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 = 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 ? (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, BIO_MAX_PAGES, 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, page->mapping->host,
 253                                                 0, page->index, META, WRITE);
 254
 255         unlock_page(page);
 256
 257         if (unlikely(f2fs_cp_error(sbi)))
 258                 f2fs_submit_merged_bio(sbi, META, WRITE);
 259
 260         return 0;
 261
 262 redirty_out:
 263         redirty_page_for_writepage(wbc, page);
 264         return AOP_WRITEPAGE_ACTIVATE;
 265 }
 266
 267 static int f2fs_write_meta_pages(struct address_space *mapping,
 268                                 struct writeback_control *wbc)
 269 {
 270         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 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         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 284         mutex_unlock(&sbi->cp_mutex);
 285         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 286         return 0;
 287
 288 skip_write:
 289         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 290         trace_f2fs_writepages(mapping->host, wbc, META);
 291         return 0;
 292 }
 293
 294 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 295                                                 long nr_to_write)
 296 {
 297         struct address_space *mapping = META_MAPPING(sbi);
 298         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 299         struct pagevec pvec;
 300         long nwritten = 0;
 301         struct writeback_control wbc = {
 302                 .for_reclaim = 0,
 303         };
 304         struct blk_plug plug;
 305
 306         pagevec_init(&pvec, 0);
 307
 308         blk_start_plug(&plug);
 309
 310         while (index <= end) {
 311                 int i, nr_pages;
 312                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 313                                 PAGECACHE_TAG_DIRTY,
 314                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 315                 if (unlikely(nr_pages == 0))
 316                         break;
 317
 318                 for (i = 0; i < nr_pages; i++) {
 319                         struct page *page = pvec.pages[i];
 320
 321                         if (prev == ULONG_MAX)
 322                                 prev = page->index - 1;
 323                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 324                                 pagevec_release(&pvec);
 325                                 goto stop;
 326                         }
 327
 328                         lock_page(page);
 329
 330                         if (unlikely(page->mapping != mapping)) {
 331 continue_unlock:
 332                                 unlock_page(page);
 333                                 continue;
 334                         }
 335                         if (!PageDirty(page)) {
 336                                 /* someone wrote it for us */
 337                                 goto continue_unlock;
 338                         }
 339
 340                         f2fs_wait_on_page_writeback(page, META, true);
 341
 342                         BUG_ON(PageWriteback(page));
 343                         if (!clear_page_dirty_for_io(page))
 344                                 goto continue_unlock;
 345
 346                         if (mapping->a_ops->writepage(page, &wbc)) {
 347                                 unlock_page(page);
 348                                 break;
 349                         }
 350                         nwritten++;
 351                         prev = page->index;
 352                         if (unlikely(nwritten >= nr_to_write))
 353                                 break;
 354                 }
 355                 pagevec_release(&pvec);
 356                 cond_resched();
 357         }
 358 stop:
 359         if (nwritten)
 360                 f2fs_submit_merged_bio(sbi, type, WRITE);
 361
 362         blk_finish_plug(&plug);
 363
 364         return nwritten;
 365 }
 366
 367 static int f2fs_set_meta_page_dirty(struct page *page)
 368 {
 369         trace_f2fs_set_page_dirty(page, META);
 370
 371         if (!PageUptodate(page))
 372                 SetPageUptodate(page);
 373         if (!PageDirty(page)) {
 374                 f2fs_set_page_dirty_nobuffers(page);
 375                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 376                 SetPagePrivate(page);
 377                 f2fs_trace_pid(page);
 378                 return 1;
 379         }
 380         return 0;
 381 }
 382
 383 const struct address_space_operations f2fs_meta_aops = {
 384         .writepage      = f2fs_write_meta_page,
 385         .writepages     = f2fs_write_meta_pages,
 386         .set_page_dirty = f2fs_set_meta_page_dirty,
 387         .invalidatepage = f2fs_invalidate_page,
 388         .releasepage    = f2fs_release_page,
 389 #ifdef CONFIG_MIGRATION
 390         .migratepage    = f2fs_migrate_page,
 391 #endif
 392 };
 393
 394 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 395 {
 396         struct inode_management *im = &sbi->im[type];
 397         struct ino_entry *e, *tmp;
 398
 399         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 400 retry:
 401         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 402
 403         spin_lock(&im->ino_lock);
 404         e = radix_tree_lookup(&im->ino_root, ino);
 405         if (!e) {
 406                 e = tmp;
 407                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 408                         spin_unlock(&im->ino_lock);
 409                         radix_tree_preload_end();
 410                         goto retry;
 411                 }
 412                 memset(e, 0, sizeof(struct ino_entry));
 413                 e->ino = ino;
 414
 415                 list_add_tail(&e->list, &im->ino_list);
 416                 if (type != ORPHAN_INO)
 417                         im->ino_num++;
 418         }
 419         spin_unlock(&im->ino_lock);
 420         radix_tree_preload_end();
 421
 422         if (e != tmp)
 423                 kmem_cache_free(ino_entry_slab, tmp);
 424 }
 425
 426 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 427 {
 428         struct inode_management *im = &sbi->im[type];
 429         struct ino_entry *e;
 430
 431         spin_lock(&im->ino_lock);
 432         e = radix_tree_lookup(&im->ino_root, ino);
 433         if (e) {
 434                 list_del(&e->list);
 435                 radix_tree_delete(&im->ino_root, ino);
 436                 im->ino_num--;
 437                 spin_unlock(&im->ino_lock);
 438                 kmem_cache_free(ino_entry_slab, e);
 439                 return;
 440         }
 441         spin_unlock(&im->ino_lock);
 442 }
 443
 444 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 445 {
 446         /* add new dirty ino entry into list */
 447         __add_ino_entry(sbi, ino, type);
 448 }
 449
 450 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 451 {
 452         /* remove dirty ino entry from list */
 453         __remove_ino_entry(sbi, ino, type);
 454 }
 455
 456 /* mode should be APPEND_INO or UPDATE_INO */
 457 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 458 {
 459         struct inode_management *im = &sbi->im[mode];
 460         struct ino_entry *e;
 461
 462         spin_lock(&im->ino_lock);
 463         e = radix_tree_lookup(&im->ino_root, ino);
 464         spin_unlock(&im->ino_lock);
 465         return e ? true : false;
 466 }
 467
 468 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 469 {
 470         struct ino_entry *e, *tmp;
 471         int i;
 472
 473         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 474                 struct inode_management *im = &sbi->im[i];
 475
 476                 spin_lock(&im->ino_lock);
 477                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 478                         list_del(&e->list);
 479                         radix_tree_delete(&im->ino_root, e->ino);
 480                         kmem_cache_free(ino_entry_slab, e);
 481                         im->ino_num--;
 482                 }
 483                 spin_unlock(&im->ino_lock);
 484         }
 485 }
 486
 487 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 488 {
 489         struct inode_management *im = &sbi->im[ORPHAN_INO];
 490         int err = 0;
 491
 492         spin_lock(&im->ino_lock);
 493
 494 #ifdef CONFIG_F2FS_FAULT_INJECTION
 495         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 496                 spin_unlock(&im->ino_lock);
 497                 f2fs_show_injection_info(FAULT_ORPHAN);
 498                 return -ENOSPC;
 499         }
 500 #endif
 501         if (unlikely(im->ino_num >= sbi->max_orphans))
 502                 err = -ENOSPC;
 503         else
 504                 im->ino_num++;
 505         spin_unlock(&im->ino_lock);
 506
 507         return err;
 508 }
 509
 510 void release_orphan_inode(struct f2fs_sb_info *sbi)
 511 {
 512         struct inode_management *im = &sbi->im[ORPHAN_INO];
 513
 514         spin_lock(&im->ino_lock);
 515         f2fs_bug_on(sbi, im->ino_num == 0);
 516         im->ino_num--;
 517         spin_unlock(&im->ino_lock);
 518 }
 519
 520 void add_orphan_inode(struct inode *inode)
 521 {
 522         /* add new orphan ino entry into list */
 523         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
 524         update_inode_page(inode);
 525 }
 526
 527 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 528 {
 529         /* remove orphan entry from orphan list */
 530         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 531 }
 532
 533 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 534 {
 535         struct inode *inode;
 536         struct node_info ni;
 537         int err = acquire_orphan_inode(sbi);
 538
 539         if (err) {
 540                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 541                 f2fs_msg(sbi->sb, KERN_WARNING,
 542                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 543                                 __func__, ino);
 544                 return err;
 545         }
 546
 547         __add_ino_entry(sbi, ino, ORPHAN_INO);
 548
 549         inode = f2fs_iget_retry(sbi->sb, ino);
 550         if (IS_ERR(inode)) {
 551                 /*
 552                  * there should be a bug that we can't find the entry
 553                  * to orphan inode.
 554                  */
 555                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 556                 return PTR_ERR(inode);
 557         }
 558
 559         clear_nlink(inode);
 560
 561         /* truncate all the data during iput */
 562         iput(inode);
 563
 564         get_node_info(sbi, ino, &ni);
 565
 566         /* ENOMEM was fully retried in f2fs_evict_inode. */
 567         if (ni.blk_addr != NULL_ADDR) {
 568                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 569                 f2fs_msg(sbi->sb, KERN_WARNING,
 570                         "%s: orphan failed (ino=%x), run fsck to fix.",
 571                                 __func__, ino);
 572                 return -EIO;
 573         }
 574         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 575         return 0;
 576 }
 577
 578 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 579 {
 580         block_t start_blk, orphan_blocks, i, j;
 581         int err;
 582
 583         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 584                 return 0;
 585
 586         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 587         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 588
 589         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 590
 591         for (i = 0; i < orphan_blocks; i++) {
 592                 struct page *page = get_meta_page(sbi, start_blk + i);
 593                 struct f2fs_orphan_block *orphan_blk;
 594
 595                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 596                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 597                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 598                         err = recover_orphan_inode(sbi, ino);
 599                         if (err) {
 600                                 f2fs_put_page(page, 1);
 601                                 return err;
 602                         }
 603                 }
 604                 f2fs_put_page(page, 1);
 605         }
 606         /* clear Orphan Flag */
 607         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 608         return 0;
 609 }
 610
 611 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 612 {
 613         struct list_head *head;
 614         struct f2fs_orphan_block *orphan_blk = NULL;
 615         unsigned int nentries = 0;
 616         unsigned short index = 1;
 617         unsigned short orphan_blocks;
 618         struct page *page = NULL;
 619         struct ino_entry *orphan = NULL;
 620         struct inode_management *im = &sbi->im[ORPHAN_INO];
 621
 622         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 623
 624         /*
 625          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 626          * orphan inode operations are covered under f2fs_lock_op().
 627          * And, spin_lock should be avoided due to page operations below.
 628          */
 629         head = &im->ino_list;
 630
 631         /* loop for each orphan inode entry and write them in Jornal block */
 632         list_for_each_entry(orphan, head, list) {
 633                 if (!page) {
 634                         page = grab_meta_page(sbi, start_blk++);
 635                         orphan_blk =
 636                                 (struct f2fs_orphan_block *)page_address(page);
 637                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 638                 }
 639
 640                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 641
 642                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 643                         /*
 644                          * an orphan block is full of 1020 entries,
 645                          * then we need to flush current orphan blocks
 646                          * and bring another one in memory
 647                          */
 648                         orphan_blk->blk_addr = cpu_to_le16(index);
 649                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 650                         orphan_blk->entry_count = cpu_to_le32(nentries);
 651                         set_page_dirty(page);
 652                         f2fs_put_page(page, 1);
 653                         index++;
 654                         nentries = 0;
 655                         page = NULL;
 656                 }
 657         }
 658
 659         if (page) {
 660                 orphan_blk->blk_addr = cpu_to_le16(index);
 661                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 662                 orphan_blk->entry_count = cpu_to_le32(nentries);
 663                 set_page_dirty(page);
 664                 f2fs_put_page(page, 1);
 665         }
 666 }
 667
 668 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 669                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 670                 unsigned long long *version)
 671 {
 672         unsigned long blk_size = sbi->blocksize;
 673         size_t crc_offset = 0;
 674         __u32 crc = 0;
 675
 676         *cp_page = get_meta_page(sbi, cp_addr);
 677         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 678
 679         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 680         if (crc_offset >= blk_size) {
 681                 f2fs_msg(sbi->sb, KERN_WARNING,
 682                         "invalid crc_offset: %zu", crc_offset);
 683                 return -EINVAL;
 684         }
 685
 686         crc = cur_cp_crc(*cp_block);
 687         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 688                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 689                 return -EINVAL;
 690         }
 691
 692         *version = cur_cp_version(*cp_block);
 693         return 0;
 694 }
 695
 696 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 697                                 block_t cp_addr, unsigned long long *version)
 698 {
 699         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 700         struct f2fs_checkpoint *cp_block = NULL;
 701         unsigned long long cur_version = 0, pre_version = 0;
 702         int err;
 703
 704         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 705                                         &cp_page_1, version);
 706         if (err)
 707                 goto invalid_cp1;
 708         pre_version = *version;
 709
 710         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 711         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 712                                         &cp_page_2, version);
 713         if (err)
 714                 goto invalid_cp2;
 715         cur_version = *version;
 716
 717         if (cur_version == pre_version) {
 718                 *version = cur_version;
 719                 f2fs_put_page(cp_page_2, 1);
 720                 return cp_page_1;
 721         }
 722 invalid_cp2:
 723         f2fs_put_page(cp_page_2, 1);
 724 invalid_cp1:
 725         f2fs_put_page(cp_page_1, 1);
 726         return NULL;
 727 }
 728
 729 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 730 {
 731         struct f2fs_checkpoint *cp_block;
 732         struct f2fs_super_block *fsb = sbi->raw_super;
 733         struct page *cp1, *cp2, *cur_page;
 734         unsigned long blk_size = sbi->blocksize;
 735         unsigned long long cp1_version = 0, cp2_version = 0;
 736         unsigned long long cp_start_blk_no;
 737         unsigned int cp_blks = 1 + __cp_payload(sbi);
 738         block_t cp_blk_no;
 739         int i;
 740
 741         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 742         if (!sbi->ckpt)
 743                 return -ENOMEM;
 744         /*
 745          * Finding out valid cp block involves read both
 746          * sets( cp pack1 and cp pack 2)
 747          */
 748         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 749         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 750
 751         /* The second checkpoint pack should start at the next segment */
 752         cp_start_blk_no += ((unsigned long long)1) <<
 753                                 le32_to_cpu(fsb->log_blocks_per_seg);
 754         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 755
 756         if (cp1 && cp2) {
 757                 if (ver_after(cp2_version, cp1_version))
 758                         cur_page = cp2;
 759                 else
 760                         cur_page = cp1;
 761         } else if (cp1) {
 762                 cur_page = cp1;
 763         } else if (cp2) {
 764                 cur_page = cp2;
 765         } else {
 766                 goto fail_no_cp;
 767         }
 768
 769         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 770         memcpy(sbi->ckpt, cp_block, blk_size);
 771
 772         /* Sanity checking of checkpoint */
 773         if (sanity_check_ckpt(sbi))
 774                 goto free_fail_no_cp;
 775
 776         if (cur_page == cp1)
 777                 sbi->cur_cp_pack = 1;
 778         else
 779                 sbi->cur_cp_pack = 2;
 780
 781         if (cp_blks <= 1)
 782                 goto done;
 783
 784         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 785         if (cur_page == cp2)
 786                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 787
 788         for (i = 1; i < cp_blks; i++) {
 789                 void *sit_bitmap_ptr;
 790                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 791
 792                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 793                 sit_bitmap_ptr = page_address(cur_page);
 794                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 795                 f2fs_put_page(cur_page, 1);
 796         }
 797 done:
 798         f2fs_put_page(cp1, 1);
 799         f2fs_put_page(cp2, 1);
 800         return 0;
 801
 802 free_fail_no_cp:
 803         f2fs_put_page(cp1, 1);
 804         f2fs_put_page(cp2, 1);
 805 fail_no_cp:
 806         kfree(sbi->ckpt);
 807         return -EINVAL;
 808 }
 809
 810 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 811 {
 812         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 813         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 814
 815         if (is_inode_flag_set(inode, flag))
 816                 return;
 817
 818         set_inode_flag(inode, flag);
 819         list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
 820         stat_inc_dirty_inode(sbi, type);
 821 }
 822
 823 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 824 {
 825         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 826
 827         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 828                 return;
 829
 830         list_del_init(&F2FS_I(inode)->dirty_list);
 831         clear_inode_flag(inode, flag);
 832         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 833 }
 834
 835 void update_dirty_page(struct inode *inode, struct page *page)
 836 {
 837         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 838         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 839
 840         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 841                         !S_ISLNK(inode->i_mode))
 842                 return;
 843
 844         spin_lock(&sbi->inode_lock[type]);
 845         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 846                 __add_dirty_inode(inode, type);
 847         inode_inc_dirty_pages(inode);
 848         spin_unlock(&sbi->inode_lock[type]);
 849
 850         SetPagePrivate(page);
 851         f2fs_trace_pid(page);
 852 }
 853
 854 void remove_dirty_inode(struct inode *inode)
 855 {
 856         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 857         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 858
 859         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 860                         !S_ISLNK(inode->i_mode))
 861                 return;
 862
 863         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 864                 return;
 865
 866         spin_lock(&sbi->inode_lock[type]);
 867         __remove_dirty_inode(inode, type);
 868         spin_unlock(&sbi->inode_lock[type]);
 869 }
 870
 871 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 872 {
 873         struct list_head *head;
 874         struct inode *inode;
 875         struct f2fs_inode_info *fi;
 876         bool is_dir = (type == DIR_INODE);
 877
 878         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 879                                 get_pages(sbi, is_dir ?
 880                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 881 retry:
 882         if (unlikely(f2fs_cp_error(sbi)))
 883                 return -EIO;
 884
 885         spin_lock(&sbi->inode_lock[type]);
 886
 887         head = &sbi->inode_list[type];
 888         if (list_empty(head)) {
 889                 spin_unlock(&sbi->inode_lock[type]);
 890                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 891                                 get_pages(sbi, is_dir ?
 892                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 893                 return 0;
 894         }
 895         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
 896         inode = igrab(&fi->vfs_inode);
 897         spin_unlock(&sbi->inode_lock[type]);
 898         if (inode) {
 899                 filemap_fdatawrite(inode->i_mapping);
 900                 iput(inode);
 901         } else {
 902                 /*
 903                  * We should submit bio, since it exists several
 904                  * wribacking dentry pages in the freeing inode.
 905                  */
 906                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 907                 cond_resched();
 908         }
 909         goto retry;
 910 }
 911
 912 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 913 {
 914         struct list_head *head = &sbi->inode_list[DIRTY_META];
 915         struct inode *inode;
 916         struct f2fs_inode_info *fi;
 917         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 918
 919         while (total--) {
 920                 if (unlikely(f2fs_cp_error(sbi)))
 921                         return -EIO;
 922
 923                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 924                 if (list_empty(head)) {
 925                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 926                         return 0;
 927                 }
 928                 fi = list_first_entry(head, struct f2fs_inode_info,
 929                                                         gdirty_list);
 930                 inode = igrab(&fi->vfs_inode);
 931                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 932                 if (inode) {
 933                         sync_inode_metadata(inode, 0);
 934
 935                         /* it's on eviction */
 936                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
 937                                 update_inode_page(inode);
 938                         iput(inode);
 939                 }
 940         };
 941         return 0;
 942 }
 943
 944 /*
 945  * Freeze all the FS-operations for checkpoint.
 946  */
 947 static int block_operations(struct f2fs_sb_info *sbi)
 948 {
 949         struct writeback_control wbc = {
 950                 .sync_mode = WB_SYNC_ALL,
 951                 .nr_to_write = LONG_MAX,
 952                 .for_reclaim = 0,
 953         };
 954         struct blk_plug plug;
 955         int err = 0;
 956
 957         blk_start_plug(&plug);
 958
 959 retry_flush_dents:
 960         f2fs_lock_all(sbi);
 961         /* write all the dirty dentry pages */
 962         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 963                 f2fs_unlock_all(sbi);
 964                 err = sync_dirty_inodes(sbi, DIR_INODE);
 965                 if (err)
 966                         goto out;
 967                 goto retry_flush_dents;
 968         }
 969
 970         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
 971                 f2fs_unlock_all(sbi);
 972                 err = f2fs_sync_inode_meta(sbi);
 973                 if (err)
 974                         goto out;
 975                 goto retry_flush_dents;
 976         }
 977
 978         /*
 979          * POR: we should ensure that there are no dirty node pages
 980          * until finishing nat/sit flush.
 981          */
 982 retry_flush_nodes:
 983         down_write(&sbi->node_write);
 984
 985         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 986                 up_write(&sbi->node_write);
 987                 err = sync_node_pages(sbi, &wbc);
 988                 if (err) {
 989                         f2fs_unlock_all(sbi);
 990                         goto out;
 991                 }
 992                 goto retry_flush_nodes;
 993         }
 994 out:
 995         blk_finish_plug(&plug);
 996         return err;
 997 }
 998
 999 static void unblock_operations(struct f2fs_sb_info *sbi)
1000 {
1001         up_write(&sbi->node_write);
1002         f2fs_unlock_all(sbi);
1003 }
1004
1005 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1006 {
1007         DEFINE_WAIT(wait);
1008
1009         for (;;) {
1010                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1011
1012                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1013                         break;
1014
1015                 io_schedule_timeout(5*HZ);
1016         }
1017         finish_wait(&sbi->cp_wait, &wait);
1018 }
1019
1020 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1021 {
1022         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1023         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1024
1025         spin_lock(&sbi->cp_lock);
1026
1027         if (cpc->reason == CP_UMOUNT && ckpt->cp_pack_total_block_count >
1028                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1029                 disable_nat_bits(sbi, false);
1030
1031         if (cpc->reason == CP_UMOUNT)
1032                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1033         else
1034                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1035
1036         if (cpc->reason == CP_FASTBOOT)
1037                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1038         else
1039                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1040
1041         if (orphan_num)
1042                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1043         else
1044                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1045
1046         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1047                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1048
1049         /* set this flag to activate crc|cp_ver for recovery */
1050         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1051
1052         spin_unlock(&sbi->cp_lock);
1053 }
1054
1055 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1056 {
1057         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1058         struct f2fs_nm_info *nm_i = NM_I(sbi);
1059         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1060         nid_t last_nid = nm_i->next_scan_nid;
1061         block_t start_blk;
1062         unsigned int data_sum_blocks, orphan_blocks;
1063         __u32 crc32 = 0;
1064         int i;
1065         int cp_payload_blks = __cp_payload(sbi);
1066         struct super_block *sb = sbi->sb;
1067         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1068         u64 kbytes_written;
1069
1070         /* Flush all the NAT/SIT pages */
1071         while (get_pages(sbi, F2FS_DIRTY_META)) {
1072                 sync_meta_pages(sbi, META, LONG_MAX);
1073                 if (unlikely(f2fs_cp_error(sbi)))
1074                         return -EIO;
1075         }
1076
1077         next_free_nid(sbi, &last_nid);
1078
1079         /*
1080          * modify checkpoint
1081          * version number is already updated
1082          */
1083         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1084         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1085         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1086         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1087                 ckpt->cur_node_segno[i] =
1088                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1089                 ckpt->cur_node_blkoff[i] =
1090                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1091                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1092                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1093         }
1094         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1095                 ckpt->cur_data_segno[i] =
1096                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1097                 ckpt->cur_data_blkoff[i] =
1098                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1099                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1100                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1101         }
1102
1103         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1104         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1105         ckpt->next_free_nid = cpu_to_le32(last_nid);
1106
1107         /* 2 cp  + n data seg summary + orphan inode blocks */
1108         data_sum_blocks = npages_for_summary_flush(sbi, false);
1109         spin_lock(&sbi->cp_lock);
1110         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1111                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1112         else
1113                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1114         spin_unlock(&sbi->cp_lock);
1115
1116         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1117         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1118                         orphan_blocks);
1119
1120         if (__remain_node_summaries(cpc->reason))
1121                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1122                                 cp_payload_blks + data_sum_blocks +
1123                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1124         else
1125                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1126                                 cp_payload_blks + data_sum_blocks +
1127                                 orphan_blocks);
1128
1129         /* update ckpt flag for checkpoint */
1130         update_ckpt_flags(sbi, cpc);
1131
1132         /* update SIT/NAT bitmap */
1133         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1134         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1135
1136         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1137         *((__le32 *)((unsigned char *)ckpt +
1138                                 le32_to_cpu(ckpt->checksum_offset)))
1139                                 = cpu_to_le32(crc32);
1140
1141         start_blk = __start_cp_next_addr(sbi);
1142
1143         /* write nat bits */
1144         if (enabled_nat_bits(sbi, cpc)) {
1145                 __u64 cp_ver = cur_cp_version(ckpt);
1146                 unsigned int i;
1147                 block_t blk;
1148
1149                 cp_ver |= ((__u64)crc32 << 32);
1150                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1151
1152                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1153                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1154                         update_meta_page(sbi, nm_i->nat_bits +
1155                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1156
1157                 /* Flush all the NAT BITS pages */
1158                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1159                         sync_meta_pages(sbi, META, LONG_MAX);
1160                         if (unlikely(f2fs_cp_error(sbi)))
1161                                 return -EIO;
1162                 }
1163         }
1164
1165         /* need to wait for end_io results */
1166         wait_on_all_pages_writeback(sbi);
1167         if (unlikely(f2fs_cp_error(sbi)))
1168                 return -EIO;
1169
1170         /* write out checkpoint buffer at block 0 */
1171         update_meta_page(sbi, ckpt, start_blk++);
1172
1173         for (i = 1; i < 1 + cp_payload_blks; i++)
1174                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1175                                                         start_blk++);
1176
1177         if (orphan_num) {
1178                 write_orphan_inodes(sbi, start_blk);
1179                 start_blk += orphan_blocks;
1180         }
1181
1182         write_data_summaries(sbi, start_blk);
1183         start_blk += data_sum_blocks;
1184
1185         /* Record write statistics in the hot node summary */
1186         kbytes_written = sbi->kbytes_written;
1187         if (sb->s_bdev->bd_part)
1188                 kbytes_written += BD_PART_WRITTEN(sbi);
1189
1190         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1191
1192         if (__remain_node_summaries(cpc->reason)) {
1193                 write_node_summaries(sbi, start_blk);
1194                 start_blk += NR_CURSEG_NODE_TYPE;
1195         }
1196
1197         /* writeout checkpoint block */
1198         update_meta_page(sbi, ckpt, start_blk);
1199
1200         /* wait for previous submitted node/meta pages writeback */
1201         wait_on_all_pages_writeback(sbi);
1202
1203         if (unlikely(f2fs_cp_error(sbi)))
1204                 return -EIO;
1205
1206         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1207         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1208
1209         /* update user_block_counts */
1210         sbi->last_valid_block_count = sbi->total_valid_block_count;
1211         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1212
1213         /* Here, we only have one bio having CP pack */
1214         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1215
1216         /* wait for previous submitted meta pages writeback */
1217         wait_on_all_pages_writeback(sbi);
1218
1219         release_ino_entry(sbi, false);
1220
1221         if (unlikely(f2fs_cp_error(sbi)))
1222                 return -EIO;
1223
1224         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1225         clear_sbi_flag(sbi, SBI_NEED_CP);
1226         __set_cp_next_pack(sbi);
1227
1228         /*
1229          * redirty superblock if metadata like node page or inode cache is
1230          * updated during writing checkpoint.
1231          */
1232         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1233                         get_pages(sbi, F2FS_DIRTY_IMETA))
1234                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1235
1236         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1237
1238         return 0;
1239 }
1240
1241 /*
1242  * We guarantee that this checkpoint procedure will not fail.
1243  */
1244 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1245 {
1246         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1247         unsigned long long ckpt_ver;
1248         int err = 0;
1249
1250         mutex_lock(&sbi->cp_mutex);
1251
1252         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1253                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1254                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1255                 goto out;
1256         if (unlikely(f2fs_cp_error(sbi))) {
1257                 err = -EIO;
1258                 goto out;
1259         }
1260         if (f2fs_readonly(sbi->sb)) {
1261                 err = -EROFS;
1262                 goto out;
1263         }
1264
1265         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1266
1267         err = block_operations(sbi);
1268         if (err)
1269                 goto out;
1270
1271         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1272
1273         f2fs_flush_merged_bios(sbi);
1274
1275         /* this is the case of multiple fstrims without any changes */
1276         if (cpc->reason == CP_DISCARD) {
1277                 if (!exist_trim_candidates(sbi, cpc)) {
1278                         unblock_operations(sbi);
1279                         goto out;
1280                 }
1281
1282                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1283                                 SIT_I(sbi)->dirty_sentries == 0 &&
1284                                 prefree_segments(sbi) == 0) {
1285                         flush_sit_entries(sbi, cpc);
1286                         clear_prefree_segments(sbi, cpc);
1287                         unblock_operations(sbi);
1288                         goto out;
1289                 }
1290         }
1291
1292         /*
1293          * update checkpoint pack index
1294          * Increase the version number so that
1295          * SIT entries and seg summaries are written at correct place
1296          */
1297         ckpt_ver = cur_cp_version(ckpt);
1298         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1299
1300         /* write cached NAT/SIT entries to NAT/SIT area */
1301         flush_nat_entries(sbi, cpc);
1302         flush_sit_entries(sbi, cpc);
1303
1304         /* unlock all the fs_lock[] in do_checkpoint() */
1305         err = do_checkpoint(sbi, cpc);
1306         if (err)
1307                 release_discard_addrs(sbi);
1308         else
1309                 clear_prefree_segments(sbi, cpc);
1310
1311         unblock_operations(sbi);
1312         stat_inc_cp_count(sbi->stat_info);
1313
1314         if (cpc->reason == CP_RECOVERY)
1315                 f2fs_msg(sbi->sb, KERN_NOTICE,
1316                         "checkpoint: version = %llx", ckpt_ver);
1317
1318         /* do checkpoint periodically */
1319         f2fs_update_time(sbi, CP_TIME);
1320         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1321 out:
1322         mutex_unlock(&sbi->cp_mutex);
1323         return err;
1324 }
1325
1326 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1327 {
1328         int i;
1329
1330         for (i = 0; i < MAX_INO_ENTRY; i++) {
1331                 struct inode_management *im = &sbi->im[i];
1332
1333                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1334                 spin_lock_init(&im->ino_lock);
1335                 INIT_LIST_HEAD(&im->ino_list);
1336                 im->ino_num = 0;
1337         }
1338
1339         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1340                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1341                                 F2FS_ORPHANS_PER_BLOCK;
1342 }
1343
1344 int __init create_checkpoint_caches(void)
1345 {
1346         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1347                         sizeof(struct ino_entry));
1348         if (!ino_entry_slab)
1349                 return -ENOMEM;
1350         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1351                         sizeof(struct inode_entry));
1352         if (!inode_entry_slab) {
1353                 kmem_cache_destroy(ino_entry_slab);
1354                 return -ENOMEM;
1355         }
1356         return 0;
1357 }
1358
1359 void destroy_checkpoint_caches(void)
1360 {
1361         kmem_cache_destroy(ino_entry_slab);
1362         kmem_cache_destroy(inode_entry_slab);
1363 }