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         long diff, written;
 271
 272         /* collect a number of dirty meta pages and write together */
 273         if (wbc->for_kupdate ||
 274                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 275                 goto skip_write;
 276
 277         trace_f2fs_writepages(mapping->host, wbc, META);
 278
 279         /* if mounting is failed, skip writing node pages */
 280         mutex_lock(&sbi->cp_mutex);
 281         diff = nr_pages_to_write(sbi, META, wbc);
 282         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 283         mutex_unlock(&sbi->cp_mutex);
 284         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 285         return 0;
 286
 287 skip_write:
 288         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 289         trace_f2fs_writepages(mapping->host, wbc, META);
 290         return 0;
 291 }
 292
 293 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 294                                                 long nr_to_write)
 295 {
 296         struct address_space *mapping = META_MAPPING(sbi);
 297         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 298         struct pagevec pvec;
 299         long nwritten = 0;
 300         struct writeback_control wbc = {
 301                 .for_reclaim = 0,
 302         };
 303         struct blk_plug plug;
 304
 305         pagevec_init(&pvec, 0);
 306
 307         blk_start_plug(&plug);
 308
 309         while (index <= end) {
 310                 int i, nr_pages;
 311                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 312                                 PAGECACHE_TAG_DIRTY,
 313                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 314                 if (unlikely(nr_pages == 0))
 315                         break;
 316
 317                 for (i = 0; i < nr_pages; i++) {
 318                         struct page *page = pvec.pages[i];
 319
 320                         if (prev == ULONG_MAX)
 321                                 prev = page->index - 1;
 322                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 323                                 pagevec_release(&pvec);
 324                                 goto stop;
 325                         }
 326
 327                         lock_page(page);
 328
 329                         if (unlikely(page->mapping != mapping)) {
 330 continue_unlock:
 331                                 unlock_page(page);
 332                                 continue;
 333                         }
 334                         if (!PageDirty(page)) {
 335                                 /* someone wrote it for us */
 336                                 goto continue_unlock;
 337                         }
 338
 339                         f2fs_wait_on_page_writeback(page, META, true);
 340
 341                         BUG_ON(PageWriteback(page));
 342                         if (!clear_page_dirty_for_io(page))
 343                                 goto continue_unlock;
 344
 345                         if (mapping->a_ops->writepage(page, &wbc)) {
 346                                 unlock_page(page);
 347                                 break;
 348                         }
 349                         nwritten++;
 350                         prev = page->index;
 351                         if (unlikely(nwritten >= nr_to_write))
 352                                 break;
 353                 }
 354                 pagevec_release(&pvec);
 355                 cond_resched();
 356         }
 357 stop:
 358         if (nwritten)
 359                 f2fs_submit_merged_bio(sbi, type, WRITE);
 360
 361         blk_finish_plug(&plug);
 362
 363         return nwritten;
 364 }
 365
 366 static int f2fs_set_meta_page_dirty(struct page *page)
 367 {
 368         trace_f2fs_set_page_dirty(page, META);
 369
 370         if (!PageUptodate(page))
 371                 SetPageUptodate(page);
 372         if (!PageDirty(page)) {
 373                 f2fs_set_page_dirty_nobuffers(page);
 374                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 375                 SetPagePrivate(page);
 376                 f2fs_trace_pid(page);
 377                 return 1;
 378         }
 379         return 0;
 380 }
 381
 382 const struct address_space_operations f2fs_meta_aops = {
 383         .writepage      = f2fs_write_meta_page,
 384         .writepages     = f2fs_write_meta_pages,
 385         .set_page_dirty = f2fs_set_meta_page_dirty,
 386         .invalidatepage = f2fs_invalidate_page,
 387         .releasepage    = f2fs_release_page,
 388 #ifdef CONFIG_MIGRATION
 389         .migratepage    = f2fs_migrate_page,
 390 #endif
 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(sbi, 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         int err = acquire_orphan_inode(sbi);
 536
 537         if (err) {
 538                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 539                 f2fs_msg(sbi->sb, KERN_WARNING,
 540                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 541                                 __func__, ino);
 542                 return err;
 543         }
 544
 545         __add_ino_entry(sbi, ino, ORPHAN_INO);
 546
 547         inode = f2fs_iget_retry(sbi->sb, ino);
 548         if (IS_ERR(inode)) {
 549                 /*
 550                  * there should be a bug that we can't find the entry
 551                  * to orphan inode.
 552                  */
 553                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 554                 return PTR_ERR(inode);
 555         }
 556
 557         clear_nlink(inode);
 558
 559         /* truncate all the data during iput */
 560         iput(inode);
 561
 562         get_node_info(sbi, ino, &ni);
 563
 564         /* ENOMEM was fully retried in f2fs_evict_inode. */
 565         if (ni.blk_addr != NULL_ADDR) {
 566                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 567                 f2fs_msg(sbi->sb, KERN_WARNING,
 568                         "%s: orphan failed (ino=%x), run fsck to fix.",
 569                                 __func__, ino);
 570                 return -EIO;
 571         }
 572         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 573         return 0;
 574 }
 575
 576 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 577 {
 578         block_t start_blk, orphan_blocks, i, j;
 579         int err;
 580
 581         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 582                 return 0;
 583
 584         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 585         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 586
 587         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 588
 589         for (i = 0; i < orphan_blocks; i++) {
 590                 struct page *page = get_meta_page(sbi, start_blk + i);
 591                 struct f2fs_orphan_block *orphan_blk;
 592
 593                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 594                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 595                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 596                         err = recover_orphan_inode(sbi, ino);
 597                         if (err) {
 598                                 f2fs_put_page(page, 1);
 599                                 return err;
 600                         }
 601                 }
 602                 f2fs_put_page(page, 1);
 603         }
 604         /* clear Orphan Flag */
 605         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 606         return 0;
 607 }
 608
 609 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 610 {
 611         struct list_head *head;
 612         struct f2fs_orphan_block *orphan_blk = NULL;
 613         unsigned int nentries = 0;
 614         unsigned short index = 1;
 615         unsigned short orphan_blocks;
 616         struct page *page = NULL;
 617         struct ino_entry *orphan = NULL;
 618         struct inode_management *im = &sbi->im[ORPHAN_INO];
 619
 620         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 621
 622         /*
 623          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 624          * orphan inode operations are covered under f2fs_lock_op().
 625          * And, spin_lock should be avoided due to page operations below.
 626          */
 627         head = &im->ino_list;
 628
 629         /* loop for each orphan inode entry and write them in Jornal block */
 630         list_for_each_entry(orphan, head, list) {
 631                 if (!page) {
 632                         page = grab_meta_page(sbi, start_blk++);
 633                         orphan_blk =
 634                                 (struct f2fs_orphan_block *)page_address(page);
 635                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 636                 }
 637
 638                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 639
 640                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 641                         /*
 642                          * an orphan block is full of 1020 entries,
 643                          * then we need to flush current orphan blocks
 644                          * and bring another one in memory
 645                          */
 646                         orphan_blk->blk_addr = cpu_to_le16(index);
 647                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 648                         orphan_blk->entry_count = cpu_to_le32(nentries);
 649                         set_page_dirty(page);
 650                         f2fs_put_page(page, 1);
 651                         index++;
 652                         nentries = 0;
 653                         page = NULL;
 654                 }
 655         }
 656
 657         if (page) {
 658                 orphan_blk->blk_addr = cpu_to_le16(index);
 659                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 660                 orphan_blk->entry_count = cpu_to_le32(nentries);
 661                 set_page_dirty(page);
 662                 f2fs_put_page(page, 1);
 663         }
 664 }
 665
 666 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 667                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 668                 unsigned long long *version)
 669 {
 670         unsigned long blk_size = sbi->blocksize;
 671         size_t crc_offset = 0;
 672         __u32 crc = 0;
 673
 674         *cp_page = get_meta_page(sbi, cp_addr);
 675         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 676
 677         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 678         if (crc_offset >= blk_size) {
 679                 f2fs_msg(sbi->sb, KERN_WARNING,
 680                         "invalid crc_offset: %zu", crc_offset);
 681                 return -EINVAL;
 682         }
 683
 684         crc = le32_to_cpu(*((__le32 *)((unsigned char *)*cp_block
 685                                                         + crc_offset)));
 686         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 687                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 688                 return -EINVAL;
 689         }
 690
 691         *version = cur_cp_version(*cp_block);
 692         return 0;
 693 }
 694
 695 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 696                                 block_t cp_addr, unsigned long long *version)
 697 {
 698         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 699         struct f2fs_checkpoint *cp_block = NULL;
 700         unsigned long long cur_version = 0, pre_version = 0;
 701         int err;
 702
 703         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 704                                         &cp_page_1, version);
 705         if (err)
 706                 goto invalid_cp1;
 707         pre_version = *version;
 708
 709         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 710         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 711                                         &cp_page_2, version);
 712         if (err)
 713                 goto invalid_cp2;
 714         cur_version = *version;
 715
 716         if (cur_version == pre_version) {
 717                 *version = cur_version;
 718                 f2fs_put_page(cp_page_2, 1);
 719                 return cp_page_1;
 720         }
 721 invalid_cp2:
 722         f2fs_put_page(cp_page_2, 1);
 723 invalid_cp1:
 724         f2fs_put_page(cp_page_1, 1);
 725         return NULL;
 726 }
 727
 728 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 729 {
 730         struct f2fs_checkpoint *cp_block;
 731         struct f2fs_super_block *fsb = sbi->raw_super;
 732         struct page *cp1, *cp2, *cur_page;
 733         unsigned long blk_size = sbi->blocksize;
 734         unsigned long long cp1_version = 0, cp2_version = 0;
 735         unsigned long long cp_start_blk_no;
 736         unsigned int cp_blks = 1 + __cp_payload(sbi);
 737         block_t cp_blk_no;
 738         int i;
 739
 740         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 741         if (!sbi->ckpt)
 742                 return -ENOMEM;
 743         /*
 744          * Finding out valid cp block involves read both
 745          * sets( cp pack1 and cp pack 2)
 746          */
 747         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 748         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 749
 750         /* The second checkpoint pack should start at the next segment */
 751         cp_start_blk_no += ((unsigned long long)1) <<
 752                                 le32_to_cpu(fsb->log_blocks_per_seg);
 753         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 754
 755         if (cp1 && cp2) {
 756                 if (ver_after(cp2_version, cp1_version))
 757                         cur_page = cp2;
 758                 else
 759                         cur_page = cp1;
 760         } else if (cp1) {
 761                 cur_page = cp1;
 762         } else if (cp2) {
 763                 cur_page = cp2;
 764         } else {
 765                 goto fail_no_cp;
 766         }
 767
 768         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 769         memcpy(sbi->ckpt, cp_block, blk_size);
 770
 771         /* Sanity checking of checkpoint */
 772         if (sanity_check_ckpt(sbi))
 773                 goto fail_no_cp;
 774
 775         if (cp_blks <= 1)
 776                 goto done;
 777
 778         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 779         if (cur_page == cp2)
 780                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 781
 782         for (i = 1; i < cp_blks; i++) {
 783                 void *sit_bitmap_ptr;
 784                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 785
 786                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 787                 sit_bitmap_ptr = page_address(cur_page);
 788                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 789                 f2fs_put_page(cur_page, 1);
 790         }
 791 done:
 792         f2fs_put_page(cp1, 1);
 793         f2fs_put_page(cp2, 1);
 794         return 0;
 795
 796 fail_no_cp:
 797         kfree(sbi->ckpt);
 798         return -EINVAL;
 799 }
 800
 801 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 802 {
 803         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 804         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 805
 806         if (is_inode_flag_set(inode, flag))
 807                 return;
 808
 809         set_inode_flag(inode, flag);
 810         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
 811         stat_inc_dirty_inode(sbi, type);
 812 }
 813
 814 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 815 {
 816         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 817
 818         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 819                 return;
 820
 821         list_del_init(&F2FS_I(inode)->dirty_list);
 822         clear_inode_flag(inode, flag);
 823         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 824 }
 825
 826 void update_dirty_page(struct inode *inode, struct page *page)
 827 {
 828         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 829         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 830
 831         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 832                         !S_ISLNK(inode->i_mode))
 833                 return;
 834
 835         spin_lock(&sbi->inode_lock[type]);
 836         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 837                 __add_dirty_inode(inode, type);
 838         inode_inc_dirty_pages(inode);
 839         spin_unlock(&sbi->inode_lock[type]);
 840
 841         SetPagePrivate(page);
 842         f2fs_trace_pid(page);
 843 }
 844
 845 void remove_dirty_inode(struct inode *inode)
 846 {
 847         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 848         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 849
 850         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 851                         !S_ISLNK(inode->i_mode))
 852                 return;
 853
 854         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 855                 return;
 856
 857         spin_lock(&sbi->inode_lock[type]);
 858         __remove_dirty_inode(inode, type);
 859         spin_unlock(&sbi->inode_lock[type]);
 860 }
 861
 862 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 863 {
 864         struct list_head *head;
 865         struct inode *inode;
 866         struct f2fs_inode_info *fi;
 867         bool is_dir = (type == DIR_INODE);
 868
 869         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 870                                 get_pages(sbi, is_dir ?
 871                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 872 retry:
 873         if (unlikely(f2fs_cp_error(sbi)))
 874                 return -EIO;
 875
 876         spin_lock(&sbi->inode_lock[type]);
 877
 878         head = &sbi->inode_list[type];
 879         if (list_empty(head)) {
 880                 spin_unlock(&sbi->inode_lock[type]);
 881                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 882                                 get_pages(sbi, is_dir ?
 883                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 884                 return 0;
 885         }
 886         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 887         inode = igrab(&fi->vfs_inode);
 888         spin_unlock(&sbi->inode_lock[type]);
 889         if (inode) {
 890                 filemap_fdatawrite(inode->i_mapping);
 891                 iput(inode);
 892         } else {
 893                 /*
 894                  * We should submit bio, since it exists several
 895                  * wribacking dentry pages in the freeing inode.
 896                  */
 897                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 898                 cond_resched();
 899         }
 900         goto retry;
 901 }
 902
 903 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 904 {
 905         struct list_head *head = &sbi->inode_list[DIRTY_META];
 906         struct inode *inode;
 907         struct f2fs_inode_info *fi;
 908         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 909
 910         while (total--) {
 911                 if (unlikely(f2fs_cp_error(sbi)))
 912                         return -EIO;
 913
 914                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 915                 if (list_empty(head)) {
 916                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 917                         return 0;
 918                 }
 919                 fi = list_entry(head->next, struct f2fs_inode_info,
 920                                                         gdirty_list);
 921                 inode = igrab(&fi->vfs_inode);
 922                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 923                 if (inode) {
 924                         update_inode_page(inode);
 925                         iput(inode);
 926                 }
 927         };
 928         return 0;
 929 }
 930
 931 /*
 932  * Freeze all the FS-operations for checkpoint.
 933  */
 934 static int block_operations(struct f2fs_sb_info *sbi)
 935 {
 936         struct writeback_control wbc = {
 937                 .sync_mode = WB_SYNC_ALL,
 938                 .nr_to_write = LONG_MAX,
 939                 .for_reclaim = 0,
 940         };
 941         struct blk_plug plug;
 942         int err = 0;
 943
 944         blk_start_plug(&plug);
 945
 946 retry_flush_dents:
 947         f2fs_lock_all(sbi);
 948         /* write all the dirty dentry pages */
 949         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 950                 f2fs_unlock_all(sbi);
 951                 err = sync_dirty_inodes(sbi, DIR_INODE);
 952                 if (err)
 953                         goto out;
 954                 goto retry_flush_dents;
 955         }
 956
 957         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
 958                 f2fs_unlock_all(sbi);
 959                 err = f2fs_sync_inode_meta(sbi);
 960                 if (err)
 961                         goto out;
 962                 goto retry_flush_dents;
 963         }
 964
 965         /*
 966          * POR: we should ensure that there are no dirty node pages
 967          * until finishing nat/sit flush.
 968          */
 969 retry_flush_nodes:
 970         down_write(&sbi->node_write);
 971
 972         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 973                 up_write(&sbi->node_write);
 974                 err = sync_node_pages(sbi, &wbc);
 975                 if (err) {
 976                         f2fs_unlock_all(sbi);
 977                         goto out;
 978                 }
 979                 goto retry_flush_nodes;
 980         }
 981 out:
 982         blk_finish_plug(&plug);
 983         return err;
 984 }
 985
 986 static void unblock_operations(struct f2fs_sb_info *sbi)
 987 {
 988         up_write(&sbi->node_write);
 989
 990         build_free_nids(sbi);
 991         f2fs_unlock_all(sbi);
 992 }
 993
 994 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 995 {
 996         DEFINE_WAIT(wait);
 997
 998         for (;;) {
 999                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1000
1001                 if (!atomic_read(&sbi->nr_wb_bios))
1002                         break;
1003
1004                 io_schedule_timeout(5*HZ);
1005         }
1006         finish_wait(&sbi->cp_wait, &wait);
1007 }
1008
1009 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1010 {
1011         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1012         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1013
1014         spin_lock(&sbi->cp_lock);
1015
1016         if (cpc->reason == CP_UMOUNT)
1017                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1018         else
1019                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1020
1021         if (cpc->reason == CP_FASTBOOT)
1022                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1023         else
1024                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1025
1026         if (orphan_num)
1027                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1028         else
1029                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1030
1031         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1032                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1033
1034         /* set this flag to activate crc|cp_ver for recovery */
1035         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1036
1037         spin_unlock(&sbi->cp_lock);
1038 }
1039
1040 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1041 {
1042         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1043         struct f2fs_nm_info *nm_i = NM_I(sbi);
1044         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1045         nid_t last_nid = nm_i->next_scan_nid;
1046         block_t start_blk;
1047         unsigned int data_sum_blocks, orphan_blocks;
1048         __u32 crc32 = 0;
1049         int i;
1050         int cp_payload_blks = __cp_payload(sbi);
1051         struct super_block *sb = sbi->sb;
1052         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1053         u64 kbytes_written;
1054
1055         /* Flush all the NAT/SIT pages */
1056         while (get_pages(sbi, F2FS_DIRTY_META)) {
1057                 sync_meta_pages(sbi, META, LONG_MAX);
1058                 if (unlikely(f2fs_cp_error(sbi)))
1059                         return -EIO;
1060         }
1061
1062         next_free_nid(sbi, &last_nid);
1063
1064         /*
1065          * modify checkpoint
1066          * version number is already updated
1067          */
1068         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1069         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1070         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1071         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1072                 ckpt->cur_node_segno[i] =
1073                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1074                 ckpt->cur_node_blkoff[i] =
1075                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1076                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1077                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1078         }
1079         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1080                 ckpt->cur_data_segno[i] =
1081                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1082                 ckpt->cur_data_blkoff[i] =
1083                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1084                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1085                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1086         }
1087
1088         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1089         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1090         ckpt->next_free_nid = cpu_to_le32(last_nid);
1091
1092         /* 2 cp  + n data seg summary + orphan inode blocks */
1093         data_sum_blocks = npages_for_summary_flush(sbi, false);
1094         spin_lock(&sbi->cp_lock);
1095         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1096                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1097         else
1098                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1099         spin_unlock(&sbi->cp_lock);
1100
1101         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1102         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1103                         orphan_blocks);
1104
1105         if (__remain_node_summaries(cpc->reason))
1106                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1107                                 cp_payload_blks + data_sum_blocks +
1108                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1109         else
1110                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1111                                 cp_payload_blks + data_sum_blocks +
1112                                 orphan_blocks);
1113
1114         /* update ckpt flag for checkpoint */
1115         update_ckpt_flags(sbi, cpc);
1116
1117         /* update SIT/NAT bitmap */
1118         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1119         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1120
1121         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1122         *((__le32 *)((unsigned char *)ckpt +
1123                                 le32_to_cpu(ckpt->checksum_offset)))
1124                                 = cpu_to_le32(crc32);
1125
1126         start_blk = __start_cp_addr(sbi);
1127
1128         /* need to wait for end_io results */
1129         wait_on_all_pages_writeback(sbi);
1130         if (unlikely(f2fs_cp_error(sbi)))
1131                 return -EIO;
1132
1133         /* write out checkpoint buffer at block 0 */
1134         update_meta_page(sbi, ckpt, start_blk++);
1135
1136         for (i = 1; i < 1 + cp_payload_blks; i++)
1137                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1138                                                         start_blk++);
1139
1140         if (orphan_num) {
1141                 write_orphan_inodes(sbi, start_blk);
1142                 start_blk += orphan_blocks;
1143         }
1144
1145         write_data_summaries(sbi, start_blk);
1146         start_blk += data_sum_blocks;
1147
1148         /* Record write statistics in the hot node summary */
1149         kbytes_written = sbi->kbytes_written;
1150         if (sb->s_bdev->bd_part)
1151                 kbytes_written += BD_PART_WRITTEN(sbi);
1152
1153         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1154
1155         if (__remain_node_summaries(cpc->reason)) {
1156                 write_node_summaries(sbi, start_blk);
1157                 start_blk += NR_CURSEG_NODE_TYPE;
1158         }
1159
1160         /* writeout checkpoint block */
1161         update_meta_page(sbi, ckpt, start_blk);
1162
1163         /* wait for previous submitted node/meta pages writeback */
1164         wait_on_all_pages_writeback(sbi);
1165
1166         if (unlikely(f2fs_cp_error(sbi)))
1167                 return -EIO;
1168
1169         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1170         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1171
1172         /* update user_block_counts */
1173         sbi->last_valid_block_count = sbi->total_valid_block_count;
1174         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1175
1176         /* Here, we only have one bio having CP pack */
1177         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1178
1179         /* wait for previous submitted meta pages writeback */
1180         wait_on_all_pages_writeback(sbi);
1181
1182         release_ino_entry(sbi, false);
1183
1184         if (unlikely(f2fs_cp_error(sbi)))
1185                 return -EIO;
1186
1187         clear_prefree_segments(sbi, cpc);
1188         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1189         clear_sbi_flag(sbi, SBI_NEED_CP);
1190
1191         /*
1192          * redirty superblock if metadata like node page or inode cache is
1193          * updated during writing checkpoint.
1194          */
1195         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1196                         get_pages(sbi, F2FS_DIRTY_IMETA))
1197                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1198
1199         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1200
1201         return 0;
1202 }
1203
1204 /*
1205  * We guarantee that this checkpoint procedure will not fail.
1206  */
1207 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1208 {
1209         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1210         unsigned long long ckpt_ver;
1211         int err = 0;
1212
1213         mutex_lock(&sbi->cp_mutex);
1214
1215         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1216                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1217                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1218                 goto out;
1219         if (unlikely(f2fs_cp_error(sbi))) {
1220                 err = -EIO;
1221                 goto out;
1222         }
1223         if (f2fs_readonly(sbi->sb)) {
1224                 err = -EROFS;
1225                 goto out;
1226         }
1227
1228         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1229
1230         err = block_operations(sbi);
1231         if (err)
1232                 goto out;
1233
1234         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1235
1236         f2fs_flush_merged_bios(sbi);
1237
1238         /* this is the case of multiple fstrims without any changes */
1239         if (cpc->reason == CP_DISCARD && !is_sbi_flag_set(sbi, SBI_IS_DIRTY)) {
1240                 f2fs_bug_on(sbi, NM_I(sbi)->dirty_nat_cnt);
1241                 f2fs_bug_on(sbi, SIT_I(sbi)->dirty_sentries);
1242                 f2fs_bug_on(sbi, prefree_segments(sbi));
1243                 flush_sit_entries(sbi, cpc);
1244                 clear_prefree_segments(sbi, cpc);
1245                 f2fs_wait_all_discard_bio(sbi);
1246                 unblock_operations(sbi);
1247                 goto out;
1248         }
1249
1250         /*
1251          * update checkpoint pack index
1252          * Increase the version number so that
1253          * SIT entries and seg summaries are written at correct place
1254          */
1255         ckpt_ver = cur_cp_version(ckpt);
1256         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1257
1258         /* write cached NAT/SIT entries to NAT/SIT area */
1259         flush_nat_entries(sbi);
1260         flush_sit_entries(sbi, cpc);
1261
1262         /* unlock all the fs_lock[] in do_checkpoint() */
1263         err = do_checkpoint(sbi, cpc);
1264
1265         f2fs_wait_all_discard_bio(sbi);
1266
1267         unblock_operations(sbi);
1268         stat_inc_cp_count(sbi->stat_info);
1269
1270         if (cpc->reason == CP_RECOVERY)
1271                 f2fs_msg(sbi->sb, KERN_NOTICE,
1272                         "checkpoint: version = %llx", ckpt_ver);
1273
1274         /* do checkpoint periodically */
1275         f2fs_update_time(sbi, CP_TIME);
1276         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1277 out:
1278         mutex_unlock(&sbi->cp_mutex);
1279         return err;
1280 }
1281
1282 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1283 {
1284         int i;
1285
1286         for (i = 0; i < MAX_INO_ENTRY; i++) {
1287                 struct inode_management *im = &sbi->im[i];
1288
1289                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1290                 spin_lock_init(&im->ino_lock);
1291                 INIT_LIST_HEAD(&im->ino_list);
1292                 im->ino_num = 0;
1293         }
1294
1295         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1296                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1297                                 F2FS_ORPHANS_PER_BLOCK;
1298 }
1299
1300 int __init create_checkpoint_caches(void)
1301 {
1302         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1303                         sizeof(struct ino_entry));
1304         if (!ino_entry_slab)
1305                 return -ENOMEM;
1306         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1307                         sizeof(struct inode_entry));
1308         if (!inode_entry_slab) {
1309                 kmem_cache_destroy(ino_entry_slab);
1310                 return -ENOMEM;
1311         }
1312         return 0;
1313 }
1314
1315 void destroy_checkpoint_caches(void)
1316 {
1317         kmem_cache_destroy(ino_entry_slab);
1318         kmem_cache_destroy(inode_entry_slab);
1319 }