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(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                 f2fs_trace_pid(page);
 306                 return 1;
 307         }
 308         return 0;
 309 }
 310
 311 const struct address_space_operations f2fs_meta_aops = {
 312         .writepage      = f2fs_write_meta_page,
 313         .writepages     = f2fs_write_meta_pages,
 314         .set_page_dirty = f2fs_set_meta_page_dirty,
 315 };
 316
 317 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 318 {
 319         struct inode_management *im = &sbi->im[type];
 320         struct ino_entry *e;
 321 retry:
 322         if (radix_tree_preload(GFP_NOFS)) {
 323                 cond_resched();
 324                 goto retry;
 325         }
 326
 327         spin_lock(&im->ino_lock);
 328
 329         e = radix_tree_lookup(&im->ino_root, ino);
 330         if (!e) {
 331                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 332                 if (!e) {
 333                         spin_unlock(&im->ino_lock);
 334                         radix_tree_preload_end();
 335                         goto retry;
 336                 }
 337                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 338                         spin_unlock(&im->ino_lock);
 339                         kmem_cache_free(ino_entry_slab, e);
 340                         radix_tree_preload_end();
 341                         goto retry;
 342                 }
 343                 memset(e, 0, sizeof(struct ino_entry));
 344                 e->ino = ino;
 345
 346                 list_add_tail(&e->list, &im->ino_list);
 347                 if (type != ORPHAN_INO)
 348                         im->ino_num++;
 349         }
 350         spin_unlock(&im->ino_lock);
 351         radix_tree_preload_end();
 352 }
 353
 354 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 355 {
 356         struct inode_management *im = &sbi->im[type];
 357         struct ino_entry *e;
 358
 359         spin_lock(&im->ino_lock);
 360         e = radix_tree_lookup(&im->ino_root, ino);
 361         if (e) {
 362                 list_del(&e->list);
 363                 radix_tree_delete(&im->ino_root, ino);
 364                 im->ino_num--;
 365                 spin_unlock(&im->ino_lock);
 366                 kmem_cache_free(ino_entry_slab, e);
 367                 return;
 368         }
 369         spin_unlock(&im->ino_lock);
 370 }
 371
 372 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 373 {
 374         /* add new dirty ino entry into list */
 375         __add_ino_entry(sbi, ino, type);
 376 }
 377
 378 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 379 {
 380         /* remove dirty ino entry from list */
 381         __remove_ino_entry(sbi, ino, type);
 382 }
 383
 384 /* mode should be APPEND_INO or UPDATE_INO */
 385 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 386 {
 387         struct inode_management *im = &sbi->im[mode];
 388         struct ino_entry *e;
 389
 390         spin_lock(&im->ino_lock);
 391         e = radix_tree_lookup(&im->ino_root, ino);
 392         spin_unlock(&im->ino_lock);
 393         return e ? true : false;
 394 }
 395
 396 void release_dirty_inode(struct f2fs_sb_info *sbi)
 397 {
 398         struct ino_entry *e, *tmp;
 399         int i;
 400
 401         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 402                 struct inode_management *im = &sbi->im[i];
 403
 404                 spin_lock(&im->ino_lock);
 405                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 406                         list_del(&e->list);
 407                         radix_tree_delete(&im->ino_root, e->ino);
 408                         kmem_cache_free(ino_entry_slab, e);
 409                         im->ino_num--;
 410                 }
 411                 spin_unlock(&im->ino_lock);
 412         }
 413 }
 414
 415 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 416 {
 417         struct inode_management *im = &sbi->im[ORPHAN_INO];
 418         int err = 0;
 419
 420         spin_lock(&im->ino_lock);
 421         if (unlikely(im->ino_num >= sbi->max_orphans))
 422                 err = -ENOSPC;
 423         else
 424                 im->ino_num++;
 425         spin_unlock(&im->ino_lock);
 426
 427         return err;
 428 }
 429
 430 void release_orphan_inode(struct f2fs_sb_info *sbi)
 431 {
 432         struct inode_management *im = &sbi->im[ORPHAN_INO];
 433
 434         spin_lock(&im->ino_lock);
 435         f2fs_bug_on(sbi, im->ino_num == 0);
 436         im->ino_num--;
 437         spin_unlock(&im->ino_lock);
 438 }
 439
 440 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 441 {
 442         /* add new orphan ino entry into list */
 443         __add_ino_entry(sbi, ino, ORPHAN_INO);
 444 }
 445
 446 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 447 {
 448         /* remove orphan entry from orphan list */
 449         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 450 }
 451
 452 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 453 {
 454         struct inode *inode = f2fs_iget(sbi->sb, ino);
 455         f2fs_bug_on(sbi, IS_ERR(inode));
 456         clear_nlink(inode);
 457
 458         /* truncate all the data during iput */
 459         iput(inode);
 460 }
 461
 462 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 463 {
 464         block_t start_blk, orphan_blkaddr, i, j;
 465
 466         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 467                 return;
 468
 469         sbi->por_doing = true;
 470
 471         start_blk = __start_cp_addr(sbi) + 1 +
 472                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 473         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 474
 475         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 476
 477         for (i = 0; i < orphan_blkaddr; i++) {
 478                 struct page *page = get_meta_page(sbi, start_blk + i);
 479                 struct f2fs_orphan_block *orphan_blk;
 480
 481                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 482                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 483                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 484                         recover_orphan_inode(sbi, ino);
 485                 }
 486                 f2fs_put_page(page, 1);
 487         }
 488         /* clear Orphan Flag */
 489         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 490         sbi->por_doing = false;
 491         return;
 492 }
 493
 494 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 495 {
 496         struct list_head *head;
 497         struct f2fs_orphan_block *orphan_blk = NULL;
 498         unsigned int nentries = 0;
 499         unsigned short index;
 500         unsigned short orphan_blocks;
 501         struct page *page = NULL;
 502         struct ino_entry *orphan = NULL;
 503         struct inode_management *im = &sbi->im[ORPHAN_INO];
 504
 505         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 506
 507         for (index = 0; index < orphan_blocks; index++)
 508                 grab_meta_page(sbi, start_blk + index);
 509
 510         index = 1;
 511         spin_lock(&im->ino_lock);
 512         head = &im->ino_list;
 513
 514         /* loop for each orphan inode entry and write them in Jornal block */
 515         list_for_each_entry(orphan, head, list) {
 516                 if (!page) {
 517                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 518                         f2fs_bug_on(sbi, !page);
 519                         orphan_blk =
 520                                 (struct f2fs_orphan_block *)page_address(page);
 521                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 522                         f2fs_put_page(page, 0);
 523                 }
 524
 525                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 526
 527                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 528                         /*
 529                          * an orphan block is full of 1020 entries,
 530                          * then we need to flush current orphan blocks
 531                          * and bring another one in memory
 532                          */
 533                         orphan_blk->blk_addr = cpu_to_le16(index);
 534                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 535                         orphan_blk->entry_count = cpu_to_le32(nentries);
 536                         set_page_dirty(page);
 537                         f2fs_put_page(page, 1);
 538                         index++;
 539                         nentries = 0;
 540                         page = NULL;
 541                 }
 542         }
 543
 544         if (page) {
 545                 orphan_blk->blk_addr = cpu_to_le16(index);
 546                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 547                 orphan_blk->entry_count = cpu_to_le32(nentries);
 548                 set_page_dirty(page);
 549                 f2fs_put_page(page, 1);
 550         }
 551
 552         spin_unlock(&im->ino_lock);
 553 }
 554
 555 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 556                                 block_t cp_addr, unsigned long long *version)
 557 {
 558         struct page *cp_page_1, *cp_page_2 = NULL;
 559         unsigned long blk_size = sbi->blocksize;
 560         struct f2fs_checkpoint *cp_block;
 561         unsigned long long cur_version = 0, pre_version = 0;
 562         size_t crc_offset;
 563         __u32 crc = 0;
 564
 565         /* Read the 1st cp block in this CP pack */
 566         cp_page_1 = get_meta_page(sbi, cp_addr);
 567
 568         /* get the version number */
 569         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 570         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 571         if (crc_offset >= blk_size)
 572                 goto invalid_cp1;
 573
 574         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 575         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 576                 goto invalid_cp1;
 577
 578         pre_version = cur_cp_version(cp_block);
 579
 580         /* Read the 2nd cp block in this CP pack */
 581         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 582         cp_page_2 = get_meta_page(sbi, cp_addr);
 583
 584         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 585         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 586         if (crc_offset >= blk_size)
 587                 goto invalid_cp2;
 588
 589         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 590         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 591                 goto invalid_cp2;
 592
 593         cur_version = cur_cp_version(cp_block);
 594
 595         if (cur_version == pre_version) {
 596                 *version = cur_version;
 597                 f2fs_put_page(cp_page_2, 1);
 598                 return cp_page_1;
 599         }
 600 invalid_cp2:
 601         f2fs_put_page(cp_page_2, 1);
 602 invalid_cp1:
 603         f2fs_put_page(cp_page_1, 1);
 604         return NULL;
 605 }
 606
 607 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 608 {
 609         struct f2fs_checkpoint *cp_block;
 610         struct f2fs_super_block *fsb = sbi->raw_super;
 611         struct page *cp1, *cp2, *cur_page;
 612         unsigned long blk_size = sbi->blocksize;
 613         unsigned long long cp1_version = 0, cp2_version = 0;
 614         unsigned long long cp_start_blk_no;
 615         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 616         block_t cp_blk_no;
 617         int i;
 618
 619         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 620         if (!sbi->ckpt)
 621                 return -ENOMEM;
 622         /*
 623          * Finding out valid cp block involves read both
 624          * sets( cp pack1 and cp pack 2)
 625          */
 626         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 627         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 628
 629         /* The second checkpoint pack should start at the next segment */
 630         cp_start_blk_no += ((unsigned long long)1) <<
 631                                 le32_to_cpu(fsb->log_blocks_per_seg);
 632         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 633
 634         if (cp1 && cp2) {
 635                 if (ver_after(cp2_version, cp1_version))
 636                         cur_page = cp2;
 637                 else
 638                         cur_page = cp1;
 639         } else if (cp1) {
 640                 cur_page = cp1;
 641         } else if (cp2) {
 642                 cur_page = cp2;
 643         } else {
 644                 goto fail_no_cp;
 645         }
 646
 647         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 648         memcpy(sbi->ckpt, cp_block, blk_size);
 649
 650         if (cp_blks <= 1)
 651                 goto done;
 652
 653         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 654         if (cur_page == cp2)
 655                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 656
 657         for (i = 1; i < cp_blks; i++) {
 658                 void *sit_bitmap_ptr;
 659                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 660
 661                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 662                 sit_bitmap_ptr = page_address(cur_page);
 663                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 664                 f2fs_put_page(cur_page, 1);
 665         }
 666 done:
 667         f2fs_put_page(cp1, 1);
 668         f2fs_put_page(cp2, 1);
 669         return 0;
 670
 671 fail_no_cp:
 672         kfree(sbi->ckpt);
 673         return -EINVAL;
 674 }
 675
 676 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
 677 {
 678         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 679
 680         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 681                 return -EEXIST;
 682
 683         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 684         F2FS_I(inode)->dirty_dir = new;
 685         list_add_tail(&new->list, &sbi->dir_inode_list);
 686         stat_inc_dirty_dir(sbi);
 687         return 0;
 688 }
 689
 690 void update_dirty_page(struct inode *inode, struct page *page)
 691 {
 692         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 693         struct inode_entry *new;
 694         int ret = 0;
 695
 696         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
 697                 return;
 698
 699         if (!S_ISDIR(inode->i_mode)) {
 700                 inode_inc_dirty_pages(inode);
 701                 goto out;
 702         }
 703
 704         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 705         new->inode = inode;
 706         INIT_LIST_HEAD(&new->list);
 707
 708         spin_lock(&sbi->dir_inode_lock);
 709         ret = __add_dirty_inode(inode, new);
 710         inode_inc_dirty_pages(inode);
 711         spin_unlock(&sbi->dir_inode_lock);
 712
 713         if (ret)
 714                 kmem_cache_free(inode_entry_slab, new);
 715 out:
 716         SetPagePrivate(page);
 717         f2fs_trace_pid(page);
 718 }
 719
 720 void add_dirty_dir_inode(struct inode *inode)
 721 {
 722         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 723         struct inode_entry *new =
 724                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 725         int ret = 0;
 726
 727         new->inode = inode;
 728         INIT_LIST_HEAD(&new->list);
 729
 730         spin_lock(&sbi->dir_inode_lock);
 731         ret = __add_dirty_inode(inode, new);
 732         spin_unlock(&sbi->dir_inode_lock);
 733
 734         if (ret)
 735                 kmem_cache_free(inode_entry_slab, new);
 736 }
 737
 738 void remove_dirty_dir_inode(struct inode *inode)
 739 {
 740         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 741         struct inode_entry *entry;
 742
 743         if (!S_ISDIR(inode->i_mode))
 744                 return;
 745
 746         spin_lock(&sbi->dir_inode_lock);
 747         if (get_dirty_pages(inode) ||
 748                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 749                 spin_unlock(&sbi->dir_inode_lock);
 750                 return;
 751         }
 752
 753         entry = F2FS_I(inode)->dirty_dir;
 754         list_del(&entry->list);
 755         F2FS_I(inode)->dirty_dir = NULL;
 756         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 757         stat_dec_dirty_dir(sbi);
 758         spin_unlock(&sbi->dir_inode_lock);
 759         kmem_cache_free(inode_entry_slab, entry);
 760
 761         /* Only from the recovery routine */
 762         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 763                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 764                 iput(inode);
 765         }
 766 }
 767
 768 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 769 {
 770         struct list_head *head;
 771         struct inode_entry *entry;
 772         struct inode *inode;
 773 retry:
 774         if (unlikely(f2fs_cp_error(sbi)))
 775                 return;
 776
 777         spin_lock(&sbi->dir_inode_lock);
 778
 779         head = &sbi->dir_inode_list;
 780         if (list_empty(head)) {
 781                 spin_unlock(&sbi->dir_inode_lock);
 782                 return;
 783         }
 784         entry = list_entry(head->next, struct inode_entry, list);
 785         inode = igrab(entry->inode);
 786         spin_unlock(&sbi->dir_inode_lock);
 787         if (inode) {
 788                 filemap_fdatawrite(inode->i_mapping);
 789                 iput(inode);
 790         } else {
 791                 /*
 792                  * We should submit bio, since it exists several
 793                  * wribacking dentry pages in the freeing inode.
 794                  */
 795                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 796         }
 797         goto retry;
 798 }
 799
 800 /*
 801  * Freeze all the FS-operations for checkpoint.
 802  */
 803 static int block_operations(struct f2fs_sb_info *sbi)
 804 {
 805         struct writeback_control wbc = {
 806                 .sync_mode = WB_SYNC_ALL,
 807                 .nr_to_write = LONG_MAX,
 808                 .for_reclaim = 0,
 809         };
 810         struct blk_plug plug;
 811         int err = 0;
 812
 813         blk_start_plug(&plug);
 814
 815 retry_flush_dents:
 816         f2fs_lock_all(sbi);
 817         /* write all the dirty dentry pages */
 818         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 819                 f2fs_unlock_all(sbi);
 820                 sync_dirty_dir_inodes(sbi);
 821                 if (unlikely(f2fs_cp_error(sbi))) {
 822                         err = -EIO;
 823                         goto out;
 824                 }
 825                 goto retry_flush_dents;
 826         }
 827
 828         /*
 829          * POR: we should ensure that there are no dirty node pages
 830          * until finishing nat/sit flush.
 831          */
 832 retry_flush_nodes:
 833         down_write(&sbi->node_write);
 834
 835         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 836                 up_write(&sbi->node_write);
 837                 sync_node_pages(sbi, 0, &wbc);
 838                 if (unlikely(f2fs_cp_error(sbi))) {
 839                         f2fs_unlock_all(sbi);
 840                         err = -EIO;
 841                         goto out;
 842                 }
 843                 goto retry_flush_nodes;
 844         }
 845 out:
 846         blk_finish_plug(&plug);
 847         return err;
 848 }
 849
 850 static void unblock_operations(struct f2fs_sb_info *sbi)
 851 {
 852         up_write(&sbi->node_write);
 853         f2fs_unlock_all(sbi);
 854 }
 855
 856 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 857 {
 858         DEFINE_WAIT(wait);
 859
 860         for (;;) {
 861                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 862
 863                 if (!get_pages(sbi, F2FS_WRITEBACK))
 864                         break;
 865
 866                 io_schedule();
 867         }
 868         finish_wait(&sbi->cp_wait, &wait);
 869 }
 870
 871 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 872 {
 873         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 874         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 875         struct f2fs_nm_info *nm_i = NM_I(sbi);
 876         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 877         nid_t last_nid = nm_i->next_scan_nid;
 878         block_t start_blk;
 879         struct page *cp_page;
 880         unsigned int data_sum_blocks, orphan_blocks;
 881         __u32 crc32 = 0;
 882         void *kaddr;
 883         int i;
 884         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 885
 886         /*
 887          * This avoids to conduct wrong roll-forward operations and uses
 888          * metapages, so should be called prior to sync_meta_pages below.
 889          */
 890         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 891
 892         /* Flush all the NAT/SIT pages */
 893         while (get_pages(sbi, F2FS_DIRTY_META)) {
 894                 sync_meta_pages(sbi, META, LONG_MAX);
 895                 if (unlikely(f2fs_cp_error(sbi)))
 896                         return;
 897         }
 898
 899         next_free_nid(sbi, &last_nid);
 900
 901         /*
 902          * modify checkpoint
 903          * version number is already updated
 904          */
 905         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 906         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 907         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 908         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 909                 ckpt->cur_node_segno[i] =
 910                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 911                 ckpt->cur_node_blkoff[i] =
 912                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 913                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 914                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 915         }
 916         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 917                 ckpt->cur_data_segno[i] =
 918                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 919                 ckpt->cur_data_blkoff[i] =
 920                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 921                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 922                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 923         }
 924
 925         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 926         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 927         ckpt->next_free_nid = cpu_to_le32(last_nid);
 928
 929         /* 2 cp  + n data seg summary + orphan inode blocks */
 930         data_sum_blocks = npages_for_summary_flush(sbi, false);
 931         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 932                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 933         else
 934                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 935
 936         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 937         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 938                         orphan_blocks);
 939
 940         if (cpc->reason == CP_UMOUNT) {
 941                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 942                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 943                                 cp_payload_blks + data_sum_blocks +
 944                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 945         } else {
 946                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 947                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 948                                 cp_payload_blks + data_sum_blocks +
 949                                 orphan_blocks);
 950         }
 951
 952         if (orphan_num)
 953                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 954         else
 955                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 956
 957         if (sbi->need_fsck)
 958                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
 959
 960         /* update SIT/NAT bitmap */
 961         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 962         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 963
 964         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 965         *((__le32 *)((unsigned char *)ckpt +
 966                                 le32_to_cpu(ckpt->checksum_offset)))
 967                                 = cpu_to_le32(crc32);
 968
 969         start_blk = __start_cp_addr(sbi);
 970
 971         /* write out checkpoint buffer at block 0 */
 972         cp_page = grab_meta_page(sbi, start_blk++);
 973         kaddr = page_address(cp_page);
 974         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 975         set_page_dirty(cp_page);
 976         f2fs_put_page(cp_page, 1);
 977
 978         for (i = 1; i < 1 + cp_payload_blks; i++) {
 979                 cp_page = grab_meta_page(sbi, start_blk++);
 980                 kaddr = page_address(cp_page);
 981                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 982                                 (1 << sbi->log_blocksize));
 983                 set_page_dirty(cp_page);
 984                 f2fs_put_page(cp_page, 1);
 985         }
 986
 987         if (orphan_num) {
 988                 write_orphan_inodes(sbi, start_blk);
 989                 start_blk += orphan_blocks;
 990         }
 991
 992         write_data_summaries(sbi, start_blk);
 993         start_blk += data_sum_blocks;
 994         if (cpc->reason == CP_UMOUNT) {
 995                 write_node_summaries(sbi, start_blk);
 996                 start_blk += NR_CURSEG_NODE_TYPE;
 997         }
 998
 999         /* writeout checkpoint block */
1000         cp_page = grab_meta_page(sbi, start_blk);
1001         kaddr = page_address(cp_page);
1002         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
1003         set_page_dirty(cp_page);
1004         f2fs_put_page(cp_page, 1);
1005
1006         /* wait for previous submitted node/meta pages writeback */
1007         wait_on_all_pages_writeback(sbi);
1008
1009         if (unlikely(f2fs_cp_error(sbi)))
1010                 return;
1011
1012         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1013         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1014
1015         /* update user_block_counts */
1016         sbi->last_valid_block_count = sbi->total_valid_block_count;
1017         sbi->alloc_valid_block_count = 0;
1018
1019         /* Here, we only have one bio having CP pack */
1020         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1021
1022         /* wait for previous submitted meta pages writeback */
1023         wait_on_all_pages_writeback(sbi);
1024
1025         release_dirty_inode(sbi);
1026
1027         if (unlikely(f2fs_cp_error(sbi)))
1028                 return;
1029
1030         clear_prefree_segments(sbi);
1031         F2FS_RESET_SB_DIRT(sbi);
1032 }
1033
1034 /*
1035  * We guarantee that this checkpoint procedure will not fail.
1036  */
1037 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1038 {
1039         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1040         unsigned long long ckpt_ver;
1041
1042         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1043
1044         mutex_lock(&sbi->cp_mutex);
1045
1046         if (!sbi->s_dirty && cpc->reason != CP_DISCARD)
1047                 goto out;
1048         if (unlikely(f2fs_cp_error(sbi)))
1049                 goto out;
1050         if (block_operations(sbi))
1051                 goto out;
1052
1053         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1054
1055         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1056         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1057         f2fs_submit_merged_bio(sbi, META, WRITE);
1058
1059         /*
1060          * update checkpoint pack index
1061          * Increase the version number so that
1062          * SIT entries and seg summaries are written at correct place
1063          */
1064         ckpt_ver = cur_cp_version(ckpt);
1065         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1066
1067         /* write cached NAT/SIT entries to NAT/SIT area */
1068         flush_nat_entries(sbi);
1069         flush_sit_entries(sbi, cpc);
1070
1071         /* unlock all the fs_lock[] in do_checkpoint() */
1072         do_checkpoint(sbi, cpc);
1073
1074         unblock_operations(sbi);
1075         stat_inc_cp_count(sbi->stat_info);
1076 out:
1077         mutex_unlock(&sbi->cp_mutex);
1078         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1079 }
1080
1081 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1082 {
1083         int i;
1084
1085         for (i = 0; i < MAX_INO_ENTRY; i++) {
1086                 struct inode_management *im = &sbi->im[i];
1087
1088                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1089                 spin_lock_init(&im->ino_lock);
1090                 INIT_LIST_HEAD(&im->ino_list);
1091                 im->ino_num = 0;
1092         }
1093
1094         /*
1095          * considering 512 blocks in a segment 8 blocks are needed for cp
1096          * and log segment summaries. Remaining blocks are used to keep
1097          * orphan entries with the limitation one reserved segment
1098          * for cp pack we can have max 1020*504 orphan entries
1099          */
1100         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1101                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1102 }
1103
1104 int __init create_checkpoint_caches(void)
1105 {
1106         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1107                         sizeof(struct ino_entry));
1108         if (!ino_entry_slab)
1109                 return -ENOMEM;
1110         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1111                         sizeof(struct inode_entry));
1112         if (!inode_entry_slab) {
1113                 kmem_cache_destroy(ino_entry_slab);
1114                 return -ENOMEM;
1115         }
1116         return 0;
1117 }
1118
1119 void destroy_checkpoint_caches(void)
1120 {
1121         kmem_cache_destroy(ino_entry_slab);
1122         kmem_cache_destroy(inode_entry_slab);
1123 }