fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/prefetch.h>
  18 #include <linux/uio.h>
  19 #include <linux/cleancache.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include "trace.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define NUM_PREALLOC_POST_READ_CTXS     128
  29
  30 static struct kmem_cache *bio_post_read_ctx_cache;
  31 static mempool_t *bio_post_read_ctx_pool;
  32
  33 static bool __is_cp_guaranteed(struct page *page)
  34 {
  35         struct address_space *mapping = page->mapping;
  36         struct inode *inode;
  37         struct f2fs_sb_info *sbi;
  38
  39         if (!mapping)
  40                 return false;
  41
  42         inode = mapping->host;
  43         sbi = F2FS_I_SB(inode);
  44
  45         if (inode->i_ino == F2FS_META_INO(sbi) ||
  46                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47                         S_ISDIR(inode->i_mode) ||
  48                         (S_ISREG(inode->i_mode) &&
  49                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  50                         is_cold_data(page))
  51                 return true;
  52         return false;
  53 }
  54
  55 static enum count_type __read_io_type(struct page *page)
  56 {
  57         struct address_space *mapping = page->mapping;
  58
  59         if (mapping) {
  60                 struct inode *inode = mapping->host;
  61                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  62
  63                 if (inode->i_ino == F2FS_META_INO(sbi))
  64                         return F2FS_RD_META;
  65
  66                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  67                         return F2FS_RD_NODE;
  68         }
  69         return F2FS_RD_DATA;
  70 }
  71
  72 /* postprocessing steps for read bios */
  73 enum bio_post_read_step {
  74         STEP_INITIAL = 0,
  75         STEP_DECRYPT,
  76 };
  77
  78 struct bio_post_read_ctx {
  79         struct bio *bio;
  80         struct work_struct work;
  81         unsigned int cur_step;
  82         unsigned int enabled_steps;
  83 };
  84
  85 static void __read_end_io(struct bio *bio)
  86 {
  87         struct page *page;
  88         struct bio_vec *bv;
  89         struct bvec_iter_all iter_all;
  90
  91         bio_for_each_segment_all(bv, bio, iter_all) {
  92                 page = bv->bv_page;
  93
  94                 /* PG_error was set if any post_read step failed */
  95                 if (bio->bi_status || PageError(page)) {
  96                         ClearPageUptodate(page);
  97                         /* will re-read again later */
  98                         ClearPageError(page);
  99                 } else {
 100                         SetPageUptodate(page);
 101                 }
 102                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 103                 unlock_page(page);
 104         }
 105         if (bio->bi_private)
 106                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 107         bio_put(bio);
 108 }
 109
 110 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 111
 112 static void decrypt_work(struct work_struct *work)
 113 {
 114         struct bio_post_read_ctx *ctx =
 115                 container_of(work, struct bio_post_read_ctx, work);
 116
 117         fscrypt_decrypt_bio(ctx->bio);
 118
 119         bio_post_read_processing(ctx);
 120 }
 121
 122 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 123 {
 124         switch (++ctx->cur_step) {
 125         case STEP_DECRYPT:
 126                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 127                         INIT_WORK(&ctx->work, decrypt_work);
 128                         fscrypt_enqueue_decrypt_work(&ctx->work);
 129                         return;
 130                 }
 131                 ctx->cur_step++;
 132                 /* fall-through */
 133         default:
 134                 __read_end_io(ctx->bio);
 135         }
 136 }
 137
 138 static bool f2fs_bio_post_read_required(struct bio *bio)
 139 {
 140         return bio->bi_private && !bio->bi_status;
 141 }
 142
 143 static void f2fs_read_end_io(struct bio *bio)
 144 {
 145         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
 146                                                 FAULT_READ_IO)) {
 147                 f2fs_show_injection_info(FAULT_READ_IO);
 148                 bio->bi_status = BLK_STS_IOERR;
 149         }
 150
 151         if (f2fs_bio_post_read_required(bio)) {
 152                 struct bio_post_read_ctx *ctx = bio->bi_private;
 153
 154                 ctx->cur_step = STEP_INITIAL;
 155                 bio_post_read_processing(ctx);
 156                 return;
 157         }
 158
 159         __read_end_io(bio);
 160 }
 161
 162 static void f2fs_write_end_io(struct bio *bio)
 163 {
 164         struct f2fs_sb_info *sbi = bio->bi_private;
 165         struct bio_vec *bvec;
 166         struct bvec_iter_all iter_all;
 167
 168         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 169                 f2fs_show_injection_info(FAULT_WRITE_IO);
 170                 bio->bi_status = BLK_STS_IOERR;
 171         }
 172
 173         bio_for_each_segment_all(bvec, bio, iter_all) {
 174                 struct page *page = bvec->bv_page;
 175                 enum count_type type = WB_DATA_TYPE(page);
 176
 177                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 178                         set_page_private(page, (unsigned long)NULL);
 179                         ClearPagePrivate(page);
 180                         unlock_page(page);
 181                         mempool_free(page, sbi->write_io_dummy);
 182
 183                         if (unlikely(bio->bi_status))
 184                                 f2fs_stop_checkpoint(sbi, true);
 185                         continue;
 186                 }
 187
 188                 fscrypt_pullback_bio_page(&page, true);
 189
 190                 if (unlikely(bio->bi_status)) {
 191                         mapping_set_error(page->mapping, -EIO);
 192                         if (type == F2FS_WB_CP_DATA)
 193                                 f2fs_stop_checkpoint(sbi, true);
 194                 }
 195
 196                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 197                                         page->index != nid_of_node(page));
 198
 199                 dec_page_count(sbi, type);
 200                 if (f2fs_in_warm_node_list(sbi, page))
 201                         f2fs_del_fsync_node_entry(sbi, page);
 202                 clear_cold_data(page);
 203                 end_page_writeback(page);
 204         }
 205         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 206                                 wq_has_sleeper(&sbi->cp_wait))
 207                 wake_up(&sbi->cp_wait);
 208
 209         bio_put(bio);
 210 }
 211
 212 /*
 213  * Return true, if pre_bio's bdev is same as its target device.
 214  */
 215 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 216                                 block_t blk_addr, struct bio *bio)
 217 {
 218         struct block_device *bdev = sbi->sb->s_bdev;
 219         int i;
 220
 221         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         struct bvec_iter_all iter_all;
 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, iter_all) {
 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 (IS_ENCRYPTED(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                 /* just zeroing out page which is beyond EOF */
1555                 if (block_in_file >= last_block)
1556                         goto zero_out;
1557                 /*
1558                  * Map blocks using the previous result first.
1559                  */
1560                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1561                                 block_in_file > map.m_lblk &&
1562                                 block_in_file < (map.m_lblk + map.m_len))
1563                         goto got_it;
1564
1565                 /*
1566                  * Then do more f2fs_map_blocks() calls until we are
1567                  * done with this page.
1568                  */
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, F2FS_GET_BLOCK_DEFAULT))
1573                         goto set_error_page;
1574 got_it:
1575                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1576                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1577                         SetPageMappedToDisk(page);
1578
1579                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1580                                 SetPageUptodate(page);
1581                                 goto confused;
1582                         }
1583
1584                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1585                                                                 DATA_GENERIC))
1586                                 goto set_error_page;
1587                 } else {
1588 zero_out:
1589                         zero_user_segment(page, 0, PAGE_SIZE);
1590                         if (!PageUptodate(page))
1591                                 SetPageUptodate(page);
1592                         unlock_page(page);
1593                         goto next_page;
1594                 }
1595
1596                 /*
1597                  * This page will go to BIO.  Do we need to send this
1598                  * BIO off first?
1599                  */
1600                 if (bio && (last_block_in_bio != block_nr - 1 ||
1601                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1602 submit_and_realloc:
1603                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1604                         bio = NULL;
1605                 }
1606                 if (bio == NULL) {
1607                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1608                                         is_readahead ? REQ_RAHEAD : 0);
1609                         if (IS_ERR(bio)) {
1610                                 bio = NULL;
1611                                 goto set_error_page;
1612                         }
1613                 }
1614
1615                 /*
1616                  * If the page is under writeback, we need to wait for
1617                  * its completion to see the correct decrypted data.
1618                  */
1619                 f2fs_wait_on_block_writeback(inode, block_nr);
1620
1621                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1622                         goto submit_and_realloc;
1623
1624                 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1625                 ClearPageError(page);
1626                 last_block_in_bio = block_nr;
1627                 goto next_page;
1628 set_error_page:
1629                 SetPageError(page);
1630                 zero_user_segment(page, 0, PAGE_SIZE);
1631                 unlock_page(page);
1632                 goto next_page;
1633 confused:
1634                 if (bio) {
1635                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1636                         bio = NULL;
1637                 }
1638                 unlock_page(page);
1639 next_page:
1640                 if (pages)
1641                         put_page(page);
1642         }
1643         BUG_ON(pages && !list_empty(pages));
1644         if (bio)
1645                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1646         return 0;
1647 }
1648
1649 static int f2fs_read_data_page(struct file *file, struct page *page)
1650 {
1651         struct inode *inode = page->mapping->host;
1652         int ret = -EAGAIN;
1653
1654         trace_f2fs_readpage(page, DATA);
1655
1656         /* If the file has inline data, try to read it directly */
1657         if (f2fs_has_inline_data(inode))
1658                 ret = f2fs_read_inline_data(inode, page);
1659         if (ret == -EAGAIN)
1660                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1661         return ret;
1662 }
1663
1664 static int f2fs_read_data_pages(struct file *file,
1665                         struct address_space *mapping,
1666                         struct list_head *pages, unsigned nr_pages)
1667 {
1668         struct inode *inode = mapping->host;
1669         struct page *page = list_last_entry(pages, struct page, lru);
1670
1671         trace_f2fs_readpages(inode, page, nr_pages);
1672
1673         /* If the file has inline data, skip readpages */
1674         if (f2fs_has_inline_data(inode))
1675                 return 0;
1676
1677         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1678 }
1679
1680 static int encrypt_one_page(struct f2fs_io_info *fio)
1681 {
1682         struct inode *inode = fio->page->mapping->host;
1683         struct page *mpage;
1684         gfp_t gfp_flags = GFP_NOFS;
1685
1686         if (!f2fs_encrypted_file(inode))
1687                 return 0;
1688
1689         /* wait for GCed page writeback via META_MAPPING */
1690         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1691
1692 retry_encrypt:
1693         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1694                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1695         if (IS_ERR(fio->encrypted_page)) {
1696                 /* flush pending IOs and wait for a while in the ENOMEM case */
1697                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1698                         f2fs_flush_merged_writes(fio->sbi);
1699                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1700                         gfp_flags |= __GFP_NOFAIL;
1701                         goto retry_encrypt;
1702                 }
1703                 return PTR_ERR(fio->encrypted_page);
1704         }
1705
1706         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1707         if (mpage) {
1708                 if (PageUptodate(mpage))
1709                         memcpy(page_address(mpage),
1710                                 page_address(fio->encrypted_page), PAGE_SIZE);
1711                 f2fs_put_page(mpage, 1);
1712         }
1713         return 0;
1714 }
1715
1716 static inline bool check_inplace_update_policy(struct inode *inode,
1717                                 struct f2fs_io_info *fio)
1718 {
1719         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1720         unsigned int policy = SM_I(sbi)->ipu_policy;
1721
1722         if (policy & (0x1 << F2FS_IPU_FORCE))
1723                 return true;
1724         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1725                 return true;
1726         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1727                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1728                 return true;
1729         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1730                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1731                 return true;
1732
1733         /*
1734          * IPU for rewrite async pages
1735          */
1736         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1737                         fio && fio->op == REQ_OP_WRITE &&
1738                         !(fio->op_flags & REQ_SYNC) &&
1739                         !IS_ENCRYPTED(inode))
1740                 return true;
1741
1742         /* this is only set during fdatasync */
1743         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1744                         is_inode_flag_set(inode, FI_NEED_IPU))
1745                 return true;
1746
1747         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1748                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1749                 return true;
1750
1751         return false;
1752 }
1753
1754 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1755 {
1756         if (f2fs_is_pinned_file(inode))
1757                 return true;
1758
1759         /* if this is cold file, we should overwrite to avoid fragmentation */
1760         if (file_is_cold(inode))
1761                 return true;
1762
1763         return check_inplace_update_policy(inode, fio);
1764 }
1765
1766 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1767 {
1768         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1769
1770         if (test_opt(sbi, LFS))
1771                 return true;
1772         if (S_ISDIR(inode->i_mode))
1773                 return true;
1774         if (IS_NOQUOTA(inode))
1775                 return true;
1776         if (f2fs_is_atomic_file(inode))
1777                 return true;
1778         if (fio) {
1779                 if (is_cold_data(fio->page))
1780                         return true;
1781                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1782                         return true;
1783                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1784                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1785                         return true;
1786         }
1787         return false;
1788 }
1789
1790 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1791 {
1792         struct inode *inode = fio->page->mapping->host;
1793
1794         if (f2fs_should_update_outplace(inode, fio))
1795                 return false;
1796
1797         return f2fs_should_update_inplace(inode, fio);
1798 }
1799
1800 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1801 {
1802         struct page *page = fio->page;
1803         struct inode *inode = page->mapping->host;
1804         struct dnode_of_data dn;
1805         struct extent_info ei = {0,0,0};
1806         struct node_info ni;
1807         bool ipu_force = false;
1808         int err = 0;
1809
1810         set_new_dnode(&dn, inode, NULL, NULL, 0);
1811         if (need_inplace_update(fio) &&
1812                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1813                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1814
1815                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1816                                                         DATA_GENERIC))
1817                         return -EFAULT;
1818
1819                 ipu_force = true;
1820                 fio->need_lock = LOCK_DONE;
1821                 goto got_it;
1822         }
1823
1824         /* Deadlock due to between page->lock and f2fs_lock_op */
1825         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1826                 return -EAGAIN;
1827
1828         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1829         if (err)
1830                 goto out;
1831
1832         fio->old_blkaddr = dn.data_blkaddr;
1833
1834         /* This page is already truncated */
1835         if (fio->old_blkaddr == NULL_ADDR) {
1836                 ClearPageUptodate(page);
1837                 clear_cold_data(page);
1838                 goto out_writepage;
1839         }
1840 got_it:
1841         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1842                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1843                                                         DATA_GENERIC)) {
1844                 err = -EFAULT;
1845                 goto out_writepage;
1846         }
1847         /*
1848          * If current allocation needs SSR,
1849          * it had better in-place writes for updated data.
1850          */
1851         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1852                                         need_inplace_update(fio))) {
1853                 err = encrypt_one_page(fio);
1854                 if (err)
1855                         goto out_writepage;
1856
1857                 set_page_writeback(page);
1858                 ClearPageError(page);
1859                 f2fs_put_dnode(&dn);
1860                 if (fio->need_lock == LOCK_REQ)
1861                         f2fs_unlock_op(fio->sbi);
1862                 err = f2fs_inplace_write_data(fio);
1863                 if (err) {
1864                         if (f2fs_encrypted_file(inode))
1865                                 fscrypt_pullback_bio_page(&fio->encrypted_page,
1866                                                                         true);
1867                         if (PageWriteback(page))
1868                                 end_page_writeback(page);
1869                 }
1870                 trace_f2fs_do_write_data_page(fio->page, IPU);
1871                 set_inode_flag(inode, FI_UPDATE_WRITE);
1872                 return err;
1873         }
1874
1875         if (fio->need_lock == LOCK_RETRY) {
1876                 if (!f2fs_trylock_op(fio->sbi)) {
1877                         err = -EAGAIN;
1878                         goto out_writepage;
1879                 }
1880                 fio->need_lock = LOCK_REQ;
1881         }
1882
1883         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1884         if (err)
1885                 goto out_writepage;
1886
1887         fio->version = ni.version;
1888
1889         err = encrypt_one_page(fio);
1890         if (err)
1891                 goto out_writepage;
1892
1893         set_page_writeback(page);
1894         ClearPageError(page);
1895
1896         /* LFS mode write path */
1897         f2fs_outplace_write_data(&dn, fio);
1898         trace_f2fs_do_write_data_page(page, OPU);
1899         set_inode_flag(inode, FI_APPEND_WRITE);
1900         if (page->index == 0)
1901                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1902 out_writepage:
1903         f2fs_put_dnode(&dn);
1904 out:
1905         if (fio->need_lock == LOCK_REQ)
1906                 f2fs_unlock_op(fio->sbi);
1907         return err;
1908 }
1909
1910 static int __write_data_page(struct page *page, bool *submitted,
1911                                 struct writeback_control *wbc,
1912                                 enum iostat_type io_type)
1913 {
1914         struct inode *inode = page->mapping->host;
1915         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1916         loff_t i_size = i_size_read(inode);
1917         const pgoff_t end_index = ((unsigned long long) i_size)
1918                                                         >> PAGE_SHIFT;
1919         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1920         unsigned offset = 0;
1921         bool need_balance_fs = false;
1922         int err = 0;
1923         struct f2fs_io_info fio = {
1924                 .sbi = sbi,
1925                 .ino = inode->i_ino,
1926                 .type = DATA,
1927                 .op = REQ_OP_WRITE,
1928                 .op_flags = wbc_to_write_flags(wbc),
1929                 .old_blkaddr = NULL_ADDR,
1930                 .page = page,
1931                 .encrypted_page = NULL,
1932                 .submitted = false,
1933                 .need_lock = LOCK_RETRY,
1934                 .io_type = io_type,
1935                 .io_wbc = wbc,
1936         };
1937
1938         trace_f2fs_writepage(page, DATA);
1939
1940         /* we should bypass data pages to proceed the kworkder jobs */
1941         if (unlikely(f2fs_cp_error(sbi))) {
1942                 mapping_set_error(page->mapping, -EIO);
1943                 /*
1944                  * don't drop any dirty dentry pages for keeping lastest
1945                  * directory structure.
1946                  */
1947                 if (S_ISDIR(inode->i_mode))
1948                         goto redirty_out;
1949                 goto out;
1950         }
1951
1952         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1953                 goto redirty_out;
1954
1955         if (page->index < end_index)
1956                 goto write;
1957
1958         /*
1959          * If the offset is out-of-range of file size,
1960          * this page does not have to be written to disk.
1961          */
1962         offset = i_size & (PAGE_SIZE - 1);
1963         if ((page->index >= end_index + 1) || !offset)
1964                 goto out;
1965
1966         zero_user_segment(page, offset, PAGE_SIZE);
1967 write:
1968         if (f2fs_is_drop_cache(inode))
1969                 goto out;
1970         /* we should not write 0'th page having journal header */
1971         if (f2fs_is_volatile_file(inode) && (!page->index ||
1972                         (!wbc->for_reclaim &&
1973                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1974                 goto redirty_out;
1975
1976         /* Dentry blocks are controlled by checkpoint */
1977         if (S_ISDIR(inode->i_mode)) {
1978                 fio.need_lock = LOCK_DONE;
1979                 err = f2fs_do_write_data_page(&fio);
1980                 goto done;
1981         }
1982
1983         if (!wbc->for_reclaim)
1984                 need_balance_fs = true;
1985         else if (has_not_enough_free_secs(sbi, 0, 0))
1986                 goto redirty_out;
1987         else
1988                 set_inode_flag(inode, FI_HOT_DATA);
1989
1990         err = -EAGAIN;
1991         if (f2fs_has_inline_data(inode)) {
1992                 err = f2fs_write_inline_data(inode, page);
1993                 if (!err)
1994                         goto out;
1995         }
1996
1997         if (err == -EAGAIN) {
1998                 err = f2fs_do_write_data_page(&fio);
1999                 if (err == -EAGAIN) {
2000                         fio.need_lock = LOCK_REQ;
2001                         err = f2fs_do_write_data_page(&fio);
2002                 }
2003         }
2004
2005         if (err) {
2006                 file_set_keep_isize(inode);
2007         } else {
2008                 down_write(&F2FS_I(inode)->i_sem);
2009                 if (F2FS_I(inode)->last_disk_size < psize)
2010                         F2FS_I(inode)->last_disk_size = psize;
2011                 up_write(&F2FS_I(inode)->i_sem);
2012         }
2013
2014 done:
2015         if (err && err != -ENOENT)
2016                 goto redirty_out;
2017
2018 out:
2019         inode_dec_dirty_pages(inode);
2020         if (err) {
2021                 ClearPageUptodate(page);
2022                 clear_cold_data(page);
2023         }
2024
2025         if (wbc->for_reclaim) {
2026                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2027                 clear_inode_flag(inode, FI_HOT_DATA);
2028                 f2fs_remove_dirty_inode(inode);
2029                 submitted = NULL;
2030         }
2031
2032         unlock_page(page);
2033         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2034                 f2fs_balance_fs(sbi, need_balance_fs);
2035
2036         if (unlikely(f2fs_cp_error(sbi))) {
2037                 f2fs_submit_merged_write(sbi, DATA);
2038                 submitted = NULL;
2039         }
2040
2041         if (submitted)
2042                 *submitted = fio.submitted;
2043
2044         return 0;
2045
2046 redirty_out:
2047         redirty_page_for_writepage(wbc, page);
2048         /*
2049          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2050          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2051          * file_write_and_wait_range() will see EIO error, which is critical
2052          * to return value of fsync() followed by atomic_write failure to user.
2053          */
2054         if (!err || wbc->for_reclaim)
2055                 return AOP_WRITEPAGE_ACTIVATE;
2056         unlock_page(page);
2057         return err;
2058 }
2059
2060 static int f2fs_write_data_page(struct page *page,
2061                                         struct writeback_control *wbc)
2062 {
2063         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2064 }
2065
2066 /*
2067  * This function was copied from write_cche_pages from mm/page-writeback.c.
2068  * The major change is making write step of cold data page separately from
2069  * warm/hot data page.
2070  */
2071 static int f2fs_write_cache_pages(struct address_space *mapping,
2072                                         struct writeback_control *wbc,
2073                                         enum iostat_type io_type)
2074 {
2075         int ret = 0;
2076         int done = 0;
2077         struct pagevec pvec;
2078         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2079         int nr_pages;
2080         pgoff_t uninitialized_var(writeback_index);
2081         pgoff_t index;
2082         pgoff_t end;            /* Inclusive */
2083         pgoff_t done_index;
2084         int cycled;
2085         int range_whole = 0;
2086         xa_mark_t tag;
2087         int nwritten = 0;
2088
2089         pagevec_init(&pvec);
2090
2091         if (get_dirty_pages(mapping->host) <=
2092                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2093                 set_inode_flag(mapping->host, FI_HOT_DATA);
2094         else
2095                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2096
2097         if (wbc->range_cyclic) {
2098                 writeback_index = mapping->writeback_index; /* prev offset */
2099                 index = writeback_index;
2100                 if (index == 0)
2101                         cycled = 1;
2102                 else
2103                         cycled = 0;
2104                 end = -1;
2105         } else {
2106                 index = wbc->range_start >> PAGE_SHIFT;
2107                 end = wbc->range_end >> PAGE_SHIFT;
2108                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2109                         range_whole = 1;
2110                 cycled = 1; /* ignore range_cyclic tests */
2111         }
2112         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2113                 tag = PAGECACHE_TAG_TOWRITE;
2114         else
2115                 tag = PAGECACHE_TAG_DIRTY;
2116 retry:
2117         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2118                 tag_pages_for_writeback(mapping, index, end);
2119         done_index = index;
2120         while (!done && (index <= end)) {
2121                 int i;
2122
2123                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2124                                 tag);
2125                 if (nr_pages == 0)
2126                         break;
2127
2128                 for (i = 0; i < nr_pages; i++) {
2129                         struct page *page = pvec.pages[i];
2130                         bool submitted = false;
2131
2132                         /* give a priority to WB_SYNC threads */
2133                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2134                                         wbc->sync_mode == WB_SYNC_NONE) {
2135                                 done = 1;
2136                                 break;
2137                         }
2138
2139                         done_index = page->index;
2140 retry_write:
2141                         lock_page(page);
2142
2143                         if (unlikely(page->mapping != mapping)) {
2144 continue_unlock:
2145                                 unlock_page(page);
2146                                 continue;
2147                         }
2148
2149                         if (!PageDirty(page)) {
2150                                 /* someone wrote it for us */
2151                                 goto continue_unlock;
2152                         }
2153
2154                         if (PageWriteback(page)) {
2155                                 if (wbc->sync_mode != WB_SYNC_NONE)
2156                                         f2fs_wait_on_page_writeback(page,
2157                                                         DATA, true, true);
2158                                 else
2159                                         goto continue_unlock;
2160                         }
2161
2162                         if (!clear_page_dirty_for_io(page))
2163                                 goto continue_unlock;
2164
2165                         ret = __write_data_page(page, &submitted, wbc, io_type);
2166                         if (unlikely(ret)) {
2167                                 /*
2168                                  * keep nr_to_write, since vfs uses this to
2169                                  * get # of written pages.
2170                                  */
2171                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2172                                         unlock_page(page);
2173                                         ret = 0;
2174                                         continue;
2175                                 } else if (ret == -EAGAIN) {
2176                                         ret = 0;
2177                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2178                                                 cond_resched();
2179                                                 congestion_wait(BLK_RW_ASYNC,
2180                                                                         HZ/50);
2181                                                 goto retry_write;
2182                                         }
2183                                         continue;
2184                                 }
2185                                 done_index = page->index + 1;
2186                                 done = 1;
2187                                 break;
2188                         } else if (submitted) {
2189                                 nwritten++;
2190                         }
2191
2192                         if (--wbc->nr_to_write <= 0 &&
2193                                         wbc->sync_mode == WB_SYNC_NONE) {
2194                                 done = 1;
2195                                 break;
2196                         }
2197                 }
2198                 pagevec_release(&pvec);
2199                 cond_resched();
2200         }
2201
2202         if (!cycled && !done) {
2203                 cycled = 1;
2204                 index = 0;
2205                 end = writeback_index - 1;
2206                 goto retry;
2207         }
2208         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2209                 mapping->writeback_index = done_index;
2210
2211         if (nwritten)
2212                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2213                                                                 NULL, 0, DATA);
2214
2215         return ret;
2216 }
2217
2218 static inline bool __should_serialize_io(struct inode *inode,
2219                                         struct writeback_control *wbc)
2220 {
2221         if (!S_ISREG(inode->i_mode))
2222                 return false;
2223         if (IS_NOQUOTA(inode))
2224                 return false;
2225         if (wbc->sync_mode != WB_SYNC_ALL)
2226                 return true;
2227         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2228                 return true;
2229         return false;
2230 }
2231
2232 static int __f2fs_write_data_pages(struct address_space *mapping,
2233                                                 struct writeback_control *wbc,
2234                                                 enum iostat_type io_type)
2235 {
2236         struct inode *inode = mapping->host;
2237         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2238         struct blk_plug plug;
2239         int ret;
2240         bool locked = false;
2241
2242         /* deal with chardevs and other special file */
2243         if (!mapping->a_ops->writepage)
2244                 return 0;
2245
2246         /* skip writing if there is no dirty page in this inode */
2247         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2248                 return 0;
2249
2250         /* during POR, we don't need to trigger writepage at all. */
2251         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2252                 goto skip_write;
2253
2254         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2255                         wbc->sync_mode == WB_SYNC_NONE &&
2256                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2257                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2258                 goto skip_write;
2259
2260         /* skip writing during file defragment */
2261         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2262                 goto skip_write;
2263
2264         trace_f2fs_writepages(mapping->host, wbc, DATA);
2265
2266         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2267         if (wbc->sync_mode == WB_SYNC_ALL)
2268                 atomic_inc(&sbi->wb_sync_req[DATA]);
2269         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2270                 goto skip_write;
2271
2272         if (__should_serialize_io(inode, wbc)) {
2273                 mutex_lock(&sbi->writepages);
2274                 locked = true;
2275         }
2276
2277         blk_start_plug(&plug);
2278         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2279         blk_finish_plug(&plug);
2280
2281         if (locked)
2282                 mutex_unlock(&sbi->writepages);
2283
2284         if (wbc->sync_mode == WB_SYNC_ALL)
2285                 atomic_dec(&sbi->wb_sync_req[DATA]);
2286         /*
2287          * if some pages were truncated, we cannot guarantee its mapping->host
2288          * to detect pending bios.
2289          */
2290
2291         f2fs_remove_dirty_inode(inode);
2292         return ret;
2293
2294 skip_write:
2295         wbc->pages_skipped += get_dirty_pages(inode);
2296         trace_f2fs_writepages(mapping->host, wbc, DATA);
2297         return 0;
2298 }
2299
2300 static int f2fs_write_data_pages(struct address_space *mapping,
2301                             struct writeback_control *wbc)
2302 {
2303         struct inode *inode = mapping->host;
2304
2305         return __f2fs_write_data_pages(mapping, wbc,
2306                         F2FS_I(inode)->cp_task == current ?
2307                         FS_CP_DATA_IO : FS_DATA_IO);
2308 }
2309
2310 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2311 {
2312         struct inode *inode = mapping->host;
2313         loff_t i_size = i_size_read(inode);
2314
2315         if (to > i_size) {
2316                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2317                 down_write(&F2FS_I(inode)->i_mmap_sem);
2318
2319                 truncate_pagecache(inode, i_size);
2320                 if (!IS_NOQUOTA(inode))
2321                         f2fs_truncate_blocks(inode, i_size, true);
2322
2323                 up_write(&F2FS_I(inode)->i_mmap_sem);
2324                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2325         }
2326 }
2327
2328 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2329                         struct page *page, loff_t pos, unsigned len,
2330                         block_t *blk_addr, bool *node_changed)
2331 {
2332         struct inode *inode = page->mapping->host;
2333         pgoff_t index = page->index;
2334         struct dnode_of_data dn;
2335         struct page *ipage;
2336         bool locked = false;
2337         struct extent_info ei = {0,0,0};
2338         int err = 0;
2339         int flag;
2340
2341         /*
2342          * we already allocated all the blocks, so we don't need to get
2343          * the block addresses when there is no need to fill the page.
2344          */
2345         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2346                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2347                 return 0;
2348
2349         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2350         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2351                 flag = F2FS_GET_BLOCK_DEFAULT;
2352         else
2353                 flag = F2FS_GET_BLOCK_PRE_AIO;
2354
2355         if (f2fs_has_inline_data(inode) ||
2356                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2357                 __do_map_lock(sbi, flag, true);
2358                 locked = true;
2359         }
2360 restart:
2361         /* check inline_data */
2362         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2363         if (IS_ERR(ipage)) {
2364                 err = PTR_ERR(ipage);
2365                 goto unlock_out;
2366         }
2367
2368         set_new_dnode(&dn, inode, ipage, ipage, 0);
2369
2370         if (f2fs_has_inline_data(inode)) {
2371                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2372                         f2fs_do_read_inline_data(page, ipage);
2373                         set_inode_flag(inode, FI_DATA_EXIST);
2374                         if (inode->i_nlink)
2375                                 set_inline_node(ipage);
2376                 } else {
2377                         err = f2fs_convert_inline_page(&dn, page);
2378                         if (err)
2379                                 goto out;
2380                         if (dn.data_blkaddr == NULL_ADDR)
2381                                 err = f2fs_get_block(&dn, index);
2382                 }
2383         } else if (locked) {
2384                 err = f2fs_get_block(&dn, index);
2385         } else {
2386                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2387                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2388                 } else {
2389                         /* hole case */
2390                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2391                         if (err || dn.data_blkaddr == NULL_ADDR) {
2392                                 f2fs_put_dnode(&dn);
2393                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2394                                                                 true);
2395                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2396                                 locked = true;
2397                                 goto restart;
2398                         }
2399                 }
2400         }
2401
2402         /* convert_inline_page can make node_changed */
2403         *blk_addr = dn.data_blkaddr;
2404         *node_changed = dn.node_changed;
2405 out:
2406         f2fs_put_dnode(&dn);
2407 unlock_out:
2408         if (locked)
2409                 __do_map_lock(sbi, flag, false);
2410         return err;
2411 }
2412
2413 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2414                 loff_t pos, unsigned len, unsigned flags,
2415                 struct page **pagep, void **fsdata)
2416 {
2417         struct inode *inode = mapping->host;
2418         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2419         struct page *page = NULL;
2420         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2421         bool need_balance = false, drop_atomic = false;
2422         block_t blkaddr = NULL_ADDR;
2423         int err = 0;
2424
2425         trace_f2fs_write_begin(inode, pos, len, flags);
2426
2427         err = f2fs_is_checkpoint_ready(sbi);
2428         if (err)
2429                 goto fail;
2430
2431         if ((f2fs_is_atomic_file(inode) &&
2432                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2433                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2434                 err = -ENOMEM;
2435                 drop_atomic = true;
2436                 goto fail;
2437         }
2438
2439         /*
2440          * We should check this at this moment to avoid deadlock on inode page
2441          * and #0 page. The locking rule for inline_data conversion should be:
2442          * lock_page(page #0) -> lock_page(inode_page)
2443          */
2444         if (index != 0) {
2445                 err = f2fs_convert_inline_inode(inode);
2446                 if (err)
2447                         goto fail;
2448         }
2449 repeat:
2450         /*
2451          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2452          * wait_for_stable_page. Will wait that below with our IO control.
2453          */
2454         page = f2fs_pagecache_get_page(mapping, index,
2455                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2456         if (!page) {
2457                 err = -ENOMEM;
2458                 goto fail;
2459         }
2460
2461         *pagep = page;
2462
2463         err = prepare_write_begin(sbi, page, pos, len,
2464                                         &blkaddr, &need_balance);
2465         if (err)
2466                 goto fail;
2467
2468         if (need_balance && !IS_NOQUOTA(inode) &&
2469                         has_not_enough_free_secs(sbi, 0, 0)) {
2470                 unlock_page(page);
2471                 f2fs_balance_fs(sbi, true);
2472                 lock_page(page);
2473                 if (page->mapping != mapping) {
2474                         /* The page got truncated from under us */
2475                         f2fs_put_page(page, 1);
2476                         goto repeat;
2477                 }
2478         }
2479
2480         f2fs_wait_on_page_writeback(page, DATA, false, true);
2481
2482         if (len == PAGE_SIZE || PageUptodate(page))
2483                 return 0;
2484
2485         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2486                 zero_user_segment(page, len, PAGE_SIZE);
2487                 return 0;
2488         }
2489
2490         if (blkaddr == NEW_ADDR) {
2491                 zero_user_segment(page, 0, PAGE_SIZE);
2492                 SetPageUptodate(page);
2493         } else {
2494                 err = f2fs_submit_page_read(inode, page, blkaddr);
2495                 if (err)
2496                         goto fail;
2497
2498                 lock_page(page);
2499                 if (unlikely(page->mapping != mapping)) {
2500                         f2fs_put_page(page, 1);
2501                         goto repeat;
2502                 }
2503                 if (unlikely(!PageUptodate(page))) {
2504                         err = -EIO;
2505                         goto fail;
2506                 }
2507         }
2508         return 0;
2509
2510 fail:
2511         f2fs_put_page(page, 1);
2512         f2fs_write_failed(mapping, pos + len);
2513         if (drop_atomic)
2514                 f2fs_drop_inmem_pages_all(sbi, false);
2515         return err;
2516 }
2517
2518 static int f2fs_write_end(struct file *file,
2519                         struct address_space *mapping,
2520                         loff_t pos, unsigned len, unsigned copied,
2521                         struct page *page, void *fsdata)
2522 {
2523         struct inode *inode = page->mapping->host;
2524
2525         trace_f2fs_write_end(inode, pos, len, copied);
2526
2527         /*
2528          * This should be come from len == PAGE_SIZE, and we expect copied
2529          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2530          * let generic_perform_write() try to copy data again through copied=0.
2531          */
2532         if (!PageUptodate(page)) {
2533                 if (unlikely(copied != len))
2534                         copied = 0;
2535                 else
2536                         SetPageUptodate(page);
2537         }
2538         if (!copied)
2539                 goto unlock_out;
2540
2541         set_page_dirty(page);
2542
2543         if (pos + copied > i_size_read(inode))
2544                 f2fs_i_size_write(inode, pos + copied);
2545 unlock_out:
2546         f2fs_put_page(page, 1);
2547         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2548         return copied;
2549 }
2550
2551 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2552                            loff_t offset)
2553 {
2554         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2555         unsigned blkbits = i_blkbits;
2556         unsigned blocksize_mask = (1 << blkbits) - 1;
2557         unsigned long align = offset | iov_iter_alignment(iter);
2558         struct block_device *bdev = inode->i_sb->s_bdev;
2559
2560         if (align & blocksize_mask) {
2561                 if (bdev)
2562                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2563                 blocksize_mask = (1 << blkbits) - 1;
2564                 if (align & blocksize_mask)
2565                         return -EINVAL;
2566                 return 1;
2567         }
2568         return 0;
2569 }
2570
2571 static void f2fs_dio_end_io(struct bio *bio)
2572 {
2573         struct f2fs_private_dio *dio = bio->bi_private;
2574
2575         dec_page_count(F2FS_I_SB(dio->inode),
2576                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2577
2578         bio->bi_private = dio->orig_private;
2579         bio->bi_end_io = dio->orig_end_io;
2580
2581         kvfree(dio);
2582
2583         bio_endio(bio);
2584 }
2585
2586 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2587                                                         loff_t file_offset)
2588 {
2589         struct f2fs_private_dio *dio;
2590         bool write = (bio_op(bio) == REQ_OP_WRITE);
2591
2592         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2593                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2594         if (!dio)
2595                 goto out;
2596
2597         dio->inode = inode;
2598         dio->orig_end_io = bio->bi_end_io;
2599         dio->orig_private = bio->bi_private;
2600         dio->write = write;
2601
2602         bio->bi_end_io = f2fs_dio_end_io;
2603         bio->bi_private = dio;
2604
2605         inc_page_count(F2FS_I_SB(inode),
2606                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2607
2608         submit_bio(bio);
2609         return;
2610 out:
2611         bio->bi_status = BLK_STS_IOERR;
2612         bio_endio(bio);
2613 }
2614
2615 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2616 {
2617         struct address_space *mapping = iocb->ki_filp->f_mapping;
2618         struct inode *inode = mapping->host;
2619         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2620         struct f2fs_inode_info *fi = F2FS_I(inode);
2621         size_t count = iov_iter_count(iter);
2622         loff_t offset = iocb->ki_pos;
2623         int rw = iov_iter_rw(iter);
2624         int err;
2625         enum rw_hint hint = iocb->ki_hint;
2626         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2627         bool do_opu;
2628
2629         err = check_direct_IO(inode, iter, offset);
2630         if (err)
2631                 return err < 0 ? err : 0;
2632
2633         if (f2fs_force_buffered_io(inode, iocb, iter))
2634                 return 0;
2635
2636         do_opu = allow_outplace_dio(inode, iocb, iter);
2637
2638         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2639
2640         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2641                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2642
2643         if (iocb->ki_flags & IOCB_NOWAIT) {
2644                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2645                         iocb->ki_hint = hint;
2646                         err = -EAGAIN;
2647                         goto out;
2648                 }
2649                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2650                         up_read(&fi->i_gc_rwsem[rw]);
2651                         iocb->ki_hint = hint;
2652                         err = -EAGAIN;
2653                         goto out;
2654                 }
2655         } else {
2656                 down_read(&fi->i_gc_rwsem[rw]);
2657                 if (do_opu)
2658                         down_read(&fi->i_gc_rwsem[READ]);
2659         }
2660
2661         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2662                         iter, rw == WRITE ? get_data_block_dio_write :
2663                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2664                         DIO_LOCKING | DIO_SKIP_HOLES);
2665
2666         if (do_opu)
2667                 up_read(&fi->i_gc_rwsem[READ]);
2668
2669         up_read(&fi->i_gc_rwsem[rw]);
2670
2671         if (rw == WRITE) {
2672                 if (whint_mode == WHINT_MODE_OFF)
2673                         iocb->ki_hint = hint;
2674                 if (err > 0) {
2675                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2676                                                                         err);
2677                         if (!do_opu)
2678                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2679                 } else if (err < 0) {
2680                         f2fs_write_failed(mapping, offset + count);
2681                 }
2682         }
2683
2684 out:
2685         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2686
2687         return err;
2688 }
2689
2690 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2691                                                         unsigned int length)
2692 {
2693         struct inode *inode = page->mapping->host;
2694         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2695
2696         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2697                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2698                 return;
2699
2700         if (PageDirty(page)) {
2701                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2702                         dec_page_count(sbi, F2FS_DIRTY_META);
2703                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2704                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2705                 } else {
2706                         inode_dec_dirty_pages(inode);
2707                         f2fs_remove_dirty_inode(inode);
2708                 }
2709         }
2710
2711         clear_cold_data(page);
2712
2713         if (IS_ATOMIC_WRITTEN_PAGE(page))
2714                 return f2fs_drop_inmem_page(inode, page);
2715
2716         f2fs_clear_page_private(page);
2717 }
2718
2719 int f2fs_release_page(struct page *page, gfp_t wait)
2720 {
2721         /* If this is dirty page, keep PagePrivate */
2722         if (PageDirty(page))
2723                 return 0;
2724
2725         /* This is atomic written page, keep Private */
2726         if (IS_ATOMIC_WRITTEN_PAGE(page))
2727                 return 0;
2728
2729         clear_cold_data(page);
2730         f2fs_clear_page_private(page);
2731         return 1;
2732 }
2733
2734 static int f2fs_set_data_page_dirty(struct page *page)
2735 {
2736         struct address_space *mapping = page->mapping;
2737         struct inode *inode = mapping->host;
2738
2739         trace_f2fs_set_page_dirty(page, DATA);
2740
2741         if (!PageUptodate(page))
2742                 SetPageUptodate(page);
2743
2744         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2745                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2746                         f2fs_register_inmem_page(inode, page);
2747                         return 1;
2748                 }
2749                 /*
2750                  * Previously, this page has been registered, we just
2751                  * return here.
2752                  */
2753                 return 0;
2754         }
2755
2756         if (!PageDirty(page)) {
2757                 __set_page_dirty_nobuffers(page);
2758                 f2fs_update_dirty_page(inode, page);
2759                 return 1;
2760         }
2761         return 0;
2762 }
2763
2764 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2765 {
2766         struct inode *inode = mapping->host;
2767
2768         if (f2fs_has_inline_data(inode))
2769                 return 0;
2770
2771         /* make sure allocating whole blocks */
2772         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2773                 filemap_write_and_wait(mapping);
2774
2775         return generic_block_bmap(mapping, block, get_data_block_bmap);
2776 }
2777
2778 #ifdef CONFIG_MIGRATION
2779 #include <linux/migrate.h>
2780
2781 int f2fs_migrate_page(struct address_space *mapping,
2782                 struct page *newpage, struct page *page, enum migrate_mode mode)
2783 {
2784         int rc, extra_count;
2785         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2786         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2787
2788         BUG_ON(PageWriteback(page));
2789
2790         /* migrating an atomic written page is safe with the inmem_lock hold */
2791         if (atomic_written) {
2792                 if (mode != MIGRATE_SYNC)
2793                         return -EBUSY;
2794                 if (!mutex_trylock(&fi->inmem_lock))
2795                         return -EAGAIN;
2796         }
2797
2798         /* one extra reference was held for atomic_write page */
2799         extra_count = atomic_written ? 1 : 0;
2800         rc = migrate_page_move_mapping(mapping, newpage,
2801                                 page, mode, extra_count);
2802         if (rc != MIGRATEPAGE_SUCCESS) {
2803                 if (atomic_written)
2804                         mutex_unlock(&fi->inmem_lock);
2805                 return rc;
2806         }
2807
2808         if (atomic_written) {
2809                 struct inmem_pages *cur;
2810                 list_for_each_entry(cur, &fi->inmem_pages, list)
2811                         if (cur->page == page) {
2812                                 cur->page = newpage;
2813                                 break;
2814                         }
2815                 mutex_unlock(&fi->inmem_lock);
2816                 put_page(page);
2817                 get_page(newpage);
2818         }
2819
2820         if (PagePrivate(page)) {
2821                 f2fs_set_page_private(newpage, page_private(page));
2822                 f2fs_clear_page_private(page);
2823         }
2824
2825         if (mode != MIGRATE_SYNC_NO_COPY)
2826                 migrate_page_copy(newpage, page);
2827         else
2828                 migrate_page_states(newpage, page);
2829
2830         return MIGRATEPAGE_SUCCESS;
2831 }
2832 #endif
2833
2834 const struct address_space_operations f2fs_dblock_aops = {
2835         .readpage       = f2fs_read_data_page,
2836         .readpages      = f2fs_read_data_pages,
2837         .writepage      = f2fs_write_data_page,
2838         .writepages     = f2fs_write_data_pages,
2839         .write_begin    = f2fs_write_begin,
2840         .write_end      = f2fs_write_end,
2841         .set_page_dirty = f2fs_set_data_page_dirty,
2842         .invalidatepage = f2fs_invalidate_page,
2843         .releasepage    = f2fs_release_page,
2844         .direct_IO      = f2fs_direct_IO,
2845         .bmap           = f2fs_bmap,
2846 #ifdef CONFIG_MIGRATION
2847         .migratepage    = f2fs_migrate_page,
2848 #endif
2849 };
2850
2851 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2852 {
2853         struct address_space *mapping = page_mapping(page);
2854         unsigned long flags;
2855
2856         xa_lock_irqsave(&mapping->i_pages, flags);
2857         __xa_clear_mark(&mapping->i_pages, page_index(page),
2858                                                 PAGECACHE_TAG_DIRTY);
2859         xa_unlock_irqrestore(&mapping->i_pages, flags);
2860 }
2861
2862 int __init f2fs_init_post_read_processing(void)
2863 {
2864         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2865         if (!bio_post_read_ctx_cache)
2866                 goto fail;
2867         bio_post_read_ctx_pool =
2868                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2869                                          bio_post_read_ctx_cache);
2870         if (!bio_post_read_ctx_pool)
2871                 goto fail_free_cache;
2872         return 0;
2873
2874 fail_free_cache:
2875         kmem_cache_destroy(bio_post_read_ctx_cache);
2876 fail:
2877         return -ENOMEM;
2878 }
2879
2880 void __exit f2fs_destroy_post_read_processing(void)
2881 {
2882         mempool_destroy(bio_post_read_ctx_pool);
2883         kmem_cache_destroy(bio_post_read_ctx_cache);
2884 }