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         account_page_redirty(page);
 182         set_page_dirty(page);
 183         return AOP_WRITEPAGE_ACTIVATE;
 184 }
 185
 186 static int f2fs_write_meta_pages(struct address_space *mapping,
 187                                 struct writeback_control *wbc)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 190         int nrpages = nr_pages_to_skip(sbi, META);
 191         long written;
 192
 193         if (wbc->for_kupdate)
 194                 return 0;
 195
 196         /* collect a number of dirty meta pages and write together */
 197         if (get_pages(sbi, F2FS_DIRTY_META) < nrpages)
 198                 return 0;
 199
 200         /* if mounting is failed, skip writing node pages */
 201         mutex_lock(&sbi->cp_mutex);
 202         written = sync_meta_pages(sbi, META, nrpages);
 203         mutex_unlock(&sbi->cp_mutex);
 204         wbc->nr_to_write -= written;
 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                         kmem_cache_free(orphan_entry_slab, orphan);
 351                         f2fs_bug_on(sbi->n_orphans == 0);
 352                         sbi->n_orphans--;
 353                         break;
 354                 }
 355         }
 356         spin_unlock(&sbi->orphan_inode_lock);
 357 }
 358
 359 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 360 {
 361         struct inode *inode = f2fs_iget(sbi->sb, ino);
 362         f2fs_bug_on(IS_ERR(inode));
 363         clear_nlink(inode);
 364
 365         /* truncate all the data during iput */
 366         iput(inode);
 367 }
 368
 369 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 370 {
 371         block_t start_blk, orphan_blkaddr, i, j;
 372
 373         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 374                 return;
 375
 376         sbi->por_doing = true;
 377         start_blk = __start_cp_addr(sbi) + 1;
 378         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 379
 380         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 381
 382         for (i = 0; i < orphan_blkaddr; i++) {
 383                 struct page *page = get_meta_page(sbi, start_blk + i);
 384                 struct f2fs_orphan_block *orphan_blk;
 385
 386                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 387                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 388                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 389                         recover_orphan_inode(sbi, ino);
 390                 }
 391                 f2fs_put_page(page, 1);
 392         }
 393         /* clear Orphan Flag */
 394         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 395         sbi->por_doing = false;
 396         return;
 397 }
 398
 399 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 400 {
 401         struct list_head *head;
 402         struct f2fs_orphan_block *orphan_blk = NULL;
 403         unsigned int nentries = 0;
 404         unsigned short index;
 405         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 406                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 407         struct page *page = NULL;
 408         struct orphan_inode_entry *orphan = NULL;
 409
 410         for (index = 0; index < orphan_blocks; index++)
 411                 grab_meta_page(sbi, start_blk + index);
 412
 413         index = 1;
 414         spin_lock(&sbi->orphan_inode_lock);
 415         head = &sbi->orphan_inode_list;
 416
 417         /* loop for each orphan inode entry and write them in Jornal block */
 418         list_for_each_entry(orphan, head, list) {
 419                 if (!page) {
 420                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 421                         f2fs_bug_on(!page);
 422                         orphan_blk =
 423                                 (struct f2fs_orphan_block *)page_address(page);
 424                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 425                         f2fs_put_page(page, 0);
 426                 }
 427
 428                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 429
 430                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 431                         /*
 432                          * an orphan block is full of 1020 entries,
 433                          * then we need to flush current orphan blocks
 434                          * and bring another one in memory
 435                          */
 436                         orphan_blk->blk_addr = cpu_to_le16(index);
 437                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 438                         orphan_blk->entry_count = cpu_to_le32(nentries);
 439                         set_page_dirty(page);
 440                         f2fs_put_page(page, 1);
 441                         index++;
 442                         nentries = 0;
 443                         page = NULL;
 444                 }
 445         }
 446
 447         if (page) {
 448                 orphan_blk->blk_addr = cpu_to_le16(index);
 449                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 450                 orphan_blk->entry_count = cpu_to_le32(nentries);
 451                 set_page_dirty(page);
 452                 f2fs_put_page(page, 1);
 453         }
 454
 455         spin_unlock(&sbi->orphan_inode_lock);
 456 }
 457
 458 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 459                                 block_t cp_addr, unsigned long long *version)
 460 {
 461         struct page *cp_page_1, *cp_page_2 = NULL;
 462         unsigned long blk_size = sbi->blocksize;
 463         struct f2fs_checkpoint *cp_block;
 464         unsigned long long cur_version = 0, pre_version = 0;
 465         size_t crc_offset;
 466         __u32 crc = 0;
 467
 468         /* Read the 1st cp block in this CP pack */
 469         cp_page_1 = get_meta_page(sbi, cp_addr);
 470
 471         /* get the version number */
 472         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 473         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 474         if (crc_offset >= blk_size)
 475                 goto invalid_cp1;
 476
 477         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 478         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 479                 goto invalid_cp1;
 480
 481         pre_version = cur_cp_version(cp_block);
 482
 483         /* Read the 2nd cp block in this CP pack */
 484         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 485         cp_page_2 = get_meta_page(sbi, cp_addr);
 486
 487         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 488         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 489         if (crc_offset >= blk_size)
 490                 goto invalid_cp2;
 491
 492         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 493         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 494                 goto invalid_cp2;
 495
 496         cur_version = cur_cp_version(cp_block);
 497
 498         if (cur_version == pre_version) {
 499                 *version = cur_version;
 500                 f2fs_put_page(cp_page_2, 1);
 501                 return cp_page_1;
 502         }
 503 invalid_cp2:
 504         f2fs_put_page(cp_page_2, 1);
 505 invalid_cp1:
 506         f2fs_put_page(cp_page_1, 1);
 507         return NULL;
 508 }
 509
 510 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 511 {
 512         struct f2fs_checkpoint *cp_block;
 513         struct f2fs_super_block *fsb = sbi->raw_super;
 514         struct page *cp1, *cp2, *cur_page;
 515         unsigned long blk_size = sbi->blocksize;
 516         unsigned long long cp1_version = 0, cp2_version = 0;
 517         unsigned long long cp_start_blk_no;
 518
 519         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 520         if (!sbi->ckpt)
 521                 return -ENOMEM;
 522         /*
 523          * Finding out valid cp block involves read both
 524          * sets( cp pack1 and cp pack 2)
 525          */
 526         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 527         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 528
 529         /* The second checkpoint pack should start at the next segment */
 530         cp_start_blk_no += ((unsigned long long)1) <<
 531                                 le32_to_cpu(fsb->log_blocks_per_seg);
 532         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 533
 534         if (cp1 && cp2) {
 535                 if (ver_after(cp2_version, cp1_version))
 536                         cur_page = cp2;
 537                 else
 538                         cur_page = cp1;
 539         } else if (cp1) {
 540                 cur_page = cp1;
 541         } else if (cp2) {
 542                 cur_page = cp2;
 543         } else {
 544                 goto fail_no_cp;
 545         }
 546
 547         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 548         memcpy(sbi->ckpt, cp_block, blk_size);
 549
 550         f2fs_put_page(cp1, 1);
 551         f2fs_put_page(cp2, 1);
 552         return 0;
 553
 554 fail_no_cp:
 555         kfree(sbi->ckpt);
 556         return -EINVAL;
 557 }
 558
 559 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 560 {
 561         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 562         struct list_head *head = &sbi->dir_inode_list;
 563         struct list_head *this;
 564
 565         list_for_each(this, head) {
 566                 struct dir_inode_entry *entry;
 567                 entry = list_entry(this, struct dir_inode_entry, list);
 568                 if (unlikely(entry->inode == inode))
 569                         return -EEXIST;
 570         }
 571         list_add_tail(&new->list, head);
 572         stat_inc_dirty_dir(sbi);
 573         return 0;
 574 }
 575
 576 void set_dirty_dir_page(struct inode *inode, struct page *page)
 577 {
 578         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 579         struct dir_inode_entry *new;
 580
 581         if (!S_ISDIR(inode->i_mode))
 582                 return;
 583
 584         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 585         new->inode = inode;
 586         INIT_LIST_HEAD(&new->list);
 587
 588         spin_lock(&sbi->dir_inode_lock);
 589         if (__add_dirty_inode(inode, new))
 590                 kmem_cache_free(inode_entry_slab, new);
 591
 592         inode_inc_dirty_dents(inode);
 593         SetPagePrivate(page);
 594         spin_unlock(&sbi->dir_inode_lock);
 595 }
 596
 597 void add_dirty_dir_inode(struct inode *inode)
 598 {
 599         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 600         struct dir_inode_entry *new =
 601                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 602
 603         new->inode = inode;
 604         INIT_LIST_HEAD(&new->list);
 605
 606         spin_lock(&sbi->dir_inode_lock);
 607         if (__add_dirty_inode(inode, new))
 608                 kmem_cache_free(inode_entry_slab, new);
 609         spin_unlock(&sbi->dir_inode_lock);
 610 }
 611
 612 void remove_dirty_dir_inode(struct inode *inode)
 613 {
 614         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 615
 616         struct list_head *this, *head;
 617
 618         if (!S_ISDIR(inode->i_mode))
 619                 return;
 620
 621         spin_lock(&sbi->dir_inode_lock);
 622         if (get_dirty_dents(inode)) {
 623                 spin_unlock(&sbi->dir_inode_lock);
 624                 return;
 625         }
 626
 627         head = &sbi->dir_inode_list;
 628         list_for_each(this, head) {
 629                 struct dir_inode_entry *entry;
 630                 entry = list_entry(this, struct dir_inode_entry, list);
 631                 if (entry->inode == inode) {
 632                         list_del(&entry->list);
 633                         kmem_cache_free(inode_entry_slab, entry);
 634                         stat_dec_dirty_dir(sbi);
 635                         break;
 636                 }
 637         }
 638         spin_unlock(&sbi->dir_inode_lock);
 639
 640         /* Only from the recovery routine */
 641         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 642                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 643                 iput(inode);
 644         }
 645 }
 646
 647 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 648 {
 649
 650         struct list_head *this, *head;
 651         struct inode *inode = NULL;
 652
 653         spin_lock(&sbi->dir_inode_lock);
 654
 655         head = &sbi->dir_inode_list;
 656         list_for_each(this, head) {
 657                 struct dir_inode_entry *entry;
 658                 entry = list_entry(this, struct dir_inode_entry, list);
 659                 if (entry->inode->i_ino == ino) {
 660                         inode = entry->inode;
 661                         break;
 662                 }
 663         }
 664         spin_unlock(&sbi->dir_inode_lock);
 665         return inode;
 666 }
 667
 668 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 669 {
 670         struct list_head *head;
 671         struct dir_inode_entry *entry;
 672         struct inode *inode;
 673 retry:
 674         spin_lock(&sbi->dir_inode_lock);
 675
 676         head = &sbi->dir_inode_list;
 677         if (list_empty(head)) {
 678                 spin_unlock(&sbi->dir_inode_lock);
 679                 return;
 680         }
 681         entry = list_entry(head->next, struct dir_inode_entry, list);
 682         inode = igrab(entry->inode);
 683         spin_unlock(&sbi->dir_inode_lock);
 684         if (inode) {
 685                 filemap_fdatawrite(inode->i_mapping);
 686                 iput(inode);
 687         } else {
 688                 /*
 689                  * We should submit bio, since it exists several
 690                  * wribacking dentry pages in the freeing inode.
 691                  */
 692                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 693         }
 694         goto retry;
 695 }
 696
 697 /*
 698  * Freeze all the FS-operations for checkpoint.
 699  */
 700 static void block_operations(struct f2fs_sb_info *sbi)
 701 {
 702         struct writeback_control wbc = {
 703                 .sync_mode = WB_SYNC_ALL,
 704                 .nr_to_write = LONG_MAX,
 705                 .for_reclaim = 0,
 706         };
 707         struct blk_plug plug;
 708
 709         blk_start_plug(&plug);
 710
 711 retry_flush_dents:
 712         f2fs_lock_all(sbi);
 713         /* write all the dirty dentry pages */
 714         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 715                 f2fs_unlock_all(sbi);
 716                 sync_dirty_dir_inodes(sbi);
 717                 goto retry_flush_dents;
 718         }
 719
 720         /*
 721          * POR: we should ensure that there is no dirty node pages
 722          * until finishing nat/sit flush.
 723          */
 724 retry_flush_nodes:
 725         mutex_lock(&sbi->node_write);
 726
 727         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 728                 mutex_unlock(&sbi->node_write);
 729                 sync_node_pages(sbi, 0, &wbc);
 730                 goto retry_flush_nodes;
 731         }
 732         blk_finish_plug(&plug);
 733 }
 734
 735 static void unblock_operations(struct f2fs_sb_info *sbi)
 736 {
 737         mutex_unlock(&sbi->node_write);
 738         f2fs_unlock_all(sbi);
 739 }
 740
 741 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 742 {
 743         DEFINE_WAIT(wait);
 744
 745         for (;;) {
 746                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 747
 748                 if (!get_pages(sbi, F2FS_WRITEBACK))
 749                         break;
 750
 751                 io_schedule();
 752         }
 753         finish_wait(&sbi->cp_wait, &wait);
 754 }
 755
 756 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 757 {
 758         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 759         nid_t last_nid = 0;
 760         block_t start_blk;
 761         struct page *cp_page;
 762         unsigned int data_sum_blocks, orphan_blocks;
 763         __u32 crc32 = 0;
 764         void *kaddr;
 765         int i;
 766
 767         /* Flush all the NAT/SIT pages */
 768         while (get_pages(sbi, F2FS_DIRTY_META))
 769                 sync_meta_pages(sbi, META, LONG_MAX);
 770
 771         next_free_nid(sbi, &last_nid);
 772
 773         /*
 774          * modify checkpoint
 775          * version number is already updated
 776          */
 777         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 778         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 779         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 780         for (i = 0; i < 3; i++) {
 781                 ckpt->cur_node_segno[i] =
 782                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 783                 ckpt->cur_node_blkoff[i] =
 784                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 785                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 786                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 787         }
 788         for (i = 0; i < 3; i++) {
 789                 ckpt->cur_data_segno[i] =
 790                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 791                 ckpt->cur_data_blkoff[i] =
 792                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 793                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 794                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 795         }
 796
 797         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 798         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 799         ckpt->next_free_nid = cpu_to_le32(last_nid);
 800
 801         /* 2 cp  + n data seg summary + orphan inode blocks */
 802         data_sum_blocks = npages_for_summary_flush(sbi);
 803         if (data_sum_blocks < 3)
 804                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 805         else
 806                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 807
 808         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 809                                         / F2FS_ORPHANS_PER_BLOCK;
 810         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 811
 812         if (is_umount) {
 813                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 814                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 815                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 816         } else {
 817                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 818                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 819                         data_sum_blocks + orphan_blocks);
 820         }
 821
 822         if (sbi->n_orphans)
 823                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 824         else
 825                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 826
 827         /* update SIT/NAT bitmap */
 828         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 829         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 830
 831         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 832         *((__le32 *)((unsigned char *)ckpt +
 833                                 le32_to_cpu(ckpt->checksum_offset)))
 834                                 = cpu_to_le32(crc32);
 835
 836         start_blk = __start_cp_addr(sbi);
 837
 838         /* write out checkpoint buffer at block 0 */
 839         cp_page = grab_meta_page(sbi, start_blk++);
 840         kaddr = page_address(cp_page);
 841         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 842         set_page_dirty(cp_page);
 843         f2fs_put_page(cp_page, 1);
 844
 845         if (sbi->n_orphans) {
 846                 write_orphan_inodes(sbi, start_blk);
 847                 start_blk += orphan_blocks;
 848         }
 849
 850         write_data_summaries(sbi, start_blk);
 851         start_blk += data_sum_blocks;
 852         if (is_umount) {
 853                 write_node_summaries(sbi, start_blk);
 854                 start_blk += NR_CURSEG_NODE_TYPE;
 855         }
 856
 857         /* writeout checkpoint block */
 858         cp_page = grab_meta_page(sbi, start_blk);
 859         kaddr = page_address(cp_page);
 860         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 861         set_page_dirty(cp_page);
 862         f2fs_put_page(cp_page, 1);
 863
 864         /* wait for previous submitted node/meta pages writeback */
 865         wait_on_all_pages_writeback(sbi);
 866
 867         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 868         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 869
 870         /* update user_block_counts */
 871         sbi->last_valid_block_count = sbi->total_valid_block_count;
 872         sbi->alloc_valid_block_count = 0;
 873
 874         /* Here, we only have one bio having CP pack */
 875         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 876
 877         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 878                 clear_prefree_segments(sbi);
 879                 F2FS_RESET_SB_DIRT(sbi);
 880         }
 881 }
 882
 883 /*
 884  * We guarantee that this checkpoint procedure should not fail.
 885  */
 886 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 887 {
 888         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 889         unsigned long long ckpt_ver;
 890
 891         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 892
 893         mutex_lock(&sbi->cp_mutex);
 894         block_operations(sbi);
 895
 896         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 897
 898         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 899         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 900         f2fs_submit_merged_bio(sbi, META, WRITE);
 901
 902         /*
 903          * update checkpoint pack index
 904          * Increase the version number so that
 905          * SIT entries and seg summaries are written at correct place
 906          */
 907         ckpt_ver = cur_cp_version(ckpt);
 908         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 909
 910         /* write cached NAT/SIT entries to NAT/SIT area */
 911         flush_nat_entries(sbi);
 912         flush_sit_entries(sbi);
 913
 914         /* unlock all the fs_lock[] in do_checkpoint() */
 915         do_checkpoint(sbi, is_umount);
 916
 917         unblock_operations(sbi);
 918         mutex_unlock(&sbi->cp_mutex);
 919
 920         stat_inc_cp_count(sbi->stat_info);
 921         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 922 }
 923
 924 void init_orphan_info(struct f2fs_sb_info *sbi)
 925 {
 926         spin_lock_init(&sbi->orphan_inode_lock);
 927         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 928         sbi->n_orphans = 0;
 929         /*
 930          * considering 512 blocks in a segment 8 blocks are needed for cp
 931          * and log segment summaries. Remaining blocks are used to keep
 932          * orphan entries with the limitation one reserved segment
 933          * for cp pack we can have max 1020*504 orphan entries
 934          */
 935         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 936                                 * F2FS_ORPHANS_PER_BLOCK;
 937 }
 938
 939 int __init create_checkpoint_caches(void)
 940 {
 941         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 942                         sizeof(struct orphan_inode_entry));
 943         if (!orphan_entry_slab)
 944                 return -ENOMEM;
 945         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 946                         sizeof(struct dir_inode_entry));
 947         if (!inode_entry_slab) {
 948                 kmem_cache_destroy(orphan_entry_slab);
 949                 return -ENOMEM;
 950         }
 951         return 0;
 952 }
 953
 954 void destroy_checkpoint_caches(void)
 955 {
 956         kmem_cache_destroy(orphan_entry_slab);
 957         kmem_cache_destroy(inode_entry_slab);
 958 }