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 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 330 {
 331         /* add new dirty ino entry into list */
 332         __add_ino_entry(sbi, ino, type);
 333 }
 334
 335 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 336 {
 337         /* remove dirty ino entry from list */
 338         __remove_ino_entry(sbi, ino, type);
 339 }
 340
 341 /* mode should be APPEND_INO or UPDATE_INO */
 342 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 343 {
 344         struct ino_entry *e;
 345         spin_lock(&sbi->ino_lock[mode]);
 346         e = radix_tree_lookup(&sbi->ino_root[mode], ino);
 347         spin_unlock(&sbi->ino_lock[mode]);
 348         return e ? true : false;
 349 }
 350
 351 static void release_dirty_inode(struct f2fs_sb_info *sbi)
 352 {
 353         struct ino_entry *e, *tmp;
 354         int i;
 355
 356         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 357                 spin_lock(&sbi->ino_lock[i]);
 358                 list_for_each_entry_safe(e, tmp, &sbi->ino_list[i], list) {
 359                         list_del(&e->list);
 360                         radix_tree_delete(&sbi->ino_root[i], e->ino);
 361                         kmem_cache_free(ino_entry_slab, e);
 362                 }
 363                 spin_unlock(&sbi->ino_lock[i]);
 364         }
 365 }
 366
 367 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 368 {
 369         int err = 0;
 370
 371         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 372         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 373                 err = -ENOSPC;
 374         else
 375                 sbi->n_orphans++;
 376         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 377
 378         return err;
 379 }
 380
 381 void release_orphan_inode(struct f2fs_sb_info *sbi)
 382 {
 383         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 384         f2fs_bug_on(sbi->n_orphans == 0);
 385         sbi->n_orphans--;
 386         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 387 }
 388
 389 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 390 {
 391         /* add new orphan ino entry into list */
 392         __add_ino_entry(sbi, ino, ORPHAN_INO);
 393 }
 394
 395 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 396 {
 397         /* remove orphan entry from orphan list */
 398         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 399 }
 400
 401 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 402 {
 403         struct inode *inode = f2fs_iget(sbi->sb, ino);
 404         f2fs_bug_on(IS_ERR(inode));
 405         clear_nlink(inode);
 406
 407         /* truncate all the data during iput */
 408         iput(inode);
 409 }
 410
 411 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 412 {
 413         block_t start_blk, orphan_blkaddr, i, j;
 414
 415         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 416                 return;
 417
 418         sbi->por_doing = true;
 419
 420         start_blk = __start_cp_addr(sbi) + 1 +
 421                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 422         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 423
 424         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 425
 426         for (i = 0; i < orphan_blkaddr; i++) {
 427                 struct page *page = get_meta_page(sbi, start_blk + i);
 428                 struct f2fs_orphan_block *orphan_blk;
 429
 430                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 431                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 432                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 433                         recover_orphan_inode(sbi, ino);
 434                 }
 435                 f2fs_put_page(page, 1);
 436         }
 437         /* clear Orphan Flag */
 438         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 439         sbi->por_doing = false;
 440         return;
 441 }
 442
 443 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 444 {
 445         struct list_head *head;
 446         struct f2fs_orphan_block *orphan_blk = NULL;
 447         unsigned int nentries = 0;
 448         unsigned short index;
 449         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 450                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 451         struct page *page = NULL;
 452         struct ino_entry *orphan = NULL;
 453
 454         for (index = 0; index < orphan_blocks; index++)
 455                 grab_meta_page(sbi, start_blk + index);
 456
 457         index = 1;
 458         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 459         head = &sbi->ino_list[ORPHAN_INO];
 460
 461         /* loop for each orphan inode entry and write them in Jornal block */
 462         list_for_each_entry(orphan, head, list) {
 463                 if (!page) {
 464                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 465                         f2fs_bug_on(!page);
 466                         orphan_blk =
 467                                 (struct f2fs_orphan_block *)page_address(page);
 468                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 469                         f2fs_put_page(page, 0);
 470                 }
 471
 472                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 473
 474                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 475                         /*
 476                          * an orphan block is full of 1020 entries,
 477                          * then we need to flush current orphan blocks
 478                          * and bring another one in memory
 479                          */
 480                         orphan_blk->blk_addr = cpu_to_le16(index);
 481                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 482                         orphan_blk->entry_count = cpu_to_le32(nentries);
 483                         set_page_dirty(page);
 484                         f2fs_put_page(page, 1);
 485                         index++;
 486                         nentries = 0;
 487                         page = NULL;
 488                 }
 489         }
 490
 491         if (page) {
 492                 orphan_blk->blk_addr = cpu_to_le16(index);
 493                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 494                 orphan_blk->entry_count = cpu_to_le32(nentries);
 495                 set_page_dirty(page);
 496                 f2fs_put_page(page, 1);
 497         }
 498
 499         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 500 }
 501
 502 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 503                                 block_t cp_addr, unsigned long long *version)
 504 {
 505         struct page *cp_page_1, *cp_page_2 = NULL;
 506         unsigned long blk_size = sbi->blocksize;
 507         struct f2fs_checkpoint *cp_block;
 508         unsigned long long cur_version = 0, pre_version = 0;
 509         size_t crc_offset;
 510         __u32 crc = 0;
 511
 512         /* Read the 1st cp block in this CP pack */
 513         cp_page_1 = get_meta_page(sbi, cp_addr);
 514
 515         /* get the version number */
 516         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 517         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 518         if (crc_offset >= blk_size)
 519                 goto invalid_cp1;
 520
 521         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 522         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 523                 goto invalid_cp1;
 524
 525         pre_version = cur_cp_version(cp_block);
 526
 527         /* Read the 2nd cp block in this CP pack */
 528         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 529         cp_page_2 = get_meta_page(sbi, cp_addr);
 530
 531         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 532         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 533         if (crc_offset >= blk_size)
 534                 goto invalid_cp2;
 535
 536         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 537         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 538                 goto invalid_cp2;
 539
 540         cur_version = cur_cp_version(cp_block);
 541
 542         if (cur_version == pre_version) {
 543                 *version = cur_version;
 544                 f2fs_put_page(cp_page_2, 1);
 545                 return cp_page_1;
 546         }
 547 invalid_cp2:
 548         f2fs_put_page(cp_page_2, 1);
 549 invalid_cp1:
 550         f2fs_put_page(cp_page_1, 1);
 551         return NULL;
 552 }
 553
 554 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 555 {
 556         struct f2fs_checkpoint *cp_block;
 557         struct f2fs_super_block *fsb = sbi->raw_super;
 558         struct page *cp1, *cp2, *cur_page;
 559         unsigned long blk_size = sbi->blocksize;
 560         unsigned long long cp1_version = 0, cp2_version = 0;
 561         unsigned long long cp_start_blk_no;
 562         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 563         block_t cp_blk_no;
 564         int i;
 565
 566         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 567         if (!sbi->ckpt)
 568                 return -ENOMEM;
 569         /*
 570          * Finding out valid cp block involves read both
 571          * sets( cp pack1 and cp pack 2)
 572          */
 573         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 574         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 575
 576         /* The second checkpoint pack should start at the next segment */
 577         cp_start_blk_no += ((unsigned long long)1) <<
 578                                 le32_to_cpu(fsb->log_blocks_per_seg);
 579         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 580
 581         if (cp1 && cp2) {
 582                 if (ver_after(cp2_version, cp1_version))
 583                         cur_page = cp2;
 584                 else
 585                         cur_page = cp1;
 586         } else if (cp1) {
 587                 cur_page = cp1;
 588         } else if (cp2) {
 589                 cur_page = cp2;
 590         } else {
 591                 goto fail_no_cp;
 592         }
 593
 594         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 595         memcpy(sbi->ckpt, cp_block, blk_size);
 596
 597         if (cp_blks <= 1)
 598                 goto done;
 599
 600         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 601         if (cur_page == cp2)
 602                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 603
 604         for (i = 1; i < cp_blks; i++) {
 605                 void *sit_bitmap_ptr;
 606                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 607
 608                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 609                 sit_bitmap_ptr = page_address(cur_page);
 610                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 611                 f2fs_put_page(cur_page, 1);
 612         }
 613 done:
 614         f2fs_put_page(cp1, 1);
 615         f2fs_put_page(cp2, 1);
 616         return 0;
 617
 618 fail_no_cp:
 619         kfree(sbi->ckpt);
 620         return -EINVAL;
 621 }
 622
 623 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 624 {
 625         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 626
 627         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 628                 return -EEXIST;
 629
 630         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 631         F2FS_I(inode)->dirty_dir = new;
 632         list_add_tail(&new->list, &sbi->dir_inode_list);
 633         stat_inc_dirty_dir(sbi);
 634         return 0;
 635 }
 636
 637 void set_dirty_dir_page(struct inode *inode, struct page *page)
 638 {
 639         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 640         struct dir_inode_entry *new;
 641         int ret = 0;
 642
 643         if (!S_ISDIR(inode->i_mode))
 644                 return;
 645
 646         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 647         new->inode = inode;
 648         INIT_LIST_HEAD(&new->list);
 649
 650         spin_lock(&sbi->dir_inode_lock);
 651         ret = __add_dirty_inode(inode, new);
 652         inode_inc_dirty_dents(inode);
 653         SetPagePrivate(page);
 654         spin_unlock(&sbi->dir_inode_lock);
 655
 656         if (ret)
 657                 kmem_cache_free(inode_entry_slab, new);
 658 }
 659
 660 void add_dirty_dir_inode(struct inode *inode)
 661 {
 662         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 663         struct dir_inode_entry *new =
 664                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 665         int ret = 0;
 666
 667         new->inode = inode;
 668         INIT_LIST_HEAD(&new->list);
 669
 670         spin_lock(&sbi->dir_inode_lock);
 671         ret = __add_dirty_inode(inode, new);
 672         spin_unlock(&sbi->dir_inode_lock);
 673
 674         if (ret)
 675                 kmem_cache_free(inode_entry_slab, new);
 676 }
 677
 678 void remove_dirty_dir_inode(struct inode *inode)
 679 {
 680         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 681         struct dir_inode_entry *entry;
 682
 683         if (!S_ISDIR(inode->i_mode))
 684                 return;
 685
 686         spin_lock(&sbi->dir_inode_lock);
 687         if (get_dirty_dents(inode) ||
 688                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 689                 spin_unlock(&sbi->dir_inode_lock);
 690                 return;
 691         }
 692
 693         entry = F2FS_I(inode)->dirty_dir;
 694         list_del(&entry->list);
 695         F2FS_I(inode)->dirty_dir = NULL;
 696         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 697         stat_dec_dirty_dir(sbi);
 698         spin_unlock(&sbi->dir_inode_lock);
 699         kmem_cache_free(inode_entry_slab, entry);
 700
 701         /* Only from the recovery routine */
 702         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 703                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 704                 iput(inode);
 705         }
 706 }
 707
 708 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 709 {
 710         struct list_head *head;
 711         struct dir_inode_entry *entry;
 712         struct inode *inode;
 713 retry:
 714         spin_lock(&sbi->dir_inode_lock);
 715
 716         head = &sbi->dir_inode_list;
 717         if (list_empty(head)) {
 718                 spin_unlock(&sbi->dir_inode_lock);
 719                 return;
 720         }
 721         entry = list_entry(head->next, struct dir_inode_entry, list);
 722         inode = igrab(entry->inode);
 723         spin_unlock(&sbi->dir_inode_lock);
 724         if (inode) {
 725                 filemap_fdatawrite(inode->i_mapping);
 726                 iput(inode);
 727         } else {
 728                 /*
 729                  * We should submit bio, since it exists several
 730                  * wribacking dentry pages in the freeing inode.
 731                  */
 732                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 733         }
 734         goto retry;
 735 }
 736
 737 /*
 738  * Freeze all the FS-operations for checkpoint.
 739  */
 740 static void block_operations(struct f2fs_sb_info *sbi)
 741 {
 742         struct writeback_control wbc = {
 743                 .sync_mode = WB_SYNC_ALL,
 744                 .nr_to_write = LONG_MAX,
 745                 .for_reclaim = 0,
 746         };
 747         struct blk_plug plug;
 748
 749         blk_start_plug(&plug);
 750
 751 retry_flush_dents:
 752         f2fs_lock_all(sbi);
 753         /* write all the dirty dentry pages */
 754         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 755                 f2fs_unlock_all(sbi);
 756                 sync_dirty_dir_inodes(sbi);
 757                 goto retry_flush_dents;
 758         }
 759
 760         /*
 761          * POR: we should ensure that there is no dirty node pages
 762          * until finishing nat/sit flush.
 763          */
 764 retry_flush_nodes:
 765         down_write(&sbi->node_write);
 766
 767         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 768                 up_write(&sbi->node_write);
 769                 sync_node_pages(sbi, 0, &wbc);
 770                 goto retry_flush_nodes;
 771         }
 772         blk_finish_plug(&plug);
 773 }
 774
 775 static void unblock_operations(struct f2fs_sb_info *sbi)
 776 {
 777         up_write(&sbi->node_write);
 778         f2fs_unlock_all(sbi);
 779 }
 780
 781 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 782 {
 783         DEFINE_WAIT(wait);
 784
 785         for (;;) {
 786                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 787
 788                 if (!get_pages(sbi, F2FS_WRITEBACK))
 789                         break;
 790
 791                 io_schedule();
 792         }
 793         finish_wait(&sbi->cp_wait, &wait);
 794 }
 795
 796 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 797 {
 798         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 799         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 800         nid_t last_nid = 0;
 801         block_t start_blk;
 802         struct page *cp_page;
 803         unsigned int data_sum_blocks, orphan_blocks;
 804         __u32 crc32 = 0;
 805         void *kaddr;
 806         int i;
 807         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 808
 809         /*
 810          * This avoids to conduct wrong roll-forward operations and uses
 811          * metapages, so should be called prior to sync_meta_pages below.
 812          */
 813         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 814
 815         /* Flush all the NAT/SIT pages */
 816         while (get_pages(sbi, F2FS_DIRTY_META))
 817                 sync_meta_pages(sbi, META, LONG_MAX);
 818
 819         next_free_nid(sbi, &last_nid);
 820
 821         /*
 822          * modify checkpoint
 823          * version number is already updated
 824          */
 825         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 826         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 827         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 828         for (i = 0; i < 3; i++) {
 829                 ckpt->cur_node_segno[i] =
 830                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 831                 ckpt->cur_node_blkoff[i] =
 832                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 833                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 834                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 835         }
 836         for (i = 0; i < 3; i++) {
 837                 ckpt->cur_data_segno[i] =
 838                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 839                 ckpt->cur_data_blkoff[i] =
 840                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 841                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 842                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 843         }
 844
 845         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 846         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 847         ckpt->next_free_nid = cpu_to_le32(last_nid);
 848
 849         /* 2 cp  + n data seg summary + orphan inode blocks */
 850         data_sum_blocks = npages_for_summary_flush(sbi);
 851         if (data_sum_blocks < 3)
 852                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 853         else
 854                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 855
 856         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 857                                         / F2FS_ORPHANS_PER_BLOCK;
 858         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 859                         orphan_blocks);
 860
 861         if (is_umount) {
 862                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 863                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 864                                 cp_payload_blks + data_sum_blocks +
 865                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 866         } else {
 867                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 868                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 869                                 cp_payload_blks + data_sum_blocks +
 870                                 orphan_blocks);
 871         }
 872
 873         if (sbi->n_orphans)
 874                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 875         else
 876                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 877
 878         /* update SIT/NAT bitmap */
 879         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 880         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 881
 882         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 883         *((__le32 *)((unsigned char *)ckpt +
 884                                 le32_to_cpu(ckpt->checksum_offset)))
 885                                 = cpu_to_le32(crc32);
 886
 887         start_blk = __start_cp_addr(sbi);
 888
 889         /* write out checkpoint buffer at block 0 */
 890         cp_page = grab_meta_page(sbi, start_blk++);
 891         kaddr = page_address(cp_page);
 892         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 893         set_page_dirty(cp_page);
 894         f2fs_put_page(cp_page, 1);
 895
 896         for (i = 1; i < 1 + cp_payload_blks; i++) {
 897                 cp_page = grab_meta_page(sbi, start_blk++);
 898                 kaddr = page_address(cp_page);
 899                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 900                                 (1 << sbi->log_blocksize));
 901                 set_page_dirty(cp_page);
 902                 f2fs_put_page(cp_page, 1);
 903         }
 904
 905         if (sbi->n_orphans) {
 906                 write_orphan_inodes(sbi, start_blk);
 907                 start_blk += orphan_blocks;
 908         }
 909
 910         write_data_summaries(sbi, start_blk);
 911         start_blk += data_sum_blocks;
 912         if (is_umount) {
 913                 write_node_summaries(sbi, start_blk);
 914                 start_blk += NR_CURSEG_NODE_TYPE;
 915         }
 916
 917         /* writeout checkpoint block */
 918         cp_page = grab_meta_page(sbi, start_blk);
 919         kaddr = page_address(cp_page);
 920         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 921         set_page_dirty(cp_page);
 922         f2fs_put_page(cp_page, 1);
 923
 924         /* wait for previous submitted node/meta pages writeback */
 925         wait_on_all_pages_writeback(sbi);
 926
 927         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 928         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 929
 930         /* update user_block_counts */
 931         sbi->last_valid_block_count = sbi->total_valid_block_count;
 932         sbi->alloc_valid_block_count = 0;
 933
 934         /* Here, we only have one bio having CP pack */
 935         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 936
 937         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 938                 clear_prefree_segments(sbi);
 939                 release_dirty_inode(sbi);
 940                 F2FS_RESET_SB_DIRT(sbi);
 941         }
 942 }
 943
 944 /*
 945  * We guarantee that this checkpoint procedure should not fail.
 946  */
 947 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 948 {
 949         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 950         unsigned long long ckpt_ver;
 951
 952         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 953
 954         mutex_lock(&sbi->cp_mutex);
 955         block_operations(sbi);
 956
 957         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 958
 959         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 960         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 961         f2fs_submit_merged_bio(sbi, META, WRITE);
 962
 963         /*
 964          * update checkpoint pack index
 965          * Increase the version number so that
 966          * SIT entries and seg summaries are written at correct place
 967          */
 968         ckpt_ver = cur_cp_version(ckpt);
 969         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 970
 971         /* write cached NAT/SIT entries to NAT/SIT area */
 972         flush_nat_entries(sbi);
 973         flush_sit_entries(sbi);
 974
 975         /* unlock all the fs_lock[] in do_checkpoint() */
 976         do_checkpoint(sbi, is_umount);
 977
 978         unblock_operations(sbi);
 979         mutex_unlock(&sbi->cp_mutex);
 980
 981         stat_inc_cp_count(sbi->stat_info);
 982         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 983 }
 984
 985 void init_ino_entry_info(struct f2fs_sb_info *sbi)
 986 {
 987         int i;
 988
 989         for (i = 0; i < MAX_INO_ENTRY; i++) {
 990                 INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
 991                 spin_lock_init(&sbi->ino_lock[i]);
 992                 INIT_LIST_HEAD(&sbi->ino_list[i]);
 993         }
 994
 995         /*
 996          * considering 512 blocks in a segment 8 blocks are needed for cp
 997          * and log segment summaries. Remaining blocks are used to keep
 998          * orphan entries with the limitation one reserved segment
 999          * for cp pack we can have max 1020*504 orphan entries
1000          */
1001         sbi->n_orphans = 0;
1002         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
1003                                 * F2FS_ORPHANS_PER_BLOCK;
1004 }
1005
1006 int __init create_checkpoint_caches(void)
1007 {
1008         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1009                         sizeof(struct ino_entry));
1010         if (!ino_entry_slab)
1011                 return -ENOMEM;
1012         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1013                         sizeof(struct dir_inode_entry));
1014         if (!inode_entry_slab) {
1015                 kmem_cache_destroy(ino_entry_slab);
1016                 return -ENOMEM;
1017         }
1018         return 0;
1019 }
1020
1021 void destroy_checkpoint_caches(void)
1022 {
1023         kmem_cache_destroy(ino_entry_slab);
1024         kmem_cache_destroy(inode_entry_slab);
1025 }