fs/f2fs/checkpoint.c

   1 /*
   2  * fs/f2fs/checkpoint.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/bio.h>
  13 #include <linux/mpage.h>
  14 #include <linux/writeback.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/f2fs_fs.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/swap.h>
  19
  20 #include "f2fs.h"
  21 #include "node.h"
  22 #include "segment.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *inode_entry_slab;
  28
  29 /*
  30  * We guarantee no failure on the returned page.
  31  */
  32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  33 {
  34         struct address_space *mapping = META_MAPPING(sbi);
  35         struct page *page = NULL;
  36 repeat:
  37         page = grab_cache_page(mapping, index);
  38         if (!page) {
  39                 cond_resched();
  40                 goto repeat;
  41         }
  42         f2fs_wait_on_page_writeback(page, META);
  43         SetPageUptodate(page);
  44         return page;
  45 }
  46
  47 /*
  48  * We guarantee no failure on the returned page.
  49  */
  50 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  51                                                         bool is_meta)
  52 {
  53         struct address_space *mapping = META_MAPPING(sbi);
  54         struct page *page;
  55         struct f2fs_io_info fio = {
  56                 .sbi = sbi,
  57                 .type = META,
  58                 .rw = READ_SYNC | REQ_META | REQ_PRIO,
  59                 .blk_addr = index,
  60                 .encrypted_page = NULL,
  61                 .is_meta = is_meta,
  62         };
  63
  64         if (unlikely(!is_meta))
  65                 fio.rw &= ~REQ_META;
  66 repeat:
  67         page = grab_cache_page(mapping, index);
  68         if (!page) {
  69                 cond_resched();
  70                 goto repeat;
  71         }
  72         if (PageUptodate(page))
  73                 goto out;
  74
  75         fio.page = page;
  76
  77         if (f2fs_submit_page_bio(&fio)) {
  78                 memset(page_address(page), 0, PAGE_SIZE);
  79                 f2fs_stop_checkpoint(sbi);
  80                 f2fs_bug_on(sbi, 1);
  81                 return page;
  82         }
  83
  84         lock_page(page);
  85         if (unlikely(page->mapping != mapping)) {
  86                 f2fs_put_page(page, 1);
  87                 goto repeat;
  88         }
  89
  90         /*
  91          * if there is any IO error when accessing device, make our filesystem
  92          * readonly and make sure do not write checkpoint with non-uptodate
  93          * meta page.
  94          */
  95         if (unlikely(!PageUptodate(page)))
  96                 f2fs_stop_checkpoint(sbi);
  97 out:
  98         return page;
  99 }
 100
 101 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 102 {
 103         return __get_meta_page(sbi, index, true);
 104 }
 105
 106 /* for POR only */
 107 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 108 {
 109         return __get_meta_page(sbi, index, false);
 110 }
 111
 112 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 113                                         block_t blkaddr, int type)
 114 {
 115         switch (type) {
 116         case META_NAT:
 117                 break;
 118         case META_SIT:
 119                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 120                         return false;
 121                 break;
 122         case META_SSA:
 123                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 124                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 125                         return false;
 126                 break;
 127         case META_CP:
 128                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 129                         blkaddr < __start_cp_addr(sbi)))
 130                         return false;
 131                 break;
 132         case META_POR:
 133         case DATA_GENERIC:
 134                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 135                         blkaddr < MAIN_BLKADDR(sbi))) {
 136                         if (type == DATA_GENERIC) {
 137                                 f2fs_msg(sbi->sb, KERN_WARNING,
 138                                         "access invalid blkaddr:%u", blkaddr);
 139                                 WARN_ON(1);
 140                         }
 141                         return false;
 142                 }
 143                 break;
 144         case META_GENERIC:
 145                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 146                         blkaddr >= MAIN_BLKADDR(sbi)))
 147                         return false;
 148                 break;
 149         default:
 150                 BUG();
 151         }
 152
 153         return true;
 154 }
 155
 156 /*
 157  * Readahead CP/NAT/SIT/SSA pages
 158  */
 159 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 160                                                         int type, bool sync)
 161 {
 162         block_t prev_blk_addr = 0;
 163         struct page *page;
 164         block_t blkno = start;
 165         struct f2fs_io_info fio = {
 166                 .sbi = sbi,
 167                 .type = META,
 168                 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
 169                 .encrypted_page = NULL,
 170                 .is_meta = (type != META_POR),
 171         };
 172
 173         if (unlikely(type == META_POR))
 174                 fio.rw &= ~REQ_META;
 175
 176         for (; nrpages-- > 0; blkno++) {
 177
 178                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 179                         goto out;
 180
 181                 switch (type) {
 182                 case META_NAT:
 183                         if (unlikely(blkno >=
 184                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 185                                 blkno = 0;
 186                         /* get nat block addr */
 187                         fio.blk_addr = current_nat_addr(sbi,
 188                                         blkno * NAT_ENTRY_PER_BLOCK);
 189                         break;
 190                 case META_SIT:
 191                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 192                                 goto out;
 193                         /* get sit block addr */
 194                         fio.blk_addr = current_sit_addr(sbi,
 195                                         blkno * SIT_ENTRY_PER_BLOCK);
 196                         if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
 197                                 goto out;
 198                         prev_blk_addr = fio.blk_addr;
 199                         break;
 200                 case META_SSA:
 201                 case META_CP:
 202                 case META_POR:
 203                         fio.blk_addr = blkno;
 204                         break;
 205                 default:
 206                         BUG();
 207                 }
 208
 209                 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
 210                 if (!page)
 211                         continue;
 212                 if (PageUptodate(page)) {
 213                         f2fs_put_page(page, 1);
 214                         continue;
 215                 }
 216
 217                 fio.page = page;
 218                 f2fs_submit_page_mbio(&fio);
 219                 f2fs_put_page(page, 0);
 220         }
 221 out:
 222         f2fs_submit_merged_bio(sbi, META, READ);
 223         return blkno - start;
 224 }
 225
 226 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 227 {
 228         struct page *page;
 229         bool readahead = false;
 230
 231         page = find_get_page(META_MAPPING(sbi), index);
 232         if (!page || (page && !PageUptodate(page)))
 233                 readahead = true;
 234         f2fs_put_page(page, 0);
 235
 236         if (readahead)
 237                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 238 }
 239
 240 static int f2fs_write_meta_page(struct page *page,
 241                                 struct writeback_control *wbc)
 242 {
 243         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 244
 245         trace_f2fs_writepage(page, META);
 246
 247         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 248                 goto redirty_out;
 249         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 250                 goto redirty_out;
 251         if (unlikely(f2fs_cp_error(sbi)))
 252                 goto redirty_out;
 253
 254         f2fs_wait_on_page_writeback(page, META);
 255         write_meta_page(sbi, page);
 256         dec_page_count(sbi, F2FS_DIRTY_META);
 257         unlock_page(page);
 258
 259         if (wbc->for_reclaim)
 260                 f2fs_submit_merged_bio(sbi, META, WRITE);
 261         return 0;
 262
 263 redirty_out:
 264         redirty_page_for_writepage(wbc, page);
 265         return AOP_WRITEPAGE_ACTIVATE;
 266 }
 267
 268 static int f2fs_write_meta_pages(struct address_space *mapping,
 269                                 struct writeback_control *wbc)
 270 {
 271         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 272         long diff, written;
 273
 274         trace_f2fs_writepages(mapping->host, wbc, META);
 275
 276         /* collect a number of dirty meta pages and write together */
 277         if (wbc->for_kupdate ||
 278                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 279                 goto skip_write;
 280
 281         /* if mounting is failed, skip writing node pages */
 282         mutex_lock(&sbi->cp_mutex);
 283         diff = nr_pages_to_write(sbi, META, wbc);
 284         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 285         mutex_unlock(&sbi->cp_mutex);
 286         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 287         return 0;
 288
 289 skip_write:
 290         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 291         return 0;
 292 }
 293
 294 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 295                                                 long nr_to_write)
 296 {
 297         struct address_space *mapping = META_MAPPING(sbi);
 298         pgoff_t index = 0, end = LONG_MAX, prev = LONG_MAX;
 299         struct pagevec pvec;
 300         long nwritten = 0;
 301         struct writeback_control wbc = {
 302                 .for_reclaim = 0,
 303         };
 304
 305         pagevec_init(&pvec, 0);
 306
 307         while (index <= end) {
 308                 int i, nr_pages;
 309                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 310                                 PAGECACHE_TAG_DIRTY,
 311                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 312                 if (unlikely(nr_pages == 0))
 313                         break;
 314
 315                 for (i = 0; i < nr_pages; i++) {
 316                         struct page *page = pvec.pages[i];
 317
 318                         if (prev == LONG_MAX)
 319                                 prev = page->index - 1;
 320                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 321                                 pagevec_release(&pvec);
 322                                 goto stop;
 323                         }
 324
 325                         lock_page(page);
 326
 327                         if (unlikely(page->mapping != mapping)) {
 328 continue_unlock:
 329                                 unlock_page(page);
 330                                 continue;
 331                         }
 332                         if (!PageDirty(page)) {
 333                                 /* someone wrote it for us */
 334                                 goto continue_unlock;
 335                         }
 336
 337                         if (!clear_page_dirty_for_io(page))
 338                                 goto continue_unlock;
 339
 340                         if (mapping->a_ops->writepage(page, &wbc)) {
 341                                 unlock_page(page);
 342                                 break;
 343                         }
 344                         nwritten++;
 345                         prev = page->index;
 346                         if (unlikely(nwritten >= nr_to_write))
 347                                 break;
 348                 }
 349                 pagevec_release(&pvec);
 350                 cond_resched();
 351         }
 352 stop:
 353         if (nwritten)
 354                 f2fs_submit_merged_bio(sbi, type, WRITE);
 355
 356         return nwritten;
 357 }
 358
 359 static int f2fs_set_meta_page_dirty(struct page *page)
 360 {
 361         trace_f2fs_set_page_dirty(page, META);
 362
 363         SetPageUptodate(page);
 364         if (!PageDirty(page)) {
 365                 __set_page_dirty_nobuffers(page);
 366                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 367                 SetPagePrivate(page);
 368                 f2fs_trace_pid(page);
 369                 return 1;
 370         }
 371         return 0;
 372 }
 373
 374 const struct address_space_operations f2fs_meta_aops = {
 375         .writepage      = f2fs_write_meta_page,
 376         .writepages     = f2fs_write_meta_pages,
 377         .set_page_dirty = f2fs_set_meta_page_dirty,
 378         .invalidatepage = f2fs_invalidate_page,
 379         .releasepage    = f2fs_release_page,
 380 };
 381
 382 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 383 {
 384         struct inode_management *im = &sbi->im[type];
 385         struct ino_entry *e, *tmp;
 386
 387         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 388 retry:
 389         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 390
 391         spin_lock(&im->ino_lock);
 392         e = radix_tree_lookup(&im->ino_root, ino);
 393         if (!e) {
 394                 e = tmp;
 395                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 396                         spin_unlock(&im->ino_lock);
 397                         radix_tree_preload_end();
 398                         goto retry;
 399                 }
 400                 memset(e, 0, sizeof(struct ino_entry));
 401                 e->ino = ino;
 402
 403                 list_add_tail(&e->list, &im->ino_list);
 404                 if (type != ORPHAN_INO)
 405                         im->ino_num++;
 406         }
 407         spin_unlock(&im->ino_lock);
 408         radix_tree_preload_end();
 409
 410         if (e != tmp)
 411                 kmem_cache_free(ino_entry_slab, tmp);
 412 }
 413
 414 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 415 {
 416         struct inode_management *im = &sbi->im[type];
 417         struct ino_entry *e;
 418
 419         spin_lock(&im->ino_lock);
 420         e = radix_tree_lookup(&im->ino_root, ino);
 421         if (e) {
 422                 list_del(&e->list);
 423                 radix_tree_delete(&im->ino_root, ino);
 424                 im->ino_num--;
 425                 spin_unlock(&im->ino_lock);
 426                 kmem_cache_free(ino_entry_slab, e);
 427                 return;
 428         }
 429         spin_unlock(&im->ino_lock);
 430 }
 431
 432 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 433 {
 434         /* add new dirty ino entry into list */
 435         __add_ino_entry(sbi, ino, type);
 436 }
 437
 438 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
 439 {
 440         /* remove dirty ino entry from list */
 441         __remove_ino_entry(sbi, ino, type);
 442 }
 443
 444 /* mode should be APPEND_INO or UPDATE_INO */
 445 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 446 {
 447         struct inode_management *im = &sbi->im[mode];
 448         struct ino_entry *e;
 449
 450         spin_lock(&im->ino_lock);
 451         e = radix_tree_lookup(&im->ino_root, ino);
 452         spin_unlock(&im->ino_lock);
 453         return e ? true : false;
 454 }
 455
 456 void release_dirty_inode(struct f2fs_sb_info *sbi)
 457 {
 458         struct ino_entry *e, *tmp;
 459         int i;
 460
 461         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 462                 struct inode_management *im = &sbi->im[i];
 463
 464                 spin_lock(&im->ino_lock);
 465                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 466                         list_del(&e->list);
 467                         radix_tree_delete(&im->ino_root, e->ino);
 468                         kmem_cache_free(ino_entry_slab, e);
 469                         im->ino_num--;
 470                 }
 471                 spin_unlock(&im->ino_lock);
 472         }
 473 }
 474
 475 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 476 {
 477         struct inode_management *im = &sbi->im[ORPHAN_INO];
 478         int err = 0;
 479
 480         spin_lock(&im->ino_lock);
 481         if (unlikely(im->ino_num >= sbi->max_orphans))
 482                 err = -ENOSPC;
 483         else
 484                 im->ino_num++;
 485         spin_unlock(&im->ino_lock);
 486
 487         return err;
 488 }
 489
 490 void release_orphan_inode(struct f2fs_sb_info *sbi)
 491 {
 492         struct inode_management *im = &sbi->im[ORPHAN_INO];
 493
 494         spin_lock(&im->ino_lock);
 495         f2fs_bug_on(sbi, im->ino_num == 0);
 496         im->ino_num--;
 497         spin_unlock(&im->ino_lock);
 498 }
 499
 500 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 501 {
 502         /* add new orphan ino entry into list */
 503         __add_ino_entry(sbi, ino, ORPHAN_INO);
 504 }
 505
 506 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 507 {
 508         /* remove orphan entry from orphan list */
 509         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 510 }
 511
 512 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 513 {
 514         struct inode *inode;
 515
 516         inode = f2fs_iget(sbi->sb, ino);
 517         if (IS_ERR(inode)) {
 518                 /*
 519                  * there should be a bug that we can't find the entry
 520                  * to orphan inode.
 521                  */
 522                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 523                 return PTR_ERR(inode);
 524         }
 525
 526         clear_nlink(inode);
 527
 528         /* truncate all the data during iput */
 529         iput(inode);
 530         return 0;
 531 }
 532
 533 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 534 {
 535         block_t start_blk, orphan_blocks, i, j;
 536         int err;
 537
 538         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 539                 return 0;
 540
 541         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 542         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 543
 544         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 545
 546         for (i = 0; i < orphan_blocks; i++) {
 547                 struct page *page = get_meta_page(sbi, start_blk + i);
 548                 struct f2fs_orphan_block *orphan_blk;
 549
 550                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 551                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 552                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 553                         err = recover_orphan_inode(sbi, ino);
 554                         if (err) {
 555                                 f2fs_put_page(page, 1);
 556                                 return err;
 557                         }
 558                 }
 559                 f2fs_put_page(page, 1);
 560         }
 561         /* clear Orphan Flag */
 562         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 563         return 0;
 564 }
 565
 566 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 567 {
 568         struct list_head *head;
 569         struct f2fs_orphan_block *orphan_blk = NULL;
 570         unsigned int nentries = 0;
 571         unsigned short index = 1;
 572         unsigned short orphan_blocks;
 573         struct page *page = NULL;
 574         struct ino_entry *orphan = NULL;
 575         struct inode_management *im = &sbi->im[ORPHAN_INO];
 576
 577         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 578
 579         /*
 580          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 581          * orphan inode operations are covered under f2fs_lock_op().
 582          * And, spin_lock should be avoided due to page operations below.
 583          */
 584         head = &im->ino_list;
 585
 586         /* loop for each orphan inode entry and write them in Jornal block */
 587         list_for_each_entry(orphan, head, list) {
 588                 if (!page) {
 589                         page = grab_meta_page(sbi, start_blk++);
 590                         orphan_blk =
 591                                 (struct f2fs_orphan_block *)page_address(page);
 592                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 593                 }
 594
 595                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 596
 597                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 598                         /*
 599                          * an orphan block is full of 1020 entries,
 600                          * then we need to flush current orphan blocks
 601                          * and bring another one in memory
 602                          */
 603                         orphan_blk->blk_addr = cpu_to_le16(index);
 604                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 605                         orphan_blk->entry_count = cpu_to_le32(nentries);
 606                         set_page_dirty(page);
 607                         f2fs_put_page(page, 1);
 608                         index++;
 609                         nentries = 0;
 610                         page = NULL;
 611                 }
 612         }
 613
 614         if (page) {
 615                 orphan_blk->blk_addr = cpu_to_le16(index);
 616                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 617                 orphan_blk->entry_count = cpu_to_le32(nentries);
 618                 set_page_dirty(page);
 619                 f2fs_put_page(page, 1);
 620         }
 621 }
 622
 623 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 624                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 625                 unsigned long long *version)
 626 {
 627         unsigned long blk_size = sbi->blocksize;
 628         size_t crc_offset = 0;
 629         __u32 crc = 0;
 630
 631         *cp_page = get_meta_page(sbi, cp_addr);
 632         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 633
 634         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 635         if (crc_offset >= blk_size) {
 636                 f2fs_put_page(*cp_page, 1);
 637                 f2fs_msg(sbi->sb, KERN_WARNING,
 638                         "invalid crc_offset: %zu", crc_offset);
 639                 return -EINVAL;
 640         }
 641
 642         crc = le32_to_cpu(*((__le32 *)((unsigned char *)*cp_block
 643                                                         + crc_offset)));
 644         if (!f2fs_crc_valid(crc, *cp_block, crc_offset)) {
 645                 f2fs_put_page(*cp_page, 1);
 646                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 647                 return -EINVAL;
 648         }
 649
 650         *version = cur_cp_version(*cp_block);
 651         return 0;
 652 }
 653
 654 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 655                                 block_t cp_addr, unsigned long long *version)
 656 {
 657         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 658         struct f2fs_checkpoint *cp_block = NULL;
 659         unsigned long long cur_version = 0, pre_version = 0;
 660         int err;
 661
 662         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 663                                         &cp_page_1, version);
 664         if (err)
 665                 return NULL;
 666
 667         if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
 668                                         sbi->blocks_per_seg) {
 669                 f2fs_msg(sbi->sb, KERN_WARNING,
 670                         "invalid cp_pack_total_block_count:%u",
 671                         le32_to_cpu(cp_block->cp_pack_total_block_count));
 672                 goto invalid_cp;
 673         }
 674         pre_version = *version;
 675
 676         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 677         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 678                                         &cp_page_2, version);
 679         if (err)
 680                 goto invalid_cp;
 681         cur_version = *version;
 682
 683         if (cur_version == pre_version) {
 684                 *version = cur_version;
 685                 f2fs_put_page(cp_page_2, 1);
 686                 return cp_page_1;
 687         }
 688         f2fs_put_page(cp_page_2, 1);
 689 invalid_cp:
 690         f2fs_put_page(cp_page_1, 1);
 691         return NULL;
 692 }
 693
 694 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 695 {
 696         struct f2fs_checkpoint *cp_block;
 697         struct f2fs_super_block *fsb = sbi->raw_super;
 698         struct page *cp1, *cp2, *cur_page;
 699         unsigned long blk_size = sbi->blocksize;
 700         unsigned long long cp1_version = 0, cp2_version = 0;
 701         unsigned long long cp_start_blk_no;
 702         unsigned int cp_blks = 1 + __cp_payload(sbi);
 703         block_t cp_blk_no;
 704         int i;
 705
 706         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 707         if (!sbi->ckpt)
 708                 return -ENOMEM;
 709         /*
 710          * Finding out valid cp block involves read both
 711          * sets( cp pack1 and cp pack 2)
 712          */
 713         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 714         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 715
 716         /* The second checkpoint pack should start at the next segment */
 717         cp_start_blk_no += ((unsigned long long)1) <<
 718                                 le32_to_cpu(fsb->log_blocks_per_seg);
 719         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 720
 721         if (cp1 && cp2) {
 722                 if (ver_after(cp2_version, cp1_version))
 723                         cur_page = cp2;
 724                 else
 725                         cur_page = cp1;
 726         } else if (cp1) {
 727                 cur_page = cp1;
 728         } else if (cp2) {
 729                 cur_page = cp2;
 730         } else {
 731                 goto fail_no_cp;
 732         }
 733
 734         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 735         memcpy(sbi->ckpt, cp_block, blk_size);
 736
 737         if (cur_page == cp1)
 738                 sbi->cur_cp_pack = 1;
 739         else
 740                 sbi->cur_cp_pack = 2;
 741
 742         /* Sanity checking of checkpoint */
 743         if (sanity_check_ckpt(sbi))
 744                 goto free_fail_no_cp;
 745
 746         if (cp_blks <= 1)
 747                 goto done;
 748
 749         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 750         if (cur_page == cp2)
 751                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 752
 753         for (i = 1; i < cp_blks; i++) {
 754                 void *sit_bitmap_ptr;
 755                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 756
 757                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 758                 sit_bitmap_ptr = page_address(cur_page);
 759                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 760                 f2fs_put_page(cur_page, 1);
 761         }
 762 done:
 763         f2fs_put_page(cp1, 1);
 764         f2fs_put_page(cp2, 1);
 765         return 0;
 766
 767 free_fail_no_cp:
 768         f2fs_put_page(cp1, 1);
 769         f2fs_put_page(cp2, 1);
 770 fail_no_cp:
 771         kfree(sbi->ckpt);
 772         return -EINVAL;
 773 }
 774
 775 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
 776 {
 777         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 778
 779         if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
 780                 return -EEXIST;
 781
 782         set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 783         F2FS_I(inode)->dirty_dir = new;
 784         list_add_tail(&new->list, &sbi->dir_inode_list);
 785         stat_inc_dirty_dir(sbi);
 786         return 0;
 787 }
 788
 789 void update_dirty_page(struct inode *inode, struct page *page)
 790 {
 791         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 792         struct inode_entry *new;
 793         int ret = 0;
 794
 795         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 796                         !S_ISLNK(inode->i_mode))
 797                 return;
 798
 799         if (!S_ISDIR(inode->i_mode)) {
 800                 inode_inc_dirty_pages(inode);
 801                 goto out;
 802         }
 803
 804         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 805         new->inode = inode;
 806         INIT_LIST_HEAD(&new->list);
 807
 808         spin_lock(&sbi->dir_inode_lock);
 809         ret = __add_dirty_inode(inode, new);
 810         inode_inc_dirty_pages(inode);
 811         spin_unlock(&sbi->dir_inode_lock);
 812
 813         if (ret)
 814                 kmem_cache_free(inode_entry_slab, new);
 815 out:
 816         SetPagePrivate(page);
 817         f2fs_trace_pid(page);
 818 }
 819
 820 void remove_dirty_dir_inode(struct inode *inode)
 821 {
 822         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 823         struct inode_entry *entry;
 824
 825         if (!S_ISDIR(inode->i_mode))
 826                 return;
 827
 828         spin_lock(&sbi->dir_inode_lock);
 829         if (get_dirty_pages(inode) ||
 830                         !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
 831                 spin_unlock(&sbi->dir_inode_lock);
 832                 return;
 833         }
 834
 835         entry = F2FS_I(inode)->dirty_dir;
 836         list_del(&entry->list);
 837         F2FS_I(inode)->dirty_dir = NULL;
 838         clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
 839         stat_dec_dirty_dir(sbi);
 840         spin_unlock(&sbi->dir_inode_lock);
 841         kmem_cache_free(inode_entry_slab, entry);
 842 }
 843
 844 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 845 {
 846         struct list_head *head;
 847         struct inode_entry *entry;
 848         struct inode *inode;
 849 retry:
 850         if (unlikely(f2fs_cp_error(sbi)))
 851                 return;
 852
 853         spin_lock(&sbi->dir_inode_lock);
 854
 855         head = &sbi->dir_inode_list;
 856         if (list_empty(head)) {
 857                 spin_unlock(&sbi->dir_inode_lock);
 858                 return;
 859         }
 860         entry = list_entry(head->next, struct inode_entry, list);
 861         inode = igrab(entry->inode);
 862         spin_unlock(&sbi->dir_inode_lock);
 863         if (inode) {
 864                 filemap_fdatawrite(inode->i_mapping);
 865                 iput(inode);
 866         } else {
 867                 /*
 868                  * We should submit bio, since it exists several
 869                  * wribacking dentry pages in the freeing inode.
 870                  */
 871                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 872                 cond_resched();
 873         }
 874         goto retry;
 875 }
 876
 877 /*
 878  * Freeze all the FS-operations for checkpoint.
 879  */
 880 static int block_operations(struct f2fs_sb_info *sbi)
 881 {
 882         struct writeback_control wbc = {
 883                 .sync_mode = WB_SYNC_ALL,
 884                 .nr_to_write = LONG_MAX,
 885                 .for_reclaim = 0,
 886         };
 887         struct blk_plug plug;
 888         int err = 0;
 889
 890         blk_start_plug(&plug);
 891
 892 retry_flush_dents:
 893         f2fs_lock_all(sbi);
 894         /* write all the dirty dentry pages */
 895         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 896                 f2fs_unlock_all(sbi);
 897                 sync_dirty_dir_inodes(sbi);
 898                 if (unlikely(f2fs_cp_error(sbi))) {
 899                         err = -EIO;
 900                         goto out;
 901                 }
 902                 goto retry_flush_dents;
 903         }
 904
 905         /*
 906          * POR: we should ensure that there are no dirty node pages
 907          * until finishing nat/sit flush.
 908          */
 909 retry_flush_nodes:
 910         down_write(&sbi->node_write);
 911
 912         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 913                 up_write(&sbi->node_write);
 914                 sync_node_pages(sbi, 0, &wbc);
 915                 if (unlikely(f2fs_cp_error(sbi))) {
 916                         f2fs_unlock_all(sbi);
 917                         err = -EIO;
 918                         goto out;
 919                 }
 920                 goto retry_flush_nodes;
 921         }
 922 out:
 923         blk_finish_plug(&plug);
 924         return err;
 925 }
 926
 927 static void unblock_operations(struct f2fs_sb_info *sbi)
 928 {
 929         up_write(&sbi->node_write);
 930         f2fs_unlock_all(sbi);
 931 }
 932
 933 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 934 {
 935         DEFINE_WAIT(wait);
 936
 937         for (;;) {
 938                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 939
 940                 if (!get_pages(sbi, F2FS_WRITEBACK))
 941                         break;
 942
 943                 io_schedule();
 944         }
 945         finish_wait(&sbi->cp_wait, &wait);
 946 }
 947
 948 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 949 {
 950         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 951         struct f2fs_nm_info *nm_i = NM_I(sbi);
 952         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 953         nid_t last_nid = nm_i->next_scan_nid;
 954         block_t start_blk;
 955         unsigned int data_sum_blocks, orphan_blocks;
 956         __u32 crc32 = 0;
 957         int i;
 958         int cp_payload_blks = __cp_payload(sbi);
 959
 960         /* Flush all the NAT/SIT pages */
 961         while (get_pages(sbi, F2FS_DIRTY_META)) {
 962                 sync_meta_pages(sbi, META, LONG_MAX);
 963                 if (unlikely(f2fs_cp_error(sbi)))
 964                         return;
 965         }
 966
 967         next_free_nid(sbi, &last_nid);
 968
 969         /*
 970          * modify checkpoint
 971          * version number is already updated
 972          */
 973         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 974         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 975         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 976         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 977                 ckpt->cur_node_segno[i] =
 978                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 979                 ckpt->cur_node_blkoff[i] =
 980                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 981                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 982                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 983         }
 984         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 985                 ckpt->cur_data_segno[i] =
 986                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 987                 ckpt->cur_data_blkoff[i] =
 988                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 989                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 990                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 991         }
 992
 993         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 994         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 995         ckpt->next_free_nid = cpu_to_le32(last_nid);
 996
 997         /* 2 cp  + n data seg summary + orphan inode blocks */
 998         data_sum_blocks = npages_for_summary_flush(sbi, false);
 999         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1000                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1001         else
1002                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1003
1004         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1005         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1006                         orphan_blocks);
1007
1008         if (__remain_node_summaries(cpc->reason))
1009                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1010                                 cp_payload_blks + data_sum_blocks +
1011                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1012         else
1013                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1014                                 cp_payload_blks + data_sum_blocks +
1015                                 orphan_blocks);
1016
1017         if (cpc->reason == CP_UMOUNT)
1018                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1019         else
1020                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1021
1022         if (cpc->reason == CP_FASTBOOT)
1023                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1024         else
1025                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1026
1027         if (orphan_num)
1028                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1029         else
1030                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1031
1032         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1033                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1034
1035         /* set this flag to activate crc|cp_ver for recovery */
1036         set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1037
1038         /* update SIT/NAT bitmap */
1039         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1040         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1041
1042         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
1043         *((__le32 *)((unsigned char *)ckpt +
1044                                 le32_to_cpu(ckpt->checksum_offset)))
1045                                 = cpu_to_le32(crc32);
1046
1047         start_blk = __start_cp_next_addr(sbi);
1048
1049         /* need to wait for end_io results */
1050         wait_on_all_pages_writeback(sbi);
1051         if (unlikely(f2fs_cp_error(sbi)))
1052                 return;
1053
1054         /* write out checkpoint buffer at block 0 */
1055         update_meta_page(sbi, ckpt, start_blk++);
1056
1057         for (i = 1; i < 1 + cp_payload_blks; i++)
1058                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1059                                                         start_blk++);
1060
1061         if (orphan_num) {
1062                 write_orphan_inodes(sbi, start_blk);
1063                 start_blk += orphan_blocks;
1064         }
1065
1066         write_data_summaries(sbi, start_blk);
1067         start_blk += data_sum_blocks;
1068         if (__remain_node_summaries(cpc->reason)) {
1069                 write_node_summaries(sbi, start_blk);
1070                 start_blk += NR_CURSEG_NODE_TYPE;
1071         }
1072
1073         /* writeout checkpoint block */
1074         update_meta_page(sbi, ckpt, start_blk);
1075
1076         /* wait for previous submitted node/meta pages writeback */
1077         wait_on_all_pages_writeback(sbi);
1078
1079         if (unlikely(f2fs_cp_error(sbi)))
1080                 return;
1081
1082         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1083         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1084
1085         /* update user_block_counts */
1086         sbi->last_valid_block_count = sbi->total_valid_block_count;
1087         sbi->alloc_valid_block_count = 0;
1088
1089         /* Here, we only have one bio having CP pack */
1090         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1091
1092         /* wait for previous submitted meta pages writeback */
1093         wait_on_all_pages_writeback(sbi);
1094
1095         release_dirty_inode(sbi);
1096
1097         if (unlikely(f2fs_cp_error(sbi)))
1098                 return;
1099
1100         clear_prefree_segments(sbi, cpc);
1101         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1102         __set_cp_next_pack(sbi);
1103 }
1104
1105 /*
1106  * We guarantee that this checkpoint procedure will not fail.
1107  */
1108 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1109 {
1110         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1111         unsigned long long ckpt_ver;
1112
1113         mutex_lock(&sbi->cp_mutex);
1114
1115         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1116                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1117                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1118                 goto out;
1119         if (unlikely(f2fs_cp_error(sbi)))
1120                 goto out;
1121         if (f2fs_readonly(sbi->sb))
1122                 goto out;
1123
1124         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1125
1126         if (block_operations(sbi))
1127                 goto out;
1128
1129         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1130
1131         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1132         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1133         f2fs_submit_merged_bio(sbi, META, WRITE);
1134
1135         /*
1136          * update checkpoint pack index
1137          * Increase the version number so that
1138          * SIT entries and seg summaries are written at correct place
1139          */
1140         ckpt_ver = cur_cp_version(ckpt);
1141         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1142
1143         /* write cached NAT/SIT entries to NAT/SIT area */
1144         flush_nat_entries(sbi);
1145         flush_sit_entries(sbi, cpc);
1146
1147         /* unlock all the fs_lock[] in do_checkpoint() */
1148         do_checkpoint(sbi, cpc);
1149
1150         unblock_operations(sbi);
1151         stat_inc_cp_count(sbi->stat_info);
1152
1153         if (cpc->reason == CP_RECOVERY)
1154                 f2fs_msg(sbi->sb, KERN_NOTICE,
1155                         "checkpoint: version = %llx", ckpt_ver);
1156
1157         /* do checkpoint periodically */
1158         sbi->cp_expires = round_jiffies_up(jiffies + HZ * sbi->cp_interval);
1159 out:
1160         mutex_unlock(&sbi->cp_mutex);
1161         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1162 }
1163
1164 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1165 {
1166         int i;
1167
1168         for (i = 0; i < MAX_INO_ENTRY; i++) {
1169                 struct inode_management *im = &sbi->im[i];
1170
1171                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1172                 spin_lock_init(&im->ino_lock);
1173                 INIT_LIST_HEAD(&im->ino_list);
1174                 im->ino_num = 0;
1175         }
1176
1177         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1178                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1179                                 F2FS_ORPHANS_PER_BLOCK;
1180 }
1181
1182 int __init create_checkpoint_caches(void)
1183 {
1184         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1185                         sizeof(struct ino_entry));
1186         if (!ino_entry_slab)
1187                 return -ENOMEM;
1188         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1189                         sizeof(struct inode_entry));
1190         if (!inode_entry_slab) {
1191                 kmem_cache_destroy(ino_entry_slab);
1192                 return -ENOMEM;
1193         }
1194         return 0;
1195 }
1196
1197 void destroy_checkpoint_caches(void)
1198 {
1199         kmem_cache_destroy(ino_entry_slab);
1200         kmem_cache_destroy(inode_entry_slab);
1201 }