fs/f2fs/data.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/data.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/mpage.h>
  12 #include <linux/writeback.h>
  13 #include <linux/backing-dev.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/bio.h>
  17 #include <linux/prefetch.h>
  18 #include <linux/uio.h>
  19 #include <linux/cleancache.h>
  20 #include <linux/sched/signal.h>
  21
  22 #include "f2fs.h"
  23 #include "node.h"
  24 #include "segment.h"
  25 #include "trace.h"
  26 #include <trace/events/f2fs.h>
  27
  28 #define NUM_PREALLOC_POST_READ_CTXS     128
  29
  30 static struct kmem_cache *bio_post_read_ctx_cache;
  31 static mempool_t *bio_post_read_ctx_pool;
  32
  33 static bool __is_cp_guaranteed(struct page *page)
  34 {
  35         struct address_space *mapping = page->mapping;
  36         struct inode *inode;
  37         struct f2fs_sb_info *sbi;
  38
  39         if (!mapping)
  40                 return false;
  41
  42         inode = mapping->host;
  43         sbi = F2FS_I_SB(inode);
  44
  45         if (inode->i_ino == F2FS_META_INO(sbi) ||
  46                         inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47                         S_ISDIR(inode->i_mode) ||
  48                         (S_ISREG(inode->i_mode) &&
  49                         (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
  50                         is_cold_data(page))
  51                 return true;
  52         return false;
  53 }
  54
  55 static enum count_type __read_io_type(struct page *page)
  56 {
  57         struct address_space *mapping = page->mapping;
  58
  59         if (mapping) {
  60                 struct inode *inode = mapping->host;
  61                 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  62
  63                 if (inode->i_ino == F2FS_META_INO(sbi))
  64                         return F2FS_RD_META;
  65
  66                 if (inode->i_ino == F2FS_NODE_INO(sbi))
  67                         return F2FS_RD_NODE;
  68         }
  69         return F2FS_RD_DATA;
  70 }
  71
  72 /* postprocessing steps for read bios */
  73 enum bio_post_read_step {
  74         STEP_INITIAL = 0,
  75         STEP_DECRYPT,
  76 };
  77
  78 struct bio_post_read_ctx {
  79         struct bio *bio;
  80         struct work_struct work;
  81         unsigned int cur_step;
  82         unsigned int enabled_steps;
  83 };
  84
  85 static void __read_end_io(struct bio *bio)
  86 {
  87         struct page *page;
  88         struct bio_vec *bv;
  89         int i;
  90         struct bvec_iter_all iter_all;
  91
  92         bio_for_each_segment_all(bv, bio, i, 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         int i;
 168         struct bvec_iter_all iter_all;
 169
 170         if (time_to_inject(sbi, FAULT_WRITE_IO)) {
 171                 f2fs_show_injection_info(FAULT_WRITE_IO);
 172                 bio->bi_status = BLK_STS_IOERR;
 173         }
 174
 175         bio_for_each_segment_all(bvec, bio, i, iter_all) {
 176                 struct page *page = bvec->bv_page;
 177                 enum count_type type = WB_DATA_TYPE(page);
 178
 179                 if (IS_DUMMY_WRITTEN_PAGE(page)) {
 180                         set_page_private(page, (unsigned long)NULL);
 181                         ClearPagePrivate(page);
 182                         unlock_page(page);
 183                         mempool_free(page, sbi->write_io_dummy);
 184
 185                         if (unlikely(bio->bi_status))
 186                                 f2fs_stop_checkpoint(sbi, true);
 187                         continue;
 188                 }
 189
 190                 fscrypt_pullback_bio_page(&page, true);
 191
 192                 if (unlikely(bio->bi_status)) {
 193                         mapping_set_error(page->mapping, -EIO);
 194                         if (type == F2FS_WB_CP_DATA)
 195                                 f2fs_stop_checkpoint(sbi, true);
 196                 }
 197
 198                 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 199                                         page->index != nid_of_node(page));
 200
 201                 dec_page_count(sbi, type);
 202                 if (f2fs_in_warm_node_list(sbi, page))
 203                         f2fs_del_fsync_node_entry(sbi, page);
 204                 clear_cold_data(page);
 205                 end_page_writeback(page);
 206         }
 207         if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 208                                 wq_has_sleeper(&sbi->cp_wait))
 209                 wake_up(&sbi->cp_wait);
 210
 211         bio_put(bio);
 212 }
 213
 214 /*
 215  * Return true, if pre_bio's bdev is same as its target device.
 216  */
 217 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 218                                 block_t blk_addr, struct bio *bio)
 219 {
 220         struct block_device *bdev = sbi->sb->s_bdev;
 221         int i;
 222
 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         if (bio) {
 232                 bio_set_dev(bio, bdev);
 233                 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 234         }
 235         return bdev;
 236 }
 237
 238 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 239 {
 240         int i;
 241
 242         for (i = 0; i < sbi->s_ndevs; i++)
 243                 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 244                         return i;
 245         return 0;
 246 }
 247
 248 static bool __same_bdev(struct f2fs_sb_info *sbi,
 249                                 block_t blk_addr, struct bio *bio)
 250 {
 251         struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 252         return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 253 }
 254
 255 /*
 256  * Low-level block read/write IO operations.
 257  */
 258 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 259                                 struct writeback_control *wbc,
 260                                 int npages, bool is_read,
 261                                 enum page_type type, enum temp_type temp)
 262 {
 263         struct bio *bio;
 264
 265         bio = f2fs_bio_alloc(sbi, npages, true);
 266
 267         f2fs_target_device(sbi, blk_addr, bio);
 268         if (is_read) {
 269                 bio->bi_end_io = f2fs_read_end_io;
 270                 bio->bi_private = NULL;
 271         } else {
 272                 bio->bi_end_io = f2fs_write_end_io;
 273                 bio->bi_private = sbi;
 274                 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
 275         }
 276         if (wbc)
 277                 wbc_init_bio(wbc, bio);
 278
 279         return bio;
 280 }
 281
 282 static inline void __submit_bio(struct f2fs_sb_info *sbi,
 283                                 struct bio *bio, enum page_type type)
 284 {
 285         if (!is_read_io(bio_op(bio))) {
 286                 unsigned int start;
 287
 288                 if (type != DATA && type != NODE)
 289                         goto submit_io;
 290
 291                 if (test_opt(sbi, LFS) && current->plug)
 292                         blk_finish_plug(current->plug);
 293
 294                 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 295                 start %= F2FS_IO_SIZE(sbi);
 296
 297                 if (start == 0)
 298                         goto submit_io;
 299
 300                 /* fill dummy pages */
 301                 for (; start < F2FS_IO_SIZE(sbi); start++) {
 302                         struct page *page =
 303                                 mempool_alloc(sbi->write_io_dummy,
 304                                               GFP_NOIO | __GFP_NOFAIL);
 305                         f2fs_bug_on(sbi, !page);
 306
 307                         zero_user_segment(page, 0, PAGE_SIZE);
 308                         SetPagePrivate(page);
 309                         set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 310                         lock_page(page);
 311                         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 312                                 f2fs_bug_on(sbi, 1);
 313                 }
 314                 /*
 315                  * In the NODE case, we lose next block address chain. So, we
 316                  * need to do checkpoint in f2fs_sync_file.
 317                  */
 318                 if (type == NODE)
 319                         set_sbi_flag(sbi, SBI_NEED_CP);
 320         }
 321 submit_io:
 322         if (is_read_io(bio_op(bio)))
 323                 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 324         else
 325                 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 326         submit_bio(bio);
 327 }
 328
 329 static void __submit_merged_bio(struct f2fs_bio_info *io)
 330 {
 331         struct f2fs_io_info *fio = &io->fio;
 332
 333         if (!io->bio)
 334                 return;
 335
 336         bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 337
 338         if (is_read_io(fio->op))
 339                 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 340         else
 341                 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 342
 343         __submit_bio(io->sbi, io->bio, fio->type);
 344         io->bio = NULL;
 345 }
 346
 347 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
 348                                                 struct page *page, nid_t ino)
 349 {
 350         struct bio_vec *bvec;
 351         struct page *target;
 352         int i;
 353         struct bvec_iter_all iter_all;
 354
 355         if (!io->bio)
 356                 return false;
 357
 358         if (!inode && !page && !ino)
 359                 return true;
 360
 361         bio_for_each_segment_all(bvec, io->bio, i, iter_all) {
 362
 363                 if (bvec->bv_page->mapping)
 364                         target = bvec->bv_page;
 365                 else
 366                         target = fscrypt_control_page(bvec->bv_page);
 367
 368                 if (inode && inode == target->mapping->host)
 369                         return true;
 370                 if (page && page == target)
 371                         return true;
 372                 if (ino && ino == ino_of_node(target))
 373                         return true;
 374         }
 375
 376         return false;
 377 }
 378
 379 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 380                                 enum page_type type, enum temp_type temp)
 381 {
 382         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 383         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 384
 385         down_write(&io->io_rwsem);
 386
 387         /* change META to META_FLUSH in the checkpoint procedure */
 388         if (type >= META_FLUSH) {
 389                 io->fio.type = META_FLUSH;
 390                 io->fio.op = REQ_OP_WRITE;
 391                 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 392                 if (!test_opt(sbi, NOBARRIER))
 393                         io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 394         }
 395         __submit_merged_bio(io);
 396         up_write(&io->io_rwsem);
 397 }
 398
 399 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 400                                 struct inode *inode, struct page *page,
 401                                 nid_t ino, enum page_type type, bool force)
 402 {
 403         enum temp_type temp;
 404         bool ret = true;
 405
 406         for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 407                 if (!force)     {
 408                         enum page_type btype = PAGE_TYPE_OF_BIO(type);
 409                         struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 410
 411                         down_read(&io->io_rwsem);
 412                         ret = __has_merged_page(io, inode, page, ino);
 413                         up_read(&io->io_rwsem);
 414                 }
 415                 if (ret)
 416                         __f2fs_submit_merged_write(sbi, type, temp);
 417
 418                 /* TODO: use HOT temp only for meta pages now. */
 419                 if (type >= META)
 420                         break;
 421         }
 422 }
 423
 424 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 425 {
 426         __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 427 }
 428
 429 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 430                                 struct inode *inode, struct page *page,
 431                                 nid_t ino, enum page_type type)
 432 {
 433         __submit_merged_write_cond(sbi, inode, page, ino, type, false);
 434 }
 435
 436 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 437 {
 438         f2fs_submit_merged_write(sbi, DATA);
 439         f2fs_submit_merged_write(sbi, NODE);
 440         f2fs_submit_merged_write(sbi, META);
 441 }
 442
 443 /*
 444  * Fill the locked page with data located in the block address.
 445  * A caller needs to unlock the page on failure.
 446  */
 447 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 448 {
 449         struct bio *bio;
 450         struct page *page = fio->encrypted_page ?
 451                         fio->encrypted_page : fio->page;
 452
 453         if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 454                         __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
 455                 return -EFAULT;
 456
 457         trace_f2fs_submit_page_bio(page, fio);
 458         f2fs_trace_ios(fio, 0);
 459
 460         /* Allocate a new bio */
 461         bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 462                                 1, is_read_io(fio->op), fio->type, fio->temp);
 463
 464         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 465                 bio_put(bio);
 466                 return -EFAULT;
 467         }
 468
 469         if (fio->io_wbc && !is_read_io(fio->op))
 470                 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
 471
 472         bio_set_op_attrs(bio, fio->op, fio->op_flags);
 473
 474         inc_page_count(fio->sbi, is_read_io(fio->op) ?
 475                         __read_io_type(page): WB_DATA_TYPE(fio->page));
 476
 477         __submit_bio(fio->sbi, bio, fio->type);
 478         return 0;
 479 }
 480
 481 void f2fs_submit_page_write(struct f2fs_io_info *fio)
 482 {
 483         struct f2fs_sb_info *sbi = fio->sbi;
 484         enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 485         struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 486         struct page *bio_page;
 487
 488         f2fs_bug_on(sbi, is_read_io(fio->op));
 489
 490         down_write(&io->io_rwsem);
 491 next:
 492         if (fio->in_list) {
 493                 spin_lock(&io->io_lock);
 494                 if (list_empty(&io->io_list)) {
 495                         spin_unlock(&io->io_lock);
 496                         goto out;
 497                 }
 498                 fio = list_first_entry(&io->io_list,
 499                                                 struct f2fs_io_info, list);
 500                 list_del(&fio->list);
 501                 spin_unlock(&io->io_lock);
 502         }
 503
 504         if (__is_valid_data_blkaddr(fio->old_blkaddr))
 505                 verify_block_addr(fio, fio->old_blkaddr);
 506         verify_block_addr(fio, fio->new_blkaddr);
 507
 508         bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 509
 510         /* set submitted = true as a return value */
 511         fio->submitted = true;
 512
 513         inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 514
 515         if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 516             (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 517                         !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 518                 __submit_merged_bio(io);
 519 alloc_new:
 520         if (io->bio == NULL) {
 521                 if ((fio->type == DATA || fio->type == NODE) &&
 522                                 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 523                         dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 524                         fio->retry = true;
 525                         goto skip;
 526                 }
 527                 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 528                                                 BIO_MAX_PAGES, false,
 529                                                 fio->type, fio->temp);
 530                 io->fio = *fio;
 531         }
 532
 533         if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 534                 __submit_merged_bio(io);
 535                 goto alloc_new;
 536         }
 537
 538         if (fio->io_wbc)
 539                 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 540
 541         io->last_block_in_bio = fio->new_blkaddr;
 542         f2fs_trace_ios(fio, 0);
 543
 544         trace_f2fs_submit_page_write(fio->page, fio);
 545 skip:
 546         if (fio->in_list)
 547                 goto next;
 548 out:
 549         if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
 550                                 f2fs_is_checkpoint_ready(sbi))
 551                 __submit_merged_bio(io);
 552         up_write(&io->io_rwsem);
 553 }
 554
 555 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 556                                         unsigned nr_pages, unsigned op_flag)
 557 {
 558         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 559         struct bio *bio;
 560         struct bio_post_read_ctx *ctx;
 561         unsigned int post_read_steps = 0;
 562
 563         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
 564                 return ERR_PTR(-EFAULT);
 565
 566         bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 567         if (!bio)
 568                 return ERR_PTR(-ENOMEM);
 569         f2fs_target_device(sbi, blkaddr, bio);
 570         bio->bi_end_io = f2fs_read_end_io;
 571         bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
 572
 573         if (f2fs_encrypted_file(inode))
 574                 post_read_steps |= 1 << STEP_DECRYPT;
 575         if (post_read_steps) {
 576                 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 577                 if (!ctx) {
 578                         bio_put(bio);
 579                         return ERR_PTR(-ENOMEM);
 580                 }
 581                 ctx->bio = bio;
 582                 ctx->enabled_steps = post_read_steps;
 583                 bio->bi_private = ctx;
 584         }
 585
 586         return bio;
 587 }
 588
 589 /* This can handle encryption stuffs */
 590 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 591                                                         block_t blkaddr)
 592 {
 593         struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
 594
 595         if (IS_ERR(bio))
 596                 return PTR_ERR(bio);
 597
 598         /* wait for GCed page writeback via META_MAPPING */
 599         f2fs_wait_on_block_writeback(inode, blkaddr);
 600
 601         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 602                 bio_put(bio);
 603                 return -EFAULT;
 604         }
 605         ClearPageError(page);
 606         inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
 607         __submit_bio(F2FS_I_SB(inode), bio, DATA);
 608         return 0;
 609 }
 610
 611 static void __set_data_blkaddr(struct dnode_of_data *dn)
 612 {
 613         struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 614         __le32 *addr_array;
 615         int base = 0;
 616
 617         if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 618                 base = get_extra_isize(dn->inode);
 619
 620         /* Get physical address of data block */
 621         addr_array = blkaddr_in_node(rn);
 622         addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 623 }
 624
 625 /*
 626  * Lock ordering for the change of data block address:
 627  * ->data_page
 628  *  ->node_page
 629  *    update block addresses in the node page
 630  */
 631 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
 632 {
 633         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 634         __set_data_blkaddr(dn);
 635         if (set_page_dirty(dn->node_page))
 636                 dn->node_changed = true;
 637 }
 638
 639 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 640 {
 641         dn->data_blkaddr = blkaddr;
 642         f2fs_set_data_blkaddr(dn);
 643         f2fs_update_extent_cache(dn);
 644 }
 645
 646 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
 647 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 648 {
 649         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 650         int err;
 651
 652         if (!count)
 653                 return 0;
 654
 655         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 656                 return -EPERM;
 657         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 658                 return err;
 659
 660         trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 661                                                 dn->ofs_in_node, count);
 662
 663         f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
 664
 665         for (; count > 0; dn->ofs_in_node++) {
 666                 block_t blkaddr = datablock_addr(dn->inode,
 667                                         dn->node_page, dn->ofs_in_node);
 668                 if (blkaddr == NULL_ADDR) {
 669                         dn->data_blkaddr = NEW_ADDR;
 670                         __set_data_blkaddr(dn);
 671                         count--;
 672                 }
 673         }
 674
 675         if (set_page_dirty(dn->node_page))
 676                 dn->node_changed = true;
 677         return 0;
 678 }
 679
 680 /* Should keep dn->ofs_in_node unchanged */
 681 int f2fs_reserve_new_block(struct dnode_of_data *dn)
 682 {
 683         unsigned int ofs_in_node = dn->ofs_in_node;
 684         int ret;
 685
 686         ret = f2fs_reserve_new_blocks(dn, 1);
 687         dn->ofs_in_node = ofs_in_node;
 688         return ret;
 689 }
 690
 691 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 692 {
 693         bool need_put = dn->inode_page ? false : true;
 694         int err;
 695
 696         err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
 697         if (err)
 698                 return err;
 699
 700         if (dn->data_blkaddr == NULL_ADDR)
 701                 err = f2fs_reserve_new_block(dn);
 702         if (err || need_put)
 703                 f2fs_put_dnode(dn);
 704         return err;
 705 }
 706
 707 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 708 {
 709         struct extent_info ei  = {0,0,0};
 710         struct inode *inode = dn->inode;
 711
 712         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 713                 dn->data_blkaddr = ei.blk + index - ei.fofs;
 714                 return 0;
 715         }
 716
 717         return f2fs_reserve_block(dn, index);
 718 }
 719
 720 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 721                                                 int op_flags, bool for_write)
 722 {
 723         struct address_space *mapping = inode->i_mapping;
 724         struct dnode_of_data dn;
 725         struct page *page;
 726         struct extent_info ei = {0,0,0};
 727         int err;
 728
 729         page = f2fs_grab_cache_page(mapping, index, for_write);
 730         if (!page)
 731                 return ERR_PTR(-ENOMEM);
 732
 733         if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 734                 dn.data_blkaddr = ei.blk + index - ei.fofs;
 735                 goto got_it;
 736         }
 737
 738         set_new_dnode(&dn, inode, NULL, NULL, 0);
 739         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
 740         if (err)
 741                 goto put_err;
 742         f2fs_put_dnode(&dn);
 743
 744         if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 745                 err = -ENOENT;
 746                 goto put_err;
 747         }
 748 got_it:
 749         if (PageUptodate(page)) {
 750                 unlock_page(page);
 751                 return page;
 752         }
 753
 754         /*
 755          * A new dentry page is allocated but not able to be written, since its
 756          * new inode page couldn't be allocated due to -ENOSPC.
 757          * In such the case, its blkaddr can be remained as NEW_ADDR.
 758          * see, f2fs_add_link -> f2fs_get_new_data_page ->
 759          * f2fs_init_inode_metadata.
 760          */
 761         if (dn.data_blkaddr == NEW_ADDR) {
 762                 zero_user_segment(page, 0, PAGE_SIZE);
 763                 if (!PageUptodate(page))
 764                         SetPageUptodate(page);
 765                 unlock_page(page);
 766                 return page;
 767         }
 768
 769         err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 770         if (err)
 771                 goto put_err;
 772         return page;
 773
 774 put_err:
 775         f2fs_put_page(page, 1);
 776         return ERR_PTR(err);
 777 }
 778
 779 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
 780 {
 781         struct address_space *mapping = inode->i_mapping;
 782         struct page *page;
 783
 784         page = find_get_page(mapping, index);
 785         if (page && PageUptodate(page))
 786                 return page;
 787         f2fs_put_page(page, 0);
 788
 789         page = f2fs_get_read_data_page(inode, index, 0, false);
 790         if (IS_ERR(page))
 791                 return page;
 792
 793         if (PageUptodate(page))
 794                 return page;
 795
 796         wait_on_page_locked(page);
 797         if (unlikely(!PageUptodate(page))) {
 798                 f2fs_put_page(page, 0);
 799                 return ERR_PTR(-EIO);
 800         }
 801         return page;
 802 }
 803
 804 /*
 805  * If it tries to access a hole, return an error.
 806  * Because, the callers, functions in dir.c and GC, should be able to know
 807  * whether this page exists or not.
 808  */
 809 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
 810                                                         bool for_write)
 811 {
 812         struct address_space *mapping = inode->i_mapping;
 813         struct page *page;
 814 repeat:
 815         page = f2fs_get_read_data_page(inode, index, 0, for_write);
 816         if (IS_ERR(page))
 817                 return page;
 818
 819         /* wait for read completion */
 820         lock_page(page);
 821         if (unlikely(page->mapping != mapping)) {
 822                 f2fs_put_page(page, 1);
 823                 goto repeat;
 824         }
 825         if (unlikely(!PageUptodate(page))) {
 826                 f2fs_put_page(page, 1);
 827                 return ERR_PTR(-EIO);
 828         }
 829         return page;
 830 }
 831
 832 /*
 833  * Caller ensures that this data page is never allocated.
 834  * A new zero-filled data page is allocated in the page cache.
 835  *
 836  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 837  * f2fs_unlock_op().
 838  * Note that, ipage is set only by make_empty_dir, and if any error occur,
 839  * ipage should be released by this function.
 840  */
 841 struct page *f2fs_get_new_data_page(struct inode *inode,
 842                 struct page *ipage, pgoff_t index, bool new_i_size)
 843 {
 844         struct address_space *mapping = inode->i_mapping;
 845         struct page *page;
 846         struct dnode_of_data dn;
 847         int err;
 848
 849         page = f2fs_grab_cache_page(mapping, index, true);
 850         if (!page) {
 851                 /*
 852                  * before exiting, we should make sure ipage will be released
 853                  * if any error occur.
 854                  */
 855                 f2fs_put_page(ipage, 1);
 856                 return ERR_PTR(-ENOMEM);
 857         }
 858
 859         set_new_dnode(&dn, inode, ipage, NULL, 0);
 860         err = f2fs_reserve_block(&dn, index);
 861         if (err) {
 862                 f2fs_put_page(page, 1);
 863                 return ERR_PTR(err);
 864         }
 865         if (!ipage)
 866                 f2fs_put_dnode(&dn);
 867
 868         if (PageUptodate(page))
 869                 goto got_it;
 870
 871         if (dn.data_blkaddr == NEW_ADDR) {
 872                 zero_user_segment(page, 0, PAGE_SIZE);
 873                 if (!PageUptodate(page))
 874                         SetPageUptodate(page);
 875         } else {
 876                 f2fs_put_page(page, 1);
 877
 878                 /* if ipage exists, blkaddr should be NEW_ADDR */
 879                 f2fs_bug_on(F2FS_I_SB(inode), ipage);
 880                 page = f2fs_get_lock_data_page(inode, index, true);
 881                 if (IS_ERR(page))
 882                         return page;
 883         }
 884 got_it:
 885         if (new_i_size && i_size_read(inode) <
 886                                 ((loff_t)(index + 1) << PAGE_SHIFT))
 887                 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 888         return page;
 889 }
 890
 891 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 892 {
 893         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 894         struct f2fs_summary sum;
 895         struct node_info ni;
 896         block_t old_blkaddr;
 897         blkcnt_t count = 1;
 898         int err;
 899
 900         if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 901                 return -EPERM;
 902
 903         err = f2fs_get_node_info(sbi, dn->nid, &ni);
 904         if (err)
 905                 return err;
 906
 907         dn->data_blkaddr = datablock_addr(dn->inode,
 908                                 dn->node_page, dn->ofs_in_node);
 909         if (dn->data_blkaddr != NULL_ADDR)
 910                 goto alloc;
 911
 912         if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 913                 return err;
 914
 915 alloc:
 916         set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 917         old_blkaddr = dn->data_blkaddr;
 918         f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
 919                                         &sum, seg_type, NULL, false);
 920         if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
 921                 invalidate_mapping_pages(META_MAPPING(sbi),
 922                                         old_blkaddr, old_blkaddr);
 923         f2fs_set_data_blkaddr(dn);
 924
 925         /*
 926          * i_size will be updated by direct_IO. Otherwise, we'll get stale
 927          * data from unwritten block via dio_read.
 928          */
 929         return 0;
 930 }
 931
 932 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 933 {
 934         struct inode *inode = file_inode(iocb->ki_filp);
 935         struct f2fs_map_blocks map;
 936         int flag;
 937         int err = 0;
 938         bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 939
 940         /* convert inline data for Direct I/O*/
 941         if (direct_io) {
 942                 err = f2fs_convert_inline_inode(inode);
 943                 if (err)
 944                         return err;
 945         }
 946
 947         if (direct_io && allow_outplace_dio(inode, iocb, from))
 948                 return 0;
 949
 950         if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 951                 return 0;
 952
 953         map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 954         map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 955         if (map.m_len > map.m_lblk)
 956                 map.m_len -= map.m_lblk;
 957         else
 958                 map.m_len = 0;
 959
 960         map.m_next_pgofs = NULL;
 961         map.m_next_extent = NULL;
 962         map.m_seg_type = NO_CHECK_TYPE;
 963         map.m_may_create = true;
 964
 965         if (direct_io) {
 966                 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
 967                 flag = f2fs_force_buffered_io(inode, iocb, from) ?
 968                                         F2FS_GET_BLOCK_PRE_AIO :
 969                                         F2FS_GET_BLOCK_PRE_DIO;
 970                 goto map_blocks;
 971         }
 972         if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 973                 err = f2fs_convert_inline_inode(inode);
 974                 if (err)
 975                         return err;
 976         }
 977         if (f2fs_has_inline_data(inode))
 978                 return err;
 979
 980         flag = F2FS_GET_BLOCK_PRE_AIO;
 981
 982 map_blocks:
 983         err = f2fs_map_blocks(inode, &map, 1, flag);
 984         if (map.m_len > 0 && err == -ENOSPC) {
 985                 if (!direct_io)
 986                         set_inode_flag(inode, FI_NO_PREALLOC);
 987                 err = 0;
 988         }
 989         return err;
 990 }
 991
 992 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 993 {
 994         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 995                 if (lock)
 996                         down_read(&sbi->node_change);
 997                 else
 998                         up_read(&sbi->node_change);
 999         } else {
1000                 if (lock)
1001                         f2fs_lock_op(sbi);
1002                 else
1003                         f2fs_unlock_op(sbi);
1004         }
1005 }
1006
1007 /*
1008  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1009  * f2fs_map_blocks structure.
1010  * If original data blocks are allocated, then give them to blockdev.
1011  * Otherwise,
1012  *     a. preallocate requested block addresses
1013  *     b. do not use extent cache for better performance
1014  *     c. give the block addresses to blockdev
1015  */
1016 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1017                                                 int create, int flag)
1018 {
1019         unsigned int maxblocks = map->m_len;
1020         struct dnode_of_data dn;
1021         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1022         int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1023         pgoff_t pgofs, end_offset, end;
1024         int err = 0, ofs = 1;
1025         unsigned int ofs_in_node, last_ofs_in_node;
1026         blkcnt_t prealloc;
1027         struct extent_info ei = {0,0,0};
1028         block_t blkaddr;
1029         unsigned int start_pgofs;
1030
1031         if (!maxblocks)
1032                 return 0;
1033
1034         map->m_len = 0;
1035         map->m_flags = 0;
1036
1037         /* it only supports block size == page size */
1038         pgofs = (pgoff_t)map->m_lblk;
1039         end = pgofs + maxblocks;
1040
1041         if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1042                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1043                                                         map->m_may_create)
1044                         goto next_dnode;
1045
1046                 map->m_pblk = ei.blk + pgofs - ei.fofs;
1047                 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1048                 map->m_flags = F2FS_MAP_MAPPED;
1049                 if (map->m_next_extent)
1050                         *map->m_next_extent = pgofs + map->m_len;
1051
1052                 /* for hardware encryption, but to avoid potential issue in future */
1053                 if (flag == F2FS_GET_BLOCK_DIO)
1054                         f2fs_wait_on_block_writeback_range(inode,
1055                                                 map->m_pblk, map->m_len);
1056                 goto out;
1057         }
1058
1059 next_dnode:
1060         if (map->m_may_create)
1061                 __do_map_lock(sbi, flag, true);
1062
1063         /* When reading holes, we need its node page */
1064         set_new_dnode(&dn, inode, NULL, NULL, 0);
1065         err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1066         if (err) {
1067                 if (flag == F2FS_GET_BLOCK_BMAP)
1068                         map->m_pblk = 0;
1069                 if (err == -ENOENT) {
1070                         err = 0;
1071                         if (map->m_next_pgofs)
1072                                 *map->m_next_pgofs =
1073                                         f2fs_get_next_page_offset(&dn, pgofs);
1074                         if (map->m_next_extent)
1075                                 *map->m_next_extent =
1076                                         f2fs_get_next_page_offset(&dn, pgofs);
1077                 }
1078                 goto unlock_out;
1079         }
1080
1081         start_pgofs = pgofs;
1082         prealloc = 0;
1083         last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1084         end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1085
1086 next_block:
1087         blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1088
1089         if (__is_valid_data_blkaddr(blkaddr) &&
1090                 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1091                 err = -EFAULT;
1092                 goto sync_out;
1093         }
1094
1095         if (is_valid_data_blkaddr(sbi, blkaddr)) {
1096                 /* use out-place-update for driect IO under LFS mode */
1097                 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1098                                                         map->m_may_create) {
1099                         err = __allocate_data_block(&dn, map->m_seg_type);
1100                         if (!err) {
1101                                 blkaddr = dn.data_blkaddr;
1102                                 set_inode_flag(inode, FI_APPEND_WRITE);
1103                         }
1104                 }
1105         } else {
1106                 if (create) {
1107                         if (unlikely(f2fs_cp_error(sbi))) {
1108                                 err = -EIO;
1109                                 goto sync_out;
1110                         }
1111                         if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1112                                 if (blkaddr == NULL_ADDR) {
1113                                         prealloc++;
1114                                         last_ofs_in_node = dn.ofs_in_node;
1115                                 }
1116                         } else {
1117                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1118                                         flag != F2FS_GET_BLOCK_DIO);
1119                                 err = __allocate_data_block(&dn,
1120                                                         map->m_seg_type);
1121                                 if (!err)
1122                                         set_inode_flag(inode, FI_APPEND_WRITE);
1123                         }
1124                         if (err)
1125                                 goto sync_out;
1126                         map->m_flags |= F2FS_MAP_NEW;
1127                         blkaddr = dn.data_blkaddr;
1128                 } else {
1129                         if (flag == F2FS_GET_BLOCK_BMAP) {
1130                                 map->m_pblk = 0;
1131                                 goto sync_out;
1132                         }
1133                         if (flag == F2FS_GET_BLOCK_PRECACHE)
1134                                 goto sync_out;
1135                         if (flag == F2FS_GET_BLOCK_FIEMAP &&
1136                                                 blkaddr == NULL_ADDR) {
1137                                 if (map->m_next_pgofs)
1138                                         *map->m_next_pgofs = pgofs + 1;
1139                                 goto sync_out;
1140                         }
1141                         if (flag != F2FS_GET_BLOCK_FIEMAP) {
1142                                 /* for defragment case */
1143                                 if (map->m_next_pgofs)
1144                                         *map->m_next_pgofs = pgofs + 1;
1145                                 goto sync_out;
1146                         }
1147                 }
1148         }
1149
1150         if (flag == F2FS_GET_BLOCK_PRE_AIO)
1151                 goto skip;
1152
1153         if (map->m_len == 0) {
1154                 /* preallocated unwritten block should be mapped for fiemap. */
1155                 if (blkaddr == NEW_ADDR)
1156                         map->m_flags |= F2FS_MAP_UNWRITTEN;
1157                 map->m_flags |= F2FS_MAP_MAPPED;
1158
1159                 map->m_pblk = blkaddr;
1160                 map->m_len = 1;
1161         } else if ((map->m_pblk != NEW_ADDR &&
1162                         blkaddr == (map->m_pblk + ofs)) ||
1163                         (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1164                         flag == F2FS_GET_BLOCK_PRE_DIO) {
1165                 ofs++;
1166                 map->m_len++;
1167         } else {
1168                 goto sync_out;
1169         }
1170
1171 skip:
1172         dn.ofs_in_node++;
1173         pgofs++;
1174
1175         /* preallocate blocks in batch for one dnode page */
1176         if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1177                         (pgofs == end || dn.ofs_in_node == end_offset)) {
1178
1179                 dn.ofs_in_node = ofs_in_node;
1180                 err = f2fs_reserve_new_blocks(&dn, prealloc);
1181                 if (err)
1182                         goto sync_out;
1183
1184                 map->m_len += dn.ofs_in_node - ofs_in_node;
1185                 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1186                         err = -ENOSPC;
1187                         goto sync_out;
1188                 }
1189                 dn.ofs_in_node = end_offset;
1190         }
1191
1192         if (pgofs >= end)
1193                 goto sync_out;
1194         else if (dn.ofs_in_node < end_offset)
1195                 goto next_block;
1196
1197         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1198                 if (map->m_flags & F2FS_MAP_MAPPED) {
1199                         unsigned int ofs = start_pgofs - map->m_lblk;
1200
1201                         f2fs_update_extent_cache_range(&dn,
1202                                 start_pgofs, map->m_pblk + ofs,
1203                                 map->m_len - ofs);
1204                 }
1205         }
1206
1207         f2fs_put_dnode(&dn);
1208
1209         if (map->m_may_create) {
1210                 __do_map_lock(sbi, flag, false);
1211                 f2fs_balance_fs(sbi, dn.node_changed);
1212         }
1213         goto next_dnode;
1214
1215 sync_out:
1216
1217         /* for hardware encryption, but to avoid potential issue in future */
1218         if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1219                 f2fs_wait_on_block_writeback_range(inode,
1220                                                 map->m_pblk, map->m_len);
1221
1222         if (flag == F2FS_GET_BLOCK_PRECACHE) {
1223                 if (map->m_flags & F2FS_MAP_MAPPED) {
1224                         unsigned int ofs = start_pgofs - map->m_lblk;
1225
1226                         f2fs_update_extent_cache_range(&dn,
1227                                 start_pgofs, map->m_pblk + ofs,
1228                                 map->m_len - ofs);
1229                 }
1230                 if (map->m_next_extent)
1231                         *map->m_next_extent = pgofs + 1;
1232         }
1233         f2fs_put_dnode(&dn);
1234 unlock_out:
1235         if (map->m_may_create) {
1236                 __do_map_lock(sbi, flag, false);
1237                 f2fs_balance_fs(sbi, dn.node_changed);
1238         }
1239 out:
1240         trace_f2fs_map_blocks(inode, map, err);
1241         return err;
1242 }
1243
1244 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1245 {
1246         struct f2fs_map_blocks map;
1247         block_t last_lblk;
1248         int err;
1249
1250         if (pos + len > i_size_read(inode))
1251                 return false;
1252
1253         map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1254         map.m_next_pgofs = NULL;
1255         map.m_next_extent = NULL;
1256         map.m_seg_type = NO_CHECK_TYPE;
1257         map.m_may_create = false;
1258         last_lblk = F2FS_BLK_ALIGN(pos + len);
1259
1260         while (map.m_lblk < last_lblk) {
1261                 map.m_len = last_lblk - map.m_lblk;
1262                 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1263                 if (err || map.m_len == 0)
1264                         return false;
1265                 map.m_lblk += map.m_len;
1266         }
1267         return true;
1268 }
1269
1270 static int __get_data_block(struct inode *inode, sector_t iblock,
1271                         struct buffer_head *bh, int create, int flag,
1272                         pgoff_t *next_pgofs, int seg_type, bool may_write)
1273 {
1274         struct f2fs_map_blocks map;
1275         int err;
1276
1277         map.m_lblk = iblock;
1278         map.m_len = bh->b_size >> inode->i_blkbits;
1279         map.m_next_pgofs = next_pgofs;
1280         map.m_next_extent = NULL;
1281         map.m_seg_type = seg_type;
1282         map.m_may_create = may_write;
1283
1284         err = f2fs_map_blocks(inode, &map, create, flag);
1285         if (!err) {
1286                 map_bh(bh, inode->i_sb, map.m_pblk);
1287                 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1288                 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1289         }
1290         return err;
1291 }
1292
1293 static int get_data_block(struct inode *inode, sector_t iblock,
1294                         struct buffer_head *bh_result, int create, int flag,
1295                         pgoff_t *next_pgofs)
1296 {
1297         return __get_data_block(inode, iblock, bh_result, create,
1298                                                         flag, next_pgofs,
1299                                                         NO_CHECK_TYPE, create);
1300 }
1301
1302 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1303                         struct buffer_head *bh_result, int create)
1304 {
1305         return __get_data_block(inode, iblock, bh_result, create,
1306                                 F2FS_GET_BLOCK_DIO, NULL,
1307                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1308                                 true);
1309 }
1310
1311 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1312                         struct buffer_head *bh_result, int create)
1313 {
1314         return __get_data_block(inode, iblock, bh_result, create,
1315                                 F2FS_GET_BLOCK_DIO, NULL,
1316                                 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1317                                 false);
1318 }
1319
1320 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1321                         struct buffer_head *bh_result, int create)
1322 {
1323         /* Block number less than F2FS MAX BLOCKS */
1324         if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1325                 return -EFBIG;
1326
1327         return __get_data_block(inode, iblock, bh_result, create,
1328                                                 F2FS_GET_BLOCK_BMAP, NULL,
1329                                                 NO_CHECK_TYPE, create);
1330 }
1331
1332 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1333 {
1334         return (offset >> inode->i_blkbits);
1335 }
1336
1337 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1338 {
1339         return (blk << inode->i_blkbits);
1340 }
1341
1342 static int f2fs_xattr_fiemap(struct inode *inode,
1343                                 struct fiemap_extent_info *fieinfo)
1344 {
1345         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1346         struct page *page;
1347         struct node_info ni;
1348         __u64 phys = 0, len;
1349         __u32 flags;
1350         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1351         int err = 0;
1352
1353         if (f2fs_has_inline_xattr(inode)) {
1354                 int offset;
1355
1356                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1357                                                 inode->i_ino, false);
1358                 if (!page)
1359                         return -ENOMEM;
1360
1361                 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1362                 if (err) {
1363                         f2fs_put_page(page, 1);
1364                         return err;
1365                 }
1366
1367                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1368                 offset = offsetof(struct f2fs_inode, i_addr) +
1369                                         sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1370                                         get_inline_xattr_addrs(inode));
1371
1372                 phys += offset;
1373                 len = inline_xattr_size(inode);
1374
1375                 f2fs_put_page(page, 1);
1376
1377                 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1378
1379                 if (!xnid)
1380                         flags |= FIEMAP_EXTENT_LAST;
1381
1382                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1383                 if (err || err == 1)
1384                         return err;
1385         }
1386
1387         if (xnid) {
1388                 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1389                 if (!page)
1390                         return -ENOMEM;
1391
1392                 err = f2fs_get_node_info(sbi, xnid, &ni);
1393                 if (err) {
1394                         f2fs_put_page(page, 1);
1395                         return err;
1396                 }
1397
1398                 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1399                 len = inode->i_sb->s_blocksize;
1400
1401                 f2fs_put_page(page, 1);
1402
1403                 flags = FIEMAP_EXTENT_LAST;
1404         }
1405
1406         if (phys)
1407                 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1408
1409         return (err < 0 ? err : 0);
1410 }
1411
1412 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1413                 u64 start, u64 len)
1414 {
1415         struct buffer_head map_bh;
1416         sector_t start_blk, last_blk;
1417         pgoff_t next_pgofs;
1418         u64 logical = 0, phys = 0, size = 0;
1419         u32 flags = 0;
1420         int ret = 0;
1421
1422         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1423                 ret = f2fs_precache_extents(inode);
1424                 if (ret)
1425                         return ret;
1426         }
1427
1428         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1429         if (ret)
1430                 return ret;
1431
1432         inode_lock(inode);
1433
1434         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1435                 ret = f2fs_xattr_fiemap(inode, fieinfo);
1436                 goto out;
1437         }
1438
1439         if (f2fs_has_inline_data(inode)) {
1440                 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1441                 if (ret != -EAGAIN)
1442                         goto out;
1443         }
1444
1445         if (logical_to_blk(inode, len) == 0)
1446                 len = blk_to_logical(inode, 1);
1447
1448         start_blk = logical_to_blk(inode, start);
1449         last_blk = logical_to_blk(inode, start + len - 1);
1450
1451 next:
1452         memset(&map_bh, 0, sizeof(struct buffer_head));
1453         map_bh.b_size = len;
1454
1455         ret = get_data_block(inode, start_blk, &map_bh, 0,
1456                                         F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1457         if (ret)
1458                 goto out;
1459
1460         /* HOLE */
1461         if (!buffer_mapped(&map_bh)) {
1462                 start_blk = next_pgofs;
1463
1464                 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1465                                         F2FS_I_SB(inode)->max_file_blocks))
1466                         goto prep_next;
1467
1468                 flags |= FIEMAP_EXTENT_LAST;
1469         }
1470
1471         if (size) {
1472                 if (IS_ENCRYPTED(inode))
1473                         flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1474
1475                 ret = fiemap_fill_next_extent(fieinfo, logical,
1476                                 phys, size, flags);
1477         }
1478
1479         if (start_blk > last_blk || ret)
1480                 goto out;
1481
1482         logical = blk_to_logical(inode, start_blk);
1483         phys = blk_to_logical(inode, map_bh.b_blocknr);
1484         size = map_bh.b_size;
1485         flags = 0;
1486         if (buffer_unwritten(&map_bh))
1487                 flags = FIEMAP_EXTENT_UNWRITTEN;
1488
1489         start_blk += logical_to_blk(inode, size);
1490
1491 prep_next:
1492         cond_resched();
1493         if (fatal_signal_pending(current))
1494                 ret = -EINTR;
1495         else
1496                 goto next;
1497 out:
1498         if (ret == 1)
1499                 ret = 0;
1500
1501         inode_unlock(inode);
1502         return ret;
1503 }
1504
1505 /*
1506  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1507  * Major change was from block_size == page_size in f2fs by default.
1508  *
1509  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1510  * this function ever deviates from doing just read-ahead, it should either
1511  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1512  * from read-ahead.
1513  */
1514 static int f2fs_mpage_readpages(struct address_space *mapping,
1515                         struct list_head *pages, struct page *page,
1516                         unsigned nr_pages, bool is_readahead)
1517 {
1518         struct bio *bio = NULL;
1519         sector_t last_block_in_bio = 0;
1520         struct inode *inode = mapping->host;
1521         const unsigned blkbits = inode->i_blkbits;
1522         const unsigned blocksize = 1 << blkbits;
1523         sector_t block_in_file;
1524         sector_t last_block;
1525         sector_t last_block_in_file;
1526         sector_t block_nr;
1527         struct f2fs_map_blocks map;
1528
1529         map.m_pblk = 0;
1530         map.m_lblk = 0;
1531         map.m_len = 0;
1532         map.m_flags = 0;
1533         map.m_next_pgofs = NULL;
1534         map.m_next_extent = NULL;
1535         map.m_seg_type = NO_CHECK_TYPE;
1536         map.m_may_create = false;
1537
1538         for (; nr_pages; nr_pages--) {
1539                 if (pages) {
1540                         page = list_last_entry(pages, struct page, lru);
1541
1542                         prefetchw(&page->flags);
1543                         list_del(&page->lru);
1544                         if (add_to_page_cache_lru(page, mapping,
1545                                                   page->index,
1546                                                   readahead_gfp_mask(mapping)))
1547                                 goto next_page;
1548                 }
1549
1550                 block_in_file = (sector_t)page->index;
1551                 last_block = block_in_file + nr_pages;
1552                 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1553                                                                 blkbits;
1554                 if (last_block > last_block_in_file)
1555                         last_block = last_block_in_file;
1556
1557                 /* just zeroing out page which is beyond EOF */
1558                 if (block_in_file >= last_block)
1559                         goto zero_out;
1560                 /*
1561                  * Map blocks using the previous result first.
1562                  */
1563                 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1564                                 block_in_file > map.m_lblk &&
1565                                 block_in_file < (map.m_lblk + map.m_len))
1566                         goto got_it;
1567
1568                 /*
1569                  * Then do more f2fs_map_blocks() calls until we are
1570                  * done with this page.
1571                  */
1572                 map.m_lblk = block_in_file;
1573                 map.m_len = last_block - block_in_file;
1574
1575                 if (f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT))
1576                         goto set_error_page;
1577 got_it:
1578                 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1579                         block_nr = map.m_pblk + block_in_file - map.m_lblk;
1580                         SetPageMappedToDisk(page);
1581
1582                         if (!PageUptodate(page) && !cleancache_get_page(page)) {
1583                                 SetPageUptodate(page);
1584                                 goto confused;
1585                         }
1586
1587                         if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1588                                                                 DATA_GENERIC))
1589                                 goto set_error_page;
1590                 } else {
1591 zero_out:
1592                         zero_user_segment(page, 0, PAGE_SIZE);
1593                         if (!PageUptodate(page))
1594                                 SetPageUptodate(page);
1595                         unlock_page(page);
1596                         goto next_page;
1597                 }
1598
1599                 /*
1600                  * This page will go to BIO.  Do we need to send this
1601                  * BIO off first?
1602                  */
1603                 if (bio && (last_block_in_bio != block_nr - 1 ||
1604                         !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1605 submit_and_realloc:
1606                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1607                         bio = NULL;
1608                 }
1609                 if (bio == NULL) {
1610                         bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1611                                         is_readahead ? REQ_RAHEAD : 0);
1612                         if (IS_ERR(bio)) {
1613                                 bio = NULL;
1614                                 goto set_error_page;
1615                         }
1616                 }
1617
1618                 /*
1619                  * If the page is under writeback, we need to wait for
1620                  * its completion to see the correct decrypted data.
1621                  */
1622                 f2fs_wait_on_block_writeback(inode, block_nr);
1623
1624                 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1625                         goto submit_and_realloc;
1626
1627                 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1628                 ClearPageError(page);
1629                 last_block_in_bio = block_nr;
1630                 goto next_page;
1631 set_error_page:
1632                 SetPageError(page);
1633                 zero_user_segment(page, 0, PAGE_SIZE);
1634                 unlock_page(page);
1635                 goto next_page;
1636 confused:
1637                 if (bio) {
1638                         __submit_bio(F2FS_I_SB(inode), bio, DATA);
1639                         bio = NULL;
1640                 }
1641                 unlock_page(page);
1642 next_page:
1643                 if (pages)
1644                         put_page(page);
1645         }
1646         BUG_ON(pages && !list_empty(pages));
1647         if (bio)
1648                 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1649         return 0;
1650 }
1651
1652 static int f2fs_read_data_page(struct file *file, struct page *page)
1653 {
1654         struct inode *inode = page->mapping->host;
1655         int ret = -EAGAIN;
1656
1657         trace_f2fs_readpage(page, DATA);
1658
1659         /* If the file has inline data, try to read it directly */
1660         if (f2fs_has_inline_data(inode))
1661                 ret = f2fs_read_inline_data(inode, page);
1662         if (ret == -EAGAIN)
1663                 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1664         return ret;
1665 }
1666
1667 static int f2fs_read_data_pages(struct file *file,
1668                         struct address_space *mapping,
1669                         struct list_head *pages, unsigned nr_pages)
1670 {
1671         struct inode *inode = mapping->host;
1672         struct page *page = list_last_entry(pages, struct page, lru);
1673
1674         trace_f2fs_readpages(inode, page, nr_pages);
1675
1676         /* If the file has inline data, skip readpages */
1677         if (f2fs_has_inline_data(inode))
1678                 return 0;
1679
1680         return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1681 }
1682
1683 static int encrypt_one_page(struct f2fs_io_info *fio)
1684 {
1685         struct inode *inode = fio->page->mapping->host;
1686         struct page *mpage;
1687         gfp_t gfp_flags = GFP_NOFS;
1688
1689         if (!f2fs_encrypted_file(inode))
1690                 return 0;
1691
1692         /* wait for GCed page writeback via META_MAPPING */
1693         f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1694
1695 retry_encrypt:
1696         fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1697                         PAGE_SIZE, 0, fio->page->index, gfp_flags);
1698         if (IS_ERR(fio->encrypted_page)) {
1699                 /* flush pending IOs and wait for a while in the ENOMEM case */
1700                 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1701                         f2fs_flush_merged_writes(fio->sbi);
1702                         congestion_wait(BLK_RW_ASYNC, HZ/50);
1703                         gfp_flags |= __GFP_NOFAIL;
1704                         goto retry_encrypt;
1705                 }
1706                 return PTR_ERR(fio->encrypted_page);
1707         }
1708
1709         mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1710         if (mpage) {
1711                 if (PageUptodate(mpage))
1712                         memcpy(page_address(mpage),
1713                                 page_address(fio->encrypted_page), PAGE_SIZE);
1714                 f2fs_put_page(mpage, 1);
1715         }
1716         return 0;
1717 }
1718
1719 static inline bool check_inplace_update_policy(struct inode *inode,
1720                                 struct f2fs_io_info *fio)
1721 {
1722         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1723         unsigned int policy = SM_I(sbi)->ipu_policy;
1724
1725         if (policy & (0x1 << F2FS_IPU_FORCE))
1726                 return true;
1727         if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1728                 return true;
1729         if (policy & (0x1 << F2FS_IPU_UTIL) &&
1730                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1731                 return true;
1732         if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1733                         utilization(sbi) > SM_I(sbi)->min_ipu_util)
1734                 return true;
1735
1736         /*
1737          * IPU for rewrite async pages
1738          */
1739         if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1740                         fio && fio->op == REQ_OP_WRITE &&
1741                         !(fio->op_flags & REQ_SYNC) &&
1742                         !IS_ENCRYPTED(inode))
1743                 return true;
1744
1745         /* this is only set during fdatasync */
1746         if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1747                         is_inode_flag_set(inode, FI_NEED_IPU))
1748                 return true;
1749
1750         if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1751                         !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1752                 return true;
1753
1754         return false;
1755 }
1756
1757 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1758 {
1759         if (f2fs_is_pinned_file(inode))
1760                 return true;
1761
1762         /* if this is cold file, we should overwrite to avoid fragmentation */
1763         if (file_is_cold(inode))
1764                 return true;
1765
1766         return check_inplace_update_policy(inode, fio);
1767 }
1768
1769 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1770 {
1771         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1772
1773         if (test_opt(sbi, LFS))
1774                 return true;
1775         if (S_ISDIR(inode->i_mode))
1776                 return true;
1777         if (IS_NOQUOTA(inode))
1778                 return true;
1779         if (f2fs_is_atomic_file(inode))
1780                 return true;
1781         if (fio) {
1782                 if (is_cold_data(fio->page))
1783                         return true;
1784                 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1785                         return true;
1786                 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1787                         f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1788                         return true;
1789         }
1790         return false;
1791 }
1792
1793 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1794 {
1795         struct inode *inode = fio->page->mapping->host;
1796
1797         if (f2fs_should_update_outplace(inode, fio))
1798                 return false;
1799
1800         return f2fs_should_update_inplace(inode, fio);
1801 }
1802
1803 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1804 {
1805         struct page *page = fio->page;
1806         struct inode *inode = page->mapping->host;
1807         struct dnode_of_data dn;
1808         struct extent_info ei = {0,0,0};
1809         struct node_info ni;
1810         bool ipu_force = false;
1811         int err = 0;
1812
1813         set_new_dnode(&dn, inode, NULL, NULL, 0);
1814         if (need_inplace_update(fio) &&
1815                         f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1816                 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1817
1818                 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1819                                                         DATA_GENERIC))
1820                         return -EFAULT;
1821
1822                 ipu_force = true;
1823                 fio->need_lock = LOCK_DONE;
1824                 goto got_it;
1825         }
1826
1827         /* Deadlock due to between page->lock and f2fs_lock_op */
1828         if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1829                 return -EAGAIN;
1830
1831         err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1832         if (err)
1833                 goto out;
1834
1835         fio->old_blkaddr = dn.data_blkaddr;
1836
1837         /* This page is already truncated */
1838         if (fio->old_blkaddr == NULL_ADDR) {
1839                 ClearPageUptodate(page);
1840                 clear_cold_data(page);
1841                 goto out_writepage;
1842         }
1843 got_it:
1844         if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1845                 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1846                                                         DATA_GENERIC)) {
1847                 err = -EFAULT;
1848                 goto out_writepage;
1849         }
1850         /*
1851          * If current allocation needs SSR,
1852          * it had better in-place writes for updated data.
1853          */
1854         if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1855                                         need_inplace_update(fio))) {
1856                 err = encrypt_one_page(fio);
1857                 if (err)
1858                         goto out_writepage;
1859
1860                 set_page_writeback(page);
1861                 ClearPageError(page);
1862                 f2fs_put_dnode(&dn);
1863                 if (fio->need_lock == LOCK_REQ)
1864                         f2fs_unlock_op(fio->sbi);
1865                 err = f2fs_inplace_write_data(fio);
1866                 if (err) {
1867                         if (f2fs_encrypted_file(inode))
1868                                 fscrypt_pullback_bio_page(&fio->encrypted_page,
1869                                                                         true);
1870                         if (PageWriteback(page))
1871                                 end_page_writeback(page);
1872                 }
1873                 trace_f2fs_do_write_data_page(fio->page, IPU);
1874                 set_inode_flag(inode, FI_UPDATE_WRITE);
1875                 return err;
1876         }
1877
1878         if (fio->need_lock == LOCK_RETRY) {
1879                 if (!f2fs_trylock_op(fio->sbi)) {
1880                         err = -EAGAIN;
1881                         goto out_writepage;
1882                 }
1883                 fio->need_lock = LOCK_REQ;
1884         }
1885
1886         err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1887         if (err)
1888                 goto out_writepage;
1889
1890         fio->version = ni.version;
1891
1892         err = encrypt_one_page(fio);
1893         if (err)
1894                 goto out_writepage;
1895
1896         set_page_writeback(page);
1897         ClearPageError(page);
1898
1899         /* LFS mode write path */
1900         f2fs_outplace_write_data(&dn, fio);
1901         trace_f2fs_do_write_data_page(page, OPU);
1902         set_inode_flag(inode, FI_APPEND_WRITE);
1903         if (page->index == 0)
1904                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1905 out_writepage:
1906         f2fs_put_dnode(&dn);
1907 out:
1908         if (fio->need_lock == LOCK_REQ)
1909                 f2fs_unlock_op(fio->sbi);
1910         return err;
1911 }
1912
1913 static int __write_data_page(struct page *page, bool *submitted,
1914                                 struct writeback_control *wbc,
1915                                 enum iostat_type io_type)
1916 {
1917         struct inode *inode = page->mapping->host;
1918         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1919         loff_t i_size = i_size_read(inode);
1920         const pgoff_t end_index = ((unsigned long long) i_size)
1921                                                         >> PAGE_SHIFT;
1922         loff_t psize = (page->index + 1) << PAGE_SHIFT;
1923         unsigned offset = 0;
1924         bool need_balance_fs = false;
1925         int err = 0;
1926         struct f2fs_io_info fio = {
1927                 .sbi = sbi,
1928                 .ino = inode->i_ino,
1929                 .type = DATA,
1930                 .op = REQ_OP_WRITE,
1931                 .op_flags = wbc_to_write_flags(wbc),
1932                 .old_blkaddr = NULL_ADDR,
1933                 .page = page,
1934                 .encrypted_page = NULL,
1935                 .submitted = false,
1936                 .need_lock = LOCK_RETRY,
1937                 .io_type = io_type,
1938                 .io_wbc = wbc,
1939         };
1940
1941         trace_f2fs_writepage(page, DATA);
1942
1943         /* we should bypass data pages to proceed the kworkder jobs */
1944         if (unlikely(f2fs_cp_error(sbi))) {
1945                 mapping_set_error(page->mapping, -EIO);
1946                 /*
1947                  * don't drop any dirty dentry pages for keeping lastest
1948                  * directory structure.
1949                  */
1950                 if (S_ISDIR(inode->i_mode))
1951                         goto redirty_out;
1952                 goto out;
1953         }
1954
1955         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1956                 goto redirty_out;
1957
1958         if (page->index < end_index)
1959                 goto write;
1960
1961         /*
1962          * If the offset is out-of-range of file size,
1963          * this page does not have to be written to disk.
1964          */
1965         offset = i_size & (PAGE_SIZE - 1);
1966         if ((page->index >= end_index + 1) || !offset)
1967                 goto out;
1968
1969         zero_user_segment(page, offset, PAGE_SIZE);
1970 write:
1971         if (f2fs_is_drop_cache(inode))
1972                 goto out;
1973         /* we should not write 0'th page having journal header */
1974         if (f2fs_is_volatile_file(inode) && (!page->index ||
1975                         (!wbc->for_reclaim &&
1976                         f2fs_available_free_memory(sbi, BASE_CHECK))))
1977                 goto redirty_out;
1978
1979         /* Dentry blocks are controlled by checkpoint */
1980         if (S_ISDIR(inode->i_mode)) {
1981                 fio.need_lock = LOCK_DONE;
1982                 err = f2fs_do_write_data_page(&fio);
1983                 goto done;
1984         }
1985
1986         if (!wbc->for_reclaim)
1987                 need_balance_fs = true;
1988         else if (has_not_enough_free_secs(sbi, 0, 0))
1989                 goto redirty_out;
1990         else
1991                 set_inode_flag(inode, FI_HOT_DATA);
1992
1993         err = -EAGAIN;
1994         if (f2fs_has_inline_data(inode)) {
1995                 err = f2fs_write_inline_data(inode, page);
1996                 if (!err)
1997                         goto out;
1998         }
1999
2000         if (err == -EAGAIN) {
2001                 err = f2fs_do_write_data_page(&fio);
2002                 if (err == -EAGAIN) {
2003                         fio.need_lock = LOCK_REQ;
2004                         err = f2fs_do_write_data_page(&fio);
2005                 }
2006         }
2007
2008         if (err) {
2009                 file_set_keep_isize(inode);
2010         } else {
2011                 down_write(&F2FS_I(inode)->i_sem);
2012                 if (F2FS_I(inode)->last_disk_size < psize)
2013                         F2FS_I(inode)->last_disk_size = psize;
2014                 up_write(&F2FS_I(inode)->i_sem);
2015         }
2016
2017 done:
2018         if (err && err != -ENOENT)
2019                 goto redirty_out;
2020
2021 out:
2022         inode_dec_dirty_pages(inode);
2023         if (err) {
2024                 ClearPageUptodate(page);
2025                 clear_cold_data(page);
2026         }
2027
2028         if (wbc->for_reclaim) {
2029                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2030                 clear_inode_flag(inode, FI_HOT_DATA);
2031                 f2fs_remove_dirty_inode(inode);
2032                 submitted = NULL;
2033         }
2034
2035         unlock_page(page);
2036         if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2037                 f2fs_balance_fs(sbi, need_balance_fs);
2038
2039         if (unlikely(f2fs_cp_error(sbi))) {
2040                 f2fs_submit_merged_write(sbi, DATA);
2041                 submitted = NULL;
2042         }
2043
2044         if (submitted)
2045                 *submitted = fio.submitted;
2046
2047         return 0;
2048
2049 redirty_out:
2050         redirty_page_for_writepage(wbc, page);
2051         /*
2052          * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2053          * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2054          * file_write_and_wait_range() will see EIO error, which is critical
2055          * to return value of fsync() followed by atomic_write failure to user.
2056          */
2057         if (!err || wbc->for_reclaim)
2058                 return AOP_WRITEPAGE_ACTIVATE;
2059         unlock_page(page);
2060         return err;
2061 }
2062
2063 static int f2fs_write_data_page(struct page *page,
2064                                         struct writeback_control *wbc)
2065 {
2066         return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2067 }
2068
2069 /*
2070  * This function was copied from write_cche_pages from mm/page-writeback.c.
2071  * The major change is making write step of cold data page separately from
2072  * warm/hot data page.
2073  */
2074 static int f2fs_write_cache_pages(struct address_space *mapping,
2075                                         struct writeback_control *wbc,
2076                                         enum iostat_type io_type)
2077 {
2078         int ret = 0;
2079         int done = 0;
2080         struct pagevec pvec;
2081         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2082         int nr_pages;
2083         pgoff_t uninitialized_var(writeback_index);
2084         pgoff_t index;
2085         pgoff_t end;            /* Inclusive */
2086         pgoff_t done_index;
2087         int cycled;
2088         int range_whole = 0;
2089         xa_mark_t tag;
2090         int nwritten = 0;
2091
2092         pagevec_init(&pvec);
2093
2094         if (get_dirty_pages(mapping->host) <=
2095                                 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2096                 set_inode_flag(mapping->host, FI_HOT_DATA);
2097         else
2098                 clear_inode_flag(mapping->host, FI_HOT_DATA);
2099
2100         if (wbc->range_cyclic) {
2101                 writeback_index = mapping->writeback_index; /* prev offset */
2102                 index = writeback_index;
2103                 if (index == 0)
2104                         cycled = 1;
2105                 else
2106                         cycled = 0;
2107                 end = -1;
2108         } else {
2109                 index = wbc->range_start >> PAGE_SHIFT;
2110                 end = wbc->range_end >> PAGE_SHIFT;
2111                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2112                         range_whole = 1;
2113                 cycled = 1; /* ignore range_cyclic tests */
2114         }
2115         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2116                 tag = PAGECACHE_TAG_TOWRITE;
2117         else
2118                 tag = PAGECACHE_TAG_DIRTY;
2119 retry:
2120         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2121                 tag_pages_for_writeback(mapping, index, end);
2122         done_index = index;
2123         while (!done && (index <= end)) {
2124                 int i;
2125
2126                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2127                                 tag);
2128                 if (nr_pages == 0)
2129                         break;
2130
2131                 for (i = 0; i < nr_pages; i++) {
2132                         struct page *page = pvec.pages[i];
2133                         bool submitted = false;
2134
2135                         /* give a priority to WB_SYNC threads */
2136                         if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2137                                         wbc->sync_mode == WB_SYNC_NONE) {
2138                                 done = 1;
2139                                 break;
2140                         }
2141
2142                         done_index = page->index;
2143 retry_write:
2144                         lock_page(page);
2145
2146                         if (unlikely(page->mapping != mapping)) {
2147 continue_unlock:
2148                                 unlock_page(page);
2149                                 continue;
2150                         }
2151
2152                         if (!PageDirty(page)) {
2153                                 /* someone wrote it for us */
2154                                 goto continue_unlock;
2155                         }
2156
2157                         if (PageWriteback(page)) {
2158                                 if (wbc->sync_mode != WB_SYNC_NONE)
2159                                         f2fs_wait_on_page_writeback(page,
2160                                                         DATA, true, true);
2161                                 else
2162                                         goto continue_unlock;
2163                         }
2164
2165                         if (!clear_page_dirty_for_io(page))
2166                                 goto continue_unlock;
2167
2168                         ret = __write_data_page(page, &submitted, wbc, io_type);
2169                         if (unlikely(ret)) {
2170                                 /*
2171                                  * keep nr_to_write, since vfs uses this to
2172                                  * get # of written pages.
2173                                  */
2174                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2175                                         unlock_page(page);
2176                                         ret = 0;
2177                                         continue;
2178                                 } else if (ret == -EAGAIN) {
2179                                         ret = 0;
2180                                         if (wbc->sync_mode == WB_SYNC_ALL) {
2181                                                 cond_resched();
2182                                                 congestion_wait(BLK_RW_ASYNC,
2183                                                                         HZ/50);
2184                                                 goto retry_write;
2185                                         }
2186                                         continue;
2187                                 }
2188                                 done_index = page->index + 1;
2189                                 done = 1;
2190                                 break;
2191                         } else if (submitted) {
2192                                 nwritten++;
2193                         }
2194
2195                         if (--wbc->nr_to_write <= 0 &&
2196                                         wbc->sync_mode == WB_SYNC_NONE) {
2197                                 done = 1;
2198                                 break;
2199                         }
2200                 }
2201                 pagevec_release(&pvec);
2202                 cond_resched();
2203         }
2204
2205         if (!cycled && !done) {
2206                 cycled = 1;
2207                 index = 0;
2208                 end = writeback_index - 1;
2209                 goto retry;
2210         }
2211         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2212                 mapping->writeback_index = done_index;
2213
2214         if (nwritten)
2215                 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2216                                                                 NULL, 0, DATA);
2217
2218         return ret;
2219 }
2220
2221 static inline bool __should_serialize_io(struct inode *inode,
2222                                         struct writeback_control *wbc)
2223 {
2224         if (!S_ISREG(inode->i_mode))
2225                 return false;
2226         if (IS_NOQUOTA(inode))
2227                 return false;
2228         if (wbc->sync_mode != WB_SYNC_ALL)
2229                 return true;
2230         if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2231                 return true;
2232         return false;
2233 }
2234
2235 static int __f2fs_write_data_pages(struct address_space *mapping,
2236                                                 struct writeback_control *wbc,
2237                                                 enum iostat_type io_type)
2238 {
2239         struct inode *inode = mapping->host;
2240         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2241         struct blk_plug plug;
2242         int ret;
2243         bool locked = false;
2244
2245         /* deal with chardevs and other special file */
2246         if (!mapping->a_ops->writepage)
2247                 return 0;
2248
2249         /* skip writing if there is no dirty page in this inode */
2250         if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2251                 return 0;
2252
2253         /* during POR, we don't need to trigger writepage at all. */
2254         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2255                 goto skip_write;
2256
2257         if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2258                         wbc->sync_mode == WB_SYNC_NONE &&
2259                         get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2260                         f2fs_available_free_memory(sbi, DIRTY_DENTS))
2261                 goto skip_write;
2262
2263         /* skip writing during file defragment */
2264         if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2265                 goto skip_write;
2266
2267         trace_f2fs_writepages(mapping->host, wbc, DATA);
2268
2269         /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2270         if (wbc->sync_mode == WB_SYNC_ALL)
2271                 atomic_inc(&sbi->wb_sync_req[DATA]);
2272         else if (atomic_read(&sbi->wb_sync_req[DATA]))
2273                 goto skip_write;
2274
2275         if (__should_serialize_io(inode, wbc)) {
2276                 mutex_lock(&sbi->writepages);
2277                 locked = true;
2278         }
2279
2280         blk_start_plug(&plug);
2281         ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2282         blk_finish_plug(&plug);
2283
2284         if (locked)
2285                 mutex_unlock(&sbi->writepages);
2286
2287         if (wbc->sync_mode == WB_SYNC_ALL)
2288                 atomic_dec(&sbi->wb_sync_req[DATA]);
2289         /*
2290          * if some pages were truncated, we cannot guarantee its mapping->host
2291          * to detect pending bios.
2292          */
2293
2294         f2fs_remove_dirty_inode(inode);
2295         return ret;
2296
2297 skip_write:
2298         wbc->pages_skipped += get_dirty_pages(inode);
2299         trace_f2fs_writepages(mapping->host, wbc, DATA);
2300         return 0;
2301 }
2302
2303 static int f2fs_write_data_pages(struct address_space *mapping,
2304                             struct writeback_control *wbc)
2305 {
2306         struct inode *inode = mapping->host;
2307
2308         return __f2fs_write_data_pages(mapping, wbc,
2309                         F2FS_I(inode)->cp_task == current ?
2310                         FS_CP_DATA_IO : FS_DATA_IO);
2311 }
2312
2313 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2314 {
2315         struct inode *inode = mapping->host;
2316         loff_t i_size = i_size_read(inode);
2317
2318         if (to > i_size) {
2319                 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2320                 down_write(&F2FS_I(inode)->i_mmap_sem);
2321
2322                 truncate_pagecache(inode, i_size);
2323                 if (!IS_NOQUOTA(inode))
2324                         f2fs_truncate_blocks(inode, i_size, true);
2325
2326                 up_write(&F2FS_I(inode)->i_mmap_sem);
2327                 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2328         }
2329 }
2330
2331 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2332                         struct page *page, loff_t pos, unsigned len,
2333                         block_t *blk_addr, bool *node_changed)
2334 {
2335         struct inode *inode = page->mapping->host;
2336         pgoff_t index = page->index;
2337         struct dnode_of_data dn;
2338         struct page *ipage;
2339         bool locked = false;
2340         struct extent_info ei = {0,0,0};
2341         int err = 0;
2342         int flag;
2343
2344         /*
2345          * we already allocated all the blocks, so we don't need to get
2346          * the block addresses when there is no need to fill the page.
2347          */
2348         if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2349                         !is_inode_flag_set(inode, FI_NO_PREALLOC))
2350                 return 0;
2351
2352         /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2353         if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2354                 flag = F2FS_GET_BLOCK_DEFAULT;
2355         else
2356                 flag = F2FS_GET_BLOCK_PRE_AIO;
2357
2358         if (f2fs_has_inline_data(inode) ||
2359                         (pos & PAGE_MASK) >= i_size_read(inode)) {
2360                 __do_map_lock(sbi, flag, true);
2361                 locked = true;
2362         }
2363 restart:
2364         /* check inline_data */
2365         ipage = f2fs_get_node_page(sbi, inode->i_ino);
2366         if (IS_ERR(ipage)) {
2367                 err = PTR_ERR(ipage);
2368                 goto unlock_out;
2369         }
2370
2371         set_new_dnode(&dn, inode, ipage, ipage, 0);
2372
2373         if (f2fs_has_inline_data(inode)) {
2374                 if (pos + len <= MAX_INLINE_DATA(inode)) {
2375                         f2fs_do_read_inline_data(page, ipage);
2376                         set_inode_flag(inode, FI_DATA_EXIST);
2377                         if (inode->i_nlink)
2378                                 set_inline_node(ipage);
2379                 } else {
2380                         err = f2fs_convert_inline_page(&dn, page);
2381                         if (err)
2382                                 goto out;
2383                         if (dn.data_blkaddr == NULL_ADDR)
2384                                 err = f2fs_get_block(&dn, index);
2385                 }
2386         } else if (locked) {
2387                 err = f2fs_get_block(&dn, index);
2388         } else {
2389                 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2390                         dn.data_blkaddr = ei.blk + index - ei.fofs;
2391                 } else {
2392                         /* hole case */
2393                         err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2394                         if (err || dn.data_blkaddr == NULL_ADDR) {
2395                                 f2fs_put_dnode(&dn);
2396                                 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2397                                                                 true);
2398                                 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2399                                 locked = true;
2400                                 goto restart;
2401                         }
2402                 }
2403         }
2404
2405         /* convert_inline_page can make node_changed */
2406         *blk_addr = dn.data_blkaddr;
2407         *node_changed = dn.node_changed;
2408 out:
2409         f2fs_put_dnode(&dn);
2410 unlock_out:
2411         if (locked)
2412                 __do_map_lock(sbi, flag, false);
2413         return err;
2414 }
2415
2416 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2417                 loff_t pos, unsigned len, unsigned flags,
2418                 struct page **pagep, void **fsdata)
2419 {
2420         struct inode *inode = mapping->host;
2421         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2422         struct page *page = NULL;
2423         pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2424         bool need_balance = false, drop_atomic = false;
2425         block_t blkaddr = NULL_ADDR;
2426         int err = 0;
2427
2428         trace_f2fs_write_begin(inode, pos, len, flags);
2429
2430         err = f2fs_is_checkpoint_ready(sbi);
2431         if (err)
2432                 goto fail;
2433
2434         if ((f2fs_is_atomic_file(inode) &&
2435                         !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2436                         is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2437                 err = -ENOMEM;
2438                 drop_atomic = true;
2439                 goto fail;
2440         }
2441
2442         /*
2443          * We should check this at this moment to avoid deadlock on inode page
2444          * and #0 page. The locking rule for inline_data conversion should be:
2445          * lock_page(page #0) -> lock_page(inode_page)
2446          */
2447         if (index != 0) {
2448                 err = f2fs_convert_inline_inode(inode);
2449                 if (err)
2450                         goto fail;
2451         }
2452 repeat:
2453         /*
2454          * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2455          * wait_for_stable_page. Will wait that below with our IO control.
2456          */
2457         page = f2fs_pagecache_get_page(mapping, index,
2458                                 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2459         if (!page) {
2460                 err = -ENOMEM;
2461                 goto fail;
2462         }
2463
2464         *pagep = page;
2465
2466         err = prepare_write_begin(sbi, page, pos, len,
2467                                         &blkaddr, &need_balance);
2468         if (err)
2469                 goto fail;
2470
2471         if (need_balance && !IS_NOQUOTA(inode) &&
2472                         has_not_enough_free_secs(sbi, 0, 0)) {
2473                 unlock_page(page);
2474                 f2fs_balance_fs(sbi, true);
2475                 lock_page(page);
2476                 if (page->mapping != mapping) {
2477                         /* The page got truncated from under us */
2478                         f2fs_put_page(page, 1);
2479                         goto repeat;
2480                 }
2481         }
2482
2483         f2fs_wait_on_page_writeback(page, DATA, false, true);
2484
2485         if (len == PAGE_SIZE || PageUptodate(page))
2486                 return 0;
2487
2488         if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2489                 zero_user_segment(page, len, PAGE_SIZE);
2490                 return 0;
2491         }
2492
2493         if (blkaddr == NEW_ADDR) {
2494                 zero_user_segment(page, 0, PAGE_SIZE);
2495                 SetPageUptodate(page);
2496         } else {
2497                 err = f2fs_submit_page_read(inode, page, blkaddr);
2498                 if (err)
2499                         goto fail;
2500
2501                 lock_page(page);
2502                 if (unlikely(page->mapping != mapping)) {
2503                         f2fs_put_page(page, 1);
2504                         goto repeat;
2505                 }
2506                 if (unlikely(!PageUptodate(page))) {
2507                         err = -EIO;
2508                         goto fail;
2509                 }
2510         }
2511         return 0;
2512
2513 fail:
2514         f2fs_put_page(page, 1);
2515         f2fs_write_failed(mapping, pos + len);
2516         if (drop_atomic)
2517                 f2fs_drop_inmem_pages_all(sbi, false);
2518         return err;
2519 }
2520
2521 static int f2fs_write_end(struct file *file,
2522                         struct address_space *mapping,
2523                         loff_t pos, unsigned len, unsigned copied,
2524                         struct page *page, void *fsdata)
2525 {
2526         struct inode *inode = page->mapping->host;
2527
2528         trace_f2fs_write_end(inode, pos, len, copied);
2529
2530         /*
2531          * This should be come from len == PAGE_SIZE, and we expect copied
2532          * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2533          * let generic_perform_write() try to copy data again through copied=0.
2534          */
2535         if (!PageUptodate(page)) {
2536                 if (unlikely(copied != len))
2537                         copied = 0;
2538                 else
2539                         SetPageUptodate(page);
2540         }
2541         if (!copied)
2542                 goto unlock_out;
2543
2544         set_page_dirty(page);
2545
2546         if (pos + copied > i_size_read(inode))
2547                 f2fs_i_size_write(inode, pos + copied);
2548 unlock_out:
2549         f2fs_put_page(page, 1);
2550         f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2551         return copied;
2552 }
2553
2554 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2555                            loff_t offset)
2556 {
2557         unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2558         unsigned blkbits = i_blkbits;
2559         unsigned blocksize_mask = (1 << blkbits) - 1;
2560         unsigned long align = offset | iov_iter_alignment(iter);
2561         struct block_device *bdev = inode->i_sb->s_bdev;
2562
2563         if (align & blocksize_mask) {
2564                 if (bdev)
2565                         blkbits = blksize_bits(bdev_logical_block_size(bdev));
2566                 blocksize_mask = (1 << blkbits) - 1;
2567                 if (align & blocksize_mask)
2568                         return -EINVAL;
2569                 return 1;
2570         }
2571         return 0;
2572 }
2573
2574 static void f2fs_dio_end_io(struct bio *bio)
2575 {
2576         struct f2fs_private_dio *dio = bio->bi_private;
2577
2578         dec_page_count(F2FS_I_SB(dio->inode),
2579                         dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2580
2581         bio->bi_private = dio->orig_private;
2582         bio->bi_end_io = dio->orig_end_io;
2583
2584         kvfree(dio);
2585
2586         bio_endio(bio);
2587 }
2588
2589 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2590                                                         loff_t file_offset)
2591 {
2592         struct f2fs_private_dio *dio;
2593         bool write = (bio_op(bio) == REQ_OP_WRITE);
2594
2595         dio = f2fs_kzalloc(F2FS_I_SB(inode),
2596                         sizeof(struct f2fs_private_dio), GFP_NOFS);
2597         if (!dio)
2598                 goto out;
2599
2600         dio->inode = inode;
2601         dio->orig_end_io = bio->bi_end_io;
2602         dio->orig_private = bio->bi_private;
2603         dio->write = write;
2604
2605         bio->bi_end_io = f2fs_dio_end_io;
2606         bio->bi_private = dio;
2607
2608         inc_page_count(F2FS_I_SB(inode),
2609                         write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2610
2611         submit_bio(bio);
2612         return;
2613 out:
2614         bio->bi_status = BLK_STS_IOERR;
2615         bio_endio(bio);
2616 }
2617
2618 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2619 {
2620         struct address_space *mapping = iocb->ki_filp->f_mapping;
2621         struct inode *inode = mapping->host;
2622         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2623         struct f2fs_inode_info *fi = F2FS_I(inode);
2624         size_t count = iov_iter_count(iter);
2625         loff_t offset = iocb->ki_pos;
2626         int rw = iov_iter_rw(iter);
2627         int err;
2628         enum rw_hint hint = iocb->ki_hint;
2629         int whint_mode = F2FS_OPTION(sbi).whint_mode;
2630         bool do_opu;
2631
2632         err = check_direct_IO(inode, iter, offset);
2633         if (err)
2634                 return err < 0 ? err : 0;
2635
2636         if (f2fs_force_buffered_io(inode, iocb, iter))
2637                 return 0;
2638
2639         do_opu = allow_outplace_dio(inode, iocb, iter);
2640
2641         trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2642
2643         if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2644                 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2645
2646         if (iocb->ki_flags & IOCB_NOWAIT) {
2647                 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2648                         iocb->ki_hint = hint;
2649                         err = -EAGAIN;
2650                         goto out;
2651                 }
2652                 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2653                         up_read(&fi->i_gc_rwsem[rw]);
2654                         iocb->ki_hint = hint;
2655                         err = -EAGAIN;
2656                         goto out;
2657                 }
2658         } else {
2659                 down_read(&fi->i_gc_rwsem[rw]);
2660                 if (do_opu)
2661                         down_read(&fi->i_gc_rwsem[READ]);
2662         }
2663
2664         err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2665                         iter, rw == WRITE ? get_data_block_dio_write :
2666                         get_data_block_dio, NULL, f2fs_dio_submit_bio,
2667                         DIO_LOCKING | DIO_SKIP_HOLES);
2668
2669         if (do_opu)
2670                 up_read(&fi->i_gc_rwsem[READ]);
2671
2672         up_read(&fi->i_gc_rwsem[rw]);
2673
2674         if (rw == WRITE) {
2675                 if (whint_mode == WHINT_MODE_OFF)
2676                         iocb->ki_hint = hint;
2677                 if (err > 0) {
2678                         f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2679                                                                         err);
2680                         if (!do_opu)
2681                                 set_inode_flag(inode, FI_UPDATE_WRITE);
2682                 } else if (err < 0) {
2683                         f2fs_write_failed(mapping, offset + count);
2684                 }
2685         }
2686
2687 out:
2688         trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2689
2690         return err;
2691 }
2692
2693 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2694                                                         unsigned int length)
2695 {
2696         struct inode *inode = page->mapping->host;
2697         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2698
2699         if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2700                 (offset % PAGE_SIZE || length != PAGE_SIZE))
2701                 return;
2702
2703         if (PageDirty(page)) {
2704                 if (inode->i_ino == F2FS_META_INO(sbi)) {
2705                         dec_page_count(sbi, F2FS_DIRTY_META);
2706                 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2707                         dec_page_count(sbi, F2FS_DIRTY_NODES);
2708                 } else {
2709                         inode_dec_dirty_pages(inode);
2710                         f2fs_remove_dirty_inode(inode);
2711                 }
2712         }
2713
2714         clear_cold_data(page);
2715
2716         if (IS_ATOMIC_WRITTEN_PAGE(page))
2717                 return f2fs_drop_inmem_page(inode, page);
2718
2719         f2fs_clear_page_private(page);
2720 }
2721
2722 int f2fs_release_page(struct page *page, gfp_t wait)
2723 {
2724         /* If this is dirty page, keep PagePrivate */
2725         if (PageDirty(page))
2726                 return 0;
2727
2728         /* This is atomic written page, keep Private */
2729         if (IS_ATOMIC_WRITTEN_PAGE(page))
2730                 return 0;
2731
2732         clear_cold_data(page);
2733         f2fs_clear_page_private(page);
2734         return 1;
2735 }
2736
2737 static int f2fs_set_data_page_dirty(struct page *page)
2738 {
2739         struct address_space *mapping = page->mapping;
2740         struct inode *inode = mapping->host;
2741
2742         trace_f2fs_set_page_dirty(page, DATA);
2743
2744         if (!PageUptodate(page))
2745                 SetPageUptodate(page);
2746
2747         if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2748                 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2749                         f2fs_register_inmem_page(inode, page);
2750                         return 1;
2751                 }
2752                 /*
2753                  * Previously, this page has been registered, we just
2754                  * return here.
2755                  */
2756                 return 0;
2757         }
2758
2759         if (!PageDirty(page)) {
2760                 __set_page_dirty_nobuffers(page);
2761                 f2fs_update_dirty_page(inode, page);
2762                 return 1;
2763         }
2764         return 0;
2765 }
2766
2767 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2768 {
2769         struct inode *inode = mapping->host;
2770
2771         if (f2fs_has_inline_data(inode))
2772                 return 0;
2773
2774         /* make sure allocating whole blocks */
2775         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2776                 filemap_write_and_wait(mapping);
2777
2778         return generic_block_bmap(mapping, block, get_data_block_bmap);
2779 }
2780
2781 #ifdef CONFIG_MIGRATION
2782 #include <linux/migrate.h>
2783
2784 int f2fs_migrate_page(struct address_space *mapping,
2785                 struct page *newpage, struct page *page, enum migrate_mode mode)
2786 {
2787         int rc, extra_count;
2788         struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2789         bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2790
2791         BUG_ON(PageWriteback(page));
2792
2793         /* migrating an atomic written page is safe with the inmem_lock hold */
2794         if (atomic_written) {
2795                 if (mode != MIGRATE_SYNC)
2796                         return -EBUSY;
2797                 if (!mutex_trylock(&fi->inmem_lock))
2798                         return -EAGAIN;
2799         }
2800
2801         /* one extra reference was held for atomic_write page */
2802         extra_count = atomic_written ? 1 : 0;
2803         rc = migrate_page_move_mapping(mapping, newpage,
2804                                 page, mode, extra_count);
2805         if (rc != MIGRATEPAGE_SUCCESS) {
2806                 if (atomic_written)
2807                         mutex_unlock(&fi->inmem_lock);
2808                 return rc;
2809         }
2810
2811         if (atomic_written) {
2812                 struct inmem_pages *cur;
2813                 list_for_each_entry(cur, &fi->inmem_pages, list)
2814                         if (cur->page == page) {
2815                                 cur->page = newpage;
2816                                 break;
2817                         }
2818                 mutex_unlock(&fi->inmem_lock);
2819                 put_page(page);
2820                 get_page(newpage);
2821         }
2822
2823         if (PagePrivate(page)) {
2824                 f2fs_set_page_private(newpage, page_private(page));
2825                 f2fs_clear_page_private(page);
2826         }
2827
2828         if (mode != MIGRATE_SYNC_NO_COPY)
2829                 migrate_page_copy(newpage, page);
2830         else
2831                 migrate_page_states(newpage, page);
2832
2833         return MIGRATEPAGE_SUCCESS;
2834 }
2835 #endif
2836
2837 const struct address_space_operations f2fs_dblock_aops = {
2838         .readpage       = f2fs_read_data_page,
2839         .readpages      = f2fs_read_data_pages,
2840         .writepage      = f2fs_write_data_page,
2841         .writepages     = f2fs_write_data_pages,
2842         .write_begin    = f2fs_write_begin,
2843         .write_end      = f2fs_write_end,
2844         .set_page_dirty = f2fs_set_data_page_dirty,
2845         .invalidatepage = f2fs_invalidate_page,
2846         .releasepage    = f2fs_release_page,
2847         .direct_IO      = f2fs_direct_IO,
2848         .bmap           = f2fs_bmap,
2849 #ifdef CONFIG_MIGRATION
2850         .migratepage    = f2fs_migrate_page,
2851 #endif
2852 };
2853
2854 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2855 {
2856         struct address_space *mapping = page_mapping(page);
2857         unsigned long flags;
2858
2859         xa_lock_irqsave(&mapping->i_pages, flags);
2860         __xa_clear_mark(&mapping->i_pages, page_index(page),
2861                                                 PAGECACHE_TAG_DIRTY);
2862         xa_unlock_irqrestore(&mapping->i_pages, flags);
2863 }
2864
2865 int __init f2fs_init_post_read_processing(void)
2866 {
2867         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2868         if (!bio_post_read_ctx_cache)
2869                 goto fail;
2870         bio_post_read_ctx_pool =
2871                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2872                                          bio_post_read_ctx_cache);
2873         if (!bio_post_read_ctx_pool)
2874                 goto fail_free_cache;
2875         return 0;
2876
2877 fail_free_cache:
2878         kmem_cache_destroy(bio_post_read_ctx_cache);
2879 fail:
2880         return -ENOMEM;
2881 }
2882
2883 void __exit f2fs_destroy_post_read_processing(void)
2884 {
2885         mempool_destroy(bio_post_read_ctx_pool);
2886         kmem_cache_destroy(bio_post_read_ctx_cache);
2887 }