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                 .blk_addr = index,
  60                 .encrypted_page = NULL,
  61         };
  62
  63         if (unlikely(!is_meta))
  64                 fio.rw &= ~REQ_META;
  65 repeat:
  66         page = grab_cache_page(mapping, index);
  67         if (!page) {
  68                 cond_resched();
  69                 goto repeat;
  70         }
  71         if (PageUptodate(page))
  72                 goto out;
  73
  74         fio.page = page;
  75
  76         if (f2fs_submit_page_bio(&fio)) {
  77                 f2fs_put_page(page, 1);
  78                 goto repeat;
  79         }
  80
  81         lock_page(page);
  82         if (unlikely(page->mapping != mapping)) {
  83                 f2fs_put_page(page, 1);
  84                 goto repeat;
  85         }
  86
  87         /*
  88          * if there is any IO error when accessing device, make our filesystem
  89          * readonly and make sure do not write checkpoint with non-uptodate
  90          * meta page.
  91          */
  92         if (unlikely(!PageUptodate(page)))
  93                 f2fs_stop_checkpoint(sbi);
  94 out:
  95         return page;
  96 }
  97
  98 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  99 {
 100         return __get_meta_page(sbi, index, true);
 101 }
 102
 103 /* for POR only */
 104 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 105 {
 106         return __get_meta_page(sbi, index, false);
 107 }
 108
 109 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 110 {
 111         switch (type) {
 112         case META_NAT:
 113                 break;
 114         case META_SIT:
 115                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 116                         return false;
 117                 break;
 118         case META_SSA:
 119                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 120                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 121                         return false;
 122                 break;
 123         case META_CP:
 124                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 125                         blkaddr < __start_cp_addr(sbi)))
 126                         return false;
 127                 break;
 128         case META_POR:
 129                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 130                         blkaddr < MAIN_BLKADDR(sbi)))
 131                         return false;
 132                 break;
 133         default:
 134                 BUG();
 135         }
 136
 137         return true;
 138 }
 139
 140 /*
 141  * Readahead CP/NAT/SIT/SSA pages
 142  */
 143 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 144                                                         int type, bool sync)
 145 {
 146         struct page *page;
 147         block_t blkno = start;
 148         struct f2fs_io_info fio = {
 149                 .sbi = sbi,
 150                 .type = META,
 151                 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
 152                 .encrypted_page = NULL,
 153         };
 154         struct blk_plug plug;
 155
 156         if (unlikely(type == META_POR))
 157                 fio.rw &= ~REQ_META;
 158
 159         blk_start_plug(&plug);
 160         for (; nrpages-- > 0; blkno++) {
 161
 162                 if (!is_valid_blkaddr(sbi, blkno, type))
 163                         goto out;
 164
 165                 switch (type) {
 166                 case META_NAT:
 167                         if (unlikely(blkno >=
 168                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 169                                 blkno = 0;
 170                         /* get nat block addr */
 171                         fio.blk_addr = current_nat_addr(sbi,
 172                                         blkno * NAT_ENTRY_PER_BLOCK);
 173                         break;
 174                 case META_SIT:
 175                         /* get sit block addr */
 176                         fio.blk_addr = current_sit_addr(sbi,
 177                                         blkno * SIT_ENTRY_PER_BLOCK);
 178                         break;
 179                 case META_SSA:
 180                 case META_CP:
 181                 case META_POR:
 182                         fio.blk_addr = blkno;
 183                         break;
 184                 default:
 185                         BUG();
 186                 }
 187
 188                 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
 189                 if (!page)
 190                         continue;
 191                 if (PageUptodate(page)) {
 192                         f2fs_put_page(page, 1);
 193                         continue;
 194                 }
 195
 196                 fio.page = page;
 197                 f2fs_submit_page_mbio(&fio);
 198                 f2fs_put_page(page, 0);
 199         }
 200 out:
 201         f2fs_submit_merged_bio(sbi, META, READ);
 202         blk_finish_plug(&plug);
 203         return blkno - start;
 204 }
 205
 206 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 207 {
 208         struct page *page;
 209         bool readahead = false;
 210
 211         page = find_get_page(META_MAPPING(sbi), index);
 212         if (!page || (page && !PageUptodate(page)))
 213                 readahead = true;
 214         f2fs_put_page(page, 0);
 215
 216         if (readahead)
 217                 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
 218 }
 219
 220 static int f2fs_write_meta_page(struct page *page,
 221                                 struct writeback_control *wbc)
 222 {
 223         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 224
 225         trace_f2fs_writepage(page, META);
 226
 227         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 228                 goto redirty_out;
 229         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 230                 goto redirty_out;
 231         if (unlikely(f2fs_cp_error(sbi)))
 232                 goto redirty_out;
 233
 234         write_meta_page(sbi, page);
 235         dec_page_count(sbi, F2FS_DIRTY_META);
 236
 237         if (wbc->for_reclaim)
 238                 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
 239
 240         unlock_page(page);
 241
 242         if (unlikely(f2fs_cp_error(sbi)))
 243                 f2fs_submit_merged_bio(sbi, META, WRITE);
 244
 245         return 0;
 246
 247 redirty_out:
 248         redirty_page_for_writepage(wbc, page);
 249         return AOP_WRITEPAGE_ACTIVATE;
 250 }
 251
 252 static int f2fs_write_meta_pages(struct address_space *mapping,
 253                                 struct writeback_control *wbc)
 254 {
 255         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 256         long diff, written;
 257
 258         /* collect a number of dirty meta pages and write together */
 259         if (wbc->for_kupdate ||
 260                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 261                 goto skip_write;
 262
 263         trace_f2fs_writepages(mapping->host, wbc, META);
 264
 265         /* if mounting is failed, skip writing node pages */
 266         mutex_lock(&sbi->cp_mutex);
 267         diff = nr_pages_to_write(sbi, META, wbc);
 268         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 269         mutex_unlock(&sbi->cp_mutex);
 270         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 271         return 0;
 272
 273 skip_write:
 274         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 275         trace_f2fs_writepages(mapping->host, wbc, META);
 276         return 0;
 277 }
 278
 279 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 280                                                 long nr_to_write)
 281 {
 282         struct address_space *mapping = META_MAPPING(sbi);
 283         pgoff_t index = 0, end = LONG_MAX, prev = LONG_MAX;
 284         struct pagevec pvec;
 285         long nwritten = 0;
 286         struct writeback_control wbc = {
 287                 .for_reclaim = 0,
 288         };
 289         struct blk_plug plug;
 290
 291         pagevec_init(&pvec, 0);
 292
 293         blk_start_plug(&plug);
 294
 295         while (index <= end) {
 296                 int i, nr_pages;
 297                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 298                                 PAGECACHE_TAG_DIRTY,
 299                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 300                 if (unlikely(nr_pages == 0))
 301                         break;
 302
 303                 for (i = 0; i < nr_pages; i++) {
 304                         struct page *page = pvec.pages[i];
 305
 306                         if (prev == LONG_MAX)
 307                                 prev = page->index - 1;
 308                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 309                                 pagevec_release(&pvec);
 310                                 goto stop;
 311                         }
 312
 313                         lock_page(page);
 314
 315                         if (unlikely(page->mapping != mapping)) {
 316 continue_unlock:
 317                                 unlock_page(page);
 318                                 continue;
 319                         }
 320                         if (!PageDirty(page)) {
 321                                 /* someone wrote it for us */
 322                                 goto continue_unlock;
 323                         }
 324
 325                         f2fs_wait_on_page_writeback(page, META, true);
 326
 327                         BUG_ON(PageWriteback(page));
 328                         if (!clear_page_dirty_for_io(page))
 329                                 goto continue_unlock;
 330
 331                         if (mapping->a_ops->writepage(page, &wbc)) {
 332                                 unlock_page(page);
 333                                 break;
 334                         }
 335                         nwritten++;
 336                         prev = page->index;
 337                         if (unlikely(nwritten >= nr_to_write))
 338                                 break;
 339                 }
 340                 pagevec_release(&pvec);
 341                 cond_resched();
 342         }
 343 stop:
 344         if (nwritten)
 345                 f2fs_submit_merged_bio(sbi, type, WRITE);
 346
 347         blk_finish_plug(&plug);
 348
 349         return nwritten;
 350 }
 351
 352 static int f2fs_set_meta_page_dirty(struct page *page)
 353 {
 354         trace_f2fs_set_page_dirty(page, META);
 355
 356         SetPageUptodate(page);
 357         if (!PageDirty(page)) {
 358                 __set_page_dirty_nobuffers(page);
 359                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 360                 SetPagePrivate(page);
 361                 f2fs_trace_pid(page);
 362                 return 1;
 363         }
 364         return 0;
 365 }
 366
 367 const struct address_space_operations f2fs_meta_aops = {
 368         .writepage      = f2fs_write_meta_page,
 369         .writepages     = f2fs_write_meta_pages,
 370         .set_page_dirty = f2fs_set_meta_page_dirty,
 371         .invalidatepage = f2fs_invalidate_page,
 372         .releasepage    = f2fs_release_page,
 373 };
 374
 375 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 376 {
 377         struct inode_management *im = &sbi->im[type];
 378         struct ino_entry *e, *tmp;
 379
 380         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 381 retry:
 382         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 383
 384         spin_lock(&im->ino_lock);
 385         e = radix_tree_lookup(&im->ino_root, ino);
 386         if (!e) {
 387                 e = tmp;
 388                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 389                         spin_unlock(&im->ino_lock);
 390                         radix_tree_preload_end();
 391                         goto retry;
 392                 }
 393                 memset(e, 0, sizeof(struct ino_entry));
 394                 e->ino = ino;
 395
 396                 list_add_tail(&e->list, &im->ino_list);
 397                 if (type != ORPHAN_INO)
 398                         im->ino_num++;
 399         }
 400         spin_unlock(&im->ino_lock);
 401         radix_tree_preload_end();
 402
 403         if (e != tmp)
 404                 kmem_cache_free(ino_entry_slab, tmp);
 405 }
 406
 407 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 408 {
 409         struct inode_management *im = &sbi->im[type];
 410         struct ino_entry *e;
 411
 412         spin_lock(&im->ino_lock);
 413         e = radix_tree_lookup(&im->ino_root, ino);
 414         if (e) {
 415                 list_del(&e->list);
 416                 radix_tree_delete(&im->ino_root, ino);
 417                 im->ino_num--;
 418                 spin_unlock(&im->ino_lock);
 419                 kmem_cache_free(ino_entry_slab, e);
 420                 return;
 421         }
 422         spin_unlock(&im->ino_lock);
 423 }
 424
 425 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 426 {
 427         /* add new dirty ino entry into list */
 428         __add_ino_entry(sbi, ino, type);
 429 }
 430
 431 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 432 {
 433         /* remove dirty ino entry from list */
 434         __remove_ino_entry(sbi, ino, type);
 435 }
 436
 437 /* mode should be APPEND_INO or UPDATE_INO */
 438 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 439 {
 440         struct inode_management *im = &sbi->im[mode];
 441         struct ino_entry *e;
 442
 443         spin_lock(&im->ino_lock);
 444         e = radix_tree_lookup(&im->ino_root, ino);
 445         spin_unlock(&im->ino_lock);
 446         return e ? true : false;
 447 }
 448
 449 void release_ino_entry(struct f2fs_sb_info *sbi)
 450 {
 451         struct ino_entry *e, *tmp;
 452         int i;
 453
 454         for (i = APPEND_INO; i <= UPDATE_INO; i++) {
 455                 struct inode_management *im = &sbi->im[i];
 456
 457                 spin_lock(&im->ino_lock);
 458                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 459                         list_del(&e->list);
 460                         radix_tree_delete(&im->ino_root, e->ino);
 461                         kmem_cache_free(ino_entry_slab, e);
 462                         im->ino_num--;
 463                 }
 464                 spin_unlock(&im->ino_lock);
 465         }
 466 }
 467
 468 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 469 {
 470         struct inode_management *im = &sbi->im[ORPHAN_INO];
 471         int err = 0;
 472
 473         spin_lock(&im->ino_lock);
 474         if (unlikely(im->ino_num >= sbi->max_orphans))
 475                 err = -ENOSPC;
 476         else
 477                 im->ino_num++;
 478         spin_unlock(&im->ino_lock);
 479
 480         return err;
 481 }
 482
 483 void release_orphan_inode(struct f2fs_sb_info *sbi)
 484 {
 485         struct inode_management *im = &sbi->im[ORPHAN_INO];
 486
 487         spin_lock(&im->ino_lock);
 488         f2fs_bug_on(sbi, im->ino_num == 0);
 489         im->ino_num--;
 490         spin_unlock(&im->ino_lock);
 491 }
 492
 493 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 494 {
 495         /* add new orphan ino entry into list */
 496         __add_ino_entry(sbi, ino, ORPHAN_INO);
 497 }
 498
 499 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 500 {
 501         /* remove orphan entry from orphan list */
 502         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 503 }
 504
 505 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 506 {
 507         struct inode *inode;
 508
 509         inode = f2fs_iget(sbi->sb, ino);
 510         if (IS_ERR(inode)) {
 511                 /*
 512                  * there should be a bug that we can't find the entry
 513                  * to orphan inode.
 514                  */
 515                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 516                 return PTR_ERR(inode);
 517         }
 518
 519         clear_nlink(inode);
 520
 521         /* truncate all the data during iput */
 522         iput(inode);
 523         return 0;
 524 }
 525
 526 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 527 {
 528         block_t start_blk, orphan_blocks, i, j;
 529         int err;
 530
 531         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 532                 return 0;
 533
 534         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 535         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 536
 537         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 538
 539         for (i = 0; i < orphan_blocks; i++) {
 540                 struct page *page = get_meta_page(sbi, start_blk + i);
 541                 struct f2fs_orphan_block *orphan_blk;
 542
 543                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 544                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 545                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 546                         err = recover_orphan_inode(sbi, ino);
 547                         if (err) {
 548                                 f2fs_put_page(page, 1);
 549                                 return err;
 550                         }
 551                 }
 552                 f2fs_put_page(page, 1);
 553         }
 554         /* clear Orphan Flag */
 555         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 556         return 0;
 557 }
 558
 559 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 560 {
 561         struct list_head *head;
 562         struct f2fs_orphan_block *orphan_blk = NULL;
 563         unsigned int nentries = 0;
 564         unsigned short index = 1;
 565         unsigned short orphan_blocks;
 566         struct page *page = NULL;
 567         struct ino_entry *orphan = NULL;
 568         struct inode_management *im = &sbi->im[ORPHAN_INO];
 569
 570         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 571
 572         /*
 573          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 574          * orphan inode operations are covered under f2fs_lock_op().
 575          * And, spin_lock should be avoided due to page operations below.
 576          */
 577         head = &im->ino_list;
 578
 579         /* loop for each orphan inode entry and write them in Jornal block */
 580         list_for_each_entry(orphan, head, list) {
 581                 if (!page) {
 582                         page = grab_meta_page(sbi, start_blk++);
 583                         orphan_blk =
 584                                 (struct f2fs_orphan_block *)page_address(page);
 585                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 586                 }
 587
 588                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 589
 590                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 591                         /*
 592                          * an orphan block is full of 1020 entries,
 593                          * then we need to flush current orphan blocks
 594                          * and bring another one in memory
 595                          */
 596                         orphan_blk->blk_addr = cpu_to_le16(index);
 597                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 598                         orphan_blk->entry_count = cpu_to_le32(nentries);
 599                         set_page_dirty(page);
 600                         f2fs_put_page(page, 1);
 601                         index++;
 602                         nentries = 0;
 603                         page = NULL;
 604                 }
 605         }
 606
 607         if (page) {
 608                 orphan_blk->blk_addr = cpu_to_le16(index);
 609                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 610                 orphan_blk->entry_count = cpu_to_le32(nentries);
 611                 set_page_dirty(page);
 612                 f2fs_put_page(page, 1);
 613         }
 614 }
 615
 616 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 617                                 block_t cp_addr, unsigned long long *version)
 618 {
 619         struct page *cp_page_1, *cp_page_2 = NULL;
 620         unsigned long blk_size = sbi->blocksize;
 621         struct f2fs_checkpoint *cp_block;
 622         unsigned long long cur_version = 0, pre_version = 0;
 623         size_t crc_offset;
 624         __u32 crc = 0;
 625
 626         /* Read the 1st cp block in this CP pack */
 627         cp_page_1 = get_meta_page(sbi, cp_addr);
 628
 629         /* get the version number */
 630         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 631         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 632         if (crc_offset >= blk_size)
 633                 goto invalid_cp1;
 634
 635         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 636         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 637                 goto invalid_cp1;
 638
 639         pre_version = cur_cp_version(cp_block);
 640
 641         /* Read the 2nd cp block in this CP pack */
 642         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 643         cp_page_2 = get_meta_page(sbi, cp_addr);
 644
 645         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 646         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 647         if (crc_offset >= blk_size)
 648                 goto invalid_cp2;
 649
 650         crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
 651         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 652                 goto invalid_cp2;
 653
 654         cur_version = cur_cp_version(cp_block);
 655
 656         if (cur_version == pre_version) {
 657                 *version = cur_version;
 658                 f2fs_put_page(cp_page_2, 1);
 659                 return cp_page_1;
 660         }
 661 invalid_cp2:
 662         f2fs_put_page(cp_page_2, 1);
 663 invalid_cp1:
 664         f2fs_put_page(cp_page_1, 1);
 665         return NULL;
 666 }
 667
 668 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 669 {
 670         struct f2fs_checkpoint *cp_block;
 671         struct f2fs_super_block *fsb = sbi->raw_super;
 672         struct page *cp1, *cp2, *cur_page;
 673         unsigned long blk_size = sbi->blocksize;
 674         unsigned long long cp1_version = 0, cp2_version = 0;
 675         unsigned long long cp_start_blk_no;
 676         unsigned int cp_blks = 1 + __cp_payload(sbi);
 677         block_t cp_blk_no;
 678         int i;
 679
 680         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 681         if (!sbi->ckpt)
 682                 return -ENOMEM;
 683         /*
 684          * Finding out valid cp block involves read both
 685          * sets( cp pack1 and cp pack 2)
 686          */
 687         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 688         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 689
 690         /* The second checkpoint pack should start at the next segment */
 691         cp_start_blk_no += ((unsigned long long)1) <<
 692                                 le32_to_cpu(fsb->log_blocks_per_seg);
 693         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 694
 695         if (cp1 && cp2) {
 696                 if (ver_after(cp2_version, cp1_version))
 697                         cur_page = cp2;
 698                 else
 699                         cur_page = cp1;
 700         } else if (cp1) {
 701                 cur_page = cp1;
 702         } else if (cp2) {
 703                 cur_page = cp2;
 704         } else {
 705                 goto fail_no_cp;
 706         }
 707
 708         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 709         memcpy(sbi->ckpt, cp_block, blk_size);
 710
 711         if (cp_blks <= 1)
 712                 goto done;
 713
 714         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 715         if (cur_page == cp2)
 716                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 717
 718         for (i = 1; i < cp_blks; i++) {
 719                 void *sit_bitmap_ptr;
 720                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 721
 722                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 723                 sit_bitmap_ptr = page_address(cur_page);
 724                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 725                 f2fs_put_page(cur_page, 1);
 726         }
 727 done:
 728         f2fs_put_page(cp1, 1);
 729         f2fs_put_page(cp2, 1);
 730         return 0;
 731
 732 fail_no_cp:
 733         kfree(sbi->ckpt);
 734         return -EINVAL;
 735 }
 736
 737 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 738 {
 739         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 740         struct f2fs_inode_info *fi = F2FS_I(inode);
 741         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 742
 743         if (is_inode_flag_set(fi, flag))
 744                 return;
 745
 746         set_inode_flag(fi, flag);
 747         list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
 748         stat_inc_dirty_inode(sbi, type);
 749 }
 750
 751 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 752 {
 753         struct f2fs_inode_info *fi = F2FS_I(inode);
 754         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 755
 756         if (get_dirty_pages(inode) ||
 757                         !is_inode_flag_set(F2FS_I(inode), flag))
 758                 return;
 759
 760         list_del_init(&fi->dirty_list);
 761         clear_inode_flag(fi, flag);
 762         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 763 }
 764
 765 void update_dirty_page(struct inode *inode, struct page *page)
 766 {
 767         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 768         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 769
 770         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 771                         !S_ISLNK(inode->i_mode))
 772                 return;
 773
 774         spin_lock(&sbi->inode_lock[type]);
 775         __add_dirty_inode(inode, type);
 776         inode_inc_dirty_pages(inode);
 777         spin_unlock(&sbi->inode_lock[type]);
 778
 779         SetPagePrivate(page);
 780         f2fs_trace_pid(page);
 781 }
 782
 783 void add_dirty_dir_inode(struct inode *inode)
 784 {
 785         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 786
 787         spin_lock(&sbi->inode_lock[DIR_INODE]);
 788         __add_dirty_inode(inode, DIR_INODE);
 789         spin_unlock(&sbi->inode_lock[DIR_INODE]);
 790 }
 791
 792 void remove_dirty_inode(struct inode *inode)
 793 {
 794         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 795         struct f2fs_inode_info *fi = F2FS_I(inode);
 796         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 797
 798         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 799                         !S_ISLNK(inode->i_mode))
 800                 return;
 801
 802         spin_lock(&sbi->inode_lock[type]);
 803         __remove_dirty_inode(inode, type);
 804         spin_unlock(&sbi->inode_lock[type]);
 805
 806         /* Only from the recovery routine */
 807         if (is_inode_flag_set(fi, FI_DELAY_IPUT)) {
 808                 clear_inode_flag(fi, FI_DELAY_IPUT);
 809                 iput(inode);
 810         }
 811 }
 812
 813 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 814 {
 815         struct list_head *head;
 816         struct inode *inode;
 817         struct f2fs_inode_info *fi;
 818         bool is_dir = (type == DIR_INODE);
 819
 820         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 821                                 get_pages(sbi, is_dir ?
 822                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 823 retry:
 824         if (unlikely(f2fs_cp_error(sbi)))
 825                 return -EIO;
 826
 827         spin_lock(&sbi->inode_lock[type]);
 828
 829         head = &sbi->inode_list[type];
 830         if (list_empty(head)) {
 831                 spin_unlock(&sbi->inode_lock[type]);
 832                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 833                                 get_pages(sbi, is_dir ?
 834                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 835                 return 0;
 836         }
 837         fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
 838         inode = igrab(&fi->vfs_inode);
 839         spin_unlock(&sbi->inode_lock[type]);
 840         if (inode) {
 841                 filemap_fdatawrite(inode->i_mapping);
 842                 iput(inode);
 843         } else {
 844                 /*
 845                  * We should submit bio, since it exists several
 846                  * wribacking dentry pages in the freeing inode.
 847                  */
 848                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 849                 cond_resched();
 850         }
 851         goto retry;
 852 }
 853
 854 /*
 855  * Freeze all the FS-operations for checkpoint.
 856  */
 857 static int block_operations(struct f2fs_sb_info *sbi)
 858 {
 859         struct writeback_control wbc = {
 860                 .sync_mode = WB_SYNC_ALL,
 861                 .nr_to_write = LONG_MAX,
 862                 .for_reclaim = 0,
 863         };
 864         struct blk_plug plug;
 865         int err = 0;
 866
 867         blk_start_plug(&plug);
 868
 869 retry_flush_dents:
 870         f2fs_lock_all(sbi);
 871         /* write all the dirty dentry pages */
 872         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 873                 f2fs_unlock_all(sbi);
 874                 err = sync_dirty_inodes(sbi, DIR_INODE);
 875                 if (err)
 876                         goto out;
 877                 goto retry_flush_dents;
 878         }
 879
 880         /*
 881          * POR: we should ensure that there are no dirty node pages
 882          * until finishing nat/sit flush.
 883          */
 884 retry_flush_nodes:
 885         down_write(&sbi->node_write);
 886
 887         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 888                 up_write(&sbi->node_write);
 889                 err = sync_node_pages(sbi, 0, &wbc);
 890                 if (err) {
 891                         f2fs_unlock_all(sbi);
 892                         goto out;
 893                 }
 894                 goto retry_flush_nodes;
 895         }
 896 out:
 897         blk_finish_plug(&plug);
 898         return err;
 899 }
 900
 901 static void unblock_operations(struct f2fs_sb_info *sbi)
 902 {
 903         up_write(&sbi->node_write);
 904         f2fs_unlock_all(sbi);
 905 }
 906
 907 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 908 {
 909         DEFINE_WAIT(wait);
 910
 911         for (;;) {
 912                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 913
 914                 if (!get_pages(sbi, F2FS_WRITEBACK))
 915                         break;
 916
 917                 io_schedule();
 918         }
 919         finish_wait(&sbi->cp_wait, &wait);
 920 }
 921
 922 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 923 {
 924         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 925         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 926         struct f2fs_nm_info *nm_i = NM_I(sbi);
 927         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
 928         nid_t last_nid = nm_i->next_scan_nid;
 929         block_t start_blk;
 930         unsigned int data_sum_blocks, orphan_blocks;
 931         __u32 crc32 = 0;
 932         int i;
 933         int cp_payload_blks = __cp_payload(sbi);
 934         block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
 935         bool invalidate = false;
 936         struct super_block *sb = sbi->sb;
 937         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
 938         u64 kbytes_written;
 939
 940         /*
 941          * This avoids to conduct wrong roll-forward operations and uses
 942          * metapages, so should be called prior to sync_meta_pages below.
 943          */
 944         if (discard_next_dnode(sbi, discard_blk))
 945                 invalidate = true;
 946
 947         /* Flush all the NAT/SIT pages */
 948         while (get_pages(sbi, F2FS_DIRTY_META)) {
 949                 sync_meta_pages(sbi, META, LONG_MAX);
 950                 if (unlikely(f2fs_cp_error(sbi)))
 951                         return -EIO;
 952         }
 953
 954         next_free_nid(sbi, &last_nid);
 955
 956         /*
 957          * modify checkpoint
 958          * version number is already updated
 959          */
 960         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 961         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 962         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 963         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
 964                 ckpt->cur_node_segno[i] =
 965                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 966                 ckpt->cur_node_blkoff[i] =
 967                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 968                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 969                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 970         }
 971         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
 972                 ckpt->cur_data_segno[i] =
 973                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 974                 ckpt->cur_data_blkoff[i] =
 975                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 976                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 977                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 978         }
 979
 980         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 981         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 982         ckpt->next_free_nid = cpu_to_le32(last_nid);
 983
 984         /* 2 cp  + n data seg summary + orphan inode blocks */
 985         data_sum_blocks = npages_for_summary_flush(sbi, false);
 986         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
 987                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 988         else
 989                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 990
 991         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
 992         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
 993                         orphan_blocks);
 994
 995         if (__remain_node_summaries(cpc->reason))
 996                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
 997                                 cp_payload_blks + data_sum_blocks +
 998                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
 999         else
1000                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1001                                 cp_payload_blks + data_sum_blocks +
1002                                 orphan_blocks);
1003
1004         if (cpc->reason == CP_UMOUNT)
1005                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1006         else
1007                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1008
1009         if (cpc->reason == CP_FASTBOOT)
1010                 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1011         else
1012                 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1013
1014         if (orphan_num)
1015                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1016         else
1017                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1018
1019         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1020                 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1021
1022         /* update SIT/NAT bitmap */
1023         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1024         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1025
1026         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
1027         *((__le32 *)((unsigned char *)ckpt +
1028                                 le32_to_cpu(ckpt->checksum_offset)))
1029                                 = cpu_to_le32(crc32);
1030
1031         start_blk = __start_cp_addr(sbi);
1032
1033         /* need to wait for end_io results */
1034         wait_on_all_pages_writeback(sbi);
1035         if (unlikely(f2fs_cp_error(sbi)))
1036                 return -EIO;
1037
1038         /* write out checkpoint buffer at block 0 */
1039         update_meta_page(sbi, ckpt, start_blk++);
1040
1041         for (i = 1; i < 1 + cp_payload_blks; i++)
1042                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1043                                                         start_blk++);
1044
1045         if (orphan_num) {
1046                 write_orphan_inodes(sbi, start_blk);
1047                 start_blk += orphan_blocks;
1048         }
1049
1050         write_data_summaries(sbi, start_blk);
1051         start_blk += data_sum_blocks;
1052
1053         /* Record write statistics in the hot node summary */
1054         kbytes_written = sbi->kbytes_written;
1055         if (sb->s_bdev->bd_part)
1056                 kbytes_written += BD_PART_WRITTEN(sbi);
1057
1058         seg_i->sum_blk->journal.info.kbytes_written = cpu_to_le64(kbytes_written);
1059
1060         if (__remain_node_summaries(cpc->reason)) {
1061                 write_node_summaries(sbi, start_blk);
1062                 start_blk += NR_CURSEG_NODE_TYPE;
1063         }
1064
1065         /* writeout checkpoint block */
1066         update_meta_page(sbi, ckpt, start_blk);
1067
1068         /* wait for previous submitted node/meta pages writeback */
1069         wait_on_all_pages_writeback(sbi);
1070
1071         if (unlikely(f2fs_cp_error(sbi)))
1072                 return -EIO;
1073
1074         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1075         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1076
1077         /* update user_block_counts */
1078         sbi->last_valid_block_count = sbi->total_valid_block_count;
1079         sbi->alloc_valid_block_count = 0;
1080
1081         /* Here, we only have one bio having CP pack */
1082         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1083
1084         /* wait for previous submitted meta pages writeback */
1085         wait_on_all_pages_writeback(sbi);
1086
1087         /*
1088          * invalidate meta page which is used temporarily for zeroing out
1089          * block at the end of warm node chain.
1090          */
1091         if (invalidate)
1092                 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1093                                                                 discard_blk);
1094
1095         release_ino_entry(sbi);
1096
1097         if (unlikely(f2fs_cp_error(sbi)))
1098                 return -EIO;
1099
1100         clear_prefree_segments(sbi, cpc);
1101         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1102
1103         return 0;
1104 }
1105
1106 /*
1107  * We guarantee that this checkpoint procedure will not fail.
1108  */
1109 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1110 {
1111         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1112         unsigned long long ckpt_ver;
1113         int err = 0;
1114
1115         mutex_lock(&sbi->cp_mutex);
1116
1117         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1118                 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1119                 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1120                 goto out;
1121         if (unlikely(f2fs_cp_error(sbi))) {
1122                 err = -EIO;
1123                 goto out;
1124         }
1125         if (f2fs_readonly(sbi->sb)) {
1126                 err = -EROFS;
1127                 goto out;
1128         }
1129
1130         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1131
1132         err = block_operations(sbi);
1133         if (err)
1134                 goto out;
1135
1136         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1137
1138         f2fs_submit_merged_bio(sbi, DATA, WRITE);
1139         f2fs_submit_merged_bio(sbi, NODE, WRITE);
1140         f2fs_submit_merged_bio(sbi, META, WRITE);
1141
1142         /*
1143          * update checkpoint pack index
1144          * Increase the version number so that
1145          * SIT entries and seg summaries are written at correct place
1146          */
1147         ckpt_ver = cur_cp_version(ckpt);
1148         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1149
1150         /* write cached NAT/SIT entries to NAT/SIT area */
1151         flush_nat_entries(sbi);
1152         flush_sit_entries(sbi, cpc);
1153
1154         /* unlock all the fs_lock[] in do_checkpoint() */
1155         err = do_checkpoint(sbi, cpc);
1156
1157         unblock_operations(sbi);
1158         stat_inc_cp_count(sbi->stat_info);
1159
1160         if (cpc->reason == CP_RECOVERY)
1161                 f2fs_msg(sbi->sb, KERN_NOTICE,
1162                         "checkpoint: version = %llx", ckpt_ver);
1163
1164         /* do checkpoint periodically */
1165         f2fs_update_time(sbi, CP_TIME);
1166         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1167 out:
1168         mutex_unlock(&sbi->cp_mutex);
1169         return err;
1170 }
1171
1172 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1173 {
1174         int i;
1175
1176         for (i = 0; i < MAX_INO_ENTRY; i++) {
1177                 struct inode_management *im = &sbi->im[i];
1178
1179                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1180                 spin_lock_init(&im->ino_lock);
1181                 INIT_LIST_HEAD(&im->ino_list);
1182                 im->ino_num = 0;
1183         }
1184
1185         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1186                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1187                                 F2FS_ORPHANS_PER_BLOCK;
1188 }
1189
1190 int __init create_checkpoint_caches(void)
1191 {
1192         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1193                         sizeof(struct ino_entry));
1194         if (!ino_entry_slab)
1195                 return -ENOMEM;
1196         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1197                         sizeof(struct inode_entry));
1198         if (!inode_entry_slab) {
1199                 kmem_cache_destroy(ino_entry_slab);
1200                 return -ENOMEM;
1201         }
1202         return 0;
1203 }
1204
1205 void destroy_checkpoint_caches(void)
1206 {
1207         kmem_cache_destroy(ino_entry_slab);
1208         kmem_cache_destroy(inode_entry_slab);
1209 }