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 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
  32         sbi->sb->s_flags |= MS_RDONLY;
  33         if (!end_io)
  34                 f2fs_flush_merged_bios(sbi);
  35 }
  36
  37 /*
  38  * We guarantee no failure on the returned page.
  39  */
  40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  41 {
  42         struct address_space *mapping = META_MAPPING(sbi);
  43         struct page *page = NULL;
  44 repeat:
  45         page = f2fs_grab_cache_page(mapping, index, false);
  46         if (!page) {
  47                 cond_resched();
  48                 goto repeat;
  49         }
  50         f2fs_wait_on_page_writeback(page, META, true);
  51         SetPageUptodate(page);
  52         return page;
  53 }
  54
  55 /*
  56  * We guarantee no failure on the returned page.
  57  */
  58 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  59                                                         bool is_meta)
  60 {
  61         struct address_space *mapping = META_MAPPING(sbi);
  62         struct page *page;
  63         struct f2fs_io_info fio = {
  64                 .sbi = sbi,
  65                 .type = META,
  66                 .op = REQ_OP_READ,
  67                 .op_flags = READ_SYNC | REQ_META | REQ_PRIO,
  68                 .old_blkaddr = index,
  69                 .new_blkaddr = index,
  70                 .encrypted_page = NULL,
  71         };
  72
  73         if (unlikely(!is_meta))
  74                 fio.op_flags &= ~REQ_META;
  75 repeat:
  76         page = f2fs_grab_cache_page(mapping, index, false);
  77         if (!page) {
  78                 cond_resched();
  79                 goto repeat;
  80         }
  81         if (PageUptodate(page))
  82                 goto out;
  83
  84         fio.page = page;
  85
  86         if (f2fs_submit_page_bio(&fio)) {
  87                 f2fs_put_page(page, 1);
  88                 goto repeat;
  89         }
  90
  91         lock_page(page);
  92         if (unlikely(page->mapping != mapping)) {
  93                 f2fs_put_page(page, 1);
  94                 goto repeat;
  95         }
  96
  97         /*
  98          * if there is any IO error when accessing device, make our filesystem
  99          * readonly and make sure do not write checkpoint with non-uptodate
 100          * meta page.
 101          */
 102         if (unlikely(!PageUptodate(page)))
 103                 f2fs_stop_checkpoint(sbi, false);
 104 out:
 105         return page;
 106 }
 107
 108 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 109 {
 110         return __get_meta_page(sbi, index, true);
 111 }
 112
 113 /* for POR only */
 114 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 115 {
 116         return __get_meta_page(sbi, index, false);
 117 }
 118
 119 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 120 {
 121         switch (type) {
 122         case META_NAT:
 123                 break;
 124         case META_SIT:
 125                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 126                         return false;
 127                 break;
 128         case META_SSA:
 129                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 130                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 131                         return false;
 132                 break;
 133         case META_CP:
 134                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 135                         blkaddr < __start_cp_addr(sbi)))
 136                         return false;
 137                 break;
 138         case META_POR:
 139                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 140                         blkaddr < MAIN_BLKADDR(sbi)))
 141                         return false;
 142                 break;
 143         default:
 144                 BUG();
 145         }
 146
 147         return true;
 148 }
 149
 150 /*
 151  * Readahead CP/NAT/SIT/SSA pages
 152  */
 153 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 154                                                         int type, bool sync)
 155 {
 156         struct page *page;
 157         block_t blkno = start;
 158         struct f2fs_io_info fio = {
 159                 .sbi = sbi,
 160                 .type = META,
 161                 .op = REQ_OP_READ,
 162                 .op_flags = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
 163                 .encrypted_page = NULL,
 164         };
 165         struct blk_plug plug;
 166
 167         if (unlikely(type == META_POR))
 168                 fio.op_flags &= ~REQ_META;
 169
 170         blk_start_plug(&plug);
 171         for (; nrpages-- > 0; blkno++) {
 172
 173                 if (!is_valid_blkaddr(sbi, blkno, type))
 174                         goto out;
 175
 176                 switch (type) {
 177                 case META_NAT:
 178                         if (unlikely(blkno >=
 179                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 180                                 blkno = 0;
 181                         /* get nat block addr */
 182                         fio.new_blkaddr = current_nat_addr(sbi,
 183                                         blkno * NAT_ENTRY_PER_BLOCK);
 184                         break;
 185                 case META_SIT:
 186                         /* get sit block addr */
 187                         fio.new_blkaddr = current_sit_addr(sbi,
 188                                         blkno * SIT_ENTRY_PER_BLOCK);
 189                         break;
 190                 case META_SSA:
 191                 case META_CP:
 192                 case META_POR:
 193                         fio.new_blkaddr = blkno;
 194                         break;
 195                 default:
 196                         BUG();
 197                 }
 198
 199                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 200                                                 fio.new_blkaddr, false);
 201                 if (!page)
 202                         continue;
 203                 if (PageUptodate(page)) {
 204                         f2fs_put_page(page, 1);
 205                         continue;
 206                 }
 207
 208                 fio.page = page;
 209                 fio.old_blkaddr = fio.new_blkaddr;
 210                 f2fs_submit_page_mbio(&fio);
 211                 f2fs_put_page(page, 0);
 212         }
 213 out:
 214         f2fs_submit_merged_bio(sbi, META, READ);
 215         blk_finish_plug(&plug);
 216         return blkno - start;
 217 }
 218
 219 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 220 {
 221         struct page *page;
 222         bool readahead = false;
 223
 224         page = find_get_page(META_MAPPING(sbi), index);
 225         if (!page || !PageUptodate(page))
 226                 readahead = true;
 227         f2fs_put_page(page, 0);
 228
 229         if (readahead)
 230                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 231 }
 232
 233 static int f2fs_write_meta_page(struct page *page,
 234                                 struct writeback_control *wbc)
 235 {
 236         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 237
 238         trace_f2fs_writepage(page, META);
 239
 240         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 241                 goto redirty_out;
 242         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 243                 goto redirty_out;
 244         if (unlikely(f2fs_cp_error(sbi)))
 245                 goto redirty_out;
 246
 247         write_meta_page(sbi, page);
 248         dec_page_count(sbi, F2FS_DIRTY_META);
 249
 250         if (wbc->for_reclaim)
 251                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
 252
 253         unlock_page(page);
 254
 255         if (unlikely(f2fs_cp_error(sbi)))
 256                 f2fs_submit_merged_bio(sbi, META, WRITE);
 257
 258         return 0;
 259
 260 redirty_out:
 261         redirty_page_for_writepage(wbc, page);
 262         return AOP_WRITEPAGE_ACTIVATE;
 263 }
 264
 265 static int f2fs_write_meta_pages(struct address_space *mapping,
 266                                 struct writeback_control *wbc)
 267 {
 268         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 269         long diff, written;
 270
 271         /* collect a number of dirty meta pages and write together */
 272         if (wbc->for_kupdate ||
 273                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 274                 goto skip_write;
 275
 276         trace_f2fs_writepages(mapping->host, wbc, META);
 277
 278         /* if mounting is failed, skip writing node pages */
 279         mutex_lock(&sbi->cp_mutex);
 280         diff = nr_pages_to_write(sbi, META, wbc);
 281         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 282         mutex_unlock(&sbi->cp_mutex);
 283         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 284         return 0;
 285
 286 skip_write:
 287         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 288         trace_f2fs_writepages(mapping->host, wbc, META);
 289         return 0;
 290 }
 291
 292 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 293                                                 long nr_to_write)
 294 {
 295         struct address_space *mapping = META_MAPPING(sbi);
 296         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 297         struct pagevec pvec;
 298         long nwritten = 0;
 299         struct writeback_control wbc = {
 300                 .for_reclaim = 0,
 301         };
 302         struct blk_plug plug;
 303
 304         pagevec_init(&pvec, 0);
 305
 306         blk_start_plug(&plug);
 307
 308         while (index <= end) {
 309                 int i, nr_pages;
 310                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 311                                 PAGECACHE_TAG_DIRTY,
 312                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 313                 if (unlikely(nr_pages == 0))
 314                         break;
 315
 316                 for (i = 0; i < nr_pages; i++) {
 317                         struct page *page = pvec.pages[i];
 318
 319                         if (prev == ULONG_MAX)
 320                                 prev = page->index - 1;
 321                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 322                                 pagevec_release(&pvec);
 323                                 goto stop;
 324                         }
 325
 326                         lock_page(page);
 327
 328                         if (unlikely(page->mapping != mapping)) {
 329 continue_unlock:
 330                                 unlock_page(page);
 331                                 continue;
 332                         }
 333                         if (!PageDirty(page)) {
 334                                 /* someone wrote it for us */
 335                                 goto continue_unlock;
 336                         }
 337
 338                         f2fs_wait_on_page_writeback(page, META, true);
 339
 340                         BUG_ON(PageWriteback(page));
 341                         if (!clear_page_dirty_for_io(page))
 342                                 goto continue_unlock;
 343
 344                         if (mapping->a_ops->writepage(page, &wbc)) {
 345                                 unlock_page(page);
 346                                 break;
 347                         }
 348                         nwritten++;
 349                         prev = page->index;
 350                         if (unlikely(nwritten >= nr_to_write))
 351                                 break;
 352                 }
 353                 pagevec_release(&pvec);
 354                 cond_resched();
 355         }
 356 stop:
 357         if (nwritten)
 358                 f2fs_submit_merged_bio(sbi, type, WRITE);
 359
 360         blk_finish_plug(&plug);
 361
 362         return nwritten;
 363 }
 364
 365 static int f2fs_set_meta_page_dirty(struct page *page)
 366 {
 367         trace_f2fs_set_page_dirty(page, META);
 368
 369         SetPageUptodate(page);
 370         if (!PageDirty(page)) {
 371                 __set_page_dirty_nobuffers(page);
 372                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 373                 SetPagePrivate(page);
 374                 f2fs_trace_pid(page);
 375                 return 1;
 376         }
 377         return 0;
 378 }
 379
 380 const struct address_space_operations f2fs_meta_aops = {
 381         .writepage      = f2fs_write_meta_page,
 382         .writepages     = f2fs_write_meta_pages,
 383         .set_page_dirty = f2fs_set_meta_page_dirty,
 384         .invalidatepage = f2fs_invalidate_page,
 385         .releasepage    = f2fs_release_page,
 386 };
 387
 388 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 389 {
 390         struct inode_management *im = &sbi->im[type];
 391         struct ino_entry *e, *tmp;
 392
 393         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 394 retry:
 395         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 396
 397         spin_lock(&im->ino_lock);
 398         e = radix_tree_lookup(&im->ino_root, ino);
 399         if (!e) {
 400                 e = tmp;
 401                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 402                         spin_unlock(&im->ino_lock);
 403                         radix_tree_preload_end();
 404                         goto retry;
 405                 }
 406                 memset(e, 0, sizeof(struct ino_entry));
 407                 e->ino = ino;
 408
 409                 list_add_tail(&e->list, &im->ino_list);
 410                 if (type != ORPHAN_INO)
 411                         im->ino_num++;
 412         }
 413         spin_unlock(&im->ino_lock);
 414         radix_tree_preload_end();
 415
 416         if (e != tmp)
 417                 kmem_cache_free(ino_entry_slab, tmp);
 418 }
 419
 420 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 421 {
 422         struct inode_management *im = &sbi->im[type];
 423         struct ino_entry *e;
 424
 425         spin_lock(&im->ino_lock);
 426         e = radix_tree_lookup(&im->ino_root, ino);
 427         if (e) {
 428                 list_del(&e->list);
 429                 radix_tree_delete(&im->ino_root, ino);
 430                 im->ino_num--;
 431                 spin_unlock(&im->ino_lock);
 432                 kmem_cache_free(ino_entry_slab, e);
 433                 return;
 434         }
 435         spin_unlock(&im->ino_lock);
 436 }
 437
 438 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 439 {
 440         /* add new dirty ino entry into list */
 441         __add_ino_entry(sbi, ino, type);
 442 }
 443
 444 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 445 {
 446         /* remove dirty ino entry from list */
 447         __remove_ino_entry(sbi, ino, type);
 448 }
 449
 450 /* mode should be APPEND_INO or UPDATE_INO */
 451 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 452 {
 453         struct inode_management *im = &sbi->im[mode];
 454         struct ino_entry *e;
 455
 456         spin_lock(&im->ino_lock);
 457         e = radix_tree_lookup(&im->ino_root, ino);
 458         spin_unlock(&im->ino_lock);
 459         return e ? true : false;
 460 }
 461
 462 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 463 {
 464         struct ino_entry *e, *tmp;
 465         int i;
 466
 467         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 468                 struct inode_management *im = &sbi->im[i];
 469
 470                 spin_lock(&im->ino_lock);
 471                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 472                         list_del(&e->list);
 473                         radix_tree_delete(&im->ino_root, e->ino);
 474                         kmem_cache_free(ino_entry_slab, e);
 475                         im->ino_num--;
 476                 }
 477                 spin_unlock(&im->ino_lock);
 478         }
 479 }
 480
 481 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 482 {
 483         struct inode_management *im = &sbi->im[ORPHAN_INO];
 484         int err = 0;
 485
 486         spin_lock(&im->ino_lock);
 487
 488 #ifdef CONFIG_F2FS_FAULT_INJECTION
 489         if (time_to_inject(FAULT_ORPHAN)) {
 490                 spin_unlock(&im->ino_lock);
 491                 return -ENOSPC;
 492         }
 493 #endif
 494         if (unlikely(im->ino_num >= sbi->max_orphans))
 495                 err = -ENOSPC;
 496         else
 497                 im->ino_num++;
 498         spin_unlock(&im->ino_lock);
 499
 500         return err;
 501 }
 502
 503 void release_orphan_inode(struct f2fs_sb_info *sbi)
 504 {
 505         struct inode_management *im = &sbi->im[ORPHAN_INO];
 506
 507         spin_lock(&im->ino_lock);
 508         f2fs_bug_on(sbi, im->ino_num == 0);
 509         im->ino_num--;
 510         spin_unlock(&im->ino_lock);
 511 }
 512
 513 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 514 {
 515         /* add new orphan ino entry into list */
 516         __add_ino_entry(sbi, ino, ORPHAN_INO);
 517 }
 518
 519 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 520 {
 521         /* remove orphan entry from orphan list */
 522         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 523 }
 524
 525 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 526 {
 527         struct inode *inode;
 528
 529         inode = f2fs_iget(sbi->sb, ino);
 530         if (IS_ERR(inode)) {
 531                 /*
 532                  * there should be a bug that we can't find the entry
 533                  * to orphan inode.
 534                  */
 535                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 536                 return PTR_ERR(inode);
 537         }
 538
 539         clear_nlink(inode);
 540
 541         /* truncate all the data during iput */
 542         iput(inode);
 543         return 0;
 544 }
 545
 546 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 547 {
 548         block_t start_blk, orphan_blocks, i, j;
 549         int err;
 550
 551         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 552                 return 0;
 553
 554         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 555         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 556
 557         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 558
 559         for (i = 0; i < orphan_blocks; i++) {
 560                 struct page *page = get_meta_page(sbi, start_blk + i);
 561                 struct f2fs_orphan_block *orphan_blk;
 562
 563                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 564                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 565                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 566                         err = recover_orphan_inode(sbi, ino);
 567                         if (err) {
 568                                 f2fs_put_page(page, 1);
 569                                 return err;
 570                         }
 571                 }
 572                 f2fs_put_page(page, 1);
 573         }
 574         /* clear Orphan Flag */
 575         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 576         return 0;
 577 }
 578
 579 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 580 {
 581         struct list_head *head;
 582         struct f2fs_orphan_block *orphan_blk = NULL;
 583         unsigned int nentries = 0;
 584         unsigned short index = 1;
 585         unsigned short orphan_blocks;
 586         struct page *page = NULL;
 587         struct ino_entry *orphan = NULL;
 588         struct inode_management *im = &sbi->im[ORPHAN_INO];
 589
 590         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 591
 592         /*
 593          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 594          * orphan inode operations are covered under f2fs_lock_op().
 595          * And, spin_lock should be avoided due to page operations below.
 596          */
 597         head = &im->ino_list;
 598
 599         /* loop for each orphan inode entry and write them in Jornal block */
 600         list_for_each_entry(orphan, head, list) {
 601                 if (!page) {
 602                         page = grab_meta_page(sbi, start_blk++);
 603                         orphan_blk =
 604                                 (struct f2fs_orphan_block *)page_address(page);
 605                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 606                 }
 607
 608                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 609
 610                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 611                         /*
 612                          * an orphan block is full of 1020 entries,
 613                          * then we need to flush current orphan blocks
 614                          * and bring another one in memory
 615                          */
 616                         orphan_blk->blk_addr = cpu_to_le16(index);
 617                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 618                         orphan_blk->entry_count = cpu_to_le32(nentries);
 619                         set_page_dirty(page);
 620                         f2fs_put_page(page, 1);
 621                         index++;
 622                         nentries = 0;
 623                         page = NULL;
 624                 }
 625         }
 626
 627         if (page) {
 628                 orphan_blk->blk_addr = cpu_to_le16(index);
 629                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 630                 orphan_blk->entry_count = cpu_to_le32(nentries);
 631                 set_page_dirty(page);
 632                 f2fs_put_page(page, 1);
 633         }
 634 }
 635
 636 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 637                                 block_t cp_addr, unsigned long long *version)
 638 {
 639         struct page *cp_page_1, *cp_page_2 = NULL;
 640         unsigned long blk_size = sbi->blocksize;
 641         struct f2fs_checkpoint *cp_block;
 642         unsigned long long cur_version = 0, pre_version = 0;
 643         size_t crc_offset;
 644         __u32 crc = 0;
 645
 646         /* Read the 1st cp block in this CP pack */
 647         cp_page_1 = get_meta_page(sbi, cp_addr);
 648
 649         /* get the version number */
 650         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 651         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 652         if (crc_offset >= blk_size)
 653                 goto invalid_cp1;
 654
 655         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 656         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 657                 goto invalid_cp1;
 658
 659         pre_version = cur_cp_version(cp_block);
 660
 661         /* Read the 2nd cp block in this CP pack */
 662         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 663         cp_page_2 = get_meta_page(sbi, cp_addr);
 664
 665         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 666         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 667         if (crc_offset >= blk_size)
 668                 goto invalid_cp2;
 669
 670         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 671         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 672                 goto invalid_cp2;
 673
 674         cur_version = cur_cp_version(cp_block);
 675
 676         if (cur_version == pre_version) {
 677                 *version = cur_version;
 678                 f2fs_put_page(cp_page_2, 1);
 679                 return cp_page_1;
 680         }
 681 invalid_cp2:
 682         f2fs_put_page(cp_page_2, 1);
 683 invalid_cp1:
 684         f2fs_put_page(cp_page_1, 1);
 685         return NULL;
 686 }
 687
 688 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 689 {
 690         struct f2fs_checkpoint *cp_block;
 691         struct f2fs_super_block *fsb = sbi->raw_super;
 692         struct page *cp1, *cp2, *cur_page;
 693         unsigned long blk_size = sbi->blocksize;
 694         unsigned long long cp1_version = 0, cp2_version = 0;
 695         unsigned long long cp_start_blk_no;
 696         unsigned int cp_blks = 1 + __cp_payload(sbi);
 697         block_t cp_blk_no;
 698         int i;
 699
 700         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 701         if (!sbi->ckpt)
 702                 return -ENOMEM;
 703         /*
 704          * Finding out valid cp block involves read both
 705          * sets( cp pack1 and cp pack 2)
 706          */
 707         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 708         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 709
 710         /* The second checkpoint pack should start at the next segment */
 711         cp_start_blk_no += ((unsigned long long)1) <<
 712                                 le32_to_cpu(fsb->log_blocks_per_seg);
 713         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 714
 715         if (cp1 && cp2) {
 716                 if (ver_after(cp2_version, cp1_version))
 717                         cur_page = cp2;
 718                 else
 719                         cur_page = cp1;
 720         } else if (cp1) {
 721                 cur_page = cp1;
 722         } else if (cp2) {
 723                 cur_page = cp2;
 724         } else {
 725                 goto fail_no_cp;
 726         }
 727
 728         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 729         memcpy(sbi->ckpt, cp_block, blk_size);
 730
 731         /* Sanity checking of checkpoint */
 732         if (sanity_check_ckpt(sbi))
 733                 goto fail_no_cp;
 734
 735         if (cp_blks <= 1)
 736                 goto done;
 737
 738         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 739         if (cur_page == cp2)
 740                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 741
 742         for (i = 1; i < cp_blks; i++) {
 743                 void *sit_bitmap_ptr;
 744                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 745
 746                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 747                 sit_bitmap_ptr = page_address(cur_page);
 748                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 749                 f2fs_put_page(cur_page, 1);
 750         }
 751 done:
 752         f2fs_put_page(cp1, 1);
 753         f2fs_put_page(cp2, 1);
 754         return 0;
 755
 756 fail_no_cp:
 757         kfree(sbi->ckpt);
 758         return -EINVAL;
 759 }
 760
 761 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 762 {
 763         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 764         struct f2fs_inode_info *fi = F2FS_I(inode);
 765         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 766
 767         if (is_inode_flag_set(fi, flag))
 768                 return;
 769
 770         set_inode_flag(fi, flag);
 771         list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
 772         stat_inc_dirty_inode(sbi, type);
 773 }
 774
 775 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 776 {
 777         struct f2fs_inode_info *fi = F2FS_I(inode);
 778         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 779
 780         if (get_dirty_pages(inode) ||
 781                         !is_inode_flag_set(F2FS_I(inode), flag))
 782                 return;
 783
 784         list_del_init(&fi->dirty_list);
 785         clear_inode_flag(fi, flag);
 786         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 787 }
 788
 789 void update_dirty_page(struct inode *inode, struct page *page)
 790 {
 791         struct f2fs_sb_info *sbi = F2FS_I_SB(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         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) {
 799                 spin_lock(&sbi->inode_lock[type]);
 800                 __add_dirty_inode(inode, type);
 801                 spin_unlock(&sbi->inode_lock[type]);
 802         }
 803
 804         inode_inc_dirty_pages(inode);
 805         SetPagePrivate(page);
 806         f2fs_trace_pid(page);
 807 }
 808
 809 void remove_dirty_inode(struct inode *inode)
 810 {
 811         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 812         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 813
 814         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 815                         !S_ISLNK(inode->i_mode))
 816                 return;
 817
 818         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 819                 return;
 820
 821         spin_lock(&sbi->inode_lock[type]);
 822         __remove_dirty_inode(inode, type);
 823         spin_unlock(&sbi->inode_lock[type]);
 824 }
 825
 826 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 827 {
 828         struct list_head *head;
 829         struct inode *inode;
 830         struct f2fs_inode_info *fi;
 831         bool is_dir = (type == DIR_INODE);
 832
 833         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 834                                 get_pages(sbi, is_dir ?
 835                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 836 retry:
 837         if (unlikely(f2fs_cp_error(sbi)))
 838                 return -EIO;
 839
 840         spin_lock(&sbi->inode_lock[type]);
 841
 842         head = &sbi->inode_list[type];
 843         if (list_empty(head)) {
 844                 spin_unlock(&sbi->inode_lock[type]);
 845                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 846                                 get_pages(sbi, is_dir ?
 847                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 848                 return 0;
 849         }
 850         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 851         inode = igrab(&fi->vfs_inode);
 852         spin_unlock(&sbi->inode_lock[type]);
 853         if (inode) {
 854                 filemap_fdatawrite(inode->i_mapping);
 855                 iput(inode);
 856         } else {
 857                 /*
 858                  * We should submit bio, since it exists several
 859                  * wribacking dentry pages in the freeing inode.
 860                  */
 861                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 862                 cond_resched();
 863         }
 864         goto retry;
 865 }
 866
 867 /*
 868  * Freeze all the FS-operations for checkpoint.
 869  */
 870 static int block_operations(struct f2fs_sb_info *sbi)
 871 {
 872         struct writeback_control wbc = {
 873                 .sync_mode = WB_SYNC_ALL,
 874                 .nr_to_write = LONG_MAX,
 875                 .for_reclaim = 0,
 876         };
 877         struct blk_plug plug;
 878         int err = 0;
 879
 880         blk_start_plug(&plug);
 881
 882 retry_flush_dents:
 883         f2fs_lock_all(sbi);
 884         /* write all the dirty dentry pages */
 885         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 886                 f2fs_unlock_all(sbi);
 887                 err = sync_dirty_inodes(sbi, DIR_INODE);
 888                 if (err)
 889                         goto out;
 890                 goto retry_flush_dents;
 891         }
 892
 893         /*
 894          * POR: we should ensure that there are no dirty node pages
 895          * until finishing nat/sit flush.
 896          */
 897 retry_flush_nodes:
 898         down_write(&sbi->node_write);
 899
 900         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 901                 up_write(&sbi->node_write);
 902                 err = sync_node_pages(sbi, &wbc);
 903                 if (err) {
 904                         f2fs_unlock_all(sbi);
 905                         goto out;
 906                 }
 907                 goto retry_flush_nodes;
 908         }
 909 out:
 910         blk_finish_plug(&plug);
 911         return err;
 912 }
 913
 914 static void unblock_operations(struct f2fs_sb_info *sbi)
 915 {
 916         up_write(&sbi->node_write);
 917         f2fs_unlock_all(sbi);
 918 }
 919
 920 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 921 {
 922         DEFINE_WAIT(wait);
 923
 924         for (;;) {
 925                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 926
 927                 if (!atomic_read(&sbi->nr_wb_bios))
 928                         break;
 929
 930                 io_schedule_timeout(5*HZ);
 931         }
 932         finish_wait(&sbi->cp_wait, &wait);
 933 }
 934
 935 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 936 {
 937         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 938         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 939         struct f2fs_nm_info *nm_i = NM_I(sbi);
 940         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 941         nid_t last_nid = nm_i->next_scan_nid;
 942         block_t start_blk;
 943         unsigned int data_sum_blocks, orphan_blocks;
 944         __u32 crc32 = 0;
 945         int i;
 946         int cp_payload_blks = __cp_payload(sbi);
 947         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
 948         bool invalidate = false;
 949         struct super_block *sb = sbi->sb;
 950         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
 951         u64 kbytes_written;
 952
 953         /*
 954          * This avoids to conduct wrong roll-forward operations and uses
 955          * metapages, so should be called prior to sync_meta_pages below.
 956          */
 957         if (discard_next_dnode(sbi, discard_blk))
 958                 invalidate = true;
 959
 960         /* Flush all the NAT/SIT pages */
 961         while (get_pages(sbi, F2FS_DIRTY_META)) {
 962                 sync_meta_pages(sbi, META, LONG_MAX);
 963                 if (unlikely(f2fs_cp_error(sbi)))
 964                         return -EIO;
 965         }
 966
 967         next_free_nid(sbi, &last_nid);
 968
 969         /*
 970          * modify checkpoint
 971          * version number is already updated
 972          */
 973         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 974         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 975         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 976         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 977                 ckpt->cur_node_segno[i] =
 978                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 979                 ckpt->cur_node_blkoff[i] =
 980                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 981                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 982                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 983         }
 984         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 985                 ckpt->cur_data_segno[i] =
 986                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 987                 ckpt->cur_data_blkoff[i] =
 988                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 989                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 990                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 991         }
 992
 993         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 994         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 995         ckpt->next_free_nid = cpu_to_le32(last_nid);
 996
 997         /* 2 cp  + n data seg summary + orphan inode blocks */
 998         data_sum_blocks = npages_for_summary_flush(sbi, false);
 999         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1000                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1001         else
1002                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1003
1004         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1005         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1006                         orphan_blocks);
1007
1008         if (__remain_node_summaries(cpc->reason))
1009                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1010                                 cp_payload_blks + data_sum_blocks +
1011                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1012         else
1013                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1014                                 cp_payload_blks + data_sum_blocks +
1015                                 orphan_blocks);
1016
1017         if (cpc->reason == CP_UMOUNT)
1018                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1019         else
1020                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1021
1022         if (cpc->reason == CP_FASTBOOT)
1023                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1024         else
1025                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1026
1027         if (orphan_num)
1028                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1029         else
1030                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1031
1032         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1033                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1034
1035         /* update SIT/NAT bitmap */
1036         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1037         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1038
1039         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1040         *((__le32 *)((unsigned char *)ckpt +
1041                                 le32_to_cpu(ckpt->checksum_offset)))
1042                                 = cpu_to_le32(crc32);
1043
1044         start_blk = __start_cp_addr(sbi);
1045
1046         /* need to wait for end_io results */
1047         wait_on_all_pages_writeback(sbi);
1048         if (unlikely(f2fs_cp_error(sbi)))
1049                 return -EIO;
1050
1051         /* write out checkpoint buffer at block 0 */
1052         update_meta_page(sbi, ckpt, start_blk++);
1053
1054         for (i = 1; i < 1 + cp_payload_blks; i++)
1055                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1056                                                         start_blk++);
1057
1058         if (orphan_num) {
1059                 write_orphan_inodes(sbi, start_blk);
1060                 start_blk += orphan_blocks;
1061         }
1062
1063         write_data_summaries(sbi, start_blk);
1064         start_blk += data_sum_blocks;
1065
1066         /* Record write statistics in the hot node summary */
1067         kbytes_written = sbi->kbytes_written;
1068         if (sb->s_bdev->bd_part)
1069                 kbytes_written += BD_PART_WRITTEN(sbi);
1070
1071         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1072
1073         if (__remain_node_summaries(cpc->reason)) {
1074                 write_node_summaries(sbi, start_blk);
1075                 start_blk += NR_CURSEG_NODE_TYPE;
1076         }
1077
1078         /* writeout checkpoint block */
1079         update_meta_page(sbi, ckpt, start_blk);
1080
1081         /* wait for previous submitted node/meta pages writeback */
1082         wait_on_all_pages_writeback(sbi);
1083
1084         if (unlikely(f2fs_cp_error(sbi)))
1085                 return -EIO;
1086
1087         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1088         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1089
1090         /* update user_block_counts */
1091         sbi->last_valid_block_count = sbi->total_valid_block_count;
1092         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1093
1094         /* Here, we only have one bio having CP pack */
1095         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1096
1097         /* wait for previous submitted meta pages writeback */
1098         wait_on_all_pages_writeback(sbi);
1099
1100         /*
1101          * invalidate meta page which is used temporarily for zeroing out
1102          * block at the end of warm node chain.
1103          */
1104         if (invalidate)
1105                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1106                                                                 discard_blk);
1107
1108         release_ino_entry(sbi, false);
1109
1110         if (unlikely(f2fs_cp_error(sbi)))
1111                 return -EIO;
1112
1113         clear_prefree_segments(sbi, cpc);
1114         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1115
1116         return 0;
1117 }
1118
1119 /*
1120  * We guarantee that this checkpoint procedure will not fail.
1121  */
1122 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1123 {
1124         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1125         unsigned long long ckpt_ver;
1126         int err = 0;
1127
1128         mutex_lock(&sbi->cp_mutex);
1129
1130         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1131                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1132                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1133                 goto out;
1134         if (unlikely(f2fs_cp_error(sbi))) {
1135                 err = -EIO;
1136                 goto out;
1137         }
1138         if (f2fs_readonly(sbi->sb)) {
1139                 err = -EROFS;
1140                 goto out;
1141         }
1142
1143         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1144
1145         err = block_operations(sbi);
1146         if (err)
1147                 goto out;
1148
1149         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1150
1151         f2fs_flush_merged_bios(sbi);
1152
1153         /*
1154          * update checkpoint pack index
1155          * Increase the version number so that
1156          * SIT entries and seg summaries are written at correct place
1157          */
1158         ckpt_ver = cur_cp_version(ckpt);
1159         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1160
1161         /* write cached NAT/SIT entries to NAT/SIT area */
1162         flush_nat_entries(sbi);
1163         flush_sit_entries(sbi, cpc);
1164
1165         /* unlock all the fs_lock[] in do_checkpoint() */
1166         err = do_checkpoint(sbi, cpc);
1167
1168         unblock_operations(sbi);
1169         stat_inc_cp_count(sbi->stat_info);
1170
1171         if (cpc->reason == CP_RECOVERY)
1172                 f2fs_msg(sbi->sb, KERN_NOTICE,
1173                         "checkpoint: version = %llx", ckpt_ver);
1174
1175         /* do checkpoint periodically */
1176         f2fs_update_time(sbi, CP_TIME);
1177         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1178 out:
1179         mutex_unlock(&sbi->cp_mutex);
1180         return err;
1181 }
1182
1183 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1184 {
1185         int i;
1186
1187         for (i = 0; i < MAX_INO_ENTRY; i++) {
1188                 struct inode_management *im = &sbi->im[i];
1189
1190                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1191                 spin_lock_init(&im->ino_lock);
1192                 INIT_LIST_HEAD(&im->ino_list);
1193                 im->ino_num = 0;
1194         }
1195
1196         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1197                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1198                                 F2FS_ORPHANS_PER_BLOCK;
1199 }
1200
1201 int __init create_checkpoint_caches(void)
1202 {
1203         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1204                         sizeof(struct ino_entry));
1205         if (!ino_entry_slab)
1206                 return -ENOMEM;
1207         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1208                         sizeof(struct inode_entry));
1209         if (!inode_entry_slab) {
1210                 kmem_cache_destroy(ino_entry_slab);
1211                 return -ENOMEM;
1212         }
1213         return 0;
1214 }
1215
1216 void destroy_checkpoint_caches(void)
1217 {
1218         kmem_cache_destroy(ino_entry_slab);
1219         kmem_cache_destroy(inode_entry_slab);
1220 }