f2fs_lock_op(sbi);
 683
 684         /* When reading holes, we need its node page */
 685         set_new_dnode(&dn, inode, NULL, NULL, 0);
 686         err = get_dnode_of_data(&dn, pgofs, mode);
 687         if (err) {
 688                 if (flag == F2FS_GET_BLOCK_BMAP)
 689                         map->m_pblk = 0;
 690                 if (err == -ENOENT) {
 691                         err = 0;
 692                         if (map->m_next_pgofs)
 693                                 *map->m_next_pgofs =
 694                                         get_next_page_offset(&dn, pgofs);
 695                 }
 696                 goto unlock_out;
 697         }
 698
 699         prealloc = 0;
 700         ofs_in_node = dn.ofs_in_node;
 701         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 702
 703 next_block:
 704         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 705
 706         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 707                 if (create) {
 708                         if (unlikely(f2fs_cp_error(sbi))) {
 709                                 err = -EIO;
 710                                 goto sync_out;
 711                         }
 712                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 713                                 if (blkaddr == NULL_ADDR) {
 714                                         prealloc++;
 715                                         last_ofs_in_node = dn.ofs_in_node;
 716                                 }
 717                         } else {
 718                                 err = __allocate_data_block(&dn);
 719                                 if (!err) {
 720                                         set_inode_flag(F2FS_I(inode),
 721                                                         FI_APPEND_WRITE);
 722                                         allocated = true;
 723                                 }
 724                         }
 725                         if (err)
 726                                 goto sync_out;
 727                         map->m_flags = F2FS_MAP_NEW;
 728                         blkaddr = dn.data_blkaddr;
 729                 } else {
 730                         if (flag == F2FS_GET_BLOCK_BMAP) {
 731                                 map->m_pblk = 0;
 732                                 goto sync_out;
 733                         }
 734                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 735                                                 blkaddr == NULL_ADDR) {
 736                                 if (map->m_next_pgofs)
 737                                         *map->m_next_pgofs = pgofs + 1;
 738                         }
 739                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 740                                                 blkaddr != NEW_ADDR)
 741                                 goto sync_out;
 742                 }
 743         }
 744
 745         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 746                 goto skip;
 747
 748         if (map->m_len == 0) {
 749                 /* preallocated unwritten block should be mapped for fiemap. */
 750                 if (blkaddr == NEW_ADDR)
 751                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 752                 map->m_flags |= F2FS_MAP_MAPPED;
 753
 754                 map->m_pblk = blkaddr;
 755                 map->m_len = 1;
 756         } else if ((map->m_pblk != NEW_ADDR &&
 757                         blkaddr == (map->m_pblk + ofs)) ||
 758                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 759                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 760                 ofs++;
 761                 map->m_len++;
 762         } else {
 763                 goto sync_out;
 764         }
 765
 766 skip:
 767         dn.ofs_in_node++;
 768         pgofs++;
 769
 770         /* preallocate blocks in batch for one dnode page */
 771         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 772                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 773
 774                 dn.ofs_in_node = ofs_in_node;
 775                 err = reserve_new_blocks(&dn, prealloc);
 776                 if (err)
 777                         goto sync_out;
 778
 779                 map->m_len += dn.ofs_in_node - ofs_in_node;
 780                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 781                         err = -ENOSPC;
 782                         goto sync_out;
 783                 }
 784                 dn.ofs_in_node = end_offset;
 785         }
 786
 787         if (pgofs >= end)
 788                 goto sync_out;
 789         else if (dn.ofs_in_node < end_offset)
 790                 goto next_block;
 791
 792         if (allocated)
 793                 sync_inode_page(&dn);
 794         f2fs_put_dnode(&dn);
 795
 796         if (create) {
 797                 f2fs_unlock_op(sbi);
 798                 f2fs_balance_fs(sbi, allocated);
 799         }
 800         allocated = false;
 801         goto next_dnode;
 802
 803 sync_out:
 804         if (allocated)
 805                 sync_inode_page(&dn);
 806         f2fs_put_dnode(&dn);
 807 unlock_out:
 808         if (create) {
 809                 f2fs_unlock_op(sbi);
 810                 f2fs_balance_fs(sbi, allocated);
 811         }
 812 out:
 813         trace_f2fs_map_blocks(inode, map, err);
 814         return err;
 815 }
 816
 817 static int __get_data_block(struct inode *inode, sector_t iblock,
 818                         struct buffer_head *bh, int create, int flag,
 819                         pgoff_t *next_pgofs)
 820 {
 821         struct f2fs_map_blocks map;
 822         int ret;
 823
 824         map.m_lblk = iblock;
 825         map.m_len = bh->b_size >> inode->i_blkbits;
 826         map.m_next_pgofs = next_pgofs;
 827
 828         ret = f2fs_map_blocks(inode, &map, create, flag);
 829         if (!ret) {
 830                 map_bh(bh, inode->i_sb, map.m_pblk);
 831                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 832                 bh->b_size = map.m_len << inode->i_blkbits;
 833         }
 834         return ret;
 835 }
 836
 837 static int get_data_block(struct inode *inode, sector_t iblock,
 838                         struct buffer_head *bh_result, int create, int flag,
 839                         pgoff_t *next_pgofs)
 840 {
 841         return __get_data_block(inode, iblock, bh_result, create,
 842                                                         flag, next_pgofs);
 843 }
 844
 845 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 846                         struct buffer_head *bh_result, int create)
 847 {
 848         return __get_data_block(inode, iblock, bh_result, create,
 849                                                 F2FS_GET_BLOCK_DIO, NULL);
 850 }
 851
 852 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 853                         struct buffer_head *bh_result, int create)
 854 {
 855         /* Block number less than F2FS MAX BLOCKS */
 856         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 857                 return -EFBIG;
 858
 859         return __get_data_block(inode, iblock, bh_result, create,
 860                                                 F2FS_GET_BLOCK_BMAP, NULL);
 861 }
 862
 863 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 864 {
 865         return (offset >> inode->i_blkbits);
 866 }
 867
 868 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 869 {
 870         return (blk << inode->i_blkbits);
 871 }
 872
 873 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 874                 u64 start, u64 len)
 875 {
 876         struct buffer_head map_bh;
 877         sector_t start_blk, last_blk;
 878         pgoff_t next_pgofs;
 879         loff_t isize;
 880         u64 logical = 0, phys = 0, size = 0;
 881         u32 flags = 0;
 882         int ret = 0;
 883
 884         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 885         if (ret)
 886                 return ret;
 887
 888         if (f2fs_has_inline_data(inode)) {
 889                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 890                 if (ret != -EAGAIN)
 891                         return ret;
 892         }
 893
 894         inode_lock(inode);
 895
 896         isize = i_size_read(inode);
 897         if (start >= isize)
 898                 goto out;
 899
 900         if (start + len > isize)
 901                 len = isize - start;
 902
 903         if (logical_to_blk(inode, len) == 0)
 904                 len = blk_to_logical(inode, 1);
 905
 906         start_blk = logical_to_blk(inode, start);
 907         last_blk = logical_to_blk(inode, start + len - 1);
 908
 909 next:
 910         memset(&map_bh, 0, sizeof(struct buffer_head));
 911         map_bh.b_size = len;
 912
 913         ret = get_data_block(inode, start_blk, &map_bh, 0,
 914                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 915         if (ret)
 916                 goto out;
 917
 918         /* HOLE */
 919         if (!buffer_mapped(&map_bh)) {
 920                 start_blk = next_pgofs;
 921                 /* Go through holes util pass the EOF */
 922                 if (blk_to_logical(inode, start_blk) < isize)
 923                         goto prep_next;
 924                 /* Found a hole beyond isize means no more extents.
 925                  * Note that the premise is that filesystems don't
 926                  * punch holes beyond isize and keep size unchanged.
 927                  */
 928                 flags |= FIEMAP_EXTENT_LAST;
 929         }
 930
 931         if (size) {
 932                 if (f2fs_encrypted_inode(inode))
 933                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 934
 935                 ret = fiemap_fill_next_extent(fieinfo, logical,
 936                                 phys, size, flags);
 937         }
 938
 939         if (start_blk > last_blk || ret)
 940                 goto out;
 941
 942         logical = blk_to_logical(inode, start_blk);
 943         phys = blk_to_logical(inode, map_bh.b_blocknr);
 944         size = map_bh.b_size;
 945         flags = 0;
 946         if (buffer_unwritten(&map_bh))
 947                 flags = FIEMAP_EXTENT_UNWRITTEN;
 948
 949         start_blk += logical_to_blk(inode, size);
 950
 951 prep_next:
 952         cond_resched();
 953         if (fatal_signal_pending(current))
 954                 ret = -EINTR;
 955         else
 956                 goto next;
 957 out:
 958         if (ret == 1)
 959                 ret = 0;
 960
 961         inode_unlock(inode);
 962         return ret;
 963 }
 964
 965 /*
 966  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 967  * Major change was from block_size == page_size in f2fs by default.
 968  */
 969 static int f2fs_mpage_readpages(struct address_space *mapping,
 970                         struct list_head *pages, struct page *page,
 971                         unsigned nr_pages)
 972 {
 973         struct bio *bio = NULL;
 974         unsigned page_idx;
 975         sector_t last_block_in_bio = 0;
 976         struct inode *inode = mapping->host;
 977         const unsigned blkbits = inode->i_blkbits;
 978         const unsigned blocksize = 1 << blkbits;
 979         sector_t block_in_file;
 980         sector_t last_block;
 981         sector_t last_block_in_file;
 982         sector_t block_nr;
 983         struct block_device *bdev = inode->i_sb->s_bdev;
 984         struct f2fs_map_blocks map;
 985
 986         map.m_pblk = 0;
 987         map.m_lblk = 0;
 988         map.m_len = 0;
 989         map.m_flags = 0;
 990         map.m_next_pgofs = NULL;
 991
 992         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 993
 994                 prefetchw(&page->flags);
 995                 if (pages) {
 996                         page = list_entry(pages->prev, struct page, lru);
 997                         list_del(&page->lru);
 998                         if (add_to_page_cache_lru(page, mapping,
 999                                                   page->index, GFP_KERNEL))
1000                                 goto next_page;
1001                 }
1002
1003                 block_in_file = (sector_t)page->index;
1004                 last_block = block_in_file + nr_pages;
1005                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1006                                                                 blkbits;
1007                 if (last_block > last_block_in_file)
1008                         last_block = last_block_in_file;
1009
1010                 /*
1011                  * Map blocks using the previous result first.
1012                  */
1013                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1014                                 block_in_file > map.m_lblk &&
1015                                 block_in_file < (map.m_lblk + map.m_len))
1016                         goto got_it;
1017
1018                 /*
1019                  * Then do more f2fs_map_blocks() calls until we are
1020                  * done with this page.
1021                  */
1022                 map.m_flags = 0;
1023
1024                 if (block_in_file < last_block) {
1025                         map.m_lblk = block_in_file;
1026                         map.m_len = last_block - block_in_file;
1027
1028                         if (f2fs_map_blocks(inode, &map, 0,
1029                                                 F2FS_GET_BLOCK_READ))
1030                                 goto set_error_page;
1031                 }
1032 got_it:
1033                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1034                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1035                         SetPageMappedToDisk(page);
1036
1037                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1038                                 SetPageUptodate(page);
1039                                 goto confused;
1040                         }
1041                 } else {
1042                         zero_user_segment(page, 0, PAGE_SIZE);
1043                         SetPageUptodate(page);
1044                         unlock_page(page);
1045                         goto next_page;
1046                 }
1047
1048                 /*
1049                  * This page will go to BIO.  Do we need to send this
1050                  * BIO off first?
1051                  */
1052                 if (bio && (last_block_in_bio != block_nr - 1)) {
1053 submit_and_realloc:
1054                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1055                         bio = NULL;
1056                 }
1057                 if (bio == NULL) {
1058                         struct fscrypt_ctx *ctx = NULL;
1059
1060                         if (f2fs_encrypted_inode(inode) &&
1061                                         S_ISREG(inode->i_mode)) {
1062
1063                                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1064                                 if (IS_ERR(ctx))
1065                                         goto set_error_page;
1066
1067                                 /* wait the page to be moved by cleaning */
1068                                 f2fs_wait_on_encrypted_page_writeback(
1069                                                 F2FS_I_SB(inode), block_nr);
1070                         }
1071
1072                         bio = bio_alloc(GFP_KERNEL,
1073                                 min_t(int, nr_pages, BIO_MAX_PAGES));
1074                         if (!bio) {
1075                                 if (ctx)
1076                                         fscrypt_release_ctx(ctx);
1077                                 goto set_error_page;
1078                         }
1079                         bio->bi_bdev = bdev;
1080                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1081                         bio->bi_end_io = f2fs_read_end_io;
1082                         bio->bi_private = ctx;
1083                 }
1084
1085                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1086                         goto submit_and_realloc;
1087
1088                 last_block_in_bio = block_nr;
1089                 goto next_page;
1090 set_error_page:
1091                 SetPageError(page);
1092                 zero_user_segment(page, 0, PAGE_SIZE);
1093                 unlock_page(page);
1094                 goto next_page;
1095 confused:
1096                 if (bio) {
1097                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1098                         bio = NULL;
1099                 }
1100                 unlock_page(page);
1101 next_page:
1102                 if (pages)
1103                         put_page(page);
1104         }
1105         BUG_ON(pages && !list_empty(pages));
1106         if (bio)
1107                 __submit_bio(F2FS_I_SB(inode), READ, bio);
1108         return 0;
1109 }
1110
1111 static int f2fs_read_data_page(struct file *file, struct page *page)
1112 {
1113         struct inode *inode = page->mapping->host;
1114         int ret = -EAGAIN;
1115
1116         trace_f2fs_readpage(page, DATA);
1117
1118         /* If the file has inline data, try to read it directly */
1119         if (f2fs_has_inline_data(inode))
1120                 ret = f2fs_read_inline_data(inode, page);
1121         if (ret == -EAGAIN)
1122                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1123         return ret;
1124 }
1125
1126 static int f2fs_read_data_pages(struct file *file,
1127                         struct address_space *mapping,
1128                         struct list_head *pages, unsigned nr_pages)
1129 {
1130         struct inode *inode = file->f_mapping->host;
1131         struct page *page = list_entry(pages->prev, struct page, lru);
1132
1133         trace_f2fs_readpages(inode, page, nr_pages);
1134
1135         /* If the file has inline data, skip readpages */
1136         if (f2fs_has_inline_data(inode))
1137                 return 0;
1138
1139         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1140 }
1141
1142 int do_write_data_page(struct f2fs_io_info *fio)
1143 {
1144         struct page *page = fio->page;
1145         struct inode *inode = page->mapping->host;
1146         struct dnode_of_data dn;
1147         int err = 0;
1148
1149         set_new_dnode(&dn, inode, NULL, NULL, 0);
1150         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1151         if (err)
1152                 return err;
1153
1154         fio->old_blkaddr = dn.data_blkaddr;
1155
1156         /* This page is already truncated */
1157         if (fio->old_blkaddr == NULL_ADDR) {
1158                 ClearPageUptodate(page);
1159                 goto out_writepage;
1160         }
1161
1162         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1163                 gfp_t gfp_flags = GFP_NOFS;
1164
1165                 /* wait for GCed encrypted page writeback */
1166                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1167                                                         fio->old_blkaddr);
1168 retry_encrypt:
1169                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1170                                                                 gfp_flags);
1171                 if (IS_ERR(fio->encrypted_page)) {
1172                         err = PTR_ERR(fio->encrypted_page);
1173                         if (err == -ENOMEM) {
1174                                 /* flush pending ios and wait for a while */
1175                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1176                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1177                                 gfp_flags |= __GFP_NOFAIL;
1178                                 err = 0;
1179                                 goto retry_encrypt;
1180                         }
1181                         goto out_writepage;
1182                 }
1183         }
1184
1185         set_page_writeback(page);
1186
1187         /*
1188          * If current allocation needs SSR,
1189          * it had better in-place writes for updated data.
1190          */
1191         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1192                         !is_cold_data(page) &&
1193                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1194                         need_inplace_update(inode))) {
1195                 rewrite_data_page(fio);
1196                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1197                 trace_f2fs_do_write_data_page(page, IPU);
1198         } else {
1199                 write_data_page(&dn, fio);
1200                 trace_f2fs_do_write_data_page(page, OPU);
1201                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1202                 if (page->index == 0)
1203                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1204         }
1205 out_writepage:
1206         f2fs_put_dnode(&dn);
1207         return err;
1208 }
1209
1210 static int f2fs_write_data_page(struct page *page,
1211                                         struct writeback_control *wbc)
1212 {
1213         struct inode *inode = page->mapping->host;
1214         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1215         loff_t i_size = i_size_read(inode);
1216         const pgoff_t end_index = ((unsigned long long) i_size)
1217                                                         >> PAGE_SHIFT;
1218         unsigned offset = 0;
1219         bool need_balance_fs = false;
1220         int err = 0;
1221         struct f2fs_io_info fio = {
1222                 .sbi = sbi,
1223                 .type = DATA,
1224                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1225                 .page = page,
1226                 .encrypted_page = NULL,
1227         };
1228
1229         trace_f2fs_writepage(page, DATA);
1230
1231         if (page->index < end_index)
1232                 goto write;
1233
1234         /*
1235          * If the offset is out-of-range of file size,
1236          * this page does not have to be written to disk.
1237          */
1238         offset = i_size & (PAGE_SIZE - 1);
1239         if ((page->index >= end_index + 1) || !offset)
1240                 goto out;
1241
1242         zero_user_segment(page, offset, PAGE_SIZE);
1243 write:
1244         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1245                 goto redirty_out;
1246         if (f2fs_is_drop_cache(inode))
1247                 goto out;
1248         /* we should not write 0'th page having journal header */
1249         if (f2fs_is_volatile_file(inode) && (!page->index ||
1250                         (!wbc->for_reclaim &&
1251                         available_free_memory(sbi, BASE_CHECK))))
1252                 goto redirty_out;
1253
1254         /* Dentry blocks are controlled by checkpoint */
1255         if (S_ISDIR(inode->i_mode)) {
1256                 if (unlikely(f2fs_cp_error(sbi)))
1257                         goto redirty_out;
1258                 err = do_write_data_page(&fio);
1259                 goto done;
1260         }
1261
1262         /* we should bypass data pages to proceed the kworkder jobs */
1263         if (unlikely(f2fs_cp_error(sbi))) {
1264                 SetPageError(page);
1265                 goto out;
1266         }
1267
1268         if (!wbc->for_reclaim)
1269                 need_balance_fs = true;
1270         else if (has_not_enough_free_secs(sbi, 0))
1271                 goto redirty_out;
1272
1273         err = -EAGAIN;
1274         f2fs_lock_op(sbi);
1275         if (f2fs_has_inline_data(inode))
1276                 err = f2fs_write_inline_data(inode, page);
1277         if (err == -EAGAIN)
1278                 err = do_write_data_page(&fio);
1279         f2fs_unlock_op(sbi);
1280 done:
1281         if (err && err != -ENOENT)
1282                 goto redirty_out;
1283
1284         clear_cold_data(page);
1285 out:
1286         inode_dec_dirty_pages(inode);
1287         if (err)
1288                 ClearPageUptodate(page);
1289
1290         if (wbc->for_reclaim) {
1291                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1292                 remove_dirty_inode(inode);
1293         }
1294
1295         unlock_page(page);
1296         f2fs_balance_fs(sbi, need_balance_fs);
1297
1298         if (unlikely(f2fs_cp_error(sbi)))
1299                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1300
1301         return 0;
1302
1303 redirty_out:
1304         redirty_page_for_writepage(wbc, page);
1305         return AOP_WRITEPAGE_ACTIVATE;
1306 }
1307
1308 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1309                         void *data)
1310 {
1311         struct address_space *mapping = data;
1312         int ret = mapping->a_ops->writepage(page, wbc);
1313         mapping_set_error(mapping, ret);
1314         return ret;
1315 }
1316
1317 /*
1318  * This function was copied from write_cche_pages from mm/page-writeback.c.
1319  * The major change is making write step of cold data page separately from
1320  * warm/hot data page.
1321  */
1322 static int f2fs_write_cache_pages(struct address_space *mapping,
1323                         struct writeback_control *wbc, writepage_t writepage,
1324                         void *data)
1325 {
1326         int ret = 0;
1327         int done = 0;
1328         struct pagevec pvec;
1329         int nr_pages;
1330         pgoff_t uninitialized_var(writeback_index);
1331         pgoff_t index;
1332         pgoff_t end;            /* Inclusive */
1333         pgoff_t done_index;
1334         int cycled;
1335         int range_whole = 0;
1336         int tag;
1337         int step = 0;
1338
1339         pagevec_init(&pvec, 0);
1340 next:
1341         if (wbc->range_cyclic) {
1342                 writeback_index = mapping->writeback_index; /* prev offset */
1343                 index = writeback_index;
1344                 if (index == 0)
1345                         cycled = 1;
1346                 else
1347                         cycled = 0;
1348                 end = -1;
1349         } else {
1350                 index = wbc->range_start >> PAGE_SHIFT;
1351                 end = wbc->range_end >> PAGE_SHIFT;
1352                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1353                         range_whole = 1;
1354                 cycled = 1; /* ignore range_cyclic tests */
1355         }
1356         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1357                 tag = PAGECACHE_TAG_TOWRITE;
1358         else
1359                 tag = PAGECACHE_TAG_DIRTY;
1360 retry:
1361         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1362                 tag_pages_for_writeback(mapping, index, end);
1363         done_index = index;
1364         while (!done && (index <= end)) {
1365                 int i;
1366
1367                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1368                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1369                 if (nr_pages == 0)
1370                         break;
1371
1372                 for (i = 0; i < nr_pages; i++) {
1373                         struct page *page = pvec.pages[i];
1374
1375                         if (page->index > end) {
1376                                 done = 1;
1377                                 break;
1378                         }
1379
1380                         done_index = page->index;
1381
1382                         lock_page(page);
1383
1384                         if (unlikely(page->mapping != mapping)) {
1385 continue_unlock:
1386                                 unlock_page(page);
1387                                 continue;
1388                         }
1389
1390                         if (!PageDirty(page)) {
1391                                 /* someone wrote it for us */
1392                                 goto continue_unlock;
1393                         }
1394
1395                         if (step == is_cold_data(page))
1396                                 goto continue_unlock;
1397
1398                         if (PageWriteback(page)) {
1399                                 if (wbc->sync_mode != WB_SYNC_NONE)
1400                                         f2fs_wait_on_page_writeback(page,
1401                                                                 DATA, true);
1402                                 else
1403                                         goto continue_unlock;
1404                         }
1405
1406                         BUG_ON(PageWriteback(page));
1407                         if (!clear_page_dirty_for_io(page))
1408                                 goto continue_unlock;
1409
1410                         ret = (*writepage)(page, wbc, data);
1411                         if (unlikely(ret)) {
1412                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1413                                         unlock_page(page);
1414                                         ret = 0;
1415                                 } else {
1416                                         done_index = page->index + 1;
1417                                         done = 1;
1418                                         break;
1419                                 }
1420                         }
1421
1422                         if (--wbc->nr_to_write <= 0 &&
1423                             wbc->sync_mode == WB_SYNC_NONE) {
1424                                 done = 1;
1425                                 break;
1426                         }
1427                 }
1428                 pagevec_release(&pvec);
1429                 cond_resched();
1430         }
1431
1432         if (step < 1) {
1433                 step++;
1434                 goto next;
1435         }
1436
1437         if (!cycled && !done) {
1438                 cycled = 1;
1439                 index = 0;
1440                 end = writeback_index - 1;
1441                 goto retry;
1442         }
1443         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1444                 mapping->writeback_index = done_index;
1445
1446         return ret;
1447 }
1448
1449 static int f2fs_write_data_pages(struct address_space *mapping,
1450                             struct writeback_control *wbc)
1451 {
1452         struct inode *inode = mapping->host;
1453         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1454         bool locked = false;
1455         int ret;
1456         long diff;
1457
1458         /* deal with chardevs and other special file */
1459         if (!mapping->a_ops->writepage)
1460                 return 0;
1461
1462         /* skip writing if there is no dirty page in this inode */
1463         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1464                 return 0;
1465
1466         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1467                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1468                         available_free_memory(sbi, DIRTY_DENTS))
1469                 goto skip_write;
1470
1471         /* skip writing during file defragment */
1472         if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1473                 goto skip_write;
1474
1475         /* during POR, we don't need to trigger writepage at all. */
1476         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1477                 goto skip_write;
1478
1479         trace_f2fs_writepages(mapping->host, wbc, DATA);
1480
1481         diff = nr_pages_to_write(sbi, DATA, wbc);
1482
1483         if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1484                 mutex_lock(&sbi->writepages);
1485                 locked = true;
1486         }
1487         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1488         f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1489         if (locked)
1490                 mutex_unlock(&sbi->writepages);
1491
1492         remove_dirty_inode(inode);
1493
1494         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1495         return ret;
1496
1497 skip_write:
1498         wbc->pages_skipped += get_dirty_pages(inode);
1499         trace_f2fs_writepages(mapping->host, wbc, DATA);
1500         return 0;
1501 }
1502
1503 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1504 {
1505         struct inode *inode = mapping->host;
1506         loff_t i_size = i_size_read(inode);
1507
1508         if (to > i_size) {
1509                 truncate_pagecache(inode, i_size);
1510                 truncate_blocks(inode, i_size, true);
1511         }
1512 }
1513
1514 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1515                         struct page *page, loff_t pos, unsigned len,
1516                         block_t *blk_addr, bool *node_changed)
1517 {
1518         struct inode *inode = page->mapping->host;
1519         pgoff_t index = page->index;
1520         struct dnode_of_data dn;
1521         struct page *ipage;
1522         bool locked = false;
1523         struct extent_info ei;
1524         int err = 0;
1525
1526         /*
1527          * we already allocated all the blocks, so we don't need to get
1528          * the block addresses when there is no need to fill the page.
1529          */
1530         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1531                                         len == PAGE_SIZE)
1532                 return 0;
1533
1534         if (f2fs_has_inline_data(inode) ||
1535                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1536                 f2fs_lock_op(sbi);
1537                 locked = true;
1538         }
1539 restart:
1540         /* check inline_data */
1541         ipage = get_node_page(sbi, inode->i_ino);
1542         if (IS_ERR(ipage)) {
1543                 err = PTR_ERR(ipage);
1544                 goto unlock_out;
1545         }
1546
1547         set_new_dnode(&dn, inode, ipage, ipage, 0);
1548
1549         if (f2fs_has_inline_data(inode)) {
1550                 if (pos + len <= MAX_INLINE_DATA) {
1551                         read_inline_data(page, ipage);
1552                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1553                         if (inode->i_nlink)
1554                                 set_inline_node(ipage);
1555                 } else {
1556                         err = f2fs_convert_inline_page(&dn, page);
1557                         if (err)
1558                                 goto out;
1559                         if (dn.data_blkaddr == NULL_ADDR)
1560                                 err = f2fs_get_block(&dn, index);
1561                 }
1562         } else if (locked) {
1563                 err = f2fs_get_block(&dn, index);
1564         } else {
1565                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1566                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1567                 } else {
1568                         /* hole case */
1569                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1570                         if (err || dn.data_blkaddr == NULL_ADDR) {
1571                                 f2fs_put_dnode(&dn);
1572                                 f2fs_lock_op(sbi);
1573                                 locked = true;
1574                                 goto restart;
1575                         }
1576                 }
1577         }
1578
1579         /* convert_inline_page can make node_changed */
1580         *blk_addr = dn.data_blkaddr;
1581         *node_changed = dn.node_changed;
1582 out:
1583         f2fs_put_dnode(&dn);
1584 unlock_out:
1585         if (locked)
1586                 f2fs_unlock_op(sbi);
1587         return err;
1588 }
1589
1590 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1591                 loff_t pos, unsigned len, unsigned flags,
1592                 struct page **pagep, void **fsdata)
1593 {
1594         struct inode *inode = mapping->host;
1595         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1596         struct page *page = NULL;
1597         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1598         bool need_balance = false;
1599         block_t blkaddr = NULL_ADDR;
1600         int err = 0;
1601
1602         trace_f2fs_write_begin(inode, pos, len, flags);
1603
1604         /*
1605          * We should check this at this moment to avoid deadlock on inode page
1606          * and #0 page. The locking rule for inline_data conversion should be:
1607          * lock_page(page #0) -> lock_page(inode_page)
1608          */
1609         if (index != 0) {
1610                 err = f2fs_convert_inline_inode(inode);
1611                 if (err)
1612                         goto fail;
1613         }
1614 repeat:
1615         page = grab_cache_page_write_begin(mapping, index, flags);
1616         if (!page) {
1617                 err = -ENOMEM;
1618                 goto fail;
1619         }
1620
1621         *pagep = page;
1622
1623         err = prepare_write_begin(sbi, page, pos, len,
1624                                         &blkaddr, &need_balance);
1625         if (err)
1626                 goto fail;
1627
1628         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1629                 unlock_page(page);
1630                 f2fs_balance_fs(sbi, true);
1631                 lock_page(page);
1632                 if (page->mapping != mapping) {
1633                         /* The page got truncated from under us */
1634                         f2fs_put_page(page, 1);
1635                         goto repeat;
1636                 }
1637         }
1638
1639         f2fs_wait_on_page_writeback(page, DATA, false);
1640
1641         /* wait for GCed encrypted page writeback */
1642         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1643                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1644
1645         if (len == PAGE_SIZE)
1646                 goto out_update;
1647         if (PageUptodate(page))
1648                 goto out_clear;
1649
1650         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1651                 unsigned start = pos & (PAGE_SIZE - 1);
1652                 unsigned end = start + len;
1653
1654                 /* Reading beyond i_size is simple: memset to zero */
1655                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1656                 goto out_update;
1657         }
1658
1659         if (blkaddr == NEW_ADDR) {
1660                 zero_user_segment(page, 0, PAGE_SIZE);
1661         } else {
1662                 struct f2fs_io_info fio = {
1663                         .sbi = sbi,
1664                         .type = DATA,
1665                         .rw = READ_SYNC,
1666                         .old_blkaddr = blkaddr,
1667                         .new_blkaddr = blkaddr,
1668                         .page = page,
1669                         .encrypted_page = NULL,
1670                 };
1671                 err = f2fs_submit_page_bio(&fio);
1672                 if (err)
1673                         goto fail;
1674
1675                 lock_page(page);
1676                 if (unlikely(!PageUptodate(page))) {
1677                         err = -EIO;
1678                         goto fail;
1679                 }
1680                 if (unlikely(page->mapping != mapping)) {
1681                         f2fs_put_page(page, 1);
1682                         goto repeat;
1683                 }
1684
1685                 /* avoid symlink page */
1686                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1687                         err = fscrypt_decrypt_page(page);
1688                         if (err)
1689                                 goto fail;
1690                 }
1691         }
1692 out_update:
1693         SetPageUptodate(page);
1694 out_clear:
1695         clear_cold_data(page);
1696         return 0;
1697
1698 fail:
1699         f2fs_put_page(page, 1);
1700         f2fs_write_failed(mapping, pos + len);
1701         return err;
1702 }
1703
1704 static int f2fs_write_end(struct file *file,
1705                         struct address_space *mapping,
1706                         loff_t pos, unsigned len, unsigned copied,
1707                         struct page *page, void *fsdata)
1708 {
1709         struct inode *inode = page->mapping->host;
1710
1711         trace_f2fs_write_end(inode, pos, len, copied);
1712
1713         set_page_dirty(page);
1714
1715         if (pos + copied > i_size_read(inode)) {
1716                 i_size_write(inode, pos + copied);
1717                 mark_inode_dirty(inode);
1718         }
1719
1720         f2fs_put_page(page, 1);
1721         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1722         return copied;
1723 }
1724
1725 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1726                            loff_t offset)
1727 {
1728         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1729
1730         if (offset & blocksize_mask)
1731                 return -EINVAL;
1732
1733         if (iov_iter_alignment(iter) & blocksize_mask)
1734                 return -EINVAL;
1735
1736         return 0;
1737 }
1738
1739 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1740 {
1741         struct address_space *mapping = iocb->ki_filp->f_mapping;
1742         struct inode *inode = mapping->host;
1743         size_t count = iov_iter_count(iter);
1744         loff_t offset = iocb->ki_pos;
1745         int err;
1746
1747         err = check_direct_IO(inode, iter, offset);
1748         if (err)
1749                 return err;
1750
1751         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1752                 return 0;
1753
1754         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1755
1756         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1757         if (iov_iter_rw(iter) == WRITE) {
1758                 if (err > 0)
1759                         set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1760                 else if (err < 0)
1761                         f2fs_write_failed(mapping, offset + count);
1762         }
1763
1764         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1765
1766         return err;
1767 }
1768
1769 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1770                                                         unsigned int length)
1771 {
1772         struct inode *inode = page->mapping->host;
1773         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1774
1775         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1776                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1777                 return;
1778
1779         if (PageDirty(page)) {
1780                 if (inode->i_ino == F2FS_META_INO(sbi))
1781                         dec_page_count(sbi, F2FS_DIRTY_META);
1782                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1783                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1784                 else
1785                         inode_dec_dirty_pages(inode);
1786         }
1787
1788         /* This is atomic written page, keep Private */
1789         if (IS_ATOMIC_WRITTEN_PAGE(page))
1790                 return;
1791
1792         set_page_private(page, 0);
1793         ClearPagePrivate(page);
1794 }
1795
1796 int f2fs_release_page(struct page *page, gfp_t wait)
1797 {
1798         /* If this is dirty page, keep PagePrivate */
1799         if (PageDirty(page))
1800                 return 0;
1801
1802         /* This is atomic written page, keep Private */
1803         if (IS_ATOMIC_WRITTEN_PAGE(page))
1804                 return 0;
1805
1806         set_page_private(page, 0);
1807         ClearPagePrivate(page);
1808         return 1;
1809 }
1810
1811 static int f2fs_set_data_page_dirty(struct page *page)
1812 {
1813         struct address_space *mapping = page->mapping;
1814         struct inode *inode = mapping->host;
1815
1816         trace_f2fs_set_page_dirty(page, DATA);
1817
1818         SetPageUptodate(page);
1819
1820         if (f2fs_is_atomic_file(inode)) {
1821                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1822                         register_inmem_page(inode, page);
1823                         return 1;
1824                 }
1825                 /*
1826                  * Previously, this page has been registered, we just
1827                  * return here.
1828                  */
1829                 return 0;
1830         }
1831
1832         if (!PageDirty(page)) {
1833                 __set_page_dirty_nobuffers(page);
1834                 update_dirty_page(inode, page);
1835                 return 1;
1836         }
1837         return 0;
1838 }
1839
1840 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1841 {
1842         struct inode *inode = mapping->host;
1843
1844         if (f2fs_has_inline_data(inode))
1845                 return 0;
1846
1847         /* make sure allocating whole blocks */
1848         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1849                 filemap_write_and_wait(mapping);
1850
1851         return generic_block_bmap(mapping, block, get_data_block_bmap);
1852 }
1853
1854 const struct address_space_operations f2fs_dblock_aops = {
1855         .readpage       = f2fs_read_data_page,
1856         .readpages      = f2fs_read_data_pages,
1857         .writepage      = f2fs_write_data_page,
1858         .writepages     = f2fs_write_data_pages,
1859         .write_begin    = f2fs_write_begin,
1860         .write_end      = f2fs_write_end,
1861         .set_page_dirty = f2fs_set_data_page_dirty,
1862         .invalidatepage = f2fs_invalidate_page,
1863         .releasepage    = f2fs_release_page,
1864         .direct_IO      = f2fs_direct_IO,
1865         .bmap           = f2fs_bmap,
1866 };