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 bool __is_cp_guaranteed(struct page *page)
  33 {
  34         struct address_space *mapping = page->mapping;
  35         struct inode *inode;
  36         struct f2fs_sb_info *sbi;
  37
  38         if (!mapping)
  39                 return false;
  40
  41         inode = mapping->host;
  42         sbi = F2FS_I_SB(inode);
  43
  44         if (inode->i_ino == F2FS_META_INO(sbi) ||
  45                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  46                         S_ISDIR(inode->i_mode) ||
  47                         is_cold_data(page))
  48                 return true;
  49         return false;
  50 }
  51
  52 static void f2fs_read_end_io(struct bio *bio)
  53 {
  54         struct bio_vec *bvec;
  55         int i;
  56
  57 #ifdef CONFIG_F2FS_FAULT_INJECTION
  58         if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) {
  59                 f2fs_show_injection_info(FAULT_IO);
  60                 bio->bi_error = -EIO;
  61         }
  62 #endif
  63
  64         if (f2fs_bio_encrypted(bio)) {
  65                 if (bio->bi_error) {
  66                         fscrypt_release_ctx(bio->bi_private);
  67                 } else {
  68                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  69                         return;
  70                 }
  71         }
  72
  73         bio_for_each_segment_all(bvec, bio, i) {
  74                 struct page *page = bvec->bv_page;
  75
  76                 if (!bio->bi_error) {
  77                         if (!PageUptodate(page))
  78                                 SetPageUptodate(page);
  79                 } else {
  80                         ClearPageUptodate(page);
  81                         SetPageError(page);
  82                 }
  83                 unlock_page(page);
  84         }
  85         bio_put(bio);
  86 }
  87
  88 static void f2fs_write_end_io(struct bio *bio)
  89 {
  90         struct f2fs_sb_info *sbi = bio->bi_private;
  91         struct bio_vec *bvec;
  92         int i;
  93
  94         bio_for_each_segment_all(bvec, bio, i) {
  95                 struct page *page = bvec->bv_page;
  96                 enum count_type type = WB_DATA_TYPE(page);
  97
  98                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
  99                         set_page_private(page, (unsigned long)NULL);
 100                         ClearPagePrivate(page);
 101                         unlock_page(page);
 102                         mempool_free(page, sbi->write_io_dummy);
 103
 104                         if (unlikely(bio->bi_error))
 105                                 f2fs_stop_checkpoint(sbi, true);
 106                         continue;
 107                 }
 108
 109                 fscrypt_pullback_bio_page(&page, true);
 110
 111                 if (unlikely(bio->bi_error)) {
 112                         mapping_set_error(page->mapping, -EIO);
 113                         f2fs_stop_checkpoint(sbi, true);
 114                 }
 115                 dec_page_count(sbi, type);
 116                 clear_cold_data(page);
 117                 end_page_writeback(page);
 118         }
 119         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 120                                 wq_has_sleeper(&sbi->cp_wait))
 121                 wake_up(&sbi->cp_wait);
 122
 123         bio_put(bio);
 124 }
 125
 126 /*
 127  * Return true, if pre_bio's bdev is same as its target device.
 128  */
 129 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 130                                 block_t blk_addr, struct bio *bio)
 131 {
 132         struct block_device *bdev = sbi->sb->s_bdev;
 133         int i;
 134
 135         for (i = 0; i < sbi->s_ndevs; i++) {
 136                 if (FDEV(i).start_blk <= blk_addr &&
 137                                         FDEV(i).end_blk >= blk_addr) {
 138                         blk_addr -= FDEV(i).start_blk;
 139                         bdev = FDEV(i).bdev;
 140                         break;
 141                 }
 142         }
 143         if (bio) {
 144                 bio->bi_bdev = bdev;
 145                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 146         }
 147         return bdev;
 148 }
 149
 150 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 151 {
 152         int i;
 153
 154         for (i = 0; i < sbi->s_ndevs; i++)
 155                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 156                         return i;
 157         return 0;
 158 }
 159
 160 static bool __same_bdev(struct f2fs_sb_info *sbi,
 161                                 block_t blk_addr, struct bio *bio)
 162 {
 163         return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
 164 }
 165
 166 /*
 167  * Low-level block read/write IO operations.
 168  */
 169 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 170                                 int npages, bool is_read)
 171 {
 172         struct bio *bio;
 173
 174         bio = f2fs_bio_alloc(npages);
 175
 176         f2fs_target_device(sbi, blk_addr, bio);
 177         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
 178         bio->bi_private = is_read ? NULL : sbi;
 179
 180         return bio;
 181 }
 182
 183 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 184                                 struct bio *bio, enum page_type type)
 185 {
 186         if (!is_read_io(bio_op(bio))) {
 187                 unsigned int start;
 188
 189                 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
 190                         current->plug && (type == DATA || type == NODE))
 191                         blk_finish_plug(current->plug);
 192
 193                 if (type != DATA && type != NODE)
 194                         goto submit_io;
 195
 196                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 197                 start %= F2FS_IO_SIZE(sbi);
 198
 199                 if (start == 0)
 200                         goto submit_io;
 201
 202                 /* fill dummy pages */
 203                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 204                         struct page *page =
 205                                 mempool_alloc(sbi->write_io_dummy,
 206                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 207                         f2fs_bug_on(sbi, !page);
 208
 209                         SetPagePrivate(page);
 210                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 211                         lock_page(page);
 212                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 213                                 f2fs_bug_on(sbi, 1);
 214                 }
 215                 /*
 216                  * In the NODE case, we lose next block address chain. So, we
 217                  * need to do checkpoint in f2fs_sync_file.
 218                  */
 219                 if (type == NODE)
 220                         set_sbi_flag(sbi, SBI_NEED_CP);
 221         }
 222 submit_io:
 223         if (is_read_io(bio_op(bio)))
 224                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 225         else
 226                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 227         submit_bio(bio);
 228 }
 229
 230 static void __submit_merged_bio(struct f2fs_bio_info *io)
 231 {
 232         struct f2fs_io_info *fio = &io->fio;
 233
 234         if (!io->bio)
 235                 return;
 236
 237         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 238
 239         if (is_read_io(fio->op))
 240                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 241         else
 242                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 243
 244         __submit_bio(io->sbi, io->bio, fio->type);
 245         io->bio = NULL;
 246 }
 247
 248 static bool __has_merged_page(struct f2fs_bio_info *io,
 249                                 struct inode *inode, nid_t ino, pgoff_t idx)
 250 {
 251         struct bio_vec *bvec;
 252         struct page *target;
 253         int i;
 254
 255         if (!io->bio)
 256                 return false;
 257
 258         if (!inode && !ino)
 259                 return true;
 260
 261         bio_for_each_segment_all(bvec, io->bio, i) {
 262
 263                 if (bvec->bv_page->mapping)
 264                         target = bvec->bv_page;
 265                 else
 266                         target = fscrypt_control_page(bvec->bv_page);
 267
 268                 if (idx != target->index)
 269                         continue;
 270
 271                 if (inode && inode == target->mapping->host)
 272                         return true;
 273                 if (ino && ino == ino_of_node(target))
 274                         return true;
 275         }
 276
 277         return false;
 278 }
 279
 280 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 281                                 nid_t ino, pgoff_t idx, enum page_type type)
 282 {
 283         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 284         struct f2fs_bio_info *io = &sbi->write_io[btype];
 285         bool ret;
 286
 287         down_read(&io->io_rwsem);
 288         ret = __has_merged_page(io, inode, ino, idx);
 289         up_read(&io->io_rwsem);
 290         return ret;
 291 }
 292
 293 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 294                                 struct inode *inode, nid_t ino, pgoff_t idx,
 295                                 enum page_type type, int rw)
 296 {
 297         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 298         struct f2fs_bio_info *io;
 299
 300         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 301
 302         down_write(&io->io_rwsem);
 303
 304         if (!__has_merged_page(io, inode, ino, idx))
 305                 goto out;
 306
 307         /* change META to META_FLUSH in the checkpoint procedure */
 308         if (type >= META_FLUSH) {
 309                 io->fio.type = META_FLUSH;
 310                 io->fio.op = REQ_OP_WRITE;
 311                 io->fio.op_flags = REQ_META | REQ_PRIO;
 312                 if (!test_opt(sbi, NOBARRIER))
 313                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 314         }
 315         __submit_merged_bio(io);
 316 out:
 317         up_write(&io->io_rwsem);
 318 }
 319
 320 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 321                                                                         int rw)
 322 {
 323         __f2fs_submit_merged_bio(sbi, NULL, 0, 0, type, rw);
 324 }
 325
 326 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 327                                 struct inode *inode, nid_t ino, pgoff_t idx,
 328                                 enum page_type type, int rw)
 329 {
 330         if (has_merged_page(sbi, inode, ino, idx, type))
 331                 __f2fs_submit_merged_bio(sbi, inode, ino, idx, type, rw);
 332 }
 333
 334 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 335 {
 336         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 337         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 338         f2fs_submit_merged_bio(sbi, META, WRITE);
 339 }
 340
 341 /*
 342  * Fill the locked page with data located in the block address.
 343  * Return unlocked page.
 344  */
 345 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 346 {
 347         struct bio *bio;
 348         struct page *page = fio->encrypted_page ?
 349                         fio->encrypted_page : fio->page;
 350
 351         trace_f2fs_submit_page_bio(page, fio);
 352         f2fs_trace_ios(fio, 0);
 353
 354         /* Allocate a new bio */
 355         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
 356
 357         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 358                 bio_put(bio);
 359                 return -EFAULT;
 360         }
 361         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 362
 363         __submit_bio(fio->sbi, bio, fio->type);
 364         return 0;
 365 }
 366
 367 int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 368 {
 369         struct f2fs_sb_info *sbi = fio->sbi;
 370         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 371         struct f2fs_bio_info *io;
 372         bool is_read = is_read_io(fio->op);
 373         struct page *bio_page;
 374         int err = 0;
 375
 376         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 377
 378         if (fio->old_blkaddr != NEW_ADDR)
 379                 verify_block_addr(sbi, fio->old_blkaddr);
 380         verify_block_addr(sbi, fio->new_blkaddr);
 381
 382         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 383
 384         /* set submitted = 1 as a return value */
 385         fio->submitted = 1;
 386
 387         if (!is_read)
 388                 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 389
 390         down_write(&io->io_rwsem);
 391
 392         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 393             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 394                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 395                 __submit_merged_bio(io);
 396 alloc_new:
 397         if (io->bio == NULL) {
 398                 if ((fio->type == DATA || fio->type == NODE) &&
 399                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 400                         err = -EAGAIN;
 401                         if (!is_read)
 402                                 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 403                         goto out_fail;
 404                 }
 405                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 406                                                 BIO_MAX_PAGES, is_read);
 407                 io->fio = *fio;
 408         }
 409
 410         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 411                                                         PAGE_SIZE) {
 412                 __submit_merged_bio(io);
 413                 goto alloc_new;
 414         }
 415
 416         io->last_block_in_bio = fio->new_blkaddr;
 417         f2fs_trace_ios(fio, 0);
 418 out_fail:
 419         up_write(&io->io_rwsem);
 420         trace_f2fs_submit_page_mbio(fio->page, fio);
 421         return err;
 422 }
 423
 424 static void __set_data_blkaddr(struct dnode_of_data *dn)
 425 {
 426         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 427         __le32 *addr_array;
 428
 429         /* Get physical address of data block */
 430         addr_array = blkaddr_in_node(rn);
 431         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 432 }
 433
 434 /*
 435  * Lock ordering for the change of data block address:
 436  * ->data_page
 437  *  ->node_page
 438  *    update block addresses in the node page
 439  */
 440 void set_data_blkaddr(struct dnode_of_data *dn)
 441 {
 442         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 443         __set_data_blkaddr(dn);
 444         if (set_page_dirty(dn->node_page))
 445                 dn->node_changed = true;
 446 }
 447
 448 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 449 {
 450         dn->data_blkaddr = blkaddr;
 451         set_data_blkaddr(dn);
 452         f2fs_update_extent_cache(dn);
 453 }
 454
 455 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 456 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 457 {
 458         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 459
 460         if (!count)
 461                 return 0;
 462
 463         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 464                 return -EPERM;
 465         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 466                 return -ENOSPC;
 467
 468         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 469                                                 dn->ofs_in_node, count);
 470
 471         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 472
 473         for (; count > 0; dn->ofs_in_node++) {
 474                 block_t blkaddr =
 475                         datablock_addr(dn->node_page, dn->ofs_in_node);
 476                 if (blkaddr == NULL_ADDR) {
 477                         dn->data_blkaddr = NEW_ADDR;
 478                         __set_data_blkaddr(dn);
 479                         count--;
 480                 }
 481         }
 482
 483         if (set_page_dirty(dn->node_page))
 484                 dn->node_changed = true;
 485         return 0;
 486 }
 487
 488 /* Should keep dn->ofs_in_node unchanged */
 489 int reserve_new_block(struct dnode_of_data *dn)
 490 {
 491         unsigned int ofs_in_node = dn->ofs_in_node;
 492         int ret;
 493
 494         ret = reserve_new_blocks(dn, 1);
 495         dn->ofs_in_node = ofs_in_node;
 496         return ret;
 497 }
 498
 499 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 500 {
 501         bool need_put = dn->inode_page ? false : true;
 502         int err;
 503
 504         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 505         if (err)
 506                 return err;
 507
 508         if (dn->data_blkaddr == NULL_ADDR)
 509                 err = reserve_new_block(dn);
 510         if (err || need_put)
 511                 f2fs_put_dnode(dn);
 512         return err;
 513 }
 514
 515 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 516 {
 517         struct extent_info ei  = {0,0,0};
 518         struct inode *inode = dn->inode;
 519
 520         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 521                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 522                 return 0;
 523         }
 524
 525         return f2fs_reserve_block(dn, index);
 526 }
 527
 528 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 529                                                 int op_flags, bool for_write)
 530 {
 531         struct address_space *mapping = inode->i_mapping;
 532         struct dnode_of_data dn;
 533         struct page *page;
 534         struct extent_info ei = {0,0,0};
 535         int err;
 536         struct f2fs_io_info fio = {
 537                 .sbi = F2FS_I_SB(inode),
 538                 .type = DATA,
 539                 .op = REQ_OP_READ,
 540                 .op_flags = op_flags,
 541                 .encrypted_page = NULL,
 542         };
 543
 544         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 545                 return read_mapping_page(mapping, index, NULL);
 546
 547         page = f2fs_grab_cache_page(mapping, index, for_write);
 548         if (!page)
 549                 return ERR_PTR(-ENOMEM);
 550
 551         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 552                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 553                 goto got_it;
 554         }
 555
 556         set_new_dnode(&dn, inode, NULL, NULL, 0);
 557         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 558         if (err)
 559                 goto put_err;
 560         f2fs_put_dnode(&dn);
 561
 562         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 563                 err = -ENOENT;
 564                 goto put_err;
 565         }
 566 got_it:
 567         if (PageUptodate(page)) {
 568                 unlock_page(page);
 569                 return page;
 570         }
 571
 572         /*
 573          * A new dentry page is allocated but not able to be written, since its
 574          * new inode page couldn't be allocated due to -ENOSPC.
 575          * In such the case, its blkaddr can be remained as NEW_ADDR.
 576          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 577          */
 578         if (dn.data_blkaddr == NEW_ADDR) {
 579                 zero_user_segment(page, 0, PAGE_SIZE);
 580                 if (!PageUptodate(page))
 581                         SetPageUptodate(page);
 582                 unlock_page(page);
 583                 return page;
 584         }
 585
 586         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 587         fio.page = page;
 588         err = f2fs_submit_page_bio(&fio);
 589         if (err)
 590                 goto put_err;
 591         return page;
 592
 593 put_err:
 594         f2fs_put_page(page, 1);
 595         return ERR_PTR(err);
 596 }
 597
 598 struct page *find_data_page(struct inode *inode, pgoff_t index)
 599 {
 600         struct address_space *mapping = inode->i_mapping;
 601         struct page *page;
 602
 603         page = find_get_page(mapping, index);
 604         if (page && PageUptodate(page))
 605                 return page;
 606         f2fs_put_page(page, 0);
 607
 608         page = get_read_data_page(inode, index, 0, false);
 609         if (IS_ERR(page))
 610                 return page;
 611
 612         if (PageUptodate(page))
 613                 return page;
 614
 615         wait_on_page_locked(page);
 616         if (unlikely(!PageUptodate(page))) {
 617                 f2fs_put_page(page, 0);
 618                 return ERR_PTR(-EIO);
 619         }
 620         return page;
 621 }
 622
 623 /*
 624  * If it tries to access a hole, return an error.
 625  * Because, the callers, functions in dir.c and GC, should be able to know
 626  * whether this page exists or not.
 627  */
 628 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 629                                                         bool for_write)
 630 {
 631         struct address_space *mapping = inode->i_mapping;
 632         struct page *page;
 633 repeat:
 634         page = get_read_data_page(inode, index, 0, for_write);
 635         if (IS_ERR(page))
 636                 return page;
 637
 638         /* wait for read completion */
 639         lock_page(page);
 640         if (unlikely(page->mapping != mapping)) {
 641                 f2fs_put_page(page, 1);
 642                 goto repeat;
 643         }
 644         if (unlikely(!PageUptodate(page))) {
 645                 f2fs_put_page(page, 1);
 646                 return ERR_PTR(-EIO);
 647         }
 648         return page;
 649 }
 650
 651 /*
 652  * Caller ensures that this data page is never allocated.
 653  * A new zero-filled data page is allocated in the page cache.
 654  *
 655  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 656  * f2fs_unlock_op().
 657  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 658  * ipage should be released by this function.
 659  */
 660 struct page *get_new_data_page(struct inode *inode,
 661                 struct page *ipage, pgoff_t index, bool new_i_size)
 662 {
 663         struct address_space *mapping = inode->i_mapping;
 664         struct page *page;
 665         struct dnode_of_data dn;
 666         int err;
 667
 668         page = f2fs_grab_cache_page(mapping, index, true);
 669         if (!page) {
 670                 /*
 671                  * before exiting, we should make sure ipage will be released
 672                  * if any error occur.
 673                  */
 674                 f2fs_put_page(ipage, 1);
 675                 return ERR_PTR(-ENOMEM);
 676         }
 677
 678         set_new_dnode(&dn, inode, ipage, NULL, 0);
 679         err = f2fs_reserve_block(&dn, index);
 680         if (err) {
 681                 f2fs_put_page(page, 1);
 682                 return ERR_PTR(err);
 683         }
 684         if (!ipage)
 685                 f2fs_put_dnode(&dn);
 686
 687         if (PageUptodate(page))
 688                 goto got_it;
 689
 690         if (dn.data_blkaddr == NEW_ADDR) {
 691                 zero_user_segment(page, 0, PAGE_SIZE);
 692                 if (!PageUptodate(page))
 693                         SetPageUptodate(page);
 694         } else {
 695                 f2fs_put_page(page, 1);
 696
 697                 /* if ipage exists, blkaddr should be NEW_ADDR */
 698                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 699                 page = get_lock_data_page(inode, index, true);
 700                 if (IS_ERR(page))
 701                         return page;
 702         }
 703 got_it:
 704         if (new_i_size && i_size_read(inode) <
 705                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 706                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 707         return page;
 708 }
 709
 710 static int __allocate_data_block(struct dnode_of_data *dn)
 711 {
 712         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 713         struct f2fs_summary sum;
 714         struct node_info ni;
 715         pgoff_t fofs;
 716         blkcnt_t count = 1;
 717
 718         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 719                 return -EPERM;
 720
 721         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 722         if (dn->data_blkaddr == NEW_ADDR)
 723                 goto alloc;
 724
 725         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 726                 return -ENOSPC;
 727
 728 alloc:
 729         get_node_info(sbi, dn->nid, &ni);
 730         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 731
 732         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 733                                                 &sum, CURSEG_WARM_DATA);
 734         set_data_blkaddr(dn);
 735
 736         /* update i_size */
 737         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 738                                                         dn->ofs_in_node;
 739         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 740                 f2fs_i_size_write(dn->inode,
 741                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 742         return 0;
 743 }
 744
 745 static inline bool __force_buffered_io(struct inode *inode, int rw)
 746 {
 747         return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) ||
 748                         (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
 749                         F2FS_I_SB(inode)->s_ndevs);
 750 }
 751
 752 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 753 {
 754         struct inode *inode = file_inode(iocb->ki_filp);
 755         struct f2fs_map_blocks map;
 756         int err = 0;
 757
 758         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 759                 return 0;
 760
 761         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 762         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 763         if (map.m_len > map.m_lblk)
 764                 map.m_len -= map.m_lblk;
 765         else
 766                 map.m_len = 0;
 767
 768         map.m_next_pgofs = NULL;
 769
 770         if (iocb->ki_flags & IOCB_DIRECT) {
 771                 err = f2fs_convert_inline_inode(inode);
 772                 if (err)
 773                         return err;
 774                 return f2fs_map_blocks(inode, &map, 1,
 775                         __force_buffered_io(inode, WRITE) ?
 776                                 F2FS_GET_BLOCK_PRE_AIO :
 777                                 F2FS_GET_BLOCK_PRE_DIO);
 778         }
 779         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 780                 err = f2fs_convert_inline_inode(inode);
 781                 if (err)
 782                         return err;
 783         }
 784         if (!f2fs_has_inline_data(inode))
 785                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 786         return err;
 787 }
 788
 789 /*
 790  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 791  * f2fs_map_blocks structure.
 792  * If original data blocks are allocated, then give them to blockdev.
 793  * Otherwise,
 794  *     a. preallocate requested block addresses
 795  *     b. do not use extent cache for better performance
 796  *     c. give the block addresses to blockdev
 797  */
 798 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 799                                                 int create, int flag)
 800 {
 801         unsigned int maxblocks = map->m_len;
 802         struct dnode_of_data dn;
 803         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 804         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 805         pgoff_t pgofs, end_offset, end;
 806         int err = 0, ofs = 1;
 807         unsigned int ofs_in_node, last_ofs_in_node;
 808         blkcnt_t prealloc;
 809         struct extent_info ei = {0,0,0};
 810         block_t blkaddr;
 811
 812         if (!maxblocks)
 813                 return 0;
 814
 815         map->m_len = 0;
 816         map->m_flags = 0;
 817
 818         /* it only supports block size == page size */
 819         pgofs = (pgoff_t)map->m_lblk;
 820         end = pgofs + maxblocks;
 821
 822         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 823                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 824                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 825                 map->m_flags = F2FS_MAP_MAPPED;
 826                 goto out;
 827         }
 828
 829 next_dnode:
 830         if (create)
 831                 f2fs_lock_op(sbi);
 832
 833         /* When reading holes, we need its node page */
 834         set_new_dnode(&dn, inode, NULL, NULL, 0);
 835         err = get_dnode_of_data(&dn, pgofs, mode);
 836         if (err) {
 837                 if (flag == F2FS_GET_BLOCK_BMAP)
 838                         map->m_pblk = 0;
 839                 if (err == -ENOENT) {
 840                         err = 0;
 841                         if (map->m_next_pgofs)
 842                                 *map->m_next_pgofs =
 843                                         get_next_page_offset(&dn, pgofs);
 844                 }
 845                 goto unlock_out;
 846         }
 847
 848         prealloc = 0;
 849         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
 850         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 851
 852 next_block:
 853         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 854
 855         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 856                 if (create) {
 857                         if (unlikely(f2fs_cp_error(sbi))) {
 858                                 err = -EIO;
 859                                 goto sync_out;
 860                         }
 861                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 862                                 if (blkaddr == NULL_ADDR) {
 863                                         prealloc++;
 864                                         last_ofs_in_node = dn.ofs_in_node;
 865                                 }
 866                         } else {
 867                                 err = __allocate_data_block(&dn);
 868                                 if (!err)
 869                                         set_inode_flag(inode, FI_APPEND_WRITE);
 870                         }
 871                         if (err)
 872                                 goto sync_out;
 873                         map->m_flags |= F2FS_MAP_NEW;
 874                         blkaddr = dn.data_blkaddr;
 875                 } else {
 876                         if (flag == F2FS_GET_BLOCK_BMAP) {
 877                                 map->m_pblk = 0;
 878                                 goto sync_out;
 879                         }
 880                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 881                                                 blkaddr == NULL_ADDR) {
 882                                 if (map->m_next_pgofs)
 883                                         *map->m_next_pgofs = pgofs + 1;
 884                         }
 885                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 886                                                 blkaddr != NEW_ADDR)
 887                                 goto sync_out;
 888                 }
 889         }
 890
 891         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 892                 goto skip;
 893
 894         if (map->m_len == 0) {
 895                 /* preallocated unwritten block should be mapped for fiemap. */
 896                 if (blkaddr == NEW_ADDR)
 897                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 898                 map->m_flags |= F2FS_MAP_MAPPED;
 899
 900                 map->m_pblk = blkaddr;
 901                 map->m_len = 1;
 902         } else if ((map->m_pblk != NEW_ADDR &&
 903                         blkaddr == (map->m_pblk + ofs)) ||
 904                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 905                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 906                 ofs++;
 907                 map->m_len++;
 908         } else {
 909                 goto sync_out;
 910         }
 911
 912 skip:
 913         dn.ofs_in_node++;
 914         pgofs++;
 915
 916         /* preallocate blocks in batch for one dnode page */
 917         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 918                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 919
 920                 dn.ofs_in_node = ofs_in_node;
 921                 err = reserve_new_blocks(&dn, prealloc);
 922                 if (err)
 923                         goto sync_out;
 924
 925                 map->m_len += dn.ofs_in_node - ofs_in_node;
 926                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 927                         err = -ENOSPC;
 928                         goto sync_out;
 929                 }
 930                 dn.ofs_in_node = end_offset;
 931         }
 932
 933         if (pgofs >= end)
 934                 goto sync_out;
 935         else if (dn.ofs_in_node < end_offset)
 936                 goto next_block;
 937
 938         f2fs_put_dnode(&dn);
 939
 940         if (create) {
 941                 f2fs_unlock_op(sbi);
 942                 f2fs_balance_fs(sbi, dn.node_changed);
 943         }
 944         goto next_dnode;
 945
 946 sync_out:
 947         f2fs_put_dnode(&dn);
 948 unlock_out:
 949         if (create) {
 950                 f2fs_unlock_op(sbi);
 951                 f2fs_balance_fs(sbi, dn.node_changed);
 952         }
 953 out:
 954         trace_f2fs_map_blocks(inode, map, err);
 955         return err;
 956 }
 957
 958 static int __get_data_block(struct inode *inode, sector_t iblock,
 959                         struct buffer_head *bh, int create, int flag,
 960                         pgoff_t *next_pgofs)
 961 {
 962         struct f2fs_map_blocks map;
 963         int err;
 964
 965         map.m_lblk = iblock;
 966         map.m_len = bh->b_size >> inode->i_blkbits;
 967         map.m_next_pgofs = next_pgofs;
 968
 969         err = f2fs_map_blocks(inode, &map, create, flag);
 970         if (!err) {
 971                 map_bh(bh, inode->i_sb, map.m_pblk);
 972                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 973                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
 974         }
 975         return err;
 976 }
 977
 978 static int get_data_block(struct inode *inode, sector_t iblock,
 979                         struct buffer_head *bh_result, int create, int flag,
 980                         pgoff_t *next_pgofs)
 981 {
 982         return __get_data_block(inode, iblock, bh_result, create,
 983                                                         flag, next_pgofs);
 984 }
 985
 986 static int get_data_block_dio(struct inode *inode, sector_t iblock,
 987                         struct buffer_head *bh_result, int create)
 988 {
 989         return __get_data_block(inode, iblock, bh_result, create,
 990                                                 F2FS_GET_BLOCK_DIO, NULL);
 991 }
 992
 993 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
 994                         struct buffer_head *bh_result, int create)
 995 {
 996         /* Block number less than F2FS MAX BLOCKS */
 997         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
 998                 return -EFBIG;
 999
1000         return __get_data_block(inode, iblock, bh_result, create,
1001                                                 F2FS_GET_BLOCK_BMAP, NULL);
1002 }
1003
1004 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1005 {
1006         return (offset >> inode->i_blkbits);
1007 }
1008
1009 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1010 {
1011         return (blk << inode->i_blkbits);
1012 }
1013
1014 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1015                 u64 start, u64 len)
1016 {
1017         struct buffer_head map_bh;
1018         sector_t start_blk, last_blk;
1019         pgoff_t next_pgofs;
1020         u64 logical = 0, phys = 0, size = 0;
1021         u32 flags = 0;
1022         int ret = 0;
1023
1024         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1025         if (ret)
1026                 return ret;
1027
1028         if (f2fs_has_inline_data(inode)) {
1029                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1030                 if (ret != -EAGAIN)
1031                         return ret;
1032         }
1033
1034         inode_lock(inode);
1035
1036         if (logical_to_blk(inode, len) == 0)
1037                 len = blk_to_logical(inode, 1);
1038
1039         start_blk = logical_to_blk(inode, start);
1040         last_blk = logical_to_blk(inode, start + len - 1);
1041
1042 next:
1043         memset(&map_bh, 0, sizeof(struct buffer_head));
1044         map_bh.b_size = len;
1045
1046         ret = get_data_block(inode, start_blk, &map_bh, 0,
1047                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1048         if (ret)
1049                 goto out;
1050
1051         /* HOLE */
1052         if (!buffer_mapped(&map_bh)) {
1053                 start_blk = next_pgofs;
1054
1055                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1056                                         F2FS_I_SB(inode)->max_file_blocks))
1057                         goto prep_next;
1058
1059                 flags |= FIEMAP_EXTENT_LAST;
1060         }
1061
1062         if (size) {
1063                 if (f2fs_encrypted_inode(inode))
1064                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1065
1066                 ret = fiemap_fill_next_extent(fieinfo, logical,
1067                                 phys, size, flags);
1068         }
1069
1070         if (start_blk > last_blk || ret)
1071                 goto out;
1072
1073         logical = blk_to_logical(inode, start_blk);
1074         phys = blk_to_logical(inode, map_bh.b_blocknr);
1075         size = map_bh.b_size;
1076         flags = 0;
1077         if (buffer_unwritten(&map_bh))
1078                 flags = FIEMAP_EXTENT_UNWRITTEN;
1079
1080         start_blk += logical_to_blk(inode, size);
1081
1082 prep_next:
1083         cond_resched();
1084         if (fatal_signal_pending(current))
1085                 ret = -EINTR;
1086         else
1087                 goto next;
1088 out:
1089         if (ret == 1)
1090                 ret = 0;
1091
1092         inode_unlock(inode);
1093         return ret;
1094 }
1095
1096 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1097                                  unsigned nr_pages)
1098 {
1099         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1100         struct fscrypt_ctx *ctx = NULL;
1101         struct bio *bio;
1102
1103         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1104                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1105                 if (IS_ERR(ctx))
1106                         return ERR_CAST(ctx);
1107
1108                 /* wait the page to be moved by cleaning */
1109                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1110         }
1111
1112         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1113         if (!bio) {
1114                 if (ctx)
1115                         fscrypt_release_ctx(ctx);
1116                 return ERR_PTR(-ENOMEM);
1117         }
1118         f2fs_target_device(sbi, blkaddr, bio);
1119         bio->bi_end_io = f2fs_read_end_io;
1120         bio->bi_private = ctx;
1121
1122         return bio;
1123 }
1124
1125 /*
1126  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1127  * Major change was from block_size == page_size in f2fs by default.
1128  */
1129 static int f2fs_mpage_readpages(struct address_space *mapping,
1130                         struct list_head *pages, struct page *page,
1131                         unsigned nr_pages)
1132 {
1133         struct bio *bio = NULL;
1134         unsigned page_idx;
1135         sector_t last_block_in_bio = 0;
1136         struct inode *inode = mapping->host;
1137         const unsigned blkbits = inode->i_blkbits;
1138         const unsigned blocksize = 1 << blkbits;
1139         sector_t block_in_file;
1140         sector_t last_block;
1141         sector_t last_block_in_file;
1142         sector_t block_nr;
1143         struct f2fs_map_blocks map;
1144
1145         map.m_pblk = 0;
1146         map.m_lblk = 0;
1147         map.m_len = 0;
1148         map.m_flags = 0;
1149         map.m_next_pgofs = NULL;
1150
1151         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1152
1153                 prefetchw(&page->flags);
1154                 if (pages) {
1155                         page = list_last_entry(pages, struct page, lru);
1156                         list_del(&page->lru);
1157                         if (add_to_page_cache_lru(page, mapping,
1158                                                   page->index,
1159                                                   readahead_gfp_mask(mapping)))
1160                                 goto next_page;
1161                 }
1162
1163                 block_in_file = (sector_t)page->index;
1164                 last_block = block_in_file + nr_pages;
1165                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1166                                                                 blkbits;
1167                 if (last_block > last_block_in_file)
1168                         last_block = last_block_in_file;
1169
1170                 /*
1171                  * Map blocks using the previous result first.
1172                  */
1173                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1174                                 block_in_file > map.m_lblk &&
1175                                 block_in_file < (map.m_lblk + map.m_len))
1176                         goto got_it;
1177
1178                 /*
1179                  * Then do more f2fs_map_blocks() calls until we are
1180                  * done with this page.
1181                  */
1182                 map.m_flags = 0;
1183
1184                 if (block_in_file < last_block) {
1185                         map.m_lblk = block_in_file;
1186                         map.m_len = last_block - block_in_file;
1187
1188                         if (f2fs_map_blocks(inode, &map, 0,
1189                                                 F2FS_GET_BLOCK_READ))
1190                                 goto set_error_page;
1191                 }
1192 got_it:
1193                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1194                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1195                         SetPageMappedToDisk(page);
1196
1197                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1198                                 SetPageUptodate(page);
1199                                 goto confused;
1200                         }
1201                 } else {
1202                         zero_user_segment(page, 0, PAGE_SIZE);
1203                         if (!PageUptodate(page))
1204                                 SetPageUptodate(page);
1205                         unlock_page(page);
1206                         goto next_page;
1207                 }
1208
1209                 /*
1210                  * This page will go to BIO.  Do we need to send this
1211                  * BIO off first?
1212                  */
1213                 if (bio && (last_block_in_bio != block_nr - 1 ||
1214                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1215 submit_and_realloc:
1216                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1217                         bio = NULL;
1218                 }
1219                 if (bio == NULL) {
1220                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1221                         if (IS_ERR(bio)) {
1222                                 bio = NULL;
1223                                 goto set_error_page;
1224                         }
1225                         bio_set_op_attrs(bio, REQ_OP_READ, 0);
1226                 }
1227
1228                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1229                         goto submit_and_realloc;
1230
1231                 last_block_in_bio = block_nr;
1232                 goto next_page;
1233 set_error_page:
1234                 SetPageError(page);
1235                 zero_user_segment(page, 0, PAGE_SIZE);
1236                 unlock_page(page);
1237                 goto next_page;
1238 confused:
1239                 if (bio) {
1240                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1241                         bio = NULL;
1242                 }
1243                 unlock_page(page);
1244 next_page:
1245                 if (pages)
1246                         put_page(page);
1247         }
1248         BUG_ON(pages && !list_empty(pages));
1249         if (bio)
1250                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1251         return 0;
1252 }
1253
1254 static int f2fs_read_data_page(struct file *file, struct page *page)
1255 {
1256         struct inode *inode = page->mapping->host;
1257         int ret = -EAGAIN;
1258
1259         trace_f2fs_readpage(page, DATA);
1260
1261         /* If the file has inline data, try to read it directly */
1262         if (f2fs_has_inline_data(inode))
1263                 ret = f2fs_read_inline_data(inode, page);
1264         if (ret == -EAGAIN)
1265                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1266         return ret;
1267 }
1268
1269 static int f2fs_read_data_pages(struct file *file,
1270                         struct address_space *mapping,
1271                         struct list_head *pages, unsigned nr_pages)
1272 {
1273         struct inode *inode = file->f_mapping->host;
1274         struct page *page = list_last_entry(pages, struct page, lru);
1275
1276         trace_f2fs_readpages(inode, page, nr_pages);
1277
1278         /* If the file has inline data, skip readpages */
1279         if (f2fs_has_inline_data(inode))
1280                 return 0;
1281
1282         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1283 }
1284
1285 int do_write_data_page(struct f2fs_io_info *fio)
1286 {
1287         struct page *page = fio->page;
1288         struct inode *inode = page->mapping->host;
1289         struct dnode_of_data dn;
1290         int err = 0;
1291
1292         set_new_dnode(&dn, inode, NULL, NULL, 0);
1293         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1294         if (err)
1295                 return err;
1296
1297         fio->old_blkaddr = dn.data_blkaddr;
1298
1299         /* This page is already truncated */
1300         if (fio->old_blkaddr == NULL_ADDR) {
1301                 ClearPageUptodate(page);
1302                 goto out_writepage;
1303         }
1304
1305         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1306                 gfp_t gfp_flags = GFP_NOFS;
1307
1308                 /* wait for GCed encrypted page writeback */
1309                 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1310                                                         fio->old_blkaddr);
1311 retry_encrypt:
1312                 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1313                                                         PAGE_SIZE, 0,
1314                                                         fio->page->index,
1315                                                         gfp_flags);
1316                 if (IS_ERR(fio->encrypted_page)) {
1317                         err = PTR_ERR(fio->encrypted_page);
1318                         if (err == -ENOMEM) {
1319                                 /* flush pending ios and wait for a while */
1320                                 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1321                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1322                                 gfp_flags |= __GFP_NOFAIL;
1323                                 err = 0;
1324                                 goto retry_encrypt;
1325                         }
1326                         goto out_writepage;
1327                 }
1328         }
1329
1330         set_page_writeback(page);
1331
1332         /*
1333          * If current allocation needs SSR,
1334          * it had better in-place writes for updated data.
1335          */
1336         if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1337                         !is_cold_data(page) &&
1338                         !IS_ATOMIC_WRITTEN_PAGE(page) &&
1339                         need_inplace_update(inode))) {
1340                 rewrite_data_page(fio);
1341                 set_inode_flag(inode, FI_UPDATE_WRITE);
1342                 trace_f2fs_do_write_data_page(page, IPU);
1343         } else {
1344                 write_data_page(&dn, fio);
1345                 trace_f2fs_do_write_data_page(page, OPU);
1346                 set_inode_flag(inode, FI_APPEND_WRITE);
1347                 if (page->index == 0)
1348                         set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1349         }
1350 out_writepage:
1351         f2fs_put_dnode(&dn);
1352         return err;
1353 }
1354
1355 static int __write_data_page(struct page *page, bool *submitted,
1356                                 struct writeback_control *wbc)
1357 {
1358         struct inode *inode = page->mapping->host;
1359         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1360         loff_t i_size = i_size_read(inode);
1361         const pgoff_t end_index = ((unsigned long long) i_size)
1362                                                         >> PAGE_SHIFT;
1363         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1364         unsigned offset = 0;
1365         bool need_balance_fs = false;
1366         int err = 0;
1367         struct f2fs_io_info fio = {
1368                 .sbi = sbi,
1369                 .type = DATA,
1370                 .op = REQ_OP_WRITE,
1371                 .op_flags = wbc_to_write_flags(wbc),
1372                 .page = page,
1373                 .encrypted_page = NULL,
1374                 .submitted = false,
1375         };
1376
1377         trace_f2fs_writepage(page, DATA);
1378
1379         if (page->index < end_index)
1380                 goto write;
1381
1382         /*
1383          * If the offset is out-of-range of file size,
1384          * this page does not have to be written to disk.
1385          */
1386         offset = i_size & (PAGE_SIZE - 1);
1387         if ((page->index >= end_index + 1) || !offset)
1388                 goto out;
1389
1390         zero_user_segment(page, offset, PAGE_SIZE);
1391 write:
1392         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1393                 goto redirty_out;
1394         if (f2fs_is_drop_cache(inode))
1395                 goto out;
1396         /* we should not write 0'th page having journal header */
1397         if (f2fs_is_volatile_file(inode) && (!page->index ||
1398                         (!wbc->for_reclaim &&
1399                         available_free_memory(sbi, BASE_CHECK))))
1400                 goto redirty_out;
1401
1402         /* we should bypass data pages to proceed the kworkder jobs */
1403         if (unlikely(f2fs_cp_error(sbi))) {
1404                 mapping_set_error(page->mapping, -EIO);
1405                 goto out;
1406         }
1407
1408         /* Dentry blocks are controlled by checkpoint */
1409         if (S_ISDIR(inode->i_mode)) {
1410                 err = do_write_data_page(&fio);
1411                 goto done;
1412         }
1413
1414         if (!wbc->for_reclaim)
1415                 need_balance_fs = true;
1416         else if (has_not_enough_free_secs(sbi, 0, 0))
1417                 goto redirty_out;
1418
1419         err = -EAGAIN;
1420         if (f2fs_has_inline_data(inode)) {
1421                 err = f2fs_write_inline_data(inode, page);
1422                 if (!err)
1423                         goto out;
1424         }
1425         f2fs_lock_op(sbi);
1426         if (err == -EAGAIN)
1427                 err = do_write_data_page(&fio);
1428         if (F2FS_I(inode)->last_disk_size < psize)
1429                 F2FS_I(inode)->last_disk_size = psize;
1430         f2fs_unlock_op(sbi);
1431 done:
1432         if (err && err != -ENOENT)
1433                 goto redirty_out;
1434
1435 out:
1436         inode_dec_dirty_pages(inode);
1437         if (err)
1438                 ClearPageUptodate(page);
1439
1440         if (wbc->for_reclaim) {
1441                 f2fs_submit_merged_bio_cond(sbi, inode, 0, page->index,
1442                                                 DATA, WRITE);
1443                 remove_dirty_inode(inode);
1444                 submitted = NULL;
1445         }
1446
1447         unlock_page(page);
1448         f2fs_balance_fs(sbi, need_balance_fs);
1449
1450         if (unlikely(f2fs_cp_error(sbi))) {
1451                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1452                 submitted = NULL;
1453         }
1454
1455         if (submitted)
1456                 *submitted = fio.submitted;
1457
1458         return 0;
1459
1460 redirty_out:
1461         redirty_page_for_writepage(wbc, page);
1462         if (!err)
1463                 return AOP_WRITEPAGE_ACTIVATE;
1464         unlock_page(page);
1465         return err;
1466 }
1467
1468 static int f2fs_write_data_page(struct page *page,
1469                                         struct writeback_control *wbc)
1470 {
1471         return __write_data_page(page, NULL, wbc);
1472 }
1473
1474 /*
1475  * This function was copied from write_cche_pages from mm/page-writeback.c.
1476  * The major change is making write step of cold data page separately from
1477  * warm/hot data page.
1478  */
1479 static int f2fs_write_cache_pages(struct address_space *mapping,
1480                                         struct writeback_control *wbc)
1481 {
1482         int ret = 0;
1483         int done = 0;
1484         struct pagevec pvec;
1485         int nr_pages;
1486         pgoff_t uninitialized_var(writeback_index);
1487         pgoff_t index;
1488         pgoff_t end;            /* Inclusive */
1489         pgoff_t done_index;
1490         pgoff_t last_idx = ULONG_MAX;
1491         int cycled;
1492         int range_whole = 0;
1493         int tag;
1494
1495         pagevec_init(&pvec, 0);
1496
1497         if (wbc->range_cyclic) {
1498                 writeback_index = mapping->writeback_index; /* prev offset */
1499                 index = writeback_index;
1500                 if (index == 0)
1501                         cycled = 1;
1502                 else
1503                         cycled = 0;
1504                 end = -1;
1505         } else {
1506                 index = wbc->range_start >> PAGE_SHIFT;
1507                 end = wbc->range_end >> PAGE_SHIFT;
1508                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1509                         range_whole = 1;
1510                 cycled = 1; /* ignore range_cyclic tests */
1511         }
1512         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1513                 tag = PAGECACHE_TAG_TOWRITE;
1514         else
1515                 tag = PAGECACHE_TAG_DIRTY;
1516 retry:
1517         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1518                 tag_pages_for_writeback(mapping, index, end);
1519         done_index = index;
1520         while (!done && (index <= end)) {
1521                 int i;
1522
1523                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1524                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1525                 if (nr_pages == 0)
1526                         break;
1527
1528                 for (i = 0; i < nr_pages; i++) {
1529                         struct page *page = pvec.pages[i];
1530                         bool submitted = false;
1531
1532                         if (page->index > end) {
1533                                 done = 1;
1534                                 break;
1535                         }
1536
1537                         done_index = page->index;
1538
1539                         lock_page(page);
1540
1541                         if (unlikely(page->mapping != mapping)) {
1542 continue_unlock:
1543                                 unlock_page(page);
1544                                 continue;
1545                         }
1546
1547                         if (!PageDirty(page)) {
1548                                 /* someone wrote it for us */
1549                                 goto continue_unlock;
1550                         }
1551
1552                         if (PageWriteback(page)) {
1553                                 if (wbc->sync_mode != WB_SYNC_NONE)
1554                                         f2fs_wait_on_page_writeback(page,
1555                                                                 DATA, true);
1556                                 else
1557                                         goto continue_unlock;
1558                         }
1559
1560                         BUG_ON(PageWriteback(page));
1561                         if (!clear_page_dirty_for_io(page))
1562                                 goto continue_unlock;
1563
1564                         ret = __write_data_page(page, &submitted, wbc);
1565                         if (unlikely(ret)) {
1566                                 /*
1567                                  * keep nr_to_write, since vfs uses this to
1568                                  * get # of written pages.
1569                                  */
1570                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1571                                         unlock_page(page);
1572                                         ret = 0;
1573                                         continue;
1574                                 }
1575                                 done_index = page->index + 1;
1576                                 done = 1;
1577                                 break;
1578                         } else if (submitted) {
1579                                 last_idx = page->index;
1580                         }
1581
1582                         if (--wbc->nr_to_write <= 0 &&
1583                             wbc->sync_mode == WB_SYNC_NONE) {
1584                                 done = 1;
1585                                 break;
1586                         }
1587                 }
1588                 pagevec_release(&pvec);
1589                 cond_resched();
1590         }
1591
1592         if (!cycled && !done) {
1593                 cycled = 1;
1594                 index = 0;
1595                 end = writeback_index - 1;
1596                 goto retry;
1597         }
1598         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1599                 mapping->writeback_index = done_index;
1600
1601         if (last_idx != ULONG_MAX)
1602                 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1603                                                 0, last_idx, DATA, WRITE);
1604
1605         return ret;
1606 }
1607
1608 static int f2fs_write_data_pages(struct address_space *mapping,
1609                             struct writeback_control *wbc)
1610 {
1611         struct inode *inode = mapping->host;
1612         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1613         struct blk_plug plug;
1614         int ret;
1615
1616         /* deal with chardevs and other special file */
1617         if (!mapping->a_ops->writepage)
1618                 return 0;
1619
1620         /* skip writing if there is no dirty page in this inode */
1621         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1622                 return 0;
1623
1624         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1625                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1626                         available_free_memory(sbi, DIRTY_DENTS))
1627                 goto skip_write;
1628
1629         /* skip writing during file defragment */
1630         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1631                 goto skip_write;
1632
1633         /* during POR, we don't need to trigger writepage at all. */
1634         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1635                 goto skip_write;
1636
1637         trace_f2fs_writepages(mapping->host, wbc, DATA);
1638
1639         blk_start_plug(&plug);
1640         ret = f2fs_write_cache_pages(mapping, wbc);
1641         blk_finish_plug(&plug);
1642         /*
1643          * if some pages were truncated, we cannot guarantee its mapping->host
1644          * to detect pending bios.
1645          */
1646
1647         remove_dirty_inode(inode);
1648         return ret;
1649
1650 skip_write:
1651         wbc->pages_skipped += get_dirty_pages(inode);
1652         trace_f2fs_writepages(mapping->host, wbc, DATA);
1653         return 0;
1654 }
1655
1656 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1657 {
1658         struct inode *inode = mapping->host;
1659         loff_t i_size = i_size_read(inode);
1660
1661         if (to > i_size) {
1662                 truncate_pagecache(inode, i_size);
1663                 truncate_blocks(inode, i_size, true);
1664         }
1665 }
1666
1667 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1668                         struct page *page, loff_t pos, unsigned len,
1669                         block_t *blk_addr, bool *node_changed)
1670 {
1671         struct inode *inode = page->mapping->host;
1672         pgoff_t index = page->index;
1673         struct dnode_of_data dn;
1674         struct page *ipage;
1675         bool locked = false;
1676         struct extent_info ei = {0,0,0};
1677         int err = 0;
1678
1679         /*
1680          * we already allocated all the blocks, so we don't need to get
1681          * the block addresses when there is no need to fill the page.
1682          */
1683         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1684                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
1685                 return 0;
1686
1687         if (f2fs_has_inline_data(inode) ||
1688                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1689                 f2fs_lock_op(sbi);
1690                 locked = true;
1691         }
1692 restart:
1693         /* check inline_data */
1694         ipage = get_node_page(sbi, inode->i_ino);
1695         if (IS_ERR(ipage)) {
1696                 err = PTR_ERR(ipage);
1697                 goto unlock_out;
1698         }
1699
1700         set_new_dnode(&dn, inode, ipage, ipage, 0);
1701
1702         if (f2fs_has_inline_data(inode)) {
1703                 if (pos + len <= MAX_INLINE_DATA) {
1704                         read_inline_data(page, ipage);
1705                         set_inode_flag(inode, FI_DATA_EXIST);
1706                         if (inode->i_nlink)
1707                                 set_inline_node(ipage);
1708                 } else {
1709                         err = f2fs_convert_inline_page(&dn, page);
1710                         if (err)
1711                                 goto out;
1712                         if (dn.data_blkaddr == NULL_ADDR)
1713                                 err = f2fs_get_block(&dn, index);
1714                 }
1715         } else if (locked) {
1716                 err = f2fs_get_block(&dn, index);
1717         } else {
1718                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1719                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1720                 } else {
1721                         /* hole case */
1722                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1723                         if (err || dn.data_blkaddr == NULL_ADDR) {
1724                                 f2fs_put_dnode(&dn);
1725                                 f2fs_lock_op(sbi);
1726                                 locked = true;
1727                                 goto restart;
1728                         }
1729                 }
1730         }
1731
1732         /* convert_inline_page can make node_changed */
1733         *blk_addr = dn.data_blkaddr;
1734         *node_changed = dn.node_changed;
1735 out:
1736         f2fs_put_dnode(&dn);
1737 unlock_out:
1738         if (locked)
1739                 f2fs_unlock_op(sbi);
1740         return err;
1741 }
1742
1743 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1744                 loff_t pos, unsigned len, unsigned flags,
1745                 struct page **pagep, void **fsdata)
1746 {
1747         struct inode *inode = mapping->host;
1748         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1749         struct page *page = NULL;
1750         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1751         bool need_balance = false;
1752         block_t blkaddr = NULL_ADDR;
1753         int err = 0;
1754
1755         trace_f2fs_write_begin(inode, pos, len, flags);
1756
1757         /*
1758          * We should check this at this moment to avoid deadlock on inode page
1759          * and #0 page. The locking rule for inline_data conversion should be:
1760          * lock_page(page #0) -> lock_page(inode_page)
1761          */
1762         if (index != 0) {
1763                 err = f2fs_convert_inline_inode(inode);
1764                 if (err)
1765                         goto fail;
1766         }
1767 repeat:
1768         /*
1769          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1770          * wait_for_stable_page. Will wait that below with our IO control.
1771          */
1772         page = pagecache_get_page(mapping, index,
1773                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1774         if (!page) {
1775                 err = -ENOMEM;
1776                 goto fail;
1777         }
1778
1779         *pagep = page;
1780
1781         err = prepare_write_begin(sbi, page, pos, len,
1782                                         &blkaddr, &need_balance);
1783         if (err)
1784                 goto fail;
1785
1786         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1787                 unlock_page(page);
1788                 f2fs_balance_fs(sbi, true);
1789                 lock_page(page);
1790                 if (page->mapping != mapping) {
1791                         /* The page got truncated from under us */
1792                         f2fs_put_page(page, 1);
1793                         goto repeat;
1794                 }
1795         }
1796
1797         f2fs_wait_on_page_writeback(page, DATA, false);
1798
1799         /* wait for GCed encrypted page writeback */
1800         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1801                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1802
1803         if (len == PAGE_SIZE || PageUptodate(page))
1804                 return 0;
1805
1806         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1807                 zero_user_segment(page, len, PAGE_SIZE);
1808                 return 0;
1809         }
1810
1811         if (blkaddr == NEW_ADDR) {
1812                 zero_user_segment(page, 0, PAGE_SIZE);
1813                 SetPageUptodate(page);
1814         } else {
1815                 struct bio *bio;
1816
1817                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1818                 if (IS_ERR(bio)) {
1819                         err = PTR_ERR(bio);
1820                         goto fail;
1821                 }
1822                 bio->bi_opf = REQ_OP_READ;
1823                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1824                         bio_put(bio);
1825                         err = -EFAULT;
1826                         goto fail;
1827                 }
1828
1829                 __submit_bio(sbi, bio, DATA);
1830
1831                 lock_page(page);
1832                 if (unlikely(page->mapping != mapping)) {
1833                         f2fs_put_page(page, 1);
1834                         goto repeat;
1835                 }
1836                 if (unlikely(!PageUptodate(page))) {
1837                         err = -EIO;
1838                         goto fail;
1839                 }
1840         }
1841         return 0;
1842
1843 fail:
1844         f2fs_put_page(page, 1);
1845         f2fs_write_failed(mapping, pos + len);
1846         return err;
1847 }
1848
1849 static int f2fs_write_end(struct file *file,
1850                         struct address_space *mapping,
1851                         loff_t pos, unsigned len, unsigned copied,
1852                         struct page *page, void *fsdata)
1853 {
1854         struct inode *inode = page->mapping->host;
1855
1856         trace_f2fs_write_end(inode, pos, len, copied);
1857
1858         /*
1859          * This should be come from len == PAGE_SIZE, and we expect copied
1860          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1861          * let generic_perform_write() try to copy data again through copied=0.
1862          */
1863         if (!PageUptodate(page)) {
1864                 if (unlikely(copied != len))
1865                         copied = 0;
1866                 else
1867                         SetPageUptodate(page);
1868         }
1869         if (!copied)
1870                 goto unlock_out;
1871
1872         set_page_dirty(page);
1873
1874         if (pos + copied > i_size_read(inode))
1875                 f2fs_i_size_write(inode, pos + copied);
1876 unlock_out:
1877         f2fs_put_page(page, 1);
1878         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1879         return copied;
1880 }
1881
1882 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1883                            loff_t offset)
1884 {
1885         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1886
1887         if (offset & blocksize_mask)
1888                 return -EINVAL;
1889
1890         if (iov_iter_alignment(iter) & blocksize_mask)
1891                 return -EINVAL;
1892
1893         return 0;
1894 }
1895
1896 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1897 {
1898         struct address_space *mapping = iocb->ki_filp->f_mapping;
1899         struct inode *inode = mapping->host;
1900         size_t count = iov_iter_count(iter);
1901         loff_t offset = iocb->ki_pos;
1902         int rw = iov_iter_rw(iter);
1903         int err;
1904
1905         err = check_direct_IO(inode, iter, offset);
1906         if (err)
1907                 return err;
1908
1909         if (__force_buffered_io(inode, rw))
1910                 return 0;
1911
1912         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1913
1914         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1915         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1916         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1917
1918         if (rw == WRITE) {
1919                 if (err > 0)
1920                         set_inode_flag(inode, FI_UPDATE_WRITE);
1921                 else if (err < 0)
1922                         f2fs_write_failed(mapping, offset + count);
1923         }
1924
1925         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1926
1927         return err;
1928 }
1929
1930 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1931                                                         unsigned int length)
1932 {
1933         struct inode *inode = page->mapping->host;
1934         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1935
1936         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1937                 (offset % PAGE_SIZE || length != PAGE_SIZE))
1938                 return;
1939
1940         if (PageDirty(page)) {
1941                 if (inode->i_ino == F2FS_META_INO(sbi)) {
1942                         dec_page_count(sbi, F2FS_DIRTY_META);
1943                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
1944                         dec_page_count(sbi, F2FS_DIRTY_NODES);
1945                 } else {
1946                         inode_dec_dirty_pages(inode);
1947                         remove_dirty_inode(inode);
1948                 }
1949         }
1950
1951         /* This is atomic written page, keep Private */
1952         if (IS_ATOMIC_WRITTEN_PAGE(page))
1953                 return;
1954
1955         set_page_private(page, 0);
1956         ClearPagePrivate(page);
1957 }
1958
1959 int f2fs_release_page(struct page *page, gfp_t wait)
1960 {
1961         /* If this is dirty page, keep PagePrivate */
1962         if (PageDirty(page))
1963                 return 0;
1964
1965         /* This is atomic written page, keep Private */
1966         if (IS_ATOMIC_WRITTEN_PAGE(page))
1967                 return 0;
1968
1969         set_page_private(page, 0);
1970         ClearPagePrivate(page);
1971         return 1;
1972 }
1973
1974 /*
1975  * This was copied from __set_page_dirty_buffers which gives higher performance
1976  * in very high speed storages. (e.g., pmem)
1977  */
1978 void f2fs_set_page_dirty_nobuffers(struct page *page)
1979 {
1980         struct address_space *mapping = page->mapping;
1981         unsigned long flags;
1982
1983         if (unlikely(!mapping))
1984                 return;
1985
1986         spin_lock(&mapping->private_lock);
1987         lock_page_memcg(page);
1988         SetPageDirty(page);
1989         spin_unlock(&mapping->private_lock);
1990
1991         spin_lock_irqsave(&mapping->tree_lock, flags);
1992         WARN_ON_ONCE(!PageUptodate(page));
1993         account_page_dirtied(page, mapping);
1994         radix_tree_tag_set(&mapping->page_tree,
1995                         page_index(page), PAGECACHE_TAG_DIRTY);
1996         spin_unlock_irqrestore(&mapping->tree_lock, flags);
1997         unlock_page_memcg(page);
1998
1999         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2000         return;
2001 }
2002
2003 static int f2fs_set_data_page_dirty(struct page *page)
2004 {
2005         struct address_space *mapping = page->mapping;
2006         struct inode *inode = mapping->host;
2007
2008         trace_f2fs_set_page_dirty(page, DATA);
2009
2010         if (!PageUptodate(page))
2011                 SetPageUptodate(page);
2012
2013         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2014                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2015                         register_inmem_page(inode, page);
2016                         return 1;
2017                 }
2018                 /*
2019                  * Previously, this page has been registered, we just
2020                  * return here.
2021                  */
2022                 return 0;
2023         }
2024
2025         if (!PageDirty(page)) {
2026                 f2fs_set_page_dirty_nobuffers(page);
2027                 update_dirty_page(inode, page);
2028                 return 1;
2029         }
2030         return 0;
2031 }
2032
2033 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2034 {
2035         struct inode *inode = mapping->host;
2036
2037         if (f2fs_has_inline_data(inode))
2038                 return 0;
2039
2040         /* make sure allocating whole blocks */
2041         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2042                 filemap_write_and_wait(mapping);
2043
2044         return generic_block_bmap(mapping, block, get_data_block_bmap);
2045 }
2046
2047 #ifdef CONFIG_MIGRATION
2048 #include <linux/migrate.h>
2049
2050 int f2fs_migrate_page(struct address_space *mapping,
2051                 struct page *newpage, struct page *page, enum migrate_mode mode)
2052 {
2053         int rc, extra_count;
2054         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2055         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2056
2057         BUG_ON(PageWriteback(page));
2058
2059         /* migrating an atomic written page is safe with the inmem_lock hold */
2060         if (atomic_written && !mutex_trylock(&fi->inmem_lock))
2061                 return -EAGAIN;
2062
2063         /*
2064          * A reference is expected if PagePrivate set when move mapping,
2065          * however F2FS breaks this for maintaining dirty page counts when
2066          * truncating pages. So here adjusting the 'extra_count' make it work.
2067          */
2068         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2069         rc = migrate_page_move_mapping(mapping, newpage,
2070                                 page, NULL, mode, extra_count);
2071         if (rc != MIGRATEPAGE_SUCCESS) {
2072                 if (atomic_written)
2073                         mutex_unlock(&fi->inmem_lock);
2074                 return rc;
2075         }
2076
2077         if (atomic_written) {
2078                 struct inmem_pages *cur;
2079                 list_for_each_entry(cur, &fi->inmem_pages, list)
2080                         if (cur->page == page) {
2081                                 cur->page = newpage;
2082                                 break;
2083                         }
2084                 mutex_unlock(&fi->inmem_lock);
2085                 put_page(page);
2086                 get_page(newpage);
2087         }
2088
2089         if (PagePrivate(page))
2090                 SetPagePrivate(newpage);
2091         set_page_private(newpage, page_private(page));
2092
2093         migrate_page_copy(newpage, page);
2094
2095         return MIGRATEPAGE_SUCCESS;
2096 }
2097 #endif
2098
2099 const struct address_space_operations f2fs_dblock_aops = {
2100         .readpage       = f2fs_read_data_page,
2101         .readpages      = f2fs_read_data_pages,
2102         .writepage      = f2fs_write_data_page,
2103         .writepages     = f2fs_write_data_pages,
2104         .write_begin    = f2fs_write_begin,
2105         .write_end      = f2fs_write_end,
2106         .set_page_dirty = f2fs_set_data_page_dirty,
2107         .invalidatepage = f2fs_invalidate_page,
2108         .releasepage    = f2fs_release_page,
2109         .direct_IO      = f2fs_direct_IO,
2110         .bmap           = f2fs_bmap,
2111 #ifdef CONFIG_MIGRATION
2112         .migratepage    = f2fs_migrate_page,
2113 #endif
2114 };