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 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  32         sbi->sb->s_flags |= MS_RDONLY;
  33         if (!end_io)
  34                 f2fs_flush_merged_writes(sbi);
  35 }
  36
  37 /*
  38  * We guarantee no failure on the returned page.
  39  */
  40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  41 {
  42         struct address_space *mapping = META_MAPPING(sbi);
  43         struct page *page = NULL;
  44 repeat:
  45         page = f2fs_grab_cache_page(mapping, index, false);
  46         if (!page) {
  47                 cond_resched();
  48                 goto repeat;
  49         }
  50         f2fs_wait_on_page_writeback(page, META, true);
  51         if (!PageUptodate(page))
  52                 SetPageUptodate(page);
  53         return page;
  54 }
  55
  56 /*
  57  * We guarantee no failure on the returned page.
  58  */
  59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  60                                                         bool is_meta)
  61 {
  62         struct address_space *mapping = META_MAPPING(sbi);
  63         struct page *page;
  64         struct f2fs_io_info fio = {
  65                 .sbi = sbi,
  66                 .type = META,
  67                 .op = REQ_OP_READ,
  68                 .op_flags = REQ_META | REQ_PRIO,
  69                 .old_blkaddr = index,
  70                 .new_blkaddr = index,
  71                 .encrypted_page = NULL,
  72         };
  73
  74         if (unlikely(!is_meta))
  75                 fio.op_flags &= ~REQ_META;
  76 repeat:
  77         page = f2fs_grab_cache_page(mapping, index, false);
  78         if (!page) {
  79                 cond_resched();
  80                 goto repeat;
  81         }
  82         if (PageUptodate(page))
  83                 goto out;
  84
  85         fio.page = page;
  86
  87         if (f2fs_submit_page_bio(&fio)) {
  88                 f2fs_put_page(page, 1);
  89                 goto repeat;
  90         }
  91
  92         lock_page(page);
  93         if (unlikely(page->mapping != mapping)) {
  94                 f2fs_put_page(page, 1);
  95                 goto repeat;
  96         }
  97
  98         /*
  99          * if there is any IO error when accessing device, make our filesystem
 100          * readonly and make sure do not write checkpoint with non-uptodate
 101          * meta page.
 102          */
 103         if (unlikely(!PageUptodate(page)))
 104                 f2fs_stop_checkpoint(sbi, false);
 105 out:
 106         return page;
 107 }
 108
 109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 110 {
 111         return __get_meta_page(sbi, index, true);
 112 }
 113
 114 /* for POR only */
 115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 116 {
 117         return __get_meta_page(sbi, index, false);
 118 }
 119
 120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 121 {
 122         switch (type) {
 123         case META_NAT:
 124                 break;
 125         case META_SIT:
 126                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 127                         return false;
 128                 break;
 129         case META_SSA:
 130                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 131                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 132                         return false;
 133                 break;
 134         case META_CP:
 135                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 136                         blkaddr < __start_cp_addr(sbi)))
 137                         return false;
 138                 break;
 139         case META_POR:
 140                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 141                         blkaddr < MAIN_BLKADDR(sbi)))
 142                         return false;
 143                 break;
 144         default:
 145                 BUG();
 146         }
 147
 148         return true;
 149 }
 150
 151 /*
 152  * Readahead CP/NAT/SIT/SSA pages
 153  */
 154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 155                                                         int type, bool sync)
 156 {
 157         struct page *page;
 158         block_t blkno = start;
 159         struct f2fs_io_info fio = {
 160                 .sbi = sbi,
 161                 .type = META,
 162                 .op = REQ_OP_READ,
 163                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 164                 .encrypted_page = NULL,
 165                 .in_list = false,
 166         };
 167         struct blk_plug plug;
 168
 169         if (unlikely(type == META_POR))
 170                 fio.op_flags &= ~REQ_META;
 171
 172         blk_start_plug(&plug);
 173         for (; nrpages-- > 0; blkno++) {
 174
 175                 if (!is_valid_blkaddr(sbi, blkno, type))
 176                         goto out;
 177
 178                 switch (type) {
 179                 case META_NAT:
 180                         if (unlikely(blkno >=
 181                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 182                                 blkno = 0;
 183                         /* get nat block addr */
 184                         fio.new_blkaddr = current_nat_addr(sbi,
 185                                         blkno * NAT_ENTRY_PER_BLOCK);
 186                         break;
 187                 case META_SIT:
 188                         /* get sit block addr */
 189                         fio.new_blkaddr = current_sit_addr(sbi,
 190                                         blkno * SIT_ENTRY_PER_BLOCK);
 191                         break;
 192                 case META_SSA:
 193                 case META_CP:
 194                 case META_POR:
 195                         fio.new_blkaddr = blkno;
 196                         break;
 197                 default:
 198                         BUG();
 199                 }
 200
 201                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 202                                                 fio.new_blkaddr, false);
 203                 if (!page)
 204                         continue;
 205                 if (PageUptodate(page)) {
 206                         f2fs_put_page(page, 1);
 207                         continue;
 208                 }
 209
 210                 fio.page = page;
 211                 f2fs_submit_page_bio(&fio);
 212                 f2fs_put_page(page, 0);
 213         }
 214 out:
 215         blk_finish_plug(&plug);
 216         return blkno - start;
 217 }
 218
 219 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 220 {
 221         struct page *page;
 222         bool readahead = false;
 223
 224         page = find_get_page(META_MAPPING(sbi), index);
 225         if (!page || !PageUptodate(page))
 226                 readahead = true;
 227         f2fs_put_page(page, 0);
 228
 229         if (readahead)
 230                 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 231 }
 232
 233 static int f2fs_write_meta_page(struct page *page,
 234                                 struct writeback_control *wbc)
 235 {
 236         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 237
 238         trace_f2fs_writepage(page, META);
 239
 240         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 241                 goto redirty_out;
 242         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 243                 goto redirty_out;
 244         if (unlikely(f2fs_cp_error(sbi)))
 245                 goto redirty_out;
 246
 247         write_meta_page(sbi, page);
 248         dec_page_count(sbi, F2FS_DIRTY_META);
 249
 250         if (wbc->for_reclaim)
 251                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
 252                                                 0, page->index, META);
 253
 254         unlock_page(page);
 255
 256         if (unlikely(f2fs_cp_error(sbi)))
 257                 f2fs_submit_merged_write(sbi, META);
 258
 259         return 0;
 260
 261 redirty_out:
 262         redirty_page_for_writepage(wbc, page);
 263         return AOP_WRITEPAGE_ACTIVATE;
 264 }
 265
 266 static int f2fs_write_meta_pages(struct address_space *mapping,
 267                                 struct writeback_control *wbc)
 268 {
 269         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 270         long diff, written;
 271
 272         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 273                 goto skip_write;
 274
 275         /* collect a number of dirty meta pages and write together */
 276         if (wbc->for_kupdate ||
 277                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 278                 goto skip_write;
 279
 280         /* if locked failed, cp will flush dirty pages instead */
 281         if (!mutex_trylock(&sbi->cp_mutex))
 282                 goto skip_write;
 283
 284         trace_f2fs_writepages(mapping->host, wbc, META);
 285         diff = nr_pages_to_write(sbi, META, wbc);
 286         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 287         mutex_unlock(&sbi->cp_mutex);
 288         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 289         return 0;
 290
 291 skip_write:
 292         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 293         trace_f2fs_writepages(mapping->host, wbc, META);
 294         return 0;
 295 }
 296
 297 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 298                                                 long nr_to_write)
 299 {
 300         struct address_space *mapping = META_MAPPING(sbi);
 301         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 302         struct pagevec pvec;
 303         long nwritten = 0;
 304         struct writeback_control wbc = {
 305                 .for_reclaim = 0,
 306         };
 307         struct blk_plug plug;
 308
 309         pagevec_init(&pvec, 0);
 310
 311         blk_start_plug(&plug);
 312
 313         while (index <= end) {
 314                 int i, nr_pages;
 315                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 316                                 PAGECACHE_TAG_DIRTY,
 317                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 318                 if (unlikely(nr_pages == 0))
 319                         break;
 320
 321                 for (i = 0; i < nr_pages; i++) {
 322                         struct page *page = pvec.pages[i];
 323
 324                         if (prev == ULONG_MAX)
 325                                 prev = page->index - 1;
 326                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 327                                 pagevec_release(&pvec);
 328                                 goto stop;
 329                         }
 330
 331                         lock_page(page);
 332
 333                         if (unlikely(page->mapping != mapping)) {
 334 continue_unlock:
 335                                 unlock_page(page);
 336                                 continue;
 337                         }
 338                         if (!PageDirty(page)) {
 339                                 /* someone wrote it for us */
 340                                 goto continue_unlock;
 341                         }
 342
 343                         f2fs_wait_on_page_writeback(page, META, true);
 344
 345                         BUG_ON(PageWriteback(page));
 346                         if (!clear_page_dirty_for_io(page))
 347                                 goto continue_unlock;
 348
 349                         if (mapping->a_ops->writepage(page, &wbc)) {
 350                                 unlock_page(page);
 351                                 break;
 352                         }
 353                         nwritten++;
 354                         prev = page->index;
 355                         if (unlikely(nwritten >= nr_to_write))
 356                                 break;
 357                 }
 358                 pagevec_release(&pvec);
 359                 cond_resched();
 360         }
 361 stop:
 362         if (nwritten)
 363                 f2fs_submit_merged_write(sbi, type);
 364
 365         blk_finish_plug(&plug);
 366
 367         return nwritten;
 368 }
 369
 370 static int f2fs_set_meta_page_dirty(struct page *page)
 371 {
 372         trace_f2fs_set_page_dirty(page, META);
 373
 374         if (!PageUptodate(page))
 375                 SetPageUptodate(page);
 376         if (!PageDirty(page)) {
 377                 f2fs_set_page_dirty_nobuffers(page);
 378                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 379                 SetPagePrivate(page);
 380                 f2fs_trace_pid(page);
 381                 return 1;
 382         }
 383         return 0;
 384 }
 385
 386 const struct address_space_operations f2fs_meta_aops = {
 387         .writepage      = f2fs_write_meta_page,
 388         .writepages     = f2fs_write_meta_pages,
 389         .set_page_dirty = f2fs_set_meta_page_dirty,
 390         .invalidatepage = f2fs_invalidate_page,
 391         .releasepage    = f2fs_release_page,
 392 #ifdef CONFIG_MIGRATION
 393         .migratepage    = f2fs_migrate_page,
 394 #endif
 395 };
 396
 397 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 398 {
 399         struct inode_management *im = &sbi->im[type];
 400         struct ino_entry *e, *tmp;
 401
 402         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 403 retry:
 404         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 405
 406         spin_lock(&im->ino_lock);
 407         e = radix_tree_lookup(&im->ino_root, ino);
 408         if (!e) {
 409                 e = tmp;
 410                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 411                         spin_unlock(&im->ino_lock);
 412                         radix_tree_preload_end();
 413                         goto retry;
 414                 }
 415                 memset(e, 0, sizeof(struct ino_entry));
 416                 e->ino = ino;
 417
 418                 list_add_tail(&e->list, &im->ino_list);
 419                 if (type != ORPHAN_INO)
 420                         im->ino_num++;
 421         }
 422         spin_unlock(&im->ino_lock);
 423         radix_tree_preload_end();
 424
 425         if (e != tmp)
 426                 kmem_cache_free(ino_entry_slab, tmp);
 427 }
 428
 429 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 430 {
 431         struct inode_management *im = &sbi->im[type];
 432         struct ino_entry *e;
 433
 434         spin_lock(&im->ino_lock);
 435         e = radix_tree_lookup(&im->ino_root, ino);
 436         if (e) {
 437                 list_del(&e->list);
 438                 radix_tree_delete(&im->ino_root, ino);
 439                 im->ino_num--;
 440                 spin_unlock(&im->ino_lock);
 441                 kmem_cache_free(ino_entry_slab, e);
 442                 return;
 443         }
 444         spin_unlock(&im->ino_lock);
 445 }
 446
 447 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 448 {
 449         /* add new dirty ino entry into list */
 450         __add_ino_entry(sbi, ino, type);
 451 }
 452
 453 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 454 {
 455         /* remove dirty ino entry from list */
 456         __remove_ino_entry(sbi, ino, type);
 457 }
 458
 459 /* mode should be APPEND_INO or UPDATE_INO */
 460 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 461 {
 462         struct inode_management *im = &sbi->im[mode];
 463         struct ino_entry *e;
 464
 465         spin_lock(&im->ino_lock);
 466         e = radix_tree_lookup(&im->ino_root, ino);
 467         spin_unlock(&im->ino_lock);
 468         return e ? true : false;
 469 }
 470
 471 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 472 {
 473         struct ino_entry *e, *tmp;
 474         int i;
 475
 476         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 477                 struct inode_management *im = &sbi->im[i];
 478
 479                 spin_lock(&im->ino_lock);
 480                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 481                         list_del(&e->list);
 482                         radix_tree_delete(&im->ino_root, e->ino);
 483                         kmem_cache_free(ino_entry_slab, e);
 484                         im->ino_num--;
 485                 }
 486                 spin_unlock(&im->ino_lock);
 487         }
 488 }
 489
 490 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 491 {
 492         struct inode_management *im = &sbi->im[ORPHAN_INO];
 493         int err = 0;
 494
 495         spin_lock(&im->ino_lock);
 496
 497 #ifdef CONFIG_F2FS_FAULT_INJECTION
 498         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 499                 spin_unlock(&im->ino_lock);
 500                 f2fs_show_injection_info(FAULT_ORPHAN);
 501                 return -ENOSPC;
 502         }
 503 #endif
 504         if (unlikely(im->ino_num >= sbi->max_orphans))
 505                 err = -ENOSPC;
 506         else
 507                 im->ino_num++;
 508         spin_unlock(&im->ino_lock);
 509
 510         return err;
 511 }
 512
 513 void release_orphan_inode(struct f2fs_sb_info *sbi)
 514 {
 515         struct inode_management *im = &sbi->im[ORPHAN_INO];
 516
 517         spin_lock(&im->ino_lock);
 518         f2fs_bug_on(sbi, im->ino_num == 0);
 519         im->ino_num--;
 520         spin_unlock(&im->ino_lock);
 521 }
 522
 523 void add_orphan_inode(struct inode *inode)
 524 {
 525         /* add new orphan ino entry into list */
 526         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
 527         update_inode_page(inode);
 528 }
 529
 530 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 531 {
 532         /* remove orphan entry from orphan list */
 533         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 534 }
 535
 536 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 537 {
 538         struct inode *inode;
 539         struct node_info ni;
 540         int err = acquire_orphan_inode(sbi);
 541
 542         if (err) {
 543                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 544                 f2fs_msg(sbi->sb, KERN_WARNING,
 545                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 546                                 __func__, ino);
 547                 return err;
 548         }
 549
 550         __add_ino_entry(sbi, ino, ORPHAN_INO);
 551
 552         inode = f2fs_iget_retry(sbi->sb, ino);
 553         if (IS_ERR(inode)) {
 554                 /*
 555                  * there should be a bug that we can't find the entry
 556                  * to orphan inode.
 557                  */
 558                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 559                 return PTR_ERR(inode);
 560         }
 561
 562         clear_nlink(inode);
 563
 564         /* truncate all the data during iput */
 565         iput(inode);
 566
 567         get_node_info(sbi, ino, &ni);
 568
 569         /* ENOMEM was fully retried in f2fs_evict_inode. */
 570         if (ni.blk_addr != NULL_ADDR) {
 571                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 572                 f2fs_msg(sbi->sb, KERN_WARNING,
 573                         "%s: orphan failed (ino=%x) by kernel, retry mount.",
 574                                 __func__, ino);
 575                 return -EIO;
 576         }
 577         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 578         return 0;
 579 }
 580
 581 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 582 {
 583         block_t start_blk, orphan_blocks, i, j;
 584         int err;
 585
 586         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 587                 return 0;
 588
 589         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 590         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 591
 592         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 593
 594         for (i = 0; i < orphan_blocks; i++) {
 595                 struct page *page = get_meta_page(sbi, start_blk + i);
 596                 struct f2fs_orphan_block *orphan_blk;
 597
 598                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 599                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 600                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 601                         err = recover_orphan_inode(sbi, ino);
 602                         if (err) {
 603                                 f2fs_put_page(page, 1);
 604                                 return err;
 605                         }
 606                 }
 607                 f2fs_put_page(page, 1);
 608         }
 609         /* clear Orphan Flag */
 610         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 611         return 0;
 612 }
 613
 614 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 615 {
 616         struct list_head *head;
 617         struct f2fs_orphan_block *orphan_blk = NULL;
 618         unsigned int nentries = 0;
 619         unsigned short index = 1;
 620         unsigned short orphan_blocks;
 621         struct page *page = NULL;
 622         struct ino_entry *orphan = NULL;
 623         struct inode_management *im = &sbi->im[ORPHAN_INO];
 624
 625         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 626
 627         /*
 628          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 629          * orphan inode operations are covered under f2fs_lock_op().
 630          * And, spin_lock should be avoided due to page operations below.
 631          */
 632         head = &im->ino_list;
 633
 634         /* loop for each orphan inode entry and write them in Jornal block */
 635         list_for_each_entry(orphan, head, list) {
 636                 if (!page) {
 637                         page = grab_meta_page(sbi, start_blk++);
 638                         orphan_blk =
 639                                 (struct f2fs_orphan_block *)page_address(page);
 640                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 641                 }
 642
 643                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 644
 645                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 646                         /*
 647                          * an orphan block is full of 1020 entries,
 648                          * then we need to flush current orphan blocks
 649                          * and bring another one in memory
 650                          */
 651                         orphan_blk->blk_addr = cpu_to_le16(index);
 652                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 653                         orphan_blk->entry_count = cpu_to_le32(nentries);
 654                         set_page_dirty(page);
 655                         f2fs_put_page(page, 1);
 656                         index++;
 657                         nentries = 0;
 658                         page = NULL;
 659                 }
 660         }
 661
 662         if (page) {
 663                 orphan_blk->blk_addr = cpu_to_le16(index);
 664                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 665                 orphan_blk->entry_count = cpu_to_le32(nentries);
 666                 set_page_dirty(page);
 667                 f2fs_put_page(page, 1);
 668         }
 669 }
 670
 671 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 672                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 673                 unsigned long long *version)
 674 {
 675         unsigned long blk_size = sbi->blocksize;
 676         size_t crc_offset = 0;
 677         __u32 crc = 0;
 678
 679         *cp_page = get_meta_page(sbi, cp_addr);
 680         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 681
 682         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 683         if (crc_offset > (blk_size - sizeof(__le32))) {
 684                 f2fs_msg(sbi->sb, KERN_WARNING,
 685                         "invalid crc_offset: %zu", crc_offset);
 686                 return -EINVAL;
 687         }
 688
 689         crc = cur_cp_crc(*cp_block);
 690         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 691                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 692                 return -EINVAL;
 693         }
 694
 695         *version = cur_cp_version(*cp_block);
 696         return 0;
 697 }
 698
 699 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 700                                 block_t cp_addr, unsigned long long *version)
 701 {
 702         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 703         struct f2fs_checkpoint *cp_block = NULL;
 704         unsigned long long cur_version = 0, pre_version = 0;
 705         int err;
 706
 707         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 708                                         &cp_page_1, version);
 709         if (err)
 710                 goto invalid_cp1;
 711         pre_version = *version;
 712
 713         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 714         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 715                                         &cp_page_2, version);
 716         if (err)
 717                 goto invalid_cp2;
 718         cur_version = *version;
 719
 720         if (cur_version == pre_version) {
 721                 *version = cur_version;
 722                 f2fs_put_page(cp_page_2, 1);
 723                 return cp_page_1;
 724         }
 725 invalid_cp2:
 726         f2fs_put_page(cp_page_2, 1);
 727 invalid_cp1:
 728         f2fs_put_page(cp_page_1, 1);
 729         return NULL;
 730 }
 731
 732 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 733 {
 734         struct f2fs_checkpoint *cp_block;
 735         struct f2fs_super_block *fsb = sbi->raw_super;
 736         struct page *cp1, *cp2, *cur_page;
 737         unsigned long blk_size = sbi->blocksize;
 738         unsigned long long cp1_version = 0, cp2_version = 0;
 739         unsigned long long cp_start_blk_no;
 740         unsigned int cp_blks = 1 + __cp_payload(sbi);
 741         block_t cp_blk_no;
 742         int i;
 743
 744         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 745         if (!sbi->ckpt)
 746                 return -ENOMEM;
 747         /*
 748          * Finding out valid cp block involves read both
 749          * sets( cp pack1 and cp pack 2)
 750          */
 751         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 752         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 753
 754         /* The second checkpoint pack should start at the next segment */
 755         cp_start_blk_no += ((unsigned long long)1) <<
 756                                 le32_to_cpu(fsb->log_blocks_per_seg);
 757         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 758
 759         if (cp1 && cp2) {
 760                 if (ver_after(cp2_version, cp1_version))
 761                         cur_page = cp2;
 762                 else
 763                         cur_page = cp1;
 764         } else if (cp1) {
 765                 cur_page = cp1;
 766         } else if (cp2) {
 767                 cur_page = cp2;
 768         } else {
 769                 goto fail_no_cp;
 770         }
 771
 772         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 773         memcpy(sbi->ckpt, cp_block, blk_size);
 774
 775         /* Sanity checking of checkpoint */
 776         if (sanity_check_ckpt(sbi))
 777                 goto free_fail_no_cp;
 778
 779         if (cur_page == cp1)
 780                 sbi->cur_cp_pack = 1;
 781         else
 782                 sbi->cur_cp_pack = 2;
 783
 784         if (cp_blks <= 1)
 785                 goto done;
 786
 787         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 788         if (cur_page == cp2)
 789                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 790
 791         for (i = 1; i < cp_blks; i++) {
 792                 void *sit_bitmap_ptr;
 793                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 794
 795                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 796                 sit_bitmap_ptr = page_address(cur_page);
 797                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 798                 f2fs_put_page(cur_page, 1);
 799         }
 800 done:
 801         f2fs_put_page(cp1, 1);
 802         f2fs_put_page(cp2, 1);
 803         return 0;
 804
 805 free_fail_no_cp:
 806         f2fs_put_page(cp1, 1);
 807         f2fs_put_page(cp2, 1);
 808 fail_no_cp:
 809         kfree(sbi->ckpt);
 810         return -EINVAL;
 811 }
 812
 813 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 814 {
 815         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 816         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 817
 818         if (is_inode_flag_set(inode, flag))
 819                 return;
 820
 821         set_inode_flag(inode, flag);
 822         if (!f2fs_is_volatile_file(inode))
 823                 list_add_tail(&F2FS_I(inode)->dirty_list,
 824                                                 &sbi->inode_list[type]);
 825         stat_inc_dirty_inode(sbi, type);
 826 }
 827
 828 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 829 {
 830         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 831
 832         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 833                 return;
 834
 835         list_del_init(&F2FS_I(inode)->dirty_list);
 836         clear_inode_flag(inode, flag);
 837         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 838 }
 839
 840 void update_dirty_page(struct inode *inode, struct page *page)
 841 {
 842         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 843         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 844
 845         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 846                         !S_ISLNK(inode->i_mode))
 847                 return;
 848
 849         spin_lock(&sbi->inode_lock[type]);
 850         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 851                 __add_dirty_inode(inode, type);
 852         inode_inc_dirty_pages(inode);
 853         spin_unlock(&sbi->inode_lock[type]);
 854
 855         SetPagePrivate(page);
 856         f2fs_trace_pid(page);
 857 }
 858
 859 void remove_dirty_inode(struct inode *inode)
 860 {
 861         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 862         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 863
 864         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 865                         !S_ISLNK(inode->i_mode))
 866                 return;
 867
 868         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 869                 return;
 870
 871         spin_lock(&sbi->inode_lock[type]);
 872         __remove_dirty_inode(inode, type);
 873         spin_unlock(&sbi->inode_lock[type]);
 874 }
 875
 876 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 877 {
 878         struct list_head *head;
 879         struct inode *inode;
 880         struct f2fs_inode_info *fi;
 881         bool is_dir = (type == DIR_INODE);
 882
 883         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 884                                 get_pages(sbi, is_dir ?
 885                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 886 retry:
 887         if (unlikely(f2fs_cp_error(sbi)))
 888                 return -EIO;
 889
 890         spin_lock(&sbi->inode_lock[type]);
 891
 892         head = &sbi->inode_list[type];
 893         if (list_empty(head)) {
 894                 spin_unlock(&sbi->inode_lock[type]);
 895                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 896                                 get_pages(sbi, is_dir ?
 897                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 898                 return 0;
 899         }
 900         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
 901         inode = igrab(&fi->vfs_inode);
 902         spin_unlock(&sbi->inode_lock[type]);
 903         if (inode) {
 904                 filemap_fdatawrite(inode->i_mapping);
 905                 iput(inode);
 906         } else {
 907                 /*
 908                  * We should submit bio, since it exists several
 909                  * wribacking dentry pages in the freeing inode.
 910                  */
 911                 f2fs_submit_merged_write(sbi, DATA);
 912                 cond_resched();
 913         }
 914         goto retry;
 915 }
 916
 917 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 918 {
 919         struct list_head *head = &sbi->inode_list[DIRTY_META];
 920         struct inode *inode;
 921         struct f2fs_inode_info *fi;
 922         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 923
 924         while (total--) {
 925                 if (unlikely(f2fs_cp_error(sbi)))
 926                         return -EIO;
 927
 928                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 929                 if (list_empty(head)) {
 930                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 931                         return 0;
 932                 }
 933                 fi = list_first_entry(head, struct f2fs_inode_info,
 934                                                         gdirty_list);
 935                 inode = igrab(&fi->vfs_inode);
 936                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 937                 if (inode) {
 938                         sync_inode_metadata(inode, 0);
 939
 940                         /* it's on eviction */
 941                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
 942                                 update_inode_page(inode);
 943                         iput(inode);
 944                 }
 945         };
 946         return 0;
 947 }
 948
 949 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
 950 {
 951         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 952         struct f2fs_nm_info *nm_i = NM_I(sbi);
 953         nid_t last_nid = nm_i->next_scan_nid;
 954
 955         next_free_nid(sbi, &last_nid);
 956         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 957         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 958         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 959         ckpt->next_free_nid = cpu_to_le32(last_nid);
 960 }
 961
 962 /*
 963  * Freeze all the FS-operations for checkpoint.
 964  */
 965 static int block_operations(struct f2fs_sb_info *sbi)
 966 {
 967         struct writeback_control wbc = {
 968                 .sync_mode = WB_SYNC_ALL,
 969                 .nr_to_write = LONG_MAX,
 970                 .for_reclaim = 0,
 971         };
 972         struct blk_plug plug;
 973         int err = 0;
 974
 975         blk_start_plug(&plug);
 976
 977 retry_flush_dents:
 978         f2fs_lock_all(sbi);
 979         /* write all the dirty dentry pages */
 980         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 981                 f2fs_unlock_all(sbi);
 982                 err = sync_dirty_inodes(sbi, DIR_INODE);
 983                 if (err)
 984                         goto out;
 985                 cond_resched();
 986                 goto retry_flush_dents;
 987         }
 988
 989         /*
 990          * POR: we should ensure that there are no dirty node pages
 991          * until finishing nat/sit flush. inode->i_blocks can be updated.
 992          */
 993         down_write(&sbi->node_change);
 994
 995         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
 996                 up_write(&sbi->node_change);
 997                 f2fs_unlock_all(sbi);
 998                 err = f2fs_sync_inode_meta(sbi);
 999                 if (err)
1000                         goto out;
1001                 cond_resched();
1002                 goto retry_flush_dents;
1003         }
1004
1005 retry_flush_nodes:
1006         down_write(&sbi->node_write);
1007
1008         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1009                 up_write(&sbi->node_write);
1010                 err = sync_node_pages(sbi, &wbc);
1011                 if (err) {
1012                         up_write(&sbi->node_change);
1013                         f2fs_unlock_all(sbi);
1014                         goto out;
1015                 }
1016                 cond_resched();
1017                 goto retry_flush_nodes;
1018         }
1019
1020         /*
1021          * sbi->node_change is used only for AIO write_begin path which produces
1022          * dirty node blocks and some checkpoint values by block allocation.
1023          */
1024         __prepare_cp_block(sbi);
1025         up_write(&sbi->node_change);
1026 out:
1027         blk_finish_plug(&plug);
1028         return err;
1029 }
1030
1031 static void unblock_operations(struct f2fs_sb_info *sbi)
1032 {
1033         up_write(&sbi->node_write);
1034         f2fs_unlock_all(sbi);
1035 }
1036
1037 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1038 {
1039         DEFINE_WAIT(wait);
1040
1041         for (;;) {
1042                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1043
1044                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1045                         break;
1046
1047                 io_schedule_timeout(5*HZ);
1048         }
1049         finish_wait(&sbi->cp_wait, &wait);
1050 }
1051
1052 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1053 {
1054         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1055         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1056         unsigned long flags;
1057
1058         spin_lock_irqsave(&sbi->cp_lock, flags);
1059
1060         if ((cpc->reason & CP_UMOUNT) &&
1061                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1062                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1063                 disable_nat_bits(sbi, false);
1064
1065         if (cpc->reason & CP_TRIMMED)
1066                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1067
1068         if (cpc->reason & CP_UMOUNT)
1069                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1070         else
1071                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1072
1073         if (cpc->reason & CP_FASTBOOT)
1074                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1075         else
1076                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1077
1078         if (orphan_num)
1079                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1080         else
1081                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1082
1083         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1084                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1085
1086         /* set this flag to activate crc|cp_ver for recovery */
1087         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1088
1089         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1090 }
1091
1092 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1093 {
1094         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1095         struct f2fs_nm_info *nm_i = NM_I(sbi);
1096         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1097         block_t start_blk;
1098         unsigned int data_sum_blocks, orphan_blocks;
1099         __u32 crc32 = 0;
1100         int i;
1101         int cp_payload_blks = __cp_payload(sbi);
1102         struct super_block *sb = sbi->sb;
1103         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1104         u64 kbytes_written;
1105
1106         /* Flush all the NAT/SIT pages */
1107         while (get_pages(sbi, F2FS_DIRTY_META)) {
1108                 sync_meta_pages(sbi, META, LONG_MAX);
1109                 if (unlikely(f2fs_cp_error(sbi)))
1110                         return -EIO;
1111         }
1112
1113         /*
1114          * modify checkpoint
1115          * version number is already updated
1116          */
1117         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1118         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1119         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1120                 ckpt->cur_node_segno[i] =
1121                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1122                 ckpt->cur_node_blkoff[i] =
1123                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1124                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1125                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1126         }
1127         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1128                 ckpt->cur_data_segno[i] =
1129                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1130                 ckpt->cur_data_blkoff[i] =
1131                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1132                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1133                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1134         }
1135
1136         /* 2 cp  + n data seg summary + orphan inode blocks */
1137         data_sum_blocks = npages_for_summary_flush(sbi, false);
1138         spin_lock_irqsave(&sbi->cp_lock, flags);
1139         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1140                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1141         else
1142                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1143         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1144
1145         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1146         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1147                         orphan_blocks);
1148
1149         if (__remain_node_summaries(cpc->reason))
1150                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1151                                 cp_payload_blks + data_sum_blocks +
1152                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1153         else
1154                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1155                                 cp_payload_blks + data_sum_blocks +
1156                                 orphan_blocks);
1157
1158         /* update ckpt flag for checkpoint */
1159         update_ckpt_flags(sbi, cpc);
1160
1161         /* update SIT/NAT bitmap */
1162         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1163         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1164
1165         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1166         *((__le32 *)((unsigned char *)ckpt +
1167                                 le32_to_cpu(ckpt->checksum_offset)))
1168                                 = cpu_to_le32(crc32);
1169
1170         start_blk = __start_cp_next_addr(sbi);
1171
1172         /* write nat bits */
1173         if (enabled_nat_bits(sbi, cpc)) {
1174                 __u64 cp_ver = cur_cp_version(ckpt);
1175                 block_t blk;
1176
1177                 cp_ver |= ((__u64)crc32 << 32);
1178                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1179
1180                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1181                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1182                         update_meta_page(sbi, nm_i->nat_bits +
1183                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1184
1185                 /* Flush all the NAT BITS pages */
1186                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1187                         sync_meta_pages(sbi, META, LONG_MAX);
1188                         if (unlikely(f2fs_cp_error(sbi)))
1189                                 return -EIO;
1190                 }
1191         }
1192
1193         /* need to wait for end_io results */
1194         wait_on_all_pages_writeback(sbi);
1195         if (unlikely(f2fs_cp_error(sbi)))
1196                 return -EIO;
1197
1198         /* write out checkpoint buffer at block 0 */
1199         update_meta_page(sbi, ckpt, start_blk++);
1200
1201         for (i = 1; i < 1 + cp_payload_blks; i++)
1202                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1203                                                         start_blk++);
1204
1205         if (orphan_num) {
1206                 write_orphan_inodes(sbi, start_blk);
1207                 start_blk += orphan_blocks;
1208         }
1209
1210         write_data_summaries(sbi, start_blk);
1211         start_blk += data_sum_blocks;
1212
1213         /* Record write statistics in the hot node summary */
1214         kbytes_written = sbi->kbytes_written;
1215         if (sb->s_bdev->bd_part)
1216                 kbytes_written += BD_PART_WRITTEN(sbi);
1217
1218         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1219
1220         if (__remain_node_summaries(cpc->reason)) {
1221                 write_node_summaries(sbi, start_blk);
1222                 start_blk += NR_CURSEG_NODE_TYPE;
1223         }
1224
1225         /* writeout checkpoint block */
1226         update_meta_page(sbi, ckpt, start_blk);
1227
1228         /* wait for previous submitted node/meta pages writeback */
1229         wait_on_all_pages_writeback(sbi);
1230
1231         if (unlikely(f2fs_cp_error(sbi)))
1232                 return -EIO;
1233
1234         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1235         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1236
1237         /* update user_block_counts */
1238         sbi->last_valid_block_count = sbi->total_valid_block_count;
1239         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1240
1241         /* Here, we only have one bio having CP pack */
1242         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1243
1244         /* wait for previous submitted meta pages writeback */
1245         wait_on_all_pages_writeback(sbi);
1246
1247         release_ino_entry(sbi, false);
1248
1249         if (unlikely(f2fs_cp_error(sbi)))
1250                 return -EIO;
1251
1252         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1253         clear_sbi_flag(sbi, SBI_NEED_CP);
1254         __set_cp_next_pack(sbi);
1255
1256         /*
1257          * redirty superblock if metadata like node page or inode cache is
1258          * updated during writing checkpoint.
1259          */
1260         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1261                         get_pages(sbi, F2FS_DIRTY_IMETA))
1262                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1263
1264         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1265
1266         return 0;
1267 }
1268
1269 /*
1270  * We guarantee that this checkpoint procedure will not fail.
1271  */
1272 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1273 {
1274         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1275         unsigned long long ckpt_ver;
1276         int err = 0;
1277
1278         mutex_lock(&sbi->cp_mutex);
1279
1280         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1281                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1282                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1283                 goto out;
1284         if (unlikely(f2fs_cp_error(sbi))) {
1285                 err = -EIO;
1286                 goto out;
1287         }
1288         if (f2fs_readonly(sbi->sb)) {
1289                 err = -EROFS;
1290                 goto out;
1291         }
1292
1293         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1294
1295         err = block_operations(sbi);
1296         if (err)
1297                 goto out;
1298
1299         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1300
1301         f2fs_flush_merged_writes(sbi);
1302
1303         /* this is the case of multiple fstrims without any changes */
1304         if (cpc->reason & CP_DISCARD) {
1305                 if (!exist_trim_candidates(sbi, cpc)) {
1306                         unblock_operations(sbi);
1307                         goto out;
1308                 }
1309
1310                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1311                                 SIT_I(sbi)->dirty_sentries == 0 &&
1312                                 prefree_segments(sbi) == 0) {
1313                         flush_sit_entries(sbi, cpc);
1314                         clear_prefree_segments(sbi, cpc);
1315                         unblock_operations(sbi);
1316                         goto out;
1317                 }
1318         }
1319
1320         /*
1321          * update checkpoint pack index
1322          * Increase the version number so that
1323          * SIT entries and seg summaries are written at correct place
1324          */
1325         ckpt_ver = cur_cp_version(ckpt);
1326         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1327
1328         /* write cached NAT/SIT entries to NAT/SIT area */
1329         flush_nat_entries(sbi, cpc);
1330         flush_sit_entries(sbi, cpc);
1331
1332         /* unlock all the fs_lock[] in do_checkpoint() */
1333         err = do_checkpoint(sbi, cpc);
1334         if (err)
1335                 release_discard_addrs(sbi);
1336         else
1337                 clear_prefree_segments(sbi, cpc);
1338
1339         unblock_operations(sbi);
1340         stat_inc_cp_count(sbi->stat_info);
1341
1342         if (cpc->reason & CP_RECOVERY)
1343                 f2fs_msg(sbi->sb, KERN_NOTICE,
1344                         "checkpoint: version = %llx", ckpt_ver);
1345
1346         /* do checkpoint periodically */
1347         f2fs_update_time(sbi, CP_TIME);
1348         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1349 out:
1350         mutex_unlock(&sbi->cp_mutex);
1351         return err;
1352 }
1353
1354 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1355 {
1356         int i;
1357
1358         for (i = 0; i < MAX_INO_ENTRY; i++) {
1359                 struct inode_management *im = &sbi->im[i];
1360
1361                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1362                 spin_lock_init(&im->ino_lock);
1363                 INIT_LIST_HEAD(&im->ino_list);
1364                 im->ino_num = 0;
1365         }
1366
1367         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1368                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1369                                 F2FS_ORPHANS_PER_BLOCK;
1370 }
1371
1372 int __init create_checkpoint_caches(void)
1373 {
1374         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1375                         sizeof(struct ino_entry));
1376         if (!ino_entry_slab)
1377                 return -ENOMEM;
1378         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1379                         sizeof(struct inode_entry));
1380         if (!inode_entry_slab) {
1381                 kmem_cache_destroy(ino_entry_slab);
1382                 return -ENOMEM;
1383         }
1384         return 0;
1385 }
1386
1387 void destroy_checkpoint_caches(void)
1388 {
1389         kmem_cache_destroy(ino_entry_slab);
1390         kmem_cache_destroy(inode_entry_slab);
1391 }