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/events/f2fs.h>
  24
  25 static struct kmem_cache *ino_entry_slab;
  26 static struct kmem_cache *inode_entry_slab;
  27
  28 /*
  29  * We guarantee no failure on the returned page.
  30  */
  31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  32 {
  33         struct address_space *mapping = META_MAPPING(sbi);
  34         struct page *page = NULL;
  35 repeat:
  36         page = grab_cache_page(mapping, index);
  37         if (!page) {
  38                 cond_resched();
  39                 goto repeat;
  40         }
  41         f2fs_wait_on_page_writeback(page, META);
  42         SetPageUptodate(page);
  43         return page;
  44 }
  45
  46 /*
  47  * We guarantee no failure on the returned page.
  48  */
  49 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  50 {
  51         struct address_space *mapping = META_MAPPING(sbi);
  52         struct page *page;
  53         struct f2fs_io_info fio = {
  54                 .type = META,
  55                 .rw = READ_SYNC | REQ_META | REQ_PRIO,
  56                 .blk_addr = index,
  57         };
  58 repeat:
  59         page = grab_cache_page(mapping, index);
  60         if (!page) {
  61                 cond_resched();
  62                 goto repeat;
  63         }
  64         if (PageUptodate(page))
  65                 goto out;
  66
  67         if (f2fs_submit_page_bio(sbi, page, &fio))
  68                 goto repeat;
  69
  70         lock_page(page);
  71         if (unlikely(page->mapping != mapping)) {
  72                 f2fs_put_page(page, 1);
  73                 goto repeat;
  74         }
  75 out:
  76         return page;
  77 }
  78
  79 static inline bool is_valid_blkaddr(struct f2fs_sb_info *sbi,
  80                                                 block_t blkaddr, int type)
  81 {
  82         switch (type) {
  83         case META_NAT:
  84                 break;
  85         case META_SIT:
  86                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
  87                         return false;
  88                 break;
  89         case META_SSA:
  90                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
  91                         blkaddr < SM_I(sbi)->ssa_blkaddr))
  92                         return false;
  93                 break;
  94         case META_CP:
  95                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
  96                         blkaddr < __start_cp_addr(sbi)))
  97                         return false;
  98                 break;
  99         case META_POR:
 100                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 101                         blkaddr < MAIN_BLKADDR(sbi)))
 102                         return false;
 103                 break;
 104         default:
 105                 BUG();
 106         }
 107
 108         return true;
 109 }
 110
 111 /*
 112  * Readahead CP/NAT/SIT/SSA pages
 113  */
 114 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
 115 {
 116         block_t prev_blk_addr = 0;
 117         struct page *page;
 118         block_t blkno = start;
 119         struct f2fs_io_info fio = {
 120                 .type = META,
 121                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 122         };
 123
 124         for (; nrpages-- > 0; blkno++) {
 125
 126                 if (!is_valid_blkaddr(sbi, blkno, type))
 127                         goto out;
 128
 129                 switch (type) {
 130                 case META_NAT:
 131                         if (unlikely(blkno >=
 132                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 133                                 blkno = 0;
 134                         /* get nat block addr */
 135                         fio.blk_addr = current_nat_addr(sbi,
 136                                         blkno * NAT_ENTRY_PER_BLOCK);
 137                         break;
 138                 case META_SIT:
 139                         /* get sit block addr */
 140                         fio.blk_addr = current_sit_addr(sbi,
 141                                         blkno * SIT_ENTRY_PER_BLOCK);
 142                         if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
 143                                 goto out;
 144                         prev_blk_addr = fio.blk_addr;
 145                         break;
 146                 case META_SSA:
 147                 case META_CP:
 148                 case META_POR:
 149                         fio.blk_addr = blkno;
 150                         break;
 151                 default:
 152                         BUG();
 153                 }
 154
 155                 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
 156                 if (!page)
 157                         continue;
 158                 if (PageUptodate(page)) {
 159                         f2fs_put_page(page, 1);
 160                         continue;
 161                 }
 162
 163                 f2fs_submit_page_mbio(sbi, page, &fio);
 164                 f2fs_put_page(page, 0);
 165         }
 166 out:
 167         f2fs_submit_merged_bio(sbi, META, READ);
 168         return blkno - start;
 169 }
 170
 171 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 172 {
 173         struct page *page;
 174         bool readahead = false;
 175
 176         page = find_get_page(META_MAPPING(sbi), index);
 177         if (!page || (page && !PageUptodate(page)))
 178                 readahead = true;
 179         f2fs_put_page(page, 0);
 180
 181         if (readahead)
 182                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
 183 }
 184
 185 static int f2fs_write_meta_page(struct page *page,
 186                                 struct writeback_control *wbc)
 187 {
 188         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 189
 190         trace_f2fs_writepage(page, META);
 191
 192         if (unlikely(sbi->por_doing))
 193                 goto redirty_out;
 194         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 195                 goto redirty_out;
 196         if (unlikely(f2fs_cp_error(sbi)))
 197                 goto redirty_out;
 198
 199         f2fs_wait_on_page_writeback(page, META);
 200         write_meta_page(sbi, page);
 201         dec_page_count(sbi, F2FS_DIRTY_META);
 202         unlock_page(page);
 203
 204         if (wbc->for_reclaim)
 205                 f2fs_submit_merged_bio(sbi, META, WRITE);
 206         return 0;
 207
 208 redirty_out:
 209         redirty_page_for_writepage(wbc, page);
 210         return AOP_WRITEPAGE_ACTIVATE;
 211 }
 212
 213 static int f2fs_write_meta_pages(struct address_space *mapping,
 214                                 struct writeback_control *wbc)
 215 {
 216         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 217         long diff, written;
 218
 219         trace_f2fs_writepages(mapping->host, wbc, META);
 220
 221         /* collect a number of dirty meta pages and write together */
 222         if (wbc->for_kupdate ||
 223                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 224                 goto skip_write;
 225
 226         /* if mounting is failed, skip writing node pages */
 227         mutex_lock(&sbi->cp_mutex);
 228         diff = nr_pages_to_write(sbi, META, wbc);
 229         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 230         mutex_unlock(&sbi->cp_mutex);
 231         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 232         return 0;
 233
 234 skip_write:
 235         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 236         return 0;
 237 }
 238
 239 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 240                                                 long nr_to_write)
 241 {
 242         struct address_space *mapping = META_MAPPING(sbi);
 243         pgoff_t index = 0, end = LONG_MAX;
 244         struct pagevec pvec;
 245         long nwritten = 0;
 246         struct writeback_control wbc = {
 247                 .for_reclaim = 0,
 248         };
 249
 250         pagevec_init(&pvec, 0);
 251
 252         while (index <= end) {
 253                 int i, nr_pages;
 254                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 255                                 PAGECACHE_TAG_DIRTY,
 256                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 257                 if (unlikely(nr_pages == 0))
 258                         break;
 259
 260                 for (i = 0; i < nr_pages; i++) {
 261                         struct page *page = pvec.pages[i];
 262
 263                         lock_page(page);
 264
 265                         if (unlikely(page->mapping != mapping)) {
 266 continue_unlock:
 267                                 unlock_page(page);
 268                                 continue;
 269                         }
 270                         if (!PageDirty(page)) {
 271                                 /* someone wrote it for us */
 272                                 goto continue_unlock;
 273                         }
 274
 275                         if (!clear_page_dirty_for_io(page))
 276                                 goto continue_unlock;
 277
 278                         if (f2fs_write_meta_page(page, &wbc)) {
 279                                 unlock_page(page);
 280                                 break;
 281                         }
 282                         nwritten++;
 283                         if (unlikely(nwritten >= nr_to_write))
 284                                 break;
 285                 }
 286                 pagevec_release(&pvec);
 287                 cond_resched();
 288         }
 289
 290         if (nwritten)
 291                 f2fs_submit_merged_bio(sbi, type, WRITE);
 292
 293         return nwritten;
 294 }
 295
 296 static int f2fs_set_meta_page_dirty(struct page *page)
 297 {
 298         trace_f2fs_set_page_dirty(page, META);
 299
 300         SetPageUptodate(page);
 301         if (!PageDirty(page)) {
 302                 __set_page_dirty_nobuffers(page);
 303                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 304                 return 1;
 305         }
 306         return 0;
 307 }
 308
 309 const struct address_space_operations f2fs_meta_aops = {
 310         .writepage      = f2fs_write_meta_page,
 311         .writepages     = f2fs_write_meta_pages,
 312         .set_page_dirty = f2fs_set_meta_page_dirty,
 313 };
 314
 315 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 316 {
 317         struct inode_management *im = &sbi->im[type];
 318         struct ino_entry *e;
 319 retry:
 320         if (radix_tree_preload(GFP_NOFS)) {
 321                 cond_resched();
 322                 goto retry;
 323         }
 324
 325         spin_lock(&im->ino_lock);
 326
 327         e = radix_tree_lookup(&im->ino_root, ino);
 328         if (!e) {
 329                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 330                 if (!e) {
 331                         spin_unlock(&im->ino_lock);
 332                         radix_tree_preload_end();
 333                         goto retry;
 334                 }
 335                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 336                         spin_unlock(&im->ino_lock);
 337                         kmem_cache_free(ino_entry_slab, e);
 338                         radix_tree_preload_end();
 339                         goto retry;
 340                 }
 341                 memset(e, 0, sizeof(struct ino_entry));
 342                 e->ino = ino;
 343
 344                 list_add_tail(&e->list, &im->ino_list);
 345                 if (type != ORPHAN_INO)
 346                         im->ino_num++;
 347         }
 348         spin_unlock(&im->ino_lock);
 349         radix_tree_preload_end();
 350 }
 351
 352 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 353 {
 354         struct inode_management *im = &sbi->im[type];
 355         struct ino_entry *e;
 356
 357         spin_lock(&im->ino_lock);
 358         e = radix_tree_lookup(&im->ino_root, ino);
 359         if (e) {
 360                 list_del(&e->list);
 361                 radix_tree_delete(&im->ino_root, ino);
 362                 im->ino_num--;
 363                 spin_unlock(&im->ino_lock);
 364                 kmem_cache_free(ino_entry_slab, e);
 365                 return;
 366         }
 367         spin_unlock(&im->ino_lock);
 368 }
 369
 370 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 371 {
 372         /* add new dirty ino entry into list */
 373         __add_ino_entry(sbi, ino, type);
 374 }
 375
 376 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 377 {
 378         /* remove dirty ino entry from list */
 379         __remove_ino_entry(sbi, ino, type);
 380 }
 381
 382 /* mode should be APPEND_INO or UPDATE_INO */
 383 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 384 {
 385         struct inode_management *im = &sbi->im[mode];
 386         struct ino_entry *e;
 387
 388         spin_lock(&im->ino_lock);
 389         e = radix_tree_lookup(&im->ino_root, ino);
 390         spin_unlock(&im->ino_lock);
 391         return e ? true : false;
 392 }
 393
 394 void release_dirty_inode(struct f2fs_sb_info *sbi)
 395 {
 396         struct ino_entry *e, *tmp;
 397         int i;
 398
 399         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 400                 struct inode_management *im = &sbi->im[i];
 401
 402                 spin_lock(&im->ino_lock);
 403                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 404                         list_del(&e->list);
 405                         radix_tree_delete(&im->ino_root, e->ino);
 406                         kmem_cache_free(ino_entry_slab, e);
 407                         im->ino_num--;
 408                 }
 409                 spin_unlock(&im->ino_lock);
 410         }
 411 }
 412
 413 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 414 {
 415         struct inode_management *im = &sbi->im[ORPHAN_INO];
 416         int err = 0;
 417
 418         spin_lock(&im->ino_lock);
 419         if (unlikely(im->ino_num >= sbi->max_orphans))
 420                 err = -ENOSPC;
 421         else
 422                 im->ino_num++;
 423         spin_unlock(&im->ino_lock);
 424
 425         return err;
 426 }
 427
 428 void release_orphan_inode(struct f2fs_sb_info *sbi)
 429 {
 430         struct inode_management *im = &sbi->im[ORPHAN_INO];
 431
 432         spin_lock(&im->ino_lock);
 433         f2fs_bug_on(sbi, im->ino_num == 0);
 434         im->ino_num--;
 435         spin_unlock(&im->ino_lock);
 436 }
 437
 438 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 439 {
 440         /* add new orphan ino entry into list */
 441         __add_ino_entry(sbi, ino, ORPHAN_INO);
 442 }
 443
 444 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 445 {
 446         /* remove orphan entry from orphan list */
 447         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 448 }
 449
 450 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 451 {
 452         struct inode *inode = f2fs_iget(sbi->sb, ino);
 453         f2fs_bug_on(sbi, IS_ERR(inode));
 454         clear_nlink(inode);
 455
 456         /* truncate all the data during iput */
 457         iput(inode);
 458 }
 459
 460 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 461 {
 462         block_t start_blk, orphan_blkaddr, i, j;
 463
 464         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 465                 return;
 466
 467         sbi->por_doing = true;
 468
 469         start_blk = __start_cp_addr(sbi) + 1 +
 470                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 471         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 472
 473         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 474
 475         for (i = 0; i < orphan_blkaddr; i++) {
 476                 struct page *page = get_meta_page(sbi, start_blk + i);
 477                 struct f2fs_orphan_block *orphan_blk;
 478
 479                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 480                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 481                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 482                         recover_orphan_inode(sbi, ino);
 483                 }
 484                 f2fs_put_page(page, 1);
 485         }
 486         /* clear Orphan Flag */
 487         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 488         sbi->por_doing = false;
 489         return;
 490 }
 491
 492 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 493 {
 494         struct list_head *head;
 495         struct f2fs_orphan_block *orphan_blk = NULL;
 496         unsigned int nentries = 0;
 497         unsigned short index;
 498         unsigned short orphan_blocks;
 499         struct page *page = NULL;
 500         struct ino_entry *orphan = NULL;
 501         struct inode_management *im = &sbi->im[ORPHAN_INO];
 502
 503         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 504
 505         for (index = 0; index < orphan_blocks; index++)
 506                 grab_meta_page(sbi, start_blk + index);
 507
 508         index = 1;
 509         spin_lock(&im->ino_lock);
 510         head = &im->ino_list;
 511
 512         /* loop for each orphan inode entry and write them in Jornal block */
 513         list_for_each_entry(orphan, head, list) {
 514                 if (!page) {
 515                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 516                         f2fs_bug_on(sbi, !page);
 517                         orphan_blk =
 518                                 (struct f2fs_orphan_block *)page_address(page);
 519                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 520                         f2fs_put_page(page, 0);
 521                 }
 522
 523                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 524
 525                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 526                         /*
 527                          * an orphan block is full of 1020 entries,
 528                          * then we need to flush current orphan blocks
 529                          * and bring another one in memory
 530                          */
 531                         orphan_blk->blk_addr = cpu_to_le16(index);
 532                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 533                         orphan_blk->entry_count = cpu_to_le32(nentries);
 534                         set_page_dirty(page);
 535                         f2fs_put_page(page, 1);
 536                         index++;
 537                         nentries = 0;
 538                         page = NULL;
 539                 }
 540         }
 541
 542         if (page) {
 543                 orphan_blk->blk_addr = cpu_to_le16(index);
 544                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 545                 orphan_blk->entry_count = cpu_to_le32(nentries);
 546                 set_page_dirty(page);
 547                 f2fs_put_page(page, 1);
 548         }
 549
 550         spin_unlock(&im->ino_lock);
 551 }
 552
 553 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 554                                 block_t cp_addr, unsigned long long *version)
 555 {
 556         struct page *cp_page_1, *cp_page_2 = NULL;
 557         unsigned long blk_size = sbi->blocksize;
 558         struct f2fs_checkpoint *cp_block;
 559         unsigned long long cur_version = 0, pre_version = 0;
 560         size_t crc_offset;
 561         __u32 crc = 0;
 562
 563         /* Read the 1st cp block in this CP pack */
 564         cp_page_1 = get_meta_page(sbi, cp_addr);
 565
 566         /* get the version number */
 567         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 568         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 569         if (crc_offset >= blk_size)
 570                 goto invalid_cp1;
 571
 572         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 573         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 574                 goto invalid_cp1;
 575
 576         pre_version = cur_cp_version(cp_block);
 577
 578         /* Read the 2nd cp block in this CP pack */
 579         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 580         cp_page_2 = get_meta_page(sbi, cp_addr);
 581
 582         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 583         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 584         if (crc_offset >= blk_size)
 585                 goto invalid_cp2;
 586
 587         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 588         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 589                 goto invalid_cp2;
 590
 591         cur_version = cur_cp_version(cp_block);
 592
 593         if (cur_version == pre_version) {
 594                 *version = cur_version;
 595                 f2fs_put_page(cp_page_2, 1);
 596                 return cp_page_1;
 597         }
 598 invalid_cp2:
 599         f2fs_put_page(cp_page_2, 1);
 600 invalid_cp1:
 601         f2fs_put_page(cp_page_1, 1);
 602         return NULL;
 603 }
 604
 605 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 606 {
 607         struct f2fs_checkpoint *cp_block;
 608         struct f2fs_super_block *fsb = sbi->raw_super;
 609         struct page *cp1, *cp2, *cur_page;
 610         unsigned long blk_size = sbi->blocksize;
 611         unsigned long long cp1_version = 0, cp2_version = 0;
 612         unsigned long long cp_start_blk_no;
 613         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 614         block_t cp_blk_no;
 615         int i;
 616
 617         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 618         if (!sbi->ckpt)
 619                 return -ENOMEM;
 620         /*
 621          * Finding out valid cp block involves read both
 622          * sets( cp pack1 and cp pack 2)
 623          */
 624         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 625         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 626
 627         /* The second checkpoint pack should start at the next segment */
 628         cp_start_blk_no += ((unsigned long long)1) <<
 629                                 le32_to_cpu(fsb->log_blocks_per_seg);
 630         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 631
 632         if (cp1 && cp2) {
 633                 if (ver_after(cp2_version, cp1_version))
 634                         cur_page = cp2;
 635                 else
 636                         cur_page = cp1;
 637         } else if (cp1) {
 638                 cur_page = cp1;
 639         } else if (cp2) {
 640                 cur_page = cp2;
 641         } else {
 642                 goto fail_no_cp;
 643         }
 644
 645         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 646         memcpy(sbi->ckpt, cp_block, blk_size);
 647
 648         if (cp_blks <= 1)
 649                 goto done;
 650
 651         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 652         if (cur_page == cp2)
 653                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 654
 655         for (i = 1; i < cp_blks; i++) {
 656                 void *sit_bitmap_ptr;
 657                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 658
 659                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 660                 sit_bitmap_ptr = page_address(cur_page);
 661                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 662                 f2fs_put_page(cur_page, 1);
 663         }
 664 done:
 665         f2fs_put_page(cp1, 1);
 666         f2fs_put_page(cp2, 1);
 667         return 0;
 668
 669 fail_no_cp:
 670         kfree(sbi->ckpt);
 671         return -EINVAL;
 672 }
 673
 674 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 675 {
 676         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 677
 678         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 679                 return -EEXIST;
 680
 681         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 682         F2FS_I(inode)->dirty_dir = new;
 683         list_add_tail(&new->list, &sbi->dir_inode_list);
 684         stat_inc_dirty_dir(sbi);
 685         return 0;
 686 }
 687
 688 void update_dirty_page(struct inode *inode, struct page *page)
 689 {
 690         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 691         struct dir_inode_entry *new;
 692         int ret = 0;
 693
 694         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
 695                 return;
 696
 697         if (!S_ISDIR(inode->i_mode)) {
 698                 inode_inc_dirty_pages(inode);
 699                 goto out;
 700         }
 701
 702         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 703         new->inode = inode;
 704         INIT_LIST_HEAD(&new->list);
 705
 706         spin_lock(&sbi->dir_inode_lock);
 707         ret = __add_dirty_inode(inode, new);
 708         inode_inc_dirty_pages(inode);
 709         spin_unlock(&sbi->dir_inode_lock);
 710
 711         if (ret)
 712                 kmem_cache_free(inode_entry_slab, new);
 713 out:
 714         SetPagePrivate(page);
 715 }
 716
 717 void add_dirty_dir_inode(struct inode *inode)
 718 {
 719         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 720         struct dir_inode_entry *new =
 721                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 722         int ret = 0;
 723
 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         spin_unlock(&sbi->dir_inode_lock);
 730
 731         if (ret)
 732                 kmem_cache_free(inode_entry_slab, new);
 733 }
 734
 735 void remove_dirty_dir_inode(struct inode *inode)
 736 {
 737         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 738         struct dir_inode_entry *entry;
 739
 740         if (!S_ISDIR(inode->i_mode))
 741                 return;
 742
 743         spin_lock(&sbi->dir_inode_lock);
 744         if (get_dirty_pages(inode) ||
 745                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 746                 spin_unlock(&sbi->dir_inode_lock);
 747                 return;
 748         }
 749
 750         entry = F2FS_I(inode)->dirty_dir;
 751         list_del(&entry->list);
 752         F2FS_I(inode)->dirty_dir = NULL;
 753         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 754         stat_dec_dirty_dir(sbi);
 755         spin_unlock(&sbi->dir_inode_lock);
 756         kmem_cache_free(inode_entry_slab, entry);
 757
 758         /* Only from the recovery routine */
 759         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 760                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 761                 iput(inode);
 762         }
 763 }
 764
 765 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 766 {
 767         struct list_head *head;
 768         struct dir_inode_entry *entry;
 769         struct inode *inode;
 770 retry:
 771         if (unlikely(f2fs_cp_error(sbi)))
 772                 return;
 773
 774         spin_lock(&sbi->dir_inode_lock);
 775
 776         head = &sbi->dir_inode_list;
 777         if (list_empty(head)) {
 778                 spin_unlock(&sbi->dir_inode_lock);
 779                 return;
 780         }
 781         entry = list_entry(head->next, struct dir_inode_entry, list);
 782         inode = igrab(entry->inode);
 783         spin_unlock(&sbi->dir_inode_lock);
 784         if (inode) {
 785                 filemap_fdatawrite(inode->i_mapping);
 786                 iput(inode);
 787         } else {
 788                 /*
 789                  * We should submit bio, since it exists several
 790                  * wribacking dentry pages in the freeing inode.
 791                  */
 792                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 793         }
 794         goto retry;
 795 }
 796
 797 /*
 798  * Freeze all the FS-operations for checkpoint.
 799  */
 800 static int block_operations(struct f2fs_sb_info *sbi)
 801 {
 802         struct writeback_control wbc = {
 803                 .sync_mode = WB_SYNC_ALL,
 804                 .nr_to_write = LONG_MAX,
 805                 .for_reclaim = 0,
 806         };
 807         struct blk_plug plug;
 808         int err = 0;
 809
 810         blk_start_plug(&plug);
 811
 812 retry_flush_dents:
 813         f2fs_lock_all(sbi);
 814         /* write all the dirty dentry pages */
 815         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 816                 f2fs_unlock_all(sbi);
 817                 sync_dirty_dir_inodes(sbi);
 818                 if (unlikely(f2fs_cp_error(sbi))) {
 819                         err = -EIO;
 820                         goto out;
 821                 }
 822                 goto retry_flush_dents;
 823         }
 824
 825         /*
 826          * POR: we should ensure that there are no dirty node pages
 827          * until finishing nat/sit flush.
 828          */
 829 retry_flush_nodes:
 830         down_write(&sbi->node_write);
 831
 832         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 833                 up_write(&sbi->node_write);
 834                 sync_node_pages(sbi, 0, &wbc);
 835                 if (unlikely(f2fs_cp_error(sbi))) {
 836                         f2fs_unlock_all(sbi);
 837                         err = -EIO;
 838                         goto out;
 839                 }
 840                 goto retry_flush_nodes;
 841         }
 842 out:
 843         blk_finish_plug(&plug);
 844         return err;
 845 }
 846
 847 static void unblock_operations(struct f2fs_sb_info *sbi)
 848 {
 849         up_write(&sbi->node_write);
 850         f2fs_unlock_all(sbi);
 851 }
 852
 853 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 854 {
 855         DEFINE_WAIT(wait);
 856
 857         for (;;) {
 858                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 859
 860                 if (!get_pages(sbi, F2FS_WRITEBACK))
 861                         break;
 862
 863                 io_schedule();
 864         }
 865         finish_wait(&sbi->cp_wait, &wait);
 866 }
 867
 868 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 869 {
 870         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 871         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 872         struct f2fs_nm_info *nm_i = NM_I(sbi);
 873         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 874         nid_t last_nid = nm_i->next_scan_nid;
 875         block_t start_blk;
 876         struct page *cp_page;
 877         unsigned int data_sum_blocks, orphan_blocks;
 878         __u32 crc32 = 0;
 879         void *kaddr;
 880         int i;
 881         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 882
 883         /*
 884          * This avoids to conduct wrong roll-forward operations and uses
 885          * metapages, so should be called prior to sync_meta_pages below.
 886          */
 887         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 888
 889         /* Flush all the NAT/SIT pages */
 890         while (get_pages(sbi, F2FS_DIRTY_META)) {
 891                 sync_meta_pages(sbi, META, LONG_MAX);
 892                 if (unlikely(f2fs_cp_error(sbi)))
 893                         return;
 894         }
 895
 896         next_free_nid(sbi, &last_nid);
 897
 898         /*
 899          * modify checkpoint
 900          * version number is already updated
 901          */
 902         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 903         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 904         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 905         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 906                 ckpt->cur_node_segno[i] =
 907                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 908                 ckpt->cur_node_blkoff[i] =
 909                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 910                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 911                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 912         }
 913         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 914                 ckpt->cur_data_segno[i] =
 915                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 916                 ckpt->cur_data_blkoff[i] =
 917                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 918                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 919                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 920         }
 921
 922         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 923         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 924         ckpt->next_free_nid = cpu_to_le32(last_nid);
 925
 926         /* 2 cp  + n data seg summary + orphan inode blocks */
 927         data_sum_blocks = npages_for_summary_flush(sbi, false);
 928         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 929                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 930         else
 931                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 932
 933         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 934         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 935                         orphan_blocks);
 936
 937         if (cpc->reason == CP_UMOUNT) {
 938                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 939                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 940                                 cp_payload_blks + data_sum_blocks +
 941                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 942         } else {
 943                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 944                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 945                                 cp_payload_blks + data_sum_blocks +
 946                                 orphan_blocks);
 947         }
 948
 949         if (orphan_num)
 950                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 951         else
 952                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 953
 954         if (sbi->need_fsck)
 955                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
 956
 957         /* update SIT/NAT bitmap */
 958         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 959         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 960
 961         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 962         *((__le32 *)((unsigned char *)ckpt +
 963                                 le32_to_cpu(ckpt->checksum_offset)))
 964                                 = cpu_to_le32(crc32);
 965
 966         start_blk = __start_cp_addr(sbi);
 967
 968         /* write out checkpoint buffer at block 0 */
 969         cp_page = grab_meta_page(sbi, start_blk++);
 970         kaddr = page_address(cp_page);
 971         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 972         set_page_dirty(cp_page);
 973         f2fs_put_page(cp_page, 1);
 974
 975         for (i = 1; i < 1 + cp_payload_blks; i++) {
 976                 cp_page = grab_meta_page(sbi, start_blk++);
 977                 kaddr = page_address(cp_page);
 978                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 979                                 (1 << sbi->log_blocksize));
 980                 set_page_dirty(cp_page);
 981                 f2fs_put_page(cp_page, 1);
 982         }
 983
 984         if (orphan_num) {
 985                 write_orphan_inodes(sbi, start_blk);
 986                 start_blk += orphan_blocks;
 987         }
 988
 989         write_data_summaries(sbi, start_blk);
 990         start_blk += data_sum_blocks;
 991         if (cpc->reason == CP_UMOUNT) {
 992                 write_node_summaries(sbi, start_blk);
 993                 start_blk += NR_CURSEG_NODE_TYPE;
 994         }
 995
 996         /* writeout checkpoint block */
 997         cp_page = grab_meta_page(sbi, start_blk);
 998         kaddr = page_address(cp_page);
 999         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
1000         set_page_dirty(cp_page);
1001         f2fs_put_page(cp_page, 1);
1002
1003         /* wait for previous submitted node/meta pages writeback */
1004         wait_on_all_pages_writeback(sbi);
1005
1006         if (unlikely(f2fs_cp_error(sbi)))
1007                 return;
1008
1009         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1010         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1011
1012         /* update user_block_counts */
1013         sbi->last_valid_block_count = sbi->total_valid_block_count;
1014         sbi->alloc_valid_block_count = 0;
1015
1016         /* Here, we only have one bio having CP pack */
1017         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1018
1019         /* wait for previous submitted meta pages writeback */
1020         wait_on_all_pages_writeback(sbi);
1021
1022         release_dirty_inode(sbi);
1023
1024         if (unlikely(f2fs_cp_error(sbi)))
1025                 return;
1026
1027         clear_prefree_segments(sbi);
1028         F2FS_RESET_SB_DIRT(sbi);
1029 }
1030
1031 /*
1032  * We guarantee that this checkpoint procedure will not fail.
1033  */
1034 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1035 {
1036         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1037         unsigned long long ckpt_ver;
1038
1039         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1040
1041         mutex_lock(&sbi->cp_mutex);
1042
1043         if (!sbi->s_dirty && cpc->reason != CP_DISCARD)
1044                 goto out;
1045         if (unlikely(f2fs_cp_error(sbi)))
1046                 goto out;
1047         if (block_operations(sbi))
1048                 goto out;
1049
1050         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1051
1052         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1053         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1054         f2fs_submit_merged_bio(sbi, META, WRITE);
1055
1056         /*
1057          * update checkpoint pack index
1058          * Increase the version number so that
1059          * SIT entries and seg summaries are written at correct place
1060          */
1061         ckpt_ver = cur_cp_version(ckpt);
1062         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1063
1064         /* write cached NAT/SIT entries to NAT/SIT area */
1065         flush_nat_entries(sbi);
1066         flush_sit_entries(sbi, cpc);
1067
1068         /* unlock all the fs_lock[] in do_checkpoint() */
1069         do_checkpoint(sbi, cpc);
1070
1071         unblock_operations(sbi);
1072         stat_inc_cp_count(sbi->stat_info);
1073 out:
1074         mutex_unlock(&sbi->cp_mutex);
1075         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1076 }
1077
1078 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1079 {
1080         int i;
1081
1082         for (i = 0; i < MAX_INO_ENTRY; i++) {
1083                 struct inode_management *im = &sbi->im[i];
1084
1085                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1086                 spin_lock_init(&im->ino_lock);
1087                 INIT_LIST_HEAD(&im->ino_list);
1088                 im->ino_num = 0;
1089         }
1090
1091         /*
1092          * considering 512 blocks in a segment 8 blocks are needed for cp
1093          * and log segment summaries. Remaining blocks are used to keep
1094          * orphan entries with the limitation one reserved segment
1095          * for cp pack we can have max 1020*504 orphan entries
1096          */
1097         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1098                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1099 }
1100
1101 int __init create_checkpoint_caches(void)
1102 {
1103         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1104                         sizeof(struct ino_entry));
1105         if (!ino_entry_slab)
1106                 return -ENOMEM;
1107         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1108                         sizeof(struct dir_inode_entry));
1109         if (!inode_entry_slab) {
1110                 kmem_cache_destroy(ino_entry_slab);
1111                 return -ENOMEM;
1112         }
1113         return 0;
1114 }
1115
1116 void destroy_checkpoint_caches(void)
1117 {
1118         kmem_cache_destroy(ino_entry_slab);
1119         kmem_cache_destroy(inode_entry_slab);
1120 }