fs/f2fs/checkpoint.c

   1 /*
   2  * fs/f2fs/checkpoint.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/bio.h>
  13 #include <linux/mpage.h>
  14 #include <linux/writeback.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/f2fs_fs.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/swap.h>
  19
  20 #include "f2fs.h"
  21 #include "node.h"
  22 #include "segment.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *inode_entry_slab;
  28
  29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  32         sbi->sb->s_flags |= MS_RDONLY;
  33         if (!end_io)
  34                 f2fs_flush_merged_bios(sbi);
  35 }
  36
  37 /*
  38  * We guarantee no failure on the returned page.
  39  */
  40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  41 {
  42         struct address_space *mapping = META_MAPPING(sbi);
  43         struct page *page = NULL;
  44 repeat:
  45         page = f2fs_grab_cache_page(mapping, index, false);
  46         if (!page) {
  47                 cond_resched();
  48                 goto repeat;
  49         }
  50         f2fs_wait_on_page_writeback(page, META, true);
  51         if (!PageUptodate(page))
  52                 SetPageUptodate(page);
  53         return page;
  54 }
  55
  56 /*
  57  * We guarantee no failure on the returned page.
  58  */
  59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  60                                                         bool is_meta)
  61 {
  62         struct address_space *mapping = META_MAPPING(sbi);
  63         struct page *page;
  64         struct f2fs_io_info fio = {
  65                 .sbi = sbi,
  66                 .type = META,
  67                 .op = REQ_OP_READ,
  68                 .op_flags = READ_SYNC | REQ_META | REQ_PRIO,
  69                 .old_blkaddr = index,
  70                 .new_blkaddr = index,
  71                 .encrypted_page = NULL,
  72         };
  73
  74         if (unlikely(!is_meta))
  75                 fio.op_flags &= ~REQ_META;
  76 repeat:
  77         page = f2fs_grab_cache_page(mapping, index, false);
  78         if (!page) {
  79                 cond_resched();
  80                 goto repeat;
  81         }
  82         if (PageUptodate(page))
  83                 goto out;
  84
  85         fio.page = page;
  86
  87         if (f2fs_submit_page_bio(&fio)) {
  88                 f2fs_put_page(page, 1);
  89                 goto repeat;
  90         }
  91
  92         lock_page(page);
  93         if (unlikely(page->mapping != mapping)) {
  94                 f2fs_put_page(page, 1);
  95                 goto repeat;
  96         }
  97
  98         /*
  99          * if there is any IO error when accessing device, make our filesystem
 100          * readonly and make sure do not write checkpoint with non-uptodate
 101          * meta page.
 102          */
 103         if (unlikely(!PageUptodate(page)))
 104                 f2fs_stop_checkpoint(sbi, false);
 105 out:
 106         return page;
 107 }
 108
 109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 110 {
 111         return __get_meta_page(sbi, index, true);
 112 }
 113
 114 /* for POR only */
 115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 116 {
 117         return __get_meta_page(sbi, index, false);
 118 }
 119
 120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 121 {
 122         switch (type) {
 123         case META_NAT:
 124                 break;
 125         case META_SIT:
 126                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 127                         return false;
 128                 break;
 129         case META_SSA:
 130                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 131                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 132                         return false;
 133                 break;
 134         case META_CP:
 135                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 136                         blkaddr < __start_cp_addr(sbi)))
 137                         return false;
 138                 break;
 139         case META_POR:
 140                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 141                         blkaddr < MAIN_BLKADDR(sbi)))
 142                         return false;
 143                 break;
 144         default:
 145                 BUG();
 146         }
 147
 148         return true;
 149 }
 150
 151 /*
 152  * Readahead CP/NAT/SIT/SSA pages
 153  */
 154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 155                                                         int type, bool sync)
 156 {
 157         struct page *page;
 158         block_t blkno = start;
 159         struct f2fs_io_info fio = {
 160                 .sbi = sbi,
 161                 .type = META,
 162                 .op = REQ_OP_READ,
 163                 .op_flags = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : REQ_RAHEAD,
 164                 .encrypted_page = NULL,
 165         };
 166         struct blk_plug plug;
 167
 168         if (unlikely(type == META_POR))
 169                 fio.op_flags &= ~REQ_META;
 170
 171         blk_start_plug(&plug);
 172         for (; nrpages-- > 0; blkno++) {
 173
 174                 if (!is_valid_blkaddr(sbi, blkno, type))
 175                         goto out;
 176
 177                 switch (type) {
 178                 case META_NAT:
 179                         if (unlikely(blkno >=
 180                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 181                                 blkno = 0;
 182                         /* get nat block addr */
 183                         fio.new_blkaddr = current_nat_addr(sbi,
 184                                         blkno * NAT_ENTRY_PER_BLOCK);
 185                         break;
 186                 case META_SIT:
 187                         /* get sit block addr */
 188                         fio.new_blkaddr = current_sit_addr(sbi,
 189                                         blkno * SIT_ENTRY_PER_BLOCK);
 190                         break;
 191                 case META_SSA:
 192                 case META_CP:
 193                 case META_POR:
 194                         fio.new_blkaddr = blkno;
 195                         break;
 196                 default:
 197                         BUG();
 198                 }
 199
 200                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 201                                                 fio.new_blkaddr, false);
 202                 if (!page)
 203                         continue;
 204                 if (PageUptodate(page)) {
 205                         f2fs_put_page(page, 1);
 206                         continue;
 207                 }
 208
 209                 fio.page = page;
 210                 fio.old_blkaddr = fio.new_blkaddr;
 211                 f2fs_submit_page_mbio(&fio);
 212                 f2fs_put_page(page, 0);
 213         }
 214 out:
 215         f2fs_submit_merged_bio(sbi, META, READ);
 216         blk_finish_plug(&plug);
 217         return blkno - start;
 218 }
 219
 220 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 221 {
 222         struct page *page;
 223         bool readahead = false;
 224
 225         page = find_get_page(META_MAPPING(sbi), index);
 226         if (!page || !PageUptodate(page))
 227                 readahead = true;
 228         f2fs_put_page(page, 0);
 229
 230         if (readahead)
 231                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 232 }
 233
 234 static int f2fs_write_meta_page(struct page *page,
 235                                 struct writeback_control *wbc)
 236 {
 237         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 238
 239         trace_f2fs_writepage(page, META);
 240
 241         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 242                 goto redirty_out;
 243         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 244                 goto redirty_out;
 245         if (unlikely(f2fs_cp_error(sbi)))
 246                 goto redirty_out;
 247
 248         write_meta_page(sbi, page);
 249         dec_page_count(sbi, F2FS_DIRTY_META);
 250
 251         if (wbc->for_reclaim)
 252                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
 253
 254         unlock_page(page);
 255
 256         if (unlikely(f2fs_cp_error(sbi)))
 257                 f2fs_submit_merged_bio(sbi, META, WRITE);
 258
 259         return 0;
 260
 261 redirty_out:
 262         redirty_page_for_writepage(wbc, page);
 263         return AOP_WRITEPAGE_ACTIVATE;
 264 }
 265
 266 static int f2fs_write_meta_pages(struct address_space *mapping,
 267                                 struct writeback_control *wbc)
 268 {
 269         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 270         struct blk_plug plug;
 271         long diff, written;
 272
 273         /* collect a number of dirty meta pages and write together */
 274         if (wbc->for_kupdate ||
 275                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 276                 goto skip_write;
 277
 278         trace_f2fs_writepages(mapping->host, wbc, META);
 279
 280         /* if mounting is failed, skip writing node pages */
 281         mutex_lock(&sbi->cp_mutex);
 282         diff = nr_pages_to_write(sbi, META, wbc);
 283         blk_start_plug(&plug);
 284         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 285         blk_finish_plug(&plug);
 286         mutex_unlock(&sbi->cp_mutex);
 287         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 288         return 0;
 289
 290 skip_write:
 291         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 292         trace_f2fs_writepages(mapping->host, wbc, META);
 293         return 0;
 294 }
 295
 296 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 297                                                 long nr_to_write)
 298 {
 299         struct address_space *mapping = META_MAPPING(sbi);
 300         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 301         struct pagevec pvec;
 302         long nwritten = 0;
 303         struct writeback_control wbc = {
 304                 .for_reclaim = 0,
 305         };
 306         struct blk_plug plug;
 307
 308         pagevec_init(&pvec, 0);
 309
 310         blk_start_plug(&plug);
 311
 312         while (index <= end) {
 313                 int i, nr_pages;
 314                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 315                                 PAGECACHE_TAG_DIRTY,
 316                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 317                 if (unlikely(nr_pages == 0))
 318                         break;
 319
 320                 for (i = 0; i < nr_pages; i++) {
 321                         struct page *page = pvec.pages[i];
 322
 323                         if (prev == ULONG_MAX)
 324                                 prev = page->index - 1;
 325                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 326                                 pagevec_release(&pvec);
 327                                 goto stop;
 328                         }
 329
 330                         lock_page(page);
 331
 332                         if (unlikely(page->mapping != mapping)) {
 333 continue_unlock:
 334                                 unlock_page(page);
 335                                 continue;
 336                         }
 337                         if (!PageDirty(page)) {
 338                                 /* someone wrote it for us */
 339                                 goto continue_unlock;
 340                         }
 341
 342                         f2fs_wait_on_page_writeback(page, META, true);
 343
 344                         BUG_ON(PageWriteback(page));
 345                         if (!clear_page_dirty_for_io(page))
 346                                 goto continue_unlock;
 347
 348                         if (mapping->a_ops->writepage(page, &wbc)) {
 349                                 unlock_page(page);
 350                                 break;
 351                         }
 352                         nwritten++;
 353                         prev = page->index;
 354                         if (unlikely(nwritten >= nr_to_write))
 355                                 break;
 356                 }
 357                 pagevec_release(&pvec);
 358                 cond_resched();
 359         }
 360 stop:
 361         if (nwritten)
 362                 f2fs_submit_merged_bio(sbi, type, WRITE);
 363
 364         blk_finish_plug(&plug);
 365
 366         return nwritten;
 367 }
 368
 369 static int f2fs_set_meta_page_dirty(struct page *page)
 370 {
 371         trace_f2fs_set_page_dirty(page, META);
 372
 373         if (!PageUptodate(page))
 374                 SetPageUptodate(page);
 375         if (!PageDirty(page)) {
 376                 f2fs_set_page_dirty_nobuffers(page);
 377                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 378                 SetPagePrivate(page);
 379                 f2fs_trace_pid(page);
 380                 return 1;
 381         }
 382         return 0;
 383 }
 384
 385 const struct address_space_operations f2fs_meta_aops = {
 386         .writepage      = f2fs_write_meta_page,
 387         .writepages     = f2fs_write_meta_pages,
 388         .set_page_dirty = f2fs_set_meta_page_dirty,
 389         .invalidatepage = f2fs_invalidate_page,
 390         .releasepage    = f2fs_release_page,
 391 };
 392
 393 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 394 {
 395         struct inode_management *im = &sbi->im[type];
 396         struct ino_entry *e, *tmp;
 397
 398         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 399 retry:
 400         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 401
 402         spin_lock(&im->ino_lock);
 403         e = radix_tree_lookup(&im->ino_root, ino);
 404         if (!e) {
 405                 e = tmp;
 406                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 407                         spin_unlock(&im->ino_lock);
 408                         radix_tree_preload_end();
 409                         goto retry;
 410                 }
 411                 memset(e, 0, sizeof(struct ino_entry));
 412                 e->ino = ino;
 413
 414                 list_add_tail(&e->list, &im->ino_list);
 415                 if (type != ORPHAN_INO)
 416                         im->ino_num++;
 417         }
 418         spin_unlock(&im->ino_lock);
 419         radix_tree_preload_end();
 420
 421         if (e != tmp)
 422                 kmem_cache_free(ino_entry_slab, tmp);
 423 }
 424
 425 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 426 {
 427         struct inode_management *im = &sbi->im[type];
 428         struct ino_entry *e;
 429
 430         spin_lock(&im->ino_lock);
 431         e = radix_tree_lookup(&im->ino_root, ino);
 432         if (e) {
 433                 list_del(&e->list);
 434                 radix_tree_delete(&im->ino_root, ino);
 435                 im->ino_num--;
 436                 spin_unlock(&im->ino_lock);
 437                 kmem_cache_free(ino_entry_slab, e);
 438                 return;
 439         }
 440         spin_unlock(&im->ino_lock);
 441 }
 442
 443 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 444 {
 445         /* add new dirty ino entry into list */
 446         __add_ino_entry(sbi, ino, type);
 447 }
 448
 449 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 450 {
 451         /* remove dirty ino entry from list */
 452         __remove_ino_entry(sbi, ino, type);
 453 }
 454
 455 /* mode should be APPEND_INO or UPDATE_INO */
 456 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 457 {
 458         struct inode_management *im = &sbi->im[mode];
 459         struct ino_entry *e;
 460
 461         spin_lock(&im->ino_lock);
 462         e = radix_tree_lookup(&im->ino_root, ino);
 463         spin_unlock(&im->ino_lock);
 464         return e ? true : false;
 465 }
 466
 467 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 468 {
 469         struct ino_entry *e, *tmp;
 470         int i;
 471
 472         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 473                 struct inode_management *im = &sbi->im[i];
 474
 475                 spin_lock(&im->ino_lock);
 476                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 477                         list_del(&e->list);
 478                         radix_tree_delete(&im->ino_root, e->ino);
 479                         kmem_cache_free(ino_entry_slab, e);
 480                         im->ino_num--;
 481                 }
 482                 spin_unlock(&im->ino_lock);
 483         }
 484 }
 485
 486 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 487 {
 488         struct inode_management *im = &sbi->im[ORPHAN_INO];
 489         int err = 0;
 490
 491         spin_lock(&im->ino_lock);
 492
 493 #ifdef CONFIG_F2FS_FAULT_INJECTION
 494         if (time_to_inject(FAULT_ORPHAN)) {
 495                 spin_unlock(&im->ino_lock);
 496                 return -ENOSPC;
 497         }
 498 #endif
 499         if (unlikely(im->ino_num >= sbi->max_orphans))
 500                 err = -ENOSPC;
 501         else
 502                 im->ino_num++;
 503         spin_unlock(&im->ino_lock);
 504
 505         return err;
 506 }
 507
 508 void release_orphan_inode(struct f2fs_sb_info *sbi)
 509 {
 510         struct inode_management *im = &sbi->im[ORPHAN_INO];
 511
 512         spin_lock(&im->ino_lock);
 513         f2fs_bug_on(sbi, im->ino_num == 0);
 514         im->ino_num--;
 515         spin_unlock(&im->ino_lock);
 516 }
 517
 518 void add_orphan_inode(struct inode *inode)
 519 {
 520         /* add new orphan ino entry into list */
 521         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
 522         update_inode_page(inode);
 523 }
 524
 525 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 526 {
 527         /* remove orphan entry from orphan list */
 528         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 529 }
 530
 531 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 532 {
 533         struct inode *inode;
 534         struct node_info ni;
 535
 536         inode = f2fs_iget_retry(sbi->sb, ino);
 537         if (IS_ERR(inode)) {
 538                 /*
 539                  * there should be a bug that we can't find the entry
 540                  * to orphan inode.
 541                  */
 542                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 543                 return PTR_ERR(inode);
 544         }
 545
 546         clear_nlink(inode);
 547
 548         /* truncate all the data during iput */
 549         iput(inode);
 550
 551         get_node_info(sbi, ino, &ni);
 552
 553         /* ENOMEM was fully retried in f2fs_evict_inode. */
 554         if (ni.blk_addr != NULL_ADDR) {
 555                 int err = acquire_orphan_inode(sbi);
 556
 557                 if (err) {
 558                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 559                         f2fs_msg(sbi->sb, KERN_WARNING,
 560                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 561                                         __func__, ino);
 562                         return err;
 563                 }
 564                 __add_ino_entry(sbi, ino, ORPHAN_INO);
 565         }
 566         return 0;
 567 }
 568
 569 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 570 {
 571         block_t start_blk, orphan_blocks, i, j;
 572         int err;
 573
 574         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 575                 return 0;
 576
 577         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 578         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 579
 580         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 581
 582         for (i = 0; i < orphan_blocks; i++) {
 583                 struct page *page = get_meta_page(sbi, start_blk + i);
 584                 struct f2fs_orphan_block *orphan_blk;
 585
 586                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 587                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 588                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 589                         err = recover_orphan_inode(sbi, ino);
 590                         if (err) {
 591                                 f2fs_put_page(page, 1);
 592                                 return err;
 593                         }
 594                 }
 595                 f2fs_put_page(page, 1);
 596         }
 597         /* clear Orphan Flag */
 598         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 599         return 0;
 600 }
 601
 602 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 603 {
 604         struct list_head *head;
 605         struct f2fs_orphan_block *orphan_blk = NULL;
 606         unsigned int nentries = 0;
 607         unsigned short index = 1;
 608         unsigned short orphan_blocks;
 609         struct page *page = NULL;
 610         struct ino_entry *orphan = NULL;
 611         struct inode_management *im = &sbi->im[ORPHAN_INO];
 612
 613         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 614
 615         /*
 616          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 617          * orphan inode operations are covered under f2fs_lock_op().
 618          * And, spin_lock should be avoided due to page operations below.
 619          */
 620         head = &im->ino_list;
 621
 622         /* loop for each orphan inode entry and write them in Jornal block */
 623         list_for_each_entry(orphan, head, list) {
 624                 if (!page) {
 625                         page = grab_meta_page(sbi, start_blk++);
 626                         orphan_blk =
 627                                 (struct f2fs_orphan_block *)page_address(page);
 628                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 629                 }
 630
 631                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 632
 633                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 634                         /*
 635                          * an orphan block is full of 1020 entries,
 636                          * then we need to flush current orphan blocks
 637                          * and bring another one in memory
 638                          */
 639                         orphan_blk->blk_addr = cpu_to_le16(index);
 640                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 641                         orphan_blk->entry_count = cpu_to_le32(nentries);
 642                         set_page_dirty(page);
 643                         f2fs_put_page(page, 1);
 644                         index++;
 645                         nentries = 0;
 646                         page = NULL;
 647                 }
 648         }
 649
 650         if (page) {
 651                 orphan_blk->blk_addr = cpu_to_le16(index);
 652                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 653                 orphan_blk->entry_count = cpu_to_le32(nentries);
 654                 set_page_dirty(page);
 655                 f2fs_put_page(page, 1);
 656         }
 657 }
 658
 659 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 660                                 block_t cp_addr, unsigned long long *version)
 661 {
 662         struct page *cp_page_1, *cp_page_2 = NULL;
 663         unsigned long blk_size = sbi->blocksize;
 664         struct f2fs_checkpoint *cp_block;
 665         unsigned long long cur_version = 0, pre_version = 0;
 666         size_t crc_offset;
 667         __u32 crc = 0;
 668
 669         /* Read the 1st cp block in this CP pack */
 670         cp_page_1 = get_meta_page(sbi, cp_addr);
 671
 672         /* get the version number */
 673         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 674         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 675         if (crc_offset >= blk_size)
 676                 goto invalid_cp1;
 677
 678         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 679         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 680                 goto invalid_cp1;
 681
 682         pre_version = cur_cp_version(cp_block);
 683
 684         /* Read the 2nd cp block in this CP pack */
 685         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 686         cp_page_2 = get_meta_page(sbi, cp_addr);
 687
 688         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 689         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 690         if (crc_offset >= blk_size)
 691                 goto invalid_cp2;
 692
 693         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 694         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 695                 goto invalid_cp2;
 696
 697         cur_version = cur_cp_version(cp_block);
 698
 699         if (cur_version == pre_version) {
 700                 *version = cur_version;
 701                 f2fs_put_page(cp_page_2, 1);
 702                 return cp_page_1;
 703         }
 704 invalid_cp2:
 705         f2fs_put_page(cp_page_2, 1);
 706 invalid_cp1:
 707         f2fs_put_page(cp_page_1, 1);
 708         return NULL;
 709 }
 710
 711 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 712 {
 713         struct f2fs_checkpoint *cp_block;
 714         struct f2fs_super_block *fsb = sbi->raw_super;
 715         struct page *cp1, *cp2, *cur_page;
 716         unsigned long blk_size = sbi->blocksize;
 717         unsigned long long cp1_version = 0, cp2_version = 0;
 718         unsigned long long cp_start_blk_no;
 719         unsigned int cp_blks = 1 + __cp_payload(sbi);
 720         block_t cp_blk_no;
 721         int i;
 722
 723         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 724         if (!sbi->ckpt)
 725                 return -ENOMEM;
 726         /*
 727          * Finding out valid cp block involves read both
 728          * sets( cp pack1 and cp pack 2)
 729          */
 730         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 731         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 732
 733         /* The second checkpoint pack should start at the next segment */
 734         cp_start_blk_no += ((unsigned long long)1) <<
 735                                 le32_to_cpu(fsb->log_blocks_per_seg);
 736         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 737
 738         if (cp1 && cp2) {
 739                 if (ver_after(cp2_version, cp1_version))
 740                         cur_page = cp2;
 741                 else
 742                         cur_page = cp1;
 743         } else if (cp1) {
 744                 cur_page = cp1;
 745         } else if (cp2) {
 746                 cur_page = cp2;
 747         } else {
 748                 goto fail_no_cp;
 749         }
 750
 751         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 752         memcpy(sbi->ckpt, cp_block, blk_size);
 753
 754         /* Sanity checking of checkpoint */
 755         if (sanity_check_ckpt(sbi))
 756                 goto fail_no_cp;
 757
 758         if (cp_blks <= 1)
 759                 goto done;
 760
 761         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 762         if (cur_page == cp2)
 763                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 764
 765         for (i = 1; i < cp_blks; i++) {
 766                 void *sit_bitmap_ptr;
 767                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 768
 769                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 770                 sit_bitmap_ptr = page_address(cur_page);
 771                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 772                 f2fs_put_page(cur_page, 1);
 773         }
 774 done:
 775         f2fs_put_page(cp1, 1);
 776         f2fs_put_page(cp2, 1);
 777         return 0;
 778
 779 fail_no_cp:
 780         kfree(sbi->ckpt);
 781         return -EINVAL;
 782 }
 783
 784 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 785 {
 786         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 787         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 788
 789         if (is_inode_flag_set(inode, flag))
 790                 return;
 791
 792         set_inode_flag(inode, flag);
 793         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
 794         stat_inc_dirty_inode(sbi, type);
 795 }
 796
 797 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 798 {
 799         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 800
 801         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 802                 return;
 803
 804         list_del_init(&F2FS_I(inode)->dirty_list);
 805         clear_inode_flag(inode, flag);
 806         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 807 }
 808
 809 void update_dirty_page(struct inode *inode, struct page *page)
 810 {
 811         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 812         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 813
 814         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 815                         !S_ISLNK(inode->i_mode))
 816                 return;
 817
 818         spin_lock(&sbi->inode_lock[type]);
 819         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 820                 __add_dirty_inode(inode, type);
 821         inode_inc_dirty_pages(inode);
 822         spin_unlock(&sbi->inode_lock[type]);
 823
 824         SetPagePrivate(page);
 825         f2fs_trace_pid(page);
 826 }
 827
 828 void remove_dirty_inode(struct inode *inode)
 829 {
 830         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 831         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 832
 833         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 834                         !S_ISLNK(inode->i_mode))
 835                 return;
 836
 837         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 838                 return;
 839
 840         spin_lock(&sbi->inode_lock[type]);
 841         __remove_dirty_inode(inode, type);
 842         spin_unlock(&sbi->inode_lock[type]);
 843 }
 844
 845 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 846 {
 847         struct list_head *head;
 848         struct inode *inode;
 849         struct f2fs_inode_info *fi;
 850         bool is_dir = (type == DIR_INODE);
 851
 852         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 853                                 get_pages(sbi, is_dir ?
 854                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 855 retry:
 856         if (unlikely(f2fs_cp_error(sbi)))
 857                 return -EIO;
 858
 859         spin_lock(&sbi->inode_lock[type]);
 860
 861         head = &sbi->inode_list[type];
 862         if (list_empty(head)) {
 863                 spin_unlock(&sbi->inode_lock[type]);
 864                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 865                                 get_pages(sbi, is_dir ?
 866                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 867                 return 0;
 868         }
 869         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 870         inode = igrab(&fi->vfs_inode);
 871         spin_unlock(&sbi->inode_lock[type]);
 872         if (inode) {
 873                 filemap_fdatawrite(inode->i_mapping);
 874                 iput(inode);
 875         } else {
 876                 /*
 877                  * We should submit bio, since it exists several
 878                  * wribacking dentry pages in the freeing inode.
 879                  */
 880                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 881                 cond_resched();
 882         }
 883         goto retry;
 884 }
 885
 886 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 887 {
 888         struct list_head *head = &sbi->inode_list[DIRTY_META];
 889         struct inode *inode;
 890         struct f2fs_inode_info *fi;
 891         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 892
 893         while (total--) {
 894                 if (unlikely(f2fs_cp_error(sbi)))
 895                         return -EIO;
 896
 897                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 898                 if (list_empty(head)) {
 899                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 900                         return 0;
 901                 }
 902                 fi = list_entry(head->next, struct f2fs_inode_info,
 903                                                         gdirty_list);
 904                 inode = igrab(&fi->vfs_inode);
 905                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 906                 if (inode) {
 907                         update_inode_page(inode);
 908                         iput(inode);
 909                 }
 910         };
 911         return 0;
 912 }
 913
 914 /*
 915  * Freeze all the FS-operations for checkpoint.
 916  */
 917 static int block_operations(struct f2fs_sb_info *sbi)
 918 {
 919         struct writeback_control wbc = {
 920                 .sync_mode = WB_SYNC_ALL,
 921                 .nr_to_write = LONG_MAX,
 922                 .for_reclaim = 0,
 923         };
 924         struct blk_plug plug;
 925         int err = 0;
 926
 927         blk_start_plug(&plug);
 928
 929 retry_flush_dents:
 930         f2fs_lock_all(sbi);
 931         /* write all the dirty dentry pages */
 932         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 933                 f2fs_unlock_all(sbi);
 934                 err = sync_dirty_inodes(sbi, DIR_INODE);
 935                 if (err)
 936                         goto out;
 937                 goto retry_flush_dents;
 938         }
 939
 940         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
 941                 f2fs_unlock_all(sbi);
 942                 err = f2fs_sync_inode_meta(sbi);
 943                 if (err)
 944                         goto out;
 945                 goto retry_flush_dents;
 946         }
 947
 948         /*
 949          * POR: we should ensure that there are no dirty node pages
 950          * until finishing nat/sit flush.
 951          */
 952 retry_flush_nodes:
 953         down_write(&sbi->node_write);
 954
 955         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 956                 up_write(&sbi->node_write);
 957                 err = sync_node_pages(sbi, &wbc);
 958                 if (err) {
 959                         f2fs_unlock_all(sbi);
 960                         goto out;
 961                 }
 962                 goto retry_flush_nodes;
 963         }
 964 out:
 965         blk_finish_plug(&plug);
 966         return err;
 967 }
 968
 969 static void unblock_operations(struct f2fs_sb_info *sbi)
 970 {
 971         up_write(&sbi->node_write);
 972
 973         build_free_nids(sbi);
 974         f2fs_unlock_all(sbi);
 975 }
 976
 977 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 978 {
 979         DEFINE_WAIT(wait);
 980
 981         for (;;) {
 982                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 983
 984                 if (!atomic_read(&sbi->nr_wb_bios))
 985                         break;
 986
 987                 io_schedule_timeout(5*HZ);
 988         }
 989         finish_wait(&sbi->cp_wait, &wait);
 990 }
 991
 992 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 993 {
 994         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 995         struct f2fs_nm_info *nm_i = NM_I(sbi);
 996         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 997         nid_t last_nid = nm_i->next_scan_nid;
 998         block_t start_blk;
 999         unsigned int data_sum_blocks, orphan_blocks;
1000         __u32 crc32 = 0;
1001         int i;
1002         int cp_payload_blks = __cp_payload(sbi);
1003         struct super_block *sb = sbi->sb;
1004         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1005         u64 kbytes_written;
1006
1007         /* Flush all the NAT/SIT pages */
1008         while (get_pages(sbi, F2FS_DIRTY_META)) {
1009                 sync_meta_pages(sbi, META, LONG_MAX);
1010                 if (unlikely(f2fs_cp_error(sbi)))
1011                         return -EIO;
1012         }
1013
1014         next_free_nid(sbi, &last_nid);
1015
1016         /*
1017          * modify checkpoint
1018          * version number is already updated
1019          */
1020         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1021         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1022         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1023         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1024                 ckpt->cur_node_segno[i] =
1025                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1026                 ckpt->cur_node_blkoff[i] =
1027                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1028                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1029                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1030         }
1031         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1032                 ckpt->cur_data_segno[i] =
1033                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1034                 ckpt->cur_data_blkoff[i] =
1035                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1036                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1037                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1038         }
1039
1040         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1041         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1042         ckpt->next_free_nid = cpu_to_le32(last_nid);
1043
1044         /* 2 cp  + n data seg summary + orphan inode blocks */
1045         data_sum_blocks = npages_for_summary_flush(sbi, false);
1046         spin_lock(&sbi->cp_lock);
1047         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1048                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1049         else
1050                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1051         spin_unlock(&sbi->cp_lock);
1052
1053         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1054         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1055                         orphan_blocks);
1056
1057         if (__remain_node_summaries(cpc->reason))
1058                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1059                                 cp_payload_blks + data_sum_blocks +
1060                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1061         else
1062                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1063                                 cp_payload_blks + data_sum_blocks +
1064                                 orphan_blocks);
1065
1066         spin_lock(&sbi->cp_lock);
1067         if (cpc->reason == CP_UMOUNT)
1068                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1069         else
1070                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1071
1072         if (cpc->reason == CP_FASTBOOT)
1073                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1074         else
1075                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1076
1077         if (orphan_num)
1078                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1079         else
1080                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1081
1082         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1083                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1084
1085         /* set this flag to activate crc|cp_ver for recovery */
1086         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1087
1088         spin_unlock(&sbi->cp_lock);
1089
1090         /* update SIT/NAT bitmap */
1091         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1092         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1093
1094         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1095         *((__le32 *)((unsigned char *)ckpt +
1096                                 le32_to_cpu(ckpt->checksum_offset)))
1097                                 = cpu_to_le32(crc32);
1098
1099         start_blk = __start_cp_addr(sbi);
1100
1101         /* need to wait for end_io results */
1102         wait_on_all_pages_writeback(sbi);
1103         if (unlikely(f2fs_cp_error(sbi)))
1104                 return -EIO;
1105
1106         /* write out checkpoint buffer at block 0 */
1107         update_meta_page(sbi, ckpt, start_blk++);
1108
1109         for (i = 1; i < 1 + cp_payload_blks; i++)
1110                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1111                                                         start_blk++);
1112
1113         if (orphan_num) {
1114                 write_orphan_inodes(sbi, start_blk);
1115                 start_blk += orphan_blocks;
1116         }
1117
1118         write_data_summaries(sbi, start_blk);
1119         start_blk += data_sum_blocks;
1120
1121         /* Record write statistics in the hot node summary */
1122         kbytes_written = sbi->kbytes_written;
1123         if (sb->s_bdev->bd_part)
1124                 kbytes_written += BD_PART_WRITTEN(sbi);
1125
1126         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1127
1128         if (__remain_node_summaries(cpc->reason)) {
1129                 write_node_summaries(sbi, start_blk);
1130                 start_blk += NR_CURSEG_NODE_TYPE;
1131         }
1132
1133         /* writeout checkpoint block */
1134         update_meta_page(sbi, ckpt, start_blk);
1135
1136         /* wait for previous submitted node/meta pages writeback */
1137         wait_on_all_pages_writeback(sbi);
1138
1139         if (unlikely(f2fs_cp_error(sbi)))
1140                 return -EIO;
1141
1142         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1143         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1144
1145         /* update user_block_counts */
1146         sbi->last_valid_block_count = sbi->total_valid_block_count;
1147         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1148
1149         /* Here, we only have one bio having CP pack */
1150         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1151
1152         /* wait for previous submitted meta pages writeback */
1153         wait_on_all_pages_writeback(sbi);
1154
1155         release_ino_entry(sbi, false);
1156
1157         if (unlikely(f2fs_cp_error(sbi)))
1158                 return -EIO;
1159
1160         clear_prefree_segments(sbi, cpc);
1161         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1162         clear_sbi_flag(sbi, SBI_NEED_CP);
1163
1164         /*
1165          * redirty superblock if metadata like node page or inode cache is
1166          * updated during writing checkpoint.
1167          */
1168         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1169                         get_pages(sbi, F2FS_DIRTY_IMETA))
1170                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1171
1172         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1173
1174         return 0;
1175 }
1176
1177 /*
1178  * We guarantee that this checkpoint procedure will not fail.
1179  */
1180 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1181 {
1182         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1183         unsigned long long ckpt_ver;
1184         int err = 0;
1185
1186         mutex_lock(&sbi->cp_mutex);
1187
1188         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1189                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1190                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1191                 goto out;
1192         if (unlikely(f2fs_cp_error(sbi))) {
1193                 err = -EIO;
1194                 goto out;
1195         }
1196         if (f2fs_readonly(sbi->sb)) {
1197                 err = -EROFS;
1198                 goto out;
1199         }
1200
1201         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1202
1203         err = block_operations(sbi);
1204         if (err)
1205                 goto out;
1206
1207         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1208
1209         f2fs_flush_merged_bios(sbi);
1210
1211         /* this is the case of multiple fstrims without any changes */
1212         if (cpc->reason == CP_DISCARD && !is_sbi_flag_set(sbi, SBI_IS_DIRTY)) {
1213                 f2fs_bug_on(sbi, NM_I(sbi)->dirty_nat_cnt);
1214                 f2fs_bug_on(sbi, SIT_I(sbi)->dirty_sentries);
1215                 f2fs_bug_on(sbi, prefree_segments(sbi));
1216                 flush_sit_entries(sbi, cpc);
1217                 clear_prefree_segments(sbi, cpc);
1218                 f2fs_wait_all_discard_bio(sbi);
1219                 unblock_operations(sbi);
1220                 goto out;
1221         }
1222
1223         /*
1224          * update checkpoint pack index
1225          * Increase the version number so that
1226          * SIT entries and seg summaries are written at correct place
1227          */
1228         ckpt_ver = cur_cp_version(ckpt);
1229         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1230
1231         /* write cached NAT/SIT entries to NAT/SIT area */
1232         flush_nat_entries(sbi);
1233         flush_sit_entries(sbi, cpc);
1234
1235         /* unlock all the fs_lock[] in do_checkpoint() */
1236         err = do_checkpoint(sbi, cpc);
1237
1238         f2fs_wait_all_discard_bio(sbi);
1239
1240         unblock_operations(sbi);
1241         stat_inc_cp_count(sbi->stat_info);
1242
1243         if (cpc->reason == CP_RECOVERY)
1244                 f2fs_msg(sbi->sb, KERN_NOTICE,
1245                         "checkpoint: version = %llx", ckpt_ver);
1246
1247         /* do checkpoint periodically */
1248         f2fs_update_time(sbi, CP_TIME);
1249         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1250 out:
1251         mutex_unlock(&sbi->cp_mutex);
1252         return err;
1253 }
1254
1255 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1256 {
1257         int i;
1258
1259         for (i = 0; i < MAX_INO_ENTRY; i++) {
1260                 struct inode_management *im = &sbi->im[i];
1261
1262                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1263                 spin_lock_init(&im->ino_lock);
1264                 INIT_LIST_HEAD(&im->ino_list);
1265                 im->ino_num = 0;
1266         }
1267
1268         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1269                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1270                                 F2FS_ORPHANS_PER_BLOCK;
1271 }
1272
1273 int __init create_checkpoint_caches(void)
1274 {
1275         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1276                         sizeof(struct ino_entry));
1277         if (!ino_entry_slab)
1278                 return -ENOMEM;
1279         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1280                         sizeof(struct inode_entry));
1281         if (!inode_entry_slab) {
1282                 kmem_cache_destroy(ino_entry_slab);
1283                 return -ENOMEM;
1284         }
1285         return 0;
1286 }
1287
1288 void destroy_checkpoint_caches(void)
1289 {
1290         kmem_cache_destroy(ino_entry_slab);
1291         kmem_cache_destroy(inode_entry_slab);
1292 }