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 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  51 {
  52         struct address_space *mapping = META_MAPPING(sbi);
  53         struct page *page;
  54         struct f2fs_io_info fio = {
  55                 .type = META,
  56                 .rw = READ_SYNC | REQ_META | REQ_PRIO,
  57                 .blk_addr = index,
  58         };
  59 repeat:
  60         page = grab_cache_page(mapping, index);
  61         if (!page) {
  62                 cond_resched();
  63                 goto repeat;
  64         }
  65         if (PageUptodate(page))
  66                 goto out;
  67
  68         if (f2fs_submit_page_bio(sbi, page, &fio))
  69                 goto repeat;
  70
  71         lock_page(page);
  72         if (unlikely(page->mapping != mapping)) {
  73                 f2fs_put_page(page, 1);
  74                 goto repeat;
  75         }
  76 out:
  77         return page;
  78 }
  79
  80 static inline bool is_valid_blkaddr(struct f2fs_sb_info *sbi,
  81                                                 block_t blkaddr, int type)
  82 {
  83         switch (type) {
  84         case META_NAT:
  85                 break;
  86         case META_SIT:
  87                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
  88                         return false;
  89                 break;
  90         case META_SSA:
  91                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
  92                         blkaddr < SM_I(sbi)->ssa_blkaddr))
  93                         return false;
  94                 break;
  95         case META_CP:
  96                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
  97                         blkaddr < __start_cp_addr(sbi)))
  98                         return false;
  99                 break;
 100         case META_POR:
 101                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 102                         blkaddr < MAIN_BLKADDR(sbi)))
 103                         return false;
 104                 break;
 105         default:
 106                 BUG();
 107         }
 108
 109         return true;
 110 }
 111
 112 /*
 113  * Readahead CP/NAT/SIT/SSA pages
 114  */
 115 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
 116 {
 117         block_t prev_blk_addr = 0;
 118         struct page *page;
 119         block_t blkno = start;
 120         struct f2fs_io_info fio = {
 121                 .type = META,
 122                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 123         };
 124
 125         for (; nrpages-- > 0; blkno++) {
 126
 127                 if (!is_valid_blkaddr(sbi, blkno, type))
 128                         goto out;
 129
 130                 switch (type) {
 131                 case META_NAT:
 132                         if (unlikely(blkno >=
 133                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 134                                 blkno = 0;
 135                         /* get nat block addr */
 136                         fio.blk_addr = current_nat_addr(sbi,
 137                                         blkno * NAT_ENTRY_PER_BLOCK);
 138                         break;
 139                 case META_SIT:
 140                         /* get sit block addr */
 141                         fio.blk_addr = current_sit_addr(sbi,
 142                                         blkno * SIT_ENTRY_PER_BLOCK);
 143                         if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
 144                                 goto out;
 145                         prev_blk_addr = fio.blk_addr;
 146                         break;
 147                 case META_SSA:
 148                 case META_CP:
 149                 case META_POR:
 150                         fio.blk_addr = blkno;
 151                         break;
 152                 default:
 153                         BUG();
 154                 }
 155
 156                 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
 157                 if (!page)
 158                         continue;
 159                 if (PageUptodate(page)) {
 160                         f2fs_put_page(page, 1);
 161                         continue;
 162                 }
 163
 164                 f2fs_submit_page_mbio(sbi, page, &fio);
 165                 f2fs_put_page(page, 0);
 166         }
 167 out:
 168         f2fs_submit_merged_bio(sbi, META, READ);
 169         return blkno - start;
 170 }
 171
 172 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 173 {
 174         struct page *page;
 175         bool readahead = false;
 176
 177         page = find_get_page(META_MAPPING(sbi), index);
 178         if (!page || (page && !PageUptodate(page)))
 179                 readahead = true;
 180         f2fs_put_page(page, 0);
 181
 182         if (readahead)
 183                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
 184 }
 185
 186 static int f2fs_write_meta_page(struct page *page,
 187                                 struct writeback_control *wbc)
 188 {
 189         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 190
 191         trace_f2fs_writepage(page, META);
 192
 193         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 194                 goto redirty_out;
 195         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 196                 goto redirty_out;
 197         if (unlikely(f2fs_cp_error(sbi)))
 198                 goto redirty_out;
 199
 200         f2fs_wait_on_page_writeback(page, META);
 201         write_meta_page(sbi, page);
 202         dec_page_count(sbi, F2FS_DIRTY_META);
 203         unlock_page(page);
 204
 205         if (wbc->for_reclaim)
 206                 f2fs_submit_merged_bio(sbi, META, WRITE);
 207         return 0;
 208
 209 redirty_out:
 210         redirty_page_for_writepage(wbc, page);
 211         return AOP_WRITEPAGE_ACTIVATE;
 212 }
 213
 214 static int f2fs_write_meta_pages(struct address_space *mapping,
 215                                 struct writeback_control *wbc)
 216 {
 217         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 218         long diff, written;
 219
 220         trace_f2fs_writepages(mapping->host, wbc, META);
 221
 222         /* collect a number of dirty meta pages and write together */
 223         if (wbc->for_kupdate ||
 224                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 225                 goto skip_write;
 226
 227         /* if mounting is failed, skip writing node pages */
 228         mutex_lock(&sbi->cp_mutex);
 229         diff = nr_pages_to_write(sbi, META, wbc);
 230         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 231         mutex_unlock(&sbi->cp_mutex);
 232         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 233         return 0;
 234
 235 skip_write:
 236         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 237         return 0;
 238 }
 239
 240 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 241                                                 long nr_to_write)
 242 {
 243         struct address_space *mapping = META_MAPPING(sbi);
 244         pgoff_t index = 0, end = LONG_MAX;
 245         struct pagevec pvec;
 246         long nwritten = 0;
 247         struct writeback_control wbc = {
 248                 .for_reclaim = 0,
 249         };
 250
 251         pagevec_init(&pvec, 0);
 252
 253         while (index <= end) {
 254                 int i, nr_pages;
 255                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 256                                 PAGECACHE_TAG_DIRTY,
 257                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 258                 if (unlikely(nr_pages == 0))
 259                         break;
 260
 261                 for (i = 0; i < nr_pages; i++) {
 262                         struct page *page = pvec.pages[i];
 263
 264                         lock_page(page);
 265
 266                         if (unlikely(page->mapping != mapping)) {
 267 continue_unlock:
 268                                 unlock_page(page);
 269                                 continue;
 270                         }
 271                         if (!PageDirty(page)) {
 272                                 /* someone wrote it for us */
 273                                 goto continue_unlock;
 274                         }
 275
 276                         if (!clear_page_dirty_for_io(page))
 277                                 goto continue_unlock;
 278
 279                         if (f2fs_write_meta_page(page, &wbc)) {
 280                                 unlock_page(page);
 281                                 break;
 282                         }
 283                         nwritten++;
 284                         if (unlikely(nwritten >= nr_to_write))
 285                                 break;
 286                 }
 287                 pagevec_release(&pvec);
 288                 cond_resched();
 289         }
 290
 291         if (nwritten)
 292                 f2fs_submit_merged_bio(sbi, type, WRITE);
 293
 294         return nwritten;
 295 }
 296
 297 static int f2fs_set_meta_page_dirty(struct page *page)
 298 {
 299         trace_f2fs_set_page_dirty(page, META);
 300
 301         SetPageUptodate(page);
 302         if (!PageDirty(page)) {
 303                 __set_page_dirty_nobuffers(page);
 304                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 305                 SetPagePrivate(page);
 306                 f2fs_trace_pid(page);
 307                 return 1;
 308         }
 309         return 0;
 310 }
 311
 312 static void f2fs_invalidate_meta_page(struct page *page, unsigned int offset,
 313                                       unsigned int length)
 314 {
 315         struct inode *inode = page->mapping->host;
 316
 317         if (PageDirty(page))
 318                 dec_page_count(F2FS_I_SB(inode), F2FS_DIRTY_META);
 319         ClearPagePrivate(page);
 320 }
 321
 322 static int f2fs_release_meta_page(struct page *page, gfp_t wait)
 323 {
 324         ClearPagePrivate(page);
 325         return 1;
 326 }
 327
 328 const struct address_space_operations f2fs_meta_aops = {
 329         .writepage      = f2fs_write_meta_page,
 330         .writepages     = f2fs_write_meta_pages,
 331         .set_page_dirty = f2fs_set_meta_page_dirty,
 332         .invalidatepage = f2fs_invalidate_meta_page,
 333         .releasepage    = f2fs_release_meta_page,
 334 };
 335
 336 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 337 {
 338         struct inode_management *im = &sbi->im[type];
 339         struct ino_entry *e;
 340 retry:
 341         if (radix_tree_preload(GFP_NOFS)) {
 342                 cond_resched();
 343                 goto retry;
 344         }
 345
 346         spin_lock(&im->ino_lock);
 347
 348         e = radix_tree_lookup(&im->ino_root, ino);
 349         if (!e) {
 350                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 351                 if (!e) {
 352                         spin_unlock(&im->ino_lock);
 353                         radix_tree_preload_end();
 354                         goto retry;
 355                 }
 356                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 357                         spin_unlock(&im->ino_lock);
 358                         kmem_cache_free(ino_entry_slab, e);
 359                         radix_tree_preload_end();
 360                         goto retry;
 361                 }
 362                 memset(e, 0, sizeof(struct ino_entry));
 363                 e->ino = ino;
 364
 365                 list_add_tail(&e->list, &im->ino_list);
 366                 if (type != ORPHAN_INO)
 367                         im->ino_num++;
 368         }
 369         spin_unlock(&im->ino_lock);
 370         radix_tree_preload_end();
 371 }
 372
 373 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 374 {
 375         struct inode_management *im = &sbi->im[type];
 376         struct ino_entry *e;
 377
 378         spin_lock(&im->ino_lock);
 379         e = radix_tree_lookup(&im->ino_root, ino);
 380         if (e) {
 381                 list_del(&e->list);
 382                 radix_tree_delete(&im->ino_root, ino);
 383                 im->ino_num--;
 384                 spin_unlock(&im->ino_lock);
 385                 kmem_cache_free(ino_entry_slab, e);
 386                 return;
 387         }
 388         spin_unlock(&im->ino_lock);
 389 }
 390
 391 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 392 {
 393         /* add new dirty ino entry into list */
 394         __add_ino_entry(sbi, ino, type);
 395 }
 396
 397 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 398 {
 399         /* remove dirty ino entry from list */
 400         __remove_ino_entry(sbi, ino, type);
 401 }
 402
 403 /* mode should be APPEND_INO or UPDATE_INO */
 404 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 405 {
 406         struct inode_management *im = &sbi->im[mode];
 407         struct ino_entry *e;
 408
 409         spin_lock(&im->ino_lock);
 410         e = radix_tree_lookup(&im->ino_root, ino);
 411         spin_unlock(&im->ino_lock);
 412         return e ? true : false;
 413 }
 414
 415 void release_dirty_inode(struct f2fs_sb_info *sbi)
 416 {
 417         struct ino_entry *e, *tmp;
 418         int i;
 419
 420         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 421                 struct inode_management *im = &sbi->im[i];
 422
 423                 spin_lock(&im->ino_lock);
 424                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 425                         list_del(&e->list);
 426                         radix_tree_delete(&im->ino_root, e->ino);
 427                         kmem_cache_free(ino_entry_slab, e);
 428                         im->ino_num--;
 429                 }
 430                 spin_unlock(&im->ino_lock);
 431         }
 432 }
 433
 434 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 435 {
 436         struct inode_management *im = &sbi->im[ORPHAN_INO];
 437         int err = 0;
 438
 439         spin_lock(&im->ino_lock);
 440         if (unlikely(im->ino_num >= sbi->max_orphans))
 441                 err = -ENOSPC;
 442         else
 443                 im->ino_num++;
 444         spin_unlock(&im->ino_lock);
 445
 446         return err;
 447 }
 448
 449 void release_orphan_inode(struct f2fs_sb_info *sbi)
 450 {
 451         struct inode_management *im = &sbi->im[ORPHAN_INO];
 452
 453         spin_lock(&im->ino_lock);
 454         f2fs_bug_on(sbi, im->ino_num == 0);
 455         im->ino_num--;
 456         spin_unlock(&im->ino_lock);
 457 }
 458
 459 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 460 {
 461         /* add new orphan ino entry into list */
 462         __add_ino_entry(sbi, ino, ORPHAN_INO);
 463 }
 464
 465 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 466 {
 467         /* remove orphan entry from orphan list */
 468         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 469 }
 470
 471 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 472 {
 473         struct inode *inode = f2fs_iget(sbi->sb, ino);
 474         f2fs_bug_on(sbi, IS_ERR(inode));
 475         clear_nlink(inode);
 476
 477         /* truncate all the data during iput */
 478         iput(inode);
 479 }
 480
 481 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 482 {
 483         block_t start_blk, orphan_blkaddr, i, j;
 484
 485         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 486                 return;
 487
 488         set_sbi_flag(sbi, SBI_POR_DOING);
 489
 490         start_blk = __start_cp_addr(sbi) + 1 +
 491                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 492         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 493
 494         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 495
 496         for (i = 0; i < orphan_blkaddr; i++) {
 497                 struct page *page = get_meta_page(sbi, start_blk + i);
 498                 struct f2fs_orphan_block *orphan_blk;
 499
 500                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 501                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 502                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 503                         recover_orphan_inode(sbi, ino);
 504                 }
 505                 f2fs_put_page(page, 1);
 506         }
 507         /* clear Orphan Flag */
 508         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 509         clear_sbi_flag(sbi, SBI_POR_DOING);
 510         return;
 511 }
 512
 513 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 514 {
 515         struct list_head *head;
 516         struct f2fs_orphan_block *orphan_blk = NULL;
 517         unsigned int nentries = 0;
 518         unsigned short index;
 519         unsigned short orphan_blocks;
 520         struct page *page = NULL;
 521         struct ino_entry *orphan = NULL;
 522         struct inode_management *im = &sbi->im[ORPHAN_INO];
 523
 524         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 525
 526         for (index = 0; index < orphan_blocks; index++)
 527                 grab_meta_page(sbi, start_blk + index);
 528
 529         index = 1;
 530         spin_lock(&im->ino_lock);
 531         head = &im->ino_list;
 532
 533         /* loop for each orphan inode entry and write them in Jornal block */
 534         list_for_each_entry(orphan, head, list) {
 535                 if (!page) {
 536                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 537                         f2fs_bug_on(sbi, !page);
 538                         orphan_blk =
 539                                 (struct f2fs_orphan_block *)page_address(page);
 540                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 541                         f2fs_put_page(page, 0);
 542                 }
 543
 544                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 545
 546                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 547                         /*
 548                          * an orphan block is full of 1020 entries,
 549                          * then we need to flush current orphan blocks
 550                          * and bring another one in memory
 551                          */
 552                         orphan_blk->blk_addr = cpu_to_le16(index);
 553                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 554                         orphan_blk->entry_count = cpu_to_le32(nentries);
 555                         set_page_dirty(page);
 556                         f2fs_put_page(page, 1);
 557                         index++;
 558                         nentries = 0;
 559                         page = NULL;
 560                 }
 561         }
 562
 563         if (page) {
 564                 orphan_blk->blk_addr = cpu_to_le16(index);
 565                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 566                 orphan_blk->entry_count = cpu_to_le32(nentries);
 567                 set_page_dirty(page);
 568                 f2fs_put_page(page, 1);
 569         }
 570
 571         spin_unlock(&im->ino_lock);
 572 }
 573
 574 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 575                                 block_t cp_addr, unsigned long long *version)
 576 {
 577         struct page *cp_page_1, *cp_page_2 = NULL;
 578         unsigned long blk_size = sbi->blocksize;
 579         struct f2fs_checkpoint *cp_block;
 580         unsigned long long cur_version = 0, pre_version = 0;
 581         size_t crc_offset;
 582         __u32 crc = 0;
 583
 584         /* Read the 1st cp block in this CP pack */
 585         cp_page_1 = get_meta_page(sbi, cp_addr);
 586
 587         /* get the version number */
 588         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 589         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 590         if (crc_offset >= blk_size)
 591                 goto invalid_cp1;
 592
 593         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 594         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 595                 goto invalid_cp1;
 596
 597         pre_version = cur_cp_version(cp_block);
 598
 599         /* Read the 2nd cp block in this CP pack */
 600         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 601         cp_page_2 = get_meta_page(sbi, cp_addr);
 602
 603         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 604         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 605         if (crc_offset >= blk_size)
 606                 goto invalid_cp2;
 607
 608         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 609         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 610                 goto invalid_cp2;
 611
 612         cur_version = cur_cp_version(cp_block);
 613
 614         if (cur_version == pre_version) {
 615                 *version = cur_version;
 616                 f2fs_put_page(cp_page_2, 1);
 617                 return cp_page_1;
 618         }
 619 invalid_cp2:
 620         f2fs_put_page(cp_page_2, 1);
 621 invalid_cp1:
 622         f2fs_put_page(cp_page_1, 1);
 623         return NULL;
 624 }
 625
 626 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 627 {
 628         struct f2fs_checkpoint *cp_block;
 629         struct f2fs_super_block *fsb = sbi->raw_super;
 630         struct page *cp1, *cp2, *cur_page;
 631         unsigned long blk_size = sbi->blocksize;
 632         unsigned long long cp1_version = 0, cp2_version = 0;
 633         unsigned long long cp_start_blk_no;
 634         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 635         block_t cp_blk_no;
 636         int i;
 637
 638         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 639         if (!sbi->ckpt)
 640                 return -ENOMEM;
 641         /*
 642          * Finding out valid cp block involves read both
 643          * sets( cp pack1 and cp pack 2)
 644          */
 645         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 646         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 647
 648         /* The second checkpoint pack should start at the next segment */
 649         cp_start_blk_no += ((unsigned long long)1) <<
 650                                 le32_to_cpu(fsb->log_blocks_per_seg);
 651         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 652
 653         if (cp1 && cp2) {
 654                 if (ver_after(cp2_version, cp1_version))
 655                         cur_page = cp2;
 656                 else
 657                         cur_page = cp1;
 658         } else if (cp1) {
 659                 cur_page = cp1;
 660         } else if (cp2) {
 661                 cur_page = cp2;
 662         } else {
 663                 goto fail_no_cp;
 664         }
 665
 666         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 667         memcpy(sbi->ckpt, cp_block, blk_size);
 668
 669         if (cp_blks <= 1)
 670                 goto done;
 671
 672         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 673         if (cur_page == cp2)
 674                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 675
 676         for (i = 1; i < cp_blks; i++) {
 677                 void *sit_bitmap_ptr;
 678                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 679
 680                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 681                 sit_bitmap_ptr = page_address(cur_page);
 682                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 683                 f2fs_put_page(cur_page, 1);
 684         }
 685 done:
 686         f2fs_put_page(cp1, 1);
 687         f2fs_put_page(cp2, 1);
 688         return 0;
 689
 690 fail_no_cp:
 691         kfree(sbi->ckpt);
 692         return -EINVAL;
 693 }
 694
 695 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
 696 {
 697         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 698
 699         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 700                 return -EEXIST;
 701
 702         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 703         F2FS_I(inode)->dirty_dir = new;
 704         list_add_tail(&new->list, &sbi->dir_inode_list);
 705         stat_inc_dirty_dir(sbi);
 706         return 0;
 707 }
 708
 709 void update_dirty_page(struct inode *inode, struct page *page)
 710 {
 711         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 712         struct inode_entry *new;
 713         int ret = 0;
 714
 715         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
 716                 return;
 717
 718         if (!S_ISDIR(inode->i_mode)) {
 719                 inode_inc_dirty_pages(inode);
 720                 goto out;
 721         }
 722
 723         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 724         new->inode = inode;
 725         INIT_LIST_HEAD(&new->list);
 726
 727         spin_lock(&sbi->dir_inode_lock);
 728         ret = __add_dirty_inode(inode, new);
 729         inode_inc_dirty_pages(inode);
 730         spin_unlock(&sbi->dir_inode_lock);
 731
 732         if (ret)
 733                 kmem_cache_free(inode_entry_slab, new);
 734 out:
 735         SetPagePrivate(page);
 736         f2fs_trace_pid(page);
 737 }
 738
 739 void add_dirty_dir_inode(struct inode *inode)
 740 {
 741         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 742         struct inode_entry *new =
 743                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 744         int ret = 0;
 745
 746         new->inode = inode;
 747         INIT_LIST_HEAD(&new->list);
 748
 749         spin_lock(&sbi->dir_inode_lock);
 750         ret = __add_dirty_inode(inode, new);
 751         spin_unlock(&sbi->dir_inode_lock);
 752
 753         if (ret)
 754                 kmem_cache_free(inode_entry_slab, new);
 755 }
 756
 757 void remove_dirty_dir_inode(struct inode *inode)
 758 {
 759         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 760         struct inode_entry *entry;
 761
 762         if (!S_ISDIR(inode->i_mode))
 763                 return;
 764
 765         spin_lock(&sbi->dir_inode_lock);
 766         if (get_dirty_pages(inode) ||
 767                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 768                 spin_unlock(&sbi->dir_inode_lock);
 769                 return;
 770         }
 771
 772         entry = F2FS_I(inode)->dirty_dir;
 773         list_del(&entry->list);
 774         F2FS_I(inode)->dirty_dir = NULL;
 775         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 776         stat_dec_dirty_dir(sbi);
 777         spin_unlock(&sbi->dir_inode_lock);
 778         kmem_cache_free(inode_entry_slab, entry);
 779
 780         /* Only from the recovery routine */
 781         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 782                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 783                 iput(inode);
 784         }
 785 }
 786
 787 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 788 {
 789         struct list_head *head;
 790         struct inode_entry *entry;
 791         struct inode *inode;
 792 retry:
 793         if (unlikely(f2fs_cp_error(sbi)))
 794                 return;
 795
 796         spin_lock(&sbi->dir_inode_lock);
 797
 798         head = &sbi->dir_inode_list;
 799         if (list_empty(head)) {
 800                 spin_unlock(&sbi->dir_inode_lock);
 801                 return;
 802         }
 803         entry = list_entry(head->next, struct inode_entry, list);
 804         inode = igrab(entry->inode);
 805         spin_unlock(&sbi->dir_inode_lock);
 806         if (inode) {
 807                 filemap_fdatawrite(inode->i_mapping);
 808                 iput(inode);
 809         } else {
 810                 /*
 811                  * We should submit bio, since it exists several
 812                  * wribacking dentry pages in the freeing inode.
 813                  */
 814                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 815         }
 816         goto retry;
 817 }
 818
 819 /*
 820  * Freeze all the FS-operations for checkpoint.
 821  */
 822 static int block_operations(struct f2fs_sb_info *sbi)
 823 {
 824         struct writeback_control wbc = {
 825                 .sync_mode = WB_SYNC_ALL,
 826                 .nr_to_write = LONG_MAX,
 827                 .for_reclaim = 0,
 828         };
 829         struct blk_plug plug;
 830         int err = 0;
 831
 832         blk_start_plug(&plug);
 833
 834 retry_flush_dents:
 835         f2fs_lock_all(sbi);
 836         /* write all the dirty dentry pages */
 837         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 838                 f2fs_unlock_all(sbi);
 839                 sync_dirty_dir_inodes(sbi);
 840                 if (unlikely(f2fs_cp_error(sbi))) {
 841                         err = -EIO;
 842                         goto out;
 843                 }
 844                 goto retry_flush_dents;
 845         }
 846
 847         /*
 848          * POR: we should ensure that there are no dirty node pages
 849          * until finishing nat/sit flush.
 850          */
 851 retry_flush_nodes:
 852         down_write(&sbi->node_write);
 853
 854         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 855                 up_write(&sbi->node_write);
 856                 sync_node_pages(sbi, 0, &wbc);
 857                 if (unlikely(f2fs_cp_error(sbi))) {
 858                         f2fs_unlock_all(sbi);
 859                         err = -EIO;
 860                         goto out;
 861                 }
 862                 goto retry_flush_nodes;
 863         }
 864 out:
 865         blk_finish_plug(&plug);
 866         return err;
 867 }
 868
 869 static void unblock_operations(struct f2fs_sb_info *sbi)
 870 {
 871         up_write(&sbi->node_write);
 872         f2fs_unlock_all(sbi);
 873 }
 874
 875 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 876 {
 877         DEFINE_WAIT(wait);
 878
 879         for (;;) {
 880                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 881
 882                 if (!get_pages(sbi, F2FS_WRITEBACK))
 883                         break;
 884
 885                 io_schedule();
 886         }
 887         finish_wait(&sbi->cp_wait, &wait);
 888 }
 889
 890 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 891 {
 892         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 893         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 894         struct f2fs_nm_info *nm_i = NM_I(sbi);
 895         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 896         nid_t last_nid = nm_i->next_scan_nid;
 897         block_t start_blk;
 898         struct page *cp_page;
 899         unsigned int data_sum_blocks, orphan_blocks;
 900         __u32 crc32 = 0;
 901         void *kaddr;
 902         int i;
 903         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 904
 905         /*
 906          * This avoids to conduct wrong roll-forward operations and uses
 907          * metapages, so should be called prior to sync_meta_pages below.
 908          */
 909         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 910
 911         /* Flush all the NAT/SIT pages */
 912         while (get_pages(sbi, F2FS_DIRTY_META)) {
 913                 sync_meta_pages(sbi, META, LONG_MAX);
 914                 if (unlikely(f2fs_cp_error(sbi)))
 915                         return;
 916         }
 917
 918         next_free_nid(sbi, &last_nid);
 919
 920         /*
 921          * modify checkpoint
 922          * version number is already updated
 923          */
 924         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 925         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 926         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 927         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 928                 ckpt->cur_node_segno[i] =
 929                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 930                 ckpt->cur_node_blkoff[i] =
 931                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 932                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 933                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 934         }
 935         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 936                 ckpt->cur_data_segno[i] =
 937                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 938                 ckpt->cur_data_blkoff[i] =
 939                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 940                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 941                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 942         }
 943
 944         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 945         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 946         ckpt->next_free_nid = cpu_to_le32(last_nid);
 947
 948         /* 2 cp  + n data seg summary + orphan inode blocks */
 949         data_sum_blocks = npages_for_summary_flush(sbi, false);
 950         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 951                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 952         else
 953                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 954
 955         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 956         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 957                         orphan_blocks);
 958
 959         if (__remain_node_summaries(cpc->reason))
 960                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 961                                 cp_payload_blks + data_sum_blocks +
 962                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 963         else
 964                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 965                                 cp_payload_blks + data_sum_blocks +
 966                                 orphan_blocks);
 967
 968         if (cpc->reason == CP_UMOUNT)
 969                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 970         else
 971                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 972
 973         if (cpc->reason == CP_FASTBOOT)
 974                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
 975         else
 976                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
 977
 978         if (orphan_num)
 979                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 980         else
 981                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 982
 983         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
 984                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
 985
 986         /* update SIT/NAT bitmap */
 987         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 988         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 989
 990         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 991         *((__le32 *)((unsigned char *)ckpt +
 992                                 le32_to_cpu(ckpt->checksum_offset)))
 993                                 = cpu_to_le32(crc32);
 994
 995         start_blk = __start_cp_addr(sbi);
 996
 997         /* write out checkpoint buffer at block 0 */
 998         cp_page = grab_meta_page(sbi, start_blk++);
 999         kaddr = page_address(cp_page);
1000         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
1001         set_page_dirty(cp_page);
1002         f2fs_put_page(cp_page, 1);
1003
1004         for (i = 1; i < 1 + cp_payload_blks; i++) {
1005                 cp_page = grab_meta_page(sbi, start_blk++);
1006                 kaddr = page_address(cp_page);
1007                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
1008                                 (1 << sbi->log_blocksize));
1009                 set_page_dirty(cp_page);
1010                 f2fs_put_page(cp_page, 1);
1011         }
1012
1013         if (orphan_num) {
1014                 write_orphan_inodes(sbi, start_blk);
1015                 start_blk += orphan_blocks;
1016         }
1017
1018         write_data_summaries(sbi, start_blk);
1019         start_blk += data_sum_blocks;
1020         if (__remain_node_summaries(cpc->reason)) {
1021                 write_node_summaries(sbi, start_blk);
1022                 start_blk += NR_CURSEG_NODE_TYPE;
1023         }
1024
1025         /* writeout checkpoint block */
1026         cp_page = grab_meta_page(sbi, start_blk);
1027         kaddr = page_address(cp_page);
1028         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
1029         set_page_dirty(cp_page);
1030         f2fs_put_page(cp_page, 1);
1031
1032         /* wait for previous submitted node/meta pages writeback */
1033         wait_on_all_pages_writeback(sbi);
1034
1035         if (unlikely(f2fs_cp_error(sbi)))
1036                 return;
1037
1038         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1039         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1040
1041         /* update user_block_counts */
1042         sbi->last_valid_block_count = sbi->total_valid_block_count;
1043         sbi->alloc_valid_block_count = 0;
1044
1045         /* Here, we only have one bio having CP pack */
1046         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1047
1048         /* wait for previous submitted meta pages writeback */
1049         wait_on_all_pages_writeback(sbi);
1050
1051         release_dirty_inode(sbi);
1052
1053         if (unlikely(f2fs_cp_error(sbi)))
1054                 return;
1055
1056         clear_prefree_segments(sbi);
1057         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1058 }
1059
1060 /*
1061  * We guarantee that this checkpoint procedure will not fail.
1062  */
1063 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1064 {
1065         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1066         unsigned long long ckpt_ver;
1067
1068         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1069
1070         mutex_lock(&sbi->cp_mutex);
1071
1072         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1073                         cpc->reason != CP_DISCARD && cpc->reason != CP_UMOUNT)
1074                 goto out;
1075         if (unlikely(f2fs_cp_error(sbi)))
1076                 goto out;
1077         if (f2fs_readonly(sbi->sb))
1078                 goto out;
1079         if (block_operations(sbi))
1080                 goto out;
1081
1082         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1083
1084         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1085         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1086         f2fs_submit_merged_bio(sbi, META, WRITE);
1087
1088         /*
1089          * update checkpoint pack index
1090          * Increase the version number so that
1091          * SIT entries and seg summaries are written at correct place
1092          */
1093         ckpt_ver = cur_cp_version(ckpt);
1094         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1095
1096         /* write cached NAT/SIT entries to NAT/SIT area */
1097         flush_nat_entries(sbi);
1098         flush_sit_entries(sbi, cpc);
1099
1100         /* unlock all the fs_lock[] in do_checkpoint() */
1101         do_checkpoint(sbi, cpc);
1102
1103         unblock_operations(sbi);
1104         stat_inc_cp_count(sbi->stat_info);
1105 out:
1106         mutex_unlock(&sbi->cp_mutex);
1107         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1108 }
1109
1110 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1111 {
1112         int i;
1113
1114         for (i = 0; i < MAX_INO_ENTRY; i++) {
1115                 struct inode_management *im = &sbi->im[i];
1116
1117                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1118                 spin_lock_init(&im->ino_lock);
1119                 INIT_LIST_HEAD(&im->ino_list);
1120                 im->ino_num = 0;
1121         }
1122
1123         /*
1124          * considering 512 blocks in a segment 8 blocks are needed for cp
1125          * and log segment summaries. Remaining blocks are used to keep
1126          * orphan entries with the limitation one reserved segment
1127          * for cp pack we can have max 1020*504 orphan entries
1128          */
1129         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1130                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1131 }
1132
1133 int __init create_checkpoint_caches(void)
1134 {
1135         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1136                         sizeof(struct ino_entry));
1137         if (!ino_entry_slab)
1138                 return -ENOMEM;
1139         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1140                         sizeof(struct inode_entry));
1141         if (!inode_entry_slab) {
1142                 kmem_cache_destroy(ino_entry_slab);
1143                 return -ENOMEM;
1144         }
1145         return 0;
1146 }
1147
1148 void destroy_checkpoint_caches(void)
1149 {
1150         kmem_cache_destroy(ino_entry_slab);
1151         kmem_cache_destroy(inode_entry_slab);
1152 }