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 acquire_orphan_inode(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         else
 201                 sbi->n_orphans++;
 202         mutex_unlock(&sbi->orphan_inode_mutex);
 203         return err;
 204 }
 205
 206 void release_orphan_inode(struct f2fs_sb_info *sbi)
 207 {
 208         mutex_lock(&sbi->orphan_inode_mutex);
 209         BUG_ON(sbi->n_orphans == 0);
 210         sbi->n_orphans--;
 211         mutex_unlock(&sbi->orphan_inode_mutex);
 212 }
 213
 214 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 215 {
 216         struct list_head *head, *this;
 217         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 218
 219         mutex_lock(&sbi->orphan_inode_mutex);
 220         head = &sbi->orphan_inode_list;
 221         list_for_each(this, head) {
 222                 orphan = list_entry(this, struct orphan_inode_entry, list);
 223                 if (orphan->ino == ino)
 224                         goto out;
 225                 if (orphan->ino > ino)
 226                         break;
 227                 orphan = NULL;
 228         }
 229 retry:
 230         new = kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 231         if (!new) {
 232                 cond_resched();
 233                 goto retry;
 234         }
 235         new->ino = ino;
 236
 237         /* add new_oentry into list which is sorted by inode number */
 238         if (orphan)
 239                 list_add(&new->list, this->prev);
 240         else
 241                 list_add_tail(&new->list, head);
 242 out:
 243         mutex_unlock(&sbi->orphan_inode_mutex);
 244 }
 245
 246 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 247 {
 248         struct list_head *head;
 249         struct orphan_inode_entry *orphan;
 250
 251         mutex_lock(&sbi->orphan_inode_mutex);
 252         head = &sbi->orphan_inode_list;
 253         list_for_each_entry(orphan, head, list) {
 254                 if (orphan->ino == ino) {
 255                         list_del(&orphan->list);
 256                         kmem_cache_free(orphan_entry_slab, orphan);
 257                         BUG_ON(sbi->n_orphans == 0);
 258                         sbi->n_orphans--;
 259                         break;
 260                 }
 261         }
 262         mutex_unlock(&sbi->orphan_inode_mutex);
 263 }
 264
 265 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 266 {
 267         struct inode *inode = f2fs_iget(sbi->sb, ino);
 268         BUG_ON(IS_ERR(inode));
 269         clear_nlink(inode);
 270
 271         /* truncate all the data during iput */
 272         iput(inode);
 273 }
 274
 275 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 276 {
 277         block_t start_blk, orphan_blkaddr, i, j;
 278
 279         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 280                 return 0;
 281
 282         sbi->por_doing = 1;
 283         start_blk = __start_cp_addr(sbi) + 1;
 284         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 285
 286         for (i = 0; i < orphan_blkaddr; i++) {
 287                 struct page *page = get_meta_page(sbi, start_blk + i);
 288                 struct f2fs_orphan_block *orphan_blk;
 289
 290                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 291                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 292                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 293                         recover_orphan_inode(sbi, ino);
 294                 }
 295                 f2fs_put_page(page, 1);
 296         }
 297         /* clear Orphan Flag */
 298         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 299         sbi->por_doing = 0;
 300         return 0;
 301 }
 302
 303 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 304 {
 305         struct list_head *head, *this, *next;
 306         struct f2fs_orphan_block *orphan_blk = NULL;
 307         struct page *page = NULL;
 308         unsigned int nentries = 0;
 309         unsigned short index = 1;
 310         unsigned short orphan_blocks;
 311
 312         orphan_blocks = (unsigned short)((sbi->n_orphans +
 313                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 314
 315         mutex_lock(&sbi->orphan_inode_mutex);
 316         head = &sbi->orphan_inode_list;
 317
 318         /* loop for each orphan inode entry and write them in Jornal block */
 319         list_for_each_safe(this, next, head) {
 320                 struct orphan_inode_entry *orphan;
 321
 322                 orphan = list_entry(this, struct orphan_inode_entry, list);
 323
 324                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 325                         /*
 326                          * an orphan block is full of 1020 entries,
 327                          * then we need to flush current orphan blocks
 328                          * and bring another one in memory
 329                          */
 330                         orphan_blk->blk_addr = cpu_to_le16(index);
 331                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 332                         orphan_blk->entry_count = cpu_to_le32(nentries);
 333                         set_page_dirty(page);
 334                         f2fs_put_page(page, 1);
 335                         index++;
 336                         start_blk++;
 337                         nentries = 0;
 338                         page = NULL;
 339                 }
 340                 if (page)
 341                         goto page_exist;
 342
 343                 page = grab_meta_page(sbi, start_blk);
 344                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 345                 memset(orphan_blk, 0, sizeof(*orphan_blk));
 346 page_exist:
 347                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 348         }
 349         if (!page)
 350                 goto end;
 351
 352         orphan_blk->blk_addr = cpu_to_le16(index);
 353         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 354         orphan_blk->entry_count = cpu_to_le32(nentries);
 355         set_page_dirty(page);
 356         f2fs_put_page(page, 1);
 357 end:
 358         mutex_unlock(&sbi->orphan_inode_mutex);
 359 }
 360
 361 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 362                                 block_t cp_addr, unsigned long long *version)
 363 {
 364         struct page *cp_page_1, *cp_page_2 = NULL;
 365         unsigned long blk_size = sbi->blocksize;
 366         struct f2fs_checkpoint *cp_block;
 367         unsigned long long cur_version = 0, pre_version = 0;
 368         size_t crc_offset;
 369         __u32 crc = 0;
 370
 371         /* Read the 1st cp block in this CP pack */
 372         cp_page_1 = get_meta_page(sbi, cp_addr);
 373
 374         /* get the version number */
 375         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 376         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 377         if (crc_offset >= blk_size)
 378                 goto invalid_cp1;
 379
 380         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 381         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 382                 goto invalid_cp1;
 383
 384         pre_version = cur_cp_version(cp_block);
 385
 386         /* Read the 2nd cp block in this CP pack */
 387         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 388         cp_page_2 = get_meta_page(sbi, cp_addr);
 389
 390         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 391         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 392         if (crc_offset >= blk_size)
 393                 goto invalid_cp2;
 394
 395         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 396         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 397                 goto invalid_cp2;
 398
 399         cur_version = cur_cp_version(cp_block);
 400
 401         if (cur_version == pre_version) {
 402                 *version = cur_version;
 403                 f2fs_put_page(cp_page_2, 1);
 404                 return cp_page_1;
 405         }
 406 invalid_cp2:
 407         f2fs_put_page(cp_page_2, 1);
 408 invalid_cp1:
 409         f2fs_put_page(cp_page_1, 1);
 410         return NULL;
 411 }
 412
 413 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 414 {
 415         struct f2fs_checkpoint *cp_block;
 416         struct f2fs_super_block *fsb = sbi->raw_super;
 417         struct page *cp1, *cp2, *cur_page;
 418         unsigned long blk_size = sbi->blocksize;
 419         unsigned long long cp1_version = 0, cp2_version = 0;
 420         unsigned long long cp_start_blk_no;
 421
 422         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 423         if (!sbi->ckpt)
 424                 return -ENOMEM;
 425         /*
 426          * Finding out valid cp block involves read both
 427          * sets( cp pack1 and cp pack 2)
 428          */
 429         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 430         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 431
 432         /* The second checkpoint pack should start at the next segment */
 433         cp_start_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 434         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 435
 436         if (cp1 && cp2) {
 437                 if (ver_after(cp2_version, cp1_version))
 438                         cur_page = cp2;
 439                 else
 440                         cur_page = cp1;
 441         } else if (cp1) {
 442                 cur_page = cp1;
 443         } else if (cp2) {
 444                 cur_page = cp2;
 445         } else {
 446                 goto fail_no_cp;
 447         }
 448
 449         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 450         memcpy(sbi->ckpt, cp_block, blk_size);
 451
 452         f2fs_put_page(cp1, 1);
 453         f2fs_put_page(cp2, 1);
 454         return 0;
 455
 456 fail_no_cp:
 457         kfree(sbi->ckpt);
 458         return -EINVAL;
 459 }
 460
 461 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 462 {
 463         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 464         struct list_head *head = &sbi->dir_inode_list;
 465         struct list_head *this;
 466
 467         list_for_each(this, head) {
 468                 struct dir_inode_entry *entry;
 469                 entry = list_entry(this, struct dir_inode_entry, list);
 470                 if (entry->inode == inode)
 471                         return -EEXIST;
 472         }
 473         list_add_tail(&new->list, head);
 474 #ifdef CONFIG_F2FS_STAT_FS
 475         sbi->n_dirty_dirs++;
 476 #endif
 477         return 0;
 478 }
 479
 480 void set_dirty_dir_page(struct inode *inode, struct page *page)
 481 {
 482         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 483         struct dir_inode_entry *new;
 484
 485         if (!S_ISDIR(inode->i_mode))
 486                 return;
 487 retry:
 488         new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 489         if (!new) {
 490                 cond_resched();
 491                 goto retry;
 492         }
 493         new->inode = inode;
 494         INIT_LIST_HEAD(&new->list);
 495
 496         spin_lock(&sbi->dir_inode_lock);
 497         if (__add_dirty_inode(inode, new))
 498                 kmem_cache_free(inode_entry_slab, new);
 499
 500         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 501         inode_inc_dirty_dents(inode);
 502         SetPagePrivate(page);
 503         spin_unlock(&sbi->dir_inode_lock);
 504 }
 505
 506 void add_dirty_dir_inode(struct inode *inode)
 507 {
 508         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 509         struct dir_inode_entry *new;
 510 retry:
 511         new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 512         if (!new) {
 513                 cond_resched();
 514                 goto retry;
 515         }
 516         new->inode = inode;
 517         INIT_LIST_HEAD(&new->list);
 518
 519         spin_lock(&sbi->dir_inode_lock);
 520         if (__add_dirty_inode(inode, new))
 521                 kmem_cache_free(inode_entry_slab, new);
 522         spin_unlock(&sbi->dir_inode_lock);
 523 }
 524
 525 void remove_dirty_dir_inode(struct inode *inode)
 526 {
 527         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 528         struct list_head *head = &sbi->dir_inode_list;
 529         struct list_head *this;
 530
 531         if (!S_ISDIR(inode->i_mode))
 532                 return;
 533
 534         spin_lock(&sbi->dir_inode_lock);
 535         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 536                 spin_unlock(&sbi->dir_inode_lock);
 537                 return;
 538         }
 539
 540         list_for_each(this, head) {
 541                 struct dir_inode_entry *entry;
 542                 entry = list_entry(this, struct dir_inode_entry, list);
 543                 if (entry->inode == inode) {
 544                         list_del(&entry->list);
 545                         kmem_cache_free(inode_entry_slab, entry);
 546 #ifdef CONFIG_F2FS_STAT_FS
 547                         sbi->n_dirty_dirs--;
 548 #endif
 549                         break;
 550                 }
 551         }
 552         spin_unlock(&sbi->dir_inode_lock);
 553
 554         /* Only from the recovery routine */
 555         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 556                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 557                 iput(inode);
 558         }
 559 }
 560
 561 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 562 {
 563         struct list_head *head = &sbi->dir_inode_list;
 564         struct list_head *this;
 565         struct inode *inode = NULL;
 566
 567         spin_lock(&sbi->dir_inode_lock);
 568         list_for_each(this, head) {
 569                 struct dir_inode_entry *entry;
 570                 entry = list_entry(this, struct dir_inode_entry, list);
 571                 if (entry->inode->i_ino == ino) {
 572                         inode = entry->inode;
 573                         break;
 574                 }
 575         }
 576         spin_unlock(&sbi->dir_inode_lock);
 577         return inode;
 578 }
 579
 580 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 581 {
 582         struct list_head *head = &sbi->dir_inode_list;
 583         struct dir_inode_entry *entry;
 584         struct inode *inode;
 585 retry:
 586         spin_lock(&sbi->dir_inode_lock);
 587         if (list_empty(head)) {
 588                 spin_unlock(&sbi->dir_inode_lock);
 589                 return;
 590         }
 591         entry = list_entry(head->next, struct dir_inode_entry, list);
 592         inode = igrab(entry->inode);
 593         spin_unlock(&sbi->dir_inode_lock);
 594         if (inode) {
 595                 filemap_flush(inode->i_mapping);
 596                 iput(inode);
 597         } else {
 598                 /*
 599                  * We should submit bio, since it exists several
 600                  * wribacking dentry pages in the freeing inode.
 601                  */
 602                 f2fs_submit_bio(sbi, DATA, true);
 603         }
 604         goto retry;
 605 }
 606
 607 /*
 608  * Freeze all the FS-operations for checkpoint.
 609  */
 610 static void block_operations(struct f2fs_sb_info *sbi)
 611 {
 612         struct writeback_control wbc = {
 613                 .sync_mode = WB_SYNC_ALL,
 614                 .nr_to_write = LONG_MAX,
 615                 .for_reclaim = 0,
 616         };
 617         struct blk_plug plug;
 618
 619         blk_start_plug(&plug);
 620
 621 retry_flush_dents:
 622         f2fs_lock_all(sbi);
 623         /* write all the dirty dentry pages */
 624         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 625                 f2fs_unlock_all(sbi);
 626                 sync_dirty_dir_inodes(sbi);
 627                 goto retry_flush_dents;
 628         }
 629
 630         /*
 631          * POR: we should ensure that there is no dirty node pages
 632          * until finishing nat/sit flush.
 633          */
 634 retry_flush_nodes:
 635         mutex_lock(&sbi->node_write);
 636
 637         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 638                 mutex_unlock(&sbi->node_write);
 639                 sync_node_pages(sbi, 0, &wbc);
 640                 goto retry_flush_nodes;
 641         }
 642         blk_finish_plug(&plug);
 643 }
 644
 645 static void unblock_operations(struct f2fs_sb_info *sbi)
 646 {
 647         mutex_unlock(&sbi->node_write);
 648         f2fs_unlock_all(sbi);
 649 }
 650
 651 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 652 {
 653         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 654         nid_t last_nid = 0;
 655         block_t start_blk;
 656         struct page *cp_page;
 657         unsigned int data_sum_blocks, orphan_blocks;
 658         __u32 crc32 = 0;
 659         void *kaddr;
 660         int i;
 661
 662         /* Flush all the NAT/SIT pages */
 663         while (get_pages(sbi, F2FS_DIRTY_META))
 664                 sync_meta_pages(sbi, META, LONG_MAX);
 665
 666         next_free_nid(sbi, &last_nid);
 667
 668         /*
 669          * modify checkpoint
 670          * version number is already updated
 671          */
 672         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 673         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 674         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 675         for (i = 0; i < 3; i++) {
 676                 ckpt->cur_node_segno[i] =
 677                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 678                 ckpt->cur_node_blkoff[i] =
 679                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 680                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 681                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 682         }
 683         for (i = 0; i < 3; i++) {
 684                 ckpt->cur_data_segno[i] =
 685                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 686                 ckpt->cur_data_blkoff[i] =
 687                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 688                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 689                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 690         }
 691
 692         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 693         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 694         ckpt->next_free_nid = cpu_to_le32(last_nid);
 695
 696         /* 2 cp  + n data seg summary + orphan inode blocks */
 697         data_sum_blocks = npages_for_summary_flush(sbi);
 698         if (data_sum_blocks < 3)
 699                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 700         else
 701                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 702
 703         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 704                                         / F2FS_ORPHANS_PER_BLOCK;
 705         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 706
 707         if (is_umount) {
 708                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 709                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 710                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 711         } else {
 712                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 713                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 714                         data_sum_blocks + orphan_blocks);
 715         }
 716
 717         if (sbi->n_orphans)
 718                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 719         else
 720                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 721
 722         /* update SIT/NAT bitmap */
 723         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 724         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 725
 726         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 727         *((__le32 *)((unsigned char *)ckpt +
 728                                 le32_to_cpu(ckpt->checksum_offset)))
 729                                 = cpu_to_le32(crc32);
 730
 731         start_blk = __start_cp_addr(sbi);
 732
 733         /* write out checkpoint buffer at block 0 */
 734         cp_page = grab_meta_page(sbi, start_blk++);
 735         kaddr = page_address(cp_page);
 736         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 737         set_page_dirty(cp_page);
 738         f2fs_put_page(cp_page, 1);
 739
 740         if (sbi->n_orphans) {
 741                 write_orphan_inodes(sbi, start_blk);
 742                 start_blk += orphan_blocks;
 743         }
 744
 745         write_data_summaries(sbi, start_blk);
 746         start_blk += data_sum_blocks;
 747         if (is_umount) {
 748                 write_node_summaries(sbi, start_blk);
 749                 start_blk += NR_CURSEG_NODE_TYPE;
 750         }
 751
 752         /* writeout checkpoint block */
 753         cp_page = grab_meta_page(sbi, start_blk);
 754         kaddr = page_address(cp_page);
 755         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 756         set_page_dirty(cp_page);
 757         f2fs_put_page(cp_page, 1);
 758
 759         /* wait for previous submitted node/meta pages writeback */
 760         while (get_pages(sbi, F2FS_WRITEBACK))
 761                 congestion_wait(BLK_RW_ASYNC, HZ / 50);
 762
 763         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 764         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 765
 766         /* update user_block_counts */
 767         sbi->last_valid_block_count = sbi->total_valid_block_count;
 768         sbi->alloc_valid_block_count = 0;
 769
 770         /* Here, we only have one bio having CP pack */
 771         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 772
 773         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 774                 clear_prefree_segments(sbi);
 775                 F2FS_RESET_SB_DIRT(sbi);
 776         }
 777 }
 778
 779 /*
 780  * We guarantee that this checkpoint procedure should not fail.
 781  */
 782 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 783 {
 784         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 785         unsigned long long ckpt_ver;
 786
 787         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 788
 789         mutex_lock(&sbi->cp_mutex);
 790         block_operations(sbi);
 791
 792         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 793
 794         f2fs_submit_bio(sbi, DATA, true);
 795         f2fs_submit_bio(sbi, NODE, true);
 796         f2fs_submit_bio(sbi, META, true);
 797
 798         /*
 799          * update checkpoint pack index
 800          * Increase the version number so that
 801          * SIT entries and seg summaries are written at correct place
 802          */
 803         ckpt_ver = cur_cp_version(ckpt);
 804         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 805
 806         /* write cached NAT/SIT entries to NAT/SIT area */
 807         flush_nat_entries(sbi);
 808         flush_sit_entries(sbi);
 809
 810         /* unlock all the fs_lock[] in do_checkpoint() */
 811         do_checkpoint(sbi, is_umount);
 812
 813         unblock_operations(sbi);
 814         mutex_unlock(&sbi->cp_mutex);
 815
 816         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 817 }
 818
 819 void init_orphan_info(struct f2fs_sb_info *sbi)
 820 {
 821         mutex_init(&sbi->orphan_inode_mutex);
 822         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 823         sbi->n_orphans = 0;
 824 }
 825
 826 int __init create_checkpoint_caches(void)
 827 {
 828         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 829                         sizeof(struct orphan_inode_entry), NULL);
 830         if (unlikely(!orphan_entry_slab))
 831                 return -ENOMEM;
 832         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 833                         sizeof(struct dir_inode_entry), NULL);
 834         if (unlikely(!inode_entry_slab)) {
 835                 kmem_cache_destroy(orphan_entry_slab);
 836                 return -ENOMEM;
 837         }
 838         return 0;
 839 }
 840
 841 void destroy_checkpoint_caches(void)
 842 {
 843         kmem_cache_destroy(orphan_entry_slab);
 844         kmem_cache_destroy(inode_entry_slab);
 845 }