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 #include <linux/sched/signal.h>
  26
  27 #include "f2fs.h"
  28 #include "node.h"
  29 #include "segment.h"
  30 #include "trace.h"
  31 #include <trace/events/f2fs.h>
  32
  33 static bool __is_cp_guaranteed(struct page *page)
  34 {
  35         struct address_space *mapping = page->mapping;
  36         struct inode *inode;
  37         struct f2fs_sb_info *sbi;
  38
  39         if (!mapping)
  40                 return false;
  41
  42         inode = mapping->host;
  43         sbi = F2FS_I_SB(inode);
  44
  45         if (inode->i_ino == F2FS_META_INO(sbi) ||
  46                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47                         S_ISDIR(inode->i_mode) ||
  48                         is_cold_data(page))
  49                 return true;
  50         return false;
  51 }
  52
  53 static void f2fs_read_end_io(struct bio *bio)
  54 {
  55         struct bio_vec *bvec;
  56         int i;
  57
  58 #ifdef CONFIG_F2FS_FAULT_INJECTION
  59         if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) {
  60                 f2fs_show_injection_info(FAULT_IO);
  61                 bio->bi_status = BLK_STS_IOERR;
  62         }
  63 #endif
  64
  65         if (f2fs_bio_encrypted(bio)) {
  66                 if (bio->bi_status) {
  67                         fscrypt_release_ctx(bio->bi_private);
  68                 } else {
  69                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  70                         return;
  71                 }
  72         }
  73
  74         bio_for_each_segment_all(bvec, bio, i) {
  75                 struct page *page = bvec->bv_page;
  76
  77                 if (!bio->bi_status) {
  78                         if (!PageUptodate(page))
  79                                 SetPageUptodate(page);
  80                 } else {
  81                         ClearPageUptodate(page);
  82                         SetPageError(page);
  83                 }
  84                 unlock_page(page);
  85         }
  86         bio_put(bio);
  87 }
  88
  89 static void f2fs_write_end_io(struct bio *bio)
  90 {
  91         struct f2fs_sb_info *sbi = bio->bi_private;
  92         struct bio_vec *bvec;
  93         int i;
  94
  95         bio_for_each_segment_all(bvec, bio, i) {
  96                 struct page *page = bvec->bv_page;
  97                 enum count_type type = WB_DATA_TYPE(page);
  98
  99                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 100                         set_page_private(page, (unsigned long)NULL);
 101                         ClearPagePrivate(page);
 102                         unlock_page(page);
 103                         mempool_free(page, sbi->write_io_dummy);
 104
 105                         if (unlikely(bio->bi_status))
 106                                 f2fs_stop_checkpoint(sbi, true);
 107                         continue;
 108                 }
 109
 110                 fscrypt_pullback_bio_page(&page, true);
 111
 112                 if (unlikely(bio->bi_status)) {
 113                         mapping_set_error(page->mapping, -EIO);
 114                         f2fs_stop_checkpoint(sbi, true);
 115                 }
 116                 dec_page_count(sbi, type);
 117                 clear_cold_data(page);
 118                 end_page_writeback(page);
 119         }
 120         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 121                                 wq_has_sleeper(&sbi->cp_wait))
 122                 wake_up(&sbi->cp_wait);
 123
 124         bio_put(bio);
 125 }
 126
 127 /*
 128  * Return true, if pre_bio's bdev is same as its target device.
 129  */
 130 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 131                                 block_t blk_addr, struct bio *bio)
 132 {
 133         struct block_device *bdev = sbi->sb->s_bdev;
 134         int i;
 135
 136         for (i = 0; i < sbi->s_ndevs; i++) {
 137                 if (FDEV(i).start_blk <= blk_addr &&
 138                                         FDEV(i).end_blk >= blk_addr) {
 139                         blk_addr -= FDEV(i).start_blk;
 140                         bdev = FDEV(i).bdev;
 141                         break;
 142                 }
 143         }
 144         if (bio) {
 145                 bio->bi_bdev = bdev;
 146                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 147         }
 148         return bdev;
 149 }
 150
 151 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 152 {
 153         int i;
 154
 155         for (i = 0; i < sbi->s_ndevs; i++)
 156                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 157                         return i;
 158         return 0;
 159 }
 160
 161 static bool __same_bdev(struct f2fs_sb_info *sbi,
 162                                 block_t blk_addr, struct bio *bio)
 163 {
 164         return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
 165 }
 166
 167 /*
 168  * Low-level block read/write IO operations.
 169  */
 170 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 171                                 int npages, bool is_read)
 172 {
 173         struct bio *bio;
 174
 175         bio = f2fs_bio_alloc(npages);
 176
 177         f2fs_target_device(sbi, blk_addr, bio);
 178         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
 179         bio->bi_private = is_read ? NULL : sbi;
 180
 181         return bio;
 182 }
 183
 184 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 185                                 struct bio *bio, enum page_type type)
 186 {
 187         if (!is_read_io(bio_op(bio))) {
 188                 unsigned int start;
 189
 190                 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
 191                         current->plug && (type == DATA || type == NODE))
 192                         blk_finish_plug(current->plug);
 193
 194                 if (type != DATA && type != NODE)
 195                         goto submit_io;
 196
 197                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 198                 start %= F2FS_IO_SIZE(sbi);
 199
 200                 if (start == 0)
 201                         goto submit_io;
 202
 203                 /* fill dummy pages */
 204                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 205                         struct page *page =
 206                                 mempool_alloc(sbi->write_io_dummy,
 207                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 208                         f2fs_bug_on(sbi, !page);
 209
 210                         SetPagePrivate(page);
 211                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 212                         lock_page(page);
 213                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 214                                 f2fs_bug_on(sbi, 1);
 215                 }
 216                 /*
 217                  * In the NODE case, we lose next block address chain. So, we
 218                  * need to do checkpoint in f2fs_sync_file.
 219                  */
 220                 if (type == NODE)
 221                         set_sbi_flag(sbi, SBI_NEED_CP);
 222         }
 223 submit_io:
 224         if (is_read_io(bio_op(bio)))
 225                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 226         else
 227                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 228         submit_bio(bio);
 229 }
 230
 231 static void __submit_merged_bio(struct f2fs_bio_info *io)
 232 {
 233         struct f2fs_io_info *fio = &io->fio;
 234
 235         if (!io->bio)
 236                 return;
 237
 238         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 239
 240         if (is_read_io(fio->op))
 241                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 242         else
 243                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 244
 245         __submit_bio(io->sbi, io->bio, fio->type);
 246         io->bio = NULL;
 247 }
 248
 249 static bool __has_merged_page(struct f2fs_bio_info *io,
 250                                 struct inode *inode, nid_t ino, pgoff_t idx)
 251 {
 252         struct bio_vec *bvec;
 253         struct page *target;
 254         int i;
 255
 256         if (!io->bio)
 257                 return false;
 258
 259         if (!inode && !ino)
 260                 return true;
 261
 262         bio_for_each_segment_all(bvec, io->bio, i) {
 263
 264                 if (bvec->bv_page->mapping)
 265                         target = bvec->bv_page;
 266                 else
 267                         target = fscrypt_control_page(bvec->bv_page);
 268
 269                 if (idx != target->index)
 270                         continue;
 271
 272                 if (inode && inode == target->mapping->host)
 273                         return true;
 274                 if (ino && ino == ino_of_node(target))
 275                         return true;
 276         }
 277
 278         return false;
 279 }
 280
 281 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 282                                 nid_t ino, pgoff_t idx, enum page_type type)
 283 {
 284         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 285         struct f2fs_bio_info *io = &sbi->write_io[btype];
 286         bool ret;
 287
 288         down_read(&io->io_rwsem);
 289         ret = __has_merged_page(io, inode, ino, idx);
 290         up_read(&io->io_rwsem);
 291         return ret;
 292 }
 293
 294 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
 295                                 struct inode *inode, nid_t ino, pgoff_t idx,
 296                                 enum page_type type, int rw)
 297 {
 298         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 299         struct f2fs_bio_info *io;
 300
 301         io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
 302
 303         down_write(&io->io_rwsem);
 304
 305         if (!__has_merged_page(io, inode, ino, idx))
 306                 goto out;
 307
 308         /* change META to META_FLUSH in the checkpoint procedure */
 309         if (type >= META_FLUSH) {
 310                 io->fio.type = META_FLUSH;
 311                 io->fio.op = REQ_OP_WRITE;
 312                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 313                 if (!test_opt(sbi, NOBARRIER))
 314                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 315         }
 316         __submit_merged_bio(io);
 317 out:
 318         up_write(&io->io_rwsem);
 319 }
 320
 321 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
 322                                                                         int rw)
 323 {
 324         __f2fs_submit_merged_bio(sbi, NULL, 0, 0, type, rw);
 325 }
 326
 327 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
 328                                 struct inode *inode, nid_t ino, pgoff_t idx,
 329                                 enum page_type type, int rw)
 330 {
 331         if (has_merged_page(sbi, inode, ino, idx, type))
 332                 __f2fs_submit_merged_bio(sbi, inode, ino, idx, type, rw);
 333 }
 334
 335 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
 336 {
 337         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 338         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 339         f2fs_submit_merged_bio(sbi, META, WRITE);
 340 }
 341
 342 /*
 343  * Fill the locked page with data located in the block address.
 344  * A caller needs to unlock the page on failure.
 345  */
 346 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 347 {
 348         struct bio *bio;
 349         struct page *page = fio->encrypted_page ?
 350                         fio->encrypted_page : fio->page;
 351
 352         trace_f2fs_submit_page_bio(page, fio);
 353         f2fs_trace_ios(fio, 0);
 354
 355         /* Allocate a new bio */
 356         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
 357
 358         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 359                 bio_put(bio);
 360                 return -EFAULT;
 361         }
 362         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 363
 364         __submit_bio(fio->sbi, bio, fio->type);
 365
 366         if (!is_read_io(fio->op))
 367                 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 368         return 0;
 369 }
 370
 371 int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
 372 {
 373         struct f2fs_sb_info *sbi = fio->sbi;
 374         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 375         struct f2fs_bio_info *io;
 376         bool is_read = is_read_io(fio->op);
 377         struct page *bio_page;
 378         int err = 0;
 379
 380         io = is_read ? &sbi->read_io : &sbi->write_io[btype];
 381
 382         if (fio->old_blkaddr != NEW_ADDR)
 383                 verify_block_addr(sbi, fio->old_blkaddr);
 384         verify_block_addr(sbi, fio->new_blkaddr);
 385
 386         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 387
 388         /* set submitted = 1 as a return value */
 389         fio->submitted = 1;
 390
 391         if (!is_read)
 392                 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 393
 394         down_write(&io->io_rwsem);
 395
 396         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 397             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 398                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 399                 __submit_merged_bio(io);
 400 alloc_new:
 401         if (io->bio == NULL) {
 402                 if ((fio->type == DATA || fio->type == NODE) &&
 403                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 404                         err = -EAGAIN;
 405                         if (!is_read)
 406                                 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 407                         goto out_fail;
 408                 }
 409                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 410                                                 BIO_MAX_PAGES, is_read);
 411                 io->fio = *fio;
 412         }
 413
 414         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
 415                                                         PAGE_SIZE) {
 416                 __submit_merged_bio(io);
 417                 goto alloc_new;
 418         }
 419
 420         io->last_block_in_bio = fio->new_blkaddr;
 421         f2fs_trace_ios(fio, 0);
 422 out_fail:
 423         up_write(&io->io_rwsem);
 424         trace_f2fs_submit_page_mbio(fio->page, fio);
 425         return err;
 426 }
 427
 428 static void __set_data_blkaddr(struct dnode_of_data *dn)
 429 {
 430         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 431         __le32 *addr_array;
 432
 433         /* Get physical address of data block */
 434         addr_array = blkaddr_in_node(rn);
 435         addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 436 }
 437
 438 /*
 439  * Lock ordering for the change of data block address:
 440  * ->data_page
 441  *  ->node_page
 442  *    update block addresses in the node page
 443  */
 444 void set_data_blkaddr(struct dnode_of_data *dn)
 445 {
 446         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 447         __set_data_blkaddr(dn);
 448         if (set_page_dirty(dn->node_page))
 449                 dn->node_changed = true;
 450 }
 451
 452 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 453 {
 454         dn->data_blkaddr = blkaddr;
 455         set_data_blkaddr(dn);
 456         f2fs_update_extent_cache(dn);
 457 }
 458
 459 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 460 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 461 {
 462         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 463
 464         if (!count)
 465                 return 0;
 466
 467         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 468                 return -EPERM;
 469         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 470                 return -ENOSPC;
 471
 472         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 473                                                 dn->ofs_in_node, count);
 474
 475         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 476
 477         for (; count > 0; dn->ofs_in_node++) {
 478                 block_t blkaddr =
 479                         datablock_addr(dn->node_page, dn->ofs_in_node);
 480                 if (blkaddr == NULL_ADDR) {
 481                         dn->data_blkaddr = NEW_ADDR;
 482                         __set_data_blkaddr(dn);
 483                         count--;
 484                 }
 485         }
 486
 487         if (set_page_dirty(dn->node_page))
 488                 dn->node_changed = true;
 489         return 0;
 490 }
 491
 492 /* Should keep dn->ofs_in_node unchanged */
 493 int reserve_new_block(struct dnode_of_data *dn)
 494 {
 495         unsigned int ofs_in_node = dn->ofs_in_node;
 496         int ret;
 497
 498         ret = reserve_new_blocks(dn, 1);
 499         dn->ofs_in_node = ofs_in_node;
 500         return ret;
 501 }
 502
 503 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 504 {
 505         bool need_put = dn->inode_page ? false : true;
 506         int err;
 507
 508         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 509         if (err)
 510                 return err;
 511
 512         if (dn->data_blkaddr == NULL_ADDR)
 513                 err = reserve_new_block(dn);
 514         if (err || need_put)
 515                 f2fs_put_dnode(dn);
 516         return err;
 517 }
 518
 519 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 520 {
 521         struct extent_info ei  = {0,0,0};
 522         struct inode *inode = dn->inode;
 523
 524         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 525                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 526                 return 0;
 527         }
 528
 529         return f2fs_reserve_block(dn, index);
 530 }
 531
 532 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 533                                                 int op_flags, bool for_write)
 534 {
 535         struct address_space *mapping = inode->i_mapping;
 536         struct dnode_of_data dn;
 537         struct page *page;
 538         struct extent_info ei = {0,0,0};
 539         int err;
 540         struct f2fs_io_info fio = {
 541                 .sbi = F2FS_I_SB(inode),
 542                 .type = DATA,
 543                 .op = REQ_OP_READ,
 544                 .op_flags = op_flags,
 545                 .encrypted_page = NULL,
 546         };
 547
 548         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
 549                 return read_mapping_page(mapping, index, NULL);
 550
 551         page = f2fs_grab_cache_page(mapping, index, for_write);
 552         if (!page)
 553                 return ERR_PTR(-ENOMEM);
 554
 555         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 556                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 557                 goto got_it;
 558         }
 559
 560         set_new_dnode(&dn, inode, NULL, NULL, 0);
 561         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 562         if (err)
 563                 goto put_err;
 564         f2fs_put_dnode(&dn);
 565
 566         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 567                 err = -ENOENT;
 568                 goto put_err;
 569         }
 570 got_it:
 571         if (PageUptodate(page)) {
 572                 unlock_page(page);
 573                 return page;
 574         }
 575
 576         /*
 577          * A new dentry page is allocated but not able to be written, since its
 578          * new inode page couldn't be allocated due to -ENOSPC.
 579          * In such the case, its blkaddr can be remained as NEW_ADDR.
 580          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 581          */
 582         if (dn.data_blkaddr == NEW_ADDR) {
 583                 zero_user_segment(page, 0, PAGE_SIZE);
 584                 if (!PageUptodate(page))
 585                         SetPageUptodate(page);
 586                 unlock_page(page);
 587                 return page;
 588         }
 589
 590         fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 591         fio.page = page;
 592         err = f2fs_submit_page_bio(&fio);
 593         if (err)
 594                 goto put_err;
 595         return page;
 596
 597 put_err:
 598         f2fs_put_page(page, 1);
 599         return ERR_PTR(err);
 600 }
 601
 602 struct page *find_data_page(struct inode *inode, pgoff_t index)
 603 {
 604         struct address_space *mapping = inode->i_mapping;
 605         struct page *page;
 606
 607         page = find_get_page(mapping, index);
 608         if (page && PageUptodate(page))
 609                 return page;
 610         f2fs_put_page(page, 0);
 611
 612         page = get_read_data_page(inode, index, 0, false);
 613         if (IS_ERR(page))
 614                 return page;
 615
 616         if (PageUptodate(page))
 617                 return page;
 618
 619         wait_on_page_locked(page);
 620         if (unlikely(!PageUptodate(page))) {
 621                 f2fs_put_page(page, 0);
 622                 return ERR_PTR(-EIO);
 623         }
 624         return page;
 625 }
 626
 627 /*
 628  * If it tries to access a hole, return an error.
 629  * Because, the callers, functions in dir.c and GC, should be able to know
 630  * whether this page exists or not.
 631  */
 632 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 633                                                         bool for_write)
 634 {
 635         struct address_space *mapping = inode->i_mapping;
 636         struct page *page;
 637 repeat:
 638         page = get_read_data_page(inode, index, 0, for_write);
 639         if (IS_ERR(page))
 640                 return page;
 641
 642         /* wait for read completion */
 643         lock_page(page);
 644         if (unlikely(page->mapping != mapping)) {
 645                 f2fs_put_page(page, 1);
 646                 goto repeat;
 647         }
 648         if (unlikely(!PageUptodate(page))) {
 649                 f2fs_put_page(page, 1);
 650                 return ERR_PTR(-EIO);
 651         }
 652         return page;
 653 }
 654
 655 /*
 656  * Caller ensures that this data page is never allocated.
 657  * A new zero-filled data page is allocated in the page cache.
 658  *
 659  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 660  * f2fs_unlock_op().
 661  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 662  * ipage should be released by this function.
 663  */
 664 struct page *get_new_data_page(struct inode *inode,
 665                 struct page *ipage, pgoff_t index, bool new_i_size)
 666 {
 667         struct address_space *mapping = inode->i_mapping;
 668         struct page *page;
 669         struct dnode_of_data dn;
 670         int err;
 671
 672         page = f2fs_grab_cache_page(mapping, index, true);
 673         if (!page) {
 674                 /*
 675                  * before exiting, we should make sure ipage will be released
 676                  * if any error occur.
 677                  */
 678                 f2fs_put_page(ipage, 1);
 679                 return ERR_PTR(-ENOMEM);
 680         }
 681
 682         set_new_dnode(&dn, inode, ipage, NULL, 0);
 683         err = f2fs_reserve_block(&dn, index);
 684         if (err) {
 685                 f2fs_put_page(page, 1);
 686                 return ERR_PTR(err);
 687         }
 688         if (!ipage)
 689                 f2fs_put_dnode(&dn);
 690
 691         if (PageUptodate(page))
 692                 goto got_it;
 693
 694         if (dn.data_blkaddr == NEW_ADDR) {
 695                 zero_user_segment(page, 0, PAGE_SIZE);
 696                 if (!PageUptodate(page))
 697                         SetPageUptodate(page);
 698         } else {
 699                 f2fs_put_page(page, 1);
 700
 701                 /* if ipage exists, blkaddr should be NEW_ADDR */
 702                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 703                 page = get_lock_data_page(inode, index, true);
 704                 if (IS_ERR(page))
 705                         return page;
 706         }
 707 got_it:
 708         if (new_i_size && i_size_read(inode) <
 709                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 710                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 711         return page;
 712 }
 713
 714 static int __allocate_data_block(struct dnode_of_data *dn)
 715 {
 716         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 717         struct f2fs_summary sum;
 718         struct node_info ni;
 719         pgoff_t fofs;
 720         blkcnt_t count = 1;
 721
 722         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 723                 return -EPERM;
 724
 725         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 726         if (dn->data_blkaddr == NEW_ADDR)
 727                 goto alloc;
 728
 729         if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
 730                 return -ENOSPC;
 731
 732 alloc:
 733         get_node_info(sbi, dn->nid, &ni);
 734         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 735
 736         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 737                                                 &sum, CURSEG_WARM_DATA);
 738         set_data_blkaddr(dn);
 739
 740         /* update i_size */
 741         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 742                                                         dn->ofs_in_node;
 743         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 744                 f2fs_i_size_write(dn->inode,
 745                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 746         return 0;
 747 }
 748
 749 static inline bool __force_buffered_io(struct inode *inode, int rw)
 750 {
 751         return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) ||
 752                         (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
 753                         F2FS_I_SB(inode)->s_ndevs);
 754 }
 755
 756 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 757 {
 758         struct inode *inode = file_inode(iocb->ki_filp);
 759         struct f2fs_map_blocks map;
 760         int err = 0;
 761
 762         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 763                 return 0;
 764
 765         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 766         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 767         if (map.m_len > map.m_lblk)
 768                 map.m_len -= map.m_lblk;
 769         else
 770                 map.m_len = 0;
 771
 772         map.m_next_pgofs = NULL;
 773
 774         if (iocb->ki_flags & IOCB_DIRECT) {
 775                 err = f2fs_convert_inline_inode(inode);
 776                 if (err)
 777                         return err;
 778                 return f2fs_map_blocks(inode, &map, 1,
 779                         __force_buffered_io(inode, WRITE) ?
 780                                 F2FS_GET_BLOCK_PRE_AIO :
 781                                 F2FS_GET_BLOCK_PRE_DIO);
 782         }
 783         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
 784                 err = f2fs_convert_inline_inode(inode);
 785                 if (err)
 786                         return err;
 787         }
 788         if (!f2fs_has_inline_data(inode))
 789                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 790         return err;
 791 }
 792
 793 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 794 {
 795         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 796                 if (lock)
 797                         down_read(&sbi->node_change);
 798                 else
 799                         up_read(&sbi->node_change);
 800         } else {
 801                 if (lock)
 802                         f2fs_lock_op(sbi);
 803                 else
 804                         f2fs_unlock_op(sbi);
 805         }
 806 }
 807
 808 /*
 809  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 810  * f2fs_map_blocks structure.
 811  * If original data blocks are allocated, then give them to blockdev.
 812  * Otherwise,
 813  *     a. preallocate requested block addresses
 814  *     b. do not use extent cache for better performance
 815  *     c. give the block addresses to blockdev
 816  */
 817 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 818                                                 int create, int flag)
 819 {
 820         unsigned int maxblocks = map->m_len;
 821         struct dnode_of_data dn;
 822         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 823         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 824         pgoff_t pgofs, end_offset, end;
 825         int err = 0, ofs = 1;
 826         unsigned int ofs_in_node, last_ofs_in_node;
 827         blkcnt_t prealloc;
 828         struct extent_info ei = {0,0,0};
 829         block_t blkaddr;
 830
 831         if (!maxblocks)
 832                 return 0;
 833
 834         map->m_len = 0;
 835         map->m_flags = 0;
 836
 837         /* it only supports block size == page size */
 838         pgofs = (pgoff_t)map->m_lblk;
 839         end = pgofs + maxblocks;
 840
 841         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 842                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 843                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 844                 map->m_flags = F2FS_MAP_MAPPED;
 845                 goto out;
 846         }
 847
 848 next_dnode:
 849         if (create)
 850                 __do_map_lock(sbi, flag, true);
 851
 852         /* When reading holes, we need its node page */
 853         set_new_dnode(&dn, inode, NULL, NULL, 0);
 854         err = get_dnode_of_data(&dn, pgofs, mode);
 855         if (err) {
 856                 if (flag == F2FS_GET_BLOCK_BMAP)
 857                         map->m_pblk = 0;
 858                 if (err == -ENOENT) {
 859                         err = 0;
 860                         if (map->m_next_pgofs)
 861                                 *map->m_next_pgofs =
 862                                         get_next_page_offset(&dn, pgofs);
 863                 }
 864                 goto unlock_out;
 865         }
 866
 867         prealloc = 0;
 868         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
 869         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 870
 871 next_block:
 872         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 873
 874         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 875                 if (create) {
 876                         if (unlikely(f2fs_cp_error(sbi))) {
 877                                 err = -EIO;
 878                                 goto sync_out;
 879                         }
 880                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 881                                 if (blkaddr == NULL_ADDR) {
 882                                         prealloc++;
 883                                         last_ofs_in_node = dn.ofs_in_node;
 884                                 }
 885                         } else {
 886                                 err = __allocate_data_block(&dn);
 887                                 if (!err)
 888                                         set_inode_flag(inode, FI_APPEND_WRITE);
 889                         }
 890                         if (err)
 891                                 goto sync_out;
 892                         map->m_flags |= F2FS_MAP_NEW;
 893                         blkaddr = dn.data_blkaddr;
 894                 } else {
 895                         if (flag == F2FS_GET_BLOCK_BMAP) {
 896                                 map->m_pblk = 0;
 897                                 goto sync_out;
 898                         }
 899                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 900                                                 blkaddr == NULL_ADDR) {
 901                                 if (map->m_next_pgofs)
 902                                         *map->m_next_pgofs = pgofs + 1;
 903                         }
 904                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 905                                                 blkaddr != NEW_ADDR)
 906                                 goto sync_out;
 907                 }
 908         }
 909
 910         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 911                 goto skip;
 912
 913         if (map->m_len == 0) {
 914                 /* preallocated unwritten block should be mapped for fiemap. */
 915                 if (blkaddr == NEW_ADDR)
 916                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 917                 map->m_flags |= F2FS_MAP_MAPPED;
 918
 919                 map->m_pblk = blkaddr;
 920                 map->m_len = 1;
 921         } else if ((map->m_pblk != NEW_ADDR &&
 922                         blkaddr == (map->m_pblk + ofs)) ||
 923                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 924                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 925                 ofs++;
 926                 map->m_len++;
 927         } else {
 928                 goto sync_out;
 929         }
 930
 931 skip:
 932         dn.ofs_in_node++;
 933         pgofs++;
 934
 935         /* preallocate blocks in batch for one dnode page */
 936         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
 937                         (pgofs == end || dn.ofs_in_node == end_offset)) {
 938
 939                 dn.ofs_in_node = ofs_in_node;
 940                 err = reserve_new_blocks(&dn, prealloc);
 941                 if (err)
 942                         goto sync_out;
 943
 944                 map->m_len += dn.ofs_in_node - ofs_in_node;
 945                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
 946                         err = -ENOSPC;
 947                         goto sync_out;
 948                 }
 949                 dn.ofs_in_node = end_offset;
 950         }
 951
 952         if (pgofs >= end)
 953                 goto sync_out;
 954         else if (dn.ofs_in_node < end_offset)
 955                 goto next_block;
 956
 957         f2fs_put_dnode(&dn);
 958
 959         if (create) {
 960                 __do_map_lock(sbi, flag, false);
 961                 f2fs_balance_fs(sbi, dn.node_changed);
 962         }
 963         goto next_dnode;
 964
 965 sync_out:
 966         f2fs_put_dnode(&dn);
 967 unlock_out:
 968         if (create) {
 969                 __do_map_lock(sbi, flag, false);
 970                 f2fs_balance_fs(sbi, dn.node_changed);
 971         }
 972 out:
 973         trace_f2fs_map_blocks(inode, map, err);
 974         return err;
 975 }
 976
 977 static int __get_data_block(struct inode *inode, sector_t iblock,
 978                         struct buffer_head *bh, int create, int flag,
 979                         pgoff_t *next_pgofs)
 980 {
 981         struct f2fs_map_blocks map;
 982         int err;
 983
 984         map.m_lblk = iblock;
 985         map.m_len = bh->b_size >> inode->i_blkbits;
 986         map.m_next_pgofs = next_pgofs;
 987
 988         err = f2fs_map_blocks(inode, &map, create, flag);
 989         if (!err) {
 990                 map_bh(bh, inode->i_sb, map.m_pblk);
 991                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
 992                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
 993         }
 994         return err;
 995 }
 996
 997 static int get_data_block(struct inode *inode, sector_t iblock,
 998                         struct buffer_head *bh_result, int create, int flag,
 999                         pgoff_t *next_pgofs)
1000 {
1001         return __get_data_block(inode, iblock, bh_result, create,
1002                                                         flag, next_pgofs);
1003 }
1004
1005 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1006                         struct buffer_head *bh_result, int create)
1007 {
1008         return __get_data_block(inode, iblock, bh_result, create,
1009                                                 F2FS_GET_BLOCK_DIO, NULL);
1010 }
1011
1012 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1013                         struct buffer_head *bh_result, int create)
1014 {
1015         /* Block number less than F2FS MAX BLOCKS */
1016         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1017                 return -EFBIG;
1018
1019         return __get_data_block(inode, iblock, bh_result, create,
1020                                                 F2FS_GET_BLOCK_BMAP, NULL);
1021 }
1022
1023 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1024 {
1025         return (offset >> inode->i_blkbits);
1026 }
1027
1028 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1029 {
1030         return (blk << inode->i_blkbits);
1031 }
1032
1033 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1034                 u64 start, u64 len)
1035 {
1036         struct buffer_head map_bh;
1037         sector_t start_blk, last_blk;
1038         pgoff_t next_pgofs;
1039         u64 logical = 0, phys = 0, size = 0;
1040         u32 flags = 0;
1041         int ret = 0;
1042
1043         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1044         if (ret)
1045                 return ret;
1046
1047         if (f2fs_has_inline_data(inode)) {
1048                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1049                 if (ret != -EAGAIN)
1050                         return ret;
1051         }
1052
1053         inode_lock(inode);
1054
1055         if (logical_to_blk(inode, len) == 0)
1056                 len = blk_to_logical(inode, 1);
1057
1058         start_blk = logical_to_blk(inode, start);
1059         last_blk = logical_to_blk(inode, start + len - 1);
1060
1061 next:
1062         memset(&map_bh, 0, sizeof(struct buffer_head));
1063         map_bh.b_size = len;
1064
1065         ret = get_data_block(inode, start_blk, &map_bh, 0,
1066                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1067         if (ret)
1068                 goto out;
1069
1070         /* HOLE */
1071         if (!buffer_mapped(&map_bh)) {
1072                 start_blk = next_pgofs;
1073
1074                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1075                                         F2FS_I_SB(inode)->max_file_blocks))
1076                         goto prep_next;
1077
1078                 flags |= FIEMAP_EXTENT_LAST;
1079         }
1080
1081         if (size) {
1082                 if (f2fs_encrypted_inode(inode))
1083                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1084
1085                 ret = fiemap_fill_next_extent(fieinfo, logical,
1086                                 phys, size, flags);
1087         }
1088
1089         if (start_blk > last_blk || ret)
1090                 goto out;
1091
1092         logical = blk_to_logical(inode, start_blk);
1093         phys = blk_to_logical(inode, map_bh.b_blocknr);
1094         size = map_bh.b_size;
1095         flags = 0;
1096         if (buffer_unwritten(&map_bh))
1097                 flags = FIEMAP_EXTENT_UNWRITTEN;
1098
1099         start_blk += logical_to_blk(inode, size);
1100
1101 prep_next:
1102         cond_resched();
1103         if (fatal_signal_pending(current))
1104                 ret = -EINTR;
1105         else
1106                 goto next;
1107 out:
1108         if (ret == 1)
1109                 ret = 0;
1110
1111         inode_unlock(inode);
1112         return ret;
1113 }
1114
1115 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1116                                  unsigned nr_pages)
1117 {
1118         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1119         struct fscrypt_ctx *ctx = NULL;
1120         struct bio *bio;
1121
1122         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1123                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1124                 if (IS_ERR(ctx))
1125                         return ERR_CAST(ctx);
1126
1127                 /* wait the page to be moved by cleaning */
1128                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1129         }
1130
1131         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1132         if (!bio) {
1133                 if (ctx)
1134                         fscrypt_release_ctx(ctx);
1135                 return ERR_PTR(-ENOMEM);
1136         }
1137         f2fs_target_device(sbi, blkaddr, bio);
1138         bio->bi_end_io = f2fs_read_end_io;
1139         bio->bi_private = ctx;
1140
1141         return bio;
1142 }
1143
1144 /*
1145  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1146  * Major change was from block_size == page_size in f2fs by default.
1147  */
1148 static int f2fs_mpage_readpages(struct address_space *mapping,
1149                         struct list_head *pages, struct page *page,
1150                         unsigned nr_pages)
1151 {
1152         struct bio *bio = NULL;
1153         unsigned page_idx;
1154         sector_t last_block_in_bio = 0;
1155         struct inode *inode = mapping->host;
1156         const unsigned blkbits = inode->i_blkbits;
1157         const unsigned blocksize = 1 << blkbits;
1158         sector_t block_in_file;
1159         sector_t last_block;
1160         sector_t last_block_in_file;
1161         sector_t block_nr;
1162         struct f2fs_map_blocks map;
1163
1164         map.m_pblk = 0;
1165         map.m_lblk = 0;
1166         map.m_len = 0;
1167         map.m_flags = 0;
1168         map.m_next_pgofs = NULL;
1169
1170         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1171
1172                 if (pages) {
1173                         page = list_last_entry(pages, struct page, lru);
1174
1175                         prefetchw(&page->flags);
1176                         list_del(&page->lru);
1177                         if (add_to_page_cache_lru(page, mapping,
1178                                                   page->index,
1179                                                   readahead_gfp_mask(mapping)))
1180                                 goto next_page;
1181                 }
1182
1183                 block_in_file = (sector_t)page->index;
1184                 last_block = block_in_file + nr_pages;
1185                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1186                                                                 blkbits;
1187                 if (last_block > last_block_in_file)
1188                         last_block = last_block_in_file;
1189
1190                 /*
1191                  * Map blocks using the previous result first.
1192                  */
1193                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1194                                 block_in_file > map.m_lblk &&
1195                                 block_in_file < (map.m_lblk + map.m_len))
1196                         goto got_it;
1197
1198                 /*
1199                  * Then do more f2fs_map_blocks() calls until we are
1200                  * done with this page.
1201                  */
1202                 map.m_flags = 0;
1203
1204                 if (block_in_file < last_block) {
1205                         map.m_lblk = block_in_file;
1206                         map.m_len = last_block - block_in_file;
1207
1208                         if (f2fs_map_blocks(inode, &map, 0,
1209                                                 F2FS_GET_BLOCK_READ))
1210                                 goto set_error_page;
1211                 }
1212 got_it:
1213                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1214                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1215                         SetPageMappedToDisk(page);
1216
1217                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1218                                 SetPageUptodate(page);
1219                                 goto confused;
1220                         }
1221                 } else {
1222                         zero_user_segment(page, 0, PAGE_SIZE);
1223                         if (!PageUptodate(page))
1224                                 SetPageUptodate(page);
1225                         unlock_page(page);
1226                         goto next_page;
1227                 }
1228
1229                 /*
1230                  * This page will go to BIO.  Do we need to send this
1231                  * BIO off first?
1232                  */
1233                 if (bio && (last_block_in_bio != block_nr - 1 ||
1234                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1235 submit_and_realloc:
1236                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1237                         bio = NULL;
1238                 }
1239                 if (bio == NULL) {
1240                         bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1241                         if (IS_ERR(bio)) {
1242                                 bio = NULL;
1243                                 goto set_error_page;
1244                         }
1245                         bio_set_op_attrs(bio, REQ_OP_READ, 0);
1246                 }
1247
1248                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1249                         goto submit_and_realloc;
1250
1251                 last_block_in_bio = block_nr;
1252                 goto next_page;
1253 set_error_page:
1254                 SetPageError(page);
1255                 zero_user_segment(page, 0, PAGE_SIZE);
1256                 unlock_page(page);
1257                 goto next_page;
1258 confused:
1259                 if (bio) {
1260                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1261                         bio = NULL;
1262                 }
1263                 unlock_page(page);
1264 next_page:
1265                 if (pages)
1266                         put_page(page);
1267         }
1268         BUG_ON(pages && !list_empty(pages));
1269         if (bio)
1270                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1271         return 0;
1272 }
1273
1274 static int f2fs_read_data_page(struct file *file, struct page *page)
1275 {
1276         struct inode *inode = page->mapping->host;
1277         int ret = -EAGAIN;
1278
1279         trace_f2fs_readpage(page, DATA);
1280
1281         /* If the file has inline data, try to read it directly */
1282         if (f2fs_has_inline_data(inode))
1283                 ret = f2fs_read_inline_data(inode, page);
1284         if (ret == -EAGAIN)
1285                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1286         return ret;
1287 }
1288
1289 static int f2fs_read_data_pages(struct file *file,
1290                         struct address_space *mapping,
1291                         struct list_head *pages, unsigned nr_pages)
1292 {
1293         struct inode *inode = file->f_mapping->host;
1294         struct page *page = list_last_entry(pages, struct page, lru);
1295
1296         trace_f2fs_readpages(inode, page, nr_pages);
1297
1298         /* If the file has inline data, skip readpages */
1299         if (f2fs_has_inline_data(inode))
1300                 return 0;
1301
1302         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1303 }
1304
1305 static int encrypt_one_page(struct f2fs_io_info *fio)
1306 {
1307         struct inode *inode = fio->page->mapping->host;
1308         gfp_t gfp_flags = GFP_NOFS;
1309
1310         if (!f2fs_encrypted_inode(inode) || !S_ISREG(inode->i_mode))
1311                 return 0;
1312
1313         /* wait for GCed encrypted page writeback */
1314         f2fs_wait_on_encrypted_page_writeback(fio->sbi, fio->old_blkaddr);
1315
1316 retry_encrypt:
1317         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1318                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1319         if (!IS_ERR(fio->encrypted_page))
1320                 return 0;
1321
1322         /* flush pending IOs and wait for a while in the ENOMEM case */
1323         if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1324                 f2fs_flush_merged_bios(fio->sbi);
1325                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1326                 gfp_flags |= __GFP_NOFAIL;
1327                 goto retry_encrypt;
1328         }
1329         return PTR_ERR(fio->encrypted_page);
1330 }
1331
1332 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1333 {
1334         struct inode *inode = fio->page->mapping->host;
1335
1336         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1337                 return false;
1338         if (is_cold_data(fio->page))
1339                 return false;
1340         if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1341                 return false;
1342
1343         return need_inplace_update_policy(inode, fio);
1344 }
1345
1346 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1347 {
1348         if (fio->old_blkaddr == NEW_ADDR)
1349                 return false;
1350         if (fio->old_blkaddr == NULL_ADDR)
1351                 return false;
1352         return true;
1353 }
1354
1355 int do_write_data_page(struct f2fs_io_info *fio)
1356 {
1357         struct page *page = fio->page;
1358         struct inode *inode = page->mapping->host;
1359         struct dnode_of_data dn;
1360         struct extent_info ei = {0,0,0};
1361         bool ipu_force = false;
1362         int err = 0;
1363
1364         set_new_dnode(&dn, inode, NULL, NULL, 0);
1365         if (need_inplace_update(fio) &&
1366                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1367                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1368
1369                 if (valid_ipu_blkaddr(fio)) {
1370                         ipu_force = true;
1371                         fio->need_lock = false;
1372                         goto got_it;
1373                 }
1374         }
1375
1376         if (fio->need_lock)
1377                 f2fs_lock_op(fio->sbi);
1378
1379         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1380         if (err)
1381                 goto out;
1382
1383         fio->old_blkaddr = dn.data_blkaddr;
1384
1385         /* This page is already truncated */
1386         if (fio->old_blkaddr == NULL_ADDR) {
1387                 ClearPageUptodate(page);
1388                 goto out_writepage;
1389         }
1390 got_it:
1391         err = encrypt_one_page(fio);
1392         if (err)
1393                 goto out_writepage;
1394
1395         set_page_writeback(page);
1396
1397         /*
1398          * If current allocation needs SSR,
1399          * it had better in-place writes for updated data.
1400          */
1401         if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1402                 f2fs_put_dnode(&dn);
1403                 if (fio->need_lock)
1404                         f2fs_unlock_op(fio->sbi);
1405                 err = rewrite_data_page(fio);
1406                 trace_f2fs_do_write_data_page(fio->page, IPU);
1407                 set_inode_flag(inode, FI_UPDATE_WRITE);
1408                 return err;
1409         }
1410
1411         /* LFS mode write path */
1412         write_data_page(&dn, fio);
1413         trace_f2fs_do_write_data_page(page, OPU);
1414         set_inode_flag(inode, FI_APPEND_WRITE);
1415         if (page->index == 0)
1416                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1417 out_writepage:
1418         f2fs_put_dnode(&dn);
1419 out:
1420         if (fio->need_lock)
1421                 f2fs_unlock_op(fio->sbi);
1422         return err;
1423 }
1424
1425 static int __write_data_page(struct page *page, bool *submitted,
1426                                 struct writeback_control *wbc)
1427 {
1428         struct inode *inode = page->mapping->host;
1429         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1430         loff_t i_size = i_size_read(inode);
1431         const pgoff_t end_index = ((unsigned long long) i_size)
1432                                                         >> PAGE_SHIFT;
1433         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1434         unsigned offset = 0;
1435         bool need_balance_fs = false;
1436         int err = 0;
1437         struct f2fs_io_info fio = {
1438                 .sbi = sbi,
1439                 .type = DATA,
1440                 .op = REQ_OP_WRITE,
1441                 .op_flags = wbc_to_write_flags(wbc),
1442                 .old_blkaddr = NULL_ADDR,
1443                 .page = page,
1444                 .encrypted_page = NULL,
1445                 .submitted = false,
1446                 .need_lock = true,
1447         };
1448
1449         trace_f2fs_writepage(page, DATA);
1450
1451         if (page->index < end_index)
1452                 goto write;
1453
1454         /*
1455          * If the offset is out-of-range of file size,
1456          * this page does not have to be written to disk.
1457          */
1458         offset = i_size & (PAGE_SIZE - 1);
1459         if ((page->index >= end_index + 1) || !offset)
1460                 goto out;
1461
1462         zero_user_segment(page, offset, PAGE_SIZE);
1463 write:
1464         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1465                 goto redirty_out;
1466         if (f2fs_is_drop_cache(inode))
1467                 goto out;
1468         /* we should not write 0'th page having journal header */
1469         if (f2fs_is_volatile_file(inode) && (!page->index ||
1470                         (!wbc->for_reclaim &&
1471                         available_free_memory(sbi, BASE_CHECK))))
1472                 goto redirty_out;
1473
1474         /* we should bypass data pages to proceed the kworkder jobs */
1475         if (unlikely(f2fs_cp_error(sbi))) {
1476                 mapping_set_error(page->mapping, -EIO);
1477                 goto out;
1478         }
1479
1480         /* Dentry blocks are controlled by checkpoint */
1481         if (S_ISDIR(inode->i_mode)) {
1482                 fio.need_lock = false;
1483                 err = do_write_data_page(&fio);
1484                 goto done;
1485         }
1486
1487         if (!wbc->for_reclaim)
1488                 need_balance_fs = true;
1489         else if (has_not_enough_free_secs(sbi, 0, 0))
1490                 goto redirty_out;
1491         else
1492                 set_inode_flag(inode, FI_HOT_DATA);
1493
1494         err = -EAGAIN;
1495         if (f2fs_has_inline_data(inode)) {
1496                 err = f2fs_write_inline_data(inode, page);
1497                 if (!err)
1498                         goto out;
1499         }
1500
1501         if (err == -EAGAIN)
1502                 err = do_write_data_page(&fio);
1503         if (F2FS_I(inode)->last_disk_size < psize)
1504                 F2FS_I(inode)->last_disk_size = psize;
1505
1506 done:
1507         if (err && err != -ENOENT)
1508                 goto redirty_out;
1509
1510 out:
1511         inode_dec_dirty_pages(inode);
1512         if (err)
1513                 ClearPageUptodate(page);
1514
1515         if (wbc->for_reclaim) {
1516                 f2fs_submit_merged_bio_cond(sbi, inode, 0, page->index,
1517                                                 DATA, WRITE);
1518                 clear_inode_flag(inode, FI_HOT_DATA);
1519                 remove_dirty_inode(inode);
1520                 submitted = NULL;
1521         }
1522
1523         unlock_page(page);
1524         if (!S_ISDIR(inode->i_mode))
1525                 f2fs_balance_fs(sbi, need_balance_fs);
1526
1527         if (unlikely(f2fs_cp_error(sbi))) {
1528                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1529                 submitted = NULL;
1530         }
1531
1532         if (submitted)
1533                 *submitted = fio.submitted;
1534
1535         return 0;
1536
1537 redirty_out:
1538         redirty_page_for_writepage(wbc, page);
1539         if (!err)
1540                 return AOP_WRITEPAGE_ACTIVATE;
1541         unlock_page(page);
1542         return err;
1543 }
1544
1545 static int f2fs_write_data_page(struct page *page,
1546                                         struct writeback_control *wbc)
1547 {
1548         return __write_data_page(page, NULL, wbc);
1549 }
1550
1551 /*
1552  * This function was copied from write_cche_pages from mm/page-writeback.c.
1553  * The major change is making write step of cold data page separately from
1554  * warm/hot data page.
1555  */
1556 static int f2fs_write_cache_pages(struct address_space *mapping,
1557                                         struct writeback_control *wbc)
1558 {
1559         int ret = 0;
1560         int done = 0;
1561         struct pagevec pvec;
1562         int nr_pages;
1563         pgoff_t uninitialized_var(writeback_index);
1564         pgoff_t index;
1565         pgoff_t end;            /* Inclusive */
1566         pgoff_t done_index;
1567         pgoff_t last_idx = ULONG_MAX;
1568         int cycled;
1569         int range_whole = 0;
1570         int tag;
1571
1572         pagevec_init(&pvec, 0);
1573
1574         if (get_dirty_pages(mapping->host) <=
1575                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1576                 set_inode_flag(mapping->host, FI_HOT_DATA);
1577         else
1578                 clear_inode_flag(mapping->host, FI_HOT_DATA);
1579
1580         if (wbc->range_cyclic) {
1581                 writeback_index = mapping->writeback_index; /* prev offset */
1582                 index = writeback_index;
1583                 if (index == 0)
1584                         cycled = 1;
1585                 else
1586                         cycled = 0;
1587                 end = -1;
1588         } else {
1589                 index = wbc->range_start >> PAGE_SHIFT;
1590                 end = wbc->range_end >> PAGE_SHIFT;
1591                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1592                         range_whole = 1;
1593                 cycled = 1; /* ignore range_cyclic tests */
1594         }
1595         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1596                 tag = PAGECACHE_TAG_TOWRITE;
1597         else
1598                 tag = PAGECACHE_TAG_DIRTY;
1599 retry:
1600         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1601                 tag_pages_for_writeback(mapping, index, end);
1602         done_index = index;
1603         while (!done && (index <= end)) {
1604                 int i;
1605
1606                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1607                               min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1608                 if (nr_pages == 0)
1609                         break;
1610
1611                 for (i = 0; i < nr_pages; i++) {
1612                         struct page *page = pvec.pages[i];
1613                         bool submitted = false;
1614
1615                         if (page->index > end) {
1616                                 done = 1;
1617                                 break;
1618                         }
1619
1620                         done_index = page->index;
1621
1622                         lock_page(page);
1623
1624                         if (unlikely(page->mapping != mapping)) {
1625 continue_unlock:
1626                                 unlock_page(page);
1627                                 continue;
1628                         }
1629
1630                         if (!PageDirty(page)) {
1631                                 /* someone wrote it for us */
1632                                 goto continue_unlock;
1633                         }
1634
1635                         if (PageWriteback(page)) {
1636                                 if (wbc->sync_mode != WB_SYNC_NONE)
1637                                         f2fs_wait_on_page_writeback(page,
1638                                                                 DATA, true);
1639                                 else
1640                                         goto continue_unlock;
1641                         }
1642
1643                         BUG_ON(PageWriteback(page));
1644                         if (!clear_page_dirty_for_io(page))
1645                                 goto continue_unlock;
1646
1647                         ret = __write_data_page(page, &submitted, wbc);
1648                         if (unlikely(ret)) {
1649                                 /*
1650                                  * keep nr_to_write, since vfs uses this to
1651                                  * get # of written pages.
1652                                  */
1653                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1654                                         unlock_page(page);
1655                                         ret = 0;
1656                                         continue;
1657                                 }
1658                                 done_index = page->index + 1;
1659                                 done = 1;
1660                                 break;
1661                         } else if (submitted) {
1662                                 last_idx = page->index;
1663                         }
1664
1665                         /* give a priority to WB_SYNC threads */
1666                         if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1667                                         --wbc->nr_to_write <= 0) &&
1668                                         wbc->sync_mode == WB_SYNC_NONE) {
1669                                 done = 1;
1670                                 break;
1671                         }
1672                 }
1673                 pagevec_release(&pvec);
1674                 cond_resched();
1675         }
1676
1677         if (!cycled && !done) {
1678                 cycled = 1;
1679                 index = 0;
1680                 end = writeback_index - 1;
1681                 goto retry;
1682         }
1683         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1684                 mapping->writeback_index = done_index;
1685
1686         if (last_idx != ULONG_MAX)
1687                 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1688                                                 0, last_idx, DATA, WRITE);
1689
1690         return ret;
1691 }
1692
1693 static int f2fs_write_data_pages(struct address_space *mapping,
1694                             struct writeback_control *wbc)
1695 {
1696         struct inode *inode = mapping->host;
1697         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1698         struct blk_plug plug;
1699         int ret;
1700
1701         /* deal with chardevs and other special file */
1702         if (!mapping->a_ops->writepage)
1703                 return 0;
1704
1705         /* skip writing if there is no dirty page in this inode */
1706         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1707                 return 0;
1708
1709         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1710                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1711                         available_free_memory(sbi, DIRTY_DENTS))
1712                 goto skip_write;
1713
1714         /* skip writing during file defragment */
1715         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1716                 goto skip_write;
1717
1718         /* during POR, we don't need to trigger writepage at all. */
1719         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1720                 goto skip_write;
1721
1722         trace_f2fs_writepages(mapping->host, wbc, DATA);
1723
1724         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1725         if (wbc->sync_mode == WB_SYNC_ALL)
1726                 atomic_inc(&sbi->wb_sync_req);
1727         else if (atomic_read(&sbi->wb_sync_req))
1728                 goto skip_write;
1729
1730         blk_start_plug(&plug);
1731         ret = f2fs_write_cache_pages(mapping, wbc);
1732         blk_finish_plug(&plug);
1733
1734         if (wbc->sync_mode == WB_SYNC_ALL)
1735                 atomic_dec(&sbi->wb_sync_req);
1736         /*
1737          * if some pages were truncated, we cannot guarantee its mapping->host
1738          * to detect pending bios.
1739          */
1740
1741         remove_dirty_inode(inode);
1742         return ret;
1743
1744 skip_write:
1745         wbc->pages_skipped += get_dirty_pages(inode);
1746         trace_f2fs_writepages(mapping->host, wbc, DATA);
1747         return 0;
1748 }
1749
1750 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1751 {
1752         struct inode *inode = mapping->host;
1753         loff_t i_size = i_size_read(inode);
1754
1755         if (to > i_size) {
1756                 truncate_pagecache(inode, i_size);
1757                 truncate_blocks(inode, i_size, true);
1758         }
1759 }
1760
1761 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1762                         struct page *page, loff_t pos, unsigned len,
1763                         block_t *blk_addr, bool *node_changed)
1764 {
1765         struct inode *inode = page->mapping->host;
1766         pgoff_t index = page->index;
1767         struct dnode_of_data dn;
1768         struct page *ipage;
1769         bool locked = false;
1770         struct extent_info ei = {0,0,0};
1771         int err = 0;
1772
1773         /*
1774          * we already allocated all the blocks, so we don't need to get
1775          * the block addresses when there is no need to fill the page.
1776          */
1777         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1778                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
1779                 return 0;
1780
1781         if (f2fs_has_inline_data(inode) ||
1782                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1783                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1784                 locked = true;
1785         }
1786 restart:
1787         /* check inline_data */
1788         ipage = get_node_page(sbi, inode->i_ino);
1789         if (IS_ERR(ipage)) {
1790                 err = PTR_ERR(ipage);
1791                 goto unlock_out;
1792         }
1793
1794         set_new_dnode(&dn, inode, ipage, ipage, 0);
1795
1796         if (f2fs_has_inline_data(inode)) {
1797                 if (pos + len <= MAX_INLINE_DATA) {
1798                         read_inline_data(page, ipage);
1799                         set_inode_flag(inode, FI_DATA_EXIST);
1800                         if (inode->i_nlink)
1801                                 set_inline_node(ipage);
1802                 } else {
1803                         err = f2fs_convert_inline_page(&dn, page);
1804                         if (err)
1805                                 goto out;
1806                         if (dn.data_blkaddr == NULL_ADDR)
1807                                 err = f2fs_get_block(&dn, index);
1808                 }
1809         } else if (locked) {
1810                 err = f2fs_get_block(&dn, index);
1811         } else {
1812                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1813                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1814                 } else {
1815                         /* hole case */
1816                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1817                         if (err || dn.data_blkaddr == NULL_ADDR) {
1818                                 f2fs_put_dnode(&dn);
1819                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1820                                                                 true);
1821                                 locked = true;
1822                                 goto restart;
1823                         }
1824                 }
1825         }
1826
1827         /* convert_inline_page can make node_changed */
1828         *blk_addr = dn.data_blkaddr;
1829         *node_changed = dn.node_changed;
1830 out:
1831         f2fs_put_dnode(&dn);
1832 unlock_out:
1833         if (locked)
1834                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1835         return err;
1836 }
1837
1838 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1839                 loff_t pos, unsigned len, unsigned flags,
1840                 struct page **pagep, void **fsdata)
1841 {
1842         struct inode *inode = mapping->host;
1843         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1844         struct page *page = NULL;
1845         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1846         bool need_balance = false;
1847         block_t blkaddr = NULL_ADDR;
1848         int err = 0;
1849
1850         trace_f2fs_write_begin(inode, pos, len, flags);
1851
1852         /*
1853          * We should check this at this moment to avoid deadlock on inode page
1854          * and #0 page. The locking rule for inline_data conversion should be:
1855          * lock_page(page #0) -> lock_page(inode_page)
1856          */
1857         if (index != 0) {
1858                 err = f2fs_convert_inline_inode(inode);
1859                 if (err)
1860                         goto fail;
1861         }
1862 repeat:
1863         /*
1864          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1865          * wait_for_stable_page. Will wait that below with our IO control.
1866          */
1867         page = pagecache_get_page(mapping, index,
1868                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1869         if (!page) {
1870                 err = -ENOMEM;
1871                 goto fail;
1872         }
1873
1874         *pagep = page;
1875
1876         err = prepare_write_begin(sbi, page, pos, len,
1877                                         &blkaddr, &need_balance);
1878         if (err)
1879                 goto fail;
1880
1881         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1882                 unlock_page(page);
1883                 f2fs_balance_fs(sbi, true);
1884                 lock_page(page);
1885                 if (page->mapping != mapping) {
1886                         /* The page got truncated from under us */
1887                         f2fs_put_page(page, 1);
1888                         goto repeat;
1889                 }
1890         }
1891
1892         f2fs_wait_on_page_writeback(page, DATA, false);
1893
1894         /* wait for GCed encrypted page writeback */
1895         if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1896                 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1897
1898         if (len == PAGE_SIZE || PageUptodate(page))
1899                 return 0;
1900
1901         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1902                 zero_user_segment(page, len, PAGE_SIZE);
1903                 return 0;
1904         }
1905
1906         if (blkaddr == NEW_ADDR) {
1907                 zero_user_segment(page, 0, PAGE_SIZE);
1908                 SetPageUptodate(page);
1909         } else {
1910                 struct bio *bio;
1911
1912                 bio = f2fs_grab_bio(inode, blkaddr, 1);
1913                 if (IS_ERR(bio)) {
1914                         err = PTR_ERR(bio);
1915                         goto fail;
1916                 }
1917                 bio->bi_opf = REQ_OP_READ;
1918                 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1919                         bio_put(bio);
1920                         err = -EFAULT;
1921                         goto fail;
1922                 }
1923
1924                 __submit_bio(sbi, bio, DATA);
1925
1926                 lock_page(page);
1927                 if (unlikely(page->mapping != mapping)) {
1928                         f2fs_put_page(page, 1);
1929                         goto repeat;
1930                 }
1931                 if (unlikely(!PageUptodate(page))) {
1932                         err = -EIO;
1933                         goto fail;
1934                 }
1935         }
1936         return 0;
1937
1938 fail:
1939         f2fs_put_page(page, 1);
1940         f2fs_write_failed(mapping, pos + len);
1941         return err;
1942 }
1943
1944 static int f2fs_write_end(struct file *file,
1945                         struct address_space *mapping,
1946                         loff_t pos, unsigned len, unsigned copied,
1947                         struct page *page, void *fsdata)
1948 {
1949         struct inode *inode = page->mapping->host;
1950
1951         trace_f2fs_write_end(inode, pos, len, copied);
1952
1953         /*
1954          * This should be come from len == PAGE_SIZE, and we expect copied
1955          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1956          * let generic_perform_write() try to copy data again through copied=0.
1957          */
1958         if (!PageUptodate(page)) {
1959                 if (unlikely(copied != len))
1960                         copied = 0;
1961                 else
1962                         SetPageUptodate(page);
1963         }
1964         if (!copied)
1965                 goto unlock_out;
1966
1967         set_page_dirty(page);
1968
1969         if (pos + copied > i_size_read(inode))
1970                 f2fs_i_size_write(inode, pos + copied);
1971 unlock_out:
1972         f2fs_put_page(page, 1);
1973         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1974         return copied;
1975 }
1976
1977 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1978                            loff_t offset)
1979 {
1980         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1981
1982         if (offset & blocksize_mask)
1983                 return -EINVAL;
1984
1985         if (iov_iter_alignment(iter) & blocksize_mask)
1986                 return -EINVAL;
1987
1988         return 0;
1989 }
1990
1991 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1992 {
1993         struct address_space *mapping = iocb->ki_filp->f_mapping;
1994         struct inode *inode = mapping->host;
1995         size_t count = iov_iter_count(iter);
1996         loff_t offset = iocb->ki_pos;
1997         int rw = iov_iter_rw(iter);
1998         int err;
1999
2000         err = check_direct_IO(inode, iter, offset);
2001         if (err)
2002                 return err;
2003
2004         if (__force_buffered_io(inode, rw))
2005                 return 0;
2006
2007         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2008
2009         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2010         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2011         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2012
2013         if (rw == WRITE) {
2014                 if (err > 0)
2015                         set_inode_flag(inode, FI_UPDATE_WRITE);
2016                 else if (err < 0)
2017                         f2fs_write_failed(mapping, offset + count);
2018         }
2019
2020         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2021
2022         return err;
2023 }
2024
2025 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2026                                                         unsigned int length)
2027 {
2028         struct inode *inode = page->mapping->host;
2029         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2030
2031         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2032                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2033                 return;
2034
2035         if (PageDirty(page)) {
2036                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2037                         dec_page_count(sbi, F2FS_DIRTY_META);
2038                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2039                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2040                 } else {
2041                         inode_dec_dirty_pages(inode);
2042                         remove_dirty_inode(inode);
2043                 }
2044         }
2045
2046         /* This is atomic written page, keep Private */
2047         if (IS_ATOMIC_WRITTEN_PAGE(page))
2048                 return drop_inmem_page(inode, page);
2049
2050         set_page_private(page, 0);
2051         ClearPagePrivate(page);
2052 }
2053
2054 int f2fs_release_page(struct page *page, gfp_t wait)
2055 {
2056         /* If this is dirty page, keep PagePrivate */
2057         if (PageDirty(page))
2058                 return 0;
2059
2060         /* This is atomic written page, keep Private */
2061         if (IS_ATOMIC_WRITTEN_PAGE(page))
2062                 return 0;
2063
2064         set_page_private(page, 0);
2065         ClearPagePrivate(page);
2066         return 1;
2067 }
2068
2069 /*
2070  * This was copied from __set_page_dirty_buffers which gives higher performance
2071  * in very high speed storages. (e.g., pmem)
2072  */
2073 void f2fs_set_page_dirty_nobuffers(struct page *page)
2074 {
2075         struct address_space *mapping = page->mapping;
2076         unsigned long flags;
2077
2078         if (unlikely(!mapping))
2079                 return;
2080
2081         spin_lock(&mapping->private_lock);
2082         lock_page_memcg(page);
2083         SetPageDirty(page);
2084         spin_unlock(&mapping->private_lock);
2085
2086         spin_lock_irqsave(&mapping->tree_lock, flags);
2087         WARN_ON_ONCE(!PageUptodate(page));
2088         account_page_dirtied(page, mapping);
2089         radix_tree_tag_set(&mapping->page_tree,
2090                         page_index(page), PAGECACHE_TAG_DIRTY);
2091         spin_unlock_irqrestore(&mapping->tree_lock, flags);
2092         unlock_page_memcg(page);
2093
2094         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2095         return;
2096 }
2097
2098 static int f2fs_set_data_page_dirty(struct page *page)
2099 {
2100         struct address_space *mapping = page->mapping;
2101         struct inode *inode = mapping->host;
2102
2103         trace_f2fs_set_page_dirty(page, DATA);
2104
2105         if (!PageUptodate(page))
2106                 SetPageUptodate(page);
2107
2108         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2109                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2110                         register_inmem_page(inode, page);
2111                         return 1;
2112                 }
2113                 /*
2114                  * Previously, this page has been registered, we just
2115                  * return here.
2116                  */
2117                 return 0;
2118         }
2119
2120         if (!PageDirty(page)) {
2121                 f2fs_set_page_dirty_nobuffers(page);
2122                 update_dirty_page(inode, page);
2123                 return 1;
2124         }
2125         return 0;
2126 }
2127
2128 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2129 {
2130         struct inode *inode = mapping->host;
2131
2132         if (f2fs_has_inline_data(inode))
2133                 return 0;
2134
2135         /* make sure allocating whole blocks */
2136         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2137                 filemap_write_and_wait(mapping);
2138
2139         return generic_block_bmap(mapping, block, get_data_block_bmap);
2140 }
2141
2142 #ifdef CONFIG_MIGRATION
2143 #include <linux/migrate.h>
2144
2145 int f2fs_migrate_page(struct address_space *mapping,
2146                 struct page *newpage, struct page *page, enum migrate_mode mode)
2147 {
2148         int rc, extra_count;
2149         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2150         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2151
2152         BUG_ON(PageWriteback(page));
2153
2154         /* migrating an atomic written page is safe with the inmem_lock hold */
2155         if (atomic_written && !mutex_trylock(&fi->inmem_lock))
2156                 return -EAGAIN;
2157
2158         /*
2159          * A reference is expected if PagePrivate set when move mapping,
2160          * however F2FS breaks this for maintaining dirty page counts when
2161          * truncating pages. So here adjusting the 'extra_count' make it work.
2162          */
2163         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2164         rc = migrate_page_move_mapping(mapping, newpage,
2165                                 page, NULL, mode, extra_count);
2166         if (rc != MIGRATEPAGE_SUCCESS) {
2167                 if (atomic_written)
2168                         mutex_unlock(&fi->inmem_lock);
2169                 return rc;
2170         }
2171
2172         if (atomic_written) {
2173                 struct inmem_pages *cur;
2174                 list_for_each_entry(cur, &fi->inmem_pages, list)
2175                         if (cur->page == page) {
2176                                 cur->page = newpage;
2177                                 break;
2178                         }
2179                 mutex_unlock(&fi->inmem_lock);
2180                 put_page(page);
2181                 get_page(newpage);
2182         }
2183
2184         if (PagePrivate(page))
2185                 SetPagePrivate(newpage);
2186         set_page_private(newpage, page_private(page));
2187
2188         migrate_page_copy(newpage, page);
2189
2190         return MIGRATEPAGE_SUCCESS;
2191 }
2192 #endif
2193
2194 const struct address_space_operations f2fs_dblock_aops = {
2195         .readpage       = f2fs_read_data_page,
2196         .readpages      = f2fs_read_data_pages,
2197         .writepage      = f2fs_write_data_page,
2198         .writepages     = f2fs_write_data_pages,
2199         .write_begin    = f2fs_write_begin,
2200         .write_end      = f2fs_write_end,
2201         .set_page_dirty = f2fs_set_data_page_dirty,
2202         .invalidatepage = f2fs_invalidate_page,
2203         .releasepage    = f2fs_release_page,
2204         .direct_IO      = f2fs_direct_IO,
2205         .bmap           = f2fs_bmap,
2206 #ifdef CONFIG_MIGRATION
2207         .migratepage    = f2fs_migrate_page,
2208 #endif
2209 };