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