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
  24 static struct kmem_cache *orphan_entry_slab;
  25 static struct kmem_cache *inode_entry_slab;
  26
  27 /*
  28  * We guarantee no failure on the returned page.
  29  */
  30 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  31 {
  32         struct address_space *mapping = sbi->meta_inode->i_mapping;
  33         struct page *page = NULL;
  34 repeat:
  35         page = grab_cache_page(mapping, index);
  36         if (!page) {
  37                 cond_resched();
  38                 goto repeat;
  39         }
  40
  41         /* We wait writeback only inside grab_meta_page() */
  42         wait_on_page_writeback(page);
  43         SetPageUptodate(page);
  44         return page;
  45 }
  46
  47 /*
  48  * We guarantee no failure on the returned page.
  49  */
  50 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  51 {
  52         struct address_space *mapping = sbi->meta_inode->i_mapping;
  53         struct page *page;
  54 repeat:
  55         page = grab_cache_page(mapping, index);
  56         if (!page) {
  57                 cond_resched();
  58                 goto repeat;
  59         }
  60         if (f2fs_readpage(sbi, page, index, READ_SYNC)) {
  61                 f2fs_put_page(page, 1);
  62                 goto repeat;
  63         }
  64         mark_page_accessed(page);
  65
  66         /* We do not allow returning an errorneous page */
  67         return page;
  68 }
  69
  70 static int f2fs_write_meta_page(struct page *page,
  71                                 struct writeback_control *wbc)
  72 {
  73         struct inode *inode = page->mapping->host;
  74         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
  75         int err;
  76
  77         wait_on_page_writeback(page);
  78
  79         err = write_meta_page(sbi, page, wbc);
  80         if (err) {
  81                 wbc->pages_skipped++;
  82                 set_page_dirty(page);
  83         }
  84
  85         dec_page_count(sbi, F2FS_DIRTY_META);
  86
  87         /* In this case, we should not unlock this page */
  88         if (err != AOP_WRITEPAGE_ACTIVATE)
  89                 unlock_page(page);
  90         return err;
  91 }
  92
  93 static int f2fs_write_meta_pages(struct address_space *mapping,
  94                                 struct writeback_control *wbc)
  95 {
  96         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  97         struct block_device *bdev = sbi->sb->s_bdev;
  98         long written;
  99
 100         if (wbc->for_kupdate)
 101                 return 0;
 102
 103         if (get_pages(sbi, F2FS_DIRTY_META) == 0)
 104                 return 0;
 105
 106         /* if mounting is failed, skip writing node pages */
 107         mutex_lock(&sbi->cp_mutex);
 108         written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev));
 109         mutex_unlock(&sbi->cp_mutex);
 110         wbc->nr_to_write -= written;
 111         return 0;
 112 }
 113
 114 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 115                                                 long nr_to_write)
 116 {
 117         struct address_space *mapping = sbi->meta_inode->i_mapping;
 118         pgoff_t index = 0, end = LONG_MAX;
 119         struct pagevec pvec;
 120         long nwritten = 0;
 121         struct writeback_control wbc = {
 122                 .for_reclaim = 0,
 123         };
 124
 125         pagevec_init(&pvec, 0);
 126
 127         while (index <= end) {
 128                 int i, nr_pages;
 129                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 130                                 PAGECACHE_TAG_DIRTY,
 131                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 132                 if (nr_pages == 0)
 133                         break;
 134
 135                 for (i = 0; i < nr_pages; i++) {
 136                         struct page *page = pvec.pages[i];
 137                         lock_page(page);
 138                         BUG_ON(page->mapping != mapping);
 139                         BUG_ON(!PageDirty(page));
 140                         clear_page_dirty_for_io(page);
 141                         f2fs_write_meta_page(page, &wbc);
 142                         if (nwritten++ >= nr_to_write)
 143                                 break;
 144                 }
 145                 pagevec_release(&pvec);
 146                 cond_resched();
 147         }
 148
 149         if (nwritten)
 150                 f2fs_submit_bio(sbi, type, nr_to_write == LONG_MAX);
 151
 152         return nwritten;
 153 }
 154
 155 static int f2fs_set_meta_page_dirty(struct page *page)
 156 {
 157         struct address_space *mapping = page->mapping;
 158         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 159
 160         SetPageUptodate(page);
 161         if (!PageDirty(page)) {
 162                 __set_page_dirty_nobuffers(page);
 163                 inc_page_count(sbi, F2FS_DIRTY_META);
 164                 F2FS_SET_SB_DIRT(sbi);
 165                 return 1;
 166         }
 167         return 0;
 168 }
 169
 170 const struct address_space_operations f2fs_meta_aops = {
 171         .writepage      = f2fs_write_meta_page,
 172         .writepages     = f2fs_write_meta_pages,
 173         .set_page_dirty = f2fs_set_meta_page_dirty,
 174 };
 175
 176 int check_orphan_space(struct f2fs_sb_info *sbi)
 177 {
 178         unsigned int max_orphans;
 179         int err = 0;
 180
 181         /*
 182          * considering 512 blocks in a segment 5 blocks are needed for cp
 183          * and log segment summaries. Remaining blocks are used to keep
 184          * orphan entries with the limitation one reserved segment
 185          * for cp pack we can have max 1020*507 orphan entries
 186          */
 187         max_orphans = (sbi->blocks_per_seg - 5) * F2FS_ORPHANS_PER_BLOCK;
 188         mutex_lock(&sbi->orphan_inode_mutex);
 189         if (sbi->n_orphans >= max_orphans)
 190                 err = -ENOSPC;
 191         mutex_unlock(&sbi->orphan_inode_mutex);
 192         return err;
 193 }
 194
 195 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 196 {
 197         struct list_head *head, *this;
 198         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 199
 200         mutex_lock(&sbi->orphan_inode_mutex);
 201         head = &sbi->orphan_inode_list;
 202         list_for_each(this, head) {
 203                 orphan = list_entry(this, struct orphan_inode_entry, list);
 204                 if (orphan->ino == ino)
 205                         goto out;
 206                 if (orphan->ino > ino)
 207                         break;
 208                 orphan = NULL;
 209         }
 210 retry:
 211         new = kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 212         if (!new) {
 213                 cond_resched();
 214                 goto retry;
 215         }
 216         new->ino = ino;
 217         INIT_LIST_HEAD(&new->list);
 218
 219         /* add new_oentry into list which is sorted by inode number */
 220         if (orphan) {
 221                 struct orphan_inode_entry *prev;
 222
 223                 /* get previous entry */
 224                 prev = list_entry(orphan->list.prev, typeof(*prev), list);
 225                 if (&prev->list != head)
 226                         /* insert new orphan inode entry */
 227                         list_add(&new->list, &prev->list);
 228                 else
 229                         list_add(&new->list, head);
 230         } else {
 231                 list_add_tail(&new->list, head);
 232         }
 233         sbi->n_orphans++;
 234 out:
 235         mutex_unlock(&sbi->orphan_inode_mutex);
 236 }
 237
 238 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 239 {
 240         struct list_head *this, *next, *head;
 241         struct orphan_inode_entry *orphan;
 242
 243         mutex_lock(&sbi->orphan_inode_mutex);
 244         head = &sbi->orphan_inode_list;
 245         list_for_each_safe(this, next, head) {
 246                 orphan = list_entry(this, struct orphan_inode_entry, list);
 247                 if (orphan->ino == ino) {
 248                         list_del(&orphan->list);
 249                         kmem_cache_free(orphan_entry_slab, orphan);
 250                         sbi->n_orphans--;
 251                         break;
 252                 }
 253         }
 254         mutex_unlock(&sbi->orphan_inode_mutex);
 255 }
 256
 257 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 258 {
 259         struct inode *inode = f2fs_iget(sbi->sb, ino);
 260         BUG_ON(IS_ERR(inode));
 261         clear_nlink(inode);
 262
 263         /* truncate all the data during iput */
 264         iput(inode);
 265 }
 266
 267 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 268 {
 269         block_t start_blk, orphan_blkaddr, i, j;
 270
 271         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 272                 return 0;
 273
 274         sbi->por_doing = 1;
 275         start_blk = __start_cp_addr(sbi) + 1;
 276         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 277
 278         for (i = 0; i < orphan_blkaddr; i++) {
 279                 struct page *page = get_meta_page(sbi, start_blk + i);
 280                 struct f2fs_orphan_block *orphan_blk;
 281
 282                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 283                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 284                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 285                         recover_orphan_inode(sbi, ino);
 286                 }
 287                 f2fs_put_page(page, 1);
 288         }
 289         /* clear Orphan Flag */
 290         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 291         sbi->por_doing = 0;
 292         return 0;
 293 }
 294
 295 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 296 {
 297         struct list_head *head, *this, *next;
 298         struct f2fs_orphan_block *orphan_blk = NULL;
 299         struct page *page = NULL;
 300         unsigned int nentries = 0;
 301         unsigned short index = 1;
 302         unsigned short orphan_blocks;
 303
 304         orphan_blocks = (unsigned short)((sbi->n_orphans +
 305                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 306
 307         mutex_lock(&sbi->orphan_inode_mutex);
 308         head = &sbi->orphan_inode_list;
 309
 310         /* loop for each orphan inode entry and write them in Jornal block */
 311         list_for_each_safe(this, next, head) {
 312                 struct orphan_inode_entry *orphan;
 313
 314                 orphan = list_entry(this, struct orphan_inode_entry, list);
 315
 316                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 317                         /*
 318                          * an orphan block is full of 1020 entries,
 319                          * then we need to flush current orphan blocks
 320                          * and bring another one in memory
 321                          */
 322                         orphan_blk->blk_addr = cpu_to_le16(index);
 323                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 324                         orphan_blk->entry_count = cpu_to_le32(nentries);
 325                         set_page_dirty(page);
 326                         f2fs_put_page(page, 1);
 327                         index++;
 328                         start_blk++;
 329                         nentries = 0;
 330                         page = NULL;
 331                 }
 332                 if (page)
 333                         goto page_exist;
 334
 335                 page = grab_meta_page(sbi, start_blk);
 336                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 337                 memset(orphan_blk, 0, sizeof(*orphan_blk));
 338 page_exist:
 339                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 340         }
 341         if (!page)
 342                 goto end;
 343
 344         orphan_blk->blk_addr = cpu_to_le16(index);
 345         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 346         orphan_blk->entry_count = cpu_to_le32(nentries);
 347         set_page_dirty(page);
 348         f2fs_put_page(page, 1);
 349 end:
 350         mutex_unlock(&sbi->orphan_inode_mutex);
 351 }
 352
 353 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 354                                 block_t cp_addr, unsigned long long *version)
 355 {
 356         struct page *cp_page_1, *cp_page_2 = NULL;
 357         unsigned long blk_size = sbi->blocksize;
 358         struct f2fs_checkpoint *cp_block;
 359         unsigned long long cur_version = 0, pre_version = 0;
 360         unsigned int crc = 0;
 361         size_t crc_offset;
 362
 363         /* Read the 1st cp block in this CP pack */
 364         cp_page_1 = get_meta_page(sbi, cp_addr);
 365
 366         /* get the version number */
 367         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 368         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 369         if (crc_offset >= blk_size)
 370                 goto invalid_cp1;
 371
 372         crc = *(unsigned int *)((unsigned char *)cp_block + crc_offset);
 373         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 374                 goto invalid_cp1;
 375
 376         pre_version = le64_to_cpu(cp_block->checkpoint_ver);
 377
 378         /* Read the 2nd cp block in this CP pack */
 379         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 380         cp_page_2 = get_meta_page(sbi, cp_addr);
 381
 382         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 383         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 384         if (crc_offset >= blk_size)
 385                 goto invalid_cp2;
 386
 387         crc = *(unsigned int *)((unsigned char *)cp_block + crc_offset);
 388         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 389                 goto invalid_cp2;
 390
 391         cur_version = le64_to_cpu(cp_block->checkpoint_ver);
 392
 393         if (cur_version == pre_version) {
 394                 *version = cur_version;
 395                 f2fs_put_page(cp_page_2, 1);
 396                 return cp_page_1;
 397         }
 398 invalid_cp2:
 399         f2fs_put_page(cp_page_2, 1);
 400 invalid_cp1:
 401         f2fs_put_page(cp_page_1, 1);
 402         return NULL;
 403 }
 404
 405 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 406 {
 407         struct f2fs_checkpoint *cp_block;
 408         struct f2fs_super_block *fsb = sbi->raw_super;
 409         struct page *cp1, *cp2, *cur_page;
 410         unsigned long blk_size = sbi->blocksize;
 411         unsigned long long cp1_version = 0, cp2_version = 0;
 412         unsigned long long cp_start_blk_no;
 413
 414         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 415         if (!sbi->ckpt)
 416                 return -ENOMEM;
 417         /*
 418          * Finding out valid cp block involves read both
 419          * sets( cp pack1 and cp pack 2)
 420          */
 421         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 422         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 423
 424         /* The second checkpoint pack should start at the next segment */
 425         cp_start_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 426         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 427
 428         if (cp1 && cp2) {
 429                 if (ver_after(cp2_version, cp1_version))
 430                         cur_page = cp2;
 431                 else
 432                         cur_page = cp1;
 433         } else if (cp1) {
 434                 cur_page = cp1;
 435         } else if (cp2) {
 436                 cur_page = cp2;
 437         } else {
 438                 goto fail_no_cp;
 439         }
 440
 441         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 442         memcpy(sbi->ckpt, cp_block, blk_size);
 443
 444         f2fs_put_page(cp1, 1);
 445         f2fs_put_page(cp2, 1);
 446         return 0;
 447
 448 fail_no_cp:
 449         kfree(sbi->ckpt);
 450         return -EINVAL;
 451 }
 452
 453 void set_dirty_dir_page(struct inode *inode, struct page *page)
 454 {
 455         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 456         struct list_head *head = &sbi->dir_inode_list;
 457         struct dir_inode_entry *new;
 458         struct list_head *this;
 459
 460         if (!S_ISDIR(inode->i_mode))
 461                 return;
 462 retry:
 463         new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 464         if (!new) {
 465                 cond_resched();
 466                 goto retry;
 467         }
 468         new->inode = inode;
 469         INIT_LIST_HEAD(&new->list);
 470
 471         spin_lock(&sbi->dir_inode_lock);
 472         list_for_each(this, head) {
 473                 struct dir_inode_entry *entry;
 474                 entry = list_entry(this, struct dir_inode_entry, list);
 475                 if (entry->inode == inode) {
 476                         kmem_cache_free(inode_entry_slab, new);
 477                         goto out;
 478                 }
 479         }
 480         list_add_tail(&new->list, head);
 481         sbi->n_dirty_dirs++;
 482
 483         BUG_ON(!S_ISDIR(inode->i_mode));
 484 out:
 485         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 486         inode_inc_dirty_dents(inode);
 487         SetPagePrivate(page);
 488
 489         spin_unlock(&sbi->dir_inode_lock);
 490 }
 491
 492 void remove_dirty_dir_inode(struct inode *inode)
 493 {
 494         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 495         struct list_head *head = &sbi->dir_inode_list;
 496         struct list_head *this;
 497
 498         if (!S_ISDIR(inode->i_mode))
 499                 return;
 500
 501         spin_lock(&sbi->dir_inode_lock);
 502         if (atomic_read(&F2FS_I(inode)->dirty_dents))
 503                 goto out;
 504
 505         list_for_each(this, head) {
 506                 struct dir_inode_entry *entry;
 507                 entry = list_entry(this, struct dir_inode_entry, list);
 508                 if (entry->inode == inode) {
 509                         list_del(&entry->list);
 510                         kmem_cache_free(inode_entry_slab, entry);
 511                         sbi->n_dirty_dirs--;
 512                         break;
 513                 }
 514         }
 515 out:
 516         spin_unlock(&sbi->dir_inode_lock);
 517 }
 518
 519 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 520 {
 521         struct list_head *head = &sbi->dir_inode_list;
 522         struct dir_inode_entry *entry;
 523         struct inode *inode;
 524 retry:
 525         spin_lock(&sbi->dir_inode_lock);
 526         if (list_empty(head)) {
 527                 spin_unlock(&sbi->dir_inode_lock);
 528                 return;
 529         }
 530         entry = list_entry(head->next, struct dir_inode_entry, list);
 531         inode = igrab(entry->inode);
 532         spin_unlock(&sbi->dir_inode_lock);
 533         if (inode) {
 534                 filemap_flush(inode->i_mapping);
 535                 iput(inode);
 536         } else {
 537                 /*
 538                  * We should submit bio, since it exists several
 539                  * wribacking dentry pages in the freeing inode.
 540                  */
 541                 f2fs_submit_bio(sbi, DATA, true);
 542         }
 543         goto retry;
 544 }
 545
 546 /*
 547  * Freeze all the FS-operations for checkpoint.
 548  */
 549 void block_operations(struct f2fs_sb_info *sbi)
 550 {
 551         int t;
 552         struct writeback_control wbc = {
 553                 .sync_mode = WB_SYNC_ALL,
 554                 .nr_to_write = LONG_MAX,
 555                 .for_reclaim = 0,
 556         };
 557
 558         /* Stop renaming operation */
 559         mutex_lock_op(sbi, RENAME);
 560         mutex_lock_op(sbi, DENTRY_OPS);
 561
 562 retry_dents:
 563         /* write all the dirty dentry pages */
 564         sync_dirty_dir_inodes(sbi);
 565
 566         mutex_lock_op(sbi, DATA_WRITE);
 567         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 568                 mutex_unlock_op(sbi, DATA_WRITE);
 569                 goto retry_dents;
 570         }
 571
 572         /* block all the operations */
 573         for (t = DATA_NEW; t <= NODE_TRUNC; t++)
 574                 mutex_lock_op(sbi, t);
 575
 576         mutex_lock(&sbi->write_inode);
 577
 578         /*
 579          * POR: we should ensure that there is no dirty node pages
 580          * until finishing nat/sit flush.
 581          */
 582 retry:
 583         sync_node_pages(sbi, 0, &wbc);
 584
 585         mutex_lock_op(sbi, NODE_WRITE);
 586
 587         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 588                 mutex_unlock_op(sbi, NODE_WRITE);
 589                 goto retry;
 590         }
 591         mutex_unlock(&sbi->write_inode);
 592 }
 593
 594 static void unblock_operations(struct f2fs_sb_info *sbi)
 595 {
 596         int t;
 597         for (t = NODE_WRITE; t >= RENAME; t--)
 598                 mutex_unlock_op(sbi, t);
 599 }
 600
 601 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 602 {
 603         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 604         nid_t last_nid = 0;
 605         block_t start_blk;
 606         struct page *cp_page;
 607         unsigned int data_sum_blocks, orphan_blocks;
 608         unsigned int crc32 = 0;
 609         void *kaddr;
 610         int i;
 611
 612         /* Flush all the NAT/SIT pages */
 613         while (get_pages(sbi, F2FS_DIRTY_META))
 614                 sync_meta_pages(sbi, META, LONG_MAX);
 615
 616         next_free_nid(sbi, &last_nid);
 617
 618         /*
 619          * modify checkpoint
 620          * version number is already updated
 621          */
 622         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 623         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 624         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 625         for (i = 0; i < 3; i++) {
 626                 ckpt->cur_node_segno[i] =
 627                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 628                 ckpt->cur_node_blkoff[i] =
 629                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 630                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 631                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 632         }
 633         for (i = 0; i < 3; i++) {
 634                 ckpt->cur_data_segno[i] =
 635                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 636                 ckpt->cur_data_blkoff[i] =
 637                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 638                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 639                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 640         }
 641
 642         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 643         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 644         ckpt->next_free_nid = cpu_to_le32(last_nid);
 645
 646         /* 2 cp  + n data seg summary + orphan inode blocks */
 647         data_sum_blocks = npages_for_summary_flush(sbi);
 648         if (data_sum_blocks < 3)
 649                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 650         else
 651                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 652
 653         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 654                                         / F2FS_ORPHANS_PER_BLOCK;
 655         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 656
 657         if (is_umount) {
 658                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 659                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 660                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 661         } else {
 662                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 663                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 664                         data_sum_blocks + orphan_blocks);
 665         }
 666
 667         if (sbi->n_orphans)
 668                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 669         else
 670                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 671
 672         /* update SIT/NAT bitmap */
 673         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 674         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 675
 676         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 677         *(__le32 *)((unsigned char *)ckpt +
 678                                 le32_to_cpu(ckpt->checksum_offset))
 679                                 = cpu_to_le32(crc32);
 680
 681         start_blk = __start_cp_addr(sbi);
 682
 683         /* write out checkpoint buffer at block 0 */
 684         cp_page = grab_meta_page(sbi, start_blk++);
 685         kaddr = page_address(cp_page);
 686         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 687         set_page_dirty(cp_page);
 688         f2fs_put_page(cp_page, 1);
 689
 690         if (sbi->n_orphans) {
 691                 write_orphan_inodes(sbi, start_blk);
 692                 start_blk += orphan_blocks;
 693         }
 694
 695         write_data_summaries(sbi, start_blk);
 696         start_blk += data_sum_blocks;
 697         if (is_umount) {
 698                 write_node_summaries(sbi, start_blk);
 699                 start_blk += NR_CURSEG_NODE_TYPE;
 700         }
 701
 702         /* writeout checkpoint block */
 703         cp_page = grab_meta_page(sbi, start_blk);
 704         kaddr = page_address(cp_page);
 705         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 706         set_page_dirty(cp_page);
 707         f2fs_put_page(cp_page, 1);
 708
 709         /* wait for previous submitted node/meta pages writeback */
 710         while (get_pages(sbi, F2FS_WRITEBACK))
 711                 congestion_wait(BLK_RW_ASYNC, HZ / 50);
 712
 713         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 714         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 715
 716         /* update user_block_counts */
 717         sbi->last_valid_block_count = sbi->total_valid_block_count;
 718         sbi->alloc_valid_block_count = 0;
 719
 720         /* Here, we only have one bio having CP pack */
 721         if (is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))
 722                 sbi->sb->s_flags |= MS_RDONLY;
 723         else
 724                 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 725
 726         clear_prefree_segments(sbi);
 727         F2FS_RESET_SB_DIRT(sbi);
 728 }
 729
 730 /*
 731  * We guarantee that this checkpoint procedure should not fail.
 732  */
 733 void write_checkpoint(struct f2fs_sb_info *sbi, bool blocked, bool is_umount)
 734 {
 735         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 736         unsigned long long ckpt_ver;
 737
 738         if (!blocked) {
 739                 mutex_lock(&sbi->cp_mutex);
 740                 block_operations(sbi);
 741         }
 742
 743         f2fs_submit_bio(sbi, DATA, true);
 744         f2fs_submit_bio(sbi, NODE, true);
 745         f2fs_submit_bio(sbi, META, true);
 746
 747         /*
 748          * update checkpoint pack index
 749          * Increase the version number so that
 750          * SIT entries and seg summaries are written at correct place
 751          */
 752         ckpt_ver = le64_to_cpu(ckpt->checkpoint_ver);
 753         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 754
 755         /* write cached NAT/SIT entries to NAT/SIT area */
 756         flush_nat_entries(sbi);
 757         flush_sit_entries(sbi);
 758
 759         reset_victim_segmap(sbi);
 760
 761         /* unlock all the fs_lock[] in do_checkpoint() */
 762         do_checkpoint(sbi, is_umount);
 763
 764         unblock_operations(sbi);
 765         mutex_unlock(&sbi->cp_mutex);
 766 }
 767
 768 void init_orphan_info(struct f2fs_sb_info *sbi)
 769 {
 770         mutex_init(&sbi->orphan_inode_mutex);
 771         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 772         sbi->n_orphans = 0;
 773 }
 774
 775 int create_checkpoint_caches(void)
 776 {
 777         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 778                         sizeof(struct orphan_inode_entry), NULL);
 779         if (unlikely(!orphan_entry_slab))
 780                 return -ENOMEM;
 781         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 782                         sizeof(struct dir_inode_entry), NULL);
 783         if (unlikely(!inode_entry_slab)) {
 784                 kmem_cache_destroy(orphan_entry_slab);
 785                 return -ENOMEM;
 786         }
 787         return 0;
 788 }
 789
 790 void destroy_checkpoint_caches(void)
 791 {
 792         kmem_cache_destroy(orphan_entry_slab);
 793         kmem_cache_destroy(inode_entry_slab);
 794 }