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_write_begin(mapping, index, AOP_FLAG_NOFS);
  37         if (!page) {
  38                 cond_resched();
  39                 goto repeat;
  40         }
  41
  42         SetPageUptodate(page);
  43         return page;
  44 }
  45
  46 /*
  47  * We guarantee no failure on the returned page.
  48  */
  49 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  50 {
  51         struct address_space *mapping = META_MAPPING(sbi);
  52         struct page *page;
  53 repeat:
  54         page = grab_cache_page(mapping, index);
  55         if (!page) {
  56                 cond_resched();
  57                 goto repeat;
  58         }
  59         if (PageUptodate(page))
  60                 goto out;
  61
  62         if (f2fs_submit_page_bio(sbi, page, index,
  63                                 READ_SYNC | REQ_META | REQ_PRIO))
  64                 goto repeat;
  65
  66         lock_page(page);
  67         if (unlikely(page->mapping != mapping)) {
  68                 f2fs_put_page(page, 1);
  69                 goto repeat;
  70         }
  71 out:
  72         mark_page_accessed(page);
  73         return page;
  74 }
  75
  76 inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  77 {
  78         switch (type) {
  79         case META_NAT:
  80                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  81         case META_SIT:
  82                 return SIT_BLK_CNT(sbi);
  83         case META_SSA:
  84         case META_CP:
  85                 return 0;
  86         default:
  87                 BUG();
  88         }
  89 }
  90
  91 /*
  92  * Readahead CP/NAT/SIT/SSA pages
  93  */
  94 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
  95 {
  96         block_t prev_blk_addr = 0;
  97         struct page *page;
  98         int blkno = start;
  99         int max_blks = get_max_meta_blks(sbi, type);
 100
 101         struct f2fs_io_info fio = {
 102                 .type = META,
 103                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 104         };
 105
 106         for (; nrpages-- > 0; blkno++) {
 107                 block_t blk_addr;
 108
 109                 switch (type) {
 110                 case META_NAT:
 111                         /* get nat block addr */
 112                         if (unlikely(blkno >= max_blks))
 113                                 blkno = 0;
 114                         blk_addr = current_nat_addr(sbi,
 115                                         blkno * NAT_ENTRY_PER_BLOCK);
 116                         break;
 117                 case META_SIT:
 118                         /* get sit block addr */
 119                         if (unlikely(blkno >= max_blks))
 120                                 goto out;
 121                         blk_addr = current_sit_addr(sbi,
 122                                         blkno * SIT_ENTRY_PER_BLOCK);
 123                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 124                                 goto out;
 125                         prev_blk_addr = blk_addr;
 126                         break;
 127                 case META_SSA:
 128                 case META_CP:
 129                         /* get ssa/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         f2fs_wait_on_page_writeback(page, META);
 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         account_page_redirty(page);
 180         set_page_dirty(page);
 181         return AOP_WRITEPAGE_ACTIVATE;
 182 }
 183
 184 static int f2fs_write_meta_pages(struct address_space *mapping,
 185                                 struct writeback_control *wbc)
 186 {
 187         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 188         long diff, written;
 189
 190         /* collect a number of dirty meta pages and write together */
 191         if (wbc->for_kupdate ||
 192                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 193                 goto skip_write;
 194
 195         /* if mounting is failed, skip writing node pages */
 196         mutex_lock(&sbi->cp_mutex);
 197         diff = nr_pages_to_write(sbi, META, wbc);
 198         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 199         mutex_unlock(&sbi->cp_mutex);
 200         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 201         return 0;
 202
 203 skip_write:
 204         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 205         return 0;
 206 }
 207
 208 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 209                                                 long nr_to_write)
 210 {
 211         struct address_space *mapping = META_MAPPING(sbi);
 212         pgoff_t index = 0, end = LONG_MAX;
 213         struct pagevec pvec;
 214         long nwritten = 0;
 215         struct writeback_control wbc = {
 216                 .for_reclaim = 0,
 217         };
 218
 219         pagevec_init(&pvec, 0);
 220
 221         while (index <= end) {
 222                 int i, nr_pages;
 223                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 224                                 PAGECACHE_TAG_DIRTY,
 225                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 226                 if (unlikely(nr_pages == 0))
 227                         break;
 228
 229                 for (i = 0; i < nr_pages; i++) {
 230                         struct page *page = pvec.pages[i];
 231
 232                         lock_page(page);
 233
 234                         if (unlikely(page->mapping != mapping)) {
 235 continue_unlock:
 236                                 unlock_page(page);
 237                                 continue;
 238                         }
 239                         if (!PageDirty(page)) {
 240                                 /* someone wrote it for us */
 241                                 goto continue_unlock;
 242                         }
 243
 244                         if (!clear_page_dirty_for_io(page))
 245                                 goto continue_unlock;
 246
 247                         if (f2fs_write_meta_page(page, &wbc)) {
 248                                 unlock_page(page);
 249                                 break;
 250                         }
 251                         nwritten++;
 252                         if (unlikely(nwritten >= nr_to_write))
 253                                 break;
 254                 }
 255                 pagevec_release(&pvec);
 256                 cond_resched();
 257         }
 258
 259         if (nwritten)
 260                 f2fs_submit_merged_bio(sbi, type, WRITE);
 261
 262         return nwritten;
 263 }
 264
 265 static int f2fs_set_meta_page_dirty(struct page *page)
 266 {
 267         struct address_space *mapping = page->mapping;
 268         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 269
 270         trace_f2fs_set_page_dirty(page, META);
 271
 272         SetPageUptodate(page);
 273         if (!PageDirty(page)) {
 274                 __set_page_dirty_nobuffers(page);
 275                 inc_page_count(sbi, F2FS_DIRTY_META);
 276                 return 1;
 277         }
 278         return 0;
 279 }
 280
 281 const struct address_space_operations f2fs_meta_aops = {
 282         .writepage      = f2fs_write_meta_page,
 283         .writepages     = f2fs_write_meta_pages,
 284         .set_page_dirty = f2fs_set_meta_page_dirty,
 285 };
 286
 287 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 288 {
 289         int err = 0;
 290
 291         spin_lock(&sbi->orphan_inode_lock);
 292         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 293                 err = -ENOSPC;
 294         else
 295                 sbi->n_orphans++;
 296         spin_unlock(&sbi->orphan_inode_lock);
 297
 298         return err;
 299 }
 300
 301 void release_orphan_inode(struct f2fs_sb_info *sbi)
 302 {
 303         spin_lock(&sbi->orphan_inode_lock);
 304         f2fs_bug_on(sbi->n_orphans == 0);
 305         sbi->n_orphans--;
 306         spin_unlock(&sbi->orphan_inode_lock);
 307 }
 308
 309 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 310 {
 311         struct list_head *head, *this;
 312         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 313
 314         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 315         new->ino = ino;
 316
 317         spin_lock(&sbi->orphan_inode_lock);
 318         head = &sbi->orphan_inode_list;
 319         list_for_each(this, head) {
 320                 orphan = list_entry(this, struct orphan_inode_entry, list);
 321                 if (orphan->ino == ino) {
 322                         spin_unlock(&sbi->orphan_inode_lock);
 323                         kmem_cache_free(orphan_entry_slab, new);
 324                         return;
 325                 }
 326
 327                 if (orphan->ino > ino)
 328                         break;
 329                 orphan = NULL;
 330         }
 331
 332         /* add new_oentry into list which is sorted by inode number */
 333         if (orphan)
 334                 list_add(&new->list, this->prev);
 335         else
 336                 list_add_tail(&new->list, head);
 337         spin_unlock(&sbi->orphan_inode_lock);
 338 }
 339
 340 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 341 {
 342         struct list_head *head;
 343         struct orphan_inode_entry *orphan;
 344
 345         spin_lock(&sbi->orphan_inode_lock);
 346         head = &sbi->orphan_inode_list;
 347         list_for_each_entry(orphan, head, list) {
 348                 if (orphan->ino == ino) {
 349                         list_del(&orphan->list);
 350                         f2fs_bug_on(sbi->n_orphans == 0);
 351                         sbi->n_orphans--;
 352                         spin_unlock(&sbi->orphan_inode_lock);
 353                         kmem_cache_free(orphan_entry_slab, orphan);
 354                         return;
 355                 }
 356         }
 357         spin_unlock(&sbi->orphan_inode_lock);
 358 }
 359
 360 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 361 {
 362         struct inode *inode = f2fs_iget(sbi->sb, ino);
 363         f2fs_bug_on(IS_ERR(inode));
 364         clear_nlink(inode);
 365
 366         /* truncate all the data during iput */
 367         iput(inode);
 368 }
 369
 370 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 371 {
 372         block_t start_blk, orphan_blkaddr, i, j;
 373
 374         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 375                 return;
 376
 377         sbi->por_doing = true;
 378         start_blk = __start_cp_addr(sbi) + 1;
 379         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 380
 381         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 382
 383         for (i = 0; i < orphan_blkaddr; i++) {
 384                 struct page *page = get_meta_page(sbi, start_blk + i);
 385                 struct f2fs_orphan_block *orphan_blk;
 386
 387                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 388                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 389                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 390                         recover_orphan_inode(sbi, ino);
 391                 }
 392                 f2fs_put_page(page, 1);
 393         }
 394         /* clear Orphan Flag */
 395         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 396         sbi->por_doing = false;
 397         return;
 398 }
 399
 400 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 401 {
 402         struct list_head *head;
 403         struct f2fs_orphan_block *orphan_blk = NULL;
 404         unsigned int nentries = 0;
 405         unsigned short index;
 406         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 407                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 408         struct page *page = NULL;
 409         struct orphan_inode_entry *orphan = NULL;
 410
 411         for (index = 0; index < orphan_blocks; index++)
 412                 grab_meta_page(sbi, start_blk + index);
 413
 414         index = 1;
 415         spin_lock(&sbi->orphan_inode_lock);
 416         head = &sbi->orphan_inode_list;
 417
 418         /* loop for each orphan inode entry and write them in Jornal block */
 419         list_for_each_entry(orphan, head, list) {
 420                 if (!page) {
 421                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 422                         f2fs_bug_on(!page);
 423                         orphan_blk =
 424                                 (struct f2fs_orphan_block *)page_address(page);
 425                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 426                         f2fs_put_page(page, 0);
 427                 }
 428
 429                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 430
 431                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 432                         /*
 433                          * an orphan block is full of 1020 entries,
 434                          * then we need to flush current orphan blocks
 435                          * and bring another one in memory
 436                          */
 437                         orphan_blk->blk_addr = cpu_to_le16(index);
 438                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 439                         orphan_blk->entry_count = cpu_to_le32(nentries);
 440                         set_page_dirty(page);
 441                         f2fs_put_page(page, 1);
 442                         index++;
 443                         nentries = 0;
 444                         page = NULL;
 445                 }
 446         }
 447
 448         if (page) {
 449                 orphan_blk->blk_addr = cpu_to_le16(index);
 450                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 451                 orphan_blk->entry_count = cpu_to_le32(nentries);
 452                 set_page_dirty(page);
 453                 f2fs_put_page(page, 1);
 454         }
 455
 456         spin_unlock(&sbi->orphan_inode_lock);
 457 }
 458
 459 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 460                                 block_t cp_addr, unsigned long long *version)
 461 {
 462         struct page *cp_page_1, *cp_page_2 = NULL;
 463         unsigned long blk_size = sbi->blocksize;
 464         struct f2fs_checkpoint *cp_block;
 465         unsigned long long cur_version = 0, pre_version = 0;
 466         size_t crc_offset;
 467         __u32 crc = 0;
 468
 469         /* Read the 1st cp block in this CP pack */
 470         cp_page_1 = get_meta_page(sbi, cp_addr);
 471
 472         /* get the version number */
 473         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 474         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 475         if (crc_offset >= blk_size)
 476                 goto invalid_cp1;
 477
 478         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 479         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 480                 goto invalid_cp1;
 481
 482         pre_version = cur_cp_version(cp_block);
 483
 484         /* Read the 2nd cp block in this CP pack */
 485         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 486         cp_page_2 = get_meta_page(sbi, cp_addr);
 487
 488         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 489         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 490         if (crc_offset >= blk_size)
 491                 goto invalid_cp2;
 492
 493         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 494         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 495                 goto invalid_cp2;
 496
 497         cur_version = cur_cp_version(cp_block);
 498
 499         if (cur_version == pre_version) {
 500                 *version = cur_version;
 501                 f2fs_put_page(cp_page_2, 1);
 502                 return cp_page_1;
 503         }
 504 invalid_cp2:
 505         f2fs_put_page(cp_page_2, 1);
 506 invalid_cp1:
 507         f2fs_put_page(cp_page_1, 1);
 508         return NULL;
 509 }
 510
 511 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 512 {
 513         struct f2fs_checkpoint *cp_block;
 514         struct f2fs_super_block *fsb = sbi->raw_super;
 515         struct page *cp1, *cp2, *cur_page;
 516         unsigned long blk_size = sbi->blocksize;
 517         unsigned long long cp1_version = 0, cp2_version = 0;
 518         unsigned long long cp_start_blk_no;
 519
 520         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 521         if (!sbi->ckpt)
 522                 return -ENOMEM;
 523         /*
 524          * Finding out valid cp block involves read both
 525          * sets( cp pack1 and cp pack 2)
 526          */
 527         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 528         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 529
 530         /* The second checkpoint pack should start at the next segment */
 531         cp_start_blk_no += ((unsigned long long)1) <<
 532                                 le32_to_cpu(fsb->log_blocks_per_seg);
 533         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 534
 535         if (cp1 && cp2) {
 536                 if (ver_after(cp2_version, cp1_version))
 537                         cur_page = cp2;
 538                 else
 539                         cur_page = cp1;
 540         } else if (cp1) {
 541                 cur_page = cp1;
 542         } else if (cp2) {
 543                 cur_page = cp2;
 544         } else {
 545                 goto fail_no_cp;
 546         }
 547
 548         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 549         memcpy(sbi->ckpt, cp_block, blk_size);
 550
 551         f2fs_put_page(cp1, 1);
 552         f2fs_put_page(cp2, 1);
 553         return 0;
 554
 555 fail_no_cp:
 556         kfree(sbi->ckpt);
 557         return -EINVAL;
 558 }
 559
 560 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 561 {
 562         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 563         struct list_head *head = &sbi->dir_inode_list;
 564         struct list_head *this;
 565
 566         list_for_each(this, head) {
 567                 struct dir_inode_entry *entry;
 568                 entry = list_entry(this, struct dir_inode_entry, list);
 569                 if (unlikely(entry->inode == inode))
 570                         return -EEXIST;
 571         }
 572         list_add_tail(&new->list, head);
 573         stat_inc_dirty_dir(sbi);
 574         return 0;
 575 }
 576
 577 void set_dirty_dir_page(struct inode *inode, struct page *page)
 578 {
 579         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 580         struct dir_inode_entry *new;
 581         int ret = 0;
 582
 583         if (!S_ISDIR(inode->i_mode))
 584                 return;
 585
 586         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 587         new->inode = inode;
 588         INIT_LIST_HEAD(&new->list);
 589
 590         spin_lock(&sbi->dir_inode_lock);
 591         ret = __add_dirty_inode(inode, new);
 592         inode_inc_dirty_dents(inode);
 593         SetPagePrivate(page);
 594         spin_unlock(&sbi->dir_inode_lock);
 595
 596         if (ret)
 597                 kmem_cache_free(inode_entry_slab, new);
 598 }
 599
 600 void add_dirty_dir_inode(struct inode *inode)
 601 {
 602         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 603         struct dir_inode_entry *new =
 604                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 605         int ret = 0;
 606
 607         new->inode = inode;
 608         INIT_LIST_HEAD(&new->list);
 609
 610         spin_lock(&sbi->dir_inode_lock);
 611         ret = __add_dirty_inode(inode, new);
 612         spin_unlock(&sbi->dir_inode_lock);
 613
 614         if (ret)
 615                 kmem_cache_free(inode_entry_slab, new);
 616 }
 617
 618 void remove_dirty_dir_inode(struct inode *inode)
 619 {
 620         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 621         struct list_head *this, *head;
 622
 623         if (!S_ISDIR(inode->i_mode))
 624                 return;
 625
 626         spin_lock(&sbi->dir_inode_lock);
 627         if (get_dirty_dents(inode)) {
 628                 spin_unlock(&sbi->dir_inode_lock);
 629                 return;
 630         }
 631
 632         head = &sbi->dir_inode_list;
 633         list_for_each(this, head) {
 634                 struct dir_inode_entry *entry;
 635                 entry = list_entry(this, struct dir_inode_entry, list);
 636                 if (entry->inode == inode) {
 637                         list_del(&entry->list);
 638                         stat_dec_dirty_dir(sbi);
 639                         spin_unlock(&sbi->dir_inode_lock);
 640                         kmem_cache_free(inode_entry_slab, entry);
 641                         goto done;
 642                 }
 643         }
 644         spin_unlock(&sbi->dir_inode_lock);
 645
 646 done:
 647         /* Only from the recovery routine */
 648         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 649                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 650                 iput(inode);
 651         }
 652 }
 653
 654 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 655 {
 656
 657         struct list_head *this, *head;
 658         struct inode *inode = NULL;
 659
 660         spin_lock(&sbi->dir_inode_lock);
 661
 662         head = &sbi->dir_inode_list;
 663         list_for_each(this, head) {
 664                 struct dir_inode_entry *entry;
 665                 entry = list_entry(this, struct dir_inode_entry, list);
 666                 if (entry->inode->i_ino == ino) {
 667                         inode = entry->inode;
 668                         break;
 669                 }
 670         }
 671         spin_unlock(&sbi->dir_inode_lock);
 672         return inode;
 673 }
 674
 675 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 676 {
 677         struct list_head *head;
 678         struct dir_inode_entry *entry;
 679         struct inode *inode;
 680 retry:
 681         spin_lock(&sbi->dir_inode_lock);
 682
 683         head = &sbi->dir_inode_list;
 684         if (list_empty(head)) {
 685                 spin_unlock(&sbi->dir_inode_lock);
 686                 return;
 687         }
 688         entry = list_entry(head->next, struct dir_inode_entry, list);
 689         inode = igrab(entry->inode);
 690         spin_unlock(&sbi->dir_inode_lock);
 691         if (inode) {
 692                 filemap_fdatawrite(inode->i_mapping);
 693                 iput(inode);
 694         } else {
 695                 /*
 696                  * We should submit bio, since it exists several
 697                  * wribacking dentry pages in the freeing inode.
 698                  */
 699                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 700         }
 701         goto retry;
 702 }
 703
 704 /*
 705  * Freeze all the FS-operations for checkpoint.
 706  */
 707 static void block_operations(struct f2fs_sb_info *sbi)
 708 {
 709         struct writeback_control wbc = {
 710                 .sync_mode = WB_SYNC_ALL,
 711                 .nr_to_write = LONG_MAX,
 712                 .for_reclaim = 0,
 713         };
 714         struct blk_plug plug;
 715
 716         blk_start_plug(&plug);
 717
 718 retry_flush_dents:
 719         f2fs_lock_all(sbi);
 720         /* write all the dirty dentry pages */
 721         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 722                 f2fs_unlock_all(sbi);
 723                 sync_dirty_dir_inodes(sbi);
 724                 goto retry_flush_dents;
 725         }
 726
 727         /*
 728          * POR: we should ensure that there is no dirty node pages
 729          * until finishing nat/sit flush.
 730          */
 731 retry_flush_nodes:
 732         mutex_lock(&sbi->node_write);
 733
 734         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 735                 mutex_unlock(&sbi->node_write);
 736                 sync_node_pages(sbi, 0, &wbc);
 737                 goto retry_flush_nodes;
 738         }
 739         blk_finish_plug(&plug);
 740 }
 741
 742 static void unblock_operations(struct f2fs_sb_info *sbi)
 743 {
 744         mutex_unlock(&sbi->node_write);
 745         f2fs_unlock_all(sbi);
 746 }
 747
 748 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 749 {
 750         DEFINE_WAIT(wait);
 751
 752         for (;;) {
 753                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 754
 755                 if (!get_pages(sbi, F2FS_WRITEBACK))
 756                         break;
 757
 758                 io_schedule();
 759         }
 760         finish_wait(&sbi->cp_wait, &wait);
 761 }
 762
 763 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 764 {
 765         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 766         nid_t last_nid = 0;
 767         block_t start_blk;
 768         struct page *cp_page;
 769         unsigned int data_sum_blocks, orphan_blocks;
 770         __u32 crc32 = 0;
 771         void *kaddr;
 772         int i;
 773
 774         /* Flush all the NAT/SIT pages */
 775         while (get_pages(sbi, F2FS_DIRTY_META))
 776                 sync_meta_pages(sbi, META, LONG_MAX);
 777
 778         next_free_nid(sbi, &last_nid);
 779
 780         /*
 781          * modify checkpoint
 782          * version number is already updated
 783          */
 784         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 785         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 786         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 787         for (i = 0; i < 3; i++) {
 788                 ckpt->cur_node_segno[i] =
 789                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 790                 ckpt->cur_node_blkoff[i] =
 791                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 792                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 793                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 794         }
 795         for (i = 0; i < 3; i++) {
 796                 ckpt->cur_data_segno[i] =
 797                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 798                 ckpt->cur_data_blkoff[i] =
 799                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 800                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 801                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 802         }
 803
 804         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 805         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 806         ckpt->next_free_nid = cpu_to_le32(last_nid);
 807
 808         /* 2 cp  + n data seg summary + orphan inode blocks */
 809         data_sum_blocks = npages_for_summary_flush(sbi);
 810         if (data_sum_blocks < 3)
 811                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 812         else
 813                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 814
 815         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 816                                         / F2FS_ORPHANS_PER_BLOCK;
 817         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 818
 819         if (is_umount) {
 820                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 821                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 822                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 823         } else {
 824                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 825                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 826                         data_sum_blocks + orphan_blocks);
 827         }
 828
 829         if (sbi->n_orphans)
 830                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 831         else
 832                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 833
 834         /* update SIT/NAT bitmap */
 835         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 836         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 837
 838         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 839         *((__le32 *)((unsigned char *)ckpt +
 840                                 le32_to_cpu(ckpt->checksum_offset)))
 841                                 = cpu_to_le32(crc32);
 842
 843         start_blk = __start_cp_addr(sbi);
 844
 845         /* write out checkpoint buffer at block 0 */
 846         cp_page = grab_meta_page(sbi, start_blk++);
 847         kaddr = page_address(cp_page);
 848         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 849         set_page_dirty(cp_page);
 850         f2fs_put_page(cp_page, 1);
 851
 852         if (sbi->n_orphans) {
 853                 write_orphan_inodes(sbi, start_blk);
 854                 start_blk += orphan_blocks;
 855         }
 856
 857         write_data_summaries(sbi, start_blk);
 858         start_blk += data_sum_blocks;
 859         if (is_umount) {
 860                 write_node_summaries(sbi, start_blk);
 861                 start_blk += NR_CURSEG_NODE_TYPE;
 862         }
 863
 864         /* writeout checkpoint block */
 865         cp_page = grab_meta_page(sbi, start_blk);
 866         kaddr = page_address(cp_page);
 867         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 868         set_page_dirty(cp_page);
 869         f2fs_put_page(cp_page, 1);
 870
 871         /* wait for previous submitted node/meta pages writeback */
 872         wait_on_all_pages_writeback(sbi);
 873
 874         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 875         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 876
 877         /* update user_block_counts */
 878         sbi->last_valid_block_count = sbi->total_valid_block_count;
 879         sbi->alloc_valid_block_count = 0;
 880
 881         /* Here, we only have one bio having CP pack */
 882         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 883
 884         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 885                 clear_prefree_segments(sbi);
 886                 F2FS_RESET_SB_DIRT(sbi);
 887         }
 888 }
 889
 890 /*
 891  * We guarantee that this checkpoint procedure should not fail.
 892  */
 893 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 894 {
 895         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 896         unsigned long long ckpt_ver;
 897
 898         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 899
 900         mutex_lock(&sbi->cp_mutex);
 901         block_operations(sbi);
 902
 903         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 904
 905         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 906         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 907         f2fs_submit_merged_bio(sbi, META, WRITE);
 908
 909         /*
 910          * update checkpoint pack index
 911          * Increase the version number so that
 912          * SIT entries and seg summaries are written at correct place
 913          */
 914         ckpt_ver = cur_cp_version(ckpt);
 915         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 916
 917         /* write cached NAT/SIT entries to NAT/SIT area */
 918         flush_nat_entries(sbi);
 919         flush_sit_entries(sbi);
 920
 921         /* unlock all the fs_lock[] in do_checkpoint() */
 922         do_checkpoint(sbi, is_umount);
 923
 924         unblock_operations(sbi);
 925         mutex_unlock(&sbi->cp_mutex);
 926
 927         stat_inc_cp_count(sbi->stat_info);
 928         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 929 }
 930
 931 void init_orphan_info(struct f2fs_sb_info *sbi)
 932 {
 933         spin_lock_init(&sbi->orphan_inode_lock);
 934         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 935         sbi->n_orphans = 0;
 936         /*
 937          * considering 512 blocks in a segment 8 blocks are needed for cp
 938          * and log segment summaries. Remaining blocks are used to keep
 939          * orphan entries with the limitation one reserved segment
 940          * for cp pack we can have max 1020*504 orphan entries
 941          */
 942         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 943                                 * F2FS_ORPHANS_PER_BLOCK;
 944 }
 945
 946 int __init create_checkpoint_caches(void)
 947 {
 948         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 949                         sizeof(struct orphan_inode_entry));
 950         if (!orphan_entry_slab)
 951                 return -ENOMEM;
 952         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 953                         sizeof(struct dir_inode_entry));
 954         if (!inode_entry_slab) {
 955                 kmem_cache_destroy(orphan_entry_slab);
 956                 return -ENOMEM;
 957         }
 958         return 0;
 959 }
 960
 961 void destroy_checkpoint_caches(void)
 962 {
 963         kmem_cache_destroy(orphan_entry_slab);
 964         kmem_cache_destroy(inode_entry_slab);
 965 }