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 = sbi->meta_inode->i_mapping;
  34         struct page *page = NULL;
  35 repeat:
  36         page = grab_cache_page(mapping, index);
  37         if (!page) {
  38                 cond_resched();
  39                 goto repeat;
  40         }
  41
  42         /* We wait writeback only inside grab_meta_page() */
  43         wait_on_page_writeback(page);
  44         SetPageUptodate(page);
  45         return page;
  46 }
  47
  48 /*
  49  * We guarantee no failure on the returned page.
  50  */
  51 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  52 {
  53         struct address_space *mapping = sbi->meta_inode->i_mapping;
  54         struct page *page;
  55 repeat:
  56         page = grab_cache_page(mapping, index);
  57         if (!page) {
  58                 cond_resched();
  59                 goto repeat;
  60         }
  61         if (PageUptodate(page))
  62                 goto out;
  63
  64         if (f2fs_readpage(sbi, page, index, READ_SYNC))
  65                 goto repeat;
  66
  67         lock_page(page);
  68         if (page->mapping != mapping) {
  69                 f2fs_put_page(page, 1);
  70                 goto repeat;
  71         }
  72 out:
  73         mark_page_accessed(page);
  74         return page;
  75 }
  76
  77 static int f2fs_write_meta_page(struct page *page,
  78                                 struct writeback_control *wbc)
  79 {
  80         struct inode *inode = page->mapping->host;
  81         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
  82
  83         /* Should not write any meta pages, if any IO error was occurred */
  84         if (wbc->for_reclaim ||
  85                         is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)) {
  86                 dec_page_count(sbi, F2FS_DIRTY_META);
  87                 wbc->pages_skipped++;
  88                 set_page_dirty(page);
  89                 return AOP_WRITEPAGE_ACTIVATE;
  90         }
  91
  92         wait_on_page_writeback(page);
  93
  94         write_meta_page(sbi, page);
  95         dec_page_count(sbi, F2FS_DIRTY_META);
  96         unlock_page(page);
  97         return 0;
  98 }
  99
 100 static int f2fs_write_meta_pages(struct address_space *mapping,
 101                                 struct writeback_control *wbc)
 102 {
 103         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 104         struct block_device *bdev = sbi->sb->s_bdev;
 105         long written;
 106
 107         if (wbc->for_kupdate)
 108                 return 0;
 109
 110         if (get_pages(sbi, F2FS_DIRTY_META) == 0)
 111                 return 0;
 112
 113         /* if mounting is failed, skip writing node pages */
 114         mutex_lock(&sbi->cp_mutex);
 115         written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev));
 116         mutex_unlock(&sbi->cp_mutex);
 117         wbc->nr_to_write -= written;
 118         return 0;
 119 }
 120
 121 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 122                                                 long nr_to_write)
 123 {
 124         struct address_space *mapping = sbi->meta_inode->i_mapping;
 125         pgoff_t index = 0, end = LONG_MAX;
 126         struct pagevec pvec;
 127         long nwritten = 0;
 128         struct writeback_control wbc = {
 129                 .for_reclaim = 0,
 130         };
 131
 132         pagevec_init(&pvec, 0);
 133
 134         while (index <= end) {
 135                 int i, nr_pages;
 136                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 137                                 PAGECACHE_TAG_DIRTY,
 138                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 139                 if (nr_pages == 0)
 140                         break;
 141
 142                 for (i = 0; i < nr_pages; i++) {
 143                         struct page *page = pvec.pages[i];
 144                         lock_page(page);
 145                         BUG_ON(page->mapping != mapping);
 146                         BUG_ON(!PageDirty(page));
 147                         clear_page_dirty_for_io(page);
 148                         if (f2fs_write_meta_page(page, &wbc)) {
 149                                 unlock_page(page);
 150                                 break;
 151                         }
 152                         if (nwritten++ >= nr_to_write)
 153                                 break;
 154                 }
 155                 pagevec_release(&pvec);
 156                 cond_resched();
 157         }
 158
 159         if (nwritten)
 160                 f2fs_submit_bio(sbi, type, nr_to_write == LONG_MAX);
 161
 162         return nwritten;
 163 }
 164
 165 static int f2fs_set_meta_page_dirty(struct page *page)
 166 {
 167         struct address_space *mapping = page->mapping;
 168         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 169
 170         SetPageUptodate(page);
 171         if (!PageDirty(page)) {
 172                 __set_page_dirty_nobuffers(page);
 173                 inc_page_count(sbi, F2FS_DIRTY_META);
 174                 return 1;
 175         }
 176         return 0;
 177 }
 178
 179 const struct address_space_operations f2fs_meta_aops = {
 180         .writepage      = f2fs_write_meta_page,
 181         .writepages     = f2fs_write_meta_pages,
 182         .set_page_dirty = f2fs_set_meta_page_dirty,
 183 };
 184
 185 int check_orphan_space(struct f2fs_sb_info *sbi)
 186 {
 187         unsigned int max_orphans;
 188         int err = 0;
 189
 190         /*
 191          * considering 512 blocks in a segment 5 blocks are needed for cp
 192          * and log segment summaries. Remaining blocks are used to keep
 193          * orphan entries with the limitation one reserved segment
 194          * for cp pack we can have max 1020*507 orphan entries
 195          */
 196         max_orphans = (sbi->blocks_per_seg - 5) * F2FS_ORPHANS_PER_BLOCK;
 197         mutex_lock(&sbi->orphan_inode_mutex);
 198         if (sbi->n_orphans >= max_orphans)
 199                 err = -ENOSPC;
 200         mutex_unlock(&sbi->orphan_inode_mutex);
 201         return err;
 202 }
 203
 204 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 205 {
 206         struct list_head *head, *this;
 207         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 208
 209         mutex_lock(&sbi->orphan_inode_mutex);
 210         head = &sbi->orphan_inode_list;
 211         list_for_each(this, head) {
 212                 orphan = list_entry(this, struct orphan_inode_entry, list);
 213                 if (orphan->ino == ino)
 214                         goto out;
 215                 if (orphan->ino > ino)
 216                         break;
 217                 orphan = NULL;
 218         }
 219 retry:
 220         new = kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 221         if (!new) {
 222                 cond_resched();
 223                 goto retry;
 224         }
 225         new->ino = ino;
 226
 227         /* add new_oentry into list which is sorted by inode number */
 228         if (orphan)
 229                 list_add(&new->list, this->prev);
 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         size_t crc_offset;
 361         __u32 crc = 0;
 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 = le32_to_cpu(*((__u32 *)((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 = le32_to_cpu(*((__u32 *)((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 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 454 {
 455         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 456         struct list_head *head = &sbi->dir_inode_list;
 457         struct list_head *this;
 458
 459         list_for_each(this, head) {
 460                 struct dir_inode_entry *entry;
 461                 entry = list_entry(this, struct dir_inode_entry, list);
 462                 if (entry->inode == inode)
 463                         return -EEXIST;
 464         }
 465         list_add_tail(&new->list, head);
 466 #ifdef CONFIG_F2FS_STAT_FS
 467         sbi->n_dirty_dirs++;
 468 #endif
 469         return 0;
 470 }
 471
 472 void set_dirty_dir_page(struct inode *inode, struct page *page)
 473 {
 474         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 475         struct dir_inode_entry *new;
 476
 477         if (!S_ISDIR(inode->i_mode))
 478                 return;
 479 retry:
 480         new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 481         if (!new) {
 482                 cond_resched();
 483                 goto retry;
 484         }
 485         new->inode = inode;
 486         INIT_LIST_HEAD(&new->list);
 487
 488         spin_lock(&sbi->dir_inode_lock);
 489         if (__add_dirty_inode(inode, new))
 490                 kmem_cache_free(inode_entry_slab, new);
 491
 492         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 493         inode_inc_dirty_dents(inode);
 494         SetPagePrivate(page);
 495         spin_unlock(&sbi->dir_inode_lock);
 496 }
 497
 498 void add_dirty_dir_inode(struct inode *inode)
 499 {
 500         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 501         struct dir_inode_entry *new;
 502 retry:
 503         new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 504         if (!new) {
 505                 cond_resched();
 506                 goto retry;
 507         }
 508         new->inode = inode;
 509         INIT_LIST_HEAD(&new->list);
 510
 511         spin_lock(&sbi->dir_inode_lock);
 512         if (__add_dirty_inode(inode, new))
 513                 kmem_cache_free(inode_entry_slab, new);
 514         spin_unlock(&sbi->dir_inode_lock);
 515 }
 516
 517 void remove_dirty_dir_inode(struct inode *inode)
 518 {
 519         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 520         struct list_head *head = &sbi->dir_inode_list;
 521         struct list_head *this;
 522
 523         if (!S_ISDIR(inode->i_mode))
 524                 return;
 525
 526         spin_lock(&sbi->dir_inode_lock);
 527         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 528                 spin_unlock(&sbi->dir_inode_lock);
 529                 return;
 530         }
 531
 532         list_for_each(this, head) {
 533                 struct dir_inode_entry *entry;
 534                 entry = list_entry(this, struct dir_inode_entry, list);
 535                 if (entry->inode == inode) {
 536                         list_del(&entry->list);
 537                         kmem_cache_free(inode_entry_slab, entry);
 538 #ifdef CONFIG_F2FS_STAT_FS
 539                         sbi->n_dirty_dirs--;
 540 #endif
 541                         break;
 542                 }
 543         }
 544         spin_unlock(&sbi->dir_inode_lock);
 545
 546         /* Only from the recovery routine */
 547         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 548                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 549                 iput(inode);
 550         }
 551 }
 552
 553 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 554 {
 555         struct list_head *head = &sbi->dir_inode_list;
 556         struct list_head *this;
 557         struct inode *inode = NULL;
 558
 559         spin_lock(&sbi->dir_inode_lock);
 560         list_for_each(this, head) {
 561                 struct dir_inode_entry *entry;
 562                 entry = list_entry(this, struct dir_inode_entry, list);
 563                 if (entry->inode->i_ino == ino) {
 564                         inode = entry->inode;
 565                         break;
 566                 }
 567         }
 568         spin_unlock(&sbi->dir_inode_lock);
 569         return inode;
 570 }
 571
 572 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 573 {
 574         struct list_head *head = &sbi->dir_inode_list;
 575         struct dir_inode_entry *entry;
 576         struct inode *inode;
 577 retry:
 578         spin_lock(&sbi->dir_inode_lock);
 579         if (list_empty(head)) {
 580                 spin_unlock(&sbi->dir_inode_lock);
 581                 return;
 582         }
 583         entry = list_entry(head->next, struct dir_inode_entry, list);
 584         inode = igrab(entry->inode);
 585         spin_unlock(&sbi->dir_inode_lock);
 586         if (inode) {
 587                 filemap_flush(inode->i_mapping);
 588                 iput(inode);
 589         } else {
 590                 /*
 591                  * We should submit bio, since it exists several
 592                  * wribacking dentry pages in the freeing inode.
 593                  */
 594                 f2fs_submit_bio(sbi, DATA, true);
 595         }
 596         goto retry;
 597 }
 598
 599 /*
 600  * Freeze all the FS-operations for checkpoint.
 601  */
 602 static void block_operations(struct f2fs_sb_info *sbi)
 603 {
 604         struct writeback_control wbc = {
 605                 .sync_mode = WB_SYNC_ALL,
 606                 .nr_to_write = LONG_MAX,
 607                 .for_reclaim = 0,
 608         };
 609         struct blk_plug plug;
 610
 611         blk_start_plug(&plug);
 612
 613 retry_flush_dents:
 614         mutex_lock_all(sbi);
 615
 616         /* write all the dirty dentry pages */
 617         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 618                 mutex_unlock_all(sbi);
 619                 sync_dirty_dir_inodes(sbi);
 620                 goto retry_flush_dents;
 621         }
 622
 623         /*
 624          * POR: we should ensure that there is no dirty node pages
 625          * until finishing nat/sit flush.
 626          */
 627 retry_flush_nodes:
 628         mutex_lock(&sbi->node_write);
 629
 630         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 631                 mutex_unlock(&sbi->node_write);
 632                 sync_node_pages(sbi, 0, &wbc);
 633                 goto retry_flush_nodes;
 634         }
 635         blk_finish_plug(&plug);
 636 }
 637
 638 static void unblock_operations(struct f2fs_sb_info *sbi)
 639 {
 640         mutex_unlock(&sbi->node_write);
 641         mutex_unlock_all(sbi);
 642 }
 643
 644 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 645 {
 646         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 647         nid_t last_nid = 0;
 648         block_t start_blk;
 649         struct page *cp_page;
 650         unsigned int data_sum_blocks, orphan_blocks;
 651         __u32 crc32 = 0;
 652         void *kaddr;
 653         int i;
 654
 655         /* Flush all the NAT/SIT pages */
 656         while (get_pages(sbi, F2FS_DIRTY_META))
 657                 sync_meta_pages(sbi, META, LONG_MAX);
 658
 659         next_free_nid(sbi, &last_nid);
 660
 661         /*
 662          * modify checkpoint
 663          * version number is already updated
 664          */
 665         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 666         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 667         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 668         for (i = 0; i < 3; i++) {
 669                 ckpt->cur_node_segno[i] =
 670                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 671                 ckpt->cur_node_blkoff[i] =
 672                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 673                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 674                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 675         }
 676         for (i = 0; i < 3; i++) {
 677                 ckpt->cur_data_segno[i] =
 678                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 679                 ckpt->cur_data_blkoff[i] =
 680                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 681                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 682                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 683         }
 684
 685         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 686         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 687         ckpt->next_free_nid = cpu_to_le32(last_nid);
 688
 689         /* 2 cp  + n data seg summary + orphan inode blocks */
 690         data_sum_blocks = npages_for_summary_flush(sbi);
 691         if (data_sum_blocks < 3)
 692                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 693         else
 694                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 695
 696         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 697                                         / F2FS_ORPHANS_PER_BLOCK;
 698         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 699
 700         if (is_umount) {
 701                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 702                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 703                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 704         } else {
 705                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 706                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 707                         data_sum_blocks + orphan_blocks);
 708         }
 709
 710         if (sbi->n_orphans)
 711                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 712         else
 713                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 714
 715         /* update SIT/NAT bitmap */
 716         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 717         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 718
 719         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 720         *((__le32 *)((unsigned char *)ckpt +
 721                                 le32_to_cpu(ckpt->checksum_offset)))
 722                                 = cpu_to_le32(crc32);
 723
 724         start_blk = __start_cp_addr(sbi);
 725
 726         /* write out checkpoint buffer at block 0 */
 727         cp_page = grab_meta_page(sbi, start_blk++);
 728         kaddr = page_address(cp_page);
 729         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 730         set_page_dirty(cp_page);
 731         f2fs_put_page(cp_page, 1);
 732
 733         if (sbi->n_orphans) {
 734                 write_orphan_inodes(sbi, start_blk);
 735                 start_blk += orphan_blocks;
 736         }
 737
 738         write_data_summaries(sbi, start_blk);
 739         start_blk += data_sum_blocks;
 740         if (is_umount) {
 741                 write_node_summaries(sbi, start_blk);
 742                 start_blk += NR_CURSEG_NODE_TYPE;
 743         }
 744
 745         /* writeout checkpoint block */
 746         cp_page = grab_meta_page(sbi, start_blk);
 747         kaddr = page_address(cp_page);
 748         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 749         set_page_dirty(cp_page);
 750         f2fs_put_page(cp_page, 1);
 751
 752         /* wait for previous submitted node/meta pages writeback */
 753         while (get_pages(sbi, F2FS_WRITEBACK))
 754                 congestion_wait(BLK_RW_ASYNC, HZ / 50);
 755
 756         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 757         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 758
 759         /* update user_block_counts */
 760         sbi->last_valid_block_count = sbi->total_valid_block_count;
 761         sbi->alloc_valid_block_count = 0;
 762
 763         /* Here, we only have one bio having CP pack */
 764         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 765
 766         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 767                 clear_prefree_segments(sbi);
 768                 F2FS_RESET_SB_DIRT(sbi);
 769         }
 770 }
 771
 772 /*
 773  * We guarantee that this checkpoint procedure should not fail.
 774  */
 775 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 776 {
 777         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 778         unsigned long long ckpt_ver;
 779
 780         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 781
 782         mutex_lock(&sbi->cp_mutex);
 783         block_operations(sbi);
 784
 785         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 786
 787         f2fs_submit_bio(sbi, DATA, true);
 788         f2fs_submit_bio(sbi, NODE, true);
 789         f2fs_submit_bio(sbi, META, true);
 790
 791         /*
 792          * update checkpoint pack index
 793          * Increase the version number so that
 794          * SIT entries and seg summaries are written at correct place
 795          */
 796         ckpt_ver = le64_to_cpu(ckpt->checkpoint_ver);
 797         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 798
 799         /* write cached NAT/SIT entries to NAT/SIT area */
 800         flush_nat_entries(sbi);
 801         flush_sit_entries(sbi);
 802
 803         /* unlock all the fs_lock[] in do_checkpoint() */
 804         do_checkpoint(sbi, is_umount);
 805
 806         unblock_operations(sbi);
 807         mutex_unlock(&sbi->cp_mutex);
 808
 809         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 810 }
 811
 812 void init_orphan_info(struct f2fs_sb_info *sbi)
 813 {
 814         mutex_init(&sbi->orphan_inode_mutex);
 815         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 816         sbi->n_orphans = 0;
 817 }
 818
 819 int __init create_checkpoint_caches(void)
 820 {
 821         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 822                         sizeof(struct orphan_inode_entry), NULL);
 823         if (unlikely(!orphan_entry_slab))
 824                 return -ENOMEM;
 825         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 826                         sizeof(struct dir_inode_entry), NULL);
 827         if (unlikely(!inode_entry_slab)) {
 828                 kmem_cache_destroy(orphan_entry_slab);
 829                 return -ENOMEM;
 830         }
 831         return 0;
 832 }
 833
 834 void destroy_checkpoint_caches(void)
 835 {
 836         kmem_cache_destroy(orphan_entry_slab);
 837         kmem_cache_destroy(inode_entry_slab);
 838 }