fs/f2fs/checkpoint.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/f2fs/checkpoint.c
   4  *
   5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6  *             http://www.samsung.com/
   7  */
   8 #include <linux/fs.h>
   9 #include <linux/bio.h>
  10 #include <linux/mpage.h>
  11 #include <linux/writeback.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/f2fs_fs.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/swap.h>
  16
  17 #include "f2fs.h"
  18 #include "node.h"
  19 #include "segment.h"
  20 #include "trace.h"
  21 #include <trace/events/f2fs.h>
  22
  23 static struct kmem_cache *ino_entry_slab;
  24 struct kmem_cache *f2fs_inode_entry_slab;
  25
  26 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  27 {
  28         f2fs_build_fault_attr(sbi, 0, 0);
  29         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  30         if (!end_io)
  31                 f2fs_flush_merged_writes(sbi);
  32 }
  33
  34 /*
  35  * We guarantee no failure on the returned page.
  36  */
  37 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  38 {
  39         struct address_space *mapping = META_MAPPING(sbi);
  40         struct page *page = NULL;
  41 repeat:
  42         page = f2fs_grab_cache_page(mapping, index, false);
  43         if (!page) {
  44                 cond_resched();
  45                 goto repeat;
  46         }
  47         f2fs_wait_on_page_writeback(page, META, true, true);
  48         if (!PageUptodate(page))
  49                 SetPageUptodate(page);
  50         return page;
  51 }
  52
  53 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  54                                                         bool is_meta)
  55 {
  56         struct address_space *mapping = META_MAPPING(sbi);
  57         struct page *page;
  58         struct f2fs_io_info fio = {
  59                 .sbi = sbi,
  60                 .type = META,
  61                 .op = REQ_OP_READ,
  62                 .op_flags = REQ_META | REQ_PRIO,
  63                 .old_blkaddr = index,
  64                 .new_blkaddr = index,
  65                 .encrypted_page = NULL,
  66                 .is_por = !is_meta,
  67         };
  68         int err;
  69
  70         if (unlikely(!is_meta))
  71                 fio.op_flags &= ~REQ_META;
  72 repeat:
  73         page = f2fs_grab_cache_page(mapping, index, false);
  74         if (!page) {
  75                 cond_resched();
  76                 goto repeat;
  77         }
  78         if (PageUptodate(page))
  79                 goto out;
  80
  81         fio.page = page;
  82
  83         err = f2fs_submit_page_bio(&fio);
  84         if (err) {
  85                 f2fs_put_page(page, 1);
  86                 return ERR_PTR(err);
  87         }
  88
  89         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
  90
  91         lock_page(page);
  92         if (unlikely(page->mapping != mapping)) {
  93                 f2fs_put_page(page, 1);
  94                 goto repeat;
  95         }
  96
  97         if (unlikely(!PageUptodate(page))) {
  98                 f2fs_put_page(page, 1);
  99                 return ERR_PTR(-EIO);
 100         }
 101 out:
 102         return page;
 103 }
 104
 105 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 106 {
 107         return __get_meta_page(sbi, index, true);
 108 }
 109
 110 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 111 {
 112         struct page *page;
 113         int count = 0;
 114
 115 retry:
 116         page = __get_meta_page(sbi, index, true);
 117         if (IS_ERR(page)) {
 118                 if (PTR_ERR(page) == -EIO &&
 119                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 120                         goto retry;
 121                 f2fs_stop_checkpoint(sbi, false);
 122         }
 123         return page;
 124 }
 125
 126 /* for POR only */
 127 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 128 {
 129         return __get_meta_page(sbi, index, false);
 130 }
 131
 132 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 133                                                         int type)
 134 {
 135         struct seg_entry *se;
 136         unsigned int segno, offset;
 137         bool exist;
 138
 139         if (type == DATA_GENERIC)
 140                 return true;
 141
 142         segno = GET_SEGNO(sbi, blkaddr);
 143         offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 144         se = get_seg_entry(sbi, segno);
 145
 146         exist = f2fs_test_bit(offset, se->cur_valid_map);
 147         if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
 148                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 149                          blkaddr, exist);
 150                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 151                 return exist;
 152         }
 153
 154         if (!exist && type == DATA_GENERIC_ENHANCE) {
 155                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 156                          blkaddr, exist);
 157                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 158                 dump_stack();
 159         }
 160         return exist;
 161 }
 162
 163 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 164                                         block_t blkaddr, int type)
 165 {
 166         switch (type) {
 167         case META_NAT:
 168                 break;
 169         case META_SIT:
 170                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 171                         return false;
 172                 break;
 173         case META_SSA:
 174                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 175                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 176                         return false;
 177                 break;
 178         case META_CP:
 179                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 180                         blkaddr < __start_cp_addr(sbi)))
 181                         return false;
 182                 break;
 183         case META_POR:
 184                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 185                         blkaddr < MAIN_BLKADDR(sbi)))
 186                         return false;
 187                 break;
 188         case DATA_GENERIC:
 189         case DATA_GENERIC_ENHANCE:
 190         case DATA_GENERIC_ENHANCE_READ:
 191         case DATA_GENERIC_ENHANCE_UPDATE:
 192                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 193                                 blkaddr < MAIN_BLKADDR(sbi))) {
 194                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 195                                   blkaddr);
 196                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 197                         dump_stack();
 198                         return false;
 199                 } else {
 200                         return __is_bitmap_valid(sbi, blkaddr, type);
 201                 }
 202                 break;
 203         case META_GENERIC:
 204                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 205                         blkaddr >= MAIN_BLKADDR(sbi)))
 206                         return false;
 207                 break;
 208         default:
 209                 BUG();
 210         }
 211
 212         return true;
 213 }
 214
 215 /*
 216  * Readahead CP/NAT/SIT/SSA/POR pages
 217  */
 218 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 219                                                         int type, bool sync)
 220 {
 221         struct page *page;
 222         block_t blkno = start;
 223         struct f2fs_io_info fio = {
 224                 .sbi = sbi,
 225                 .type = META,
 226                 .op = REQ_OP_READ,
 227                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 228                 .encrypted_page = NULL,
 229                 .in_list = false,
 230                 .is_por = (type == META_POR),
 231         };
 232         struct blk_plug plug;
 233         int err;
 234
 235         if (unlikely(type == META_POR))
 236                 fio.op_flags &= ~REQ_META;
 237
 238         blk_start_plug(&plug);
 239         for (; nrpages-- > 0; blkno++) {
 240
 241                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 242                         goto out;
 243
 244                 switch (type) {
 245                 case META_NAT:
 246                         if (unlikely(blkno >=
 247                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 248                                 blkno = 0;
 249                         /* get nat block addr */
 250                         fio.new_blkaddr = current_nat_addr(sbi,
 251                                         blkno * NAT_ENTRY_PER_BLOCK);
 252                         break;
 253                 case META_SIT:
 254                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 255                                 goto out;
 256                         /* get sit block addr */
 257                         fio.new_blkaddr = current_sit_addr(sbi,
 258                                         blkno * SIT_ENTRY_PER_BLOCK);
 259                         break;
 260                 case META_SSA:
 261                 case META_CP:
 262                 case META_POR:
 263                         fio.new_blkaddr = blkno;
 264                         break;
 265                 default:
 266                         BUG();
 267                 }
 268
 269                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 270                                                 fio.new_blkaddr, false);
 271                 if (!page)
 272                         continue;
 273                 if (PageUptodate(page)) {
 274                         f2fs_put_page(page, 1);
 275                         continue;
 276                 }
 277
 278                 fio.page = page;
 279                 err = f2fs_submit_page_bio(&fio);
 280                 f2fs_put_page(page, err ? 1 : 0);
 281
 282                 if (!err)
 283                         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
 284         }
 285 out:
 286         blk_finish_plug(&plug);
 287         return blkno - start;
 288 }
 289
 290 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 291 {
 292         struct page *page;
 293         bool readahead = false;
 294
 295         page = find_get_page(META_MAPPING(sbi), index);
 296         if (!page || !PageUptodate(page))
 297                 readahead = true;
 298         f2fs_put_page(page, 0);
 299
 300         if (readahead)
 301                 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 302 }
 303
 304 static int __f2fs_write_meta_page(struct page *page,
 305                                 struct writeback_control *wbc,
 306                                 enum iostat_type io_type)
 307 {
 308         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 309
 310         trace_f2fs_writepage(page, META);
 311
 312         if (unlikely(f2fs_cp_error(sbi))) {
 313                 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
 314                         ClearPageUptodate(page);
 315                         dec_page_count(sbi, F2FS_DIRTY_META);
 316                         unlock_page(page);
 317                         return 0;
 318                 }
 319                 goto redirty_out;
 320         }
 321         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 322                 goto redirty_out;
 323         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 324                 goto redirty_out;
 325
 326         f2fs_do_write_meta_page(sbi, page, io_type);
 327         dec_page_count(sbi, F2FS_DIRTY_META);
 328
 329         if (wbc->for_reclaim)
 330                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 331
 332         unlock_page(page);
 333
 334         if (unlikely(f2fs_cp_error(sbi)))
 335                 f2fs_submit_merged_write(sbi, META);
 336
 337         return 0;
 338
 339 redirty_out:
 340         redirty_page_for_writepage(wbc, page);
 341         return AOP_WRITEPAGE_ACTIVATE;
 342 }
 343
 344 static int f2fs_write_meta_page(struct page *page,
 345                                 struct writeback_control *wbc)
 346 {
 347         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 348 }
 349
 350 static int f2fs_write_meta_pages(struct address_space *mapping,
 351                                 struct writeback_control *wbc)
 352 {
 353         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 354         long diff, written;
 355
 356         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 357                 goto skip_write;
 358
 359         /* collect a number of dirty meta pages and write together */
 360         if (wbc->sync_mode != WB_SYNC_ALL &&
 361                         get_pages(sbi, F2FS_DIRTY_META) <
 362                                         nr_pages_to_skip(sbi, META))
 363                 goto skip_write;
 364
 365         /* if locked failed, cp will flush dirty pages instead */
 366         if (!mutex_trylock(&sbi->cp_mutex))
 367                 goto skip_write;
 368
 369         trace_f2fs_writepages(mapping->host, wbc, META);
 370         diff = nr_pages_to_write(sbi, META, wbc);
 371         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 372         mutex_unlock(&sbi->cp_mutex);
 373         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 374         return 0;
 375
 376 skip_write:
 377         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 378         trace_f2fs_writepages(mapping->host, wbc, META);
 379         return 0;
 380 }
 381
 382 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 383                                 long nr_to_write, enum iostat_type io_type)
 384 {
 385         struct address_space *mapping = META_MAPPING(sbi);
 386         pgoff_t index = 0, prev = ULONG_MAX;
 387         struct pagevec pvec;
 388         long nwritten = 0;
 389         int nr_pages;
 390         struct writeback_control wbc = {
 391                 .for_reclaim = 0,
 392         };
 393         struct blk_plug plug;
 394
 395         pagevec_init(&pvec);
 396
 397         blk_start_plug(&plug);
 398
 399         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 400                                 PAGECACHE_TAG_DIRTY))) {
 401                 int i;
 402
 403                 for (i = 0; i < nr_pages; i++) {
 404                         struct page *page = pvec.pages[i];
 405
 406                         if (prev == ULONG_MAX)
 407                                 prev = page->index - 1;
 408                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 409                                 pagevec_release(&pvec);
 410                                 goto stop;
 411                         }
 412
 413                         lock_page(page);
 414
 415                         if (unlikely(page->mapping != mapping)) {
 416 continue_unlock:
 417                                 unlock_page(page);
 418                                 continue;
 419                         }
 420                         if (!PageDirty(page)) {
 421                                 /* someone wrote it for us */
 422                                 goto continue_unlock;
 423                         }
 424
 425                         f2fs_wait_on_page_writeback(page, META, true, true);
 426
 427                         if (!clear_page_dirty_for_io(page))
 428                                 goto continue_unlock;
 429
 430                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 431                                 unlock_page(page);
 432                                 break;
 433                         }
 434                         nwritten++;
 435                         prev = page->index;
 436                         if (unlikely(nwritten >= nr_to_write))
 437                                 break;
 438                 }
 439                 pagevec_release(&pvec);
 440                 cond_resched();
 441         }
 442 stop:
 443         if (nwritten)
 444                 f2fs_submit_merged_write(sbi, type);
 445
 446         blk_finish_plug(&plug);
 447
 448         return nwritten;
 449 }
 450
 451 static int f2fs_set_meta_page_dirty(struct page *page)
 452 {
 453         trace_f2fs_set_page_dirty(page, META);
 454
 455         if (!PageUptodate(page))
 456                 SetPageUptodate(page);
 457         if (!PageDirty(page)) {
 458                 __set_page_dirty_nobuffers(page);
 459                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 460                 f2fs_set_page_private(page, 0);
 461                 f2fs_trace_pid(page);
 462                 return 1;
 463         }
 464         return 0;
 465 }
 466
 467 const struct address_space_operations f2fs_meta_aops = {
 468         .writepage      = f2fs_write_meta_page,
 469         .writepages     = f2fs_write_meta_pages,
 470         .set_page_dirty = f2fs_set_meta_page_dirty,
 471         .invalidatepage = f2fs_invalidate_page,
 472         .releasepage    = f2fs_release_page,
 473 #ifdef CONFIG_MIGRATION
 474         .migratepage    = f2fs_migrate_page,
 475 #endif
 476 };
 477
 478 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 479                                                 unsigned int devidx, int type)
 480 {
 481         struct inode_management *im = &sbi->im[type];
 482         struct ino_entry *e, *tmp;
 483
 484         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 485
 486         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 487
 488         spin_lock(&im->ino_lock);
 489         e = radix_tree_lookup(&im->ino_root, ino);
 490         if (!e) {
 491                 e = tmp;
 492                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 493                         f2fs_bug_on(sbi, 1);
 494
 495                 memset(e, 0, sizeof(struct ino_entry));
 496                 e->ino = ino;
 497
 498                 list_add_tail(&e->list, &im->ino_list);
 499                 if (type != ORPHAN_INO)
 500                         im->ino_num++;
 501         }
 502
 503         if (type == FLUSH_INO)
 504                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 505
 506         spin_unlock(&im->ino_lock);
 507         radix_tree_preload_end();
 508
 509         if (e != tmp)
 510                 kmem_cache_free(ino_entry_slab, tmp);
 511 }
 512
 513 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 514 {
 515         struct inode_management *im = &sbi->im[type];
 516         struct ino_entry *e;
 517
 518         spin_lock(&im->ino_lock);
 519         e = radix_tree_lookup(&im->ino_root, ino);
 520         if (e) {
 521                 list_del(&e->list);
 522                 radix_tree_delete(&im->ino_root, ino);
 523                 im->ino_num--;
 524                 spin_unlock(&im->ino_lock);
 525                 kmem_cache_free(ino_entry_slab, e);
 526                 return;
 527         }
 528         spin_unlock(&im->ino_lock);
 529 }
 530
 531 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 532 {
 533         /* add new dirty ino entry into list */
 534         __add_ino_entry(sbi, ino, 0, type);
 535 }
 536
 537 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 538 {
 539         /* remove dirty ino entry from list */
 540         __remove_ino_entry(sbi, ino, type);
 541 }
 542
 543 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 544 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 545 {
 546         struct inode_management *im = &sbi->im[mode];
 547         struct ino_entry *e;
 548
 549         spin_lock(&im->ino_lock);
 550         e = radix_tree_lookup(&im->ino_root, ino);
 551         spin_unlock(&im->ino_lock);
 552         return e ? true : false;
 553 }
 554
 555 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 556 {
 557         struct ino_entry *e, *tmp;
 558         int i;
 559
 560         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 561                 struct inode_management *im = &sbi->im[i];
 562
 563                 spin_lock(&im->ino_lock);
 564                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 565                         list_del(&e->list);
 566                         radix_tree_delete(&im->ino_root, e->ino);
 567                         kmem_cache_free(ino_entry_slab, e);
 568                         im->ino_num--;
 569                 }
 570                 spin_unlock(&im->ino_lock);
 571         }
 572 }
 573
 574 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 575                                         unsigned int devidx, int type)
 576 {
 577         __add_ino_entry(sbi, ino, devidx, type);
 578 }
 579
 580 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 581                                         unsigned int devidx, int type)
 582 {
 583         struct inode_management *im = &sbi->im[type];
 584         struct ino_entry *e;
 585         bool is_dirty = false;
 586
 587         spin_lock(&im->ino_lock);
 588         e = radix_tree_lookup(&im->ino_root, ino);
 589         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 590                 is_dirty = true;
 591         spin_unlock(&im->ino_lock);
 592         return is_dirty;
 593 }
 594
 595 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 596 {
 597         struct inode_management *im = &sbi->im[ORPHAN_INO];
 598         int err = 0;
 599
 600         spin_lock(&im->ino_lock);
 601
 602         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 603                 spin_unlock(&im->ino_lock);
 604                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 605                 return -ENOSPC;
 606         }
 607
 608         if (unlikely(im->ino_num >= sbi->max_orphans))
 609                 err = -ENOSPC;
 610         else
 611                 im->ino_num++;
 612         spin_unlock(&im->ino_lock);
 613
 614         return err;
 615 }
 616
 617 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 618 {
 619         struct inode_management *im = &sbi->im[ORPHAN_INO];
 620
 621         spin_lock(&im->ino_lock);
 622         f2fs_bug_on(sbi, im->ino_num == 0);
 623         im->ino_num--;
 624         spin_unlock(&im->ino_lock);
 625 }
 626
 627 void f2fs_add_orphan_inode(struct inode *inode)
 628 {
 629         /* add new orphan ino entry into list */
 630         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 631         f2fs_update_inode_page(inode);
 632 }
 633
 634 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 635 {
 636         /* remove orphan entry from orphan list */
 637         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 638 }
 639
 640 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 641 {
 642         struct inode *inode;
 643         struct node_info ni;
 644         int err;
 645
 646         inode = f2fs_iget_retry(sbi->sb, ino);
 647         if (IS_ERR(inode)) {
 648                 /*
 649                  * there should be a bug that we can't find the entry
 650                  * to orphan inode.
 651                  */
 652                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 653                 return PTR_ERR(inode);
 654         }
 655
 656         err = dquot_initialize(inode);
 657         if (err) {
 658                 iput(inode);
 659                 goto err_out;
 660         }
 661
 662         clear_nlink(inode);
 663
 664         /* truncate all the data during iput */
 665         iput(inode);
 666
 667         err = f2fs_get_node_info(sbi, ino, &ni);
 668         if (err)
 669                 goto err_out;
 670
 671         /* ENOMEM was fully retried in f2fs_evict_inode. */
 672         if (ni.blk_addr != NULL_ADDR) {
 673                 err = -EIO;
 674                 goto err_out;
 675         }
 676         return 0;
 677
 678 err_out:
 679         set_sbi_flag(sbi, SBI_NEED_FSCK);
 680         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 681                   __func__, ino);
 682         return err;
 683 }
 684
 685 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 686 {
 687         block_t start_blk, orphan_blocks, i, j;
 688         unsigned int s_flags = sbi->sb->s_flags;
 689         int err = 0;
 690 #ifdef CONFIG_QUOTA
 691         int quota_enabled;
 692 #endif
 693
 694         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 695                 return 0;
 696
 697         if (bdev_read_only(sbi->sb->s_bdev)) {
 698                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 699                 return 0;
 700         }
 701
 702         if (s_flags & SB_RDONLY) {
 703                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 704                 sbi->sb->s_flags &= ~SB_RDONLY;
 705         }
 706
 707 #ifdef CONFIG_QUOTA
 708         /* Needed for iput() to work correctly and not trash data */
 709         sbi->sb->s_flags |= SB_ACTIVE;
 710
 711         /*
 712          * Turn on quotas which were not enabled for read-only mounts if
 713          * filesystem has quota feature, so that they are updated correctly.
 714          */
 715         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 716 #endif
 717
 718         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 719         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 720
 721         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 722
 723         for (i = 0; i < orphan_blocks; i++) {
 724                 struct page *page;
 725                 struct f2fs_orphan_block *orphan_blk;
 726
 727                 page = f2fs_get_meta_page(sbi, start_blk + i);
 728                 if (IS_ERR(page)) {
 729                         err = PTR_ERR(page);
 730                         goto out;
 731                 }
 732
 733                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 734                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 735                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 736                         err = recover_orphan_inode(sbi, ino);
 737                         if (err) {
 738                                 f2fs_put_page(page, 1);
 739                                 goto out;
 740                         }
 741                 }
 742                 f2fs_put_page(page, 1);
 743         }
 744         /* clear Orphan Flag */
 745         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 746 out:
 747         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 748
 749 #ifdef CONFIG_QUOTA
 750         /* Turn quotas off */
 751         if (quota_enabled)
 752                 f2fs_quota_off_umount(sbi->sb);
 753 #endif
 754         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 755
 756         return err;
 757 }
 758
 759 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 760 {
 761         struct list_head *head;
 762         struct f2fs_orphan_block *orphan_blk = NULL;
 763         unsigned int nentries = 0;
 764         unsigned short index = 1;
 765         unsigned short orphan_blocks;
 766         struct page *page = NULL;
 767         struct ino_entry *orphan = NULL;
 768         struct inode_management *im = &sbi->im[ORPHAN_INO];
 769
 770         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 771
 772         /*
 773          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 774          * orphan inode operations are covered under f2fs_lock_op().
 775          * And, spin_lock should be avoided due to page operations below.
 776          */
 777         head = &im->ino_list;
 778
 779         /* loop for each orphan inode entry and write them in journal block */
 780         list_for_each_entry(orphan, head, list) {
 781                 if (!page) {
 782                         page = f2fs_grab_meta_page(sbi, start_blk++);
 783                         orphan_blk =
 784                                 (struct f2fs_orphan_block *)page_address(page);
 785                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 786                 }
 787
 788                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 789
 790                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 791                         /*
 792                          * an orphan block is full of 1020 entries,
 793                          * then we need to flush current orphan blocks
 794                          * and bring another one in memory
 795                          */
 796                         orphan_blk->blk_addr = cpu_to_le16(index);
 797                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 798                         orphan_blk->entry_count = cpu_to_le32(nentries);
 799                         set_page_dirty(page);
 800                         f2fs_put_page(page, 1);
 801                         index++;
 802                         nentries = 0;
 803                         page = NULL;
 804                 }
 805         }
 806
 807         if (page) {
 808                 orphan_blk->blk_addr = cpu_to_le16(index);
 809                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 810                 orphan_blk->entry_count = cpu_to_le32(nentries);
 811                 set_page_dirty(page);
 812                 f2fs_put_page(page, 1);
 813         }
 814 }
 815
 816 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 817                                                 struct f2fs_checkpoint *ckpt)
 818 {
 819         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 820         __u32 chksum;
 821
 822         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 823         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 824                 chksum_ofs += sizeof(chksum);
 825                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 826                                                 F2FS_BLKSIZE - chksum_ofs);
 827         }
 828         return chksum;
 829 }
 830
 831 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 832                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 833                 unsigned long long *version)
 834 {
 835         size_t crc_offset = 0;
 836         __u32 crc;
 837
 838         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 839         if (IS_ERR(*cp_page))
 840                 return PTR_ERR(*cp_page);
 841
 842         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 843
 844         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 845         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 846                         crc_offset > CP_CHKSUM_OFFSET) {
 847                 f2fs_put_page(*cp_page, 1);
 848                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 849                 return -EINVAL;
 850         }
 851
 852         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 853         if (crc != cur_cp_crc(*cp_block)) {
 854                 f2fs_put_page(*cp_page, 1);
 855                 f2fs_warn(sbi, "invalid crc value");
 856                 return -EINVAL;
 857         }
 858
 859         *version = cur_cp_version(*cp_block);
 860         return 0;
 861 }
 862
 863 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 864                                 block_t cp_addr, unsigned long long *version)
 865 {
 866         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 867         struct f2fs_checkpoint *cp_block = NULL;
 868         unsigned long long cur_version = 0, pre_version = 0;
 869         unsigned int cp_blocks;
 870         int err;
 871
 872         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 873                                         &cp_page_1, version);
 874         if (err)
 875                 return NULL;
 876
 877         cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 878
 879         if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
 880                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 881                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 882                 goto invalid_cp;
 883         }
 884         pre_version = *version;
 885
 886         cp_addr += cp_blocks - 1;
 887         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 888                                         &cp_page_2, version);
 889         if (err)
 890                 goto invalid_cp;
 891         cur_version = *version;
 892
 893         if (cur_version == pre_version) {
 894                 *version = cur_version;
 895                 f2fs_put_page(cp_page_2, 1);
 896                 return cp_page_1;
 897         }
 898         f2fs_put_page(cp_page_2, 1);
 899 invalid_cp:
 900         f2fs_put_page(cp_page_1, 1);
 901         return NULL;
 902 }
 903
 904 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 905 {
 906         struct f2fs_checkpoint *cp_block;
 907         struct f2fs_super_block *fsb = sbi->raw_super;
 908         struct page *cp1, *cp2, *cur_page;
 909         unsigned long blk_size = sbi->blocksize;
 910         unsigned long long cp1_version = 0, cp2_version = 0;
 911         unsigned long long cp_start_blk_no;
 912         unsigned int cp_blks = 1 + __cp_payload(sbi);
 913         block_t cp_blk_no;
 914         int i;
 915         int err;
 916
 917         sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 918                                   GFP_KERNEL);
 919         if (!sbi->ckpt)
 920                 return -ENOMEM;
 921         /*
 922          * Finding out valid cp block involves read both
 923          * sets( cp pack 1 and cp pack 2)
 924          */
 925         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 926         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 927
 928         /* The second checkpoint pack should start at the next segment */
 929         cp_start_blk_no += ((unsigned long long)1) <<
 930                                 le32_to_cpu(fsb->log_blocks_per_seg);
 931         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 932
 933         if (cp1 && cp2) {
 934                 if (ver_after(cp2_version, cp1_version))
 935                         cur_page = cp2;
 936                 else
 937                         cur_page = cp1;
 938         } else if (cp1) {
 939                 cur_page = cp1;
 940         } else if (cp2) {
 941                 cur_page = cp2;
 942         } else {
 943                 err = -EFSCORRUPTED;
 944                 goto fail_no_cp;
 945         }
 946
 947         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 948         memcpy(sbi->ckpt, cp_block, blk_size);
 949
 950         if (cur_page == cp1)
 951                 sbi->cur_cp_pack = 1;
 952         else
 953                 sbi->cur_cp_pack = 2;
 954
 955         /* Sanity checking of checkpoint */
 956         if (f2fs_sanity_check_ckpt(sbi)) {
 957                 err = -EFSCORRUPTED;
 958                 goto free_fail_no_cp;
 959         }
 960
 961         if (cp_blks <= 1)
 962                 goto done;
 963
 964         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 965         if (cur_page == cp2)
 966                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 967
 968         for (i = 1; i < cp_blks; i++) {
 969                 void *sit_bitmap_ptr;
 970                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 971
 972                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 973                 if (IS_ERR(cur_page)) {
 974                         err = PTR_ERR(cur_page);
 975                         goto free_fail_no_cp;
 976                 }
 977                 sit_bitmap_ptr = page_address(cur_page);
 978                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 979                 f2fs_put_page(cur_page, 1);
 980         }
 981 done:
 982         f2fs_put_page(cp1, 1);
 983         f2fs_put_page(cp2, 1);
 984         return 0;
 985
 986 free_fail_no_cp:
 987         f2fs_put_page(cp1, 1);
 988         f2fs_put_page(cp2, 1);
 989 fail_no_cp:
 990         kvfree(sbi->ckpt);
 991         return err;
 992 }
 993
 994 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 995 {
 996         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 997         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 998
 999         if (is_inode_flag_set(inode, flag))
1000                 return;
1001
1002         set_inode_flag(inode, flag);
1003         if (!f2fs_is_volatile_file(inode))
1004                 list_add_tail(&F2FS_I(inode)->dirty_list,
1005                                                 &sbi->inode_list[type]);
1006         stat_inc_dirty_inode(sbi, type);
1007 }
1008
1009 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1010 {
1011         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1012
1013         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1014                 return;
1015
1016         list_del_init(&F2FS_I(inode)->dirty_list);
1017         clear_inode_flag(inode, flag);
1018         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1019 }
1020
1021 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1022 {
1023         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1024         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1025
1026         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1027                         !S_ISLNK(inode->i_mode))
1028                 return;
1029
1030         spin_lock(&sbi->inode_lock[type]);
1031         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1032                 __add_dirty_inode(inode, type);
1033         inode_inc_dirty_pages(inode);
1034         spin_unlock(&sbi->inode_lock[type]);
1035
1036         f2fs_set_page_private(page, 0);
1037         f2fs_trace_pid(page);
1038 }
1039
1040 void f2fs_remove_dirty_inode(struct inode *inode)
1041 {
1042         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1043         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1044
1045         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1046                         !S_ISLNK(inode->i_mode))
1047                 return;
1048
1049         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1050                 return;
1051
1052         spin_lock(&sbi->inode_lock[type]);
1053         __remove_dirty_inode(inode, type);
1054         spin_unlock(&sbi->inode_lock[type]);
1055 }
1056
1057 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1058                                                 bool from_cp)
1059 {
1060         struct list_head *head;
1061         struct inode *inode;
1062         struct f2fs_inode_info *fi;
1063         bool is_dir = (type == DIR_INODE);
1064         unsigned long ino = 0;
1065
1066         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1067                                 get_pages(sbi, is_dir ?
1068                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1069 retry:
1070         if (unlikely(f2fs_cp_error(sbi))) {
1071                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1072                                 get_pages(sbi, is_dir ?
1073                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1074                 return -EIO;
1075         }
1076
1077         spin_lock(&sbi->inode_lock[type]);
1078
1079         head = &sbi->inode_list[type];
1080         if (list_empty(head)) {
1081                 spin_unlock(&sbi->inode_lock[type]);
1082                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1083                                 get_pages(sbi, is_dir ?
1084                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085                 return 0;
1086         }
1087         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1088         inode = igrab(&fi->vfs_inode);
1089         spin_unlock(&sbi->inode_lock[type]);
1090         if (inode) {
1091                 unsigned long cur_ino = inode->i_ino;
1092
1093                 if (from_cp)
1094                         F2FS_I(inode)->cp_task = current;
1095                 F2FS_I(inode)->wb_task = current;
1096
1097                 filemap_fdatawrite(inode->i_mapping);
1098
1099                 F2FS_I(inode)->wb_task = NULL;
1100                 if (from_cp)
1101                         F2FS_I(inode)->cp_task = NULL;
1102
1103                 iput(inode);
1104                 /* We need to give cpu to another writers. */
1105                 if (ino == cur_ino)
1106                         cond_resched();
1107                 else
1108                         ino = cur_ino;
1109         } else {
1110                 /*
1111                  * We should submit bio, since it exists several
1112                  * writebacking dentry pages in the freeing inode.
1113                  */
1114                 f2fs_submit_merged_write(sbi, DATA);
1115                 cond_resched();
1116         }
1117         goto retry;
1118 }
1119
1120 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1121 {
1122         struct list_head *head = &sbi->inode_list[DIRTY_META];
1123         struct inode *inode;
1124         struct f2fs_inode_info *fi;
1125         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1126
1127         while (total--) {
1128                 if (unlikely(f2fs_cp_error(sbi)))
1129                         return -EIO;
1130
1131                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1132                 if (list_empty(head)) {
1133                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1134                         return 0;
1135                 }
1136                 fi = list_first_entry(head, struct f2fs_inode_info,
1137                                                         gdirty_list);
1138                 inode = igrab(&fi->vfs_inode);
1139                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1140                 if (inode) {
1141                         sync_inode_metadata(inode, 0);
1142
1143                         /* it's on eviction */
1144                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1145                                 f2fs_update_inode_page(inode);
1146                         iput(inode);
1147                 }
1148         }
1149         return 0;
1150 }
1151
1152 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1153 {
1154         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1155         struct f2fs_nm_info *nm_i = NM_I(sbi);
1156         nid_t last_nid = nm_i->next_scan_nid;
1157
1158         next_free_nid(sbi, &last_nid);
1159         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1160         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1161         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1162         ckpt->next_free_nid = cpu_to_le32(last_nid);
1163 }
1164
1165 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1166 {
1167         bool ret = false;
1168
1169         if (!is_journalled_quota(sbi))
1170                 return false;
1171
1172         if (!down_write_trylock(&sbi->quota_sem))
1173                 return true;
1174         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1175                 ret = false;
1176         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1177                 ret = false;
1178         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1179                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1180                 ret = true;
1181         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1182                 ret = true;
1183         }
1184         up_write(&sbi->quota_sem);
1185         return ret;
1186 }
1187
1188 /*
1189  * Freeze all the FS-operations for checkpoint.
1190  */
1191 static int block_operations(struct f2fs_sb_info *sbi)
1192 {
1193         struct writeback_control wbc = {
1194                 .sync_mode = WB_SYNC_ALL,
1195                 .nr_to_write = LONG_MAX,
1196                 .for_reclaim = 0,
1197         };
1198         int err = 0, cnt = 0;
1199
1200         /*
1201          * Let's flush inline_data in dirty node pages.
1202          */
1203         f2fs_flush_inline_data(sbi);
1204
1205 retry_flush_quotas:
1206         f2fs_lock_all(sbi);
1207         if (__need_flush_quota(sbi)) {
1208                 int locked;
1209
1210                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1211                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1212                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1213                         goto retry_flush_dents;
1214                 }
1215                 f2fs_unlock_all(sbi);
1216
1217                 /* only failed during mount/umount/freeze/quotactl */
1218                 locked = down_read_trylock(&sbi->sb->s_umount);
1219                 f2fs_quota_sync(sbi->sb, -1);
1220                 if (locked)
1221                         up_read(&sbi->sb->s_umount);
1222                 cond_resched();
1223                 goto retry_flush_quotas;
1224         }
1225
1226 retry_flush_dents:
1227         /* write all the dirty dentry pages */
1228         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1229                 f2fs_unlock_all(sbi);
1230                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1231                 if (err)
1232                         return err;
1233                 cond_resched();
1234                 goto retry_flush_quotas;
1235         }
1236
1237         /*
1238          * POR: we should ensure that there are no dirty node pages
1239          * until finishing nat/sit flush. inode->i_blocks can be updated.
1240          */
1241         down_write(&sbi->node_change);
1242
1243         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1244                 up_write(&sbi->node_change);
1245                 f2fs_unlock_all(sbi);
1246                 err = f2fs_sync_inode_meta(sbi);
1247                 if (err)
1248                         return err;
1249                 cond_resched();
1250                 goto retry_flush_quotas;
1251         }
1252
1253 retry_flush_nodes:
1254         down_write(&sbi->node_write);
1255
1256         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1257                 up_write(&sbi->node_write);
1258                 atomic_inc(&sbi->wb_sync_req[NODE]);
1259                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1260                 atomic_dec(&sbi->wb_sync_req[NODE]);
1261                 if (err) {
1262                         up_write(&sbi->node_change);
1263                         f2fs_unlock_all(sbi);
1264                         return err;
1265                 }
1266                 cond_resched();
1267                 goto retry_flush_nodes;
1268         }
1269
1270         /*
1271          * sbi->node_change is used only for AIO write_begin path which produces
1272          * dirty node blocks and some checkpoint values by block allocation.
1273          */
1274         __prepare_cp_block(sbi);
1275         up_write(&sbi->node_change);
1276         return err;
1277 }
1278
1279 static void unblock_operations(struct f2fs_sb_info *sbi)
1280 {
1281         up_write(&sbi->node_write);
1282         f2fs_unlock_all(sbi);
1283 }
1284
1285 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1286 {
1287         DEFINE_WAIT(wait);
1288
1289         for (;;) {
1290                 if (!get_pages(sbi, type))
1291                         break;
1292
1293                 if (unlikely(f2fs_cp_error(sbi) &&
1294                         !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1295                         break;
1296
1297                 if (type == F2FS_DIRTY_META)
1298                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1299                                                         FS_CP_META_IO);
1300                 else if (type == F2FS_WB_CP_DATA)
1301                         f2fs_submit_merged_write(sbi, DATA);
1302
1303                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1304                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1305         }
1306         finish_wait(&sbi->cp_wait, &wait);
1307 }
1308
1309 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1310 {
1311         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1312         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1313         unsigned long flags;
1314
1315         spin_lock_irqsave(&sbi->cp_lock, flags);
1316
1317         if ((cpc->reason & CP_UMOUNT) &&
1318                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1319                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1320                 disable_nat_bits(sbi, false);
1321
1322         if (cpc->reason & CP_TRIMMED)
1323                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1324         else
1325                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1326
1327         if (cpc->reason & CP_UMOUNT)
1328                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1329         else
1330                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1331
1332         if (cpc->reason & CP_FASTBOOT)
1333                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1334         else
1335                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1336
1337         if (orphan_num)
1338                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1339         else
1340                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1341
1342         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1343                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1344
1345         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1346                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1347         else
1348                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1349
1350         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1351                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1352         else
1353                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1354
1355         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1356                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1357         else
1358                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1359
1360         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1361                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1362         else
1363                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1364
1365         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1366                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1367
1368         /* set this flag to activate crc|cp_ver for recovery */
1369         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1370         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1371
1372         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1373 }
1374
1375 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1376         void *src, block_t blk_addr)
1377 {
1378         struct writeback_control wbc = {
1379                 .for_reclaim = 0,
1380         };
1381
1382         /*
1383          * pagevec_lookup_tag and lock_page again will take
1384          * some extra time. Therefore, f2fs_update_meta_pages and
1385          * f2fs_sync_meta_pages are combined in this function.
1386          */
1387         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1388         int err;
1389
1390         f2fs_wait_on_page_writeback(page, META, true, true);
1391
1392         memcpy(page_address(page), src, PAGE_SIZE);
1393
1394         set_page_dirty(page);
1395         if (unlikely(!clear_page_dirty_for_io(page)))
1396                 f2fs_bug_on(sbi, 1);
1397
1398         /* writeout cp pack 2 page */
1399         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1400         if (unlikely(err && f2fs_cp_error(sbi))) {
1401                 f2fs_put_page(page, 1);
1402                 return;
1403         }
1404
1405         f2fs_bug_on(sbi, err);
1406         f2fs_put_page(page, 0);
1407
1408         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1409         f2fs_submit_merged_write(sbi, META_FLUSH);
1410 }
1411
1412 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1413 {
1414         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1415         struct f2fs_nm_info *nm_i = NM_I(sbi);
1416         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1417         block_t start_blk;
1418         unsigned int data_sum_blocks, orphan_blocks;
1419         __u32 crc32 = 0;
1420         int i;
1421         int cp_payload_blks = __cp_payload(sbi);
1422         struct super_block *sb = sbi->sb;
1423         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1424         u64 kbytes_written;
1425         int err;
1426
1427         /* Flush all the NAT/SIT pages */
1428         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1429
1430         /* start to update checkpoint, cp ver is already updated previously */
1431         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1432         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1433         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1434                 ckpt->cur_node_segno[i] =
1435                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1436                 ckpt->cur_node_blkoff[i] =
1437                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1438                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1439                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1440         }
1441         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1442                 ckpt->cur_data_segno[i] =
1443                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1444                 ckpt->cur_data_blkoff[i] =
1445                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1446                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1447                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1448         }
1449
1450         /* 2 cp + n data seg summary + orphan inode blocks */
1451         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1452         spin_lock_irqsave(&sbi->cp_lock, flags);
1453         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1454                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1455         else
1456                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1457         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1458
1459         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1460         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1461                         orphan_blocks);
1462
1463         if (__remain_node_summaries(cpc->reason))
1464                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1465                                 cp_payload_blks + data_sum_blocks +
1466                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1467         else
1468                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1469                                 cp_payload_blks + data_sum_blocks +
1470                                 orphan_blocks);
1471
1472         /* update ckpt flag for checkpoint */
1473         update_ckpt_flags(sbi, cpc);
1474
1475         /* update SIT/NAT bitmap */
1476         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1477         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1478
1479         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1480         *((__le32 *)((unsigned char *)ckpt +
1481                                 le32_to_cpu(ckpt->checksum_offset)))
1482                                 = cpu_to_le32(crc32);
1483
1484         start_blk = __start_cp_next_addr(sbi);
1485
1486         /* write nat bits */
1487         if (enabled_nat_bits(sbi, cpc)) {
1488                 __u64 cp_ver = cur_cp_version(ckpt);
1489                 block_t blk;
1490
1491                 cp_ver |= ((__u64)crc32 << 32);
1492                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1493
1494                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1495                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1496                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1497                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1498         }
1499
1500         /* write out checkpoint buffer at block 0 */
1501         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1502
1503         for (i = 1; i < 1 + cp_payload_blks; i++)
1504                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1505                                                         start_blk++);
1506
1507         if (orphan_num) {
1508                 write_orphan_inodes(sbi, start_blk);
1509                 start_blk += orphan_blocks;
1510         }
1511
1512         f2fs_write_data_summaries(sbi, start_blk);
1513         start_blk += data_sum_blocks;
1514
1515         /* Record write statistics in the hot node summary */
1516         kbytes_written = sbi->kbytes_written;
1517         if (sb->s_bdev->bd_part)
1518                 kbytes_written += BD_PART_WRITTEN(sbi);
1519
1520         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1521
1522         if (__remain_node_summaries(cpc->reason)) {
1523                 f2fs_write_node_summaries(sbi, start_blk);
1524                 start_blk += NR_CURSEG_NODE_TYPE;
1525         }
1526
1527         /* update user_block_counts */
1528         sbi->last_valid_block_count = sbi->total_valid_block_count;
1529         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1530
1531         /* Here, we have one bio having CP pack except cp pack 2 page */
1532         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1533         /* Wait for all dirty meta pages to be submitted for IO */
1534         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1535
1536         /* wait for previous submitted meta pages writeback */
1537         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1538
1539         /* flush all device cache */
1540         err = f2fs_flush_device_cache(sbi);
1541         if (err)
1542                 return err;
1543
1544         /* barrier and flush checkpoint cp pack 2 page if it can */
1545         commit_checkpoint(sbi, ckpt, start_blk);
1546         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1547
1548         /*
1549          * invalidate intermediate page cache borrowed from meta inode which are
1550          * used for migration of encrypted, verity or compressed inode's blocks.
1551          */
1552         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1553                 f2fs_sb_has_compression(sbi))
1554                 invalidate_mapping_pages(META_MAPPING(sbi),
1555                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1556
1557         f2fs_release_ino_entry(sbi, false);
1558
1559         f2fs_reset_fsync_node_info(sbi);
1560
1561         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1562         clear_sbi_flag(sbi, SBI_NEED_CP);
1563         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1564
1565         spin_lock(&sbi->stat_lock);
1566         sbi->unusable_block_count = 0;
1567         spin_unlock(&sbi->stat_lock);
1568
1569         __set_cp_next_pack(sbi);
1570
1571         /*
1572          * redirty superblock if metadata like node page or inode cache is
1573          * updated during writing checkpoint.
1574          */
1575         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1576                         get_pages(sbi, F2FS_DIRTY_IMETA))
1577                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1578
1579         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1580
1581         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1582 }
1583
1584 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1585 {
1586         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1587         unsigned long long ckpt_ver;
1588         int err = 0;
1589
1590         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1591                 return -EROFS;
1592
1593         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1594                 if (cpc->reason != CP_PAUSE)
1595                         return 0;
1596                 f2fs_warn(sbi, "Start checkpoint disabled!");
1597         }
1598         if (cpc->reason != CP_RESIZE)
1599                 mutex_lock(&sbi->cp_mutex);
1600
1601         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1602                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1603                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1604                 goto out;
1605         if (unlikely(f2fs_cp_error(sbi))) {
1606                 err = -EIO;
1607                 goto out;
1608         }
1609
1610         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1611
1612         err = block_operations(sbi);
1613         if (err)
1614                 goto out;
1615
1616         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1617
1618         f2fs_flush_merged_writes(sbi);
1619
1620         /* this is the case of multiple fstrims without any changes */
1621         if (cpc->reason & CP_DISCARD) {
1622                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1623                         unblock_operations(sbi);
1624                         goto out;
1625                 }
1626
1627                 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1628                                 SIT_I(sbi)->dirty_sentries == 0 &&
1629                                 prefree_segments(sbi) == 0) {
1630                         f2fs_flush_sit_entries(sbi, cpc);
1631                         f2fs_clear_prefree_segments(sbi, cpc);
1632                         unblock_operations(sbi);
1633                         goto out;
1634                 }
1635         }
1636
1637         /*
1638          * update checkpoint pack index
1639          * Increase the version number so that
1640          * SIT entries and seg summaries are written at correct place
1641          */
1642         ckpt_ver = cur_cp_version(ckpt);
1643         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1644
1645         /* write cached NAT/SIT entries to NAT/SIT area */
1646         err = f2fs_flush_nat_entries(sbi, cpc);
1647         if (err)
1648                 goto stop;
1649
1650         f2fs_flush_sit_entries(sbi, cpc);
1651
1652         /* save inmem log status */
1653         f2fs_save_inmem_curseg(sbi);
1654
1655         err = do_checkpoint(sbi, cpc);
1656         if (err)
1657                 f2fs_release_discard_addrs(sbi);
1658         else
1659                 f2fs_clear_prefree_segments(sbi, cpc);
1660
1661         f2fs_restore_inmem_curseg(sbi);
1662 stop:
1663         unblock_operations(sbi);
1664         stat_inc_cp_count(sbi->stat_info);
1665
1666         if (cpc->reason & CP_RECOVERY)
1667                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1668
1669         /* update CP_TIME to trigger checkpoint periodically */
1670         f2fs_update_time(sbi, CP_TIME);
1671         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1672 out:
1673         if (cpc->reason != CP_RESIZE)
1674                 mutex_unlock(&sbi->cp_mutex);
1675         return err;
1676 }
1677
1678 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1679 {
1680         int i;
1681
1682         for (i = 0; i < MAX_INO_ENTRY; i++) {
1683                 struct inode_management *im = &sbi->im[i];
1684
1685                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1686                 spin_lock_init(&im->ino_lock);
1687                 INIT_LIST_HEAD(&im->ino_list);
1688                 im->ino_num = 0;
1689         }
1690
1691         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1692                         NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1693                                 F2FS_ORPHANS_PER_BLOCK;
1694 }
1695
1696 int __init f2fs_create_checkpoint_caches(void)
1697 {
1698         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1699                         sizeof(struct ino_entry));
1700         if (!ino_entry_slab)
1701                 return -ENOMEM;
1702         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1703                         sizeof(struct inode_entry));
1704         if (!f2fs_inode_entry_slab) {
1705                 kmem_cache_destroy(ino_entry_slab);
1706                 return -ENOMEM;
1707         }
1708         return 0;
1709 }
1710
1711 void f2fs_destroy_checkpoint_caches(void)
1712 {
1713         kmem_cache_destroy(ino_entry_slab);
1714         kmem_cache_destroy(f2fs_inode_entry_slab);
1715 }