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         int i;
  90
  91         bio_for_each_segment_all(bv, bio, i) {
  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         int i;
 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, i) {
 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         for (i = 0; i < sbi->s_ndevs; i++) {
 222                 if (FDEV(i).start_blk <= blk_addr &&
 223                                         FDEV(i).end_blk >= blk_addr) {
 224                         blk_addr -= FDEV(i).start_blk;
 225                         bdev = FDEV(i).bdev;
 226                         break;
 227                 }
 228         }
 229         if (bio) {
 230                 bio_set_dev(bio, bdev);
 231                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 232         }
 233         return bdev;
 234 }
 235
 236 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 237 {
 238         int i;
 239
 240         for (i = 0; i < sbi->s_ndevs; i++)
 241                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 242                         return i;
 243         return 0;
 244 }
 245
 246 static bool __same_bdev(struct f2fs_sb_info *sbi,
 247                                 block_t blk_addr, struct bio *bio)
 248 {
 249         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 250         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 251 }
 252
 253 /*
 254  * Low-level block read/write IO operations.
 255  */
 256 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 257                                 struct writeback_control *wbc,
 258                                 int npages, bool is_read,
 259                                 enum page_type type, enum temp_type temp)
 260 {
 261         struct bio *bio;
 262
 263         bio = f2fs_bio_alloc(sbi, npages, true);
 264
 265         f2fs_target_device(sbi, blk_addr, bio);
 266         if (is_read) {
 267                 bio->bi_end_io = f2fs_read_end_io;
 268                 bio->bi_private = NULL;
 269         } else {
 270                 bio->bi_end_io = f2fs_write_end_io;
 271                 bio->bi_private = sbi;
 272                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 273         }
 274         if (wbc)
 275                 wbc_init_bio(wbc, bio);
 276
 277         return bio;
 278 }
 279
 280 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 281                                 struct bio *bio, enum page_type type)
 282 {
 283         if (!is_read_io(bio_op(bio))) {
 284                 unsigned int start;
 285
 286                 if (type != DATA && type != NODE)
 287                         goto submit_io;
 288
 289                 if (test_opt(sbi, LFS) && current->plug)
 290                         blk_finish_plug(current->plug);
 291
 292                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 293                 start %= F2FS_IO_SIZE(sbi);
 294
 295                 if (start == 0)
 296                         goto submit_io;
 297
 298                 /* fill dummy pages */
 299                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 300                         struct page *page =
 301                                 mempool_alloc(sbi->write_io_dummy,
 302                                               GFP_NOIO | __GFP_NOFAIL);
 303                         f2fs_bug_on(sbi, !page);
 304
 305                         zero_user_segment(page, 0, PAGE_SIZE);
 306                         SetPagePrivate(page);
 307                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 308                         lock_page(page);
 309                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 310                                 f2fs_bug_on(sbi, 1);
 311                 }
 312                 /*
 313                  * In the NODE case, we lose next block address chain. So, we
 314                  * need to do checkpoint in f2fs_sync_file.
 315                  */
 316                 if (type == NODE)
 317                         set_sbi_flag(sbi, SBI_NEED_CP);
 318         }
 319 submit_io:
 320         if (is_read_io(bio_op(bio)))
 321                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 322         else
 323                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 324         submit_bio(bio);
 325 }
 326
 327 static void __submit_merged_bio(struct f2fs_bio_info *io)
 328 {
 329         struct f2fs_io_info *fio = &io->fio;
 330
 331         if (!io->bio)
 332                 return;
 333
 334         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 335
 336         if (is_read_io(fio->op))
 337                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 338         else
 339                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 340
 341         __submit_bio(io->sbi, io->bio, fio->type);
 342         io->bio = NULL;
 343 }
 344
 345 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 346                                                 struct page *page, nid_t ino)
 347 {
 348         struct bio_vec *bvec;
 349         struct page *target;
 350         int i;
 351
 352         if (!io->bio)
 353                 return false;
 354
 355         if (!inode && !page && !ino)
 356                 return true;
 357
 358         bio_for_each_segment_all(bvec, io->bio, i) {
 359
 360                 if (bvec->bv_page->mapping)
 361                         target = bvec->bv_page;
 362                 else
 363                         target = fscrypt_control_page(bvec->bv_page);
 364
 365                 if (inode && inode == target->mapping->host)
 366                         return true;
 367                 if (page && page == target)
 368                         return true;
 369                 if (ino && ino == ino_of_node(target))
 370                         return true;
 371         }
 372
 373         return false;
 374 }
 375
 376 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 377                                 enum page_type type, enum temp_type temp)
 378 {
 379         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 380         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 381
 382         down_write(&io->io_rwsem);
 383
 384         /* change META to META_FLUSH in the checkpoint procedure */
 385         if (type >= META_FLUSH) {
 386                 io->fio.type = META_FLUSH;
 387                 io->fio.op = REQ_OP_WRITE;
 388                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 389                 if (!test_opt(sbi, NOBARRIER))
 390                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 391         }
 392         __submit_merged_bio(io);
 393         up_write(&io->io_rwsem);
 394 }
 395
 396 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 397                                 struct inode *inode, struct page *page,
 398                                 nid_t ino, enum page_type type, bool force)
 399 {
 400         enum temp_type temp;
 401         bool ret = true;
 402
 403         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 404                 if (!force)     {
 405                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 406                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 407
 408                         down_read(&io->io_rwsem);
 409                         ret = __has_merged_page(io, inode, page, ino);
 410                         up_read(&io->io_rwsem);
 411                 }
 412                 if (ret)
 413                         __f2fs_submit_merged_write(sbi, type, temp);
 414
 415                 /* TODO: use HOT temp only for meta pages now. */
 416                 if (type >= META)
 417                         break;
 418         }
 419 }
 420
 421 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 422 {
 423         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 424 }
 425
 426 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 427                                 struct inode *inode, struct page *page,
 428                                 nid_t ino, enum page_type type)
 429 {
 430         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 431 }
 432
 433 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 434 {
 435         f2fs_submit_merged_write(sbi, DATA);
 436         f2fs_submit_merged_write(sbi, NODE);
 437         f2fs_submit_merged_write(sbi, META);
 438 }
 439
 440 /*
 441  * Fill the locked page with data located in the block address.
 442  * A caller needs to unlock the page on failure.
 443  */
 444 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 445 {
 446         struct bio *bio;
 447         struct page *page = fio->encrypted_page ?
 448                         fio->encrypted_page : fio->page;
 449
 450         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 451                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 452                 return -EFAULT;
 453
 454         trace_f2fs_submit_page_bio(page, fio);
 455         f2fs_trace_ios(fio, 0);
 456
 457         /* Allocate a new bio */
 458         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 459                                 1, is_read_io(fio->op), fio->type, fio->temp);
 460
 461         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 462                 bio_put(bio);
 463                 return -EFAULT;
 464         }
 465
 466         if (fio->io_wbc && !is_read_io(fio->op))
 467                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 468
 469         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 470
 471         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 472                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 473
 474         __submit_bio(fio->sbi, bio, fio->type);
 475         return 0;
 476 }
 477
 478 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 479 {
 480         struct f2fs_sb_info *sbi = fio->sbi;
 481         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 482         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 483         struct page *bio_page;
 484
 485         f2fs_bug_on(sbi, is_read_io(fio->op));
 486
 487         down_write(&io->io_rwsem);
 488 next:
 489         if (fio->in_list) {
 490                 spin_lock(&io->io_lock);
 491                 if (list_empty(&io->io_list)) {
 492                         spin_unlock(&io->io_lock);
 493                         goto out;
 494                 }
 495                 fio = list_first_entry(&io->io_list,
 496                                                 struct f2fs_io_info, list);
 497                 list_del(&fio->list);
 498                 spin_unlock(&io->io_lock);
 499         }
 500
 501         if (__is_valid_data_blkaddr(fio->old_blkaddr))
 502                 verify_block_addr(fio, fio->old_blkaddr);
 503         verify_block_addr(fio, fio->new_blkaddr);
 504
 505         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 506
 507         /* set submitted = true as a return value */
 508         fio->submitted = true;
 509
 510         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 511
 512         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 513             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 514                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 515                 __submit_merged_bio(io);
 516 alloc_new:
 517         if (io->bio == NULL) {
 518                 if ((fio->type == DATA || fio->type == NODE) &&
 519                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 520                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 521                         fio->retry = true;
 522                         goto skip;
 523                 }
 524                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 525                                                 BIO_MAX_PAGES, false,
 526                                                 fio->type, fio->temp);
 527                 io->fio = *fio;
 528         }
 529
 530         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 531                 __submit_merged_bio(io);
 532                 goto alloc_new;
 533         }
 534
 535         if (fio->io_wbc)
 536                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 537
 538         io->last_block_in_bio = fio->new_blkaddr;
 539         f2fs_trace_ios(fio, 0);
 540
 541         trace_f2fs_submit_page_write(fio->page, fio);
 542 skip:
 543         if (fio->in_list)
 544                 goto next;
 545 out:
 546         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 547                                 f2fs_is_checkpoint_ready(sbi))
 548                 __submit_merged_bio(io);
 549         up_write(&io->io_rwsem);
 550 }
 551
 552 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 553                                         unsigned nr_pages, unsigned op_flag)
 554 {
 555         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 556         struct bio *bio;
 557         struct bio_post_read_ctx *ctx;
 558         unsigned int post_read_steps = 0;
 559
 560         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
 561                 return ERR_PTR(-EFAULT);
 562
 563         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 564         if (!bio)
 565                 return ERR_PTR(-ENOMEM);
 566         f2fs_target_device(sbi, blkaddr, bio);
 567         bio->bi_end_io = f2fs_read_end_io;
 568         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 569
 570         if (f2fs_encrypted_file(inode))
 571                 post_read_steps |= 1 << STEP_DECRYPT;
 572         if (post_read_steps) {
 573                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 574                 if (!ctx) {
 575                         bio_put(bio);
 576                         return ERR_PTR(-ENOMEM);
 577                 }
 578                 ctx->bio = bio;
 579                 ctx->enabled_steps = post_read_steps;
 580                 bio->bi_private = ctx;
 581         }
 582
 583         return bio;
 584 }
 585
 586 /* This can handle encryption stuffs */
 587 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 588                                                         block_t blkaddr)
 589 {
 590         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 591
 592         if (IS_ERR(bio))
 593                 return PTR_ERR(bio);
 594
 595         /* wait for GCed page writeback via META_MAPPING */
 596         f2fs_wait_on_block_writeback(inode, blkaddr);
 597
 598         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 599                 bio_put(bio);
 600                 return -EFAULT;
 601         }
 602         ClearPageError(page);
 603         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
 604         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 605         return 0;
 606 }
 607
 608 static void __set_data_blkaddr(struct dnode_of_data *dn)
 609 {
 610         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 611         __le32 *addr_array;
 612         int base = 0;
 613
 614         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 615                 base = get_extra_isize(dn->inode);
 616
 617         /* Get physical address of data block */
 618         addr_array = blkaddr_in_node(rn);
 619         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 620 }
 621
 622 /*
 623  * Lock ordering for the change of data block address:
 624  * ->data_page
 625  *  ->node_page
 626  *    update block addresses in the node page
 627  */
 628 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 629 {
 630         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 631         __set_data_blkaddr(dn);
 632         if (set_page_dirty(dn->node_page))
 633                 dn->node_changed = true;
 634 }
 635
 636 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 637 {
 638         dn->data_blkaddr = blkaddr;
 639         f2fs_set_data_blkaddr(dn);
 640         f2fs_update_extent_cache(dn);
 641 }
 642
 643 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 644 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 645 {
 646         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 647         int err;
 648
 649         if (!count)
 650                 return 0;
 651
 652         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 653                 return -EPERM;
 654         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 655                 return err;
 656
 657         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 658                                                 dn->ofs_in_node, count);
 659
 660         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 661
 662         for (; count > 0; dn->ofs_in_node++) {
 663                 block_t blkaddr = datablock_addr(dn->inode,
 664                                         dn->node_page, dn->ofs_in_node);
 665                 if (blkaddr == NULL_ADDR) {
 666                         dn->data_blkaddr = NEW_ADDR;
 667                         __set_data_blkaddr(dn);
 668                         count--;
 669                 }
 670         }
 671
 672         if (set_page_dirty(dn->node_page))
 673                 dn->node_changed = true;
 674         return 0;
 675 }
 676
 677 /* Should keep dn->ofs_in_node unchanged */
 678 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 679 {
 680         unsigned int ofs_in_node = dn->ofs_in_node;
 681         int ret;
 682
 683         ret = f2fs_reserve_new_blocks(dn, 1);
 684         dn->ofs_in_node = ofs_in_node;
 685         return ret;
 686 }
 687
 688 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 689 {
 690         bool need_put = dn->inode_page ? false : true;
 691         int err;
 692
 693         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 694         if (err)
 695                 return err;
 696
 697         if (dn->data_blkaddr == NULL_ADDR)
 698                 err = f2fs_reserve_new_block(dn);
 699         if (err || need_put)
 700                 f2fs_put_dnode(dn);
 701         return err;
 702 }
 703
 704 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 705 {
 706         struct extent_info ei  = {0,0,0};
 707         struct inode *inode = dn->inode;
 708
 709         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 710                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 711                 return 0;
 712         }
 713
 714         return f2fs_reserve_block(dn, index);
 715 }
 716
 717 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 718                                                 int op_flags, bool for_write)
 719 {
 720         struct address_space *mapping = inode->i_mapping;
 721         struct dnode_of_data dn;
 722         struct page *page;
 723         struct extent_info ei = {0,0,0};
 724         int err;
 725
 726         page = f2fs_grab_cache_page(mapping, index, for_write);
 727         if (!page)
 728                 return ERR_PTR(-ENOMEM);
 729
 730         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 731                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 732                 goto got_it;
 733         }
 734
 735         set_new_dnode(&dn, inode, NULL, NULL, 0);
 736         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 737         if (err)
 738                 goto put_err;
 739         f2fs_put_dnode(&dn);
 740
 741         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 742                 err = -ENOENT;
 743                 goto put_err;
 744         }
 745 got_it:
 746         if (PageUptodate(page)) {
 747                 unlock_page(page);
 748                 return page;
 749         }
 750
 751         /*
 752          * A new dentry page is allocated but not able to be written, since its
 753          * new inode page couldn't be allocated due to -ENOSPC.
 754          * In such the case, its blkaddr can be remained as NEW_ADDR.
 755          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 756          * f2fs_init_inode_metadata.
 757          */
 758         if (dn.data_blkaddr == NEW_ADDR) {
 759                 zero_user_segment(page, 0, PAGE_SIZE);
 760                 if (!PageUptodate(page))
 761                         SetPageUptodate(page);
 762                 unlock_page(page);
 763                 return page;
 764         }
 765
 766         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 767         if (err)
 768                 goto put_err;
 769         return page;
 770
 771 put_err:
 772         f2fs_put_page(page, 1);
 773         return ERR_PTR(err);
 774 }
 775
 776 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 777 {
 778         struct address_space *mapping = inode->i_mapping;
 779         struct page *page;
 780
 781         page = find_get_page(mapping, index);
 782         if (page && PageUptodate(page))
 783                 return page;
 784         f2fs_put_page(page, 0);
 785
 786         page = f2fs_get_read_data_page(inode, index, 0, false);
 787         if (IS_ERR(page))
 788                 return page;
 789
 790         if (PageUptodate(page))
 791                 return page;
 792
 793         wait_on_page_locked(page);
 794         if (unlikely(!PageUptodate(page))) {
 795                 f2fs_put_page(page, 0);
 796                 return ERR_PTR(-EIO);
 797         }
 798         return page;
 799 }
 800
 801 /*
 802  * If it tries to access a hole, return an error.
 803  * Because, the callers, functions in dir.c and GC, should be able to know
 804  * whether this page exists or not.
 805  */
 806 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 807                                                         bool for_write)
 808 {
 809         struct address_space *mapping = inode->i_mapping;
 810         struct page *page;
 811 repeat:
 812         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 813         if (IS_ERR(page))
 814                 return page;
 815
 816         /* wait for read completion */
 817         lock_page(page);
 818         if (unlikely(page->mapping != mapping)) {
 819                 f2fs_put_page(page, 1);
 820                 goto repeat;
 821         }
 822         if (unlikely(!PageUptodate(page))) {
 823                 f2fs_put_page(page, 1);
 824                 return ERR_PTR(-EIO);
 825         }
 826         return page;
 827 }
 828
 829 /*
 830  * Caller ensures that this data page is never allocated.
 831  * A new zero-filled data page is allocated in the page cache.
 832  *
 833  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 834  * f2fs_unlock_op().
 835  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 836  * ipage should be released by this function.
 837  */
 838 struct page *f2fs_get_new_data_page(struct inode *inode,
 839                 struct page *ipage, pgoff_t index, bool new_i_size)
 840 {
 841         struct address_space *mapping = inode->i_mapping;
 842         struct page *page;
 843         struct dnode_of_data dn;
 844         int err;
 845
 846         page = f2fs_grab_cache_page(mapping, index, true);
 847         if (!page) {
 848                 /*
 849                  * before exiting, we should make sure ipage will be released
 850                  * if any error occur.
 851                  */
 852                 f2fs_put_page(ipage, 1);
 853                 return ERR_PTR(-ENOMEM);
 854         }
 855
 856         set_new_dnode(&dn, inode, ipage, NULL, 0);
 857         err = f2fs_reserve_block(&dn, index);
 858         if (err) {
 859                 f2fs_put_page(page, 1);
 860                 return ERR_PTR(err);
 861         }
 862         if (!ipage)
 863                 f2fs_put_dnode(&dn);
 864
 865         if (PageUptodate(page))
 866                 goto got_it;
 867
 868         if (dn.data_blkaddr == NEW_ADDR) {
 869                 zero_user_segment(page, 0, PAGE_SIZE);
 870                 if (!PageUptodate(page))
 871                         SetPageUptodate(page);
 872         } else {
 873                 f2fs_put_page(page, 1);
 874
 875                 /* if ipage exists, blkaddr should be NEW_ADDR */
 876                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 877                 page = f2fs_get_lock_data_page(inode, index, true);
 878                 if (IS_ERR(page))
 879                         return page;
 880         }
 881 got_it:
 882         if (new_i_size && i_size_read(inode) <
 883                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 884                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 885         return page;
 886 }
 887
 888 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 889 {
 890         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 891         struct f2fs_summary sum;
 892         struct node_info ni;
 893         block_t old_blkaddr;
 894         blkcnt_t count = 1;
 895         int err;
 896
 897         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 898                 return -EPERM;
 899
 900         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 901         if (err)
 902                 return err;
 903
 904         dn->data_blkaddr = datablock_addr(dn->inode,
 905                                 dn->node_page, dn->ofs_in_node);
 906         if (dn->data_blkaddr != NULL_ADDR)
 907                 goto alloc;
 908
 909         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 910                 return err;
 911
 912 alloc:
 913         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 914         old_blkaddr = dn->data_blkaddr;
 915         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 916                                         &sum, seg_type, NULL, false);
 917         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 918                 invalidate_mapping_pages(META_MAPPING(sbi),
 919                                         old_blkaddr, old_blkaddr);
 920         f2fs_set_data_blkaddr(dn);
 921
 922         /*
 923          * i_size will be updated by direct_IO. Otherwise, we'll get stale
 924          * data from unwritten block via dio_read.
 925          */
 926         return 0;
 927 }
 928
 929 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 930 {
 931         struct inode *inode = file_inode(iocb->ki_filp);
 932         struct f2fs_map_blocks map;
 933         int flag;
 934         int err = 0;
 935         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 936
 937         /* convert inline data for Direct I/O*/
 938         if (direct_io) {
 939                 err = f2fs_convert_inline_inode(inode);
 940                 if (err)
 941                         return err;
 942         }
 943
 944         if (direct_io && allow_outplace_dio(inode, iocb, from))
 945                 return 0;
 946
 947         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 948                 return 0;
 949
 950         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 951         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 952         if (map.m_len > map.m_lblk)
 953                 map.m_len -= map.m_lblk;
 954         else
 955                 map.m_len = 0;
 956
 957         map.m_next_pgofs = NULL;
 958         map.m_next_extent = NULL;
 959         map.m_seg_type = NO_CHECK_TYPE;
 960         map.m_may_create = true;
 961
 962         if (direct_io) {
 963                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 964                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
 965                                         F2FS_GET_BLOCK_PRE_AIO :
 966                                         F2FS_GET_BLOCK_PRE_DIO;
 967                 goto map_blocks;
 968         }
 969         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 970                 err = f2fs_convert_inline_inode(inode);
 971                 if (err)
 972                         return err;
 973         }
 974         if (f2fs_has_inline_data(inode))
 975                 return err;
 976
 977         flag = F2FS_GET_BLOCK_PRE_AIO;
 978
 979 map_blocks:
 980         err = f2fs_map_blocks(inode, &map, 1, flag);
 981         if (map.m_len > 0 && err == -ENOSPC) {
 982                 if (!direct_io)
 983                         set_inode_flag(inode, FI_NO_PREALLOC);
 984                 err = 0;
 985         }
 986         return err;
 987 }
 988
 989 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 990 {
 991         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 992                 if (lock)
 993                         down_read(&sbi->node_change);
 994                 else
 995                         up_read(&sbi->node_change);
 996         } else {
 997                 if (lock)
 998                         f2fs_lock_op(sbi);
 999                 else
1000                         f2fs_unlock_op(sbi);
1001         }
1002 }
1003
1004 /*
1005  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1006  * f2fs_map_blocks structure.
1007  * If original data blocks are allocated, then give them to blockdev.
1008  * Otherwise,
1009  *     a. preallocate requested block addresses
1010  *     b. do not use extent cache for better performance
1011  *     c. give the block addresses to blockdev
1012  */
1013 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1014                                                 int create, int flag)
1015 {
1016         unsigned int maxblocks = map->m_len;
1017         struct dnode_of_data dn;
1018         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1019         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1020         pgoff_t pgofs, end_offset, end;
1021         int err = 0, ofs = 1;
1022         unsigned int ofs_in_node, last_ofs_in_node;
1023         blkcnt_t prealloc;
1024         struct extent_info ei = {0,0,0};
1025         block_t blkaddr;
1026         unsigned int start_pgofs;
1027
1028         if (!maxblocks)
1029                 return 0;
1030
1031         map->m_len = 0;
1032         map->m_flags = 0;
1033
1034         /* it only supports block size == page size */
1035         pgofs = (pgoff_t)map->m_lblk;
1036         end = pgofs + maxblocks;
1037
1038         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1039                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1040                                                         map->m_may_create)
1041                         goto next_dnode;
1042
1043                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1044                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1045                 map->m_flags = F2FS_MAP_MAPPED;
1046                 if (map->m_next_extent)
1047                         *map->m_next_extent = pgofs + map->m_len;
1048
1049                 /* for hardware encryption, but to avoid potential issue in future */
1050                 if (flag == F2FS_GET_BLOCK_DIO)
1051                         f2fs_wait_on_block_writeback_range(inode,
1052                                                 map->m_pblk, map->m_len);
1053                 goto out;
1054         }
1055
1056 next_dnode:
1057         if (map->m_may_create)
1058                 __do_map_lock(sbi, flag, true);
1059
1060         /* When reading holes, we need its node page */
1061         set_new_dnode(&dn, inode, NULL, NULL, 0);
1062         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1063         if (err) {
1064                 if (flag == F2FS_GET_BLOCK_BMAP)
1065                         map->m_pblk = 0;
1066                 if (err == -ENOENT) {
1067                         err = 0;
1068                         if (map->m_next_pgofs)
1069                                 *map->m_next_pgofs =
1070                                         f2fs_get_next_page_offset(&dn, pgofs);
1071                         if (map->m_next_extent)
1072                                 *map->m_next_extent =
1073                                         f2fs_get_next_page_offset(&dn, pgofs);
1074                 }
1075                 goto unlock_out;
1076         }
1077
1078         start_pgofs = pgofs;
1079         prealloc = 0;
1080         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1081         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1082
1083 next_block:
1084         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1085
1086         if (__is_valid_data_blkaddr(blkaddr) &&
1087                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1088                 err = -EFAULT;
1089                 goto sync_out;
1090         }
1091
1092         if (is_valid_data_blkaddr(sbi, blkaddr)) {
1093                 /* use out-place-update for driect IO under LFS mode */
1094                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1095                                                         map->m_may_create) {
1096                         err = __allocate_data_block(&dn, map->m_seg_type);
1097                         if (!err) {
1098                                 blkaddr = dn.data_blkaddr;
1099                                 set_inode_flag(inode, FI_APPEND_WRITE);
1100                         }
1101                 }
1102         } else {
1103                 if (create) {
1104                         if (unlikely(f2fs_cp_error(sbi))) {
1105                                 err = -EIO;
1106                                 goto sync_out;
1107                         }
1108                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1109                                 if (blkaddr == NULL_ADDR) {
1110                                         prealloc++;
1111                                         last_ofs_in_node = dn.ofs_in_node;
1112                                 }
1113                         } else {
1114                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1115                                         flag != F2FS_GET_BLOCK_DIO);
1116                                 err = __allocate_data_block(&dn,
1117                                                         map->m_seg_type);
1118                                 if (!err)
1119                                         set_inode_flag(inode, FI_APPEND_WRITE);
1120                         }
1121                         if (err)
1122                                 goto sync_out;
1123                         map->m_flags |= F2FS_MAP_NEW;
1124                         blkaddr = dn.data_blkaddr;
1125                 } else {
1126                         if (flag == F2FS_GET_BLOCK_BMAP) {
1127                                 map->m_pblk = 0;
1128                                 goto sync_out;
1129                         }
1130                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1131                                 goto sync_out;
1132                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1133                                                 blkaddr == NULL_ADDR) {
1134                                 if (map->m_next_pgofs)
1135                                         *map->m_next_pgofs = pgofs + 1;
1136                                 goto sync_out;
1137                         }
1138                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1139                                 /* for defragment case */
1140                                 if (map->m_next_pgofs)
1141                                         *map->m_next_pgofs = pgofs + 1;
1142                                 goto sync_out;
1143                         }
1144                 }
1145         }
1146
1147         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1148                 goto skip;
1149
1150         if (map->m_len == 0) {
1151                 /* preallocated unwritten block should be mapped for fiemap. */
1152                 if (blkaddr == NEW_ADDR)
1153                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1154                 map->m_flags |= F2FS_MAP_MAPPED;
1155
1156                 map->m_pblk = blkaddr;
1157                 map->m_len = 1;
1158         } else if ((map->m_pblk != NEW_ADDR &&
1159                         blkaddr == (map->m_pblk + ofs)) ||
1160                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1161                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1162                 ofs++;
1163                 map->m_len++;
1164         } else {
1165                 goto sync_out;
1166         }
1167
1168 skip:
1169         dn.ofs_in_node++;
1170         pgofs++;
1171
1172         /* preallocate blocks in batch for one dnode page */
1173         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1174                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1175
1176                 dn.ofs_in_node = ofs_in_node;
1177                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1178                 if (err)
1179                         goto sync_out;
1180
1181                 map->m_len += dn.ofs_in_node - ofs_in_node;
1182                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1183                         err = -ENOSPC;
1184                         goto sync_out;
1185                 }
1186                 dn.ofs_in_node = end_offset;
1187         }
1188
1189         if (pgofs >= end)
1190                 goto sync_out;
1191         else if (dn.ofs_in_node < end_offset)
1192                 goto next_block;
1193
1194         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1195                 if (map->m_flags & F2FS_MAP_MAPPED) {
1196                         unsigned int ofs = start_pgofs - map->m_lblk;
1197
1198                         f2fs_update_extent_cache_range(&dn,
1199                                 start_pgofs, map->m_pblk + ofs,
1200                                 map->m_len - ofs);
1201                 }
1202         }
1203
1204         f2fs_put_dnode(&dn);
1205
1206         if (map->m_may_create) {
1207                 __do_map_lock(sbi, flag, false);
1208                 f2fs_balance_fs(sbi, dn.node_changed);
1209         }
1210         goto next_dnode;
1211
1212 sync_out:
1213
1214         /* for hardware encryption, but to avoid potential issue in future */
1215         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1216                 f2fs_wait_on_block_writeback_range(inode,
1217                                                 map->m_pblk, map->m_len);
1218
1219         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1220                 if (map->m_flags & F2FS_MAP_MAPPED) {
1221                         unsigned int ofs = start_pgofs - map->m_lblk;
1222
1223                         f2fs_update_extent_cache_range(&dn,
1224                                 start_pgofs, map->m_pblk + ofs,
1225                                 map->m_len - ofs);
1226                 }
1227                 if (map->m_next_extent)
1228                         *map->m_next_extent = pgofs + 1;
1229         }
1230         f2fs_put_dnode(&dn);
1231 unlock_out:
1232         if (map->m_may_create) {
1233                 __do_map_lock(sbi, flag, false);
1234                 f2fs_balance_fs(sbi, dn.node_changed);
1235         }
1236 out:
1237         trace_f2fs_map_blocks(inode, map, err);
1238         return err;
1239 }
1240
1241 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1242 {
1243         struct f2fs_map_blocks map;
1244         block_t last_lblk;
1245         int err;
1246
1247         if (pos + len > i_size_read(inode))
1248                 return false;
1249
1250         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1251         map.m_next_pgofs = NULL;
1252         map.m_next_extent = NULL;
1253         map.m_seg_type = NO_CHECK_TYPE;
1254         map.m_may_create = false;
1255         last_lblk = F2FS_BLK_ALIGN(pos + len);
1256
1257         while (map.m_lblk < last_lblk) {
1258                 map.m_len = last_lblk - map.m_lblk;
1259                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1260                 if (err || map.m_len == 0)
1261                         return false;
1262                 map.m_lblk += map.m_len;
1263         }
1264         return true;
1265 }
1266
1267 static int __get_data_block(struct inode *inode, sector_t iblock,
1268                         struct buffer_head *bh, int create, int flag,
1269                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1270 {
1271         struct f2fs_map_blocks map;
1272         int err;
1273
1274         map.m_lblk = iblock;
1275         map.m_len = bh->b_size >> inode->i_blkbits;
1276         map.m_next_pgofs = next_pgofs;
1277         map.m_next_extent = NULL;
1278         map.m_seg_type = seg_type;
1279         map.m_may_create = may_write;
1280
1281         err = f2fs_map_blocks(inode, &map, create, flag);
1282         if (!err) {
1283                 map_bh(bh, inode->i_sb, map.m_pblk);
1284                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1285                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1286         }
1287         return err;
1288 }
1289
1290 static int get_data_block(struct inode *inode, sector_t iblock,
1291                         struct buffer_head *bh_result, int create, int flag,
1292                         pgoff_t *next_pgofs)
1293 {
1294         return __get_data_block(inode, iblock, bh_result, create,
1295                                                         flag, next_pgofs,
1296                                                         NO_CHECK_TYPE, create);
1297 }
1298
1299 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1300                         struct buffer_head *bh_result, int create)
1301 {
1302         return __get_data_block(inode, iblock, bh_result, create,
1303                                 F2FS_GET_BLOCK_DIO, NULL,
1304                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1305                                 true);
1306 }
1307
1308 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1309                         struct buffer_head *bh_result, int create)
1310 {
1311         return __get_data_block(inode, iblock, bh_result, create,
1312                                 F2FS_GET_BLOCK_DIO, NULL,
1313                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1314                                 false);
1315 }
1316
1317 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1318                         struct buffer_head *bh_result, int create)
1319 {
1320         /* Block number less than F2FS MAX BLOCKS */
1321         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1322                 return -EFBIG;
1323
1324         return __get_data_block(inode, iblock, bh_result, create,
1325                                                 F2FS_GET_BLOCK_BMAP, NULL,
1326                                                 NO_CHECK_TYPE, create);
1327 }
1328
1329 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1330 {
1331         return (offset >> inode->i_blkbits);
1332 }
1333
1334 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1335 {
1336         return (blk << inode->i_blkbits);
1337 }
1338
1339 static int f2fs_xattr_fiemap(struct inode *inode,
1340                                 struct fiemap_extent_info *fieinfo)
1341 {
1342         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1343         struct page *page;
1344         struct node_info ni;
1345         __u64 phys = 0, len;
1346         __u32 flags;
1347         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1348         int err = 0;
1349
1350         if (f2fs_has_inline_xattr(inode)) {
1351                 int offset;
1352
1353                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1354                                                 inode->i_ino, false);
1355                 if (!page)
1356                         return -ENOMEM;
1357
1358                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1359                 if (err) {
1360                         f2fs_put_page(page, 1);
1361                         return err;
1362                 }
1363
1364                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1365                 offset = offsetof(struct f2fs_inode, i_addr) +
1366                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1367                                         get_inline_xattr_addrs(inode));
1368
1369                 phys += offset;
1370                 len = inline_xattr_size(inode);
1371
1372                 f2fs_put_page(page, 1);
1373
1374                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1375
1376                 if (!xnid)
1377                         flags |= FIEMAP_EXTENT_LAST;
1378
1379                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1380                 if (err || err == 1)
1381                         return err;
1382         }
1383
1384         if (xnid) {
1385                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1386                 if (!page)
1387                         return -ENOMEM;
1388
1389                 err = f2fs_get_node_info(sbi, xnid, &ni);
1390                 if (err) {
1391                         f2fs_put_page(page, 1);
1392                         return err;
1393                 }
1394
1395                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1396                 len = inode->i_sb->s_blocksize;
1397
1398                 f2fs_put_page(page, 1);
1399
1400                 flags = FIEMAP_EXTENT_LAST;
1401         }
1402
1403         if (phys)
1404                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1405
1406         return (err < 0 ? err : 0);
1407 }
1408
1409 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1410                 u64 start, u64 len)
1411 {
1412         struct buffer_head map_bh;
1413         sector_t start_blk, last_blk;
1414         pgoff_t next_pgofs;
1415         u64 logical = 0, phys = 0, size = 0;
1416         u32 flags = 0;
1417         int ret = 0;
1418
1419         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1420                 ret = f2fs_precache_extents(inode);
1421                 if (ret)
1422                         return ret;
1423         }
1424
1425         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1426         if (ret)
1427                 return ret;
1428
1429         inode_lock(inode);
1430
1431         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1432                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1433                 goto out;
1434         }
1435
1436         if (f2fs_has_inline_data(inode)) {
1437                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1438                 if (ret != -EAGAIN)
1439                         goto out;
1440         }
1441
1442         if (logical_to_blk(inode, len) == 0)
1443                 len = blk_to_logical(inode, 1);
1444
1445         start_blk = logical_to_blk(inode, start);
1446         last_blk = logical_to_blk(inode, start + len - 1);
1447
1448 next:
1449         memset(&map_bh, 0, sizeof(struct buffer_head));
1450         map_bh.b_size = len;
1451
1452         ret = get_data_block(inode, start_blk, &map_bh, 0,
1453                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1454         if (ret)
1455                 goto out;
1456
1457         /* HOLE */
1458         if (!buffer_mapped(&map_bh)) {
1459                 start_blk = next_pgofs;
1460
1461                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1462                                         F2FS_I_SB(inode)->max_file_blocks))
1463                         goto prep_next;
1464
1465                 flags |= FIEMAP_EXTENT_LAST;
1466         }
1467
1468         if (size) {
1469                 if (f2fs_encrypted_inode(inode))
1470                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1471
1472                 ret = fiemap_fill_next_extent(fieinfo, logical,
1473                                 phys, size, flags);
1474         }
1475
1476         if (start_blk > last_blk || ret)
1477                 goto out;
1478
1479         logical = blk_to_logical(inode, start_blk);
1480         phys = blk_to_logical(inode, map_bh.b_blocknr);
1481         size = map_bh.b_size;
1482         flags = 0;
1483         if (buffer_unwritten(&map_bh))
1484                 flags = FIEMAP_EXTENT_UNWRITTEN;
1485
1486         start_blk += logical_to_blk(inode, size);
1487
1488 prep_next:
1489         cond_resched();
1490         if (fatal_signal_pending(current))
1491                 ret = -EINTR;
1492         else
1493                 goto next;
1494 out:
1495         if (ret == 1)
1496                 ret = 0;
1497
1498         inode_unlock(inode);
1499         return ret;
1500 }
1501
1502 /*
1503  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1504  * Major change was from block_size == page_size in f2fs by default.
1505  *
1506  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1507  * this function ever deviates from doing just read-ahead, it should either
1508  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1509  * from read-ahead.
1510  */
1511 static int f2fs_mpage_readpages(struct address_space *mapping,
1512                         struct list_head *pages, struct page *page,
1513                         unsigned nr_pages, bool is_readahead)
1514 {
1515         struct bio *bio = NULL;
1516         sector_t last_block_in_bio = 0;
1517         struct inode *inode = mapping->host;
1518         const unsigned blkbits = inode->i_blkbits;
1519         const unsigned blocksize = 1 << blkbits;
1520         sector_t block_in_file;
1521         sector_t last_block;
1522         sector_t last_block_in_file;
1523         sector_t block_nr;
1524         struct f2fs_map_blocks map;
1525
1526         map.m_pblk = 0;
1527         map.m_lblk = 0;
1528         map.m_len = 0;
1529         map.m_flags = 0;
1530         map.m_next_pgofs = NULL;
1531         map.m_next_extent = NULL;
1532         map.m_seg_type = NO_CHECK_TYPE;
1533         map.m_may_create = false;
1534
1535         for (; nr_pages; nr_pages--) {
1536                 if (pages) {
1537                         page = list_last_entry(pages, struct page, lru);
1538
1539                         prefetchw(&page->flags);
1540                         list_del(&page->lru);
1541                         if (add_to_page_cache_lru(page, mapping,
1542                                                   page->index,
1543                                                   readahead_gfp_mask(mapping)))
1544                                 goto next_page;
1545                 }
1546
1547                 block_in_file = (sector_t)page->index;
1548                 last_block = block_in_file + nr_pages;
1549                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1550                                                                 blkbits;
1551                 if (last_block > last_block_in_file)
1552                         last_block = last_block_in_file;
1553
1554                 /*
1555                  * Map blocks using the previous result first.
1556                  */
1557                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1558                                 block_in_file > map.m_lblk &&
1559                                 block_in_file < (map.m_lblk + map.m_len))
1560                         goto got_it;
1561
1562                 /*
1563                  * Then do more f2fs_map_blocks() calls until we are
1564                  * done with this page.
1565                  */
1566                 map.m_flags = 0;
1567
1568                 if (block_in_file < last_block) {
1569                         map.m_lblk = block_in_file;
1570                         map.m_len = last_block - block_in_file;
1571
1572                         if (f2fs_map_blocks(inode, &map, 0,
1573                                                 F2FS_GET_BLOCK_DEFAULT))
1574                                 goto set_error_page;
1575                 }
1576 got_it:
1577                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1578                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1579                         SetPageMappedToDisk(page);
1580
1581                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1582                                 SetPageUptodate(page);
1583                                 goto confused;
1584                         }
1585
1586                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1587                                                                 DATA_GENERIC))
1588                                 goto set_error_page;
1589                 } else {
1590                         zero_user_segment(page, 0, PAGE_SIZE);
1591                         if (!PageUptodate(page))
1592                                 SetPageUptodate(page);
1593                         unlock_page(page);
1594                         goto next_page;
1595                 }
1596
1597                 /*
1598                  * This page will go to BIO.  Do we need to send this
1599                  * BIO off first?
1600                  */
1601                 if (bio && (last_block_in_bio != block_nr - 1 ||
1602                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1603 submit_and_realloc:
1604                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1605                         bio = NULL;
1606                 }
1607                 if (bio == NULL) {
1608                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1609                                         is_readahead ? REQ_RAHEAD : 0);
1610                         if (IS_ERR(bio)) {
1611                                 bio = NULL;
1612                                 goto set_error_page;
1613                         }
1614                 }
1615
1616                 /*
1617                  * If the page is under writeback, we need to wait for
1618                  * its completion to see the correct decrypted data.
1619                  */
1620                 f2fs_wait_on_block_writeback(inode, block_nr);
1621
1622                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1623                         goto submit_and_realloc;
1624
1625                 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1626                 ClearPageError(page);
1627                 last_block_in_bio = block_nr;
1628                 goto next_page;
1629 set_error_page:
1630                 SetPageError(page);
1631                 zero_user_segment(page, 0, PAGE_SIZE);
1632                 unlock_page(page);
1633                 goto next_page;
1634 confused:
1635                 if (bio) {
1636                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1637                         bio = NULL;
1638                 }
1639                 unlock_page(page);
1640 next_page:
1641                 if (pages)
1642                         put_page(page);
1643         }
1644         BUG_ON(pages && !list_empty(pages));
1645         if (bio)
1646                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1647         return 0;
1648 }
1649
1650 static int f2fs_read_data_page(struct file *file, struct page *page)
1651 {
1652         struct inode *inode = page->mapping->host;
1653         int ret = -EAGAIN;
1654
1655         trace_f2fs_readpage(page, DATA);
1656
1657         /* If the file has inline data, try to read it directly */
1658         if (f2fs_has_inline_data(inode))
1659                 ret = f2fs_read_inline_data(inode, page);
1660         if (ret == -EAGAIN)
1661                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1662         return ret;
1663 }
1664
1665 static int f2fs_read_data_pages(struct file *file,
1666                         struct address_space *mapping,
1667                         struct list_head *pages, unsigned nr_pages)
1668 {
1669         struct inode *inode = mapping->host;
1670         struct page *page = list_last_entry(pages, struct page, lru);
1671
1672         trace_f2fs_readpages(inode, page, nr_pages);
1673
1674         /* If the file has inline data, skip readpages */
1675         if (f2fs_has_inline_data(inode))
1676                 return 0;
1677
1678         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1679 }
1680
1681 static int encrypt_one_page(struct f2fs_io_info *fio)
1682 {
1683         struct inode *inode = fio->page->mapping->host;
1684         struct page *mpage;
1685         gfp_t gfp_flags = GFP_NOFS;
1686
1687         if (!f2fs_encrypted_file(inode))
1688                 return 0;
1689
1690         /* wait for GCed page writeback via META_MAPPING */
1691         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1692
1693 retry_encrypt:
1694         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1695                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1696         if (IS_ERR(fio->encrypted_page)) {
1697                 /* flush pending IOs and wait for a while in the ENOMEM case */
1698                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1699                         f2fs_flush_merged_writes(fio->sbi);
1700                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1701                         gfp_flags |= __GFP_NOFAIL;
1702                         goto retry_encrypt;
1703                 }
1704                 return PTR_ERR(fio->encrypted_page);
1705         }
1706
1707         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1708         if (mpage) {
1709                 if (PageUptodate(mpage))
1710                         memcpy(page_address(mpage),
1711                                 page_address(fio->encrypted_page), PAGE_SIZE);
1712                 f2fs_put_page(mpage, 1);
1713         }
1714         return 0;
1715 }
1716
1717 static inline bool check_inplace_update_policy(struct inode *inode,
1718                                 struct f2fs_io_info *fio)
1719 {
1720         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1721         unsigned int policy = SM_I(sbi)->ipu_policy;
1722
1723         if (policy & (0x1 << F2FS_IPU_FORCE))
1724                 return true;
1725         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1726                 return true;
1727         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1728                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1729                 return true;
1730         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1731                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1732                 return true;
1733
1734         /*
1735          * IPU for rewrite async pages
1736          */
1737         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1738                         fio && fio->op == REQ_OP_WRITE &&
1739                         !(fio->op_flags & REQ_SYNC) &&
1740                         !f2fs_encrypted_inode(inode))
1741                 return true;
1742
1743         /* this is only set during fdatasync */
1744         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1745                         is_inode_flag_set(inode, FI_NEED_IPU))
1746                 return true;
1747
1748         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1749                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1750                 return true;
1751
1752         return false;
1753 }
1754
1755 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1756 {
1757         if (f2fs_is_pinned_file(inode))
1758                 return true;
1759
1760         /* if this is cold file, we should overwrite to avoid fragmentation */
1761         if (file_is_cold(inode))
1762                 return true;
1763
1764         return check_inplace_update_policy(inode, fio);
1765 }
1766
1767 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1768 {
1769         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1770
1771         if (test_opt(sbi, LFS))
1772                 return true;
1773         if (S_ISDIR(inode->i_mode))
1774                 return true;
1775         if (IS_NOQUOTA(inode))
1776                 return true;
1777         if (f2fs_is_atomic_file(inode))
1778                 return true;
1779         if (fio) {
1780                 if (is_cold_data(fio->page))
1781                         return true;
1782                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1783                         return true;
1784                 if (unlikely(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 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1792 {
1793         struct inode *inode = fio->page->mapping->host;
1794
1795         if (f2fs_should_update_outplace(inode, fio))
1796                 return false;
1797
1798         return f2fs_should_update_inplace(inode, fio);
1799 }
1800
1801 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1802 {
1803         struct page *page = fio->page;
1804         struct inode *inode = page->mapping->host;
1805         struct dnode_of_data dn;
1806         struct extent_info ei = {0,0,0};
1807         struct node_info ni;
1808         bool ipu_force = false;
1809         int err = 0;
1810
1811         set_new_dnode(&dn, inode, NULL, NULL, 0);
1812         if (need_inplace_update(fio) &&
1813                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1814                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1815
1816                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1817                                                         DATA_GENERIC))
1818                         return -EFAULT;
1819
1820                 ipu_force = true;
1821                 fio->need_lock = LOCK_DONE;
1822                 goto got_it;
1823         }
1824
1825         /* Deadlock due to between page->lock and f2fs_lock_op */
1826         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1827                 return -EAGAIN;
1828
1829         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1830         if (err)
1831                 goto out;
1832
1833         fio->old_blkaddr = dn.data_blkaddr;
1834
1835         /* This page is already truncated */
1836         if (fio->old_blkaddr == NULL_ADDR) {
1837                 ClearPageUptodate(page);
1838                 clear_cold_data(page);
1839                 goto out_writepage;
1840         }
1841 got_it:
1842         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1843                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1844                                                         DATA_GENERIC)) {
1845                 err = -EFAULT;
1846                 goto out_writepage;
1847         }
1848         /*
1849          * If current allocation needs SSR,
1850          * it had better in-place writes for updated data.
1851          */
1852         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1853                                         need_inplace_update(fio))) {
1854                 err = encrypt_one_page(fio);
1855                 if (err)
1856                         goto out_writepage;
1857
1858                 set_page_writeback(page);
1859                 ClearPageError(page);
1860                 f2fs_put_dnode(&dn);
1861                 if (fio->need_lock == LOCK_REQ)
1862                         f2fs_unlock_op(fio->sbi);
1863                 err = f2fs_inplace_write_data(fio);
1864                 if (err && PageWriteback(page))
1865                         end_page_writeback(page);
1866                 trace_f2fs_do_write_data_page(fio->page, IPU);
1867                 set_inode_flag(inode, FI_UPDATE_WRITE);
1868                 return err;
1869         }
1870
1871         if (fio->need_lock == LOCK_RETRY) {
1872                 if (!f2fs_trylock_op(fio->sbi)) {
1873                         err = -EAGAIN;
1874                         goto out_writepage;
1875                 }
1876                 fio->need_lock = LOCK_REQ;
1877         }
1878
1879         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1880         if (err)
1881                 goto out_writepage;
1882
1883         fio->version = ni.version;
1884
1885         err = encrypt_one_page(fio);
1886         if (err)
1887                 goto out_writepage;
1888
1889         set_page_writeback(page);
1890         ClearPageError(page);
1891
1892         /* LFS mode write path */
1893         f2fs_outplace_write_data(&dn, fio);
1894         trace_f2fs_do_write_data_page(page, OPU);
1895         set_inode_flag(inode, FI_APPEND_WRITE);
1896         if (page->index == 0)
1897                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1898 out_writepage:
1899         f2fs_put_dnode(&dn);
1900 out:
1901         if (fio->need_lock == LOCK_REQ)
1902                 f2fs_unlock_op(fio->sbi);
1903         return err;
1904 }
1905
1906 static int __write_data_page(struct page *page, bool *submitted,
1907                                 struct writeback_control *wbc,
1908                                 enum iostat_type io_type)
1909 {
1910         struct inode *inode = page->mapping->host;
1911         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1912         loff_t i_size = i_size_read(inode);
1913         const pgoff_t end_index = ((unsigned long long) i_size)
1914                                                         >> PAGE_SHIFT;
1915         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1916         unsigned offset = 0;
1917         bool need_balance_fs = false;
1918         int err = 0;
1919         struct f2fs_io_info fio = {
1920                 .sbi = sbi,
1921                 .ino = inode->i_ino,
1922                 .type = DATA,
1923                 .op = REQ_OP_WRITE,
1924                 .op_flags = wbc_to_write_flags(wbc),
1925                 .old_blkaddr = NULL_ADDR,
1926                 .page = page,
1927                 .encrypted_page = NULL,
1928                 .submitted = false,
1929                 .need_lock = LOCK_RETRY,
1930                 .io_type = io_type,
1931                 .io_wbc = wbc,
1932         };
1933
1934         trace_f2fs_writepage(page, DATA);
1935
1936         /* we should bypass data pages to proceed the kworkder jobs */
1937         if (unlikely(f2fs_cp_error(sbi))) {
1938                 mapping_set_error(page->mapping, -EIO);
1939                 /*
1940                  * don't drop any dirty dentry pages for keeping lastest
1941                  * directory structure.
1942                  */
1943                 if (S_ISDIR(inode->i_mode))
1944                         goto redirty_out;
1945                 goto out;
1946         }
1947
1948         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1949                 goto redirty_out;
1950
1951         if (page->index < end_index)
1952                 goto write;
1953
1954         /*
1955          * If the offset is out-of-range of file size,
1956          * this page does not have to be written to disk.
1957          */
1958         offset = i_size & (PAGE_SIZE - 1);
1959         if ((page->index >= end_index + 1) || !offset)
1960                 goto out;
1961
1962         zero_user_segment(page, offset, PAGE_SIZE);
1963 write:
1964         if (f2fs_is_drop_cache(inode))
1965                 goto out;
1966         /* we should not write 0'th page having journal header */
1967         if (f2fs_is_volatile_file(inode) && (!page->index ||
1968                         (!wbc->for_reclaim &&
1969                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1970                 goto redirty_out;
1971
1972         /* Dentry blocks are controlled by checkpoint */
1973         if (S_ISDIR(inode->i_mode)) {
1974                 fio.need_lock = LOCK_DONE;
1975                 err = f2fs_do_write_data_page(&fio);
1976                 goto done;
1977         }
1978
1979         if (!wbc->for_reclaim)
1980                 need_balance_fs = true;
1981         else if (has_not_enough_free_secs(sbi, 0, 0))
1982                 goto redirty_out;
1983         else
1984                 set_inode_flag(inode, FI_HOT_DATA);
1985
1986         err = -EAGAIN;
1987         if (f2fs_has_inline_data(inode)) {
1988                 err = f2fs_write_inline_data(inode, page);
1989                 if (!err)
1990                         goto out;
1991         }
1992
1993         if (err == -EAGAIN) {
1994                 err = f2fs_do_write_data_page(&fio);
1995                 if (err == -EAGAIN) {
1996                         fio.need_lock = LOCK_REQ;
1997                         err = f2fs_do_write_data_page(&fio);
1998                 }
1999         }
2000
2001         if (err) {
2002                 file_set_keep_isize(inode);
2003         } else {
2004                 down_write(&F2FS_I(inode)->i_sem);
2005                 if (F2FS_I(inode)->last_disk_size < psize)
2006                         F2FS_I(inode)->last_disk_size = psize;
2007                 up_write(&F2FS_I(inode)->i_sem);
2008         }
2009
2010 done:
2011         if (err && err != -ENOENT)
2012                 goto redirty_out;
2013
2014 out:
2015         inode_dec_dirty_pages(inode);
2016         if (err) {
2017                 ClearPageUptodate(page);
2018                 clear_cold_data(page);
2019         }
2020
2021         if (wbc->for_reclaim) {
2022                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2023                 clear_inode_flag(inode, FI_HOT_DATA);
2024                 f2fs_remove_dirty_inode(inode);
2025                 submitted = NULL;
2026         }
2027
2028         unlock_page(page);
2029         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2030                 f2fs_balance_fs(sbi, need_balance_fs);
2031
2032         if (unlikely(f2fs_cp_error(sbi))) {
2033                 f2fs_submit_merged_write(sbi, DATA);
2034                 submitted = NULL;
2035         }
2036
2037         if (submitted)
2038                 *submitted = fio.submitted;
2039
2040         return 0;
2041
2042 redirty_out:
2043         redirty_page_for_writepage(wbc, page);
2044         /*
2045          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2046          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2047          * file_write_and_wait_range() will see EIO error, which is critical
2048          * to return value of fsync() followed by atomic_write failure to user.
2049          */
2050         if (!err || wbc->for_reclaim)
2051                 return AOP_WRITEPAGE_ACTIVATE;
2052         unlock_page(page);
2053         return err;
2054 }
2055
2056 static int f2fs_write_data_page(struct page *page,
2057                                         struct writeback_control *wbc)
2058 {
2059         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2060 }
2061
2062 /*
2063  * This function was copied from write_cche_pages from mm/page-writeback.c.
2064  * The major change is making write step of cold data page separately from
2065  * warm/hot data page.
2066  */
2067 static int f2fs_write_cache_pages(struct address_space *mapping,
2068                                         struct writeback_control *wbc,
2069                                         enum iostat_type io_type)
2070 {
2071         int ret = 0;
2072         int done = 0;
2073         struct pagevec pvec;
2074         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2075         int nr_pages;
2076         pgoff_t uninitialized_var(writeback_index);
2077         pgoff_t index;
2078         pgoff_t end;            /* Inclusive */
2079         pgoff_t done_index;
2080         int cycled;
2081         int range_whole = 0;
2082         xa_mark_t tag;
2083         int nwritten = 0;
2084
2085         pagevec_init(&pvec);
2086
2087         if (get_dirty_pages(mapping->host) <=
2088                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2089                 set_inode_flag(mapping->host, FI_HOT_DATA);
2090         else
2091                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2092
2093         if (wbc->range_cyclic) {
2094                 writeback_index = mapping->writeback_index; /* prev offset */
2095                 index = writeback_index;
2096                 if (index == 0)
2097                         cycled = 1;
2098                 else
2099                         cycled = 0;
2100                 end = -1;
2101         } else {
2102                 index = wbc->range_start >> PAGE_SHIFT;
2103                 end = wbc->range_end >> PAGE_SHIFT;
2104                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2105                         range_whole = 1;
2106                 cycled = 1; /* ignore range_cyclic tests */
2107         }
2108         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2109                 tag = PAGECACHE_TAG_TOWRITE;
2110         else
2111                 tag = PAGECACHE_TAG_DIRTY;
2112 retry:
2113         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2114                 tag_pages_for_writeback(mapping, index, end);
2115         done_index = index;
2116         while (!done && (index <= end)) {
2117                 int i;
2118
2119                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2120                                 tag);
2121                 if (nr_pages == 0)
2122                         break;
2123
2124                 for (i = 0; i < nr_pages; i++) {
2125                         struct page *page = pvec.pages[i];
2126                         bool submitted = false;
2127
2128                         /* give a priority to WB_SYNC threads */
2129                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2130                                         wbc->sync_mode == WB_SYNC_NONE) {
2131                                 done = 1;
2132                                 break;
2133                         }
2134
2135                         done_index = page->index;
2136 retry_write:
2137                         lock_page(page);
2138
2139                         if (unlikely(page->mapping != mapping)) {
2140 continue_unlock:
2141                                 unlock_page(page);
2142                                 continue;
2143                         }
2144
2145                         if (!PageDirty(page)) {
2146                                 /* someone wrote it for us */
2147                                 goto continue_unlock;
2148                         }
2149
2150                         if (PageWriteback(page)) {
2151                                 if (wbc->sync_mode != WB_SYNC_NONE)
2152                                         f2fs_wait_on_page_writeback(page,
2153                                                         DATA, true, true);
2154                                 else
2155                                         goto continue_unlock;
2156                         }
2157
2158                         if (!clear_page_dirty_for_io(page))
2159                                 goto continue_unlock;
2160
2161                         ret = __write_data_page(page, &submitted, wbc, io_type);
2162                         if (unlikely(ret)) {
2163                                 /*
2164                                  * keep nr_to_write, since vfs uses this to
2165                                  * get # of written pages.
2166                                  */
2167                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2168                                         unlock_page(page);
2169                                         ret = 0;
2170                                         continue;
2171                                 } else if (ret == -EAGAIN) {
2172                                         ret = 0;
2173                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2174                                                 cond_resched();
2175                                                 congestion_wait(BLK_RW_ASYNC,
2176                                                                         HZ/50);
2177                                                 goto retry_write;
2178                                         }
2179                                         continue;
2180                                 }
2181                                 done_index = page->index + 1;
2182                                 done = 1;
2183                                 break;
2184                         } else if (submitted) {
2185                                 nwritten++;
2186                         }
2187
2188                         if (--wbc->nr_to_write <= 0 &&
2189                                         wbc->sync_mode == WB_SYNC_NONE) {
2190                                 done = 1;
2191                                 break;
2192                         }
2193                 }
2194                 pagevec_release(&pvec);
2195                 cond_resched();
2196         }
2197
2198         if (!cycled && !done) {
2199                 cycled = 1;
2200                 index = 0;
2201                 end = writeback_index - 1;
2202                 goto retry;
2203         }
2204         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2205                 mapping->writeback_index = done_index;
2206
2207         if (nwritten)
2208                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2209                                                                 NULL, 0, DATA);
2210
2211         return ret;
2212 }
2213
2214 static inline bool __should_serialize_io(struct inode *inode,
2215                                         struct writeback_control *wbc)
2216 {
2217         if (!S_ISREG(inode->i_mode))
2218                 return false;
2219         if (IS_NOQUOTA(inode))
2220                 return false;
2221         if (wbc->sync_mode != WB_SYNC_ALL)
2222                 return true;
2223         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2224                 return true;
2225         return false;
2226 }
2227
2228 static int __f2fs_write_data_pages(struct address_space *mapping,
2229                                                 struct writeback_control *wbc,
2230                                                 enum iostat_type io_type)
2231 {
2232         struct inode *inode = mapping->host;
2233         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2234         struct blk_plug plug;
2235         int ret;
2236         bool locked = false;
2237
2238         /* deal with chardevs and other special file */
2239         if (!mapping->a_ops->writepage)
2240                 return 0;
2241
2242         /* skip writing if there is no dirty page in this inode */
2243         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2244                 return 0;
2245
2246         /* during POR, we don't need to trigger writepage at all. */
2247         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2248                 goto skip_write;
2249
2250         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2251                         wbc->sync_mode == WB_SYNC_NONE &&
2252                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2253                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2254                 goto skip_write;
2255
2256         /* skip writing during file defragment */
2257         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2258                 goto skip_write;
2259
2260         trace_f2fs_writepages(mapping->host, wbc, DATA);
2261
2262         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2263         if (wbc->sync_mode == WB_SYNC_ALL)
2264                 atomic_inc(&sbi->wb_sync_req[DATA]);
2265         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2266                 goto skip_write;
2267
2268         if (__should_serialize_io(inode, wbc)) {
2269                 mutex_lock(&sbi->writepages);
2270                 locked = true;
2271         }
2272
2273         blk_start_plug(&plug);
2274         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2275         blk_finish_plug(&plug);
2276
2277         if (locked)
2278                 mutex_unlock(&sbi->writepages);
2279
2280         if (wbc->sync_mode == WB_SYNC_ALL)
2281                 atomic_dec(&sbi->wb_sync_req[DATA]);
2282         /*
2283          * if some pages were truncated, we cannot guarantee its mapping->host
2284          * to detect pending bios.
2285          */
2286
2287         f2fs_remove_dirty_inode(inode);
2288         return ret;
2289
2290 skip_write:
2291         wbc->pages_skipped += get_dirty_pages(inode);
2292         trace_f2fs_writepages(mapping->host, wbc, DATA);
2293         return 0;
2294 }
2295
2296 static int f2fs_write_data_pages(struct address_space *mapping,
2297                             struct writeback_control *wbc)
2298 {
2299         struct inode *inode = mapping->host;
2300
2301         return __f2fs_write_data_pages(mapping, wbc,
2302                         F2FS_I(inode)->cp_task == current ?
2303                         FS_CP_DATA_IO : FS_DATA_IO);
2304 }
2305
2306 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2307 {
2308         struct inode *inode = mapping->host;
2309         loff_t i_size = i_size_read(inode);
2310
2311         if (to > i_size) {
2312                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2313                 down_write(&F2FS_I(inode)->i_mmap_sem);
2314
2315                 truncate_pagecache(inode, i_size);
2316                 if (!IS_NOQUOTA(inode))
2317                         f2fs_truncate_blocks(inode, i_size, true);
2318
2319                 up_write(&F2FS_I(inode)->i_mmap_sem);
2320                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2321         }
2322 }
2323
2324 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2325                         struct page *page, loff_t pos, unsigned len,
2326                         block_t *blk_addr, bool *node_changed)
2327 {
2328         struct inode *inode = page->mapping->host;
2329         pgoff_t index = page->index;
2330         struct dnode_of_data dn;
2331         struct page *ipage;
2332         bool locked = false;
2333         struct extent_info ei = {0,0,0};
2334         int err = 0;
2335         int flag;
2336
2337         /*
2338          * we already allocated all the blocks, so we don't need to get
2339          * the block addresses when there is no need to fill the page.
2340          */
2341         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2342                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2343                 return 0;
2344
2345         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2346         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2347                 flag = F2FS_GET_BLOCK_DEFAULT;
2348         else
2349                 flag = F2FS_GET_BLOCK_PRE_AIO;
2350
2351         if (f2fs_has_inline_data(inode) ||
2352                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2353                 __do_map_lock(sbi, flag, true);
2354                 locked = true;
2355         }
2356 restart:
2357         /* check inline_data */
2358         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2359         if (IS_ERR(ipage)) {
2360                 err = PTR_ERR(ipage);
2361                 goto unlock_out;
2362         }
2363
2364         set_new_dnode(&dn, inode, ipage, ipage, 0);
2365
2366         if (f2fs_has_inline_data(inode)) {
2367                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2368                         f2fs_do_read_inline_data(page, ipage);
2369                         set_inode_flag(inode, FI_DATA_EXIST);
2370                         if (inode->i_nlink)
2371                                 set_inline_node(ipage);
2372                 } else {
2373                         err = f2fs_convert_inline_page(&dn, page);
2374                         if (err)
2375                                 goto out;
2376                         if (dn.data_blkaddr == NULL_ADDR)
2377                                 err = f2fs_get_block(&dn, index);
2378                 }
2379         } else if (locked) {
2380                 err = f2fs_get_block(&dn, index);
2381         } else {
2382                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2383                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2384                 } else {
2385                         /* hole case */
2386                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2387                         if (err || dn.data_blkaddr == NULL_ADDR) {
2388                                 f2fs_put_dnode(&dn);
2389                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2390                                                                 true);
2391                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2392                                 locked = true;
2393                                 goto restart;
2394                         }
2395                 }
2396         }
2397
2398         /* convert_inline_page can make node_changed */
2399         *blk_addr = dn.data_blkaddr;
2400         *node_changed = dn.node_changed;
2401 out:
2402         f2fs_put_dnode(&dn);
2403 unlock_out:
2404         if (locked)
2405                 __do_map_lock(sbi, flag, false);
2406         return err;
2407 }
2408
2409 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2410                 loff_t pos, unsigned len, unsigned flags,
2411                 struct page **pagep, void **fsdata)
2412 {
2413         struct inode *inode = mapping->host;
2414         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2415         struct page *page = NULL;
2416         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2417         bool need_balance = false, drop_atomic = false;
2418         block_t blkaddr = NULL_ADDR;
2419         int err = 0;
2420
2421         trace_f2fs_write_begin(inode, pos, len, flags);
2422
2423         err = f2fs_is_checkpoint_ready(sbi);
2424         if (err)
2425                 goto fail;
2426
2427         if ((f2fs_is_atomic_file(inode) &&
2428                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2429                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2430                 err = -ENOMEM;
2431                 drop_atomic = true;
2432                 goto fail;
2433         }
2434
2435         /*
2436          * We should check this at this moment to avoid deadlock on inode page
2437          * and #0 page. The locking rule for inline_data conversion should be:
2438          * lock_page(page #0) -> lock_page(inode_page)
2439          */
2440         if (index != 0) {
2441                 err = f2fs_convert_inline_inode(inode);
2442                 if (err)
2443                         goto fail;
2444         }
2445 repeat:
2446         /*
2447          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2448          * wait_for_stable_page. Will wait that below with our IO control.
2449          */
2450         page = f2fs_pagecache_get_page(mapping, index,
2451                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2452         if (!page) {
2453                 err = -ENOMEM;
2454                 goto fail;
2455         }
2456
2457         *pagep = page;
2458
2459         err = prepare_write_begin(sbi, page, pos, len,
2460                                         &blkaddr, &need_balance);
2461         if (err)
2462                 goto fail;
2463
2464         if (need_balance && !IS_NOQUOTA(inode) &&
2465                         has_not_enough_free_secs(sbi, 0, 0)) {
2466                 unlock_page(page);
2467                 f2fs_balance_fs(sbi, true);
2468                 lock_page(page);
2469                 if (page->mapping != mapping) {
2470                         /* The page got truncated from under us */
2471                         f2fs_put_page(page, 1);
2472                         goto repeat;
2473                 }
2474         }
2475
2476         f2fs_wait_on_page_writeback(page, DATA, false, true);
2477
2478         if (len == PAGE_SIZE || PageUptodate(page))
2479                 return 0;
2480
2481         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2482                 zero_user_segment(page, len, PAGE_SIZE);
2483                 return 0;
2484         }
2485
2486         if (blkaddr == NEW_ADDR) {
2487                 zero_user_segment(page, 0, PAGE_SIZE);
2488                 SetPageUptodate(page);
2489         } else {
2490                 err = f2fs_submit_page_read(inode, page, blkaddr);
2491                 if (err)
2492                         goto fail;
2493
2494                 lock_page(page);
2495                 if (unlikely(page->mapping != mapping)) {
2496                         f2fs_put_page(page, 1);
2497                         goto repeat;
2498                 }
2499                 if (unlikely(!PageUptodate(page))) {
2500                         err = -EIO;
2501                         goto fail;
2502                 }
2503         }
2504         return 0;
2505
2506 fail:
2507         f2fs_put_page(page, 1);
2508         f2fs_write_failed(mapping, pos + len);
2509         if (drop_atomic)
2510                 f2fs_drop_inmem_pages_all(sbi, false);
2511         return err;
2512 }
2513
2514 static int f2fs_write_end(struct file *file,
2515                         struct address_space *mapping,
2516                         loff_t pos, unsigned len, unsigned copied,
2517                         struct page *page, void *fsdata)
2518 {
2519         struct inode *inode = page->mapping->host;
2520
2521         trace_f2fs_write_end(inode, pos, len, copied);
2522
2523         /*
2524          * This should be come from len == PAGE_SIZE, and we expect copied
2525          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2526          * let generic_perform_write() try to copy data again through copied=0.
2527          */
2528         if (!PageUptodate(page)) {
2529                 if (unlikely(copied != len))
2530                         copied = 0;
2531                 else
2532                         SetPageUptodate(page);
2533         }
2534         if (!copied)
2535                 goto unlock_out;
2536
2537         set_page_dirty(page);
2538
2539         if (pos + copied > i_size_read(inode))
2540                 f2fs_i_size_write(inode, pos + copied);
2541 unlock_out:
2542         f2fs_put_page(page, 1);
2543         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2544         return copied;
2545 }
2546
2547 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2548                            loff_t offset)
2549 {
2550         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2551         unsigned blkbits = i_blkbits;
2552         unsigned blocksize_mask = (1 << blkbits) - 1;
2553         unsigned long align = offset | iov_iter_alignment(iter);
2554         struct block_device *bdev = inode->i_sb->s_bdev;
2555
2556         if (align & blocksize_mask) {
2557                 if (bdev)
2558                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2559                 blocksize_mask = (1 << blkbits) - 1;
2560                 if (align & blocksize_mask)
2561                         return -EINVAL;
2562                 return 1;
2563         }
2564         return 0;
2565 }
2566
2567 static void f2fs_dio_end_io(struct bio *bio)
2568 {
2569         struct f2fs_private_dio *dio = bio->bi_private;
2570
2571         dec_page_count(F2FS_I_SB(dio->inode),
2572                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2573
2574         bio->bi_private = dio->orig_private;
2575         bio->bi_end_io = dio->orig_end_io;
2576
2577         kvfree(dio);
2578
2579         bio_endio(bio);
2580 }
2581
2582 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2583                                                         loff_t file_offset)
2584 {
2585         struct f2fs_private_dio *dio;
2586         bool write = (bio_op(bio) == REQ_OP_WRITE);
2587
2588         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2589                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2590         if (!dio)
2591                 goto out;
2592
2593         dio->inode = inode;
2594         dio->orig_end_io = bio->bi_end_io;
2595         dio->orig_private = bio->bi_private;
2596         dio->write = write;
2597
2598         bio->bi_end_io = f2fs_dio_end_io;
2599         bio->bi_private = dio;
2600
2601         inc_page_count(F2FS_I_SB(inode),
2602                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2603
2604         submit_bio(bio);
2605         return;
2606 out:
2607         bio->bi_status = BLK_STS_IOERR;
2608         bio_endio(bio);
2609 }
2610
2611 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2612 {
2613         struct address_space *mapping = iocb->ki_filp->f_mapping;
2614         struct inode *inode = mapping->host;
2615         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2616         struct f2fs_inode_info *fi = F2FS_I(inode);
2617         size_t count = iov_iter_count(iter);
2618         loff_t offset = iocb->ki_pos;
2619         int rw = iov_iter_rw(iter);
2620         int err;
2621         enum rw_hint hint = iocb->ki_hint;
2622         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2623         bool do_opu;
2624
2625         err = check_direct_IO(inode, iter, offset);
2626         if (err)
2627                 return err < 0 ? err : 0;
2628
2629         if (f2fs_force_buffered_io(inode, iocb, iter))
2630                 return 0;
2631
2632         do_opu = allow_outplace_dio(inode, iocb, iter);
2633
2634         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2635
2636         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2637                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2638
2639         if (iocb->ki_flags & IOCB_NOWAIT) {
2640                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2641                         iocb->ki_hint = hint;
2642                         err = -EAGAIN;
2643                         goto out;
2644                 }
2645                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2646                         up_read(&fi->i_gc_rwsem[rw]);
2647                         iocb->ki_hint = hint;
2648                         err = -EAGAIN;
2649                         goto out;
2650                 }
2651         } else {
2652                 down_read(&fi->i_gc_rwsem[rw]);
2653                 if (do_opu)
2654                         down_read(&fi->i_gc_rwsem[READ]);
2655         }
2656
2657         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2658                         iter, rw == WRITE ? get_data_block_dio_write :
2659                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2660                         DIO_LOCKING | DIO_SKIP_HOLES);
2661
2662         if (do_opu)
2663                 up_read(&fi->i_gc_rwsem[READ]);
2664
2665         up_read(&fi->i_gc_rwsem[rw]);
2666
2667         if (rw == WRITE) {
2668                 if (whint_mode == WHINT_MODE_OFF)
2669                         iocb->ki_hint = hint;
2670                 if (err > 0) {
2671                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2672                                                                         err);
2673                         if (!do_opu)
2674                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2675                 } else if (err < 0) {
2676                         f2fs_write_failed(mapping, offset + count);
2677                 }
2678         }
2679
2680 out:
2681         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2682
2683         return err;
2684 }
2685
2686 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2687                                                         unsigned int length)
2688 {
2689         struct inode *inode = page->mapping->host;
2690         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2691
2692         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2693                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2694                 return;
2695
2696         if (PageDirty(page)) {
2697                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2698                         dec_page_count(sbi, F2FS_DIRTY_META);
2699                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2700                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2701                 } else {
2702                         inode_dec_dirty_pages(inode);
2703                         f2fs_remove_dirty_inode(inode);
2704                 }
2705         }
2706
2707         clear_cold_data(page);
2708
2709         /* This is atomic written page, keep Private */
2710         if (IS_ATOMIC_WRITTEN_PAGE(page))
2711                 return f2fs_drop_inmem_page(inode, page);
2712
2713         set_page_private(page, 0);
2714         ClearPagePrivate(page);
2715 }
2716
2717 int f2fs_release_page(struct page *page, gfp_t wait)
2718 {
2719         /* If this is dirty page, keep PagePrivate */
2720         if (PageDirty(page))
2721                 return 0;
2722
2723         /* This is atomic written page, keep Private */
2724         if (IS_ATOMIC_WRITTEN_PAGE(page))
2725                 return 0;
2726
2727         clear_cold_data(page);
2728         set_page_private(page, 0);
2729         ClearPagePrivate(page);
2730         return 1;
2731 }
2732
2733 static int f2fs_set_data_page_dirty(struct page *page)
2734 {
2735         struct address_space *mapping = page->mapping;
2736         struct inode *inode = mapping->host;
2737
2738         trace_f2fs_set_page_dirty(page, DATA);
2739
2740         if (!PageUptodate(page))
2741                 SetPageUptodate(page);
2742
2743         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2744                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2745                         f2fs_register_inmem_page(inode, page);
2746                         return 1;
2747                 }
2748                 /*
2749                  * Previously, this page has been registered, we just
2750                  * return here.
2751                  */
2752                 return 0;
2753         }
2754
2755         if (!PageDirty(page)) {
2756                 __set_page_dirty_nobuffers(page);
2757                 f2fs_update_dirty_page(inode, page);
2758                 return 1;
2759         }
2760         return 0;
2761 }
2762
2763 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2764 {
2765         struct inode *inode = mapping->host;
2766
2767         if (f2fs_has_inline_data(inode))
2768                 return 0;
2769
2770         /* make sure allocating whole blocks */
2771         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2772                 filemap_write_and_wait(mapping);
2773
2774         return generic_block_bmap(mapping, block, get_data_block_bmap);
2775 }
2776
2777 #ifdef CONFIG_MIGRATION
2778 #include <linux/migrate.h>
2779
2780 int f2fs_migrate_page(struct address_space *mapping,
2781                 struct page *newpage, struct page *page, enum migrate_mode mode)
2782 {
2783         int rc, extra_count;
2784         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2785         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2786
2787         BUG_ON(PageWriteback(page));
2788
2789         /* migrating an atomic written page is safe with the inmem_lock hold */
2790         if (atomic_written) {
2791                 if (mode != MIGRATE_SYNC)
2792                         return -EBUSY;
2793                 if (!mutex_trylock(&fi->inmem_lock))
2794                         return -EAGAIN;
2795         }
2796
2797         /*
2798          * A reference is expected if PagePrivate set when move mapping,
2799          * however F2FS breaks this for maintaining dirty page counts when
2800          * truncating pages. So here adjusting the 'extra_count' make it work.
2801          */
2802         extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2803         rc = migrate_page_move_mapping(mapping, newpage,
2804                                 page, mode, extra_count);
2805         if (rc != MIGRATEPAGE_SUCCESS) {
2806                 if (atomic_written)
2807                         mutex_unlock(&fi->inmem_lock);
2808                 return rc;
2809         }
2810
2811         if (atomic_written) {
2812                 struct inmem_pages *cur;
2813                 list_for_each_entry(cur, &fi->inmem_pages, list)
2814                         if (cur->page == page) {
2815                                 cur->page = newpage;
2816                                 break;
2817                         }
2818                 mutex_unlock(&fi->inmem_lock);
2819                 put_page(page);
2820                 get_page(newpage);
2821         }
2822
2823         if (PagePrivate(page))
2824                 SetPagePrivate(newpage);
2825         set_page_private(newpage, page_private(page));
2826
2827         if (mode != MIGRATE_SYNC_NO_COPY)
2828                 migrate_page_copy(newpage, page);
2829         else
2830                 migrate_page_states(newpage, page);
2831
2832         return MIGRATEPAGE_SUCCESS;
2833 }
2834 #endif
2835
2836 const struct address_space_operations f2fs_dblock_aops = {
2837         .readpage       = f2fs_read_data_page,
2838         .readpages      = f2fs_read_data_pages,
2839         .writepage      = f2fs_write_data_page,
2840         .writepages     = f2fs_write_data_pages,
2841         .write_begin    = f2fs_write_begin,
2842         .write_end      = f2fs_write_end,
2843         .set_page_dirty = f2fs_set_data_page_dirty,
2844         .invalidatepage = f2fs_invalidate_page,
2845         .releasepage    = f2fs_release_page,
2846         .direct_IO      = f2fs_direct_IO,
2847         .bmap           = f2fs_bmap,
2848 #ifdef CONFIG_MIGRATION
2849         .migratepage    = f2fs_migrate_page,
2850 #endif
2851 };
2852
2853 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2854 {
2855         struct address_space *mapping = page_mapping(page);
2856         unsigned long flags;
2857
2858         xa_lock_irqsave(&mapping->i_pages, flags);
2859         __xa_clear_mark(&mapping->i_pages, page_index(page),
2860                                                 PAGECACHE_TAG_DIRTY);
2861         xa_unlock_irqrestore(&mapping->i_pages, flags);
2862 }
2863
2864 int __init f2fs_init_post_read_processing(void)
2865 {
2866         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2867         if (!bio_post_read_ctx_cache)
2868                 goto fail;
2869         bio_post_read_ctx_pool =
2870                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2871                                          bio_post_read_ctx_cache);
2872         if (!bio_post_read_ctx_pool)
2873                 goto fail_free_cache;
2874         return 0;
2875
2876 fail_free_cache:
2877         kmem_cache_destroy(bio_post_read_ctx_cache);
2878 fail:
2879         return -ENOMEM;
2880 }
2881
2882 void __exit f2fs_destroy_post_read_processing(void)
2883 {
2884         mempool_destroy(bio_post_read_ctx_pool);
2885         kmem_cache_destroy(bio_post_read_ctx_cache);
2886 }