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, enum page_type type)
 102 {
 103         if (!is_read_io(rw)) {
 104                 atomic_inc(&sbi->nr_wb_bios);
 105                 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
 106                         current->plug && (type == DATA || type == NODE))
 107                         blk_finish_plug(current->plug);
 108         }
 109         submit_bio(rw, bio);
 110 }
 111
 112 static void __submit_merged_bio(struct f2fs_bio_info *io)
 113 {
 114         struct f2fs_io_info *fio = &io->fio;
 115
 116         if (!io->bio)
 117                 return;
 118
 119         if (is_read_io(fio->rw))
 120                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 121         else
 122                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 123
 124         __submit_bio(io->sbi, fio->rw, io->bio, fio->type);
 125         io->bio = NULL;
 126 }
 127
 128 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 129                                                 struct page *page, nid_t ino)
 130 {
 131         struct bio_vec *bvec;
 132         struct page *target;
 133         int i;
 134
 135         if (!io->bio)
 136                 return false;
 137
 138         if (!inode && !page && !ino)
 139                 return true;
 140
 141         bio_for_each_segment_all(bvec, io->bio, i) {
 142
 143                 if (bvec->bv_page->mapping)
 144                         target = bvec->bv_page;
 145                 else
 146                         target = fscrypt_control_page(bvec->bv_page);
 147
 148                 if (inode && inode == target->mapping->host)
 149                         return true;
 150                 if (page && page == target)
 151                         return true;
 152                 if (ino && ino == ino_of_node(target))
 153                         return true;
 154         }
 155
 156         return false;
 157 }
 158
 159 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 160                                                 struct page *page, nid_t ino,
 161                                                 enum page_type type)
 162 {
 163         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 164         struct f2fs_bio_info *io = &sbi->write_io[btype];
 165         bool ret;
 166
 167         down_read(&io->io_rwsem);
 168         ret = __has_merged_page(io, inode, page, ino);
 169         up_read(&io->io_rwsem);
 170         return ret;
 171 }
 172
 173 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 174                                 struct inode *inode, struct page *page,
 175                                 nid_t ino, enum page_type type, int rw)
 176 {
 177         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 178         struct f2fs_bio_info *io;
 179
 180         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 181
 182         down_write(&io->io_rwsem);
 183
 184         if (!__has_merged_page(io, inode, page, ino))
 185                 goto out;
 186
 187         /* change META to META_FLUSH in the checkpoint procedure */
 188         if (type >= META_FLUSH) {
 189                 io->fio.type = META_FLUSH;
 190                 if (test_opt(sbi, NOBARRIER))
 191                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 192                 else
 193                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 194         }
 195         __submit_merged_bio(io);
 196 out:
 197         up_write(&io->io_rwsem);
 198 }
 199
 200 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 201                                                                         int rw)
 202 {
 203         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 204 }
 205
 206 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 207                                 struct inode *inode, struct page *page,
 208                                 nid_t ino, enum page_type type, int rw)
 209 {
 210         if (has_merged_page(sbi, inode, page, ino, type))
 211                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 212 }
 213
 214 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 215 {
 216         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 217         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 218         f2fs_submit_merged_bio(sbi, META, WRITE);
 219 }
 220
 221 /*
 222  * Fill the locked page with data located in the block address.
 223  * Return unlocked page.
 224  */
 225 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 226 {
 227         struct bio *bio;
 228         struct page *page = fio->encrypted_page ?
 229                         fio->encrypted_page : fio->page;
 230
 231         trace_f2fs_submit_page_bio(page, fio);
 232         f2fs_trace_ios(fio, 0);
 233
 234         /* Allocate a new bio */
 235         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
 236
 237         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 238                 bio_put(bio);
 239                 return -EFAULT;
 240         }
 241
 242         __submit_bio(fio->sbi, fio->rw, bio, fio->type);
 243         return 0;
 244 }
 245
 246 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 247 {
 248         struct f2fs_sb_info *sbi = fio->sbi;
 249         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 250         struct f2fs_bio_info *io;
 251         bool is_read = is_read_io(fio->rw);
 252         struct page *bio_page;
 253
 254         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 255
 256         if (fio->old_blkaddr != NEW_ADDR)
 257                 verify_block_addr(sbi, fio->old_blkaddr);
 258         verify_block_addr(sbi, fio->new_blkaddr);
 259
 260         down_write(&io->io_rwsem);
 261
 262         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 263                                                 io->fio.rw != fio->rw))
 264                 __submit_merged_bio(io);
 265 alloc_new:
 266         if (io->bio == NULL) {
 267                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 268
 269                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 270                                                 bio_blocks, is_read);
 271                 io->fio = *fio;
 272         }
 273
 274         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 275
 276         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 277                                                         PAGE_SIZE) {
 278                 __submit_merged_bio(io);
 279                 goto alloc_new;
 280         }
 281
 282         io->last_block_in_bio = fio->new_blkaddr;
 283         f2fs_trace_ios(fio, 0);
 284
 285         up_write(&io->io_rwsem);
 286         trace_f2fs_submit_page_mbio(fio->page, fio);
 287 }
 288
 289 static void __set_data_blkaddr(struct dnode_of_data *dn)
 290 {
 291         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 292         __le32 *addr_array;
 293
 294         /* Get physical address of data block */
 295         addr_array = blkaddr_in_node(rn);
 296         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 297 }
 298
 299 /*
 300  * Lock ordering for the change of data block address:
 301  * ->data_page
 302  *  ->node_page
 303  *    update block addresses in the node page
 304  */
 305 void set_data_blkaddr(struct dnode_of_data *dn)
 306 {
 307         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 308         __set_data_blkaddr(dn);
 309         if (set_page_dirty(dn->node_page))
 310                 dn->node_changed = true;
 311 }
 312
 313 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 314 {
 315         dn->data_blkaddr = blkaddr;
 316         set_data_blkaddr(dn);
 317         f2fs_update_extent_cache(dn);
 318 }
 319
 320 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 321 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 322 {
 323         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 324
 325         if (!count)
 326                 return 0;
 327
 328         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 329                 return -EPERM;
 330         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 331                 return -ENOSPC;
 332
 333         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 334                                                 dn->ofs_in_node, count);
 335
 336         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 337
 338         for (; count > 0; dn->ofs_in_node++) {
 339                 block_t blkaddr =
 340                         datablock_addr(dn->node_page, dn->ofs_in_node);
 341                 if (blkaddr == NULL_ADDR) {
 342                         dn->data_blkaddr = NEW_ADDR;
 343                         __set_data_blkaddr(dn);
 344                         count--;
 345                 }
 346         }
 347
 348         if (set_page_dirty(dn->node_page))
 349                 dn->node_changed = true;
 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                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 569         return page;
 570 }
 571
 572 static int __allocate_data_block(struct dnode_of_data *dn)
 573 {
 574         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 575         struct f2fs_summary sum;
 576         struct node_info ni;
 577         int seg = CURSEG_WARM_DATA;
 578         pgoff_t fofs;
 579         blkcnt_t count = 1;
 580
 581         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 582                 return -EPERM;
 583
 584         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 585         if (dn->data_blkaddr == NEW_ADDR)
 586                 goto alloc;
 587
 588         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 589                 return -ENOSPC;
 590
 591 alloc:
 592         get_node_info(sbi, dn->nid, &ni);
 593         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 594
 595         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 596                 seg = CURSEG_DIRECT_IO;
 597
 598         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 599                                                                 &sum, seg);
 600         set_data_blkaddr(dn);
 601
 602         /* update i_size */
 603         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 604                                                         dn->ofs_in_node;
 605         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 606                 f2fs_i_size_write(dn->inode,
 607                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 608         return 0;
 609 }
 610
 611 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 612 {
 613         struct inode *inode = file_inode(iocb->ki_filp);
 614         struct f2fs_map_blocks map;
 615         ssize_t ret = 0;
 616
 617         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 618         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 619         map.m_next_pgofs = NULL;
 620
 621         if (f2fs_encrypted_inode(inode))
 622                 return 0;
 623
 624         if (iocb->ki_flags & IOCB_DIRECT) {
 625                 ret = f2fs_convert_inline_inode(inode);
 626                 if (ret)
 627                         return ret;
 628                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 629         }
 630         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 631                 ret = f2fs_convert_inline_inode(inode);
 632                 if (ret)
 633                         return ret;
 634         }
 635         if (!f2fs_has_inline_data(inode))
 636                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 637         return ret;
 638 }
 639
 640 /*
 641  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 642  * f2fs_map_blocks structure.
 643  * If original data blocks are allocated, then give them to blockdev.
 644  * Otherwise,
 645  *     a. preallocate requested block addresses
 646  *     b. do not use extent cache for better performance
 647  *     c. give the block addresses to blockdev
 648  */
 649 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 650                                                 int create, int flag)
 651 {
 652         unsigned int maxblocks = map->m_len;
 653         struct dnode_of_data dn;
 654         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 655         int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
 656         pgoff_t pgofs, end_offset, end;
 657         int err = 0, ofs = 1;
 658         unsigned int ofs_in_node, last_ofs_in_node;
 659         blkcnt_t prealloc;
 660         struct extent_info ei;
 661         bool allocated = false;
 662         block_t blkaddr;
 663
 664         map->m_len = 0;
 665         map->m_flags = 0;
 666
 667         /* it only supports block size == page size */
 668         pgofs = (pgoff_t)map->m_lblk;
 669         end = pgofs + maxblocks;
 670
 671         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 672                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 673                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 674                 map->m_flags = F2FS_MAP_MAPPED;
 675                 goto out;
 676         }
 677
 678 next_dnode:
 679         if (create)
 680                 f2fs_lock_op(sbi);
 681
 682         /* When reading holes, we need its node page */
 683         set_new_dnode(&dn, inode, NULL, NULL, 0);
 684         err = get_dnode_of_data(&dn, pgofs, mode);
 685         if (err) {
 686                 if (flag == F2FS_GET_BLOCK_BMAP)
 687                         map->m_pblk = 0;
 688                 if (err == -ENOENT) {
 689                         err = 0;
 690                         if (map->m_next_pgofs)
 691                                 *map->m_next_pgofs =
 692                                         get_next_page_offset(&dn, pgofs);
 693                 }
 694                 goto unlock_out;
 695         }
 696
 697         prealloc = 0;
 698         ofs_in_node = dn.ofs_in_node;
 699         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 700
 701 next_block:
 702         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 703
 704         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 705                 if (create) {
 706                         if (unlikely(f2fs_cp_error(sbi))) {
 707                                 err = -EIO;
 708                                 goto sync_out;
 709                         }
 710                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 711                                 if (blkaddr == NULL_ADDR) {
 712                                         prealloc++;
 713                                         last_ofs_in_node = dn.ofs_in_node;
 714                                 }
 715                         } else {
 716                                 err = __allocate_data_block(&dn);
 717                                 if (!err) {
 718                                         set_inode_flag(inode, FI_APPEND_WRITE);
 719                                         allocated = true;
 720                                 }
 721                         }
 722                         if (err)
 723                                 goto sync_out;
 724                         map->m_flags = F2FS_MAP_NEW;
 725                         blkaddr = dn.data_blkaddr;
 726                 } else {
 727                         if (flag == F2FS_GET_BLOCK_BMAP) {
 728                                 map->m_pblk = 0;
 729                                 goto sync_out;
 730                         }
 731                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 732                                                 blkaddr == NULL_ADDR) {
 733                                 if (map->m_next_pgofs)
 734                                         *map->m_next_pgofs = pgofs + 1;
 735                         }
 736                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 737                                                 blkaddr != NEW_ADDR)
 738                                 goto sync_out;
 739                 }
 740         }
 741
 742         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 743                 goto skip;
 744
 745         if (map->m_len == 0) {
 746                 /* preallocated unwritten block should be mapped for fiemap. */
 747                 if (blkaddr == NEW_ADDR)
 748                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 749                 map->m_flags |= F2FS_MAP_MAPPED;
 750
 751                 map->m_pblk = blkaddr;
 752                 map->m_len = 1;
 753         } else if ((map->m_pblk != NEW_ADDR &&
 754                         blkaddr == (map->m_pblk + ofs)) ||
 755                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 756                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 757                 ofs++;
 758                 map->m_len++;
 759         } else {
 760                 goto sync_out;
 761         }
 762
 763 skip:
 764         dn.ofs_in_node++;
 765         pgofs++;
 766
 767         /* preallocate blocks in batch for one dnode page */
 768         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 769                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 770
 771                 dn.ofs_in_node = ofs_in_node;
 772                 err = reserve_new_blocks(&dn, prealloc);
 773                 if (err)
 774                         goto sync_out;
 775
 776                 map->m_len += dn.ofs_in_node - ofs_in_node;
 777                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 778                         err = -ENOSPC;
 779                         goto sync_out;
 780                 }
 781                 dn.ofs_in_node = end_offset;
 782         }
 783
 784         if (pgofs >= end)
 785                 goto sync_out;
 786         else if (dn.ofs_in_node < end_offset)
 787                 goto next_block;
 788
 789         f2fs_put_dnode(&dn);
 790
 791         if (create) {
 792                 f2fs_unlock_op(sbi);
 793                 f2fs_balance_fs(sbi, allocated);
 794         }
 795         allocated = false;
 796         goto next_dnode;
 797
 798 sync_out:
 799         f2fs_put_dnode(&dn);
 800 unlock_out:
 801         if (create) {
 802                 f2fs_unlock_op(sbi);
 803                 f2fs_balance_fs(sbi, allocated);
 804         }
 805 out:
 806         trace_f2fs_map_blocks(inode, map, err);
 807         return err;
 808 }
 809
 810 static int __get_data_block(struct inode *inode, sector_t iblock,
 811                         struct buffer_head *bh, int create, int flag,
 812                         pgoff_t *next_pgofs)
 813 {
 814         struct f2fs_map_blocks map;
 815         int ret;
 816
 817         map.m_lblk = iblock;
 818         map.m_len = bh->b_size >> inode->i_blkbits;
 819         map.m_next_pgofs = next_pgofs;
 820
 821         ret = f2fs_map_blocks(inode, &map, create, flag);
 822         if (!ret) {
 823                 map_bh(bh, inode->i_sb, map.m_pblk);
 824                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 825                 bh->b_size = map.m_len << inode->i_blkbits;
 826         }
 827         return ret;
 828 }
 829
 830 static int get_data_block(struct inode *inode, sector_t iblock,
 831                         struct buffer_head *bh_result, int create, int flag,
 832                         pgoff_t *next_pgofs)
 833 {
 834         return __get_data_block(inode, iblock, bh_result, create,
 835                                                         flag, next_pgofs);
 836 }
 837
 838 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 839                         struct buffer_head *bh_result, int create)
 840 {
 841         return __get_data_block(inode, iblock, bh_result, create,
 842                                                 F2FS_GET_BLOCK_DIO, NULL);
 843 }
 844
 845 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 846                         struct buffer_head *bh_result, int create)
 847 {
 848         /* Block number less than F2FS MAX BLOCKS */
 849         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 850                 return -EFBIG;
 851
 852         return __get_data_block(inode, iblock, bh_result, create,
 853                                                 F2FS_GET_BLOCK_BMAP, NULL);
 854 }
 855
 856 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 857 {
 858         return (offset >> inode->i_blkbits);
 859 }
 860
 861 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 862 {
 863         return (blk << inode->i_blkbits);
 864 }
 865
 866 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 867                 u64 start, u64 len)
 868 {
 869         struct buffer_head map_bh;
 870         sector_t start_blk, last_blk;
 871         pgoff_t next_pgofs;
 872         loff_t isize;
 873         u64 logical = 0, phys = 0, size = 0;
 874         u32 flags = 0;
 875         int ret = 0;
 876
 877         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 878         if (ret)
 879                 return ret;
 880
 881         if (f2fs_has_inline_data(inode)) {
 882                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 883                 if (ret != -EAGAIN)
 884                         return ret;
 885         }
 886
 887         inode_lock(inode);
 888
 889         isize = i_size_read(inode);
 890         if (start >= isize)
 891                 goto out;
 892
 893         if (start + len > isize)
 894                 len = isize - start;
 895
 896         if (logical_to_blk(inode, len) == 0)
 897                 len = blk_to_logical(inode, 1);
 898
 899         start_blk = logical_to_blk(inode, start);
 900         last_blk = logical_to_blk(inode, start + len - 1);
 901
 902 next:
 903         memset(&map_bh, 0, sizeof(struct buffer_head));
 904         map_bh.b_size = len;
 905
 906         ret = get_data_block(inode, start_blk, &map_bh, 0,
 907                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 908         if (ret)
 909                 goto out;
 910
 911         /* HOLE */
 912         if (!buffer_mapped(&map_bh)) {
 913                 start_blk = next_pgofs;
 914                 /* Go through holes util pass the EOF */
 915                 if (blk_to_logical(inode, start_blk) < isize)
 916                         goto prep_next;
 917                 /* Found a hole beyond isize means no more extents.
 918                  * Note that the premise is that filesystems don't
 919                  * punch holes beyond isize and keep size unchanged.
 920                  */
 921                 flags |= FIEMAP_EXTENT_LAST;
 922         }
 923
 924         if (size) {
 925                 if (f2fs_encrypted_inode(inode))
 926                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 927
 928                 ret = fiemap_fill_next_extent(fieinfo, logical,
 929                                 phys, size, flags);
 930         }
 931
 932         if (start_blk > last_blk || ret)
 933                 goto out;
 934
 935         logical = blk_to_logical(inode, start_blk);
 936         phys = blk_to_logical(inode, map_bh.b_blocknr);
 937         size = map_bh.b_size;
 938         flags = 0;
 939         if (buffer_unwritten(&map_bh))
 940                 flags = FIEMAP_EXTENT_UNWRITTEN;
 941
 942         start_blk += logical_to_blk(inode, size);
 943
 944 prep_next:
 945         cond_resched();
 946         if (fatal_signal_pending(current))
 947                 ret = -EINTR;
 948         else
 949                 goto next;
 950 out:
 951         if (ret == 1)
 952                 ret = 0;
 953
 954         inode_unlock(inode);
 955         return ret;
 956 }
 957
 958 /*
 959  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 960  * Major change was from block_size == page_size in f2fs by default.
 961  */
 962 static int f2fs_mpage_readpages(struct address_space *mapping,
 963                         struct list_head *pages, struct page *page,
 964                         unsigned nr_pages)
 965 {
 966         struct bio *bio = NULL;
 967         unsigned page_idx;
 968         sector_t last_block_in_bio = 0;
 969         struct inode *inode = mapping->host;
 970         const unsigned blkbits = inode->i_blkbits;
 971         const unsigned blocksize = 1 << blkbits;
 972         sector_t block_in_file;
 973         sector_t last_block;
 974         sector_t last_block_in_file;
 975         sector_t block_nr;
 976         struct block_device *bdev = inode->i_sb->s_bdev;
 977         struct f2fs_map_blocks map;
 978
 979         map.m_pblk = 0;
 980         map.m_lblk = 0;
 981         map.m_len = 0;
 982         map.m_flags = 0;
 983         map.m_next_pgofs = NULL;
 984
 985         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 986
 987                 prefetchw(&page->flags);
 988                 if (pages) {
 989                         page = list_entry(pages->prev, struct page, lru);
 990                         list_del(&page->lru);
 991                         if (add_to_page_cache_lru(page, mapping,
 992                                                   page->index, GFP_KERNEL))
 993                                 goto next_page;
 994                 }
 995
 996                 block_in_file = (sector_t)page->index;
 997                 last_block = block_in_file + nr_pages;
 998                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
 999                                                                 blkbits;
1000                 if (last_block > last_block_in_file)
1001                         last_block = last_block_in_file;
1002
1003                 /*
1004                  * Map blocks using the previous result first.
1005                  */
1006                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1007                                 block_in_file > map.m_lblk &&
1008                                 block_in_file < (map.m_lblk + map.m_len))
1009                         goto got_it;
1010
1011                 /*
1012                  * Then do more f2fs_map_blocks() calls until we are
1013                  * done with this page.
1014                  */
1015                 map.m_flags = 0;
1016
1017                 if (block_in_file < last_block) {
1018                         map.m_lblk = block_in_file;
1019                         map.m_len = last_block - block_in_file;
1020
1021                         if (f2fs_map_blocks(inode, &map, 0,
1022                                                 F2FS_GET_BLOCK_READ))
1023                                 goto set_error_page;
1024                 }
1025 got_it:
1026                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1027                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1028                         SetPageMappedToDisk(page);
1029
1030                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1031                                 SetPageUptodate(page);
1032                                 goto confused;
1033                         }
1034                 } else {
1035                         zero_user_segment(page, 0, PAGE_SIZE);
1036                         SetPageUptodate(page);
1037                         unlock_page(page);
1038                         goto next_page;
1039                 }
1040
1041                 /*
1042                  * This page will go to BIO.  Do we need to send this
1043                  * BIO off first?
1044                  */
1045                 if (bio && (last_block_in_bio != block_nr - 1)) {
1046 submit_and_realloc:
1047                         __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1048                         bio = NULL;
1049                 }
1050                 if (bio == NULL) {
1051                         struct fscrypt_ctx *ctx = NULL;
1052
1053                         if (f2fs_encrypted_inode(inode) &&
1054                                         S_ISREG(inode->i_mode)) {
1055
1056                                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1057                                 if (IS_ERR(ctx))
1058                                         goto set_error_page;
1059
1060                                 /* wait the page to be moved by cleaning */
1061                                 f2fs_wait_on_encrypted_page_writeback(
1062                                                 F2FS_I_SB(inode), block_nr);
1063                         }
1064
1065                         bio = bio_alloc(GFP_KERNEL,
1066                                 min_t(int, nr_pages, BIO_MAX_PAGES));
1067                         if (!bio) {
1068                                 if (ctx)
1069                                         fscrypt_release_ctx(ctx);
1070                                 goto set_error_page;
1071                         }
1072                         bio->bi_bdev = bdev;
1073                         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1074                         bio->bi_end_io = f2fs_read_end_io;
1075                         bio->bi_private = ctx;
1076                 }
1077
1078                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1079                         goto submit_and_realloc;
1080
1081                 last_block_in_bio = block_nr;
1082                 goto next_page;
1083 set_error_page:
1084                 SetPageError(page);
1085                 zero_user_segment(page, 0, PAGE_SIZE);
1086                 unlock_page(page);
1087                 goto next_page;
1088 confused:
1089                 if (bio) {
1090                         __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1091                         bio = NULL;
1092                 }
1093                 unlock_page(page);
1094 next_page:
1095                 if (pages)
1096                         put_page(page);
1097         }
1098         BUG_ON(pages && !list_empty(pages));
1099         if (bio)
1100                 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1101         return 0;
1102 }
1103
1104 static int f2fs_read_data_page(struct file *file, struct page *page)
1105 {
1106         struct inode *inode = page->mapping->host;
1107         int ret = -EAGAIN;
1108
1109         trace_f2fs_readpage(page, DATA);
1110
1111         /* If the file has inline data, try to read it directly */
1112         if (f2fs_has_inline_data(inode))
1113                 ret = f2fs_read_inline_data(inode, page);
1114         if (ret == -EAGAIN)
1115                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1116         return ret;
1117 }
1118
1119 static int f2fs_read_data_pages(struct file *file,
1120                         struct address_space *mapping,
1121                         struct list_head *pages, unsigned nr_pages)
1122 {
1123         struct inode *inode = file->f_mapping->host;
1124         struct page *page = list_entry(pages->prev, struct page, lru);
1125
1126         trace_f2fs_readpages(inode, page, nr_pages);
1127
1128         /* If the file has inline data, skip readpages */
1129         if (f2fs_has_inline_data(inode))
1130                 return 0;
1131
1132         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1133 }
1134
1135 int do_write_data_page(struct f2fs_io_info *fio)
1136 {
1137         struct page *page = fio->page;
1138         struct inode *inode = page->mapping->host;
1139         struct dnode_of_data dn;
1140         int err = 0;
1141
1142         set_new_dnode(&dn, inode, NULL, NULL, 0);
1143         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1144         if (err)
1145                 return err;
1146
1147         fio->old_blkaddr = dn.data_blkaddr;
1148
1149         /* This page is already truncated */
1150         if (fio->old_blkaddr == NULL_ADDR) {
1151                 ClearPageUptodate(page);
1152                 goto out_writepage;
1153         }
1154
1155         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1156                 gfp_t gfp_flags = GFP_NOFS;
1157
1158                 /* wait for GCed encrypted page writeback */
1159                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1160                                                         fio->old_blkaddr);
1161 retry_encrypt:
1162                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1163                                                                 gfp_flags);
1164                 if (IS_ERR(fio->encrypted_page)) {
1165                         err = PTR_ERR(fio->encrypted_page);
1166                         if (err == -ENOMEM) {
1167                                 /* flush pending ios and wait for a while */
1168                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1169                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1170                                 gfp_flags |= __GFP_NOFAIL;
1171                                 err = 0;
1172                                 goto retry_encrypt;
1173                         }
1174                         goto out_writepage;
1175                 }
1176         }
1177
1178         set_page_writeback(page);
1179
1180         /*
1181          * If current allocation needs SSR,
1182          * it had better in-place writes for updated data.
1183          */
1184         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1185                         !is_cold_data(page) &&
1186                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1187                         need_inplace_update(inode))) {
1188                 rewrite_data_page(fio);
1189                 set_inode_flag(inode, FI_UPDATE_WRITE);
1190                 trace_f2fs_do_write_data_page(page, IPU);
1191         } else {
1192                 write_data_page(&dn, fio);
1193                 trace_f2fs_do_write_data_page(page, OPU);
1194                 set_inode_flag(inode, FI_APPEND_WRITE);
1195                 if (page->index == 0)
1196                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1197         }
1198 out_writepage:
1199         f2fs_put_dnode(&dn);
1200         return err;
1201 }
1202
1203 static int f2fs_write_data_page(struct page *page,
1204                                         struct writeback_control *wbc)
1205 {
1206         struct inode *inode = page->mapping->host;
1207         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1208         loff_t i_size = i_size_read(inode);
1209         const pgoff_t end_index = ((unsigned long long) i_size)
1210                                                         >> PAGE_SHIFT;
1211         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1212         unsigned offset = 0;
1213         bool need_balance_fs = false;
1214         int err = 0;
1215         struct f2fs_io_info fio = {
1216                 .sbi = sbi,
1217                 .type = DATA,
1218                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1219                 .page = page,
1220                 .encrypted_page = NULL,
1221         };
1222
1223         trace_f2fs_writepage(page, DATA);
1224
1225         if (page->index < end_index)
1226                 goto write;
1227
1228         /*
1229          * If the offset is out-of-range of file size,
1230          * this page does not have to be written to disk.
1231          */
1232         offset = i_size & (PAGE_SIZE - 1);
1233         if ((page->index >= end_index + 1) || !offset)
1234                 goto out;
1235
1236         zero_user_segment(page, offset, PAGE_SIZE);
1237 write:
1238         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1239                 goto redirty_out;
1240         if (f2fs_is_drop_cache(inode))
1241                 goto out;
1242         /* we should not write 0'th page having journal header */
1243         if (f2fs_is_volatile_file(inode) && (!page->index ||
1244                         (!wbc->for_reclaim &&
1245                         available_free_memory(sbi, BASE_CHECK))))
1246                 goto redirty_out;
1247
1248         /* we should bypass data pages to proceed the kworkder jobs */
1249         if (unlikely(f2fs_cp_error(sbi))) {
1250                 mapping_set_error(page->mapping, -EIO);
1251                 goto out;
1252         }
1253
1254         /* Dentry blocks are controlled by checkpoint */
1255         if (S_ISDIR(inode->i_mode)) {
1256                 err = do_write_data_page(&fio);
1257                 goto done;
1258         }
1259
1260         if (!wbc->for_reclaim)
1261                 need_balance_fs = true;
1262         else if (has_not_enough_free_secs(sbi, 0))
1263                 goto redirty_out;
1264
1265         err = -EAGAIN;
1266         f2fs_lock_op(sbi);
1267         if (f2fs_has_inline_data(inode))
1268                 err = f2fs_write_inline_data(inode, page);
1269         if (err == -EAGAIN)
1270                 err = do_write_data_page(&fio);
1271         if (F2FS_I(inode)->last_disk_size < psize)
1272                 F2FS_I(inode)->last_disk_size = psize;
1273         f2fs_unlock_op(sbi);
1274 done:
1275         if (err && err != -ENOENT)
1276                 goto redirty_out;
1277
1278         clear_cold_data(page);
1279 out:
1280         inode_dec_dirty_pages(inode);
1281         if (err)
1282                 ClearPageUptodate(page);
1283
1284         if (wbc->for_reclaim) {
1285                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1286                 remove_dirty_inode(inode);
1287         }
1288
1289         unlock_page(page);
1290         f2fs_balance_fs(sbi, need_balance_fs);
1291
1292         if (unlikely(f2fs_cp_error(sbi)))
1293                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1294
1295         return 0;
1296
1297 redirty_out:
1298         redirty_page_for_writepage(wbc, page);
1299         unlock_page(page);
1300         return err;
1301 }
1302
1303 /*
1304  * This function was copied from write_cche_pages from mm/page-writeback.c.
1305  * The major change is making write step of cold data page separately from
1306  * warm/hot data page.
1307  */
1308 static int f2fs_write_cache_pages(struct address_space *mapping,
1309                                         struct writeback_control *wbc)
1310 {
1311         int ret = 0;
1312         int done = 0;
1313         struct pagevec pvec;
1314         int nr_pages;
1315         pgoff_t uninitialized_var(writeback_index);
1316         pgoff_t index;
1317         pgoff_t end;            /* Inclusive */
1318         pgoff_t done_index;
1319         int cycled;
1320         int range_whole = 0;
1321         int tag;
1322
1323         pagevec_init(&pvec, 0);
1324
1325         if (wbc->range_cyclic) {
1326                 writeback_index = mapping->writeback_index; /* prev offset */
1327                 index = writeback_index;
1328                 if (index == 0)
1329                         cycled = 1;
1330                 else
1331                         cycled = 0;
1332                 end = -1;
1333         } else {
1334                 index = wbc->range_start >> PAGE_SHIFT;
1335                 end = wbc->range_end >> PAGE_SHIFT;
1336                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1337                         range_whole = 1;
1338                 cycled = 1; /* ignore range_cyclic tests */
1339         }
1340         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1341                 tag = PAGECACHE_TAG_TOWRITE;
1342         else
1343                 tag = PAGECACHE_TAG_DIRTY;
1344 retry:
1345         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1346                 tag_pages_for_writeback(mapping, index, end);
1347         done_index = index;
1348         while (!done && (index <= end)) {
1349                 int i;
1350
1351                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1352                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1353                 if (nr_pages == 0)
1354                         break;
1355
1356                 for (i = 0; i < nr_pages; i++) {
1357                         struct page *page = pvec.pages[i];
1358
1359                         if (page->index > end) {
1360                                 done = 1;
1361                                 break;
1362                         }
1363
1364                         done_index = page->index;
1365
1366                         lock_page(page);
1367
1368                         if (unlikely(page->mapping != mapping)) {
1369 continue_unlock:
1370                                 unlock_page(page);
1371                                 continue;
1372                         }
1373
1374                         if (!PageDirty(page)) {
1375                                 /* someone wrote it for us */
1376                                 goto continue_unlock;
1377                         }
1378
1379                         if (PageWriteback(page)) {
1380                                 if (wbc->sync_mode != WB_SYNC_NONE)
1381                                         f2fs_wait_on_page_writeback(page,
1382                                                                 DATA, true);
1383                                 else
1384                                         goto continue_unlock;
1385                         }
1386
1387                         BUG_ON(PageWriteback(page));
1388                         if (!clear_page_dirty_for_io(page))
1389                                 goto continue_unlock;
1390
1391                         ret = mapping->a_ops->writepage(page, wbc);
1392                         if (unlikely(ret)) {
1393                                 done_index = page->index + 1;
1394                                 done = 1;
1395                                 break;
1396                         }
1397
1398                         if (--wbc->nr_to_write <= 0 &&
1399                             wbc->sync_mode == WB_SYNC_NONE) {
1400                                 done = 1;
1401                                 break;
1402                         }
1403                 }
1404                 pagevec_release(&pvec);
1405                 cond_resched();
1406         }
1407
1408         if (!cycled && !done) {
1409                 cycled = 1;
1410                 index = 0;
1411                 end = writeback_index - 1;
1412                 goto retry;
1413         }
1414         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1415                 mapping->writeback_index = done_index;
1416
1417         return ret;
1418 }
1419
1420 static int f2fs_write_data_pages(struct address_space *mapping,
1421                             struct writeback_control *wbc)
1422 {
1423         struct inode *inode = mapping->host;
1424         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1425         int ret;
1426
1427         /* deal with chardevs and other special file */
1428         if (!mapping->a_ops->writepage)
1429                 return 0;
1430
1431         /* skip writing if there is no dirty page in this inode */
1432         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1433                 return 0;
1434
1435         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1436                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1437                         available_free_memory(sbi, DIRTY_DENTS))
1438                 goto skip_write;
1439
1440         /* skip writing during file defragment */
1441         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1442                 goto skip_write;
1443
1444         /* during POR, we don't need to trigger writepage at all. */
1445         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1446                 goto skip_write;
1447
1448         trace_f2fs_writepages(mapping->host, wbc, DATA);
1449
1450         ret = f2fs_write_cache_pages(mapping, wbc);
1451         /*
1452          * if some pages were truncated, we cannot guarantee its mapping->host
1453          * to detect pending bios.
1454          */
1455         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1456
1457         remove_dirty_inode(inode);
1458         return ret;
1459
1460 skip_write:
1461         wbc->pages_skipped += get_dirty_pages(inode);
1462         trace_f2fs_writepages(mapping->host, wbc, DATA);
1463         return 0;
1464 }
1465
1466 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1467 {
1468         struct inode *inode = mapping->host;
1469         loff_t i_size = i_size_read(inode);
1470
1471         if (to > i_size) {
1472                 truncate_pagecache(inode, i_size);
1473                 truncate_blocks(inode, i_size, true);
1474         }
1475 }
1476
1477 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1478                         struct page *page, loff_t pos, unsigned len,
1479                         block_t *blk_addr, bool *node_changed)
1480 {
1481         struct inode *inode = page->mapping->host;
1482         pgoff_t index = page->index;
1483         struct dnode_of_data dn;
1484         struct page *ipage;
1485         bool locked = false;
1486         struct extent_info ei;
1487         int err = 0;
1488
1489         /*
1490          * we already allocated all the blocks, so we don't need to get
1491          * the block addresses when there is no need to fill the page.
1492          */
1493         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1494                                         len == PAGE_SIZE)
1495                 return 0;
1496
1497         if (f2fs_has_inline_data(inode) ||
1498                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1499                 f2fs_lock_op(sbi);
1500                 locked = true;
1501         }
1502 restart:
1503         /* check inline_data */
1504         ipage = get_node_page(sbi, inode->i_ino);
1505         if (IS_ERR(ipage)) {
1506                 err = PTR_ERR(ipage);
1507                 goto unlock_out;
1508         }
1509
1510         set_new_dnode(&dn, inode, ipage, ipage, 0);
1511
1512         if (f2fs_has_inline_data(inode)) {
1513                 if (pos + len <= MAX_INLINE_DATA) {
1514                         read_inline_data(page, ipage);
1515                         set_inode_flag(inode, FI_DATA_EXIST);
1516                         if (inode->i_nlink)
1517                                 set_inline_node(ipage);
1518                 } else {
1519                         err = f2fs_convert_inline_page(&dn, page);
1520                         if (err)
1521                                 goto out;
1522                         if (dn.data_blkaddr == NULL_ADDR)
1523                                 err = f2fs_get_block(&dn, index);
1524                 }
1525         } else if (locked) {
1526                 err = f2fs_get_block(&dn, index);
1527         } else {
1528                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1529                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1530                 } else {
1531                         /* hole case */
1532                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1533                         if (err || dn.data_blkaddr == NULL_ADDR) {
1534                                 f2fs_put_dnode(&dn);
1535                                 f2fs_lock_op(sbi);
1536                                 locked = true;
1537                                 goto restart;
1538                         }
1539                 }
1540         }
1541
1542         /* convert_inline_page can make node_changed */
1543         *blk_addr = dn.data_blkaddr;
1544         *node_changed = dn.node_changed;
1545 out:
1546         f2fs_put_dnode(&dn);
1547 unlock_out:
1548         if (locked)
1549                 f2fs_unlock_op(sbi);
1550         return err;
1551 }
1552
1553 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1554                 loff_t pos, unsigned len, unsigned flags,
1555                 struct page **pagep, void **fsdata)
1556 {
1557         struct inode *inode = mapping->host;
1558         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1559         struct page *page = NULL;
1560         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1561         bool need_balance = false;
1562         block_t blkaddr = NULL_ADDR;
1563         int err = 0;
1564
1565         trace_f2fs_write_begin(inode, pos, len, flags);
1566
1567         /*
1568          * We should check this at this moment to avoid deadlock on inode page
1569          * and #0 page. The locking rule for inline_data conversion should be:
1570          * lock_page(page #0) -> lock_page(inode_page)
1571          */
1572         if (index != 0) {
1573                 err = f2fs_convert_inline_inode(inode);
1574                 if (err)
1575                         goto fail;
1576         }
1577 repeat:
1578         page = grab_cache_page_write_begin(mapping, index, flags);
1579         if (!page) {
1580                 err = -ENOMEM;
1581                 goto fail;
1582         }
1583
1584         *pagep = page;
1585
1586         err = prepare_write_begin(sbi, page, pos, len,
1587                                         &blkaddr, &need_balance);
1588         if (err)
1589                 goto fail;
1590
1591         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1592                 unlock_page(page);
1593                 f2fs_balance_fs(sbi, true);
1594                 lock_page(page);
1595                 if (page->mapping != mapping) {
1596                         /* The page got truncated from under us */
1597                         f2fs_put_page(page, 1);
1598                         goto repeat;
1599                 }
1600         }
1601
1602         f2fs_wait_on_page_writeback(page, DATA, false);
1603
1604         /* wait for GCed encrypted page writeback */
1605         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1606                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1607
1608         if (len == PAGE_SIZE)
1609                 goto out_update;
1610         if (PageUptodate(page))
1611                 goto out_clear;
1612
1613         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1614                 unsigned start = pos & (PAGE_SIZE - 1);
1615                 unsigned end = start + len;
1616
1617                 /* Reading beyond i_size is simple: memset to zero */
1618                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1619                 goto out_update;
1620         }
1621
1622         if (blkaddr == NEW_ADDR) {
1623                 zero_user_segment(page, 0, PAGE_SIZE);
1624         } else {
1625                 struct f2fs_io_info fio = {
1626                         .sbi = sbi,
1627                         .type = DATA,
1628                         .rw = READ_SYNC,
1629                         .old_blkaddr = blkaddr,
1630                         .new_blkaddr = blkaddr,
1631                         .page = page,
1632                         .encrypted_page = NULL,
1633                 };
1634                 err = f2fs_submit_page_bio(&fio);
1635                 if (err)
1636                         goto fail;
1637
1638                 lock_page(page);
1639                 if (unlikely(!PageUptodate(page))) {
1640                         err = -EIO;
1641                         goto fail;
1642                 }
1643                 if (unlikely(page->mapping != mapping)) {
1644                         f2fs_put_page(page, 1);
1645                         goto repeat;
1646                 }
1647
1648                 /* avoid symlink page */
1649                 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1650                         err = fscrypt_decrypt_page(page);
1651                         if (err)
1652                                 goto fail;
1653                 }
1654         }
1655 out_update:
1656         SetPageUptodate(page);
1657 out_clear:
1658         clear_cold_data(page);
1659         return 0;
1660
1661 fail:
1662         f2fs_put_page(page, 1);
1663         f2fs_write_failed(mapping, pos + len);
1664         return err;
1665 }
1666
1667 static int f2fs_write_end(struct file *file,
1668                         struct address_space *mapping,
1669                         loff_t pos, unsigned len, unsigned copied,
1670                         struct page *page, void *fsdata)
1671 {
1672         struct inode *inode = page->mapping->host;
1673
1674         trace_f2fs_write_end(inode, pos, len, copied);
1675
1676         set_page_dirty(page);
1677
1678         if (pos + copied > i_size_read(inode))
1679                 f2fs_i_size_write(inode, pos + copied);
1680
1681         f2fs_put_page(page, 1);
1682         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1683         return copied;
1684 }
1685
1686 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1687                            loff_t offset)
1688 {
1689         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1690
1691         if (offset & blocksize_mask)
1692                 return -EINVAL;
1693
1694         if (iov_iter_alignment(iter) & blocksize_mask)
1695                 return -EINVAL;
1696
1697         return 0;
1698 }
1699
1700 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1701 {
1702         struct address_space *mapping = iocb->ki_filp->f_mapping;
1703         struct inode *inode = mapping->host;
1704         size_t count = iov_iter_count(iter);
1705         loff_t offset = iocb->ki_pos;
1706         int err;
1707
1708         err = check_direct_IO(inode, iter, offset);
1709         if (err)
1710                 return err;
1711
1712         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1713                 return 0;
1714         if (test_opt(F2FS_I_SB(inode), LFS))
1715                 return 0;
1716
1717         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1718
1719         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1720         if (iov_iter_rw(iter) == WRITE) {
1721                 if (err > 0)
1722                         set_inode_flag(inode, FI_UPDATE_WRITE);
1723                 else if (err < 0)
1724                         f2fs_write_failed(mapping, offset + count);
1725         }
1726
1727         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1728
1729         return err;
1730 }
1731
1732 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1733                                                         unsigned int length)
1734 {
1735         struct inode *inode = page->mapping->host;
1736         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1737
1738         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1739                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1740                 return;
1741
1742         if (PageDirty(page)) {
1743                 if (inode->i_ino == F2FS_META_INO(sbi))
1744                         dec_page_count(sbi, F2FS_DIRTY_META);
1745                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1746                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1747                 else
1748                         inode_dec_dirty_pages(inode);
1749         }
1750
1751         /* This is atomic written page, keep Private */
1752         if (IS_ATOMIC_WRITTEN_PAGE(page))
1753                 return;
1754
1755         set_page_private(page, 0);
1756         ClearPagePrivate(page);
1757 }
1758
1759 int f2fs_release_page(struct page *page, gfp_t wait)
1760 {
1761         /* If this is dirty page, keep PagePrivate */
1762         if (PageDirty(page))
1763                 return 0;
1764
1765         /* This is atomic written page, keep Private */
1766         if (IS_ATOMIC_WRITTEN_PAGE(page))
1767                 return 0;
1768
1769         set_page_private(page, 0);
1770         ClearPagePrivate(page);
1771         return 1;
1772 }
1773
1774 static int f2fs_set_data_page_dirty(struct page *page)
1775 {
1776         struct address_space *mapping = page->mapping;
1777         struct inode *inode = mapping->host;
1778
1779         trace_f2fs_set_page_dirty(page, DATA);
1780
1781         SetPageUptodate(page);
1782
1783         if (f2fs_is_atomic_file(inode)) {
1784                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1785                         register_inmem_page(inode, page);
1786                         return 1;
1787                 }
1788                 /*
1789                  * Previously, this page has been registered, we just
1790                  * return here.
1791                  */
1792                 return 0;
1793         }
1794
1795         if (!PageDirty(page)) {
1796                 __set_page_dirty_nobuffers(page);
1797                 update_dirty_page(inode, page);
1798                 return 1;
1799         }
1800         return 0;
1801 }
1802
1803 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1804 {
1805         struct inode *inode = mapping->host;
1806
1807         if (f2fs_has_inline_data(inode))
1808                 return 0;
1809
1810         /* make sure allocating whole blocks */
1811         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1812                 filemap_write_and_wait(mapping);
1813
1814         return generic_block_bmap(mapping, block, get_data_block_bmap);
1815 }
1816
1817 const struct address_space_operations f2fs_dblock_aops = {
1818         .readpage       = f2fs_read_data_page,
1819         .readpages      = f2fs_read_data_pages,
1820         .writepage      = f2fs_write_data_page,
1821         .writepages     = f2fs_write_data_pages,
1822         .write_begin    = f2fs_write_begin,
1823         .write_end      = f2fs_write_end,
1824         .set_page_dirty = f2fs_set_data_page_dirty,
1825         .invalidatepage = f2fs_invalidate_page,
1826         .releasepage    = f2fs_release_page,
1827         .direct_IO      = f2fs_direct_IO,
1828         .bmap           = f2fs_bmap,
1829 };