fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/prefetch.h>
  18 #include <linux/uio.h>
  19 #include <linux/cleancache.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include "trace.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define NUM_PREALLOC_POST_READ_CTXS     128
  29
  30 static struct kmem_cache *bio_post_read_ctx_cache;
  31 static mempool_t *bio_post_read_ctx_pool;
  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                         (S_ISREG(inode->i_mode) &&
  49                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  50                         is_cold_data(page))
  51                 return true;
  52         return false;
  53 }
  54
  55 static enum count_type __read_io_type(struct page *page)
  56 {
  57         struct address_space *mapping = page->mapping;
  58
  59         if (mapping) {
  60                 struct inode *inode = mapping->host;
  61                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  62
  63                 if (inode->i_ino == F2FS_META_INO(sbi))
  64                         return F2FS_RD_META;
  65
  66                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  67                         return F2FS_RD_NODE;
  68         }
  69         return F2FS_RD_DATA;
  70 }
  71
  72 /* postprocessing steps for read bios */
  73 enum bio_post_read_step {
  74         STEP_INITIAL = 0,
  75         STEP_DECRYPT,
  76 };
  77
  78 struct bio_post_read_ctx {
  79         struct bio *bio;
  80         struct work_struct work;
  81         unsigned int cur_step;
  82         unsigned int enabled_steps;
  83 };
  84
  85 static void __read_end_io(struct bio *bio)
  86 {
  87         struct page *page;
  88         struct bio_vec *bv;
  89         struct bvec_iter_all iter_all;
  90
  91         bio_for_each_segment_all(bv, bio, iter_all) {
  92                 page = bv->bv_page;
  93
  94                 /* PG_error was set if any post_read step failed */
  95                 if (bio->bi_status || PageError(page)) {
  96                         ClearPageUptodate(page);
  97                         /* will re-read again later */
  98                         ClearPageError(page);
  99                 } else {
 100                         SetPageUptodate(page);
 101                 }
 102                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 103                 unlock_page(page);
 104         }
 105         if (bio->bi_private)
 106                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 107         bio_put(bio);
 108 }
 109
 110 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 111
 112 static void decrypt_work(struct work_struct *work)
 113 {
 114         struct bio_post_read_ctx *ctx =
 115                 container_of(work, struct bio_post_read_ctx, work);
 116
 117         fscrypt_decrypt_bio(ctx->bio);
 118
 119         bio_post_read_processing(ctx);
 120 }
 121
 122 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 123 {
 124         switch (++ctx->cur_step) {
 125         case STEP_DECRYPT:
 126                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 127                         INIT_WORK(&ctx->work, decrypt_work);
 128                         fscrypt_enqueue_decrypt_work(&ctx->work);
 129                         return;
 130                 }
 131                 ctx->cur_step++;
 132                 /* fall-through */
 133         default:
 134                 __read_end_io(ctx->bio);
 135         }
 136 }
 137
 138 static bool f2fs_bio_post_read_required(struct bio *bio)
 139 {
 140         return bio->bi_private && !bio->bi_status;
 141 }
 142
 143 static void f2fs_read_end_io(struct bio *bio)
 144 {
 145         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
 146                                                 FAULT_READ_IO)) {
 147                 f2fs_show_injection_info(FAULT_READ_IO);
 148                 bio->bi_status = BLK_STS_IOERR;
 149         }
 150
 151         if (f2fs_bio_post_read_required(bio)) {
 152                 struct bio_post_read_ctx *ctx = bio->bi_private;
 153
 154                 ctx->cur_step = STEP_INITIAL;
 155                 bio_post_read_processing(ctx);
 156                 return;
 157         }
 158
 159         __read_end_io(bio);
 160 }
 161
 162 static void f2fs_write_end_io(struct bio *bio)
 163 {
 164         struct f2fs_sb_info *sbi = bio->bi_private;
 165         struct bio_vec *bvec;
 166         struct bvec_iter_all iter_all;
 167
 168         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 169                 f2fs_show_injection_info(FAULT_WRITE_IO);
 170                 bio->bi_status = BLK_STS_IOERR;
 171         }
 172
 173         bio_for_each_segment_all(bvec, bio, iter_all) {
 174                 struct page *page = bvec->bv_page;
 175                 enum count_type type = WB_DATA_TYPE(page);
 176
 177                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 178                         set_page_private(page, (unsigned long)NULL);
 179                         ClearPagePrivate(page);
 180                         unlock_page(page);
 181                         mempool_free(page, sbi->write_io_dummy);
 182
 183                         if (unlikely(bio->bi_status))
 184                                 f2fs_stop_checkpoint(sbi, true);
 185                         continue;
 186                 }
 187
 188                 fscrypt_pullback_bio_page(&page, true);
 189
 190                 if (unlikely(bio->bi_status)) {
 191                         mapping_set_error(page->mapping, -EIO);
 192                         if (type == F2FS_WB_CP_DATA)
 193                                 f2fs_stop_checkpoint(sbi, true);
 194                 }
 195
 196                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 197                                         page->index != nid_of_node(page));
 198
 199                 dec_page_count(sbi, type);
 200                 if (f2fs_in_warm_node_list(sbi, page))
 201                         f2fs_del_fsync_node_entry(sbi, page);
 202                 clear_cold_data(page);
 203                 end_page_writeback(page);
 204         }
 205         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 206                                 wq_has_sleeper(&sbi->cp_wait))
 207                 wake_up(&sbi->cp_wait);
 208
 209         bio_put(bio);
 210 }
 211
 212 /*
 213  * Return true, if pre_bio's bdev is same as its target device.
 214  */
 215 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 216                                 block_t blk_addr, struct bio *bio)
 217 {
 218         struct block_device *bdev = sbi->sb->s_bdev;
 219         int i;
 220
 221         if (f2fs_is_multi_device(sbi)) {
 222                 for (i = 0; i < sbi->s_ndevs; i++) {
 223                         if (FDEV(i).start_blk <= blk_addr &&
 224                             FDEV(i).end_blk >= blk_addr) {
 225                                 blk_addr -= FDEV(i).start_blk;
 226                                 bdev = FDEV(i).bdev;
 227                                 break;
 228                         }
 229                 }
 230         }
 231         if (bio) {
 232                 bio_set_dev(bio, bdev);
 233                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 234         }
 235         return bdev;
 236 }
 237
 238 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 239 {
 240         int i;
 241
 242         if (!f2fs_is_multi_device(sbi))
 243                 return 0;
 244
 245         for (i = 0; i < sbi->s_ndevs; i++)
 246                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 247                         return i;
 248         return 0;
 249 }
 250
 251 static bool __same_bdev(struct f2fs_sb_info *sbi,
 252                                 block_t blk_addr, struct bio *bio)
 253 {
 254         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 255         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 256 }
 257
 258 /*
 259  * Low-level block read/write IO operations.
 260  */
 261 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 262                                 struct writeback_control *wbc,
 263                                 int npages, bool is_read,
 264                                 enum page_type type, enum temp_type temp)
 265 {
 266         struct bio *bio;
 267
 268         bio = f2fs_bio_alloc(sbi, npages, true);
 269
 270         f2fs_target_device(sbi, blk_addr, bio);
 271         if (is_read) {
 272                 bio->bi_end_io = f2fs_read_end_io;
 273                 bio->bi_private = NULL;
 274         } else {
 275                 bio->bi_end_io = f2fs_write_end_io;
 276                 bio->bi_private = sbi;
 277                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 278         }
 279         if (wbc)
 280                 wbc_init_bio(wbc, bio);
 281
 282         return bio;
 283 }
 284
 285 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 286                                 struct bio *bio, enum page_type type)
 287 {
 288         if (!is_read_io(bio_op(bio))) {
 289                 unsigned int start;
 290
 291                 if (type != DATA && type != NODE)
 292                         goto submit_io;
 293
 294                 if (test_opt(sbi, LFS) && current->plug)
 295                         blk_finish_plug(current->plug);
 296
 297                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 298                 start %= F2FS_IO_SIZE(sbi);
 299
 300                 if (start == 0)
 301                         goto submit_io;
 302
 303                 /* fill dummy pages */
 304                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 305                         struct page *page =
 306                                 mempool_alloc(sbi->write_io_dummy,
 307                                               GFP_NOIO | __GFP_NOFAIL);
 308                         f2fs_bug_on(sbi, !page);
 309
 310                         zero_user_segment(page, 0, PAGE_SIZE);
 311                         SetPagePrivate(page);
 312                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 313                         lock_page(page);
 314                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 315                                 f2fs_bug_on(sbi, 1);
 316                 }
 317                 /*
 318                  * In the NODE case, we lose next block address chain. So, we
 319                  * need to do checkpoint in f2fs_sync_file.
 320                  */
 321                 if (type == NODE)
 322                         set_sbi_flag(sbi, SBI_NEED_CP);
 323         }
 324 submit_io:
 325         if (is_read_io(bio_op(bio)))
 326                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 327         else
 328                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 329         submit_bio(bio);
 330 }
 331
 332 static void __submit_merged_bio(struct f2fs_bio_info *io)
 333 {
 334         struct f2fs_io_info *fio = &io->fio;
 335
 336         if (!io->bio)
 337                 return;
 338
 339         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 340
 341         if (is_read_io(fio->op))
 342                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 343         else
 344                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 345
 346         __submit_bio(io->sbi, io->bio, fio->type);
 347         io->bio = NULL;
 348 }
 349
 350 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 351                                                 struct page *page, nid_t ino)
 352 {
 353         struct bio_vec *bvec;
 354         struct page *target;
 355         struct bvec_iter_all iter_all;
 356
 357         if (!io->bio)
 358                 return false;
 359
 360         if (!inode && !page && !ino)
 361                 return true;
 362
 363         bio_for_each_segment_all(bvec, io->bio, iter_all) {
 364
 365                 if (bvec->bv_page->mapping)
 366                         target = bvec->bv_page;
 367                 else
 368                         target = fscrypt_control_page(bvec->bv_page);
 369
 370                 if (inode && inode == target->mapping->host)
 371                         return true;
 372                 if (page && page == target)
 373                         return true;
 374                 if (ino && ino == ino_of_node(target))
 375                         return true;
 376         }
 377
 378         return false;
 379 }
 380
 381 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 382                                 enum page_type type, enum temp_type temp)
 383 {
 384         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 385         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 386
 387         down_write(&io->io_rwsem);
 388
 389         /* change META to META_FLUSH in the checkpoint procedure */
 390         if (type >= META_FLUSH) {
 391                 io->fio.type = META_FLUSH;
 392                 io->fio.op = REQ_OP_WRITE;
 393                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 394                 if (!test_opt(sbi, NOBARRIER))
 395                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 396         }
 397         __submit_merged_bio(io);
 398         up_write(&io->io_rwsem);
 399 }
 400
 401 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 402                                 struct inode *inode, struct page *page,
 403                                 nid_t ino, enum page_type type, bool force)
 404 {
 405         enum temp_type temp;
 406         bool ret = true;
 407
 408         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 409                 if (!force)     {
 410                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 411                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 412
 413                         down_read(&io->io_rwsem);
 414                         ret = __has_merged_page(io, inode, page, ino);
 415                         up_read(&io->io_rwsem);
 416                 }
 417                 if (ret)
 418                         __f2fs_submit_merged_write(sbi, type, temp);
 419
 420                 /* TODO: use HOT temp only for meta pages now. */
 421                 if (type >= META)
 422                         break;
 423         }
 424 }
 425
 426 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 427 {
 428         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 429 }
 430
 431 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 432                                 struct inode *inode, struct page *page,
 433                                 nid_t ino, enum page_type type)
 434 {
 435         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 436 }
 437
 438 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 439 {
 440         f2fs_submit_merged_write(sbi, DATA);
 441         f2fs_submit_merged_write(sbi, NODE);
 442         f2fs_submit_merged_write(sbi, META);
 443 }
 444
 445 /*
 446  * Fill the locked page with data located in the block address.
 447  * A caller needs to unlock the page on failure.
 448  */
 449 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 450 {
 451         struct bio *bio;
 452         struct page *page = fio->encrypted_page ?
 453                         fio->encrypted_page : fio->page;
 454
 455         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 456                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 457                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 458                 return -EFAULT;
 459
 460         trace_f2fs_submit_page_bio(page, fio);
 461         f2fs_trace_ios(fio, 0);
 462
 463         /* Allocate a new bio */
 464         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 465                                 1, is_read_io(fio->op), fio->type, fio->temp);
 466
 467         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 468                 bio_put(bio);
 469                 return -EFAULT;
 470         }
 471
 472         if (fio->io_wbc && !is_read_io(fio->op))
 473                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 474
 475         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 476
 477         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 478                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 479
 480         __submit_bio(fio->sbi, bio, fio->type);
 481         return 0;
 482 }
 483
 484 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 485 {
 486         struct f2fs_sb_info *sbi = fio->sbi;
 487         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 488         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 489         struct page *bio_page;
 490
 491         f2fs_bug_on(sbi, is_read_io(fio->op));
 492
 493         down_write(&io->io_rwsem);
 494 next:
 495         if (fio->in_list) {
 496                 spin_lock(&io->io_lock);
 497                 if (list_empty(&io->io_list)) {
 498                         spin_unlock(&io->io_lock);
 499                         goto out;
 500                 }
 501                 fio = list_first_entry(&io->io_list,
 502                                                 struct f2fs_io_info, list);
 503                 list_del(&fio->list);
 504                 spin_unlock(&io->io_lock);
 505         }
 506
 507         verify_fio_blkaddr(fio);
 508
 509         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 510
 511         /* set submitted = true as a return value */
 512         fio->submitted = true;
 513
 514         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 515
 516         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 517             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 518                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 519                 __submit_merged_bio(io);
 520 alloc_new:
 521         if (io->bio == NULL) {
 522                 if ((fio->type == DATA || fio->type == NODE) &&
 523                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 524                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 525                         fio->retry = true;
 526                         goto skip;
 527                 }
 528                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 529                                                 BIO_MAX_PAGES, false,
 530                                                 fio->type, fio->temp);
 531                 io->fio = *fio;
 532         }
 533
 534         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 535                 __submit_merged_bio(io);
 536                 goto alloc_new;
 537         }
 538
 539         if (fio->io_wbc)
 540                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 541
 542         io->last_block_in_bio = fio->new_blkaddr;
 543         f2fs_trace_ios(fio, 0);
 544
 545         trace_f2fs_submit_page_write(fio->page, fio);
 546 skip:
 547         if (fio->in_list)
 548                 goto next;
 549 out:
 550         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 551                                 f2fs_is_checkpoint_ready(sbi))
 552                 __submit_merged_bio(io);
 553         up_write(&io->io_rwsem);
 554 }
 555
 556 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 557                                         unsigned nr_pages, unsigned op_flag)
 558 {
 559         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 560         struct bio *bio;
 561         struct bio_post_read_ctx *ctx;
 562         unsigned int post_read_steps = 0;
 563
 564         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 565         if (!bio)
 566                 return ERR_PTR(-ENOMEM);
 567         f2fs_target_device(sbi, blkaddr, bio);
 568         bio->bi_end_io = f2fs_read_end_io;
 569         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 570
 571         if (f2fs_encrypted_file(inode))
 572                 post_read_steps |= 1 << STEP_DECRYPT;
 573         if (post_read_steps) {
 574                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 575                 if (!ctx) {
 576                         bio_put(bio);
 577                         return ERR_PTR(-ENOMEM);
 578                 }
 579                 ctx->bio = bio;
 580                 ctx->enabled_steps = post_read_steps;
 581                 bio->bi_private = ctx;
 582         }
 583
 584         return bio;
 585 }
 586
 587 /* This can handle encryption stuffs */
 588 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 589                                                         block_t blkaddr)
 590 {
 591         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 592         struct bio *bio;
 593
 594         bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 595         if (IS_ERR(bio))
 596                 return PTR_ERR(bio);
 597
 598         /* wait for GCed page writeback via META_MAPPING */
 599         f2fs_wait_on_block_writeback(inode, blkaddr);
 600
 601         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 602                 bio_put(bio);
 603                 return -EFAULT;
 604         }
 605         ClearPageError(page);
 606         inc_page_count(sbi, F2FS_RD_DATA);
 607         __submit_bio(sbi, bio, DATA);
 608         return 0;
 609 }
 610
 611 static void __set_data_blkaddr(struct dnode_of_data *dn)
 612 {
 613         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 614         __le32 *addr_array;
 615         int base = 0;
 616
 617         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 618                 base = get_extra_isize(dn->inode);
 619
 620         /* Get physical address of data block */
 621         addr_array = blkaddr_in_node(rn);
 622         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 623 }
 624
 625 /*
 626  * Lock ordering for the change of data block address:
 627  * ->data_page
 628  *  ->node_page
 629  *    update block addresses in the node page
 630  */
 631 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 632 {
 633         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 634         __set_data_blkaddr(dn);
 635         if (set_page_dirty(dn->node_page))
 636                 dn->node_changed = true;
 637 }
 638
 639 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 640 {
 641         dn->data_blkaddr = blkaddr;
 642         f2fs_set_data_blkaddr(dn);
 643         f2fs_update_extent_cache(dn);
 644 }
 645
 646 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 647 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 648 {
 649         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 650         int err;
 651
 652         if (!count)
 653                 return 0;
 654
 655         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 656                 return -EPERM;
 657         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 658                 return err;
 659
 660         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 661                                                 dn->ofs_in_node, count);
 662
 663         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 664
 665         for (; count > 0; dn->ofs_in_node++) {
 666                 block_t blkaddr = datablock_addr(dn->inode,
 667                                         dn->node_page, dn->ofs_in_node);
 668                 if (blkaddr == NULL_ADDR) {
 669                         dn->data_blkaddr = NEW_ADDR;
 670                         __set_data_blkaddr(dn);
 671                         count--;
 672                 }
 673         }
 674
 675         if (set_page_dirty(dn->node_page))
 676                 dn->node_changed = true;
 677         return 0;
 678 }
 679
 680 /* Should keep dn->ofs_in_node unchanged */
 681 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 682 {
 683         unsigned int ofs_in_node = dn->ofs_in_node;
 684         int ret;
 685
 686         ret = f2fs_reserve_new_blocks(dn, 1);
 687         dn->ofs_in_node = ofs_in_node;
 688         return ret;
 689 }
 690
 691 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 692 {
 693         bool need_put = dn->inode_page ? false : true;
 694         int err;
 695
 696         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 697         if (err)
 698                 return err;
 699
 700         if (dn->data_blkaddr == NULL_ADDR)
 701                 err = f2fs_reserve_new_block(dn);
 702         if (err || need_put)
 703                 f2fs_put_dnode(dn);
 704         return err;
 705 }
 706
 707 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 708 {
 709         struct extent_info ei  = {0,0,0};
 710         struct inode *inode = dn->inode;
 711
 712         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 713                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 714                 return 0;
 715         }
 716
 717         return f2fs_reserve_block(dn, index);
 718 }
 719
 720 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 721                                                 int op_flags, bool for_write)
 722 {
 723         struct address_space *mapping = inode->i_mapping;
 724         struct dnode_of_data dn;
 725         struct page *page;
 726         struct extent_info ei = {0,0,0};
 727         int err;
 728
 729         page = f2fs_grab_cache_page(mapping, index, for_write);
 730         if (!page)
 731                 return ERR_PTR(-ENOMEM);
 732
 733         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 734                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 735                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
 736                                                 DATA_GENERIC_ENHANCE_READ)) {
 737                         err = -EFAULT;
 738                         goto put_err;
 739                 }
 740                 goto got_it;
 741         }
 742
 743         set_new_dnode(&dn, inode, NULL, NULL, 0);
 744         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 745         if (err)
 746                 goto put_err;
 747         f2fs_put_dnode(&dn);
 748
 749         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 750                 err = -ENOENT;
 751                 goto put_err;
 752         }
 753         if (dn.data_blkaddr != NEW_ADDR &&
 754                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
 755                                                 dn.data_blkaddr,
 756                                                 DATA_GENERIC_ENHANCE)) {
 757                 err = -EFAULT;
 758                 goto put_err;
 759         }
 760 got_it:
 761         if (PageUptodate(page)) {
 762                 unlock_page(page);
 763                 return page;
 764         }
 765
 766         /*
 767          * A new dentry page is allocated but not able to be written, since its
 768          * new inode page couldn't be allocated due to -ENOSPC.
 769          * In such the case, its blkaddr can be remained as NEW_ADDR.
 770          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 771          * f2fs_init_inode_metadata.
 772          */
 773         if (dn.data_blkaddr == NEW_ADDR) {
 774                 zero_user_segment(page, 0, PAGE_SIZE);
 775                 if (!PageUptodate(page))
 776                         SetPageUptodate(page);
 777                 unlock_page(page);
 778                 return page;
 779         }
 780
 781         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 782         if (err)
 783                 goto put_err;
 784         return page;
 785
 786 put_err:
 787         f2fs_put_page(page, 1);
 788         return ERR_PTR(err);
 789 }
 790
 791 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 792 {
 793         struct address_space *mapping = inode->i_mapping;
 794         struct page *page;
 795
 796         page = find_get_page(mapping, index);
 797         if (page && PageUptodate(page))
 798                 return page;
 799         f2fs_put_page(page, 0);
 800
 801         page = f2fs_get_read_data_page(inode, index, 0, false);
 802         if (IS_ERR(page))
 803                 return page;
 804
 805         if (PageUptodate(page))
 806                 return page;
 807
 808         wait_on_page_locked(page);
 809         if (unlikely(!PageUptodate(page))) {
 810                 f2fs_put_page(page, 0);
 811                 return ERR_PTR(-EIO);
 812         }
 813         return page;
 814 }
 815
 816 /*
 817  * If it tries to access a hole, return an error.
 818  * Because, the callers, functions in dir.c and GC, should be able to know
 819  * whether this page exists or not.
 820  */
 821 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 822                                                         bool for_write)
 823 {
 824         struct address_space *mapping = inode->i_mapping;
 825         struct page *page;
 826 repeat:
 827         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 828         if (IS_ERR(page))
 829                 return page;
 830
 831         /* wait for read completion */
 832         lock_page(page);
 833         if (unlikely(page->mapping != mapping)) {
 834                 f2fs_put_page(page, 1);
 835                 goto repeat;
 836         }
 837         if (unlikely(!PageUptodate(page))) {
 838                 f2fs_put_page(page, 1);
 839                 return ERR_PTR(-EIO);
 840         }
 841         return page;
 842 }
 843
 844 /*
 845  * Caller ensures that this data page is never allocated.
 846  * A new zero-filled data page is allocated in the page cache.
 847  *
 848  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 849  * f2fs_unlock_op().
 850  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 851  * ipage should be released by this function.
 852  */
 853 struct page *f2fs_get_new_data_page(struct inode *inode,
 854                 struct page *ipage, pgoff_t index, bool new_i_size)
 855 {
 856         struct address_space *mapping = inode->i_mapping;
 857         struct page *page;
 858         struct dnode_of_data dn;
 859         int err;
 860
 861         page = f2fs_grab_cache_page(mapping, index, true);
 862         if (!page) {
 863                 /*
 864                  * before exiting, we should make sure ipage will be released
 865                  * if any error occur.
 866                  */
 867                 f2fs_put_page(ipage, 1);
 868                 return ERR_PTR(-ENOMEM);
 869         }
 870
 871         set_new_dnode(&dn, inode, ipage, NULL, 0);
 872         err = f2fs_reserve_block(&dn, index);
 873         if (err) {
 874                 f2fs_put_page(page, 1);
 875                 return ERR_PTR(err);
 876         }
 877         if (!ipage)
 878                 f2fs_put_dnode(&dn);
 879
 880         if (PageUptodate(page))
 881                 goto got_it;
 882
 883         if (dn.data_blkaddr == NEW_ADDR) {
 884                 zero_user_segment(page, 0, PAGE_SIZE);
 885                 if (!PageUptodate(page))
 886                         SetPageUptodate(page);
 887         } else {
 888                 f2fs_put_page(page, 1);
 889
 890                 /* if ipage exists, blkaddr should be NEW_ADDR */
 891                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 892                 page = f2fs_get_lock_data_page(inode, index, true);
 893                 if (IS_ERR(page))
 894                         return page;
 895         }
 896 got_it:
 897         if (new_i_size && i_size_read(inode) <
 898                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 899                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 900         return page;
 901 }
 902
 903 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 904 {
 905         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 906         struct f2fs_summary sum;
 907         struct node_info ni;
 908         block_t old_blkaddr;
 909         blkcnt_t count = 1;
 910         int err;
 911
 912         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 913                 return -EPERM;
 914
 915         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 916         if (err)
 917                 return err;
 918
 919         dn->data_blkaddr = datablock_addr(dn->inode,
 920                                 dn->node_page, dn->ofs_in_node);
 921         if (dn->data_blkaddr != NULL_ADDR)
 922                 goto alloc;
 923
 924         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 925                 return err;
 926
 927 alloc:
 928         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 929         old_blkaddr = dn->data_blkaddr;
 930         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 931                                         &sum, seg_type, NULL, false);
 932         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 933                 invalidate_mapping_pages(META_MAPPING(sbi),
 934                                         old_blkaddr, old_blkaddr);
 935         f2fs_set_data_blkaddr(dn);
 936
 937         /*
 938          * i_size will be updated by direct_IO. Otherwise, we'll get stale
 939          * data from unwritten block via dio_read.
 940          */
 941         return 0;
 942 }
 943
 944 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 945 {
 946         struct inode *inode = file_inode(iocb->ki_filp);
 947         struct f2fs_map_blocks map;
 948         int flag;
 949         int err = 0;
 950         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 951
 952         /* convert inline data for Direct I/O*/
 953         if (direct_io) {
 954                 err = f2fs_convert_inline_inode(inode);
 955                 if (err)
 956                         return err;
 957         }
 958
 959         if (direct_io && allow_outplace_dio(inode, iocb, from))
 960                 return 0;
 961
 962         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 963                 return 0;
 964
 965         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 966         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 967         if (map.m_len > map.m_lblk)
 968                 map.m_len -= map.m_lblk;
 969         else
 970                 map.m_len = 0;
 971
 972         map.m_next_pgofs = NULL;
 973         map.m_next_extent = NULL;
 974         map.m_seg_type = NO_CHECK_TYPE;
 975         map.m_may_create = true;
 976
 977         if (direct_io) {
 978                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 979                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
 980                                         F2FS_GET_BLOCK_PRE_AIO :
 981                                         F2FS_GET_BLOCK_PRE_DIO;
 982                 goto map_blocks;
 983         }
 984         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 985                 err = f2fs_convert_inline_inode(inode);
 986                 if (err)
 987                         return err;
 988         }
 989         if (f2fs_has_inline_data(inode))
 990                 return err;
 991
 992         flag = F2FS_GET_BLOCK_PRE_AIO;
 993
 994 map_blocks:
 995         err = f2fs_map_blocks(inode, &map, 1, flag);
 996         if (map.m_len > 0 && err == -ENOSPC) {
 997                 if (!direct_io)
 998                         set_inode_flag(inode, FI_NO_PREALLOC);
 999                 err = 0;
1000         }
1001         return err;
1002 }
1003
1004 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1005 {
1006         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1007                 if (lock)
1008                         down_read(&sbi->node_change);
1009                 else
1010                         up_read(&sbi->node_change);
1011         } else {
1012                 if (lock)
1013                         f2fs_lock_op(sbi);
1014                 else
1015                         f2fs_unlock_op(sbi);
1016         }
1017 }
1018
1019 /*
1020  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1021  * f2fs_map_blocks structure.
1022  * If original data blocks are allocated, then give them to blockdev.
1023  * Otherwise,
1024  *     a. preallocate requested block addresses
1025  *     b. do not use extent cache for better performance
1026  *     c. give the block addresses to blockdev
1027  */
1028 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1029                                                 int create, int flag)
1030 {
1031         unsigned int maxblocks = map->m_len;
1032         struct dnode_of_data dn;
1033         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1034         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1035         pgoff_t pgofs, end_offset, end;
1036         int err = 0, ofs = 1;
1037         unsigned int ofs_in_node, last_ofs_in_node;
1038         blkcnt_t prealloc;
1039         struct extent_info ei = {0,0,0};
1040         block_t blkaddr;
1041         unsigned int start_pgofs;
1042
1043         if (!maxblocks)
1044                 return 0;
1045
1046         map->m_len = 0;
1047         map->m_flags = 0;
1048
1049         /* it only supports block size == page size */
1050         pgofs = (pgoff_t)map->m_lblk;
1051         end = pgofs + maxblocks;
1052
1053         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1054                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1055                                                         map->m_may_create)
1056                         goto next_dnode;
1057
1058                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1059                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1060                 map->m_flags = F2FS_MAP_MAPPED;
1061                 if (map->m_next_extent)
1062                         *map->m_next_extent = pgofs + map->m_len;
1063
1064                 /* for hardware encryption, but to avoid potential issue in future */
1065                 if (flag == F2FS_GET_BLOCK_DIO)
1066                         f2fs_wait_on_block_writeback_range(inode,
1067                                                 map->m_pblk, map->m_len);
1068                 goto out;
1069         }
1070
1071 next_dnode:
1072         if (map->m_may_create)
1073                 __do_map_lock(sbi, flag, true);
1074
1075         /* When reading holes, we need its node page */
1076         set_new_dnode(&dn, inode, NULL, NULL, 0);
1077         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1078         if (err) {
1079                 if (flag == F2FS_GET_BLOCK_BMAP)
1080                         map->m_pblk = 0;
1081                 if (err == -ENOENT) {
1082                         err = 0;
1083                         if (map->m_next_pgofs)
1084                                 *map->m_next_pgofs =
1085                                         f2fs_get_next_page_offset(&dn, pgofs);
1086                         if (map->m_next_extent)
1087                                 *map->m_next_extent =
1088                                         f2fs_get_next_page_offset(&dn, pgofs);
1089                 }
1090                 goto unlock_out;
1091         }
1092
1093         start_pgofs = pgofs;
1094         prealloc = 0;
1095         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1096         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1097
1098 next_block:
1099         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1100
1101         if (__is_valid_data_blkaddr(blkaddr) &&
1102                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1103                 err = -EFAULT;
1104                 goto sync_out;
1105         }
1106
1107         if (__is_valid_data_blkaddr(blkaddr)) {
1108                 /* use out-place-update for driect IO under LFS mode */
1109                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1110                                                         map->m_may_create) {
1111                         err = __allocate_data_block(&dn, map->m_seg_type);
1112                         if (!err) {
1113                                 blkaddr = dn.data_blkaddr;
1114                                 set_inode_flag(inode, FI_APPEND_WRITE);
1115                         }
1116                 }
1117         } else {
1118                 if (create) {
1119                         if (unlikely(f2fs_cp_error(sbi))) {
1120                                 err = -EIO;
1121                                 goto sync_out;
1122                         }
1123                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1124                                 if (blkaddr == NULL_ADDR) {
1125                                         prealloc++;
1126                                         last_ofs_in_node = dn.ofs_in_node;
1127                                 }
1128                         } else {
1129                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1130                                         flag != F2FS_GET_BLOCK_DIO);
1131                                 err = __allocate_data_block(&dn,
1132                                                         map->m_seg_type);
1133                                 if (!err)
1134                                         set_inode_flag(inode, FI_APPEND_WRITE);
1135                         }
1136                         if (err)
1137                                 goto sync_out;
1138                         map->m_flags |= F2FS_MAP_NEW;
1139                         blkaddr = dn.data_blkaddr;
1140                 } else {
1141                         if (flag == F2FS_GET_BLOCK_BMAP) {
1142                                 map->m_pblk = 0;
1143                                 goto sync_out;
1144                         }
1145                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1146                                 goto sync_out;
1147                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1148                                                 blkaddr == NULL_ADDR) {
1149                                 if (map->m_next_pgofs)
1150                                         *map->m_next_pgofs = pgofs + 1;
1151                                 goto sync_out;
1152                         }
1153                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1154                                 /* for defragment case */
1155                                 if (map->m_next_pgofs)
1156                                         *map->m_next_pgofs = pgofs + 1;
1157                                 goto sync_out;
1158                         }
1159                 }
1160         }
1161
1162         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1163                 goto skip;
1164
1165         if (map->m_len == 0) {
1166                 /* preallocated unwritten block should be mapped for fiemap. */
1167                 if (blkaddr == NEW_ADDR)
1168                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1169                 map->m_flags |= F2FS_MAP_MAPPED;
1170
1171                 map->m_pblk = blkaddr;
1172                 map->m_len = 1;
1173         } else if ((map->m_pblk != NEW_ADDR &&
1174                         blkaddr == (map->m_pblk + ofs)) ||
1175                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1176                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1177                 ofs++;
1178                 map->m_len++;
1179         } else {
1180                 goto sync_out;
1181         }
1182
1183 skip:
1184         dn.ofs_in_node++;
1185         pgofs++;
1186
1187         /* preallocate blocks in batch for one dnode page */
1188         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1189                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1190
1191                 dn.ofs_in_node = ofs_in_node;
1192                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1193                 if (err)
1194                         goto sync_out;
1195
1196                 map->m_len += dn.ofs_in_node - ofs_in_node;
1197                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1198                         err = -ENOSPC;
1199                         goto sync_out;
1200                 }
1201                 dn.ofs_in_node = end_offset;
1202         }
1203
1204         if (pgofs >= end)
1205                 goto sync_out;
1206         else if (dn.ofs_in_node < end_offset)
1207                 goto next_block;
1208
1209         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1210                 if (map->m_flags & F2FS_MAP_MAPPED) {
1211                         unsigned int ofs = start_pgofs - map->m_lblk;
1212
1213                         f2fs_update_extent_cache_range(&dn,
1214                                 start_pgofs, map->m_pblk + ofs,
1215                                 map->m_len - ofs);
1216                 }
1217         }
1218
1219         f2fs_put_dnode(&dn);
1220
1221         if (map->m_may_create) {
1222                 __do_map_lock(sbi, flag, false);
1223                 f2fs_balance_fs(sbi, dn.node_changed);
1224         }
1225         goto next_dnode;
1226
1227 sync_out:
1228
1229         /* for hardware encryption, but to avoid potential issue in future */
1230         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1231                 f2fs_wait_on_block_writeback_range(inode,
1232                                                 map->m_pblk, map->m_len);
1233
1234         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1235                 if (map->m_flags & F2FS_MAP_MAPPED) {
1236                         unsigned int ofs = start_pgofs - map->m_lblk;
1237
1238                         f2fs_update_extent_cache_range(&dn,
1239                                 start_pgofs, map->m_pblk + ofs,
1240                                 map->m_len - ofs);
1241                 }
1242                 if (map->m_next_extent)
1243                         *map->m_next_extent = pgofs + 1;
1244         }
1245         f2fs_put_dnode(&dn);
1246 unlock_out:
1247         if (map->m_may_create) {
1248                 __do_map_lock(sbi, flag, false);
1249                 f2fs_balance_fs(sbi, dn.node_changed);
1250         }
1251 out:
1252         trace_f2fs_map_blocks(inode, map, err);
1253         return err;
1254 }
1255
1256 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1257 {
1258         struct f2fs_map_blocks map;
1259         block_t last_lblk;
1260         int err;
1261
1262         if (pos + len > i_size_read(inode))
1263                 return false;
1264
1265         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1266         map.m_next_pgofs = NULL;
1267         map.m_next_extent = NULL;
1268         map.m_seg_type = NO_CHECK_TYPE;
1269         map.m_may_create = false;
1270         last_lblk = F2FS_BLK_ALIGN(pos + len);
1271
1272         while (map.m_lblk < last_lblk) {
1273                 map.m_len = last_lblk - map.m_lblk;
1274                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1275                 if (err || map.m_len == 0)
1276                         return false;
1277                 map.m_lblk += map.m_len;
1278         }
1279         return true;
1280 }
1281
1282 static int __get_data_block(struct inode *inode, sector_t iblock,
1283                         struct buffer_head *bh, int create, int flag,
1284                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1285 {
1286         struct f2fs_map_blocks map;
1287         int err;
1288
1289         map.m_lblk = iblock;
1290         map.m_len = bh->b_size >> inode->i_blkbits;
1291         map.m_next_pgofs = next_pgofs;
1292         map.m_next_extent = NULL;
1293         map.m_seg_type = seg_type;
1294         map.m_may_create = may_write;
1295
1296         err = f2fs_map_blocks(inode, &map, create, flag);
1297         if (!err) {
1298                 map_bh(bh, inode->i_sb, map.m_pblk);
1299                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1300                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1301         }
1302         return err;
1303 }
1304
1305 static int get_data_block(struct inode *inode, sector_t iblock,
1306                         struct buffer_head *bh_result, int create, int flag,
1307                         pgoff_t *next_pgofs)
1308 {
1309         return __get_data_block(inode, iblock, bh_result, create,
1310                                                         flag, next_pgofs,
1311                                                         NO_CHECK_TYPE, create);
1312 }
1313
1314 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1315                         struct buffer_head *bh_result, int create)
1316 {
1317         return __get_data_block(inode, iblock, bh_result, create,
1318                                 F2FS_GET_BLOCK_DIO, NULL,
1319                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1320                                 true);
1321 }
1322
1323 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1324                         struct buffer_head *bh_result, int create)
1325 {
1326         return __get_data_block(inode, iblock, bh_result, create,
1327                                 F2FS_GET_BLOCK_DIO, NULL,
1328                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1329                                 false);
1330 }
1331
1332 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1333                         struct buffer_head *bh_result, int create)
1334 {
1335         /* Block number less than F2FS MAX BLOCKS */
1336         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1337                 return -EFBIG;
1338
1339         return __get_data_block(inode, iblock, bh_result, create,
1340                                                 F2FS_GET_BLOCK_BMAP, NULL,
1341                                                 NO_CHECK_TYPE, create);
1342 }
1343
1344 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1345 {
1346         return (offset >> inode->i_blkbits);
1347 }
1348
1349 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1350 {
1351         return (blk << inode->i_blkbits);
1352 }
1353
1354 static int f2fs_xattr_fiemap(struct inode *inode,
1355                                 struct fiemap_extent_info *fieinfo)
1356 {
1357         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1358         struct page *page;
1359         struct node_info ni;
1360         __u64 phys = 0, len;
1361         __u32 flags;
1362         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1363         int err = 0;
1364
1365         if (f2fs_has_inline_xattr(inode)) {
1366                 int offset;
1367
1368                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1369                                                 inode->i_ino, false);
1370                 if (!page)
1371                         return -ENOMEM;
1372
1373                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1374                 if (err) {
1375                         f2fs_put_page(page, 1);
1376                         return err;
1377                 }
1378
1379                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1380                 offset = offsetof(struct f2fs_inode, i_addr) +
1381                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1382                                         get_inline_xattr_addrs(inode));
1383
1384                 phys += offset;
1385                 len = inline_xattr_size(inode);
1386
1387                 f2fs_put_page(page, 1);
1388
1389                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1390
1391                 if (!xnid)
1392                         flags |= FIEMAP_EXTENT_LAST;
1393
1394                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1395                 if (err || err == 1)
1396                         return err;
1397         }
1398
1399         if (xnid) {
1400                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1401                 if (!page)
1402                         return -ENOMEM;
1403
1404                 err = f2fs_get_node_info(sbi, xnid, &ni);
1405                 if (err) {
1406                         f2fs_put_page(page, 1);
1407                         return err;
1408                 }
1409
1410                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1411                 len = inode->i_sb->s_blocksize;
1412
1413                 f2fs_put_page(page, 1);
1414
1415                 flags = FIEMAP_EXTENT_LAST;
1416         }
1417
1418         if (phys)
1419                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1420
1421         return (err < 0 ? err : 0);
1422 }
1423
1424 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1425                 u64 start, u64 len)
1426 {
1427         struct buffer_head map_bh;
1428         sector_t start_blk, last_blk;
1429         pgoff_t next_pgofs;
1430         u64 logical = 0, phys = 0, size = 0;
1431         u32 flags = 0;
1432         int ret = 0;
1433
1434         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1435                 ret = f2fs_precache_extents(inode);
1436                 if (ret)
1437                         return ret;
1438         }
1439
1440         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1441         if (ret)
1442                 return ret;
1443
1444         inode_lock(inode);
1445
1446         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1447                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1448                 goto out;
1449         }
1450
1451         if (f2fs_has_inline_data(inode)) {
1452                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1453                 if (ret != -EAGAIN)
1454                         goto out;
1455         }
1456
1457         if (logical_to_blk(inode, len) == 0)
1458                 len = blk_to_logical(inode, 1);
1459
1460         start_blk = logical_to_blk(inode, start);
1461         last_blk = logical_to_blk(inode, start + len - 1);
1462
1463 next:
1464         memset(&map_bh, 0, sizeof(struct buffer_head));
1465         map_bh.b_size = len;
1466
1467         ret = get_data_block(inode, start_blk, &map_bh, 0,
1468                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1469         if (ret)
1470                 goto out;
1471
1472         /* HOLE */
1473         if (!buffer_mapped(&map_bh)) {
1474                 start_blk = next_pgofs;
1475
1476                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1477                                         F2FS_I_SB(inode)->max_file_blocks))
1478                         goto prep_next;
1479
1480                 flags |= FIEMAP_EXTENT_LAST;
1481         }
1482
1483         if (size) {
1484                 if (IS_ENCRYPTED(inode))
1485                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1486
1487                 ret = fiemap_fill_next_extent(fieinfo, logical,
1488                                 phys, size, flags);
1489         }
1490
1491         if (start_blk > last_blk || ret)
1492                 goto out;
1493
1494         logical = blk_to_logical(inode, start_blk);
1495         phys = blk_to_logical(inode, map_bh.b_blocknr);
1496         size = map_bh.b_size;
1497         flags = 0;
1498         if (buffer_unwritten(&map_bh))
1499                 flags = FIEMAP_EXTENT_UNWRITTEN;
1500
1501         start_blk += logical_to_blk(inode, size);
1502
1503 prep_next:
1504         cond_resched();
1505         if (fatal_signal_pending(current))
1506                 ret = -EINTR;
1507         else
1508                 goto next;
1509 out:
1510         if (ret == 1)
1511                 ret = 0;
1512
1513         inode_unlock(inode);
1514         return ret;
1515 }
1516
1517 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1518                                         unsigned nr_pages,
1519                                         struct f2fs_map_blocks *map,
1520                                         struct bio **bio_ret,
1521                                         sector_t *last_block_in_bio,
1522                                         bool is_readahead)
1523 {
1524         struct bio *bio = *bio_ret;
1525         const unsigned blkbits = inode->i_blkbits;
1526         const unsigned blocksize = 1 << blkbits;
1527         sector_t block_in_file;
1528         sector_t last_block;
1529         sector_t last_block_in_file;
1530         sector_t block_nr;
1531         int ret = 0;
1532
1533         block_in_file = (sector_t)page->index;
1534         last_block = block_in_file + nr_pages;
1535         last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1536                                                         blkbits;
1537         if (last_block > last_block_in_file)
1538                 last_block = last_block_in_file;
1539
1540         /* just zeroing out page which is beyond EOF */
1541         if (block_in_file >= last_block)
1542                 goto zero_out;
1543         /*
1544          * Map blocks using the previous result first.
1545          */
1546         if ((map->m_flags & F2FS_MAP_MAPPED) &&
1547                         block_in_file > map->m_lblk &&
1548                         block_in_file < (map->m_lblk + map->m_len))
1549                 goto got_it;
1550
1551         /*
1552          * Then do more f2fs_map_blocks() calls until we are
1553          * done with this page.
1554          */
1555         map->m_lblk = block_in_file;
1556         map->m_len = last_block - block_in_file;
1557
1558         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1559         if (ret)
1560                 goto out;
1561 got_it:
1562         if ((map->m_flags & F2FS_MAP_MAPPED)) {
1563                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1564                 SetPageMappedToDisk(page);
1565
1566                 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1567                         SetPageUptodate(page);
1568                         goto confused;
1569                 }
1570
1571                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1572                                                 DATA_GENERIC_ENHANCE_READ)) {
1573                         ret = -EFAULT;
1574                         goto out;
1575                 }
1576         } else {
1577 zero_out:
1578                 zero_user_segment(page, 0, PAGE_SIZE);
1579                 if (!PageUptodate(page))
1580                         SetPageUptodate(page);
1581                 unlock_page(page);
1582                 goto out;
1583         }
1584
1585         /*
1586          * This page will go to BIO.  Do we need to send this
1587          * BIO off first?
1588          */
1589         if (bio && (*last_block_in_bio != block_nr - 1 ||
1590                 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1591 submit_and_realloc:
1592                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1593                 bio = NULL;
1594         }
1595         if (bio == NULL) {
1596                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1597                                 is_readahead ? REQ_RAHEAD : 0);
1598                 if (IS_ERR(bio)) {
1599                         ret = PTR_ERR(bio);
1600                         bio = NULL;
1601                         goto out;
1602                 }
1603         }
1604
1605         /*
1606          * If the page is under writeback, we need to wait for
1607          * its completion to see the correct decrypted data.
1608          */
1609         f2fs_wait_on_block_writeback(inode, block_nr);
1610
1611         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1612                 goto submit_and_realloc;
1613
1614         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1615         ClearPageError(page);
1616         *last_block_in_bio = block_nr;
1617         goto out;
1618 confused:
1619         if (bio) {
1620                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1621                 bio = NULL;
1622         }
1623         unlock_page(page);
1624 out:
1625         *bio_ret = bio;
1626         return ret;
1627 }
1628
1629 /*
1630  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1631  * Major change was from block_size == page_size in f2fs by default.
1632  *
1633  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1634  * this function ever deviates from doing just read-ahead, it should either
1635  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1636  * from read-ahead.
1637  */
1638 static int f2fs_mpage_readpages(struct address_space *mapping,
1639                         struct list_head *pages, struct page *page,
1640                         unsigned nr_pages, bool is_readahead)
1641 {
1642         struct bio *bio = NULL;
1643         sector_t last_block_in_bio = 0;
1644         struct inode *inode = mapping->host;
1645         struct f2fs_map_blocks map;
1646         int ret = 0;
1647
1648         map.m_pblk = 0;
1649         map.m_lblk = 0;
1650         map.m_len = 0;
1651         map.m_flags = 0;
1652         map.m_next_pgofs = NULL;
1653         map.m_next_extent = NULL;
1654         map.m_seg_type = NO_CHECK_TYPE;
1655         map.m_may_create = false;
1656
1657         for (; nr_pages; nr_pages--) {
1658                 if (pages) {
1659                         page = list_last_entry(pages, struct page, lru);
1660
1661                         prefetchw(&page->flags);
1662                         list_del(&page->lru);
1663                         if (add_to_page_cache_lru(page, mapping,
1664                                                   page->index,
1665                                                   readahead_gfp_mask(mapping)))
1666                                 goto next_page;
1667                 }
1668
1669                 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1670                                         &last_block_in_bio, is_readahead);
1671                 if (ret) {
1672                         SetPageError(page);
1673                         zero_user_segment(page, 0, PAGE_SIZE);
1674                         unlock_page(page);
1675                 }
1676 next_page:
1677                 if (pages)
1678                         put_page(page);
1679         }
1680         BUG_ON(pages && !list_empty(pages));
1681         if (bio)
1682                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1683         return pages ? 0 : ret;
1684 }
1685
1686 static int f2fs_read_data_page(struct file *file, struct page *page)
1687 {
1688         struct inode *inode = page->mapping->host;
1689         int ret = -EAGAIN;
1690
1691         trace_f2fs_readpage(page, DATA);
1692
1693         /* If the file has inline data, try to read it directly */
1694         if (f2fs_has_inline_data(inode))
1695                 ret = f2fs_read_inline_data(inode, page);
1696         if (ret == -EAGAIN)
1697                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1698         return ret;
1699 }
1700
1701 static int f2fs_read_data_pages(struct file *file,
1702                         struct address_space *mapping,
1703                         struct list_head *pages, unsigned nr_pages)
1704 {
1705         struct inode *inode = mapping->host;
1706         struct page *page = list_last_entry(pages, struct page, lru);
1707
1708         trace_f2fs_readpages(inode, page, nr_pages);
1709
1710         /* If the file has inline data, skip readpages */
1711         if (f2fs_has_inline_data(inode))
1712                 return 0;
1713
1714         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1715 }
1716
1717 static int encrypt_one_page(struct f2fs_io_info *fio)
1718 {
1719         struct inode *inode = fio->page->mapping->host;
1720         struct page *mpage;
1721         gfp_t gfp_flags = GFP_NOFS;
1722
1723         if (!f2fs_encrypted_file(inode))
1724                 return 0;
1725
1726         /* wait for GCed page writeback via META_MAPPING */
1727         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1728
1729 retry_encrypt:
1730         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1731                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1732         if (IS_ERR(fio->encrypted_page)) {
1733                 /* flush pending IOs and wait for a while in the ENOMEM case */
1734                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1735                         f2fs_flush_merged_writes(fio->sbi);
1736                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1737                         gfp_flags |= __GFP_NOFAIL;
1738                         goto retry_encrypt;
1739                 }
1740                 return PTR_ERR(fio->encrypted_page);
1741         }
1742
1743         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1744         if (mpage) {
1745                 if (PageUptodate(mpage))
1746                         memcpy(page_address(mpage),
1747                                 page_address(fio->encrypted_page), PAGE_SIZE);
1748                 f2fs_put_page(mpage, 1);
1749         }
1750         return 0;
1751 }
1752
1753 static inline bool check_inplace_update_policy(struct inode *inode,
1754                                 struct f2fs_io_info *fio)
1755 {
1756         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1757         unsigned int policy = SM_I(sbi)->ipu_policy;
1758
1759         if (policy & (0x1 << F2FS_IPU_FORCE))
1760                 return true;
1761         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1762                 return true;
1763         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1764                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1765                 return true;
1766         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1767                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1768                 return true;
1769
1770         /*
1771          * IPU for rewrite async pages
1772          */
1773         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1774                         fio && fio->op == REQ_OP_WRITE &&
1775                         !(fio->op_flags & REQ_SYNC) &&
1776                         !IS_ENCRYPTED(inode))
1777                 return true;
1778
1779         /* this is only set during fdatasync */
1780         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1781                         is_inode_flag_set(inode, FI_NEED_IPU))
1782                 return true;
1783
1784         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1785                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1786                 return true;
1787
1788         return false;
1789 }
1790
1791 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1792 {
1793         if (f2fs_is_pinned_file(inode))
1794                 return true;
1795
1796         /* if this is cold file, we should overwrite to avoid fragmentation */
1797         if (file_is_cold(inode))
1798                 return true;
1799
1800         return check_inplace_update_policy(inode, fio);
1801 }
1802
1803 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1804 {
1805         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1806
1807         if (test_opt(sbi, LFS))
1808                 return true;
1809         if (S_ISDIR(inode->i_mode))
1810                 return true;
1811         if (IS_NOQUOTA(inode))
1812                 return true;
1813         if (f2fs_is_atomic_file(inode))
1814                 return true;
1815         if (fio) {
1816                 if (is_cold_data(fio->page))
1817                         return true;
1818                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1819                         return true;
1820                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1821                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1822                         return true;
1823         }
1824         return false;
1825 }
1826
1827 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1828 {
1829         struct inode *inode = fio->page->mapping->host;
1830
1831         if (f2fs_should_update_outplace(inode, fio))
1832                 return false;
1833
1834         return f2fs_should_update_inplace(inode, fio);
1835 }
1836
1837 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1838 {
1839         struct page *page = fio->page;
1840         struct inode *inode = page->mapping->host;
1841         struct dnode_of_data dn;
1842         struct extent_info ei = {0,0,0};
1843         struct node_info ni;
1844         bool ipu_force = false;
1845         int err = 0;
1846
1847         set_new_dnode(&dn, inode, NULL, NULL, 0);
1848         if (need_inplace_update(fio) &&
1849                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1850                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1851
1852                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1853                                                 DATA_GENERIC_ENHANCE))
1854                         return -EFAULT;
1855
1856                 ipu_force = true;
1857                 fio->need_lock = LOCK_DONE;
1858                 goto got_it;
1859         }
1860
1861         /* Deadlock due to between page->lock and f2fs_lock_op */
1862         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1863                 return -EAGAIN;
1864
1865         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1866         if (err)
1867                 goto out;
1868
1869         fio->old_blkaddr = dn.data_blkaddr;
1870
1871         /* This page is already truncated */
1872         if (fio->old_blkaddr == NULL_ADDR) {
1873                 ClearPageUptodate(page);
1874                 clear_cold_data(page);
1875                 goto out_writepage;
1876         }
1877 got_it:
1878         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1879                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1880                                                 DATA_GENERIC_ENHANCE)) {
1881                 err = -EFAULT;
1882                 goto out_writepage;
1883         }
1884         /*
1885          * If current allocation needs SSR,
1886          * it had better in-place writes for updated data.
1887          */
1888         if (ipu_force ||
1889                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1890                                         need_inplace_update(fio))) {
1891                 err = encrypt_one_page(fio);
1892                 if (err)
1893                         goto out_writepage;
1894
1895                 set_page_writeback(page);
1896                 ClearPageError(page);
1897                 f2fs_put_dnode(&dn);
1898                 if (fio->need_lock == LOCK_REQ)
1899                         f2fs_unlock_op(fio->sbi);
1900                 err = f2fs_inplace_write_data(fio);
1901                 if (err) {
1902                         if (f2fs_encrypted_file(inode))
1903                                 fscrypt_pullback_bio_page(&fio->encrypted_page,
1904                                                                         true);
1905                         if (PageWriteback(page))
1906                                 end_page_writeback(page);
1907                 } else {
1908                         set_inode_flag(inode, FI_UPDATE_WRITE);
1909                 }
1910                 trace_f2fs_do_write_data_page(fio->page, IPU);
1911                 return err;
1912         }
1913
1914         if (fio->need_lock == LOCK_RETRY) {
1915                 if (!f2fs_trylock_op(fio->sbi)) {
1916                         err = -EAGAIN;
1917                         goto out_writepage;
1918                 }
1919                 fio->need_lock = LOCK_REQ;
1920         }
1921
1922         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1923         if (err)
1924                 goto out_writepage;
1925
1926         fio->version = ni.version;
1927
1928         err = encrypt_one_page(fio);
1929         if (err)
1930                 goto out_writepage;
1931
1932         set_page_writeback(page);
1933         ClearPageError(page);
1934
1935         /* LFS mode write path */
1936         f2fs_outplace_write_data(&dn, fio);
1937         trace_f2fs_do_write_data_page(page, OPU);
1938         set_inode_flag(inode, FI_APPEND_WRITE);
1939         if (page->index == 0)
1940                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1941 out_writepage:
1942         f2fs_put_dnode(&dn);
1943 out:
1944         if (fio->need_lock == LOCK_REQ)
1945                 f2fs_unlock_op(fio->sbi);
1946         return err;
1947 }
1948
1949 static int __write_data_page(struct page *page, bool *submitted,
1950                                 struct writeback_control *wbc,
1951                                 enum iostat_type io_type)
1952 {
1953         struct inode *inode = page->mapping->host;
1954         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1955         loff_t i_size = i_size_read(inode);
1956         const pgoff_t end_index = ((unsigned long long) i_size)
1957                                                         >> PAGE_SHIFT;
1958         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1959         unsigned offset = 0;
1960         bool need_balance_fs = false;
1961         int err = 0;
1962         struct f2fs_io_info fio = {
1963                 .sbi = sbi,
1964                 .ino = inode->i_ino,
1965                 .type = DATA,
1966                 .op = REQ_OP_WRITE,
1967                 .op_flags = wbc_to_write_flags(wbc),
1968                 .old_blkaddr = NULL_ADDR,
1969                 .page = page,
1970                 .encrypted_page = NULL,
1971                 .submitted = false,
1972                 .need_lock = LOCK_RETRY,
1973                 .io_type = io_type,
1974                 .io_wbc = wbc,
1975         };
1976
1977         trace_f2fs_writepage(page, DATA);
1978
1979         /* we should bypass data pages to proceed the kworkder jobs */
1980         if (unlikely(f2fs_cp_error(sbi))) {
1981                 mapping_set_error(page->mapping, -EIO);
1982                 /*
1983                  * don't drop any dirty dentry pages for keeping lastest
1984                  * directory structure.
1985                  */
1986                 if (S_ISDIR(inode->i_mode))
1987                         goto redirty_out;
1988                 goto out;
1989         }
1990
1991         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1992                 goto redirty_out;
1993
1994         if (page->index < end_index)
1995                 goto write;
1996
1997         /*
1998          * If the offset is out-of-range of file size,
1999          * this page does not have to be written to disk.
2000          */
2001         offset = i_size & (PAGE_SIZE - 1);
2002         if ((page->index >= end_index + 1) || !offset)
2003                 goto out;
2004
2005         zero_user_segment(page, offset, PAGE_SIZE);
2006 write:
2007         if (f2fs_is_drop_cache(inode))
2008                 goto out;
2009         /* we should not write 0'th page having journal header */
2010         if (f2fs_is_volatile_file(inode) && (!page->index ||
2011                         (!wbc->for_reclaim &&
2012                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2013                 goto redirty_out;
2014
2015         /* Dentry blocks are controlled by checkpoint */
2016         if (S_ISDIR(inode->i_mode)) {
2017                 fio.need_lock = LOCK_DONE;
2018                 err = f2fs_do_write_data_page(&fio);
2019                 goto done;
2020         }
2021
2022         if (!wbc->for_reclaim)
2023                 need_balance_fs = true;
2024         else if (has_not_enough_free_secs(sbi, 0, 0))
2025                 goto redirty_out;
2026         else
2027                 set_inode_flag(inode, FI_HOT_DATA);
2028
2029         err = -EAGAIN;
2030         if (f2fs_has_inline_data(inode)) {
2031                 err = f2fs_write_inline_data(inode, page);
2032                 if (!err)
2033                         goto out;
2034         }
2035
2036         if (err == -EAGAIN) {
2037                 err = f2fs_do_write_data_page(&fio);
2038                 if (err == -EAGAIN) {
2039                         fio.need_lock = LOCK_REQ;
2040                         err = f2fs_do_write_data_page(&fio);
2041                 }
2042         }
2043
2044         if (err) {
2045                 file_set_keep_isize(inode);
2046         } else {
2047                 down_write(&F2FS_I(inode)->i_sem);
2048                 if (F2FS_I(inode)->last_disk_size < psize)
2049                         F2FS_I(inode)->last_disk_size = psize;
2050                 up_write(&F2FS_I(inode)->i_sem);
2051         }
2052
2053 done:
2054         if (err && err != -ENOENT)
2055                 goto redirty_out;
2056
2057 out:
2058         inode_dec_dirty_pages(inode);
2059         if (err) {
2060                 ClearPageUptodate(page);
2061                 clear_cold_data(page);
2062         }
2063
2064         if (wbc->for_reclaim) {
2065                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2066                 clear_inode_flag(inode, FI_HOT_DATA);
2067                 f2fs_remove_dirty_inode(inode);
2068                 submitted = NULL;
2069         }
2070
2071         unlock_page(page);
2072         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2073                                         !F2FS_I(inode)->cp_task)
2074                 f2fs_balance_fs(sbi, need_balance_fs);
2075
2076         if (unlikely(f2fs_cp_error(sbi))) {
2077                 f2fs_submit_merged_write(sbi, DATA);
2078                 submitted = NULL;
2079         }
2080
2081         if (submitted)
2082                 *submitted = fio.submitted;
2083
2084         return 0;
2085
2086 redirty_out:
2087         redirty_page_for_writepage(wbc, page);
2088         /*
2089          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2090          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2091          * file_write_and_wait_range() will see EIO error, which is critical
2092          * to return value of fsync() followed by atomic_write failure to user.
2093          */
2094         if (!err || wbc->for_reclaim)
2095                 return AOP_WRITEPAGE_ACTIVATE;
2096         unlock_page(page);
2097         return err;
2098 }
2099
2100 static int f2fs_write_data_page(struct page *page,
2101                                         struct writeback_control *wbc)
2102 {
2103         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2104 }
2105
2106 /*
2107  * This function was copied from write_cche_pages from mm/page-writeback.c.
2108  * The major change is making write step of cold data page separately from
2109  * warm/hot data page.
2110  */
2111 static int f2fs_write_cache_pages(struct address_space *mapping,
2112                                         struct writeback_control *wbc,
2113                                         enum iostat_type io_type)
2114 {
2115         int ret = 0;
2116         int done = 0;
2117         struct pagevec pvec;
2118         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2119         int nr_pages;
2120         pgoff_t uninitialized_var(writeback_index);
2121         pgoff_t index;
2122         pgoff_t end;            /* Inclusive */
2123         pgoff_t done_index;
2124         int cycled;
2125         int range_whole = 0;
2126         xa_mark_t tag;
2127         int nwritten = 0;
2128
2129         pagevec_init(&pvec);
2130
2131         if (get_dirty_pages(mapping->host) <=
2132                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2133                 set_inode_flag(mapping->host, FI_HOT_DATA);
2134         else
2135                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2136
2137         if (wbc->range_cyclic) {
2138                 writeback_index = mapping->writeback_index; /* prev offset */
2139                 index = writeback_index;
2140                 if (index == 0)
2141                         cycled = 1;
2142                 else
2143                         cycled = 0;
2144                 end = -1;
2145         } else {
2146                 index = wbc->range_start >> PAGE_SHIFT;
2147                 end = wbc->range_end >> PAGE_SHIFT;
2148                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2149                         range_whole = 1;
2150                 cycled = 1; /* ignore range_cyclic tests */
2151         }
2152         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2153                 tag = PAGECACHE_TAG_TOWRITE;
2154         else
2155                 tag = PAGECACHE_TAG_DIRTY;
2156 retry:
2157         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2158                 tag_pages_for_writeback(mapping, index, end);
2159         done_index = index;
2160         while (!done && (index <= end)) {
2161                 int i;
2162
2163                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2164                                 tag);
2165                 if (nr_pages == 0)
2166                         break;
2167
2168                 for (i = 0; i < nr_pages; i++) {
2169                         struct page *page = pvec.pages[i];
2170                         bool submitted = false;
2171
2172                         /* give a priority to WB_SYNC threads */
2173                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2174                                         wbc->sync_mode == WB_SYNC_NONE) {
2175                                 done = 1;
2176                                 break;
2177                         }
2178
2179                         done_index = page->index;
2180 retry_write:
2181                         lock_page(page);
2182
2183                         if (unlikely(page->mapping != mapping)) {
2184 continue_unlock:
2185                                 unlock_page(page);
2186                                 continue;
2187                         }
2188
2189                         if (!PageDirty(page)) {
2190                                 /* someone wrote it for us */
2191                                 goto continue_unlock;
2192                         }
2193
2194                         if (PageWriteback(page)) {
2195                                 if (wbc->sync_mode != WB_SYNC_NONE)
2196                                         f2fs_wait_on_page_writeback(page,
2197                                                         DATA, true, true);
2198                                 else
2199                                         goto continue_unlock;
2200                         }
2201
2202                         if (!clear_page_dirty_for_io(page))
2203                                 goto continue_unlock;
2204
2205                         ret = __write_data_page(page, &submitted, wbc, io_type);
2206                         if (unlikely(ret)) {
2207                                 /*
2208                                  * keep nr_to_write, since vfs uses this to
2209                                  * get # of written pages.
2210                                  */
2211                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2212                                         unlock_page(page);
2213                                         ret = 0;
2214                                         continue;
2215                                 } else if (ret == -EAGAIN) {
2216                                         ret = 0;
2217                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2218                                                 cond_resched();
2219                                                 congestion_wait(BLK_RW_ASYNC,
2220                                                                         HZ/50);
2221                                                 goto retry_write;
2222                                         }
2223                                         continue;
2224                                 }
2225                                 done_index = page->index + 1;
2226                                 done = 1;
2227                                 break;
2228                         } else if (submitted) {
2229                                 nwritten++;
2230                         }
2231
2232                         if (--wbc->nr_to_write <= 0 &&
2233                                         wbc->sync_mode == WB_SYNC_NONE) {
2234                                 done = 1;
2235                                 break;
2236                         }
2237                 }
2238                 pagevec_release(&pvec);
2239                 cond_resched();
2240         }
2241
2242         if (!cycled && !done) {
2243                 cycled = 1;
2244                 index = 0;
2245                 end = writeback_index - 1;
2246                 goto retry;
2247         }
2248         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2249                 mapping->writeback_index = done_index;
2250
2251         if (nwritten)
2252                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2253                                                                 NULL, 0, DATA);
2254
2255         return ret;
2256 }
2257
2258 static inline bool __should_serialize_io(struct inode *inode,
2259                                         struct writeback_control *wbc)
2260 {
2261         if (!S_ISREG(inode->i_mode))
2262                 return false;
2263         if (IS_NOQUOTA(inode))
2264                 return false;
2265         if (wbc->sync_mode != WB_SYNC_ALL)
2266                 return true;
2267         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2268                 return true;
2269         return false;
2270 }
2271
2272 static int __f2fs_write_data_pages(struct address_space *mapping,
2273                                                 struct writeback_control *wbc,
2274                                                 enum iostat_type io_type)
2275 {
2276         struct inode *inode = mapping->host;
2277         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2278         struct blk_plug plug;
2279         int ret;
2280         bool locked = false;
2281
2282         /* deal with chardevs and other special file */
2283         if (!mapping->a_ops->writepage)
2284                 return 0;
2285
2286         /* skip writing if there is no dirty page in this inode */
2287         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2288                 return 0;
2289
2290         /* during POR, we don't need to trigger writepage at all. */
2291         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2292                 goto skip_write;
2293
2294         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2295                         wbc->sync_mode == WB_SYNC_NONE &&
2296                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2297                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2298                 goto skip_write;
2299
2300         /* skip writing during file defragment */
2301         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2302                 goto skip_write;
2303
2304         trace_f2fs_writepages(mapping->host, wbc, DATA);
2305
2306         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2307         if (wbc->sync_mode == WB_SYNC_ALL)
2308                 atomic_inc(&sbi->wb_sync_req[DATA]);
2309         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2310                 goto skip_write;
2311
2312         if (__should_serialize_io(inode, wbc)) {
2313                 mutex_lock(&sbi->writepages);
2314                 locked = true;
2315         }
2316
2317         blk_start_plug(&plug);
2318         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2319         blk_finish_plug(&plug);
2320
2321         if (locked)
2322                 mutex_unlock(&sbi->writepages);
2323
2324         if (wbc->sync_mode == WB_SYNC_ALL)
2325                 atomic_dec(&sbi->wb_sync_req[DATA]);
2326         /*
2327          * if some pages were truncated, we cannot guarantee its mapping->host
2328          * to detect pending bios.
2329          */
2330
2331         f2fs_remove_dirty_inode(inode);
2332         return ret;
2333
2334 skip_write:
2335         wbc->pages_skipped += get_dirty_pages(inode);
2336         trace_f2fs_writepages(mapping->host, wbc, DATA);
2337         return 0;
2338 }
2339
2340 static int f2fs_write_data_pages(struct address_space *mapping,
2341                             struct writeback_control *wbc)
2342 {
2343         struct inode *inode = mapping->host;
2344
2345         return __f2fs_write_data_pages(mapping, wbc,
2346                         F2FS_I(inode)->cp_task == current ?
2347                         FS_CP_DATA_IO : FS_DATA_IO);
2348 }
2349
2350 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2351 {
2352         struct inode *inode = mapping->host;
2353         loff_t i_size = i_size_read(inode);
2354
2355         if (to > i_size) {
2356                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2357                 down_write(&F2FS_I(inode)->i_mmap_sem);
2358
2359                 truncate_pagecache(inode, i_size);
2360                 if (!IS_NOQUOTA(inode))
2361                         f2fs_truncate_blocks(inode, i_size, true);
2362
2363                 up_write(&F2FS_I(inode)->i_mmap_sem);
2364                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2365         }
2366 }
2367
2368 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2369                         struct page *page, loff_t pos, unsigned len,
2370                         block_t *blk_addr, bool *node_changed)
2371 {
2372         struct inode *inode = page->mapping->host;
2373         pgoff_t index = page->index;
2374         struct dnode_of_data dn;
2375         struct page *ipage;
2376         bool locked = false;
2377         struct extent_info ei = {0,0,0};
2378         int err = 0;
2379         int flag;
2380
2381         /*
2382          * we already allocated all the blocks, so we don't need to get
2383          * the block addresses when there is no need to fill the page.
2384          */
2385         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2386                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2387                 return 0;
2388
2389         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2390         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2391                 flag = F2FS_GET_BLOCK_DEFAULT;
2392         else
2393                 flag = F2FS_GET_BLOCK_PRE_AIO;
2394
2395         if (f2fs_has_inline_data(inode) ||
2396                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2397                 __do_map_lock(sbi, flag, true);
2398                 locked = true;
2399         }
2400 restart:
2401         /* check inline_data */
2402         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2403         if (IS_ERR(ipage)) {
2404                 err = PTR_ERR(ipage);
2405                 goto unlock_out;
2406         }
2407
2408         set_new_dnode(&dn, inode, ipage, ipage, 0);
2409
2410         if (f2fs_has_inline_data(inode)) {
2411                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2412                         f2fs_do_read_inline_data(page, ipage);
2413                         set_inode_flag(inode, FI_DATA_EXIST);
2414                         if (inode->i_nlink)
2415                                 set_inline_node(ipage);
2416                 } else {
2417                         err = f2fs_convert_inline_page(&dn, page);
2418                         if (err)
2419                                 goto out;
2420                         if (dn.data_blkaddr == NULL_ADDR)
2421                                 err = f2fs_get_block(&dn, index);
2422                 }
2423         } else if (locked) {
2424                 err = f2fs_get_block(&dn, index);
2425         } else {
2426                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2427                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2428                 } else {
2429                         /* hole case */
2430                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2431                         if (err || dn.data_blkaddr == NULL_ADDR) {
2432                                 f2fs_put_dnode(&dn);
2433                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2434                                                                 true);
2435                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2436                                 locked = true;
2437                                 goto restart;
2438                         }
2439                 }
2440         }
2441
2442         /* convert_inline_page can make node_changed */
2443         *blk_addr = dn.data_blkaddr;
2444         *node_changed = dn.node_changed;
2445 out:
2446         f2fs_put_dnode(&dn);
2447 unlock_out:
2448         if (locked)
2449                 __do_map_lock(sbi, flag, false);
2450         return err;
2451 }
2452
2453 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2454                 loff_t pos, unsigned len, unsigned flags,
2455                 struct page **pagep, void **fsdata)
2456 {
2457         struct inode *inode = mapping->host;
2458         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2459         struct page *page = NULL;
2460         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2461         bool need_balance = false, drop_atomic = false;
2462         block_t blkaddr = NULL_ADDR;
2463         int err = 0;
2464
2465         trace_f2fs_write_begin(inode, pos, len, flags);
2466
2467         err = f2fs_is_checkpoint_ready(sbi);
2468         if (err)
2469                 goto fail;
2470
2471         if ((f2fs_is_atomic_file(inode) &&
2472                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2473                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2474                 err = -ENOMEM;
2475                 drop_atomic = true;
2476                 goto fail;
2477         }
2478
2479         /*
2480          * We should check this at this moment to avoid deadlock on inode page
2481          * and #0 page. The locking rule for inline_data conversion should be:
2482          * lock_page(page #0) -> lock_page(inode_page)
2483          */
2484         if (index != 0) {
2485                 err = f2fs_convert_inline_inode(inode);
2486                 if (err)
2487                         goto fail;
2488         }
2489 repeat:
2490         /*
2491          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2492          * wait_for_stable_page. Will wait that below with our IO control.
2493          */
2494         page = f2fs_pagecache_get_page(mapping, index,
2495                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2496         if (!page) {
2497                 err = -ENOMEM;
2498                 goto fail;
2499         }
2500
2501         *pagep = page;
2502
2503         err = prepare_write_begin(sbi, page, pos, len,
2504                                         &blkaddr, &need_balance);
2505         if (err)
2506                 goto fail;
2507
2508         if (need_balance && !IS_NOQUOTA(inode) &&
2509                         has_not_enough_free_secs(sbi, 0, 0)) {
2510                 unlock_page(page);
2511                 f2fs_balance_fs(sbi, true);
2512                 lock_page(page);
2513                 if (page->mapping != mapping) {
2514                         /* The page got truncated from under us */
2515                         f2fs_put_page(page, 1);
2516                         goto repeat;
2517                 }
2518         }
2519
2520         f2fs_wait_on_page_writeback(page, DATA, false, true);
2521
2522         if (len == PAGE_SIZE || PageUptodate(page))
2523                 return 0;
2524
2525         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2526                 zero_user_segment(page, len, PAGE_SIZE);
2527                 return 0;
2528         }
2529
2530         if (blkaddr == NEW_ADDR) {
2531                 zero_user_segment(page, 0, PAGE_SIZE);
2532                 SetPageUptodate(page);
2533         } else {
2534                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2535                                 DATA_GENERIC_ENHANCE_READ)) {
2536                         err = -EFAULT;
2537                         goto fail;
2538                 }
2539                 err = f2fs_submit_page_read(inode, page, blkaddr);
2540                 if (err)
2541                         goto fail;
2542
2543                 lock_page(page);
2544                 if (unlikely(page->mapping != mapping)) {
2545                         f2fs_put_page(page, 1);
2546                         goto repeat;
2547                 }
2548                 if (unlikely(!PageUptodate(page))) {
2549                         err = -EIO;
2550                         goto fail;
2551                 }
2552         }
2553         return 0;
2554
2555 fail:
2556         f2fs_put_page(page, 1);
2557         f2fs_write_failed(mapping, pos + len);
2558         if (drop_atomic)
2559                 f2fs_drop_inmem_pages_all(sbi, false);
2560         return err;
2561 }
2562
2563 static int f2fs_write_end(struct file *file,
2564                         struct address_space *mapping,
2565                         loff_t pos, unsigned len, unsigned copied,
2566                         struct page *page, void *fsdata)
2567 {
2568         struct inode *inode = page->mapping->host;
2569
2570         trace_f2fs_write_end(inode, pos, len, copied);
2571
2572         /*
2573          * This should be come from len == PAGE_SIZE, and we expect copied
2574          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2575          * let generic_perform_write() try to copy data again through copied=0.
2576          */
2577         if (!PageUptodate(page)) {
2578                 if (unlikely(copied != len))
2579                         copied = 0;
2580                 else
2581                         SetPageUptodate(page);
2582         }
2583         if (!copied)
2584                 goto unlock_out;
2585
2586         set_page_dirty(page);
2587
2588         if (pos + copied > i_size_read(inode))
2589                 f2fs_i_size_write(inode, pos + copied);
2590 unlock_out:
2591         f2fs_put_page(page, 1);
2592         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2593         return copied;
2594 }
2595
2596 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2597                            loff_t offset)
2598 {
2599         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2600         unsigned blkbits = i_blkbits;
2601         unsigned blocksize_mask = (1 << blkbits) - 1;
2602         unsigned long align = offset | iov_iter_alignment(iter);
2603         struct block_device *bdev = inode->i_sb->s_bdev;
2604
2605         if (align & blocksize_mask) {
2606                 if (bdev)
2607                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2608                 blocksize_mask = (1 << blkbits) - 1;
2609                 if (align & blocksize_mask)
2610                         return -EINVAL;
2611                 return 1;
2612         }
2613         return 0;
2614 }
2615
2616 static void f2fs_dio_end_io(struct bio *bio)
2617 {
2618         struct f2fs_private_dio *dio = bio->bi_private;
2619
2620         dec_page_count(F2FS_I_SB(dio->inode),
2621                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2622
2623         bio->bi_private = dio->orig_private;
2624         bio->bi_end_io = dio->orig_end_io;
2625
2626         kvfree(dio);
2627
2628         bio_endio(bio);
2629 }
2630
2631 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2632                                                         loff_t file_offset)
2633 {
2634         struct f2fs_private_dio *dio;
2635         bool write = (bio_op(bio) == REQ_OP_WRITE);
2636
2637         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2638                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2639         if (!dio)
2640                 goto out;
2641
2642         dio->inode = inode;
2643         dio->orig_end_io = bio->bi_end_io;
2644         dio->orig_private = bio->bi_private;
2645         dio->write = write;
2646
2647         bio->bi_end_io = f2fs_dio_end_io;
2648         bio->bi_private = dio;
2649
2650         inc_page_count(F2FS_I_SB(inode),
2651                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2652
2653         submit_bio(bio);
2654         return;
2655 out:
2656         bio->bi_status = BLK_STS_IOERR;
2657         bio_endio(bio);
2658 }
2659
2660 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2661 {
2662         struct address_space *mapping = iocb->ki_filp->f_mapping;
2663         struct inode *inode = mapping->host;
2664         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2665         struct f2fs_inode_info *fi = F2FS_I(inode);
2666         size_t count = iov_iter_count(iter);
2667         loff_t offset = iocb->ki_pos;
2668         int rw = iov_iter_rw(iter);
2669         int err;
2670         enum rw_hint hint = iocb->ki_hint;
2671         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2672         bool do_opu;
2673
2674         err = check_direct_IO(inode, iter, offset);
2675         if (err)
2676                 return err < 0 ? err : 0;
2677
2678         if (f2fs_force_buffered_io(inode, iocb, iter))
2679                 return 0;
2680
2681         do_opu = allow_outplace_dio(inode, iocb, iter);
2682
2683         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2684
2685         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2686                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2687
2688         if (iocb->ki_flags & IOCB_NOWAIT) {
2689                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2690                         iocb->ki_hint = hint;
2691                         err = -EAGAIN;
2692                         goto out;
2693                 }
2694                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2695                         up_read(&fi->i_gc_rwsem[rw]);
2696                         iocb->ki_hint = hint;
2697                         err = -EAGAIN;
2698                         goto out;
2699                 }
2700         } else {
2701                 down_read(&fi->i_gc_rwsem[rw]);
2702                 if (do_opu)
2703                         down_read(&fi->i_gc_rwsem[READ]);
2704         }
2705
2706         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2707                         iter, rw == WRITE ? get_data_block_dio_write :
2708                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2709                         DIO_LOCKING | DIO_SKIP_HOLES);
2710
2711         if (do_opu)
2712                 up_read(&fi->i_gc_rwsem[READ]);
2713
2714         up_read(&fi->i_gc_rwsem[rw]);
2715
2716         if (rw == WRITE) {
2717                 if (whint_mode == WHINT_MODE_OFF)
2718                         iocb->ki_hint = hint;
2719                 if (err > 0) {
2720                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2721                                                                         err);
2722                         if (!do_opu)
2723                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2724                 } else if (err < 0) {
2725                         f2fs_write_failed(mapping, offset + count);
2726                 }
2727         }
2728
2729 out:
2730         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2731
2732         return err;
2733 }
2734
2735 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2736                                                         unsigned int length)
2737 {
2738         struct inode *inode = page->mapping->host;
2739         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2740
2741         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2742                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2743                 return;
2744
2745         if (PageDirty(page)) {
2746                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2747                         dec_page_count(sbi, F2FS_DIRTY_META);
2748                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2749                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2750                 } else {
2751                         inode_dec_dirty_pages(inode);
2752                         f2fs_remove_dirty_inode(inode);
2753                 }
2754         }
2755
2756         clear_cold_data(page);
2757
2758         if (IS_ATOMIC_WRITTEN_PAGE(page))
2759                 return f2fs_drop_inmem_page(inode, page);
2760
2761         f2fs_clear_page_private(page);
2762 }
2763
2764 int f2fs_release_page(struct page *page, gfp_t wait)
2765 {
2766         /* If this is dirty page, keep PagePrivate */
2767         if (PageDirty(page))
2768                 return 0;
2769
2770         /* This is atomic written page, keep Private */
2771         if (IS_ATOMIC_WRITTEN_PAGE(page))
2772                 return 0;
2773
2774         clear_cold_data(page);
2775         f2fs_clear_page_private(page);
2776         return 1;
2777 }
2778
2779 static int f2fs_set_data_page_dirty(struct page *page)
2780 {
2781         struct address_space *mapping = page->mapping;
2782         struct inode *inode = mapping->host;
2783
2784         trace_f2fs_set_page_dirty(page, DATA);
2785
2786         if (!PageUptodate(page))
2787                 SetPageUptodate(page);
2788
2789         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2790                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2791                         f2fs_register_inmem_page(inode, page);
2792                         return 1;
2793                 }
2794                 /*
2795                  * Previously, this page has been registered, we just
2796                  * return here.
2797                  */
2798                 return 0;
2799         }
2800
2801         if (!PageDirty(page)) {
2802                 __set_page_dirty_nobuffers(page);
2803                 f2fs_update_dirty_page(inode, page);
2804                 return 1;
2805         }
2806         return 0;
2807 }
2808
2809 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2810 {
2811         struct inode *inode = mapping->host;
2812
2813         if (f2fs_has_inline_data(inode))
2814                 return 0;
2815
2816         /* make sure allocating whole blocks */
2817         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2818                 filemap_write_and_wait(mapping);
2819
2820         return generic_block_bmap(mapping, block, get_data_block_bmap);
2821 }
2822
2823 #ifdef CONFIG_MIGRATION
2824 #include <linux/migrate.h>
2825
2826 int f2fs_migrate_page(struct address_space *mapping,
2827                 struct page *newpage, struct page *page, enum migrate_mode mode)
2828 {
2829         int rc, extra_count;
2830         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2831         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2832
2833         BUG_ON(PageWriteback(page));
2834
2835         /* migrating an atomic written page is safe with the inmem_lock hold */
2836         if (atomic_written) {
2837                 if (mode != MIGRATE_SYNC)
2838                         return -EBUSY;
2839                 if (!mutex_trylock(&fi->inmem_lock))
2840                         return -EAGAIN;
2841         }
2842
2843         /* one extra reference was held for atomic_write page */
2844         extra_count = atomic_written ? 1 : 0;
2845         rc = migrate_page_move_mapping(mapping, newpage,
2846                                 page, mode, extra_count);
2847         if (rc != MIGRATEPAGE_SUCCESS) {
2848                 if (atomic_written)
2849                         mutex_unlock(&fi->inmem_lock);
2850                 return rc;
2851         }
2852
2853         if (atomic_written) {
2854                 struct inmem_pages *cur;
2855                 list_for_each_entry(cur, &fi->inmem_pages, list)
2856                         if (cur->page == page) {
2857                                 cur->page = newpage;
2858                                 break;
2859                         }
2860                 mutex_unlock(&fi->inmem_lock);
2861                 put_page(page);
2862                 get_page(newpage);
2863         }
2864
2865         if (PagePrivate(page)) {
2866                 f2fs_set_page_private(newpage, page_private(page));
2867                 f2fs_clear_page_private(page);
2868         }
2869
2870         if (mode != MIGRATE_SYNC_NO_COPY)
2871                 migrate_page_copy(newpage, page);
2872         else
2873                 migrate_page_states(newpage, page);
2874
2875         return MIGRATEPAGE_SUCCESS;
2876 }
2877 #endif
2878
2879 const struct address_space_operations f2fs_dblock_aops = {
2880         .readpage       = f2fs_read_data_page,
2881         .readpages      = f2fs_read_data_pages,
2882         .writepage      = f2fs_write_data_page,
2883         .writepages     = f2fs_write_data_pages,
2884         .write_begin    = f2fs_write_begin,
2885         .write_end      = f2fs_write_end,
2886         .set_page_dirty = f2fs_set_data_page_dirty,
2887         .invalidatepage = f2fs_invalidate_page,
2888         .releasepage    = f2fs_release_page,
2889         .direct_IO      = f2fs_direct_IO,
2890         .bmap           = f2fs_bmap,
2891 #ifdef CONFIG_MIGRATION
2892         .migratepage    = f2fs_migrate_page,
2893 #endif
2894 };
2895
2896 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2897 {
2898         struct address_space *mapping = page_mapping(page);
2899         unsigned long flags;
2900
2901         xa_lock_irqsave(&mapping->i_pages, flags);
2902         __xa_clear_mark(&mapping->i_pages, page_index(page),
2903                                                 PAGECACHE_TAG_DIRTY);
2904         xa_unlock_irqrestore(&mapping->i_pages, flags);
2905 }
2906
2907 int __init f2fs_init_post_read_processing(void)
2908 {
2909         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2910         if (!bio_post_read_ctx_cache)
2911                 goto fail;
2912         bio_post_read_ctx_pool =
2913                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2914                                          bio_post_read_ctx_cache);
2915         if (!bio_post_read_ctx_pool)
2916                 goto fail_free_cache;
2917         return 0;
2918
2919 fail_free_cache:
2920         kmem_cache_destroy(bio_post_read_ctx_cache);
2921 fail:
2922         return -ENOMEM;
2923 }
2924
2925 void __exit f2fs_destroy_post_read_processing(void)
2926 {
2927         mempool_destroy(bio_post_read_ctx_pool);
2928         kmem_cache_destroy(bio_post_read_ctx_cache);
2929 }