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 = META_MAPPING(sbi);
  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 = META_MAPPING(sbi);
  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_submit_page_bio(sbi, page, index,
  65                                 READ_SYNC | REQ_META | REQ_PRIO))
  66                 goto repeat;
  67
  68         lock_page(page);
  69         if (unlikely(page->mapping != mapping)) {
  70                 f2fs_put_page(page, 1);
  71                 goto repeat;
  72         }
  73 out:
  74         mark_page_accessed(page);
  75         return page;
  76 }
  77
  78 inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  79 {
  80         switch (type) {
  81         case META_NAT:
  82                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  83         case META_SIT:
  84                 return SIT_BLK_CNT(sbi);
  85         case META_CP:
  86                 return 0;
  87         default:
  88                 BUG();
  89         }
  90 }
  91
  92 /*
  93  * Readahead CP/NAT/SIT pages
  94  */
  95 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
  96 {
  97         block_t prev_blk_addr = 0;
  98         struct page *page;
  99         int blkno = start;
 100         int max_blks = get_max_meta_blks(sbi, type);
 101
 102         struct f2fs_io_info fio = {
 103                 .type = META,
 104                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 105         };
 106
 107         for (; nrpages-- > 0; blkno++) {
 108                 block_t blk_addr;
 109
 110                 switch (type) {
 111                 case META_NAT:
 112                         /* get nat block addr */
 113                         if (unlikely(blkno >= max_blks))
 114                                 blkno = 0;
 115                         blk_addr = current_nat_addr(sbi,
 116                                         blkno * NAT_ENTRY_PER_BLOCK);
 117                         break;
 118                 case META_SIT:
 119                         /* get sit block addr */
 120                         if (unlikely(blkno >= max_blks))
 121                                 goto out;
 122                         blk_addr = current_sit_addr(sbi,
 123                                         blkno * SIT_ENTRY_PER_BLOCK);
 124                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 125                                 goto out;
 126                         prev_blk_addr = blk_addr;
 127                         break;
 128                 case META_CP:
 129                         /* get cp block addr */
 130                         blk_addr = blkno;
 131                         break;
 132                 default:
 133                         BUG();
 134                 }
 135
 136                 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
 137                 if (!page)
 138                         continue;
 139                 if (PageUptodate(page)) {
 140                         mark_page_accessed(page);
 141                         f2fs_put_page(page, 1);
 142                         continue;
 143                 }
 144
 145                 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
 146                 mark_page_accessed(page);
 147                 f2fs_put_page(page, 0);
 148         }
 149 out:
 150         f2fs_submit_merged_bio(sbi, META, READ);
 151         return blkno - start;
 152 }
 153
 154 static int f2fs_write_meta_page(struct page *page,
 155                                 struct writeback_control *wbc)
 156 {
 157         struct inode *inode = page->mapping->host;
 158         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 159
 160         if (unlikely(sbi->por_doing))
 161                 goto redirty_out;
 162         if (wbc->for_reclaim)
 163                 goto redirty_out;
 164
 165         /* Should not write any meta pages, if any IO error was occurred */
 166         if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
 167                 goto no_write;
 168
 169         wait_on_page_writeback(page);
 170         write_meta_page(sbi, page);
 171 no_write:
 172         dec_page_count(sbi, F2FS_DIRTY_META);
 173         unlock_page(page);
 174         return 0;
 175
 176 redirty_out:
 177         dec_page_count(sbi, F2FS_DIRTY_META);
 178         wbc->pages_skipped++;
 179         set_page_dirty(page);
 180         return AOP_WRITEPAGE_ACTIVATE;
 181 }
 182
 183 static int f2fs_write_meta_pages(struct address_space *mapping,
 184                                 struct writeback_control *wbc)
 185 {
 186         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 187         int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
 188         long written;
 189
 190         if (wbc->for_kupdate)
 191                 return 0;
 192
 193         /* collect a number of dirty meta pages and write together */
 194         if (get_pages(sbi, F2FS_DIRTY_META) < nrpages)
 195                 return 0;
 196
 197         /* if mounting is failed, skip writing node pages */
 198         mutex_lock(&sbi->cp_mutex);
 199         written = sync_meta_pages(sbi, META, nrpages);
 200         mutex_unlock(&sbi->cp_mutex);
 201         wbc->nr_to_write -= written;
 202         return 0;
 203 }
 204
 205 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 206                                                 long nr_to_write)
 207 {
 208         struct address_space *mapping = META_MAPPING(sbi);
 209         pgoff_t index = 0, end = LONG_MAX;
 210         struct pagevec pvec;
 211         long nwritten = 0;
 212         struct writeback_control wbc = {
 213                 .for_reclaim = 0,
 214         };
 215
 216         pagevec_init(&pvec, 0);
 217
 218         while (index <= end) {
 219                 int i, nr_pages;
 220                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 221                                 PAGECACHE_TAG_DIRTY,
 222                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 223                 if (unlikely(nr_pages == 0))
 224                         break;
 225
 226                 for (i = 0; i < nr_pages; i++) {
 227                         struct page *page = pvec.pages[i];
 228
 229                         lock_page(page);
 230
 231                         if (unlikely(page->mapping != mapping)) {
 232 continue_unlock:
 233                                 unlock_page(page);
 234                                 continue;
 235                         }
 236                         if (!PageDirty(page)) {
 237                                 /* someone wrote it for us */
 238                                 goto continue_unlock;
 239                         }
 240
 241                         if (!clear_page_dirty_for_io(page))
 242                                 goto continue_unlock;
 243
 244                         if (f2fs_write_meta_page(page, &wbc)) {
 245                                 unlock_page(page);
 246                                 break;
 247                         }
 248                         nwritten++;
 249                         if (unlikely(nwritten >= nr_to_write))
 250                                 break;
 251                 }
 252                 pagevec_release(&pvec);
 253                 cond_resched();
 254         }
 255
 256         if (nwritten)
 257                 f2fs_submit_merged_bio(sbi, type, WRITE);
 258
 259         return nwritten;
 260 }
 261
 262 static int f2fs_set_meta_page_dirty(struct page *page)
 263 {
 264         struct address_space *mapping = page->mapping;
 265         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 266
 267         trace_f2fs_set_page_dirty(page, META);
 268
 269         SetPageUptodate(page);
 270         if (!PageDirty(page)) {
 271                 __set_page_dirty_nobuffers(page);
 272                 inc_page_count(sbi, F2FS_DIRTY_META);
 273                 return 1;
 274         }
 275         return 0;
 276 }
 277
 278 const struct address_space_operations f2fs_meta_aops = {
 279         .writepage      = f2fs_write_meta_page,
 280         .writepages     = f2fs_write_meta_pages,
 281         .set_page_dirty = f2fs_set_meta_page_dirty,
 282 };
 283
 284 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 285 {
 286         int err = 0;
 287
 288         spin_lock(&sbi->orphan_inode_lock);
 289         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 290                 err = -ENOSPC;
 291         else
 292                 sbi->n_orphans++;
 293         spin_unlock(&sbi->orphan_inode_lock);
 294
 295         return err;
 296 }
 297
 298 void release_orphan_inode(struct f2fs_sb_info *sbi)
 299 {
 300         spin_lock(&sbi->orphan_inode_lock);
 301         f2fs_bug_on(sbi->n_orphans == 0);
 302         sbi->n_orphans--;
 303         spin_unlock(&sbi->orphan_inode_lock);
 304 }
 305
 306 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 307 {
 308         struct list_head *head, *this;
 309         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 310
 311         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 312         new->ino = ino;
 313
 314         spin_lock(&sbi->orphan_inode_lock);
 315         head = &sbi->orphan_inode_list;
 316         list_for_each(this, head) {
 317                 orphan = list_entry(this, struct orphan_inode_entry, list);
 318                 if (orphan->ino == ino) {
 319                         spin_unlock(&sbi->orphan_inode_lock);
 320                         kmem_cache_free(orphan_entry_slab, new);
 321                         return;
 322                 }
 323
 324                 if (orphan->ino > ino)
 325                         break;
 326                 orphan = NULL;
 327         }
 328
 329         /* add new_oentry into list which is sorted by inode number */
 330         if (orphan)
 331                 list_add(&new->list, this->prev);
 332         else
 333                 list_add_tail(&new->list, head);
 334         spin_unlock(&sbi->orphan_inode_lock);
 335 }
 336
 337 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 338 {
 339         struct list_head *head;
 340         struct orphan_inode_entry *orphan;
 341
 342         spin_lock(&sbi->orphan_inode_lock);
 343         head = &sbi->orphan_inode_list;
 344         list_for_each_entry(orphan, head, list) {
 345                 if (orphan->ino == ino) {
 346                         list_del(&orphan->list);
 347                         kmem_cache_free(orphan_entry_slab, orphan);
 348                         f2fs_bug_on(sbi->n_orphans == 0);
 349                         sbi->n_orphans--;
 350                         break;
 351                 }
 352         }
 353         spin_unlock(&sbi->orphan_inode_lock);
 354 }
 355
 356 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 357 {
 358         struct inode *inode = f2fs_iget(sbi->sb, ino);
 359         f2fs_bug_on(IS_ERR(inode));
 360         clear_nlink(inode);
 361
 362         /* truncate all the data during iput */
 363         iput(inode);
 364 }
 365
 366 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 367 {
 368         block_t start_blk, orphan_blkaddr, i, j;
 369
 370         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 371                 return;
 372
 373         sbi->por_doing = true;
 374         start_blk = __start_cp_addr(sbi) + 1;
 375         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 376
 377         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 378
 379         for (i = 0; i < orphan_blkaddr; i++) {
 380                 struct page *page = get_meta_page(sbi, start_blk + i);
 381                 struct f2fs_orphan_block *orphan_blk;
 382
 383                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 384                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 385                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 386                         recover_orphan_inode(sbi, ino);
 387                 }
 388                 f2fs_put_page(page, 1);
 389         }
 390         /* clear Orphan Flag */
 391         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 392         sbi->por_doing = false;
 393         return;
 394 }
 395
 396 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 397 {
 398         struct list_head *head;
 399         struct f2fs_orphan_block *orphan_blk = NULL;
 400         unsigned int nentries = 0;
 401         unsigned short index;
 402         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 403                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 404         struct page *page = NULL;
 405         struct orphan_inode_entry *orphan = NULL;
 406
 407         for (index = 0; index < orphan_blocks; index++)
 408                 grab_meta_page(sbi, start_blk + index);
 409
 410         index = 1;
 411         spin_lock(&sbi->orphan_inode_lock);
 412         head = &sbi->orphan_inode_list;
 413
 414         /* loop for each orphan inode entry and write them in Jornal block */
 415         list_for_each_entry(orphan, head, list) {
 416                 if (!page) {
 417                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 418                         f2fs_bug_on(!page);
 419                         orphan_blk =
 420                                 (struct f2fs_orphan_block *)page_address(page);
 421                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 422                         f2fs_put_page(page, 0);
 423                 }
 424
 425                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 426
 427                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 428                         /*
 429                          * an orphan block is full of 1020 entries,
 430                          * then we need to flush current orphan blocks
 431                          * and bring another one in memory
 432                          */
 433                         orphan_blk->blk_addr = cpu_to_le16(index);
 434                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 435                         orphan_blk->entry_count = cpu_to_le32(nentries);
 436                         set_page_dirty(page);
 437                         f2fs_put_page(page, 1);
 438                         index++;
 439                         nentries = 0;
 440                         page = NULL;
 441                 }
 442         }
 443
 444         if (page) {
 445                 orphan_blk->blk_addr = cpu_to_le16(index);
 446                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 447                 orphan_blk->entry_count = cpu_to_le32(nentries);
 448                 set_page_dirty(page);
 449                 f2fs_put_page(page, 1);
 450         }
 451
 452         spin_unlock(&sbi->orphan_inode_lock);
 453 }
 454
 455 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 456                                 block_t cp_addr, unsigned long long *version)
 457 {
 458         struct page *cp_page_1, *cp_page_2 = NULL;
 459         unsigned long blk_size = sbi->blocksize;
 460         struct f2fs_checkpoint *cp_block;
 461         unsigned long long cur_version = 0, pre_version = 0;
 462         size_t crc_offset;
 463         __u32 crc = 0;
 464
 465         /* Read the 1st cp block in this CP pack */
 466         cp_page_1 = get_meta_page(sbi, cp_addr);
 467
 468         /* get the version number */
 469         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 470         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 471         if (crc_offset >= blk_size)
 472                 goto invalid_cp1;
 473
 474         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 475         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 476                 goto invalid_cp1;
 477
 478         pre_version = cur_cp_version(cp_block);
 479
 480         /* Read the 2nd cp block in this CP pack */
 481         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 482         cp_page_2 = get_meta_page(sbi, cp_addr);
 483
 484         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 485         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 486         if (crc_offset >= blk_size)
 487                 goto invalid_cp2;
 488
 489         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 490         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 491                 goto invalid_cp2;
 492
 493         cur_version = cur_cp_version(cp_block);
 494
 495         if (cur_version == pre_version) {
 496                 *version = cur_version;
 497                 f2fs_put_page(cp_page_2, 1);
 498                 return cp_page_1;
 499         }
 500 invalid_cp2:
 501         f2fs_put_page(cp_page_2, 1);
 502 invalid_cp1:
 503         f2fs_put_page(cp_page_1, 1);
 504         return NULL;
 505 }
 506
 507 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 508 {
 509         struct f2fs_checkpoint *cp_block;
 510         struct f2fs_super_block *fsb = sbi->raw_super;
 511         struct page *cp1, *cp2, *cur_page;
 512         unsigned long blk_size = sbi->blocksize;
 513         unsigned long long cp1_version = 0, cp2_version = 0;
 514         unsigned long long cp_start_blk_no;
 515
 516         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 517         if (!sbi->ckpt)
 518                 return -ENOMEM;
 519         /*
 520          * Finding out valid cp block involves read both
 521          * sets( cp pack1 and cp pack 2)
 522          */
 523         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 524         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 525
 526         /* The second checkpoint pack should start at the next segment */
 527         cp_start_blk_no += ((unsigned long long)1) <<
 528                                 le32_to_cpu(fsb->log_blocks_per_seg);
 529         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 530
 531         if (cp1 && cp2) {
 532                 if (ver_after(cp2_version, cp1_version))
 533                         cur_page = cp2;
 534                 else
 535                         cur_page = cp1;
 536         } else if (cp1) {
 537                 cur_page = cp1;
 538         } else if (cp2) {
 539                 cur_page = cp2;
 540         } else {
 541                 goto fail_no_cp;
 542         }
 543
 544         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 545         memcpy(sbi->ckpt, cp_block, blk_size);
 546
 547         f2fs_put_page(cp1, 1);
 548         f2fs_put_page(cp2, 1);
 549         return 0;
 550
 551 fail_no_cp:
 552         kfree(sbi->ckpt);
 553         return -EINVAL;
 554 }
 555
 556 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 557 {
 558         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 559         struct list_head *head = &sbi->dir_inode_list;
 560         struct list_head *this;
 561
 562         list_for_each(this, head) {
 563                 struct dir_inode_entry *entry;
 564                 entry = list_entry(this, struct dir_inode_entry, list);
 565                 if (unlikely(entry->inode == inode))
 566                         return -EEXIST;
 567         }
 568         list_add_tail(&new->list, head);
 569         stat_inc_dirty_dir(sbi);
 570         return 0;
 571 }
 572
 573 void set_dirty_dir_page(struct inode *inode, struct page *page)
 574 {
 575         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 576         struct dir_inode_entry *new;
 577
 578         if (!S_ISDIR(inode->i_mode))
 579                 return;
 580
 581         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 582         new->inode = inode;
 583         INIT_LIST_HEAD(&new->list);
 584
 585         spin_lock(&sbi->dir_inode_lock);
 586         if (__add_dirty_inode(inode, new))
 587                 kmem_cache_free(inode_entry_slab, new);
 588
 589         inode_inc_dirty_dents(inode);
 590         SetPagePrivate(page);
 591         spin_unlock(&sbi->dir_inode_lock);
 592 }
 593
 594 void add_dirty_dir_inode(struct inode *inode)
 595 {
 596         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 597         struct dir_inode_entry *new =
 598                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 599
 600         new->inode = inode;
 601         INIT_LIST_HEAD(&new->list);
 602
 603         spin_lock(&sbi->dir_inode_lock);
 604         if (__add_dirty_inode(inode, new))
 605                 kmem_cache_free(inode_entry_slab, new);
 606         spin_unlock(&sbi->dir_inode_lock);
 607 }
 608
 609 void remove_dirty_dir_inode(struct inode *inode)
 610 {
 611         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 612
 613         struct list_head *this, *head;
 614
 615         if (!S_ISDIR(inode->i_mode))
 616                 return;
 617
 618         spin_lock(&sbi->dir_inode_lock);
 619         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 620                 spin_unlock(&sbi->dir_inode_lock);
 621                 return;
 622         }
 623
 624         head = &sbi->dir_inode_list;
 625         list_for_each(this, head) {
 626                 struct dir_inode_entry *entry;
 627                 entry = list_entry(this, struct dir_inode_entry, list);
 628                 if (entry->inode == inode) {
 629                         list_del(&entry->list);
 630                         kmem_cache_free(inode_entry_slab, entry);
 631                         stat_dec_dirty_dir(sbi);
 632                         break;
 633                 }
 634         }
 635         spin_unlock(&sbi->dir_inode_lock);
 636
 637         /* Only from the recovery routine */
 638         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 639                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 640                 iput(inode);
 641         }
 642 }
 643
 644 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 645 {
 646
 647         struct list_head *this, *head;
 648         struct inode *inode = NULL;
 649
 650         spin_lock(&sbi->dir_inode_lock);
 651
 652         head = &sbi->dir_inode_list;
 653         list_for_each(this, head) {
 654                 struct dir_inode_entry *entry;
 655                 entry = list_entry(this, struct dir_inode_entry, list);
 656                 if (entry->inode->i_ino == ino) {
 657                         inode = entry->inode;
 658                         break;
 659                 }
 660         }
 661         spin_unlock(&sbi->dir_inode_lock);
 662         return inode;
 663 }
 664
 665 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 666 {
 667         struct list_head *head;
 668         struct dir_inode_entry *entry;
 669         struct inode *inode;
 670 retry:
 671         spin_lock(&sbi->dir_inode_lock);
 672
 673         head = &sbi->dir_inode_list;
 674         if (list_empty(head)) {
 675                 spin_unlock(&sbi->dir_inode_lock);
 676                 return;
 677         }
 678         entry = list_entry(head->next, struct dir_inode_entry, list);
 679         inode = igrab(entry->inode);
 680         spin_unlock(&sbi->dir_inode_lock);
 681         if (inode) {
 682                 filemap_flush(inode->i_mapping);
 683                 iput(inode);
 684         } else {
 685                 /*
 686                  * We should submit bio, since it exists several
 687                  * wribacking dentry pages in the freeing inode.
 688                  */
 689                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 690         }
 691         goto retry;
 692 }
 693
 694 /*
 695  * Freeze all the FS-operations for checkpoint.
 696  */
 697 static void block_operations(struct f2fs_sb_info *sbi)
 698 {
 699         struct writeback_control wbc = {
 700                 .sync_mode = WB_SYNC_ALL,
 701                 .nr_to_write = LONG_MAX,
 702                 .for_reclaim = 0,
 703         };
 704         struct blk_plug plug;
 705
 706         blk_start_plug(&plug);
 707
 708 retry_flush_dents:
 709         f2fs_lock_all(sbi);
 710         /* write all the dirty dentry pages */
 711         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 712                 f2fs_unlock_all(sbi);
 713                 sync_dirty_dir_inodes(sbi);
 714                 goto retry_flush_dents;
 715         }
 716
 717         /*
 718          * POR: we should ensure that there is no dirty node pages
 719          * until finishing nat/sit flush.
 720          */
 721 retry_flush_nodes:
 722         mutex_lock(&sbi->node_write);
 723
 724         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 725                 mutex_unlock(&sbi->node_write);
 726                 sync_node_pages(sbi, 0, &wbc);
 727                 goto retry_flush_nodes;
 728         }
 729         blk_finish_plug(&plug);
 730 }
 731
 732 static void unblock_operations(struct f2fs_sb_info *sbi)
 733 {
 734         mutex_unlock(&sbi->node_write);
 735         f2fs_unlock_all(sbi);
 736 }
 737
 738 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 739 {
 740         DEFINE_WAIT(wait);
 741
 742         for (;;) {
 743                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 744
 745                 if (!get_pages(sbi, F2FS_WRITEBACK))
 746                         break;
 747
 748                 io_schedule();
 749         }
 750         finish_wait(&sbi->cp_wait, &wait);
 751 }
 752
 753 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 754 {
 755         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 756         nid_t last_nid = 0;
 757         block_t start_blk;
 758         struct page *cp_page;
 759         unsigned int data_sum_blocks, orphan_blocks;
 760         __u32 crc32 = 0;
 761         void *kaddr;
 762         int i;
 763
 764         /* Flush all the NAT/SIT pages */
 765         while (get_pages(sbi, F2FS_DIRTY_META))
 766                 sync_meta_pages(sbi, META, LONG_MAX);
 767
 768         next_free_nid(sbi, &last_nid);
 769
 770         /*
 771          * modify checkpoint
 772          * version number is already updated
 773          */
 774         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 775         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 776         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 777         for (i = 0; i < 3; i++) {
 778                 ckpt->cur_node_segno[i] =
 779                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 780                 ckpt->cur_node_blkoff[i] =
 781                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 782                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 783                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 784         }
 785         for (i = 0; i < 3; i++) {
 786                 ckpt->cur_data_segno[i] =
 787                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 788                 ckpt->cur_data_blkoff[i] =
 789                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 790                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 791                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 792         }
 793
 794         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 795         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 796         ckpt->next_free_nid = cpu_to_le32(last_nid);
 797
 798         /* 2 cp  + n data seg summary + orphan inode blocks */
 799         data_sum_blocks = npages_for_summary_flush(sbi);
 800         if (data_sum_blocks < 3)
 801                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 802         else
 803                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 804
 805         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 806                                         / F2FS_ORPHANS_PER_BLOCK;
 807         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 808
 809         if (is_umount) {
 810                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 811                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 812                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 813         } else {
 814                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 815                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 816                         data_sum_blocks + orphan_blocks);
 817         }
 818
 819         if (sbi->n_orphans)
 820                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 821         else
 822                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 823
 824         /* update SIT/NAT bitmap */
 825         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 826         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 827
 828         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 829         *((__le32 *)((unsigned char *)ckpt +
 830                                 le32_to_cpu(ckpt->checksum_offset)))
 831                                 = cpu_to_le32(crc32);
 832
 833         start_blk = __start_cp_addr(sbi);
 834
 835         /* write out checkpoint buffer at block 0 */
 836         cp_page = grab_meta_page(sbi, start_blk++);
 837         kaddr = page_address(cp_page);
 838         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 839         set_page_dirty(cp_page);
 840         f2fs_put_page(cp_page, 1);
 841
 842         if (sbi->n_orphans) {
 843                 write_orphan_inodes(sbi, start_blk);
 844                 start_blk += orphan_blocks;
 845         }
 846
 847         write_data_summaries(sbi, start_blk);
 848         start_blk += data_sum_blocks;
 849         if (is_umount) {
 850                 write_node_summaries(sbi, start_blk);
 851                 start_blk += NR_CURSEG_NODE_TYPE;
 852         }
 853
 854         /* writeout checkpoint block */
 855         cp_page = grab_meta_page(sbi, start_blk);
 856         kaddr = page_address(cp_page);
 857         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 858         set_page_dirty(cp_page);
 859         f2fs_put_page(cp_page, 1);
 860
 861         /* wait for previous submitted node/meta pages writeback */
 862         wait_on_all_pages_writeback(sbi);
 863
 864         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 865         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 866
 867         /* update user_block_counts */
 868         sbi->last_valid_block_count = sbi->total_valid_block_count;
 869         sbi->alloc_valid_block_count = 0;
 870
 871         /* Here, we only have one bio having CP pack */
 872         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 873
 874         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 875                 clear_prefree_segments(sbi);
 876                 F2FS_RESET_SB_DIRT(sbi);
 877         }
 878 }
 879
 880 /*
 881  * We guarantee that this checkpoint procedure should not fail.
 882  */
 883 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 884 {
 885         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 886         unsigned long long ckpt_ver;
 887
 888         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 889
 890         mutex_lock(&sbi->cp_mutex);
 891         block_operations(sbi);
 892
 893         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 894
 895         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 896         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 897         f2fs_submit_merged_bio(sbi, META, WRITE);
 898
 899         /*
 900          * update checkpoint pack index
 901          * Increase the version number so that
 902          * SIT entries and seg summaries are written at correct place
 903          */
 904         ckpt_ver = cur_cp_version(ckpt);
 905         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 906
 907         /* write cached NAT/SIT entries to NAT/SIT area */
 908         flush_nat_entries(sbi);
 909         flush_sit_entries(sbi);
 910
 911         /* unlock all the fs_lock[] in do_checkpoint() */
 912         do_checkpoint(sbi, is_umount);
 913
 914         unblock_operations(sbi);
 915         mutex_unlock(&sbi->cp_mutex);
 916
 917         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 918 }
 919
 920 void init_orphan_info(struct f2fs_sb_info *sbi)
 921 {
 922         spin_lock_init(&sbi->orphan_inode_lock);
 923         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 924         sbi->n_orphans = 0;
 925         /*
 926          * considering 512 blocks in a segment 8 blocks are needed for cp
 927          * and log segment summaries. Remaining blocks are used to keep
 928          * orphan entries with the limitation one reserved segment
 929          * for cp pack we can have max 1020*504 orphan entries
 930          */
 931         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 932                                 * F2FS_ORPHANS_PER_BLOCK;
 933 }
 934
 935 int __init create_checkpoint_caches(void)
 936 {
 937         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 938                         sizeof(struct orphan_inode_entry), NULL);
 939         if (!orphan_entry_slab)
 940                 return -ENOMEM;
 941         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 942                         sizeof(struct dir_inode_entry), NULL);
 943         if (!inode_entry_slab) {
 944                 kmem_cache_destroy(orphan_entry_slab);
 945                 return -ENOMEM;
 946         }
 947         return 0;
 948 }
 949
 950 void destroy_checkpoint_caches(void)
 951 {
 952         kmem_cache_destroy(orphan_entry_slab);
 953         kmem_cache_destroy(inode_entry_slab);
 954 }