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, true);
  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                 .old_blkaddr = index,
  60                 .new_blkaddr = index,
  61                 .encrypted_page = NULL,
  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                 f2fs_put_page(page, 1);
  79                 goto repeat;
  80         }
  81
  82         lock_page(page);
  83         if (unlikely(page->mapping != mapping)) {
  84                 f2fs_put_page(page, 1);
  85                 goto repeat;
  86         }
  87
  88         /*
  89          * if there is any IO error when accessing device, make our filesystem
  90          * readonly and make sure do not write checkpoint with non-uptodate
  91          * meta page.
  92          */
  93         if (unlikely(!PageUptodate(page)))
  94                 f2fs_stop_checkpoint(sbi);
  95 out:
  96         return page;
  97 }
  98
  99 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 100 {
 101         return __get_meta_page(sbi, index, true);
 102 }
 103
 104 /* for POR only */
 105 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 106 {
 107         return __get_meta_page(sbi, index, false);
 108 }
 109
 110 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 111 {
 112         switch (type) {
 113         case META_NAT:
 114                 break;
 115         case META_SIT:
 116                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 117                         return false;
 118                 break;
 119         case META_SSA:
 120                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 121                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 122                         return false;
 123                 break;
 124         case META_CP:
 125                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 126                         blkaddr < __start_cp_addr(sbi)))
 127                         return false;
 128                 break;
 129         case META_POR:
 130                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 131                         blkaddr < MAIN_BLKADDR(sbi)))
 132                         return false;
 133                 break;
 134         default:
 135                 BUG();
 136         }
 137
 138         return true;
 139 }
 140
 141 /*
 142  * Readahead CP/NAT/SIT/SSA pages
 143  */
 144 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 145                                                         int type, bool sync)
 146 {
 147         struct page *page;
 148         block_t blkno = start;
 149         struct f2fs_io_info fio = {
 150                 .sbi = sbi,
 151                 .type = META,
 152                 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
 153                 .encrypted_page = NULL,
 154         };
 155         struct blk_plug plug;
 156
 157         if (unlikely(type == META_POR))
 158                 fio.rw &= ~REQ_META;
 159
 160         blk_start_plug(&plug);
 161         for (; nrpages-- > 0; blkno++) {
 162
 163                 if (!is_valid_blkaddr(sbi, blkno, type))
 164                         goto out;
 165
 166                 switch (type) {
 167                 case META_NAT:
 168                         if (unlikely(blkno >=
 169                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 170                                 blkno = 0;
 171                         /* get nat block addr */
 172                         fio.new_blkaddr = current_nat_addr(sbi,
 173                                         blkno * NAT_ENTRY_PER_BLOCK);
 174                         break;
 175                 case META_SIT:
 176                         /* get sit block addr */
 177                         fio.new_blkaddr = current_sit_addr(sbi,
 178                                         blkno * SIT_ENTRY_PER_BLOCK);
 179                         break;
 180                 case META_SSA:
 181                 case META_CP:
 182                 case META_POR:
 183                         fio.new_blkaddr = blkno;
 184                         break;
 185                 default:
 186                         BUG();
 187                 }
 188
 189                 page = grab_cache_page(META_MAPPING(sbi), fio.new_blkaddr);
 190                 if (!page)
 191                         continue;
 192                 if (PageUptodate(page)) {
 193                         f2fs_put_page(page, 1);
 194                         continue;
 195                 }
 196
 197                 fio.page = page;
 198                 fio.old_blkaddr = fio.new_blkaddr;
 199                 f2fs_submit_page_mbio(&fio);
 200                 f2fs_put_page(page, 0);
 201         }
 202 out:
 203         f2fs_submit_merged_bio(sbi, META, READ);
 204         blk_finish_plug(&plug);
 205         return blkno - start;
 206 }
 207
 208 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 209 {
 210         struct page *page;
 211         bool readahead = false;
 212
 213         page = find_get_page(META_MAPPING(sbi), index);
 214         if (!page || (page && !PageUptodate(page)))
 215                 readahead = true;
 216         f2fs_put_page(page, 0);
 217
 218         if (readahead)
 219                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 220 }
 221
 222 static int f2fs_write_meta_page(struct page *page,
 223                                 struct writeback_control *wbc)
 224 {
 225         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 226
 227         trace_f2fs_writepage(page, META);
 228
 229         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 230                 goto redirty_out;
 231         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 232                 goto redirty_out;
 233         if (unlikely(f2fs_cp_error(sbi)))
 234                 goto redirty_out;
 235
 236         write_meta_page(sbi, page);
 237         dec_page_count(sbi, F2FS_DIRTY_META);
 238
 239         if (wbc->for_reclaim)
 240                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
 241
 242         unlock_page(page);
 243
 244         if (unlikely(f2fs_cp_error(sbi)))
 245                 f2fs_submit_merged_bio(sbi, META, WRITE);
 246
 247         return 0;
 248
 249 redirty_out:
 250         redirty_page_for_writepage(wbc, page);
 251         return AOP_WRITEPAGE_ACTIVATE;
 252 }
 253
 254 static int f2fs_write_meta_pages(struct address_space *mapping,
 255                                 struct writeback_control *wbc)
 256 {
 257         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 258         long diff, written;
 259
 260         /* collect a number of dirty meta pages and write together */
 261         if (wbc->for_kupdate ||
 262                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 263                 goto skip_write;
 264
 265         trace_f2fs_writepages(mapping->host, wbc, META);
 266
 267         /* if mounting is failed, skip writing node pages */
 268         mutex_lock(&sbi->cp_mutex);
 269         diff = nr_pages_to_write(sbi, META, wbc);
 270         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 271         mutex_unlock(&sbi->cp_mutex);
 272         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 273         return 0;
 274
 275 skip_write:
 276         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 277         trace_f2fs_writepages(mapping->host, wbc, META);
 278         return 0;
 279 }
 280
 281 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 282                                                 long nr_to_write)
 283 {
 284         struct address_space *mapping = META_MAPPING(sbi);
 285         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 286         struct pagevec pvec;
 287         long nwritten = 0;
 288         struct writeback_control wbc = {
 289                 .for_reclaim = 0,
 290         };
 291         struct blk_plug plug;
 292
 293         pagevec_init(&pvec, 0);
 294
 295         blk_start_plug(&plug);
 296
 297         while (index <= end) {
 298                 int i, nr_pages;
 299                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 300                                 PAGECACHE_TAG_DIRTY,
 301                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 302                 if (unlikely(nr_pages == 0))
 303                         break;
 304
 305                 for (i = 0; i < nr_pages; i++) {
 306                         struct page *page = pvec.pages[i];
 307
 308                         if (prev == ULONG_MAX)
 309                                 prev = page->index - 1;
 310                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 311                                 pagevec_release(&pvec);
 312                                 goto stop;
 313                         }
 314
 315                         lock_page(page);
 316
 317                         if (unlikely(page->mapping != mapping)) {
 318 continue_unlock:
 319                                 unlock_page(page);
 320                                 continue;
 321                         }
 322                         if (!PageDirty(page)) {
 323                                 /* someone wrote it for us */
 324                                 goto continue_unlock;
 325                         }
 326
 327                         f2fs_wait_on_page_writeback(page, META, true);
 328
 329                         BUG_ON(PageWriteback(page));
 330                         if (!clear_page_dirty_for_io(page))
 331                                 goto continue_unlock;
 332
 333                         if (mapping->a_ops->writepage(page, &wbc)) {
 334                                 unlock_page(page);
 335                                 break;
 336                         }
 337                         nwritten++;
 338                         prev = page->index;
 339                         if (unlikely(nwritten >= nr_to_write))
 340                                 break;
 341                 }
 342                 pagevec_release(&pvec);
 343                 cond_resched();
 344         }
 345 stop:
 346         if (nwritten)
 347                 f2fs_submit_merged_bio(sbi, type, WRITE);
 348
 349         blk_finish_plug(&plug);
 350
 351         return nwritten;
 352 }
 353
 354 static int f2fs_set_meta_page_dirty(struct page *page)
 355 {
 356         trace_f2fs_set_page_dirty(page, META);
 357
 358         SetPageUptodate(page);
 359         if (!PageDirty(page)) {
 360                 __set_page_dirty_nobuffers(page);
 361                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 362                 SetPagePrivate(page);
 363                 f2fs_trace_pid(page);
 364                 return 1;
 365         }
 366         return 0;
 367 }
 368
 369 const struct address_space_operations f2fs_meta_aops = {
 370         .writepage      = f2fs_write_meta_page,
 371         .writepages     = f2fs_write_meta_pages,
 372         .set_page_dirty = f2fs_set_meta_page_dirty,
 373         .invalidatepage = f2fs_invalidate_page,
 374         .releasepage    = f2fs_release_page,
 375 };
 376
 377 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 378 {
 379         struct inode_management *im = &sbi->im[type];
 380         struct ino_entry *e, *tmp;
 381
 382         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 383 retry:
 384         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 385
 386         spin_lock(&im->ino_lock);
 387         e = radix_tree_lookup(&im->ino_root, ino);
 388         if (!e) {
 389                 e = tmp;
 390                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 391                         spin_unlock(&im->ino_lock);
 392                         radix_tree_preload_end();
 393                         goto retry;
 394                 }
 395                 memset(e, 0, sizeof(struct ino_entry));
 396                 e->ino = ino;
 397
 398                 list_add_tail(&e->list, &im->ino_list);
 399                 if (type != ORPHAN_INO)
 400                         im->ino_num++;
 401         }
 402         spin_unlock(&im->ino_lock);
 403         radix_tree_preload_end();
 404
 405         if (e != tmp)
 406                 kmem_cache_free(ino_entry_slab, tmp);
 407 }
 408
 409 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 410 {
 411         struct inode_management *im = &sbi->im[type];
 412         struct ino_entry *e;
 413
 414         spin_lock(&im->ino_lock);
 415         e = radix_tree_lookup(&im->ino_root, ino);
 416         if (e) {
 417                 list_del(&e->list);
 418                 radix_tree_delete(&im->ino_root, ino);
 419                 im->ino_num--;
 420                 spin_unlock(&im->ino_lock);
 421                 kmem_cache_free(ino_entry_slab, e);
 422                 return;
 423         }
 424         spin_unlock(&im->ino_lock);
 425 }
 426
 427 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 428 {
 429         /* add new dirty ino entry into list */
 430         __add_ino_entry(sbi, ino, type);
 431 }
 432
 433 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 434 {
 435         /* remove dirty ino entry from list */
 436         __remove_ino_entry(sbi, ino, type);
 437 }
 438
 439 /* mode should be APPEND_INO or UPDATE_INO */
 440 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 441 {
 442         struct inode_management *im = &sbi->im[mode];
 443         struct ino_entry *e;
 444
 445         spin_lock(&im->ino_lock);
 446         e = radix_tree_lookup(&im->ino_root, ino);
 447         spin_unlock(&im->ino_lock);
 448         return e ? true : false;
 449 }
 450
 451 void release_ino_entry(struct f2fs_sb_info *sbi)
 452 {
 453         struct ino_entry *e, *tmp;
 454         int i;
 455
 456         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 457                 struct inode_management *im = &sbi->im[i];
 458
 459                 spin_lock(&im->ino_lock);
 460                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 461                         list_del(&e->list);
 462                         radix_tree_delete(&im->ino_root, e->ino);
 463                         kmem_cache_free(ino_entry_slab, e);
 464                         im->ino_num--;
 465                 }
 466                 spin_unlock(&im->ino_lock);
 467         }
 468 }
 469
 470 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 471 {
 472         struct inode_management *im = &sbi->im[ORPHAN_INO];
 473         int err = 0;
 474
 475         spin_lock(&im->ino_lock);
 476         if (unlikely(im->ino_num >= sbi->max_orphans))
 477                 err = -ENOSPC;
 478         else
 479                 im->ino_num++;
 480         spin_unlock(&im->ino_lock);
 481
 482         return err;
 483 }
 484
 485 void release_orphan_inode(struct f2fs_sb_info *sbi)
 486 {
 487         struct inode_management *im = &sbi->im[ORPHAN_INO];
 488
 489         spin_lock(&im->ino_lock);
 490         f2fs_bug_on(sbi, im->ino_num == 0);
 491         im->ino_num--;
 492         spin_unlock(&im->ino_lock);
 493 }
 494
 495 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 496 {
 497         /* add new orphan ino entry into list */
 498         __add_ino_entry(sbi, ino, ORPHAN_INO);
 499 }
 500
 501 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 502 {
 503         /* remove orphan entry from orphan list */
 504         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 505 }
 506
 507 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 508 {
 509         struct inode *inode;
 510
 511         inode = f2fs_iget(sbi->sb, ino);
 512         if (IS_ERR(inode)) {
 513                 /*
 514                  * there should be a bug that we can't find the entry
 515                  * to orphan inode.
 516                  */
 517                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 518                 return PTR_ERR(inode);
 519         }
 520
 521         clear_nlink(inode);
 522
 523         /* truncate all the data during iput */
 524         iput(inode);
 525         return 0;
 526 }
 527
 528 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 529 {
 530         block_t start_blk, orphan_blocks, i, j;
 531         int err;
 532
 533         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 534                 return 0;
 535
 536         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 537         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 538
 539         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 540
 541         for (i = 0; i < orphan_blocks; i++) {
 542                 struct page *page = get_meta_page(sbi, start_blk + i);
 543                 struct f2fs_orphan_block *orphan_blk;
 544
 545                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 546                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 547                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 548                         err = recover_orphan_inode(sbi, ino);
 549                         if (err) {
 550                                 f2fs_put_page(page, 1);
 551                                 return err;
 552                         }
 553                 }
 554                 f2fs_put_page(page, 1);
 555         }
 556         /* clear Orphan Flag */
 557         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 558         return 0;
 559 }
 560
 561 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 562 {
 563         struct list_head *head;
 564         struct f2fs_orphan_block *orphan_blk = NULL;
 565         unsigned int nentries = 0;
 566         unsigned short index = 1;
 567         unsigned short orphan_blocks;
 568         struct page *page = NULL;
 569         struct ino_entry *orphan = NULL;
 570         struct inode_management *im = &sbi->im[ORPHAN_INO];
 571
 572         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 573
 574         /*
 575          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 576          * orphan inode operations are covered under f2fs_lock_op().
 577          * And, spin_lock should be avoided due to page operations below.
 578          */
 579         head = &im->ino_list;
 580
 581         /* loop for each orphan inode entry and write them in Jornal block */
 582         list_for_each_entry(orphan, head, list) {
 583                 if (!page) {
 584                         page = grab_meta_page(sbi, start_blk++);
 585                         orphan_blk =
 586                                 (struct f2fs_orphan_block *)page_address(page);
 587                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 588                 }
 589
 590                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 591
 592                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 593                         /*
 594                          * an orphan block is full of 1020 entries,
 595                          * then we need to flush current orphan blocks
 596                          * and bring another one in memory
 597                          */
 598                         orphan_blk->blk_addr = cpu_to_le16(index);
 599                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 600                         orphan_blk->entry_count = cpu_to_le32(nentries);
 601                         set_page_dirty(page);
 602                         f2fs_put_page(page, 1);
 603                         index++;
 604                         nentries = 0;
 605                         page = NULL;
 606                 }
 607         }
 608
 609         if (page) {
 610                 orphan_blk->blk_addr = cpu_to_le16(index);
 611                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 612                 orphan_blk->entry_count = cpu_to_le32(nentries);
 613                 set_page_dirty(page);
 614                 f2fs_put_page(page, 1);
 615         }
 616 }
 617
 618 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 619                                 block_t cp_addr, unsigned long long *version)
 620 {
 621         struct page *cp_page_1, *cp_page_2 = NULL;
 622         unsigned long blk_size = sbi->blocksize;
 623         struct f2fs_checkpoint *cp_block;
 624         unsigned long long cur_version = 0, pre_version = 0;
 625         size_t crc_offset;
 626         __u32 crc = 0;
 627
 628         /* Read the 1st cp block in this CP pack */
 629         cp_page_1 = get_meta_page(sbi, cp_addr);
 630
 631         /* get the version number */
 632         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 633         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 634         if (crc_offset >= blk_size)
 635                 goto invalid_cp1;
 636
 637         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 638         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 639                 goto invalid_cp1;
 640
 641         pre_version = cur_cp_version(cp_block);
 642
 643         /* Read the 2nd cp block in this CP pack */
 644         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 645         cp_page_2 = get_meta_page(sbi, cp_addr);
 646
 647         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 648         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 649         if (crc_offset >= blk_size)
 650                 goto invalid_cp2;
 651
 652         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 653         if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
 654                 goto invalid_cp2;
 655
 656         cur_version = cur_cp_version(cp_block);
 657
 658         if (cur_version == pre_version) {
 659                 *version = cur_version;
 660                 f2fs_put_page(cp_page_2, 1);
 661                 return cp_page_1;
 662         }
 663 invalid_cp2:
 664         f2fs_put_page(cp_page_2, 1);
 665 invalid_cp1:
 666         f2fs_put_page(cp_page_1, 1);
 667         return NULL;
 668 }
 669
 670 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 671 {
 672         struct f2fs_checkpoint *cp_block;
 673         struct f2fs_super_block *fsb = sbi->raw_super;
 674         struct page *cp1, *cp2, *cur_page;
 675         unsigned long blk_size = sbi->blocksize;
 676         unsigned long long cp1_version = 0, cp2_version = 0;
 677         unsigned long long cp_start_blk_no;
 678         unsigned int cp_blks = 1 + __cp_payload(sbi);
 679         block_t cp_blk_no;
 680         int i;
 681
 682         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 683         if (!sbi->ckpt)
 684                 return -ENOMEM;
 685         /*
 686          * Finding out valid cp block involves read both
 687          * sets( cp pack1 and cp pack 2)
 688          */
 689         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 690         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 691
 692         /* The second checkpoint pack should start at the next segment */
 693         cp_start_blk_no += ((unsigned long long)1) <<
 694                                 le32_to_cpu(fsb->log_blocks_per_seg);
 695         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 696
 697         if (cp1 && cp2) {
 698                 if (ver_after(cp2_version, cp1_version))
 699                         cur_page = cp2;
 700                 else
 701                         cur_page = cp1;
 702         } else if (cp1) {
 703                 cur_page = cp1;
 704         } else if (cp2) {
 705                 cur_page = cp2;
 706         } else {
 707                 goto fail_no_cp;
 708         }
 709
 710         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 711         memcpy(sbi->ckpt, cp_block, blk_size);
 712
 713         /* Sanity checking of checkpoint */
 714         if (sanity_check_ckpt(sbi))
 715                 goto fail_no_cp;
 716
 717         if (cp_blks <= 1)
 718                 goto done;
 719
 720         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 721         if (cur_page == cp2)
 722                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 723
 724         for (i = 1; i < cp_blks; i++) {
 725                 void *sit_bitmap_ptr;
 726                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 727
 728                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 729                 sit_bitmap_ptr = page_address(cur_page);
 730                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 731                 f2fs_put_page(cur_page, 1);
 732         }
 733 done:
 734         f2fs_put_page(cp1, 1);
 735         f2fs_put_page(cp2, 1);
 736         return 0;
 737
 738 fail_no_cp:
 739         kfree(sbi->ckpt);
 740         return -EINVAL;
 741 }
 742
 743 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 744 {
 745         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 746         struct f2fs_inode_info *fi = F2FS_I(inode);
 747         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 748
 749         if (is_inode_flag_set(fi, flag))
 750                 return;
 751
 752         set_inode_flag(fi, flag);
 753         list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
 754         stat_inc_dirty_inode(sbi, type);
 755 }
 756
 757 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 758 {
 759         struct f2fs_inode_info *fi = F2FS_I(inode);
 760         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 761
 762         if (get_dirty_pages(inode) ||
 763                         !is_inode_flag_set(F2FS_I(inode), flag))
 764                 return;
 765
 766         list_del_init(&fi->dirty_list);
 767         clear_inode_flag(fi, flag);
 768         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 769 }
 770
 771 void update_dirty_page(struct inode *inode, struct page *page)
 772 {
 773         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 774         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 775
 776         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 777                         !S_ISLNK(inode->i_mode))
 778                 return;
 779
 780         spin_lock(&sbi->inode_lock[type]);
 781         __add_dirty_inode(inode, type);
 782         inode_inc_dirty_pages(inode);
 783         spin_unlock(&sbi->inode_lock[type]);
 784
 785         SetPagePrivate(page);
 786         f2fs_trace_pid(page);
 787 }
 788
 789 void add_dirty_dir_inode(struct inode *inode)
 790 {
 791         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 792
 793         spin_lock(&sbi->inode_lock[DIR_INODE]);
 794         __add_dirty_inode(inode, DIR_INODE);
 795         spin_unlock(&sbi->inode_lock[DIR_INODE]);
 796 }
 797
 798 void remove_dirty_inode(struct inode *inode)
 799 {
 800         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 801         struct f2fs_inode_info *fi = F2FS_I(inode);
 802         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 803
 804         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 805                         !S_ISLNK(inode->i_mode))
 806                 return;
 807
 808         spin_lock(&sbi->inode_lock[type]);
 809         __remove_dirty_inode(inode, type);
 810         spin_unlock(&sbi->inode_lock[type]);
 811
 812         /* Only from the recovery routine */
 813         if (is_inode_flag_set(fi, FI_DELAY_IPUT)) {
 814                 clear_inode_flag(fi, FI_DELAY_IPUT);
 815                 iput(inode);
 816         }
 817 }
 818
 819 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 820 {
 821         struct list_head *head;
 822         struct inode *inode;
 823         struct f2fs_inode_info *fi;
 824         bool is_dir = (type == DIR_INODE);
 825
 826         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 827                                 get_pages(sbi, is_dir ?
 828                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 829 retry:
 830         if (unlikely(f2fs_cp_error(sbi)))
 831                 return -EIO;
 832
 833         spin_lock(&sbi->inode_lock[type]);
 834
 835         head = &sbi->inode_list[type];
 836         if (list_empty(head)) {
 837                 spin_unlock(&sbi->inode_lock[type]);
 838                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 839                                 get_pages(sbi, is_dir ?
 840                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 841                 return 0;
 842         }
 843         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 844         inode = igrab(&fi->vfs_inode);
 845         spin_unlock(&sbi->inode_lock[type]);
 846         if (inode) {
 847                 filemap_fdatawrite(inode->i_mapping);
 848                 iput(inode);
 849         } else {
 850                 /*
 851                  * We should submit bio, since it exists several
 852                  * wribacking dentry pages in the freeing inode.
 853                  */
 854                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 855                 cond_resched();
 856         }
 857         goto retry;
 858 }
 859
 860 /*
 861  * Freeze all the FS-operations for checkpoint.
 862  */
 863 static int block_operations(struct f2fs_sb_info *sbi)
 864 {
 865         struct writeback_control wbc = {
 866                 .sync_mode = WB_SYNC_ALL,
 867                 .nr_to_write = LONG_MAX,
 868                 .for_reclaim = 0,
 869         };
 870         struct blk_plug plug;
 871         int err = 0;
 872
 873         blk_start_plug(&plug);
 874
 875 retry_flush_dents:
 876         f2fs_lock_all(sbi);
 877         /* write all the dirty dentry pages */
 878         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 879                 f2fs_unlock_all(sbi);
 880                 err = sync_dirty_inodes(sbi, DIR_INODE);
 881                 if (err)
 882                         goto out;
 883                 goto retry_flush_dents;
 884         }
 885
 886         /*
 887          * POR: we should ensure that there are no dirty node pages
 888          * until finishing nat/sit flush.
 889          */
 890 retry_flush_nodes:
 891         down_write(&sbi->node_write);
 892
 893         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 894                 up_write(&sbi->node_write);
 895                 err = sync_node_pages(sbi, 0, &wbc);
 896                 if (err) {
 897                         f2fs_unlock_all(sbi);
 898                         goto out;
 899                 }
 900                 goto retry_flush_nodes;
 901         }
 902 out:
 903         blk_finish_plug(&plug);
 904         return err;
 905 }
 906
 907 static void unblock_operations(struct f2fs_sb_info *sbi)
 908 {
 909         up_write(&sbi->node_write);
 910         f2fs_unlock_all(sbi);
 911 }
 912
 913 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 914 {
 915         DEFINE_WAIT(wait);
 916
 917         for (;;) {
 918                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 919
 920                 if (!get_pages(sbi, F2FS_WRITEBACK))
 921                         break;
 922
 923                 io_schedule_timeout(5*HZ);
 924         }
 925         finish_wait(&sbi->cp_wait, &wait);
 926 }
 927
 928 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 929 {
 930         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 931         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 932         struct f2fs_nm_info *nm_i = NM_I(sbi);
 933         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 934         nid_t last_nid = nm_i->next_scan_nid;
 935         block_t start_blk;
 936         unsigned int data_sum_blocks, orphan_blocks;
 937         __u32 crc32 = 0;
 938         int i;
 939         int cp_payload_blks = __cp_payload(sbi);
 940         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
 941         bool invalidate = false;
 942         struct super_block *sb = sbi->sb;
 943         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
 944         u64 kbytes_written;
 945
 946         /*
 947          * This avoids to conduct wrong roll-forward operations and uses
 948          * metapages, so should be called prior to sync_meta_pages below.
 949          */
 950         if (discard_next_dnode(sbi, discard_blk))
 951                 invalidate = true;
 952
 953         /* Flush all the NAT/SIT pages */
 954         while (get_pages(sbi, F2FS_DIRTY_META)) {
 955                 sync_meta_pages(sbi, META, LONG_MAX);
 956                 if (unlikely(f2fs_cp_error(sbi)))
 957                         return -EIO;
 958         }
 959
 960         next_free_nid(sbi, &last_nid);
 961
 962         /*
 963          * modify checkpoint
 964          * version number is already updated
 965          */
 966         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 967         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 968         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 969         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 970                 ckpt->cur_node_segno[i] =
 971                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 972                 ckpt->cur_node_blkoff[i] =
 973                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 974                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 975                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 976         }
 977         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 978                 ckpt->cur_data_segno[i] =
 979                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 980                 ckpt->cur_data_blkoff[i] =
 981                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 982                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 983                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 984         }
 985
 986         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 987         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 988         ckpt->next_free_nid = cpu_to_le32(last_nid);
 989
 990         /* 2 cp  + n data seg summary + orphan inode blocks */
 991         data_sum_blocks = npages_for_summary_flush(sbi, false);
 992         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 993                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 994         else
 995                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 996
 997         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 998         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 999                         orphan_blocks);
1000
1001         if (__remain_node_summaries(cpc->reason))
1002                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1003                                 cp_payload_blks + data_sum_blocks +
1004                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1005         else
1006                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1007                                 cp_payload_blks + data_sum_blocks +
1008                                 orphan_blocks);
1009
1010         if (cpc->reason == CP_UMOUNT)
1011                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1012         else
1013                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1014
1015         if (cpc->reason == CP_FASTBOOT)
1016                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1017         else
1018                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1019
1020         if (orphan_num)
1021                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1022         else
1023                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1024
1025         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1026                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1027
1028         /* update SIT/NAT bitmap */
1029         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1030         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1031
1032         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1033         *((__le32 *)((unsigned char *)ckpt +
1034                                 le32_to_cpu(ckpt->checksum_offset)))
1035                                 = cpu_to_le32(crc32);
1036
1037         start_blk = __start_cp_addr(sbi);
1038
1039         /* need to wait for end_io results */
1040         wait_on_all_pages_writeback(sbi);
1041         if (unlikely(f2fs_cp_error(sbi)))
1042                 return -EIO;
1043
1044         /* write out checkpoint buffer at block 0 */
1045         update_meta_page(sbi, ckpt, start_blk++);
1046
1047         for (i = 1; i < 1 + cp_payload_blks; i++)
1048                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1049                                                         start_blk++);
1050
1051         if (orphan_num) {
1052                 write_orphan_inodes(sbi, start_blk);
1053                 start_blk += orphan_blocks;
1054         }
1055
1056         write_data_summaries(sbi, start_blk);
1057         start_blk += data_sum_blocks;
1058
1059         /* Record write statistics in the hot node summary */
1060         kbytes_written = sbi->kbytes_written;
1061         if (sb->s_bdev->bd_part)
1062                 kbytes_written += BD_PART_WRITTEN(sbi);
1063
1064         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1065
1066         if (__remain_node_summaries(cpc->reason)) {
1067                 write_node_summaries(sbi, start_blk);
1068                 start_blk += NR_CURSEG_NODE_TYPE;
1069         }
1070
1071         /* writeout checkpoint block */
1072         update_meta_page(sbi, ckpt, start_blk);
1073
1074         /* wait for previous submitted node/meta pages writeback */
1075         wait_on_all_pages_writeback(sbi);
1076
1077         if (unlikely(f2fs_cp_error(sbi)))
1078                 return -EIO;
1079
1080         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1081         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1082
1083         /* update user_block_counts */
1084         sbi->last_valid_block_count = sbi->total_valid_block_count;
1085         sbi->alloc_valid_block_count = 0;
1086
1087         /* Here, we only have one bio having CP pack */
1088         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1089
1090         /* wait for previous submitted meta pages writeback */
1091         wait_on_all_pages_writeback(sbi);
1092
1093         /*
1094          * invalidate meta page which is used temporarily for zeroing out
1095          * block at the end of warm node chain.
1096          */
1097         if (invalidate)
1098                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1099                                                                 discard_blk);
1100
1101         release_ino_entry(sbi);
1102
1103         if (unlikely(f2fs_cp_error(sbi)))
1104                 return -EIO;
1105
1106         clear_prefree_segments(sbi, cpc);
1107         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1108
1109         return 0;
1110 }
1111
1112 /*
1113  * We guarantee that this checkpoint procedure will not fail.
1114  */
1115 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1116 {
1117         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1118         unsigned long long ckpt_ver;
1119         int err = 0;
1120
1121         mutex_lock(&sbi->cp_mutex);
1122
1123         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1124                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1125                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1126                 goto out;
1127         if (unlikely(f2fs_cp_error(sbi))) {
1128                 err = -EIO;
1129                 goto out;
1130         }
1131         if (f2fs_readonly(sbi->sb)) {
1132                 err = -EROFS;
1133                 goto out;
1134         }
1135
1136         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1137
1138         err = block_operations(sbi);
1139         if (err)
1140                 goto out;
1141
1142         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1143
1144         f2fs_flush_merged_bios(sbi);
1145
1146         /*
1147          * update checkpoint pack index
1148          * Increase the version number so that
1149          * SIT entries and seg summaries are written at correct place
1150          */
1151         ckpt_ver = cur_cp_version(ckpt);
1152         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1153
1154         /* write cached NAT/SIT entries to NAT/SIT area */
1155         flush_nat_entries(sbi);
1156         flush_sit_entries(sbi, cpc);
1157
1158         /* unlock all the fs_lock[] in do_checkpoint() */
1159         err = do_checkpoint(sbi, cpc);
1160
1161         unblock_operations(sbi);
1162         stat_inc_cp_count(sbi->stat_info);
1163
1164         if (cpc->reason == CP_RECOVERY)
1165                 f2fs_msg(sbi->sb, KERN_NOTICE,
1166                         "checkpoint: version = %llx", ckpt_ver);
1167
1168         /* do checkpoint periodically */
1169         f2fs_update_time(sbi, CP_TIME);
1170         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1171 out:
1172         mutex_unlock(&sbi->cp_mutex);
1173         return err;
1174 }
1175
1176 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1177 {
1178         int i;
1179
1180         for (i = 0; i < MAX_INO_ENTRY; i++) {
1181                 struct inode_management *im = &sbi->im[i];
1182
1183                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1184                 spin_lock_init(&im->ino_lock);
1185                 INIT_LIST_HEAD(&im->ino_list);
1186                 im->ino_num = 0;
1187         }
1188
1189         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1190                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1191                                 F2FS_ORPHANS_PER_BLOCK;
1192 }
1193
1194 int __init create_checkpoint_caches(void)
1195 {
1196         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1197                         sizeof(struct ino_entry));
1198         if (!ino_entry_slab)
1199                 return -ENOMEM;
1200         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1201                         sizeof(struct inode_entry));
1202         if (!inode_entry_slab) {
1203                 kmem_cache_destroy(ino_entry_slab);
1204                 return -ENOMEM;
1205         }
1206         return 0;
1207 }
1208
1209 void destroy_checkpoint_caches(void)
1210 {
1211         kmem_cache_destroy(ino_entry_slab);
1212         kmem_cache_destroy(inode_entry_slab);
1213 }