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/events/f2fs.h>
  24
  25 static struct kmem_cache *orphan_entry_slab;
  26 static struct kmem_cache *inode_entry_slab;
  27
  28 /*
  29  * We guarantee no failure on the returned page.
  30  */
  31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  32 {
  33         struct address_space *mapping = META_MAPPING(sbi);
  34         struct page *page = NULL;
  35 repeat:
  36         page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
  37         if (!page) {
  38                 cond_resched();
  39                 goto repeat;
  40         }
  41
  42         SetPageUptodate(page);
  43         return page;
  44 }
  45
  46 /*
  47  * We guarantee no failure on the returned page.
  48  */
  49 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  50 {
  51         struct address_space *mapping = META_MAPPING(sbi);
  52         struct page *page;
  53 repeat:
  54         page = grab_cache_page(mapping, index);
  55         if (!page) {
  56                 cond_resched();
  57                 goto repeat;
  58         }
  59         if (PageUptodate(page))
  60                 goto out;
  61
  62         if (f2fs_submit_page_bio(sbi, page, index,
  63                                 READ_SYNC | REQ_META | REQ_PRIO))
  64                 goto repeat;
  65
  66         lock_page(page);
  67         if (unlikely(page->mapping != mapping)) {
  68                 f2fs_put_page(page, 1);
  69                 goto repeat;
  70         }
  71 out:
  72         mark_page_accessed(page);
  73         return page;
  74 }
  75
  76 static inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  77 {
  78         switch (type) {
  79         case META_NAT:
  80                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  81         case META_SIT:
  82                 return SIT_BLK_CNT(sbi);
  83         case META_SSA:
  84         case META_CP:
  85                 return 0;
  86         default:
  87                 BUG();
  88         }
  89 }
  90
  91 /*
  92  * Readahead CP/NAT/SIT/SSA pages
  93  */
  94 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
  95 {
  96         block_t prev_blk_addr = 0;
  97         struct page *page;
  98         int blkno = start;
  99         int max_blks = get_max_meta_blks(sbi, type);
 100
 101         struct f2fs_io_info fio = {
 102                 .type = META,
 103                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 104         };
 105
 106         for (; nrpages-- > 0; blkno++) {
 107                 block_t blk_addr;
 108
 109                 switch (type) {
 110                 case META_NAT:
 111                         /* get nat block addr */
 112                         if (unlikely(blkno >= max_blks))
 113                                 blkno = 0;
 114                         blk_addr = current_nat_addr(sbi,
 115                                         blkno * NAT_ENTRY_PER_BLOCK);
 116                         break;
 117                 case META_SIT:
 118                         /* get sit block addr */
 119                         if (unlikely(blkno >= max_blks))
 120                                 goto out;
 121                         blk_addr = current_sit_addr(sbi,
 122                                         blkno * SIT_ENTRY_PER_BLOCK);
 123                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 124                                 goto out;
 125                         prev_blk_addr = blk_addr;
 126                         break;
 127                 case META_SSA:
 128                 case META_CP:
 129                         /* get ssa/cp block addr */
 130                         blk_addr = blkno;
 131                         break;
 132                 default:
 133                         BUG();
 134                 }
 135
 136                 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
 137                 if (!page)
 138                         continue;
 139                 if (PageUptodate(page)) {
 140                         mark_page_accessed(page);
 141                         f2fs_put_page(page, 1);
 142                         continue;
 143                 }
 144
 145                 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
 146                 mark_page_accessed(page);
 147                 f2fs_put_page(page, 0);
 148         }
 149 out:
 150         f2fs_submit_merged_bio(sbi, META, READ);
 151         return blkno - start;
 152 }
 153
 154 static int f2fs_write_meta_page(struct page *page,
 155                                 struct writeback_control *wbc)
 156 {
 157         struct inode *inode = page->mapping->host;
 158         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 159
 160         if (unlikely(sbi->por_doing))
 161                 goto redirty_out;
 162         if (wbc->for_reclaim)
 163                 goto redirty_out;
 164
 165         /* Should not write any meta pages, if any IO error was occurred */
 166         if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
 167                 goto no_write;
 168
 169         f2fs_wait_on_page_writeback(page, META);
 170         write_meta_page(sbi, page);
 171 no_write:
 172         dec_page_count(sbi, F2FS_DIRTY_META);
 173         unlock_page(page);
 174         return 0;
 175
 176 redirty_out:
 177         redirty_page_for_writepage(wbc, page);
 178         return AOP_WRITEPAGE_ACTIVATE;
 179 }
 180
 181 static int f2fs_write_meta_pages(struct address_space *mapping,
 182                                 struct writeback_control *wbc)
 183 {
 184         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 185         long diff, written;
 186
 187         /* collect a number of dirty meta pages and write together */
 188         if (wbc->for_kupdate ||
 189                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 190                 goto skip_write;
 191
 192         /* if mounting is failed, skip writing node pages */
 193         mutex_lock(&sbi->cp_mutex);
 194         diff = nr_pages_to_write(sbi, META, wbc);
 195         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 196         mutex_unlock(&sbi->cp_mutex);
 197         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 198         return 0;
 199
 200 skip_write:
 201         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 202         return 0;
 203 }
 204
 205 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 206                                                 long nr_to_write)
 207 {
 208         struct address_space *mapping = META_MAPPING(sbi);
 209         pgoff_t index = 0, end = LONG_MAX;
 210         struct pagevec pvec;
 211         long nwritten = 0;
 212         struct writeback_control wbc = {
 213                 .for_reclaim = 0,
 214         };
 215
 216         pagevec_init(&pvec, 0);
 217
 218         while (index <= end) {
 219                 int i, nr_pages;
 220                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 221                                 PAGECACHE_TAG_DIRTY,
 222                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 223                 if (unlikely(nr_pages == 0))
 224                         break;
 225
 226                 for (i = 0; i < nr_pages; i++) {
 227                         struct page *page = pvec.pages[i];
 228
 229                         lock_page(page);
 230
 231                         if (unlikely(page->mapping != mapping)) {
 232 continue_unlock:
 233                                 unlock_page(page);
 234                                 continue;
 235                         }
 236                         if (!PageDirty(page)) {
 237                                 /* someone wrote it for us */
 238                                 goto continue_unlock;
 239                         }
 240
 241                         if (!clear_page_dirty_for_io(page))
 242                                 goto continue_unlock;
 243
 244                         if (f2fs_write_meta_page(page, &wbc)) {
 245                                 unlock_page(page);
 246                                 break;
 247                         }
 248                         nwritten++;
 249                         if (unlikely(nwritten >= nr_to_write))
 250                                 break;
 251                 }
 252                 pagevec_release(&pvec);
 253                 cond_resched();
 254         }
 255
 256         if (nwritten)
 257                 f2fs_submit_merged_bio(sbi, type, WRITE);
 258
 259         return nwritten;
 260 }
 261
 262 static int f2fs_set_meta_page_dirty(struct page *page)
 263 {
 264         struct address_space *mapping = page->mapping;
 265         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 266
 267         trace_f2fs_set_page_dirty(page, META);
 268
 269         SetPageUptodate(page);
 270         if (!PageDirty(page)) {
 271                 __set_page_dirty_nobuffers(page);
 272                 inc_page_count(sbi, F2FS_DIRTY_META);
 273                 return 1;
 274         }
 275         return 0;
 276 }
 277
 278 const struct address_space_operations f2fs_meta_aops = {
 279         .writepage      = f2fs_write_meta_page,
 280         .writepages     = f2fs_write_meta_pages,
 281         .set_page_dirty = f2fs_set_meta_page_dirty,
 282 };
 283
 284 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 285 {
 286         int err = 0;
 287
 288         spin_lock(&sbi->orphan_inode_lock);
 289         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 290                 err = -ENOSPC;
 291         else
 292                 sbi->n_orphans++;
 293         spin_unlock(&sbi->orphan_inode_lock);
 294
 295         return err;
 296 }
 297
 298 void release_orphan_inode(struct f2fs_sb_info *sbi)
 299 {
 300         spin_lock(&sbi->orphan_inode_lock);
 301         f2fs_bug_on(sbi->n_orphans == 0);
 302         sbi->n_orphans--;
 303         spin_unlock(&sbi->orphan_inode_lock);
 304 }
 305
 306 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 307 {
 308         struct list_head *head;
 309         struct orphan_inode_entry *new, *orphan;
 310
 311         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 312         new->ino = ino;
 313
 314         spin_lock(&sbi->orphan_inode_lock);
 315         head = &sbi->orphan_inode_list;
 316         list_for_each_entry(orphan, head, list) {
 317                 if (orphan->ino == ino) {
 318                         spin_unlock(&sbi->orphan_inode_lock);
 319                         kmem_cache_free(orphan_entry_slab, new);
 320                         return;
 321                 }
 322
 323                 if (orphan->ino > ino)
 324                         break;
 325         }
 326
 327         /* add new orphan entry into list which is sorted by inode number */
 328         list_add_tail(&new->list, &orphan->list);
 329         spin_unlock(&sbi->orphan_inode_lock);
 330 }
 331
 332 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 333 {
 334         struct list_head *head;
 335         struct orphan_inode_entry *orphan;
 336
 337         spin_lock(&sbi->orphan_inode_lock);
 338         head = &sbi->orphan_inode_list;
 339         list_for_each_entry(orphan, head, list) {
 340                 if (orphan->ino == ino) {
 341                         list_del(&orphan->list);
 342                         f2fs_bug_on(sbi->n_orphans == 0);
 343                         sbi->n_orphans--;
 344                         spin_unlock(&sbi->orphan_inode_lock);
 345                         kmem_cache_free(orphan_entry_slab, orphan);
 346                         return;
 347                 }
 348         }
 349         spin_unlock(&sbi->orphan_inode_lock);
 350 }
 351
 352 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 353 {
 354         struct inode *inode = f2fs_iget(sbi->sb, ino);
 355         f2fs_bug_on(IS_ERR(inode));
 356         clear_nlink(inode);
 357
 358         /* truncate all the data during iput */
 359         iput(inode);
 360 }
 361
 362 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 363 {
 364         block_t start_blk, orphan_blkaddr, i, j;
 365
 366         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 367                 return;
 368
 369         sbi->por_doing = true;
 370         start_blk = __start_cp_addr(sbi) + 1;
 371         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 372
 373         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 374
 375         for (i = 0; i < orphan_blkaddr; i++) {
 376                 struct page *page = get_meta_page(sbi, start_blk + i);
 377                 struct f2fs_orphan_block *orphan_blk;
 378
 379                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 380                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 381                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 382                         recover_orphan_inode(sbi, ino);
 383                 }
 384                 f2fs_put_page(page, 1);
 385         }
 386         /* clear Orphan Flag */
 387         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 388         sbi->por_doing = false;
 389         return;
 390 }
 391
 392 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 393 {
 394         struct list_head *head;
 395         struct f2fs_orphan_block *orphan_blk = NULL;
 396         unsigned int nentries = 0;
 397         unsigned short index;
 398         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 399                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 400         struct page *page = NULL;
 401         struct orphan_inode_entry *orphan = NULL;
 402
 403         for (index = 0; index < orphan_blocks; index++)
 404                 grab_meta_page(sbi, start_blk + index);
 405
 406         index = 1;
 407         spin_lock(&sbi->orphan_inode_lock);
 408         head = &sbi->orphan_inode_list;
 409
 410         /* loop for each orphan inode entry and write them in Jornal block */
 411         list_for_each_entry(orphan, head, list) {
 412                 if (!page) {
 413                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 414                         f2fs_bug_on(!page);
 415                         orphan_blk =
 416                                 (struct f2fs_orphan_block *)page_address(page);
 417                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 418                         f2fs_put_page(page, 0);
 419                 }
 420
 421                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 422
 423                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 424                         /*
 425                          * an orphan block is full of 1020 entries,
 426                          * then we need to flush current orphan blocks
 427                          * and bring another one in memory
 428                          */
 429                         orphan_blk->blk_addr = cpu_to_le16(index);
 430                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 431                         orphan_blk->entry_count = cpu_to_le32(nentries);
 432                         set_page_dirty(page);
 433                         f2fs_put_page(page, 1);
 434                         index++;
 435                         nentries = 0;
 436                         page = NULL;
 437                 }
 438         }
 439
 440         if (page) {
 441                 orphan_blk->blk_addr = cpu_to_le16(index);
 442                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 443                 orphan_blk->entry_count = cpu_to_le32(nentries);
 444                 set_page_dirty(page);
 445                 f2fs_put_page(page, 1);
 446         }
 447
 448         spin_unlock(&sbi->orphan_inode_lock);
 449 }
 450
 451 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 452                                 block_t cp_addr, unsigned long long *version)
 453 {
 454         struct page *cp_page_1, *cp_page_2 = NULL;
 455         unsigned long blk_size = sbi->blocksize;
 456         struct f2fs_checkpoint *cp_block;
 457         unsigned long long cur_version = 0, pre_version = 0;
 458         size_t crc_offset;
 459         __u32 crc = 0;
 460
 461         /* Read the 1st cp block in this CP pack */
 462         cp_page_1 = get_meta_page(sbi, cp_addr);
 463
 464         /* get the version number */
 465         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 466         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 467         if (crc_offset >= blk_size)
 468                 goto invalid_cp1;
 469
 470         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 471         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 472                 goto invalid_cp1;
 473
 474         pre_version = cur_cp_version(cp_block);
 475
 476         /* Read the 2nd cp block in this CP pack */
 477         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 478         cp_page_2 = get_meta_page(sbi, cp_addr);
 479
 480         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 481         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 482         if (crc_offset >= blk_size)
 483                 goto invalid_cp2;
 484
 485         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 486         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 487                 goto invalid_cp2;
 488
 489         cur_version = cur_cp_version(cp_block);
 490
 491         if (cur_version == pre_version) {
 492                 *version = cur_version;
 493                 f2fs_put_page(cp_page_2, 1);
 494                 return cp_page_1;
 495         }
 496 invalid_cp2:
 497         f2fs_put_page(cp_page_2, 1);
 498 invalid_cp1:
 499         f2fs_put_page(cp_page_1, 1);
 500         return NULL;
 501 }
 502
 503 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 504 {
 505         struct f2fs_checkpoint *cp_block;
 506         struct f2fs_super_block *fsb = sbi->raw_super;
 507         struct page *cp1, *cp2, *cur_page;
 508         unsigned long blk_size = sbi->blocksize;
 509         unsigned long long cp1_version = 0, cp2_version = 0;
 510         unsigned long long cp_start_blk_no;
 511
 512         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 513         if (!sbi->ckpt)
 514                 return -ENOMEM;
 515         /*
 516          * Finding out valid cp block involves read both
 517          * sets( cp pack1 and cp pack 2)
 518          */
 519         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 520         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 521
 522         /* The second checkpoint pack should start at the next segment */
 523         cp_start_blk_no += ((unsigned long long)1) <<
 524                                 le32_to_cpu(fsb->log_blocks_per_seg);
 525         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 526
 527         if (cp1 && cp2) {
 528                 if (ver_after(cp2_version, cp1_version))
 529                         cur_page = cp2;
 530                 else
 531                         cur_page = cp1;
 532         } else if (cp1) {
 533                 cur_page = cp1;
 534         } else if (cp2) {
 535                 cur_page = cp2;
 536         } else {
 537                 goto fail_no_cp;
 538         }
 539
 540         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 541         memcpy(sbi->ckpt, cp_block, blk_size);
 542
 543         f2fs_put_page(cp1, 1);
 544         f2fs_put_page(cp2, 1);
 545         return 0;
 546
 547 fail_no_cp:
 548         kfree(sbi->ckpt);
 549         return -EINVAL;
 550 }
 551
 552 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 553 {
 554         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 555
 556         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 557                 return -EEXIST;
 558
 559         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 560         F2FS_I(inode)->dirty_dir = new;
 561         list_add_tail(&new->list, &sbi->dir_inode_list);
 562         stat_inc_dirty_dir(sbi);
 563         return 0;
 564 }
 565
 566 void set_dirty_dir_page(struct inode *inode, struct page *page)
 567 {
 568         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 569         struct dir_inode_entry *new;
 570         int ret = 0;
 571
 572         if (!S_ISDIR(inode->i_mode))
 573                 return;
 574
 575         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 576         new->inode = inode;
 577         INIT_LIST_HEAD(&new->list);
 578
 579         spin_lock(&sbi->dir_inode_lock);
 580         ret = __add_dirty_inode(inode, new);
 581         inode_inc_dirty_dents(inode);
 582         SetPagePrivate(page);
 583         spin_unlock(&sbi->dir_inode_lock);
 584
 585         if (ret)
 586                 kmem_cache_free(inode_entry_slab, new);
 587 }
 588
 589 void add_dirty_dir_inode(struct inode *inode)
 590 {
 591         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 592         struct dir_inode_entry *new =
 593                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 594         int ret = 0;
 595
 596         new->inode = inode;
 597         INIT_LIST_HEAD(&new->list);
 598
 599         spin_lock(&sbi->dir_inode_lock);
 600         ret = __add_dirty_inode(inode, new);
 601         spin_unlock(&sbi->dir_inode_lock);
 602
 603         if (ret)
 604                 kmem_cache_free(inode_entry_slab, new);
 605 }
 606
 607 void remove_dirty_dir_inode(struct inode *inode)
 608 {
 609         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 610         struct dir_inode_entry *entry;
 611
 612         if (!S_ISDIR(inode->i_mode))
 613                 return;
 614
 615         spin_lock(&sbi->dir_inode_lock);
 616         if (get_dirty_dents(inode) ||
 617                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 618                 spin_unlock(&sbi->dir_inode_lock);
 619                 return;
 620         }
 621
 622         entry = F2FS_I(inode)->dirty_dir;
 623         list_del(&entry->list);
 624         F2FS_I(inode)->dirty_dir = NULL;
 625         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 626         stat_dec_dirty_dir(sbi);
 627         spin_unlock(&sbi->dir_inode_lock);
 628         kmem_cache_free(inode_entry_slab, entry);
 629
 630         /* Only from the recovery routine */
 631         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 632                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 633                 iput(inode);
 634         }
 635 }
 636
 637 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 638 {
 639         struct list_head *head;
 640         struct dir_inode_entry *entry;
 641         struct inode *inode;
 642 retry:
 643         spin_lock(&sbi->dir_inode_lock);
 644
 645         head = &sbi->dir_inode_list;
 646         if (list_empty(head)) {
 647                 spin_unlock(&sbi->dir_inode_lock);
 648                 return;
 649         }
 650         entry = list_entry(head->next, struct dir_inode_entry, list);
 651         inode = igrab(entry->inode);
 652         spin_unlock(&sbi->dir_inode_lock);
 653         if (inode) {
 654                 filemap_fdatawrite(inode->i_mapping);
 655                 iput(inode);
 656         } else {
 657                 /*
 658                  * We should submit bio, since it exists several
 659                  * wribacking dentry pages in the freeing inode.
 660                  */
 661                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 662         }
 663         goto retry;
 664 }
 665
 666 /*
 667  * Freeze all the FS-operations for checkpoint.
 668  */
 669 static void block_operations(struct f2fs_sb_info *sbi)
 670 {
 671         struct writeback_control wbc = {
 672                 .sync_mode = WB_SYNC_ALL,
 673                 .nr_to_write = LONG_MAX,
 674                 .for_reclaim = 0,
 675         };
 676         struct blk_plug plug;
 677
 678         blk_start_plug(&plug);
 679
 680 retry_flush_dents:
 681         f2fs_lock_all(sbi);
 682         /* write all the dirty dentry pages */
 683         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 684                 f2fs_unlock_all(sbi);
 685                 sync_dirty_dir_inodes(sbi);
 686                 goto retry_flush_dents;
 687         }
 688
 689         /*
 690          * POR: we should ensure that there is no dirty node pages
 691          * until finishing nat/sit flush.
 692          */
 693 retry_flush_nodes:
 694         mutex_lock(&sbi->node_write);
 695
 696         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 697                 mutex_unlock(&sbi->node_write);
 698                 sync_node_pages(sbi, 0, &wbc);
 699                 goto retry_flush_nodes;
 700         }
 701         blk_finish_plug(&plug);
 702 }
 703
 704 static void unblock_operations(struct f2fs_sb_info *sbi)
 705 {
 706         mutex_unlock(&sbi->node_write);
 707         f2fs_unlock_all(sbi);
 708 }
 709
 710 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 711 {
 712         DEFINE_WAIT(wait);
 713
 714         for (;;) {
 715                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 716
 717                 if (!get_pages(sbi, F2FS_WRITEBACK))
 718                         break;
 719
 720                 io_schedule();
 721         }
 722         finish_wait(&sbi->cp_wait, &wait);
 723 }
 724
 725 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 726 {
 727         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 728         nid_t last_nid = 0;
 729         block_t start_blk;
 730         struct page *cp_page;
 731         unsigned int data_sum_blocks, orphan_blocks;
 732         __u32 crc32 = 0;
 733         void *kaddr;
 734         int i;
 735
 736         /*
 737          * This avoids to conduct wrong roll-forward operations and uses
 738          * metapages, so should be called prior to sync_meta_pages below.
 739          */
 740         discard_next_dnode(sbi);
 741
 742         /* Flush all the NAT/SIT pages */
 743         while (get_pages(sbi, F2FS_DIRTY_META))
 744                 sync_meta_pages(sbi, META, LONG_MAX);
 745
 746         next_free_nid(sbi, &last_nid);
 747
 748         /*
 749          * modify checkpoint
 750          * version number is already updated
 751          */
 752         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 753         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 754         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 755         for (i = 0; i < 3; i++) {
 756                 ckpt->cur_node_segno[i] =
 757                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 758                 ckpt->cur_node_blkoff[i] =
 759                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 760                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 761                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 762         }
 763         for (i = 0; i < 3; i++) {
 764                 ckpt->cur_data_segno[i] =
 765                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 766                 ckpt->cur_data_blkoff[i] =
 767                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 768                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 769                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 770         }
 771
 772         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 773         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 774         ckpt->next_free_nid = cpu_to_le32(last_nid);
 775
 776         /* 2 cp  + n data seg summary + orphan inode blocks */
 777         data_sum_blocks = npages_for_summary_flush(sbi);
 778         if (data_sum_blocks < 3)
 779                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 780         else
 781                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 782
 783         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 784                                         / F2FS_ORPHANS_PER_BLOCK;
 785         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 786
 787         if (is_umount) {
 788                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 789                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 790                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 791         } else {
 792                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 793                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 794                         data_sum_blocks + orphan_blocks);
 795         }
 796
 797         if (sbi->n_orphans)
 798                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 799         else
 800                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 801
 802         /* update SIT/NAT bitmap */
 803         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 804         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 805
 806         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 807         *((__le32 *)((unsigned char *)ckpt +
 808                                 le32_to_cpu(ckpt->checksum_offset)))
 809                                 = cpu_to_le32(crc32);
 810
 811         start_blk = __start_cp_addr(sbi);
 812
 813         /* write out checkpoint buffer at block 0 */
 814         cp_page = grab_meta_page(sbi, start_blk++);
 815         kaddr = page_address(cp_page);
 816         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 817         set_page_dirty(cp_page);
 818         f2fs_put_page(cp_page, 1);
 819
 820         if (sbi->n_orphans) {
 821                 write_orphan_inodes(sbi, start_blk);
 822                 start_blk += orphan_blocks;
 823         }
 824
 825         write_data_summaries(sbi, start_blk);
 826         start_blk += data_sum_blocks;
 827         if (is_umount) {
 828                 write_node_summaries(sbi, start_blk);
 829                 start_blk += NR_CURSEG_NODE_TYPE;
 830         }
 831
 832         /* writeout checkpoint block */
 833         cp_page = grab_meta_page(sbi, start_blk);
 834         kaddr = page_address(cp_page);
 835         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 836         set_page_dirty(cp_page);
 837         f2fs_put_page(cp_page, 1);
 838
 839         /* wait for previous submitted node/meta pages writeback */
 840         wait_on_all_pages_writeback(sbi);
 841
 842         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 843         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 844
 845         /* update user_block_counts */
 846         sbi->last_valid_block_count = sbi->total_valid_block_count;
 847         sbi->alloc_valid_block_count = 0;
 848
 849         /* Here, we only have one bio having CP pack */
 850         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 851
 852         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 853                 clear_prefree_segments(sbi);
 854                 F2FS_RESET_SB_DIRT(sbi);
 855         }
 856 }
 857
 858 /*
 859  * We guarantee that this checkpoint procedure should not fail.
 860  */
 861 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 862 {
 863         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 864         unsigned long long ckpt_ver;
 865
 866         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 867
 868         mutex_lock(&sbi->cp_mutex);
 869         block_operations(sbi);
 870
 871         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 872
 873         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 874         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 875         f2fs_submit_merged_bio(sbi, META, WRITE);
 876
 877         /*
 878          * update checkpoint pack index
 879          * Increase the version number so that
 880          * SIT entries and seg summaries are written at correct place
 881          */
 882         ckpt_ver = cur_cp_version(ckpt);
 883         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 884
 885         /* write cached NAT/SIT entries to NAT/SIT area */
 886         flush_nat_entries(sbi);
 887         flush_sit_entries(sbi);
 888
 889         /* unlock all the fs_lock[] in do_checkpoint() */
 890         do_checkpoint(sbi, is_umount);
 891
 892         unblock_operations(sbi);
 893         mutex_unlock(&sbi->cp_mutex);
 894
 895         stat_inc_cp_count(sbi->stat_info);
 896         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 897 }
 898
 899 void init_orphan_info(struct f2fs_sb_info *sbi)
 900 {
 901         spin_lock_init(&sbi->orphan_inode_lock);
 902         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 903         sbi->n_orphans = 0;
 904         /*
 905          * considering 512 blocks in a segment 8 blocks are needed for cp
 906          * and log segment summaries. Remaining blocks are used to keep
 907          * orphan entries with the limitation one reserved segment
 908          * for cp pack we can have max 1020*504 orphan entries
 909          */
 910         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 911                                 * F2FS_ORPHANS_PER_BLOCK;
 912 }
 913
 914 int __init create_checkpoint_caches(void)
 915 {
 916         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 917                         sizeof(struct orphan_inode_entry));
 918         if (!orphan_entry_slab)
 919                 return -ENOMEM;
 920         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 921                         sizeof(struct dir_inode_entry));
 922         if (!inode_entry_slab) {
 923                 kmem_cache_destroy(orphan_entry_slab);
 924                 return -ENOMEM;
 925         }
 926         return 0;
 927 }
 928
 929 void destroy_checkpoint_caches(void)
 930 {
 931         kmem_cache_destroy(orphan_entry_slab);
 932         kmem_cache_destroy(inode_entry_slab);
 933 }