fs/f2fs/data.c

   1 /*
   2  * fs/f2fs/data.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/mpage.h>
  15 #include <linux/writeback.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21 #include <linux/uio.h>
  22 #include <linux/cleancache.h>
  23 #include <linux/sched/signal.h>
  24
  25 #include "f2fs.h"
  26 #include "node.h"
  27 #include "segment.h"
  28 #include "trace.h"
  29 #include <trace/events/f2fs.h>
  30
  31 #define NUM_PREALLOC_POST_READ_CTXS     128
  32
  33 static struct kmem_cache *bio_post_read_ctx_cache;
  34 static mempool_t *bio_post_read_ctx_pool;
  35
  36 static bool __is_cp_guaranteed(struct page *page)
  37 {
  38         struct address_space *mapping = page->mapping;
  39         struct inode *inode;
  40         struct f2fs_sb_info *sbi;
  41
  42         if (!mapping)
  43                 return false;
  44
  45         inode = mapping->host;
  46         sbi = F2FS_I_SB(inode);
  47
  48         if (inode->i_ino == F2FS_META_INO(sbi) ||
  49                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  50                         S_ISDIR(inode->i_mode) ||
  51                         (S_ISREG(inode->i_mode) &&
  52                         is_inode_flag_set(inode, FI_ATOMIC_FILE)) ||
  53                         is_cold_data(page))
  54                 return true;
  55         return false;
  56 }
  57
  58 /* postprocessing steps for read bios */
  59 enum bio_post_read_step {
  60         STEP_INITIAL = 0,
  61         STEP_DECRYPT,
  62 };
  63
  64 struct bio_post_read_ctx {
  65         struct bio *bio;
  66         struct work_struct work;
  67         unsigned int cur_step;
  68         unsigned int enabled_steps;
  69 };
  70
  71 static void __read_end_io(struct bio *bio)
  72 {
  73         struct page *page;
  74         struct bio_vec *bv;
  75         int i;
  76
  77         bio_for_each_segment_all(bv, bio, i) {
  78                 page = bv->bv_page;
  79
  80                 /* PG_error was set if any post_read step failed */
  81                 if (bio->bi_status || PageError(page)) {
  82                         ClearPageUptodate(page);
  83                         SetPageError(page);
  84                 } else {
  85                         SetPageUptodate(page);
  86                 }
  87                 unlock_page(page);
  88         }
  89         if (bio->bi_private)
  90                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
  91         bio_put(bio);
  92 }
  93
  94 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
  95
  96 static void decrypt_work(struct work_struct *work)
  97 {
  98         struct bio_post_read_ctx *ctx =
  99                 container_of(work, struct bio_post_read_ctx, work);
 100
 101         fscrypt_decrypt_bio(ctx->bio);
 102
 103         bio_post_read_processing(ctx);
 104 }
 105
 106 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 107 {
 108         switch (++ctx->cur_step) {
 109         case STEP_DECRYPT:
 110                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 111                         INIT_WORK(&ctx->work, decrypt_work);
 112                         fscrypt_enqueue_decrypt_work(&ctx->work);
 113                         return;
 114                 }
 115                 ctx->cur_step++;
 116                 /* fall-through */
 117         default:
 118                 __read_end_io(ctx->bio);
 119         }
 120 }
 121
 122 static bool f2fs_bio_post_read_required(struct bio *bio)
 123 {
 124         return bio->bi_private && !bio->bi_status;
 125 }
 126
 127 static void f2fs_read_end_io(struct bio *bio)
 128 {
 129         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)), FAULT_IO)) {
 130                 f2fs_show_injection_info(FAULT_IO);
 131                 bio->bi_status = BLK_STS_IOERR;
 132         }
 133
 134         if (f2fs_bio_post_read_required(bio)) {
 135                 struct bio_post_read_ctx *ctx = bio->bi_private;
 136
 137                 ctx->cur_step = STEP_INITIAL;
 138                 bio_post_read_processing(ctx);
 139                 return;
 140         }
 141
 142         __read_end_io(bio);
 143 }
 144
 145 static void f2fs_write_end_io(struct bio *bio)
 146 {
 147         struct f2fs_sb_info *sbi = bio->bi_private;
 148         struct bio_vec *bvec;
 149         int i;
 150
 151         bio_for_each_segment_all(bvec, bio, i) {
 152                 struct page *page = bvec->bv_page;
 153                 enum count_type type = WB_DATA_TYPE(page);
 154
 155                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 156                         set_page_private(page, (unsigned long)NULL);
 157                         ClearPagePrivate(page);
 158                         unlock_page(page);
 159                         mempool_free(page, sbi->write_io_dummy);
 160
 161                         if (unlikely(bio->bi_status))
 162                                 f2fs_stop_checkpoint(sbi, true);
 163                         continue;
 164                 }
 165
 166                 fscrypt_pullback_bio_page(&page, true);
 167
 168                 if (unlikely(bio->bi_status)) {
 169                         mapping_set_error(page->mapping, -EIO);
 170                         if (type == F2FS_WB_CP_DATA)
 171                                 f2fs_stop_checkpoint(sbi, true);
 172                 }
 173
 174                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 175                                         page->index != nid_of_node(page));
 176
 177                 dec_page_count(sbi, type);
 178                 if (f2fs_in_warm_node_list(sbi, page))
 179                         f2fs_del_fsync_node_entry(sbi, page);
 180                 clear_cold_data(page);
 181                 end_page_writeback(page);
 182         }
 183         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 184                                 wq_has_sleeper(&sbi->cp_wait))
 185                 wake_up(&sbi->cp_wait);
 186
 187         bio_put(bio);
 188 }
 189
 190 /*
 191  * Return true, if pre_bio's bdev is same as its target device.
 192  */
 193 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 194                                 block_t blk_addr, struct bio *bio)
 195 {
 196         struct block_device *bdev = sbi->sb->s_bdev;
 197         int i;
 198
 199         for (i = 0; i < sbi->s_ndevs; i++) {
 200                 if (FDEV(i).start_blk <= blk_addr &&
 201                                         FDEV(i).end_blk >= blk_addr) {
 202                         blk_addr -= FDEV(i).start_blk;
 203                         bdev = FDEV(i).bdev;
 204                         break;
 205                 }
 206         }
 207         if (bio) {
 208                 bio_set_dev(bio, bdev);
 209                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 210         }
 211         return bdev;
 212 }
 213
 214 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 215 {
 216         int i;
 217
 218         for (i = 0; i < sbi->s_ndevs; i++)
 219                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 220                         return i;
 221         return 0;
 222 }
 223
 224 static bool __same_bdev(struct f2fs_sb_info *sbi,
 225                                 block_t blk_addr, struct bio *bio)
 226 {
 227         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 228         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 229 }
 230
 231 /*
 232  * Low-level block read/write IO operations.
 233  */
 234 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 235                                 struct writeback_control *wbc,
 236                                 int npages, bool is_read,
 237                                 enum page_type type, enum temp_type temp)
 238 {
 239         struct bio *bio;
 240
 241         bio = f2fs_bio_alloc(sbi, npages, true);
 242
 243         f2fs_target_device(sbi, blk_addr, bio);
 244         if (is_read) {
 245                 bio->bi_end_io = f2fs_read_end_io;
 246                 bio->bi_private = NULL;
 247         } else {
 248                 bio->bi_end_io = f2fs_write_end_io;
 249                 bio->bi_private = sbi;
 250                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 251         }
 252         if (wbc)
 253                 wbc_init_bio(wbc, bio);
 254
 255         return bio;
 256 }
 257
 258 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 259                                 struct bio *bio, enum page_type type)
 260 {
 261         if (!is_read_io(bio_op(bio))) {
 262                 unsigned int start;
 263
 264                 if (type != DATA && type != NODE)
 265                         goto submit_io;
 266
 267                 if (test_opt(sbi, LFS) && current->plug)
 268                         blk_finish_plug(current->plug);
 269
 270                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 271                 start %= F2FS_IO_SIZE(sbi);
 272
 273                 if (start == 0)
 274                         goto submit_io;
 275
 276                 /* fill dummy pages */
 277                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 278                         struct page *page =
 279                                 mempool_alloc(sbi->write_io_dummy,
 280                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 281                         f2fs_bug_on(sbi, !page);
 282
 283                         SetPagePrivate(page);
 284                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 285                         lock_page(page);
 286                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 287                                 f2fs_bug_on(sbi, 1);
 288                 }
 289                 /*
 290                  * In the NODE case, we lose next block address chain. So, we
 291                  * need to do checkpoint in f2fs_sync_file.
 292                  */
 293                 if (type == NODE)
 294                         set_sbi_flag(sbi, SBI_NEED_CP);
 295         }
 296 submit_io:
 297         if (is_read_io(bio_op(bio)))
 298                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 299         else
 300                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 301         submit_bio(bio);
 302 }
 303
 304 static void __submit_merged_bio(struct f2fs_bio_info *io)
 305 {
 306         struct f2fs_io_info *fio = &io->fio;
 307
 308         if (!io->bio)
 309                 return;
 310
 311         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 312
 313         if (is_read_io(fio->op))
 314                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 315         else
 316                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 317
 318         __submit_bio(io->sbi, io->bio, fio->type);
 319         io->bio = NULL;
 320 }
 321
 322 static bool __has_merged_page(struct f2fs_bio_info *io,
 323                                 struct inode *inode, nid_t ino, pgoff_t idx)
 324 {
 325         struct bio_vec *bvec;
 326         struct page *target;
 327         int i;
 328
 329         if (!io->bio)
 330                 return false;
 331
 332         if (!inode && !ino)
 333                 return true;
 334
 335         bio_for_each_segment_all(bvec, io->bio, i) {
 336
 337                 if (bvec->bv_page->mapping)
 338                         target = bvec->bv_page;
 339                 else
 340                         target = fscrypt_control_page(bvec->bv_page);
 341
 342                 if (idx != target->index)
 343                         continue;
 344
 345                 if (inode && inode == target->mapping->host)
 346                         return true;
 347                 if (ino && ino == ino_of_node(target))
 348                         return true;
 349         }
 350
 351         return false;
 352 }
 353
 354 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 355                                 nid_t ino, pgoff_t idx, enum page_type type)
 356 {
 357         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 358         enum temp_type temp;
 359         struct f2fs_bio_info *io;
 360         bool ret = false;
 361
 362         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 363                 io = sbi->write_io[btype] + temp;
 364
 365                 down_read(&io->io_rwsem);
 366                 ret = __has_merged_page(io, inode, ino, idx);
 367                 up_read(&io->io_rwsem);
 368
 369                 /* TODO: use HOT temp only for meta pages now. */
 370                 if (ret || btype == META)
 371                         break;
 372         }
 373         return ret;
 374 }
 375
 376 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 377                                 enum page_type type, enum temp_type temp)
 378 {
 379         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 380         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 381
 382         down_write(&io->io_rwsem);
 383
 384         /* change META to META_FLUSH in the checkpoint procedure */
 385         if (type >= META_FLUSH) {
 386                 io->fio.type = META_FLUSH;
 387                 io->fio.op = REQ_OP_WRITE;
 388                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 389                 if (!test_opt(sbi, NOBARRIER))
 390                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 391         }
 392         __submit_merged_bio(io);
 393         up_write(&io->io_rwsem);
 394 }
 395
 396 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 397                                 struct inode *inode, nid_t ino, pgoff_t idx,
 398                                 enum page_type type, bool force)
 399 {
 400         enum temp_type temp;
 401
 402         if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 403                 return;
 404
 405         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 406
 407                 __f2fs_submit_merged_write(sbi, type, temp);
 408
 409                 /* TODO: use HOT temp only for meta pages now. */
 410                 if (type >= META)
 411                         break;
 412         }
 413 }
 414
 415 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 416 {
 417         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 418 }
 419
 420 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 421                                 struct inode *inode, nid_t ino, pgoff_t idx,
 422                                 enum page_type type)
 423 {
 424         __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 425 }
 426
 427 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 428 {
 429         f2fs_submit_merged_write(sbi, DATA);
 430         f2fs_submit_merged_write(sbi, NODE);
 431         f2fs_submit_merged_write(sbi, META);
 432 }
 433
 434 /*
 435  * Fill the locked page with data located in the block address.
 436  * A caller needs to unlock the page on failure.
 437  */
 438 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 439 {
 440         struct bio *bio;
 441         struct page *page = fio->encrypted_page ?
 442                         fio->encrypted_page : fio->page;
 443
 444         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 445                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 446                 return -EFAULT;
 447
 448         trace_f2fs_submit_page_bio(page, fio);
 449         f2fs_trace_ios(fio, 0);
 450
 451         /* Allocate a new bio */
 452         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 453                                 1, is_read_io(fio->op), fio->type, fio->temp);
 454
 455         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 456                 bio_put(bio);
 457                 return -EFAULT;
 458         }
 459         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 460
 461         __submit_bio(fio->sbi, bio, fio->type);
 462
 463         if (!is_read_io(fio->op))
 464                 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 465         return 0;
 466 }
 467
 468 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 469 {
 470         struct f2fs_sb_info *sbi = fio->sbi;
 471         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 472         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 473         struct page *bio_page;
 474
 475         f2fs_bug_on(sbi, is_read_io(fio->op));
 476
 477         down_write(&io->io_rwsem);
 478 next:
 479         if (fio->in_list) {
 480                 spin_lock(&io->io_lock);
 481                 if (list_empty(&io->io_list)) {
 482                         spin_unlock(&io->io_lock);
 483                         goto out;
 484                 }
 485                 fio = list_first_entry(&io->io_list,
 486                                                 struct f2fs_io_info, list);
 487                 list_del(&fio->list);
 488                 spin_unlock(&io->io_lock);
 489         }
 490
 491         if (__is_valid_data_blkaddr(fio->old_blkaddr))
 492                 verify_block_addr(fio, fio->old_blkaddr);
 493         verify_block_addr(fio, fio->new_blkaddr);
 494
 495         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 496
 497         /* set submitted = true as a return value */
 498         fio->submitted = true;
 499
 500         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 501
 502         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 503             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 504                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 505                 __submit_merged_bio(io);
 506 alloc_new:
 507         if (io->bio == NULL) {
 508                 if ((fio->type == DATA || fio->type == NODE) &&
 509                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 510                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 511                         fio->retry = true;
 512                         goto skip;
 513                 }
 514                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 515                                                 BIO_MAX_PAGES, false,
 516                                                 fio->type, fio->temp);
 517                 io->fio = *fio;
 518         }
 519
 520         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 521                 __submit_merged_bio(io);
 522                 goto alloc_new;
 523         }
 524
 525         if (fio->io_wbc)
 526                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 527
 528         io->last_block_in_bio = fio->new_blkaddr;
 529         f2fs_trace_ios(fio, 0);
 530
 531         trace_f2fs_submit_page_write(fio->page, fio);
 532 skip:
 533         if (fio->in_list)
 534                 goto next;
 535 out:
 536         up_write(&io->io_rwsem);
 537 }
 538
 539 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 540                                         unsigned nr_pages, unsigned op_flag)
 541 {
 542         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 543         struct bio *bio;
 544         struct bio_post_read_ctx *ctx;
 545         unsigned int post_read_steps = 0;
 546
 547         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
 548                 return ERR_PTR(-EFAULT);
 549
 550         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 551         if (!bio)
 552                 return ERR_PTR(-ENOMEM);
 553         f2fs_target_device(sbi, blkaddr, bio);
 554         bio->bi_end_io = f2fs_read_end_io;
 555         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 556
 557         if (f2fs_encrypted_file(inode))
 558                 post_read_steps |= 1 << STEP_DECRYPT;
 559         if (post_read_steps) {
 560                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 561                 if (!ctx) {
 562                         bio_put(bio);
 563                         return ERR_PTR(-ENOMEM);
 564                 }
 565                 ctx->bio = bio;
 566                 ctx->enabled_steps = post_read_steps;
 567                 bio->bi_private = ctx;
 568
 569                 /* wait the page to be moved by cleaning */
 570                 f2fs_wait_on_block_writeback(sbi, blkaddr);
 571         }
 572
 573         return bio;
 574 }
 575
 576 /* This can handle encryption stuffs */
 577 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 578                                                         block_t blkaddr)
 579 {
 580         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 581
 582         if (IS_ERR(bio))
 583                 return PTR_ERR(bio);
 584
 585         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 586                 bio_put(bio);
 587                 return -EFAULT;
 588         }
 589         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 590         return 0;
 591 }
 592
 593 static void __set_data_blkaddr(struct dnode_of_data *dn)
 594 {
 595         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 596         __le32 *addr_array;
 597         int base = 0;
 598
 599         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 600                 base = get_extra_isize(dn->inode);
 601
 602         /* Get physical address of data block */
 603         addr_array = blkaddr_in_node(rn);
 604         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 605 }
 606
 607 /*
 608  * Lock ordering for the change of data block address:
 609  * ->data_page
 610  *  ->node_page
 611  *    update block addresses in the node page
 612  */
 613 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 614 {
 615         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 616         __set_data_blkaddr(dn);
 617         if (set_page_dirty(dn->node_page))
 618                 dn->node_changed = true;
 619 }
 620
 621 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 622 {
 623         dn->data_blkaddr = blkaddr;
 624         f2fs_set_data_blkaddr(dn);
 625         f2fs_update_extent_cache(dn);
 626 }
 627
 628 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 629 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 630 {
 631         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 632         int err;
 633
 634         if (!count)
 635                 return 0;
 636
 637         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 638                 return -EPERM;
 639         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 640                 return err;
 641
 642         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 643                                                 dn->ofs_in_node, count);
 644
 645         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 646
 647         for (; count > 0; dn->ofs_in_node++) {
 648                 block_t blkaddr = datablock_addr(dn->inode,
 649                                         dn->node_page, dn->ofs_in_node);
 650                 if (blkaddr == NULL_ADDR) {
 651                         dn->data_blkaddr = NEW_ADDR;
 652                         __set_data_blkaddr(dn);
 653                         count--;
 654                 }
 655         }
 656
 657         if (set_page_dirty(dn->node_page))
 658                 dn->node_changed = true;
 659         return 0;
 660 }
 661
 662 /* Should keep dn->ofs_in_node unchanged */
 663 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 664 {
 665         unsigned int ofs_in_node = dn->ofs_in_node;
 666         int ret;
 667
 668         ret = f2fs_reserve_new_blocks(dn, 1);
 669         dn->ofs_in_node = ofs_in_node;
 670         return ret;
 671 }
 672
 673 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 674 {
 675         bool need_put = dn->inode_page ? false : true;
 676         int err;
 677
 678         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 679         if (err)
 680                 return err;
 681
 682         if (dn->data_blkaddr == NULL_ADDR)
 683                 err = f2fs_reserve_new_block(dn);
 684         if (err || need_put)
 685                 f2fs_put_dnode(dn);
 686         return err;
 687 }
 688
 689 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 690 {
 691         struct extent_info ei  = {0,0,0};
 692         struct inode *inode = dn->inode;
 693
 694         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 695                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 696                 return 0;
 697         }
 698
 699         return f2fs_reserve_block(dn, index);
 700 }
 701
 702 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 703                                                 int op_flags, bool for_write)
 704 {
 705         struct address_space *mapping = inode->i_mapping;
 706         struct dnode_of_data dn;
 707         struct page *page;
 708         struct extent_info ei = {0,0,0};
 709         int err;
 710
 711         page = f2fs_grab_cache_page(mapping, index, for_write);
 712         if (!page)
 713                 return ERR_PTR(-ENOMEM);
 714
 715         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 716                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 717                 goto got_it;
 718         }
 719
 720         set_new_dnode(&dn, inode, NULL, NULL, 0);
 721         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 722         if (err)
 723                 goto put_err;
 724         f2fs_put_dnode(&dn);
 725
 726         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 727                 err = -ENOENT;
 728                 goto put_err;
 729         }
 730 got_it:
 731         if (PageUptodate(page)) {
 732                 unlock_page(page);
 733                 return page;
 734         }
 735
 736         /*
 737          * A new dentry page is allocated but not able to be written, since its
 738          * new inode page couldn't be allocated due to -ENOSPC.
 739          * In such the case, its blkaddr can be remained as NEW_ADDR.
 740          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 741          * f2fs_init_inode_metadata.
 742          */
 743         if (dn.data_blkaddr == NEW_ADDR) {
 744                 zero_user_segment(page, 0, PAGE_SIZE);
 745                 if (!PageUptodate(page))
 746                         SetPageUptodate(page);
 747                 unlock_page(page);
 748                 return page;
 749         }
 750
 751         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 752         if (err)
 753                 goto put_err;
 754         return page;
 755
 756 put_err:
 757         f2fs_put_page(page, 1);
 758         return ERR_PTR(err);
 759 }
 760
 761 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 762 {
 763         struct address_space *mapping = inode->i_mapping;
 764         struct page *page;
 765
 766         page = find_get_page(mapping, index);
 767         if (page && PageUptodate(page))
 768                 return page;
 769         f2fs_put_page(page, 0);
 770
 771         page = f2fs_get_read_data_page(inode, index, 0, false);
 772         if (IS_ERR(page))
 773                 return page;
 774
 775         if (PageUptodate(page))
 776                 return page;
 777
 778         wait_on_page_locked(page);
 779         if (unlikely(!PageUptodate(page))) {
 780                 f2fs_put_page(page, 0);
 781                 return ERR_PTR(-EIO);
 782         }
 783         return page;
 784 }
 785
 786 /*
 787  * If it tries to access a hole, return an error.
 788  * Because, the callers, functions in dir.c and GC, should be able to know
 789  * whether this page exists or not.
 790  */
 791 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 792                                                         bool for_write)
 793 {
 794         struct address_space *mapping = inode->i_mapping;
 795         struct page *page;
 796 repeat:
 797         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 798         if (IS_ERR(page))
 799                 return page;
 800
 801         /* wait for read completion */
 802         lock_page(page);
 803         if (unlikely(page->mapping != mapping)) {
 804                 f2fs_put_page(page, 1);
 805                 goto repeat;
 806         }
 807         if (unlikely(!PageUptodate(page))) {
 808                 f2fs_put_page(page, 1);
 809                 return ERR_PTR(-EIO);
 810         }
 811         return page;
 812 }
 813
 814 /*
 815  * Caller ensures that this data page is never allocated.
 816  * A new zero-filled data page is allocated in the page cache.
 817  *
 818  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 819  * f2fs_unlock_op().
 820  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 821  * ipage should be released by this function.
 822  */
 823 struct page *f2fs_get_new_data_page(struct inode *inode,
 824                 struct page *ipage, pgoff_t index, bool new_i_size)
 825 {
 826         struct address_space *mapping = inode->i_mapping;
 827         struct page *page;
 828         struct dnode_of_data dn;
 829         int err;
 830
 831         page = f2fs_grab_cache_page(mapping, index, true);
 832         if (!page) {
 833                 /*
 834                  * before exiting, we should make sure ipage will be released
 835                  * if any error occur.
 836                  */
 837                 f2fs_put_page(ipage, 1);
 838                 return ERR_PTR(-ENOMEM);
 839         }
 840
 841         set_new_dnode(&dn, inode, ipage, NULL, 0);
 842         err = f2fs_reserve_block(&dn, index);
 843         if (err) {
 844                 f2fs_put_page(page, 1);
 845                 return ERR_PTR(err);
 846         }
 847         if (!ipage)
 848                 f2fs_put_dnode(&dn);
 849
 850         if (PageUptodate(page))
 851                 goto got_it;
 852
 853         if (dn.data_blkaddr == NEW_ADDR) {
 854                 zero_user_segment(page, 0, PAGE_SIZE);
 855                 if (!PageUptodate(page))
 856                         SetPageUptodate(page);
 857         } else {
 858                 f2fs_put_page(page, 1);
 859
 860                 /* if ipage exists, blkaddr should be NEW_ADDR */
 861                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 862                 page = f2fs_get_lock_data_page(inode, index, true);
 863                 if (IS_ERR(page))
 864                         return page;
 865         }
 866 got_it:
 867         if (new_i_size && i_size_read(inode) <
 868                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 869                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 870         return page;
 871 }
 872
 873 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 874 {
 875         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 876         struct f2fs_summary sum;
 877         struct node_info ni;
 878         block_t old_blkaddr;
 879         pgoff_t fofs;
 880         blkcnt_t count = 1;
 881         int err;
 882
 883         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 884                 return -EPERM;
 885
 886         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 887         if (err)
 888                 return err;
 889
 890         dn->data_blkaddr = datablock_addr(dn->inode,
 891                                 dn->node_page, dn->ofs_in_node);
 892         if (dn->data_blkaddr == NEW_ADDR)
 893                 goto alloc;
 894
 895         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 896                 return err;
 897
 898 alloc:
 899         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 900         old_blkaddr = dn->data_blkaddr;
 901         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 902                                         &sum, seg_type, NULL, false);
 903         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 904                 invalidate_mapping_pages(META_MAPPING(sbi),
 905                                         old_blkaddr, old_blkaddr);
 906         f2fs_set_data_blkaddr(dn);
 907
 908         /* update i_size */
 909         fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 910                                                         dn->ofs_in_node;
 911         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 912                 f2fs_i_size_write(dn->inode,
 913                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 914         return 0;
 915 }
 916
 917 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 918 {
 919         struct inode *inode = file_inode(iocb->ki_filp);
 920         struct f2fs_map_blocks map;
 921         int flag;
 922         int err = 0;
 923         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 924
 925         /* convert inline data for Direct I/O*/
 926         if (direct_io) {
 927                 err = f2fs_convert_inline_inode(inode);
 928                 if (err)
 929                         return err;
 930         }
 931
 932         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 933                 return 0;
 934
 935         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 936         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 937         if (map.m_len > map.m_lblk)
 938                 map.m_len -= map.m_lblk;
 939         else
 940                 map.m_len = 0;
 941
 942         map.m_next_pgofs = NULL;
 943         map.m_next_extent = NULL;
 944         map.m_seg_type = NO_CHECK_TYPE;
 945
 946         if (direct_io) {
 947                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 948                 flag = f2fs_force_buffered_io(inode, WRITE) ?
 949                                         F2FS_GET_BLOCK_PRE_AIO :
 950                                         F2FS_GET_BLOCK_PRE_DIO;
 951                 goto map_blocks;
 952         }
 953         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 954                 err = f2fs_convert_inline_inode(inode);
 955                 if (err)
 956                         return err;
 957         }
 958         if (f2fs_has_inline_data(inode))
 959                 return err;
 960
 961         flag = F2FS_GET_BLOCK_PRE_AIO;
 962
 963 map_blocks:
 964         err = f2fs_map_blocks(inode, &map, 1, flag);
 965         if (map.m_len > 0 && err == -ENOSPC) {
 966                 if (!direct_io)
 967                         set_inode_flag(inode, FI_NO_PREALLOC);
 968                 err = 0;
 969         }
 970         return err;
 971 }
 972
 973 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 974 {
 975         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 976                 if (lock)
 977                         down_read(&sbi->node_change);
 978                 else
 979                         up_read(&sbi->node_change);
 980         } else {
 981                 if (lock)
 982                         f2fs_lock_op(sbi);
 983                 else
 984                         f2fs_unlock_op(sbi);
 985         }
 986 }
 987
 988 /*
 989  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 990  * f2fs_map_blocks structure.
 991  * If original data blocks are allocated, then give them to blockdev.
 992  * Otherwise,
 993  *     a. preallocate requested block addresses
 994  *     b. do not use extent cache for better performance
 995  *     c. give the block addresses to blockdev
 996  */
 997 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 998                                                 int create, int flag)
 999 {
1000         unsigned int maxblocks = map->m_len;
1001         struct dnode_of_data dn;
1002         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1003         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
1004         pgoff_t pgofs, end_offset, end;
1005         int err = 0, ofs = 1;
1006         unsigned int ofs_in_node, last_ofs_in_node;
1007         blkcnt_t prealloc;
1008         struct extent_info ei = {0,0,0};
1009         block_t blkaddr;
1010         unsigned int start_pgofs;
1011
1012         if (!maxblocks)
1013                 return 0;
1014
1015         map->m_len = 0;
1016         map->m_flags = 0;
1017
1018         /* it only supports block size == page size */
1019         pgofs = (pgoff_t)map->m_lblk;
1020         end = pgofs + maxblocks;
1021
1022         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1023                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1024                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1025                 map->m_flags = F2FS_MAP_MAPPED;
1026                 if (map->m_next_extent)
1027                         *map->m_next_extent = pgofs + map->m_len;
1028                 goto out;
1029         }
1030
1031 next_dnode:
1032         if (create)
1033                 __do_map_lock(sbi, flag, true);
1034
1035         /* When reading holes, we need its node page */
1036         set_new_dnode(&dn, inode, NULL, NULL, 0);
1037         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1038         if (err) {
1039                 if (flag == F2FS_GET_BLOCK_BMAP)
1040                         map->m_pblk = 0;
1041                 if (err == -ENOENT) {
1042                         err = 0;
1043                         if (map->m_next_pgofs)
1044                                 *map->m_next_pgofs =
1045                                         f2fs_get_next_page_offset(&dn, pgofs);
1046                         if (map->m_next_extent)
1047                                 *map->m_next_extent =
1048                                         f2fs_get_next_page_offset(&dn, pgofs);
1049                 }
1050                 goto unlock_out;
1051         }
1052
1053         start_pgofs = pgofs;
1054         prealloc = 0;
1055         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1056         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1057
1058 next_block:
1059         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1060
1061         if (__is_valid_data_blkaddr(blkaddr) &&
1062                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1063                 err = -EFAULT;
1064                 goto sync_out;
1065         }
1066
1067         if (!is_valid_data_blkaddr(sbi, blkaddr)) {
1068                 if (create) {
1069                         if (unlikely(f2fs_cp_error(sbi))) {
1070                                 err = -EIO;
1071                                 goto sync_out;
1072                         }
1073                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1074                                 if (blkaddr == NULL_ADDR) {
1075                                         prealloc++;
1076                                         last_ofs_in_node = dn.ofs_in_node;
1077                                 }
1078                         } else {
1079                                 err = __allocate_data_block(&dn,
1080                                                         map->m_seg_type);
1081                                 if (!err)
1082                                         set_inode_flag(inode, FI_APPEND_WRITE);
1083                         }
1084                         if (err)
1085                                 goto sync_out;
1086                         map->m_flags |= F2FS_MAP_NEW;
1087                         blkaddr = dn.data_blkaddr;
1088                 } else {
1089                         if (flag == F2FS_GET_BLOCK_BMAP) {
1090                                 map->m_pblk = 0;
1091                                 goto sync_out;
1092                         }
1093                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1094                                 goto sync_out;
1095                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1096                                                 blkaddr == NULL_ADDR) {
1097                                 if (map->m_next_pgofs)
1098                                         *map->m_next_pgofs = pgofs + 1;
1099                                 goto sync_out;
1100                         }
1101                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1102                                 /* for defragment case */
1103                                 if (map->m_next_pgofs)
1104                                         *map->m_next_pgofs = pgofs + 1;
1105                                 goto sync_out;
1106                         }
1107                 }
1108         }
1109
1110         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1111                 goto skip;
1112
1113         if (map->m_len == 0) {
1114                 /* preallocated unwritten block should be mapped for fiemap. */
1115                 if (blkaddr == NEW_ADDR)
1116                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1117                 map->m_flags |= F2FS_MAP_MAPPED;
1118
1119                 map->m_pblk = blkaddr;
1120                 map->m_len = 1;
1121         } else if ((map->m_pblk != NEW_ADDR &&
1122                         blkaddr == (map->m_pblk + ofs)) ||
1123                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1124                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1125                 ofs++;
1126                 map->m_len++;
1127         } else {
1128                 goto sync_out;
1129         }
1130
1131 skip:
1132         dn.ofs_in_node++;
1133         pgofs++;
1134
1135         /* preallocate blocks in batch for one dnode page */
1136         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1137                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1138
1139                 dn.ofs_in_node = ofs_in_node;
1140                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1141                 if (err)
1142                         goto sync_out;
1143
1144                 map->m_len += dn.ofs_in_node - ofs_in_node;
1145                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1146                         err = -ENOSPC;
1147                         goto sync_out;
1148                 }
1149                 dn.ofs_in_node = end_offset;
1150         }
1151
1152         if (pgofs >= end)
1153                 goto sync_out;
1154         else if (dn.ofs_in_node < end_offset)
1155                 goto next_block;
1156
1157         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1158                 if (map->m_flags & F2FS_MAP_MAPPED) {
1159                         unsigned int ofs = start_pgofs - map->m_lblk;
1160
1161                         f2fs_update_extent_cache_range(&dn,
1162                                 start_pgofs, map->m_pblk + ofs,
1163                                 map->m_len - ofs);
1164                 }
1165         }
1166
1167         f2fs_put_dnode(&dn);
1168
1169         if (create) {
1170                 __do_map_lock(sbi, flag, false);
1171                 f2fs_balance_fs(sbi, dn.node_changed);
1172         }
1173         goto next_dnode;
1174
1175 sync_out:
1176         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1177                 if (map->m_flags & F2FS_MAP_MAPPED) {
1178                         unsigned int ofs = start_pgofs - map->m_lblk;
1179
1180                         f2fs_update_extent_cache_range(&dn,
1181                                 start_pgofs, map->m_pblk + ofs,
1182                                 map->m_len - ofs);
1183                 }
1184                 if (map->m_next_extent)
1185                         *map->m_next_extent = pgofs + 1;
1186         }
1187         f2fs_put_dnode(&dn);
1188 unlock_out:
1189         if (create) {
1190                 __do_map_lock(sbi, flag, false);
1191                 f2fs_balance_fs(sbi, dn.node_changed);
1192         }
1193 out:
1194         trace_f2fs_map_blocks(inode, map, err);
1195         return err;
1196 }
1197
1198 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1199 {
1200         struct f2fs_map_blocks map;
1201         block_t last_lblk;
1202         int err;
1203
1204         if (pos + len > i_size_read(inode))
1205                 return false;
1206
1207         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1208         map.m_next_pgofs = NULL;
1209         map.m_next_extent = NULL;
1210         map.m_seg_type = NO_CHECK_TYPE;
1211         last_lblk = F2FS_BLK_ALIGN(pos + len);
1212
1213         while (map.m_lblk < last_lblk) {
1214                 map.m_len = last_lblk - map.m_lblk;
1215                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1216                 if (err || map.m_len == 0)
1217                         return false;
1218                 map.m_lblk += map.m_len;
1219         }
1220         return true;
1221 }
1222
1223 static int __get_data_block(struct inode *inode, sector_t iblock,
1224                         struct buffer_head *bh, int create, int flag,
1225                         pgoff_t *next_pgofs, int seg_type)
1226 {
1227         struct f2fs_map_blocks map;
1228         int err;
1229
1230         map.m_lblk = iblock;
1231         map.m_len = bh->b_size >> inode->i_blkbits;
1232         map.m_next_pgofs = next_pgofs;
1233         map.m_next_extent = NULL;
1234         map.m_seg_type = seg_type;
1235
1236         err = f2fs_map_blocks(inode, &map, create, flag);
1237         if (!err) {
1238                 map_bh(bh, inode->i_sb, map.m_pblk);
1239                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1240                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1241         }
1242         return err;
1243 }
1244
1245 static int get_data_block(struct inode *inode, sector_t iblock,
1246                         struct buffer_head *bh_result, int create, int flag,
1247                         pgoff_t *next_pgofs)
1248 {
1249         return __get_data_block(inode, iblock, bh_result, create,
1250                                                         flag, next_pgofs,
1251                                                         NO_CHECK_TYPE);
1252 }
1253
1254 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1255                         struct buffer_head *bh_result, int create)
1256 {
1257         return __get_data_block(inode, iblock, bh_result, create,
1258                                                 F2FS_GET_BLOCK_DEFAULT, NULL,
1259                                                 f2fs_rw_hint_to_seg_type(
1260                                                         inode->i_write_hint));
1261 }
1262
1263 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1264                         struct buffer_head *bh_result, int create)
1265 {
1266         /* Block number less than F2FS MAX BLOCKS */
1267         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1268                 return -EFBIG;
1269
1270         return __get_data_block(inode, iblock, bh_result, create,
1271                                                 F2FS_GET_BLOCK_BMAP, NULL,
1272                                                 NO_CHECK_TYPE);
1273 }
1274
1275 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1276 {
1277         return (offset >> inode->i_blkbits);
1278 }
1279
1280 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1281 {
1282         return (blk << inode->i_blkbits);
1283 }
1284
1285 static int f2fs_xattr_fiemap(struct inode *inode,
1286                                 struct fiemap_extent_info *fieinfo)
1287 {
1288         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1289         struct page *page;
1290         struct node_info ni;
1291         __u64 phys = 0, len;
1292         __u32 flags;
1293         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1294         int err = 0;
1295
1296         if (f2fs_has_inline_xattr(inode)) {
1297                 int offset;
1298
1299                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1300                                                 inode->i_ino, false);
1301                 if (!page)
1302                         return -ENOMEM;
1303
1304                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1305                 if (err) {
1306                         f2fs_put_page(page, 1);
1307                         return err;
1308                 }
1309
1310                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1311                 offset = offsetof(struct f2fs_inode, i_addr) +
1312                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1313                                         get_inline_xattr_addrs(inode));
1314
1315                 phys += offset;
1316                 len = inline_xattr_size(inode);
1317
1318                 f2fs_put_page(page, 1);
1319
1320                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1321
1322                 if (!xnid)
1323                         flags |= FIEMAP_EXTENT_LAST;
1324
1325                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1326                 if (err || err == 1)
1327                         return err;
1328         }
1329
1330         if (xnid) {
1331                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1332                 if (!page)
1333                         return -ENOMEM;
1334
1335                 err = f2fs_get_node_info(sbi, xnid, &ni);
1336                 if (err) {
1337                         f2fs_put_page(page, 1);
1338                         return err;
1339                 }
1340
1341                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1342                 len = inode->i_sb->s_blocksize;
1343
1344                 f2fs_put_page(page, 1);
1345
1346                 flags = FIEMAP_EXTENT_LAST;
1347         }
1348
1349         if (phys)
1350                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1351
1352         return (err < 0 ? err : 0);
1353 }
1354
1355 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1356                 u64 start, u64 len)
1357 {
1358         struct buffer_head map_bh;
1359         sector_t start_blk, last_blk;
1360         pgoff_t next_pgofs;
1361         u64 logical = 0, phys = 0, size = 0;
1362         u32 flags = 0;
1363         int ret = 0;
1364
1365         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1366                 ret = f2fs_precache_extents(inode);
1367                 if (ret)
1368                         return ret;
1369         }
1370
1371         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1372         if (ret)
1373                 return ret;
1374
1375         inode_lock(inode);
1376
1377         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1378                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1379                 goto out;
1380         }
1381
1382         if (f2fs_has_inline_data(inode)) {
1383                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1384                 if (ret != -EAGAIN)
1385                         goto out;
1386         }
1387
1388         if (logical_to_blk(inode, len) == 0)
1389                 len = blk_to_logical(inode, 1);
1390
1391         start_blk = logical_to_blk(inode, start);
1392         last_blk = logical_to_blk(inode, start + len - 1);
1393
1394 next:
1395         memset(&map_bh, 0, sizeof(struct buffer_head));
1396         map_bh.b_size = len;
1397
1398         ret = get_data_block(inode, start_blk, &map_bh, 0,
1399                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1400         if (ret)
1401                 goto out;
1402
1403         /* HOLE */
1404         if (!buffer_mapped(&map_bh)) {
1405                 start_blk = next_pgofs;
1406
1407                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1408                                         F2FS_I_SB(inode)->max_file_blocks))
1409                         goto prep_next;
1410
1411                 flags |= FIEMAP_EXTENT_LAST;
1412         }
1413
1414         if (size) {
1415                 if (f2fs_encrypted_inode(inode))
1416                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1417
1418                 ret = fiemap_fill_next_extent(fieinfo, logical,
1419                                 phys, size, flags);
1420         }
1421
1422         if (start_blk > last_blk || ret)
1423                 goto out;
1424
1425         logical = blk_to_logical(inode, start_blk);
1426         phys = blk_to_logical(inode, map_bh.b_blocknr);
1427         size = map_bh.b_size;
1428         flags = 0;
1429         if (buffer_unwritten(&map_bh))
1430                 flags = FIEMAP_EXTENT_UNWRITTEN;
1431
1432         start_blk += logical_to_blk(inode, size);
1433
1434 prep_next:
1435         cond_resched();
1436         if (fatal_signal_pending(current))
1437                 ret = -EINTR;
1438         else
1439                 goto next;
1440 out:
1441         if (ret == 1)
1442                 ret = 0;
1443
1444         inode_unlock(inode);
1445         return ret;
1446 }
1447
1448 /*
1449  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1450  * Major change was from block_size == page_size in f2fs by default.
1451  *
1452  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1453  * this function ever deviates from doing just read-ahead, it should either
1454  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1455  * from read-ahead.
1456  */
1457 static int f2fs_mpage_readpages(struct address_space *mapping,
1458                         struct list_head *pages, struct page *page,
1459                         unsigned nr_pages, bool is_readahead)
1460 {
1461         struct bio *bio = NULL;
1462         sector_t last_block_in_bio = 0;
1463         struct inode *inode = mapping->host;
1464         const unsigned blkbits = inode->i_blkbits;
1465         const unsigned blocksize = 1 << blkbits;
1466         sector_t block_in_file;
1467         sector_t last_block;
1468         sector_t last_block_in_file;
1469         sector_t block_nr;
1470         struct f2fs_map_blocks map;
1471
1472         map.m_pblk = 0;
1473         map.m_lblk = 0;
1474         map.m_len = 0;
1475         map.m_flags = 0;
1476         map.m_next_pgofs = NULL;
1477         map.m_next_extent = NULL;
1478         map.m_seg_type = NO_CHECK_TYPE;
1479
1480         for (; nr_pages; nr_pages--) {
1481                 if (pages) {
1482                         page = list_last_entry(pages, struct page, lru);
1483
1484                         prefetchw(&page->flags);
1485                         list_del(&page->lru);
1486                         if (add_to_page_cache_lru(page, mapping,
1487                                                   page->index,
1488                                                   readahead_gfp_mask(mapping)))
1489                                 goto next_page;
1490                 }
1491
1492                 block_in_file = (sector_t)page->index;
1493                 last_block = block_in_file + nr_pages;
1494                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1495                                                                 blkbits;
1496                 if (last_block > last_block_in_file)
1497                         last_block = last_block_in_file;
1498
1499                 /*
1500                  * Map blocks using the previous result first.
1501                  */
1502                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1503                                 block_in_file > map.m_lblk &&
1504                                 block_in_file < (map.m_lblk + map.m_len))
1505                         goto got_it;
1506
1507                 /*
1508                  * Then do more f2fs_map_blocks() calls until we are
1509                  * done with this page.
1510                  */
1511                 map.m_flags = 0;
1512
1513                 if (block_in_file < last_block) {
1514                         map.m_lblk = block_in_file;
1515                         map.m_len = last_block - block_in_file;
1516
1517                         if (f2fs_map_blocks(inode, &map, 0,
1518                                                 F2FS_GET_BLOCK_DEFAULT))
1519                                 goto set_error_page;
1520                 }
1521 got_it:
1522                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1523                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1524                         SetPageMappedToDisk(page);
1525
1526                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1527                                 SetPageUptodate(page);
1528                                 goto confused;
1529                         }
1530
1531                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1532                                                                 DATA_GENERIC))
1533                                 goto set_error_page;
1534                 } else {
1535                         zero_user_segment(page, 0, PAGE_SIZE);
1536                         if (!PageUptodate(page))
1537                                 SetPageUptodate(page);
1538                         unlock_page(page);
1539                         goto next_page;
1540                 }
1541
1542                 /*
1543                  * This page will go to BIO.  Do we need to send this
1544                  * BIO off first?
1545                  */
1546                 if (bio && (last_block_in_bio != block_nr - 1 ||
1547                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1548 submit_and_realloc:
1549                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1550                         bio = NULL;
1551                 }
1552                 if (bio == NULL) {
1553                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1554                                         is_readahead ? REQ_RAHEAD : 0);
1555                         if (IS_ERR(bio)) {
1556                                 bio = NULL;
1557                                 goto set_error_page;
1558                         }
1559                 }
1560
1561                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1562                         goto submit_and_realloc;
1563
1564                 last_block_in_bio = block_nr;
1565                 goto next_page;
1566 set_error_page:
1567                 SetPageError(page);
1568                 zero_user_segment(page, 0, PAGE_SIZE);
1569                 unlock_page(page);
1570                 goto next_page;
1571 confused:
1572                 if (bio) {
1573                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1574                         bio = NULL;
1575                 }
1576                 unlock_page(page);
1577 next_page:
1578                 if (pages)
1579                         put_page(page);
1580         }
1581         BUG_ON(pages && !list_empty(pages));
1582         if (bio)
1583                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1584         return 0;
1585 }
1586
1587 static int f2fs_read_data_page(struct file *file, struct page *page)
1588 {
1589         struct inode *inode = page->mapping->host;
1590         int ret = -EAGAIN;
1591
1592         trace_f2fs_readpage(page, DATA);
1593
1594         /* If the file has inline data, try to read it directly */
1595         if (f2fs_has_inline_data(inode))
1596                 ret = f2fs_read_inline_data(inode, page);
1597         if (ret == -EAGAIN)
1598                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1599         return ret;
1600 }
1601
1602 static int f2fs_read_data_pages(struct file *file,
1603                         struct address_space *mapping,
1604                         struct list_head *pages, unsigned nr_pages)
1605 {
1606         struct inode *inode = mapping->host;
1607         struct page *page = list_last_entry(pages, struct page, lru);
1608
1609         trace_f2fs_readpages(inode, page, nr_pages);
1610
1611         /* If the file has inline data, skip readpages */
1612         if (f2fs_has_inline_data(inode))
1613                 return 0;
1614
1615         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1616 }
1617
1618 static int encrypt_one_page(struct f2fs_io_info *fio)
1619 {
1620         struct inode *inode = fio->page->mapping->host;
1621         struct page *mpage;
1622         gfp_t gfp_flags = GFP_NOFS;
1623
1624         if (!f2fs_encrypted_file(inode))
1625                 return 0;
1626
1627         /* wait for GCed page writeback via META_MAPPING */
1628         f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1629
1630 retry_encrypt:
1631         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1632                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1633         if (IS_ERR(fio->encrypted_page)) {
1634                 /* flush pending IOs and wait for a while in the ENOMEM case */
1635                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1636                         f2fs_flush_merged_writes(fio->sbi);
1637                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1638                         gfp_flags |= __GFP_NOFAIL;
1639                         goto retry_encrypt;
1640                 }
1641                 return PTR_ERR(fio->encrypted_page);
1642         }
1643
1644         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1645         if (mpage) {
1646                 if (PageUptodate(mpage))
1647                         memcpy(page_address(mpage),
1648                                 page_address(fio->encrypted_page), PAGE_SIZE);
1649                 f2fs_put_page(mpage, 1);
1650         }
1651         return 0;
1652 }
1653
1654 static inline bool check_inplace_update_policy(struct inode *inode,
1655                                 struct f2fs_io_info *fio)
1656 {
1657         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1658         unsigned int policy = SM_I(sbi)->ipu_policy;
1659
1660         if (policy & (0x1 << F2FS_IPU_FORCE))
1661                 return true;
1662         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1663                 return true;
1664         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1665                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1666                 return true;
1667         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1668                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1669                 return true;
1670
1671         /*
1672          * IPU for rewrite async pages
1673          */
1674         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1675                         fio && fio->op == REQ_OP_WRITE &&
1676                         !(fio->op_flags & REQ_SYNC) &&
1677                         !f2fs_encrypted_inode(inode))
1678                 return true;
1679
1680         /* this is only set during fdatasync */
1681         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1682                         is_inode_flag_set(inode, FI_NEED_IPU))
1683                 return true;
1684
1685         return false;
1686 }
1687
1688 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1689 {
1690         if (f2fs_is_pinned_file(inode))
1691                 return true;
1692
1693         /* if this is cold file, we should overwrite to avoid fragmentation */
1694         if (file_is_cold(inode))
1695                 return true;
1696
1697         return check_inplace_update_policy(inode, fio);
1698 }
1699
1700 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1701 {
1702         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1703
1704         if (test_opt(sbi, LFS))
1705                 return true;
1706         if (S_ISDIR(inode->i_mode))
1707                 return true;
1708         if (f2fs_is_atomic_file(inode))
1709                 return true;
1710         if (fio) {
1711                 if (is_cold_data(fio->page))
1712                         return true;
1713                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1714                         return true;
1715         }
1716         return false;
1717 }
1718
1719 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1720 {
1721         struct inode *inode = fio->page->mapping->host;
1722
1723         if (f2fs_should_update_outplace(inode, fio))
1724                 return false;
1725
1726         return f2fs_should_update_inplace(inode, fio);
1727 }
1728
1729 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1730 {
1731         struct page *page = fio->page;
1732         struct inode *inode = page->mapping->host;
1733         struct dnode_of_data dn;
1734         struct extent_info ei = {0,0,0};
1735         struct node_info ni;
1736         bool ipu_force = false;
1737         int err = 0;
1738
1739         set_new_dnode(&dn, inode, NULL, NULL, 0);
1740         if (need_inplace_update(fio) &&
1741                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1742                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1743
1744                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1745                                                         DATA_GENERIC))
1746                         return -EFAULT;
1747
1748                 ipu_force = true;
1749                 fio->need_lock = LOCK_DONE;
1750                 goto got_it;
1751         }
1752
1753         /* Deadlock due to between page->lock and f2fs_lock_op */
1754         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1755                 return -EAGAIN;
1756
1757         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1758         if (err)
1759                 goto out;
1760
1761         fio->old_blkaddr = dn.data_blkaddr;
1762
1763         /* This page is already truncated */
1764         if (fio->old_blkaddr == NULL_ADDR) {
1765                 ClearPageUptodate(page);
1766                 goto out_writepage;
1767         }
1768 got_it:
1769         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1770                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1771                                                         DATA_GENERIC)) {
1772                 err = -EFAULT;
1773                 goto out_writepage;
1774         }
1775         /*
1776          * If current allocation needs SSR,
1777          * it had better in-place writes for updated data.
1778          */
1779         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1780                                         need_inplace_update(fio))) {
1781                 err = encrypt_one_page(fio);
1782                 if (err)
1783                         goto out_writepage;
1784
1785                 set_page_writeback(page);
1786                 ClearPageError(page);
1787                 f2fs_put_dnode(&dn);
1788                 if (fio->need_lock == LOCK_REQ)
1789                         f2fs_unlock_op(fio->sbi);
1790                 err = f2fs_inplace_write_data(fio);
1791                 trace_f2fs_do_write_data_page(fio->page, IPU);
1792                 set_inode_flag(inode, FI_UPDATE_WRITE);
1793                 return err;
1794         }
1795
1796         if (fio->need_lock == LOCK_RETRY) {
1797                 if (!f2fs_trylock_op(fio->sbi)) {
1798                         err = -EAGAIN;
1799                         goto out_writepage;
1800                 }
1801                 fio->need_lock = LOCK_REQ;
1802         }
1803
1804         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1805         if (err)
1806                 goto out_writepage;
1807
1808         fio->version = ni.version;
1809
1810         err = encrypt_one_page(fio);
1811         if (err)
1812                 goto out_writepage;
1813
1814         set_page_writeback(page);
1815         ClearPageError(page);
1816
1817         /* LFS mode write path */
1818         f2fs_outplace_write_data(&dn, fio);
1819         trace_f2fs_do_write_data_page(page, OPU);
1820         set_inode_flag(inode, FI_APPEND_WRITE);
1821         if (page->index == 0)
1822                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1823 out_writepage:
1824         f2fs_put_dnode(&dn);
1825 out:
1826         if (fio->need_lock == LOCK_REQ)
1827                 f2fs_unlock_op(fio->sbi);
1828         return err;
1829 }
1830
1831 static int __write_data_page(struct page *page, bool *submitted,
1832                                 struct writeback_control *wbc,
1833                                 enum iostat_type io_type)
1834 {
1835         struct inode *inode = page->mapping->host;
1836         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1837         loff_t i_size = i_size_read(inode);
1838         const pgoff_t end_index = ((unsigned long long) i_size)
1839                                                         >> PAGE_SHIFT;
1840         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1841         unsigned offset = 0;
1842         bool need_balance_fs = false;
1843         int err = 0;
1844         struct f2fs_io_info fio = {
1845                 .sbi = sbi,
1846                 .ino = inode->i_ino,
1847                 .type = DATA,
1848                 .op = REQ_OP_WRITE,
1849                 .op_flags = wbc_to_write_flags(wbc),
1850                 .old_blkaddr = NULL_ADDR,
1851                 .page = page,
1852                 .encrypted_page = NULL,
1853                 .submitted = false,
1854                 .need_lock = LOCK_RETRY,
1855                 .io_type = io_type,
1856                 .io_wbc = wbc,
1857         };
1858
1859         trace_f2fs_writepage(page, DATA);
1860
1861         /* we should bypass data pages to proceed the kworkder jobs */
1862         if (unlikely(f2fs_cp_error(sbi))) {
1863                 mapping_set_error(page->mapping, -EIO);
1864                 /*
1865                  * don't drop any dirty dentry pages for keeping lastest
1866                  * directory structure.
1867                  */
1868                 if (S_ISDIR(inode->i_mode))
1869                         goto redirty_out;
1870                 goto out;
1871         }
1872
1873         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1874                 goto redirty_out;
1875
1876         if (page->index < end_index)
1877                 goto write;
1878
1879         /*
1880          * If the offset is out-of-range of file size,
1881          * this page does not have to be written to disk.
1882          */
1883         offset = i_size & (PAGE_SIZE - 1);
1884         if ((page->index >= end_index + 1) || !offset)
1885                 goto out;
1886
1887         zero_user_segment(page, offset, PAGE_SIZE);
1888 write:
1889         if (f2fs_is_drop_cache(inode))
1890                 goto out;
1891         /* we should not write 0'th page having journal header */
1892         if (f2fs_is_volatile_file(inode) && (!page->index ||
1893                         (!wbc->for_reclaim &&
1894                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1895                 goto redirty_out;
1896
1897         /* Dentry blocks are controlled by checkpoint */
1898         if (S_ISDIR(inode->i_mode)) {
1899                 fio.need_lock = LOCK_DONE;
1900                 err = f2fs_do_write_data_page(&fio);
1901                 goto done;
1902         }
1903
1904         if (!wbc->for_reclaim)
1905                 need_balance_fs = true;
1906         else if (has_not_enough_free_secs(sbi, 0, 0))
1907                 goto redirty_out;
1908         else
1909                 set_inode_flag(inode, FI_HOT_DATA);
1910
1911         err = -EAGAIN;
1912         if (f2fs_has_inline_data(inode)) {
1913                 err = f2fs_write_inline_data(inode, page);
1914                 if (!err)
1915                         goto out;
1916         }
1917
1918         if (err == -EAGAIN) {
1919                 err = f2fs_do_write_data_page(&fio);
1920                 if (err == -EAGAIN) {
1921                         fio.need_lock = LOCK_REQ;
1922                         err = f2fs_do_write_data_page(&fio);
1923                 }
1924         }
1925
1926         if (err) {
1927                 file_set_keep_isize(inode);
1928         } else {
1929                 down_write(&F2FS_I(inode)->i_sem);
1930                 if (F2FS_I(inode)->last_disk_size < psize)
1931                         F2FS_I(inode)->last_disk_size = psize;
1932                 up_write(&F2FS_I(inode)->i_sem);
1933         }
1934
1935 done:
1936         if (err && err != -ENOENT)
1937                 goto redirty_out;
1938
1939 out:
1940         inode_dec_dirty_pages(inode);
1941         if (err)
1942                 ClearPageUptodate(page);
1943
1944         if (wbc->for_reclaim) {
1945                 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1946                 clear_inode_flag(inode, FI_HOT_DATA);
1947                 f2fs_remove_dirty_inode(inode);
1948                 submitted = NULL;
1949         }
1950
1951         unlock_page(page);
1952         if (!S_ISDIR(inode->i_mode))
1953                 f2fs_balance_fs(sbi, need_balance_fs);
1954
1955         if (unlikely(f2fs_cp_error(sbi))) {
1956                 f2fs_submit_merged_write(sbi, DATA);
1957                 submitted = NULL;
1958         }
1959
1960         if (submitted)
1961                 *submitted = fio.submitted;
1962
1963         return 0;
1964
1965 redirty_out:
1966         redirty_page_for_writepage(wbc, page);
1967         /*
1968          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
1969          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
1970          * file_write_and_wait_range() will see EIO error, which is critical
1971          * to return value of fsync() followed by atomic_write failure to user.
1972          */
1973         if (!err || wbc->for_reclaim)
1974                 return AOP_WRITEPAGE_ACTIVATE;
1975         unlock_page(page);
1976         return err;
1977 }
1978
1979 static int f2fs_write_data_page(struct page *page,
1980                                         struct writeback_control *wbc)
1981 {
1982         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1983 }
1984
1985 /*
1986  * This function was copied from write_cche_pages from mm/page-writeback.c.
1987  * The major change is making write step of cold data page separately from
1988  * warm/hot data page.
1989  */
1990 static int f2fs_write_cache_pages(struct address_space *mapping,
1991                                         struct writeback_control *wbc,
1992                                         enum iostat_type io_type)
1993 {
1994         int ret = 0;
1995         int done = 0;
1996         struct pagevec pvec;
1997         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1998         int nr_pages;
1999         pgoff_t uninitialized_var(writeback_index);
2000         pgoff_t index;
2001         pgoff_t end;            /* Inclusive */
2002         pgoff_t done_index;
2003         pgoff_t last_idx = ULONG_MAX;
2004         int cycled;
2005         int range_whole = 0;
2006         int tag;
2007
2008         pagevec_init(&pvec);
2009
2010         if (get_dirty_pages(mapping->host) <=
2011                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2012                 set_inode_flag(mapping->host, FI_HOT_DATA);
2013         else
2014                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2015
2016         if (wbc->range_cyclic) {
2017                 writeback_index = mapping->writeback_index; /* prev offset */
2018                 index = writeback_index;
2019                 if (index == 0)
2020                         cycled = 1;
2021                 else
2022                         cycled = 0;
2023                 end = -1;
2024         } else {
2025                 index = wbc->range_start >> PAGE_SHIFT;
2026                 end = wbc->range_end >> PAGE_SHIFT;
2027                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2028                         range_whole = 1;
2029                 cycled = 1; /* ignore range_cyclic tests */
2030         }
2031         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2032                 tag = PAGECACHE_TAG_TOWRITE;
2033         else
2034                 tag = PAGECACHE_TAG_DIRTY;
2035 retry:
2036         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2037                 tag_pages_for_writeback(mapping, index, end);
2038         done_index = index;
2039         while (!done && (index <= end)) {
2040                 int i;
2041
2042                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2043                                 tag);
2044                 if (nr_pages == 0)
2045                         break;
2046
2047                 for (i = 0; i < nr_pages; i++) {
2048                         struct page *page = pvec.pages[i];
2049                         bool submitted = false;
2050
2051                         /* give a priority to WB_SYNC threads */
2052                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2053                                         wbc->sync_mode == WB_SYNC_NONE) {
2054                                 done = 1;
2055                                 break;
2056                         }
2057
2058                         done_index = page->index;
2059 retry_write:
2060                         lock_page(page);
2061
2062                         if (unlikely(page->mapping != mapping)) {
2063 continue_unlock:
2064                                 unlock_page(page);
2065                                 continue;
2066                         }
2067
2068                         if (!PageDirty(page)) {
2069                                 /* someone wrote it for us */
2070                                 goto continue_unlock;
2071                         }
2072
2073                         if (PageWriteback(page)) {
2074                                 if (wbc->sync_mode != WB_SYNC_NONE)
2075                                         f2fs_wait_on_page_writeback(page,
2076                                                                 DATA, true);
2077                                 else
2078                                         goto continue_unlock;
2079                         }
2080
2081                         BUG_ON(PageWriteback(page));
2082                         if (!clear_page_dirty_for_io(page))
2083                                 goto continue_unlock;
2084
2085                         ret = __write_data_page(page, &submitted, wbc, io_type);
2086                         if (unlikely(ret)) {
2087                                 /*
2088                                  * keep nr_to_write, since vfs uses this to
2089                                  * get # of written pages.
2090                                  */
2091                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2092                                         unlock_page(page);
2093                                         ret = 0;
2094                                         continue;
2095                                 } else if (ret == -EAGAIN) {
2096                                         ret = 0;
2097                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2098                                                 cond_resched();
2099                                                 congestion_wait(BLK_RW_ASYNC,
2100                                                                         HZ/50);
2101                                                 goto retry_write;
2102                                         }
2103                                         continue;
2104                                 }
2105                                 done_index = page->index + 1;
2106                                 done = 1;
2107                                 break;
2108                         } else if (submitted) {
2109                                 last_idx = page->index;
2110                         }
2111
2112                         if (--wbc->nr_to_write <= 0 &&
2113                                         wbc->sync_mode == WB_SYNC_NONE) {
2114                                 done = 1;
2115                                 break;
2116                         }
2117                 }
2118                 pagevec_release(&pvec);
2119                 cond_resched();
2120         }
2121
2122         if (!cycled && !done) {
2123                 cycled = 1;
2124                 index = 0;
2125                 end = writeback_index - 1;
2126                 goto retry;
2127         }
2128         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2129                 mapping->writeback_index = done_index;
2130
2131         if (last_idx != ULONG_MAX)
2132                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2133                                                 0, last_idx, DATA);
2134
2135         return ret;
2136 }
2137
2138 static inline bool __should_serialize_io(struct inode *inode,
2139                                         struct writeback_control *wbc)
2140 {
2141         if (!S_ISREG(inode->i_mode))
2142                 return false;
2143         if (wbc->sync_mode != WB_SYNC_ALL)
2144                 return true;
2145         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2146                 return true;
2147         return false;
2148 }
2149
2150 static int __f2fs_write_data_pages(struct address_space *mapping,
2151                                                 struct writeback_control *wbc,
2152                                                 enum iostat_type io_type)
2153 {
2154         struct inode *inode = mapping->host;
2155         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2156         struct blk_plug plug;
2157         int ret;
2158         bool locked = false;
2159
2160         /* deal with chardevs and other special file */
2161         if (!mapping->a_ops->writepage)
2162                 return 0;
2163
2164         /* skip writing if there is no dirty page in this inode */
2165         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2166                 return 0;
2167
2168         /* during POR, we don't need to trigger writepage at all. */
2169         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2170                 goto skip_write;
2171
2172         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2173                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2174                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2175                 goto skip_write;
2176
2177         /* skip writing during file defragment */
2178         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2179                 goto skip_write;
2180
2181         trace_f2fs_writepages(mapping->host, wbc, DATA);
2182
2183         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2184         if (wbc->sync_mode == WB_SYNC_ALL)
2185                 atomic_inc(&sbi->wb_sync_req[DATA]);
2186         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2187                 goto skip_write;
2188
2189         if (__should_serialize_io(inode, wbc)) {
2190                 mutex_lock(&sbi->writepages);
2191                 locked = true;
2192         }
2193
2194         blk_start_plug(&plug);
2195         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2196         blk_finish_plug(&plug);
2197
2198         if (locked)
2199                 mutex_unlock(&sbi->writepages);
2200
2201         if (wbc->sync_mode == WB_SYNC_ALL)
2202                 atomic_dec(&sbi->wb_sync_req[DATA]);
2203         /*
2204          * if some pages were truncated, we cannot guarantee its mapping->host
2205          * to detect pending bios.
2206          */
2207
2208         f2fs_remove_dirty_inode(inode);
2209         return ret;
2210
2211 skip_write:
2212         wbc->pages_skipped += get_dirty_pages(inode);
2213         trace_f2fs_writepages(mapping->host, wbc, DATA);
2214         return 0;
2215 }
2216
2217 static int f2fs_write_data_pages(struct address_space *mapping,
2218                             struct writeback_control *wbc)
2219 {
2220         struct inode *inode = mapping->host;
2221
2222         return __f2fs_write_data_pages(mapping, wbc,
2223                         F2FS_I(inode)->cp_task == current ?
2224                         FS_CP_DATA_IO : FS_DATA_IO);
2225 }
2226
2227 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2228 {
2229         struct inode *inode = mapping->host;
2230         loff_t i_size = i_size_read(inode);
2231
2232         if (to > i_size) {
2233                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2234                 down_write(&F2FS_I(inode)->i_mmap_sem);
2235
2236                 truncate_pagecache(inode, i_size);
2237                 f2fs_truncate_blocks(inode, i_size, true);
2238
2239                 up_write(&F2FS_I(inode)->i_mmap_sem);
2240                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2241         }
2242 }
2243
2244 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2245                         struct page *page, loff_t pos, unsigned len,
2246                         block_t *blk_addr, bool *node_changed)
2247 {
2248         struct inode *inode = page->mapping->host;
2249         pgoff_t index = page->index;
2250         struct dnode_of_data dn;
2251         struct page *ipage;
2252         bool locked = false;
2253         struct extent_info ei = {0,0,0};
2254         int err = 0;
2255
2256         /*
2257          * we already allocated all the blocks, so we don't need to get
2258          * the block addresses when there is no need to fill the page.
2259          */
2260         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2261                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2262                 return 0;
2263
2264         if (f2fs_has_inline_data(inode) ||
2265                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2266                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
2267                 locked = true;
2268         }
2269 restart:
2270         /* check inline_data */
2271         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2272         if (IS_ERR(ipage)) {
2273                 err = PTR_ERR(ipage);
2274                 goto unlock_out;
2275         }
2276
2277         set_new_dnode(&dn, inode, ipage, ipage, 0);
2278
2279         if (f2fs_has_inline_data(inode)) {
2280                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2281                         f2fs_do_read_inline_data(page, ipage);
2282                         set_inode_flag(inode, FI_DATA_EXIST);
2283                         if (inode->i_nlink)
2284                                 set_inline_node(ipage);
2285                 } else {
2286                         err = f2fs_convert_inline_page(&dn, page);
2287                         if (err)
2288                                 goto out;
2289                         if (dn.data_blkaddr == NULL_ADDR)
2290                                 err = f2fs_get_block(&dn, index);
2291                 }
2292         } else if (locked) {
2293                 err = f2fs_get_block(&dn, index);
2294         } else {
2295                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2296                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2297                 } else {
2298                         /* hole case */
2299                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2300                         if (err || dn.data_blkaddr == NULL_ADDR) {
2301                                 f2fs_put_dnode(&dn);
2302                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2303                                                                 true);
2304                                 locked = true;
2305                                 goto restart;
2306                         }
2307                 }
2308         }
2309
2310         /* convert_inline_page can make node_changed */
2311         *blk_addr = dn.data_blkaddr;
2312         *node_changed = dn.node_changed;
2313 out:
2314         f2fs_put_dnode(&dn);
2315 unlock_out:
2316         if (locked)
2317                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
2318         return err;
2319 }
2320
2321 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2322                 loff_t pos, unsigned len, unsigned flags,
2323                 struct page **pagep, void **fsdata)
2324 {
2325         struct inode *inode = mapping->host;
2326         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2327         struct page *page = NULL;
2328         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2329         bool need_balance = false, drop_atomic = false;
2330         block_t blkaddr = NULL_ADDR;
2331         int err = 0;
2332
2333         trace_f2fs_write_begin(inode, pos, len, flags);
2334
2335         if ((f2fs_is_atomic_file(inode) &&
2336                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2337                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2338                 err = -ENOMEM;
2339                 drop_atomic = true;
2340                 goto fail;
2341         }
2342
2343         /*
2344          * We should check this at this moment to avoid deadlock on inode page
2345          * and #0 page. The locking rule for inline_data conversion should be:
2346          * lock_page(page #0) -> lock_page(inode_page)
2347          */
2348         if (index != 0) {
2349                 err = f2fs_convert_inline_inode(inode);
2350                 if (err)
2351                         goto fail;
2352         }
2353 repeat:
2354         /*
2355          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2356          * wait_for_stable_page. Will wait that below with our IO control.
2357          */
2358         page = f2fs_pagecache_get_page(mapping, index,
2359                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2360         if (!page) {
2361                 err = -ENOMEM;
2362                 goto fail;
2363         }
2364
2365         *pagep = page;
2366
2367         err = prepare_write_begin(sbi, page, pos, len,
2368                                         &blkaddr, &need_balance);
2369         if (err)
2370                 goto fail;
2371
2372         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2373                 unlock_page(page);
2374                 f2fs_balance_fs(sbi, true);
2375                 lock_page(page);
2376                 if (page->mapping != mapping) {
2377                         /* The page got truncated from under us */
2378                         f2fs_put_page(page, 1);
2379                         goto repeat;
2380                 }
2381         }
2382
2383         f2fs_wait_on_page_writeback(page, DATA, false);
2384
2385         /* wait for GCed page writeback via META_MAPPING */
2386         if (f2fs_post_read_required(inode))
2387                 f2fs_wait_on_block_writeback(sbi, blkaddr);
2388
2389         if (len == PAGE_SIZE || PageUptodate(page))
2390                 return 0;
2391
2392         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2393                 zero_user_segment(page, len, PAGE_SIZE);
2394                 return 0;
2395         }
2396
2397         if (blkaddr == NEW_ADDR) {
2398                 zero_user_segment(page, 0, PAGE_SIZE);
2399                 SetPageUptodate(page);
2400         } else {
2401                 err = f2fs_submit_page_read(inode, page, blkaddr);
2402                 if (err)
2403                         goto fail;
2404
2405                 lock_page(page);
2406                 if (unlikely(page->mapping != mapping)) {
2407                         f2fs_put_page(page, 1);
2408                         goto repeat;
2409                 }
2410                 if (unlikely(!PageUptodate(page))) {
2411                         err = -EIO;
2412                         goto fail;
2413                 }
2414         }
2415         return 0;
2416
2417 fail:
2418         f2fs_put_page(page, 1);
2419         f2fs_write_failed(mapping, pos + len);
2420         if (drop_atomic)
2421                 f2fs_drop_inmem_pages_all(sbi, false);
2422         return err;
2423 }
2424
2425 static int f2fs_write_end(struct file *file,
2426                         struct address_space *mapping,
2427                         loff_t pos, unsigned len, unsigned copied,
2428                         struct page *page, void *fsdata)
2429 {
2430         struct inode *inode = page->mapping->host;
2431
2432         trace_f2fs_write_end(inode, pos, len, copied);
2433
2434         /*
2435          * This should be come from len == PAGE_SIZE, and we expect copied
2436          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2437          * let generic_perform_write() try to copy data again through copied=0.
2438          */
2439         if (!PageUptodate(page)) {
2440                 if (unlikely(copied != len))
2441                         copied = 0;
2442                 else
2443                         SetPageUptodate(page);
2444         }
2445         if (!copied)
2446                 goto unlock_out;
2447
2448         set_page_dirty(page);
2449
2450         if (pos + copied > i_size_read(inode))
2451                 f2fs_i_size_write(inode, pos + copied);
2452 unlock_out:
2453         f2fs_put_page(page, 1);
2454         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2455         return copied;
2456 }
2457
2458 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2459                            loff_t offset)
2460 {
2461         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2462         unsigned blkbits = i_blkbits;
2463         unsigned blocksize_mask = (1 << blkbits) - 1;
2464         unsigned long align = offset | iov_iter_alignment(iter);
2465         struct block_device *bdev = inode->i_sb->s_bdev;
2466
2467         if (align & blocksize_mask) {
2468                 if (bdev)
2469                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2470                 blocksize_mask = (1 << blkbits) - 1;
2471                 if (align & blocksize_mask)
2472                         return -EINVAL;
2473                 return 1;
2474         }
2475         return 0;
2476 }
2477
2478 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2479 {
2480         struct address_space *mapping = iocb->ki_filp->f_mapping;
2481         struct inode *inode = mapping->host;
2482         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2483         size_t count = iov_iter_count(iter);
2484         loff_t offset = iocb->ki_pos;
2485         int rw = iov_iter_rw(iter);
2486         int err;
2487         enum rw_hint hint = iocb->ki_hint;
2488         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2489
2490         err = check_direct_IO(inode, iter, offset);
2491         if (err)
2492                 return err < 0 ? err : 0;
2493
2494         if (f2fs_force_buffered_io(inode, rw))
2495                 return 0;
2496
2497         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2498
2499         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2500                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2501
2502         if (!down_read_trylock(&F2FS_I(inode)->i_gc_rwsem[rw])) {
2503                 if (iocb->ki_flags & IOCB_NOWAIT) {
2504                         iocb->ki_hint = hint;
2505                         err = -EAGAIN;
2506                         goto out;
2507                 }
2508                 down_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2509         }
2510
2511         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2512         up_read(&F2FS_I(inode)->i_gc_rwsem[rw]);
2513
2514         if (rw == WRITE) {
2515                 if (whint_mode == WHINT_MODE_OFF)
2516                         iocb->ki_hint = hint;
2517                 if (err > 0) {
2518                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2519                                                                         err);
2520                         set_inode_flag(inode, FI_UPDATE_WRITE);
2521                 } else if (err < 0) {
2522                         f2fs_write_failed(mapping, offset + count);
2523                 }
2524         }
2525
2526 out:
2527         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2528
2529         return err;
2530 }
2531
2532 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2533                                                         unsigned int length)
2534 {
2535         struct inode *inode = page->mapping->host;
2536         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2537
2538         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2539                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2540                 return;
2541
2542         if (PageDirty(page)) {
2543                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2544                         dec_page_count(sbi, F2FS_DIRTY_META);
2545                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2546                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2547                 } else {
2548                         inode_dec_dirty_pages(inode);
2549                         f2fs_remove_dirty_inode(inode);
2550                 }
2551         }
2552
2553         /* This is atomic written page, keep Private */
2554         if (IS_ATOMIC_WRITTEN_PAGE(page))
2555                 return f2fs_drop_inmem_page(inode, page);
2556
2557         set_page_private(page, 0);
2558         ClearPagePrivate(page);
2559 }
2560
2561 int f2fs_release_page(struct page *page, gfp_t wait)
2562 {
2563         /* If this is dirty page, keep PagePrivate */
2564         if (PageDirty(page))
2565                 return 0;
2566
2567         /* This is atomic written page, keep Private */
2568         if (IS_ATOMIC_WRITTEN_PAGE(page))
2569                 return 0;
2570
2571         set_page_private(page, 0);
2572         ClearPagePrivate(page);
2573         return 1;
2574 }
2575
2576 static int f2fs_set_data_page_dirty(struct page *page)
2577 {
2578         struct address_space *mapping = page->mapping;
2579         struct inode *inode = mapping->host;
2580
2581         trace_f2fs_set_page_dirty(page, DATA);
2582
2583         if (!PageUptodate(page))
2584                 SetPageUptodate(page);
2585
2586         /* don't remain PG_checked flag which was set during GC */
2587         if (is_cold_data(page))
2588                 clear_cold_data(page);
2589
2590         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2591                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2592                         f2fs_register_inmem_page(inode, page);
2593                         return 1;
2594                 }
2595                 /*
2596                  * Previously, this page has been registered, we just
2597                  * return here.
2598                  */
2599                 return 0;
2600         }
2601
2602         if (!PageDirty(page)) {
2603                 __set_page_dirty_nobuffers(page);
2604                 f2fs_update_dirty_page(inode, page);
2605                 return 1;
2606         }
2607         return 0;
2608 }
2609
2610 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2611 {
2612         struct inode *inode = mapping->host;
2613
2614         if (f2fs_has_inline_data(inode))
2615                 return 0;
2616
2617         /* make sure allocating whole blocks */
2618         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2619                 filemap_write_and_wait(mapping);
2620
2621         return generic_block_bmap(mapping, block, get_data_block_bmap);
2622 }
2623
2624 #ifdef CONFIG_MIGRATION
2625 #include <linux/migrate.h>
2626
2627 int f2fs_migrate_page(struct address_space *mapping,
2628                 struct page *newpage, struct page *page, enum migrate_mode mode)
2629 {
2630         int rc, extra_count;
2631         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2632         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2633
2634         BUG_ON(PageWriteback(page));
2635
2636         /* migrating an atomic written page is safe with the inmem_lock hold */
2637         if (atomic_written) {
2638                 if (mode != MIGRATE_SYNC)
2639                         return -EBUSY;
2640                 if (!mutex_trylock(&fi->inmem_lock))
2641                         return -EAGAIN;
2642         }
2643
2644         /*
2645          * A reference is expected if PagePrivate set when move mapping,
2646          * however F2FS breaks this for maintaining dirty page counts when
2647          * truncating pages. So here adjusting the 'extra_count' make it work.
2648          */
2649         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2650         rc = migrate_page_move_mapping(mapping, newpage,
2651                                 page, NULL, mode, extra_count);
2652         if (rc != MIGRATEPAGE_SUCCESS) {
2653                 if (atomic_written)
2654                         mutex_unlock(&fi->inmem_lock);
2655                 return rc;
2656         }
2657
2658         if (atomic_written) {
2659                 struct inmem_pages *cur;
2660                 list_for_each_entry(cur, &fi->inmem_pages, list)
2661                         if (cur->page == page) {
2662                                 cur->page = newpage;
2663                                 break;
2664                         }
2665                 mutex_unlock(&fi->inmem_lock);
2666                 put_page(page);
2667                 get_page(newpage);
2668         }
2669
2670         if (PagePrivate(page))
2671                 SetPagePrivate(newpage);
2672         set_page_private(newpage, page_private(page));
2673
2674         if (mode != MIGRATE_SYNC_NO_COPY)
2675                 migrate_page_copy(newpage, page);
2676         else
2677                 migrate_page_states(newpage, page);
2678
2679         return MIGRATEPAGE_SUCCESS;
2680 }
2681 #endif
2682
2683 const struct address_space_operations f2fs_dblock_aops = {
2684         .readpage       = f2fs_read_data_page,
2685         .readpages      = f2fs_read_data_pages,
2686         .writepage      = f2fs_write_data_page,
2687         .writepages     = f2fs_write_data_pages,
2688         .write_begin    = f2fs_write_begin,
2689         .write_end      = f2fs_write_end,
2690         .set_page_dirty = f2fs_set_data_page_dirty,
2691         .invalidatepage = f2fs_invalidate_page,
2692         .releasepage    = f2fs_release_page,
2693         .direct_IO      = f2fs_direct_IO,
2694         .bmap           = f2fs_bmap,
2695 #ifdef CONFIG_MIGRATION
2696         .migratepage    = f2fs_migrate_page,
2697 #endif
2698 };
2699
2700 void f2fs_clear_radix_tree_dirty_tag(struct page *page)
2701 {
2702         struct address_space *mapping = page_mapping(page);
2703         unsigned long flags;
2704
2705         xa_lock_irqsave(&mapping->i_pages, flags);
2706         radix_tree_tag_clear(&mapping->i_pages, page_index(page),
2707                                                 PAGECACHE_TAG_DIRTY);
2708         xa_unlock_irqrestore(&mapping->i_pages, flags);
2709 }
2710
2711 int __init f2fs_init_post_read_processing(void)
2712 {
2713         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2714         if (!bio_post_read_ctx_cache)
2715                 goto fail;
2716         bio_post_read_ctx_pool =
2717                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2718                                          bio_post_read_ctx_cache);
2719         if (!bio_post_read_ctx_pool)
2720                 goto fail_free_cache;
2721         return 0;
2722
2723 fail_free_cache:
2724         kmem_cache_destroy(bio_post_read_ctx_cache);
2725 fail:
2726         return -ENOMEM;
2727 }
2728
2729 void __exit f2fs_destroy_post_read_processing(void)
2730 {
2731         mempool_destroy(bio_post_read_ctx_pool);
2732         kmem_cache_destroy(bio_post_read_ctx_cache);
2733 }