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_SSA:
  86         case META_CP:
  87                 return 0;
  88         default:
  89                 BUG();
  90         }
  91 }
  92
  93 /*
  94  * Readahead CP/NAT/SIT/SSA pages
  95  */
  96 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
  97 {
  98         block_t prev_blk_addr = 0;
  99         struct page *page;
 100         int blkno = start;
 101         int max_blks = get_max_meta_blks(sbi, type);
 102
 103         struct f2fs_io_info fio = {
 104                 .type = META,
 105                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 106         };
 107
 108         for (; nrpages-- > 0; blkno++) {
 109                 block_t blk_addr;
 110
 111                 switch (type) {
 112                 case META_NAT:
 113                         /* get nat block addr */
 114                         if (unlikely(blkno >= max_blks))
 115                                 blkno = 0;
 116                         blk_addr = current_nat_addr(sbi,
 117                                         blkno * NAT_ENTRY_PER_BLOCK);
 118                         break;
 119                 case META_SIT:
 120                         /* get sit block addr */
 121                         if (unlikely(blkno >= max_blks))
 122                                 goto out;
 123                         blk_addr = current_sit_addr(sbi,
 124                                         blkno * SIT_ENTRY_PER_BLOCK);
 125                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 126                                 goto out;
 127                         prev_blk_addr = blk_addr;
 128                         break;
 129                 case META_SSA:
 130                 case META_CP:
 131                         /* get ssa/cp block addr */
 132                         blk_addr = blkno;
 133                         break;
 134                 default:
 135                         BUG();
 136                 }
 137
 138                 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
 139                 if (!page)
 140                         continue;
 141                 if (PageUptodate(page)) {
 142                         mark_page_accessed(page);
 143                         f2fs_put_page(page, 1);
 144                         continue;
 145                 }
 146
 147                 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
 148                 mark_page_accessed(page);
 149                 f2fs_put_page(page, 0);
 150         }
 151 out:
 152         f2fs_submit_merged_bio(sbi, META, READ);
 153         return blkno - start;
 154 }
 155
 156 static int f2fs_write_meta_page(struct page *page,
 157                                 struct writeback_control *wbc)
 158 {
 159         struct inode *inode = page->mapping->host;
 160         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 161
 162         if (unlikely(sbi->por_doing))
 163                 goto redirty_out;
 164         if (wbc->for_reclaim)
 165                 goto redirty_out;
 166
 167         /* Should not write any meta pages, if any IO error was occurred */
 168         if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
 169                 goto no_write;
 170
 171         wait_on_page_writeback(page);
 172         write_meta_page(sbi, page);
 173 no_write:
 174         dec_page_count(sbi, F2FS_DIRTY_META);
 175         unlock_page(page);
 176         return 0;
 177
 178 redirty_out:
 179         dec_page_count(sbi, F2FS_DIRTY_META);
 180         wbc->pages_skipped++;
 181         set_page_dirty(page);
 182         return AOP_WRITEPAGE_ACTIVATE;
 183 }
 184
 185 static int f2fs_write_meta_pages(struct address_space *mapping,
 186                                 struct writeback_control *wbc)
 187 {
 188         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 189         int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
 190         long written;
 191
 192         if (wbc->for_kupdate)
 193                 return 0;
 194
 195         /* collect a number of dirty meta pages and write together */
 196         if (get_pages(sbi, F2FS_DIRTY_META) < nrpages)
 197                 return 0;
 198
 199         /* if mounting is failed, skip writing node pages */
 200         mutex_lock(&sbi->cp_mutex);
 201         written = sync_meta_pages(sbi, META, nrpages);
 202         mutex_unlock(&sbi->cp_mutex);
 203         wbc->nr_to_write -= written;
 204         return 0;
 205 }
 206
 207 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 208                                                 long nr_to_write)
 209 {
 210         struct address_space *mapping = META_MAPPING(sbi);
 211         pgoff_t index = 0, end = LONG_MAX;
 212         struct pagevec pvec;
 213         long nwritten = 0;
 214         struct writeback_control wbc = {
 215                 .for_reclaim = 0,
 216         };
 217
 218         pagevec_init(&pvec, 0);
 219
 220         while (index <= end) {
 221                 int i, nr_pages;
 222                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 223                                 PAGECACHE_TAG_DIRTY,
 224                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 225                 if (unlikely(nr_pages == 0))
 226                         break;
 227
 228                 for (i = 0; i < nr_pages; i++) {
 229                         struct page *page = pvec.pages[i];
 230
 231                         lock_page(page);
 232
 233                         if (unlikely(page->mapping != mapping)) {
 234 continue_unlock:
 235                                 unlock_page(page);
 236                                 continue;
 237                         }
 238                         if (!PageDirty(page)) {
 239                                 /* someone wrote it for us */
 240                                 goto continue_unlock;
 241                         }
 242
 243                         if (!clear_page_dirty_for_io(page))
 244                                 goto continue_unlock;
 245
 246                         if (f2fs_write_meta_page(page, &wbc)) {
 247                                 unlock_page(page);
 248                                 break;
 249                         }
 250                         nwritten++;
 251                         if (unlikely(nwritten >= nr_to_write))
 252                                 break;
 253                 }
 254                 pagevec_release(&pvec);
 255                 cond_resched();
 256         }
 257
 258         if (nwritten)
 259                 f2fs_submit_merged_bio(sbi, type, WRITE);
 260
 261         return nwritten;
 262 }
 263
 264 static int f2fs_set_meta_page_dirty(struct page *page)
 265 {
 266         struct address_space *mapping = page->mapping;
 267         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 268
 269         trace_f2fs_set_page_dirty(page, META);
 270
 271         SetPageUptodate(page);
 272         if (!PageDirty(page)) {
 273                 __set_page_dirty_nobuffers(page);
 274                 inc_page_count(sbi, F2FS_DIRTY_META);
 275                 return 1;
 276         }
 277         return 0;
 278 }
 279
 280 const struct address_space_operations f2fs_meta_aops = {
 281         .writepage      = f2fs_write_meta_page,
 282         .writepages     = f2fs_write_meta_pages,
 283         .set_page_dirty = f2fs_set_meta_page_dirty,
 284 };
 285
 286 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 287 {
 288         int err = 0;
 289
 290         spin_lock(&sbi->orphan_inode_lock);
 291         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 292                 err = -ENOSPC;
 293         else
 294                 sbi->n_orphans++;
 295         spin_unlock(&sbi->orphan_inode_lock);
 296
 297         return err;
 298 }
 299
 300 void release_orphan_inode(struct f2fs_sb_info *sbi)
 301 {
 302         spin_lock(&sbi->orphan_inode_lock);
 303         f2fs_bug_on(sbi->n_orphans == 0);
 304         sbi->n_orphans--;
 305         spin_unlock(&sbi->orphan_inode_lock);
 306 }
 307
 308 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 309 {
 310         struct list_head *head, *this;
 311         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 312
 313         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 314         new->ino = ino;
 315
 316         spin_lock(&sbi->orphan_inode_lock);
 317         head = &sbi->orphan_inode_list;
 318         list_for_each(this, head) {
 319                 orphan = list_entry(this, struct orphan_inode_entry, list);
 320                 if (orphan->ino == ino) {
 321                         spin_unlock(&sbi->orphan_inode_lock);
 322                         kmem_cache_free(orphan_entry_slab, new);
 323                         return;
 324                 }
 325
 326                 if (orphan->ino > ino)
 327                         break;
 328                 orphan = NULL;
 329         }
 330
 331         /* add new_oentry into list which is sorted by inode number */
 332         if (orphan)
 333                 list_add(&new->list, this->prev);
 334         else
 335                 list_add_tail(&new->list, head);
 336         spin_unlock(&sbi->orphan_inode_lock);
 337 }
 338
 339 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 340 {
 341         struct list_head *head;
 342         struct orphan_inode_entry *orphan;
 343
 344         spin_lock(&sbi->orphan_inode_lock);
 345         head = &sbi->orphan_inode_list;
 346         list_for_each_entry(orphan, head, list) {
 347                 if (orphan->ino == ino) {
 348                         list_del(&orphan->list);
 349                         kmem_cache_free(orphan_entry_slab, orphan);
 350                         f2fs_bug_on(sbi->n_orphans == 0);
 351                         sbi->n_orphans--;
 352                         break;
 353                 }
 354         }
 355         spin_unlock(&sbi->orphan_inode_lock);
 356 }
 357
 358 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 359 {
 360         struct inode *inode = f2fs_iget(sbi->sb, ino);
 361         f2fs_bug_on(IS_ERR(inode));
 362         clear_nlink(inode);
 363
 364         /* truncate all the data during iput */
 365         iput(inode);
 366 }
 367
 368 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 369 {
 370         block_t start_blk, orphan_blkaddr, i, j;
 371
 372         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 373                 return;
 374
 375         sbi->por_doing = true;
 376         start_blk = __start_cp_addr(sbi) + 1;
 377         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 378
 379         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 380
 381         for (i = 0; i < orphan_blkaddr; i++) {
 382                 struct page *page = get_meta_page(sbi, start_blk + i);
 383                 struct f2fs_orphan_block *orphan_blk;
 384
 385                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 386                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 387                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 388                         recover_orphan_inode(sbi, ino);
 389                 }
 390                 f2fs_put_page(page, 1);
 391         }
 392         /* clear Orphan Flag */
 393         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 394         sbi->por_doing = false;
 395         return;
 396 }
 397
 398 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 399 {
 400         struct list_head *head;
 401         struct f2fs_orphan_block *orphan_blk = NULL;
 402         unsigned int nentries = 0;
 403         unsigned short index;
 404         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 405                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 406         struct page *page = NULL;
 407         struct orphan_inode_entry *orphan = NULL;
 408
 409         for (index = 0; index < orphan_blocks; index++)
 410                 grab_meta_page(sbi, start_blk + index);
 411
 412         index = 1;
 413         spin_lock(&sbi->orphan_inode_lock);
 414         head = &sbi->orphan_inode_list;
 415
 416         /* loop for each orphan inode entry and write them in Jornal block */
 417         list_for_each_entry(orphan, head, list) {
 418                 if (!page) {
 419                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 420                         f2fs_bug_on(!page);
 421                         orphan_blk =
 422                                 (struct f2fs_orphan_block *)page_address(page);
 423                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 424                         f2fs_put_page(page, 0);
 425                 }
 426
 427                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 428
 429                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 430                         /*
 431                          * an orphan block is full of 1020 entries,
 432                          * then we need to flush current orphan blocks
 433                          * and bring another one in memory
 434                          */
 435                         orphan_blk->blk_addr = cpu_to_le16(index);
 436                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 437                         orphan_blk->entry_count = cpu_to_le32(nentries);
 438                         set_page_dirty(page);
 439                         f2fs_put_page(page, 1);
 440                         index++;
 441                         nentries = 0;
 442                         page = NULL;
 443                 }
 444         }
 445
 446         if (page) {
 447                 orphan_blk->blk_addr = cpu_to_le16(index);
 448                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 449                 orphan_blk->entry_count = cpu_to_le32(nentries);
 450                 set_page_dirty(page);
 451                 f2fs_put_page(page, 1);
 452         }
 453
 454         spin_unlock(&sbi->orphan_inode_lock);
 455 }
 456
 457 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 458                                 block_t cp_addr, unsigned long long *version)
 459 {
 460         struct page *cp_page_1, *cp_page_2 = NULL;
 461         unsigned long blk_size = sbi->blocksize;
 462         struct f2fs_checkpoint *cp_block;
 463         unsigned long long cur_version = 0, pre_version = 0;
 464         size_t crc_offset;
 465         __u32 crc = 0;
 466
 467         /* Read the 1st cp block in this CP pack */
 468         cp_page_1 = get_meta_page(sbi, cp_addr);
 469
 470         /* get the version number */
 471         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 472         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 473         if (crc_offset >= blk_size)
 474                 goto invalid_cp1;
 475
 476         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 477         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 478                 goto invalid_cp1;
 479
 480         pre_version = cur_cp_version(cp_block);
 481
 482         /* Read the 2nd cp block in this CP pack */
 483         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 484         cp_page_2 = get_meta_page(sbi, cp_addr);
 485
 486         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 487         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 488         if (crc_offset >= blk_size)
 489                 goto invalid_cp2;
 490
 491         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 492         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 493                 goto invalid_cp2;
 494
 495         cur_version = cur_cp_version(cp_block);
 496
 497         if (cur_version == pre_version) {
 498                 *version = cur_version;
 499                 f2fs_put_page(cp_page_2, 1);
 500                 return cp_page_1;
 501         }
 502 invalid_cp2:
 503         f2fs_put_page(cp_page_2, 1);
 504 invalid_cp1:
 505         f2fs_put_page(cp_page_1, 1);
 506         return NULL;
 507 }
 508
 509 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 510 {
 511         struct f2fs_checkpoint *cp_block;
 512         struct f2fs_super_block *fsb = sbi->raw_super;
 513         struct page *cp1, *cp2, *cur_page;
 514         unsigned long blk_size = sbi->blocksize;
 515         unsigned long long cp1_version = 0, cp2_version = 0;
 516         unsigned long long cp_start_blk_no;
 517
 518         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 519         if (!sbi->ckpt)
 520                 return -ENOMEM;
 521         /*
 522          * Finding out valid cp block involves read both
 523          * sets( cp pack1 and cp pack 2)
 524          */
 525         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 526         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 527
 528         /* The second checkpoint pack should start at the next segment */
 529         cp_start_blk_no += ((unsigned long long)1) <<
 530                                 le32_to_cpu(fsb->log_blocks_per_seg);
 531         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 532
 533         if (cp1 && cp2) {
 534                 if (ver_after(cp2_version, cp1_version))
 535                         cur_page = cp2;
 536                 else
 537                         cur_page = cp1;
 538         } else if (cp1) {
 539                 cur_page = cp1;
 540         } else if (cp2) {
 541                 cur_page = cp2;
 542         } else {
 543                 goto fail_no_cp;
 544         }
 545
 546         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 547         memcpy(sbi->ckpt, cp_block, blk_size);
 548
 549         f2fs_put_page(cp1, 1);
 550         f2fs_put_page(cp2, 1);
 551         return 0;
 552
 553 fail_no_cp:
 554         kfree(sbi->ckpt);
 555         return -EINVAL;
 556 }
 557
 558 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 559 {
 560         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 561         struct list_head *head = &sbi->dir_inode_list;
 562         struct list_head *this;
 563
 564         list_for_each(this, head) {
 565                 struct dir_inode_entry *entry;
 566                 entry = list_entry(this, struct dir_inode_entry, list);
 567                 if (unlikely(entry->inode == inode))
 568                         return -EEXIST;
 569         }
 570         list_add_tail(&new->list, head);
 571         stat_inc_dirty_dir(sbi);
 572         return 0;
 573 }
 574
 575 void set_dirty_dir_page(struct inode *inode, struct page *page)
 576 {
 577         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 578         struct dir_inode_entry *new;
 579
 580         if (!S_ISDIR(inode->i_mode))
 581                 return;
 582
 583         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 584         new->inode = inode;
 585         INIT_LIST_HEAD(&new->list);
 586
 587         spin_lock(&sbi->dir_inode_lock);
 588         if (__add_dirty_inode(inode, new))
 589                 kmem_cache_free(inode_entry_slab, new);
 590
 591         inode_inc_dirty_dents(inode);
 592         SetPagePrivate(page);
 593         spin_unlock(&sbi->dir_inode_lock);
 594 }
 595
 596 void add_dirty_dir_inode(struct inode *inode)
 597 {
 598         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 599         struct dir_inode_entry *new =
 600                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 601
 602         new->inode = inode;
 603         INIT_LIST_HEAD(&new->list);
 604
 605         spin_lock(&sbi->dir_inode_lock);
 606         if (__add_dirty_inode(inode, new))
 607                 kmem_cache_free(inode_entry_slab, new);
 608         spin_unlock(&sbi->dir_inode_lock);
 609 }
 610
 611 void remove_dirty_dir_inode(struct inode *inode)
 612 {
 613         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 614
 615         struct list_head *this, *head;
 616
 617         if (!S_ISDIR(inode->i_mode))
 618                 return;
 619
 620         spin_lock(&sbi->dir_inode_lock);
 621         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 622                 spin_unlock(&sbi->dir_inode_lock);
 623                 return;
 624         }
 625
 626         head = &sbi->dir_inode_list;
 627         list_for_each(this, head) {
 628                 struct dir_inode_entry *entry;
 629                 entry = list_entry(this, struct dir_inode_entry, list);
 630                 if (entry->inode == inode) {
 631                         list_del(&entry->list);
 632                         kmem_cache_free(inode_entry_slab, entry);
 633                         stat_dec_dirty_dir(sbi);
 634                         break;
 635                 }
 636         }
 637         spin_unlock(&sbi->dir_inode_lock);
 638
 639         /* Only from the recovery routine */
 640         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 641                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 642                 iput(inode);
 643         }
 644 }
 645
 646 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 647 {
 648
 649         struct list_head *this, *head;
 650         struct inode *inode = NULL;
 651
 652         spin_lock(&sbi->dir_inode_lock);
 653
 654         head = &sbi->dir_inode_list;
 655         list_for_each(this, head) {
 656                 struct dir_inode_entry *entry;
 657                 entry = list_entry(this, struct dir_inode_entry, list);
 658                 if (entry->inode->i_ino == ino) {
 659                         inode = entry->inode;
 660                         break;
 661                 }
 662         }
 663         spin_unlock(&sbi->dir_inode_lock);
 664         return inode;
 665 }
 666
 667 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 668 {
 669         struct list_head *head;
 670         struct dir_inode_entry *entry;
 671         struct inode *inode;
 672 retry:
 673         spin_lock(&sbi->dir_inode_lock);
 674
 675         head = &sbi->dir_inode_list;
 676         if (list_empty(head)) {
 677                 spin_unlock(&sbi->dir_inode_lock);
 678                 return;
 679         }
 680         entry = list_entry(head->next, struct dir_inode_entry, list);
 681         inode = igrab(entry->inode);
 682         spin_unlock(&sbi->dir_inode_lock);
 683         if (inode) {
 684                 filemap_flush(inode->i_mapping);
 685                 iput(inode);
 686         } else {
 687                 /*
 688                  * We should submit bio, since it exists several
 689                  * wribacking dentry pages in the freeing inode.
 690                  */
 691                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 692         }
 693         goto retry;
 694 }
 695
 696 /*
 697  * Freeze all the FS-operations for checkpoint.
 698  */
 699 static void block_operations(struct f2fs_sb_info *sbi)
 700 {
 701         struct writeback_control wbc = {
 702                 .sync_mode = WB_SYNC_ALL,
 703                 .nr_to_write = LONG_MAX,
 704                 .for_reclaim = 0,
 705         };
 706         struct blk_plug plug;
 707
 708         blk_start_plug(&plug);
 709
 710 retry_flush_dents:
 711         f2fs_lock_all(sbi);
 712         /* write all the dirty dentry pages */
 713         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 714                 f2fs_unlock_all(sbi);
 715                 sync_dirty_dir_inodes(sbi);
 716                 goto retry_flush_dents;
 717         }
 718
 719         /*
 720          * POR: we should ensure that there is no dirty node pages
 721          * until finishing nat/sit flush.
 722          */
 723 retry_flush_nodes:
 724         mutex_lock(&sbi->node_write);
 725
 726         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 727                 mutex_unlock(&sbi->node_write);
 728                 sync_node_pages(sbi, 0, &wbc);
 729                 goto retry_flush_nodes;
 730         }
 731         blk_finish_plug(&plug);
 732 }
 733
 734 static void unblock_operations(struct f2fs_sb_info *sbi)
 735 {
 736         mutex_unlock(&sbi->node_write);
 737         f2fs_unlock_all(sbi);
 738 }
 739
 740 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 741 {
 742         DEFINE_WAIT(wait);
 743
 744         for (;;) {
 745                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 746
 747                 if (!get_pages(sbi, F2FS_WRITEBACK))
 748                         break;
 749
 750                 io_schedule();
 751         }
 752         finish_wait(&sbi->cp_wait, &wait);
 753 }
 754
 755 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 756 {
 757         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 758         nid_t last_nid = 0;
 759         block_t start_blk;
 760         struct page *cp_page;
 761         unsigned int data_sum_blocks, orphan_blocks;
 762         __u32 crc32 = 0;
 763         void *kaddr;
 764         int i;
 765
 766         /* Flush all the NAT/SIT pages */
 767         while (get_pages(sbi, F2FS_DIRTY_META))
 768                 sync_meta_pages(sbi, META, LONG_MAX);
 769
 770         next_free_nid(sbi, &last_nid);
 771
 772         /*
 773          * modify checkpoint
 774          * version number is already updated
 775          */
 776         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 777         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 778         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 779         for (i = 0; i < 3; i++) {
 780                 ckpt->cur_node_segno[i] =
 781                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 782                 ckpt->cur_node_blkoff[i] =
 783                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 784                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 785                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 786         }
 787         for (i = 0; i < 3; i++) {
 788                 ckpt->cur_data_segno[i] =
 789                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 790                 ckpt->cur_data_blkoff[i] =
 791                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 792                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 793                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 794         }
 795
 796         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 797         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 798         ckpt->next_free_nid = cpu_to_le32(last_nid);
 799
 800         /* 2 cp  + n data seg summary + orphan inode blocks */
 801         data_sum_blocks = npages_for_summary_flush(sbi);
 802         if (data_sum_blocks < 3)
 803                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 804         else
 805                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 806
 807         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 808                                         / F2FS_ORPHANS_PER_BLOCK;
 809         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 810
 811         if (is_umount) {
 812                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 813                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 814                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 815         } else {
 816                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 817                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 818                         data_sum_blocks + orphan_blocks);
 819         }
 820
 821         if (sbi->n_orphans)
 822                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 823         else
 824                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 825
 826         /* update SIT/NAT bitmap */
 827         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 828         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 829
 830         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 831         *((__le32 *)((unsigned char *)ckpt +
 832                                 le32_to_cpu(ckpt->checksum_offset)))
 833                                 = cpu_to_le32(crc32);
 834
 835         start_blk = __start_cp_addr(sbi);
 836
 837         /* write out checkpoint buffer at block 0 */
 838         cp_page = grab_meta_page(sbi, start_blk++);
 839         kaddr = page_address(cp_page);
 840         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 841         set_page_dirty(cp_page);
 842         f2fs_put_page(cp_page, 1);
 843
 844         if (sbi->n_orphans) {
 845                 write_orphan_inodes(sbi, start_blk);
 846                 start_blk += orphan_blocks;
 847         }
 848
 849         write_data_summaries(sbi, start_blk);
 850         start_blk += data_sum_blocks;
 851         if (is_umount) {
 852                 write_node_summaries(sbi, start_blk);
 853                 start_blk += NR_CURSEG_NODE_TYPE;
 854         }
 855
 856         /* writeout checkpoint block */
 857         cp_page = grab_meta_page(sbi, start_blk);
 858         kaddr = page_address(cp_page);
 859         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 860         set_page_dirty(cp_page);
 861         f2fs_put_page(cp_page, 1);
 862
 863         /* wait for previous submitted node/meta pages writeback */
 864         wait_on_all_pages_writeback(sbi);
 865
 866         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 867         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 868
 869         /* update user_block_counts */
 870         sbi->last_valid_block_count = sbi->total_valid_block_count;
 871         sbi->alloc_valid_block_count = 0;
 872
 873         /* Here, we only have one bio having CP pack */
 874         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 875
 876         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 877                 clear_prefree_segments(sbi);
 878                 F2FS_RESET_SB_DIRT(sbi);
 879         }
 880 }
 881
 882 /*
 883  * We guarantee that this checkpoint procedure should not fail.
 884  */
 885 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 886 {
 887         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 888         unsigned long long ckpt_ver;
 889
 890         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 891
 892         mutex_lock(&sbi->cp_mutex);
 893         block_operations(sbi);
 894
 895         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 896
 897         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 898         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 899         f2fs_submit_merged_bio(sbi, META, WRITE);
 900
 901         /*
 902          * update checkpoint pack index
 903          * Increase the version number so that
 904          * SIT entries and seg summaries are written at correct place
 905          */
 906         ckpt_ver = cur_cp_version(ckpt);
 907         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 908
 909         /* write cached NAT/SIT entries to NAT/SIT area */
 910         flush_nat_entries(sbi);
 911         flush_sit_entries(sbi);
 912
 913         /* unlock all the fs_lock[] in do_checkpoint() */
 914         do_checkpoint(sbi, is_umount);
 915
 916         unblock_operations(sbi);
 917         mutex_unlock(&sbi->cp_mutex);
 918
 919         stat_inc_cp_count(sbi->stat_info);
 920         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 921 }
 922
 923 void init_orphan_info(struct f2fs_sb_info *sbi)
 924 {
 925         spin_lock_init(&sbi->orphan_inode_lock);
 926         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 927         sbi->n_orphans = 0;
 928         /*
 929          * considering 512 blocks in a segment 8 blocks are needed for cp
 930          * and log segment summaries. Remaining blocks are used to keep
 931          * orphan entries with the limitation one reserved segment
 932          * for cp pack we can have max 1020*504 orphan entries
 933          */
 934         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 935                                 * F2FS_ORPHANS_PER_BLOCK;
 936 }
 937
 938 int __init create_checkpoint_caches(void)
 939 {
 940         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 941                         sizeof(struct orphan_inode_entry), NULL);
 942         if (!orphan_entry_slab)
 943                 return -ENOMEM;
 944         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 945                         sizeof(struct dir_inode_entry), NULL);
 946         if (!inode_entry_slab) {
 947                 kmem_cache_destroy(orphan_entry_slab);
 948                 return -ENOMEM;
 949         }
 950         return 0;
 951 }
 952
 953 void destroy_checkpoint_caches(void)
 954 {
 955         kmem_cache_destroy(orphan_entry_slab);
 956         kmem_cache_destroy(inode_entry_slab);
 957 }