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                 .rw = READ_SYNC | REQ_META | REQ_PRIO,
  67                 .old_blkaddr = index,
  68                 .new_blkaddr = index,
  69                 .encrypted_page = NULL,
  70         };
  71
  72         if (unlikely(!is_meta))
  73                 fio.rw &= ~REQ_META;
  74 repeat:
  75         page = f2fs_grab_cache_page(mapping, index, false);
  76         if (!page) {
  77                 cond_resched();
  78                 goto repeat;
  79         }
  80         if (PageUptodate(page))
  81                 goto out;
  82
  83         fio.page = page;
  84
  85         if (f2fs_submit_page_bio(&fio)) {
  86                 f2fs_put_page(page, 1);
  87                 goto repeat;
  88         }
  89
  90         lock_page(page);
  91         if (unlikely(page->mapping != mapping)) {
  92                 f2fs_put_page(page, 1);
  93                 goto repeat;
  94         }
  95
  96         /*
  97          * if there is any IO error when accessing device, make our filesystem
  98          * readonly and make sure do not write checkpoint with non-uptodate
  99          * meta page.
 100          */
 101         if (unlikely(!PageUptodate(page)))
 102                 f2fs_stop_checkpoint(sbi, false);
 103 out:
 104         return page;
 105 }
 106
 107 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 108 {
 109         return __get_meta_page(sbi, index, true);
 110 }
 111
 112 /* for POR only */
 113 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 114 {
 115         return __get_meta_page(sbi, index, false);
 116 }
 117
 118 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 119 {
 120         switch (type) {
 121         case META_NAT:
 122                 break;
 123         case META_SIT:
 124                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 125                         return false;
 126                 break;
 127         case META_SSA:
 128                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 129                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 130                         return false;
 131                 break;
 132         case META_CP:
 133                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 134                         blkaddr < __start_cp_addr(sbi)))
 135                         return false;
 136                 break;
 137         case META_POR:
 138                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 139                         blkaddr < MAIN_BLKADDR(sbi)))
 140                         return false;
 141                 break;
 142         default:
 143                 BUG();
 144         }
 145
 146         return true;
 147 }
 148
 149 /*
 150  * Readahead CP/NAT/SIT/SSA pages
 151  */
 152 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 153                                                         int type, bool sync)
 154 {
 155         struct page *page;
 156         block_t blkno = start;
 157         struct f2fs_io_info fio = {
 158                 .sbi = sbi,
 159                 .type = META,
 160                 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
 161                 .encrypted_page = NULL,
 162         };
 163         struct blk_plug plug;
 164
 165         if (unlikely(type == META_POR))
 166                 fio.rw &= ~REQ_META;
 167
 168         blk_start_plug(&plug);
 169         for (; nrpages-- > 0; blkno++) {
 170
 171                 if (!is_valid_blkaddr(sbi, blkno, type))
 172                         goto out;
 173
 174                 switch (type) {
 175                 case META_NAT:
 176                         if (unlikely(blkno >=
 177                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 178                                 blkno = 0;
 179                         /* get nat block addr */
 180                         fio.new_blkaddr = current_nat_addr(sbi,
 181                                         blkno * NAT_ENTRY_PER_BLOCK);
 182                         break;
 183                 case META_SIT:
 184                         /* get sit block addr */
 185                         fio.new_blkaddr = current_sit_addr(sbi,
 186                                         blkno * SIT_ENTRY_PER_BLOCK);
 187                         break;
 188                 case META_SSA:
 189                 case META_CP:
 190                 case META_POR:
 191                         fio.new_blkaddr = blkno;
 192                         break;
 193                 default:
 194                         BUG();
 195                 }
 196
 197                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 198                                                 fio.new_blkaddr, false);
 199                 if (!page)
 200                         continue;
 201                 if (PageUptodate(page)) {
 202                         f2fs_put_page(page, 1);
 203                         continue;
 204                 }
 205
 206                 fio.page = page;
 207                 fio.old_blkaddr = fio.new_blkaddr;
 208                 f2fs_submit_page_mbio(&fio);
 209                 f2fs_put_page(page, 0);
 210         }
 211 out:
 212         f2fs_submit_merged_bio(sbi, META, READ);
 213         blk_finish_plug(&plug);
 214         return blkno - start;
 215 }
 216
 217 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 218 {
 219         struct page *page;
 220         bool readahead = false;
 221
 222         page = find_get_page(META_MAPPING(sbi), index);
 223         if (!page || !PageUptodate(page))
 224                 readahead = true;
 225         f2fs_put_page(page, 0);
 226
 227         if (readahead)
 228                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 229 }
 230
 231 static int f2fs_write_meta_page(struct page *page,
 232                                 struct writeback_control *wbc)
 233 {
 234         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 235
 236         trace_f2fs_writepage(page, META);
 237
 238         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 239                 goto redirty_out;
 240         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 241                 goto redirty_out;
 242         if (unlikely(f2fs_cp_error(sbi)))
 243                 goto redirty_out;
 244
 245         write_meta_page(sbi, page);
 246         dec_page_count(sbi, F2FS_DIRTY_META);
 247
 248         if (wbc->for_reclaim)
 249                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
 250
 251         unlock_page(page);
 252
 253         if (unlikely(f2fs_cp_error(sbi)))
 254                 f2fs_submit_merged_bio(sbi, META, WRITE);
 255
 256         return 0;
 257
 258 redirty_out:
 259         redirty_page_for_writepage(wbc, page);
 260         return AOP_WRITEPAGE_ACTIVATE;
 261 }
 262
 263 static int f2fs_write_meta_pages(struct address_space *mapping,
 264                                 struct writeback_control *wbc)
 265 {
 266         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 267         long diff, written;
 268
 269         /* collect a number of dirty meta pages and write together */
 270         if (wbc->for_kupdate ||
 271                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 272                 goto skip_write;
 273
 274         trace_f2fs_writepages(mapping->host, wbc, META);
 275
 276         /* if mounting is failed, skip writing node pages */
 277         mutex_lock(&sbi->cp_mutex);
 278         diff = nr_pages_to_write(sbi, META, wbc);
 279         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 280         mutex_unlock(&sbi->cp_mutex);
 281         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 282         return 0;
 283
 284 skip_write:
 285         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 286         trace_f2fs_writepages(mapping->host, wbc, META);
 287         return 0;
 288 }
 289
 290 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 291                                                 long nr_to_write)
 292 {
 293         struct address_space *mapping = META_MAPPING(sbi);
 294         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 295         struct pagevec pvec;
 296         long nwritten = 0;
 297         struct writeback_control wbc = {
 298                 .for_reclaim = 0,
 299         };
 300         struct blk_plug plug;
 301
 302         pagevec_init(&pvec, 0);
 303
 304         blk_start_plug(&plug);
 305
 306         while (index <= end) {
 307                 int i, nr_pages;
 308                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 309                                 PAGECACHE_TAG_DIRTY,
 310                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 311                 if (unlikely(nr_pages == 0))
 312                         break;
 313
 314                 for (i = 0; i < nr_pages; i++) {
 315                         struct page *page = pvec.pages[i];
 316
 317                         if (prev == ULONG_MAX)
 318                                 prev = page->index - 1;
 319                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 320                                 pagevec_release(&pvec);
 321                                 goto stop;
 322                         }
 323
 324                         lock_page(page);
 325
 326                         if (unlikely(page->mapping != mapping)) {
 327 continue_unlock:
 328                                 unlock_page(page);
 329                                 continue;
 330                         }
 331                         if (!PageDirty(page)) {
 332                                 /* someone wrote it for us */
 333                                 goto continue_unlock;
 334                         }
 335
 336                         f2fs_wait_on_page_writeback(page, META, true);
 337
 338                         BUG_ON(PageWriteback(page));
 339                         if (!clear_page_dirty_for_io(page))
 340                                 goto continue_unlock;
 341
 342                         if (mapping->a_ops->writepage(page, &wbc)) {
 343                                 unlock_page(page);
 344                                 break;
 345                         }
 346                         nwritten++;
 347                         prev = page->index;
 348                         if (unlikely(nwritten >= nr_to_write))
 349                                 break;
 350                 }
 351                 pagevec_release(&pvec);
 352                 cond_resched();
 353         }
 354 stop:
 355         if (nwritten)
 356                 f2fs_submit_merged_bio(sbi, type, WRITE);
 357
 358         blk_finish_plug(&plug);
 359
 360         return nwritten;
 361 }
 362
 363 static int f2fs_set_meta_page_dirty(struct page *page)
 364 {
 365         trace_f2fs_set_page_dirty(page, META);
 366
 367         SetPageUptodate(page);
 368         if (!PageDirty(page)) {
 369                 __set_page_dirty_nobuffers(page);
 370                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 371                 SetPagePrivate(page);
 372                 f2fs_trace_pid(page);
 373                 return 1;
 374         }
 375         return 0;
 376 }
 377
 378 const struct address_space_operations f2fs_meta_aops = {
 379         .writepage      = f2fs_write_meta_page,
 380         .writepages     = f2fs_write_meta_pages,
 381         .set_page_dirty = f2fs_set_meta_page_dirty,
 382         .invalidatepage = f2fs_invalidate_page,
 383         .releasepage    = f2fs_release_page,
 384 };
 385
 386 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 387 {
 388         struct inode_management *im = &sbi->im[type];
 389         struct ino_entry *e, *tmp;
 390
 391         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 392 retry:
 393         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 394
 395         spin_lock(&im->ino_lock);
 396         e = radix_tree_lookup(&im->ino_root, ino);
 397         if (!e) {
 398                 e = tmp;
 399                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 400                         spin_unlock(&im->ino_lock);
 401                         radix_tree_preload_end();
 402                         goto retry;
 403                 }
 404                 memset(e, 0, sizeof(struct ino_entry));
 405                 e->ino = ino;
 406
 407                 list_add_tail(&e->list, &im->ino_list);
 408                 if (type != ORPHAN_INO)
 409                         im->ino_num++;
 410         }
 411         spin_unlock(&im->ino_lock);
 412         radix_tree_preload_end();
 413
 414         if (e != tmp)
 415                 kmem_cache_free(ino_entry_slab, tmp);
 416 }
 417
 418 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 419 {
 420         struct inode_management *im = &sbi->im[type];
 421         struct ino_entry *e;
 422
 423         spin_lock(&im->ino_lock);
 424         e = radix_tree_lookup(&im->ino_root, ino);
 425         if (e) {
 426                 list_del(&e->list);
 427                 radix_tree_delete(&im->ino_root, ino);
 428                 im->ino_num--;
 429                 spin_unlock(&im->ino_lock);
 430                 kmem_cache_free(ino_entry_slab, e);
 431                 return;
 432         }
 433         spin_unlock(&im->ino_lock);
 434 }
 435
 436 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 437 {
 438         /* add new dirty ino entry into list */
 439         __add_ino_entry(sbi, ino, type);
 440 }
 441
 442 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 443 {
 444         /* remove dirty ino entry from list */
 445         __remove_ino_entry(sbi, ino, type);
 446 }
 447
 448 /* mode should be APPEND_INO or UPDATE_INO */
 449 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 450 {
 451         struct inode_management *im = &sbi->im[mode];
 452         struct ino_entry *e;
 453
 454         spin_lock(&im->ino_lock);
 455         e = radix_tree_lookup(&im->ino_root, ino);
 456         spin_unlock(&im->ino_lock);
 457         return e ? true : false;
 458 }
 459
 460 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 461 {
 462         struct ino_entry *e, *tmp;
 463         int i;
 464
 465         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 466                 struct inode_management *im = &sbi->im[i];
 467
 468                 spin_lock(&im->ino_lock);
 469                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 470                         list_del(&e->list);
 471                         radix_tree_delete(&im->ino_root, e->ino);
 472                         kmem_cache_free(ino_entry_slab, e);
 473                         im->ino_num--;
 474                 }
 475                 spin_unlock(&im->ino_lock);
 476         }
 477 }
 478
 479 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 480 {
 481         struct inode_management *im = &sbi->im[ORPHAN_INO];
 482         int err = 0;
 483
 484         spin_lock(&im->ino_lock);
 485
 486 #ifdef CONFIG_F2FS_FAULT_INJECTION
 487         if (time_to_inject(FAULT_ORPHAN)) {
 488                 spin_unlock(&im->ino_lock);
 489                 return -ENOSPC;
 490         }
 491 #endif
 492         if (unlikely(im->ino_num >= sbi->max_orphans))
 493                 err = -ENOSPC;
 494         else
 495                 im->ino_num++;
 496         spin_unlock(&im->ino_lock);
 497
 498         return err;
 499 }
 500
 501 void release_orphan_inode(struct f2fs_sb_info *sbi)
 502 {
 503         struct inode_management *im = &sbi->im[ORPHAN_INO];
 504
 505         spin_lock(&im->ino_lock);
 506         f2fs_bug_on(sbi, im->ino_num == 0);
 507         im->ino_num--;
 508         spin_unlock(&im->ino_lock);
 509 }
 510
 511 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 512 {
 513         /* add new orphan ino entry into list */
 514         __add_ino_entry(sbi, ino, ORPHAN_INO);
 515 }
 516
 517 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 518 {
 519         /* remove orphan entry from orphan list */
 520         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 521 }
 522
 523 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 524 {
 525         struct inode *inode;
 526
 527         inode = f2fs_iget(sbi->sb, ino);
 528         if (IS_ERR(inode)) {
 529                 /*
 530                  * there should be a bug that we can't find the entry
 531                  * to orphan inode.
 532                  */
 533                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 534                 return PTR_ERR(inode);
 535         }
 536
 537         clear_nlink(inode);
 538
 539         /* truncate all the data during iput */
 540         iput(inode);
 541         return 0;
 542 }
 543
 544 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 545 {
 546         block_t start_blk, orphan_blocks, i, j;
 547         int err;
 548
 549         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 550                 return 0;
 551
 552         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 553         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 554
 555         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 556
 557         for (i = 0; i < orphan_blocks; i++) {
 558                 struct page *page = get_meta_page(sbi, start_blk + i);
 559                 struct f2fs_orphan_block *orphan_blk;
 560
 561                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 562                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 563                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 564                         err = recover_orphan_inode(sbi, ino);
 565                         if (err) {
 566                                 f2fs_put_page(page, 1);
 567                                 return err;
 568                         }
 569                 }
 570                 f2fs_put_page(page, 1);
 571         }
 572         /* clear Orphan Flag */
 573         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 574         return 0;
 575 }
 576
 577 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 578 {
 579         struct list_head *head;
 580         struct f2fs_orphan_block *orphan_blk = NULL;
 581         unsigned int nentries = 0;
 582         unsigned short index = 1;
 583         unsigned short orphan_blocks;
 584         struct page *page = NULL;
 585         struct ino_entry *orphan = NULL;
 586         struct inode_management *im = &sbi->im[ORPHAN_INO];
 587
 588         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 589
 590         /*
 591          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 592          * orphan inode operations are covered under f2fs_lock_op().
 593          * And, spin_lock should be avoided due to page operations below.
 594          */
 595         head = &im->ino_list;
 596
 597         /* loop for each orphan inode entry and write them in Jornal block */
 598         list_for_each_entry(orphan, head, list) {
 599                 if (!page) {
 600                         page = grab_meta_page(sbi, start_blk++);
 601                         orphan_blk =
 602                                 (struct f2fs_orphan_block *)page_address(page);
 603                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 604                 }
 605
 606                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 607
 608                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 609                         /*
 610                          * an orphan block is full of 1020 entries,
 611                          * then we need to flush current orphan blocks
 612                          * and bring another one in memory
 613                          */
 614                         orphan_blk->blk_addr = cpu_to_le16(index);
 615                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 616                         orphan_blk->entry_count = cpu_to_le32(nentries);
 617                         set_page_dirty(page);
 618                         f2fs_put_page(page, 1);
 619                         index++;
 620                         nentries = 0;
 621                         page = NULL;
 622                 }
 623         }
 624
 625         if (page) {
 626                 orphan_blk->blk_addr = cpu_to_le16(index);
 627                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 628                 orphan_blk->entry_count = cpu_to_le32(nentries);
 629                 set_page_dirty(page);
 630                 f2fs_put_page(page, 1);
 631         }
 632 }
 633
 634 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 635                                 block_t cp_addr, unsigned long long *version)
 636 {
 637         struct page *cp_page_1, *cp_page_2 = NULL;
 638         unsigned long blk_size = sbi->blocksize;
 639         struct f2fs_checkpoint *cp_block;
 640         unsigned long long cur_version = 0, pre_version = 0;
 641         size_t crc_offset;
 642         __u32 crc = 0;
 643
 644         /* Read the 1st cp block in this CP pack */
 645         cp_page_1 = get_meta_page(sbi, cp_addr);
 646
 647         /* get the version number */
 648         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 649         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 650         if (crc_offset >= blk_size)
 651                 goto invalid_cp1;
 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_cp1;
 656
 657         pre_version = cur_cp_version(cp_block);
 658
 659         /* Read the 2nd cp block in this CP pack */
 660         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 661         cp_page_2 = get_meta_page(sbi, cp_addr);
 662
 663         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 664         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 665         if (crc_offset >= blk_size)
 666                 goto invalid_cp2;
 667
 668         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 669         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 670                 goto invalid_cp2;
 671
 672         cur_version = cur_cp_version(cp_block);
 673
 674         if (cur_version == pre_version) {
 675                 *version = cur_version;
 676                 f2fs_put_page(cp_page_2, 1);
 677                 return cp_page_1;
 678         }
 679 invalid_cp2:
 680         f2fs_put_page(cp_page_2, 1);
 681 invalid_cp1:
 682         f2fs_put_page(cp_page_1, 1);
 683         return NULL;
 684 }
 685
 686 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 687 {
 688         struct f2fs_checkpoint *cp_block;
 689         struct f2fs_super_block *fsb = sbi->raw_super;
 690         struct page *cp1, *cp2, *cur_page;
 691         unsigned long blk_size = sbi->blocksize;
 692         unsigned long long cp1_version = 0, cp2_version = 0;
 693         unsigned long long cp_start_blk_no;
 694         unsigned int cp_blks = 1 + __cp_payload(sbi);
 695         block_t cp_blk_no;
 696         int i;
 697
 698         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 699         if (!sbi->ckpt)
 700                 return -ENOMEM;
 701         /*
 702          * Finding out valid cp block involves read both
 703          * sets( cp pack1 and cp pack 2)
 704          */
 705         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 706         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 707
 708         /* The second checkpoint pack should start at the next segment */
 709         cp_start_blk_no += ((unsigned long long)1) <<
 710                                 le32_to_cpu(fsb->log_blocks_per_seg);
 711         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 712
 713         if (cp1 && cp2) {
 714                 if (ver_after(cp2_version, cp1_version))
 715                         cur_page = cp2;
 716                 else
 717                         cur_page = cp1;
 718         } else if (cp1) {
 719                 cur_page = cp1;
 720         } else if (cp2) {
 721                 cur_page = cp2;
 722         } else {
 723                 goto fail_no_cp;
 724         }
 725
 726         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 727         memcpy(sbi->ckpt, cp_block, blk_size);
 728
 729         /* Sanity checking of checkpoint */
 730         if (sanity_check_ckpt(sbi))
 731                 goto fail_no_cp;
 732
 733         if (cp_blks <= 1)
 734                 goto done;
 735
 736         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 737         if (cur_page == cp2)
 738                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 739
 740         for (i = 1; i < cp_blks; i++) {
 741                 void *sit_bitmap_ptr;
 742                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 743
 744                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 745                 sit_bitmap_ptr = page_address(cur_page);
 746                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 747                 f2fs_put_page(cur_page, 1);
 748         }
 749 done:
 750         f2fs_put_page(cp1, 1);
 751         f2fs_put_page(cp2, 1);
 752         return 0;
 753
 754 fail_no_cp:
 755         kfree(sbi->ckpt);
 756         return -EINVAL;
 757 }
 758
 759 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 760 {
 761         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 762         struct f2fs_inode_info *fi = F2FS_I(inode);
 763         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 764
 765         if (is_inode_flag_set(fi, flag))
 766                 return;
 767
 768         set_inode_flag(fi, flag);
 769         list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
 770         stat_inc_dirty_inode(sbi, type);
 771 }
 772
 773 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 774 {
 775         struct f2fs_inode_info *fi = F2FS_I(inode);
 776         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 777
 778         if (get_dirty_pages(inode) ||
 779                         !is_inode_flag_set(F2FS_I(inode), flag))
 780                 return;
 781
 782         list_del_init(&fi->dirty_list);
 783         clear_inode_flag(fi, flag);
 784         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 785 }
 786
 787 void update_dirty_page(struct inode *inode, struct page *page)
 788 {
 789         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 790         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 791
 792         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 793                         !S_ISLNK(inode->i_mode))
 794                 return;
 795
 796         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) {
 797                 spin_lock(&sbi->inode_lock[type]);
 798                 __add_dirty_inode(inode, type);
 799                 spin_unlock(&sbi->inode_lock[type]);
 800         }
 801
 802         inode_inc_dirty_pages(inode);
 803         SetPagePrivate(page);
 804         f2fs_trace_pid(page);
 805 }
 806
 807 void remove_dirty_inode(struct inode *inode)
 808 {
 809         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 810         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 811
 812         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 813                         !S_ISLNK(inode->i_mode))
 814                 return;
 815
 816         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 817                 return;
 818
 819         spin_lock(&sbi->inode_lock[type]);
 820         __remove_dirty_inode(inode, type);
 821         spin_unlock(&sbi->inode_lock[type]);
 822 }
 823
 824 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 825 {
 826         struct list_head *head;
 827         struct inode *inode;
 828         struct f2fs_inode_info *fi;
 829         bool is_dir = (type == DIR_INODE);
 830
 831         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 832                                 get_pages(sbi, is_dir ?
 833                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 834 retry:
 835         if (unlikely(f2fs_cp_error(sbi)))
 836                 return -EIO;
 837
 838         spin_lock(&sbi->inode_lock[type]);
 839
 840         head = &sbi->inode_list[type];
 841         if (list_empty(head)) {
 842                 spin_unlock(&sbi->inode_lock[type]);
 843                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 844                                 get_pages(sbi, is_dir ?
 845                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 846                 return 0;
 847         }
 848         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 849         inode = igrab(&fi->vfs_inode);
 850         spin_unlock(&sbi->inode_lock[type]);
 851         if (inode) {
 852                 filemap_fdatawrite(inode->i_mapping);
 853                 iput(inode);
 854         } else {
 855                 /*
 856                  * We should submit bio, since it exists several
 857                  * wribacking dentry pages in the freeing inode.
 858                  */
 859                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 860                 cond_resched();
 861         }
 862         goto retry;
 863 }
 864
 865 /*
 866  * Freeze all the FS-operations for checkpoint.
 867  */
 868 static int block_operations(struct f2fs_sb_info *sbi)
 869 {
 870         struct writeback_control wbc = {
 871                 .sync_mode = WB_SYNC_ALL,
 872                 .nr_to_write = LONG_MAX,
 873                 .for_reclaim = 0,
 874         };
 875         struct blk_plug plug;
 876         int err = 0;
 877
 878         blk_start_plug(&plug);
 879
 880 retry_flush_dents:
 881         f2fs_lock_all(sbi);
 882         /* write all the dirty dentry pages */
 883         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 884                 f2fs_unlock_all(sbi);
 885                 err = sync_dirty_inodes(sbi, DIR_INODE);
 886                 if (err)
 887                         goto out;
 888                 goto retry_flush_dents;
 889         }
 890
 891         /*
 892          * POR: we should ensure that there are no dirty node pages
 893          * until finishing nat/sit flush.
 894          */
 895 retry_flush_nodes:
 896         down_write(&sbi->node_write);
 897
 898         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 899                 up_write(&sbi->node_write);
 900                 err = sync_node_pages(sbi, &wbc);
 901                 if (err) {
 902                         f2fs_unlock_all(sbi);
 903                         goto out;
 904                 }
 905                 goto retry_flush_nodes;
 906         }
 907 out:
 908         blk_finish_plug(&plug);
 909         return err;
 910 }
 911
 912 static void unblock_operations(struct f2fs_sb_info *sbi)
 913 {
 914         up_write(&sbi->node_write);
 915         f2fs_unlock_all(sbi);
 916 }
 917
 918 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 919 {
 920         DEFINE_WAIT(wait);
 921
 922         for (;;) {
 923                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 924
 925                 if (!atomic_read(&sbi->nr_wb_bios))
 926                         break;
 927
 928                 io_schedule_timeout(5*HZ);
 929         }
 930         finish_wait(&sbi->cp_wait, &wait);
 931 }
 932
 933 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 934 {
 935         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 936         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 937         struct f2fs_nm_info *nm_i = NM_I(sbi);
 938         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 939         nid_t last_nid = nm_i->next_scan_nid;
 940         block_t start_blk;
 941         unsigned int data_sum_blocks, orphan_blocks;
 942         __u32 crc32 = 0;
 943         int i;
 944         int cp_payload_blks = __cp_payload(sbi);
 945         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
 946         bool invalidate = false;
 947         struct super_block *sb = sbi->sb;
 948         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
 949         u64 kbytes_written;
 950
 951         /*
 952          * This avoids to conduct wrong roll-forward operations and uses
 953          * metapages, so should be called prior to sync_meta_pages below.
 954          */
 955         if (discard_next_dnode(sbi, discard_blk))
 956                 invalidate = true;
 957
 958         /* Flush all the NAT/SIT pages */
 959         while (get_pages(sbi, F2FS_DIRTY_META)) {
 960                 sync_meta_pages(sbi, META, LONG_MAX);
 961                 if (unlikely(f2fs_cp_error(sbi)))
 962                         return -EIO;
 963         }
 964
 965         next_free_nid(sbi, &last_nid);
 966
 967         /*
 968          * modify checkpoint
 969          * version number is already updated
 970          */
 971         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 972         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 973         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 974         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 975                 ckpt->cur_node_segno[i] =
 976                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 977                 ckpt->cur_node_blkoff[i] =
 978                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 979                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 980                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 981         }
 982         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 983                 ckpt->cur_data_segno[i] =
 984                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 985                 ckpt->cur_data_blkoff[i] =
 986                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 987                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 988                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 989         }
 990
 991         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 992         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 993         ckpt->next_free_nid = cpu_to_le32(last_nid);
 994
 995         /* 2 cp  + n data seg summary + orphan inode blocks */
 996         data_sum_blocks = npages_for_summary_flush(sbi, false);
 997         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 998                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 999         else
1000                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1001
1002         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1003         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1004                         orphan_blocks);
1005
1006         if (__remain_node_summaries(cpc->reason))
1007                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1008                                 cp_payload_blks + data_sum_blocks +
1009                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1010         else
1011                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1012                                 cp_payload_blks + data_sum_blocks +
1013                                 orphan_blocks);
1014
1015         if (cpc->reason == CP_UMOUNT)
1016                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1017         else
1018                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1019
1020         if (cpc->reason == CP_FASTBOOT)
1021                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1022         else
1023                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1024
1025         if (orphan_num)
1026                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1027         else
1028                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1029
1030         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1031                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1032
1033         /* update SIT/NAT bitmap */
1034         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1035         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1036
1037         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1038         *((__le32 *)((unsigned char *)ckpt +
1039                                 le32_to_cpu(ckpt->checksum_offset)))
1040                                 = cpu_to_le32(crc32);
1041
1042         start_blk = __start_cp_addr(sbi);
1043
1044         /* need to wait for end_io results */
1045         wait_on_all_pages_writeback(sbi);
1046         if (unlikely(f2fs_cp_error(sbi)))
1047                 return -EIO;
1048
1049         /* write out checkpoint buffer at block 0 */
1050         update_meta_page(sbi, ckpt, start_blk++);
1051
1052         for (i = 1; i < 1 + cp_payload_blks; i++)
1053                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1054                                                         start_blk++);
1055
1056         if (orphan_num) {
1057                 write_orphan_inodes(sbi, start_blk);
1058                 start_blk += orphan_blocks;
1059         }
1060
1061         write_data_summaries(sbi, start_blk);
1062         start_blk += data_sum_blocks;
1063
1064         /* Record write statistics in the hot node summary */
1065         kbytes_written = sbi->kbytes_written;
1066         if (sb->s_bdev->bd_part)
1067                 kbytes_written += BD_PART_WRITTEN(sbi);
1068
1069         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1070
1071         if (__remain_node_summaries(cpc->reason)) {
1072                 write_node_summaries(sbi, start_blk);
1073                 start_blk += NR_CURSEG_NODE_TYPE;
1074         }
1075
1076         /* writeout checkpoint block */
1077         update_meta_page(sbi, ckpt, start_blk);
1078
1079         /* wait for previous submitted node/meta pages writeback */
1080         wait_on_all_pages_writeback(sbi);
1081
1082         if (unlikely(f2fs_cp_error(sbi)))
1083                 return -EIO;
1084
1085         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1086         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1087
1088         /* update user_block_counts */
1089         sbi->last_valid_block_count = sbi->total_valid_block_count;
1090         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1091
1092         /* Here, we only have one bio having CP pack */
1093         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1094
1095         /* wait for previous submitted meta pages writeback */
1096         wait_on_all_pages_writeback(sbi);
1097
1098         /*
1099          * invalidate meta page which is used temporarily for zeroing out
1100          * block at the end of warm node chain.
1101          */
1102         if (invalidate)
1103                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1104                                                                 discard_blk);
1105
1106         release_ino_entry(sbi, false);
1107
1108         if (unlikely(f2fs_cp_error(sbi)))
1109                 return -EIO;
1110
1111         clear_prefree_segments(sbi, cpc);
1112         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1113
1114         return 0;
1115 }
1116
1117 /*
1118  * We guarantee that this checkpoint procedure will not fail.
1119  */
1120 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1121 {
1122         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1123         unsigned long long ckpt_ver;
1124         int err = 0;
1125
1126         mutex_lock(&sbi->cp_mutex);
1127
1128         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1129                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1130                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1131                 goto out;
1132         if (unlikely(f2fs_cp_error(sbi))) {
1133                 err = -EIO;
1134                 goto out;
1135         }
1136         if (f2fs_readonly(sbi->sb)) {
1137                 err = -EROFS;
1138                 goto out;
1139         }
1140
1141         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1142
1143         err = block_operations(sbi);
1144         if (err)
1145                 goto out;
1146
1147         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1148
1149         f2fs_flush_merged_bios(sbi);
1150
1151         /*
1152          * update checkpoint pack index
1153          * Increase the version number so that
1154          * SIT entries and seg summaries are written at correct place
1155          */
1156         ckpt_ver = cur_cp_version(ckpt);
1157         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1158
1159         /* write cached NAT/SIT entries to NAT/SIT area */
1160         flush_nat_entries(sbi);
1161         flush_sit_entries(sbi, cpc);
1162
1163         /* unlock all the fs_lock[] in do_checkpoint() */
1164         err = do_checkpoint(sbi, cpc);
1165
1166         unblock_operations(sbi);
1167         stat_inc_cp_count(sbi->stat_info);
1168
1169         if (cpc->reason == CP_RECOVERY)
1170                 f2fs_msg(sbi->sb, KERN_NOTICE,
1171                         "checkpoint: version = %llx", ckpt_ver);
1172
1173         /* do checkpoint periodically */
1174         f2fs_update_time(sbi, CP_TIME);
1175         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1176 out:
1177         mutex_unlock(&sbi->cp_mutex);
1178         return err;
1179 }
1180
1181 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1182 {
1183         int i;
1184
1185         for (i = 0; i < MAX_INO_ENTRY; i++) {
1186                 struct inode_management *im = &sbi->im[i];
1187
1188                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1189                 spin_lock_init(&im->ino_lock);
1190                 INIT_LIST_HEAD(&im->ino_list);
1191                 im->ino_num = 0;
1192         }
1193
1194         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1195                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1196                                 F2FS_ORPHANS_PER_BLOCK;
1197 }
1198
1199 int __init create_checkpoint_caches(void)
1200 {
1201         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1202                         sizeof(struct ino_entry));
1203         if (!ino_entry_slab)
1204                 return -ENOMEM;
1205         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1206                         sizeof(struct inode_entry));
1207         if (!inode_entry_slab) {
1208                 kmem_cache_destroy(ino_entry_slab);
1209                 return -ENOMEM;
1210         }
1211         return 0;
1212 }
1213
1214 void destroy_checkpoint_caches(void)
1215 {
1216         kmem_cache_destroy(ino_entry_slab);
1217         kmem_cache_destroy(inode_entry_slab);
1218 }