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/swap.h>
  18 #include <linux/prefetch.h>
  19 #include <linux/uio.h>
  20 #include <linux/cleancache.h>
  21 #include <linux/sched/signal.h>
  22
  23 #include "f2fs.h"
  24 #include "node.h"
  25 #include "segment.h"
  26 #include "trace.h"
  27 #include <trace/events/f2fs.h>
  28
  29 #define NUM_PREALLOC_POST_READ_CTXS     128
  30
  31 static struct kmem_cache *bio_post_read_ctx_cache;
  32 static mempool_t *bio_post_read_ctx_pool;
  33
  34 static bool __is_cp_guaranteed(struct page *page)
  35 {
  36         struct address_space *mapping = page->mapping;
  37         struct inode *inode;
  38         struct f2fs_sb_info *sbi;
  39
  40         if (!mapping)
  41                 return false;
  42
  43         inode = mapping->host;
  44         sbi = F2FS_I_SB(inode);
  45
  46         if (inode->i_ino == F2FS_META_INO(sbi) ||
  47                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  48                         S_ISDIR(inode->i_mode) ||
  49                         (S_ISREG(inode->i_mode) &&
  50                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  51                         is_cold_data(page))
  52                 return true;
  53         return false;
  54 }
  55
  56 static enum count_type __read_io_type(struct page *page)
  57 {
  58         struct address_space *mapping = page_file_mapping(page);
  59
  60         if (mapping) {
  61                 struct inode *inode = mapping->host;
  62                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  63
  64                 if (inode->i_ino == F2FS_META_INO(sbi))
  65                         return F2FS_RD_META;
  66
  67                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  68                         return F2FS_RD_NODE;
  69         }
  70         return F2FS_RD_DATA;
  71 }
  72
  73 /* postprocessing steps for read bios */
  74 enum bio_post_read_step {
  75         STEP_INITIAL = 0,
  76         STEP_DECRYPT,
  77 };
  78
  79 struct bio_post_read_ctx {
  80         struct bio *bio;
  81         struct work_struct work;
  82         unsigned int cur_step;
  83         unsigned int enabled_steps;
  84 };
  85
  86 static void __read_end_io(struct bio *bio)
  87 {
  88         struct page *page;
  89         struct bio_vec *bv;
  90         struct bvec_iter_all iter_all;
  91
  92         bio_for_each_segment_all(bv, bio, iter_all) {
  93                 page = bv->bv_page;
  94
  95                 /* PG_error was set if any post_read step failed */
  96                 if (bio->bi_status || PageError(page)) {
  97                         ClearPageUptodate(page);
  98                         /* will re-read again later */
  99                         ClearPageError(page);
 100                 } else {
 101                         SetPageUptodate(page);
 102                 }
 103                 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 104                 unlock_page(page);
 105         }
 106         if (bio->bi_private)
 107                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 108         bio_put(bio);
 109 }
 110
 111 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 112
 113 static void decrypt_work(struct work_struct *work)
 114 {
 115         struct bio_post_read_ctx *ctx =
 116                 container_of(work, struct bio_post_read_ctx, work);
 117
 118         fscrypt_decrypt_bio(ctx->bio);
 119
 120         bio_post_read_processing(ctx);
 121 }
 122
 123 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 124 {
 125         switch (++ctx->cur_step) {
 126         case STEP_DECRYPT:
 127                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 128                         INIT_WORK(&ctx->work, decrypt_work);
 129                         fscrypt_enqueue_decrypt_work(&ctx->work);
 130                         return;
 131                 }
 132                 ctx->cur_step++;
 133                 /* fall-through */
 134         default:
 135                 __read_end_io(ctx->bio);
 136         }
 137 }
 138
 139 static bool f2fs_bio_post_read_required(struct bio *bio)
 140 {
 141         return bio->bi_private && !bio->bi_status;
 142 }
 143
 144 static void f2fs_read_end_io(struct bio *bio)
 145 {
 146         if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
 147                                                 FAULT_READ_IO)) {
 148                 f2fs_show_injection_info(FAULT_READ_IO);
 149                 bio->bi_status = BLK_STS_IOERR;
 150         }
 151
 152         if (f2fs_bio_post_read_required(bio)) {
 153                 struct bio_post_read_ctx *ctx = bio->bi_private;
 154
 155                 ctx->cur_step = STEP_INITIAL;
 156                 bio_post_read_processing(ctx);
 157                 return;
 158         }
 159
 160         __read_end_io(bio);
 161 }
 162
 163 static void f2fs_write_end_io(struct bio *bio)
 164 {
 165         struct f2fs_sb_info *sbi = bio->bi_private;
 166         struct bio_vec *bvec;
 167         struct bvec_iter_all iter_all;
 168
 169         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 170                 f2fs_show_injection_info(FAULT_WRITE_IO);
 171                 bio->bi_status = BLK_STS_IOERR;
 172         }
 173
 174         bio_for_each_segment_all(bvec, bio, iter_all) {
 175                 struct page *page = bvec->bv_page;
 176                 enum count_type type = WB_DATA_TYPE(page);
 177
 178                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 179                         set_page_private(page, (unsigned long)NULL);
 180                         ClearPagePrivate(page);
 181                         unlock_page(page);
 182                         mempool_free(page, sbi->write_io_dummy);
 183
 184                         if (unlikely(bio->bi_status))
 185                                 f2fs_stop_checkpoint(sbi, true);
 186                         continue;
 187                 }
 188
 189                 fscrypt_pullback_bio_page(&page, true);
 190
 191                 if (unlikely(bio->bi_status)) {
 192                         mapping_set_error(page->mapping, -EIO);
 193                         if (type == F2FS_WB_CP_DATA)
 194                                 f2fs_stop_checkpoint(sbi, true);
 195                 }
 196
 197                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 198                                         page->index != nid_of_node(page));
 199
 200                 dec_page_count(sbi, type);
 201                 if (f2fs_in_warm_node_list(sbi, page))
 202                         f2fs_del_fsync_node_entry(sbi, page);
 203                 clear_cold_data(page);
 204                 end_page_writeback(page);
 205         }
 206         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 207                                 wq_has_sleeper(&sbi->cp_wait))
 208                 wake_up(&sbi->cp_wait);
 209
 210         bio_put(bio);
 211 }
 212
 213 /*
 214  * Return true, if pre_bio's bdev is same as its target device.
 215  */
 216 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 217                                 block_t blk_addr, struct bio *bio)
 218 {
 219         struct block_device *bdev = sbi->sb->s_bdev;
 220         int i;
 221
 222         if (f2fs_is_multi_device(sbi)) {
 223                 for (i = 0; i < sbi->s_ndevs; i++) {
 224                         if (FDEV(i).start_blk <= blk_addr &&
 225                             FDEV(i).end_blk >= blk_addr) {
 226                                 blk_addr -= FDEV(i).start_blk;
 227                                 bdev = FDEV(i).bdev;
 228                                 break;
 229                         }
 230                 }
 231         }
 232         if (bio) {
 233                 bio_set_dev(bio, bdev);
 234                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 235         }
 236         return bdev;
 237 }
 238
 239 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 240 {
 241         int i;
 242
 243         if (!f2fs_is_multi_device(sbi))
 244                 return 0;
 245
 246         for (i = 0; i < sbi->s_ndevs; i++)
 247                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 248                         return i;
 249         return 0;
 250 }
 251
 252 static bool __same_bdev(struct f2fs_sb_info *sbi,
 253                                 block_t blk_addr, struct bio *bio)
 254 {
 255         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 256         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 257 }
 258
 259 /*
 260  * Low-level block read/write IO operations.
 261  */
 262 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 263                                 struct writeback_control *wbc,
 264                                 int npages, bool is_read,
 265                                 enum page_type type, enum temp_type temp)
 266 {
 267         struct bio *bio;
 268
 269         bio = f2fs_bio_alloc(sbi, npages, true);
 270
 271         f2fs_target_device(sbi, blk_addr, bio);
 272         if (is_read) {
 273                 bio->bi_end_io = f2fs_read_end_io;
 274                 bio->bi_private = NULL;
 275         } else {
 276                 bio->bi_end_io = f2fs_write_end_io;
 277                 bio->bi_private = sbi;
 278                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 279         }
 280         if (wbc)
 281                 wbc_init_bio(wbc, bio);
 282
 283         return bio;
 284 }
 285
 286 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 287                                 struct bio *bio, enum page_type type)
 288 {
 289         if (!is_read_io(bio_op(bio))) {
 290                 unsigned int start;
 291
 292                 if (type != DATA && type != NODE)
 293                         goto submit_io;
 294
 295                 if (test_opt(sbi, LFS) && current->plug)
 296                         blk_finish_plug(current->plug);
 297
 298                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 299                 start %= F2FS_IO_SIZE(sbi);
 300
 301                 if (start == 0)
 302                         goto submit_io;
 303
 304                 /* fill dummy pages */
 305                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 306                         struct page *page =
 307                                 mempool_alloc(sbi->write_io_dummy,
 308                                               GFP_NOIO | __GFP_NOFAIL);
 309                         f2fs_bug_on(sbi, !page);
 310
 311                         zero_user_segment(page, 0, PAGE_SIZE);
 312                         SetPagePrivate(page);
 313                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 314                         lock_page(page);
 315                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 316                                 f2fs_bug_on(sbi, 1);
 317                 }
 318                 /*
 319                  * In the NODE case, we lose next block address chain. So, we
 320                  * need to do checkpoint in f2fs_sync_file.
 321                  */
 322                 if (type == NODE)
 323                         set_sbi_flag(sbi, SBI_NEED_CP);
 324         }
 325 submit_io:
 326         if (is_read_io(bio_op(bio)))
 327                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 328         else
 329                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 330         submit_bio(bio);
 331 }
 332
 333 static void __submit_merged_bio(struct f2fs_bio_info *io)
 334 {
 335         struct f2fs_io_info *fio = &io->fio;
 336
 337         if (!io->bio)
 338                 return;
 339
 340         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 341
 342         if (is_read_io(fio->op))
 343                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 344         else
 345                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 346
 347         __submit_bio(io->sbi, io->bio, fio->type);
 348         io->bio = NULL;
 349 }
 350
 351 static bool __has_merged_page(struct bio *bio, struct inode *inode,
 352                                                 struct page *page, nid_t ino)
 353 {
 354         struct bio_vec *bvec;
 355         struct page *target;
 356         struct bvec_iter_all iter_all;
 357
 358         if (!bio)
 359                 return false;
 360
 361         if (!inode && !page && !ino)
 362                 return true;
 363
 364         bio_for_each_segment_all(bvec, bio, iter_all) {
 365
 366                 if (bvec->bv_page->mapping)
 367                         target = bvec->bv_page;
 368                 else
 369                         target = fscrypt_control_page(bvec->bv_page);
 370
 371                 if (inode && inode == target->mapping->host)
 372                         return true;
 373                 if (page && page == target)
 374                         return true;
 375                 if (ino && ino == ino_of_node(target))
 376                         return true;
 377         }
 378
 379         return false;
 380 }
 381
 382 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 383                                 enum page_type type, enum temp_type temp)
 384 {
 385         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 386         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 387
 388         down_write(&io->io_rwsem);
 389
 390         /* change META to META_FLUSH in the checkpoint procedure */
 391         if (type >= META_FLUSH) {
 392                 io->fio.type = META_FLUSH;
 393                 io->fio.op = REQ_OP_WRITE;
 394                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 395                 if (!test_opt(sbi, NOBARRIER))
 396                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 397         }
 398         __submit_merged_bio(io);
 399         up_write(&io->io_rwsem);
 400 }
 401
 402 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 403                                 struct inode *inode, struct page *page,
 404                                 nid_t ino, enum page_type type, bool force)
 405 {
 406         enum temp_type temp;
 407         bool ret = true;
 408
 409         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 410                 if (!force)     {
 411                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 412                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 413
 414                         down_read(&io->io_rwsem);
 415                         ret = __has_merged_page(io->bio, inode, page, ino);
 416                         up_read(&io->io_rwsem);
 417                 }
 418                 if (ret)
 419                         __f2fs_submit_merged_write(sbi, type, temp);
 420
 421                 /* TODO: use HOT temp only for meta pages now. */
 422                 if (type >= META)
 423                         break;
 424         }
 425 }
 426
 427 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 428 {
 429         __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 430 }
 431
 432 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 433                                 struct inode *inode, struct page *page,
 434                                 nid_t ino, enum page_type type)
 435 {
 436         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 437 }
 438
 439 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 440 {
 441         f2fs_submit_merged_write(sbi, DATA);
 442         f2fs_submit_merged_write(sbi, NODE);
 443         f2fs_submit_merged_write(sbi, META);
 444 }
 445
 446 /*
 447  * Fill the locked page with data located in the block address.
 448  * A caller needs to unlock the page on failure.
 449  */
 450 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 451 {
 452         struct bio *bio;
 453         struct page *page = fio->encrypted_page ?
 454                         fio->encrypted_page : fio->page;
 455
 456         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 457                         fio->is_por ? META_POR : (__is_meta_io(fio) ?
 458                         META_GENERIC : DATA_GENERIC_ENHANCE)))
 459                 return -EFSCORRUPTED;
 460
 461         trace_f2fs_submit_page_bio(page, fio);
 462         f2fs_trace_ios(fio, 0);
 463
 464         /* Allocate a new bio */
 465         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 466                                 1, is_read_io(fio->op), fio->type, fio->temp);
 467
 468         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 469                 bio_put(bio);
 470                 return -EFAULT;
 471         }
 472
 473         if (fio->io_wbc && !is_read_io(fio->op))
 474                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 475
 476         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 477
 478         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 479                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 480
 481         __submit_bio(fio->sbi, bio, fio->type);
 482         return 0;
 483 }
 484
 485 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 486 {
 487         struct bio *bio = *fio->bio;
 488         struct page *page = fio->encrypted_page ?
 489                         fio->encrypted_page : fio->page;
 490
 491         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 492                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 493                 return -EFSCORRUPTED;
 494
 495         trace_f2fs_submit_page_bio(page, fio);
 496         f2fs_trace_ios(fio, 0);
 497
 498         if (bio && (*fio->last_block + 1 != fio->new_blkaddr ||
 499                         !__same_bdev(fio->sbi, fio->new_blkaddr, bio))) {
 500                 __submit_bio(fio->sbi, bio, fio->type);
 501                 bio = NULL;
 502         }
 503 alloc_new:
 504         if (!bio) {
 505                 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 506                                 BIO_MAX_PAGES, false, fio->type, fio->temp);
 507                 bio_set_op_attrs(bio, fio->op, fio->op_flags);
 508         }
 509
 510         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 511                 __submit_bio(fio->sbi, bio, fio->type);
 512                 bio = NULL;
 513                 goto alloc_new;
 514         }
 515
 516         if (fio->io_wbc)
 517                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 518
 519         inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 520
 521         *fio->last_block = fio->new_blkaddr;
 522         *fio->bio = bio;
 523
 524         return 0;
 525 }
 526
 527 static void f2fs_submit_ipu_bio(struct f2fs_sb_info *sbi, struct bio **bio,
 528                                                         struct page *page)
 529 {
 530         if (!bio)
 531                 return;
 532
 533         if (!__has_merged_page(*bio, NULL, page, 0))
 534                 return;
 535
 536         __submit_bio(sbi, *bio, DATA);
 537         *bio = NULL;
 538 }
 539
 540 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 541 {
 542         struct f2fs_sb_info *sbi = fio->sbi;
 543         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 544         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 545         struct page *bio_page;
 546
 547         f2fs_bug_on(sbi, is_read_io(fio->op));
 548
 549         down_write(&io->io_rwsem);
 550 next:
 551         if (fio->in_list) {
 552                 spin_lock(&io->io_lock);
 553                 if (list_empty(&io->io_list)) {
 554                         spin_unlock(&io->io_lock);
 555                         goto out;
 556                 }
 557                 fio = list_first_entry(&io->io_list,
 558                                                 struct f2fs_io_info, list);
 559                 list_del(&fio->list);
 560                 spin_unlock(&io->io_lock);
 561         }
 562
 563         verify_fio_blkaddr(fio);
 564
 565         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 566
 567         /* set submitted = true as a return value */
 568         fio->submitted = true;
 569
 570         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 571
 572         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 573             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 574                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 575                 __submit_merged_bio(io);
 576 alloc_new:
 577         if (io->bio == NULL) {
 578                 if ((fio->type == DATA || fio->type == NODE) &&
 579                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 580                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 581                         fio->retry = true;
 582                         goto skip;
 583                 }
 584                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 585                                                 BIO_MAX_PAGES, false,
 586                                                 fio->type, fio->temp);
 587                 io->fio = *fio;
 588         }
 589
 590         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 591                 __submit_merged_bio(io);
 592                 goto alloc_new;
 593         }
 594
 595         if (fio->io_wbc)
 596                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 597
 598         io->last_block_in_bio = fio->new_blkaddr;
 599         f2fs_trace_ios(fio, 0);
 600
 601         trace_f2fs_submit_page_write(fio->page, fio);
 602 skip:
 603         if (fio->in_list)
 604                 goto next;
 605 out:
 606         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 607                                 f2fs_is_checkpoint_ready(sbi))
 608                 __submit_merged_bio(io);
 609         up_write(&io->io_rwsem);
 610 }
 611
 612 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 613                                         unsigned nr_pages, unsigned op_flag)
 614 {
 615         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 616         struct bio *bio;
 617         struct bio_post_read_ctx *ctx;
 618         unsigned int post_read_steps = 0;
 619
 620         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 621         if (!bio)
 622                 return ERR_PTR(-ENOMEM);
 623         f2fs_target_device(sbi, blkaddr, bio);
 624         bio->bi_end_io = f2fs_read_end_io;
 625         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 626
 627         if (f2fs_encrypted_file(inode))
 628                 post_read_steps |= 1 << STEP_DECRYPT;
 629         if (post_read_steps) {
 630                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 631                 if (!ctx) {
 632                         bio_put(bio);
 633                         return ERR_PTR(-ENOMEM);
 634                 }
 635                 ctx->bio = bio;
 636                 ctx->enabled_steps = post_read_steps;
 637                 bio->bi_private = ctx;
 638         }
 639
 640         return bio;
 641 }
 642
 643 /* This can handle encryption stuffs */
 644 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 645                                                         block_t blkaddr)
 646 {
 647         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 648         struct bio *bio;
 649
 650         bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 651         if (IS_ERR(bio))
 652                 return PTR_ERR(bio);
 653
 654         /* wait for GCed page writeback via META_MAPPING */
 655         f2fs_wait_on_block_writeback(inode, blkaddr);
 656
 657         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 658                 bio_put(bio);
 659                 return -EFAULT;
 660         }
 661         ClearPageError(page);
 662         inc_page_count(sbi, F2FS_RD_DATA);
 663         __submit_bio(sbi, bio, DATA);
 664         return 0;
 665 }
 666
 667 static void __set_data_blkaddr(struct dnode_of_data *dn)
 668 {
 669         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 670         __le32 *addr_array;
 671         int base = 0;
 672
 673         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 674                 base = get_extra_isize(dn->inode);
 675
 676         /* Get physical address of data block */
 677         addr_array = blkaddr_in_node(rn);
 678         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 679 }
 680
 681 /*
 682  * Lock ordering for the change of data block address:
 683  * ->data_page
 684  *  ->node_page
 685  *    update block addresses in the node page
 686  */
 687 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 688 {
 689         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 690         __set_data_blkaddr(dn);
 691         if (set_page_dirty(dn->node_page))
 692                 dn->node_changed = true;
 693 }
 694
 695 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 696 {
 697         dn->data_blkaddr = blkaddr;
 698         f2fs_set_data_blkaddr(dn);
 699         f2fs_update_extent_cache(dn);
 700 }
 701
 702 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 703 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 704 {
 705         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 706         int err;
 707
 708         if (!count)
 709                 return 0;
 710
 711         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 712                 return -EPERM;
 713         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 714                 return err;
 715
 716         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 717                                                 dn->ofs_in_node, count);
 718
 719         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 720
 721         for (; count > 0; dn->ofs_in_node++) {
 722                 block_t blkaddr = datablock_addr(dn->inode,
 723                                         dn->node_page, dn->ofs_in_node);
 724                 if (blkaddr == NULL_ADDR) {
 725                         dn->data_blkaddr = NEW_ADDR;
 726                         __set_data_blkaddr(dn);
 727                         count--;
 728                 }
 729         }
 730
 731         if (set_page_dirty(dn->node_page))
 732                 dn->node_changed = true;
 733         return 0;
 734 }
 735
 736 /* Should keep dn->ofs_in_node unchanged */
 737 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 738 {
 739         unsigned int ofs_in_node = dn->ofs_in_node;
 740         int ret;
 741
 742         ret = f2fs_reserve_new_blocks(dn, 1);
 743         dn->ofs_in_node = ofs_in_node;
 744         return ret;
 745 }
 746
 747 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 748 {
 749         bool need_put = dn->inode_page ? false : true;
 750         int err;
 751
 752         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 753         if (err)
 754                 return err;
 755
 756         if (dn->data_blkaddr == NULL_ADDR)
 757                 err = f2fs_reserve_new_block(dn);
 758         if (err || need_put)
 759                 f2fs_put_dnode(dn);
 760         return err;
 761 }
 762
 763 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 764 {
 765         struct extent_info ei  = {0,0,0};
 766         struct inode *inode = dn->inode;
 767
 768         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 769                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 770                 return 0;
 771         }
 772
 773         return f2fs_reserve_block(dn, index);
 774 }
 775
 776 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 777                                                 int op_flags, bool for_write)
 778 {
 779         struct address_space *mapping = inode->i_mapping;
 780         struct dnode_of_data dn;
 781         struct page *page;
 782         struct extent_info ei = {0,0,0};
 783         int err;
 784
 785         page = f2fs_grab_cache_page(mapping, index, for_write);
 786         if (!page)
 787                 return ERR_PTR(-ENOMEM);
 788
 789         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 790                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 791                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
 792                                                 DATA_GENERIC_ENHANCE_READ)) {
 793                         err = -EFSCORRUPTED;
 794                         goto put_err;
 795                 }
 796                 goto got_it;
 797         }
 798
 799         set_new_dnode(&dn, inode, NULL, NULL, 0);
 800         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 801         if (err)
 802                 goto put_err;
 803         f2fs_put_dnode(&dn);
 804
 805         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 806                 err = -ENOENT;
 807                 goto put_err;
 808         }
 809         if (dn.data_blkaddr != NEW_ADDR &&
 810                         !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
 811                                                 dn.data_blkaddr,
 812                                                 DATA_GENERIC_ENHANCE)) {
 813                 err = -EFSCORRUPTED;
 814                 goto put_err;
 815         }
 816 got_it:
 817         if (PageUptodate(page)) {
 818                 unlock_page(page);
 819                 return page;
 820         }
 821
 822         /*
 823          * A new dentry page is allocated but not able to be written, since its
 824          * new inode page couldn't be allocated due to -ENOSPC.
 825          * In such the case, its blkaddr can be remained as NEW_ADDR.
 826          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 827          * f2fs_init_inode_metadata.
 828          */
 829         if (dn.data_blkaddr == NEW_ADDR) {
 830                 zero_user_segment(page, 0, PAGE_SIZE);
 831                 if (!PageUptodate(page))
 832                         SetPageUptodate(page);
 833                 unlock_page(page);
 834                 return page;
 835         }
 836
 837         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 838         if (err)
 839                 goto put_err;
 840         return page;
 841
 842 put_err:
 843         f2fs_put_page(page, 1);
 844         return ERR_PTR(err);
 845 }
 846
 847 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 848 {
 849         struct address_space *mapping = inode->i_mapping;
 850         struct page *page;
 851
 852         page = find_get_page(mapping, index);
 853         if (page && PageUptodate(page))
 854                 return page;
 855         f2fs_put_page(page, 0);
 856
 857         page = f2fs_get_read_data_page(inode, index, 0, false);
 858         if (IS_ERR(page))
 859                 return page;
 860
 861         if (PageUptodate(page))
 862                 return page;
 863
 864         wait_on_page_locked(page);
 865         if (unlikely(!PageUptodate(page))) {
 866                 f2fs_put_page(page, 0);
 867                 return ERR_PTR(-EIO);
 868         }
 869         return page;
 870 }
 871
 872 /*
 873  * If it tries to access a hole, return an error.
 874  * Because, the callers, functions in dir.c and GC, should be able to know
 875  * whether this page exists or not.
 876  */
 877 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 878                                                         bool for_write)
 879 {
 880         struct address_space *mapping = inode->i_mapping;
 881         struct page *page;
 882 repeat:
 883         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 884         if (IS_ERR(page))
 885                 return page;
 886
 887         /* wait for read completion */
 888         lock_page(page);
 889         if (unlikely(page->mapping != mapping)) {
 890                 f2fs_put_page(page, 1);
 891                 goto repeat;
 892         }
 893         if (unlikely(!PageUptodate(page))) {
 894                 f2fs_put_page(page, 1);
 895                 return ERR_PTR(-EIO);
 896         }
 897         return page;
 898 }
 899
 900 /*
 901  * Caller ensures that this data page is never allocated.
 902  * A new zero-filled data page is allocated in the page cache.
 903  *
 904  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 905  * f2fs_unlock_op().
 906  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 907  * ipage should be released by this function.
 908  */
 909 struct page *f2fs_get_new_data_page(struct inode *inode,
 910                 struct page *ipage, pgoff_t index, bool new_i_size)
 911 {
 912         struct address_space *mapping = inode->i_mapping;
 913         struct page *page;
 914         struct dnode_of_data dn;
 915         int err;
 916
 917         page = f2fs_grab_cache_page(mapping, index, true);
 918         if (!page) {
 919                 /*
 920                  * before exiting, we should make sure ipage will be released
 921                  * if any error occur.
 922                  */
 923                 f2fs_put_page(ipage, 1);
 924                 return ERR_PTR(-ENOMEM);
 925         }
 926
 927         set_new_dnode(&dn, inode, ipage, NULL, 0);
 928         err = f2fs_reserve_block(&dn, index);
 929         if (err) {
 930                 f2fs_put_page(page, 1);
 931                 return ERR_PTR(err);
 932         }
 933         if (!ipage)
 934                 f2fs_put_dnode(&dn);
 935
 936         if (PageUptodate(page))
 937                 goto got_it;
 938
 939         if (dn.data_blkaddr == NEW_ADDR) {
 940                 zero_user_segment(page, 0, PAGE_SIZE);
 941                 if (!PageUptodate(page))
 942                         SetPageUptodate(page);
 943         } else {
 944                 f2fs_put_page(page, 1);
 945
 946                 /* if ipage exists, blkaddr should be NEW_ADDR */
 947                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 948                 page = f2fs_get_lock_data_page(inode, index, true);
 949                 if (IS_ERR(page))
 950                         return page;
 951         }
 952 got_it:
 953         if (new_i_size && i_size_read(inode) <
 954                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 955                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 956         return page;
 957 }
 958
 959 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 960 {
 961         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 962         struct f2fs_summary sum;
 963         struct node_info ni;
 964         block_t old_blkaddr;
 965         blkcnt_t count = 1;
 966         int err;
 967
 968         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 969                 return -EPERM;
 970
 971         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 972         if (err)
 973                 return err;
 974
 975         dn->data_blkaddr = datablock_addr(dn->inode,
 976                                 dn->node_page, dn->ofs_in_node);
 977         if (dn->data_blkaddr != NULL_ADDR)
 978                 goto alloc;
 979
 980         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 981                 return err;
 982
 983 alloc:
 984         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 985         old_blkaddr = dn->data_blkaddr;
 986         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 987                                         &sum, seg_type, NULL, false);
 988         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 989                 invalidate_mapping_pages(META_MAPPING(sbi),
 990                                         old_blkaddr, old_blkaddr);
 991         f2fs_set_data_blkaddr(dn);
 992
 993         /*
 994          * i_size will be updated by direct_IO. Otherwise, we'll get stale
 995          * data from unwritten block via dio_read.
 996          */
 997         return 0;
 998 }
 999
1000 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1001 {
1002         struct inode *inode = file_inode(iocb->ki_filp);
1003         struct f2fs_map_blocks map;
1004         int flag;
1005         int err = 0;
1006         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1007
1008         /* convert inline data for Direct I/O*/
1009         if (direct_io) {
1010                 err = f2fs_convert_inline_inode(inode);
1011                 if (err)
1012                         return err;
1013         }
1014
1015         if (direct_io && allow_outplace_dio(inode, iocb, from))
1016                 return 0;
1017
1018         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
1019                 return 0;
1020
1021         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1022         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1023         if (map.m_len > map.m_lblk)
1024                 map.m_len -= map.m_lblk;
1025         else
1026                 map.m_len = 0;
1027
1028         map.m_next_pgofs = NULL;
1029         map.m_next_extent = NULL;
1030         map.m_seg_type = NO_CHECK_TYPE;
1031         map.m_may_create = true;
1032
1033         if (direct_io) {
1034                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1035                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1036                                         F2FS_GET_BLOCK_PRE_AIO :
1037                                         F2FS_GET_BLOCK_PRE_DIO;
1038                 goto map_blocks;
1039         }
1040         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1041                 err = f2fs_convert_inline_inode(inode);
1042                 if (err)
1043                         return err;
1044         }
1045         if (f2fs_has_inline_data(inode))
1046                 return err;
1047
1048         flag = F2FS_GET_BLOCK_PRE_AIO;
1049
1050 map_blocks:
1051         err = f2fs_map_blocks(inode, &map, 1, flag);
1052         if (map.m_len > 0 && err == -ENOSPC) {
1053                 if (!direct_io)
1054                         set_inode_flag(inode, FI_NO_PREALLOC);
1055                 err = 0;
1056         }
1057         return err;
1058 }
1059
1060 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1061 {
1062         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1063                 if (lock)
1064                         down_read(&sbi->node_change);
1065                 else
1066                         up_read(&sbi->node_change);
1067         } else {
1068                 if (lock)
1069                         f2fs_lock_op(sbi);
1070                 else
1071                         f2fs_unlock_op(sbi);
1072         }
1073 }
1074
1075 /*
1076  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1077  * f2fs_map_blocks structure.
1078  * If original data blocks are allocated, then give them to blockdev.
1079  * Otherwise,
1080  *     a. preallocate requested block addresses
1081  *     b. do not use extent cache for better performance
1082  *     c. give the block addresses to blockdev
1083  */
1084 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1085                                                 int create, int flag)
1086 {
1087         unsigned int maxblocks = map->m_len;
1088         struct dnode_of_data dn;
1089         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1090         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1091         pgoff_t pgofs, end_offset, end;
1092         int err = 0, ofs = 1;
1093         unsigned int ofs_in_node, last_ofs_in_node;
1094         blkcnt_t prealloc;
1095         struct extent_info ei = {0,0,0};
1096         block_t blkaddr;
1097         unsigned int start_pgofs;
1098
1099         if (!maxblocks)
1100                 return 0;
1101
1102         map->m_len = 0;
1103         map->m_flags = 0;
1104
1105         /* it only supports block size == page size */
1106         pgofs = (pgoff_t)map->m_lblk;
1107         end = pgofs + maxblocks;
1108
1109         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1110                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1111                                                         map->m_may_create)
1112                         goto next_dnode;
1113
1114                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1115                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1116                 map->m_flags = F2FS_MAP_MAPPED;
1117                 if (map->m_next_extent)
1118                         *map->m_next_extent = pgofs + map->m_len;
1119
1120                 /* for hardware encryption, but to avoid potential issue in future */
1121                 if (flag == F2FS_GET_BLOCK_DIO)
1122                         f2fs_wait_on_block_writeback_range(inode,
1123                                                 map->m_pblk, map->m_len);
1124                 goto out;
1125         }
1126
1127 next_dnode:
1128         if (map->m_may_create)
1129                 __do_map_lock(sbi, flag, true);
1130
1131         /* When reading holes, we need its node page */
1132         set_new_dnode(&dn, inode, NULL, NULL, 0);
1133         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1134         if (err) {
1135                 if (flag == F2FS_GET_BLOCK_BMAP)
1136                         map->m_pblk = 0;
1137                 if (err == -ENOENT) {
1138                         err = 0;
1139                         if (map->m_next_pgofs)
1140                                 *map->m_next_pgofs =
1141                                         f2fs_get_next_page_offset(&dn, pgofs);
1142                         if (map->m_next_extent)
1143                                 *map->m_next_extent =
1144                                         f2fs_get_next_page_offset(&dn, pgofs);
1145                 }
1146                 goto unlock_out;
1147         }
1148
1149         start_pgofs = pgofs;
1150         prealloc = 0;
1151         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1152         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1153
1154 next_block:
1155         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1156
1157         if (__is_valid_data_blkaddr(blkaddr) &&
1158                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1159                 err = -EFSCORRUPTED;
1160                 goto sync_out;
1161         }
1162
1163         if (__is_valid_data_blkaddr(blkaddr)) {
1164                 /* use out-place-update for driect IO under LFS mode */
1165                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1166                                                         map->m_may_create) {
1167                         err = __allocate_data_block(&dn, map->m_seg_type);
1168                         if (!err) {
1169                                 blkaddr = dn.data_blkaddr;
1170                                 set_inode_flag(inode, FI_APPEND_WRITE);
1171                         }
1172                 }
1173         } else {
1174                 if (create) {
1175                         if (unlikely(f2fs_cp_error(sbi))) {
1176                                 err = -EIO;
1177                                 goto sync_out;
1178                         }
1179                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1180                                 if (blkaddr == NULL_ADDR) {
1181                                         prealloc++;
1182                                         last_ofs_in_node = dn.ofs_in_node;
1183                                 }
1184                         } else {
1185                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1186                                         flag != F2FS_GET_BLOCK_DIO);
1187                                 err = __allocate_data_block(&dn,
1188                                                         map->m_seg_type);
1189                                 if (!err)
1190                                         set_inode_flag(inode, FI_APPEND_WRITE);
1191                         }
1192                         if (err)
1193                                 goto sync_out;
1194                         map->m_flags |= F2FS_MAP_NEW;
1195                         blkaddr = dn.data_blkaddr;
1196                 } else {
1197                         if (flag == F2FS_GET_BLOCK_BMAP) {
1198                                 map->m_pblk = 0;
1199                                 goto sync_out;
1200                         }
1201                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1202                                 goto sync_out;
1203                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1204                                                 blkaddr == NULL_ADDR) {
1205                                 if (map->m_next_pgofs)
1206                                         *map->m_next_pgofs = pgofs + 1;
1207                                 goto sync_out;
1208                         }
1209                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1210                                 /* for defragment case */
1211                                 if (map->m_next_pgofs)
1212                                         *map->m_next_pgofs = pgofs + 1;
1213                                 goto sync_out;
1214                         }
1215                 }
1216         }
1217
1218         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1219                 goto skip;
1220
1221         if (map->m_len == 0) {
1222                 /* preallocated unwritten block should be mapped for fiemap. */
1223                 if (blkaddr == NEW_ADDR)
1224                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1225                 map->m_flags |= F2FS_MAP_MAPPED;
1226
1227                 map->m_pblk = blkaddr;
1228                 map->m_len = 1;
1229         } else if ((map->m_pblk != NEW_ADDR &&
1230                         blkaddr == (map->m_pblk + ofs)) ||
1231                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1232                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1233                 ofs++;
1234                 map->m_len++;
1235         } else {
1236                 goto sync_out;
1237         }
1238
1239 skip:
1240         dn.ofs_in_node++;
1241         pgofs++;
1242
1243         /* preallocate blocks in batch for one dnode page */
1244         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1245                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1246
1247                 dn.ofs_in_node = ofs_in_node;
1248                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1249                 if (err)
1250                         goto sync_out;
1251
1252                 map->m_len += dn.ofs_in_node - ofs_in_node;
1253                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1254                         err = -ENOSPC;
1255                         goto sync_out;
1256                 }
1257                 dn.ofs_in_node = end_offset;
1258         }
1259
1260         if (pgofs >= end)
1261                 goto sync_out;
1262         else if (dn.ofs_in_node < end_offset)
1263                 goto next_block;
1264
1265         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1266                 if (map->m_flags & F2FS_MAP_MAPPED) {
1267                         unsigned int ofs = start_pgofs - map->m_lblk;
1268
1269                         f2fs_update_extent_cache_range(&dn,
1270                                 start_pgofs, map->m_pblk + ofs,
1271                                 map->m_len - ofs);
1272                 }
1273         }
1274
1275         f2fs_put_dnode(&dn);
1276
1277         if (map->m_may_create) {
1278                 __do_map_lock(sbi, flag, false);
1279                 f2fs_balance_fs(sbi, dn.node_changed);
1280         }
1281         goto next_dnode;
1282
1283 sync_out:
1284
1285         /* for hardware encryption, but to avoid potential issue in future */
1286         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1287                 f2fs_wait_on_block_writeback_range(inode,
1288                                                 map->m_pblk, map->m_len);
1289
1290         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1291                 if (map->m_flags & F2FS_MAP_MAPPED) {
1292                         unsigned int ofs = start_pgofs - map->m_lblk;
1293
1294                         f2fs_update_extent_cache_range(&dn,
1295                                 start_pgofs, map->m_pblk + ofs,
1296                                 map->m_len - ofs);
1297                 }
1298                 if (map->m_next_extent)
1299                         *map->m_next_extent = pgofs + 1;
1300         }
1301         f2fs_put_dnode(&dn);
1302 unlock_out:
1303         if (map->m_may_create) {
1304                 __do_map_lock(sbi, flag, false);
1305                 f2fs_balance_fs(sbi, dn.node_changed);
1306         }
1307 out:
1308         trace_f2fs_map_blocks(inode, map, err);
1309         return err;
1310 }
1311
1312 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1313 {
1314         struct f2fs_map_blocks map;
1315         block_t last_lblk;
1316         int err;
1317
1318         if (pos + len > i_size_read(inode))
1319                 return false;
1320
1321         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1322         map.m_next_pgofs = NULL;
1323         map.m_next_extent = NULL;
1324         map.m_seg_type = NO_CHECK_TYPE;
1325         map.m_may_create = false;
1326         last_lblk = F2FS_BLK_ALIGN(pos + len);
1327
1328         while (map.m_lblk < last_lblk) {
1329                 map.m_len = last_lblk - map.m_lblk;
1330                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1331                 if (err || map.m_len == 0)
1332                         return false;
1333                 map.m_lblk += map.m_len;
1334         }
1335         return true;
1336 }
1337
1338 static int __get_data_block(struct inode *inode, sector_t iblock,
1339                         struct buffer_head *bh, int create, int flag,
1340                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1341 {
1342         struct f2fs_map_blocks map;
1343         int err;
1344
1345         map.m_lblk = iblock;
1346         map.m_len = bh->b_size >> inode->i_blkbits;
1347         map.m_next_pgofs = next_pgofs;
1348         map.m_next_extent = NULL;
1349         map.m_seg_type = seg_type;
1350         map.m_may_create = may_write;
1351
1352         err = f2fs_map_blocks(inode, &map, create, flag);
1353         if (!err) {
1354                 map_bh(bh, inode->i_sb, map.m_pblk);
1355                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1356                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1357         }
1358         return err;
1359 }
1360
1361 static int get_data_block(struct inode *inode, sector_t iblock,
1362                         struct buffer_head *bh_result, int create, int flag,
1363                         pgoff_t *next_pgofs)
1364 {
1365         return __get_data_block(inode, iblock, bh_result, create,
1366                                                         flag, next_pgofs,
1367                                                         NO_CHECK_TYPE, create);
1368 }
1369
1370 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1371                         struct buffer_head *bh_result, int create)
1372 {
1373         return __get_data_block(inode, iblock, bh_result, create,
1374                                 F2FS_GET_BLOCK_DIO, NULL,
1375                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1376                                 true);
1377 }
1378
1379 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1380                         struct buffer_head *bh_result, int create)
1381 {
1382         return __get_data_block(inode, iblock, bh_result, create,
1383                                 F2FS_GET_BLOCK_DIO, NULL,
1384                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1385                                 false);
1386 }
1387
1388 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1389                         struct buffer_head *bh_result, int create)
1390 {
1391         /* Block number less than F2FS MAX BLOCKS */
1392         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1393                 return -EFBIG;
1394
1395         return __get_data_block(inode, iblock, bh_result, create,
1396                                                 F2FS_GET_BLOCK_BMAP, NULL,
1397                                                 NO_CHECK_TYPE, create);
1398 }
1399
1400 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1401 {
1402         return (offset >> inode->i_blkbits);
1403 }
1404
1405 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1406 {
1407         return (blk << inode->i_blkbits);
1408 }
1409
1410 static int f2fs_xattr_fiemap(struct inode *inode,
1411                                 struct fiemap_extent_info *fieinfo)
1412 {
1413         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1414         struct page *page;
1415         struct node_info ni;
1416         __u64 phys = 0, len;
1417         __u32 flags;
1418         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1419         int err = 0;
1420
1421         if (f2fs_has_inline_xattr(inode)) {
1422                 int offset;
1423
1424                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1425                                                 inode->i_ino, false);
1426                 if (!page)
1427                         return -ENOMEM;
1428
1429                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1430                 if (err) {
1431                         f2fs_put_page(page, 1);
1432                         return err;
1433                 }
1434
1435                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1436                 offset = offsetof(struct f2fs_inode, i_addr) +
1437                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1438                                         get_inline_xattr_addrs(inode));
1439
1440                 phys += offset;
1441                 len = inline_xattr_size(inode);
1442
1443                 f2fs_put_page(page, 1);
1444
1445                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1446
1447                 if (!xnid)
1448                         flags |= FIEMAP_EXTENT_LAST;
1449
1450                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1451                 if (err || err == 1)
1452                         return err;
1453         }
1454
1455         if (xnid) {
1456                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1457                 if (!page)
1458                         return -ENOMEM;
1459
1460                 err = f2fs_get_node_info(sbi, xnid, &ni);
1461                 if (err) {
1462                         f2fs_put_page(page, 1);
1463                         return err;
1464                 }
1465
1466                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1467                 len = inode->i_sb->s_blocksize;
1468
1469                 f2fs_put_page(page, 1);
1470
1471                 flags = FIEMAP_EXTENT_LAST;
1472         }
1473
1474         if (phys)
1475                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1476
1477         return (err < 0 ? err : 0);
1478 }
1479
1480 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1481                 u64 start, u64 len)
1482 {
1483         struct buffer_head map_bh;
1484         sector_t start_blk, last_blk;
1485         pgoff_t next_pgofs;
1486         u64 logical = 0, phys = 0, size = 0;
1487         u32 flags = 0;
1488         int ret = 0;
1489
1490         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1491                 ret = f2fs_precache_extents(inode);
1492                 if (ret)
1493                         return ret;
1494         }
1495
1496         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1497         if (ret)
1498                 return ret;
1499
1500         inode_lock(inode);
1501
1502         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1503                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1504                 goto out;
1505         }
1506
1507         if (f2fs_has_inline_data(inode)) {
1508                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1509                 if (ret != -EAGAIN)
1510                         goto out;
1511         }
1512
1513         if (logical_to_blk(inode, len) == 0)
1514                 len = blk_to_logical(inode, 1);
1515
1516         start_blk = logical_to_blk(inode, start);
1517         last_blk = logical_to_blk(inode, start + len - 1);
1518
1519 next:
1520         memset(&map_bh, 0, sizeof(struct buffer_head));
1521         map_bh.b_size = len;
1522
1523         ret = get_data_block(inode, start_blk, &map_bh, 0,
1524                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1525         if (ret)
1526                 goto out;
1527
1528         /* HOLE */
1529         if (!buffer_mapped(&map_bh)) {
1530                 start_blk = next_pgofs;
1531
1532                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1533                                         F2FS_I_SB(inode)->max_file_blocks))
1534                         goto prep_next;
1535
1536                 flags |= FIEMAP_EXTENT_LAST;
1537         }
1538
1539         if (size) {
1540                 if (IS_ENCRYPTED(inode))
1541                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1542
1543                 ret = fiemap_fill_next_extent(fieinfo, logical,
1544                                 phys, size, flags);
1545         }
1546
1547         if (start_blk > last_blk || ret)
1548                 goto out;
1549
1550         logical = blk_to_logical(inode, start_blk);
1551         phys = blk_to_logical(inode, map_bh.b_blocknr);
1552         size = map_bh.b_size;
1553         flags = 0;
1554         if (buffer_unwritten(&map_bh))
1555                 flags = FIEMAP_EXTENT_UNWRITTEN;
1556
1557         start_blk += logical_to_blk(inode, size);
1558
1559 prep_next:
1560         cond_resched();
1561         if (fatal_signal_pending(current))
1562                 ret = -EINTR;
1563         else
1564                 goto next;
1565 out:
1566         if (ret == 1)
1567                 ret = 0;
1568
1569         inode_unlock(inode);
1570         return ret;
1571 }
1572
1573 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1574                                         unsigned nr_pages,
1575                                         struct f2fs_map_blocks *map,
1576                                         struct bio **bio_ret,
1577                                         sector_t *last_block_in_bio,
1578                                         bool is_readahead)
1579 {
1580         struct bio *bio = *bio_ret;
1581         const unsigned blkbits = inode->i_blkbits;
1582         const unsigned blocksize = 1 << blkbits;
1583         sector_t block_in_file;
1584         sector_t last_block;
1585         sector_t last_block_in_file;
1586         sector_t block_nr;
1587         int ret = 0;
1588
1589         block_in_file = (sector_t)page_index(page);
1590         last_block = block_in_file + nr_pages;
1591         last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1592                                                         blkbits;
1593         if (last_block > last_block_in_file)
1594                 last_block = last_block_in_file;
1595
1596         /* just zeroing out page which is beyond EOF */
1597         if (block_in_file >= last_block)
1598                 goto zero_out;
1599         /*
1600          * Map blocks using the previous result first.
1601          */
1602         if ((map->m_flags & F2FS_MAP_MAPPED) &&
1603                         block_in_file > map->m_lblk &&
1604                         block_in_file < (map->m_lblk + map->m_len))
1605                 goto got_it;
1606
1607         /*
1608          * Then do more f2fs_map_blocks() calls until we are
1609          * done with this page.
1610          */
1611         map->m_lblk = block_in_file;
1612         map->m_len = last_block - block_in_file;
1613
1614         ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1615         if (ret)
1616                 goto out;
1617 got_it:
1618         if ((map->m_flags & F2FS_MAP_MAPPED)) {
1619                 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1620                 SetPageMappedToDisk(page);
1621
1622                 if (!PageUptodate(page) && (!PageSwapCache(page) &&
1623                                         !cleancache_get_page(page))) {
1624                         SetPageUptodate(page);
1625                         goto confused;
1626                 }
1627
1628                 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1629                                                 DATA_GENERIC_ENHANCE_READ)) {
1630                         ret = -EFSCORRUPTED;
1631                         goto out;
1632                 }
1633         } else {
1634 zero_out:
1635                 zero_user_segment(page, 0, PAGE_SIZE);
1636                 if (!PageUptodate(page))
1637                         SetPageUptodate(page);
1638                 unlock_page(page);
1639                 goto out;
1640         }
1641
1642         /*
1643          * This page will go to BIO.  Do we need to send this
1644          * BIO off first?
1645          */
1646         if (bio && (*last_block_in_bio != block_nr - 1 ||
1647                 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1648 submit_and_realloc:
1649                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1650                 bio = NULL;
1651         }
1652         if (bio == NULL) {
1653                 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1654                                 is_readahead ? REQ_RAHEAD : 0);
1655                 if (IS_ERR(bio)) {
1656                         ret = PTR_ERR(bio);
1657                         bio = NULL;
1658                         goto out;
1659                 }
1660         }
1661
1662         /*
1663          * If the page is under writeback, we need to wait for
1664          * its completion to see the correct decrypted data.
1665          */
1666         f2fs_wait_on_block_writeback(inode, block_nr);
1667
1668         if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1669                 goto submit_and_realloc;
1670
1671         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1672         ClearPageError(page);
1673         *last_block_in_bio = block_nr;
1674         goto out;
1675 confused:
1676         if (bio) {
1677                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1678                 bio = NULL;
1679         }
1680         unlock_page(page);
1681 out:
1682         *bio_ret = bio;
1683         return ret;
1684 }
1685
1686 /*
1687  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1688  * Major change was from block_size == page_size in f2fs by default.
1689  *
1690  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1691  * this function ever deviates from doing just read-ahead, it should either
1692  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1693  * from read-ahead.
1694  */
1695 static int f2fs_mpage_readpages(struct address_space *mapping,
1696                         struct list_head *pages, struct page *page,
1697                         unsigned nr_pages, bool is_readahead)
1698 {
1699         struct bio *bio = NULL;
1700         sector_t last_block_in_bio = 0;
1701         struct inode *inode = mapping->host;
1702         struct f2fs_map_blocks map;
1703         int ret = 0;
1704
1705         map.m_pblk = 0;
1706         map.m_lblk = 0;
1707         map.m_len = 0;
1708         map.m_flags = 0;
1709         map.m_next_pgofs = NULL;
1710         map.m_next_extent = NULL;
1711         map.m_seg_type = NO_CHECK_TYPE;
1712         map.m_may_create = false;
1713
1714         for (; nr_pages; nr_pages--) {
1715                 if (pages) {
1716                         page = list_last_entry(pages, struct page, lru);
1717
1718                         prefetchw(&page->flags);
1719                         list_del(&page->lru);
1720                         if (add_to_page_cache_lru(page, mapping,
1721                                                   page_index(page),
1722                                                   readahead_gfp_mask(mapping)))
1723                                 goto next_page;
1724                 }
1725
1726                 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1727                                         &last_block_in_bio, is_readahead);
1728                 if (ret) {
1729                         SetPageError(page);
1730                         zero_user_segment(page, 0, PAGE_SIZE);
1731                         unlock_page(page);
1732                 }
1733 next_page:
1734                 if (pages)
1735                         put_page(page);
1736         }
1737         BUG_ON(pages && !list_empty(pages));
1738         if (bio)
1739                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1740         return pages ? 0 : ret;
1741 }
1742
1743 static int f2fs_read_data_page(struct file *file, struct page *page)
1744 {
1745         struct inode *inode = page_file_mapping(page)->host;
1746         int ret = -EAGAIN;
1747
1748         trace_f2fs_readpage(page, DATA);
1749
1750         /* If the file has inline data, try to read it directly */
1751         if (f2fs_has_inline_data(inode))
1752                 ret = f2fs_read_inline_data(inode, page);
1753         if (ret == -EAGAIN)
1754                 ret = f2fs_mpage_readpages(page_file_mapping(page),
1755                                                 NULL, page, 1, false);
1756         return ret;
1757 }
1758
1759 static int f2fs_read_data_pages(struct file *file,
1760                         struct address_space *mapping,
1761                         struct list_head *pages, unsigned nr_pages)
1762 {
1763         struct inode *inode = mapping->host;
1764         struct page *page = list_last_entry(pages, struct page, lru);
1765
1766         trace_f2fs_readpages(inode, page, nr_pages);
1767
1768         /* If the file has inline data, skip readpages */
1769         if (f2fs_has_inline_data(inode))
1770                 return 0;
1771
1772         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1773 }
1774
1775 static int encrypt_one_page(struct f2fs_io_info *fio)
1776 {
1777         struct inode *inode = fio->page->mapping->host;
1778         struct page *mpage;
1779         gfp_t gfp_flags = GFP_NOFS;
1780
1781         if (!f2fs_encrypted_file(inode))
1782                 return 0;
1783
1784         /* wait for GCed page writeback via META_MAPPING */
1785         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1786
1787 retry_encrypt:
1788         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1789                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1790         if (IS_ERR(fio->encrypted_page)) {
1791                 /* flush pending IOs and wait for a while in the ENOMEM case */
1792                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1793                         f2fs_flush_merged_writes(fio->sbi);
1794                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1795                         gfp_flags |= __GFP_NOFAIL;
1796                         goto retry_encrypt;
1797                 }
1798                 return PTR_ERR(fio->encrypted_page);
1799         }
1800
1801         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1802         if (mpage) {
1803                 if (PageUptodate(mpage))
1804                         memcpy(page_address(mpage),
1805                                 page_address(fio->encrypted_page), PAGE_SIZE);
1806                 f2fs_put_page(mpage, 1);
1807         }
1808         return 0;
1809 }
1810
1811 static inline bool check_inplace_update_policy(struct inode *inode,
1812                                 struct f2fs_io_info *fio)
1813 {
1814         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1815         unsigned int policy = SM_I(sbi)->ipu_policy;
1816
1817         if (policy & (0x1 << F2FS_IPU_FORCE))
1818                 return true;
1819         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1820                 return true;
1821         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1822                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1823                 return true;
1824         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1825                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1826                 return true;
1827
1828         /*
1829          * IPU for rewrite async pages
1830          */
1831         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1832                         fio && fio->op == REQ_OP_WRITE &&
1833                         !(fio->op_flags & REQ_SYNC) &&
1834                         !IS_ENCRYPTED(inode))
1835                 return true;
1836
1837         /* this is only set during fdatasync */
1838         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1839                         is_inode_flag_set(inode, FI_NEED_IPU))
1840                 return true;
1841
1842         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1843                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1844                 return true;
1845
1846         return false;
1847 }
1848
1849 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1850 {
1851         if (f2fs_is_pinned_file(inode))
1852                 return true;
1853
1854         /* if this is cold file, we should overwrite to avoid fragmentation */
1855         if (file_is_cold(inode))
1856                 return true;
1857
1858         return check_inplace_update_policy(inode, fio);
1859 }
1860
1861 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1862 {
1863         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1864
1865         if (test_opt(sbi, LFS))
1866                 return true;
1867         if (S_ISDIR(inode->i_mode))
1868                 return true;
1869         if (IS_NOQUOTA(inode))
1870                 return true;
1871         if (f2fs_is_atomic_file(inode))
1872                 return true;
1873         if (fio) {
1874                 if (is_cold_data(fio->page))
1875                         return true;
1876                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1877                         return true;
1878                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1879                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1880                         return true;
1881         }
1882         return false;
1883 }
1884
1885 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1886 {
1887         struct inode *inode = fio->page->mapping->host;
1888
1889         if (f2fs_should_update_outplace(inode, fio))
1890                 return false;
1891
1892         return f2fs_should_update_inplace(inode, fio);
1893 }
1894
1895 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1896 {
1897         struct page *page = fio->page;
1898         struct inode *inode = page->mapping->host;
1899         struct dnode_of_data dn;
1900         struct extent_info ei = {0,0,0};
1901         struct node_info ni;
1902         bool ipu_force = false;
1903         int err = 0;
1904
1905         set_new_dnode(&dn, inode, NULL, NULL, 0);
1906         if (need_inplace_update(fio) &&
1907                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1908                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1909
1910                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1911                                                 DATA_GENERIC_ENHANCE))
1912                         return -EFSCORRUPTED;
1913
1914                 ipu_force = true;
1915                 fio->need_lock = LOCK_DONE;
1916                 goto got_it;
1917         }
1918
1919         /* Deadlock due to between page->lock and f2fs_lock_op */
1920         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1921                 return -EAGAIN;
1922
1923         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1924         if (err)
1925                 goto out;
1926
1927         fio->old_blkaddr = dn.data_blkaddr;
1928
1929         /* This page is already truncated */
1930         if (fio->old_blkaddr == NULL_ADDR) {
1931                 ClearPageUptodate(page);
1932                 clear_cold_data(page);
1933                 goto out_writepage;
1934         }
1935 got_it:
1936         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1937                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1938                                                 DATA_GENERIC_ENHANCE)) {
1939                 err = -EFSCORRUPTED;
1940                 goto out_writepage;
1941         }
1942         /*
1943          * If current allocation needs SSR,
1944          * it had better in-place writes for updated data.
1945          */
1946         if (ipu_force ||
1947                 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1948                                         need_inplace_update(fio))) {
1949                 err = encrypt_one_page(fio);
1950                 if (err)
1951                         goto out_writepage;
1952
1953                 set_page_writeback(page);
1954                 ClearPageError(page);
1955                 f2fs_put_dnode(&dn);
1956                 if (fio->need_lock == LOCK_REQ)
1957                         f2fs_unlock_op(fio->sbi);
1958                 err = f2fs_inplace_write_data(fio);
1959                 if (err) {
1960                         if (f2fs_encrypted_file(inode))
1961                                 fscrypt_pullback_bio_page(&fio->encrypted_page,
1962                                                                         true);
1963                         if (PageWriteback(page))
1964                                 end_page_writeback(page);
1965                 } else {
1966                         set_inode_flag(inode, FI_UPDATE_WRITE);
1967                 }
1968                 trace_f2fs_do_write_data_page(fio->page, IPU);
1969                 return err;
1970         }
1971
1972         if (fio->need_lock == LOCK_RETRY) {
1973                 if (!f2fs_trylock_op(fio->sbi)) {
1974                         err = -EAGAIN;
1975                         goto out_writepage;
1976                 }
1977                 fio->need_lock = LOCK_REQ;
1978         }
1979
1980         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1981         if (err)
1982                 goto out_writepage;
1983
1984         fio->version = ni.version;
1985
1986         err = encrypt_one_page(fio);
1987         if (err)
1988                 goto out_writepage;
1989
1990         set_page_writeback(page);
1991         ClearPageError(page);
1992
1993         /* LFS mode write path */
1994         f2fs_outplace_write_data(&dn, fio);
1995         trace_f2fs_do_write_data_page(page, OPU);
1996         set_inode_flag(inode, FI_APPEND_WRITE);
1997         if (page->index == 0)
1998                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1999 out_writepage:
2000         f2fs_put_dnode(&dn);
2001 out:
2002         if (fio->need_lock == LOCK_REQ)
2003                 f2fs_unlock_op(fio->sbi);
2004         return err;
2005 }
2006
2007 static int __write_data_page(struct page *page, bool *submitted,
2008                                 struct bio **bio,
2009                                 sector_t *last_block,
2010                                 struct writeback_control *wbc,
2011                                 enum iostat_type io_type)
2012 {
2013         struct inode *inode = page->mapping->host;
2014         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2015         loff_t i_size = i_size_read(inode);
2016         const pgoff_t end_index = ((unsigned long long) i_size)
2017                                                         >> PAGE_SHIFT;
2018         loff_t psize = (page->index + 1) << PAGE_SHIFT;
2019         unsigned offset = 0;
2020         bool need_balance_fs = false;
2021         int err = 0;
2022         struct f2fs_io_info fio = {
2023                 .sbi = sbi,
2024                 .ino = inode->i_ino,
2025                 .type = DATA,
2026                 .op = REQ_OP_WRITE,
2027                 .op_flags = wbc_to_write_flags(wbc),
2028                 .old_blkaddr = NULL_ADDR,
2029                 .page = page,
2030                 .encrypted_page = NULL,
2031                 .submitted = false,
2032                 .need_lock = LOCK_RETRY,
2033                 .io_type = io_type,
2034                 .io_wbc = wbc,
2035                 .bio = bio,
2036                 .last_block = last_block,
2037         };
2038
2039         trace_f2fs_writepage(page, DATA);
2040
2041         /* we should bypass data pages to proceed the kworkder jobs */
2042         if (unlikely(f2fs_cp_error(sbi))) {
2043                 mapping_set_error(page->mapping, -EIO);
2044                 /*
2045                  * don't drop any dirty dentry pages for keeping lastest
2046                  * directory structure.
2047                  */
2048                 if (S_ISDIR(inode->i_mode))
2049                         goto redirty_out;
2050                 goto out;
2051         }
2052
2053         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2054                 goto redirty_out;
2055
2056         if (page->index < end_index)
2057                 goto write;
2058
2059         /*
2060          * If the offset is out-of-range of file size,
2061          * this page does not have to be written to disk.
2062          */
2063         offset = i_size & (PAGE_SIZE - 1);
2064         if ((page->index >= end_index + 1) || !offset)
2065                 goto out;
2066
2067         zero_user_segment(page, offset, PAGE_SIZE);
2068 write:
2069         if (f2fs_is_drop_cache(inode))
2070                 goto out;
2071         /* we should not write 0'th page having journal header */
2072         if (f2fs_is_volatile_file(inode) && (!page->index ||
2073                         (!wbc->for_reclaim &&
2074                         f2fs_available_free_memory(sbi, BASE_CHECK))))
2075                 goto redirty_out;
2076
2077         /* Dentry blocks are controlled by checkpoint */
2078         if (S_ISDIR(inode->i_mode)) {
2079                 fio.need_lock = LOCK_DONE;
2080                 err = f2fs_do_write_data_page(&fio);
2081                 goto done;
2082         }
2083
2084         if (!wbc->for_reclaim)
2085                 need_balance_fs = true;
2086         else if (has_not_enough_free_secs(sbi, 0, 0))
2087                 goto redirty_out;
2088         else
2089                 set_inode_flag(inode, FI_HOT_DATA);
2090
2091         err = -EAGAIN;
2092         if (f2fs_has_inline_data(inode)) {
2093                 err = f2fs_write_inline_data(inode, page);
2094                 if (!err)
2095                         goto out;
2096         }
2097
2098         if (err == -EAGAIN) {
2099                 err = f2fs_do_write_data_page(&fio);
2100                 if (err == -EAGAIN) {
2101                         fio.need_lock = LOCK_REQ;
2102                         err = f2fs_do_write_data_page(&fio);
2103                 }
2104         }
2105
2106         if (err) {
2107                 file_set_keep_isize(inode);
2108         } else {
2109                 down_write(&F2FS_I(inode)->i_sem);
2110                 if (F2FS_I(inode)->last_disk_size < psize)
2111                         F2FS_I(inode)->last_disk_size = psize;
2112                 up_write(&F2FS_I(inode)->i_sem);
2113         }
2114
2115 done:
2116         if (err && err != -ENOENT)
2117                 goto redirty_out;
2118
2119 out:
2120         inode_dec_dirty_pages(inode);
2121         if (err) {
2122                 ClearPageUptodate(page);
2123                 clear_cold_data(page);
2124         }
2125
2126         if (wbc->for_reclaim) {
2127                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2128                 clear_inode_flag(inode, FI_HOT_DATA);
2129                 f2fs_remove_dirty_inode(inode);
2130                 submitted = NULL;
2131         }
2132
2133         unlock_page(page);
2134         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2135                                         !F2FS_I(inode)->cp_task) {
2136                 f2fs_submit_ipu_bio(sbi, bio, page);
2137                 f2fs_balance_fs(sbi, need_balance_fs);
2138         }
2139
2140         if (unlikely(f2fs_cp_error(sbi))) {
2141                 f2fs_submit_ipu_bio(sbi, bio, page);
2142                 f2fs_submit_merged_write(sbi, DATA);
2143                 submitted = NULL;
2144         }
2145
2146         if (submitted)
2147                 *submitted = fio.submitted;
2148
2149         return 0;
2150
2151 redirty_out:
2152         redirty_page_for_writepage(wbc, page);
2153         /*
2154          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2155          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2156          * file_write_and_wait_range() will see EIO error, which is critical
2157          * to return value of fsync() followed by atomic_write failure to user.
2158          */
2159         if (!err || wbc->for_reclaim)
2160                 return AOP_WRITEPAGE_ACTIVATE;
2161         unlock_page(page);
2162         return err;
2163 }
2164
2165 static int f2fs_write_data_page(struct page *page,
2166                                         struct writeback_control *wbc)
2167 {
2168         return __write_data_page(page, NULL, NULL, NULL, wbc, FS_DATA_IO);
2169 }
2170
2171 /*
2172  * This function was copied from write_cche_pages from mm/page-writeback.c.
2173  * The major change is making write step of cold data page separately from
2174  * warm/hot data page.
2175  */
2176 static int f2fs_write_cache_pages(struct address_space *mapping,
2177                                         struct writeback_control *wbc,
2178                                         enum iostat_type io_type)
2179 {
2180         int ret = 0;
2181         int done = 0;
2182         struct pagevec pvec;
2183         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2184         struct bio *bio = NULL;
2185         sector_t last_block;
2186         int nr_pages;
2187         pgoff_t uninitialized_var(writeback_index);
2188         pgoff_t index;
2189         pgoff_t end;            /* Inclusive */
2190         pgoff_t done_index;
2191         int cycled;
2192         int range_whole = 0;
2193         xa_mark_t tag;
2194         int nwritten = 0;
2195
2196         pagevec_init(&pvec);
2197
2198         if (get_dirty_pages(mapping->host) <=
2199                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2200                 set_inode_flag(mapping->host, FI_HOT_DATA);
2201         else
2202                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2203
2204         if (wbc->range_cyclic) {
2205                 writeback_index = mapping->writeback_index; /* prev offset */
2206                 index = writeback_index;
2207                 if (index == 0)
2208                         cycled = 1;
2209                 else
2210                         cycled = 0;
2211                 end = -1;
2212         } else {
2213                 index = wbc->range_start >> PAGE_SHIFT;
2214                 end = wbc->range_end >> PAGE_SHIFT;
2215                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2216                         range_whole = 1;
2217                 cycled = 1; /* ignore range_cyclic tests */
2218         }
2219         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2220                 tag = PAGECACHE_TAG_TOWRITE;
2221         else
2222                 tag = PAGECACHE_TAG_DIRTY;
2223 retry:
2224         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2225                 tag_pages_for_writeback(mapping, index, end);
2226         done_index = index;
2227         while (!done && (index <= end)) {
2228                 int i;
2229
2230                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2231                                 tag);
2232                 if (nr_pages == 0)
2233                         break;
2234
2235                 for (i = 0; i < nr_pages; i++) {
2236                         struct page *page = pvec.pages[i];
2237                         bool submitted = false;
2238
2239                         /* give a priority to WB_SYNC threads */
2240                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2241                                         wbc->sync_mode == WB_SYNC_NONE) {
2242                                 done = 1;
2243                                 break;
2244                         }
2245
2246                         done_index = page->index;
2247 retry_write:
2248                         lock_page(page);
2249
2250                         if (unlikely(page->mapping != mapping)) {
2251 continue_unlock:
2252                                 unlock_page(page);
2253                                 continue;
2254                         }
2255
2256                         if (!PageDirty(page)) {
2257                                 /* someone wrote it for us */
2258                                 goto continue_unlock;
2259                         }
2260
2261                         if (PageWriteback(page)) {
2262                                 if (wbc->sync_mode != WB_SYNC_NONE) {
2263                                         f2fs_wait_on_page_writeback(page,
2264                                                         DATA, true, true);
2265                                         f2fs_submit_ipu_bio(sbi, &bio, page);
2266                                 } else {
2267                                         goto continue_unlock;
2268                                 }
2269                         }
2270
2271                         if (!clear_page_dirty_for_io(page))
2272                                 goto continue_unlock;
2273
2274                         ret = __write_data_page(page, &submitted, &bio,
2275                                         &last_block, wbc, io_type);
2276                         if (unlikely(ret)) {
2277                                 /*
2278                                  * keep nr_to_write, since vfs uses this to
2279                                  * get # of written pages.
2280                                  */
2281                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2282                                         unlock_page(page);
2283                                         ret = 0;
2284                                         continue;
2285                                 } else if (ret == -EAGAIN) {
2286                                         ret = 0;
2287                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2288                                                 cond_resched();
2289                                                 congestion_wait(BLK_RW_ASYNC,
2290                                                                         HZ/50);
2291                                                 goto retry_write;
2292                                         }
2293                                         continue;
2294                                 }
2295                                 done_index = page->index + 1;
2296                                 done = 1;
2297                                 break;
2298                         } else if (submitted) {
2299                                 nwritten++;
2300                         }
2301
2302                         if (--wbc->nr_to_write <= 0 &&
2303                                         wbc->sync_mode == WB_SYNC_NONE) {
2304                                 done = 1;
2305                                 break;
2306                         }
2307                 }
2308                 pagevec_release(&pvec);
2309                 cond_resched();
2310         }
2311
2312         if (!cycled && !done) {
2313                 cycled = 1;
2314                 index = 0;
2315                 end = writeback_index - 1;
2316                 goto retry;
2317         }
2318         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2319                 mapping->writeback_index = done_index;
2320
2321         if (nwritten)
2322                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2323                                                                 NULL, 0, DATA);
2324         /* submit cached bio of IPU write */
2325         if (bio)
2326                 __submit_bio(sbi, bio, DATA);
2327
2328         return ret;
2329 }
2330
2331 static inline bool __should_serialize_io(struct inode *inode,
2332                                         struct writeback_control *wbc)
2333 {
2334         if (!S_ISREG(inode->i_mode))
2335                 return false;
2336         if (IS_NOQUOTA(inode))
2337                 return false;
2338         /* to avoid deadlock in path of data flush */
2339         if (F2FS_I(inode)->cp_task)
2340                 return false;
2341         if (wbc->sync_mode != WB_SYNC_ALL)
2342                 return true;
2343         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2344                 return true;
2345         return false;
2346 }
2347
2348 static int __f2fs_write_data_pages(struct address_space *mapping,
2349                                                 struct writeback_control *wbc,
2350                                                 enum iostat_type io_type)
2351 {
2352         struct inode *inode = mapping->host;
2353         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2354         struct blk_plug plug;
2355         int ret;
2356         bool locked = false;
2357
2358         /* deal with chardevs and other special file */
2359         if (!mapping->a_ops->writepage)
2360                 return 0;
2361
2362         /* skip writing if there is no dirty page in this inode */
2363         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2364                 return 0;
2365
2366         /* during POR, we don't need to trigger writepage at all. */
2367         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2368                 goto skip_write;
2369
2370         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2371                         wbc->sync_mode == WB_SYNC_NONE &&
2372                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2373                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2374                 goto skip_write;
2375
2376         /* skip writing during file defragment */
2377         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2378                 goto skip_write;
2379
2380         trace_f2fs_writepages(mapping->host, wbc, DATA);
2381
2382         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2383         if (wbc->sync_mode == WB_SYNC_ALL)
2384                 atomic_inc(&sbi->wb_sync_req[DATA]);
2385         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2386                 goto skip_write;
2387
2388         if (__should_serialize_io(inode, wbc)) {
2389                 mutex_lock(&sbi->writepages);
2390                 locked = true;
2391         }
2392
2393         blk_start_plug(&plug);
2394         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2395         blk_finish_plug(&plug);
2396
2397         if (locked)
2398                 mutex_unlock(&sbi->writepages);
2399
2400         if (wbc->sync_mode == WB_SYNC_ALL)
2401                 atomic_dec(&sbi->wb_sync_req[DATA]);
2402         /*
2403          * if some pages were truncated, we cannot guarantee its mapping->host
2404          * to detect pending bios.
2405          */
2406
2407         f2fs_remove_dirty_inode(inode);
2408         return ret;
2409
2410 skip_write:
2411         wbc->pages_skipped += get_dirty_pages(inode);
2412         trace_f2fs_writepages(mapping->host, wbc, DATA);
2413         return 0;
2414 }
2415
2416 static int f2fs_write_data_pages(struct address_space *mapping,
2417                             struct writeback_control *wbc)
2418 {
2419         struct inode *inode = mapping->host;
2420
2421         return __f2fs_write_data_pages(mapping, wbc,
2422                         F2FS_I(inode)->cp_task == current ?
2423                         FS_CP_DATA_IO : FS_DATA_IO);
2424 }
2425
2426 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2427 {
2428         struct inode *inode = mapping->host;
2429         loff_t i_size = i_size_read(inode);
2430
2431         if (to > i_size) {
2432                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2433                 down_write(&F2FS_I(inode)->i_mmap_sem);
2434
2435                 truncate_pagecache(inode, i_size);
2436                 if (!IS_NOQUOTA(inode))
2437                         f2fs_truncate_blocks(inode, i_size, true);
2438
2439                 up_write(&F2FS_I(inode)->i_mmap_sem);
2440                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2441         }
2442 }
2443
2444 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2445                         struct page *page, loff_t pos, unsigned len,
2446                         block_t *blk_addr, bool *node_changed)
2447 {
2448         struct inode *inode = page->mapping->host;
2449         pgoff_t index = page->index;
2450         struct dnode_of_data dn;
2451         struct page *ipage;
2452         bool locked = false;
2453         struct extent_info ei = {0,0,0};
2454         int err = 0;
2455         int flag;
2456
2457         /*
2458          * we already allocated all the blocks, so we don't need to get
2459          * the block addresses when there is no need to fill the page.
2460          */
2461         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2462                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2463                 return 0;
2464
2465         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2466         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2467                 flag = F2FS_GET_BLOCK_DEFAULT;
2468         else
2469                 flag = F2FS_GET_BLOCK_PRE_AIO;
2470
2471         if (f2fs_has_inline_data(inode) ||
2472                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2473                 __do_map_lock(sbi, flag, true);
2474                 locked = true;
2475         }
2476 restart:
2477         /* check inline_data */
2478         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2479         if (IS_ERR(ipage)) {
2480                 err = PTR_ERR(ipage);
2481                 goto unlock_out;
2482         }
2483
2484         set_new_dnode(&dn, inode, ipage, ipage, 0);
2485
2486         if (f2fs_has_inline_data(inode)) {
2487                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2488                         f2fs_do_read_inline_data(page, ipage);
2489                         set_inode_flag(inode, FI_DATA_EXIST);
2490                         if (inode->i_nlink)
2491                                 set_inline_node(ipage);
2492                 } else {
2493                         err = f2fs_convert_inline_page(&dn, page);
2494                         if (err)
2495                                 goto out;
2496                         if (dn.data_blkaddr == NULL_ADDR)
2497                                 err = f2fs_get_block(&dn, index);
2498                 }
2499         } else if (locked) {
2500                 err = f2fs_get_block(&dn, index);
2501         } else {
2502                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2503                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2504                 } else {
2505                         /* hole case */
2506                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2507                         if (err || dn.data_blkaddr == NULL_ADDR) {
2508                                 f2fs_put_dnode(&dn);
2509                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2510                                                                 true);
2511                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2512                                 locked = true;
2513                                 goto restart;
2514                         }
2515                 }
2516         }
2517
2518         /* convert_inline_page can make node_changed */
2519         *blk_addr = dn.data_blkaddr;
2520         *node_changed = dn.node_changed;
2521 out:
2522         f2fs_put_dnode(&dn);
2523 unlock_out:
2524         if (locked)
2525                 __do_map_lock(sbi, flag, false);
2526         return err;
2527 }
2528
2529 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2530                 loff_t pos, unsigned len, unsigned flags,
2531                 struct page **pagep, void **fsdata)
2532 {
2533         struct inode *inode = mapping->host;
2534         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2535         struct page *page = NULL;
2536         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2537         bool need_balance = false, drop_atomic = false;
2538         block_t blkaddr = NULL_ADDR;
2539         int err = 0;
2540
2541         trace_f2fs_write_begin(inode, pos, len, flags);
2542
2543         err = f2fs_is_checkpoint_ready(sbi);
2544         if (err)
2545                 goto fail;
2546
2547         if ((f2fs_is_atomic_file(inode) &&
2548                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2549                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2550                 err = -ENOMEM;
2551                 drop_atomic = true;
2552                 goto fail;
2553         }
2554
2555         /*
2556          * We should check this at this moment to avoid deadlock on inode page
2557          * and #0 page. The locking rule for inline_data conversion should be:
2558          * lock_page(page #0) -> lock_page(inode_page)
2559          */
2560         if (index != 0) {
2561                 err = f2fs_convert_inline_inode(inode);
2562                 if (err)
2563                         goto fail;
2564         }
2565 repeat:
2566         /*
2567          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2568          * wait_for_stable_page. Will wait that below with our IO control.
2569          */
2570         page = f2fs_pagecache_get_page(mapping, index,
2571                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2572         if (!page) {
2573                 err = -ENOMEM;
2574                 goto fail;
2575         }
2576
2577         *pagep = page;
2578
2579         err = prepare_write_begin(sbi, page, pos, len,
2580                                         &blkaddr, &need_balance);
2581         if (err)
2582                 goto fail;
2583
2584         if (need_balance && !IS_NOQUOTA(inode) &&
2585                         has_not_enough_free_secs(sbi, 0, 0)) {
2586                 unlock_page(page);
2587                 f2fs_balance_fs(sbi, true);
2588                 lock_page(page);
2589                 if (page->mapping != mapping) {
2590                         /* The page got truncated from under us */
2591                         f2fs_put_page(page, 1);
2592                         goto repeat;
2593                 }
2594         }
2595
2596         f2fs_wait_on_page_writeback(page, DATA, false, true);
2597
2598         if (len == PAGE_SIZE || PageUptodate(page))
2599                 return 0;
2600
2601         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2602                 zero_user_segment(page, len, PAGE_SIZE);
2603                 return 0;
2604         }
2605
2606         if (blkaddr == NEW_ADDR) {
2607                 zero_user_segment(page, 0, PAGE_SIZE);
2608                 SetPageUptodate(page);
2609         } else {
2610                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2611                                 DATA_GENERIC_ENHANCE_READ)) {
2612                         err = -EFSCORRUPTED;
2613                         goto fail;
2614                 }
2615                 err = f2fs_submit_page_read(inode, page, blkaddr);
2616                 if (err)
2617                         goto fail;
2618
2619                 lock_page(page);
2620                 if (unlikely(page->mapping != mapping)) {
2621                         f2fs_put_page(page, 1);
2622                         goto repeat;
2623                 }
2624                 if (unlikely(!PageUptodate(page))) {
2625                         err = -EIO;
2626                         goto fail;
2627                 }
2628         }
2629         return 0;
2630
2631 fail:
2632         f2fs_put_page(page, 1);
2633         f2fs_write_failed(mapping, pos + len);
2634         if (drop_atomic)
2635                 f2fs_drop_inmem_pages_all(sbi, false);
2636         return err;
2637 }
2638
2639 static int f2fs_write_end(struct file *file,
2640                         struct address_space *mapping,
2641                         loff_t pos, unsigned len, unsigned copied,
2642                         struct page *page, void *fsdata)
2643 {
2644         struct inode *inode = page->mapping->host;
2645
2646         trace_f2fs_write_end(inode, pos, len, copied);
2647
2648         /*
2649          * This should be come from len == PAGE_SIZE, and we expect copied
2650          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2651          * let generic_perform_write() try to copy data again through copied=0.
2652          */
2653         if (!PageUptodate(page)) {
2654                 if (unlikely(copied != len))
2655                         copied = 0;
2656                 else
2657                         SetPageUptodate(page);
2658         }
2659         if (!copied)
2660                 goto unlock_out;
2661
2662         set_page_dirty(page);
2663
2664         if (pos + copied > i_size_read(inode))
2665                 f2fs_i_size_write(inode, pos + copied);
2666 unlock_out:
2667         f2fs_put_page(page, 1);
2668         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2669         return copied;
2670 }
2671
2672 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2673                            loff_t offset)
2674 {
2675         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2676         unsigned blkbits = i_blkbits;
2677         unsigned blocksize_mask = (1 << blkbits) - 1;
2678         unsigned long align = offset | iov_iter_alignment(iter);
2679         struct block_device *bdev = inode->i_sb->s_bdev;
2680
2681         if (align & blocksize_mask) {
2682                 if (bdev)
2683                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2684                 blocksize_mask = (1 << blkbits) - 1;
2685                 if (align & blocksize_mask)
2686                         return -EINVAL;
2687                 return 1;
2688         }
2689         return 0;
2690 }
2691
2692 static void f2fs_dio_end_io(struct bio *bio)
2693 {
2694         struct f2fs_private_dio *dio = bio->bi_private;
2695
2696         dec_page_count(F2FS_I_SB(dio->inode),
2697                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2698
2699         bio->bi_private = dio->orig_private;
2700         bio->bi_end_io = dio->orig_end_io;
2701
2702         kvfree(dio);
2703
2704         bio_endio(bio);
2705 }
2706
2707 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2708                                                         loff_t file_offset)
2709 {
2710         struct f2fs_private_dio *dio;
2711         bool write = (bio_op(bio) == REQ_OP_WRITE);
2712
2713         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2714                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2715         if (!dio)
2716                 goto out;
2717
2718         dio->inode = inode;
2719         dio->orig_end_io = bio->bi_end_io;
2720         dio->orig_private = bio->bi_private;
2721         dio->write = write;
2722
2723         bio->bi_end_io = f2fs_dio_end_io;
2724         bio->bi_private = dio;
2725
2726         inc_page_count(F2FS_I_SB(inode),
2727                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2728
2729         submit_bio(bio);
2730         return;
2731 out:
2732         bio->bi_status = BLK_STS_IOERR;
2733         bio_endio(bio);
2734 }
2735
2736 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2737 {
2738         struct address_space *mapping = iocb->ki_filp->f_mapping;
2739         struct inode *inode = mapping->host;
2740         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2741         struct f2fs_inode_info *fi = F2FS_I(inode);
2742         size_t count = iov_iter_count(iter);
2743         loff_t offset = iocb->ki_pos;
2744         int rw = iov_iter_rw(iter);
2745         int err;
2746         enum rw_hint hint = iocb->ki_hint;
2747         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2748         bool do_opu;
2749
2750         err = check_direct_IO(inode, iter, offset);
2751         if (err)
2752                 return err < 0 ? err : 0;
2753
2754         if (f2fs_force_buffered_io(inode, iocb, iter))
2755                 return 0;
2756
2757         do_opu = allow_outplace_dio(inode, iocb, iter);
2758
2759         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2760
2761         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2762                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2763
2764         if (iocb->ki_flags & IOCB_NOWAIT) {
2765                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2766                         iocb->ki_hint = hint;
2767                         err = -EAGAIN;
2768                         goto out;
2769                 }
2770                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2771                         up_read(&fi->i_gc_rwsem[rw]);
2772                         iocb->ki_hint = hint;
2773                         err = -EAGAIN;
2774                         goto out;
2775                 }
2776         } else {
2777                 down_read(&fi->i_gc_rwsem[rw]);
2778                 if (do_opu)
2779                         down_read(&fi->i_gc_rwsem[READ]);
2780         }
2781
2782         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2783                         iter, rw == WRITE ? get_data_block_dio_write :
2784                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2785                         DIO_LOCKING | DIO_SKIP_HOLES);
2786
2787         if (do_opu)
2788                 up_read(&fi->i_gc_rwsem[READ]);
2789
2790         up_read(&fi->i_gc_rwsem[rw]);
2791
2792         if (rw == WRITE) {
2793                 if (whint_mode == WHINT_MODE_OFF)
2794                         iocb->ki_hint = hint;
2795                 if (err > 0) {
2796                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2797                                                                         err);
2798                         if (!do_opu)
2799                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2800                 } else if (err < 0) {
2801                         f2fs_write_failed(mapping, offset + count);
2802                 }
2803         }
2804
2805 out:
2806         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2807
2808         return err;
2809 }
2810
2811 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2812                                                         unsigned int length)
2813 {
2814         struct inode *inode = page->mapping->host;
2815         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2816
2817         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2818                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2819                 return;
2820
2821         if (PageDirty(page)) {
2822                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2823                         dec_page_count(sbi, F2FS_DIRTY_META);
2824                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2825                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2826                 } else {
2827                         inode_dec_dirty_pages(inode);
2828                         f2fs_remove_dirty_inode(inode);
2829                 }
2830         }
2831
2832         clear_cold_data(page);
2833
2834         if (IS_ATOMIC_WRITTEN_PAGE(page))
2835                 return f2fs_drop_inmem_page(inode, page);
2836
2837         f2fs_clear_page_private(page);
2838 }
2839
2840 int f2fs_release_page(struct page *page, gfp_t wait)
2841 {
2842         /* If this is dirty page, keep PagePrivate */
2843         if (PageDirty(page))
2844                 return 0;
2845
2846         /* This is atomic written page, keep Private */
2847         if (IS_ATOMIC_WRITTEN_PAGE(page))
2848                 return 0;
2849
2850         clear_cold_data(page);
2851         f2fs_clear_page_private(page);
2852         return 1;
2853 }
2854
2855 static int f2fs_set_data_page_dirty(struct page *page)
2856 {
2857         struct inode *inode = page_file_mapping(page)->host;
2858
2859         trace_f2fs_set_page_dirty(page, DATA);
2860
2861         if (!PageUptodate(page))
2862                 SetPageUptodate(page);
2863         if (PageSwapCache(page))
2864                 return __set_page_dirty_nobuffers(page);
2865
2866         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2867                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2868                         f2fs_register_inmem_page(inode, page);
2869                         return 1;
2870                 }
2871                 /*
2872                  * Previously, this page has been registered, we just
2873                  * return here.
2874                  */
2875                 return 0;
2876         }
2877
2878         if (!PageDirty(page)) {
2879                 __set_page_dirty_nobuffers(page);
2880                 f2fs_update_dirty_page(inode, page);
2881                 return 1;
2882         }
2883         return 0;
2884 }
2885
2886 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2887 {
2888         struct inode *inode = mapping->host;
2889
2890         if (f2fs_has_inline_data(inode))
2891                 return 0;
2892
2893         /* make sure allocating whole blocks */
2894         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2895                 filemap_write_and_wait(mapping);
2896
2897         return generic_block_bmap(mapping, block, get_data_block_bmap);
2898 }
2899
2900 #ifdef CONFIG_MIGRATION
2901 #include <linux/migrate.h>
2902
2903 int f2fs_migrate_page(struct address_space *mapping,
2904                 struct page *newpage, struct page *page, enum migrate_mode mode)
2905 {
2906         int rc, extra_count;
2907         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2908         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2909
2910         BUG_ON(PageWriteback(page));
2911
2912         /* migrating an atomic written page is safe with the inmem_lock hold */
2913         if (atomic_written) {
2914                 if (mode != MIGRATE_SYNC)
2915                         return -EBUSY;
2916                 if (!mutex_trylock(&fi->inmem_lock))
2917                         return -EAGAIN;
2918         }
2919
2920         /* one extra reference was held for atomic_write page */
2921         extra_count = atomic_written ? 1 : 0;
2922         rc = migrate_page_move_mapping(mapping, newpage,
2923                                 page, mode, extra_count);
2924         if (rc != MIGRATEPAGE_SUCCESS) {
2925                 if (atomic_written)
2926                         mutex_unlock(&fi->inmem_lock);
2927                 return rc;
2928         }
2929
2930         if (atomic_written) {
2931                 struct inmem_pages *cur;
2932                 list_for_each_entry(cur, &fi->inmem_pages, list)
2933                         if (cur->page == page) {
2934                                 cur->page = newpage;
2935                                 break;
2936                         }
2937                 mutex_unlock(&fi->inmem_lock);
2938                 put_page(page);
2939                 get_page(newpage);
2940         }
2941
2942         if (PagePrivate(page)) {
2943                 f2fs_set_page_private(newpage, page_private(page));
2944                 f2fs_clear_page_private(page);
2945         }
2946
2947         if (mode != MIGRATE_SYNC_NO_COPY)
2948                 migrate_page_copy(newpage, page);
2949         else
2950                 migrate_page_states(newpage, page);
2951
2952         return MIGRATEPAGE_SUCCESS;
2953 }
2954 #endif
2955
2956 #ifdef CONFIG_SWAP
2957 /* Copied from generic_swapfile_activate() to check any holes */
2958 static int check_swap_activate(struct file *swap_file, unsigned int max)
2959 {
2960         struct address_space *mapping = swap_file->f_mapping;
2961         struct inode *inode = mapping->host;
2962         unsigned blocks_per_page;
2963         unsigned long page_no;
2964         unsigned blkbits;
2965         sector_t probe_block;
2966         sector_t last_block;
2967         sector_t lowest_block = -1;
2968         sector_t highest_block = 0;
2969
2970         blkbits = inode->i_blkbits;
2971         blocks_per_page = PAGE_SIZE >> blkbits;
2972
2973         /*
2974          * Map all the blocks into the extent list.  This code doesn't try
2975          * to be very smart.
2976          */
2977         probe_block = 0;
2978         page_no = 0;
2979         last_block = i_size_read(inode) >> blkbits;
2980         while ((probe_block + blocks_per_page) <= last_block && page_no < max) {
2981                 unsigned block_in_page;
2982                 sector_t first_block;
2983
2984                 cond_resched();
2985
2986                 first_block = bmap(inode, probe_block);
2987                 if (first_block == 0)
2988                         goto bad_bmap;
2989
2990                 /*
2991                  * It must be PAGE_SIZE aligned on-disk
2992                  */
2993                 if (first_block & (blocks_per_page - 1)) {
2994                         probe_block++;
2995                         goto reprobe;
2996                 }
2997
2998                 for (block_in_page = 1; block_in_page < blocks_per_page;
2999                                         block_in_page++) {
3000                         sector_t block;
3001
3002                         block = bmap(inode, probe_block + block_in_page);
3003                         if (block == 0)
3004                                 goto bad_bmap;
3005                         if (block != first_block + block_in_page) {
3006                                 /* Discontiguity */
3007                                 probe_block++;
3008                                 goto reprobe;
3009                         }
3010                 }
3011
3012                 first_block >>= (PAGE_SHIFT - blkbits);
3013                 if (page_no) {  /* exclude the header page */
3014                         if (first_block < lowest_block)
3015                                 lowest_block = first_block;
3016                         if (first_block > highest_block)
3017                                 highest_block = first_block;
3018                 }
3019
3020                 page_no++;
3021                 probe_block += blocks_per_page;
3022 reprobe:
3023                 continue;
3024         }
3025         return 0;
3026
3027 bad_bmap:
3028         pr_err("swapon: swapfile has holes\n");
3029         return -EINVAL;
3030 }
3031
3032 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3033                                 sector_t *span)
3034 {
3035         struct inode *inode = file_inode(file);
3036         int ret;
3037
3038         if (!S_ISREG(inode->i_mode))
3039                 return -EINVAL;
3040
3041         if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3042                 return -EROFS;
3043
3044         ret = f2fs_convert_inline_inode(inode);
3045         if (ret)
3046                 return ret;
3047
3048         ret = check_swap_activate(file, sis->max);
3049         if (ret)
3050                 return ret;
3051
3052         set_inode_flag(inode, FI_PIN_FILE);
3053         f2fs_precache_extents(inode);
3054         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3055         return 0;
3056 }
3057
3058 static void f2fs_swap_deactivate(struct file *file)
3059 {
3060         struct inode *inode = file_inode(file);
3061
3062         clear_inode_flag(inode, FI_PIN_FILE);
3063 }
3064 #else
3065 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3066                                 sector_t *span)
3067 {
3068         return -EOPNOTSUPP;
3069 }
3070
3071 static void f2fs_swap_deactivate(struct file *file)
3072 {
3073 }
3074 #endif
3075
3076 const struct address_space_operations f2fs_dblock_aops = {
3077         .readpage       = f2fs_read_data_page,
3078         .readpages      = f2fs_read_data_pages,
3079         .writepage      = f2fs_write_data_page,
3080         .writepages     = f2fs_write_data_pages,
3081         .write_begin    = f2fs_write_begin,
3082         .write_end      = f2fs_write_end,
3083         .set_page_dirty = f2fs_set_data_page_dirty,
3084         .invalidatepage = f2fs_invalidate_page,
3085         .releasepage    = f2fs_release_page,
3086         .direct_IO      = f2fs_direct_IO,
3087         .bmap           = f2fs_bmap,
3088         .swap_activate  = f2fs_swap_activate,
3089         .swap_deactivate = f2fs_swap_deactivate,
3090 #ifdef CONFIG_MIGRATION
3091         .migratepage    = f2fs_migrate_page,
3092 #endif
3093 };
3094
3095 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3096 {
3097         struct address_space *mapping = page_mapping(page);
3098         unsigned long flags;
3099
3100         xa_lock_irqsave(&mapping->i_pages, flags);
3101         __xa_clear_mark(&mapping->i_pages, page_index(page),
3102                                                 PAGECACHE_TAG_DIRTY);
3103         xa_unlock_irqrestore(&mapping->i_pages, flags);
3104 }
3105
3106 int __init f2fs_init_post_read_processing(void)
3107 {
3108         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
3109         if (!bio_post_read_ctx_cache)
3110                 goto fail;
3111         bio_post_read_ctx_pool =
3112                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3113                                          bio_post_read_ctx_cache);
3114         if (!bio_post_read_ctx_pool)
3115                 goto fail_free_cache;
3116         return 0;
3117
3118 fail_free_cache:
3119         kmem_cache_destroy(bio_post_read_ctx_cache);
3120 fail:
3121         return -ENOMEM;
3122 }
3123
3124 void __exit f2fs_destroy_post_read_processing(void)
3125 {
3126         mempool_destroy(bio_post_read_ctx_pool);
3127         kmem_cache_destroy(bio_post_read_ctx_cache);
3128 }