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         bio->bi_rw = rw;
 106         submit_bio(bio);
 107 }
 108
 109 static void __submit_merged_bio(struct f2fs_bio_info *io)
 110 {
 111         struct f2fs_io_info *fio = &io->fio;
 112
 113         if (!io->bio)
 114                 return;
 115
 116         if (is_read_io(fio->rw))
 117                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 118         else
 119                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 120
 121         __submit_bio(io->sbi, fio->rw, io->bio);
 122         io->bio = NULL;
 123 }
 124
 125 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 126                                                 struct page *page, nid_t ino)
 127 {
 128         struct bio_vec *bvec;
 129         struct page *target;
 130         int i;
 131
 132         if (!io->bio)
 133                 return false;
 134
 135         if (!inode && !page && !ino)
 136                 return true;
 137
 138         bio_for_each_segment_all(bvec, io->bio, i) {
 139
 140                 if (bvec->bv_page->mapping)
 141                         target = bvec->bv_page;
 142                 else
 143                         target = fscrypt_control_page(bvec->bv_page);
 144
 145                 if (inode && inode == target->mapping->host)
 146                         return true;
 147                 if (page && page == target)
 148                         return true;
 149                 if (ino && ino == ino_of_node(target))
 150                         return true;
 151         }
 152
 153         return false;
 154 }
 155
 156 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 157                                                 struct page *page, nid_t ino,
 158                                                 enum page_type type)
 159 {
 160         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 161         struct f2fs_bio_info *io = &sbi->write_io[btype];
 162         bool ret;
 163
 164         down_read(&io->io_rwsem);
 165         ret = __has_merged_page(io, inode, page, ino);
 166         up_read(&io->io_rwsem);
 167         return ret;
 168 }
 169
 170 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 171                                 struct inode *inode, struct page *page,
 172                                 nid_t ino, enum page_type type, int rw)
 173 {
 174         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 175         struct f2fs_bio_info *io;
 176
 177         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 178
 179         down_write(&io->io_rwsem);
 180
 181         if (!__has_merged_page(io, inode, page, ino))
 182                 goto out;
 183
 184         /* change META to META_FLUSH in the checkpoint procedure */
 185         if (type >= META_FLUSH) {
 186                 io->fio.type = META_FLUSH;
 187                 if (test_opt(sbi, NOBARRIER))
 188                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 189                 else
 190                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 191         }
 192         __submit_merged_bio(io);
 193 out:
 194         up_write(&io->io_rwsem);
 195 }
 196
 197 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 198                                                                         int rw)
 199 {
 200         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 201 }
 202
 203 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 204                                 struct inode *inode, struct page *page,
 205                                 nid_t ino, enum page_type type, int rw)
 206 {
 207         if (has_merged_page(sbi, inode, page, ino, type))
 208                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 209 }
 210
 211 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 212 {
 213         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 214         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 215         f2fs_submit_merged_bio(sbi, META, WRITE);
 216 }
 217
 218 /*
 219  * Fill the locked page with data located in the block address.
 220  * Return unlocked page.
 221  */
 222 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 223 {
 224         struct bio *bio;
 225         struct page *page = fio->encrypted_page ?
 226                         fio->encrypted_page : fio->page;
 227
 228         trace_f2fs_submit_page_bio(page, fio);
 229         f2fs_trace_ios(fio, 0);
 230
 231         /* Allocate a new bio */
 232         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
 233
 234         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 235                 bio_put(bio);
 236                 return -EFAULT;
 237         }
 238
 239         __submit_bio(fio->sbi, fio->rw, bio);
 240         return 0;
 241 }
 242
 243 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 244 {
 245         struct f2fs_sb_info *sbi = fio->sbi;
 246         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 247         struct f2fs_bio_info *io;
 248         bool is_read = is_read_io(fio->rw);
 249         struct page *bio_page;
 250
 251         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 252
 253         if (fio->old_blkaddr != NEW_ADDR)
 254                 verify_block_addr(sbi, fio->old_blkaddr);
 255         verify_block_addr(sbi, fio->new_blkaddr);
 256
 257         down_write(&io->io_rwsem);
 258
 259         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 260                                                 io->fio.rw != fio->rw))
 261                 __submit_merged_bio(io);
 262 alloc_new:
 263         if (io->bio == NULL) {
 264                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 265
 266                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 267                                                 bio_blocks, is_read);
 268                 io->fio = *fio;
 269         }
 270
 271         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 272
 273         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 274                                                         PAGE_SIZE) {
 275                 __submit_merged_bio(io);
 276                 goto alloc_new;
 277         }
 278
 279         io->last_block_in_bio = fio->new_blkaddr;
 280         f2fs_trace_ios(fio, 0);
 281
 282         up_write(&io->io_rwsem);
 283         trace_f2fs_submit_page_mbio(fio->page, fio);
 284 }
 285
 286 static void __set_data_blkaddr(struct dnode_of_data *dn)
 287 {
 288         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 289         __le32 *addr_array;
 290
 291         /* Get physical address of data block */
 292         addr_array = blkaddr_in_node(rn);
 293         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 294 }
 295
 296 /*
 297  * Lock ordering for the change of data block address:
 298  * ->data_page
 299  *  ->node_page
 300  *    update block addresses in the node page
 301  */
 302 void set_data_blkaddr(struct dnode_of_data *dn)
 303 {
 304         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 305         __set_data_blkaddr(dn);
 306         if (set_page_dirty(dn->node_page))
 307                 dn->node_changed = true;
 308 }
 309
 310 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 311 {
 312         dn->data_blkaddr = blkaddr;
 313         set_data_blkaddr(dn);
 314         f2fs_update_extent_cache(dn);
 315 }
 316
 317 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 318 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 319 {
 320         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 321
 322         if (!count)
 323                 return 0;
 324
 325         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 326                 return -EPERM;
 327         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 328                 return -ENOSPC;
 329
 330         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 331                                                 dn->ofs_in_node, count);
 332
 333         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 334
 335         for (; count > 0; dn->ofs_in_node++) {
 336                 block_t blkaddr =
 337                         datablock_addr(dn->node_page, dn->ofs_in_node);
 338                 if (blkaddr == NULL_ADDR) {
 339                         dn->data_blkaddr = NEW_ADDR;
 340                         __set_data_blkaddr(dn);
 341                         count--;
 342                 }
 343         }
 344
 345         if (set_page_dirty(dn->node_page))
 346                 dn->node_changed = true;
 347
 348         mark_inode_dirty(dn->inode);
 349         sync_inode_page(dn);
 350         return 0;
 351 }
 352
 353 /* Should keep dn->ofs_in_node unchanged */
 354 int reserve_new_block(struct dnode_of_data *dn)
 355 {
 356         unsigned int ofs_in_node = dn->ofs_in_node;
 357         int ret;
 358
 359         ret = reserve_new_blocks(dn, 1);
 360         dn->ofs_in_node = ofs_in_node;
 361         return ret;
 362 }
 363
 364 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 365 {
 366         bool need_put = dn->inode_page ? false : true;
 367         int err;
 368
 369         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 370         if (err)
 371                 return err;
 372
 373         if (dn->data_blkaddr == NULL_ADDR)
 374                 err = reserve_new_block(dn);
 375         if (err || need_put)
 376                 f2fs_put_dnode(dn);
 377         return err;
 378 }
 379
 380 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 381 {
 382         struct extent_info ei;
 383         struct inode *inode = dn->inode;
 384
 385         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 386                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 387                 return 0;
 388         }
 389
 390         return f2fs_reserve_block(dn, index);
 391 }
 392
 393 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 394                                                 int rw, bool for_write)
 395 {
 396         struct address_space *mapping = inode->i_mapping;
 397         struct dnode_of_data dn;
 398         struct page *page;
 399         struct extent_info ei;
 400         int err;
 401         struct f2fs_io_info fio = {
 402                 .sbi = F2FS_I_SB(inode),
 403                 .type = DATA,
 404                 .rw = rw,
 405                 .encrypted_page = NULL,
 406         };
 407
 408         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 409                 return read_mapping_page(mapping, index, NULL);
 410
 411         page = f2fs_grab_cache_page(mapping, index, for_write);
 412         if (!page)
 413                 return ERR_PTR(-ENOMEM);
 414
 415         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 416                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 417                 goto got_it;
 418         }
 419
 420         set_new_dnode(&dn, inode, NULL, NULL, 0);
 421         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 422         if (err)
 423                 goto put_err;
 424         f2fs_put_dnode(&dn);
 425
 426         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 427                 err = -ENOENT;
 428                 goto put_err;
 429         }
 430 got_it:
 431         if (PageUptodate(page)) {
 432                 unlock_page(page);
 433                 return page;
 434         }
 435
 436         /*
 437          * A new dentry page is allocated but not able to be written, since its
 438          * new inode page couldn't be allocated due to -ENOSPC.
 439          * In such the case, its blkaddr can be remained as NEW_ADDR.
 440          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 441          */
 442         if (dn.data_blkaddr == NEW_ADDR) {
 443                 zero_user_segment(page, 0, PAGE_SIZE);
 444                 SetPageUptodate(page);
 445                 unlock_page(page);
 446                 return page;
 447         }
 448
 449         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 450         fio.page = page;
 451         err = f2fs_submit_page_bio(&fio);
 452         if (err)
 453                 goto put_err;
 454         return page;
 455
 456 put_err:
 457         f2fs_put_page(page, 1);
 458         return ERR_PTR(err);
 459 }
 460
 461 struct page *find_data_page(struct inode *inode, pgoff_t index)
 462 {
 463         struct address_space *mapping = inode->i_mapping;
 464         struct page *page;
 465
 466         page = find_get_page(mapping, index);
 467         if (page && PageUptodate(page))
 468                 return page;
 469         f2fs_put_page(page, 0);
 470
 471         page = get_read_data_page(inode, index, READ_SYNC, false);
 472         if (IS_ERR(page))
 473                 return page;
 474
 475         if (PageUptodate(page))
 476                 return page;
 477
 478         wait_on_page_locked(page);
 479         if (unlikely(!PageUptodate(page))) {
 480                 f2fs_put_page(page, 0);
 481                 return ERR_PTR(-EIO);
 482         }
 483         return page;
 484 }
 485
 486 /*
 487  * If it tries to access a hole, return an error.
 488  * Because, the callers, functions in dir.c and GC, should be able to know
 489  * whether this page exists or not.
 490  */
 491 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 492                                                         bool for_write)
 493 {
 494         struct address_space *mapping = inode->i_mapping;
 495         struct page *page;
 496 repeat:
 497         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 498         if (IS_ERR(page))
 499                 return page;
 500
 501         /* wait for read completion */
 502         lock_page(page);
 503         if (unlikely(!PageUptodate(page))) {
 504                 f2fs_put_page(page, 1);
 505                 return ERR_PTR(-EIO);
 506         }
 507         if (unlikely(page->mapping != mapping)) {
 508                 f2fs_put_page(page, 1);
 509                 goto repeat;
 510         }
 511         return page;
 512 }
 513
 514 /*
 515  * Caller ensures that this data page is never allocated.
 516  * A new zero-filled data page is allocated in the page cache.
 517  *
 518  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 519  * f2fs_unlock_op().
 520  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 521  * ipage should be released by this function.
 522  */
 523 struct page *get_new_data_page(struct inode *inode,
 524                 struct page *ipage, pgoff_t index, bool new_i_size)
 525 {
 526         struct address_space *mapping = inode->i_mapping;
 527         struct page *page;
 528         struct dnode_of_data dn;
 529         int err;
 530
 531         page = f2fs_grab_cache_page(mapping, index, true);
 532         if (!page) {
 533                 /*
 534                  * before exiting, we should make sure ipage will be released
 535                  * if any error occur.
 536                  */
 537                 f2fs_put_page(ipage, 1);
 538                 return ERR_PTR(-ENOMEM);
 539         }
 540
 541         set_new_dnode(&dn, inode, ipage, NULL, 0);
 542         err = f2fs_reserve_block(&dn, index);
 543         if (err) {
 544                 f2fs_put_page(page, 1);
 545                 return ERR_PTR(err);
 546         }
 547         if (!ipage)
 548                 f2fs_put_dnode(&dn);
 549
 550         if (PageUptodate(page))
 551                 goto got_it;
 552
 553         if (dn.data_blkaddr == NEW_ADDR) {
 554                 zero_user_segment(page, 0, PAGE_SIZE);
 555                 SetPageUptodate(page);
 556         } else {
 557                 f2fs_put_page(page, 1);
 558
 559                 /* if ipage exists, blkaddr should be NEW_ADDR */
 560                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 561                 page = get_lock_data_page(inode, index, true);
 562                 if (IS_ERR(page))
 563                         return page;
 564         }
 565 got_it:
 566         if (new_i_size && i_size_read(inode) <
 567                                 ((loff_t)(index + 1) << PAGE_SHIFT)) {
 568                 i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 569                 /* Only the directory inode sets new_i_size */
 570                 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
 571         }
 572         return page;
 573 }
 574
 575 static int __allocate_data_block(struct dnode_of_data *dn)
 576 {
 577         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 578         struct f2fs_summary sum;
 579         struct node_info ni;
 580         int seg = CURSEG_WARM_DATA;
 581         pgoff_t fofs;
 582         blkcnt_t count = 1;
 583
 584         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 585                 return -EPERM;
 586
 587         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 588         if (dn->data_blkaddr == NEW_ADDR)
 589                 goto alloc;
 590
 591         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 592                 return -ENOSPC;
 593
 594 alloc:
 595         get_node_info(sbi, dn->nid, &ni);
 596         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 597
 598         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 599                 seg = CURSEG_DIRECT_IO;
 600
 601         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 602                                                                 &sum, seg);
 603         set_data_blkaddr(dn);
 604
 605         /* update i_size */
 606         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 607                                                         dn->ofs_in_node;
 608         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 609                 i_size_write(dn->inode,
 610                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 611         return 0;
 612 }
 613
 614 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 615 {
 616         struct inode *inode = file_inode(iocb->ki_filp);
 617         struct f2fs_map_blocks map;
 618         ssize_t ret = 0;
 619
 620         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 621         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 622         map.m_next_pgofs = NULL;
 623
 624         if (f2fs_encrypted_inode(inode))
 625                 return 0;
 626
 627         if (iocb->ki_flags & IOCB_DIRECT) {
 628                 ret = f2fs_convert_inline_inode(inode);
 629                 if (ret)
 630                         return ret;
 631                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 632         }
 633         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 634                 ret = f2fs_convert_inline_inode(inode);
 635                 if (ret)
 636                         return ret;
 637         }
 638         if (!f2fs_has_inline_data(inode))
 639                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 640         return ret;
 641 }
 642
 643 /*
 644  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 645  * f2fs_map_blocks structure.
 646  * If original data blocks are allocated, then give them to blockdev.
 647  * Otherwise,
 648  *     a. preallocate requested block addresses
 649  *     b. do not use extent cache for better performance
 650  *     c. give the block addresses to blockdev
 651  */
 652 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 653                                                 int create, int flag)
 654 {
 655         unsigned int maxblocks = map->m_len;
 656         struct dnode_of_data dn;
 657         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 658         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 659         pgoff_t pgofs, end_offset, end;
 660         int err = 0, ofs = 1;
 661         unsigned int ofs_in_node, last_ofs_in_node;
 662         blkcnt_t prealloc;
 663         struct extent_info ei;
 664         bool allocated = false;
 665         block_t blkaddr;
 666
 667         map->m_len = 0;
 668         map->m_flags = 0;
 669
 670         /* it only supports block size == page size */
 671         pgofs = (pgoff_t)map->m_lblk;
 672         end = pgofs + maxblocks;
 673
 674         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 675                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 676                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 677                 map->m_flags = F2FS_MAP_MAPPED;
 678                 goto out;
 679         }
 680
 681 next_dnode:
 682         if (create)
 683                 f2fs_lock_op(sbi);
 684
 685         /* When reading holes, we need its node page */
 686         set_new_dnode(&dn, inode, NULL, NULL, 0);
 687         err = get_dnode_of_data(&dn, pgofs, mode);
 688         if (err) {
 689                 if (flag == F2FS_GET_BLOCK_BMAP)
 690                         map->m_pblk = 0;
 691                 if (err == -ENOENT) {
 692                         err = 0;
 693                         if (map->m_next_pgofs)
 694                                 *map->m_next_pgofs =
 695                                         get_next_page_offset(&dn, pgofs);
 696                 }
 697                 goto unlock_out;
 698         }
 699
 700         prealloc = 0;
 701         ofs_in_node = dn.ofs_in_node;
 702         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 703
 704 next_block:
 705         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 706
 707         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 708                 if (create) {
 709                         if (unlikely(f2fs_cp_error(sbi))) {
 710                                 err = -EIO;
 711                                 goto sync_out;
 712                         }
 713                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 714                                 if (blkaddr == NULL_ADDR) {
 715                                         prealloc++;
 716                                         last_ofs_in_node = dn.ofs_in_node;
 717                                 }
 718                         } else {
 719                                 err = __allocate_data_block(&dn);
 720                                 if (!err) {
 721                                         set_inode_flag(F2FS_I(inode),
 722                                                         FI_APPEND_WRITE);
 723                                         allocated = true;
 724                                 }
 725                         }
 726                         if (err)
 727                                 goto sync_out;
 728                         map->m_flags = F2FS_MAP_NEW;
 729                         blkaddr = dn.data_blkaddr;
 730                 } else {
 731                         if (flag == F2FS_GET_BLOCK_BMAP) {
 732                                 map->m_pblk = 0;
 733                                 goto sync_out;
 734                         }
 735                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 736                                                 blkaddr == NULL_ADDR) {
 737                                 if (map->m_next_pgofs)
 738                                         *map->m_next_pgofs = pgofs + 1;
 739                         }
 740                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 741                                                 blkaddr != NEW_ADDR)
 742                                 goto sync_out;
 743                 }
 744         }
 745
 746         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 747                 goto skip;
 748
 749         if (map->m_len == 0) {
 750                 /* preallocated unwritten block should be mapped for fiemap. */
 751                 if (blkaddr == NEW_ADDR)
 752                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 753                 map->m_flags |= F2FS_MAP_MAPPED;
 754
 755                 map->m_pblk = blkaddr;
 756                 map->m_len = 1;
 757         } else if ((map->m_pblk != NEW_ADDR &&
 758                         blkaddr == (map->m_pblk + ofs)) ||
 759                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 760                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 761                 ofs++;
 762                 map->m_len++;
 763         } else {
 764                 goto sync_out;
 765         }
 766
 767 skip:
 768         dn.ofs_in_node++;
 769         pgofs++;
 770
 771         /* preallocate blocks in batch for one dnode page */
 772         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 773                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 774
 775                 dn.ofs_in_node = ofs_in_node;
 776                 err = reserve_new_blocks(&dn, prealloc);
 777                 if (err)
 778                         goto sync_out;
 779
 780                 map->m_len += dn.ofs_in_node - ofs_in_node;
 781                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 782                         err = -ENOSPC;
 783                         goto sync_out;
 784                 }
 785                 dn.ofs_in_node = end_offset;
 786         }
 787
 788         if (pgofs >= end)
 789                 goto sync_out;
 790         else if (dn.ofs_in_node < end_offset)
 791                 goto next_block;
 792
 793         if (allocated)
 794                 sync_inode_page(&dn);
 795         f2fs_put_dnode(&dn);
 796
 797         if (create) {
 798                 f2fs_unlock_op(sbi);
 799                 f2fs_balance_fs(sbi, allocated);
 800         }
 801         allocated = false;
 802         goto next_dnode;
 803
 804 sync_out:
 805         if (allocated)
 806                 sync_inode_page(&dn);
 807         f2fs_put_dnode(&dn);
 808 unlock_out:
 809         if (create) {
 810                 f2fs_unlock_op(sbi);
 811                 f2fs_balance_fs(sbi, allocated);
 812         }
 813 out:
 814         trace_f2fs_map_blocks(inode, map, err);
 815         return err;
 816 }
 817
 818 static int __get_data_block(struct inode *inode, sector_t iblock,
 819                         struct buffer_head *bh, int create, int flag,
 820                         pgoff_t *next_pgofs)
 821 {
 822         struct f2fs_map_blocks map;
 823         int ret;
 824
 825         map.m_lblk = iblock;
 826         map.m_len = bh->b_size >> inode->i_blkbits;
 827         map.m_next_pgofs = next_pgofs;
 828
 829         ret = f2fs_map_blocks(inode, &map, create, flag);
 830         if (!ret) {
 831                 map_bh(bh, inode->i_sb, map.m_pblk);
 832                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 833                 bh->b_size = map.m_len << inode->i_blkbits;
 834         }
 835         return ret;
 836 }
 837
 838 static int get_data_block(struct inode *inode, sector_t iblock,
 839                         struct buffer_head *bh_result, int create, int flag,
 840                         pgoff_t *next_pgofs)
 841 {
 842         return __get_data_block(inode, iblock, bh_result, create,
 843                                                         flag, next_pgofs);
 844 }
 845
 846 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 847                         struct buffer_head *bh_result, int create)
 848 {
 849         return __get_data_block(inode, iblock, bh_result, create,
 850                                                 F2FS_GET_BLOCK_DIO, NULL);
 851 }
 852
 853 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 854                         struct buffer_head *bh_result, int create)
 855 {
 856         /* Block number less than F2FS MAX BLOCKS */
 857         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 858                 return -EFBIG;
 859
 860         return __get_data_block(inode, iblock, bh_result, create,
 861                                                 F2FS_GET_BLOCK_BMAP, NULL);
 862 }
 863
 864 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 865 {
 866         return (offset >> inode->i_blkbits);
 867 }
 868
 869 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 870 {
 871         return (blk << inode->i_blkbits);
 872 }
 873
 874 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 875                 u64 start, u64 len)
 876 {
 877         struct buffer_head map_bh;
 878         sector_t start_blk, last_blk;
 879         pgoff_t next_pgofs;
 880         loff_t isize;
 881         u64 logical = 0, phys = 0, size = 0;
 882         u32 flags = 0;
 883         int ret = 0;
 884
 885         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 886         if (ret)
 887                 return ret;
 888
 889         if (f2fs_has_inline_data(inode)) {
 890                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 891                 if (ret != -EAGAIN)
 892                         return ret;
 893         }
 894
 895         inode_lock(inode);
 896
 897         isize = i_size_read(inode);
 898         if (start >= isize)
 899                 goto out;
 900
 901         if (start + len > isize)
 902                 len = isize - start;
 903
 904         if (logical_to_blk(inode, len) == 0)
 905                 len = blk_to_logical(inode, 1);
 906
 907         start_blk = logical_to_blk(inode, start);
 908         last_blk = logical_to_blk(inode, start + len - 1);
 909
 910 next:
 911         memset(&map_bh, 0, sizeof(struct buffer_head));
 912         map_bh.b_size = len;
 913
 914         ret = get_data_block(inode, start_blk, &map_bh, 0,
 915                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 916         if (ret)
 917                 goto out;
 918
 919         /* HOLE */
 920         if (!buffer_mapped(&map_bh)) {
 921                 start_blk = next_pgofs;
 922                 /* Go through holes util pass the EOF */
 923                 if (blk_to_logical(inode, start_blk) < isize)
 924                         goto prep_next;
 925                 /* Found a hole beyond isize means no more extents.
 926                  * Note that the premise is that filesystems don't
 927                  * punch holes beyond isize and keep size unchanged.
 928                  */
 929                 flags |= FIEMAP_EXTENT_LAST;
 930         }
 931
 932         if (size) {
 933                 if (f2fs_encrypted_inode(inode))
 934                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 935
 936                 ret = fiemap_fill_next_extent(fieinfo, logical,
 937                                 phys, size, flags);
 938         }
 939
 940         if (start_blk > last_blk || ret)
 941                 goto out;
 942
 943         logical = blk_to_logical(inode, start_blk);
 944         phys = blk_to_logical(inode, map_bh.b_blocknr);
 945         size = map_bh.b_size;
 946         flags = 0;
 947         if (buffer_unwritten(&map_bh))
 948                 flags = FIEMAP_EXTENT_UNWRITTEN;
 949
 950         start_blk += logical_to_blk(inode, size);
 951
 952 prep_next:
 953         cond_resched();
 954         if (fatal_signal_pending(current))
 955                 ret = -EINTR;
 956         else
 957                 goto next;
 958 out:
 959         if (ret == 1)
 960                 ret = 0;
 961
 962         inode_unlock(inode);
 963         return ret;
 964 }
 965
 966 /*
 967  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 968  * Major change was from block_size == page_size in f2fs by default.
 969  */
 970 static int f2fs_mpage_readpages(struct address_space *mapping,
 971                         struct list_head *pages, struct page *page,
 972                         unsigned nr_pages)
 973 {
 974         struct bio *bio = NULL;
 975         unsigned page_idx;
 976         sector_t last_block_in_bio = 0;
 977         struct inode *inode = mapping->host;
 978         const unsigned blkbits = inode->i_blkbits;
 979         const unsigned blocksize = 1 << blkbits;
 980         sector_t block_in_file;
 981         sector_t last_block;
 982         sector_t last_block_in_file;
 983         sector_t block_nr;
 984         struct block_device *bdev = inode->i_sb->s_bdev;
 985         struct f2fs_map_blocks map;
 986
 987         map.m_pblk = 0;
 988         map.m_lblk = 0;
 989         map.m_len = 0;
 990         map.m_flags = 0;
 991         map.m_next_pgofs = NULL;
 992
 993         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 994
 995                 prefetchw(&page->flags);
 996                 if (pages) {
 997                         page = list_entry(pages->prev, struct page, lru);
 998                         list_del(&page->lru);
 999                         if (add_to_page_cache_lru(page, mapping,
1000                                                   page->index, GFP_KERNEL))
1001                                 goto next_page;
1002                 }
1003
1004                 block_in_file = (sector_t)page->index;
1005                 last_block = block_in_file + nr_pages;
1006                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1007                                                                 blkbits;
1008                 if (last_block > last_block_in_file)
1009                         last_block = last_block_in_file;
1010
1011                 /*
1012                  * Map blocks using the previous result first.
1013                  */
1014                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1015                                 block_in_file > map.m_lblk &&
1016                                 block_in_file < (map.m_lblk + map.m_len))
1017                         goto got_it;
1018
1019                 /*
1020                  * Then do more f2fs_map_blocks() calls until we are
1021                  * done with this page.
1022                  */
1023                 map.m_flags = 0;
1024
1025                 if (block_in_file < last_block) {
1026                         map.m_lblk = block_in_file;
1027                         map.m_len = last_block - block_in_file;
1028
1029                         if (f2fs_map_blocks(inode, &map, 0,
1030                                                 F2FS_GET_BLOCK_READ))
1031                                 goto set_error_page;
1032                 }
1033 got_it:
1034                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1035                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1036                         SetPageMappedToDisk(page);
1037
1038                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1039                                 SetPageUptodate(page);
1040                                 goto confused;
1041                         }
1042                 } else {
1043                         zero_user_segment(page, 0, PAGE_SIZE);
1044                         SetPageUptodate(page);
1045                         unlock_page(page);
1046                         goto next_page;
1047                 }
1048
1049                 /*
1050                  * This page will go to BIO.  Do we need to send this
1051                  * BIO off first?
1052                  */
1053                 if (bio && (last_block_in_bio != block_nr - 1)) {
1054 submit_and_realloc:
1055                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1056                         bio = NULL;
1057                 }
1058                 if (bio == NULL) {
1059                         struct fscrypt_ctx *ctx = NULL;
1060
1061                         if (f2fs_encrypted_inode(inode) &&
1062                                         S_ISREG(inode->i_mode)) {
1063
1064                                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1065                                 if (IS_ERR(ctx))
1066                                         goto set_error_page;
1067
1068                                 /* wait the page to be moved by cleaning */
1069                                 f2fs_wait_on_encrypted_page_writeback(
1070                                                 F2FS_I_SB(inode), block_nr);
1071                         }
1072
1073                         bio = bio_alloc(GFP_KERNEL,
1074                                 min_t(int, nr_pages, BIO_MAX_PAGES));
1075                         if (!bio) {
1076                                 if (ctx)
1077                                         fscrypt_release_ctx(ctx);
1078                                 goto set_error_page;
1079                         }
1080                         bio->bi_bdev = bdev;
1081                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1082                         bio->bi_end_io = f2fs_read_end_io;
1083                         bio->bi_private = ctx;
1084                         bio->bi_rw = READ;
1085                 }
1086
1087                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1088                         goto submit_and_realloc;
1089
1090                 last_block_in_bio = block_nr;
1091                 goto next_page;
1092 set_error_page:
1093                 SetPageError(page);
1094                 zero_user_segment(page, 0, PAGE_SIZE);
1095                 unlock_page(page);
1096                 goto next_page;
1097 confused:
1098                 if (bio) {
1099                         __submit_bio(F2FS_I_SB(inode), READ, bio);
1100                         bio = NULL;
1101                 }
1102                 unlock_page(page);
1103 next_page:
1104                 if (pages)
1105                         put_page(page);
1106         }
1107         BUG_ON(pages && !list_empty(pages));
1108         if (bio)
1109                 __submit_bio(F2FS_I_SB(inode), READ, bio);
1110         return 0;
1111 }
1112
1113 static int f2fs_read_data_page(struct file *file, struct page *page)
1114 {
1115         struct inode *inode = page->mapping->host;
1116         int ret = -EAGAIN;
1117
1118         trace_f2fs_readpage(page, DATA);
1119
1120         /* If the file has inline data, try to read it directly */
1121         if (f2fs_has_inline_data(inode))
1122                 ret = f2fs_read_inline_data(inode, page);
1123         if (ret == -EAGAIN)
1124                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1125         return ret;
1126 }
1127
1128 static int f2fs_read_data_pages(struct file *file,
1129                         struct address_space *mapping,
1130                         struct list_head *pages, unsigned nr_pages)
1131 {
1132         struct inode *inode = file->f_mapping->host;
1133         struct page *page = list_entry(pages->prev, struct page, lru);
1134
1135         trace_f2fs_readpages(inode, page, nr_pages);
1136
1137         /* If the file has inline data, skip readpages */
1138         if (f2fs_has_inline_data(inode))
1139                 return 0;
1140
1141         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1142 }
1143
1144 int do_write_data_page(struct f2fs_io_info *fio)
1145 {
1146         struct page *page = fio->page;
1147         struct inode *inode = page->mapping->host;
1148         struct dnode_of_data dn;
1149         int err = 0;
1150
1151         set_new_dnode(&dn, inode, NULL, NULL, 0);
1152         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1153         if (err)
1154                 return err;
1155
1156         fio->old_blkaddr = dn.data_blkaddr;
1157
1158         /* This page is already truncated */
1159         if (fio->old_blkaddr == NULL_ADDR) {
1160                 ClearPageUptodate(page);
1161                 goto out_writepage;
1162         }
1163
1164         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1165                 gfp_t gfp_flags = GFP_NOFS;
1166
1167                 /* wait for GCed encrypted page writeback */
1168                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1169                                                         fio->old_blkaddr);
1170 retry_encrypt:
1171                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1172                                                                 gfp_flags);
1173                 if (IS_ERR(fio->encrypted_page)) {
1174                         err = PTR_ERR(fio->encrypted_page);
1175                         if (err == -ENOMEM) {
1176                                 /* flush pending ios and wait for a while */
1177                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1178                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1179                                 gfp_flags |= __GFP_NOFAIL;
1180                                 err = 0;
1181                                 goto retry_encrypt;
1182                         }
1183                         goto out_writepage;
1184                 }
1185         }
1186
1187         set_page_writeback(page);
1188
1189         /*
1190          * If current allocation needs SSR,
1191          * it had better in-place writes for updated data.
1192          */
1193         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1194                         !is_cold_data(page) &&
1195                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1196                         need_inplace_update(inode))) {
1197                 rewrite_data_page(fio);
1198                 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1199                 trace_f2fs_do_write_data_page(page, IPU);
1200         } else {
1201                 write_data_page(&dn, fio);
1202                 trace_f2fs_do_write_data_page(page, OPU);
1203                 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1204                 if (page->index == 0)
1205                         set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1206         }
1207 out_writepage:
1208         f2fs_put_dnode(&dn);
1209         return err;
1210 }
1211
1212 static int f2fs_write_data_page(struct page *page,
1213                                         struct writeback_control *wbc)
1214 {
1215         struct inode *inode = page->mapping->host;
1216         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1217         loff_t i_size = i_size_read(inode);
1218         const pgoff_t end_index = ((unsigned long long) i_size)
1219                                                         >> PAGE_SHIFT;
1220         unsigned offset = 0;
1221         bool need_balance_fs = false;
1222         int err = 0;
1223         struct f2fs_io_info fio = {
1224                 .sbi = sbi,
1225                 .type = DATA,
1226                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1227                 .page = page,
1228                 .encrypted_page = NULL,
1229         };
1230
1231         trace_f2fs_writepage(page, DATA);
1232
1233         if (page->index < end_index)
1234                 goto write;
1235
1236         /*
1237          * If the offset is out-of-range of file size,
1238          * this page does not have to be written to disk.
1239          */
1240         offset = i_size & (PAGE_SIZE - 1);
1241         if ((page->index >= end_index + 1) || !offset)
1242                 goto out;
1243
1244         zero_user_segment(page, offset, PAGE_SIZE);
1245 write:
1246         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1247                 goto redirty_out;
1248         if (f2fs_is_drop_cache(inode))
1249                 goto out;
1250         /* we should not write 0'th page having journal header */
1251         if (f2fs_is_volatile_file(inode) && (!page->index ||
1252                         (!wbc->for_reclaim &&
1253                         available_free_memory(sbi, BASE_CHECK))))
1254                 goto redirty_out;
1255
1256         /* Dentry blocks are controlled by checkpoint */
1257         if (S_ISDIR(inode->i_mode)) {
1258                 if (unlikely(f2fs_cp_error(sbi)))
1259                         goto redirty_out;
1260                 err = do_write_data_page(&fio);
1261                 goto done;
1262         }
1263
1264         /* we should bypass data pages to proceed the kworkder jobs */
1265         if (unlikely(f2fs_cp_error(sbi))) {
1266                 SetPageError(page);
1267                 goto out;
1268         }
1269
1270         if (!wbc->for_reclaim)
1271                 need_balance_fs = true;
1272         else if (has_not_enough_free_secs(sbi, 0))
1273                 goto redirty_out;
1274
1275         err = -EAGAIN;
1276         f2fs_lock_op(sbi);
1277         if (f2fs_has_inline_data(inode))
1278                 err = f2fs_write_inline_data(inode, page);
1279         if (err == -EAGAIN)
1280                 err = do_write_data_page(&fio);
1281         f2fs_unlock_op(sbi);
1282 done:
1283         if (err && err != -ENOENT)
1284                 goto redirty_out;
1285
1286         clear_cold_data(page);
1287 out:
1288         inode_dec_dirty_pages(inode);
1289         if (err)
1290                 ClearPageUptodate(page);
1291
1292         if (wbc->for_reclaim) {
1293                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1294                 remove_dirty_inode(inode);
1295         }
1296
1297         unlock_page(page);
1298         f2fs_balance_fs(sbi, need_balance_fs);
1299
1300         if (unlikely(f2fs_cp_error(sbi)))
1301                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1302
1303         return 0;
1304
1305 redirty_out:
1306         redirty_page_for_writepage(wbc, page);
1307         return AOP_WRITEPAGE_ACTIVATE;
1308 }
1309
1310 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1311                         void *data)
1312 {
1313         struct address_space *mapping = data;
1314         int ret = mapping->a_ops->writepage(page, wbc);
1315         mapping_set_error(mapping, ret);
1316         return ret;
1317 }
1318
1319 /*
1320  * This function was copied from write_cche_pages from mm/page-writeback.c.
1321  * The major change is making write step of cold data page separately from
1322  * warm/hot data page.
1323  */
1324 static int f2fs_write_cache_pages(struct address_space *mapping,
1325                         struct writeback_control *wbc, writepage_t writepage,
1326                         void *data)
1327 {
1328         int ret = 0;
1329         int done = 0;
1330         struct pagevec pvec;
1331         int nr_pages;
1332         pgoff_t uninitialized_var(writeback_index);
1333         pgoff_t index;
1334         pgoff_t end;            /* Inclusive */
1335         pgoff_t done_index;
1336         int cycled;
1337         int range_whole = 0;
1338         int tag;
1339         int step = 0;
1340
1341         pagevec_init(&pvec, 0);
1342 next:
1343         if (wbc->range_cyclic) {
1344                 writeback_index = mapping->writeback_index; /* prev offset */
1345                 index = writeback_index;
1346                 if (index == 0)
1347                         cycled = 1;
1348                 else
1349                         cycled = 0;
1350                 end = -1;
1351         } else {
1352                 index = wbc->range_start >> PAGE_SHIFT;
1353                 end = wbc->range_end >> PAGE_SHIFT;
1354                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1355                         range_whole = 1;
1356                 cycled = 1; /* ignore range_cyclic tests */
1357         }
1358         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1359                 tag = PAGECACHE_TAG_TOWRITE;
1360         else
1361                 tag = PAGECACHE_TAG_DIRTY;
1362 retry:
1363         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1364                 tag_pages_for_writeback(mapping, index, end);
1365         done_index = index;
1366         while (!done && (index <= end)) {
1367                 int i;
1368
1369                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1370                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1371                 if (nr_pages == 0)
1372                         break;
1373
1374                 for (i = 0; i < nr_pages; i++) {
1375                         struct page *page = pvec.pages[i];
1376
1377                         if (page->index > end) {
1378                                 done = 1;
1379                                 break;
1380                         }
1381
1382                         done_index = page->index;
1383
1384                         lock_page(page);
1385
1386                         if (unlikely(page->mapping != mapping)) {
1387 continue_unlock:
1388                                 unlock_page(page);
1389                                 continue;
1390                         }
1391
1392                         if (!PageDirty(page)) {
1393                                 /* someone wrote it for us */
1394                                 goto continue_unlock;
1395                         }
1396
1397                         if (step == is_cold_data(page))
1398                                 goto continue_unlock;
1399
1400                         if (PageWriteback(page)) {
1401                                 if (wbc->sync_mode != WB_SYNC_NONE)
1402                                         f2fs_wait_on_page_writeback(page,
1403                                                                 DATA, true);
1404                                 else
1405                                         goto continue_unlock;
1406                         }
1407
1408                         BUG_ON(PageWriteback(page));
1409                         if (!clear_page_dirty_for_io(page))
1410                                 goto continue_unlock;
1411
1412                         ret = (*writepage)(page, wbc, data);
1413                         if (unlikely(ret)) {
1414                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1415                                         unlock_page(page);
1416                                         ret = 0;
1417                                 } else {
1418                                         done_index = page->index + 1;
1419                                         done = 1;
1420                                         break;
1421                                 }
1422                         }
1423
1424                         if (--wbc->nr_to_write <= 0 &&
1425                             wbc->sync_mode == WB_SYNC_NONE) {
1426                                 done = 1;
1427                                 break;
1428                         }
1429                 }
1430                 pagevec_release(&pvec);
1431                 cond_resched();
1432         }
1433
1434         if (step < 1) {
1435                 step++;
1436                 goto next;
1437         }
1438
1439         if (!cycled && !done) {
1440                 cycled = 1;
1441                 index = 0;
1442                 end = writeback_index - 1;
1443                 goto retry;
1444         }
1445         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1446                 mapping->writeback_index = done_index;
1447
1448         return ret;
1449 }
1450
1451 static int f2fs_write_data_pages(struct address_space *mapping,
1452                             struct writeback_control *wbc)
1453 {
1454         struct inode *inode = mapping->host;
1455         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1456         bool locked = false;
1457         int ret;
1458         long diff;
1459
1460         /* deal with chardevs and other special file */
1461         if (!mapping->a_ops->writepage)
1462                 return 0;
1463
1464         /* skip writing if there is no dirty page in this inode */
1465         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1466                 return 0;
1467
1468         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1469                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1470                         available_free_memory(sbi, DIRTY_DENTS))
1471                 goto skip_write;
1472
1473         /* skip writing during file defragment */
1474         if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1475                 goto skip_write;
1476
1477         /* during POR, we don't need to trigger writepage at all. */
1478         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1479                 goto skip_write;
1480
1481         trace_f2fs_writepages(mapping->host, wbc, DATA);
1482
1483         diff = nr_pages_to_write(sbi, DATA, wbc);
1484
1485         if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1486                 mutex_lock(&sbi->writepages);
1487                 locked = true;
1488         }
1489         ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1490         f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1491         if (locked)
1492                 mutex_unlock(&sbi->writepages);
1493
1494         remove_dirty_inode(inode);
1495
1496         wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1497         return ret;
1498
1499 skip_write:
1500         wbc->pages_skipped += get_dirty_pages(inode);
1501         trace_f2fs_writepages(mapping->host, wbc, DATA);
1502         return 0;
1503 }
1504
1505 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1506 {
1507         struct inode *inode = mapping->host;
1508         loff_t i_size = i_size_read(inode);
1509
1510         if (to > i_size) {
1511                 truncate_pagecache(inode, i_size);
1512                 truncate_blocks(inode, i_size, true);
1513         }
1514 }
1515
1516 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1517                         struct page *page, loff_t pos, unsigned len,
1518                         block_t *blk_addr, bool *node_changed)
1519 {
1520         struct inode *inode = page->mapping->host;
1521         pgoff_t index = page->index;
1522         struct dnode_of_data dn;
1523         struct page *ipage;
1524         bool locked = false;
1525         struct extent_info ei;
1526         int err = 0;
1527
1528         /*
1529          * we already allocated all the blocks, so we don't need to get
1530          * the block addresses when there is no need to fill the page.
1531          */
1532         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1533                                         len == PAGE_SIZE)
1534                 return 0;
1535
1536         if (f2fs_has_inline_data(inode) ||
1537                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1538                 f2fs_lock_op(sbi);
1539                 locked = true;
1540         }
1541 restart:
1542         /* check inline_data */
1543         ipage = get_node_page(sbi, inode->i_ino);
1544         if (IS_ERR(ipage)) {
1545                 err = PTR_ERR(ipage);
1546                 goto unlock_out;
1547         }
1548
1549         set_new_dnode(&dn, inode, ipage, ipage, 0);
1550
1551         if (f2fs_has_inline_data(inode)) {
1552                 if (pos + len <= MAX_INLINE_DATA) {
1553                         read_inline_data(page, ipage);
1554                         set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1555                         if (inode->i_nlink)
1556                                 set_inline_node(ipage);
1557                 } else {
1558                         err = f2fs_convert_inline_page(&dn, page);
1559                         if (err)
1560                                 goto out;
1561                         if (dn.data_blkaddr == NULL_ADDR)
1562                                 err = f2fs_get_block(&dn, index);
1563                 }
1564         } else if (locked) {
1565                 err = f2fs_get_block(&dn, index);
1566         } else {
1567                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1568                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1569                 } else {
1570                         /* hole case */
1571                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1572                         if (err || dn.data_blkaddr == NULL_ADDR) {
1573                                 f2fs_put_dnode(&dn);
1574                                 f2fs_lock_op(sbi);
1575                                 locked = true;
1576                                 goto restart;
1577                         }
1578                 }
1579         }
1580
1581         /* convert_inline_page can make node_changed */
1582         *blk_addr = dn.data_blkaddr;
1583         *node_changed = dn.node_changed;
1584 out:
1585         f2fs_put_dnode(&dn);
1586 unlock_out:
1587         if (locked)
1588                 f2fs_unlock_op(sbi);
1589         return err;
1590 }
1591
1592 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1593                 loff_t pos, unsigned len, unsigned flags,
1594                 struct page **pagep, void **fsdata)
1595 {
1596         struct inode *inode = mapping->host;
1597         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1598         struct page *page = NULL;
1599         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1600         bool need_balance = false;
1601         block_t blkaddr = NULL_ADDR;
1602         int err = 0;
1603
1604         trace_f2fs_write_begin(inode, pos, len, flags);
1605
1606         /*
1607          * We should check this at this moment to avoid deadlock on inode page
1608          * and #0 page. The locking rule for inline_data conversion should be:
1609          * lock_page(page #0) -> lock_page(inode_page)
1610          */
1611         if (index != 0) {
1612                 err = f2fs_convert_inline_inode(inode);
1613                 if (err)
1614                         goto fail;
1615         }
1616 repeat:
1617         page = grab_cache_page_write_begin(mapping, index, flags);
1618         if (!page) {
1619                 err = -ENOMEM;
1620                 goto fail;
1621         }
1622
1623         *pagep = page;
1624
1625         err = prepare_write_begin(sbi, page, pos, len,
1626                                         &blkaddr, &need_balance);
1627         if (err)
1628                 goto fail;
1629
1630         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1631                 unlock_page(page);
1632                 f2fs_balance_fs(sbi, true);
1633                 lock_page(page);
1634                 if (page->mapping != mapping) {
1635                         /* The page got truncated from under us */
1636                         f2fs_put_page(page, 1);
1637                         goto repeat;
1638                 }
1639         }
1640
1641         f2fs_wait_on_page_writeback(page, DATA, false);
1642
1643         /* wait for GCed encrypted page writeback */
1644         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1645                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1646
1647         if (len == PAGE_SIZE)
1648                 goto out_update;
1649         if (PageUptodate(page))
1650                 goto out_clear;
1651
1652         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1653                 unsigned start = pos & (PAGE_SIZE - 1);
1654                 unsigned end = start + len;
1655
1656                 /* Reading beyond i_size is simple: memset to zero */
1657                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1658                 goto out_update;
1659         }
1660
1661         if (blkaddr == NEW_ADDR) {
1662                 zero_user_segment(page, 0, PAGE_SIZE);
1663         } else {
1664                 struct f2fs_io_info fio = {
1665                         .sbi = sbi,
1666                         .type = DATA,
1667                         .rw = READ_SYNC,
1668                         .old_blkaddr = blkaddr,
1669                         .new_blkaddr = blkaddr,
1670                         .page = page,
1671                         .encrypted_page = NULL,
1672                 };
1673                 err = f2fs_submit_page_bio(&fio);
1674                 if (err)
1675                         goto fail;
1676
1677                 lock_page(page);
1678                 if (unlikely(!PageUptodate(page))) {
1679                         err = -EIO;
1680                         goto fail;
1681                 }
1682                 if (unlikely(page->mapping != mapping)) {
1683                         f2fs_put_page(page, 1);
1684                         goto repeat;
1685                 }
1686
1687                 /* avoid symlink page */
1688                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1689                         err = fscrypt_decrypt_page(page);
1690                         if (err)
1691                                 goto fail;
1692                 }
1693         }
1694 out_update:
1695         SetPageUptodate(page);
1696 out_clear:
1697         clear_cold_data(page);
1698         return 0;
1699
1700 fail:
1701         f2fs_put_page(page, 1);
1702         f2fs_write_failed(mapping, pos + len);
1703         return err;
1704 }
1705
1706 static int f2fs_write_end(struct file *file,
1707                         struct address_space *mapping,
1708                         loff_t pos, unsigned len, unsigned copied,
1709                         struct page *page, void *fsdata)
1710 {
1711         struct inode *inode = page->mapping->host;
1712
1713         trace_f2fs_write_end(inode, pos, len, copied);
1714
1715         set_page_dirty(page);
1716
1717         if (pos + copied > i_size_read(inode)) {
1718                 i_size_write(inode, pos + copied);
1719                 mark_inode_dirty(inode);
1720         }
1721
1722         f2fs_put_page(page, 1);
1723         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1724         return copied;
1725 }
1726
1727 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1728                            loff_t offset)
1729 {
1730         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1731
1732         if (offset & blocksize_mask)
1733                 return -EINVAL;
1734
1735         if (iov_iter_alignment(iter) & blocksize_mask)
1736                 return -EINVAL;
1737
1738         return 0;
1739 }
1740
1741 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1742 {
1743         struct address_space *mapping = iocb->ki_filp->f_mapping;
1744         struct inode *inode = mapping->host;
1745         size_t count = iov_iter_count(iter);
1746         loff_t offset = iocb->ki_pos;
1747         int err;
1748
1749         err = check_direct_IO(inode, iter, offset);
1750         if (err)
1751                 return err;
1752
1753         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1754                 return 0;
1755
1756         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1757
1758         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1759         if (iov_iter_rw(iter) == WRITE) {
1760                 if (err > 0)
1761                         set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1762                 else if (err < 0)
1763                         f2fs_write_failed(mapping, offset + count);
1764         }
1765
1766         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1767
1768         return err;
1769 }
1770
1771 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1772                                                         unsigned int length)
1773 {
1774         struct inode *inode = page->mapping->host;
1775         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1776
1777         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1778                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1779                 return;
1780
1781         if (PageDirty(page)) {
1782                 if (inode->i_ino == F2FS_META_INO(sbi))
1783                         dec_page_count(sbi, F2FS_DIRTY_META);
1784                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1785                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1786                 else
1787                         inode_dec_dirty_pages(inode);
1788         }
1789
1790         /* This is atomic written page, keep Private */
1791         if (IS_ATOMIC_WRITTEN_PAGE(page))
1792                 return;
1793
1794         set_page_private(page, 0);
1795         ClearPagePrivate(page);
1796 }
1797
1798 int f2fs_release_page(struct page *page, gfp_t wait)
1799 {
1800         /* If this is dirty page, keep PagePrivate */
1801         if (PageDirty(page))
1802                 return 0;
1803
1804         /* This is atomic written page, keep Private */
1805         if (IS_ATOMIC_WRITTEN_PAGE(page))
1806                 return 0;
1807
1808         set_page_private(page, 0);
1809         ClearPagePrivate(page);
1810         return 1;
1811 }
1812
1813 static int f2fs_set_data_page_dirty(struct page *page)
1814 {
1815         struct address_space *mapping = page->mapping;
1816         struct inode *inode = mapping->host;
1817
1818         trace_f2fs_set_page_dirty(page, DATA);
1819
1820         SetPageUptodate(page);
1821
1822         if (f2fs_is_atomic_file(inode)) {
1823                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1824                         register_inmem_page(inode, page);
1825                         return 1;
1826                 }
1827                 /*
1828                  * Previously, this page has been registered, we just
1829                  * return here.
1830                  */
1831                 return 0;
1832         }
1833
1834         if (!PageDirty(page)) {
1835                 __set_page_dirty_nobuffers(page);
1836                 update_dirty_page(inode, page);
1837                 return 1;
1838         }
1839         return 0;
1840 }
1841
1842 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1843 {
1844         struct inode *inode = mapping->host;
1845
1846         if (f2fs_has_inline_data(inode))
1847                 return 0;
1848
1849         /* make sure allocating whole blocks */
1850         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1851                 filemap_write_and_wait(mapping);
1852
1853         return generic_block_bmap(mapping, block, get_data_block_bmap);
1854 }
1855
1856 const struct address_space_operations f2fs_dblock_aops = {
1857         .readpage       = f2fs_read_data_page,
1858         .readpages      = f2fs_read_data_pages,
1859         .writepage      = f2fs_write_data_page,
1860         .writepages     = f2fs_write_data_pages,
1861         .write_begin    = f2fs_write_begin,
1862         .write_end      = f2fs_write_end,
1863         .set_page_dirty = f2fs_set_data_page_dirty,
1864         .invalidatepage = f2fs_invalidate_page,
1865         .releasepage    = f2fs_release_page,
1866         .direct_IO      = f2fs_direct_IO,
1867         .bmap           = f2fs_bmap,
1868 };