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 *ino_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         f2fs_wait_on_page_writeback(page, META);
  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         return page;
  73 }
  74
  75 static inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  76 {
  77         switch (type) {
  78         case META_NAT:
  79                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  80         case META_SIT:
  81                 return SIT_BLK_CNT(sbi);
  82         case META_SSA:
  83         case META_CP:
  84                 return 0;
  85         default:
  86                 BUG();
  87         }
  88 }
  89
  90 /*
  91  * Readahead CP/NAT/SIT/SSA pages
  92  */
  93 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
  94 {
  95         block_t prev_blk_addr = 0;
  96         struct page *page;
  97         int blkno = start;
  98         int max_blks = get_max_meta_blks(sbi, type);
  99
 100         struct f2fs_io_info fio = {
 101                 .type = META,
 102                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 103         };
 104
 105         for (; nrpages-- > 0; blkno++) {
 106                 block_t blk_addr;
 107
 108                 switch (type) {
 109                 case META_NAT:
 110                         /* get nat block addr */
 111                         if (unlikely(blkno >= max_blks))
 112                                 blkno = 0;
 113                         blk_addr = current_nat_addr(sbi,
 114                                         blkno * NAT_ENTRY_PER_BLOCK);
 115                         break;
 116                 case META_SIT:
 117                         /* get sit block addr */
 118                         if (unlikely(blkno >= max_blks))
 119                                 goto out;
 120                         blk_addr = current_sit_addr(sbi,
 121                                         blkno * SIT_ENTRY_PER_BLOCK);
 122                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 123                                 goto out;
 124                         prev_blk_addr = blk_addr;
 125                         break;
 126                 case META_SSA:
 127                 case META_CP:
 128                         /* get ssa/cp block addr */
 129                         blk_addr = blkno;
 130                         break;
 131                 default:
 132                         BUG();
 133                 }
 134
 135                 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
 136                 if (!page)
 137                         continue;
 138                 if (PageUptodate(page)) {
 139                         f2fs_put_page(page, 1);
 140                         continue;
 141                 }
 142
 143                 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
 144                 f2fs_put_page(page, 0);
 145         }
 146 out:
 147         f2fs_submit_merged_bio(sbi, META, READ);
 148         return blkno - start;
 149 }
 150
 151 static int f2fs_write_meta_page(struct page *page,
 152                                 struct writeback_control *wbc)
 153 {
 154         struct inode *inode = page->mapping->host;
 155         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 156
 157         trace_f2fs_writepage(page, META);
 158
 159         if (unlikely(sbi->por_doing))
 160                 goto redirty_out;
 161         if (wbc->for_reclaim)
 162                 goto redirty_out;
 163
 164         /* Should not write any meta pages, if any IO error was occurred */
 165         if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
 166                 goto no_write;
 167
 168         f2fs_wait_on_page_writeback(page, META);
 169         write_meta_page(sbi, page);
 170 no_write:
 171         dec_page_count(sbi, F2FS_DIRTY_META);
 172         unlock_page(page);
 173         return 0;
 174
 175 redirty_out:
 176         redirty_page_for_writepage(wbc, page);
 177         return AOP_WRITEPAGE_ACTIVATE;
 178 }
 179
 180 static int f2fs_write_meta_pages(struct address_space *mapping,
 181                                 struct writeback_control *wbc)
 182 {
 183         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 184         long diff, written;
 185
 186         trace_f2fs_writepages(mapping->host, wbc, META);
 187
 188         /* collect a number of dirty meta pages and write together */
 189         if (wbc->for_kupdate ||
 190                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 191                 goto skip_write;
 192
 193         /* if mounting is failed, skip writing node pages */
 194         mutex_lock(&sbi->cp_mutex);
 195         diff = nr_pages_to_write(sbi, META, wbc);
 196         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 197         mutex_unlock(&sbi->cp_mutex);
 198         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 199         return 0;
 200
 201 skip_write:
 202         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 203         return 0;
 204 }
 205
 206 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 207                                                 long nr_to_write)
 208 {
 209         struct address_space *mapping = META_MAPPING(sbi);
 210         pgoff_t index = 0, end = LONG_MAX;
 211         struct pagevec pvec;
 212         long nwritten = 0;
 213         struct writeback_control wbc = {
 214                 .for_reclaim = 0,
 215         };
 216
 217         pagevec_init(&pvec, 0);
 218
 219         while (index <= end) {
 220                 int i, nr_pages;
 221                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 222                                 PAGECACHE_TAG_DIRTY,
 223                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 224                 if (unlikely(nr_pages == 0))
 225                         break;
 226
 227                 for (i = 0; i < nr_pages; i++) {
 228                         struct page *page = pvec.pages[i];
 229
 230                         lock_page(page);
 231
 232                         if (unlikely(page->mapping != mapping)) {
 233 continue_unlock:
 234                                 unlock_page(page);
 235                                 continue;
 236                         }
 237                         if (!PageDirty(page)) {
 238                                 /* someone wrote it for us */
 239                                 goto continue_unlock;
 240                         }
 241
 242                         if (!clear_page_dirty_for_io(page))
 243                                 goto continue_unlock;
 244
 245                         if (f2fs_write_meta_page(page, &wbc)) {
 246                                 unlock_page(page);
 247                                 break;
 248                         }
 249                         nwritten++;
 250                         if (unlikely(nwritten >= nr_to_write))
 251                                 break;
 252                 }
 253                 pagevec_release(&pvec);
 254                 cond_resched();
 255         }
 256
 257         if (nwritten)
 258                 f2fs_submit_merged_bio(sbi, type, WRITE);
 259
 260         return nwritten;
 261 }
 262
 263 static int f2fs_set_meta_page_dirty(struct page *page)
 264 {
 265         struct address_space *mapping = page->mapping;
 266         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 267
 268         trace_f2fs_set_page_dirty(page, META);
 269
 270         SetPageUptodate(page);
 271         if (!PageDirty(page)) {
 272                 __set_page_dirty_nobuffers(page);
 273                 inc_page_count(sbi, F2FS_DIRTY_META);
 274                 return 1;
 275         }
 276         return 0;
 277 }
 278
 279 const struct address_space_operations f2fs_meta_aops = {
 280         .writepage      = f2fs_write_meta_page,
 281         .writepages     = f2fs_write_meta_pages,
 282         .set_page_dirty = f2fs_set_meta_page_dirty,
 283 };
 284
 285 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 286 {
 287         struct ino_entry *e;
 288 retry:
 289         spin_lock(&sbi->ino_lock[type]);
 290
 291         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 292         if (!e) {
 293                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 294                 if (!e) {
 295                         spin_unlock(&sbi->ino_lock[type]);
 296                         goto retry;
 297                 }
 298                 if (radix_tree_insert(&sbi->ino_root[type], ino, e)) {
 299                         spin_unlock(&sbi->ino_lock[type]);
 300                         kmem_cache_free(ino_entry_slab, e);
 301                         goto retry;
 302                 }
 303                 memset(e, 0, sizeof(struct ino_entry));
 304                 e->ino = ino;
 305
 306                 list_add_tail(&e->list, &sbi->ino_list[type]);
 307         }
 308         spin_unlock(&sbi->ino_lock[type]);
 309 }
 310
 311 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 312 {
 313         struct ino_entry *e;
 314
 315         spin_lock(&sbi->ino_lock[type]);
 316         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 317         if (e) {
 318                 list_del(&e->list);
 319                 radix_tree_delete(&sbi->ino_root[type], ino);
 320                 if (type == ORPHAN_INO)
 321                         sbi->n_orphans--;
 322                 spin_unlock(&sbi->ino_lock[type]);
 323                 kmem_cache_free(ino_entry_slab, e);
 324                 return;
 325         }
 326         spin_unlock(&sbi->ino_lock[type]);
 327 }
 328
 329 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 330 {
 331         int err = 0;
 332
 333         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 334         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 335                 err = -ENOSPC;
 336         else
 337                 sbi->n_orphans++;
 338         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 339
 340         return err;
 341 }
 342
 343 void release_orphan_inode(struct f2fs_sb_info *sbi)
 344 {
 345         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 346         f2fs_bug_on(sbi->n_orphans == 0);
 347         sbi->n_orphans--;
 348         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 349 }
 350
 351 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 352 {
 353         /* add new orphan ino entry into list */
 354         __add_ino_entry(sbi, ino, ORPHAN_INO);
 355 }
 356
 357 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 358 {
 359         /* remove orphan entry from orphan list */
 360         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 361 }
 362
 363 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 364 {
 365         struct inode *inode = f2fs_iget(sbi->sb, ino);
 366         f2fs_bug_on(IS_ERR(inode));
 367         clear_nlink(inode);
 368
 369         /* truncate all the data during iput */
 370         iput(inode);
 371 }
 372
 373 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 374 {
 375         block_t start_blk, orphan_blkaddr, i, j;
 376
 377         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 378                 return;
 379
 380         sbi->por_doing = true;
 381
 382         start_blk = __start_cp_addr(sbi) + 1 +
 383                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 384         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 385
 386         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 387
 388         for (i = 0; i < orphan_blkaddr; i++) {
 389                 struct page *page = get_meta_page(sbi, start_blk + i);
 390                 struct f2fs_orphan_block *orphan_blk;
 391
 392                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 393                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 394                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 395                         recover_orphan_inode(sbi, ino);
 396                 }
 397                 f2fs_put_page(page, 1);
 398         }
 399         /* clear Orphan Flag */
 400         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 401         sbi->por_doing = false;
 402         return;
 403 }
 404
 405 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 406 {
 407         struct list_head *head;
 408         struct f2fs_orphan_block *orphan_blk = NULL;
 409         unsigned int nentries = 0;
 410         unsigned short index;
 411         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 412                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 413         struct page *page = NULL;
 414         struct ino_entry *orphan = NULL;
 415
 416         for (index = 0; index < orphan_blocks; index++)
 417                 grab_meta_page(sbi, start_blk + index);
 418
 419         index = 1;
 420         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 421         head = &sbi->ino_list[ORPHAN_INO];
 422
 423         /* loop for each orphan inode entry and write them in Jornal block */
 424         list_for_each_entry(orphan, head, list) {
 425                 if (!page) {
 426                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 427                         f2fs_bug_on(!page);
 428                         orphan_blk =
 429                                 (struct f2fs_orphan_block *)page_address(page);
 430                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 431                         f2fs_put_page(page, 0);
 432                 }
 433
 434                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 435
 436                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 437                         /*
 438                          * an orphan block is full of 1020 entries,
 439                          * then we need to flush current orphan blocks
 440                          * and bring another one in memory
 441                          */
 442                         orphan_blk->blk_addr = cpu_to_le16(index);
 443                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 444                         orphan_blk->entry_count = cpu_to_le32(nentries);
 445                         set_page_dirty(page);
 446                         f2fs_put_page(page, 1);
 447                         index++;
 448                         nentries = 0;
 449                         page = NULL;
 450                 }
 451         }
 452
 453         if (page) {
 454                 orphan_blk->blk_addr = cpu_to_le16(index);
 455                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 456                 orphan_blk->entry_count = cpu_to_le32(nentries);
 457                 set_page_dirty(page);
 458                 f2fs_put_page(page, 1);
 459         }
 460
 461         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 462 }
 463
 464 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 465                                 block_t cp_addr, unsigned long long *version)
 466 {
 467         struct page *cp_page_1, *cp_page_2 = NULL;
 468         unsigned long blk_size = sbi->blocksize;
 469         struct f2fs_checkpoint *cp_block;
 470         unsigned long long cur_version = 0, pre_version = 0;
 471         size_t crc_offset;
 472         __u32 crc = 0;
 473
 474         /* Read the 1st cp block in this CP pack */
 475         cp_page_1 = get_meta_page(sbi, cp_addr);
 476
 477         /* get the version number */
 478         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 479         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 480         if (crc_offset >= blk_size)
 481                 goto invalid_cp1;
 482
 483         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 484         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 485                 goto invalid_cp1;
 486
 487         pre_version = cur_cp_version(cp_block);
 488
 489         /* Read the 2nd cp block in this CP pack */
 490         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 491         cp_page_2 = get_meta_page(sbi, cp_addr);
 492
 493         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 494         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 495         if (crc_offset >= blk_size)
 496                 goto invalid_cp2;
 497
 498         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 499         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 500                 goto invalid_cp2;
 501
 502         cur_version = cur_cp_version(cp_block);
 503
 504         if (cur_version == pre_version) {
 505                 *version = cur_version;
 506                 f2fs_put_page(cp_page_2, 1);
 507                 return cp_page_1;
 508         }
 509 invalid_cp2:
 510         f2fs_put_page(cp_page_2, 1);
 511 invalid_cp1:
 512         f2fs_put_page(cp_page_1, 1);
 513         return NULL;
 514 }
 515
 516 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 517 {
 518         struct f2fs_checkpoint *cp_block;
 519         struct f2fs_super_block *fsb = sbi->raw_super;
 520         struct page *cp1, *cp2, *cur_page;
 521         unsigned long blk_size = sbi->blocksize;
 522         unsigned long long cp1_version = 0, cp2_version = 0;
 523         unsigned long long cp_start_blk_no;
 524         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 525         block_t cp_blk_no;
 526         int i;
 527
 528         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 529         if (!sbi->ckpt)
 530                 return -ENOMEM;
 531         /*
 532          * Finding out valid cp block involves read both
 533          * sets( cp pack1 and cp pack 2)
 534          */
 535         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 536         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 537
 538         /* The second checkpoint pack should start at the next segment */
 539         cp_start_blk_no += ((unsigned long long)1) <<
 540                                 le32_to_cpu(fsb->log_blocks_per_seg);
 541         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 542
 543         if (cp1 && cp2) {
 544                 if (ver_after(cp2_version, cp1_version))
 545                         cur_page = cp2;
 546                 else
 547                         cur_page = cp1;
 548         } else if (cp1) {
 549                 cur_page = cp1;
 550         } else if (cp2) {
 551                 cur_page = cp2;
 552         } else {
 553                 goto fail_no_cp;
 554         }
 555
 556         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 557         memcpy(sbi->ckpt, cp_block, blk_size);
 558
 559         if (cp_blks <= 1)
 560                 goto done;
 561
 562         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 563         if (cur_page == cp2)
 564                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 565
 566         for (i = 1; i < cp_blks; i++) {
 567                 void *sit_bitmap_ptr;
 568                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 569
 570                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 571                 sit_bitmap_ptr = page_address(cur_page);
 572                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 573                 f2fs_put_page(cur_page, 1);
 574         }
 575 done:
 576         f2fs_put_page(cp1, 1);
 577         f2fs_put_page(cp2, 1);
 578         return 0;
 579
 580 fail_no_cp:
 581         kfree(sbi->ckpt);
 582         return -EINVAL;
 583 }
 584
 585 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 586 {
 587         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 588
 589         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 590                 return -EEXIST;
 591
 592         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 593         F2FS_I(inode)->dirty_dir = new;
 594         list_add_tail(&new->list, &sbi->dir_inode_list);
 595         stat_inc_dirty_dir(sbi);
 596         return 0;
 597 }
 598
 599 void set_dirty_dir_page(struct inode *inode, struct page *page)
 600 {
 601         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 602         struct dir_inode_entry *new;
 603         int ret = 0;
 604
 605         if (!S_ISDIR(inode->i_mode))
 606                 return;
 607
 608         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 609         new->inode = inode;
 610         INIT_LIST_HEAD(&new->list);
 611
 612         spin_lock(&sbi->dir_inode_lock);
 613         ret = __add_dirty_inode(inode, new);
 614         inode_inc_dirty_dents(inode);
 615         SetPagePrivate(page);
 616         spin_unlock(&sbi->dir_inode_lock);
 617
 618         if (ret)
 619                 kmem_cache_free(inode_entry_slab, new);
 620 }
 621
 622 void add_dirty_dir_inode(struct inode *inode)
 623 {
 624         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 625         struct dir_inode_entry *new =
 626                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 627         int ret = 0;
 628
 629         new->inode = inode;
 630         INIT_LIST_HEAD(&new->list);
 631
 632         spin_lock(&sbi->dir_inode_lock);
 633         ret = __add_dirty_inode(inode, new);
 634         spin_unlock(&sbi->dir_inode_lock);
 635
 636         if (ret)
 637                 kmem_cache_free(inode_entry_slab, new);
 638 }
 639
 640 void remove_dirty_dir_inode(struct inode *inode)
 641 {
 642         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 643         struct dir_inode_entry *entry;
 644
 645         if (!S_ISDIR(inode->i_mode))
 646                 return;
 647
 648         spin_lock(&sbi->dir_inode_lock);
 649         if (get_dirty_dents(inode) ||
 650                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 651                 spin_unlock(&sbi->dir_inode_lock);
 652                 return;
 653         }
 654
 655         entry = F2FS_I(inode)->dirty_dir;
 656         list_del(&entry->list);
 657         F2FS_I(inode)->dirty_dir = NULL;
 658         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 659         stat_dec_dirty_dir(sbi);
 660         spin_unlock(&sbi->dir_inode_lock);
 661         kmem_cache_free(inode_entry_slab, entry);
 662
 663         /* Only from the recovery routine */
 664         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 665                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 666                 iput(inode);
 667         }
 668 }
 669
 670 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 671 {
 672         struct list_head *head;
 673         struct dir_inode_entry *entry;
 674         struct inode *inode;
 675 retry:
 676         spin_lock(&sbi->dir_inode_lock);
 677
 678         head = &sbi->dir_inode_list;
 679         if (list_empty(head)) {
 680                 spin_unlock(&sbi->dir_inode_lock);
 681                 return;
 682         }
 683         entry = list_entry(head->next, struct dir_inode_entry, list);
 684         inode = igrab(entry->inode);
 685         spin_unlock(&sbi->dir_inode_lock);
 686         if (inode) {
 687                 filemap_fdatawrite(inode->i_mapping);
 688                 iput(inode);
 689         } else {
 690                 /*
 691                  * We should submit bio, since it exists several
 692                  * wribacking dentry pages in the freeing inode.
 693                  */
 694                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 695         }
 696         goto retry;
 697 }
 698
 699 /*
 700  * Freeze all the FS-operations for checkpoint.
 701  */
 702 static void block_operations(struct f2fs_sb_info *sbi)
 703 {
 704         struct writeback_control wbc = {
 705                 .sync_mode = WB_SYNC_ALL,
 706                 .nr_to_write = LONG_MAX,
 707                 .for_reclaim = 0,
 708         };
 709         struct blk_plug plug;
 710
 711         blk_start_plug(&plug);
 712
 713 retry_flush_dents:
 714         f2fs_lock_all(sbi);
 715         /* write all the dirty dentry pages */
 716         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 717                 f2fs_unlock_all(sbi);
 718                 sync_dirty_dir_inodes(sbi);
 719                 goto retry_flush_dents;
 720         }
 721
 722         /*
 723          * POR: we should ensure that there is no dirty node pages
 724          * until finishing nat/sit flush.
 725          */
 726 retry_flush_nodes:
 727         mutex_lock(&sbi->node_write);
 728
 729         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 730                 mutex_unlock(&sbi->node_write);
 731                 sync_node_pages(sbi, 0, &wbc);
 732                 goto retry_flush_nodes;
 733         }
 734         blk_finish_plug(&plug);
 735 }
 736
 737 static void unblock_operations(struct f2fs_sb_info *sbi)
 738 {
 739         mutex_unlock(&sbi->node_write);
 740         f2fs_unlock_all(sbi);
 741 }
 742
 743 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 744 {
 745         DEFINE_WAIT(wait);
 746
 747         for (;;) {
 748                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 749
 750                 if (!get_pages(sbi, F2FS_WRITEBACK))
 751                         break;
 752
 753                 io_schedule();
 754         }
 755         finish_wait(&sbi->cp_wait, &wait);
 756 }
 757
 758 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 759 {
 760         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 761         nid_t last_nid = 0;
 762         block_t start_blk;
 763         struct page *cp_page;
 764         unsigned int data_sum_blocks, orphan_blocks;
 765         __u32 crc32 = 0;
 766         void *kaddr;
 767         int i;
 768         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 769
 770         /*
 771          * This avoids to conduct wrong roll-forward operations and uses
 772          * metapages, so should be called prior to sync_meta_pages below.
 773          */
 774         discard_next_dnode(sbi);
 775
 776         /* Flush all the NAT/SIT pages */
 777         while (get_pages(sbi, F2FS_DIRTY_META))
 778                 sync_meta_pages(sbi, META, LONG_MAX);
 779
 780         next_free_nid(sbi, &last_nid);
 781
 782         /*
 783          * modify checkpoint
 784          * version number is already updated
 785          */
 786         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 787         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 788         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 789         for (i = 0; i < 3; i++) {
 790                 ckpt->cur_node_segno[i] =
 791                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 792                 ckpt->cur_node_blkoff[i] =
 793                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 794                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 795                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 796         }
 797         for (i = 0; i < 3; i++) {
 798                 ckpt->cur_data_segno[i] =
 799                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 800                 ckpt->cur_data_blkoff[i] =
 801                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 802                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 803                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 804         }
 805
 806         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 807         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 808         ckpt->next_free_nid = cpu_to_le32(last_nid);
 809
 810         /* 2 cp  + n data seg summary + orphan inode blocks */
 811         data_sum_blocks = npages_for_summary_flush(sbi);
 812         if (data_sum_blocks < 3)
 813                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 814         else
 815                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 816
 817         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 818                                         / F2FS_ORPHANS_PER_BLOCK;
 819         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 820                         orphan_blocks);
 821
 822         if (is_umount) {
 823                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 824                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 825                                 cp_payload_blks + data_sum_blocks +
 826                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 827         } else {
 828                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 829                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 830                                 cp_payload_blks + data_sum_blocks +
 831                                 orphan_blocks);
 832         }
 833
 834         if (sbi->n_orphans)
 835                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 836         else
 837                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 838
 839         /* update SIT/NAT bitmap */
 840         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 841         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 842
 843         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 844         *((__le32 *)((unsigned char *)ckpt +
 845                                 le32_to_cpu(ckpt->checksum_offset)))
 846                                 = cpu_to_le32(crc32);
 847
 848         start_blk = __start_cp_addr(sbi);
 849
 850         /* write out checkpoint buffer at block 0 */
 851         cp_page = grab_meta_page(sbi, start_blk++);
 852         kaddr = page_address(cp_page);
 853         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 854         set_page_dirty(cp_page);
 855         f2fs_put_page(cp_page, 1);
 856
 857         for (i = 1; i < 1 + cp_payload_blks; i++) {
 858                 cp_page = grab_meta_page(sbi, start_blk++);
 859                 kaddr = page_address(cp_page);
 860                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 861                                 (1 << sbi->log_blocksize));
 862                 set_page_dirty(cp_page);
 863                 f2fs_put_page(cp_page, 1);
 864         }
 865
 866         if (sbi->n_orphans) {
 867                 write_orphan_inodes(sbi, start_blk);
 868                 start_blk += orphan_blocks;
 869         }
 870
 871         write_data_summaries(sbi, start_blk);
 872         start_blk += data_sum_blocks;
 873         if (is_umount) {
 874                 write_node_summaries(sbi, start_blk);
 875                 start_blk += NR_CURSEG_NODE_TYPE;
 876         }
 877
 878         /* writeout checkpoint block */
 879         cp_page = grab_meta_page(sbi, start_blk);
 880         kaddr = page_address(cp_page);
 881         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 882         set_page_dirty(cp_page);
 883         f2fs_put_page(cp_page, 1);
 884
 885         /* wait for previous submitted node/meta pages writeback */
 886         wait_on_all_pages_writeback(sbi);
 887
 888         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 889         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 890
 891         /* update user_block_counts */
 892         sbi->last_valid_block_count = sbi->total_valid_block_count;
 893         sbi->alloc_valid_block_count = 0;
 894
 895         /* Here, we only have one bio having CP pack */
 896         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 897
 898         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 899                 clear_prefree_segments(sbi);
 900                 F2FS_RESET_SB_DIRT(sbi);
 901         }
 902 }
 903
 904 /*
 905  * We guarantee that this checkpoint procedure should not fail.
 906  */
 907 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 908 {
 909         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 910         unsigned long long ckpt_ver;
 911
 912         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 913
 914         mutex_lock(&sbi->cp_mutex);
 915         block_operations(sbi);
 916
 917         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 918
 919         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 920         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 921         f2fs_submit_merged_bio(sbi, META, WRITE);
 922
 923         /*
 924          * update checkpoint pack index
 925          * Increase the version number so that
 926          * SIT entries and seg summaries are written at correct place
 927          */
 928         ckpt_ver = cur_cp_version(ckpt);
 929         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 930
 931         /* write cached NAT/SIT entries to NAT/SIT area */
 932         flush_nat_entries(sbi);
 933         flush_sit_entries(sbi);
 934
 935         /* unlock all the fs_lock[] in do_checkpoint() */
 936         do_checkpoint(sbi, is_umount);
 937
 938         unblock_operations(sbi);
 939         mutex_unlock(&sbi->cp_mutex);
 940
 941         stat_inc_cp_count(sbi->stat_info);
 942         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 943 }
 944
 945 void init_ino_entry_info(struct f2fs_sb_info *sbi)
 946 {
 947         int i;
 948
 949         for (i = 0; i < MAX_INO_ENTRY; i++) {
 950                 INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
 951                 spin_lock_init(&sbi->ino_lock[i]);
 952                 INIT_LIST_HEAD(&sbi->ino_list[i]);
 953         }
 954
 955         /*
 956          * considering 512 blocks in a segment 8 blocks are needed for cp
 957          * and log segment summaries. Remaining blocks are used to keep
 958          * orphan entries with the limitation one reserved segment
 959          * for cp pack we can have max 1020*504 orphan entries
 960          */
 961         sbi->n_orphans = 0;
 962         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 963                                 * F2FS_ORPHANS_PER_BLOCK;
 964 }
 965
 966 int __init create_checkpoint_caches(void)
 967 {
 968         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
 969                         sizeof(struct ino_entry));
 970         if (!ino_entry_slab)
 971                 return -ENOMEM;
 972         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 973                         sizeof(struct dir_inode_entry));
 974         if (!inode_entry_slab) {
 975                 kmem_cache_destroy(ino_entry_slab);
 976                 return -ENOMEM;
 977         }
 978         return 0;
 979 }
 980
 981 void destroy_checkpoint_caches(void)
 982 {
 983         kmem_cache_destroy(ino_entry_slab);
 984         kmem_cache_destroy(inode_entry_slab);
 985 }