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