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         mark_page_accessed(page);
  73         return page;
  74 }
  75
  76 static inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  77 {
  78         switch (type) {
  79         case META_NAT:
  80                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  81         case META_SIT:
  82                 return SIT_BLK_CNT(sbi);
  83         case META_SSA:
  84         case META_CP:
  85                 return 0;
  86         default:
  87                 BUG();
  88         }
  89 }
  90
  91 /*
  92  * Readahead CP/NAT/SIT/SSA pages
  93  */
  94 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
  95 {
  96         block_t prev_blk_addr = 0;
  97         struct page *page;
  98         int blkno = start;
  99         int max_blks = get_max_meta_blks(sbi, type);
 100
 101         struct f2fs_io_info fio = {
 102                 .type = META,
 103                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 104         };
 105
 106         for (; nrpages-- > 0; blkno++) {
 107                 block_t blk_addr;
 108
 109                 switch (type) {
 110                 case META_NAT:
 111                         /* get nat block addr */
 112                         if (unlikely(blkno >= max_blks))
 113                                 blkno = 0;
 114                         blk_addr = current_nat_addr(sbi,
 115                                         blkno * NAT_ENTRY_PER_BLOCK);
 116                         break;
 117                 case META_SIT:
 118                         /* get sit block addr */
 119                         if (unlikely(blkno >= max_blks))
 120                                 goto out;
 121                         blk_addr = current_sit_addr(sbi,
 122                                         blkno * SIT_ENTRY_PER_BLOCK);
 123                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 124                                 goto out;
 125                         prev_blk_addr = blk_addr;
 126                         break;
 127                 case META_SSA:
 128                 case META_CP:
 129                         /* get ssa/cp block addr */
 130                         blk_addr = blkno;
 131                         break;
 132                 default:
 133                         BUG();
 134                 }
 135
 136                 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
 137                 if (!page)
 138                         continue;
 139                 if (PageUptodate(page)) {
 140                         mark_page_accessed(page);
 141                         f2fs_put_page(page, 1);
 142                         continue;
 143                 }
 144
 145                 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
 146                 mark_page_accessed(page);
 147                 f2fs_put_page(page, 0);
 148         }
 149 out:
 150         f2fs_submit_merged_bio(sbi, META, READ);
 151         return blkno - start;
 152 }
 153
 154 static int f2fs_write_meta_page(struct page *page,
 155                                 struct writeback_control *wbc)
 156 {
 157         struct inode *inode = page->mapping->host;
 158         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 159
 160         trace_f2fs_writepage(page, META);
 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         f2fs_wait_on_page_writeback(page, META);
 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         redirty_page_for_writepage(wbc, page);
 180         return AOP_WRITEPAGE_ACTIVATE;
 181 }
 182
 183 static int f2fs_write_meta_pages(struct address_space *mapping,
 184                                 struct writeback_control *wbc)
 185 {
 186         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 187         long diff, written;
 188
 189         trace_f2fs_writepages(mapping->host, wbc, META);
 190
 191         /* collect a number of dirty meta pages and write together */
 192         if (wbc->for_kupdate ||
 193                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 194                 goto skip_write;
 195
 196         /* if mounting is failed, skip writing node pages */
 197         mutex_lock(&sbi->cp_mutex);
 198         diff = nr_pages_to_write(sbi, META, wbc);
 199         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 200         mutex_unlock(&sbi->cp_mutex);
 201         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 202         return 0;
 203
 204 skip_write:
 205         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 206         return 0;
 207 }
 208
 209 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 210                                                 long nr_to_write)
 211 {
 212         struct address_space *mapping = META_MAPPING(sbi);
 213         pgoff_t index = 0, end = LONG_MAX;
 214         struct pagevec pvec;
 215         long nwritten = 0;
 216         struct writeback_control wbc = {
 217                 .for_reclaim = 0,
 218         };
 219
 220         pagevec_init(&pvec, 0);
 221
 222         while (index <= end) {
 223                 int i, nr_pages;
 224                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 225                                 PAGECACHE_TAG_DIRTY,
 226                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 227                 if (unlikely(nr_pages == 0))
 228                         break;
 229
 230                 for (i = 0; i < nr_pages; i++) {
 231                         struct page *page = pvec.pages[i];
 232
 233                         lock_page(page);
 234
 235                         if (unlikely(page->mapping != mapping)) {
 236 continue_unlock:
 237                                 unlock_page(page);
 238                                 continue;
 239                         }
 240                         if (!PageDirty(page)) {
 241                                 /* someone wrote it for us */
 242                                 goto continue_unlock;
 243                         }
 244
 245                         if (!clear_page_dirty_for_io(page))
 246                                 goto continue_unlock;
 247
 248                         if (f2fs_write_meta_page(page, &wbc)) {
 249                                 unlock_page(page);
 250                                 break;
 251                         }
 252                         nwritten++;
 253                         if (unlikely(nwritten >= nr_to_write))
 254                                 break;
 255                 }
 256                 pagevec_release(&pvec);
 257                 cond_resched();
 258         }
 259
 260         if (nwritten)
 261                 f2fs_submit_merged_bio(sbi, type, WRITE);
 262
 263         return nwritten;
 264 }
 265
 266 static int f2fs_set_meta_page_dirty(struct page *page)
 267 {
 268         struct address_space *mapping = page->mapping;
 269         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 270
 271         trace_f2fs_set_page_dirty(page, META);
 272
 273         SetPageUptodate(page);
 274         if (!PageDirty(page)) {
 275                 __set_page_dirty_nobuffers(page);
 276                 inc_page_count(sbi, F2FS_DIRTY_META);
 277                 return 1;
 278         }
 279         return 0;
 280 }
 281
 282 const struct address_space_operations f2fs_meta_aops = {
 283         .writepage      = f2fs_write_meta_page,
 284         .writepages     = f2fs_write_meta_pages,
 285         .set_page_dirty = f2fs_set_meta_page_dirty,
 286 };
 287
 288 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 289 {
 290         struct ino_entry *new, *e;
 291
 292         new = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 293         new->ino = ino;
 294
 295         spin_lock(&sbi->ino_lock[type]);
 296         list_for_each_entry(e, &sbi->ino_list[type], list) {
 297                 if (e->ino == ino) {
 298                         spin_unlock(&sbi->ino_lock[type]);
 299                         kmem_cache_free(ino_entry_slab, new);
 300                         return;
 301                 }
 302                 if (e->ino > ino)
 303                         break;
 304         }
 305
 306         /* add new entry into list which is sorted by inode number */
 307         list_add_tail(&new->list, &e->list);
 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         list_for_each_entry(e, &sbi->ino_list[type], list) {
 317                 if (e->ino == ino) {
 318                         list_del(&e->list);
 319                         sbi->n_orphans--;
 320                         spin_unlock(&sbi->ino_lock[type]);
 321                         kmem_cache_free(ino_entry_slab, e);
 322                         return;
 323                 }
 324         }
 325         spin_unlock(&sbi->ino_lock[type]);
 326 }
 327
 328 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 329 {
 330         int err = 0;
 331
 332         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 333         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 334                 err = -ENOSPC;
 335         else
 336                 sbi->n_orphans++;
 337         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 338
 339         return err;
 340 }
 341
 342 void release_orphan_inode(struct f2fs_sb_info *sbi)
 343 {
 344         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 345         if (sbi->n_orphans == 0) {
 346                 f2fs_msg(sbi->sb, KERN_ERR, "releasing "
 347                         "unacquired orphan inode");
 348                 f2fs_handle_error(sbi);
 349         } else
 350                 sbi->n_orphans--;
 351         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 352 }
 353
 354 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 355 {
 356         /* add new orphan entry into list which is sorted by inode number */
 357         __add_ino_entry(sbi, ino, ORPHAN_INO);
 358 }
 359
 360 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 361 {
 362         /* remove orphan entry from orphan list */
 363         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 364 }
 365
 366 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 367 {
 368         struct inode *inode = f2fs_iget(sbi->sb, ino);
 369         if (IS_ERR(inode)) {
 370                 f2fs_msg(sbi->sb, KERN_ERR, "unable to recover orphan inode %d",
 371                                 ino);
 372                 f2fs_handle_error(sbi);
 373                 return;
 374         }
 375         clear_nlink(inode);
 376
 377         /* truncate all the data during iput */
 378         iput(inode);
 379 }
 380
 381 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 382 {
 383         block_t start_blk, orphan_blkaddr, i, j;
 384
 385         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 386                 return;
 387
 388         sbi->por_doing = true;
 389
 390         start_blk = __start_cp_addr(sbi) + 1 +
 391                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 392         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 393
 394         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 395
 396         for (i = 0; i < orphan_blkaddr; i++) {
 397                 struct page *page = get_meta_page(sbi, start_blk + i);
 398                 struct f2fs_orphan_block *orphan_blk;
 399
 400                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 401                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 402                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 403                         recover_orphan_inode(sbi, ino);
 404                 }
 405                 f2fs_put_page(page, 1);
 406         }
 407         /* clear Orphan Flag */
 408         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 409         sbi->por_doing = false;
 410         return;
 411 }
 412
 413 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 414 {
 415         struct list_head *head;
 416         struct f2fs_orphan_block *orphan_blk = NULL;
 417         unsigned int nentries = 0;
 418         unsigned short index;
 419         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 420                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 421         struct page *page = NULL;
 422         struct ino_entry *orphan = NULL;
 423
 424         for (index = 0; index < orphan_blocks; index++)
 425                 grab_meta_page(sbi, start_blk + index);
 426
 427         index = 1;
 428         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 429         head = &sbi->ino_list[ORPHAN_INO];
 430
 431         /* loop for each orphan inode entry and write them in Jornal block */
 432         list_for_each_entry(orphan, head, list) {
 433                 if (!page) {
 434                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 435                         f2fs_bug_on(!page);
 436                         orphan_blk =
 437                                 (struct f2fs_orphan_block *)page_address(page);
 438                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 439                         f2fs_put_page(page, 0);
 440                 }
 441
 442                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 443
 444                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 445                         /*
 446                          * an orphan block is full of 1020 entries,
 447                          * then we need to flush current orphan blocks
 448                          * and bring another one in memory
 449                          */
 450                         orphan_blk->blk_addr = cpu_to_le16(index);
 451                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 452                         orphan_blk->entry_count = cpu_to_le32(nentries);
 453                         set_page_dirty(page);
 454                         f2fs_put_page(page, 1);
 455                         index++;
 456                         nentries = 0;
 457                         page = NULL;
 458                 }
 459         }
 460
 461         if (page) {
 462                 orphan_blk->blk_addr = cpu_to_le16(index);
 463                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 464                 orphan_blk->entry_count = cpu_to_le32(nentries);
 465                 set_page_dirty(page);
 466                 f2fs_put_page(page, 1);
 467         }
 468
 469         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 470 }
 471
 472 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 473                                 block_t cp_addr, unsigned long long *version)
 474 {
 475         struct page *cp_page_1, *cp_page_2 = NULL;
 476         unsigned long blk_size = sbi->blocksize;
 477         struct f2fs_checkpoint *cp_block;
 478         unsigned long long cur_version = 0, pre_version = 0;
 479         size_t crc_offset;
 480         __u32 crc = 0;
 481
 482         /* Read the 1st cp block in this CP pack */
 483         cp_page_1 = get_meta_page(sbi, cp_addr);
 484
 485         /* get the version number */
 486         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 487         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 488         if (crc_offset >= blk_size)
 489                 goto invalid_cp1;
 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_cp1;
 494
 495         pre_version = cur_cp_version(cp_block);
 496
 497         /* Read the 2nd cp block in this CP pack */
 498         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 499         cp_page_2 = get_meta_page(sbi, cp_addr);
 500
 501         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 502         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 503         if (crc_offset >= blk_size)
 504                 goto invalid_cp2;
 505
 506         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 507         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 508                 goto invalid_cp2;
 509
 510         cur_version = cur_cp_version(cp_block);
 511
 512         if (cur_version == pre_version) {
 513                 *version = cur_version;
 514                 f2fs_put_page(cp_page_2, 1);
 515                 return cp_page_1;
 516         }
 517 invalid_cp2:
 518         f2fs_put_page(cp_page_2, 1);
 519 invalid_cp1:
 520         f2fs_put_page(cp_page_1, 1);
 521         return NULL;
 522 }
 523
 524 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 525 {
 526         struct f2fs_checkpoint *cp_block;
 527         struct f2fs_super_block *fsb = sbi->raw_super;
 528         struct page *cp1, *cp2, *cur_page;
 529         unsigned long blk_size = sbi->blocksize;
 530         unsigned long long cp1_version = 0, cp2_version = 0;
 531         unsigned long long cp_start_blk_no;
 532         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 533         block_t cp_blk_no;
 534         int i;
 535
 536         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 537         if (!sbi->ckpt)
 538                 return -ENOMEM;
 539         /*
 540          * Finding out valid cp block involves read both
 541          * sets( cp pack1 and cp pack 2)
 542          */
 543         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 544         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 545
 546         /* The second checkpoint pack should start at the next segment */
 547         cp_start_blk_no += ((unsigned long long)1) <<
 548                                 le32_to_cpu(fsb->log_blocks_per_seg);
 549         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 550
 551         if (cp1 && cp2) {
 552                 if (ver_after(cp2_version, cp1_version))
 553                         cur_page = cp2;
 554                 else
 555                         cur_page = cp1;
 556         } else if (cp1) {
 557                 cur_page = cp1;
 558         } else if (cp2) {
 559                 cur_page = cp2;
 560         } else {
 561                 goto fail_no_cp;
 562         }
 563
 564         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 565         memcpy(sbi->ckpt, cp_block, blk_size);
 566
 567         if (cp_blks <= 1)
 568                 goto done;
 569
 570         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 571         if (cur_page == cp2)
 572                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 573
 574         for (i = 1; i < cp_blks; i++) {
 575                 void *sit_bitmap_ptr;
 576                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 577
 578                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 579                 sit_bitmap_ptr = page_address(cur_page);
 580                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 581                 f2fs_put_page(cur_page, 1);
 582         }
 583 done:
 584         f2fs_put_page(cp1, 1);
 585         f2fs_put_page(cp2, 1);
 586         return 0;
 587
 588 fail_no_cp:
 589         kfree(sbi->ckpt);
 590         return -EINVAL;
 591 }
 592
 593 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 594 {
 595         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 596
 597         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 598                 return -EEXIST;
 599
 600         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 601         F2FS_I(inode)->dirty_dir = new;
 602         list_add_tail(&new->list, &sbi->dir_inode_list);
 603         stat_inc_dirty_dir(sbi);
 604         return 0;
 605 }
 606
 607 void set_dirty_dir_page(struct inode *inode, struct page *page)
 608 {
 609         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 610         struct dir_inode_entry *new;
 611         int ret = 0;
 612
 613         if (!S_ISDIR(inode->i_mode))
 614                 return;
 615
 616         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 617         new->inode = inode;
 618         INIT_LIST_HEAD(&new->list);
 619
 620         spin_lock(&sbi->dir_inode_lock);
 621         ret = __add_dirty_inode(inode, new);
 622         inode_inc_dirty_dents(inode);
 623         SetPagePrivate(page);
 624         spin_unlock(&sbi->dir_inode_lock);
 625
 626         if (ret)
 627                 kmem_cache_free(inode_entry_slab, new);
 628 }
 629
 630 void add_dirty_dir_inode(struct inode *inode)
 631 {
 632         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 633         struct dir_inode_entry *new =
 634                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 635         int ret = 0;
 636
 637         new->inode = inode;
 638         INIT_LIST_HEAD(&new->list);
 639
 640         spin_lock(&sbi->dir_inode_lock);
 641         ret = __add_dirty_inode(inode, new);
 642         spin_unlock(&sbi->dir_inode_lock);
 643
 644         if (ret)
 645                 kmem_cache_free(inode_entry_slab, new);
 646 }
 647
 648 void remove_dirty_dir_inode(struct inode *inode)
 649 {
 650         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 651         struct dir_inode_entry *entry;
 652
 653         if (!S_ISDIR(inode->i_mode))
 654                 return;
 655
 656         spin_lock(&sbi->dir_inode_lock);
 657         if (get_dirty_dents(inode) ||
 658                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 659                 spin_unlock(&sbi->dir_inode_lock);
 660                 return;
 661         }
 662
 663         entry = F2FS_I(inode)->dirty_dir;
 664         list_del(&entry->list);
 665         F2FS_I(inode)->dirty_dir = NULL;
 666         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 667         stat_dec_dirty_dir(sbi);
 668         spin_unlock(&sbi->dir_inode_lock);
 669         kmem_cache_free(inode_entry_slab, entry);
 670
 671         /* Only from the recovery routine */
 672         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 673                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 674                 iput(inode);
 675         }
 676 }
 677
 678 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 679 {
 680         struct list_head *head;
 681         struct dir_inode_entry *entry;
 682         struct inode *inode;
 683 retry:
 684         spin_lock(&sbi->dir_inode_lock);
 685
 686         head = &sbi->dir_inode_list;
 687         if (list_empty(head)) {
 688                 spin_unlock(&sbi->dir_inode_lock);
 689                 return;
 690         }
 691         entry = list_entry(head->next, struct dir_inode_entry, list);
 692         inode = igrab(entry->inode);
 693         spin_unlock(&sbi->dir_inode_lock);
 694         if (inode) {
 695                 filemap_fdatawrite(inode->i_mapping);
 696                 iput(inode);
 697         } else {
 698                 /*
 699                  * We should submit bio, since it exists several
 700                  * wribacking dentry pages in the freeing inode.
 701                  */
 702                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 703         }
 704         goto retry;
 705 }
 706
 707 /*
 708  * Freeze all the FS-operations for checkpoint.
 709  */
 710 static void block_operations(struct f2fs_sb_info *sbi)
 711 {
 712         struct writeback_control wbc = {
 713                 .sync_mode = WB_SYNC_ALL,
 714                 .nr_to_write = LONG_MAX,
 715                 .for_reclaim = 0,
 716         };
 717         struct blk_plug plug;
 718
 719         blk_start_plug(&plug);
 720
 721 retry_flush_dents:
 722         f2fs_lock_all(sbi);
 723         /* write all the dirty dentry pages */
 724         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 725                 f2fs_unlock_all(sbi);
 726                 sync_dirty_dir_inodes(sbi);
 727                 goto retry_flush_dents;
 728         }
 729
 730         /*
 731          * POR: we should ensure that there is no dirty node pages
 732          * until finishing nat/sit flush.
 733          */
 734 retry_flush_nodes:
 735         mutex_lock(&sbi->node_write);
 736
 737         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 738                 mutex_unlock(&sbi->node_write);
 739                 sync_node_pages(sbi, 0, &wbc);
 740                 goto retry_flush_nodes;
 741         }
 742         blk_finish_plug(&plug);
 743 }
 744
 745 static void unblock_operations(struct f2fs_sb_info *sbi)
 746 {
 747         mutex_unlock(&sbi->node_write);
 748         f2fs_unlock_all(sbi);
 749 }
 750
 751 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 752 {
 753         DEFINE_WAIT(wait);
 754
 755         for (;;) {
 756                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 757
 758                 if (!get_pages(sbi, F2FS_WRITEBACK))
 759                         break;
 760
 761                 io_schedule();
 762         }
 763         finish_wait(&sbi->cp_wait, &wait);
 764 }
 765
 766 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 767 {
 768         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 769         nid_t last_nid = 0;
 770         block_t start_blk;
 771         struct page *cp_page;
 772         unsigned int data_sum_blocks, orphan_blocks;
 773         __u32 crc32 = 0;
 774         void *kaddr;
 775         int i;
 776         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 777
 778         /*
 779          * This avoids to conduct wrong roll-forward operations and uses
 780          * metapages, so should be called prior to sync_meta_pages below.
 781          */
 782         discard_next_dnode(sbi);
 783
 784         /* Flush all the NAT/SIT pages */
 785         while (get_pages(sbi, F2FS_DIRTY_META))
 786                 sync_meta_pages(sbi, META, LONG_MAX);
 787
 788         next_free_nid(sbi, &last_nid);
 789
 790         /*
 791          * modify checkpoint
 792          * version number is already updated
 793          */
 794         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 795         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 796         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 797         for (i = 0; i < 3; i++) {
 798                 ckpt->cur_node_segno[i] =
 799                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 800                 ckpt->cur_node_blkoff[i] =
 801                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 802                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 803                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 804         }
 805         for (i = 0; i < 3; i++) {
 806                 ckpt->cur_data_segno[i] =
 807                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 808                 ckpt->cur_data_blkoff[i] =
 809                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 810                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 811                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 812         }
 813
 814         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 815         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 816         ckpt->next_free_nid = cpu_to_le32(last_nid);
 817
 818         /* 2 cp  + n data seg summary + orphan inode blocks */
 819         data_sum_blocks = npages_for_summary_flush(sbi);
 820         if (data_sum_blocks < 3)
 821                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 822         else
 823                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 824
 825         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 826                                         / F2FS_ORPHANS_PER_BLOCK;
 827         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 828                         orphan_blocks);
 829
 830         if (is_umount) {
 831                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 832                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 833                                 cp_payload_blks + data_sum_blocks +
 834                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 835         } else {
 836                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 837                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 838                                 cp_payload_blks + data_sum_blocks +
 839                                 orphan_blocks);
 840         }
 841
 842         if (sbi->n_orphans)
 843                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 844         else
 845                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 846
 847         /* update SIT/NAT bitmap */
 848         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 849         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 850
 851         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 852         *((__le32 *)((unsigned char *)ckpt +
 853                                 le32_to_cpu(ckpt->checksum_offset)))
 854                                 = cpu_to_le32(crc32);
 855
 856         start_blk = __start_cp_addr(sbi);
 857
 858         /* write out checkpoint buffer at block 0 */
 859         cp_page = grab_meta_page(sbi, start_blk++);
 860         kaddr = page_address(cp_page);
 861         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 862         set_page_dirty(cp_page);
 863         f2fs_put_page(cp_page, 1);
 864
 865         for (i = 1; i < 1 + cp_payload_blks; i++) {
 866                 cp_page = grab_meta_page(sbi, start_blk++);
 867                 kaddr = page_address(cp_page);
 868                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 869                                 (1 << sbi->log_blocksize));
 870                 set_page_dirty(cp_page);
 871                 f2fs_put_page(cp_page, 1);
 872         }
 873
 874         if (sbi->n_orphans) {
 875                 write_orphan_inodes(sbi, start_blk);
 876                 start_blk += orphan_blocks;
 877         }
 878
 879         write_data_summaries(sbi, start_blk);
 880         start_blk += data_sum_blocks;
 881         if (is_umount) {
 882                 write_node_summaries(sbi, start_blk);
 883                 start_blk += NR_CURSEG_NODE_TYPE;
 884         }
 885
 886         /* writeout checkpoint block */
 887         cp_page = grab_meta_page(sbi, start_blk);
 888         kaddr = page_address(cp_page);
 889         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 890         set_page_dirty(cp_page);
 891         f2fs_put_page(cp_page, 1);
 892
 893         /* wait for previous submitted node/meta pages writeback */
 894         wait_on_all_pages_writeback(sbi);
 895
 896         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 897         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 898
 899         /* update user_block_counts */
 900         sbi->last_valid_block_count = sbi->total_valid_block_count;
 901         sbi->alloc_valid_block_count = 0;
 902
 903         /* Here, we only have one bio having CP pack */
 904         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 905
 906         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 907                 clear_prefree_segments(sbi);
 908                 F2FS_RESET_SB_DIRT(sbi);
 909         }
 910 }
 911
 912 /*
 913  * We guarantee that this checkpoint procedure should not fail.
 914  */
 915 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 916 {
 917         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 918         unsigned long long ckpt_ver;
 919
 920         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 921
 922         mutex_lock(&sbi->cp_mutex);
 923         block_operations(sbi);
 924
 925         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 926
 927         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 928         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 929         f2fs_submit_merged_bio(sbi, META, WRITE);
 930
 931         /*
 932          * update checkpoint pack index
 933          * Increase the version number so that
 934          * SIT entries and seg summaries are written at correct place
 935          */
 936         ckpt_ver = cur_cp_version(ckpt);
 937         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 938
 939         /* write cached NAT/SIT entries to NAT/SIT area */
 940         flush_nat_entries(sbi);
 941         flush_sit_entries(sbi);
 942
 943         /* unlock all the fs_lock[] in do_checkpoint() */
 944         do_checkpoint(sbi, is_umount);
 945
 946         unblock_operations(sbi);
 947         mutex_unlock(&sbi->cp_mutex);
 948
 949         stat_inc_cp_count(sbi->stat_info);
 950         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 951 }
 952
 953 void init_ino_entry_info(struct f2fs_sb_info *sbi)
 954 {
 955         int i;
 956
 957         for (i = 0; i < MAX_INO_ENTRY; i++) {
 958                 spin_lock_init(&sbi->ino_lock[i]);
 959                 INIT_LIST_HEAD(&sbi->ino_list[i]);
 960         }
 961
 962         /*
 963          * considering 512 blocks in a segment 8 blocks are needed for cp
 964          * and log segment summaries. Remaining blocks are used to keep
 965          * orphan entries with the limitation one reserved segment
 966          * for cp pack we can have max 1020*504 orphan entries
 967          */
 968         sbi->n_orphans = 0;
 969         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 970                                 * F2FS_ORPHANS_PER_BLOCK;
 971 }
 972
 973 int __init create_checkpoint_caches(void)
 974 {
 975         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
 976                         sizeof(struct ino_entry));
 977         if (!ino_entry_slab)
 978                 return -ENOMEM;
 979         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 980                         sizeof(struct dir_inode_entry));
 981         if (!inode_entry_slab) {
 982                 kmem_cache_destroy(ino_entry_slab);
 983                 return -ENOMEM;
 984         }
 985         return 0;
 986 }
 987
 988 void destroy_checkpoint_caches(void)
 989 {
 990         kmem_cache_destroy(ino_entry_slab);
 991         kmem_cache_destroy(inode_entry_slab);
 992 }