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