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 #ifdef CONFIG_MIGRATION
 392         .migratepage    = f2fs_migrate_page,
 393 #endif
 394 };
 395
 396 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 397 {
 398         struct inode_management *im = &sbi->im[type];
 399         struct ino_entry *e, *tmp;
 400
 401         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 402 retry:
 403         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 404
 405         spin_lock(&im->ino_lock);
 406         e = radix_tree_lookup(&im->ino_root, ino);
 407         if (!e) {
 408                 e = tmp;
 409                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 410                         spin_unlock(&im->ino_lock);
 411                         radix_tree_preload_end();
 412                         goto retry;
 413                 }
 414                 memset(e, 0, sizeof(struct ino_entry));
 415                 e->ino = ino;
 416
 417                 list_add_tail(&e->list, &im->ino_list);
 418                 if (type != ORPHAN_INO)
 419                         im->ino_num++;
 420         }
 421         spin_unlock(&im->ino_lock);
 422         radix_tree_preload_end();
 423
 424         if (e != tmp)
 425                 kmem_cache_free(ino_entry_slab, tmp);
 426 }
 427
 428 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 429 {
 430         struct inode_management *im = &sbi->im[type];
 431         struct ino_entry *e;
 432
 433         spin_lock(&im->ino_lock);
 434         e = radix_tree_lookup(&im->ino_root, ino);
 435         if (e) {
 436                 list_del(&e->list);
 437                 radix_tree_delete(&im->ino_root, ino);
 438                 im->ino_num--;
 439                 spin_unlock(&im->ino_lock);
 440                 kmem_cache_free(ino_entry_slab, e);
 441                 return;
 442         }
 443         spin_unlock(&im->ino_lock);
 444 }
 445
 446 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 447 {
 448         /* add new dirty ino entry into list */
 449         __add_ino_entry(sbi, ino, type);
 450 }
 451
 452 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 453 {
 454         /* remove dirty ino entry from list */
 455         __remove_ino_entry(sbi, ino, type);
 456 }
 457
 458 /* mode should be APPEND_INO or UPDATE_INO */
 459 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 460 {
 461         struct inode_management *im = &sbi->im[mode];
 462         struct ino_entry *e;
 463
 464         spin_lock(&im->ino_lock);
 465         e = radix_tree_lookup(&im->ino_root, ino);
 466         spin_unlock(&im->ino_lock);
 467         return e ? true : false;
 468 }
 469
 470 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 471 {
 472         struct ino_entry *e, *tmp;
 473         int i;
 474
 475         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 476                 struct inode_management *im = &sbi->im[i];
 477
 478                 spin_lock(&im->ino_lock);
 479                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 480                         list_del(&e->list);
 481                         radix_tree_delete(&im->ino_root, e->ino);
 482                         kmem_cache_free(ino_entry_slab, e);
 483                         im->ino_num--;
 484                 }
 485                 spin_unlock(&im->ino_lock);
 486         }
 487 }
 488
 489 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 490 {
 491         struct inode_management *im = &sbi->im[ORPHAN_INO];
 492         int err = 0;
 493
 494         spin_lock(&im->ino_lock);
 495
 496 #ifdef CONFIG_F2FS_FAULT_INJECTION
 497         if (time_to_inject(FAULT_ORPHAN)) {
 498                 spin_unlock(&im->ino_lock);
 499                 return -ENOSPC;
 500         }
 501 #endif
 502         if (unlikely(im->ino_num >= sbi->max_orphans))
 503                 err = -ENOSPC;
 504         else
 505                 im->ino_num++;
 506         spin_unlock(&im->ino_lock);
 507
 508         return err;
 509 }
 510
 511 void release_orphan_inode(struct f2fs_sb_info *sbi)
 512 {
 513         struct inode_management *im = &sbi->im[ORPHAN_INO];
 514
 515         spin_lock(&im->ino_lock);
 516         f2fs_bug_on(sbi, im->ino_num == 0);
 517         im->ino_num--;
 518         spin_unlock(&im->ino_lock);
 519 }
 520
 521 void add_orphan_inode(struct inode *inode)
 522 {
 523         /* add new orphan ino entry into list */
 524         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
 525         update_inode_page(inode);
 526 }
 527
 528 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 529 {
 530         /* remove orphan entry from orphan list */
 531         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 532 }
 533
 534 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 535 {
 536         struct inode *inode;
 537         struct node_info ni;
 538
 539         inode = f2fs_iget_retry(sbi->sb, ino);
 540         if (IS_ERR(inode)) {
 541                 /*
 542                  * there should be a bug that we can't find the entry
 543                  * to orphan inode.
 544                  */
 545                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 546                 return PTR_ERR(inode);
 547         }
 548
 549         clear_nlink(inode);
 550
 551         /* truncate all the data during iput */
 552         iput(inode);
 553
 554         get_node_info(sbi, ino, &ni);
 555
 556         /* ENOMEM was fully retried in f2fs_evict_inode. */
 557         if (ni.blk_addr != NULL_ADDR) {
 558                 int err = acquire_orphan_inode(sbi);
 559
 560                 if (err) {
 561                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 562                         f2fs_msg(sbi->sb, KERN_WARNING,
 563                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 564                                         __func__, ino);
 565                         return err;
 566                 }
 567                 __add_ino_entry(sbi, ino, ORPHAN_INO);
 568         }
 569         return 0;
 570 }
 571
 572 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 573 {
 574         block_t start_blk, orphan_blocks, i, j;
 575         int err;
 576
 577         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 578                 return 0;
 579
 580         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 581         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 582
 583         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 584
 585         for (i = 0; i < orphan_blocks; i++) {
 586                 struct page *page = get_meta_page(sbi, start_blk + i);
 587                 struct f2fs_orphan_block *orphan_blk;
 588
 589                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 590                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 591                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 592                         err = recover_orphan_inode(sbi, ino);
 593                         if (err) {
 594                                 f2fs_put_page(page, 1);
 595                                 return err;
 596                         }
 597                 }
 598                 f2fs_put_page(page, 1);
 599         }
 600         /* clear Orphan Flag */
 601         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 602         return 0;
 603 }
 604
 605 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 606 {
 607         struct list_head *head;
 608         struct f2fs_orphan_block *orphan_blk = NULL;
 609         unsigned int nentries = 0;
 610         unsigned short index = 1;
 611         unsigned short orphan_blocks;
 612         struct page *page = NULL;
 613         struct ino_entry *orphan = NULL;
 614         struct inode_management *im = &sbi->im[ORPHAN_INO];
 615
 616         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 617
 618         /*
 619          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 620          * orphan inode operations are covered under f2fs_lock_op().
 621          * And, spin_lock should be avoided due to page operations below.
 622          */
 623         head = &im->ino_list;
 624
 625         /* loop for each orphan inode entry and write them in Jornal block */
 626         list_for_each_entry(orphan, head, list) {
 627                 if (!page) {
 628                         page = grab_meta_page(sbi, start_blk++);
 629                         orphan_blk =
 630                                 (struct f2fs_orphan_block *)page_address(page);
 631                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 632                 }
 633
 634                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 635
 636                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 637                         /*
 638                          * an orphan block is full of 1020 entries,
 639                          * then we need to flush current orphan blocks
 640                          * and bring another one in memory
 641                          */
 642                         orphan_blk->blk_addr = cpu_to_le16(index);
 643                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 644                         orphan_blk->entry_count = cpu_to_le32(nentries);
 645                         set_page_dirty(page);
 646                         f2fs_put_page(page, 1);
 647                         index++;
 648                         nentries = 0;
 649                         page = NULL;
 650                 }
 651         }
 652
 653         if (page) {
 654                 orphan_blk->blk_addr = cpu_to_le16(index);
 655                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 656                 orphan_blk->entry_count = cpu_to_le32(nentries);
 657                 set_page_dirty(page);
 658                 f2fs_put_page(page, 1);
 659         }
 660 }
 661
 662 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 663                                 block_t cp_addr, unsigned long long *version)
 664 {
 665         struct page *cp_page_1, *cp_page_2 = NULL;
 666         unsigned long blk_size = sbi->blocksize;
 667         struct f2fs_checkpoint *cp_block;
 668         unsigned long long cur_version = 0, pre_version = 0;
 669         size_t crc_offset;
 670         __u32 crc = 0;
 671
 672         /* Read the 1st cp block in this CP pack */
 673         cp_page_1 = get_meta_page(sbi, cp_addr);
 674
 675         /* get the version number */
 676         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 677         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 678         if (crc_offset >= blk_size)
 679                 goto invalid_cp1;
 680
 681         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 682         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 683                 goto invalid_cp1;
 684
 685         pre_version = cur_cp_version(cp_block);
 686
 687         /* Read the 2nd cp block in this CP pack */
 688         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 689         cp_page_2 = get_meta_page(sbi, cp_addr);
 690
 691         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 692         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 693         if (crc_offset >= blk_size)
 694                 goto invalid_cp2;
 695
 696         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 697         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 698                 goto invalid_cp2;
 699
 700         cur_version = cur_cp_version(cp_block);
 701
 702         if (cur_version == pre_version) {
 703                 *version = cur_version;
 704                 f2fs_put_page(cp_page_2, 1);
 705                 return cp_page_1;
 706         }
 707 invalid_cp2:
 708         f2fs_put_page(cp_page_2, 1);
 709 invalid_cp1:
 710         f2fs_put_page(cp_page_1, 1);
 711         return NULL;
 712 }
 713
 714 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 715 {
 716         struct f2fs_checkpoint *cp_block;
 717         struct f2fs_super_block *fsb = sbi->raw_super;
 718         struct page *cp1, *cp2, *cur_page;
 719         unsigned long blk_size = sbi->blocksize;
 720         unsigned long long cp1_version = 0, cp2_version = 0;
 721         unsigned long long cp_start_blk_no;
 722         unsigned int cp_blks = 1 + __cp_payload(sbi);
 723         block_t cp_blk_no;
 724         int i;
 725
 726         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 727         if (!sbi->ckpt)
 728                 return -ENOMEM;
 729         /*
 730          * Finding out valid cp block involves read both
 731          * sets( cp pack1 and cp pack 2)
 732          */
 733         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 734         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 735
 736         /* The second checkpoint pack should start at the next segment */
 737         cp_start_blk_no += ((unsigned long long)1) <<
 738                                 le32_to_cpu(fsb->log_blocks_per_seg);
 739         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 740
 741         if (cp1 && cp2) {
 742                 if (ver_after(cp2_version, cp1_version))
 743                         cur_page = cp2;
 744                 else
 745                         cur_page = cp1;
 746         } else if (cp1) {
 747                 cur_page = cp1;
 748         } else if (cp2) {
 749                 cur_page = cp2;
 750         } else {
 751                 goto fail_no_cp;
 752         }
 753
 754         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 755         memcpy(sbi->ckpt, cp_block, blk_size);
 756
 757         /* Sanity checking of checkpoint */
 758         if (sanity_check_ckpt(sbi))
 759                 goto fail_no_cp;
 760
 761         if (cp_blks <= 1)
 762                 goto done;
 763
 764         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 765         if (cur_page == cp2)
 766                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 767
 768         for (i = 1; i < cp_blks; i++) {
 769                 void *sit_bitmap_ptr;
 770                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 771
 772                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 773                 sit_bitmap_ptr = page_address(cur_page);
 774                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 775                 f2fs_put_page(cur_page, 1);
 776         }
 777 done:
 778         f2fs_put_page(cp1, 1);
 779         f2fs_put_page(cp2, 1);
 780         return 0;
 781
 782 fail_no_cp:
 783         kfree(sbi->ckpt);
 784         return -EINVAL;
 785 }
 786
 787 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 788 {
 789         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 790         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 791
 792         if (is_inode_flag_set(inode, flag))
 793                 return;
 794
 795         set_inode_flag(inode, flag);
 796         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
 797         stat_inc_dirty_inode(sbi, type);
 798 }
 799
 800 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 801 {
 802         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 803
 804         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 805                 return;
 806
 807         list_del_init(&F2FS_I(inode)->dirty_list);
 808         clear_inode_flag(inode, flag);
 809         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 810 }
 811
 812 void update_dirty_page(struct inode *inode, struct page *page)
 813 {
 814         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 815         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 816
 817         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 818                         !S_ISLNK(inode->i_mode))
 819                 return;
 820
 821         spin_lock(&sbi->inode_lock[type]);
 822         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 823                 __add_dirty_inode(inode, type);
 824         inode_inc_dirty_pages(inode);
 825         spin_unlock(&sbi->inode_lock[type]);
 826
 827         SetPagePrivate(page);
 828         f2fs_trace_pid(page);
 829 }
 830
 831 void remove_dirty_inode(struct inode *inode)
 832 {
 833         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 834         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 835
 836         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 837                         !S_ISLNK(inode->i_mode))
 838                 return;
 839
 840         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 841                 return;
 842
 843         spin_lock(&sbi->inode_lock[type]);
 844         __remove_dirty_inode(inode, type);
 845         spin_unlock(&sbi->inode_lock[type]);
 846 }
 847
 848 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 849 {
 850         struct list_head *head;
 851         struct inode *inode;
 852         struct f2fs_inode_info *fi;
 853         bool is_dir = (type == DIR_INODE);
 854
 855         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 856                                 get_pages(sbi, is_dir ?
 857                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 858 retry:
 859         if (unlikely(f2fs_cp_error(sbi)))
 860                 return -EIO;
 861
 862         spin_lock(&sbi->inode_lock[type]);
 863
 864         head = &sbi->inode_list[type];
 865         if (list_empty(head)) {
 866                 spin_unlock(&sbi->inode_lock[type]);
 867                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 868                                 get_pages(sbi, is_dir ?
 869                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 870                 return 0;
 871         }
 872         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 873         inode = igrab(&fi->vfs_inode);
 874         spin_unlock(&sbi->inode_lock[type]);
 875         if (inode) {
 876                 filemap_fdatawrite(inode->i_mapping);
 877                 iput(inode);
 878         } else {
 879                 /*
 880                  * We should submit bio, since it exists several
 881                  * wribacking dentry pages in the freeing inode.
 882                  */
 883                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 884                 cond_resched();
 885         }
 886         goto retry;
 887 }
 888
 889 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 890 {
 891         struct list_head *head = &sbi->inode_list[DIRTY_META];
 892         struct inode *inode;
 893         struct f2fs_inode_info *fi;
 894         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 895
 896         while (total--) {
 897                 if (unlikely(f2fs_cp_error(sbi)))
 898                         return -EIO;
 899
 900                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 901                 if (list_empty(head)) {
 902                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 903                         return 0;
 904                 }
 905                 fi = list_entry(head->next, struct f2fs_inode_info,
 906                                                         gdirty_list);
 907                 inode = igrab(&fi->vfs_inode);
 908                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 909                 if (inode) {
 910                         update_inode_page(inode);
 911                         iput(inode);
 912                 }
 913         };
 914         return 0;
 915 }
 916
 917 /*
 918  * Freeze all the FS-operations for checkpoint.
 919  */
 920 static int block_operations(struct f2fs_sb_info *sbi)
 921 {
 922         struct writeback_control wbc = {
 923                 .sync_mode = WB_SYNC_ALL,
 924                 .nr_to_write = LONG_MAX,
 925                 .for_reclaim = 0,
 926         };
 927         struct blk_plug plug;
 928         int err = 0;
 929
 930         blk_start_plug(&plug);
 931
 932 retry_flush_dents:
 933         f2fs_lock_all(sbi);
 934         /* write all the dirty dentry pages */
 935         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 936                 f2fs_unlock_all(sbi);
 937                 err = sync_dirty_inodes(sbi, DIR_INODE);
 938                 if (err)
 939                         goto out;
 940                 goto retry_flush_dents;
 941         }
 942
 943         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
 944                 f2fs_unlock_all(sbi);
 945                 err = f2fs_sync_inode_meta(sbi);
 946                 if (err)
 947                         goto out;
 948                 goto retry_flush_dents;
 949         }
 950
 951         /*
 952          * POR: we should ensure that there are no dirty node pages
 953          * until finishing nat/sit flush.
 954          */
 955 retry_flush_nodes:
 956         down_write(&sbi->node_write);
 957
 958         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 959                 up_write(&sbi->node_write);
 960                 err = sync_node_pages(sbi, &wbc);
 961                 if (err) {
 962                         f2fs_unlock_all(sbi);
 963                         goto out;
 964                 }
 965                 goto retry_flush_nodes;
 966         }
 967 out:
 968         blk_finish_plug(&plug);
 969         return err;
 970 }
 971
 972 static void unblock_operations(struct f2fs_sb_info *sbi)
 973 {
 974         up_write(&sbi->node_write);
 975
 976         build_free_nids(sbi);
 977         f2fs_unlock_all(sbi);
 978 }
 979
 980 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 981 {
 982         DEFINE_WAIT(wait);
 983
 984         for (;;) {
 985                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 986
 987                 if (!atomic_read(&sbi->nr_wb_bios))
 988                         break;
 989
 990                 io_schedule_timeout(5*HZ);
 991         }
 992         finish_wait(&sbi->cp_wait, &wait);
 993 }
 994
 995 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 996 {
 997         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 998         struct f2fs_nm_info *nm_i = NM_I(sbi);
 999         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1000         nid_t last_nid = nm_i->next_scan_nid;
1001         block_t start_blk;
1002         unsigned int data_sum_blocks, orphan_blocks;
1003         __u32 crc32 = 0;
1004         int i;
1005         int cp_payload_blks = __cp_payload(sbi);
1006         struct super_block *sb = sbi->sb;
1007         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1008         u64 kbytes_written;
1009
1010         /* Flush all the NAT/SIT pages */
1011         while (get_pages(sbi, F2FS_DIRTY_META)) {
1012                 sync_meta_pages(sbi, META, LONG_MAX);
1013                 if (unlikely(f2fs_cp_error(sbi)))
1014                         return -EIO;
1015         }
1016
1017         next_free_nid(sbi, &last_nid);
1018
1019         /*
1020          * modify checkpoint
1021          * version number is already updated
1022          */
1023         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1024         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1025         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1026         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1027                 ckpt->cur_node_segno[i] =
1028                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1029                 ckpt->cur_node_blkoff[i] =
1030                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1031                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1032                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1033         }
1034         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1035                 ckpt->cur_data_segno[i] =
1036                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1037                 ckpt->cur_data_blkoff[i] =
1038                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1039                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1040                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1041         }
1042
1043         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1044         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1045         ckpt->next_free_nid = cpu_to_le32(last_nid);
1046
1047         /* 2 cp  + n data seg summary + orphan inode blocks */
1048         data_sum_blocks = npages_for_summary_flush(sbi, false);
1049         spin_lock(&sbi->cp_lock);
1050         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1051                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1052         else
1053                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1054         spin_unlock(&sbi->cp_lock);
1055
1056         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1057         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1058                         orphan_blocks);
1059
1060         if (__remain_node_summaries(cpc->reason))
1061                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1062                                 cp_payload_blks + data_sum_blocks +
1063                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1064         else
1065                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1066                                 cp_payload_blks + data_sum_blocks +
1067                                 orphan_blocks);
1068
1069         spin_lock(&sbi->cp_lock);
1070         if (cpc->reason == CP_UMOUNT)
1071                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1072         else
1073                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1074
1075         if (cpc->reason == CP_FASTBOOT)
1076                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1077         else
1078                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1079
1080         if (orphan_num)
1081                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1082         else
1083                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1084
1085         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1086                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1087
1088         /* set this flag to activate crc|cp_ver for recovery */
1089         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1090
1091         spin_unlock(&sbi->cp_lock);
1092
1093         /* update SIT/NAT bitmap */
1094         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1095         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1096
1097         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1098         *((__le32 *)((unsigned char *)ckpt +
1099                                 le32_to_cpu(ckpt->checksum_offset)))
1100                                 = cpu_to_le32(crc32);
1101
1102         start_blk = __start_cp_addr(sbi);
1103
1104         /* need to wait for end_io results */
1105         wait_on_all_pages_writeback(sbi);
1106         if (unlikely(f2fs_cp_error(sbi)))
1107                 return -EIO;
1108
1109         /* write out checkpoint buffer at block 0 */
1110         update_meta_page(sbi, ckpt, start_blk++);
1111
1112         for (i = 1; i < 1 + cp_payload_blks; i++)
1113                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1114                                                         start_blk++);
1115
1116         if (orphan_num) {
1117                 write_orphan_inodes(sbi, start_blk);
1118                 start_blk += orphan_blocks;
1119         }
1120
1121         write_data_summaries(sbi, start_blk);
1122         start_blk += data_sum_blocks;
1123
1124         /* Record write statistics in the hot node summary */
1125         kbytes_written = sbi->kbytes_written;
1126         if (sb->s_bdev->bd_part)
1127                 kbytes_written += BD_PART_WRITTEN(sbi);
1128
1129         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1130
1131         if (__remain_node_summaries(cpc->reason)) {
1132                 write_node_summaries(sbi, start_blk);
1133                 start_blk += NR_CURSEG_NODE_TYPE;
1134         }
1135
1136         /* writeout checkpoint block */
1137         update_meta_page(sbi, ckpt, start_blk);
1138
1139         /* wait for previous submitted node/meta pages writeback */
1140         wait_on_all_pages_writeback(sbi);
1141
1142         if (unlikely(f2fs_cp_error(sbi)))
1143                 return -EIO;
1144
1145         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1146         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1147
1148         /* update user_block_counts */
1149         sbi->last_valid_block_count = sbi->total_valid_block_count;
1150         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1151
1152         /* Here, we only have one bio having CP pack */
1153         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1154
1155         /* wait for previous submitted meta pages writeback */
1156         wait_on_all_pages_writeback(sbi);
1157
1158         release_ino_entry(sbi, false);
1159
1160         if (unlikely(f2fs_cp_error(sbi)))
1161                 return -EIO;
1162
1163         clear_prefree_segments(sbi, cpc);
1164         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1165         clear_sbi_flag(sbi, SBI_NEED_CP);
1166
1167         /*
1168          * redirty superblock if metadata like node page or inode cache is
1169          * updated during writing checkpoint.
1170          */
1171         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1172                         get_pages(sbi, F2FS_DIRTY_IMETA))
1173                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1174
1175         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1176
1177         return 0;
1178 }
1179
1180 /*
1181  * We guarantee that this checkpoint procedure will not fail.
1182  */
1183 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1184 {
1185         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1186         unsigned long long ckpt_ver;
1187         int err = 0;
1188
1189         mutex_lock(&sbi->cp_mutex);
1190
1191         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1192                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1193                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1194                 goto out;
1195         if (unlikely(f2fs_cp_error(sbi))) {
1196                 err = -EIO;
1197                 goto out;
1198         }
1199         if (f2fs_readonly(sbi->sb)) {
1200                 err = -EROFS;
1201                 goto out;
1202         }
1203
1204         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1205
1206         err = block_operations(sbi);
1207         if (err)
1208                 goto out;
1209
1210         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1211
1212         f2fs_flush_merged_bios(sbi);
1213
1214         /* this is the case of multiple fstrims without any changes */
1215         if (cpc->reason == CP_DISCARD && !is_sbi_flag_set(sbi, SBI_IS_DIRTY)) {
1216                 f2fs_bug_on(sbi, NM_I(sbi)->dirty_nat_cnt);
1217                 f2fs_bug_on(sbi, SIT_I(sbi)->dirty_sentries);
1218                 f2fs_bug_on(sbi, prefree_segments(sbi));
1219                 flush_sit_entries(sbi, cpc);
1220                 clear_prefree_segments(sbi, cpc);
1221                 f2fs_wait_all_discard_bio(sbi);
1222                 unblock_operations(sbi);
1223                 goto out;
1224         }
1225
1226         /*
1227          * update checkpoint pack index
1228          * Increase the version number so that
1229          * SIT entries and seg summaries are written at correct place
1230          */
1231         ckpt_ver = cur_cp_version(ckpt);
1232         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1233
1234         /* write cached NAT/SIT entries to NAT/SIT area */
1235         flush_nat_entries(sbi);
1236         flush_sit_entries(sbi, cpc);
1237
1238         /* unlock all the fs_lock[] in do_checkpoint() */
1239         err = do_checkpoint(sbi, cpc);
1240
1241         f2fs_wait_all_discard_bio(sbi);
1242
1243         unblock_operations(sbi);
1244         stat_inc_cp_count(sbi->stat_info);
1245
1246         if (cpc->reason == CP_RECOVERY)
1247                 f2fs_msg(sbi->sb, KERN_NOTICE,
1248                         "checkpoint: version = %llx", ckpt_ver);
1249
1250         /* do checkpoint periodically */
1251         f2fs_update_time(sbi, CP_TIME);
1252         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1253 out:
1254         mutex_unlock(&sbi->cp_mutex);
1255         return err;
1256 }
1257
1258 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1259 {
1260         int i;
1261
1262         for (i = 0; i < MAX_INO_ENTRY; i++) {
1263                 struct inode_management *im = &sbi->im[i];
1264
1265                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1266                 spin_lock_init(&im->ino_lock);
1267                 INIT_LIST_HEAD(&im->ino_list);
1268                 im->ino_num = 0;
1269         }
1270
1271         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1272                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1273                                 F2FS_ORPHANS_PER_BLOCK;
1274 }
1275
1276 int __init create_checkpoint_caches(void)
1277 {
1278         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1279                         sizeof(struct ino_entry));
1280         if (!ino_entry_slab)
1281                 return -ENOMEM;
1282         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1283                         sizeof(struct inode_entry));
1284         if (!inode_entry_slab) {
1285                 kmem_cache_destroy(ino_entry_slab);
1286                 return -ENOMEM;
1287         }
1288         return 0;
1289 }
1290
1291 void destroy_checkpoint_caches(void)
1292 {
1293         kmem_cache_destroy(ino_entry_slab);
1294         kmem_cache_destroy(inode_entry_slab);
1295 }