fs/f2fs/checkpoint.c

   1 /*
   2  * fs/f2fs/checkpoint.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/bio.h>
  13 #include <linux/mpage.h>
  14 #include <linux/writeback.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/f2fs_fs.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/swap.h>
  19
  20 #include "f2fs.h"
  21 #include "node.h"
  22 #include "segment.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *inode_entry_slab;
  28
  29 /*
  30  * We guarantee no failure on the returned page.
  31  */
  32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  33 {
  34         struct address_space *mapping = META_MAPPING(sbi);
  35         struct page *page = NULL;
  36 repeat:
  37         page = grab_cache_page(mapping, index);
  38         if (!page) {
  39                 cond_resched();
  40                 goto repeat;
  41         }
  42         f2fs_wait_on_page_writeback(page, META);
  43         SetPageUptodate(page);
  44         return page;
  45 }
  46
  47 /*
  48  * We guarantee no failure on the returned page.
  49  */
  50 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         f2fs_wait_on_page_writeback(page, META);
 236         write_meta_page(sbi, page);
 237         dec_page_count(sbi, F2FS_DIRTY_META);
 238         unlock_page(page);
 239
 240         if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi)))
 241                 f2fs_submit_merged_bio(sbi, META, WRITE);
 242         return 0;
 243
 244 redirty_out:
 245         redirty_page_for_writepage(wbc, page);
 246         return AOP_WRITEPAGE_ACTIVATE;
 247 }
 248
 249 static int f2fs_write_meta_pages(struct address_space *mapping,
 250                                 struct writeback_control *wbc)
 251 {
 252         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 253         long diff, written;
 254
 255         trace_f2fs_writepages(mapping->host, wbc, META);
 256
 257         /* collect a number of dirty meta pages and write together */
 258         if (wbc->for_kupdate ||
 259                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 260                 goto skip_write;
 261
 262         /* if mounting is failed, skip writing node pages */
 263         mutex_lock(&sbi->cp_mutex);
 264         diff = nr_pages_to_write(sbi, META, wbc);
 265         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 266         mutex_unlock(&sbi->cp_mutex);
 267         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 268         return 0;
 269
 270 skip_write:
 271         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 272         return 0;
 273 }
 274
 275 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 276                                                 long nr_to_write)
 277 {
 278         struct address_space *mapping = META_MAPPING(sbi);
 279         pgoff_t index = 0, end = LONG_MAX, prev = LONG_MAX;
 280         struct pagevec pvec;
 281         long nwritten = 0;
 282         struct writeback_control wbc = {
 283                 .for_reclaim = 0,
 284         };
 285
 286         pagevec_init(&pvec, 0);
 287
 288         while (index <= end) {
 289                 int i, nr_pages;
 290                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 291                                 PAGECACHE_TAG_DIRTY,
 292                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 293                 if (unlikely(nr_pages == 0))
 294                         break;
 295
 296                 for (i = 0; i < nr_pages; i++) {
 297                         struct page *page = pvec.pages[i];
 298
 299                         if (prev == LONG_MAX)
 300                                 prev = page->index - 1;
 301                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 302                                 pagevec_release(&pvec);
 303                                 goto stop;
 304                         }
 305
 306                         lock_page(page);
 307
 308                         if (unlikely(page->mapping != mapping)) {
 309 continue_unlock:
 310                                 unlock_page(page);
 311                                 continue;
 312                         }
 313                         if (!PageDirty(page)) {
 314                                 /* someone wrote it for us */
 315                                 goto continue_unlock;
 316                         }
 317
 318                         if (!clear_page_dirty_for_io(page))
 319                                 goto continue_unlock;
 320
 321                         if (mapping->a_ops->writepage(page, &wbc)) {
 322                                 unlock_page(page);
 323                                 break;
 324                         }
 325                         nwritten++;
 326                         prev = page->index;
 327                         if (unlikely(nwritten >= nr_to_write))
 328                                 break;
 329                 }
 330                 pagevec_release(&pvec);
 331                 cond_resched();
 332         }
 333 stop:
 334         if (nwritten)
 335                 f2fs_submit_merged_bio(sbi, type, WRITE);
 336
 337         return nwritten;
 338 }
 339
 340 static int f2fs_set_meta_page_dirty(struct page *page)
 341 {
 342         trace_f2fs_set_page_dirty(page, META);
 343
 344         SetPageUptodate(page);
 345         if (!PageDirty(page)) {
 346                 __set_page_dirty_nobuffers(page);
 347                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 348                 SetPagePrivate(page);
 349                 f2fs_trace_pid(page);
 350                 return 1;
 351         }
 352         return 0;
 353 }
 354
 355 const struct address_space_operations f2fs_meta_aops = {
 356         .writepage      = f2fs_write_meta_page,
 357         .writepages     = f2fs_write_meta_pages,
 358         .set_page_dirty = f2fs_set_meta_page_dirty,
 359         .invalidatepage = f2fs_invalidate_page,
 360         .releasepage    = f2fs_release_page,
 361 };
 362
 363 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 364 {
 365         struct inode_management *im = &sbi->im[type];
 366         struct ino_entry *e, *tmp;
 367
 368         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 369 retry:
 370         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 371
 372         spin_lock(&im->ino_lock);
 373         e = radix_tree_lookup(&im->ino_root, ino);
 374         if (!e) {
 375                 e = tmp;
 376                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 377                         spin_unlock(&im->ino_lock);
 378                         radix_tree_preload_end();
 379                         goto retry;
 380                 }
 381                 memset(e, 0, sizeof(struct ino_entry));
 382                 e->ino = ino;
 383
 384                 list_add_tail(&e->list, &im->ino_list);
 385                 if (type != ORPHAN_INO)
 386                         im->ino_num++;
 387         }
 388         spin_unlock(&im->ino_lock);
 389         radix_tree_preload_end();
 390
 391         if (e != tmp)
 392                 kmem_cache_free(ino_entry_slab, tmp);
 393 }
 394
 395 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 396 {
 397         struct inode_management *im = &sbi->im[type];
 398         struct ino_entry *e;
 399
 400         spin_lock(&im->ino_lock);
 401         e = radix_tree_lookup(&im->ino_root, ino);
 402         if (e) {
 403                 list_del(&e->list);
 404                 radix_tree_delete(&im->ino_root, ino);
 405                 im->ino_num--;
 406                 spin_unlock(&im->ino_lock);
 407                 kmem_cache_free(ino_entry_slab, e);
 408                 return;
 409         }
 410         spin_unlock(&im->ino_lock);
 411 }
 412
 413 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 414 {
 415         /* add new dirty ino entry into list */
 416         __add_ino_entry(sbi, ino, type);
 417 }
 418
 419 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 420 {
 421         /* remove dirty ino entry from list */
 422         __remove_ino_entry(sbi, ino, type);
 423 }
 424
 425 /* mode should be APPEND_INO or UPDATE_INO */
 426 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 427 {
 428         struct inode_management *im = &sbi->im[mode];
 429         struct ino_entry *e;
 430
 431         spin_lock(&im->ino_lock);
 432         e = radix_tree_lookup(&im->ino_root, ino);
 433         spin_unlock(&im->ino_lock);
 434         return e ? true : false;
 435 }
 436
 437 void release_ino_entry(struct f2fs_sb_info *sbi)
 438 {
 439         struct ino_entry *e, *tmp;
 440         int i;
 441
 442         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 443                 struct inode_management *im = &sbi->im[i];
 444
 445                 spin_lock(&im->ino_lock);
 446                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 447                         list_del(&e->list);
 448                         radix_tree_delete(&im->ino_root, e->ino);
 449                         kmem_cache_free(ino_entry_slab, e);
 450                         im->ino_num--;
 451                 }
 452                 spin_unlock(&im->ino_lock);
 453         }
 454 }
 455
 456 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 457 {
 458         struct inode_management *im = &sbi->im[ORPHAN_INO];
 459         int err = 0;
 460
 461         spin_lock(&im->ino_lock);
 462         if (unlikely(im->ino_num >= sbi->max_orphans))
 463                 err = -ENOSPC;
 464         else
 465                 im->ino_num++;
 466         spin_unlock(&im->ino_lock);
 467
 468         return err;
 469 }
 470
 471 void release_orphan_inode(struct f2fs_sb_info *sbi)
 472 {
 473         struct inode_management *im = &sbi->im[ORPHAN_INO];
 474
 475         spin_lock(&im->ino_lock);
 476         f2fs_bug_on(sbi, im->ino_num == 0);
 477         im->ino_num--;
 478         spin_unlock(&im->ino_lock);
 479 }
 480
 481 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 482 {
 483         /* add new orphan ino entry into list */
 484         __add_ino_entry(sbi, ino, ORPHAN_INO);
 485 }
 486
 487 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 488 {
 489         /* remove orphan entry from orphan list */
 490         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 491 }
 492
 493 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 494 {
 495         struct inode *inode;
 496
 497         inode = f2fs_iget(sbi->sb, ino);
 498         if (IS_ERR(inode)) {
 499                 /*
 500                  * there should be a bug that we can't find the entry
 501                  * to orphan inode.
 502                  */
 503                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 504                 return PTR_ERR(inode);
 505         }
 506
 507         clear_nlink(inode);
 508
 509         /* truncate all the data during iput */
 510         iput(inode);
 511         return 0;
 512 }
 513
 514 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 515 {
 516         block_t start_blk, orphan_blocks, i, j;
 517         int err;
 518
 519         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 520                 return 0;
 521
 522         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 523         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 524
 525         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 526
 527         for (i = 0; i < orphan_blocks; i++) {
 528                 struct page *page = get_meta_page(sbi, start_blk + i);
 529                 struct f2fs_orphan_block *orphan_blk;
 530
 531                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 532                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 533                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 534                         err = recover_orphan_inode(sbi, ino);
 535                         if (err) {
 536                                 f2fs_put_page(page, 1);
 537                                 return err;
 538                         }
 539                 }
 540                 f2fs_put_page(page, 1);
 541         }
 542         /* clear Orphan Flag */
 543         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 544         return 0;
 545 }
 546
 547 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 548 {
 549         struct list_head *head;
 550         struct f2fs_orphan_block *orphan_blk = NULL;
 551         unsigned int nentries = 0;
 552         unsigned short index = 1;
 553         unsigned short orphan_blocks;
 554         struct page *page = NULL;
 555         struct ino_entry *orphan = NULL;
 556         struct inode_management *im = &sbi->im[ORPHAN_INO];
 557
 558         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 559
 560         /*
 561          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 562          * orphan inode operations are covered under f2fs_lock_op().
 563          * And, spin_lock should be avoided due to page operations below.
 564          */
 565         head = &im->ino_list;
 566
 567         /* loop for each orphan inode entry and write them in Jornal block */
 568         list_for_each_entry(orphan, head, list) {
 569                 if (!page) {
 570                         page = grab_meta_page(sbi, start_blk++);
 571                         orphan_blk =
 572                                 (struct f2fs_orphan_block *)page_address(page);
 573                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 574                 }
 575
 576                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 577
 578                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 579                         /*
 580                          * an orphan block is full of 1020 entries,
 581                          * then we need to flush current orphan blocks
 582                          * and bring another one in memory
 583                          */
 584                         orphan_blk->blk_addr = cpu_to_le16(index);
 585                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 586                         orphan_blk->entry_count = cpu_to_le32(nentries);
 587                         set_page_dirty(page);
 588                         f2fs_put_page(page, 1);
 589                         index++;
 590                         nentries = 0;
 591                         page = NULL;
 592                 }
 593         }
 594
 595         if (page) {
 596                 orphan_blk->blk_addr = cpu_to_le16(index);
 597                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 598                 orphan_blk->entry_count = cpu_to_le32(nentries);
 599                 set_page_dirty(page);
 600                 f2fs_put_page(page, 1);
 601         }
 602 }
 603
 604 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 605                                 block_t cp_addr, unsigned long long *version)
 606 {
 607         struct page *cp_page_1, *cp_page_2 = NULL;
 608         unsigned long blk_size = sbi->blocksize;
 609         struct f2fs_checkpoint *cp_block;
 610         unsigned long long cur_version = 0, pre_version = 0;
 611         size_t crc_offset;
 612         __u32 crc = 0;
 613
 614         /* Read the 1st cp block in this CP pack */
 615         cp_page_1 = get_meta_page(sbi, cp_addr);
 616
 617         /* get the version number */
 618         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 619         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 620         if (crc_offset >= blk_size)
 621                 goto invalid_cp1;
 622
 623         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 624         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 625                 goto invalid_cp1;
 626
 627         pre_version = cur_cp_version(cp_block);
 628
 629         /* Read the 2nd cp block in this CP pack */
 630         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 631         cp_page_2 = get_meta_page(sbi, cp_addr);
 632
 633         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 634         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 635         if (crc_offset >= blk_size)
 636                 goto invalid_cp2;
 637
 638         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 639         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 640                 goto invalid_cp2;
 641
 642         cur_version = cur_cp_version(cp_block);
 643
 644         if (cur_version == pre_version) {
 645                 *version = cur_version;
 646                 f2fs_put_page(cp_page_2, 1);
 647                 return cp_page_1;
 648         }
 649 invalid_cp2:
 650         f2fs_put_page(cp_page_2, 1);
 651 invalid_cp1:
 652         f2fs_put_page(cp_page_1, 1);
 653         return NULL;
 654 }
 655
 656 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 657 {
 658         struct f2fs_checkpoint *cp_block;
 659         struct f2fs_super_block *fsb = sbi->raw_super;
 660         struct page *cp1, *cp2, *cur_page;
 661         unsigned long blk_size = sbi->blocksize;
 662         unsigned long long cp1_version = 0, cp2_version = 0;
 663         unsigned long long cp_start_blk_no;
 664         unsigned int cp_blks = 1 + __cp_payload(sbi);
 665         block_t cp_blk_no;
 666         int i;
 667
 668         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 669         if (!sbi->ckpt)
 670                 return -ENOMEM;
 671         /*
 672          * Finding out valid cp block involves read both
 673          * sets( cp pack1 and cp pack 2)
 674          */
 675         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 676         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 677
 678         /* The second checkpoint pack should start at the next segment */
 679         cp_start_blk_no += ((unsigned long long)1) <<
 680                                 le32_to_cpu(fsb->log_blocks_per_seg);
 681         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 682
 683         if (cp1 && cp2) {
 684                 if (ver_after(cp2_version, cp1_version))
 685                         cur_page = cp2;
 686                 else
 687                         cur_page = cp1;
 688         } else if (cp1) {
 689                 cur_page = cp1;
 690         } else if (cp2) {
 691                 cur_page = cp2;
 692         } else {
 693                 goto fail_no_cp;
 694         }
 695
 696         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 697         memcpy(sbi->ckpt, cp_block, blk_size);
 698
 699         if (cp_blks <= 1)
 700                 goto done;
 701
 702         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 703         if (cur_page == cp2)
 704                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 705
 706         for (i = 1; i < cp_blks; i++) {
 707                 void *sit_bitmap_ptr;
 708                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 709
 710                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 711                 sit_bitmap_ptr = page_address(cur_page);
 712                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 713                 f2fs_put_page(cur_page, 1);
 714         }
 715 done:
 716         f2fs_put_page(cp1, 1);
 717         f2fs_put_page(cp2, 1);
 718         return 0;
 719
 720 fail_no_cp:
 721         kfree(sbi->ckpt);
 722         return -EINVAL;
 723 }
 724
 725 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 726 {
 727         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 728         struct f2fs_inode_info *fi = F2FS_I(inode);
 729         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 730
 731         if (is_inode_flag_set(fi, flag))
 732                 return;
 733
 734         set_inode_flag(fi, flag);
 735         list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
 736         stat_inc_dirty_inode(sbi, type);
 737 }
 738
 739 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 740 {
 741         struct f2fs_inode_info *fi = F2FS_I(inode);
 742         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 743
 744         if (get_dirty_pages(inode) ||
 745                         !is_inode_flag_set(F2FS_I(inode), flag))
 746                 return;
 747
 748         list_del_init(&fi->dirty_list);
 749         clear_inode_flag(fi, flag);
 750         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 751 }
 752
 753 void update_dirty_page(struct inode *inode, struct page *page)
 754 {
 755         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 756         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 757
 758         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 759                         !S_ISLNK(inode->i_mode))
 760                 return;
 761
 762         spin_lock(&sbi->inode_lock[type]);
 763         __add_dirty_inode(inode, type);
 764         inode_inc_dirty_pages(inode);
 765         spin_unlock(&sbi->inode_lock[type]);
 766
 767         SetPagePrivate(page);
 768         f2fs_trace_pid(page);
 769 }
 770
 771 void add_dirty_dir_inode(struct inode *inode)
 772 {
 773         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 774
 775         spin_lock(&sbi->inode_lock[DIR_INODE]);
 776         __add_dirty_inode(inode, DIR_INODE);
 777         spin_unlock(&sbi->inode_lock[DIR_INODE]);
 778 }
 779
 780 void remove_dirty_inode(struct inode *inode)
 781 {
 782         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 783         struct f2fs_inode_info *fi = F2FS_I(inode);
 784         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 785
 786         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 787                         !S_ISLNK(inode->i_mode))
 788                 return;
 789
 790         spin_lock(&sbi->inode_lock[type]);
 791         __remove_dirty_inode(inode, type);
 792         spin_unlock(&sbi->inode_lock[type]);
 793
 794         /* Only from the recovery routine */
 795         if (is_inode_flag_set(fi, FI_DELAY_IPUT)) {
 796                 clear_inode_flag(fi, FI_DELAY_IPUT);
 797                 iput(inode);
 798         }
 799 }
 800
 801 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 802 {
 803         struct list_head *head;
 804         struct inode *inode;
 805         struct f2fs_inode_info *fi;
 806         bool is_dir = (type == DIR_INODE);
 807
 808         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 809                                 get_pages(sbi, is_dir ?
 810                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 811 retry:
 812         if (unlikely(f2fs_cp_error(sbi)))
 813                 return -EIO;
 814
 815         spin_lock(&sbi->inode_lock[type]);
 816
 817         head = &sbi->inode_list[type];
 818         if (list_empty(head)) {
 819                 spin_unlock(&sbi->inode_lock[type]);
 820                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 821                                 get_pages(sbi, is_dir ?
 822                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 823                 return 0;
 824         }
 825         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 826         inode = igrab(&fi->vfs_inode);
 827         spin_unlock(&sbi->inode_lock[type]);
 828         if (inode) {
 829                 filemap_fdatawrite(inode->i_mapping);
 830                 iput(inode);
 831         } else {
 832                 /*
 833                  * We should submit bio, since it exists several
 834                  * wribacking dentry pages in the freeing inode.
 835                  */
 836                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 837                 cond_resched();
 838         }
 839         goto retry;
 840 }
 841
 842 /*
 843  * Freeze all the FS-operations for checkpoint.
 844  */
 845 static int block_operations(struct f2fs_sb_info *sbi)
 846 {
 847         struct writeback_control wbc = {
 848                 .sync_mode = WB_SYNC_ALL,
 849                 .nr_to_write = LONG_MAX,
 850                 .for_reclaim = 0,
 851         };
 852         struct blk_plug plug;
 853         int err = 0;
 854
 855         blk_start_plug(&plug);
 856
 857 retry_flush_dents:
 858         f2fs_lock_all(sbi);
 859         /* write all the dirty dentry pages */
 860         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 861                 f2fs_unlock_all(sbi);
 862                 err = sync_dirty_inodes(sbi, DIR_INODE);
 863                 if (err)
 864                         goto out;
 865                 goto retry_flush_dents;
 866         }
 867
 868         /*
 869          * POR: we should ensure that there are no dirty node pages
 870          * until finishing nat/sit flush.
 871          */
 872 retry_flush_nodes:
 873         down_write(&sbi->node_write);
 874
 875         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 876                 up_write(&sbi->node_write);
 877                 err = sync_node_pages(sbi, 0, &wbc);
 878                 if (err) {
 879                         f2fs_unlock_all(sbi);
 880                         goto out;
 881                 }
 882                 goto retry_flush_nodes;
 883         }
 884 out:
 885         blk_finish_plug(&plug);
 886         return err;
 887 }
 888
 889 static void unblock_operations(struct f2fs_sb_info *sbi)
 890 {
 891         up_write(&sbi->node_write);
 892         f2fs_unlock_all(sbi);
 893 }
 894
 895 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 896 {
 897         DEFINE_WAIT(wait);
 898
 899         for (;;) {
 900                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 901
 902                 if (!get_pages(sbi, F2FS_WRITEBACK))
 903                         break;
 904
 905                 io_schedule();
 906         }
 907         finish_wait(&sbi->cp_wait, &wait);
 908 }
 909
 910 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 911 {
 912         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 913         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 914         struct f2fs_nm_info *nm_i = NM_I(sbi);
 915         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 916         nid_t last_nid = nm_i->next_scan_nid;
 917         block_t start_blk;
 918         unsigned int data_sum_blocks, orphan_blocks;
 919         __u32 crc32 = 0;
 920         int i;
 921         int cp_payload_blks = __cp_payload(sbi);
 922         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
 923         bool invalidate = false;
 924
 925         /*
 926          * This avoids to conduct wrong roll-forward operations and uses
 927          * metapages, so should be called prior to sync_meta_pages below.
 928          */
 929         if (discard_next_dnode(sbi, discard_blk))
 930                 invalidate = true;
 931
 932         /* Flush all the NAT/SIT pages */
 933         while (get_pages(sbi, F2FS_DIRTY_META)) {
 934                 sync_meta_pages(sbi, META, LONG_MAX);
 935                 if (unlikely(f2fs_cp_error(sbi)))
 936                         return -EIO;
 937         }
 938
 939         next_free_nid(sbi, &last_nid);
 940
 941         /*
 942          * modify checkpoint
 943          * version number is already updated
 944          */
 945         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 946         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 947         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 948         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 949                 ckpt->cur_node_segno[i] =
 950                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 951                 ckpt->cur_node_blkoff[i] =
 952                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 953                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 954                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 955         }
 956         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 957                 ckpt->cur_data_segno[i] =
 958                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 959                 ckpt->cur_data_blkoff[i] =
 960                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 961                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 962                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 963         }
 964
 965         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 966         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 967         ckpt->next_free_nid = cpu_to_le32(last_nid);
 968
 969         /* 2 cp  + n data seg summary + orphan inode blocks */
 970         data_sum_blocks = npages_for_summary_flush(sbi, false);
 971         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 972                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 973         else
 974                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 975
 976         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 977         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 978                         orphan_blocks);
 979
 980         if (__remain_node_summaries(cpc->reason))
 981                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 982                                 cp_payload_blks + data_sum_blocks +
 983                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 984         else
 985                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 986                                 cp_payload_blks + data_sum_blocks +
 987                                 orphan_blocks);
 988
 989         if (cpc->reason == CP_UMOUNT)
 990                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 991         else
 992                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 993
 994         if (cpc->reason == CP_FASTBOOT)
 995                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
 996         else
 997                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
 998
 999         if (orphan_num)
1000                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1001         else
1002                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1003
1004         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1005                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1006
1007         /* update SIT/NAT bitmap */
1008         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1009         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1010
1011         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
1012         *((__le32 *)((unsigned char *)ckpt +
1013                                 le32_to_cpu(ckpt->checksum_offset)))
1014                                 = cpu_to_le32(crc32);
1015
1016         start_blk = __start_cp_addr(sbi);
1017
1018         /* need to wait for end_io results */
1019         wait_on_all_pages_writeback(sbi);
1020         if (unlikely(f2fs_cp_error(sbi)))
1021                 return -EIO;
1022
1023         /* write out checkpoint buffer at block 0 */
1024         update_meta_page(sbi, ckpt, start_blk++);
1025
1026         for (i = 1; i < 1 + cp_payload_blks; i++)
1027                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1028                                                         start_blk++);
1029
1030         if (orphan_num) {
1031                 write_orphan_inodes(sbi, start_blk);
1032                 start_blk += orphan_blocks;
1033         }
1034
1035         write_data_summaries(sbi, start_blk);
1036         start_blk += data_sum_blocks;
1037         if (__remain_node_summaries(cpc->reason)) {
1038                 write_node_summaries(sbi, start_blk);
1039                 start_blk += NR_CURSEG_NODE_TYPE;
1040         }
1041
1042         /* writeout checkpoint block */
1043         update_meta_page(sbi, ckpt, start_blk);
1044
1045         /* wait for previous submitted node/meta pages writeback */
1046         wait_on_all_pages_writeback(sbi);
1047
1048         if (unlikely(f2fs_cp_error(sbi)))
1049                 return -EIO;
1050
1051         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1052         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1053
1054         /* update user_block_counts */
1055         sbi->last_valid_block_count = sbi->total_valid_block_count;
1056         sbi->alloc_valid_block_count = 0;
1057
1058         /* Here, we only have one bio having CP pack */
1059         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1060
1061         /* wait for previous submitted meta pages writeback */
1062         wait_on_all_pages_writeback(sbi);
1063
1064         /*
1065          * invalidate meta page which is used temporarily for zeroing out
1066          * block at the end of warm node chain.
1067          */
1068         if (invalidate)
1069                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1070                                                                 discard_blk);
1071
1072         release_ino_entry(sbi);
1073
1074         if (unlikely(f2fs_cp_error(sbi)))
1075                 return -EIO;
1076
1077         clear_prefree_segments(sbi, cpc);
1078         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1079
1080         return 0;
1081 }
1082
1083 /*
1084  * We guarantee that this checkpoint procedure will not fail.
1085  */
1086 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1087 {
1088         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1089         unsigned long long ckpt_ver;
1090         int err = 0;
1091
1092         mutex_lock(&sbi->cp_mutex);
1093
1094         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1095                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1096                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1097                 goto out;
1098         if (unlikely(f2fs_cp_error(sbi))) {
1099                 err = -EIO;
1100                 goto out;
1101         }
1102         if (f2fs_readonly(sbi->sb)) {
1103                 err = -EROFS;
1104                 goto out;
1105         }
1106
1107         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1108
1109         err = block_operations(sbi);
1110         if (err)
1111                 goto out;
1112
1113         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1114
1115         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1116         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1117         f2fs_submit_merged_bio(sbi, META, WRITE);
1118
1119         /*
1120          * update checkpoint pack index
1121          * Increase the version number so that
1122          * SIT entries and seg summaries are written at correct place
1123          */
1124         ckpt_ver = cur_cp_version(ckpt);
1125         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1126
1127         /* write cached NAT/SIT entries to NAT/SIT area */
1128         flush_nat_entries(sbi);
1129         flush_sit_entries(sbi, cpc);
1130
1131         /* unlock all the fs_lock[] in do_checkpoint() */
1132         err = do_checkpoint(sbi, cpc);
1133
1134         unblock_operations(sbi);
1135         stat_inc_cp_count(sbi->stat_info);
1136
1137         if (cpc->reason == CP_RECOVERY)
1138                 f2fs_msg(sbi->sb, KERN_NOTICE,
1139                         "checkpoint: version = %llx", ckpt_ver);
1140
1141         /* do checkpoint periodically */
1142         f2fs_update_time(sbi, CP_TIME);
1143         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1144 out:
1145         mutex_unlock(&sbi->cp_mutex);
1146         return err;
1147 }
1148
1149 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1150 {
1151         int i;
1152
1153         for (i = 0; i < MAX_INO_ENTRY; i++) {
1154                 struct inode_management *im = &sbi->im[i];
1155
1156                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1157                 spin_lock_init(&im->ino_lock);
1158                 INIT_LIST_HEAD(&im->ino_list);
1159                 im->ino_num = 0;
1160         }
1161
1162         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1163                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1164                                 F2FS_ORPHANS_PER_BLOCK;
1165 }
1166
1167 int __init create_checkpoint_caches(void)
1168 {
1169         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1170                         sizeof(struct ino_entry));
1171         if (!ino_entry_slab)
1172                 return -ENOMEM;
1173         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1174                         sizeof(struct inode_entry));
1175         if (!inode_entry_slab) {
1176                 kmem_cache_destroy(ino_entry_slab);
1177                 return -ENOMEM;
1178         }
1179         return 0;
1180 }
1181
1182 void destroy_checkpoint_caches(void)
1183 {
1184         kmem_cache_destroy(ino_entry_slab);
1185         kmem_cache_destroy(inode_entry_slab);
1186 }