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 = f2fs_grab_cache_page(mapping, index, false);
  38         if (!page) {
  39                 cond_resched();
  40                 goto repeat;
  41         }
  42         f2fs_wait_on_page_writeback(page, META, true);
  43         SetPageUptodate(page);
  44         return page;
  45 }
  46
  47 /*
  48  * We guarantee no failure on the returned page.
  49  */
  50 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  51                                                         bool is_meta)
  52 {
  53         struct address_space *mapping = META_MAPPING(sbi);
  54         struct page *page;
  55         struct f2fs_io_info fio = {
  56                 .sbi = sbi,
  57                 .type = META,
  58                 .rw = READ_SYNC | REQ_META | REQ_PRIO,
  59                 .old_blkaddr = index,
  60                 .new_blkaddr = index,
  61                 .encrypted_page = NULL,
  62         };
  63
  64         if (unlikely(!is_meta))
  65                 fio.rw &= ~REQ_META;
  66 repeat:
  67         page = f2fs_grab_cache_page(mapping, index, false);
  68         if (!page) {
  69                 cond_resched();
  70                 goto repeat;
  71         }
  72         if (PageUptodate(page))
  73                 goto out;
  74
  75         fio.page = page;
  76
  77         if (f2fs_submit_page_bio(&fio)) {
  78                 f2fs_put_page(page, 1);
  79                 goto repeat;
  80         }
  81
  82         lock_page(page);
  83         if (unlikely(page->mapping != mapping)) {
  84                 f2fs_put_page(page, 1);
  85                 goto repeat;
  86         }
  87
  88         /*
  89          * if there is any IO error when accessing device, make our filesystem
  90          * readonly and make sure do not write checkpoint with non-uptodate
  91          * meta page.
  92          */
  93         if (unlikely(!PageUptodate(page)))
  94                 f2fs_stop_checkpoint(sbi);
  95 out:
  96         return page;
  97 }
  98
  99 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 100 {
 101         return __get_meta_page(sbi, index, true);
 102 }
 103
 104 /* for POR only */
 105 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 106 {
 107         return __get_meta_page(sbi, index, false);
 108 }
 109
 110 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 111 {
 112         switch (type) {
 113         case META_NAT:
 114                 break;
 115         case META_SIT:
 116                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 117                         return false;
 118                 break;
 119         case META_SSA:
 120                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 121                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 122                         return false;
 123                 break;
 124         case META_CP:
 125                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 126                         blkaddr < __start_cp_addr(sbi)))
 127                         return false;
 128                 break;
 129         case META_POR:
 130                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 131                         blkaddr < MAIN_BLKADDR(sbi)))
 132                         return false;
 133                 break;
 134         default:
 135                 BUG();
 136         }
 137
 138         return true;
 139 }
 140
 141 /*
 142  * Readahead CP/NAT/SIT/SSA pages
 143  */
 144 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 145                                                         int type, bool sync)
 146 {
 147         struct page *page;
 148         block_t blkno = start;
 149         struct f2fs_io_info fio = {
 150                 .sbi = sbi,
 151                 .type = META,
 152                 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
 153                 .encrypted_page = NULL,
 154         };
 155         struct blk_plug plug;
 156
 157         if (unlikely(type == META_POR))
 158                 fio.rw &= ~REQ_META;
 159
 160         blk_start_plug(&plug);
 161         for (; nrpages-- > 0; blkno++) {
 162
 163                 if (!is_valid_blkaddr(sbi, blkno, type))
 164                         goto out;
 165
 166                 switch (type) {
 167                 case META_NAT:
 168                         if (unlikely(blkno >=
 169                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 170                                 blkno = 0;
 171                         /* get nat block addr */
 172                         fio.new_blkaddr = current_nat_addr(sbi,
 173                                         blkno * NAT_ENTRY_PER_BLOCK);
 174                         break;
 175                 case META_SIT:
 176                         /* get sit block addr */
 177                         fio.new_blkaddr = current_sit_addr(sbi,
 178                                         blkno * SIT_ENTRY_PER_BLOCK);
 179                         break;
 180                 case META_SSA:
 181                 case META_CP:
 182                 case META_POR:
 183                         fio.new_blkaddr = blkno;
 184                         break;
 185                 default:
 186                         BUG();
 187                 }
 188
 189                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 190                                                 fio.new_blkaddr, false);
 191                 if (!page)
 192                         continue;
 193                 if (PageUptodate(page)) {
 194                         f2fs_put_page(page, 1);
 195                         continue;
 196                 }
 197
 198                 fio.page = page;
 199                 fio.old_blkaddr = fio.new_blkaddr;
 200                 f2fs_submit_page_mbio(&fio);
 201                 f2fs_put_page(page, 0);
 202         }
 203 out:
 204         f2fs_submit_merged_bio(sbi, META, READ);
 205         blk_finish_plug(&plug);
 206         return blkno - start;
 207 }
 208
 209 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 210 {
 211         struct page *page;
 212         bool readahead = false;
 213
 214         page = find_get_page(META_MAPPING(sbi), index);
 215         if (!page || !PageUptodate(page))
 216                 readahead = true;
 217         f2fs_put_page(page, 0);
 218
 219         if (readahead)
 220                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 221 }
 222
 223 static int f2fs_write_meta_page(struct page *page,
 224                                 struct writeback_control *wbc)
 225 {
 226         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 227
 228         trace_f2fs_writepage(page, META);
 229
 230         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 231                 goto redirty_out;
 232         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 233                 goto redirty_out;
 234         if (unlikely(f2fs_cp_error(sbi)))
 235                 goto redirty_out;
 236
 237         write_meta_page(sbi, page);
 238         dec_page_count(sbi, F2FS_DIRTY_META);
 239
 240         if (wbc->for_reclaim)
 241                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
 242
 243         unlock_page(page);
 244
 245         if (unlikely(f2fs_cp_error(sbi)))
 246                 f2fs_submit_merged_bio(sbi, META, WRITE);
 247
 248         return 0;
 249
 250 redirty_out:
 251         redirty_page_for_writepage(wbc, page);
 252         return AOP_WRITEPAGE_ACTIVATE;
 253 }
 254
 255 static int f2fs_write_meta_pages(struct address_space *mapping,
 256                                 struct writeback_control *wbc)
 257 {
 258         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 259         long diff, written;
 260
 261         /* collect a number of dirty meta pages and write together */
 262         if (wbc->for_kupdate ||
 263                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 264                 goto skip_write;
 265
 266         trace_f2fs_writepages(mapping->host, wbc, META);
 267
 268         /* if mounting is failed, skip writing node pages */
 269         mutex_lock(&sbi->cp_mutex);
 270         diff = nr_pages_to_write(sbi, META, wbc);
 271         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 272         mutex_unlock(&sbi->cp_mutex);
 273         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 274         return 0;
 275
 276 skip_write:
 277         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 278         trace_f2fs_writepages(mapping->host, wbc, META);
 279         return 0;
 280 }
 281
 282 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 283                                                 long nr_to_write)
 284 {
 285         struct address_space *mapping = META_MAPPING(sbi);
 286         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 287         struct pagevec pvec;
 288         long nwritten = 0;
 289         struct writeback_control wbc = {
 290                 .for_reclaim = 0,
 291         };
 292         struct blk_plug plug;
 293
 294         pagevec_init(&pvec, 0);
 295
 296         blk_start_plug(&plug);
 297
 298         while (index <= end) {
 299                 int i, nr_pages;
 300                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 301                                 PAGECACHE_TAG_DIRTY,
 302                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 303                 if (unlikely(nr_pages == 0))
 304                         break;
 305
 306                 for (i = 0; i < nr_pages; i++) {
 307                         struct page *page = pvec.pages[i];
 308
 309                         if (prev == ULONG_MAX)
 310                                 prev = page->index - 1;
 311                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 312                                 pagevec_release(&pvec);
 313                                 goto stop;
 314                         }
 315
 316                         lock_page(page);
 317
 318                         if (unlikely(page->mapping != mapping)) {
 319 continue_unlock:
 320                                 unlock_page(page);
 321                                 continue;
 322                         }
 323                         if (!PageDirty(page)) {
 324                                 /* someone wrote it for us */
 325                                 goto continue_unlock;
 326                         }
 327
 328                         f2fs_wait_on_page_writeback(page, META, true);
 329
 330                         BUG_ON(PageWriteback(page));
 331                         if (!clear_page_dirty_for_io(page))
 332                                 goto continue_unlock;
 333
 334                         if (mapping->a_ops->writepage(page, &wbc)) {
 335                                 unlock_page(page);
 336                                 break;
 337                         }
 338                         nwritten++;
 339                         prev = page->index;
 340                         if (unlikely(nwritten >= nr_to_write))
 341                                 break;
 342                 }
 343                 pagevec_release(&pvec);
 344                 cond_resched();
 345         }
 346 stop:
 347         if (nwritten)
 348                 f2fs_submit_merged_bio(sbi, type, WRITE);
 349
 350         blk_finish_plug(&plug);
 351
 352         return nwritten;
 353 }
 354
 355 static int f2fs_set_meta_page_dirty(struct page *page)
 356 {
 357         trace_f2fs_set_page_dirty(page, META);
 358
 359         SetPageUptodate(page);
 360         if (!PageDirty(page)) {
 361                 __set_page_dirty_nobuffers(page);
 362                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 363                 SetPagePrivate(page);
 364                 f2fs_trace_pid(page);
 365                 return 1;
 366         }
 367         return 0;
 368 }
 369
 370 const struct address_space_operations f2fs_meta_aops = {
 371         .writepage      = f2fs_write_meta_page,
 372         .writepages     = f2fs_write_meta_pages,
 373         .set_page_dirty = f2fs_set_meta_page_dirty,
 374         .invalidatepage = f2fs_invalidate_page,
 375         .releasepage    = f2fs_release_page,
 376 };
 377
 378 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 379 {
 380         struct inode_management *im = &sbi->im[type];
 381         struct ino_entry *e, *tmp;
 382
 383         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 384 retry:
 385         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 386
 387         spin_lock(&im->ino_lock);
 388         e = radix_tree_lookup(&im->ino_root, ino);
 389         if (!e) {
 390                 e = tmp;
 391                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 392                         spin_unlock(&im->ino_lock);
 393                         radix_tree_preload_end();
 394                         goto retry;
 395                 }
 396                 memset(e, 0, sizeof(struct ino_entry));
 397                 e->ino = ino;
 398
 399                 list_add_tail(&e->list, &im->ino_list);
 400                 if (type != ORPHAN_INO)
 401                         im->ino_num++;
 402         }
 403         spin_unlock(&im->ino_lock);
 404         radix_tree_preload_end();
 405
 406         if (e != tmp)
 407                 kmem_cache_free(ino_entry_slab, tmp);
 408 }
 409
 410 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 411 {
 412         struct inode_management *im = &sbi->im[type];
 413         struct ino_entry *e;
 414
 415         spin_lock(&im->ino_lock);
 416         e = radix_tree_lookup(&im->ino_root, ino);
 417         if (e) {
 418                 list_del(&e->list);
 419                 radix_tree_delete(&im->ino_root, ino);
 420                 im->ino_num--;
 421                 spin_unlock(&im->ino_lock);
 422                 kmem_cache_free(ino_entry_slab, e);
 423                 return;
 424         }
 425         spin_unlock(&im->ino_lock);
 426 }
 427
 428 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 429 {
 430         /* add new dirty ino entry into list */
 431         __add_ino_entry(sbi, ino, type);
 432 }
 433
 434 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 435 {
 436         /* remove dirty ino entry from list */
 437         __remove_ino_entry(sbi, ino, type);
 438 }
 439
 440 /* mode should be APPEND_INO or UPDATE_INO */
 441 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 442 {
 443         struct inode_management *im = &sbi->im[mode];
 444         struct ino_entry *e;
 445
 446         spin_lock(&im->ino_lock);
 447         e = radix_tree_lookup(&im->ino_root, ino);
 448         spin_unlock(&im->ino_lock);
 449         return e ? true : false;
 450 }
 451
 452 void release_ino_entry(struct f2fs_sb_info *sbi)
 453 {
 454         struct ino_entry *e, *tmp;
 455         int i;
 456
 457         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 458                 struct inode_management *im = &sbi->im[i];
 459
 460                 spin_lock(&im->ino_lock);
 461                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 462                         list_del(&e->list);
 463                         radix_tree_delete(&im->ino_root, e->ino);
 464                         kmem_cache_free(ino_entry_slab, e);
 465                         im->ino_num--;
 466                 }
 467                 spin_unlock(&im->ino_lock);
 468         }
 469 }
 470
 471 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 472 {
 473         struct inode_management *im = &sbi->im[ORPHAN_INO];
 474         int err = 0;
 475
 476         spin_lock(&im->ino_lock);
 477         if (unlikely(im->ino_num >= sbi->max_orphans))
 478                 err = -ENOSPC;
 479         else
 480                 im->ino_num++;
 481         spin_unlock(&im->ino_lock);
 482
 483         return err;
 484 }
 485
 486 void release_orphan_inode(struct f2fs_sb_info *sbi)
 487 {
 488         struct inode_management *im = &sbi->im[ORPHAN_INO];
 489
 490         spin_lock(&im->ino_lock);
 491         f2fs_bug_on(sbi, im->ino_num == 0);
 492         im->ino_num--;
 493         spin_unlock(&im->ino_lock);
 494 }
 495
 496 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 497 {
 498         /* add new orphan ino entry into list */
 499         __add_ino_entry(sbi, ino, ORPHAN_INO);
 500 }
 501
 502 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 503 {
 504         /* remove orphan entry from orphan list */
 505         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 506 }
 507
 508 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 509 {
 510         struct inode *inode;
 511
 512         inode = f2fs_iget(sbi->sb, ino);
 513         if (IS_ERR(inode)) {
 514                 /*
 515                  * there should be a bug that we can't find the entry
 516                  * to orphan inode.
 517                  */
 518                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 519                 return PTR_ERR(inode);
 520         }
 521
 522         clear_nlink(inode);
 523
 524         /* truncate all the data during iput */
 525         iput(inode);
 526         return 0;
 527 }
 528
 529 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 530 {
 531         block_t start_blk, orphan_blocks, i, j;
 532         int err;
 533
 534         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 535                 return 0;
 536
 537         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 538         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 539
 540         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 541
 542         for (i = 0; i < orphan_blocks; i++) {
 543                 struct page *page = get_meta_page(sbi, start_blk + i);
 544                 struct f2fs_orphan_block *orphan_blk;
 545
 546                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 547                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 548                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 549                         err = recover_orphan_inode(sbi, ino);
 550                         if (err) {
 551                                 f2fs_put_page(page, 1);
 552                                 return err;
 553                         }
 554                 }
 555                 f2fs_put_page(page, 1);
 556         }
 557         /* clear Orphan Flag */
 558         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 559         return 0;
 560 }
 561
 562 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 563 {
 564         struct list_head *head;
 565         struct f2fs_orphan_block *orphan_blk = NULL;
 566         unsigned int nentries = 0;
 567         unsigned short index = 1;
 568         unsigned short orphan_blocks;
 569         struct page *page = NULL;
 570         struct ino_entry *orphan = NULL;
 571         struct inode_management *im = &sbi->im[ORPHAN_INO];
 572
 573         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 574
 575         /*
 576          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 577          * orphan inode operations are covered under f2fs_lock_op().
 578          * And, spin_lock should be avoided due to page operations below.
 579          */
 580         head = &im->ino_list;
 581
 582         /* loop for each orphan inode entry and write them in Jornal block */
 583         list_for_each_entry(orphan, head, list) {
 584                 if (!page) {
 585                         page = grab_meta_page(sbi, start_blk++);
 586                         orphan_blk =
 587                                 (struct f2fs_orphan_block *)page_address(page);
 588                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 589                 }
 590
 591                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 592
 593                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 594                         /*
 595                          * an orphan block is full of 1020 entries,
 596                          * then we need to flush current orphan blocks
 597                          * and bring another one in memory
 598                          */
 599                         orphan_blk->blk_addr = cpu_to_le16(index);
 600                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 601                         orphan_blk->entry_count = cpu_to_le32(nentries);
 602                         set_page_dirty(page);
 603                         f2fs_put_page(page, 1);
 604                         index++;
 605                         nentries = 0;
 606                         page = NULL;
 607                 }
 608         }
 609
 610         if (page) {
 611                 orphan_blk->blk_addr = cpu_to_le16(index);
 612                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 613                 orphan_blk->entry_count = cpu_to_le32(nentries);
 614                 set_page_dirty(page);
 615                 f2fs_put_page(page, 1);
 616         }
 617 }
 618
 619 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 620                                 block_t cp_addr, unsigned long long *version)
 621 {
 622         struct page *cp_page_1, *cp_page_2 = NULL;
 623         unsigned long blk_size = sbi->blocksize;
 624         struct f2fs_checkpoint *cp_block;
 625         unsigned long long cur_version = 0, pre_version = 0;
 626         size_t crc_offset;
 627         __u32 crc = 0;
 628
 629         /* Read the 1st cp block in this CP pack */
 630         cp_page_1 = get_meta_page(sbi, cp_addr);
 631
 632         /* get the version number */
 633         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 634         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 635         if (crc_offset >= blk_size)
 636                 goto invalid_cp1;
 637
 638         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 639         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 640                 goto invalid_cp1;
 641
 642         pre_version = cur_cp_version(cp_block);
 643
 644         /* Read the 2nd cp block in this CP pack */
 645         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 646         cp_page_2 = get_meta_page(sbi, cp_addr);
 647
 648         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 649         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 650         if (crc_offset >= blk_size)
 651                 goto invalid_cp2;
 652
 653         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 654         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 655                 goto invalid_cp2;
 656
 657         cur_version = cur_cp_version(cp_block);
 658
 659         if (cur_version == pre_version) {
 660                 *version = cur_version;
 661                 f2fs_put_page(cp_page_2, 1);
 662                 return cp_page_1;
 663         }
 664 invalid_cp2:
 665         f2fs_put_page(cp_page_2, 1);
 666 invalid_cp1:
 667         f2fs_put_page(cp_page_1, 1);
 668         return NULL;
 669 }
 670
 671 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 672 {
 673         struct f2fs_checkpoint *cp_block;
 674         struct f2fs_super_block *fsb = sbi->raw_super;
 675         struct page *cp1, *cp2, *cur_page;
 676         unsigned long blk_size = sbi->blocksize;
 677         unsigned long long cp1_version = 0, cp2_version = 0;
 678         unsigned long long cp_start_blk_no;
 679         unsigned int cp_blks = 1 + __cp_payload(sbi);
 680         block_t cp_blk_no;
 681         int i;
 682
 683         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 684         if (!sbi->ckpt)
 685                 return -ENOMEM;
 686         /*
 687          * Finding out valid cp block involves read both
 688          * sets( cp pack1 and cp pack 2)
 689          */
 690         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 691         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 692
 693         /* The second checkpoint pack should start at the next segment */
 694         cp_start_blk_no += ((unsigned long long)1) <<
 695                                 le32_to_cpu(fsb->log_blocks_per_seg);
 696         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 697
 698         if (cp1 && cp2) {
 699                 if (ver_after(cp2_version, cp1_version))
 700                         cur_page = cp2;
 701                 else
 702                         cur_page = cp1;
 703         } else if (cp1) {
 704                 cur_page = cp1;
 705         } else if (cp2) {
 706                 cur_page = cp2;
 707         } else {
 708                 goto fail_no_cp;
 709         }
 710
 711         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 712         memcpy(sbi->ckpt, cp_block, blk_size);
 713
 714         /* Sanity checking of checkpoint */
 715         if (sanity_check_ckpt(sbi))
 716                 goto fail_no_cp;
 717
 718         if (cp_blks <= 1)
 719                 goto done;
 720
 721         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 722         if (cur_page == cp2)
 723                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 724
 725         for (i = 1; i < cp_blks; i++) {
 726                 void *sit_bitmap_ptr;
 727                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 728
 729                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 730                 sit_bitmap_ptr = page_address(cur_page);
 731                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 732                 f2fs_put_page(cur_page, 1);
 733         }
 734 done:
 735         f2fs_put_page(cp1, 1);
 736         f2fs_put_page(cp2, 1);
 737         return 0;
 738
 739 fail_no_cp:
 740         kfree(sbi->ckpt);
 741         return -EINVAL;
 742 }
 743
 744 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 745 {
 746         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 747         struct f2fs_inode_info *fi = F2FS_I(inode);
 748         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 749
 750         if (is_inode_flag_set(fi, flag))
 751                 return;
 752
 753         set_inode_flag(fi, flag);
 754         list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
 755         stat_inc_dirty_inode(sbi, type);
 756 }
 757
 758 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 759 {
 760         struct f2fs_inode_info *fi = F2FS_I(inode);
 761         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 762
 763         if (get_dirty_pages(inode) ||
 764                         !is_inode_flag_set(F2FS_I(inode), flag))
 765                 return;
 766
 767         list_del_init(&fi->dirty_list);
 768         clear_inode_flag(fi, flag);
 769         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 770 }
 771
 772 void update_dirty_page(struct inode *inode, struct page *page)
 773 {
 774         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 775         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 776
 777         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 778                         !S_ISLNK(inode->i_mode))
 779                 return;
 780
 781         spin_lock(&sbi->inode_lock[type]);
 782         __add_dirty_inode(inode, type);
 783         inode_inc_dirty_pages(inode);
 784         spin_unlock(&sbi->inode_lock[type]);
 785
 786         SetPagePrivate(page);
 787         f2fs_trace_pid(page);
 788 }
 789
 790 void add_dirty_dir_inode(struct inode *inode)
 791 {
 792         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 793
 794         spin_lock(&sbi->inode_lock[DIR_INODE]);
 795         __add_dirty_inode(inode, DIR_INODE);
 796         spin_unlock(&sbi->inode_lock[DIR_INODE]);
 797 }
 798
 799 void remove_dirty_inode(struct inode *inode)
 800 {
 801         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 802         struct f2fs_inode_info *fi = F2FS_I(inode);
 803         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 804
 805         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 806                         !S_ISLNK(inode->i_mode))
 807                 return;
 808
 809         spin_lock(&sbi->inode_lock[type]);
 810         __remove_dirty_inode(inode, type);
 811         spin_unlock(&sbi->inode_lock[type]);
 812
 813         /* Only from the recovery routine */
 814         if (is_inode_flag_set(fi, FI_DELAY_IPUT)) {
 815                 clear_inode_flag(fi, FI_DELAY_IPUT);
 816                 iput(inode);
 817         }
 818 }
 819
 820 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 821 {
 822         struct list_head *head;
 823         struct inode *inode;
 824         struct f2fs_inode_info *fi;
 825         bool is_dir = (type == DIR_INODE);
 826
 827         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 828                                 get_pages(sbi, is_dir ?
 829                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 830 retry:
 831         if (unlikely(f2fs_cp_error(sbi)))
 832                 return -EIO;
 833
 834         spin_lock(&sbi->inode_lock[type]);
 835
 836         head = &sbi->inode_list[type];
 837         if (list_empty(head)) {
 838                 spin_unlock(&sbi->inode_lock[type]);
 839                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 840                                 get_pages(sbi, is_dir ?
 841                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 842                 return 0;
 843         }
 844         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 845         inode = igrab(&fi->vfs_inode);
 846         spin_unlock(&sbi->inode_lock[type]);
 847         if (inode) {
 848                 filemap_fdatawrite(inode->i_mapping);
 849                 iput(inode);
 850         } else {
 851                 /*
 852                  * We should submit bio, since it exists several
 853                  * wribacking dentry pages in the freeing inode.
 854                  */
 855                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 856                 cond_resched();
 857         }
 858         goto retry;
 859 }
 860
 861 /*
 862  * Freeze all the FS-operations for checkpoint.
 863  */
 864 static int block_operations(struct f2fs_sb_info *sbi)
 865 {
 866         struct writeback_control wbc = {
 867                 .sync_mode = WB_SYNC_ALL,
 868                 .nr_to_write = LONG_MAX,
 869                 .for_reclaim = 0,
 870         };
 871         struct blk_plug plug;
 872         int err = 0;
 873
 874         blk_start_plug(&plug);
 875
 876 retry_flush_dents:
 877         f2fs_lock_all(sbi);
 878         /* write all the dirty dentry pages */
 879         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 880                 f2fs_unlock_all(sbi);
 881                 err = sync_dirty_inodes(sbi, DIR_INODE);
 882                 if (err)
 883                         goto out;
 884                 goto retry_flush_dents;
 885         }
 886
 887         /*
 888          * POR: we should ensure that there are no dirty node pages
 889          * until finishing nat/sit flush.
 890          */
 891 retry_flush_nodes:
 892         down_write(&sbi->node_write);
 893
 894         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 895                 up_write(&sbi->node_write);
 896                 err = sync_node_pages(sbi, &wbc);
 897                 if (err) {
 898                         f2fs_unlock_all(sbi);
 899                         goto out;
 900                 }
 901                 goto retry_flush_nodes;
 902         }
 903 out:
 904         blk_finish_plug(&plug);
 905         return err;
 906 }
 907
 908 static void unblock_operations(struct f2fs_sb_info *sbi)
 909 {
 910         up_write(&sbi->node_write);
 911         f2fs_unlock_all(sbi);
 912 }
 913
 914 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 915 {
 916         DEFINE_WAIT(wait);
 917
 918         for (;;) {
 919                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 920
 921                 if (!get_pages(sbi, F2FS_WRITEBACK))
 922                         break;
 923
 924                 io_schedule_timeout(5*HZ);
 925         }
 926         finish_wait(&sbi->cp_wait, &wait);
 927 }
 928
 929 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 930 {
 931         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 932         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 933         struct f2fs_nm_info *nm_i = NM_I(sbi);
 934         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 935         nid_t last_nid = nm_i->next_scan_nid;
 936         block_t start_blk;
 937         unsigned int data_sum_blocks, orphan_blocks;
 938         __u32 crc32 = 0;
 939         int i;
 940         int cp_payload_blks = __cp_payload(sbi);
 941         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
 942         bool invalidate = false;
 943         struct super_block *sb = sbi->sb;
 944         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
 945         u64 kbytes_written;
 946
 947         /*
 948          * This avoids to conduct wrong roll-forward operations and uses
 949          * metapages, so should be called prior to sync_meta_pages below.
 950          */
 951         if (discard_next_dnode(sbi, discard_blk))
 952                 invalidate = true;
 953
 954         /* Flush all the NAT/SIT pages */
 955         while (get_pages(sbi, F2FS_DIRTY_META)) {
 956                 sync_meta_pages(sbi, META, LONG_MAX);
 957                 if (unlikely(f2fs_cp_error(sbi)))
 958                         return -EIO;
 959         }
 960
 961         next_free_nid(sbi, &last_nid);
 962
 963         /*
 964          * modify checkpoint
 965          * version number is already updated
 966          */
 967         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 968         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 969         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 970         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 971                 ckpt->cur_node_segno[i] =
 972                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 973                 ckpt->cur_node_blkoff[i] =
 974                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 975                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 976                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 977         }
 978         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 979                 ckpt->cur_data_segno[i] =
 980                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 981                 ckpt->cur_data_blkoff[i] =
 982                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 983                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 984                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 985         }
 986
 987         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 988         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 989         ckpt->next_free_nid = cpu_to_le32(last_nid);
 990
 991         /* 2 cp  + n data seg summary + orphan inode blocks */
 992         data_sum_blocks = npages_for_summary_flush(sbi, false);
 993         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 994                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 995         else
 996                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 997
 998         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 999         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1000                         orphan_blocks);
1001
1002         if (__remain_node_summaries(cpc->reason))
1003                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1004                                 cp_payload_blks + data_sum_blocks +
1005                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1006         else
1007                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1008                                 cp_payload_blks + data_sum_blocks +
1009                                 orphan_blocks);
1010
1011         if (cpc->reason == CP_UMOUNT)
1012                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1013         else
1014                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1015
1016         if (cpc->reason == CP_FASTBOOT)
1017                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1018         else
1019                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1020
1021         if (orphan_num)
1022                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1023         else
1024                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1025
1026         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1027                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1028
1029         /* update SIT/NAT bitmap */
1030         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1031         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1032
1033         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1034         *((__le32 *)((unsigned char *)ckpt +
1035                                 le32_to_cpu(ckpt->checksum_offset)))
1036                                 = cpu_to_le32(crc32);
1037
1038         start_blk = __start_cp_addr(sbi);
1039
1040         /* need to wait for end_io results */
1041         wait_on_all_pages_writeback(sbi);
1042         if (unlikely(f2fs_cp_error(sbi)))
1043                 return -EIO;
1044
1045         /* write out checkpoint buffer at block 0 */
1046         update_meta_page(sbi, ckpt, start_blk++);
1047
1048         for (i = 1; i < 1 + cp_payload_blks; i++)
1049                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1050                                                         start_blk++);
1051
1052         if (orphan_num) {
1053                 write_orphan_inodes(sbi, start_blk);
1054                 start_blk += orphan_blocks;
1055         }
1056
1057         write_data_summaries(sbi, start_blk);
1058         start_blk += data_sum_blocks;
1059
1060         /* Record write statistics in the hot node summary */
1061         kbytes_written = sbi->kbytes_written;
1062         if (sb->s_bdev->bd_part)
1063                 kbytes_written += BD_PART_WRITTEN(sbi);
1064
1065         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1066
1067         if (__remain_node_summaries(cpc->reason)) {
1068                 write_node_summaries(sbi, start_blk);
1069                 start_blk += NR_CURSEG_NODE_TYPE;
1070         }
1071
1072         /* writeout checkpoint block */
1073         update_meta_page(sbi, ckpt, start_blk);
1074
1075         /* wait for previous submitted node/meta pages writeback */
1076         wait_on_all_pages_writeback(sbi);
1077
1078         if (unlikely(f2fs_cp_error(sbi)))
1079                 return -EIO;
1080
1081         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1082         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1083
1084         /* update user_block_counts */
1085         sbi->last_valid_block_count = sbi->total_valid_block_count;
1086         sbi->alloc_valid_block_count = 0;
1087
1088         /* Here, we only have one bio having CP pack */
1089         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1090
1091         /* wait for previous submitted meta pages writeback */
1092         wait_on_all_pages_writeback(sbi);
1093
1094         /*
1095          * invalidate meta page which is used temporarily for zeroing out
1096          * block at the end of warm node chain.
1097          */
1098         if (invalidate)
1099                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1100                                                                 discard_blk);
1101
1102         release_ino_entry(sbi);
1103
1104         if (unlikely(f2fs_cp_error(sbi)))
1105                 return -EIO;
1106
1107         clear_prefree_segments(sbi, cpc);
1108         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1109
1110         return 0;
1111 }
1112
1113 /*
1114  * We guarantee that this checkpoint procedure will not fail.
1115  */
1116 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1117 {
1118         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1119         unsigned long long ckpt_ver;
1120         int err = 0;
1121
1122         mutex_lock(&sbi->cp_mutex);
1123
1124         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1125                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1126                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1127                 goto out;
1128         if (unlikely(f2fs_cp_error(sbi))) {
1129                 err = -EIO;
1130                 goto out;
1131         }
1132         if (f2fs_readonly(sbi->sb)) {
1133                 err = -EROFS;
1134                 goto out;
1135         }
1136
1137         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1138
1139         err = block_operations(sbi);
1140         if (err)
1141                 goto out;
1142
1143         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1144
1145         f2fs_flush_merged_bios(sbi);
1146
1147         /*
1148          * update checkpoint pack index
1149          * Increase the version number so that
1150          * SIT entries and seg summaries are written at correct place
1151          */
1152         ckpt_ver = cur_cp_version(ckpt);
1153         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1154
1155         /* write cached NAT/SIT entries to NAT/SIT area */
1156         flush_nat_entries(sbi);
1157         flush_sit_entries(sbi, cpc);
1158
1159         /* unlock all the fs_lock[] in do_checkpoint() */
1160         err = do_checkpoint(sbi, cpc);
1161
1162         unblock_operations(sbi);
1163         stat_inc_cp_count(sbi->stat_info);
1164
1165         if (cpc->reason == CP_RECOVERY)
1166                 f2fs_msg(sbi->sb, KERN_NOTICE,
1167                         "checkpoint: version = %llx", ckpt_ver);
1168
1169         /* do checkpoint periodically */
1170         f2fs_update_time(sbi, CP_TIME);
1171         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1172 out:
1173         mutex_unlock(&sbi->cp_mutex);
1174         return err;
1175 }
1176
1177 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1178 {
1179         int i;
1180
1181         for (i = 0; i < MAX_INO_ENTRY; i++) {
1182                 struct inode_management *im = &sbi->im[i];
1183
1184                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1185                 spin_lock_init(&im->ino_lock);
1186                 INIT_LIST_HEAD(&im->ino_list);
1187                 im->ino_num = 0;
1188         }
1189
1190         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1191                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1192                                 F2FS_ORPHANS_PER_BLOCK;
1193 }
1194
1195 int __init create_checkpoint_caches(void)
1196 {
1197         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1198                         sizeof(struct ino_entry));
1199         if (!ino_entry_slab)
1200                 return -ENOMEM;
1201         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1202                         sizeof(struct inode_entry));
1203         if (!inode_entry_slab) {
1204                 kmem_cache_destroy(ino_entry_slab);
1205                 return -ENOMEM;
1206         }
1207         return 0;
1208 }
1209
1210 void destroy_checkpoint_caches(void)
1211 {
1212         kmem_cache_destroy(ino_entry_slab);
1213         kmem_cache_destroy(inode_entry_slab);
1214 }