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,
  87                                 MAX_BIO_BLOCKS(max_hw_blocks(sbi)), META_POR);
  88         return get_meta_page(sbi, index);
  89 }
  90
  91 static inline block_t get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
  92 {
  93         switch (type) {
  94         case META_NAT:
  95                 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
  96         case META_SIT:
  97                 return SIT_BLK_CNT(sbi);
  98         case META_SSA:
  99         case META_CP:
 100                 return 0;
 101         case META_POR:
 102                 return SM_I(sbi)->seg0_blkaddr + TOTAL_BLKS(sbi);
 103         default:
 104                 BUG();
 105         }
 106 }
 107
 108 /*
 109  * Readahead CP/NAT/SIT/SSA pages
 110  */
 111 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
 112 {
 113         block_t prev_blk_addr = 0;
 114         struct page *page;
 115         block_t blkno = start;
 116         block_t max_blks = get_max_meta_blks(sbi, type);
 117         block_t min_blks = SM_I(sbi)->seg0_blkaddr;
 118
 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                 block_t blk_addr;
 126
 127                 switch (type) {
 128                 case META_NAT:
 129                         /* get nat block addr */
 130                         if (unlikely(blkno >= max_blks))
 131                                 blkno = 0;
 132                         blk_addr = current_nat_addr(sbi,
 133                                         blkno * NAT_ENTRY_PER_BLOCK);
 134                         break;
 135                 case META_SIT:
 136                         /* get sit block addr */
 137                         if (unlikely(blkno >= max_blks))
 138                                 goto out;
 139                         blk_addr = current_sit_addr(sbi,
 140                                         blkno * SIT_ENTRY_PER_BLOCK);
 141                         if (blkno != start && prev_blk_addr + 1 != blk_addr)
 142                                 goto out;
 143                         prev_blk_addr = blk_addr;
 144                         break;
 145                 case META_SSA:
 146                 case META_CP:
 147                 case META_POR:
 148                         if (unlikely(blkno >= max_blks))
 149                                 goto out;
 150                         if (unlikely(blkno < min_blks))
 151                                 goto out;
 152                         blk_addr = blkno;
 153                         break;
 154                 default:
 155                         BUG();
 156                 }
 157
 158                 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
 159                 if (!page)
 160                         continue;
 161                 if (PageUptodate(page)) {
 162                         f2fs_put_page(page, 1);
 163                         continue;
 164                 }
 165
 166                 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
 167                 f2fs_put_page(page, 0);
 168         }
 169 out:
 170         f2fs_submit_merged_bio(sbi, META, READ);
 171         return blkno - start;
 172 }
 173
 174 static int f2fs_write_meta_page(struct page *page,
 175                                 struct writeback_control *wbc)
 176 {
 177         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 178
 179         trace_f2fs_writepage(page, META);
 180
 181         if (unlikely(sbi->por_doing))
 182                 goto redirty_out;
 183         if (wbc->for_reclaim)
 184                 goto redirty_out;
 185         if (unlikely(f2fs_cp_error(sbi)))
 186                 goto redirty_out;
 187
 188         f2fs_wait_on_page_writeback(page, META);
 189         write_meta_page(sbi, page);
 190         dec_page_count(sbi, F2FS_DIRTY_META);
 191         unlock_page(page);
 192         return 0;
 193
 194 redirty_out:
 195         redirty_page_for_writepage(wbc, page);
 196         return AOP_WRITEPAGE_ACTIVATE;
 197 }
 198
 199 static int f2fs_write_meta_pages(struct address_space *mapping,
 200                                 struct writeback_control *wbc)
 201 {
 202         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 203         long diff, written;
 204
 205         trace_f2fs_writepages(mapping->host, wbc, META);
 206
 207         /* collect a number of dirty meta pages and write together */
 208         if (wbc->for_kupdate ||
 209                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 210                 goto skip_write;
 211
 212         /* if mounting is failed, skip writing node pages */
 213         mutex_lock(&sbi->cp_mutex);
 214         diff = nr_pages_to_write(sbi, META, wbc);
 215         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 216         mutex_unlock(&sbi->cp_mutex);
 217         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 218         return 0;
 219
 220 skip_write:
 221         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 222         return 0;
 223 }
 224
 225 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 226                                                 long nr_to_write)
 227 {
 228         struct address_space *mapping = META_MAPPING(sbi);
 229         pgoff_t index = 0, end = LONG_MAX;
 230         struct pagevec pvec;
 231         long nwritten = 0;
 232         struct writeback_control wbc = {
 233                 .for_reclaim = 0,
 234         };
 235
 236         pagevec_init(&pvec, 0);
 237
 238         while (index <= end) {
 239                 int i, nr_pages;
 240                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 241                                 PAGECACHE_TAG_DIRTY,
 242                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 243                 if (unlikely(nr_pages == 0))
 244                         break;
 245
 246                 for (i = 0; i < nr_pages; i++) {
 247                         struct page *page = pvec.pages[i];
 248
 249                         lock_page(page);
 250
 251                         if (unlikely(page->mapping != mapping)) {
 252 continue_unlock:
 253                                 unlock_page(page);
 254                                 continue;
 255                         }
 256                         if (!PageDirty(page)) {
 257                                 /* someone wrote it for us */
 258                                 goto continue_unlock;
 259                         }
 260
 261                         if (!clear_page_dirty_for_io(page))
 262                                 goto continue_unlock;
 263
 264                         if (f2fs_write_meta_page(page, &wbc)) {
 265                                 unlock_page(page);
 266                                 break;
 267                         }
 268                         nwritten++;
 269                         if (unlikely(nwritten >= nr_to_write))
 270                                 break;
 271                 }
 272                 pagevec_release(&pvec);
 273                 cond_resched();
 274         }
 275
 276         if (nwritten)
 277                 f2fs_submit_merged_bio(sbi, type, WRITE);
 278
 279         return nwritten;
 280 }
 281
 282 static int f2fs_set_meta_page_dirty(struct page *page)
 283 {
 284         trace_f2fs_set_page_dirty(page, META);
 285
 286         SetPageUptodate(page);
 287         if (!PageDirty(page)) {
 288                 __set_page_dirty_nobuffers(page);
 289                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 290                 return 1;
 291         }
 292         return 0;
 293 }
 294
 295 const struct address_space_operations f2fs_meta_aops = {
 296         .writepage      = f2fs_write_meta_page,
 297         .writepages     = f2fs_write_meta_pages,
 298         .set_page_dirty = f2fs_set_meta_page_dirty,
 299 };
 300
 301 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 302 {
 303         struct ino_entry *e;
 304 retry:
 305         spin_lock(&sbi->ino_lock[type]);
 306
 307         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 308         if (!e) {
 309                 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
 310                 if (!e) {
 311                         spin_unlock(&sbi->ino_lock[type]);
 312                         goto retry;
 313                 }
 314                 if (radix_tree_insert(&sbi->ino_root[type], ino, e)) {
 315                         spin_unlock(&sbi->ino_lock[type]);
 316                         kmem_cache_free(ino_entry_slab, e);
 317                         goto retry;
 318                 }
 319                 memset(e, 0, sizeof(struct ino_entry));
 320                 e->ino = ino;
 321
 322                 list_add_tail(&e->list, &sbi->ino_list[type]);
 323         }
 324         spin_unlock(&sbi->ino_lock[type]);
 325 }
 326
 327 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 328 {
 329         struct ino_entry *e;
 330
 331         spin_lock(&sbi->ino_lock[type]);
 332         e = radix_tree_lookup(&sbi->ino_root[type], ino);
 333         if (e) {
 334                 list_del(&e->list);
 335                 radix_tree_delete(&sbi->ino_root[type], ino);
 336                 if (type == ORPHAN_INO)
 337                         sbi->n_orphans--;
 338                 spin_unlock(&sbi->ino_lock[type]);
 339                 kmem_cache_free(ino_entry_slab, e);
 340                 return;
 341         }
 342         spin_unlock(&sbi->ino_lock[type]);
 343 }
 344
 345 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 346 {
 347         /* add new dirty ino entry into list */
 348         __add_ino_entry(sbi, ino, type);
 349 }
 350
 351 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 352 {
 353         /* remove dirty ino entry from list */
 354         __remove_ino_entry(sbi, ino, type);
 355 }
 356
 357 /* mode should be APPEND_INO or UPDATE_INO */
 358 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 359 {
 360         struct ino_entry *e;
 361         spin_lock(&sbi->ino_lock[mode]);
 362         e = radix_tree_lookup(&sbi->ino_root[mode], ino);
 363         spin_unlock(&sbi->ino_lock[mode]);
 364         return e ? true : false;
 365 }
 366
 367 void release_dirty_inode(struct f2fs_sb_info *sbi)
 368 {
 369         struct ino_entry *e, *tmp;
 370         int i;
 371
 372         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 373                 spin_lock(&sbi->ino_lock[i]);
 374                 list_for_each_entry_safe(e, tmp, &sbi->ino_list[i], list) {
 375                         list_del(&e->list);
 376                         radix_tree_delete(&sbi->ino_root[i], e->ino);
 377                         kmem_cache_free(ino_entry_slab, e);
 378                 }
 379                 spin_unlock(&sbi->ino_lock[i]);
 380         }
 381 }
 382
 383 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 384 {
 385         int err = 0;
 386
 387         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 388         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 389                 err = -ENOSPC;
 390         else
 391                 sbi->n_orphans++;
 392         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 393
 394         return err;
 395 }
 396
 397 void release_orphan_inode(struct f2fs_sb_info *sbi)
 398 {
 399         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 400         f2fs_bug_on(sbi, sbi->n_orphans == 0);
 401         sbi->n_orphans--;
 402         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 403 }
 404
 405 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 406 {
 407         /* add new orphan ino entry into list */
 408         __add_ino_entry(sbi, ino, ORPHAN_INO);
 409 }
 410
 411 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 412 {
 413         /* remove orphan entry from orphan list */
 414         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 415 }
 416
 417 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 418 {
 419         struct inode *inode = f2fs_iget(sbi->sb, ino);
 420         f2fs_bug_on(sbi, IS_ERR(inode));
 421         clear_nlink(inode);
 422
 423         /* truncate all the data during iput */
 424         iput(inode);
 425 }
 426
 427 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 428 {
 429         block_t start_blk, orphan_blkaddr, i, j;
 430
 431         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 432                 return;
 433
 434         sbi->por_doing = true;
 435
 436         start_blk = __start_cp_addr(sbi) + 1 +
 437                 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 438         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 439
 440         ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
 441
 442         for (i = 0; i < orphan_blkaddr; i++) {
 443                 struct page *page = get_meta_page(sbi, start_blk + i);
 444                 struct f2fs_orphan_block *orphan_blk;
 445
 446                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 447                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 448                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 449                         recover_orphan_inode(sbi, ino);
 450                 }
 451                 f2fs_put_page(page, 1);
 452         }
 453         /* clear Orphan Flag */
 454         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 455         sbi->por_doing = false;
 456         return;
 457 }
 458
 459 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 460 {
 461         struct list_head *head;
 462         struct f2fs_orphan_block *orphan_blk = NULL;
 463         unsigned int nentries = 0;
 464         unsigned short index;
 465         unsigned short orphan_blocks =
 466                         (unsigned short)GET_ORPHAN_BLOCKS(sbi->n_orphans);
 467         struct page *page = NULL;
 468         struct ino_entry *orphan = NULL;
 469
 470         for (index = 0; index < orphan_blocks; index++)
 471                 grab_meta_page(sbi, start_blk + index);
 472
 473         index = 1;
 474         spin_lock(&sbi->ino_lock[ORPHAN_INO]);
 475         head = &sbi->ino_list[ORPHAN_INO];
 476
 477         /* loop for each orphan inode entry and write them in Jornal block */
 478         list_for_each_entry(orphan, head, list) {
 479                 if (!page) {
 480                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 481                         f2fs_bug_on(sbi, !page);
 482                         orphan_blk =
 483                                 (struct f2fs_orphan_block *)page_address(page);
 484                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 485                         f2fs_put_page(page, 0);
 486                 }
 487
 488                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 489
 490                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 491                         /*
 492                          * an orphan block is full of 1020 entries,
 493                          * then we need to flush current orphan blocks
 494                          * and bring another one in memory
 495                          */
 496                         orphan_blk->blk_addr = cpu_to_le16(index);
 497                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 498                         orphan_blk->entry_count = cpu_to_le32(nentries);
 499                         set_page_dirty(page);
 500                         f2fs_put_page(page, 1);
 501                         index++;
 502                         nentries = 0;
 503                         page = NULL;
 504                 }
 505         }
 506
 507         if (page) {
 508                 orphan_blk->blk_addr = cpu_to_le16(index);
 509                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 510                 orphan_blk->entry_count = cpu_to_le32(nentries);
 511                 set_page_dirty(page);
 512                 f2fs_put_page(page, 1);
 513         }
 514
 515         spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
 516 }
 517
 518 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 519                                 block_t cp_addr, unsigned long long *version)
 520 {
 521         struct page *cp_page_1, *cp_page_2 = NULL;
 522         unsigned long blk_size = sbi->blocksize;
 523         struct f2fs_checkpoint *cp_block;
 524         unsigned long long cur_version = 0, pre_version = 0;
 525         size_t crc_offset;
 526         __u32 crc = 0;
 527
 528         /* Read the 1st cp block in this CP pack */
 529         cp_page_1 = get_meta_page(sbi, cp_addr);
 530
 531         /* get the version number */
 532         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 533         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 534         if (crc_offset >= blk_size)
 535                 goto invalid_cp1;
 536
 537         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 538         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 539                 goto invalid_cp1;
 540
 541         pre_version = cur_cp_version(cp_block);
 542
 543         /* Read the 2nd cp block in this CP pack */
 544         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 545         cp_page_2 = get_meta_page(sbi, cp_addr);
 546
 547         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 548         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 549         if (crc_offset >= blk_size)
 550                 goto invalid_cp2;
 551
 552         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 553         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 554                 goto invalid_cp2;
 555
 556         cur_version = cur_cp_version(cp_block);
 557
 558         if (cur_version == pre_version) {
 559                 *version = cur_version;
 560                 f2fs_put_page(cp_page_2, 1);
 561                 return cp_page_1;
 562         }
 563 invalid_cp2:
 564         f2fs_put_page(cp_page_2, 1);
 565 invalid_cp1:
 566         f2fs_put_page(cp_page_1, 1);
 567         return NULL;
 568 }
 569
 570 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 571 {
 572         struct f2fs_checkpoint *cp_block;
 573         struct f2fs_super_block *fsb = sbi->raw_super;
 574         struct page *cp1, *cp2, *cur_page;
 575         unsigned long blk_size = sbi->blocksize;
 576         unsigned long long cp1_version = 0, cp2_version = 0;
 577         unsigned long long cp_start_blk_no;
 578         unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 579         block_t cp_blk_no;
 580         int i;
 581
 582         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 583         if (!sbi->ckpt)
 584                 return -ENOMEM;
 585         /*
 586          * Finding out valid cp block involves read both
 587          * sets( cp pack1 and cp pack 2)
 588          */
 589         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 590         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 591
 592         /* The second checkpoint pack should start at the next segment */
 593         cp_start_blk_no += ((unsigned long long)1) <<
 594                                 le32_to_cpu(fsb->log_blocks_per_seg);
 595         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 596
 597         if (cp1 && cp2) {
 598                 if (ver_after(cp2_version, cp1_version))
 599                         cur_page = cp2;
 600                 else
 601                         cur_page = cp1;
 602         } else if (cp1) {
 603                 cur_page = cp1;
 604         } else if (cp2) {
 605                 cur_page = cp2;
 606         } else {
 607                 goto fail_no_cp;
 608         }
 609
 610         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 611         memcpy(sbi->ckpt, cp_block, blk_size);
 612
 613         if (cp_blks <= 1)
 614                 goto done;
 615
 616         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 617         if (cur_page == cp2)
 618                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 619
 620         for (i = 1; i < cp_blks; i++) {
 621                 void *sit_bitmap_ptr;
 622                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 623
 624                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 625                 sit_bitmap_ptr = page_address(cur_page);
 626                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 627                 f2fs_put_page(cur_page, 1);
 628         }
 629 done:
 630         f2fs_put_page(cp1, 1);
 631         f2fs_put_page(cp2, 1);
 632         return 0;
 633
 634 fail_no_cp:
 635         kfree(sbi->ckpt);
 636         return -EINVAL;
 637 }
 638
 639 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 640 {
 641         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 642
 643         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 644                 return -EEXIST;
 645
 646         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 647         F2FS_I(inode)->dirty_dir = new;
 648         list_add_tail(&new->list, &sbi->dir_inode_list);
 649         stat_inc_dirty_dir(sbi);
 650         return 0;
 651 }
 652
 653 void update_dirty_page(struct inode *inode, struct page *page)
 654 {
 655         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 656         struct dir_inode_entry *new;
 657         int ret = 0;
 658
 659         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
 660                 return;
 661
 662         if (!S_ISDIR(inode->i_mode)) {
 663                 inode_inc_dirty_pages(inode);
 664                 goto out;
 665         }
 666
 667         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 668         new->inode = inode;
 669         INIT_LIST_HEAD(&new->list);
 670
 671         spin_lock(&sbi->dir_inode_lock);
 672         ret = __add_dirty_inode(inode, new);
 673         inode_inc_dirty_pages(inode);
 674         spin_unlock(&sbi->dir_inode_lock);
 675
 676         if (ret)
 677                 kmem_cache_free(inode_entry_slab, new);
 678 out:
 679         SetPagePrivate(page);
 680 }
 681
 682 void add_dirty_dir_inode(struct inode *inode)
 683 {
 684         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 685         struct dir_inode_entry *new =
 686                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 687         int ret = 0;
 688
 689         new->inode = inode;
 690         INIT_LIST_HEAD(&new->list);
 691
 692         spin_lock(&sbi->dir_inode_lock);
 693         ret = __add_dirty_inode(inode, new);
 694         spin_unlock(&sbi->dir_inode_lock);
 695
 696         if (ret)
 697                 kmem_cache_free(inode_entry_slab, new);
 698 }
 699
 700 void remove_dirty_dir_inode(struct inode *inode)
 701 {
 702         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 703         struct dir_inode_entry *entry;
 704
 705         if (!S_ISDIR(inode->i_mode))
 706                 return;
 707
 708         spin_lock(&sbi->dir_inode_lock);
 709         if (get_dirty_pages(inode) ||
 710                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 711                 spin_unlock(&sbi->dir_inode_lock);
 712                 return;
 713         }
 714
 715         entry = F2FS_I(inode)->dirty_dir;
 716         list_del(&entry->list);
 717         F2FS_I(inode)->dirty_dir = NULL;
 718         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 719         stat_dec_dirty_dir(sbi);
 720         spin_unlock(&sbi->dir_inode_lock);
 721         kmem_cache_free(inode_entry_slab, entry);
 722
 723         /* Only from the recovery routine */
 724         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 725                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 726                 iput(inode);
 727         }
 728 }
 729
 730 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 731 {
 732         struct list_head *head;
 733         struct dir_inode_entry *entry;
 734         struct inode *inode;
 735 retry:
 736         spin_lock(&sbi->dir_inode_lock);
 737
 738         head = &sbi->dir_inode_list;
 739         if (list_empty(head)) {
 740                 spin_unlock(&sbi->dir_inode_lock);
 741                 return;
 742         }
 743         entry = list_entry(head->next, struct dir_inode_entry, list);
 744         inode = igrab(entry->inode);
 745         spin_unlock(&sbi->dir_inode_lock);
 746         if (inode) {
 747                 filemap_fdatawrite(inode->i_mapping);
 748                 iput(inode);
 749         } else {
 750                 /*
 751                  * We should submit bio, since it exists several
 752                  * wribacking dentry pages in the freeing inode.
 753                  */
 754                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 755         }
 756         goto retry;
 757 }
 758
 759 /*
 760  * Freeze all the FS-operations for checkpoint.
 761  */
 762 static int block_operations(struct f2fs_sb_info *sbi)
 763 {
 764         struct writeback_control wbc = {
 765                 .sync_mode = WB_SYNC_ALL,
 766                 .nr_to_write = LONG_MAX,
 767                 .for_reclaim = 0,
 768         };
 769         struct blk_plug plug;
 770         int err = 0;
 771
 772         blk_start_plug(&plug);
 773
 774 retry_flush_dents:
 775         f2fs_lock_all(sbi);
 776         /* write all the dirty dentry pages */
 777         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 778                 f2fs_unlock_all(sbi);
 779                 sync_dirty_dir_inodes(sbi);
 780                 if (unlikely(f2fs_cp_error(sbi))) {
 781                         err = -EIO;
 782                         goto out;
 783                 }
 784                 goto retry_flush_dents;
 785         }
 786
 787         /*
 788          * POR: we should ensure that there are no dirty node pages
 789          * until finishing nat/sit flush.
 790          */
 791 retry_flush_nodes:
 792         down_write(&sbi->node_write);
 793
 794         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 795                 up_write(&sbi->node_write);
 796                 sync_node_pages(sbi, 0, &wbc);
 797                 if (unlikely(f2fs_cp_error(sbi))) {
 798                         f2fs_unlock_all(sbi);
 799                         err = -EIO;
 800                         goto out;
 801                 }
 802                 goto retry_flush_nodes;
 803         }
 804 out:
 805         blk_finish_plug(&plug);
 806         return err;
 807 }
 808
 809 static void unblock_operations(struct f2fs_sb_info *sbi)
 810 {
 811         up_write(&sbi->node_write);
 812         f2fs_unlock_all(sbi);
 813 }
 814
 815 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 816 {
 817         DEFINE_WAIT(wait);
 818
 819         for (;;) {
 820                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 821
 822                 if (!get_pages(sbi, F2FS_WRITEBACK))
 823                         break;
 824
 825                 io_schedule();
 826         }
 827         finish_wait(&sbi->cp_wait, &wait);
 828 }
 829
 830 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 831 {
 832         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 833         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 834         struct f2fs_nm_info *nm_i = NM_I(sbi);
 835         nid_t last_nid = nm_i->next_scan_nid;
 836         block_t start_blk;
 837         struct page *cp_page;
 838         unsigned int data_sum_blocks, orphan_blocks;
 839         __u32 crc32 = 0;
 840         void *kaddr;
 841         int i;
 842         int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
 843
 844         /*
 845          * This avoids to conduct wrong roll-forward operations and uses
 846          * metapages, so should be called prior to sync_meta_pages below.
 847          */
 848         discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 849
 850         /* Flush all the NAT/SIT pages */
 851         while (get_pages(sbi, F2FS_DIRTY_META)) {
 852                 sync_meta_pages(sbi, META, LONG_MAX);
 853                 if (unlikely(f2fs_cp_error(sbi)))
 854                         return;
 855         }
 856
 857         next_free_nid(sbi, &last_nid);
 858
 859         /*
 860          * modify checkpoint
 861          * version number is already updated
 862          */
 863         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 864         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 865         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 866         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 867                 ckpt->cur_node_segno[i] =
 868                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 869                 ckpt->cur_node_blkoff[i] =
 870                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 871                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 872                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 873         }
 874         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 875                 ckpt->cur_data_segno[i] =
 876                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 877                 ckpt->cur_data_blkoff[i] =
 878                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 879                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 880                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 881         }
 882
 883         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 884         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 885         ckpt->next_free_nid = cpu_to_le32(last_nid);
 886
 887         /* 2 cp  + n data seg summary + orphan inode blocks */
 888         data_sum_blocks = npages_for_summary_flush(sbi);
 889         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 890                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 891         else
 892                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 893
 894         orphan_blocks = GET_ORPHAN_BLOCKS(sbi->n_orphans);
 895         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 896                         orphan_blocks);
 897
 898         if (is_umount) {
 899                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 900                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 901                                 cp_payload_blks + data_sum_blocks +
 902                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 903         } else {
 904                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 905                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
 906                                 cp_payload_blks + data_sum_blocks +
 907                                 orphan_blocks);
 908         }
 909
 910         if (sbi->n_orphans)
 911                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 912         else
 913                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 914
 915         if (sbi->need_fsck)
 916                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
 917
 918         /* update SIT/NAT bitmap */
 919         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 920         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 921
 922         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 923         *((__le32 *)((unsigned char *)ckpt +
 924                                 le32_to_cpu(ckpt->checksum_offset)))
 925                                 = cpu_to_le32(crc32);
 926
 927         start_blk = __start_cp_addr(sbi);
 928
 929         /* write out checkpoint buffer at block 0 */
 930         cp_page = grab_meta_page(sbi, start_blk++);
 931         kaddr = page_address(cp_page);
 932         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 933         set_page_dirty(cp_page);
 934         f2fs_put_page(cp_page, 1);
 935
 936         for (i = 1; i < 1 + cp_payload_blks; i++) {
 937                 cp_page = grab_meta_page(sbi, start_blk++);
 938                 kaddr = page_address(cp_page);
 939                 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
 940                                 (1 << sbi->log_blocksize));
 941                 set_page_dirty(cp_page);
 942                 f2fs_put_page(cp_page, 1);
 943         }
 944
 945         if (sbi->n_orphans) {
 946                 write_orphan_inodes(sbi, start_blk);
 947                 start_blk += orphan_blocks;
 948         }
 949
 950         write_data_summaries(sbi, start_blk);
 951         start_blk += data_sum_blocks;
 952         if (is_umount) {
 953                 write_node_summaries(sbi, start_blk);
 954                 start_blk += NR_CURSEG_NODE_TYPE;
 955         }
 956
 957         /* writeout checkpoint block */
 958         cp_page = grab_meta_page(sbi, start_blk);
 959         kaddr = page_address(cp_page);
 960         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 961         set_page_dirty(cp_page);
 962         f2fs_put_page(cp_page, 1);
 963
 964         /* wait for previous submitted node/meta pages writeback */
 965         wait_on_all_pages_writeback(sbi);
 966
 967         if (unlikely(f2fs_cp_error(sbi)))
 968                 return;
 969
 970         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 971         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 972
 973         /* update user_block_counts */
 974         sbi->last_valid_block_count = sbi->total_valid_block_count;
 975         sbi->alloc_valid_block_count = 0;
 976
 977         /* Here, we only have one bio having CP pack */
 978         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 979
 980         release_dirty_inode(sbi);
 981
 982         if (unlikely(f2fs_cp_error(sbi)))
 983                 return;
 984
 985         clear_prefree_segments(sbi);
 986         F2FS_RESET_SB_DIRT(sbi);
 987 }
 988
 989 /*
 990  * We guarantee that this checkpoint procedure will not fail.
 991  */
 992 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 993 {
 994         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 995         unsigned long long ckpt_ver;
 996
 997         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 998
 999         mutex_lock(&sbi->cp_mutex);
1000
1001         if (!sbi->s_dirty)
1002                 goto out;
1003         if (unlikely(f2fs_cp_error(sbi)))
1004                 goto out;
1005         if (block_operations(sbi))
1006                 goto out;
1007
1008         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
1009
1010         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1011         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1012         f2fs_submit_merged_bio(sbi, META, WRITE);
1013
1014         /*
1015          * update checkpoint pack index
1016          * Increase the version number so that
1017          * SIT entries and seg summaries are written at correct place
1018          */
1019         ckpt_ver = cur_cp_version(ckpt);
1020         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1021
1022         /* write cached NAT/SIT entries to NAT/SIT area */
1023         flush_nat_entries(sbi);
1024         flush_sit_entries(sbi);
1025
1026         /* unlock all the fs_lock[] in do_checkpoint() */
1027         do_checkpoint(sbi, is_umount);
1028
1029         unblock_operations(sbi);
1030         stat_inc_cp_count(sbi->stat_info);
1031 out:
1032         mutex_unlock(&sbi->cp_mutex);
1033         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
1034 }
1035
1036 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1037 {
1038         int i;
1039
1040         for (i = 0; i < MAX_INO_ENTRY; i++) {
1041                 INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
1042                 spin_lock_init(&sbi->ino_lock[i]);
1043                 INIT_LIST_HEAD(&sbi->ino_list[i]);
1044         }
1045
1046         /*
1047          * considering 512 blocks in a segment 8 blocks are needed for cp
1048          * and log segment summaries. Remaining blocks are used to keep
1049          * orphan entries with the limitation one reserved segment
1050          * for cp pack we can have max 1020*504 orphan entries
1051          */
1052         sbi->n_orphans = 0;
1053         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1054                         NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1055 }
1056
1057 int __init create_checkpoint_caches(void)
1058 {
1059         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1060                         sizeof(struct ino_entry));
1061         if (!ino_entry_slab)
1062                 return -ENOMEM;
1063         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1064                         sizeof(struct dir_inode_entry));
1065         if (!inode_entry_slab) {
1066                 kmem_cache_destroy(ino_entry_slab);
1067                 return -ENOMEM;
1068         }
1069         return 0;
1070 }
1071
1072 void destroy_checkpoint_caches(void)
1073 {
1074         kmem_cache_destroy(ino_entry_slab);
1075         kmem_cache_destroy(inode_entry_slab);
1076 }