fs/f2fs/data.c

   1 /*
   2  * fs/f2fs/data.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/mpage.h>
  15 #include <linux/writeback.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/bio.h>
  20 #include <linux/prefetch.h>
  21 #include <linux/uio.h>
  22 #include <linux/mm.h>
  23 #include <linux/memcontrol.h>
  24 #include <linux/cleancache.h>
  25 #include <linux/sched/signal.h>
  26
  27 #include "f2fs.h"
  28 #include "node.h"
  29 #include "segment.h"
  30 #include "trace.h"
  31 #include <trace/events/f2fs.h>
  32
  33 static bool __is_cp_guaranteed(struct page *page)
  34 {
  35         struct address_space *mapping = page->mapping;
  36         struct inode *inode;
  37         struct f2fs_sb_info *sbi;
  38
  39         if (!mapping)
  40                 return false;
  41
  42         inode = mapping->host;
  43         sbi = F2FS_I_SB(inode);
  44
  45         if (inode->i_ino == F2FS_META_INO(sbi) ||
  46                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47                         S_ISDIR(inode->i_mode) ||
  48                         is_cold_data(page))
  49                 return true;
  50         return false;
  51 }
  52
  53 static void f2fs_read_end_io(struct bio *bio)
  54 {
  55         struct bio_vec *bvec;
  56         int i;
  57
  58 #ifdef CONFIG_F2FS_FAULT_INJECTION
  59         if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) {
  60                 f2fs_show_injection_info(FAULT_IO);
  61                 bio->bi_status = BLK_STS_IOERR;
  62         }
  63 #endif
  64
  65         if (f2fs_bio_encrypted(bio)) {
  66                 if (bio->bi_status) {
  67                         fscrypt_release_ctx(bio->bi_private);
  68                 } else {
  69                         fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  70                         return;
  71                 }
  72         }
  73
  74         bio_for_each_segment_all(bvec, bio, i) {
  75                 struct page *page = bvec->bv_page;
  76
  77                 if (!bio->bi_status) {
  78                         if (!PageUptodate(page))
  79                                 SetPageUptodate(page);
  80                 } else {
  81                         ClearPageUptodate(page);
  82                         SetPageError(page);
  83                 }
  84                 unlock_page(page);
  85         }
  86         bio_put(bio);
  87 }
  88
  89 static void f2fs_write_end_io(struct bio *bio)
  90 {
  91         struct f2fs_sb_info *sbi = bio->bi_private;
  92         struct bio_vec *bvec;
  93         int i;
  94
  95         bio_for_each_segment_all(bvec, bio, i) {
  96                 struct page *page = bvec->bv_page;
  97                 enum count_type type = WB_DATA_TYPE(page);
  98
  99                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 100                         set_page_private(page, (unsigned long)NULL);
 101                         ClearPagePrivate(page);
 102                         unlock_page(page);
 103                         mempool_free(page, sbi->write_io_dummy);
 104
 105                         if (unlikely(bio->bi_status))
 106                                 f2fs_stop_checkpoint(sbi, true);
 107                         continue;
 108                 }
 109
 110                 fscrypt_pullback_bio_page(&page, true);
 111
 112                 if (unlikely(bio->bi_status)) {
 113                         mapping_set_error(page->mapping, -EIO);
 114                         f2fs_stop_checkpoint(sbi, true);
 115                 }
 116                 dec_page_count(sbi, type);
 117                 clear_cold_data(page);
 118                 end_page_writeback(page);
 119         }
 120         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 121                                 wq_has_sleeper(&sbi->cp_wait))
 122                 wake_up(&sbi->cp_wait);
 123
 124         bio_put(bio);
 125 }
 126
 127 /*
 128  * Return true, if pre_bio's bdev is same as its target device.
 129  */
 130 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 131                                 block_t blk_addr, struct bio *bio)
 132 {
 133         struct block_device *bdev = sbi->sb->s_bdev;
 134         int i;
 135
 136         for (i = 0; i < sbi->s_ndevs; i++) {
 137                 if (FDEV(i).start_blk <= blk_addr &&
 138                                         FDEV(i).end_blk >= blk_addr) {
 139                         blk_addr -= FDEV(i).start_blk;
 140                         bdev = FDEV(i).bdev;
 141                         break;
 142                 }
 143         }
 144         if (bio) {
 145                 bio_set_dev(bio, bdev);
 146                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 147         }
 148         return bdev;
 149 }
 150
 151 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 152 {
 153         int i;
 154
 155         for (i = 0; i < sbi->s_ndevs; i++)
 156                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 157                         return i;
 158         return 0;
 159 }
 160
 161 static bool __same_bdev(struct f2fs_sb_info *sbi,
 162                                 block_t blk_addr, struct bio *bio)
 163 {
 164         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 165         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 166 }
 167
 168 /*
 169  * Low-level block read/write IO operations.
 170  */
 171 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 172                                 int npages, bool is_read)
 173 {
 174         struct bio *bio;
 175
 176         bio = f2fs_bio_alloc(npages);
 177
 178         f2fs_target_device(sbi, blk_addr, bio);
 179         bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
 180         bio->bi_private = is_read ? NULL : sbi;
 181
 182         return bio;
 183 }
 184
 185 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 186                                 struct bio *bio, enum page_type type)
 187 {
 188         if (!is_read_io(bio_op(bio))) {
 189                 unsigned int start;
 190
 191                 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
 192                         current->plug && (type == DATA || type == NODE))
 193                         blk_finish_plug(current->plug);
 194
 195                 if (type != DATA && type != NODE)
 196                         goto submit_io;
 197
 198                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 199                 start %= F2FS_IO_SIZE(sbi);
 200
 201                 if (start == 0)
 202                         goto submit_io;
 203
 204                 /* fill dummy pages */
 205                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 206                         struct page *page =
 207                                 mempool_alloc(sbi->write_io_dummy,
 208                                         GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 209                         f2fs_bug_on(sbi, !page);
 210
 211                         SetPagePrivate(page);
 212                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 213                         lock_page(page);
 214                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 215                                 f2fs_bug_on(sbi, 1);
 216                 }
 217                 /*
 218                  * In the NODE case, we lose next block address chain. So, we
 219                  * need to do checkpoint in f2fs_sync_file.
 220                  */
 221                 if (type == NODE)
 222                         set_sbi_flag(sbi, SBI_NEED_CP);
 223         }
 224 submit_io:
 225         if (is_read_io(bio_op(bio)))
 226                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 227         else
 228                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 229         submit_bio(bio);
 230 }
 231
 232 static void __submit_merged_bio(struct f2fs_bio_info *io)
 233 {
 234         struct f2fs_io_info *fio = &io->fio;
 235
 236         if (!io->bio)
 237                 return;
 238
 239         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 240
 241         if (is_read_io(fio->op))
 242                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 243         else
 244                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 245
 246         __submit_bio(io->sbi, io->bio, fio->type);
 247         io->bio = NULL;
 248 }
 249
 250 static bool __has_merged_page(struct f2fs_bio_info *io,
 251                                 struct inode *inode, nid_t ino, pgoff_t idx)
 252 {
 253         struct bio_vec *bvec;
 254         struct page *target;
 255         int i;
 256
 257         if (!io->bio)
 258                 return false;
 259
 260         if (!inode && !ino)
 261                 return true;
 262
 263         bio_for_each_segment_all(bvec, io->bio, i) {
 264
 265                 if (bvec->bv_page->mapping)
 266                         target = bvec->bv_page;
 267                 else
 268                         target = fscrypt_control_page(bvec->bv_page);
 269
 270                 if (idx != target->index)
 271                         continue;
 272
 273                 if (inode && inode == target->mapping->host)
 274                         return true;
 275                 if (ino && ino == ino_of_node(target))
 276                         return true;
 277         }
 278
 279         return false;
 280 }
 281
 282 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 283                                 nid_t ino, pgoff_t idx, enum page_type type)
 284 {
 285         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 286         enum temp_type temp;
 287         struct f2fs_bio_info *io;
 288         bool ret = false;
 289
 290         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 291                 io = sbi->write_io[btype] + temp;
 292
 293                 down_read(&io->io_rwsem);
 294                 ret = __has_merged_page(io, inode, ino, idx);
 295                 up_read(&io->io_rwsem);
 296
 297                 /* TODO: use HOT temp only for meta pages now. */
 298                 if (ret || btype == META)
 299                         break;
 300         }
 301         return ret;
 302 }
 303
 304 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 305                                 enum page_type type, enum temp_type temp)
 306 {
 307         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 308         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 309
 310         down_write(&io->io_rwsem);
 311
 312         /* change META to META_FLUSH in the checkpoint procedure */
 313         if (type >= META_FLUSH) {
 314                 io->fio.type = META_FLUSH;
 315                 io->fio.op = REQ_OP_WRITE;
 316                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 317                 if (!test_opt(sbi, NOBARRIER))
 318                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 319         }
 320         __submit_merged_bio(io);
 321         up_write(&io->io_rwsem);
 322 }
 323
 324 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 325                                 struct inode *inode, nid_t ino, pgoff_t idx,
 326                                 enum page_type type, bool force)
 327 {
 328         enum temp_type temp;
 329
 330         if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 331                 return;
 332
 333         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 334
 335                 __f2fs_submit_merged_write(sbi, type, temp);
 336
 337                 /* TODO: use HOT temp only for meta pages now. */
 338                 if (type >= META)
 339                         break;
 340         }
 341 }
 342
 343 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 344 {
 345         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 346 }
 347
 348 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 349                                 struct inode *inode, nid_t ino, pgoff_t idx,
 350                                 enum page_type type)
 351 {
 352         __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 353 }
 354
 355 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 356 {
 357         f2fs_submit_merged_write(sbi, DATA);
 358         f2fs_submit_merged_write(sbi, NODE);
 359         f2fs_submit_merged_write(sbi, META);
 360 }
 361
 362 /*
 363  * Fill the locked page with data located in the block address.
 364  * A caller needs to unlock the page on failure.
 365  */
 366 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 367 {
 368         struct bio *bio;
 369         struct page *page = fio->encrypted_page ?
 370                         fio->encrypted_page : fio->page;
 371
 372         trace_f2fs_submit_page_bio(page, fio);
 373         f2fs_trace_ios(fio, 0);
 374
 375         /* Allocate a new bio */
 376         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
 377
 378         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 379                 bio_put(bio);
 380                 return -EFAULT;
 381         }
 382         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 383
 384         __submit_bio(fio->sbi, bio, fio->type);
 385
 386         if (!is_read_io(fio->op))
 387                 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 388         return 0;
 389 }
 390
 391 int f2fs_submit_page_write(struct f2fs_io_info *fio)
 392 {
 393         struct f2fs_sb_info *sbi = fio->sbi;
 394         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 395         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 396         struct page *bio_page;
 397         int err = 0;
 398
 399         f2fs_bug_on(sbi, is_read_io(fio->op));
 400
 401         down_write(&io->io_rwsem);
 402 next:
 403         if (fio->in_list) {
 404                 spin_lock(&io->io_lock);
 405                 if (list_empty(&io->io_list)) {
 406                         spin_unlock(&io->io_lock);
 407                         goto out_fail;
 408                 }
 409                 fio = list_first_entry(&io->io_list,
 410                                                 struct f2fs_io_info, list);
 411                 list_del(&fio->list);
 412                 spin_unlock(&io->io_lock);
 413         }
 414
 415         if (fio->old_blkaddr != NEW_ADDR)
 416                 verify_block_addr(sbi, fio->old_blkaddr);
 417         verify_block_addr(sbi, fio->new_blkaddr);
 418
 419         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 420
 421         /* set submitted = 1 as a return value */
 422         fio->submitted = 1;
 423
 424         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 425
 426         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 427             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 428                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 429                 __submit_merged_bio(io);
 430 alloc_new:
 431         if (io->bio == NULL) {
 432                 if ((fio->type == DATA || fio->type == NODE) &&
 433                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 434                         err = -EAGAIN;
 435                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 436                         goto out_fail;
 437                 }
 438                 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
 439                                                 BIO_MAX_PAGES, false);
 440                 io->fio = *fio;
 441         }
 442
 443         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 444                 __submit_merged_bio(io);
 445                 goto alloc_new;
 446         }
 447
 448         io->last_block_in_bio = fio->new_blkaddr;
 449         f2fs_trace_ios(fio, 0);
 450
 451         trace_f2fs_submit_page_write(fio->page, fio);
 452
 453         if (fio->in_list)
 454                 goto next;
 455 out_fail:
 456         up_write(&io->io_rwsem);
 457         return err;
 458 }
 459
 460 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 461                                                          unsigned nr_pages)
 462 {
 463         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 464         struct fscrypt_ctx *ctx = NULL;
 465         struct bio *bio;
 466
 467         if (f2fs_encrypted_file(inode)) {
 468                 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 469                 if (IS_ERR(ctx))
 470                         return ERR_CAST(ctx);
 471
 472                 /* wait the page to be moved by cleaning */
 473                 f2fs_wait_on_block_writeback(sbi, blkaddr);
 474         }
 475
 476         bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
 477         if (!bio) {
 478                 if (ctx)
 479                         fscrypt_release_ctx(ctx);
 480                 return ERR_PTR(-ENOMEM);
 481         }
 482         f2fs_target_device(sbi, blkaddr, bio);
 483         bio->bi_end_io = f2fs_read_end_io;
 484         bio->bi_private = ctx;
 485         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 486
 487         return bio;
 488 }
 489
 490 /* This can handle encryption stuffs */
 491 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 492                                                         block_t blkaddr)
 493 {
 494         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
 495
 496         if (IS_ERR(bio))
 497                 return PTR_ERR(bio);
 498
 499         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 500                 bio_put(bio);
 501                 return -EFAULT;
 502         }
 503         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 504         return 0;
 505 }
 506
 507 static void __set_data_blkaddr(struct dnode_of_data *dn)
 508 {
 509         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 510         __le32 *addr_array;
 511         int base = 0;
 512
 513         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 514                 base = get_extra_isize(dn->inode);
 515
 516         /* Get physical address of data block */
 517         addr_array = blkaddr_in_node(rn);
 518         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 519 }
 520
 521 /*
 522  * Lock ordering for the change of data block address:
 523  * ->data_page
 524  *  ->node_page
 525  *    update block addresses in the node page
 526  */
 527 void set_data_blkaddr(struct dnode_of_data *dn)
 528 {
 529         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 530         __set_data_blkaddr(dn);
 531         if (set_page_dirty(dn->node_page))
 532                 dn->node_changed = true;
 533 }
 534
 535 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 536 {
 537         dn->data_blkaddr = blkaddr;
 538         set_data_blkaddr(dn);
 539         f2fs_update_extent_cache(dn);
 540 }
 541
 542 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 543 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 544 {
 545         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 546         int err;
 547
 548         if (!count)
 549                 return 0;
 550
 551         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 552                 return -EPERM;
 553         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 554                 return err;
 555
 556         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 557                                                 dn->ofs_in_node, count);
 558
 559         f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 560
 561         for (; count > 0; dn->ofs_in_node++) {
 562                 block_t blkaddr = datablock_addr(dn->inode,
 563                                         dn->node_page, dn->ofs_in_node);
 564                 if (blkaddr == NULL_ADDR) {
 565                         dn->data_blkaddr = NEW_ADDR;
 566                         __set_data_blkaddr(dn);
 567                         count--;
 568                 }
 569         }
 570
 571         if (set_page_dirty(dn->node_page))
 572                 dn->node_changed = true;
 573         return 0;
 574 }
 575
 576 /* Should keep dn->ofs_in_node unchanged */
 577 int reserve_new_block(struct dnode_of_data *dn)
 578 {
 579         unsigned int ofs_in_node = dn->ofs_in_node;
 580         int ret;
 581
 582         ret = reserve_new_blocks(dn, 1);
 583         dn->ofs_in_node = ofs_in_node;
 584         return ret;
 585 }
 586
 587 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 588 {
 589         bool need_put = dn->inode_page ? false : true;
 590         int err;
 591
 592         err = get_dnode_of_data(dn, index, ALLOC_NODE);
 593         if (err)
 594                 return err;
 595
 596         if (dn->data_blkaddr == NULL_ADDR)
 597                 err = reserve_new_block(dn);
 598         if (err || need_put)
 599                 f2fs_put_dnode(dn);
 600         return err;
 601 }
 602
 603 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 604 {
 605         struct extent_info ei  = {0,0,0};
 606         struct inode *inode = dn->inode;
 607
 608         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 609                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 610                 return 0;
 611         }
 612
 613         return f2fs_reserve_block(dn, index);
 614 }
 615
 616 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 617                                                 int op_flags, bool for_write)
 618 {
 619         struct address_space *mapping = inode->i_mapping;
 620         struct dnode_of_data dn;
 621         struct page *page;
 622         struct extent_info ei = {0,0,0};
 623         int err;
 624
 625         page = f2fs_grab_cache_page(mapping, index, for_write);
 626         if (!page)
 627                 return ERR_PTR(-ENOMEM);
 628
 629         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 630                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 631                 goto got_it;
 632         }
 633
 634         set_new_dnode(&dn, inode, NULL, NULL, 0);
 635         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 636         if (err)
 637                 goto put_err;
 638         f2fs_put_dnode(&dn);
 639
 640         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 641                 err = -ENOENT;
 642                 goto put_err;
 643         }
 644 got_it:
 645         if (PageUptodate(page)) {
 646                 unlock_page(page);
 647                 return page;
 648         }
 649
 650         /*
 651          * A new dentry page is allocated but not able to be written, since its
 652          * new inode page couldn't be allocated due to -ENOSPC.
 653          * In such the case, its blkaddr can be remained as NEW_ADDR.
 654          * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 655          */
 656         if (dn.data_blkaddr == NEW_ADDR) {
 657                 zero_user_segment(page, 0, PAGE_SIZE);
 658                 if (!PageUptodate(page))
 659                         SetPageUptodate(page);
 660                 unlock_page(page);
 661                 return page;
 662         }
 663
 664         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 665         if (err)
 666                 goto put_err;
 667         return page;
 668
 669 put_err:
 670         f2fs_put_page(page, 1);
 671         return ERR_PTR(err);
 672 }
 673
 674 struct page *find_data_page(struct inode *inode, pgoff_t index)
 675 {
 676         struct address_space *mapping = inode->i_mapping;
 677         struct page *page;
 678
 679         page = find_get_page(mapping, index);
 680         if (page && PageUptodate(page))
 681                 return page;
 682         f2fs_put_page(page, 0);
 683
 684         page = get_read_data_page(inode, index, 0, false);
 685         if (IS_ERR(page))
 686                 return page;
 687
 688         if (PageUptodate(page))
 689                 return page;
 690
 691         wait_on_page_locked(page);
 692         if (unlikely(!PageUptodate(page))) {
 693                 f2fs_put_page(page, 0);
 694                 return ERR_PTR(-EIO);
 695         }
 696         return page;
 697 }
 698
 699 /*
 700  * If it tries to access a hole, return an error.
 701  * Because, the callers, functions in dir.c and GC, should be able to know
 702  * whether this page exists or not.
 703  */
 704 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 705                                                         bool for_write)
 706 {
 707         struct address_space *mapping = inode->i_mapping;
 708         struct page *page;
 709 repeat:
 710         page = get_read_data_page(inode, index, 0, for_write);
 711         if (IS_ERR(page))
 712                 return page;
 713
 714         /* wait for read completion */
 715         lock_page(page);
 716         if (unlikely(page->mapping != mapping)) {
 717                 f2fs_put_page(page, 1);
 718                 goto repeat;
 719         }
 720         if (unlikely(!PageUptodate(page))) {
 721                 f2fs_put_page(page, 1);
 722                 return ERR_PTR(-EIO);
 723         }
 724         return page;
 725 }
 726
 727 /*
 728  * Caller ensures that this data page is never allocated.
 729  * A new zero-filled data page is allocated in the page cache.
 730  *
 731  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 732  * f2fs_unlock_op().
 733  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 734  * ipage should be released by this function.
 735  */
 736 struct page *get_new_data_page(struct inode *inode,
 737                 struct page *ipage, pgoff_t index, bool new_i_size)
 738 {
 739         struct address_space *mapping = inode->i_mapping;
 740         struct page *page;
 741         struct dnode_of_data dn;
 742         int err;
 743
 744         page = f2fs_grab_cache_page(mapping, index, true);
 745         if (!page) {
 746                 /*
 747                  * before exiting, we should make sure ipage will be released
 748                  * if any error occur.
 749                  */
 750                 f2fs_put_page(ipage, 1);
 751                 return ERR_PTR(-ENOMEM);
 752         }
 753
 754         set_new_dnode(&dn, inode, ipage, NULL, 0);
 755         err = f2fs_reserve_block(&dn, index);
 756         if (err) {
 757                 f2fs_put_page(page, 1);
 758                 return ERR_PTR(err);
 759         }
 760         if (!ipage)
 761                 f2fs_put_dnode(&dn);
 762
 763         if (PageUptodate(page))
 764                 goto got_it;
 765
 766         if (dn.data_blkaddr == NEW_ADDR) {
 767                 zero_user_segment(page, 0, PAGE_SIZE);
 768                 if (!PageUptodate(page))
 769                         SetPageUptodate(page);
 770         } else {
 771                 f2fs_put_page(page, 1);
 772
 773                 /* if ipage exists, blkaddr should be NEW_ADDR */
 774                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 775                 page = get_lock_data_page(inode, index, true);
 776                 if (IS_ERR(page))
 777                         return page;
 778         }
 779 got_it:
 780         if (new_i_size && i_size_read(inode) <
 781                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 782                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 783         return page;
 784 }
 785
 786 static int __allocate_data_block(struct dnode_of_data *dn)
 787 {
 788         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 789         struct f2fs_summary sum;
 790         struct node_info ni;
 791         pgoff_t fofs;
 792         blkcnt_t count = 1;
 793         int err;
 794
 795         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 796                 return -EPERM;
 797
 798         dn->data_blkaddr = datablock_addr(dn->inode,
 799                                 dn->node_page, dn->ofs_in_node);
 800         if (dn->data_blkaddr == NEW_ADDR)
 801                 goto alloc;
 802
 803         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 804                 return err;
 805
 806 alloc:
 807         get_node_info(sbi, dn->nid, &ni);
 808         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 809
 810         allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 811                                         &sum, CURSEG_WARM_DATA, NULL, false);
 812         set_data_blkaddr(dn);
 813
 814         /* update i_size */
 815         fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 816                                                         dn->ofs_in_node;
 817         if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 818                 f2fs_i_size_write(dn->inode,
 819                                 ((loff_t)(fofs + 1) << PAGE_SHIFT));
 820         return 0;
 821 }
 822
 823 static inline bool __force_buffered_io(struct inode *inode, int rw)
 824 {
 825         return (f2fs_encrypted_file(inode) ||
 826                         (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
 827                         F2FS_I_SB(inode)->s_ndevs);
 828 }
 829
 830 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 831 {
 832         struct inode *inode = file_inode(iocb->ki_filp);
 833         struct f2fs_map_blocks map;
 834         int err = 0;
 835
 836         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 837                 return 0;
 838
 839         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 840         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 841         if (map.m_len > map.m_lblk)
 842                 map.m_len -= map.m_lblk;
 843         else
 844                 map.m_len = 0;
 845
 846         map.m_next_pgofs = NULL;
 847
 848         if (iocb->ki_flags & IOCB_DIRECT) {
 849                 err = f2fs_convert_inline_inode(inode);
 850                 if (err)
 851                         return err;
 852                 return f2fs_map_blocks(inode, &map, 1,
 853                         __force_buffered_io(inode, WRITE) ?
 854                                 F2FS_GET_BLOCK_PRE_AIO :
 855                                 F2FS_GET_BLOCK_PRE_DIO);
 856         }
 857         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 858                 err = f2fs_convert_inline_inode(inode);
 859                 if (err)
 860                         return err;
 861         }
 862         if (!f2fs_has_inline_data(inode))
 863                 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
 864         return err;
 865 }
 866
 867 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 868 {
 869         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 870                 if (lock)
 871                         down_read(&sbi->node_change);
 872                 else
 873                         up_read(&sbi->node_change);
 874         } else {
 875                 if (lock)
 876                         f2fs_lock_op(sbi);
 877                 else
 878                         f2fs_unlock_op(sbi);
 879         }
 880 }
 881
 882 /*
 883  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 884  * f2fs_map_blocks structure.
 885  * If original data blocks are allocated, then give them to blockdev.
 886  * Otherwise,
 887  *     a. preallocate requested block addresses
 888  *     b. do not use extent cache for better performance
 889  *     c. give the block addresses to blockdev
 890  */
 891 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 892                                                 int create, int flag)
 893 {
 894         unsigned int maxblocks = map->m_len;
 895         struct dnode_of_data dn;
 896         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 897         int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 898         pgoff_t pgofs, end_offset, end;
 899         int err = 0, ofs = 1;
 900         unsigned int ofs_in_node, last_ofs_in_node;
 901         blkcnt_t prealloc;
 902         struct extent_info ei = {0,0,0};
 903         block_t blkaddr;
 904
 905         if (!maxblocks)
 906                 return 0;
 907
 908         map->m_len = 0;
 909         map->m_flags = 0;
 910
 911         /* it only supports block size == page size */
 912         pgofs = (pgoff_t)map->m_lblk;
 913         end = pgofs + maxblocks;
 914
 915         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 916                 map->m_pblk = ei.blk + pgofs - ei.fofs;
 917                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 918                 map->m_flags = F2FS_MAP_MAPPED;
 919                 goto out;
 920         }
 921
 922 next_dnode:
 923         if (create)
 924                 __do_map_lock(sbi, flag, true);
 925
 926         /* When reading holes, we need its node page */
 927         set_new_dnode(&dn, inode, NULL, NULL, 0);
 928         err = get_dnode_of_data(&dn, pgofs, mode);
 929         if (err) {
 930                 if (flag == F2FS_GET_BLOCK_BMAP)
 931                         map->m_pblk = 0;
 932                 if (err == -ENOENT) {
 933                         err = 0;
 934                         if (map->m_next_pgofs)
 935                                 *map->m_next_pgofs =
 936                                         get_next_page_offset(&dn, pgofs);
 937                 }
 938                 goto unlock_out;
 939         }
 940
 941         prealloc = 0;
 942         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
 943         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 944
 945 next_block:
 946         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
 947
 948         if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 949                 if (create) {
 950                         if (unlikely(f2fs_cp_error(sbi))) {
 951                                 err = -EIO;
 952                                 goto sync_out;
 953                         }
 954                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 955                                 if (blkaddr == NULL_ADDR) {
 956                                         prealloc++;
 957                                         last_ofs_in_node = dn.ofs_in_node;
 958                                 }
 959                         } else {
 960                                 err = __allocate_data_block(&dn);
 961                                 if (!err)
 962                                         set_inode_flag(inode, FI_APPEND_WRITE);
 963                         }
 964                         if (err)
 965                                 goto sync_out;
 966                         map->m_flags |= F2FS_MAP_NEW;
 967                         blkaddr = dn.data_blkaddr;
 968                 } else {
 969                         if (flag == F2FS_GET_BLOCK_BMAP) {
 970                                 map->m_pblk = 0;
 971                                 goto sync_out;
 972                         }
 973                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
 974                                                 blkaddr == NULL_ADDR) {
 975                                 if (map->m_next_pgofs)
 976                                         *map->m_next_pgofs = pgofs + 1;
 977                         }
 978                         if (flag != F2FS_GET_BLOCK_FIEMAP ||
 979                                                 blkaddr != NEW_ADDR)
 980                                 goto sync_out;
 981                 }
 982         }
 983
 984         if (flag == F2FS_GET_BLOCK_PRE_AIO)
 985                 goto skip;
 986
 987         if (map->m_len == 0) {
 988                 /* preallocated unwritten block should be mapped for fiemap. */
 989                 if (blkaddr == NEW_ADDR)
 990                         map->m_flags |= F2FS_MAP_UNWRITTEN;
 991                 map->m_flags |= F2FS_MAP_MAPPED;
 992
 993                 map->m_pblk = blkaddr;
 994                 map->m_len = 1;
 995         } else if ((map->m_pblk != NEW_ADDR &&
 996                         blkaddr == (map->m_pblk + ofs)) ||
 997                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
 998                         flag == F2FS_GET_BLOCK_PRE_DIO) {
 999                 ofs++;
1000                 map->m_len++;
1001         } else {
1002                 goto sync_out;
1003         }
1004
1005 skip:
1006         dn.ofs_in_node++;
1007         pgofs++;
1008
1009         /* preallocate blocks in batch for one dnode page */
1010         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1011                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1012
1013                 dn.ofs_in_node = ofs_in_node;
1014                 err = reserve_new_blocks(&dn, prealloc);
1015                 if (err)
1016                         goto sync_out;
1017
1018                 map->m_len += dn.ofs_in_node - ofs_in_node;
1019                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1020                         err = -ENOSPC;
1021                         goto sync_out;
1022                 }
1023                 dn.ofs_in_node = end_offset;
1024         }
1025
1026         if (pgofs >= end)
1027                 goto sync_out;
1028         else if (dn.ofs_in_node < end_offset)
1029                 goto next_block;
1030
1031         f2fs_put_dnode(&dn);
1032
1033         if (create) {
1034                 __do_map_lock(sbi, flag, false);
1035                 f2fs_balance_fs(sbi, dn.node_changed);
1036         }
1037         goto next_dnode;
1038
1039 sync_out:
1040         f2fs_put_dnode(&dn);
1041 unlock_out:
1042         if (create) {
1043                 __do_map_lock(sbi, flag, false);
1044                 f2fs_balance_fs(sbi, dn.node_changed);
1045         }
1046 out:
1047         trace_f2fs_map_blocks(inode, map, err);
1048         return err;
1049 }
1050
1051 static int __get_data_block(struct inode *inode, sector_t iblock,
1052                         struct buffer_head *bh, int create, int flag,
1053                         pgoff_t *next_pgofs)
1054 {
1055         struct f2fs_map_blocks map;
1056         int err;
1057
1058         map.m_lblk = iblock;
1059         map.m_len = bh->b_size >> inode->i_blkbits;
1060         map.m_next_pgofs = next_pgofs;
1061
1062         err = f2fs_map_blocks(inode, &map, create, flag);
1063         if (!err) {
1064                 map_bh(bh, inode->i_sb, map.m_pblk);
1065                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1066                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1067         }
1068         return err;
1069 }
1070
1071 static int get_data_block(struct inode *inode, sector_t iblock,
1072                         struct buffer_head *bh_result, int create, int flag,
1073                         pgoff_t *next_pgofs)
1074 {
1075         return __get_data_block(inode, iblock, bh_result, create,
1076                                                         flag, next_pgofs);
1077 }
1078
1079 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1080                         struct buffer_head *bh_result, int create)
1081 {
1082         return __get_data_block(inode, iblock, bh_result, create,
1083                                                 F2FS_GET_BLOCK_DEFAULT, NULL);
1084 }
1085
1086 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1087                         struct buffer_head *bh_result, int create)
1088 {
1089         /* Block number less than F2FS MAX BLOCKS */
1090         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1091                 return -EFBIG;
1092
1093         return __get_data_block(inode, iblock, bh_result, create,
1094                                                 F2FS_GET_BLOCK_BMAP, NULL);
1095 }
1096
1097 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1098 {
1099         return (offset >> inode->i_blkbits);
1100 }
1101
1102 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1103 {
1104         return (blk << inode->i_blkbits);
1105 }
1106
1107 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1108                 u64 start, u64 len)
1109 {
1110         struct buffer_head map_bh;
1111         sector_t start_blk, last_blk;
1112         pgoff_t next_pgofs;
1113         u64 logical = 0, phys = 0, size = 0;
1114         u32 flags = 0;
1115         int ret = 0;
1116
1117         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1118         if (ret)
1119                 return ret;
1120
1121         if (f2fs_has_inline_data(inode)) {
1122                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1123                 if (ret != -EAGAIN)
1124                         return ret;
1125         }
1126
1127         inode_lock(inode);
1128
1129         if (logical_to_blk(inode, len) == 0)
1130                 len = blk_to_logical(inode, 1);
1131
1132         start_blk = logical_to_blk(inode, start);
1133         last_blk = logical_to_blk(inode, start + len - 1);
1134
1135 next:
1136         memset(&map_bh, 0, sizeof(struct buffer_head));
1137         map_bh.b_size = len;
1138
1139         ret = get_data_block(inode, start_blk, &map_bh, 0,
1140                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1141         if (ret)
1142                 goto out;
1143
1144         /* HOLE */
1145         if (!buffer_mapped(&map_bh)) {
1146                 start_blk = next_pgofs;
1147
1148                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1149                                         F2FS_I_SB(inode)->max_file_blocks))
1150                         goto prep_next;
1151
1152                 flags |= FIEMAP_EXTENT_LAST;
1153         }
1154
1155         if (size) {
1156                 if (f2fs_encrypted_inode(inode))
1157                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1158
1159                 ret = fiemap_fill_next_extent(fieinfo, logical,
1160                                 phys, size, flags);
1161         }
1162
1163         if (start_blk > last_blk || ret)
1164                 goto out;
1165
1166         logical = blk_to_logical(inode, start_blk);
1167         phys = blk_to_logical(inode, map_bh.b_blocknr);
1168         size = map_bh.b_size;
1169         flags = 0;
1170         if (buffer_unwritten(&map_bh))
1171                 flags = FIEMAP_EXTENT_UNWRITTEN;
1172
1173         start_blk += logical_to_blk(inode, size);
1174
1175 prep_next:
1176         cond_resched();
1177         if (fatal_signal_pending(current))
1178                 ret = -EINTR;
1179         else
1180                 goto next;
1181 out:
1182         if (ret == 1)
1183                 ret = 0;
1184
1185         inode_unlock(inode);
1186         return ret;
1187 }
1188
1189 /*
1190  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1191  * Major change was from block_size == page_size in f2fs by default.
1192  */
1193 static int f2fs_mpage_readpages(struct address_space *mapping,
1194                         struct list_head *pages, struct page *page,
1195                         unsigned nr_pages)
1196 {
1197         struct bio *bio = NULL;
1198         unsigned page_idx;
1199         sector_t last_block_in_bio = 0;
1200         struct inode *inode = mapping->host;
1201         const unsigned blkbits = inode->i_blkbits;
1202         const unsigned blocksize = 1 << blkbits;
1203         sector_t block_in_file;
1204         sector_t last_block;
1205         sector_t last_block_in_file;
1206         sector_t block_nr;
1207         struct f2fs_map_blocks map;
1208
1209         map.m_pblk = 0;
1210         map.m_lblk = 0;
1211         map.m_len = 0;
1212         map.m_flags = 0;
1213         map.m_next_pgofs = NULL;
1214
1215         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1216
1217                 if (pages) {
1218                         page = list_last_entry(pages, struct page, lru);
1219
1220                         prefetchw(&page->flags);
1221                         list_del(&page->lru);
1222                         if (add_to_page_cache_lru(page, mapping,
1223                                                   page->index,
1224                                                   readahead_gfp_mask(mapping)))
1225                                 goto next_page;
1226                 }
1227
1228                 block_in_file = (sector_t)page->index;
1229                 last_block = block_in_file + nr_pages;
1230                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1231                                                                 blkbits;
1232                 if (last_block > last_block_in_file)
1233                         last_block = last_block_in_file;
1234
1235                 /*
1236                  * Map blocks using the previous result first.
1237                  */
1238                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1239                                 block_in_file > map.m_lblk &&
1240                                 block_in_file < (map.m_lblk + map.m_len))
1241                         goto got_it;
1242
1243                 /*
1244                  * Then do more f2fs_map_blocks() calls until we are
1245                  * done with this page.
1246                  */
1247                 map.m_flags = 0;
1248
1249                 if (block_in_file < last_block) {
1250                         map.m_lblk = block_in_file;
1251                         map.m_len = last_block - block_in_file;
1252
1253                         if (f2fs_map_blocks(inode, &map, 0,
1254                                                 F2FS_GET_BLOCK_DEFAULT))
1255                                 goto set_error_page;
1256                 }
1257 got_it:
1258                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1259                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1260                         SetPageMappedToDisk(page);
1261
1262                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1263                                 SetPageUptodate(page);
1264                                 goto confused;
1265                         }
1266                 } else {
1267                         zero_user_segment(page, 0, PAGE_SIZE);
1268                         if (!PageUptodate(page))
1269                                 SetPageUptodate(page);
1270                         unlock_page(page);
1271                         goto next_page;
1272                 }
1273
1274                 /*
1275                  * This page will go to BIO.  Do we need to send this
1276                  * BIO off first?
1277                  */
1278                 if (bio && (last_block_in_bio != block_nr - 1 ||
1279                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1280 submit_and_realloc:
1281                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1282                         bio = NULL;
1283                 }
1284                 if (bio == NULL) {
1285                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1286                         if (IS_ERR(bio)) {
1287                                 bio = NULL;
1288                                 goto set_error_page;
1289                         }
1290                 }
1291
1292                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1293                         goto submit_and_realloc;
1294
1295                 last_block_in_bio = block_nr;
1296                 goto next_page;
1297 set_error_page:
1298                 SetPageError(page);
1299                 zero_user_segment(page, 0, PAGE_SIZE);
1300                 unlock_page(page);
1301                 goto next_page;
1302 confused:
1303                 if (bio) {
1304                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1305                         bio = NULL;
1306                 }
1307                 unlock_page(page);
1308 next_page:
1309                 if (pages)
1310                         put_page(page);
1311         }
1312         BUG_ON(pages && !list_empty(pages));
1313         if (bio)
1314                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1315         return 0;
1316 }
1317
1318 static int f2fs_read_data_page(struct file *file, struct page *page)
1319 {
1320         struct inode *inode = page->mapping->host;
1321         int ret = -EAGAIN;
1322
1323         trace_f2fs_readpage(page, DATA);
1324
1325         /* If the file has inline data, try to read it directly */
1326         if (f2fs_has_inline_data(inode))
1327                 ret = f2fs_read_inline_data(inode, page);
1328         if (ret == -EAGAIN)
1329                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1330         return ret;
1331 }
1332
1333 static int f2fs_read_data_pages(struct file *file,
1334                         struct address_space *mapping,
1335                         struct list_head *pages, unsigned nr_pages)
1336 {
1337         struct inode *inode = file->f_mapping->host;
1338         struct page *page = list_last_entry(pages, struct page, lru);
1339
1340         trace_f2fs_readpages(inode, page, nr_pages);
1341
1342         /* If the file has inline data, skip readpages */
1343         if (f2fs_has_inline_data(inode))
1344                 return 0;
1345
1346         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1347 }
1348
1349 static int encrypt_one_page(struct f2fs_io_info *fio)
1350 {
1351         struct inode *inode = fio->page->mapping->host;
1352         gfp_t gfp_flags = GFP_NOFS;
1353
1354         if (!f2fs_encrypted_file(inode))
1355                 return 0;
1356
1357         /* wait for GCed encrypted page writeback */
1358         f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1359
1360 retry_encrypt:
1361         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1362                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1363         if (!IS_ERR(fio->encrypted_page))
1364                 return 0;
1365
1366         /* flush pending IOs and wait for a while in the ENOMEM case */
1367         if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1368                 f2fs_flush_merged_writes(fio->sbi);
1369                 congestion_wait(BLK_RW_ASYNC, HZ/50);
1370                 gfp_flags |= __GFP_NOFAIL;
1371                 goto retry_encrypt;
1372         }
1373         return PTR_ERR(fio->encrypted_page);
1374 }
1375
1376 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1377 {
1378         struct inode *inode = fio->page->mapping->host;
1379
1380         if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1381                 return false;
1382         if (is_cold_data(fio->page))
1383                 return false;
1384         if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1385                 return false;
1386
1387         return need_inplace_update_policy(inode, fio);
1388 }
1389
1390 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1391 {
1392         if (fio->old_blkaddr == NEW_ADDR)
1393                 return false;
1394         if (fio->old_blkaddr == NULL_ADDR)
1395                 return false;
1396         return true;
1397 }
1398
1399 int do_write_data_page(struct f2fs_io_info *fio)
1400 {
1401         struct page *page = fio->page;
1402         struct inode *inode = page->mapping->host;
1403         struct dnode_of_data dn;
1404         struct extent_info ei = {0,0,0};
1405         bool ipu_force = false;
1406         int err = 0;
1407
1408         set_new_dnode(&dn, inode, NULL, NULL, 0);
1409         if (need_inplace_update(fio) &&
1410                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1411                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1412
1413                 if (valid_ipu_blkaddr(fio)) {
1414                         ipu_force = true;
1415                         fio->need_lock = LOCK_DONE;
1416                         goto got_it;
1417                 }
1418         }
1419
1420         /* Deadlock due to between page->lock and f2fs_lock_op */
1421         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1422                 return -EAGAIN;
1423
1424         err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1425         if (err)
1426                 goto out;
1427
1428         fio->old_blkaddr = dn.data_blkaddr;
1429
1430         /* This page is already truncated */
1431         if (fio->old_blkaddr == NULL_ADDR) {
1432                 ClearPageUptodate(page);
1433                 goto out_writepage;
1434         }
1435 got_it:
1436         /*
1437          * If current allocation needs SSR,
1438          * it had better in-place writes for updated data.
1439          */
1440         if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1441                 err = encrypt_one_page(fio);
1442                 if (err)
1443                         goto out_writepage;
1444
1445                 set_page_writeback(page);
1446                 f2fs_put_dnode(&dn);
1447                 if (fio->need_lock == LOCK_REQ)
1448                         f2fs_unlock_op(fio->sbi);
1449                 err = rewrite_data_page(fio);
1450                 trace_f2fs_do_write_data_page(fio->page, IPU);
1451                 set_inode_flag(inode, FI_UPDATE_WRITE);
1452                 return err;
1453         }
1454
1455         if (fio->need_lock == LOCK_RETRY) {
1456                 if (!f2fs_trylock_op(fio->sbi)) {
1457                         err = -EAGAIN;
1458                         goto out_writepage;
1459                 }
1460                 fio->need_lock = LOCK_REQ;
1461         }
1462
1463         err = encrypt_one_page(fio);
1464         if (err)
1465                 goto out_writepage;
1466
1467         set_page_writeback(page);
1468
1469         /* LFS mode write path */
1470         write_data_page(&dn, fio);
1471         trace_f2fs_do_write_data_page(page, OPU);
1472         set_inode_flag(inode, FI_APPEND_WRITE);
1473         if (page->index == 0)
1474                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1475 out_writepage:
1476         f2fs_put_dnode(&dn);
1477 out:
1478         if (fio->need_lock == LOCK_REQ)
1479                 f2fs_unlock_op(fio->sbi);
1480         return err;
1481 }
1482
1483 static int __write_data_page(struct page *page, bool *submitted,
1484                                 struct writeback_control *wbc,
1485                                 enum iostat_type io_type)
1486 {
1487         struct inode *inode = page->mapping->host;
1488         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1489         loff_t i_size = i_size_read(inode);
1490         const pgoff_t end_index = ((unsigned long long) i_size)
1491                                                         >> PAGE_SHIFT;
1492         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1493         unsigned offset = 0;
1494         bool need_balance_fs = false;
1495         int err = 0;
1496         struct f2fs_io_info fio = {
1497                 .sbi = sbi,
1498                 .type = DATA,
1499                 .op = REQ_OP_WRITE,
1500                 .op_flags = wbc_to_write_flags(wbc),
1501                 .old_blkaddr = NULL_ADDR,
1502                 .page = page,
1503                 .encrypted_page = NULL,
1504                 .submitted = false,
1505                 .need_lock = LOCK_RETRY,
1506                 .io_type = io_type,
1507         };
1508
1509         trace_f2fs_writepage(page, DATA);
1510
1511         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1512                 goto redirty_out;
1513
1514         if (page->index < end_index)
1515                 goto write;
1516
1517         /*
1518          * If the offset is out-of-range of file size,
1519          * this page does not have to be written to disk.
1520          */
1521         offset = i_size & (PAGE_SIZE - 1);
1522         if ((page->index >= end_index + 1) || !offset)
1523                 goto out;
1524
1525         zero_user_segment(page, offset, PAGE_SIZE);
1526 write:
1527         if (f2fs_is_drop_cache(inode))
1528                 goto out;
1529         /* we should not write 0'th page having journal header */
1530         if (f2fs_is_volatile_file(inode) && (!page->index ||
1531                         (!wbc->for_reclaim &&
1532                         available_free_memory(sbi, BASE_CHECK))))
1533                 goto redirty_out;
1534
1535         /* we should bypass data pages to proceed the kworkder jobs */
1536         if (unlikely(f2fs_cp_error(sbi))) {
1537                 mapping_set_error(page->mapping, -EIO);
1538                 goto out;
1539         }
1540
1541         /* Dentry blocks are controlled by checkpoint */
1542         if (S_ISDIR(inode->i_mode)) {
1543                 fio.need_lock = LOCK_DONE;
1544                 err = do_write_data_page(&fio);
1545                 goto done;
1546         }
1547
1548         if (!wbc->for_reclaim)
1549                 need_balance_fs = true;
1550         else if (has_not_enough_free_secs(sbi, 0, 0))
1551                 goto redirty_out;
1552         else
1553                 set_inode_flag(inode, FI_HOT_DATA);
1554
1555         err = -EAGAIN;
1556         if (f2fs_has_inline_data(inode)) {
1557                 err = f2fs_write_inline_data(inode, page);
1558                 if (!err)
1559                         goto out;
1560         }
1561
1562         if (err == -EAGAIN) {
1563                 err = do_write_data_page(&fio);
1564                 if (err == -EAGAIN) {
1565                         fio.need_lock = LOCK_REQ;
1566                         err = do_write_data_page(&fio);
1567                 }
1568         }
1569         if (F2FS_I(inode)->last_disk_size < psize)
1570                 F2FS_I(inode)->last_disk_size = psize;
1571
1572 done:
1573         if (err && err != -ENOENT)
1574                 goto redirty_out;
1575
1576 out:
1577         inode_dec_dirty_pages(inode);
1578         if (err)
1579                 ClearPageUptodate(page);
1580
1581         if (wbc->for_reclaim) {
1582                 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1583                 clear_inode_flag(inode, FI_HOT_DATA);
1584                 remove_dirty_inode(inode);
1585                 submitted = NULL;
1586         }
1587
1588         unlock_page(page);
1589         if (!S_ISDIR(inode->i_mode))
1590                 f2fs_balance_fs(sbi, need_balance_fs);
1591
1592         if (unlikely(f2fs_cp_error(sbi))) {
1593                 f2fs_submit_merged_write(sbi, DATA);
1594                 submitted = NULL;
1595         }
1596
1597         if (submitted)
1598                 *submitted = fio.submitted;
1599
1600         return 0;
1601
1602 redirty_out:
1603         redirty_page_for_writepage(wbc, page);
1604         if (!err)
1605                 return AOP_WRITEPAGE_ACTIVATE;
1606         unlock_page(page);
1607         return err;
1608 }
1609
1610 static int f2fs_write_data_page(struct page *page,
1611                                         struct writeback_control *wbc)
1612 {
1613         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1614 }
1615
1616 /*
1617  * This function was copied from write_cche_pages from mm/page-writeback.c.
1618  * The major change is making write step of cold data page separately from
1619  * warm/hot data page.
1620  */
1621 static int f2fs_write_cache_pages(struct address_space *mapping,
1622                                         struct writeback_control *wbc,
1623                                         enum iostat_type io_type)
1624 {
1625         int ret = 0;
1626         int done = 0;
1627         struct pagevec pvec;
1628         int nr_pages;
1629         pgoff_t uninitialized_var(writeback_index);
1630         pgoff_t index;
1631         pgoff_t end;            /* Inclusive */
1632         pgoff_t done_index;
1633         pgoff_t last_idx = ULONG_MAX;
1634         int cycled;
1635         int range_whole = 0;
1636         int tag;
1637
1638         pagevec_init(&pvec, 0);
1639
1640         if (get_dirty_pages(mapping->host) <=
1641                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1642                 set_inode_flag(mapping->host, FI_HOT_DATA);
1643         else
1644                 clear_inode_flag(mapping->host, FI_HOT_DATA);
1645
1646         if (wbc->range_cyclic) {
1647                 writeback_index = mapping->writeback_index; /* prev offset */
1648                 index = writeback_index;
1649                 if (index == 0)
1650                         cycled = 1;
1651                 else
1652                         cycled = 0;
1653                 end = -1;
1654         } else {
1655                 index = wbc->range_start >> PAGE_SHIFT;
1656                 end = wbc->range_end >> PAGE_SHIFT;
1657                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1658                         range_whole = 1;
1659                 cycled = 1; /* ignore range_cyclic tests */
1660         }
1661         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1662                 tag = PAGECACHE_TAG_TOWRITE;
1663         else
1664                 tag = PAGECACHE_TAG_DIRTY;
1665 retry:
1666         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1667                 tag_pages_for_writeback(mapping, index, end);
1668         done_index = index;
1669         while (!done && (index <= end)) {
1670                 int i;
1671
1672                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
1673                                 tag);
1674                 if (nr_pages == 0)
1675                         break;
1676
1677                 for (i = 0; i < nr_pages; i++) {
1678                         struct page *page = pvec.pages[i];
1679                         bool submitted = false;
1680
1681                         done_index = page->index;
1682 retry_write:
1683                         lock_page(page);
1684
1685                         if (unlikely(page->mapping != mapping)) {
1686 continue_unlock:
1687                                 unlock_page(page);
1688                                 continue;
1689                         }
1690
1691                         if (!PageDirty(page)) {
1692                                 /* someone wrote it for us */
1693                                 goto continue_unlock;
1694                         }
1695
1696                         if (PageWriteback(page)) {
1697                                 if (wbc->sync_mode != WB_SYNC_NONE)
1698                                         f2fs_wait_on_page_writeback(page,
1699                                                                 DATA, true);
1700                                 else
1701                                         goto continue_unlock;
1702                         }
1703
1704                         BUG_ON(PageWriteback(page));
1705                         if (!clear_page_dirty_for_io(page))
1706                                 goto continue_unlock;
1707
1708                         ret = __write_data_page(page, &submitted, wbc, io_type);
1709                         if (unlikely(ret)) {
1710                                 /*
1711                                  * keep nr_to_write, since vfs uses this to
1712                                  * get # of written pages.
1713                                  */
1714                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1715                                         unlock_page(page);
1716                                         ret = 0;
1717                                         continue;
1718                                 } else if (ret == -EAGAIN) {
1719                                         ret = 0;
1720                                         if (wbc->sync_mode == WB_SYNC_ALL) {
1721                                                 cond_resched();
1722                                                 congestion_wait(BLK_RW_ASYNC,
1723                                                                         HZ/50);
1724                                                 goto retry_write;
1725                                         }
1726                                         continue;
1727                                 }
1728                                 done_index = page->index + 1;
1729                                 done = 1;
1730                                 break;
1731                         } else if (submitted) {
1732                                 last_idx = page->index;
1733                         }
1734
1735                         /* give a priority to WB_SYNC threads */
1736                         if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1737                                         --wbc->nr_to_write <= 0) &&
1738                                         wbc->sync_mode == WB_SYNC_NONE) {
1739                                 done = 1;
1740                                 break;
1741                         }
1742                 }
1743                 pagevec_release(&pvec);
1744                 cond_resched();
1745         }
1746
1747         if (!cycled && !done) {
1748                 cycled = 1;
1749                 index = 0;
1750                 end = writeback_index - 1;
1751                 goto retry;
1752         }
1753         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1754                 mapping->writeback_index = done_index;
1755
1756         if (last_idx != ULONG_MAX)
1757                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1758                                                 0, last_idx, DATA);
1759
1760         return ret;
1761 }
1762
1763 int __f2fs_write_data_pages(struct address_space *mapping,
1764                                                 struct writeback_control *wbc,
1765                                                 enum iostat_type io_type)
1766 {
1767         struct inode *inode = mapping->host;
1768         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1769         struct blk_plug plug;
1770         int ret;
1771
1772         /* deal with chardevs and other special file */
1773         if (!mapping->a_ops->writepage)
1774                 return 0;
1775
1776         /* skip writing if there is no dirty page in this inode */
1777         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1778                 return 0;
1779
1780         /* during POR, we don't need to trigger writepage at all. */
1781         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1782                 goto skip_write;
1783
1784         if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1785                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1786                         available_free_memory(sbi, DIRTY_DENTS))
1787                 goto skip_write;
1788
1789         /* skip writing during file defragment */
1790         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1791                 goto skip_write;
1792
1793         trace_f2fs_writepages(mapping->host, wbc, DATA);
1794
1795         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1796         if (wbc->sync_mode == WB_SYNC_ALL)
1797                 atomic_inc(&sbi->wb_sync_req);
1798         else if (atomic_read(&sbi->wb_sync_req))
1799                 goto skip_write;
1800
1801         blk_start_plug(&plug);
1802         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
1803         blk_finish_plug(&plug);
1804
1805         if (wbc->sync_mode == WB_SYNC_ALL)
1806                 atomic_dec(&sbi->wb_sync_req);
1807         /*
1808          * if some pages were truncated, we cannot guarantee its mapping->host
1809          * to detect pending bios.
1810          */
1811
1812         remove_dirty_inode(inode);
1813         return ret;
1814
1815 skip_write:
1816         wbc->pages_skipped += get_dirty_pages(inode);
1817         trace_f2fs_writepages(mapping->host, wbc, DATA);
1818         return 0;
1819 }
1820
1821 static int f2fs_write_data_pages(struct address_space *mapping,
1822                             struct writeback_control *wbc)
1823 {
1824         struct inode *inode = mapping->host;
1825
1826         return __f2fs_write_data_pages(mapping, wbc,
1827                         F2FS_I(inode)->cp_task == current ?
1828                         FS_CP_DATA_IO : FS_DATA_IO);
1829 }
1830
1831 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1832 {
1833         struct inode *inode = mapping->host;
1834         loff_t i_size = i_size_read(inode);
1835
1836         if (to > i_size) {
1837                 down_write(&F2FS_I(inode)->i_mmap_sem);
1838                 truncate_pagecache(inode, i_size);
1839                 truncate_blocks(inode, i_size, true);
1840                 up_write(&F2FS_I(inode)->i_mmap_sem);
1841         }
1842 }
1843
1844 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1845                         struct page *page, loff_t pos, unsigned len,
1846                         block_t *blk_addr, bool *node_changed)
1847 {
1848         struct inode *inode = page->mapping->host;
1849         pgoff_t index = page->index;
1850         struct dnode_of_data dn;
1851         struct page *ipage;
1852         bool locked = false;
1853         struct extent_info ei = {0,0,0};
1854         int err = 0;
1855
1856         /*
1857          * we already allocated all the blocks, so we don't need to get
1858          * the block addresses when there is no need to fill the page.
1859          */
1860         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1861                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
1862                 return 0;
1863
1864         if (f2fs_has_inline_data(inode) ||
1865                         (pos & PAGE_MASK) >= i_size_read(inode)) {
1866                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1867                 locked = true;
1868         }
1869 restart:
1870         /* check inline_data */
1871         ipage = get_node_page(sbi, inode->i_ino);
1872         if (IS_ERR(ipage)) {
1873                 err = PTR_ERR(ipage);
1874                 goto unlock_out;
1875         }
1876
1877         set_new_dnode(&dn, inode, ipage, ipage, 0);
1878
1879         if (f2fs_has_inline_data(inode)) {
1880                 if (pos + len <= MAX_INLINE_DATA(inode)) {
1881                         read_inline_data(page, ipage);
1882                         set_inode_flag(inode, FI_DATA_EXIST);
1883                         if (inode->i_nlink)
1884                                 set_inline_node(ipage);
1885                 } else {
1886                         err = f2fs_convert_inline_page(&dn, page);
1887                         if (err)
1888                                 goto out;
1889                         if (dn.data_blkaddr == NULL_ADDR)
1890                                 err = f2fs_get_block(&dn, index);
1891                 }
1892         } else if (locked) {
1893                 err = f2fs_get_block(&dn, index);
1894         } else {
1895                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1896                         dn.data_blkaddr = ei.blk + index - ei.fofs;
1897                 } else {
1898                         /* hole case */
1899                         err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1900                         if (err || dn.data_blkaddr == NULL_ADDR) {
1901                                 f2fs_put_dnode(&dn);
1902                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1903                                                                 true);
1904                                 locked = true;
1905                                 goto restart;
1906                         }
1907                 }
1908         }
1909
1910         /* convert_inline_page can make node_changed */
1911         *blk_addr = dn.data_blkaddr;
1912         *node_changed = dn.node_changed;
1913 out:
1914         f2fs_put_dnode(&dn);
1915 unlock_out:
1916         if (locked)
1917                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1918         return err;
1919 }
1920
1921 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1922                 loff_t pos, unsigned len, unsigned flags,
1923                 struct page **pagep, void **fsdata)
1924 {
1925         struct inode *inode = mapping->host;
1926         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1927         struct page *page = NULL;
1928         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1929         bool need_balance = false;
1930         block_t blkaddr = NULL_ADDR;
1931         int err = 0;
1932
1933         trace_f2fs_write_begin(inode, pos, len, flags);
1934
1935         /*
1936          * We should check this at this moment to avoid deadlock on inode page
1937          * and #0 page. The locking rule for inline_data conversion should be:
1938          * lock_page(page #0) -> lock_page(inode_page)
1939          */
1940         if (index != 0) {
1941                 err = f2fs_convert_inline_inode(inode);
1942                 if (err)
1943                         goto fail;
1944         }
1945 repeat:
1946         /*
1947          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1948          * wait_for_stable_page. Will wait that below with our IO control.
1949          */
1950         page = pagecache_get_page(mapping, index,
1951                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1952         if (!page) {
1953                 err = -ENOMEM;
1954                 goto fail;
1955         }
1956
1957         *pagep = page;
1958
1959         err = prepare_write_begin(sbi, page, pos, len,
1960                                         &blkaddr, &need_balance);
1961         if (err)
1962                 goto fail;
1963
1964         if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1965                 unlock_page(page);
1966                 f2fs_balance_fs(sbi, true);
1967                 lock_page(page);
1968                 if (page->mapping != mapping) {
1969                         /* The page got truncated from under us */
1970                         f2fs_put_page(page, 1);
1971                         goto repeat;
1972                 }
1973         }
1974
1975         f2fs_wait_on_page_writeback(page, DATA, false);
1976
1977         /* wait for GCed encrypted page writeback */
1978         if (f2fs_encrypted_file(inode))
1979                 f2fs_wait_on_block_writeback(sbi, blkaddr);
1980
1981         if (len == PAGE_SIZE || PageUptodate(page))
1982                 return 0;
1983
1984         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1985                 zero_user_segment(page, len, PAGE_SIZE);
1986                 return 0;
1987         }
1988
1989         if (blkaddr == NEW_ADDR) {
1990                 zero_user_segment(page, 0, PAGE_SIZE);
1991                 SetPageUptodate(page);
1992         } else {
1993                 err = f2fs_submit_page_read(inode, page, blkaddr);
1994                 if (err)
1995                         goto fail;
1996
1997                 lock_page(page);
1998                 if (unlikely(page->mapping != mapping)) {
1999                         f2fs_put_page(page, 1);
2000                         goto repeat;
2001                 }
2002                 if (unlikely(!PageUptodate(page))) {
2003                         err = -EIO;
2004                         goto fail;
2005                 }
2006         }
2007         return 0;
2008
2009 fail:
2010         f2fs_put_page(page, 1);
2011         f2fs_write_failed(mapping, pos + len);
2012         return err;
2013 }
2014
2015 static int f2fs_write_end(struct file *file,
2016                         struct address_space *mapping,
2017                         loff_t pos, unsigned len, unsigned copied,
2018                         struct page *page, void *fsdata)
2019 {
2020         struct inode *inode = page->mapping->host;
2021
2022         trace_f2fs_write_end(inode, pos, len, copied);
2023
2024         /*
2025          * This should be come from len == PAGE_SIZE, and we expect copied
2026          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2027          * let generic_perform_write() try to copy data again through copied=0.
2028          */
2029         if (!PageUptodate(page)) {
2030                 if (unlikely(copied != len))
2031                         copied = 0;
2032                 else
2033                         SetPageUptodate(page);
2034         }
2035         if (!copied)
2036                 goto unlock_out;
2037
2038         set_page_dirty(page);
2039
2040         if (pos + copied > i_size_read(inode))
2041                 f2fs_i_size_write(inode, pos + copied);
2042 unlock_out:
2043         f2fs_put_page(page, 1);
2044         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2045         return copied;
2046 }
2047
2048 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2049                            loff_t offset)
2050 {
2051         unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2052
2053         if (offset & blocksize_mask)
2054                 return -EINVAL;
2055
2056         if (iov_iter_alignment(iter) & blocksize_mask)
2057                 return -EINVAL;
2058
2059         return 0;
2060 }
2061
2062 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2063 {
2064         struct address_space *mapping = iocb->ki_filp->f_mapping;
2065         struct inode *inode = mapping->host;
2066         size_t count = iov_iter_count(iter);
2067         loff_t offset = iocb->ki_pos;
2068         int rw = iov_iter_rw(iter);
2069         int err;
2070
2071         err = check_direct_IO(inode, iter, offset);
2072         if (err)
2073                 return err;
2074
2075         if (__force_buffered_io(inode, rw))
2076                 return 0;
2077
2078         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2079
2080         down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2081         err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2082         up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2083
2084         if (rw == WRITE) {
2085                 if (err > 0) {
2086                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2087                                                                         err);
2088                         set_inode_flag(inode, FI_UPDATE_WRITE);
2089                 } else if (err < 0) {
2090                         f2fs_write_failed(mapping, offset + count);
2091                 }
2092         }
2093
2094         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2095
2096         return err;
2097 }
2098
2099 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2100                                                         unsigned int length)
2101 {
2102         struct inode *inode = page->mapping->host;
2103         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2104
2105         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2106                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2107                 return;
2108
2109         if (PageDirty(page)) {
2110                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2111                         dec_page_count(sbi, F2FS_DIRTY_META);
2112                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2113                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2114                 } else {
2115                         inode_dec_dirty_pages(inode);
2116                         remove_dirty_inode(inode);
2117                 }
2118         }
2119
2120         /* This is atomic written page, keep Private */
2121         if (IS_ATOMIC_WRITTEN_PAGE(page))
2122                 return drop_inmem_page(inode, page);
2123
2124         set_page_private(page, 0);
2125         ClearPagePrivate(page);
2126 }
2127
2128 int f2fs_release_page(struct page *page, gfp_t wait)
2129 {
2130         /* If this is dirty page, keep PagePrivate */
2131         if (PageDirty(page))
2132                 return 0;
2133
2134         /* This is atomic written page, keep Private */
2135         if (IS_ATOMIC_WRITTEN_PAGE(page))
2136                 return 0;
2137
2138         set_page_private(page, 0);
2139         ClearPagePrivate(page);
2140         return 1;
2141 }
2142
2143 /*
2144  * This was copied from __set_page_dirty_buffers which gives higher performance
2145  * in very high speed storages. (e.g., pmem)
2146  */
2147 void f2fs_set_page_dirty_nobuffers(struct page *page)
2148 {
2149         struct address_space *mapping = page->mapping;
2150         unsigned long flags;
2151
2152         if (unlikely(!mapping))
2153                 return;
2154
2155         spin_lock(&mapping->private_lock);
2156         lock_page_memcg(page);
2157         SetPageDirty(page);
2158         spin_unlock(&mapping->private_lock);
2159
2160         spin_lock_irqsave(&mapping->tree_lock, flags);
2161         WARN_ON_ONCE(!PageUptodate(page));
2162         account_page_dirtied(page, mapping);
2163         radix_tree_tag_set(&mapping->page_tree,
2164                         page_index(page), PAGECACHE_TAG_DIRTY);
2165         spin_unlock_irqrestore(&mapping->tree_lock, flags);
2166         unlock_page_memcg(page);
2167
2168         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2169         return;
2170 }
2171
2172 static int f2fs_set_data_page_dirty(struct page *page)
2173 {
2174         struct address_space *mapping = page->mapping;
2175         struct inode *inode = mapping->host;
2176
2177         trace_f2fs_set_page_dirty(page, DATA);
2178
2179         if (!PageUptodate(page))
2180                 SetPageUptodate(page);
2181
2182         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2183                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2184                         register_inmem_page(inode, page);
2185                         return 1;
2186                 }
2187                 /*
2188                  * Previously, this page has been registered, we just
2189                  * return here.
2190                  */
2191                 return 0;
2192         }
2193
2194         if (!PageDirty(page)) {
2195                 f2fs_set_page_dirty_nobuffers(page);
2196                 update_dirty_page(inode, page);
2197                 return 1;
2198         }
2199         return 0;
2200 }
2201
2202 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2203 {
2204         struct inode *inode = mapping->host;
2205
2206         if (f2fs_has_inline_data(inode))
2207                 return 0;
2208
2209         /* make sure allocating whole blocks */
2210         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2211                 filemap_write_and_wait(mapping);
2212
2213         return generic_block_bmap(mapping, block, get_data_block_bmap);
2214 }
2215
2216 #ifdef CONFIG_MIGRATION
2217 #include <linux/migrate.h>
2218
2219 int f2fs_migrate_page(struct address_space *mapping,
2220                 struct page *newpage, struct page *page, enum migrate_mode mode)
2221 {
2222         int rc, extra_count;
2223         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2224         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2225
2226         BUG_ON(PageWriteback(page));
2227
2228         /* migrating an atomic written page is safe with the inmem_lock hold */
2229         if (atomic_written) {
2230                 if (mode != MIGRATE_SYNC)
2231                         return -EBUSY;
2232                 if (!mutex_trylock(&fi->inmem_lock))
2233                         return -EAGAIN;
2234         }
2235
2236         /*
2237          * A reference is expected if PagePrivate set when move mapping,
2238          * however F2FS breaks this for maintaining dirty page counts when
2239          * truncating pages. So here adjusting the 'extra_count' make it work.
2240          */
2241         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2242         rc = migrate_page_move_mapping(mapping, newpage,
2243                                 page, NULL, mode, extra_count);
2244         if (rc != MIGRATEPAGE_SUCCESS) {
2245                 if (atomic_written)
2246                         mutex_unlock(&fi->inmem_lock);
2247                 return rc;
2248         }
2249
2250         if (atomic_written) {
2251                 struct inmem_pages *cur;
2252                 list_for_each_entry(cur, &fi->inmem_pages, list)
2253                         if (cur->page == page) {
2254                                 cur->page = newpage;
2255                                 break;
2256                         }
2257                 mutex_unlock(&fi->inmem_lock);
2258                 put_page(page);
2259                 get_page(newpage);
2260         }
2261
2262         if (PagePrivate(page))
2263                 SetPagePrivate(newpage);
2264         set_page_private(newpage, page_private(page));
2265
2266         if (mode != MIGRATE_SYNC_NO_COPY)
2267                 migrate_page_copy(newpage, page);
2268         else
2269                 migrate_page_states(newpage, page);
2270
2271         return MIGRATEPAGE_SUCCESS;
2272 }
2273 #endif
2274
2275 const struct address_space_operations f2fs_dblock_aops = {
2276         .readpage       = f2fs_read_data_page,
2277         .readpages      = f2fs_read_data_pages,
2278         .writepage      = f2fs_write_data_page,
2279         .writepages     = f2fs_write_data_pages,
2280         .write_begin    = f2fs_write_begin,
2281         .write_end      = f2fs_write_end,
2282         .set_page_dirty = f2fs_set_data_page_dirty,
2283         .invalidatepage = f2fs_invalidate_page,
2284         .releasepage    = f2fs_release_page,
2285         .direct_IO      = f2fs_direct_IO,
2286         .bmap           = f2fs_bmap,
2287 #ifdef CONFIG_MIGRATION
2288         .migratepage    = f2fs_migrate_page,
2289 #endif
2290 };