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_nofail(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_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
 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) {
 148                 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 149                          blkaddr, exist);
 150                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 151                 WARN_ON(1);
 152         }
 153         return exist;
 154 }
 155
 156 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 157                                         block_t blkaddr, int type)
 158 {
 159         switch (type) {
 160         case META_NAT:
 161                 break;
 162         case META_SIT:
 163                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 164                         return false;
 165                 break;
 166         case META_SSA:
 167                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 168                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 169                         return false;
 170                 break;
 171         case META_CP:
 172                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 173                         blkaddr < __start_cp_addr(sbi)))
 174                         return false;
 175                 break;
 176         case META_POR:
 177                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 178                         blkaddr < MAIN_BLKADDR(sbi)))
 179                         return false;
 180                 break;
 181         case DATA_GENERIC:
 182         case DATA_GENERIC_ENHANCE:
 183         case DATA_GENERIC_ENHANCE_READ:
 184                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 185                                 blkaddr < MAIN_BLKADDR(sbi))) {
 186                         f2fs_warn(sbi, "access invalid blkaddr:%u",
 187                                   blkaddr);
 188                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 189                         WARN_ON(1);
 190                         return false;
 191                 } else {
 192                         return __is_bitmap_valid(sbi, blkaddr, type);
 193                 }
 194                 break;
 195         case META_GENERIC:
 196                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 197                         blkaddr >= MAIN_BLKADDR(sbi)))
 198                         return false;
 199                 break;
 200         default:
 201                 BUG();
 202         }
 203
 204         return true;
 205 }
 206
 207 /*
 208  * Readahead CP/NAT/SIT/SSA/POR pages
 209  */
 210 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 211                                                         int type, bool sync)
 212 {
 213         struct page *page;
 214         block_t blkno = start;
 215         struct f2fs_io_info fio = {
 216                 .sbi = sbi,
 217                 .type = META,
 218                 .op = REQ_OP_READ,
 219                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 220                 .encrypted_page = NULL,
 221                 .in_list = false,
 222                 .is_por = (type == META_POR),
 223         };
 224         struct blk_plug plug;
 225         int err;
 226
 227         if (unlikely(type == META_POR))
 228                 fio.op_flags &= ~REQ_META;
 229
 230         blk_start_plug(&plug);
 231         for (; nrpages-- > 0; blkno++) {
 232
 233                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 234                         goto out;
 235
 236                 switch (type) {
 237                 case META_NAT:
 238                         if (unlikely(blkno >=
 239                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 240                                 blkno = 0;
 241                         /* get nat block addr */
 242                         fio.new_blkaddr = current_nat_addr(sbi,
 243                                         blkno * NAT_ENTRY_PER_BLOCK);
 244                         break;
 245                 case META_SIT:
 246                         /* get sit block addr */
 247                         fio.new_blkaddr = current_sit_addr(sbi,
 248                                         blkno * SIT_ENTRY_PER_BLOCK);
 249                         break;
 250                 case META_SSA:
 251                 case META_CP:
 252                 case META_POR:
 253                         fio.new_blkaddr = blkno;
 254                         break;
 255                 default:
 256                         BUG();
 257                 }
 258
 259                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 260                                                 fio.new_blkaddr, false);
 261                 if (!page)
 262                         continue;
 263                 if (PageUptodate(page)) {
 264                         f2fs_put_page(page, 1);
 265                         continue;
 266                 }
 267
 268                 fio.page = page;
 269                 err = f2fs_submit_page_bio(&fio);
 270                 f2fs_put_page(page, err ? 1 : 0);
 271
 272                 if (!err)
 273                         f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
 274         }
 275 out:
 276         blk_finish_plug(&plug);
 277         return blkno - start;
 278 }
 279
 280 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 281 {
 282         struct page *page;
 283         bool readahead = false;
 284
 285         page = find_get_page(META_MAPPING(sbi), index);
 286         if (!page || !PageUptodate(page))
 287                 readahead = true;
 288         f2fs_put_page(page, 0);
 289
 290         if (readahead)
 291                 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 292 }
 293
 294 static int __f2fs_write_meta_page(struct page *page,
 295                                 struct writeback_control *wbc,
 296                                 enum iostat_type io_type)
 297 {
 298         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 299
 300         trace_f2fs_writepage(page, META);
 301
 302         if (unlikely(f2fs_cp_error(sbi)))
 303                 goto redirty_out;
 304         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 305                 goto redirty_out;
 306         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 307                 goto redirty_out;
 308
 309         f2fs_do_write_meta_page(sbi, page, io_type);
 310         dec_page_count(sbi, F2FS_DIRTY_META);
 311
 312         if (wbc->for_reclaim)
 313                 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 314
 315         unlock_page(page);
 316
 317         if (unlikely(f2fs_cp_error(sbi)))
 318                 f2fs_submit_merged_write(sbi, META);
 319
 320         return 0;
 321
 322 redirty_out:
 323         redirty_page_for_writepage(wbc, page);
 324         return AOP_WRITEPAGE_ACTIVATE;
 325 }
 326
 327 static int f2fs_write_meta_page(struct page *page,
 328                                 struct writeback_control *wbc)
 329 {
 330         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 331 }
 332
 333 static int f2fs_write_meta_pages(struct address_space *mapping,
 334                                 struct writeback_control *wbc)
 335 {
 336         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 337         long diff, written;
 338
 339         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 340                 goto skip_write;
 341
 342         /* collect a number of dirty meta pages and write together */
 343         if (wbc->sync_mode != WB_SYNC_ALL &&
 344                         get_pages(sbi, F2FS_DIRTY_META) <
 345                                         nr_pages_to_skip(sbi, META))
 346                 goto skip_write;
 347
 348         /* if locked failed, cp will flush dirty pages instead */
 349         if (!mutex_trylock(&sbi->cp_mutex))
 350                 goto skip_write;
 351
 352         trace_f2fs_writepages(mapping->host, wbc, META);
 353         diff = nr_pages_to_write(sbi, META, wbc);
 354         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 355         mutex_unlock(&sbi->cp_mutex);
 356         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 357         return 0;
 358
 359 skip_write:
 360         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 361         trace_f2fs_writepages(mapping->host, wbc, META);
 362         return 0;
 363 }
 364
 365 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 366                                 long nr_to_write, enum iostat_type io_type)
 367 {
 368         struct address_space *mapping = META_MAPPING(sbi);
 369         pgoff_t index = 0, prev = ULONG_MAX;
 370         struct pagevec pvec;
 371         long nwritten = 0;
 372         int nr_pages;
 373         struct writeback_control wbc = {
 374                 .for_reclaim = 0,
 375         };
 376         struct blk_plug plug;
 377
 378         pagevec_init(&pvec);
 379
 380         blk_start_plug(&plug);
 381
 382         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 383                                 PAGECACHE_TAG_DIRTY))) {
 384                 int i;
 385
 386                 for (i = 0; i < nr_pages; i++) {
 387                         struct page *page = pvec.pages[i];
 388
 389                         if (prev == ULONG_MAX)
 390                                 prev = page->index - 1;
 391                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 392                                 pagevec_release(&pvec);
 393                                 goto stop;
 394                         }
 395
 396                         lock_page(page);
 397
 398                         if (unlikely(page->mapping != mapping)) {
 399 continue_unlock:
 400                                 unlock_page(page);
 401                                 continue;
 402                         }
 403                         if (!PageDirty(page)) {
 404                                 /* someone wrote it for us */
 405                                 goto continue_unlock;
 406                         }
 407
 408                         f2fs_wait_on_page_writeback(page, META, true, true);
 409
 410                         if (!clear_page_dirty_for_io(page))
 411                                 goto continue_unlock;
 412
 413                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 414                                 unlock_page(page);
 415                                 break;
 416                         }
 417                         nwritten++;
 418                         prev = page->index;
 419                         if (unlikely(nwritten >= nr_to_write))
 420                                 break;
 421                 }
 422                 pagevec_release(&pvec);
 423                 cond_resched();
 424         }
 425 stop:
 426         if (nwritten)
 427                 f2fs_submit_merged_write(sbi, type);
 428
 429         blk_finish_plug(&plug);
 430
 431         return nwritten;
 432 }
 433
 434 static int f2fs_set_meta_page_dirty(struct page *page)
 435 {
 436         trace_f2fs_set_page_dirty(page, META);
 437
 438         if (!PageUptodate(page))
 439                 SetPageUptodate(page);
 440         if (!PageDirty(page)) {
 441                 __set_page_dirty_nobuffers(page);
 442                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 443                 f2fs_set_page_private(page, 0);
 444                 f2fs_trace_pid(page);
 445                 return 1;
 446         }
 447         return 0;
 448 }
 449
 450 const struct address_space_operations f2fs_meta_aops = {
 451         .writepage      = f2fs_write_meta_page,
 452         .writepages     = f2fs_write_meta_pages,
 453         .set_page_dirty = f2fs_set_meta_page_dirty,
 454         .invalidatepage = f2fs_invalidate_page,
 455         .releasepage    = f2fs_release_page,
 456 #ifdef CONFIG_MIGRATION
 457         .migratepage    = f2fs_migrate_page,
 458 #endif
 459 };
 460
 461 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 462                                                 unsigned int devidx, int type)
 463 {
 464         struct inode_management *im = &sbi->im[type];
 465         struct ino_entry *e, *tmp;
 466
 467         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 468
 469         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 470
 471         spin_lock(&im->ino_lock);
 472         e = radix_tree_lookup(&im->ino_root, ino);
 473         if (!e) {
 474                 e = tmp;
 475                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 476                         f2fs_bug_on(sbi, 1);
 477
 478                 memset(e, 0, sizeof(struct ino_entry));
 479                 e->ino = ino;
 480
 481                 list_add_tail(&e->list, &im->ino_list);
 482                 if (type != ORPHAN_INO)
 483                         im->ino_num++;
 484         }
 485
 486         if (type == FLUSH_INO)
 487                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 488
 489         spin_unlock(&im->ino_lock);
 490         radix_tree_preload_end();
 491
 492         if (e != tmp)
 493                 kmem_cache_free(ino_entry_slab, tmp);
 494 }
 495
 496 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 497 {
 498         struct inode_management *im = &sbi->im[type];
 499         struct ino_entry *e;
 500
 501         spin_lock(&im->ino_lock);
 502         e = radix_tree_lookup(&im->ino_root, ino);
 503         if (e) {
 504                 list_del(&e->list);
 505                 radix_tree_delete(&im->ino_root, ino);
 506                 im->ino_num--;
 507                 spin_unlock(&im->ino_lock);
 508                 kmem_cache_free(ino_entry_slab, e);
 509                 return;
 510         }
 511         spin_unlock(&im->ino_lock);
 512 }
 513
 514 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 515 {
 516         /* add new dirty ino entry into list */
 517         __add_ino_entry(sbi, ino, 0, type);
 518 }
 519
 520 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 521 {
 522         /* remove dirty ino entry from list */
 523         __remove_ino_entry(sbi, ino, type);
 524 }
 525
 526 /* mode should be APPEND_INO or UPDATE_INO */
 527 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 528 {
 529         struct inode_management *im = &sbi->im[mode];
 530         struct ino_entry *e;
 531
 532         spin_lock(&im->ino_lock);
 533         e = radix_tree_lookup(&im->ino_root, ino);
 534         spin_unlock(&im->ino_lock);
 535         return e ? true : false;
 536 }
 537
 538 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 539 {
 540         struct ino_entry *e, *tmp;
 541         int i;
 542
 543         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 544                 struct inode_management *im = &sbi->im[i];
 545
 546                 spin_lock(&im->ino_lock);
 547                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 548                         list_del(&e->list);
 549                         radix_tree_delete(&im->ino_root, e->ino);
 550                         kmem_cache_free(ino_entry_slab, e);
 551                         im->ino_num--;
 552                 }
 553                 spin_unlock(&im->ino_lock);
 554         }
 555 }
 556
 557 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 558                                         unsigned int devidx, int type)
 559 {
 560         __add_ino_entry(sbi, ino, devidx, type);
 561 }
 562
 563 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 564                                         unsigned int devidx, int type)
 565 {
 566         struct inode_management *im = &sbi->im[type];
 567         struct ino_entry *e;
 568         bool is_dirty = false;
 569
 570         spin_lock(&im->ino_lock);
 571         e = radix_tree_lookup(&im->ino_root, ino);
 572         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 573                 is_dirty = true;
 574         spin_unlock(&im->ino_lock);
 575         return is_dirty;
 576 }
 577
 578 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 579 {
 580         struct inode_management *im = &sbi->im[ORPHAN_INO];
 581         int err = 0;
 582
 583         spin_lock(&im->ino_lock);
 584
 585         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 586                 spin_unlock(&im->ino_lock);
 587                 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
 588                 return -ENOSPC;
 589         }
 590
 591         if (unlikely(im->ino_num >= sbi->max_orphans))
 592                 err = -ENOSPC;
 593         else
 594                 im->ino_num++;
 595         spin_unlock(&im->ino_lock);
 596
 597         return err;
 598 }
 599
 600 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 601 {
 602         struct inode_management *im = &sbi->im[ORPHAN_INO];
 603
 604         spin_lock(&im->ino_lock);
 605         f2fs_bug_on(sbi, im->ino_num == 0);
 606         im->ino_num--;
 607         spin_unlock(&im->ino_lock);
 608 }
 609
 610 void f2fs_add_orphan_inode(struct inode *inode)
 611 {
 612         /* add new orphan ino entry into list */
 613         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 614         f2fs_update_inode_page(inode);
 615 }
 616
 617 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 618 {
 619         /* remove orphan entry from orphan list */
 620         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 621 }
 622
 623 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 624 {
 625         struct inode *inode;
 626         struct node_info ni;
 627         int err;
 628
 629         inode = f2fs_iget_retry(sbi->sb, ino);
 630         if (IS_ERR(inode)) {
 631                 /*
 632                  * there should be a bug that we can't find the entry
 633                  * to orphan inode.
 634                  */
 635                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 636                 return PTR_ERR(inode);
 637         }
 638
 639         err = dquot_initialize(inode);
 640         if (err) {
 641                 iput(inode);
 642                 goto err_out;
 643         }
 644
 645         clear_nlink(inode);
 646
 647         /* truncate all the data during iput */
 648         iput(inode);
 649
 650         err = f2fs_get_node_info(sbi, ino, &ni);
 651         if (err)
 652                 goto err_out;
 653
 654         /* ENOMEM was fully retried in f2fs_evict_inode. */
 655         if (ni.blk_addr != NULL_ADDR) {
 656                 err = -EIO;
 657                 goto err_out;
 658         }
 659         return 0;
 660
 661 err_out:
 662         set_sbi_flag(sbi, SBI_NEED_FSCK);
 663         f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 664                   __func__, ino);
 665         return err;
 666 }
 667
 668 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 669 {
 670         block_t start_blk, orphan_blocks, i, j;
 671         unsigned int s_flags = sbi->sb->s_flags;
 672         int err = 0;
 673 #ifdef CONFIG_QUOTA
 674         int quota_enabled;
 675 #endif
 676
 677         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 678                 return 0;
 679
 680         if (bdev_read_only(sbi->sb->s_bdev)) {
 681                 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 682                 return 0;
 683         }
 684
 685         if (s_flags & SB_RDONLY) {
 686                 f2fs_info(sbi, "orphan cleanup on readonly fs");
 687                 sbi->sb->s_flags &= ~SB_RDONLY;
 688         }
 689
 690 #ifdef CONFIG_QUOTA
 691         /* Needed for iput() to work correctly and not trash data */
 692         sbi->sb->s_flags |= SB_ACTIVE;
 693
 694         /*
 695          * Turn on quotas which were not enabled for read-only mounts if
 696          * filesystem has quota feature, so that they are updated correctly.
 697          */
 698         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 699 #endif
 700
 701         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 702         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 703
 704         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 705
 706         for (i = 0; i < orphan_blocks; i++) {
 707                 struct page *page;
 708                 struct f2fs_orphan_block *orphan_blk;
 709
 710                 page = f2fs_get_meta_page(sbi, start_blk + i);
 711                 if (IS_ERR(page)) {
 712                         err = PTR_ERR(page);
 713                         goto out;
 714                 }
 715
 716                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 717                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 718                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 719                         err = recover_orphan_inode(sbi, ino);
 720                         if (err) {
 721                                 f2fs_put_page(page, 1);
 722                                 goto out;
 723                         }
 724                 }
 725                 f2fs_put_page(page, 1);
 726         }
 727         /* clear Orphan Flag */
 728         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 729 out:
 730         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 731
 732 #ifdef CONFIG_QUOTA
 733         /* Turn quotas off */
 734         if (quota_enabled)
 735                 f2fs_quota_off_umount(sbi->sb);
 736 #endif
 737         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 738
 739         return err;
 740 }
 741
 742 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 743 {
 744         struct list_head *head;
 745         struct f2fs_orphan_block *orphan_blk = NULL;
 746         unsigned int nentries = 0;
 747         unsigned short index = 1;
 748         unsigned short orphan_blocks;
 749         struct page *page = NULL;
 750         struct ino_entry *orphan = NULL;
 751         struct inode_management *im = &sbi->im[ORPHAN_INO];
 752
 753         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 754
 755         /*
 756          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 757          * orphan inode operations are covered under f2fs_lock_op().
 758          * And, spin_lock should be avoided due to page operations below.
 759          */
 760         head = &im->ino_list;
 761
 762         /* loop for each orphan inode entry and write them in Jornal block */
 763         list_for_each_entry(orphan, head, list) {
 764                 if (!page) {
 765                         page = f2fs_grab_meta_page(sbi, start_blk++);
 766                         orphan_blk =
 767                                 (struct f2fs_orphan_block *)page_address(page);
 768                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 769                 }
 770
 771                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 772
 773                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 774                         /*
 775                          * an orphan block is full of 1020 entries,
 776                          * then we need to flush current orphan blocks
 777                          * and bring another one in memory
 778                          */
 779                         orphan_blk->blk_addr = cpu_to_le16(index);
 780                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 781                         orphan_blk->entry_count = cpu_to_le32(nentries);
 782                         set_page_dirty(page);
 783                         f2fs_put_page(page, 1);
 784                         index++;
 785                         nentries = 0;
 786                         page = NULL;
 787                 }
 788         }
 789
 790         if (page) {
 791                 orphan_blk->blk_addr = cpu_to_le16(index);
 792                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 793                 orphan_blk->entry_count = cpu_to_le32(nentries);
 794                 set_page_dirty(page);
 795                 f2fs_put_page(page, 1);
 796         }
 797 }
 798
 799 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 800                                                 struct f2fs_checkpoint *ckpt)
 801 {
 802         unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 803         __u32 chksum;
 804
 805         chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 806         if (chksum_ofs < CP_CHKSUM_OFFSET) {
 807                 chksum_ofs += sizeof(chksum);
 808                 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 809                                                 F2FS_BLKSIZE - chksum_ofs);
 810         }
 811         return chksum;
 812 }
 813
 814 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 815                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 816                 unsigned long long *version)
 817 {
 818         size_t crc_offset = 0;
 819         __u32 crc;
 820
 821         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 822         if (IS_ERR(*cp_page))
 823                 return PTR_ERR(*cp_page);
 824
 825         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 826
 827         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 828         if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 829                         crc_offset > CP_CHKSUM_OFFSET) {
 830                 f2fs_put_page(*cp_page, 1);
 831                 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 832                 return -EINVAL;
 833         }
 834
 835         crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 836         if (crc != cur_cp_crc(*cp_block)) {
 837                 f2fs_put_page(*cp_page, 1);
 838                 f2fs_warn(sbi, "invalid crc value");
 839                 return -EINVAL;
 840         }
 841
 842         *version = cur_cp_version(*cp_block);
 843         return 0;
 844 }
 845
 846 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 847                                 block_t cp_addr, unsigned long long *version)
 848 {
 849         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 850         struct f2fs_checkpoint *cp_block = NULL;
 851         unsigned long long cur_version = 0, pre_version = 0;
 852         int err;
 853
 854         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 855                                         &cp_page_1, version);
 856         if (err)
 857                 return NULL;
 858
 859         if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
 860                                         sbi->blocks_per_seg) {
 861                 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 862                           le32_to_cpu(cp_block->cp_pack_total_block_count));
 863                 goto invalid_cp;
 864         }
 865         pre_version = *version;
 866
 867         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 868         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 869                                         &cp_page_2, version);
 870         if (err)
 871                 goto invalid_cp;
 872         cur_version = *version;
 873
 874         if (cur_version == pre_version) {
 875                 *version = cur_version;
 876                 f2fs_put_page(cp_page_2, 1);
 877                 return cp_page_1;
 878         }
 879         f2fs_put_page(cp_page_2, 1);
 880 invalid_cp:
 881         f2fs_put_page(cp_page_1, 1);
 882         return NULL;
 883 }
 884
 885 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 886 {
 887         struct f2fs_checkpoint *cp_block;
 888         struct f2fs_super_block *fsb = sbi->raw_super;
 889         struct page *cp1, *cp2, *cur_page;
 890         unsigned long blk_size = sbi->blocksize;
 891         unsigned long long cp1_version = 0, cp2_version = 0;
 892         unsigned long long cp_start_blk_no;
 893         unsigned int cp_blks = 1 + __cp_payload(sbi);
 894         block_t cp_blk_no;
 895         int i;
 896         int err;
 897
 898         sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
 899                                  GFP_KERNEL);
 900         if (!sbi->ckpt)
 901                 return -ENOMEM;
 902         /*
 903          * Finding out valid cp block involves read both
 904          * sets( cp pack 1 and cp pack 2)
 905          */
 906         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 907         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 908
 909         /* The second checkpoint pack should start at the next segment */
 910         cp_start_blk_no += ((unsigned long long)1) <<
 911                                 le32_to_cpu(fsb->log_blocks_per_seg);
 912         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 913
 914         if (cp1 && cp2) {
 915                 if (ver_after(cp2_version, cp1_version))
 916                         cur_page = cp2;
 917                 else
 918                         cur_page = cp1;
 919         } else if (cp1) {
 920                 cur_page = cp1;
 921         } else if (cp2) {
 922                 cur_page = cp2;
 923         } else {
 924                 err = -EFSCORRUPTED;
 925                 goto fail_no_cp;
 926         }
 927
 928         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 929         memcpy(sbi->ckpt, cp_block, blk_size);
 930
 931         if (cur_page == cp1)
 932                 sbi->cur_cp_pack = 1;
 933         else
 934                 sbi->cur_cp_pack = 2;
 935
 936         /* Sanity checking of checkpoint */
 937         if (f2fs_sanity_check_ckpt(sbi)) {
 938                 err = -EFSCORRUPTED;
 939                 goto free_fail_no_cp;
 940         }
 941
 942         if (cp_blks <= 1)
 943                 goto done;
 944
 945         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 946         if (cur_page == cp2)
 947                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 948
 949         for (i = 1; i < cp_blks; i++) {
 950                 void *sit_bitmap_ptr;
 951                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 952
 953                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 954                 if (IS_ERR(cur_page)) {
 955                         err = PTR_ERR(cur_page);
 956                         goto free_fail_no_cp;
 957                 }
 958                 sit_bitmap_ptr = page_address(cur_page);
 959                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 960                 f2fs_put_page(cur_page, 1);
 961         }
 962 done:
 963         f2fs_put_page(cp1, 1);
 964         f2fs_put_page(cp2, 1);
 965         return 0;
 966
 967 free_fail_no_cp:
 968         f2fs_put_page(cp1, 1);
 969         f2fs_put_page(cp2, 1);
 970 fail_no_cp:
 971         kvfree(sbi->ckpt);
 972         return err;
 973 }
 974
 975 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 976 {
 977         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 978         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 979
 980         if (is_inode_flag_set(inode, flag))
 981                 return;
 982
 983         set_inode_flag(inode, flag);
 984         if (!f2fs_is_volatile_file(inode))
 985                 list_add_tail(&F2FS_I(inode)->dirty_list,
 986                                                 &sbi->inode_list[type]);
 987         stat_inc_dirty_inode(sbi, type);
 988 }
 989
 990 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 991 {
 992         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 993
 994         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 995                 return;
 996
 997         list_del_init(&F2FS_I(inode)->dirty_list);
 998         clear_inode_flag(inode, flag);
 999         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1000 }
1001
1002 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1003 {
1004         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1005         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1006
1007         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1008                         !S_ISLNK(inode->i_mode))
1009                 return;
1010
1011         spin_lock(&sbi->inode_lock[type]);
1012         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1013                 __add_dirty_inode(inode, type);
1014         inode_inc_dirty_pages(inode);
1015         spin_unlock(&sbi->inode_lock[type]);
1016
1017         f2fs_set_page_private(page, 0);
1018         f2fs_trace_pid(page);
1019 }
1020
1021 void f2fs_remove_dirty_inode(struct inode *inode)
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         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1031                 return;
1032
1033         spin_lock(&sbi->inode_lock[type]);
1034         __remove_dirty_inode(inode, type);
1035         spin_unlock(&sbi->inode_lock[type]);
1036 }
1037
1038 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1039 {
1040         struct list_head *head;
1041         struct inode *inode;
1042         struct f2fs_inode_info *fi;
1043         bool is_dir = (type == DIR_INODE);
1044         unsigned long ino = 0;
1045
1046         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1047                                 get_pages(sbi, is_dir ?
1048                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1049 retry:
1050         if (unlikely(f2fs_cp_error(sbi)))
1051                 return -EIO;
1052
1053         spin_lock(&sbi->inode_lock[type]);
1054
1055         head = &sbi->inode_list[type];
1056         if (list_empty(head)) {
1057                 spin_unlock(&sbi->inode_lock[type]);
1058                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1059                                 get_pages(sbi, is_dir ?
1060                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1061                 return 0;
1062         }
1063         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1064         inode = igrab(&fi->vfs_inode);
1065         spin_unlock(&sbi->inode_lock[type]);
1066         if (inode) {
1067                 unsigned long cur_ino = inode->i_ino;
1068
1069                 F2FS_I(inode)->cp_task = current;
1070
1071                 filemap_fdatawrite(inode->i_mapping);
1072
1073                 F2FS_I(inode)->cp_task = NULL;
1074
1075                 iput(inode);
1076                 /* We need to give cpu to another writers. */
1077                 if (ino == cur_ino)
1078                         cond_resched();
1079                 else
1080                         ino = cur_ino;
1081         } else {
1082                 /*
1083                  * We should submit bio, since it exists several
1084                  * wribacking dentry pages in the freeing inode.
1085                  */
1086                 f2fs_submit_merged_write(sbi, DATA);
1087                 cond_resched();
1088         }
1089         goto retry;
1090 }
1091
1092 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1093 {
1094         struct list_head *head = &sbi->inode_list[DIRTY_META];
1095         struct inode *inode;
1096         struct f2fs_inode_info *fi;
1097         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1098
1099         while (total--) {
1100                 if (unlikely(f2fs_cp_error(sbi)))
1101                         return -EIO;
1102
1103                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1104                 if (list_empty(head)) {
1105                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1106                         return 0;
1107                 }
1108                 fi = list_first_entry(head, struct f2fs_inode_info,
1109                                                         gdirty_list);
1110                 inode = igrab(&fi->vfs_inode);
1111                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1112                 if (inode) {
1113                         sync_inode_metadata(inode, 0);
1114
1115                         /* it's on eviction */
1116                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1117                                 f2fs_update_inode_page(inode);
1118                         iput(inode);
1119                 }
1120         }
1121         return 0;
1122 }
1123
1124 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1125 {
1126         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1127         struct f2fs_nm_info *nm_i = NM_I(sbi);
1128         nid_t last_nid = nm_i->next_scan_nid;
1129
1130         next_free_nid(sbi, &last_nid);
1131         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1132         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1133         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1134         ckpt->next_free_nid = cpu_to_le32(last_nid);
1135 }
1136
1137 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1138 {
1139         bool ret = false;
1140
1141         if (!is_journalled_quota(sbi))
1142                 return false;
1143
1144         down_write(&sbi->quota_sem);
1145         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1146                 ret = false;
1147         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1148                 ret = false;
1149         } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1150                 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1151                 ret = true;
1152         } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1153                 ret = true;
1154         }
1155         up_write(&sbi->quota_sem);
1156         return ret;
1157 }
1158
1159 /*
1160  * Freeze all the FS-operations for checkpoint.
1161  */
1162 static int block_operations(struct f2fs_sb_info *sbi)
1163 {
1164         struct writeback_control wbc = {
1165                 .sync_mode = WB_SYNC_ALL,
1166                 .nr_to_write = LONG_MAX,
1167                 .for_reclaim = 0,
1168         };
1169         struct blk_plug plug;
1170         int err = 0, cnt = 0;
1171
1172         blk_start_plug(&plug);
1173
1174 retry_flush_quotas:
1175         f2fs_lock_all(sbi);
1176         if (__need_flush_quota(sbi)) {
1177                 int locked;
1178
1179                 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1180                         set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1181                         set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1182                         goto retry_flush_dents;
1183                 }
1184                 f2fs_unlock_all(sbi);
1185
1186                 /* only failed during mount/umount/freeze/quotactl */
1187                 locked = down_read_trylock(&sbi->sb->s_umount);
1188                 f2fs_quota_sync(sbi->sb, -1);
1189                 if (locked)
1190                         up_read(&sbi->sb->s_umount);
1191                 cond_resched();
1192                 goto retry_flush_quotas;
1193         }
1194
1195 retry_flush_dents:
1196         /* write all the dirty dentry pages */
1197         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1198                 f2fs_unlock_all(sbi);
1199                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1200                 if (err)
1201                         goto out;
1202                 cond_resched();
1203                 goto retry_flush_quotas;
1204         }
1205
1206         /*
1207          * POR: we should ensure that there are no dirty node pages
1208          * until finishing nat/sit flush. inode->i_blocks can be updated.
1209          */
1210         down_write(&sbi->node_change);
1211
1212         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1213                 up_write(&sbi->node_change);
1214                 f2fs_unlock_all(sbi);
1215                 err = f2fs_sync_inode_meta(sbi);
1216                 if (err)
1217                         goto out;
1218                 cond_resched();
1219                 goto retry_flush_quotas;
1220         }
1221
1222 retry_flush_nodes:
1223         down_write(&sbi->node_write);
1224
1225         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1226                 up_write(&sbi->node_write);
1227                 atomic_inc(&sbi->wb_sync_req[NODE]);
1228                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1229                 atomic_dec(&sbi->wb_sync_req[NODE]);
1230                 if (err) {
1231                         up_write(&sbi->node_change);
1232                         f2fs_unlock_all(sbi);
1233                         goto out;
1234                 }
1235                 cond_resched();
1236                 goto retry_flush_nodes;
1237         }
1238
1239         /*
1240          * sbi->node_change is used only for AIO write_begin path which produces
1241          * dirty node blocks and some checkpoint values by block allocation.
1242          */
1243         __prepare_cp_block(sbi);
1244         up_write(&sbi->node_change);
1245 out:
1246         blk_finish_plug(&plug);
1247         return err;
1248 }
1249
1250 static void unblock_operations(struct f2fs_sb_info *sbi)
1251 {
1252         up_write(&sbi->node_write);
1253         f2fs_unlock_all(sbi);
1254 }
1255
1256 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1257 {
1258         DEFINE_WAIT(wait);
1259
1260         for (;;) {
1261                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1262
1263                 if (!get_pages(sbi, type))
1264                         break;
1265
1266                 if (unlikely(f2fs_cp_error(sbi)))
1267                         break;
1268
1269                 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1270         }
1271         finish_wait(&sbi->cp_wait, &wait);
1272 }
1273
1274 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1275 {
1276         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1277         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1278         unsigned long flags;
1279
1280         spin_lock_irqsave(&sbi->cp_lock, flags);
1281
1282         if ((cpc->reason & CP_UMOUNT) &&
1283                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1284                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1285                 disable_nat_bits(sbi, false);
1286
1287         if (cpc->reason & CP_TRIMMED)
1288                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1289         else
1290                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1291
1292         if (cpc->reason & CP_UMOUNT)
1293                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1294         else
1295                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1296
1297         if (cpc->reason & CP_FASTBOOT)
1298                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1299         else
1300                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1301
1302         if (orphan_num)
1303                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1304         else
1305                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1306
1307         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1308                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1309
1310         if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1311                 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1312         else
1313                 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1314
1315         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1316                 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1317         else
1318                 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1319
1320         if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1321                 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1322         else
1323                 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1324
1325         if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1326                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1327         else
1328                 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1329
1330         if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1331                 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1332
1333         /* set this flag to activate crc|cp_ver for recovery */
1334         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1335         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1336
1337         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1338 }
1339
1340 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1341         void *src, block_t blk_addr)
1342 {
1343         struct writeback_control wbc = {
1344                 .for_reclaim = 0,
1345         };
1346
1347         /*
1348          * pagevec_lookup_tag and lock_page again will take
1349          * some extra time. Therefore, f2fs_update_meta_pages and
1350          * f2fs_sync_meta_pages are combined in this function.
1351          */
1352         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1353         int err;
1354
1355         f2fs_wait_on_page_writeback(page, META, true, true);
1356
1357         memcpy(page_address(page), src, PAGE_SIZE);
1358
1359         set_page_dirty(page);
1360         if (unlikely(!clear_page_dirty_for_io(page)))
1361                 f2fs_bug_on(sbi, 1);
1362
1363         /* writeout cp pack 2 page */
1364         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1365         if (unlikely(err && f2fs_cp_error(sbi))) {
1366                 f2fs_put_page(page, 1);
1367                 return;
1368         }
1369
1370         f2fs_bug_on(sbi, err);
1371         f2fs_put_page(page, 0);
1372
1373         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1374         f2fs_submit_merged_write(sbi, META_FLUSH);
1375 }
1376
1377 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1378 {
1379         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1380         struct f2fs_nm_info *nm_i = NM_I(sbi);
1381         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1382         block_t start_blk;
1383         unsigned int data_sum_blocks, orphan_blocks;
1384         __u32 crc32 = 0;
1385         int i;
1386         int cp_payload_blks = __cp_payload(sbi);
1387         struct super_block *sb = sbi->sb;
1388         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1389         u64 kbytes_written;
1390         int err;
1391
1392         /* Flush all the NAT/SIT pages */
1393         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1394
1395         /* start to update checkpoint, cp ver is already updated previously */
1396         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1397         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1398         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1399                 ckpt->cur_node_segno[i] =
1400                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1401                 ckpt->cur_node_blkoff[i] =
1402                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1403                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1404                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1405         }
1406         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1407                 ckpt->cur_data_segno[i] =
1408                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1409                 ckpt->cur_data_blkoff[i] =
1410                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1411                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1412                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1413         }
1414
1415         /* 2 cp  + n data seg summary + orphan inode blocks */
1416         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1417         spin_lock_irqsave(&sbi->cp_lock, flags);
1418         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1419                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1420         else
1421                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1422         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1423
1424         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1425         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1426                         orphan_blocks);
1427
1428         if (__remain_node_summaries(cpc->reason))
1429                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1430                                 cp_payload_blks + data_sum_blocks +
1431                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1432         else
1433                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1434                                 cp_payload_blks + data_sum_blocks +
1435                                 orphan_blocks);
1436
1437         /* update ckpt flag for checkpoint */
1438         update_ckpt_flags(sbi, cpc);
1439
1440         /* update SIT/NAT bitmap */
1441         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1442         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1443
1444         crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1445         *((__le32 *)((unsigned char *)ckpt +
1446                                 le32_to_cpu(ckpt->checksum_offset)))
1447                                 = cpu_to_le32(crc32);
1448
1449         start_blk = __start_cp_next_addr(sbi);
1450
1451         /* write nat bits */
1452         if (enabled_nat_bits(sbi, cpc)) {
1453                 __u64 cp_ver = cur_cp_version(ckpt);
1454                 block_t blk;
1455
1456                 cp_ver |= ((__u64)crc32 << 32);
1457                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1458
1459                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1460                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1461                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1462                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1463         }
1464
1465         /* write out checkpoint buffer at block 0 */
1466         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1467
1468         for (i = 1; i < 1 + cp_payload_blks; i++)
1469                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1470                                                         start_blk++);
1471
1472         if (orphan_num) {
1473                 write_orphan_inodes(sbi, start_blk);
1474                 start_blk += orphan_blocks;
1475         }
1476
1477         f2fs_write_data_summaries(sbi, start_blk);
1478         start_blk += data_sum_blocks;
1479
1480         /* Record write statistics in the hot node summary */
1481         kbytes_written = sbi->kbytes_written;
1482         if (sb->s_bdev->bd_part)
1483                 kbytes_written += BD_PART_WRITTEN(sbi);
1484
1485         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1486
1487         if (__remain_node_summaries(cpc->reason)) {
1488                 f2fs_write_node_summaries(sbi, start_blk);
1489                 start_blk += NR_CURSEG_NODE_TYPE;
1490         }
1491
1492         /* update user_block_counts */
1493         sbi->last_valid_block_count = sbi->total_valid_block_count;
1494         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1495
1496         /* Here, we have one bio having CP pack except cp pack 2 page */
1497         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1498         /* Wait for all dirty meta pages to be submitted for IO */
1499         f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1500
1501         /* wait for previous submitted meta pages writeback */
1502         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1503
1504         /* flush all device cache */
1505         err = f2fs_flush_device_cache(sbi);
1506         if (err)
1507                 return err;
1508
1509         /* barrier and flush checkpoint cp pack 2 page if it can */
1510         commit_checkpoint(sbi, ckpt, start_blk);
1511         f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1512
1513         /*
1514          * invalidate intermediate page cache borrowed from meta inode which are
1515          * used for migration of encrypted or verity inode's blocks.
1516          */
1517         if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi))
1518                 invalidate_mapping_pages(META_MAPPING(sbi),
1519                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1520
1521         f2fs_release_ino_entry(sbi, false);
1522
1523         f2fs_reset_fsync_node_info(sbi);
1524
1525         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1526         clear_sbi_flag(sbi, SBI_NEED_CP);
1527         clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1528
1529         spin_lock(&sbi->stat_lock);
1530         sbi->unusable_block_count = 0;
1531         spin_unlock(&sbi->stat_lock);
1532
1533         __set_cp_next_pack(sbi);
1534
1535         /*
1536          * redirty superblock if metadata like node page or inode cache is
1537          * updated during writing checkpoint.
1538          */
1539         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1540                         get_pages(sbi, F2FS_DIRTY_IMETA))
1541                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1542
1543         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1544
1545         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1546 }
1547
1548 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1549 {
1550         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1551         unsigned long long ckpt_ver;
1552         int err = 0;
1553
1554         if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1555                 return -EROFS;
1556
1557         if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1558                 if (cpc->reason != CP_PAUSE)
1559                         return 0;
1560                 f2fs_warn(sbi, "Start checkpoint disabled!");
1561         }
1562         mutex_lock(&sbi->cp_mutex);
1563
1564         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1565                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1566                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1567                 goto out;
1568         if (unlikely(f2fs_cp_error(sbi))) {
1569                 err = -EIO;
1570                 goto out;
1571         }
1572
1573         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1574
1575         err = block_operations(sbi);
1576         if (err)
1577                 goto out;
1578
1579         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1580
1581         f2fs_flush_merged_writes(sbi);
1582
1583         /* this is the case of multiple fstrims without any changes */
1584         if (cpc->reason & CP_DISCARD) {
1585                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1586                         unblock_operations(sbi);
1587                         goto out;
1588                 }
1589
1590                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1591                                 SIT_I(sbi)->dirty_sentries == 0 &&
1592                                 prefree_segments(sbi) == 0) {
1593                         f2fs_flush_sit_entries(sbi, cpc);
1594                         f2fs_clear_prefree_segments(sbi, cpc);
1595                         unblock_operations(sbi);
1596                         goto out;
1597                 }
1598         }
1599
1600         /*
1601          * update checkpoint pack index
1602          * Increase the version number so that
1603          * SIT entries and seg summaries are written at correct place
1604          */
1605         ckpt_ver = cur_cp_version(ckpt);
1606         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1607
1608         /* write cached NAT/SIT entries to NAT/SIT area */
1609         err = f2fs_flush_nat_entries(sbi, cpc);
1610         if (err)
1611                 goto stop;
1612
1613         f2fs_flush_sit_entries(sbi, cpc);
1614
1615         err = do_checkpoint(sbi, cpc);
1616         if (err)
1617                 f2fs_release_discard_addrs(sbi);
1618         else
1619                 f2fs_clear_prefree_segments(sbi, cpc);
1620 stop:
1621         unblock_operations(sbi);
1622         stat_inc_cp_count(sbi->stat_info);
1623
1624         if (cpc->reason & CP_RECOVERY)
1625                 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1626
1627         /* update CP_TIME to trigger checkpoint periodically */
1628         f2fs_update_time(sbi, CP_TIME);
1629         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1630 out:
1631         mutex_unlock(&sbi->cp_mutex);
1632         return err;
1633 }
1634
1635 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1636 {
1637         int i;
1638
1639         for (i = 0; i < MAX_INO_ENTRY; i++) {
1640                 struct inode_management *im = &sbi->im[i];
1641
1642                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1643                 spin_lock_init(&im->ino_lock);
1644                 INIT_LIST_HEAD(&im->ino_list);
1645                 im->ino_num = 0;
1646         }
1647
1648         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1649                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1650                                 F2FS_ORPHANS_PER_BLOCK;
1651 }
1652
1653 int __init f2fs_create_checkpoint_caches(void)
1654 {
1655         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1656                         sizeof(struct ino_entry));
1657         if (!ino_entry_slab)
1658                 return -ENOMEM;
1659         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1660                         sizeof(struct inode_entry));
1661         if (!f2fs_inode_entry_slab) {
1662                 kmem_cache_destroy(ino_entry_slab);
1663                 return -ENOMEM;
1664         }
1665         return 0;
1666 }
1667
1668 void f2fs_destroy_checkpoint_caches(void)
1669 {
1670         kmem_cache_destroy(ino_entry_slab);
1671         kmem_cache_destroy(f2fs_inode_entry_slab);
1672 }