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.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *inode_entry_slab;
  28
  29 /*
  30  * We guarantee no failure on the returned page.
  31  */
  32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  33 {
  34         struct address_space *mapping = META_MAPPING(sbi);
  35         struct page *page = NULL;
  36 repeat:
  37         page = grab_cache_page(mapping, index);
  38         if (!page) {
  39                 cond_resched();
  40                 goto repeat;
  41         }
  42         f2fs_wait_on_page_writeback(page, META);
  43         SetPageUptodate(page);
  44         return page;
  45 }
  46
  47 /*
  48  * We guarantee no failure on the returned page.
  49  */
  50 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  51 {
  52         struct address_space *mapping = META_MAPPING(sbi);
  53         struct page *page;
  54         struct f2fs_io_info fio = {
  55                 .type = META,
  56                 .rw = READ_SYNC | REQ_META | REQ_PRIO,
  57                 .blk_addr = index,
  58         };
  59 repeat:
  60         page = grab_cache_page(mapping, index);
  61         if (!page) {
  62                 cond_resched();
  63                 goto repeat;
  64         }
  65         if (PageUptodate(page))
  66                 goto out;
  67
  68         if (f2fs_submit_page_bio(sbi, page, &fio))
  69                 goto repeat;
  70
  71         lock_page(page);
  72         if (unlikely(page->mapping != mapping)) {
  73                 f2fs_put_page(page, 1);
  74                 goto repeat;
  75         }
  76 out:
  77         return page;
  78 }
  79
  80 static inline bool is_valid_blkaddr(struct f2fs_sb_info *sbi,
  81                                                 block_t blkaddr, int type)
  82 {
  83         switch (type) {
  84         case META_NAT:
  85                 break;
  86         case META_SIT:
  87                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
  88                         return false;
  89                 break;
  90         case META_SSA:
  91                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
  92                         blkaddr < SM_I(sbi)->ssa_blkaddr))
  93                         return false;
  94                 break;
  95         case META_CP:
  96                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
  97                         blkaddr < __start_cp_addr(sbi)))
  98                         return false;
  99                 break;
 100         case META_POR:
 101                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 102                         blkaddr < MAIN_BLKADDR(sbi)))
 103                         return false;
 104                 break;
 105         default:
 106                 BUG();
 107         }
 108
 109         return true;
 110 }
 111
 112 /*
 113  * Readahead CP/NAT/SIT/SSA pages
 114  */
 115 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
 116 {
 117         block_t prev_blk_addr = 0;
 118         struct page *page;
 119         block_t blkno = start;
 120         struct f2fs_io_info fio = {
 121                 .type = META,
 122                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 123         };
 124
 125         for (; nrpages-- > 0; blkno++) {
 126
 127                 if (!is_valid_blkaddr(sbi, blkno, type))
 128                         goto out;
 129
 130                 switch (type) {
 131                 case META_NAT:
 132                         if (unlikely(blkno >=
 133                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 134                                 blkno = 0;
 135                         /* get nat block addr */
 136                         fio.blk_addr = current_nat_addr(sbi,
 137                                         blkno * NAT_ENTRY_PER_BLOCK);
 138                         break;
 139                 case META_SIT:
 140                         /* get sit block addr */
 141                         fio.blk_addr = current_sit_addr(sbi,
 142                                         blkno * SIT_ENTRY_PER_BLOCK);
 143                         if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
 144                                 goto out;
 145                         prev_blk_addr = fio.blk_addr;
 146                         break;
 147                 case META_SSA:
 148                 case META_CP:
 149                 case META_POR:
 150                         fio.blk_addr = blkno;
 151                         break;
 152                 default:
 153                         BUG();
 154                 }
 155
 156                 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
 157                 if (!page)
 158                         continue;
 159                 if (PageUptodate(page)) {
 160                         f2fs_put_page(page, 1);
 161                         continue;
 162                 }
 163
 164                 f2fs_submit_page_mbio(sbi, page, &fio);
 165                 f2fs_put_page(page, 0);
 166         }
 167 out:
 168         f2fs_submit_merged_bio(sbi, META, READ);
 169         return blkno - start;
 170 }
 171
 172 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 173 {
 174         struct page *page;
 175         bool readahead = false;
 176
 177         page = find_get_page(META_MAPPING(sbi), index);
 178         if (!page || (page && !PageUptodate(page)))
 179                 readahead = true;
 180         f2fs_put_page(page, 0);
 181
 182         if (readahead)
 183                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
 184 }
 185
 186 static int f2fs_write_meta_page(struct page *page,
 187                                 struct writeback_control *wbc)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 190
 191         trace_f2fs_writepage(page, META);
 192
 193         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 194                 goto redirty_out;
 195         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 196                 goto redirty_out;
 197         if (unlikely(f2fs_cp_error(sbi)))
 198                 goto redirty_out;
 199
 200         f2fs_wait_on_page_writeback(page, META);
 201         write_meta_page(sbi, page);
 202         dec_page_count(sbi, F2FS_DIRTY_META);
 203         unlock_page(page);
 204
 205         if (wbc->for_reclaim)
 206                 f2fs_submit_merged_bio(sbi, META, WRITE);
 207         return 0;
 208
 209 redirty_out:
 210         redirty_page_for_writepage(wbc, page);
 211         return AOP_WRITEPAGE_ACTIVATE;
 212 }
 213
 214 static int f2fs_write_meta_pages(struct address_space *mapping,
 215                                 struct writeback_control *wbc)
 216 {
 217         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 218         long diff, written;
 219
 220         trace_f2fs_writepages(mapping->host, wbc, META);
 221
 222         /* collect a number of dirty meta pages and write together */
 223         if (wbc->for_kupdate ||
 224                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 225                 goto skip_write;
 226
 227         /* if mounting is failed, skip writing node pages */
 228         mutex_lock(&sbi->cp_mutex);
 229         diff = nr_pages_to_write(sbi, META, wbc);
 230         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 231         mutex_unlock(&sbi->cp_mutex);
 232         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 233         return 0;
 234
 235 skip_write:
 236         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 237         return 0;
 238 }
 239
 240 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 241                                                 long nr_to_write)
 242 {
 243         struct address_space *mapping = META_MAPPING(sbi);
 244         pgoff_t index = 0, end = LONG_MAX;
 245         struct pagevec pvec;
 246         long nwritten = 0;
 247         struct writeback_control wbc = {
 248                 .for_reclaim = 0,
 249         };
 250
 251         pagevec_init(&pvec, 0);
 252
 253         while (index <= end) {
 254                 int i, nr_pages;
 255                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 256                                 PAGECACHE_TAG_DIRTY,
 257                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 258                 if (unlikely(nr_pages == 0))
 259                         break;
 260
 261                 for (i = 0; i < nr_pages; i++) {
 262                         struct page *page = pvec.pages[i];
 263
 264                         lock_page(page);
 265
 266                         if (unlikely(page->mapping != mapping)) {
 267 continue_unlock:
 268                                 unlock_page(page);
 269                                 continue;
 270                         }
 271                         if (!PageDirty(page)) {
 272                                 /* someone wrote it for us */
 273                                 goto continue_unlock;
 274                         }
 275
 276                         if (!clear_page_dirty_for_io(page))
 277                                 goto continue_unlock;
 278
 279                         if (f2fs_write_meta_page(page, &wbc)) {
 280                                 unlock_page(page);
 281                                 break;
 282                         }
 283                         nwritten++;
 284                         if (unlikely(nwritten >= nr_to_write))
 285                                 break;
 286                 }
 287                 pagevec_release(&pvec);
 288                 cond_resched();
 289         }
 290
 291         if (nwritten)
 292                 f2fs_submit_merged_bio(sbi, type, WRITE);
 293
 294         return nwritten;
 295 }
 296
 297 static int f2fs_set_meta_page_dirty(struct page *page)
 298 {
 299         trace_f2fs_set_page_dirty(page, META);
 300
 301         SetPageUptodate(page);
 302         if (!PageDirty(page)) {
 303                 __set_page_dirty_nobuffers(page);
 304                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 305                 SetPagePrivate(page);
 306                 f2fs_trace_pid(page);
 307                 return 1;
 308         }
 309         return 0;
 310 }
 311
 312 static void f2fs_invalidate_meta_page(struct page *page, unsigned int offset,
 313                                       unsigned int length)
 314 {
 315         struct inode *inode = page->mapping->host;
 316
 317         if (PageDirty(page))
 318                 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_META);
 319         ClearPagePrivate(page);
 320 }
 321
 322 static int f2fs_release_meta_page(struct page *page, gfp_t wait)
 323 {
 324         /* If this is dirty page, keep PagePrivate */
 325         if (PageDirty(page))
 326                 return 0;
 327
 328         ClearPagePrivate(page);
 329         return 1;
 330 }
 331
 332 const struct address_space_operations f2fs_meta_aops = {
 333         .writepage      = f2fs_write_meta_page,
 334         .writepages     = f2fs_write_meta_pages,
 335         .set_page_dirty = f2fs_set_meta_page_dirty,
 336         .invalidatepage = f2fs_invalidate_meta_page,
 337         .releasepage    = f2fs_release_meta_page,
 338 };
 339
 340 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 341 {
 342         struct inode_management *im = &sbi->im[type];
 343         struct ino_entry *e;
 344 retry:
 345         if (radix_tree_preload(GFP_NOFS)) {
 346                 cond_resched();
 347                 goto retry;
 348         }
 349
 350         spin_lock(&im->ino_lock);
 351
 352         e = radix_tree_lookup(&im->ino_root, ino);
 353         if (!e) {
 354                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 355                 if (!e) {
 356                         spin_unlock(&im->ino_lock);
 357                         radix_tree_preload_end();
 358                         goto retry;
 359                 }
 360                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 361                         spin_unlock(&im->ino_lock);
 362                         kmem_cache_free(ino_entry_slab, e);
 363                         radix_tree_preload_end();
 364                         goto retry;
 365                 }
 366                 memset(e, 0, sizeof(struct ino_entry));
 367                 e->ino = ino;
 368
 369                 list_add_tail(&e->list, &im->ino_list);
 370                 if (type != ORPHAN_INO)
 371                         im->ino_num++;
 372         }
 373         spin_unlock(&im->ino_lock);
 374         radix_tree_preload_end();
 375 }
 376
 377 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 378 {
 379         struct inode_management *im = &sbi->im[type];
 380         struct ino_entry *e;
 381
 382         spin_lock(&im->ino_lock);
 383         e = radix_tree_lookup(&im->ino_root, ino);
 384         if (e) {
 385                 list_del(&e->list);
 386                 radix_tree_delete(&im->ino_root, ino);
 387                 im->ino_num--;
 388                 spin_unlock(&im->ino_lock);
 389                 kmem_cache_free(ino_entry_slab, e);
 390                 return;
 391         }
 392         spin_unlock(&im->ino_lock);
 393 }
 394
 395 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 396 {
 397         /* add new dirty ino entry into list */
 398         __add_ino_entry(sbi, ino, type);
 399 }
 400
 401 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 402 {
 403         /* remove dirty ino entry from list */
 404         __remove_ino_entry(sbi, ino, type);
 405 }
 406
 407 /* mode should be APPEND_INO or UPDATE_INO */
 408 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 409 {
 410         struct inode_management *im = &sbi->im[mode];
 411         struct ino_entry *e;
 412
 413         spin_lock(&im->ino_lock);
 414         e = radix_tree_lookup(&im->ino_root, ino);
 415         spin_unlock(&im->ino_lock);
 416         return e ? true : false;
 417 }
 418
 419 void release_dirty_inode(struct f2fs_sb_info *sbi)
 420 {
 421         struct ino_entry *e, *tmp;
 422         int i;
 423
 424         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 425                 struct inode_management *im = &sbi->im[i];
 426
 427                 spin_lock(&im->ino_lock);
 428                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 429                         list_del(&e->list);
 430                         radix_tree_delete(&im->ino_root, e->ino);
 431                         kmem_cache_free(ino_entry_slab, e);
 432                         im->ino_num--;
 433                 }
 434                 spin_unlock(&im->ino_lock);
 435         }
 436 }
 437
 438 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 439 {
 440         struct inode_management *im = &sbi->im[ORPHAN_INO];
 441         int err = 0;
 442
 443         spin_lock(&im->ino_lock);
 444         if (unlikely(im->ino_num >= sbi->max_orphans))
 445                 err = -ENOSPC;
 446         else
 447                 im->ino_num++;
 448         spin_unlock(&im->ino_lock);
 449
 450         return err;
 451 }
 452
 453 void release_orphan_inode(struct f2fs_sb_info *sbi)
 454 {
 455         struct inode_management *im = &sbi->im[ORPHAN_INO];
 456
 457         spin_lock(&im->ino_lock);
 458         f2fs_bug_on(sbi, im->ino_num == 0);
 459         im->ino_num--;
 460         spin_unlock(&im->ino_lock);
 461 }
 462
 463 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 464 {
 465         /* add new orphan ino entry into list */
 466         __add_ino_entry(sbi, ino, ORPHAN_INO);
 467 }
 468
 469 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 470 {
 471         /* remove orphan entry from orphan list */
 472         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 473 }
 474
 475 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 476 {
 477         struct inode *inode = f2fs_iget(sbi->sb, ino);
 478         f2fs_bug_on(sbi, IS_ERR(inode));
 479         clear_nlink(inode);
 480
 481         /* truncate all the data during iput */
 482         iput(inode);
 483 }
 484
 485 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 486 {
 487         block_t start_blk, orphan_blkaddr, i, j;
 488
 489         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 490                 return;
 491
 492         set_sbi_flag(sbi, SBI_POR_DOING);
 493
 494         start_blk = __start_cp_addr(sbi) + 1 +
 495                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 496         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 497
 498         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 499
 500         for (i = 0; i < orphan_blkaddr; i++) {
 501                 struct page *page = get_meta_page(sbi, start_blk + i);
 502                 struct f2fs_orphan_block *orphan_blk;
 503
 504                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 505                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 506                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 507                         recover_orphan_inode(sbi, ino);
 508                 }
 509                 f2fs_put_page(page, 1);
 510         }
 511         /* clear Orphan Flag */
 512         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 513         clear_sbi_flag(sbi, SBI_POR_DOING);
 514         return;
 515 }
 516
 517 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 518 {
 519         struct list_head *head;
 520         struct f2fs_orphan_block *orphan_blk = NULL;
 521         unsigned int nentries = 0;
 522         unsigned short index;
 523         unsigned short orphan_blocks;
 524         struct page *page = NULL;
 525         struct ino_entry *orphan = NULL;
 526         struct inode_management *im = &sbi->im[ORPHAN_INO];
 527
 528         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 529
 530         for (index = 0; index < orphan_blocks; index++)
 531                 grab_meta_page(sbi, start_blk + index);
 532
 533         index = 1;
 534         spin_lock(&im->ino_lock);
 535         head = &im->ino_list;
 536
 537         /* loop for each orphan inode entry and write them in Jornal block */
 538         list_for_each_entry(orphan, head, list) {
 539                 if (!page) {
 540                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 541                         f2fs_bug_on(sbi, !page);
 542                         orphan_blk =
 543                                 (struct f2fs_orphan_block *)page_address(page);
 544                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 545                         f2fs_put_page(page, 0);
 546                 }
 547
 548                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 549
 550                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 551                         /*
 552                          * an orphan block is full of 1020 entries,
 553                          * then we need to flush current orphan blocks
 554                          * and bring another one in memory
 555                          */
 556                         orphan_blk->blk_addr = cpu_to_le16(index);
 557                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 558                         orphan_blk->entry_count = cpu_to_le32(nentries);
 559                         set_page_dirty(page);
 560                         f2fs_put_page(page, 1);
 561                         index++;
 562                         nentries = 0;
 563                         page = NULL;
 564                 }
 565         }
 566
 567         if (page) {
 568                 orphan_blk->blk_addr = cpu_to_le16(index);
 569                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 570                 orphan_blk->entry_count = cpu_to_le32(nentries);
 571                 set_page_dirty(page);
 572                 f2fs_put_page(page, 1);
 573         }
 574
 575         spin_unlock(&im->ino_lock);
 576 }
 577
 578 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 579                                 block_t cp_addr, unsigned long long *version)
 580 {
 581         struct page *cp_page_1, *cp_page_2 = NULL;
 582         unsigned long blk_size = sbi->blocksize;
 583         struct f2fs_checkpoint *cp_block;
 584         unsigned long long cur_version = 0, pre_version = 0;
 585         size_t crc_offset;
 586         __u32 crc = 0;
 587
 588         /* Read the 1st cp block in this CP pack */
 589         cp_page_1 = get_meta_page(sbi, cp_addr);
 590
 591         /* get the version number */
 592         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 593         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 594         if (crc_offset >= blk_size)
 595                 goto invalid_cp1;
 596
 597         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 598         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 599                 goto invalid_cp1;
 600
 601         pre_version = cur_cp_version(cp_block);
 602
 603         /* Read the 2nd cp block in this CP pack */
 604         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 605         cp_page_2 = get_meta_page(sbi, cp_addr);
 606
 607         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 608         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 609         if (crc_offset >= blk_size)
 610                 goto invalid_cp2;
 611
 612         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 613         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 614                 goto invalid_cp2;
 615
 616         cur_version = cur_cp_version(cp_block);
 617
 618         if (cur_version == pre_version) {
 619                 *version = cur_version;
 620                 f2fs_put_page(cp_page_2, 1);
 621                 return cp_page_1;
 622         }
 623 invalid_cp2:
 624         f2fs_put_page(cp_page_2, 1);
 625 invalid_cp1:
 626         f2fs_put_page(cp_page_1, 1);
 627         return NULL;
 628 }
 629
 630 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 631 {
 632         struct f2fs_checkpoint *cp_block;
 633         struct f2fs_super_block *fsb = sbi->raw_super;
 634         struct page *cp1, *cp2, *cur_page;
 635         unsigned long blk_size = sbi->blocksize;
 636         unsigned long long cp1_version = 0, cp2_version = 0;
 637         unsigned long long cp_start_blk_no;
 638         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 639         block_t cp_blk_no;
 640         int i;
 641
 642         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 643         if (!sbi->ckpt)
 644                 return -ENOMEM;
 645         /*
 646          * Finding out valid cp block involves read both
 647          * sets( cp pack1 and cp pack 2)
 648          */
 649         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 650         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 651
 652         /* The second checkpoint pack should start at the next segment */
 653         cp_start_blk_no += ((unsigned long long)1) <<
 654                                 le32_to_cpu(fsb->log_blocks_per_seg);
 655         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 656
 657         if (cp1 && cp2) {
 658                 if (ver_after(cp2_version, cp1_version))
 659                         cur_page = cp2;
 660                 else
 661                         cur_page = cp1;
 662         } else if (cp1) {
 663                 cur_page = cp1;
 664         } else if (cp2) {
 665                 cur_page = cp2;
 666         } else {
 667                 goto fail_no_cp;
 668         }
 669
 670         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 671         memcpy(sbi->ckpt, cp_block, blk_size);
 672
 673         if (cp_blks <= 1)
 674                 goto done;
 675
 676         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 677         if (cur_page == cp2)
 678                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 679
 680         for (i = 1; i < cp_blks; i++) {
 681                 void *sit_bitmap_ptr;
 682                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 683
 684                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 685                 sit_bitmap_ptr = page_address(cur_page);
 686                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 687                 f2fs_put_page(cur_page, 1);
 688         }
 689 done:
 690         f2fs_put_page(cp1, 1);
 691         f2fs_put_page(cp2, 1);
 692         return 0;
 693
 694 fail_no_cp:
 695         kfree(sbi->ckpt);
 696         return -EINVAL;
 697 }
 698
 699 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
 700 {
 701         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 702
 703         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 704                 return -EEXIST;
 705
 706         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 707         F2FS_I(inode)->dirty_dir = new;
 708         list_add_tail(&new->list, &sbi->dir_inode_list);
 709         stat_inc_dirty_dir(sbi);
 710         return 0;
 711 }
 712
 713 void update_dirty_page(struct inode *inode, struct page *page)
 714 {
 715         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 716         struct inode_entry *new;
 717         int ret = 0;
 718
 719         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
 720                 return;
 721
 722         if (!S_ISDIR(inode->i_mode)) {
 723                 inode_inc_dirty_pages(inode);
 724                 goto out;
 725         }
 726
 727         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 728         new->inode = inode;
 729         INIT_LIST_HEAD(&new->list);
 730
 731         spin_lock(&sbi->dir_inode_lock);
 732         ret = __add_dirty_inode(inode, new);
 733         inode_inc_dirty_pages(inode);
 734         spin_unlock(&sbi->dir_inode_lock);
 735
 736         if (ret)
 737                 kmem_cache_free(inode_entry_slab, new);
 738 out:
 739         SetPagePrivate(page);
 740         f2fs_trace_pid(page);
 741 }
 742
 743 void add_dirty_dir_inode(struct inode *inode)
 744 {
 745         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 746         struct inode_entry *new =
 747                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 748         int ret = 0;
 749
 750         new->inode = inode;
 751         INIT_LIST_HEAD(&new->list);
 752
 753         spin_lock(&sbi->dir_inode_lock);
 754         ret = __add_dirty_inode(inode, new);
 755         spin_unlock(&sbi->dir_inode_lock);
 756
 757         if (ret)
 758                 kmem_cache_free(inode_entry_slab, new);
 759 }
 760
 761 void remove_dirty_dir_inode(struct inode *inode)
 762 {
 763         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 764         struct inode_entry *entry;
 765
 766         if (!S_ISDIR(inode->i_mode))
 767                 return;
 768
 769         spin_lock(&sbi->dir_inode_lock);
 770         if (get_dirty_pages(inode) ||
 771                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 772                 spin_unlock(&sbi->dir_inode_lock);
 773                 return;
 774         }
 775
 776         entry = F2FS_I(inode)->dirty_dir;
 777         list_del(&entry->list);
 778         F2FS_I(inode)->dirty_dir = NULL;
 779         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 780         stat_dec_dirty_dir(sbi);
 781         spin_unlock(&sbi->dir_inode_lock);
 782         kmem_cache_free(inode_entry_slab, entry);
 783
 784         /* Only from the recovery routine */
 785         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 786                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 787                 iput(inode);
 788         }
 789 }
 790
 791 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 792 {
 793         struct list_head *head;
 794         struct inode_entry *entry;
 795         struct inode *inode;
 796 retry:
 797         if (unlikely(f2fs_cp_error(sbi)))
 798                 return;
 799
 800         spin_lock(&sbi->dir_inode_lock);
 801
 802         head = &sbi->dir_inode_list;
 803         if (list_empty(head)) {
 804                 spin_unlock(&sbi->dir_inode_lock);
 805                 return;
 806         }
 807         entry = list_entry(head->next, struct inode_entry, list);
 808         inode = igrab(entry->inode);
 809         spin_unlock(&sbi->dir_inode_lock);
 810         if (inode) {
 811                 filemap_fdatawrite(inode->i_mapping);
 812                 iput(inode);
 813         } else {
 814                 /*
 815                  * We should submit bio, since it exists several
 816                  * wribacking dentry pages in the freeing inode.
 817                  */
 818                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 819         }
 820         goto retry;
 821 }
 822
 823 /*
 824  * Freeze all the FS-operations for checkpoint.
 825  */
 826 static int block_operations(struct f2fs_sb_info *sbi)
 827 {
 828         struct writeback_control wbc = {
 829                 .sync_mode = WB_SYNC_ALL,
 830                 .nr_to_write = LONG_MAX,
 831                 .for_reclaim = 0,
 832         };
 833         struct blk_plug plug;
 834         int err = 0;
 835
 836         blk_start_plug(&plug);
 837
 838 retry_flush_dents:
 839         f2fs_lock_all(sbi);
 840         /* write all the dirty dentry pages */
 841         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 842                 f2fs_unlock_all(sbi);
 843                 sync_dirty_dir_inodes(sbi);
 844                 if (unlikely(f2fs_cp_error(sbi))) {
 845                         err = -EIO;
 846                         goto out;
 847                 }
 848                 goto retry_flush_dents;
 849         }
 850
 851         /*
 852          * POR: we should ensure that there are no dirty node pages
 853          * until finishing nat/sit flush.
 854          */
 855 retry_flush_nodes:
 856         down_write(&sbi->node_write);
 857
 858         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 859                 up_write(&sbi->node_write);
 860                 sync_node_pages(sbi, 0, &wbc);
 861                 if (unlikely(f2fs_cp_error(sbi))) {
 862                         f2fs_unlock_all(sbi);
 863                         err = -EIO;
 864                         goto out;
 865                 }
 866                 goto retry_flush_nodes;
 867         }
 868 out:
 869         blk_finish_plug(&plug);
 870         return err;
 871 }
 872
 873 static void unblock_operations(struct f2fs_sb_info *sbi)
 874 {
 875         up_write(&sbi->node_write);
 876         f2fs_unlock_all(sbi);
 877 }
 878
 879 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 880 {
 881         DEFINE_WAIT(wait);
 882
 883         for (;;) {
 884                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 885
 886                 if (!get_pages(sbi, F2FS_WRITEBACK))
 887                         break;
 888
 889                 io_schedule();
 890         }
 891         finish_wait(&sbi->cp_wait, &wait);
 892 }
 893
 894 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 895 {
 896         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 897         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 898         struct f2fs_nm_info *nm_i = NM_I(sbi);
 899         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 900         nid_t last_nid = nm_i->next_scan_nid;
 901         block_t start_blk;
 902         struct page *cp_page;
 903         unsigned int data_sum_blocks, orphan_blocks;
 904         __u32 crc32 = 0;
 905         void *kaddr;
 906         int i;
 907         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 908
 909         /*
 910          * This avoids to conduct wrong roll-forward operations and uses
 911          * metapages, so should be called prior to sync_meta_pages below.
 912          */
 913         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 914
 915         /* Flush all the NAT/SIT pages */
 916         while (get_pages(sbi, F2FS_DIRTY_META)) {
 917                 sync_meta_pages(sbi, META, LONG_MAX);
 918                 if (unlikely(f2fs_cp_error(sbi)))
 919                         return;
 920         }
 921
 922         next_free_nid(sbi, &last_nid);
 923
 924         /*
 925          * modify checkpoint
 926          * version number is already updated
 927          */
 928         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 929         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 930         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 931         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 932                 ckpt->cur_node_segno[i] =
 933                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 934                 ckpt->cur_node_blkoff[i] =
 935                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 936                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 937                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 938         }
 939         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 940                 ckpt->cur_data_segno[i] =
 941                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 942                 ckpt->cur_data_blkoff[i] =
 943                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 944                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 945                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 946         }
 947
 948         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 949         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 950         ckpt->next_free_nid = cpu_to_le32(last_nid);
 951
 952         /* 2 cp  + n data seg summary + orphan inode blocks */
 953         data_sum_blocks = npages_for_summary_flush(sbi, false);
 954         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 955                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 956         else
 957                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 958
 959         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 960         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 961                         orphan_blocks);
 962
 963         if (__remain_node_summaries(cpc->reason))
 964                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 965                                 cp_payload_blks + data_sum_blocks +
 966                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 967         else
 968                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 969                                 cp_payload_blks + data_sum_blocks +
 970                                 orphan_blocks);
 971
 972         if (cpc->reason == CP_UMOUNT)
 973                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 974         else
 975                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 976
 977         if (cpc->reason == CP_FASTBOOT)
 978                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
 979         else
 980                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
 981
 982         if (orphan_num)
 983                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 984         else
 985                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 986
 987         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
 988                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
 989
 990         /* update SIT/NAT bitmap */
 991         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 992         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 993
 994         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 995         *((__le32 *)((unsigned char *)ckpt +
 996                                 le32_to_cpu(ckpt->checksum_offset)))
 997                                 = cpu_to_le32(crc32);
 998
 999         start_blk = __start_cp_addr(sbi);
1000
1001         /* write out checkpoint buffer at block 0 */
1002         cp_page = grab_meta_page(sbi, start_blk++);
1003         kaddr = page_address(cp_page);
1004         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
1005         set_page_dirty(cp_page);
1006         f2fs_put_page(cp_page, 1);
1007
1008         for (i = 1; i < 1 + cp_payload_blks; i++) {
1009                 cp_page = grab_meta_page(sbi, start_blk++);
1010                 kaddr = page_address(cp_page);
1011                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
1012                                 (1 << sbi->log_blocksize));
1013                 set_page_dirty(cp_page);
1014                 f2fs_put_page(cp_page, 1);
1015         }
1016
1017         if (orphan_num) {
1018                 write_orphan_inodes(sbi, start_blk);
1019                 start_blk += orphan_blocks;
1020         }
1021
1022         write_data_summaries(sbi, start_blk);
1023         start_blk += data_sum_blocks;
1024         if (__remain_node_summaries(cpc->reason)) {
1025                 write_node_summaries(sbi, start_blk);
1026                 start_blk += NR_CURSEG_NODE_TYPE;
1027         }
1028
1029         /* writeout checkpoint block */
1030         cp_page = grab_meta_page(sbi, start_blk);
1031         kaddr = page_address(cp_page);
1032         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
1033         set_page_dirty(cp_page);
1034         f2fs_put_page(cp_page, 1);
1035
1036         /* wait for previous submitted node/meta pages writeback */
1037         wait_on_all_pages_writeback(sbi);
1038
1039         if (unlikely(f2fs_cp_error(sbi)))
1040                 return;
1041
1042         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1043         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1044
1045         /* update user_block_counts */
1046         sbi->last_valid_block_count = sbi->total_valid_block_count;
1047         sbi->alloc_valid_block_count = 0;
1048
1049         /* Here, we only have one bio having CP pack */
1050         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1051
1052         /* wait for previous submitted meta pages writeback */
1053         wait_on_all_pages_writeback(sbi);
1054
1055         release_dirty_inode(sbi);
1056
1057         if (unlikely(f2fs_cp_error(sbi)))
1058                 return;
1059
1060         clear_prefree_segments(sbi);
1061         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1062 }
1063
1064 /*
1065  * We guarantee that this checkpoint procedure will not fail.
1066  */
1067 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1068 {
1069         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1070         unsigned long long ckpt_ver;
1071
1072         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1073
1074         mutex_lock(&sbi->cp_mutex);
1075
1076         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1077                         cpc->reason != CP_DISCARD && cpc->reason != CP_UMOUNT)
1078                 goto out;
1079         if (unlikely(f2fs_cp_error(sbi)))
1080                 goto out;
1081         if (f2fs_readonly(sbi->sb))
1082                 goto out;
1083         if (block_operations(sbi))
1084                 goto out;
1085
1086         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1087
1088         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1089         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1090         f2fs_submit_merged_bio(sbi, META, WRITE);
1091
1092         /*
1093          * update checkpoint pack index
1094          * Increase the version number so that
1095          * SIT entries and seg summaries are written at correct place
1096          */
1097         ckpt_ver = cur_cp_version(ckpt);
1098         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1099
1100         /* write cached NAT/SIT entries to NAT/SIT area */
1101         flush_nat_entries(sbi);
1102         flush_sit_entries(sbi, cpc);
1103
1104         /* unlock all the fs_lock[] in do_checkpoint() */
1105         do_checkpoint(sbi, cpc);
1106
1107         unblock_operations(sbi);
1108         stat_inc_cp_count(sbi->stat_info);
1109 out:
1110         mutex_unlock(&sbi->cp_mutex);
1111         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1112 }
1113
1114 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1115 {
1116         int i;
1117
1118         for (i = 0; i < MAX_INO_ENTRY; i++) {
1119                 struct inode_management *im = &sbi->im[i];
1120
1121                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1122                 spin_lock_init(&im->ino_lock);
1123                 INIT_LIST_HEAD(&im->ino_list);
1124                 im->ino_num = 0;
1125         }
1126
1127         /*
1128          * considering 512 blocks in a segment 8 blocks are needed for cp
1129          * and log segment summaries. Remaining blocks are used to keep
1130          * orphan entries with the limitation one reserved segment
1131          * for cp pack we can have max 1020*504 orphan entries
1132          */
1133         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1134                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1135 }
1136
1137 int __init create_checkpoint_caches(void)
1138 {
1139         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1140                         sizeof(struct ino_entry));
1141         if (!ino_entry_slab)
1142                 return -ENOMEM;
1143         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1144                         sizeof(struct inode_entry));
1145         if (!inode_entry_slab) {
1146                 kmem_cache_destroy(ino_entry_slab);
1147                 return -ENOMEM;
1148         }
1149         return 0;
1150 }
1151
1152 void destroy_checkpoint_caches(void)
1153 {
1154         kmem_cache_destroy(ino_entry_slab);
1155         kmem_cache_destroy(inode_entry_slab);
1156 }