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         if (unlikely(f2fs_cp_error(sbi)))
 164                 goto redirty_out;
 165
 166         f2fs_wait_on_page_writeback(page, META);
 167         write_meta_page(sbi, page);
 168         dec_page_count(sbi, F2FS_DIRTY_META);
 169         unlock_page(page);
 170         return 0;
 171
 172 redirty_out:
 173         redirty_page_for_writepage(wbc, page);
 174         return AOP_WRITEPAGE_ACTIVATE;
 175 }
 176
 177 static int f2fs_write_meta_pages(struct address_space *mapping,
 178                                 struct writeback_control *wbc)
 179 {
 180         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 181         long diff, written;
 182
 183         trace_f2fs_writepages(mapping->host, wbc, META);
 184
 185         /* collect a number of dirty meta pages and write together */
 186         if (wbc->for_kupdate ||
 187                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 188                 goto skip_write;
 189
 190         /* if mounting is failed, skip writing node pages */
 191         mutex_lock(&sbi->cp_mutex);
 192         diff = nr_pages_to_write(sbi, META, wbc);
 193         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 194         mutex_unlock(&sbi->cp_mutex);
 195         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 196         return 0;
 197
 198 skip_write:
 199         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 200         return 0;
 201 }
 202
 203 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 204                                                 long nr_to_write)
 205 {
 206         struct address_space *mapping = META_MAPPING(sbi);
 207         pgoff_t index = 0, end = LONG_MAX;
 208         struct pagevec pvec;
 209         long nwritten = 0;
 210         struct writeback_control wbc = {
 211                 .for_reclaim = 0,
 212         };
 213
 214         pagevec_init(&pvec, 0);
 215
 216         while (index <= end) {
 217                 int i, nr_pages;
 218                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 219                                 PAGECACHE_TAG_DIRTY,
 220                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 221                 if (unlikely(nr_pages == 0))
 222                         break;
 223
 224                 for (i = 0; i < nr_pages; i++) {
 225                         struct page *page = pvec.pages[i];
 226
 227                         lock_page(page);
 228
 229                         if (unlikely(page->mapping != mapping)) {
 230 continue_unlock:
 231                                 unlock_page(page);
 232                                 continue;
 233                         }
 234                         if (!PageDirty(page)) {
 235                                 /* someone wrote it for us */
 236                                 goto continue_unlock;
 237                         }
 238
 239                         if (!clear_page_dirty_for_io(page))
 240                                 goto continue_unlock;
 241
 242                         if (f2fs_write_meta_page(page, &wbc)) {
 243                                 unlock_page(page);
 244                                 break;
 245                         }
 246                         nwritten++;
 247                         if (unlikely(nwritten >= nr_to_write))
 248                                 break;
 249                 }
 250                 pagevec_release(&pvec);
 251                 cond_resched();
 252         }
 253
 254         if (nwritten)
 255                 f2fs_submit_merged_bio(sbi, type, WRITE);
 256
 257         return nwritten;
 258 }
 259
 260 static int f2fs_set_meta_page_dirty(struct page *page)
 261 {
 262         struct address_space *mapping = page->mapping;
 263         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 264
 265         trace_f2fs_set_page_dirty(page, META);
 266
 267         SetPageUptodate(page);
 268         if (!PageDirty(page)) {
 269                 __set_page_dirty_nobuffers(page);
 270                 inc_page_count(sbi, F2FS_DIRTY_META);
 271                 return 1;
 272         }
 273         return 0;
 274 }
 275
 276 const struct address_space_operations f2fs_meta_aops = {
 277         .writepage      = f2fs_write_meta_page,
 278         .writepages     = f2fs_write_meta_pages,
 279         .set_page_dirty = f2fs_set_meta_page_dirty,
 280 };
 281
 282 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 283 {
 284         struct ino_entry *e;
 285 retry:
 286         spin_lock(&sbi->ino_lock[type]);
 287
 288         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 289         if (!e) {
 290                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 291                 if (!e) {
 292                         spin_unlock(&sbi->ino_lock[type]);
 293                         goto retry;
 294                 }
 295                 if (radix_tree_insert(&sbi->ino_root[type], ino, e)) {
 296                         spin_unlock(&sbi->ino_lock[type]);
 297                         kmem_cache_free(ino_entry_slab, e);
 298                         goto retry;
 299                 }
 300                 memset(e, 0, sizeof(struct ino_entry));
 301                 e->ino = ino;
 302
 303                 list_add_tail(&e->list, &sbi->ino_list[type]);
 304         }
 305         spin_unlock(&sbi->ino_lock[type]);
 306 }
 307
 308 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 309 {
 310         struct ino_entry *e;
 311
 312         spin_lock(&sbi->ino_lock[type]);
 313         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 314         if (e) {
 315                 list_del(&e->list);
 316                 radix_tree_delete(&sbi->ino_root[type], ino);
 317                 if (type == ORPHAN_INO)
 318                         sbi->n_orphans--;
 319                 spin_unlock(&sbi->ino_lock[type]);
 320                 kmem_cache_free(ino_entry_slab, e);
 321                 return;
 322         }
 323         spin_unlock(&sbi->ino_lock[type]);
 324 }
 325
 326 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 327 {
 328         /* add new dirty ino entry into list */
 329         __add_ino_entry(sbi, ino, type);
 330 }
 331
 332 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 333 {
 334         /* remove dirty ino entry from list */
 335         __remove_ino_entry(sbi, ino, type);
 336 }
 337
 338 /* mode should be APPEND_INO or UPDATE_INO */
 339 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 340 {
 341         struct ino_entry *e;
 342         spin_lock(&sbi->ino_lock[mode]);
 343         e = radix_tree_lookup(&sbi->ino_root[mode], ino);
 344         spin_unlock(&sbi->ino_lock[mode]);
 345         return e ? true : false;
 346 }
 347
 348 void release_dirty_inode(struct f2fs_sb_info *sbi)
 349 {
 350         struct ino_entry *e, *tmp;
 351         int i;
 352
 353         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 354                 spin_lock(&sbi->ino_lock[i]);
 355                 list_for_each_entry_safe(e, tmp, &sbi->ino_list[i], list) {
 356                         list_del(&e->list);
 357                         radix_tree_delete(&sbi->ino_root[i], e->ino);
 358                         kmem_cache_free(ino_entry_slab, e);
 359                 }
 360                 spin_unlock(&sbi->ino_lock[i]);
 361         }
 362 }
 363
 364 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 365 {
 366         int err = 0;
 367
 368         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 369         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 370                 err = -ENOSPC;
 371         else
 372                 sbi->n_orphans++;
 373         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 374
 375         return err;
 376 }
 377
 378 void release_orphan_inode(struct f2fs_sb_info *sbi)
 379 {
 380         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 381         f2fs_bug_on(sbi->n_orphans == 0);
 382         sbi->n_orphans--;
 383         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 384 }
 385
 386 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 387 {
 388         /* add new orphan ino entry into list */
 389         __add_ino_entry(sbi, ino, ORPHAN_INO);
 390 }
 391
 392 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 393 {
 394         /* remove orphan entry from orphan list */
 395         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 396 }
 397
 398 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 399 {
 400         struct inode *inode = f2fs_iget(sbi->sb, ino);
 401         f2fs_bug_on(IS_ERR(inode));
 402         clear_nlink(inode);
 403
 404         /* truncate all the data during iput */
 405         iput(inode);
 406 }
 407
 408 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 409 {
 410         block_t start_blk, orphan_blkaddr, i, j;
 411
 412         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 413                 return;
 414
 415         sbi->por_doing = true;
 416
 417         start_blk = __start_cp_addr(sbi) + 1 +
 418                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 419         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 420
 421         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 422
 423         for (i = 0; i < orphan_blkaddr; i++) {
 424                 struct page *page = get_meta_page(sbi, start_blk + i);
 425                 struct f2fs_orphan_block *orphan_blk;
 426
 427                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 428                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 429                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 430                         recover_orphan_inode(sbi, ino);
 431                 }
 432                 f2fs_put_page(page, 1);
 433         }
 434         /* clear Orphan Flag */
 435         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 436         sbi->por_doing = false;
 437         return;
 438 }
 439
 440 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 441 {
 442         struct list_head *head;
 443         struct f2fs_orphan_block *orphan_blk = NULL;
 444         unsigned int nentries = 0;
 445         unsigned short index;
 446         unsigned short orphan_blocks =
 447                         (unsigned short)GET_ORPHAN_BLOCKS(sbi->n_orphans);
 448         struct page *page = NULL;
 449         struct ino_entry *orphan = NULL;
 450
 451         for (index = 0; index < orphan_blocks; index++)
 452                 grab_meta_page(sbi, start_blk + index);
 453
 454         index = 1;
 455         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 456         head = &sbi->ino_list[ORPHAN_INO];
 457
 458         /* loop for each orphan inode entry and write them in Jornal block */
 459         list_for_each_entry(orphan, head, list) {
 460                 if (!page) {
 461                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 462                         f2fs_bug_on(!page);
 463                         orphan_blk =
 464                                 (struct f2fs_orphan_block *)page_address(page);
 465                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 466                         f2fs_put_page(page, 0);
 467                 }
 468
 469                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 470
 471                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 472                         /*
 473                          * an orphan block is full of 1020 entries,
 474                          * then we need to flush current orphan blocks
 475                          * and bring another one in memory
 476                          */
 477                         orphan_blk->blk_addr = cpu_to_le16(index);
 478                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 479                         orphan_blk->entry_count = cpu_to_le32(nentries);
 480                         set_page_dirty(page);
 481                         f2fs_put_page(page, 1);
 482                         index++;
 483                         nentries = 0;
 484                         page = NULL;
 485                 }
 486         }
 487
 488         if (page) {
 489                 orphan_blk->blk_addr = cpu_to_le16(index);
 490                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 491                 orphan_blk->entry_count = cpu_to_le32(nentries);
 492                 set_page_dirty(page);
 493                 f2fs_put_page(page, 1);
 494         }
 495
 496         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 497 }
 498
 499 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 500                                 block_t cp_addr, unsigned long long *version)
 501 {
 502         struct page *cp_page_1, *cp_page_2 = NULL;
 503         unsigned long blk_size = sbi->blocksize;
 504         struct f2fs_checkpoint *cp_block;
 505         unsigned long long cur_version = 0, pre_version = 0;
 506         size_t crc_offset;
 507         __u32 crc = 0;
 508
 509         /* Read the 1st cp block in this CP pack */
 510         cp_page_1 = get_meta_page(sbi, cp_addr);
 511
 512         /* get the version number */
 513         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 514         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 515         if (crc_offset >= blk_size)
 516                 goto invalid_cp1;
 517
 518         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 519         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 520                 goto invalid_cp1;
 521
 522         pre_version = cur_cp_version(cp_block);
 523
 524         /* Read the 2nd cp block in this CP pack */
 525         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 526         cp_page_2 = get_meta_page(sbi, cp_addr);
 527
 528         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 529         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 530         if (crc_offset >= blk_size)
 531                 goto invalid_cp2;
 532
 533         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 534         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 535                 goto invalid_cp2;
 536
 537         cur_version = cur_cp_version(cp_block);
 538
 539         if (cur_version == pre_version) {
 540                 *version = cur_version;
 541                 f2fs_put_page(cp_page_2, 1);
 542                 return cp_page_1;
 543         }
 544 invalid_cp2:
 545         f2fs_put_page(cp_page_2, 1);
 546 invalid_cp1:
 547         f2fs_put_page(cp_page_1, 1);
 548         return NULL;
 549 }
 550
 551 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 552 {
 553         struct f2fs_checkpoint *cp_block;
 554         struct f2fs_super_block *fsb = sbi->raw_super;
 555         struct page *cp1, *cp2, *cur_page;
 556         unsigned long blk_size = sbi->blocksize;
 557         unsigned long long cp1_version = 0, cp2_version = 0;
 558         unsigned long long cp_start_blk_no;
 559         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 560         block_t cp_blk_no;
 561         int i;
 562
 563         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 564         if (!sbi->ckpt)
 565                 return -ENOMEM;
 566         /*
 567          * Finding out valid cp block involves read both
 568          * sets( cp pack1 and cp pack 2)
 569          */
 570         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 571         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 572
 573         /* The second checkpoint pack should start at the next segment */
 574         cp_start_blk_no += ((unsigned long long)1) <<
 575                                 le32_to_cpu(fsb->log_blocks_per_seg);
 576         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 577
 578         if (cp1 && cp2) {
 579                 if (ver_after(cp2_version, cp1_version))
 580                         cur_page = cp2;
 581                 else
 582                         cur_page = cp1;
 583         } else if (cp1) {
 584                 cur_page = cp1;
 585         } else if (cp2) {
 586                 cur_page = cp2;
 587         } else {
 588                 goto fail_no_cp;
 589         }
 590
 591         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 592         memcpy(sbi->ckpt, cp_block, blk_size);
 593
 594         if (cp_blks <= 1)
 595                 goto done;
 596
 597         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 598         if (cur_page == cp2)
 599                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 600
 601         for (i = 1; i < cp_blks; i++) {
 602                 void *sit_bitmap_ptr;
 603                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 604
 605                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 606                 sit_bitmap_ptr = page_address(cur_page);
 607                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 608                 f2fs_put_page(cur_page, 1);
 609         }
 610 done:
 611         f2fs_put_page(cp1, 1);
 612         f2fs_put_page(cp2, 1);
 613         return 0;
 614
 615 fail_no_cp:
 616         kfree(sbi->ckpt);
 617         return -EINVAL;
 618 }
 619
 620 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 621 {
 622         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 623
 624         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 625                 return -EEXIST;
 626
 627         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 628         F2FS_I(inode)->dirty_dir = new;
 629         list_add_tail(&new->list, &sbi->dir_inode_list);
 630         stat_inc_dirty_dir(sbi);
 631         return 0;
 632 }
 633
 634 void set_dirty_dir_page(struct inode *inode, struct page *page)
 635 {
 636         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 637         struct dir_inode_entry *new;
 638         int ret = 0;
 639
 640         if (!S_ISDIR(inode->i_mode))
 641                 return;
 642
 643         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 644         new->inode = inode;
 645         INIT_LIST_HEAD(&new->list);
 646
 647         spin_lock(&sbi->dir_inode_lock);
 648         ret = __add_dirty_inode(inode, new);
 649         inode_inc_dirty_dents(inode);
 650         SetPagePrivate(page);
 651         spin_unlock(&sbi->dir_inode_lock);
 652
 653         if (ret)
 654                 kmem_cache_free(inode_entry_slab, new);
 655 }
 656
 657 void add_dirty_dir_inode(struct inode *inode)
 658 {
 659         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 660         struct dir_inode_entry *new =
 661                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 662         int ret = 0;
 663
 664         new->inode = inode;
 665         INIT_LIST_HEAD(&new->list);
 666
 667         spin_lock(&sbi->dir_inode_lock);
 668         ret = __add_dirty_inode(inode, new);
 669         spin_unlock(&sbi->dir_inode_lock);
 670
 671         if (ret)
 672                 kmem_cache_free(inode_entry_slab, new);
 673 }
 674
 675 void remove_dirty_dir_inode(struct inode *inode)
 676 {
 677         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 678         struct dir_inode_entry *entry;
 679
 680         if (!S_ISDIR(inode->i_mode))
 681                 return;
 682
 683         spin_lock(&sbi->dir_inode_lock);
 684         if (get_dirty_dents(inode) ||
 685                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 686                 spin_unlock(&sbi->dir_inode_lock);
 687                 return;
 688         }
 689
 690         entry = F2FS_I(inode)->dirty_dir;
 691         list_del(&entry->list);
 692         F2FS_I(inode)->dirty_dir = NULL;
 693         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 694         stat_dec_dirty_dir(sbi);
 695         spin_unlock(&sbi->dir_inode_lock);
 696         kmem_cache_free(inode_entry_slab, entry);
 697
 698         /* Only from the recovery routine */
 699         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 700                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 701                 iput(inode);
 702         }
 703 }
 704
 705 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 706 {
 707         struct list_head *head;
 708         struct dir_inode_entry *entry;
 709         struct inode *inode;
 710 retry:
 711         spin_lock(&sbi->dir_inode_lock);
 712
 713         head = &sbi->dir_inode_list;
 714         if (list_empty(head)) {
 715                 spin_unlock(&sbi->dir_inode_lock);
 716                 return;
 717         }
 718         entry = list_entry(head->next, struct dir_inode_entry, list);
 719         inode = igrab(entry->inode);
 720         spin_unlock(&sbi->dir_inode_lock);
 721         if (inode) {
 722                 filemap_fdatawrite(inode->i_mapping);
 723                 iput(inode);
 724         } else {
 725                 /*
 726                  * We should submit bio, since it exists several
 727                  * wribacking dentry pages in the freeing inode.
 728                  */
 729                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 730         }
 731         goto retry;
 732 }
 733
 734 /*
 735  * Freeze all the FS-operations for checkpoint.
 736  */
 737 static int block_operations(struct f2fs_sb_info *sbi)
 738 {
 739         struct writeback_control wbc = {
 740                 .sync_mode = WB_SYNC_ALL,
 741                 .nr_to_write = LONG_MAX,
 742                 .for_reclaim = 0,
 743         };
 744         struct blk_plug plug;
 745         int err = 0;
 746
 747         blk_start_plug(&plug);
 748
 749 retry_flush_dents:
 750         f2fs_lock_all(sbi);
 751         /* write all the dirty dentry pages */
 752         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 753                 f2fs_unlock_all(sbi);
 754                 sync_dirty_dir_inodes(sbi);
 755                 if (unlikely(f2fs_cp_error(sbi))) {
 756                         err = -EIO;
 757                         goto out;
 758                 }
 759                 goto retry_flush_dents;
 760         }
 761
 762         /*
 763          * POR: we should ensure that there are no dirty node pages
 764          * until finishing nat/sit flush.
 765          */
 766 retry_flush_nodes:
 767         down_write(&sbi->node_write);
 768
 769         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 770                 up_write(&sbi->node_write);
 771                 sync_node_pages(sbi, 0, &wbc);
 772                 if (unlikely(f2fs_cp_error(sbi))) {
 773                         f2fs_unlock_all(sbi);
 774                         err = -EIO;
 775                         goto out;
 776                 }
 777                 goto retry_flush_nodes;
 778         }
 779 out:
 780         blk_finish_plug(&plug);
 781         return err;
 782 }
 783
 784 static void unblock_operations(struct f2fs_sb_info *sbi)
 785 {
 786         up_write(&sbi->node_write);
 787         f2fs_unlock_all(sbi);
 788 }
 789
 790 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 791 {
 792         DEFINE_WAIT(wait);
 793
 794         for (;;) {
 795                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 796
 797                 if (!get_pages(sbi, F2FS_WRITEBACK))
 798                         break;
 799
 800                 io_schedule();
 801         }
 802         finish_wait(&sbi->cp_wait, &wait);
 803 }
 804
 805 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 806 {
 807         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 808         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 809         nid_t last_nid = 0;
 810         block_t start_blk;
 811         struct page *cp_page;
 812         unsigned int data_sum_blocks, orphan_blocks;
 813         __u32 crc32 = 0;
 814         void *kaddr;
 815         int i;
 816         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 817
 818         /*
 819          * This avoids to conduct wrong roll-forward operations and uses
 820          * metapages, so should be called prior to sync_meta_pages below.
 821          */
 822         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 823
 824         /* Flush all the NAT/SIT pages */
 825         while (get_pages(sbi, F2FS_DIRTY_META)) {
 826                 sync_meta_pages(sbi, META, LONG_MAX);
 827                 if (unlikely(f2fs_cp_error(sbi)))
 828                         return;
 829         }
 830
 831         next_free_nid(sbi, &last_nid);
 832
 833         /*
 834          * modify checkpoint
 835          * version number is already updated
 836          */
 837         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 838         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 839         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 840         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 841                 ckpt->cur_node_segno[i] =
 842                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 843                 ckpt->cur_node_blkoff[i] =
 844                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 845                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 846                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 847         }
 848         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 849                 ckpt->cur_data_segno[i] =
 850                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 851                 ckpt->cur_data_blkoff[i] =
 852                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 853                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 854                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 855         }
 856
 857         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 858         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 859         ckpt->next_free_nid = cpu_to_le32(last_nid);
 860
 861         /* 2 cp  + n data seg summary + orphan inode blocks */
 862         data_sum_blocks = npages_for_summary_flush(sbi);
 863         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 864                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 865         else
 866                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 867
 868         orphan_blocks = GET_ORPHAN_BLOCKS(sbi->n_orphans);
 869         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 870                         orphan_blocks);
 871
 872         if (is_umount) {
 873                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 874                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 875                                 cp_payload_blks + data_sum_blocks +
 876                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 877         } else {
 878                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 879                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 880                                 cp_payload_blks + data_sum_blocks +
 881                                 orphan_blocks);
 882         }
 883
 884         if (sbi->n_orphans)
 885                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 886         else
 887                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 888
 889         /* update SIT/NAT bitmap */
 890         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 891         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 892
 893         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 894         *((__le32 *)((unsigned char *)ckpt +
 895                                 le32_to_cpu(ckpt->checksum_offset)))
 896                                 = cpu_to_le32(crc32);
 897
 898         start_blk = __start_cp_addr(sbi);
 899
 900         /* write out checkpoint buffer at block 0 */
 901         cp_page = grab_meta_page(sbi, start_blk++);
 902         kaddr = page_address(cp_page);
 903         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 904         set_page_dirty(cp_page);
 905         f2fs_put_page(cp_page, 1);
 906
 907         for (i = 1; i < 1 + cp_payload_blks; i++) {
 908                 cp_page = grab_meta_page(sbi, start_blk++);
 909                 kaddr = page_address(cp_page);
 910                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 911                                 (1 << sbi->log_blocksize));
 912                 set_page_dirty(cp_page);
 913                 f2fs_put_page(cp_page, 1);
 914         }
 915
 916         if (sbi->n_orphans) {
 917                 write_orphan_inodes(sbi, start_blk);
 918                 start_blk += orphan_blocks;
 919         }
 920
 921         write_data_summaries(sbi, start_blk);
 922         start_blk += data_sum_blocks;
 923         if (is_umount) {
 924                 write_node_summaries(sbi, start_blk);
 925                 start_blk += NR_CURSEG_NODE_TYPE;
 926         }
 927
 928         /* writeout checkpoint block */
 929         cp_page = grab_meta_page(sbi, start_blk);
 930         kaddr = page_address(cp_page);
 931         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 932         set_page_dirty(cp_page);
 933         f2fs_put_page(cp_page, 1);
 934
 935         /* wait for previous submitted node/meta pages writeback */
 936         wait_on_all_pages_writeback(sbi);
 937
 938         if (unlikely(f2fs_cp_error(sbi)))
 939                 return;
 940
 941         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 942         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 943
 944         /* update user_block_counts */
 945         sbi->last_valid_block_count = sbi->total_valid_block_count;
 946         sbi->alloc_valid_block_count = 0;
 947
 948         /* Here, we only have one bio having CP pack */
 949         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 950
 951         release_dirty_inode(sbi);
 952
 953         if (unlikely(f2fs_cp_error(sbi)))
 954                 return;
 955
 956         clear_prefree_segments(sbi);
 957         F2FS_RESET_SB_DIRT(sbi);
 958 }
 959
 960 /*
 961  * We guarantee that this checkpoint procedure will not fail.
 962  */
 963 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 964 {
 965         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 966         unsigned long long ckpt_ver;
 967
 968         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 969
 970         mutex_lock(&sbi->cp_mutex);
 971
 972         if (!sbi->s_dirty)
 973                 goto out;
 974         if (unlikely(f2fs_cp_error(sbi)))
 975                 goto out;
 976         if (block_operations(sbi))
 977                 goto out;
 978
 979         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 980
 981         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 982         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 983         f2fs_submit_merged_bio(sbi, META, WRITE);
 984
 985         /*
 986          * update checkpoint pack index
 987          * Increase the version number so that
 988          * SIT entries and seg summaries are written at correct place
 989          */
 990         ckpt_ver = cur_cp_version(ckpt);
 991         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 992
 993         /* write cached NAT/SIT entries to NAT/SIT area */
 994         flush_nat_entries(sbi);
 995         flush_sit_entries(sbi);
 996
 997         /* unlock all the fs_lock[] in do_checkpoint() */
 998         do_checkpoint(sbi, is_umount);
 999
1000         unblock_operations(sbi);
1001         stat_inc_cp_count(sbi->stat_info);
1002 out:
1003         mutex_unlock(&sbi->cp_mutex);
1004         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
1005 }
1006
1007 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1008 {
1009         int i;
1010
1011         for (i = 0; i < MAX_INO_ENTRY; i++) {
1012                 INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
1013                 spin_lock_init(&sbi->ino_lock[i]);
1014                 INIT_LIST_HEAD(&sbi->ino_list[i]);
1015         }
1016
1017         /*
1018          * considering 512 blocks in a segment 8 blocks are needed for cp
1019          * and log segment summaries. Remaining blocks are used to keep
1020          * orphan entries with the limitation one reserved segment
1021          * for cp pack we can have max 1020*504 orphan entries
1022          */
1023         sbi->n_orphans = 0;
1024         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1025                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1026 }
1027
1028 int __init create_checkpoint_caches(void)
1029 {
1030         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1031                         sizeof(struct ino_entry));
1032         if (!ino_entry_slab)
1033                 return -ENOMEM;
1034         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1035                         sizeof(struct dir_inode_entry));
1036         if (!inode_entry_slab) {
1037                 kmem_cache_destroy(ino_entry_slab);
1038                 return -ENOMEM;
1039         }
1040         return 0;
1041 }
1042
1043 void destroy_checkpoint_caches(void)
1044 {
1045         kmem_cache_destroy(ino_entry_slab);
1046         kmem_cache_destroy(inode_entry_slab);
1047 }