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 repeat:
  54         page = grab_cache_page(mapping, index);
  55         if (!page) {
  56                 cond_resched();
  57                 goto repeat;
  58         }
  59         if (PageUptodate(page))
  60                 goto out;
  61
  62         if (f2fs_submit_page_bio(sbi, page, index,
  63                                 READ_SYNC | REQ_META | REQ_PRIO))
  64                 goto repeat;
  65
  66         lock_page(page);
  67         if (unlikely(page->mapping != mapping)) {
  68                 f2fs_put_page(page, 1);
  69                 goto repeat;
  70         }
  71 out:
  72         return page;
  73 }
  74
  75 struct page *get_meta_page_ra(struct f2fs_sb_info *sbi, pgoff_t index)
  76 {
  77         bool readahead = false;
  78         struct page *page;
  79
  80         page = find_get_page(META_MAPPING(sbi), index);
  81         if (!page || (page && !PageUptodate(page)))
  82                 readahead = true;
  83         f2fs_put_page(page, 0);
  84
  85         if (readahead)
  86                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
  87         return get_meta_page(sbi, index);
  88 }
  89
  90 static inline block_t get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  91 {
  92         switch (type) {
  93         case META_NAT:
  94                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  95         case META_SIT:
  96                 return SIT_BLK_CNT(sbi);
  97         case META_SSA:
  98         case META_CP:
  99                 return 0;
 100         case META_POR:
 101                 return MAX_BLKADDR(sbi);
 102         default:
 103                 BUG();
 104         }
 105 }
 106
 107 /*
 108  * Readahead CP/NAT/SIT/SSA pages
 109  */
 110 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
 111 {
 112         block_t prev_blk_addr = 0;
 113         struct page *page;
 114         block_t blkno = start;
 115         block_t max_blks = get_max_meta_blks(sbi, type);
 116
 117         struct f2fs_io_info fio = {
 118                 .type = META,
 119                 .rw = READ_SYNC | REQ_META | REQ_PRIO
 120         };
 121
 122         for (; nrpages-- > 0; blkno++) {
 123                 block_t blk_addr;
 124
 125                 switch (type) {
 126                 case META_NAT:
 127                         /* get nat block addr */
 128                         if (unlikely(blkno >= max_blks))
 129                                 blkno = 0;
 130                         blk_addr = current_nat_addr(sbi,
 131                                         blkno * NAT_ENTRY_PER_BLOCK);
 132                         break;
 133                 case META_SIT:
 134                         /* get sit block addr */
 135                         if (unlikely(blkno >= max_blks))
 136                                 goto out;
 137                         blk_addr = current_sit_addr(sbi,
 138                                         blkno * SIT_ENTRY_PER_BLOCK);
 139                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 140                                 goto out;
 141                         prev_blk_addr = blk_addr;
 142                         break;
 143                 case META_SSA:
 144                 case META_CP:
 145                 case META_POR:
 146                         if (unlikely(blkno >= max_blks))
 147                                 goto out;
 148                         if (unlikely(blkno < SEG0_BLKADDR(sbi)))
 149                                 goto out;
 150                         blk_addr = blkno;
 151                         break;
 152                 default:
 153                         BUG();
 154                 }
 155
 156                 page = grab_cache_page(META_MAPPING(sbi), 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, blk_addr, &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 static int f2fs_write_meta_page(struct page *page,
 173                                 struct writeback_control *wbc)
 174 {
 175         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 176
 177         trace_f2fs_writepage(page, META);
 178
 179         if (unlikely(sbi->por_doing))
 180                 goto redirty_out;
 181         if (wbc->for_reclaim)
 182                 goto redirty_out;
 183         if (unlikely(f2fs_cp_error(sbi)))
 184                 goto redirty_out;
 185
 186         f2fs_wait_on_page_writeback(page, META);
 187         write_meta_page(sbi, page);
 188         dec_page_count(sbi, F2FS_DIRTY_META);
 189         unlock_page(page);
 190         return 0;
 191
 192 redirty_out:
 193         redirty_page_for_writepage(wbc, page);
 194         return AOP_WRITEPAGE_ACTIVATE;
 195 }
 196
 197 static int f2fs_write_meta_pages(struct address_space *mapping,
 198                                 struct writeback_control *wbc)
 199 {
 200         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 201         long diff, written;
 202
 203         trace_f2fs_writepages(mapping->host, wbc, META);
 204
 205         /* collect a number of dirty meta pages and write together */
 206         if (wbc->for_kupdate ||
 207                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 208                 goto skip_write;
 209
 210         /* if mounting is failed, skip writing node pages */
 211         mutex_lock(&sbi->cp_mutex);
 212         diff = nr_pages_to_write(sbi, META, wbc);
 213         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 214         mutex_unlock(&sbi->cp_mutex);
 215         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 216         return 0;
 217
 218 skip_write:
 219         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 220         return 0;
 221 }
 222
 223 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 224                                                 long nr_to_write)
 225 {
 226         struct address_space *mapping = META_MAPPING(sbi);
 227         pgoff_t index = 0, end = LONG_MAX;
 228         struct pagevec pvec;
 229         long nwritten = 0;
 230         struct writeback_control wbc = {
 231                 .for_reclaim = 0,
 232         };
 233
 234         pagevec_init(&pvec, 0);
 235
 236         while (index <= end) {
 237                 int i, nr_pages;
 238                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 239                                 PAGECACHE_TAG_DIRTY,
 240                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 241                 if (unlikely(nr_pages == 0))
 242                         break;
 243
 244                 for (i = 0; i < nr_pages; i++) {
 245                         struct page *page = pvec.pages[i];
 246
 247                         lock_page(page);
 248
 249                         if (unlikely(page->mapping != mapping)) {
 250 continue_unlock:
 251                                 unlock_page(page);
 252                                 continue;
 253                         }
 254                         if (!PageDirty(page)) {
 255                                 /* someone wrote it for us */
 256                                 goto continue_unlock;
 257                         }
 258
 259                         if (!clear_page_dirty_for_io(page))
 260                                 goto continue_unlock;
 261
 262                         if (f2fs_write_meta_page(page, &wbc)) {
 263                                 unlock_page(page);
 264                                 break;
 265                         }
 266                         nwritten++;
 267                         if (unlikely(nwritten >= nr_to_write))
 268                                 break;
 269                 }
 270                 pagevec_release(&pvec);
 271                 cond_resched();
 272         }
 273
 274         if (nwritten)
 275                 f2fs_submit_merged_bio(sbi, type, WRITE);
 276
 277         return nwritten;
 278 }
 279
 280 static int f2fs_set_meta_page_dirty(struct page *page)
 281 {
 282         trace_f2fs_set_page_dirty(page, META);
 283
 284         SetPageUptodate(page);
 285         if (!PageDirty(page)) {
 286                 __set_page_dirty_nobuffers(page);
 287                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 288                 return 1;
 289         }
 290         return 0;
 291 }
 292
 293 const struct address_space_operations f2fs_meta_aops = {
 294         .writepage      = f2fs_write_meta_page,
 295         .writepages     = f2fs_write_meta_pages,
 296         .set_page_dirty = f2fs_set_meta_page_dirty,
 297 };
 298
 299 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 300 {
 301         struct ino_entry *e;
 302 retry:
 303         spin_lock(&sbi->ino_lock[type]);
 304
 305         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 306         if (!e) {
 307                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 308                 if (!e) {
 309                         spin_unlock(&sbi->ino_lock[type]);
 310                         goto retry;
 311                 }
 312                 if (radix_tree_insert(&sbi->ino_root[type], ino, e)) {
 313                         spin_unlock(&sbi->ino_lock[type]);
 314                         kmem_cache_free(ino_entry_slab, e);
 315                         goto retry;
 316                 }
 317                 memset(e, 0, sizeof(struct ino_entry));
 318                 e->ino = ino;
 319
 320                 list_add_tail(&e->list, &sbi->ino_list[type]);
 321         }
 322         spin_unlock(&sbi->ino_lock[type]);
 323 }
 324
 325 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 326 {
 327         struct ino_entry *e;
 328
 329         spin_lock(&sbi->ino_lock[type]);
 330         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 331         if (e) {
 332                 list_del(&e->list);
 333                 radix_tree_delete(&sbi->ino_root[type], ino);
 334                 if (type == ORPHAN_INO)
 335                         sbi->n_orphans--;
 336                 spin_unlock(&sbi->ino_lock[type]);
 337                 kmem_cache_free(ino_entry_slab, e);
 338                 return;
 339         }
 340         spin_unlock(&sbi->ino_lock[type]);
 341 }
 342
 343 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 344 {
 345         /* add new dirty ino entry into list */
 346         __add_ino_entry(sbi, ino, type);
 347 }
 348
 349 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 350 {
 351         /* remove dirty ino entry from list */
 352         __remove_ino_entry(sbi, ino, type);
 353 }
 354
 355 /* mode should be APPEND_INO or UPDATE_INO */
 356 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 357 {
 358         struct ino_entry *e;
 359         spin_lock(&sbi->ino_lock[mode]);
 360         e = radix_tree_lookup(&sbi->ino_root[mode], ino);
 361         spin_unlock(&sbi->ino_lock[mode]);
 362         return e ? true : false;
 363 }
 364
 365 void release_dirty_inode(struct f2fs_sb_info *sbi)
 366 {
 367         struct ino_entry *e, *tmp;
 368         int i;
 369
 370         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 371                 spin_lock(&sbi->ino_lock[i]);
 372                 list_for_each_entry_safe(e, tmp, &sbi->ino_list[i], list) {
 373                         list_del(&e->list);
 374                         radix_tree_delete(&sbi->ino_root[i], e->ino);
 375                         kmem_cache_free(ino_entry_slab, e);
 376                 }
 377                 spin_unlock(&sbi->ino_lock[i]);
 378         }
 379 }
 380
 381 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 382 {
 383         int err = 0;
 384
 385         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 386         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 387                 err = -ENOSPC;
 388         else
 389                 sbi->n_orphans++;
 390         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 391
 392         return err;
 393 }
 394
 395 void release_orphan_inode(struct f2fs_sb_info *sbi)
 396 {
 397         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 398         f2fs_bug_on(sbi, sbi->n_orphans == 0);
 399         sbi->n_orphans--;
 400         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 401 }
 402
 403 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 404 {
 405         /* add new orphan ino entry into list */
 406         __add_ino_entry(sbi, ino, ORPHAN_INO);
 407 }
 408
 409 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 410 {
 411         /* remove orphan entry from orphan list */
 412         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 413 }
 414
 415 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 416 {
 417         struct inode *inode = f2fs_iget(sbi->sb, ino);
 418         f2fs_bug_on(sbi, IS_ERR(inode));
 419         clear_nlink(inode);
 420
 421         /* truncate all the data during iput */
 422         iput(inode);
 423 }
 424
 425 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 426 {
 427         block_t start_blk, orphan_blkaddr, i, j;
 428
 429         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 430                 return;
 431
 432         sbi->por_doing = true;
 433
 434         start_blk = __start_cp_addr(sbi) + 1 +
 435                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 436         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 437
 438         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 439
 440         for (i = 0; i < orphan_blkaddr; i++) {
 441                 struct page *page = get_meta_page(sbi, start_blk + i);
 442                 struct f2fs_orphan_block *orphan_blk;
 443
 444                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 445                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 446                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 447                         recover_orphan_inode(sbi, ino);
 448                 }
 449                 f2fs_put_page(page, 1);
 450         }
 451         /* clear Orphan Flag */
 452         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 453         sbi->por_doing = false;
 454         return;
 455 }
 456
 457 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 458 {
 459         struct list_head *head;
 460         struct f2fs_orphan_block *orphan_blk = NULL;
 461         unsigned int nentries = 0;
 462         unsigned short index;
 463         unsigned short orphan_blocks =
 464                         (unsigned short)GET_ORPHAN_BLOCKS(sbi->n_orphans);
 465         struct page *page = NULL;
 466         struct ino_entry *orphan = NULL;
 467
 468         for (index = 0; index < orphan_blocks; index++)
 469                 grab_meta_page(sbi, start_blk + index);
 470
 471         index = 1;
 472         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 473         head = &sbi->ino_list[ORPHAN_INO];
 474
 475         /* loop for each orphan inode entry and write them in Jornal block */
 476         list_for_each_entry(orphan, head, list) {
 477                 if (!page) {
 478                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 479                         f2fs_bug_on(sbi, !page);
 480                         orphan_blk =
 481                                 (struct f2fs_orphan_block *)page_address(page);
 482                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 483                         f2fs_put_page(page, 0);
 484                 }
 485
 486                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 487
 488                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 489                         /*
 490                          * an orphan block is full of 1020 entries,
 491                          * then we need to flush current orphan blocks
 492                          * and bring another one in memory
 493                          */
 494                         orphan_blk->blk_addr = cpu_to_le16(index);
 495                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 496                         orphan_blk->entry_count = cpu_to_le32(nentries);
 497                         set_page_dirty(page);
 498                         f2fs_put_page(page, 1);
 499                         index++;
 500                         nentries = 0;
 501                         page = NULL;
 502                 }
 503         }
 504
 505         if (page) {
 506                 orphan_blk->blk_addr = cpu_to_le16(index);
 507                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 508                 orphan_blk->entry_count = cpu_to_le32(nentries);
 509                 set_page_dirty(page);
 510                 f2fs_put_page(page, 1);
 511         }
 512
 513         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 514 }
 515
 516 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 517                                 block_t cp_addr, unsigned long long *version)
 518 {
 519         struct page *cp_page_1, *cp_page_2 = NULL;
 520         unsigned long blk_size = sbi->blocksize;
 521         struct f2fs_checkpoint *cp_block;
 522         unsigned long long cur_version = 0, pre_version = 0;
 523         size_t crc_offset;
 524         __u32 crc = 0;
 525
 526         /* Read the 1st cp block in this CP pack */
 527         cp_page_1 = get_meta_page(sbi, cp_addr);
 528
 529         /* get the version number */
 530         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 531         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 532         if (crc_offset >= blk_size)
 533                 goto invalid_cp1;
 534
 535         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 536         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 537                 goto invalid_cp1;
 538
 539         pre_version = cur_cp_version(cp_block);
 540
 541         /* Read the 2nd cp block in this CP pack */
 542         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 543         cp_page_2 = get_meta_page(sbi, cp_addr);
 544
 545         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 546         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 547         if (crc_offset >= blk_size)
 548                 goto invalid_cp2;
 549
 550         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 551         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 552                 goto invalid_cp2;
 553
 554         cur_version = cur_cp_version(cp_block);
 555
 556         if (cur_version == pre_version) {
 557                 *version = cur_version;
 558                 f2fs_put_page(cp_page_2, 1);
 559                 return cp_page_1;
 560         }
 561 invalid_cp2:
 562         f2fs_put_page(cp_page_2, 1);
 563 invalid_cp1:
 564         f2fs_put_page(cp_page_1, 1);
 565         return NULL;
 566 }
 567
 568 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 569 {
 570         struct f2fs_checkpoint *cp_block;
 571         struct f2fs_super_block *fsb = sbi->raw_super;
 572         struct page *cp1, *cp2, *cur_page;
 573         unsigned long blk_size = sbi->blocksize;
 574         unsigned long long cp1_version = 0, cp2_version = 0;
 575         unsigned long long cp_start_blk_no;
 576         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 577         block_t cp_blk_no;
 578         int i;
 579
 580         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 581         if (!sbi->ckpt)
 582                 return -ENOMEM;
 583         /*
 584          * Finding out valid cp block involves read both
 585          * sets( cp pack1 and cp pack 2)
 586          */
 587         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 588         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 589
 590         /* The second checkpoint pack should start at the next segment */
 591         cp_start_blk_no += ((unsigned long long)1) <<
 592                                 le32_to_cpu(fsb->log_blocks_per_seg);
 593         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 594
 595         if (cp1 && cp2) {
 596                 if (ver_after(cp2_version, cp1_version))
 597                         cur_page = cp2;
 598                 else
 599                         cur_page = cp1;
 600         } else if (cp1) {
 601                 cur_page = cp1;
 602         } else if (cp2) {
 603                 cur_page = cp2;
 604         } else {
 605                 goto fail_no_cp;
 606         }
 607
 608         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 609         memcpy(sbi->ckpt, cp_block, blk_size);
 610
 611         if (cp_blks <= 1)
 612                 goto done;
 613
 614         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 615         if (cur_page == cp2)
 616                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 617
 618         for (i = 1; i < cp_blks; i++) {
 619                 void *sit_bitmap_ptr;
 620                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 621
 622                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 623                 sit_bitmap_ptr = page_address(cur_page);
 624                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 625                 f2fs_put_page(cur_page, 1);
 626         }
 627 done:
 628         f2fs_put_page(cp1, 1);
 629         f2fs_put_page(cp2, 1);
 630         return 0;
 631
 632 fail_no_cp:
 633         kfree(sbi->ckpt);
 634         return -EINVAL;
 635 }
 636
 637 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 638 {
 639         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 640
 641         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 642                 return -EEXIST;
 643
 644         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 645         F2FS_I(inode)->dirty_dir = new;
 646         list_add_tail(&new->list, &sbi->dir_inode_list);
 647         stat_inc_dirty_dir(sbi);
 648         return 0;
 649 }
 650
 651 void update_dirty_page(struct inode *inode, struct page *page)
 652 {
 653         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 654         struct dir_inode_entry *new;
 655         int ret = 0;
 656
 657         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
 658                 return;
 659
 660         if (!S_ISDIR(inode->i_mode)) {
 661                 inode_inc_dirty_pages(inode);
 662                 goto out;
 663         }
 664
 665         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 666         new->inode = inode;
 667         INIT_LIST_HEAD(&new->list);
 668
 669         spin_lock(&sbi->dir_inode_lock);
 670         ret = __add_dirty_inode(inode, new);
 671         inode_inc_dirty_pages(inode);
 672         spin_unlock(&sbi->dir_inode_lock);
 673
 674         if (ret)
 675                 kmem_cache_free(inode_entry_slab, new);
 676 out:
 677         SetPagePrivate(page);
 678 }
 679
 680 void add_dirty_dir_inode(struct inode *inode)
 681 {
 682         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 683         struct dir_inode_entry *new =
 684                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 685         int ret = 0;
 686
 687         new->inode = inode;
 688         INIT_LIST_HEAD(&new->list);
 689
 690         spin_lock(&sbi->dir_inode_lock);
 691         ret = __add_dirty_inode(inode, new);
 692         spin_unlock(&sbi->dir_inode_lock);
 693
 694         if (ret)
 695                 kmem_cache_free(inode_entry_slab, new);
 696 }
 697
 698 void remove_dirty_dir_inode(struct inode *inode)
 699 {
 700         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 701         struct dir_inode_entry *entry;
 702
 703         if (!S_ISDIR(inode->i_mode))
 704                 return;
 705
 706         spin_lock(&sbi->dir_inode_lock);
 707         if (get_dirty_pages(inode) ||
 708                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 709                 spin_unlock(&sbi->dir_inode_lock);
 710                 return;
 711         }
 712
 713         entry = F2FS_I(inode)->dirty_dir;
 714         list_del(&entry->list);
 715         F2FS_I(inode)->dirty_dir = NULL;
 716         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 717         stat_dec_dirty_dir(sbi);
 718         spin_unlock(&sbi->dir_inode_lock);
 719         kmem_cache_free(inode_entry_slab, entry);
 720
 721         /* Only from the recovery routine */
 722         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 723                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 724                 iput(inode);
 725         }
 726 }
 727
 728 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 729 {
 730         struct list_head *head;
 731         struct dir_inode_entry *entry;
 732         struct inode *inode;
 733 retry:
 734         if (unlikely(f2fs_cp_error(sbi)))
 735                 return;
 736
 737         spin_lock(&sbi->dir_inode_lock);
 738
 739         head = &sbi->dir_inode_list;
 740         if (list_empty(head)) {
 741                 spin_unlock(&sbi->dir_inode_lock);
 742                 return;
 743         }
 744         entry = list_entry(head->next, struct dir_inode_entry, list);
 745         inode = igrab(entry->inode);
 746         spin_unlock(&sbi->dir_inode_lock);
 747         if (inode) {
 748                 filemap_fdatawrite(inode->i_mapping);
 749                 iput(inode);
 750         } else {
 751                 /*
 752                  * We should submit bio, since it exists several
 753                  * wribacking dentry pages in the freeing inode.
 754                  */
 755                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 756         }
 757         goto retry;
 758 }
 759
 760 /*
 761  * Freeze all the FS-operations for checkpoint.
 762  */
 763 static int block_operations(struct f2fs_sb_info *sbi)
 764 {
 765         struct writeback_control wbc = {
 766                 .sync_mode = WB_SYNC_ALL,
 767                 .nr_to_write = LONG_MAX,
 768                 .for_reclaim = 0,
 769         };
 770         struct blk_plug plug;
 771         int err = 0;
 772
 773         blk_start_plug(&plug);
 774
 775 retry_flush_dents:
 776         f2fs_lock_all(sbi);
 777         /* write all the dirty dentry pages */
 778         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 779                 f2fs_unlock_all(sbi);
 780                 sync_dirty_dir_inodes(sbi);
 781                 if (unlikely(f2fs_cp_error(sbi))) {
 782                         err = -EIO;
 783                         goto out;
 784                 }
 785                 goto retry_flush_dents;
 786         }
 787
 788         /*
 789          * POR: we should ensure that there are no dirty node pages
 790          * until finishing nat/sit flush.
 791          */
 792 retry_flush_nodes:
 793         down_write(&sbi->node_write);
 794
 795         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 796                 up_write(&sbi->node_write);
 797                 sync_node_pages(sbi, 0, &wbc);
 798                 if (unlikely(f2fs_cp_error(sbi))) {
 799                         f2fs_unlock_all(sbi);
 800                         err = -EIO;
 801                         goto out;
 802                 }
 803                 goto retry_flush_nodes;
 804         }
 805 out:
 806         blk_finish_plug(&plug);
 807         return err;
 808 }
 809
 810 static void unblock_operations(struct f2fs_sb_info *sbi)
 811 {
 812         up_write(&sbi->node_write);
 813         f2fs_unlock_all(sbi);
 814 }
 815
 816 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 817 {
 818         DEFINE_WAIT(wait);
 819
 820         for (;;) {
 821                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 822
 823                 if (!get_pages(sbi, F2FS_WRITEBACK))
 824                         break;
 825
 826                 io_schedule();
 827         }
 828         finish_wait(&sbi->cp_wait, &wait);
 829 }
 830
 831 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 832 {
 833         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 834         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 835         struct f2fs_nm_info *nm_i = NM_I(sbi);
 836         nid_t last_nid = nm_i->next_scan_nid;
 837         block_t start_blk;
 838         struct page *cp_page;
 839         unsigned int data_sum_blocks, orphan_blocks;
 840         __u32 crc32 = 0;
 841         void *kaddr;
 842         int i;
 843         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 844
 845         /*
 846          * This avoids to conduct wrong roll-forward operations and uses
 847          * metapages, so should be called prior to sync_meta_pages below.
 848          */
 849         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 850
 851         /* Flush all the NAT/SIT pages */
 852         while (get_pages(sbi, F2FS_DIRTY_META)) {
 853                 sync_meta_pages(sbi, META, LONG_MAX);
 854                 if (unlikely(f2fs_cp_error(sbi)))
 855                         return;
 856         }
 857
 858         next_free_nid(sbi, &last_nid);
 859
 860         /*
 861          * modify checkpoint
 862          * version number is already updated
 863          */
 864         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 865         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 866         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 867         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 868                 ckpt->cur_node_segno[i] =
 869                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 870                 ckpt->cur_node_blkoff[i] =
 871                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 872                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 873                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 874         }
 875         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 876                 ckpt->cur_data_segno[i] =
 877                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 878                 ckpt->cur_data_blkoff[i] =
 879                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 880                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 881                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 882         }
 883
 884         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 885         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 886         ckpt->next_free_nid = cpu_to_le32(last_nid);
 887
 888         /* 2 cp  + n data seg summary + orphan inode blocks */
 889         data_sum_blocks = npages_for_summary_flush(sbi);
 890         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 891                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 892         else
 893                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 894
 895         orphan_blocks = GET_ORPHAN_BLOCKS(sbi->n_orphans);
 896         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 897                         orphan_blocks);
 898
 899         if (cpc->reason == CP_UMOUNT) {
 900                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 901                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 902                                 cp_payload_blks + data_sum_blocks +
 903                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 904         } else {
 905                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 906                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 907                                 cp_payload_blks + data_sum_blocks +
 908                                 orphan_blocks);
 909         }
 910
 911         if (sbi->n_orphans)
 912                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 913         else
 914                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 915
 916         if (sbi->need_fsck)
 917                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
 918
 919         /* update SIT/NAT bitmap */
 920         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 921         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 922
 923         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 924         *((__le32 *)((unsigned char *)ckpt +
 925                                 le32_to_cpu(ckpt->checksum_offset)))
 926                                 = cpu_to_le32(crc32);
 927
 928         start_blk = __start_cp_addr(sbi);
 929
 930         /* write out checkpoint buffer at block 0 */
 931         cp_page = grab_meta_page(sbi, start_blk++);
 932         kaddr = page_address(cp_page);
 933         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 934         set_page_dirty(cp_page);
 935         f2fs_put_page(cp_page, 1);
 936
 937         for (i = 1; i < 1 + cp_payload_blks; i++) {
 938                 cp_page = grab_meta_page(sbi, start_blk++);
 939                 kaddr = page_address(cp_page);
 940                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 941                                 (1 << sbi->log_blocksize));
 942                 set_page_dirty(cp_page);
 943                 f2fs_put_page(cp_page, 1);
 944         }
 945
 946         if (sbi->n_orphans) {
 947                 write_orphan_inodes(sbi, start_blk);
 948                 start_blk += orphan_blocks;
 949         }
 950
 951         write_data_summaries(sbi, start_blk);
 952         start_blk += data_sum_blocks;
 953         if (cpc->reason == CP_UMOUNT) {
 954                 write_node_summaries(sbi, start_blk);
 955                 start_blk += NR_CURSEG_NODE_TYPE;
 956         }
 957
 958         /* writeout checkpoint block */
 959         cp_page = grab_meta_page(sbi, start_blk);
 960         kaddr = page_address(cp_page);
 961         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 962         set_page_dirty(cp_page);
 963         f2fs_put_page(cp_page, 1);
 964
 965         /* wait for previous submitted node/meta pages writeback */
 966         wait_on_all_pages_writeback(sbi);
 967
 968         if (unlikely(f2fs_cp_error(sbi)))
 969                 return;
 970
 971         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 972         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 973
 974         /* update user_block_counts */
 975         sbi->last_valid_block_count = sbi->total_valid_block_count;
 976         sbi->alloc_valid_block_count = 0;
 977
 978         /* Here, we only have one bio having CP pack */
 979         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 980
 981         release_dirty_inode(sbi);
 982
 983         if (unlikely(f2fs_cp_error(sbi)))
 984                 return;
 985
 986         clear_prefree_segments(sbi);
 987         F2FS_RESET_SB_DIRT(sbi);
 988 }
 989
 990 /*
 991  * We guarantee that this checkpoint procedure will not fail.
 992  */
 993 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 994 {
 995         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 996         unsigned long long ckpt_ver;
 997
 998         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
 999
1000         mutex_lock(&sbi->cp_mutex);
1001
1002         if (!sbi->s_dirty && cpc->reason != CP_DISCARD)
1003                 goto out;
1004         if (unlikely(f2fs_cp_error(sbi)))
1005                 goto out;
1006         if (block_operations(sbi))
1007                 goto out;
1008
1009         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1010
1011         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1012         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1013         f2fs_submit_merged_bio(sbi, META, WRITE);
1014
1015         /*
1016          * update checkpoint pack index
1017          * Increase the version number so that
1018          * SIT entries and seg summaries are written at correct place
1019          */
1020         ckpt_ver = cur_cp_version(ckpt);
1021         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1022
1023         /* write cached NAT/SIT entries to NAT/SIT area */
1024         flush_nat_entries(sbi);
1025         flush_sit_entries(sbi, cpc);
1026
1027         /* unlock all the fs_lock[] in do_checkpoint() */
1028         do_checkpoint(sbi, cpc);
1029
1030         unblock_operations(sbi);
1031         stat_inc_cp_count(sbi->stat_info);
1032 out:
1033         mutex_unlock(&sbi->cp_mutex);
1034         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1035 }
1036
1037 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1038 {
1039         int i;
1040
1041         for (i = 0; i < MAX_INO_ENTRY; i++) {
1042                 INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
1043                 spin_lock_init(&sbi->ino_lock[i]);
1044                 INIT_LIST_HEAD(&sbi->ino_list[i]);
1045         }
1046
1047         /*
1048          * considering 512 blocks in a segment 8 blocks are needed for cp
1049          * and log segment summaries. Remaining blocks are used to keep
1050          * orphan entries with the limitation one reserved segment
1051          * for cp pack we can have max 1020*504 orphan entries
1052          */
1053         sbi->n_orphans = 0;
1054         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1055                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1056 }
1057
1058 int __init create_checkpoint_caches(void)
1059 {
1060         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1061                         sizeof(struct ino_entry));
1062         if (!ino_entry_slab)
1063                 return -ENOMEM;
1064         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1065                         sizeof(struct dir_inode_entry));
1066         if (!inode_entry_slab) {
1067                 kmem_cache_destroy(ino_entry_slab);
1068                 return -ENOMEM;
1069         }
1070         return 0;
1071 }
1072
1073 void destroy_checkpoint_caches(void)
1074 {
1075         kmem_cache_destroy(ino_entry_slab);
1076         kmem_cache_destroy(inode_entry_slab);
1077 }