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