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
 218         /* add new_oentry into list which is sorted by inode number */
 219         if (orphan) {
 220                 struct orphan_inode_entry *prev;
 221
 222                 /* get previous entry */
 223                 prev = list_entry(orphan->list.prev, typeof(*prev), list);
 224                 if (&prev->list != head)
 225                         /* insert new orphan inode entry */
 226                         list_add(&new->list, &prev->list);
 227                 else
 228                         list_add(&new->list, head);
 229         } else {
 230                 list_add_tail(&new->list, head);
 231         }
 232         sbi->n_orphans++;
 233 out:
 234         mutex_unlock(&sbi->orphan_inode_mutex);
 235 }
 236
 237 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 238 {
 239         struct list_head *this, *next, *head;
 240         struct orphan_inode_entry *orphan;
 241
 242         mutex_lock(&sbi->orphan_inode_mutex);
 243         head = &sbi->orphan_inode_list;
 244         list_for_each_safe(this, next, head) {
 245                 orphan = list_entry(this, struct orphan_inode_entry, list);
 246                 if (orphan->ino == ino) {
 247                         list_del(&orphan->list);
 248                         kmem_cache_free(orphan_entry_slab, orphan);
 249                         sbi->n_orphans--;
 250                         break;
 251                 }
 252         }
 253         mutex_unlock(&sbi->orphan_inode_mutex);
 254 }
 255
 256 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 257 {
 258         struct inode *inode = f2fs_iget(sbi->sb, ino);
 259         BUG_ON(IS_ERR(inode));
 260         clear_nlink(inode);
 261
 262         /* truncate all the data during iput */
 263         iput(inode);
 264 }
 265
 266 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 267 {
 268         block_t start_blk, orphan_blkaddr, i, j;
 269
 270         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 271                 return 0;
 272
 273         sbi->por_doing = 1;
 274         start_blk = __start_cp_addr(sbi) + 1;
 275         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 276
 277         for (i = 0; i < orphan_blkaddr; i++) {
 278                 struct page *page = get_meta_page(sbi, start_blk + i);
 279                 struct f2fs_orphan_block *orphan_blk;
 280
 281                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 282                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 283                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 284                         recover_orphan_inode(sbi, ino);
 285                 }
 286                 f2fs_put_page(page, 1);
 287         }
 288         /* clear Orphan Flag */
 289         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 290         sbi->por_doing = 0;
 291         return 0;
 292 }
 293
 294 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 295 {
 296         struct list_head *head, *this, *next;
 297         struct f2fs_orphan_block *orphan_blk = NULL;
 298         struct page *page = NULL;
 299         unsigned int nentries = 0;
 300         unsigned short index = 1;
 301         unsigned short orphan_blocks;
 302
 303         orphan_blocks = (unsigned short)((sbi->n_orphans +
 304                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 305
 306         mutex_lock(&sbi->orphan_inode_mutex);
 307         head = &sbi->orphan_inode_list;
 308
 309         /* loop for each orphan inode entry and write them in Jornal block */
 310         list_for_each_safe(this, next, head) {
 311                 struct orphan_inode_entry *orphan;
 312
 313                 orphan = list_entry(this, struct orphan_inode_entry, list);
 314
 315                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 316                         /*
 317                          * an orphan block is full of 1020 entries,
 318                          * then we need to flush current orphan blocks
 319                          * and bring another one in memory
 320                          */
 321                         orphan_blk->blk_addr = cpu_to_le16(index);
 322                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 323                         orphan_blk->entry_count = cpu_to_le32(nentries);
 324                         set_page_dirty(page);
 325                         f2fs_put_page(page, 1);
 326                         index++;
 327                         start_blk++;
 328                         nentries = 0;
 329                         page = NULL;
 330                 }
 331                 if (page)
 332                         goto page_exist;
 333
 334                 page = grab_meta_page(sbi, start_blk);
 335                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 336                 memset(orphan_blk, 0, sizeof(*orphan_blk));
 337 page_exist:
 338                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 339         }
 340         if (!page)
 341                 goto end;
 342
 343         orphan_blk->blk_addr = cpu_to_le16(index);
 344         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 345         orphan_blk->entry_count = cpu_to_le32(nentries);
 346         set_page_dirty(page);
 347         f2fs_put_page(page, 1);
 348 end:
 349         mutex_unlock(&sbi->orphan_inode_mutex);
 350 }
 351
 352 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 353                                 block_t cp_addr, unsigned long long *version)
 354 {
 355         struct page *cp_page_1, *cp_page_2 = NULL;
 356         unsigned long blk_size = sbi->blocksize;
 357         struct f2fs_checkpoint *cp_block;
 358         unsigned long long cur_version = 0, pre_version = 0;
 359         unsigned int crc = 0;
 360         size_t crc_offset;
 361
 362         /* Read the 1st cp block in this CP pack */
 363         cp_page_1 = get_meta_page(sbi, cp_addr);
 364
 365         /* get the version number */
 366         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 367         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 368         if (crc_offset >= blk_size)
 369                 goto invalid_cp1;
 370
 371         crc = *(unsigned int *)((unsigned char *)cp_block + crc_offset);
 372         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 373                 goto invalid_cp1;
 374
 375         pre_version = le64_to_cpu(cp_block->checkpoint_ver);
 376
 377         /* Read the 2nd cp block in this CP pack */
 378         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 379         cp_page_2 = get_meta_page(sbi, cp_addr);
 380
 381         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 382         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 383         if (crc_offset >= blk_size)
 384                 goto invalid_cp2;
 385
 386         crc = *(unsigned int *)((unsigned char *)cp_block + crc_offset);
 387         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 388                 goto invalid_cp2;
 389
 390         cur_version = le64_to_cpu(cp_block->checkpoint_ver);
 391
 392         if (cur_version == pre_version) {
 393                 *version = cur_version;
 394                 f2fs_put_page(cp_page_2, 1);
 395                 return cp_page_1;
 396         }
 397 invalid_cp2:
 398         f2fs_put_page(cp_page_2, 1);
 399 invalid_cp1:
 400         f2fs_put_page(cp_page_1, 1);
 401         return NULL;
 402 }
 403
 404 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 405 {
 406         struct f2fs_checkpoint *cp_block;
 407         struct f2fs_super_block *fsb = sbi->raw_super;
 408         struct page *cp1, *cp2, *cur_page;
 409         unsigned long blk_size = sbi->blocksize;
 410         unsigned long long cp1_version = 0, cp2_version = 0;
 411         unsigned long long cp_start_blk_no;
 412
 413         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 414         if (!sbi->ckpt)
 415                 return -ENOMEM;
 416         /*
 417          * Finding out valid cp block involves read both
 418          * sets( cp pack1 and cp pack 2)
 419          */
 420         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 421         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 422
 423         /* The second checkpoint pack should start at the next segment */
 424         cp_start_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 425         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 426
 427         if (cp1 && cp2) {
 428                 if (ver_after(cp2_version, cp1_version))
 429                         cur_page = cp2;
 430                 else
 431                         cur_page = cp1;
 432         } else if (cp1) {
 433                 cur_page = cp1;
 434         } else if (cp2) {
 435                 cur_page = cp2;
 436         } else {
 437                 goto fail_no_cp;
 438         }
 439
 440         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 441         memcpy(sbi->ckpt, cp_block, blk_size);
 442
 443         f2fs_put_page(cp1, 1);
 444         f2fs_put_page(cp2, 1);
 445         return 0;
 446
 447 fail_no_cp:
 448         kfree(sbi->ckpt);
 449         return -EINVAL;
 450 }
 451
 452 void set_dirty_dir_page(struct inode *inode, struct page *page)
 453 {
 454         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 455         struct list_head *head = &sbi->dir_inode_list;
 456         struct dir_inode_entry *new;
 457         struct list_head *this;
 458
 459         if (!S_ISDIR(inode->i_mode))
 460                 return;
 461 retry:
 462         new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 463         if (!new) {
 464                 cond_resched();
 465                 goto retry;
 466         }
 467         new->inode = inode;
 468         INIT_LIST_HEAD(&new->list);
 469
 470         spin_lock(&sbi->dir_inode_lock);
 471         list_for_each(this, head) {
 472                 struct dir_inode_entry *entry;
 473                 entry = list_entry(this, struct dir_inode_entry, list);
 474                 if (entry->inode == inode) {
 475                         kmem_cache_free(inode_entry_slab, new);
 476                         goto out;
 477                 }
 478         }
 479         list_add_tail(&new->list, head);
 480         sbi->n_dirty_dirs++;
 481
 482         BUG_ON(!S_ISDIR(inode->i_mode));
 483 out:
 484         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 485         inode_inc_dirty_dents(inode);
 486         SetPagePrivate(page);
 487
 488         spin_unlock(&sbi->dir_inode_lock);
 489 }
 490
 491 void remove_dirty_dir_inode(struct inode *inode)
 492 {
 493         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 494         struct list_head *head = &sbi->dir_inode_list;
 495         struct list_head *this;
 496
 497         if (!S_ISDIR(inode->i_mode))
 498                 return;
 499
 500         spin_lock(&sbi->dir_inode_lock);
 501         if (atomic_read(&F2FS_I(inode)->dirty_dents))
 502                 goto out;
 503
 504         list_for_each(this, head) {
 505                 struct dir_inode_entry *entry;
 506                 entry = list_entry(this, struct dir_inode_entry, list);
 507                 if (entry->inode == inode) {
 508                         list_del(&entry->list);
 509                         kmem_cache_free(inode_entry_slab, entry);
 510                         sbi->n_dirty_dirs--;
 511                         break;
 512                 }
 513         }
 514 out:
 515         spin_unlock(&sbi->dir_inode_lock);
 516 }
 517
 518 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 519 {
 520         struct list_head *head = &sbi->dir_inode_list;
 521         struct dir_inode_entry *entry;
 522         struct inode *inode;
 523 retry:
 524         spin_lock(&sbi->dir_inode_lock);
 525         if (list_empty(head)) {
 526                 spin_unlock(&sbi->dir_inode_lock);
 527                 return;
 528         }
 529         entry = list_entry(head->next, struct dir_inode_entry, list);
 530         inode = igrab(entry->inode);
 531         spin_unlock(&sbi->dir_inode_lock);
 532         if (inode) {
 533                 filemap_flush(inode->i_mapping);
 534                 iput(inode);
 535         } else {
 536                 /*
 537                  * We should submit bio, since it exists several
 538                  * wribacking dentry pages in the freeing inode.
 539                  */
 540                 f2fs_submit_bio(sbi, DATA, true);
 541         }
 542         goto retry;
 543 }
 544
 545 /*
 546  * Freeze all the FS-operations for checkpoint.
 547  */
 548 void block_operations(struct f2fs_sb_info *sbi)
 549 {
 550         int t;
 551         struct writeback_control wbc = {
 552                 .sync_mode = WB_SYNC_ALL,
 553                 .nr_to_write = LONG_MAX,
 554                 .for_reclaim = 0,
 555         };
 556
 557         /* Stop renaming operation */
 558         mutex_lock_op(sbi, RENAME);
 559         mutex_lock_op(sbi, DENTRY_OPS);
 560
 561 retry_dents:
 562         /* write all the dirty dentry pages */
 563         sync_dirty_dir_inodes(sbi);
 564
 565         mutex_lock_op(sbi, DATA_WRITE);
 566         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 567                 mutex_unlock_op(sbi, DATA_WRITE);
 568                 goto retry_dents;
 569         }
 570
 571         /* block all the operations */
 572         for (t = DATA_NEW; t <= NODE_TRUNC; t++)
 573                 mutex_lock_op(sbi, t);
 574
 575         mutex_lock(&sbi->write_inode);
 576
 577         /*
 578          * POR: we should ensure that there is no dirty node pages
 579          * until finishing nat/sit flush.
 580          */
 581 retry:
 582         sync_node_pages(sbi, 0, &wbc);
 583
 584         mutex_lock_op(sbi, NODE_WRITE);
 585
 586         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 587                 mutex_unlock_op(sbi, NODE_WRITE);
 588                 goto retry;
 589         }
 590         mutex_unlock(&sbi->write_inode);
 591 }
 592
 593 static void unblock_operations(struct f2fs_sb_info *sbi)
 594 {
 595         int t;
 596         for (t = NODE_WRITE; t >= RENAME; t--)
 597                 mutex_unlock_op(sbi, t);
 598 }
 599
 600 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 601 {
 602         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 603         nid_t last_nid = 0;
 604         block_t start_blk;
 605         struct page *cp_page;
 606         unsigned int data_sum_blocks, orphan_blocks;
 607         unsigned int crc32 = 0;
 608         void *kaddr;
 609         int i;
 610
 611         /* Flush all the NAT/SIT pages */
 612         while (get_pages(sbi, F2FS_DIRTY_META))
 613                 sync_meta_pages(sbi, META, LONG_MAX);
 614
 615         next_free_nid(sbi, &last_nid);
 616
 617         /*
 618          * modify checkpoint
 619          * version number is already updated
 620          */
 621         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 622         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 623         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 624         for (i = 0; i < 3; i++) {
 625                 ckpt->cur_node_segno[i] =
 626                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 627                 ckpt->cur_node_blkoff[i] =
 628                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 629                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 630                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 631         }
 632         for (i = 0; i < 3; i++) {
 633                 ckpt->cur_data_segno[i] =
 634                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 635                 ckpt->cur_data_blkoff[i] =
 636                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 637                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 638                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 639         }
 640
 641         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 642         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 643         ckpt->next_free_nid = cpu_to_le32(last_nid);
 644
 645         /* 2 cp  + n data seg summary + orphan inode blocks */
 646         data_sum_blocks = npages_for_summary_flush(sbi);
 647         if (data_sum_blocks < 3)
 648                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 649         else
 650                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 651
 652         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 653                                         / F2FS_ORPHANS_PER_BLOCK;
 654         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 655
 656         if (is_umount) {
 657                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 658                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 659                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 660         } else {
 661                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 662                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 663                         data_sum_blocks + orphan_blocks);
 664         }
 665
 666         if (sbi->n_orphans)
 667                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 668         else
 669                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 670
 671         /* update SIT/NAT bitmap */
 672         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 673         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 674
 675         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 676         *(__le32 *)((unsigned char *)ckpt +
 677                                 le32_to_cpu(ckpt->checksum_offset))
 678                                 = cpu_to_le32(crc32);
 679
 680         start_blk = __start_cp_addr(sbi);
 681
 682         /* write out checkpoint buffer at block 0 */
 683         cp_page = grab_meta_page(sbi, start_blk++);
 684         kaddr = page_address(cp_page);
 685         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 686         set_page_dirty(cp_page);
 687         f2fs_put_page(cp_page, 1);
 688
 689         if (sbi->n_orphans) {
 690                 write_orphan_inodes(sbi, start_blk);
 691                 start_blk += orphan_blocks;
 692         }
 693
 694         write_data_summaries(sbi, start_blk);
 695         start_blk += data_sum_blocks;
 696         if (is_umount) {
 697                 write_node_summaries(sbi, start_blk);
 698                 start_blk += NR_CURSEG_NODE_TYPE;
 699         }
 700
 701         /* writeout checkpoint block */
 702         cp_page = grab_meta_page(sbi, start_blk);
 703         kaddr = page_address(cp_page);
 704         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 705         set_page_dirty(cp_page);
 706         f2fs_put_page(cp_page, 1);
 707
 708         /* wait for previous submitted node/meta pages writeback */
 709         while (get_pages(sbi, F2FS_WRITEBACK))
 710                 congestion_wait(BLK_RW_ASYNC, HZ / 50);
 711
 712         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 713         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 714
 715         /* update user_block_counts */
 716         sbi->last_valid_block_count = sbi->total_valid_block_count;
 717         sbi->alloc_valid_block_count = 0;
 718
 719         /* Here, we only have one bio having CP pack */
 720         if (is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))
 721                 sbi->sb->s_flags |= MS_RDONLY;
 722         else
 723                 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 724
 725         clear_prefree_segments(sbi);
 726         F2FS_RESET_SB_DIRT(sbi);
 727 }
 728
 729 /*
 730  * We guarantee that this checkpoint procedure should not fail.
 731  */
 732 void write_checkpoint(struct f2fs_sb_info *sbi, bool blocked, bool is_umount)
 733 {
 734         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 735         unsigned long long ckpt_ver;
 736
 737         if (!blocked) {
 738                 mutex_lock(&sbi->cp_mutex);
 739                 block_operations(sbi);
 740         }
 741
 742         f2fs_submit_bio(sbi, DATA, true);
 743         f2fs_submit_bio(sbi, NODE, true);
 744         f2fs_submit_bio(sbi, META, true);
 745
 746         /*
 747          * update checkpoint pack index
 748          * Increase the version number so that
 749          * SIT entries and seg summaries are written at correct place
 750          */
 751         ckpt_ver = le64_to_cpu(ckpt->checkpoint_ver);
 752         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 753
 754         /* write cached NAT/SIT entries to NAT/SIT area */
 755         flush_nat_entries(sbi);
 756         flush_sit_entries(sbi);
 757
 758         reset_victim_segmap(sbi);
 759
 760         /* unlock all the fs_lock[] in do_checkpoint() */
 761         do_checkpoint(sbi, is_umount);
 762
 763         unblock_operations(sbi);
 764         mutex_unlock(&sbi->cp_mutex);
 765 }
 766
 767 void init_orphan_info(struct f2fs_sb_info *sbi)
 768 {
 769         mutex_init(&sbi->orphan_inode_mutex);
 770         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 771         sbi->n_orphans = 0;
 772 }
 773
 774 int __init create_checkpoint_caches(void)
 775 {
 776         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 777                         sizeof(struct orphan_inode_entry), NULL);
 778         if (unlikely(!orphan_entry_slab))
 779                 return -ENOMEM;
 780         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 781                         sizeof(struct dir_inode_entry), NULL);
 782         if (unlikely(!inode_entry_slab)) {
 783                 kmem_cache_destroy(orphan_entry_slab);
 784                 return -ENOMEM;
 785         }
 786         return 0;
 787 }
 788
 789 void destroy_checkpoint_caches(void)
 790 {
 791         kmem_cache_destroy(orphan_entry_slab);
 792         kmem_cache_destroy(inode_entry_slab);
 793 }