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/mm.h>
  23 #include <linux/memcontrol.h>
  24 #include <linux/cleancache.h>
  25
  26 #include "f2fs.h"
  27 #include "node.h"
  28 #include "segment.h"
  29 #include "trace.h"
  30 #include <trace/events/f2fs.h>
  31
  32 static void f2fs_read_end_io(struct bio *bio)
  33 {
  34         struct bio_vec *bvec;
  35         int i;
  36
  37         if (f2fs_bio_encrypted(bio)) {
  38                 if (bio->bi_error) {
  39                         fscrypt_release_ctx(bio->bi_private);
  40                 } else {
  41                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  42                         return;
  43                 }
  44         }
  45
  46         bio_for_each_segment_all(bvec, bio, i) {
  47                 struct page *page = bvec->bv_page;
  48
  49                 if (!bio->bi_error) {
  50                         if (!PageUptodate(page))
  51                                 SetPageUptodate(page);
  52                 } else {
  53                         ClearPageUptodate(page);
  54                         SetPageError(page);
  55                 }
  56                 unlock_page(page);
  57         }
  58         bio_put(bio);
  59 }
  60
  61 static void f2fs_write_end_io(struct bio *bio)
  62 {
  63         struct f2fs_sb_info *sbi = bio->bi_private;
  64         struct bio_vec *bvec;
  65         int i;
  66
  67         bio_for_each_segment_all(bvec, bio, i) {
  68                 struct page *page = bvec->bv_page;
  69
  70                 fscrypt_pullback_bio_page(&page, true);
  71
  72                 if (unlikely(bio->bi_error)) {
  73                         set_bit(AS_EIO, &page->mapping->flags);
  74                         f2fs_stop_checkpoint(sbi, true);
  75                 }
  76                 end_page_writeback(page);
  77         }
  78         if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
  79                                 wq_has_sleeper(&sbi->cp_wait))
  80                 wake_up(&sbi->cp_wait);
  81
  82         bio_put(bio);
  83 }
  84
  85 /*
  86  * Low-level block read/write IO operations.
  87  */
  88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
  89                                 int npages, bool is_read)
  90 {
  91         struct bio *bio;
  92
  93         bio = f2fs_bio_alloc(npages);
  94
  95         bio->bi_bdev = sbi->sb->s_bdev;
  96         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
  97         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
  98         bio->bi_private = is_read ? NULL : sbi;
  99
 100         return bio;
 101 }
 102
 103 static inline void __submit_bio(struct f2fs_sb_info *sbi, int rw,
 104                         struct bio *bio, enum page_type type)
 105 {
 106         if (!is_read_io(rw)) {
 107                 atomic_inc(&sbi->nr_wb_bios);
 108                 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
 109                         current->plug && (type == DATA || type == NODE))
 110                         blk_finish_plug(current->plug);
 111         }
 112         submit_bio(rw, bio);
 113 }
 114
 115 static void __submit_merged_bio(struct f2fs_bio_info *io)
 116 {
 117         struct f2fs_io_info *fio = &io->fio;
 118
 119         if (!io->bio)
 120                 return;
 121
 122         if (is_read_io(fio->rw))
 123                 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
 124         else
 125                 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
 126
 127         __submit_bio(io->sbi, fio->rw, io->bio, fio->type);
 128         io->bio = NULL;
 129 }
 130
 131 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 132                                                 struct page *page, nid_t ino)
 133 {
 134         struct bio_vec *bvec;
 135         struct page *target;
 136         int i;
 137
 138         if (!io->bio)
 139                 return false;
 140
 141         if (!inode && !page && !ino)
 142                 return true;
 143
 144         bio_for_each_segment_all(bvec, io->bio, i) {
 145
 146                 if (bvec->bv_page->mapping)
 147                         target = bvec->bv_page;
 148                 else
 149                         target = fscrypt_control_page(bvec->bv_page);
 150
 151                 if (inode && inode == target->mapping->host)
 152                         return true;
 153                 if (page && page == target)
 154                         return true;
 155                 if (ino && ino == ino_of_node(target))
 156                         return true;
 157         }
 158
 159         return false;
 160 }
 161
 162 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 163                                                 struct page *page, nid_t ino,
 164                                                 enum page_type type)
 165 {
 166         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 167         struct f2fs_bio_info *io = &sbi->write_io[btype];
 168         bool ret;
 169
 170         down_read(&io->io_rwsem);
 171         ret = __has_merged_page(io, inode, page, ino);
 172         up_read(&io->io_rwsem);
 173         return ret;
 174 }
 175
 176 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 177                                 struct inode *inode, struct page *page,
 178                                 nid_t ino, enum page_type type, int rw)
 179 {
 180         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 181         struct f2fs_bio_info *io;
 182
 183         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 184
 185         down_write(&io->io_rwsem);
 186
 187         if (!__has_merged_page(io, inode, page, ino))
 188                 goto out;
 189
 190         /* change META to META_FLUSH in the checkpoint procedure */
 191         if (type >= META_FLUSH) {
 192                 io->fio.type = META_FLUSH;
 193                 if (test_opt(sbi, NOBARRIER))
 194                         io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
 195                 else
 196                         io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
 197         }
 198         __submit_merged_bio(io);
 199 out:
 200         up_write(&io->io_rwsem);
 201 }
 202
 203 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 204                                                                         int rw)
 205 {
 206         __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
 207 }
 208
 209 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 210                                 struct inode *inode, struct page *page,
 211                                 nid_t ino, enum page_type type, int rw)
 212 {
 213         if (has_merged_page(sbi, inode, page, ino, type))
 214                 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
 215 }
 216
 217 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 218 {
 219         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 220         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 221         f2fs_submit_merged_bio(sbi, META, WRITE);
 222 }
 223
 224 /*
 225  * Fill the locked page with data located in the block address.
 226  * Return unlocked page.
 227  */
 228 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 229 {
 230         struct bio *bio;
 231         struct page *page = fio->encrypted_page ?
 232                         fio->encrypted_page : fio->page;
 233
 234         trace_f2fs_submit_page_bio(page, fio);
 235         f2fs_trace_ios(fio, 0);
 236
 237         /* Allocate a new bio */
 238         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
 239
 240         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 241                 bio_put(bio);
 242                 return -EFAULT;
 243         }
 244
 245         __submit_bio(fio->sbi, fio->rw, bio, fio->type);
 246         return 0;
 247 }
 248
 249 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 250 {
 251         struct f2fs_sb_info *sbi = fio->sbi;
 252         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 253         struct f2fs_bio_info *io;
 254         bool is_read = is_read_io(fio->rw);
 255         struct page *bio_page;
 256
 257         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 258
 259         if (fio->old_blkaddr != NEW_ADDR)
 260                 verify_block_addr(sbi, fio->old_blkaddr);
 261         verify_block_addr(sbi, fio->new_blkaddr);
 262
 263         down_write(&io->io_rwsem);
 264
 265         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 266                                                 io->fio.rw != fio->rw))
 267                 __submit_merged_bio(io);
 268 alloc_new:
 269         if (io->bio == NULL) {
 270                 int bio_blocks = MAX_BIO_BLOCKS(sbi);
 271
 272                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 273                                                 bio_blocks, is_read);
 274                 io->fio = *fio;
 275         }
 276
 277         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 278
 279         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 280                                                         PAGE_SIZE) {
 281                 __submit_merged_bio(io);
 282                 goto alloc_new;
 283         }
 284
 285         io->last_block_in_bio = fio->new_blkaddr;
 286         f2fs_trace_ios(fio, 0);
 287
 288         up_write(&io->io_rwsem);
 289         trace_f2fs_submit_page_mbio(fio->page, fio);
 290 }
 291
 292 static void __set_data_blkaddr(struct dnode_of_data *dn)
 293 {
 294         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 295         __le32 *addr_array;
 296
 297         /* Get physical address of data block */
 298         addr_array = blkaddr_in_node(rn);
 299         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 300 }
 301
 302 /*
 303  * Lock ordering for the change of data block address:
 304  * ->data_page
 305  *  ->node_page
 306  *    update block addresses in the node page
 307  */
 308 void set_data_blkaddr(struct dnode_of_data *dn)
 309 {
 310         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 311         __set_data_blkaddr(dn);
 312         if (set_page_dirty(dn->node_page))
 313                 dn->node_changed = true;
 314 }
 315
 316 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 317 {
 318         dn->data_blkaddr = blkaddr;
 319         set_data_blkaddr(dn);
 320         f2fs_update_extent_cache(dn);
 321 }
 322
 323 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 324 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 325 {
 326         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 327
 328         if (!count)
 329                 return 0;
 330
 331         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 332                 return -EPERM;
 333         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 334                 return -ENOSPC;
 335
 336         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 337                                                 dn->ofs_in_node, count);
 338
 339         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 340
 341         for (; count > 0; dn->ofs_in_node++) {
 342                 block_t blkaddr =
 343                         datablock_addr(dn->node_page, dn->ofs_in_node);
 344                 if (blkaddr == NULL_ADDR) {
 345                         dn->data_blkaddr = NEW_ADDR;
 346                         __set_data_blkaddr(dn);
 347                         count--;
 348                 }
 349         }
 350
 351         if (set_page_dirty(dn->node_page))
 352                 dn->node_changed = true;
 353         return 0;
 354 }
 355
 356 /* Should keep dn->ofs_in_node unchanged */
 357 int reserve_new_block(struct dnode_of_data *dn)
 358 {
 359         unsigned int ofs_in_node = dn->ofs_in_node;
 360         int ret;
 361
 362         ret = reserve_new_blocks(dn, 1);
 363         dn->ofs_in_node = ofs_in_node;
 364         return ret;
 365 }
 366
 367 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 368 {
 369         bool need_put = dn->inode_page ? false : true;
 370         int err;
 371
 372         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 373         if (err)
 374                 return err;
 375
 376         if (dn->data_blkaddr == NULL_ADDR)
 377                 err = reserve_new_block(dn);
 378         if (err || need_put)
 379                 f2fs_put_dnode(dn);
 380         return err;
 381 }
 382
 383 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 384 {
 385         struct extent_info ei;
 386         struct inode *inode = dn->inode;
 387
 388         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 389                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 390                 return 0;
 391         }
 392
 393         return f2fs_reserve_block(dn, index);
 394 }
 395
 396 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 397                                                 int rw, bool for_write)
 398 {
 399         struct address_space *mapping = inode->i_mapping;
 400         struct dnode_of_data dn;
 401         struct page *page;
 402         struct extent_info ei;
 403         int err;
 404         struct f2fs_io_info fio = {
 405                 .sbi = F2FS_I_SB(inode),
 406                 .type = DATA,
 407                 .rw = rw,
 408                 .encrypted_page = NULL,
 409         };
 410
 411         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 412                 return read_mapping_page(mapping, index, NULL);
 413
 414         page = f2fs_grab_cache_page(mapping, index, for_write);
 415         if (!page)
 416                 return ERR_PTR(-ENOMEM);
 417
 418         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 419                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 420                 goto got_it;
 421         }
 422
 423         set_new_dnode(&dn, inode, NULL, NULL, 0);
 424         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 425         if (err)
 426                 goto put_err;
 427         f2fs_put_dnode(&dn);
 428
 429         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 430                 err = -ENOENT;
 431                 goto put_err;
 432         }
 433 got_it:
 434         if (PageUptodate(page)) {
 435                 unlock_page(page);
 436                 return page;
 437         }
 438
 439         /*
 440          * A new dentry page is allocated but not able to be written, since its
 441          * new inode page couldn't be allocated due to -ENOSPC.
 442          * In such the case, its blkaddr can be remained as NEW_ADDR.
 443          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 444          */
 445         if (dn.data_blkaddr == NEW_ADDR) {
 446                 zero_user_segment(page, 0, PAGE_SIZE);
 447                 if (!PageUptodate(page))
 448                         SetPageUptodate(page);
 449                 unlock_page(page);
 450                 return page;
 451         }
 452
 453         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 454         fio.page = page;
 455         err = f2fs_submit_page_bio(&fio);
 456         if (err)
 457                 goto put_err;
 458         return page;
 459
 460 put_err:
 461         f2fs_put_page(page, 1);
 462         return ERR_PTR(err);
 463 }
 464
 465 struct page *find_data_page(struct inode *inode, pgoff_t index)
 466 {
 467         struct address_space *mapping = inode->i_mapping;
 468         struct page *page;
 469
 470         page = find_get_page(mapping, index);
 471         if (page && PageUptodate(page))
 472                 return page;
 473         f2fs_put_page(page, 0);
 474
 475         page = get_read_data_page(inode, index, READ_SYNC, false);
 476         if (IS_ERR(page))
 477                 return page;
 478
 479         if (PageUptodate(page))
 480                 return page;
 481
 482         wait_on_page_locked(page);
 483         if (unlikely(!PageUptodate(page))) {
 484                 f2fs_put_page(page, 0);
 485                 return ERR_PTR(-EIO);
 486         }
 487         return page;
 488 }
 489
 490 /*
 491  * If it tries to access a hole, return an error.
 492  * Because, the callers, functions in dir.c and GC, should be able to know
 493  * whether this page exists or not.
 494  */
 495 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 496                                                         bool for_write)
 497 {
 498         struct address_space *mapping = inode->i_mapping;
 499         struct page *page;
 500 repeat:
 501         page = get_read_data_page(inode, index, READ_SYNC, for_write);
 502         if (IS_ERR(page))
 503                 return page;
 504
 505         /* wait for read completion */
 506         lock_page(page);
 507         if (unlikely(page->mapping != mapping)) {
 508                 f2fs_put_page(page, 1);
 509                 goto repeat;
 510         }
 511         if (unlikely(!PageUptodate(page))) {
 512                 f2fs_put_page(page, 1);
 513                 return ERR_PTR(-EIO);
 514         }
 515         return page;
 516 }
 517
 518 /*
 519  * Caller ensures that this data page is never allocated.
 520  * A new zero-filled data page is allocated in the page cache.
 521  *
 522  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 523  * f2fs_unlock_op().
 524  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 525  * ipage should be released by this function.
 526  */
 527 struct page *get_new_data_page(struct inode *inode,
 528                 struct page *ipage, pgoff_t index, bool new_i_size)
 529 {
 530         struct address_space *mapping = inode->i_mapping;
 531         struct page *page;
 532         struct dnode_of_data dn;
 533         int err;
 534
 535         page = f2fs_grab_cache_page(mapping, index, true);
 536         if (!page) {
 537                 /*
 538                  * before exiting, we should make sure ipage will be released
 539                  * if any error occur.
 540                  */
 541                 f2fs_put_page(ipage, 1);
 542                 return ERR_PTR(-ENOMEM);
 543         }
 544
 545         set_new_dnode(&dn, inode, ipage, NULL, 0);
 546         err = f2fs_reserve_block(&dn, index);
 547         if (err) {
 548                 f2fs_put_page(page, 1);
 549                 return ERR_PTR(err);
 550         }
 551         if (!ipage)
 552                 f2fs_put_dnode(&dn);
 553
 554         if (PageUptodate(page))
 555                 goto got_it;
 556
 557         if (dn.data_blkaddr == NEW_ADDR) {
 558                 zero_user_segment(page, 0, PAGE_SIZE);
 559                 if (!PageUptodate(page))
 560                         SetPageUptodate(page);
 561         } else {
 562                 f2fs_put_page(page, 1);
 563
 564                 /* if ipage exists, blkaddr should be NEW_ADDR */
 565                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 566                 page = get_lock_data_page(inode, index, true);
 567                 if (IS_ERR(page))
 568                         return page;
 569         }
 570 got_it:
 571         if (new_i_size && i_size_read(inode) <
 572                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 573                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 574         return page;
 575 }
 576
 577 static int __allocate_data_block(struct dnode_of_data *dn)
 578 {
 579         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 580         struct f2fs_summary sum;
 581         struct node_info ni;
 582         int seg = CURSEG_WARM_DATA;
 583         pgoff_t fofs;
 584         blkcnt_t count = 1;
 585
 586         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 587                 return -EPERM;
 588
 589         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 590         if (dn->data_blkaddr == NEW_ADDR)
 591                 goto alloc;
 592
 593         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 594                 return -ENOSPC;
 595
 596 alloc:
 597         get_node_info(sbi, dn->nid, &ni);
 598         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 599
 600         if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
 601                 seg = CURSEG_DIRECT_IO;
 602
 603         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 604                                                                 &sum, seg);
 605         set_data_blkaddr(dn);
 606
 607         /* update i_size */
 608         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 609                                                         dn->ofs_in_node;
 610         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 611                 f2fs_i_size_write(dn->inode,
 612                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 613         return 0;
 614 }
 615
 616 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 617 {
 618         struct inode *inode = file_inode(iocb->ki_filp);
 619         struct f2fs_map_blocks map;
 620         ssize_t ret = 0;
 621
 622         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 623         map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
 624         map.m_next_pgofs = NULL;
 625
 626         if (f2fs_encrypted_inode(inode))
 627                 return 0;
 628
 629         if (iocb->ki_flags & IOCB_DIRECT) {
 630                 ret = f2fs_convert_inline_inode(inode);
 631                 if (ret)
 632                         return ret;
 633                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
 634         }
 635         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 636                 ret = f2fs_convert_inline_inode(inode);
 637                 if (ret)
 638                         return ret;
 639         }
 640         if (!f2fs_has_inline_data(inode))
 641                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 642         return ret;
 643 }
 644
 645 /*
 646  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 647  * f2fs_map_blocks structure.
 648  * If original data blocks are allocated, then give them to blockdev.
 649  * Otherwise,
 650  *     a. preallocate requested block addresses
 651  *     b. do not use extent cache for better performance
 652  *     c. give the block addresses to blockdev
 653  */
 654 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 655                                                 int create, int flag)
 656 {
 657         unsigned int maxblocks = map->m_len;
 658         struct dnode_of_data dn;
 659         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 660         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 661         pgoff_t pgofs, end_offset, end;
 662         int err = 0, ofs = 1;
 663         unsigned int ofs_in_node, last_ofs_in_node;
 664         blkcnt_t prealloc;
 665         struct extent_info ei;
 666         bool allocated = false;
 667         block_t blkaddr;
 668
 669         map->m_len = 0;
 670         map->m_flags = 0;
 671
 672         /* it only supports block size == page size */
 673         pgofs = (pgoff_t)map->m_lblk;
 674         end = pgofs + maxblocks;
 675
 676         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 677                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 678                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 679                 map->m_flags = F2FS_MAP_MAPPED;
 680                 goto out;
 681         }
 682
 683 next_dnode:
 684         if (create)
 685                 f2fs_lock_op(sbi);
 686
 687         /* When reading holes, we need its node page */
 688         set_new_dnode(&dn, inode, NULL, NULL, 0);
 689         err = get_dnode_of_data(&dn, pgofs, mode);
 690         if (err) {
 691                 if (flag == F2FS_GET_BLOCK_BMAP)
 692                         map->m_pblk = 0;
 693                 if (err == -ENOENT) {
 694                         err = 0;
 695                         if (map->m_next_pgofs)
 696                                 *map->m_next_pgofs =
 697                                         get_next_page_offset(&dn, pgofs);
 698                 }
 699                 goto unlock_out;
 700         }
 701
 702         prealloc = 0;
 703         ofs_in_node = dn.ofs_in_node;
 704         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 705
 706 next_block:
 707         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 708
 709         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 710                 if (create) {
 711                         if (unlikely(f2fs_cp_error(sbi))) {
 712                                 err = -EIO;
 713                                 goto sync_out;
 714                         }
 715                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 716                                 if (blkaddr == NULL_ADDR) {
 717                                         prealloc++;
 718                                         last_ofs_in_node = dn.ofs_in_node;
 719                                 }
 720                         } else {
 721                                 err = __allocate_data_block(&dn);
 722                                 if (!err) {
 723                                         set_inode_flag(inode, FI_APPEND_WRITE);
 724                                         allocated = true;
 725                                 }
 726                         }
 727                         if (err)
 728                                 goto sync_out;
 729                         map->m_flags = F2FS_MAP_NEW;
 730                         blkaddr = dn.data_blkaddr;
 731                 } else {
 732                         if (flag == F2FS_GET_BLOCK_BMAP) {
 733                                 map->m_pblk = 0;
 734                                 goto sync_out;
 735                         }
 736                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 737                                                 blkaddr == NULL_ADDR) {
 738                                 if (map->m_next_pgofs)
 739                                         *map->m_next_pgofs = pgofs + 1;
 740                         }
 741                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 742                                                 blkaddr != NEW_ADDR)
 743                                 goto sync_out;
 744                 }
 745         }
 746
 747         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 748                 goto skip;
 749
 750         if (map->m_len == 0) {
 751                 /* preallocated unwritten block should be mapped for fiemap. */
 752                 if (blkaddr == NEW_ADDR)
 753                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 754                 map->m_flags |= F2FS_MAP_MAPPED;
 755
 756                 map->m_pblk = blkaddr;
 757                 map->m_len = 1;
 758         } else if ((map->m_pblk != NEW_ADDR &&
 759                         blkaddr == (map->m_pblk + ofs)) ||
 760                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 761                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 762                 ofs++;
 763                 map->m_len++;
 764         } else {
 765                 goto sync_out;
 766         }
 767
 768 skip:
 769         dn.ofs_in_node++;
 770         pgofs++;
 771
 772         /* preallocate blocks in batch for one dnode page */
 773         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 774                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 775
 776                 dn.ofs_in_node = ofs_in_node;
 777                 err = reserve_new_blocks(&dn, prealloc);
 778                 if (err)
 779                         goto sync_out;
 780
 781                 map->m_len += dn.ofs_in_node - ofs_in_node;
 782                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 783                         err = -ENOSPC;
 784                         goto sync_out;
 785                 }
 786                 dn.ofs_in_node = end_offset;
 787         }
 788
 789         if (pgofs >= end)
 790                 goto sync_out;
 791         else if (dn.ofs_in_node < end_offset)
 792                 goto next_block;
 793
 794         f2fs_put_dnode(&dn);
 795
 796         if (create) {
 797                 f2fs_unlock_op(sbi);
 798                 f2fs_balance_fs(sbi, allocated);
 799         }
 800         allocated = false;
 801         goto next_dnode;
 802
 803 sync_out:
 804         f2fs_put_dnode(&dn);
 805 unlock_out:
 806         if (create) {
 807                 f2fs_unlock_op(sbi);
 808                 f2fs_balance_fs(sbi, allocated);
 809         }
 810 out:
 811         trace_f2fs_map_blocks(inode, map, err);
 812         return err;
 813 }
 814
 815 static int __get_data_block(struct inode *inode, sector_t iblock,
 816                         struct buffer_head *bh, int create, int flag,
 817                         pgoff_t *next_pgofs)
 818 {
 819         struct f2fs_map_blocks map;
 820         int ret;
 821
 822         map.m_lblk = iblock;
 823         map.m_len = bh->b_size >> inode->i_blkbits;
 824         map.m_next_pgofs = next_pgofs;
 825
 826         ret = f2fs_map_blocks(inode, &map, create, flag);
 827         if (!ret) {
 828                 map_bh(bh, inode->i_sb, map.m_pblk);
 829                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 830                 bh->b_size = map.m_len << inode->i_blkbits;
 831         }
 832         return ret;
 833 }
 834
 835 static int get_data_block(struct inode *inode, sector_t iblock,
 836                         struct buffer_head *bh_result, int create, int flag,
 837                         pgoff_t *next_pgofs)
 838 {
 839         return __get_data_block(inode, iblock, bh_result, create,
 840                                                         flag, next_pgofs);
 841 }
 842
 843 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 844                         struct buffer_head *bh_result, int create)
 845 {
 846         return __get_data_block(inode, iblock, bh_result, create,
 847                                                 F2FS_GET_BLOCK_DIO, NULL);
 848 }
 849
 850 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 851                         struct buffer_head *bh_result, int create)
 852 {
 853         /* Block number less than F2FS MAX BLOCKS */
 854         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 855                 return -EFBIG;
 856
 857         return __get_data_block(inode, iblock, bh_result, create,
 858                                                 F2FS_GET_BLOCK_BMAP, NULL);
 859 }
 860
 861 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
 862 {
 863         return (offset >> inode->i_blkbits);
 864 }
 865
 866 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
 867 {
 868         return (blk << inode->i_blkbits);
 869 }
 870
 871 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 872                 u64 start, u64 len)
 873 {
 874         struct buffer_head map_bh;
 875         sector_t start_blk, last_blk;
 876         pgoff_t next_pgofs;
 877         loff_t isize;
 878         u64 logical = 0, phys = 0, size = 0;
 879         u32 flags = 0;
 880         int ret = 0;
 881
 882         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 883         if (ret)
 884                 return ret;
 885
 886         if (f2fs_has_inline_data(inode)) {
 887                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
 888                 if (ret != -EAGAIN)
 889                         return ret;
 890         }
 891
 892         inode_lock(inode);
 893
 894         isize = i_size_read(inode);
 895         if (start >= isize)
 896                 goto out;
 897
 898         if (start + len > isize)
 899                 len = isize - start;
 900
 901         if (logical_to_blk(inode, len) == 0)
 902                 len = blk_to_logical(inode, 1);
 903
 904         start_blk = logical_to_blk(inode, start);
 905         last_blk = logical_to_blk(inode, start + len - 1);
 906
 907 next:
 908         memset(&map_bh, 0, sizeof(struct buffer_head));
 909         map_bh.b_size = len;
 910
 911         ret = get_data_block(inode, start_blk, &map_bh, 0,
 912                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
 913         if (ret)
 914                 goto out;
 915
 916         /* HOLE */
 917         if (!buffer_mapped(&map_bh)) {
 918                 start_blk = next_pgofs;
 919                 /* Go through holes util pass the EOF */
 920                 if (blk_to_logical(inode, start_blk) < isize)
 921                         goto prep_next;
 922                 /* Found a hole beyond isize means no more extents.
 923                  * Note that the premise is that filesystems don't
 924                  * punch holes beyond isize and keep size unchanged.
 925                  */
 926                 flags |= FIEMAP_EXTENT_LAST;
 927         }
 928
 929         if (size) {
 930                 if (f2fs_encrypted_inode(inode))
 931                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
 932
 933                 ret = fiemap_fill_next_extent(fieinfo, logical,
 934                                 phys, size, flags);
 935         }
 936
 937         if (start_blk > last_blk || ret)
 938                 goto out;
 939
 940         logical = blk_to_logical(inode, start_blk);
 941         phys = blk_to_logical(inode, map_bh.b_blocknr);
 942         size = map_bh.b_size;
 943         flags = 0;
 944         if (buffer_unwritten(&map_bh))
 945                 flags = FIEMAP_EXTENT_UNWRITTEN;
 946
 947         start_blk += logical_to_blk(inode, size);
 948
 949 prep_next:
 950         cond_resched();
 951         if (fatal_signal_pending(current))
 952                 ret = -EINTR;
 953         else
 954                 goto next;
 955 out:
 956         if (ret == 1)
 957                 ret = 0;
 958
 959         inode_unlock(inode);
 960         return ret;
 961 }
 962
 963 struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
 964                                                         unsigned nr_pages)
 965 {
 966         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 967         struct fscrypt_ctx *ctx = NULL;
 968         struct block_device *bdev = sbi->sb->s_bdev;
 969         struct bio *bio;
 970
 971         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
 972                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 973                 if (IS_ERR(ctx))
 974                         return ERR_CAST(ctx);
 975
 976                 /* wait the page to be moved by cleaning */
 977                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
 978         }
 979
 980         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
 981         if (!bio) {
 982                 if (ctx)
 983                         fscrypt_release_ctx(ctx);
 984                 return ERR_PTR(-ENOMEM);
 985         }
 986         bio->bi_bdev = bdev;
 987         bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
 988         bio->bi_end_io = f2fs_read_end_io;
 989         bio->bi_private = ctx;
 990
 991         return bio;
 992 }
 993
 994 /*
 995  * This function was originally taken from fs/mpage.c, and customized for f2fs.
 996  * Major change was from block_size == page_size in f2fs by default.
 997  */
 998 static int f2fs_mpage_readpages(struct address_space *mapping,
 999                         struct list_head *pages, struct page *page,
1000                         unsigned nr_pages)
1001 {
1002         struct bio *bio = NULL;
1003         unsigned page_idx;
1004         sector_t last_block_in_bio = 0;
1005         struct inode *inode = mapping->host;
1006         const unsigned blkbits = inode->i_blkbits;
1007         const unsigned blocksize = 1 << blkbits;
1008         sector_t block_in_file;
1009         sector_t last_block;
1010         sector_t last_block_in_file;
1011         sector_t block_nr;
1012         struct f2fs_map_blocks map;
1013
1014         map.m_pblk = 0;
1015         map.m_lblk = 0;
1016         map.m_len = 0;
1017         map.m_flags = 0;
1018         map.m_next_pgofs = NULL;
1019
1020         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1021
1022                 prefetchw(&page->flags);
1023                 if (pages) {
1024                         page = list_entry(pages->prev, struct page, lru);
1025                         list_del(&page->lru);
1026                         if (add_to_page_cache_lru(page, mapping,
1027                                                   page->index, GFP_KERNEL))
1028                                 goto next_page;
1029                 }
1030
1031                 block_in_file = (sector_t)page->index;
1032                 last_block = block_in_file + nr_pages;
1033                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1034                                                                 blkbits;
1035                 if (last_block > last_block_in_file)
1036                         last_block = last_block_in_file;
1037
1038                 /*
1039                  * Map blocks using the previous result first.
1040                  */
1041                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1042                                 block_in_file > map.m_lblk &&
1043                                 block_in_file < (map.m_lblk + map.m_len))
1044                         goto got_it;
1045
1046                 /*
1047                  * Then do more f2fs_map_blocks() calls until we are
1048                  * done with this page.
1049                  */
1050                 map.m_flags = 0;
1051
1052                 if (block_in_file < last_block) {
1053                         map.m_lblk = block_in_file;
1054                         map.m_len = last_block - block_in_file;
1055
1056                         if (f2fs_map_blocks(inode, &map, 0,
1057                                                 F2FS_GET_BLOCK_READ))
1058                                 goto set_error_page;
1059                 }
1060 got_it:
1061                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1062                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1063                         SetPageMappedToDisk(page);
1064
1065                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1066                                 SetPageUptodate(page);
1067                                 goto confused;
1068                         }
1069                 } else {
1070                         zero_user_segment(page, 0, PAGE_SIZE);
1071                         if (!PageUptodate(page))
1072                                 SetPageUptodate(page);
1073                         unlock_page(page);
1074                         goto next_page;
1075                 }
1076
1077                 /*
1078                  * This page will go to BIO.  Do we need to send this
1079                  * BIO off first?
1080                  */
1081                 if (bio && (last_block_in_bio != block_nr - 1)) {
1082 submit_and_realloc:
1083                         __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1084                         bio = NULL;
1085                 }
1086                 if (bio == NULL) {
1087                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1088                         if (IS_ERR(bio))
1089                                 goto set_error_page;
1090                 }
1091
1092                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1093                         goto submit_and_realloc;
1094
1095                 last_block_in_bio = block_nr;
1096                 goto next_page;
1097 set_error_page:
1098                 SetPageError(page);
1099                 zero_user_segment(page, 0, PAGE_SIZE);
1100                 unlock_page(page);
1101                 goto next_page;
1102 confused:
1103                 if (bio) {
1104                         __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1105                         bio = NULL;
1106                 }
1107                 unlock_page(page);
1108 next_page:
1109                 if (pages)
1110                         put_page(page);
1111         }
1112         BUG_ON(pages && !list_empty(pages));
1113         if (bio)
1114                 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1115         return 0;
1116 }
1117
1118 static int f2fs_read_data_page(struct file *file, struct page *page)
1119 {
1120         struct inode *inode = page->mapping->host;
1121         int ret = -EAGAIN;
1122
1123         trace_f2fs_readpage(page, DATA);
1124
1125         /* If the file has inline data, try to read it directly */
1126         if (f2fs_has_inline_data(inode))
1127                 ret = f2fs_read_inline_data(inode, page);
1128         if (ret == -EAGAIN)
1129                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1130         return ret;
1131 }
1132
1133 static int f2fs_read_data_pages(struct file *file,
1134                         struct address_space *mapping,
1135                         struct list_head *pages, unsigned nr_pages)
1136 {
1137         struct inode *inode = file->f_mapping->host;
1138         struct page *page = list_entry(pages->prev, struct page, lru);
1139
1140         trace_f2fs_readpages(inode, page, nr_pages);
1141
1142         /* If the file has inline data, skip readpages */
1143         if (f2fs_has_inline_data(inode))
1144                 return 0;
1145
1146         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1147 }
1148
1149 int do_write_data_page(struct f2fs_io_info *fio)
1150 {
1151         struct page *page = fio->page;
1152         struct inode *inode = page->mapping->host;
1153         struct dnode_of_data dn;
1154         int err = 0;
1155
1156         set_new_dnode(&dn, inode, NULL, NULL, 0);
1157         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1158         if (err)
1159                 return err;
1160
1161         fio->old_blkaddr = dn.data_blkaddr;
1162
1163         /* This page is already truncated */
1164         if (fio->old_blkaddr == NULL_ADDR) {
1165                 ClearPageUptodate(page);
1166                 goto out_writepage;
1167         }
1168
1169         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1170                 gfp_t gfp_flags = GFP_NOFS;
1171
1172                 /* wait for GCed encrypted page writeback */
1173                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1174                                                         fio->old_blkaddr);
1175 retry_encrypt:
1176                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1177                                                                 gfp_flags);
1178                 if (IS_ERR(fio->encrypted_page)) {
1179                         err = PTR_ERR(fio->encrypted_page);
1180                         if (err == -ENOMEM) {
1181                                 /* flush pending ios and wait for a while */
1182                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1183                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1184                                 gfp_flags |= __GFP_NOFAIL;
1185                                 err = 0;
1186                                 goto retry_encrypt;
1187                         }
1188                         goto out_writepage;
1189                 }
1190         }
1191
1192         set_page_writeback(page);
1193
1194         /*
1195          * If current allocation needs SSR,
1196          * it had better in-place writes for updated data.
1197          */
1198         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1199                         !is_cold_data(page) &&
1200                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1201                         need_inplace_update(inode))) {
1202                 rewrite_data_page(fio);
1203                 set_inode_flag(inode, FI_UPDATE_WRITE);
1204                 trace_f2fs_do_write_data_page(page, IPU);
1205         } else {
1206                 write_data_page(&dn, fio);
1207                 trace_f2fs_do_write_data_page(page, OPU);
1208                 set_inode_flag(inode, FI_APPEND_WRITE);
1209                 if (page->index == 0)
1210                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1211         }
1212 out_writepage:
1213         f2fs_put_dnode(&dn);
1214         return err;
1215 }
1216
1217 static int f2fs_write_data_page(struct page *page,
1218                                         struct writeback_control *wbc)
1219 {
1220         struct inode *inode = page->mapping->host;
1221         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1222         loff_t i_size = i_size_read(inode);
1223         const pgoff_t end_index = ((unsigned long long) i_size)
1224                                                         >> PAGE_SHIFT;
1225         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1226         unsigned offset = 0;
1227         bool need_balance_fs = false;
1228         int err = 0;
1229         struct f2fs_io_info fio = {
1230                 .sbi = sbi,
1231                 .type = DATA,
1232                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1233                 .page = page,
1234                 .encrypted_page = NULL,
1235         };
1236
1237         trace_f2fs_writepage(page, DATA);
1238
1239         if (page->index < end_index)
1240                 goto write;
1241
1242         /*
1243          * If the offset is out-of-range of file size,
1244          * this page does not have to be written to disk.
1245          */
1246         offset = i_size & (PAGE_SIZE - 1);
1247         if ((page->index >= end_index + 1) || !offset)
1248                 goto out;
1249
1250         zero_user_segment(page, offset, PAGE_SIZE);
1251 write:
1252         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1253                 goto redirty_out;
1254         if (f2fs_is_drop_cache(inode))
1255                 goto out;
1256         /* we should not write 0'th page having journal header */
1257         if (f2fs_is_volatile_file(inode) && (!page->index ||
1258                         (!wbc->for_reclaim &&
1259                         available_free_memory(sbi, BASE_CHECK))))
1260                 goto redirty_out;
1261
1262         /* we should bypass data pages to proceed the kworkder jobs */
1263         if (unlikely(f2fs_cp_error(sbi))) {
1264                 mapping_set_error(page->mapping, -EIO);
1265                 goto out;
1266         }
1267
1268         /* Dentry blocks are controlled by checkpoint */
1269         if (S_ISDIR(inode->i_mode)) {
1270                 err = do_write_data_page(&fio);
1271                 goto done;
1272         }
1273
1274         if (!wbc->for_reclaim)
1275                 need_balance_fs = true;
1276         else if (has_not_enough_free_secs(sbi, 0))
1277                 goto redirty_out;
1278
1279         err = -EAGAIN;
1280         f2fs_lock_op(sbi);
1281         if (f2fs_has_inline_data(inode))
1282                 err = f2fs_write_inline_data(inode, page);
1283         if (err == -EAGAIN)
1284                 err = do_write_data_page(&fio);
1285         if (F2FS_I(inode)->last_disk_size < psize)
1286                 F2FS_I(inode)->last_disk_size = psize;
1287         f2fs_unlock_op(sbi);
1288 done:
1289         if (err && err != -ENOENT)
1290                 goto redirty_out;
1291
1292         clear_cold_data(page);
1293 out:
1294         inode_dec_dirty_pages(inode);
1295         if (err)
1296                 ClearPageUptodate(page);
1297
1298         if (wbc->for_reclaim) {
1299                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1300                 remove_dirty_inode(inode);
1301         }
1302
1303         unlock_page(page);
1304         f2fs_balance_fs(sbi, need_balance_fs);
1305
1306         if (unlikely(f2fs_cp_error(sbi)))
1307                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1308
1309         return 0;
1310
1311 redirty_out:
1312         redirty_page_for_writepage(wbc, page);
1313         unlock_page(page);
1314         return err;
1315 }
1316
1317 /*
1318  * This function was copied from write_cche_pages from mm/page-writeback.c.
1319  * The major change is making write step of cold data page separately from
1320  * warm/hot data page.
1321  */
1322 static int f2fs_write_cache_pages(struct address_space *mapping,
1323                                         struct writeback_control *wbc)
1324 {
1325         int ret = 0;
1326         int done = 0;
1327         struct pagevec pvec;
1328         int nr_pages;
1329         pgoff_t uninitialized_var(writeback_index);
1330         pgoff_t index;
1331         pgoff_t end;            /* Inclusive */
1332         pgoff_t done_index;
1333         int cycled;
1334         int range_whole = 0;
1335         int tag;
1336
1337         pagevec_init(&pvec, 0);
1338
1339         if (wbc->range_cyclic) {
1340                 writeback_index = mapping->writeback_index; /* prev offset */
1341                 index = writeback_index;
1342                 if (index == 0)
1343                         cycled = 1;
1344                 else
1345                         cycled = 0;
1346                 end = -1;
1347         } else {
1348                 index = wbc->range_start >> PAGE_SHIFT;
1349                 end = wbc->range_end >> PAGE_SHIFT;
1350                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1351                         range_whole = 1;
1352                 cycled = 1; /* ignore range_cyclic tests */
1353         }
1354         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1355                 tag = PAGECACHE_TAG_TOWRITE;
1356         else
1357                 tag = PAGECACHE_TAG_DIRTY;
1358 retry:
1359         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1360                 tag_pages_for_writeback(mapping, index, end);
1361         done_index = index;
1362         while (!done && (index <= end)) {
1363                 int i;
1364
1365                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1366                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1367                 if (nr_pages == 0)
1368                         break;
1369
1370                 for (i = 0; i < nr_pages; i++) {
1371                         struct page *page = pvec.pages[i];
1372
1373                         if (page->index > end) {
1374                                 done = 1;
1375                                 break;
1376                         }
1377
1378                         done_index = page->index;
1379
1380                         lock_page(page);
1381
1382                         if (unlikely(page->mapping != mapping)) {
1383 continue_unlock:
1384                                 unlock_page(page);
1385                                 continue;
1386                         }
1387
1388                         if (!PageDirty(page)) {
1389                                 /* someone wrote it for us */
1390                                 goto continue_unlock;
1391                         }
1392
1393                         if (PageWriteback(page)) {
1394                                 if (wbc->sync_mode != WB_SYNC_NONE)
1395                                         f2fs_wait_on_page_writeback(page,
1396                                                                 DATA, true);
1397                                 else
1398                                         goto continue_unlock;
1399                         }
1400
1401                         BUG_ON(PageWriteback(page));
1402                         if (!clear_page_dirty_for_io(page))
1403                                 goto continue_unlock;
1404
1405                         ret = mapping->a_ops->writepage(page, wbc);
1406                         if (unlikely(ret)) {
1407                                 done_index = page->index + 1;
1408                                 done = 1;
1409                                 break;
1410                         }
1411
1412                         if (--wbc->nr_to_write <= 0 &&
1413                             wbc->sync_mode == WB_SYNC_NONE) {
1414                                 done = 1;
1415                                 break;
1416                         }
1417                 }
1418                 pagevec_release(&pvec);
1419                 cond_resched();
1420         }
1421
1422         if (!cycled && !done) {
1423                 cycled = 1;
1424                 index = 0;
1425                 end = writeback_index - 1;
1426                 goto retry;
1427         }
1428         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1429                 mapping->writeback_index = done_index;
1430
1431         return ret;
1432 }
1433
1434 static int f2fs_write_data_pages(struct address_space *mapping,
1435                             struct writeback_control *wbc)
1436 {
1437         struct inode *inode = mapping->host;
1438         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1439         int ret;
1440
1441         /* deal with chardevs and other special file */
1442         if (!mapping->a_ops->writepage)
1443                 return 0;
1444
1445         /* skip writing if there is no dirty page in this inode */
1446         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1447                 return 0;
1448
1449         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1450                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1451                         available_free_memory(sbi, DIRTY_DENTS))
1452                 goto skip_write;
1453
1454         /* skip writing during file defragment */
1455         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1456                 goto skip_write;
1457
1458         /* during POR, we don't need to trigger writepage at all. */
1459         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1460                 goto skip_write;
1461
1462         trace_f2fs_writepages(mapping->host, wbc, DATA);
1463
1464         ret = f2fs_write_cache_pages(mapping, wbc);
1465         /*
1466          * if some pages were truncated, we cannot guarantee its mapping->host
1467          * to detect pending bios.
1468          */
1469         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1470
1471         remove_dirty_inode(inode);
1472         return ret;
1473
1474 skip_write:
1475         wbc->pages_skipped += get_dirty_pages(inode);
1476         trace_f2fs_writepages(mapping->host, wbc, DATA);
1477         return 0;
1478 }
1479
1480 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1481 {
1482         struct inode *inode = mapping->host;
1483         loff_t i_size = i_size_read(inode);
1484
1485         if (to > i_size) {
1486                 truncate_pagecache(inode, i_size);
1487                 truncate_blocks(inode, i_size, true);
1488         }
1489 }
1490
1491 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1492                         struct page *page, loff_t pos, unsigned len,
1493                         block_t *blk_addr, bool *node_changed)
1494 {
1495         struct inode *inode = page->mapping->host;
1496         pgoff_t index = page->index;
1497         struct dnode_of_data dn;
1498         struct page *ipage;
1499         bool locked = false;
1500         struct extent_info ei;
1501         int err = 0;
1502
1503         /*
1504          * we already allocated all the blocks, so we don't need to get
1505          * the block addresses when there is no need to fill the page.
1506          */
1507         if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1508                                         len == PAGE_SIZE)
1509                 return 0;
1510
1511         if (f2fs_has_inline_data(inode) ||
1512                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1513                 f2fs_lock_op(sbi);
1514                 locked = true;
1515         }
1516 restart:
1517         /* check inline_data */
1518         ipage = get_node_page(sbi, inode->i_ino);
1519         if (IS_ERR(ipage)) {
1520                 err = PTR_ERR(ipage);
1521                 goto unlock_out;
1522         }
1523
1524         set_new_dnode(&dn, inode, ipage, ipage, 0);
1525
1526         if (f2fs_has_inline_data(inode)) {
1527                 if (pos + len <= MAX_INLINE_DATA) {
1528                         read_inline_data(page, ipage);
1529                         set_inode_flag(inode, FI_DATA_EXIST);
1530                         if (inode->i_nlink)
1531                                 set_inline_node(ipage);
1532                 } else {
1533                         err = f2fs_convert_inline_page(&dn, page);
1534                         if (err)
1535                                 goto out;
1536                         if (dn.data_blkaddr == NULL_ADDR)
1537                                 err = f2fs_get_block(&dn, index);
1538                 }
1539         } else if (locked) {
1540                 err = f2fs_get_block(&dn, index);
1541         } else {
1542                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1543                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1544                 } else {
1545                         /* hole case */
1546                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1547                         if (err || dn.data_blkaddr == NULL_ADDR) {
1548                                 f2fs_put_dnode(&dn);
1549                                 f2fs_lock_op(sbi);
1550                                 locked = true;
1551                                 goto restart;
1552                         }
1553                 }
1554         }
1555
1556         /* convert_inline_page can make node_changed */
1557         *blk_addr = dn.data_blkaddr;
1558         *node_changed = dn.node_changed;
1559 out:
1560         f2fs_put_dnode(&dn);
1561 unlock_out:
1562         if (locked)
1563                 f2fs_unlock_op(sbi);
1564         return err;
1565 }
1566
1567 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1568                 loff_t pos, unsigned len, unsigned flags,
1569                 struct page **pagep, void **fsdata)
1570 {
1571         struct inode *inode = mapping->host;
1572         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1573         struct page *page = NULL;
1574         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1575         bool need_balance = false;
1576         block_t blkaddr = NULL_ADDR;
1577         int err = 0;
1578
1579         trace_f2fs_write_begin(inode, pos, len, flags);
1580
1581         /*
1582          * We should check this at this moment to avoid deadlock on inode page
1583          * and #0 page. The locking rule for inline_data conversion should be:
1584          * lock_page(page #0) -> lock_page(inode_page)
1585          */
1586         if (index != 0) {
1587                 err = f2fs_convert_inline_inode(inode);
1588                 if (err)
1589                         goto fail;
1590         }
1591 repeat:
1592         page = grab_cache_page_write_begin(mapping, index, flags);
1593         if (!page) {
1594                 err = -ENOMEM;
1595                 goto fail;
1596         }
1597
1598         *pagep = page;
1599
1600         err = prepare_write_begin(sbi, page, pos, len,
1601                                         &blkaddr, &need_balance);
1602         if (err)
1603                 goto fail;
1604
1605         if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1606                 unlock_page(page);
1607                 f2fs_balance_fs(sbi, true);
1608                 lock_page(page);
1609                 if (page->mapping != mapping) {
1610                         /* The page got truncated from under us */
1611                         f2fs_put_page(page, 1);
1612                         goto repeat;
1613                 }
1614         }
1615
1616         f2fs_wait_on_page_writeback(page, DATA, false);
1617
1618         /* wait for GCed encrypted page writeback */
1619         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1620                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1621
1622         if (len == PAGE_SIZE)
1623                 goto out_update;
1624         if (PageUptodate(page))
1625                 goto out_clear;
1626
1627         if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1628                 unsigned start = pos & (PAGE_SIZE - 1);
1629                 unsigned end = start + len;
1630
1631                 /* Reading beyond i_size is simple: memset to zero */
1632                 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1633                 goto out_update;
1634         }
1635
1636         if (blkaddr == NEW_ADDR) {
1637                 zero_user_segment(page, 0, PAGE_SIZE);
1638         } else {
1639                 struct bio *bio;
1640
1641                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1642                 if (IS_ERR(bio)) {
1643                         err = PTR_ERR(bio);
1644                         goto fail;
1645                 }
1646
1647                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1648                         bio_put(bio);
1649                         err = -EFAULT;
1650                         goto fail;
1651                 }
1652
1653                 __submit_bio(sbi, READ_SYNC, bio, DATA);
1654
1655                 lock_page(page);
1656                 if (unlikely(page->mapping != mapping)) {
1657                         f2fs_put_page(page, 1);
1658                         goto repeat;
1659                 }
1660                 if (unlikely(!PageUptodate(page))) {
1661                         err = -EIO;
1662                         goto fail;
1663                 }
1664         }
1665 out_update:
1666         if (!PageUptodate(page))
1667                 SetPageUptodate(page);
1668 out_clear:
1669         clear_cold_data(page);
1670         return 0;
1671
1672 fail:
1673         f2fs_put_page(page, 1);
1674         f2fs_write_failed(mapping, pos + len);
1675         return err;
1676 }
1677
1678 static int f2fs_write_end(struct file *file,
1679                         struct address_space *mapping,
1680                         loff_t pos, unsigned len, unsigned copied,
1681                         struct page *page, void *fsdata)
1682 {
1683         struct inode *inode = page->mapping->host;
1684
1685         trace_f2fs_write_end(inode, pos, len, copied);
1686
1687         set_page_dirty(page);
1688
1689         if (pos + copied > i_size_read(inode))
1690                 f2fs_i_size_write(inode, pos + copied);
1691
1692         f2fs_put_page(page, 1);
1693         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1694         return copied;
1695 }
1696
1697 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1698                            loff_t offset)
1699 {
1700         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1701
1702         if (offset & blocksize_mask)
1703                 return -EINVAL;
1704
1705         if (iov_iter_alignment(iter) & blocksize_mask)
1706                 return -EINVAL;
1707
1708         return 0;
1709 }
1710
1711 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1712 {
1713         struct address_space *mapping = iocb->ki_filp->f_mapping;
1714         struct inode *inode = mapping->host;
1715         size_t count = iov_iter_count(iter);
1716         loff_t offset = iocb->ki_pos;
1717         int err;
1718
1719         err = check_direct_IO(inode, iter, offset);
1720         if (err)
1721                 return err;
1722
1723         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1724                 return 0;
1725         if (test_opt(F2FS_I_SB(inode), LFS))
1726                 return 0;
1727
1728         trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1729
1730         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1731         if (iov_iter_rw(iter) == WRITE) {
1732                 if (err > 0)
1733                         set_inode_flag(inode, FI_UPDATE_WRITE);
1734                 else if (err < 0)
1735                         f2fs_write_failed(mapping, offset + count);
1736         }
1737
1738         trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1739
1740         return err;
1741 }
1742
1743 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1744                                                         unsigned int length)
1745 {
1746         struct inode *inode = page->mapping->host;
1747         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1748
1749         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1750                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1751                 return;
1752
1753         if (PageDirty(page)) {
1754                 if (inode->i_ino == F2FS_META_INO(sbi))
1755                         dec_page_count(sbi, F2FS_DIRTY_META);
1756                 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1757                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1758                 else
1759                         inode_dec_dirty_pages(inode);
1760         }
1761
1762         /* This is atomic written page, keep Private */
1763         if (IS_ATOMIC_WRITTEN_PAGE(page))
1764                 return;
1765
1766         set_page_private(page, 0);
1767         ClearPagePrivate(page);
1768 }
1769
1770 int f2fs_release_page(struct page *page, gfp_t wait)
1771 {
1772         /* If this is dirty page, keep PagePrivate */
1773         if (PageDirty(page))
1774                 return 0;
1775
1776         /* This is atomic written page, keep Private */
1777         if (IS_ATOMIC_WRITTEN_PAGE(page))
1778                 return 0;
1779
1780         set_page_private(page, 0);
1781         ClearPagePrivate(page);
1782         return 1;
1783 }
1784
1785 /*
1786  * This was copied from __set_page_dirty_buffers which gives higher performance
1787  * in very high speed storages. (e.g., pmem)
1788  */
1789 void f2fs_set_page_dirty_nobuffers(struct page *page)
1790 {
1791         struct address_space *mapping = page->mapping;
1792         unsigned long flags;
1793
1794         if (unlikely(!mapping))
1795                 return;
1796
1797         spin_lock(&mapping->private_lock);
1798         lock_page_memcg(page);
1799         SetPageDirty(page);
1800         spin_unlock(&mapping->private_lock);
1801
1802         spin_lock_irqsave(&mapping->tree_lock, flags);
1803         WARN_ON_ONCE(!PageUptodate(page));
1804         account_page_dirtied(page, mapping);
1805         radix_tree_tag_set(&mapping->page_tree,
1806                         page_index(page), PAGECACHE_TAG_DIRTY);
1807         spin_unlock_irqrestore(&mapping->tree_lock, flags);
1808         unlock_page_memcg(page);
1809
1810         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1811         return;
1812 }
1813
1814 static int f2fs_set_data_page_dirty(struct page *page)
1815 {
1816         struct address_space *mapping = page->mapping;
1817         struct inode *inode = mapping->host;
1818
1819         trace_f2fs_set_page_dirty(page, DATA);
1820
1821         if (!PageUptodate(page))
1822                 SetPageUptodate(page);
1823
1824         if (f2fs_is_atomic_file(inode)) {
1825                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1826                         register_inmem_page(inode, page);
1827                         return 1;
1828                 }
1829                 /*
1830                  * Previously, this page has been registered, we just
1831                  * return here.
1832                  */
1833                 return 0;
1834         }
1835
1836         if (!PageDirty(page)) {
1837                 f2fs_set_page_dirty_nobuffers(page);
1838                 update_dirty_page(inode, page);
1839                 return 1;
1840         }
1841         return 0;
1842 }
1843
1844 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1845 {
1846         struct inode *inode = mapping->host;
1847
1848         if (f2fs_has_inline_data(inode))
1849                 return 0;
1850
1851         /* make sure allocating whole blocks */
1852         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1853                 filemap_write_and_wait(mapping);
1854
1855         return generic_block_bmap(mapping, block, get_data_block_bmap);
1856 }
1857
1858 const struct address_space_operations f2fs_dblock_aops = {
1859         .readpage       = f2fs_read_data_page,
1860         .readpages      = f2fs_read_data_pages,
1861         .writepage      = f2fs_write_data_page,
1862         .writepages     = f2fs_write_data_pages,
1863         .write_begin    = f2fs_write_begin,
1864         .write_end      = f2fs_write_end,
1865         .set_page_dirty = f2fs_set_data_page_dirty,
1866         .invalidatepage = f2fs_invalidate_page,
1867         .releasepage    = f2fs_release_page,
1868         .direct_IO      = f2fs_direct_IO,
1869         .bmap           = f2fs_bmap,
1870 };