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                                 enum iostat_type io_type)
 236 {
 237         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 238
 239         trace_f2fs_writepage(page, META);
 240
 241         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 242                 goto redirty_out;
 243         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 244                 goto redirty_out;
 245         if (unlikely(f2fs_cp_error(sbi)))
 246                 goto redirty_out;
 247
 248         write_meta_page(sbi, page, io_type);
 249         dec_page_count(sbi, F2FS_DIRTY_META);
 250
 251         if (wbc->for_reclaim)
 252                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
 253                                                 0, page->index, META);
 254
 255         unlock_page(page);
 256
 257         if (unlikely(f2fs_cp_error(sbi)))
 258                 f2fs_submit_merged_write(sbi, META);
 259
 260         return 0;
 261
 262 redirty_out:
 263         redirty_page_for_writepage(wbc, page);
 264         return AOP_WRITEPAGE_ACTIVATE;
 265 }
 266
 267 static int f2fs_write_meta_page(struct page *page,
 268                                 struct writeback_control *wbc)
 269 {
 270         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 271 }
 272
 273 static int f2fs_write_meta_pages(struct address_space *mapping,
 274                                 struct writeback_control *wbc)
 275 {
 276         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 277         long diff, written;
 278
 279         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 280                 goto skip_write;
 281
 282         /* collect a number of dirty meta pages and write together */
 283         if (wbc->for_kupdate ||
 284                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 285                 goto skip_write;
 286
 287         /* if locked failed, cp will flush dirty pages instead */
 288         if (!mutex_trylock(&sbi->cp_mutex))
 289                 goto skip_write;
 290
 291         trace_f2fs_writepages(mapping->host, wbc, META);
 292         diff = nr_pages_to_write(sbi, META, wbc);
 293         written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 294         mutex_unlock(&sbi->cp_mutex);
 295         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 296         return 0;
 297
 298 skip_write:
 299         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 300         trace_f2fs_writepages(mapping->host, wbc, META);
 301         return 0;
 302 }
 303
 304 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 305                                 long nr_to_write, enum iostat_type io_type)
 306 {
 307         struct address_space *mapping = META_MAPPING(sbi);
 308         pgoff_t index = 0, prev = ULONG_MAX;
 309         struct pagevec pvec;
 310         long nwritten = 0;
 311         int nr_pages;
 312         struct writeback_control wbc = {
 313                 .for_reclaim = 0,
 314         };
 315         struct blk_plug plug;
 316
 317         pagevec_init(&pvec);
 318
 319         blk_start_plug(&plug);
 320
 321         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 322                                 PAGECACHE_TAG_DIRTY))) {
 323                 int i;
 324
 325                 for (i = 0; i < nr_pages; i++) {
 326                         struct page *page = pvec.pages[i];
 327
 328                         if (prev == ULONG_MAX)
 329                                 prev = page->index - 1;
 330                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 331                                 pagevec_release(&pvec);
 332                                 goto stop;
 333                         }
 334
 335                         lock_page(page);
 336
 337                         if (unlikely(page->mapping != mapping)) {
 338 continue_unlock:
 339                                 unlock_page(page);
 340                                 continue;
 341                         }
 342                         if (!PageDirty(page)) {
 343                                 /* someone wrote it for us */
 344                                 goto continue_unlock;
 345                         }
 346
 347                         f2fs_wait_on_page_writeback(page, META, true);
 348
 349                         BUG_ON(PageWriteback(page));
 350                         if (!clear_page_dirty_for_io(page))
 351                                 goto continue_unlock;
 352
 353                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 354                                 unlock_page(page);
 355                                 break;
 356                         }
 357                         nwritten++;
 358                         prev = page->index;
 359                         if (unlikely(nwritten >= nr_to_write))
 360                                 break;
 361                 }
 362                 pagevec_release(&pvec);
 363                 cond_resched();
 364         }
 365 stop:
 366         if (nwritten)
 367                 f2fs_submit_merged_write(sbi, type);
 368
 369         blk_finish_plug(&plug);
 370
 371         return nwritten;
 372 }
 373
 374 static int f2fs_set_meta_page_dirty(struct page *page)
 375 {
 376         trace_f2fs_set_page_dirty(page, META);
 377
 378         if (!PageUptodate(page))
 379                 SetPageUptodate(page);
 380         if (!PageDirty(page)) {
 381                 f2fs_set_page_dirty_nobuffers(page);
 382                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 383                 SetPagePrivate(page);
 384                 f2fs_trace_pid(page);
 385                 return 1;
 386         }
 387         return 0;
 388 }
 389
 390 const struct address_space_operations f2fs_meta_aops = {
 391         .writepage      = f2fs_write_meta_page,
 392         .writepages     = f2fs_write_meta_pages,
 393         .set_page_dirty = f2fs_set_meta_page_dirty,
 394         .invalidatepage = f2fs_invalidate_page,
 395         .releasepage    = f2fs_release_page,
 396 #ifdef CONFIG_MIGRATION
 397         .migratepage    = f2fs_migrate_page,
 398 #endif
 399 };
 400
 401 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 402 {
 403         struct inode_management *im = &sbi->im[type];
 404         struct ino_entry *e, *tmp;
 405
 406         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 407 retry:
 408         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 409
 410         spin_lock(&im->ino_lock);
 411         e = radix_tree_lookup(&im->ino_root, ino);
 412         if (!e) {
 413                 e = tmp;
 414                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 415                         spin_unlock(&im->ino_lock);
 416                         radix_tree_preload_end();
 417                         goto retry;
 418                 }
 419                 memset(e, 0, sizeof(struct ino_entry));
 420                 e->ino = ino;
 421
 422                 list_add_tail(&e->list, &im->ino_list);
 423                 if (type != ORPHAN_INO)
 424                         im->ino_num++;
 425         }
 426         spin_unlock(&im->ino_lock);
 427         radix_tree_preload_end();
 428
 429         if (e != tmp)
 430                 kmem_cache_free(ino_entry_slab, tmp);
 431 }
 432
 433 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 434 {
 435         struct inode_management *im = &sbi->im[type];
 436         struct ino_entry *e;
 437
 438         spin_lock(&im->ino_lock);
 439         e = radix_tree_lookup(&im->ino_root, ino);
 440         if (e) {
 441                 list_del(&e->list);
 442                 radix_tree_delete(&im->ino_root, ino);
 443                 im->ino_num--;
 444                 spin_unlock(&im->ino_lock);
 445                 kmem_cache_free(ino_entry_slab, e);
 446                 return;
 447         }
 448         spin_unlock(&im->ino_lock);
 449 }
 450
 451 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 452 {
 453         /* add new dirty ino entry into list */
 454         __add_ino_entry(sbi, ino, type);
 455 }
 456
 457 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 458 {
 459         /* remove dirty ino entry from list */
 460         __remove_ino_entry(sbi, ino, type);
 461 }
 462
 463 /* mode should be APPEND_INO or UPDATE_INO */
 464 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 465 {
 466         struct inode_management *im = &sbi->im[mode];
 467         struct ino_entry *e;
 468
 469         spin_lock(&im->ino_lock);
 470         e = radix_tree_lookup(&im->ino_root, ino);
 471         spin_unlock(&im->ino_lock);
 472         return e ? true : false;
 473 }
 474
 475 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 476 {
 477         struct ino_entry *e, *tmp;
 478         int i;
 479
 480         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 481                 struct inode_management *im = &sbi->im[i];
 482
 483                 spin_lock(&im->ino_lock);
 484                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 485                         list_del(&e->list);
 486                         radix_tree_delete(&im->ino_root, e->ino);
 487                         kmem_cache_free(ino_entry_slab, e);
 488                         im->ino_num--;
 489                 }
 490                 spin_unlock(&im->ino_lock);
 491         }
 492 }
 493
 494 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 495 {
 496         struct inode_management *im = &sbi->im[ORPHAN_INO];
 497         int err = 0;
 498
 499         spin_lock(&im->ino_lock);
 500
 501 #ifdef CONFIG_F2FS_FAULT_INJECTION
 502         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 503                 spin_unlock(&im->ino_lock);
 504                 f2fs_show_injection_info(FAULT_ORPHAN);
 505                 return -ENOSPC;
 506         }
 507 #endif
 508         if (unlikely(im->ino_num >= sbi->max_orphans))
 509                 err = -ENOSPC;
 510         else
 511                 im->ino_num++;
 512         spin_unlock(&im->ino_lock);
 513
 514         return err;
 515 }
 516
 517 void release_orphan_inode(struct f2fs_sb_info *sbi)
 518 {
 519         struct inode_management *im = &sbi->im[ORPHAN_INO];
 520
 521         spin_lock(&im->ino_lock);
 522         f2fs_bug_on(sbi, im->ino_num == 0);
 523         im->ino_num--;
 524         spin_unlock(&im->ino_lock);
 525 }
 526
 527 void add_orphan_inode(struct inode *inode)
 528 {
 529         /* add new orphan ino entry into list */
 530         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
 531         update_inode_page(inode);
 532 }
 533
 534 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 535 {
 536         /* remove orphan entry from orphan list */
 537         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 538 }
 539
 540 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 541 {
 542         struct inode *inode;
 543         struct node_info ni;
 544         int err = acquire_orphan_inode(sbi);
 545
 546         if (err) {
 547                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 548                 f2fs_msg(sbi->sb, KERN_WARNING,
 549                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 550                                 __func__, ino);
 551                 return err;
 552         }
 553
 554         __add_ino_entry(sbi, ino, ORPHAN_INO);
 555
 556         inode = f2fs_iget_retry(sbi->sb, ino);
 557         if (IS_ERR(inode)) {
 558                 /*
 559                  * there should be a bug that we can't find the entry
 560                  * to orphan inode.
 561                  */
 562                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 563                 return PTR_ERR(inode);
 564         }
 565
 566         clear_nlink(inode);
 567
 568         /* truncate all the data during iput */
 569         iput(inode);
 570
 571         get_node_info(sbi, ino, &ni);
 572
 573         /* ENOMEM was fully retried in f2fs_evict_inode. */
 574         if (ni.blk_addr != NULL_ADDR) {
 575                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 576                 f2fs_msg(sbi->sb, KERN_WARNING,
 577                         "%s: orphan failed (ino=%x) by kernel, retry mount.",
 578                                 __func__, ino);
 579                 return -EIO;
 580         }
 581         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 582         return 0;
 583 }
 584
 585 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 586 {
 587         block_t start_blk, orphan_blocks, i, j;
 588         unsigned int s_flags = sbi->sb->s_flags;
 589         int err = 0;
 590
 591         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 592                 return 0;
 593
 594         if (s_flags & MS_RDONLY) {
 595                 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
 596                 sbi->sb->s_flags &= ~MS_RDONLY;
 597         }
 598
 599 #ifdef CONFIG_QUOTA
 600         /* Needed for iput() to work correctly and not trash data */
 601         sbi->sb->s_flags |= MS_ACTIVE;
 602         /* Turn on quotas so that they are updated correctly */
 603         f2fs_enable_quota_files(sbi);
 604 #endif
 605
 606         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 607         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 608
 609         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 610
 611         for (i = 0; i < orphan_blocks; i++) {
 612                 struct page *page = get_meta_page(sbi, start_blk + i);
 613                 struct f2fs_orphan_block *orphan_blk;
 614
 615                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 616                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 617                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 618                         err = recover_orphan_inode(sbi, ino);
 619                         if (err) {
 620                                 f2fs_put_page(page, 1);
 621                                 goto out;
 622                         }
 623                 }
 624                 f2fs_put_page(page, 1);
 625         }
 626         /* clear Orphan Flag */
 627         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 628 out:
 629 #ifdef CONFIG_QUOTA
 630         /* Turn quotas off */
 631         f2fs_quota_off_umount(sbi->sb);
 632 #endif
 633         sbi->sb->s_flags = s_flags; /* Restore MS_RDONLY status */
 634
 635         return err;
 636 }
 637
 638 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 639 {
 640         struct list_head *head;
 641         struct f2fs_orphan_block *orphan_blk = NULL;
 642         unsigned int nentries = 0;
 643         unsigned short index = 1;
 644         unsigned short orphan_blocks;
 645         struct page *page = NULL;
 646         struct ino_entry *orphan = NULL;
 647         struct inode_management *im = &sbi->im[ORPHAN_INO];
 648
 649         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 650
 651         /*
 652          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 653          * orphan inode operations are covered under f2fs_lock_op().
 654          * And, spin_lock should be avoided due to page operations below.
 655          */
 656         head = &im->ino_list;
 657
 658         /* loop for each orphan inode entry and write them in Jornal block */
 659         list_for_each_entry(orphan, head, list) {
 660                 if (!page) {
 661                         page = grab_meta_page(sbi, start_blk++);
 662                         orphan_blk =
 663                                 (struct f2fs_orphan_block *)page_address(page);
 664                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 665                 }
 666
 667                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 668
 669                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 670                         /*
 671                          * an orphan block is full of 1020 entries,
 672                          * then we need to flush current orphan blocks
 673                          * and bring another one in memory
 674                          */
 675                         orphan_blk->blk_addr = cpu_to_le16(index);
 676                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 677                         orphan_blk->entry_count = cpu_to_le32(nentries);
 678                         set_page_dirty(page);
 679                         f2fs_put_page(page, 1);
 680                         index++;
 681                         nentries = 0;
 682                         page = NULL;
 683                 }
 684         }
 685
 686         if (page) {
 687                 orphan_blk->blk_addr = cpu_to_le16(index);
 688                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 689                 orphan_blk->entry_count = cpu_to_le32(nentries);
 690                 set_page_dirty(page);
 691                 f2fs_put_page(page, 1);
 692         }
 693 }
 694
 695 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 696                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 697                 unsigned long long *version)
 698 {
 699         unsigned long blk_size = sbi->blocksize;
 700         size_t crc_offset = 0;
 701         __u32 crc = 0;
 702
 703         *cp_page = get_meta_page(sbi, cp_addr);
 704         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 705
 706         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 707         if (crc_offset > (blk_size - sizeof(__le32))) {
 708                 f2fs_msg(sbi->sb, KERN_WARNING,
 709                         "invalid crc_offset: %zu", crc_offset);
 710                 return -EINVAL;
 711         }
 712
 713         crc = cur_cp_crc(*cp_block);
 714         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 715                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 716                 return -EINVAL;
 717         }
 718
 719         *version = cur_cp_version(*cp_block);
 720         return 0;
 721 }
 722
 723 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 724                                 block_t cp_addr, unsigned long long *version)
 725 {
 726         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 727         struct f2fs_checkpoint *cp_block = NULL;
 728         unsigned long long cur_version = 0, pre_version = 0;
 729         int err;
 730
 731         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 732                                         &cp_page_1, version);
 733         if (err)
 734                 goto invalid_cp1;
 735         pre_version = *version;
 736
 737         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 738         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 739                                         &cp_page_2, version);
 740         if (err)
 741                 goto invalid_cp2;
 742         cur_version = *version;
 743
 744         if (cur_version == pre_version) {
 745                 *version = cur_version;
 746                 f2fs_put_page(cp_page_2, 1);
 747                 return cp_page_1;
 748         }
 749 invalid_cp2:
 750         f2fs_put_page(cp_page_2, 1);
 751 invalid_cp1:
 752         f2fs_put_page(cp_page_1, 1);
 753         return NULL;
 754 }
 755
 756 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 757 {
 758         struct f2fs_checkpoint *cp_block;
 759         struct f2fs_super_block *fsb = sbi->raw_super;
 760         struct page *cp1, *cp2, *cur_page;
 761         unsigned long blk_size = sbi->blocksize;
 762         unsigned long long cp1_version = 0, cp2_version = 0;
 763         unsigned long long cp_start_blk_no;
 764         unsigned int cp_blks = 1 + __cp_payload(sbi);
 765         block_t cp_blk_no;
 766         int i;
 767
 768         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 769         if (!sbi->ckpt)
 770                 return -ENOMEM;
 771         /*
 772          * Finding out valid cp block involves read both
 773          * sets( cp pack1 and cp pack 2)
 774          */
 775         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 776         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 777
 778         /* The second checkpoint pack should start at the next segment */
 779         cp_start_blk_no += ((unsigned long long)1) <<
 780                                 le32_to_cpu(fsb->log_blocks_per_seg);
 781         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 782
 783         if (cp1 && cp2) {
 784                 if (ver_after(cp2_version, cp1_version))
 785                         cur_page = cp2;
 786                 else
 787                         cur_page = cp1;
 788         } else if (cp1) {
 789                 cur_page = cp1;
 790         } else if (cp2) {
 791                 cur_page = cp2;
 792         } else {
 793                 goto fail_no_cp;
 794         }
 795
 796         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 797         memcpy(sbi->ckpt, cp_block, blk_size);
 798
 799         /* Sanity checking of checkpoint */
 800         if (sanity_check_ckpt(sbi))
 801                 goto free_fail_no_cp;
 802
 803         if (cur_page == cp1)
 804                 sbi->cur_cp_pack = 1;
 805         else
 806                 sbi->cur_cp_pack = 2;
 807
 808         if (cp_blks <= 1)
 809                 goto done;
 810
 811         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 812         if (cur_page == cp2)
 813                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 814
 815         for (i = 1; i < cp_blks; i++) {
 816                 void *sit_bitmap_ptr;
 817                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 818
 819                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 820                 sit_bitmap_ptr = page_address(cur_page);
 821                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 822                 f2fs_put_page(cur_page, 1);
 823         }
 824 done:
 825         f2fs_put_page(cp1, 1);
 826         f2fs_put_page(cp2, 1);
 827         return 0;
 828
 829 free_fail_no_cp:
 830         f2fs_put_page(cp1, 1);
 831         f2fs_put_page(cp2, 1);
 832 fail_no_cp:
 833         kfree(sbi->ckpt);
 834         return -EINVAL;
 835 }
 836
 837 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 838 {
 839         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 840         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 841
 842         if (is_inode_flag_set(inode, flag))
 843                 return;
 844
 845         set_inode_flag(inode, flag);
 846         if (!f2fs_is_volatile_file(inode))
 847                 list_add_tail(&F2FS_I(inode)->dirty_list,
 848                                                 &sbi->inode_list[type]);
 849         stat_inc_dirty_inode(sbi, type);
 850 }
 851
 852 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 853 {
 854         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 855
 856         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 857                 return;
 858
 859         list_del_init(&F2FS_I(inode)->dirty_list);
 860         clear_inode_flag(inode, flag);
 861         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 862 }
 863
 864 void update_dirty_page(struct inode *inode, struct page *page)
 865 {
 866         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 867         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 868
 869         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 870                         !S_ISLNK(inode->i_mode))
 871                 return;
 872
 873         spin_lock(&sbi->inode_lock[type]);
 874         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 875                 __add_dirty_inode(inode, type);
 876         inode_inc_dirty_pages(inode);
 877         spin_unlock(&sbi->inode_lock[type]);
 878
 879         SetPagePrivate(page);
 880         f2fs_trace_pid(page);
 881 }
 882
 883 void remove_dirty_inode(struct inode *inode)
 884 {
 885         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 886         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 887
 888         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 889                         !S_ISLNK(inode->i_mode))
 890                 return;
 891
 892         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 893                 return;
 894
 895         spin_lock(&sbi->inode_lock[type]);
 896         __remove_dirty_inode(inode, type);
 897         spin_unlock(&sbi->inode_lock[type]);
 898 }
 899
 900 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 901 {
 902         struct list_head *head;
 903         struct inode *inode;
 904         struct f2fs_inode_info *fi;
 905         bool is_dir = (type == DIR_INODE);
 906         unsigned long ino = 0;
 907
 908         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 909                                 get_pages(sbi, is_dir ?
 910                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 911 retry:
 912         if (unlikely(f2fs_cp_error(sbi)))
 913                 return -EIO;
 914
 915         spin_lock(&sbi->inode_lock[type]);
 916
 917         head = &sbi->inode_list[type];
 918         if (list_empty(head)) {
 919                 spin_unlock(&sbi->inode_lock[type]);
 920                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 921                                 get_pages(sbi, is_dir ?
 922                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 923                 return 0;
 924         }
 925         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
 926         inode = igrab(&fi->vfs_inode);
 927         spin_unlock(&sbi->inode_lock[type]);
 928         if (inode) {
 929                 unsigned long cur_ino = inode->i_ino;
 930
 931                 if (is_dir)
 932                         F2FS_I(inode)->cp_task = current;
 933
 934                 filemap_fdatawrite(inode->i_mapping);
 935
 936                 if (is_dir)
 937                         F2FS_I(inode)->cp_task = NULL;
 938
 939                 iput(inode);
 940                 /* We need to give cpu to another writers. */
 941                 if (ino == cur_ino) {
 942                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 943                         cond_resched();
 944                 } else {
 945                         ino = cur_ino;
 946                 }
 947         } else {
 948                 /*
 949                  * We should submit bio, since it exists several
 950                  * wribacking dentry pages in the freeing inode.
 951                  */
 952                 f2fs_submit_merged_write(sbi, DATA);
 953                 cond_resched();
 954         }
 955         goto retry;
 956 }
 957
 958 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 959 {
 960         struct list_head *head = &sbi->inode_list[DIRTY_META];
 961         struct inode *inode;
 962         struct f2fs_inode_info *fi;
 963         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 964
 965         while (total--) {
 966                 if (unlikely(f2fs_cp_error(sbi)))
 967                         return -EIO;
 968
 969                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 970                 if (list_empty(head)) {
 971                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 972                         return 0;
 973                 }
 974                 fi = list_first_entry(head, struct f2fs_inode_info,
 975                                                         gdirty_list);
 976                 inode = igrab(&fi->vfs_inode);
 977                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 978                 if (inode) {
 979                         sync_inode_metadata(inode, 0);
 980
 981                         /* it's on eviction */
 982                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
 983                                 update_inode_page(inode);
 984                         iput(inode);
 985                 }
 986         };
 987         return 0;
 988 }
 989
 990 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
 991 {
 992         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 993         struct f2fs_nm_info *nm_i = NM_I(sbi);
 994         nid_t last_nid = nm_i->next_scan_nid;
 995
 996         next_free_nid(sbi, &last_nid);
 997         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 998         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 999         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1000         ckpt->next_free_nid = cpu_to_le32(last_nid);
1001 }
1002
1003 /*
1004  * Freeze all the FS-operations for checkpoint.
1005  */
1006 static int block_operations(struct f2fs_sb_info *sbi)
1007 {
1008         struct writeback_control wbc = {
1009                 .sync_mode = WB_SYNC_ALL,
1010                 .nr_to_write = LONG_MAX,
1011                 .for_reclaim = 0,
1012         };
1013         struct blk_plug plug;
1014         int err = 0;
1015
1016         blk_start_plug(&plug);
1017
1018 retry_flush_dents:
1019         f2fs_lock_all(sbi);
1020         /* write all the dirty dentry pages */
1021         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1022                 f2fs_unlock_all(sbi);
1023                 err = sync_dirty_inodes(sbi, DIR_INODE);
1024                 if (err)
1025                         goto out;
1026                 cond_resched();
1027                 goto retry_flush_dents;
1028         }
1029
1030         /*
1031          * POR: we should ensure that there are no dirty node pages
1032          * until finishing nat/sit flush. inode->i_blocks can be updated.
1033          */
1034         down_write(&sbi->node_change);
1035
1036         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1037                 up_write(&sbi->node_change);
1038                 f2fs_unlock_all(sbi);
1039                 err = f2fs_sync_inode_meta(sbi);
1040                 if (err)
1041                         goto out;
1042                 cond_resched();
1043                 goto retry_flush_dents;
1044         }
1045
1046 retry_flush_nodes:
1047         down_write(&sbi->node_write);
1048
1049         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1050                 up_write(&sbi->node_write);
1051                 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1052                 if (err) {
1053                         up_write(&sbi->node_change);
1054                         f2fs_unlock_all(sbi);
1055                         goto out;
1056                 }
1057                 cond_resched();
1058                 goto retry_flush_nodes;
1059         }
1060
1061         /*
1062          * sbi->node_change is used only for AIO write_begin path which produces
1063          * dirty node blocks and some checkpoint values by block allocation.
1064          */
1065         __prepare_cp_block(sbi);
1066         up_write(&sbi->node_change);
1067 out:
1068         blk_finish_plug(&plug);
1069         return err;
1070 }
1071
1072 static void unblock_operations(struct f2fs_sb_info *sbi)
1073 {
1074         up_write(&sbi->node_write);
1075         f2fs_unlock_all(sbi);
1076 }
1077
1078 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1079 {
1080         DEFINE_WAIT(wait);
1081
1082         for (;;) {
1083                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1084
1085                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1086                         break;
1087
1088                 io_schedule_timeout(5*HZ);
1089         }
1090         finish_wait(&sbi->cp_wait, &wait);
1091 }
1092
1093 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1094 {
1095         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1096         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1097         unsigned long flags;
1098
1099         spin_lock_irqsave(&sbi->cp_lock, flags);
1100
1101         if ((cpc->reason & CP_UMOUNT) &&
1102                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1103                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1104                 disable_nat_bits(sbi, false);
1105
1106         if (cpc->reason & CP_TRIMMED)
1107                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1108
1109         if (cpc->reason & CP_UMOUNT)
1110                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1111         else
1112                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1113
1114         if (cpc->reason & CP_FASTBOOT)
1115                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1116         else
1117                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1118
1119         if (orphan_num)
1120                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1121         else
1122                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1123
1124         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1125                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1126
1127         /* set this flag to activate crc|cp_ver for recovery */
1128         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1129
1130         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1131 }
1132
1133 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1134 {
1135         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1136         struct f2fs_nm_info *nm_i = NM_I(sbi);
1137         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1138         block_t start_blk;
1139         unsigned int data_sum_blocks, orphan_blocks;
1140         __u32 crc32 = 0;
1141         int i;
1142         int cp_payload_blks = __cp_payload(sbi);
1143         struct super_block *sb = sbi->sb;
1144         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1145         u64 kbytes_written;
1146
1147         /* Flush all the NAT/SIT pages */
1148         while (get_pages(sbi, F2FS_DIRTY_META)) {
1149                 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1150                 if (unlikely(f2fs_cp_error(sbi)))
1151                         return -EIO;
1152         }
1153
1154         /*
1155          * modify checkpoint
1156          * version number is already updated
1157          */
1158         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1159         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1160         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1161                 ckpt->cur_node_segno[i] =
1162                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1163                 ckpt->cur_node_blkoff[i] =
1164                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1165                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1166                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1167         }
1168         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1169                 ckpt->cur_data_segno[i] =
1170                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1171                 ckpt->cur_data_blkoff[i] =
1172                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1173                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1174                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1175         }
1176
1177         /* 2 cp  + n data seg summary + orphan inode blocks */
1178         data_sum_blocks = npages_for_summary_flush(sbi, false);
1179         spin_lock_irqsave(&sbi->cp_lock, flags);
1180         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1181                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1182         else
1183                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1184         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1185
1186         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1187         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1188                         orphan_blocks);
1189
1190         if (__remain_node_summaries(cpc->reason))
1191                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1192                                 cp_payload_blks + data_sum_blocks +
1193                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1194         else
1195                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1196                                 cp_payload_blks + data_sum_blocks +
1197                                 orphan_blocks);
1198
1199         /* update ckpt flag for checkpoint */
1200         update_ckpt_flags(sbi, cpc);
1201
1202         /* update SIT/NAT bitmap */
1203         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1204         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1205
1206         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1207         *((__le32 *)((unsigned char *)ckpt +
1208                                 le32_to_cpu(ckpt->checksum_offset)))
1209                                 = cpu_to_le32(crc32);
1210
1211         start_blk = __start_cp_next_addr(sbi);
1212
1213         /* write nat bits */
1214         if (enabled_nat_bits(sbi, cpc)) {
1215                 __u64 cp_ver = cur_cp_version(ckpt);
1216                 block_t blk;
1217
1218                 cp_ver |= ((__u64)crc32 << 32);
1219                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1220
1221                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1222                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1223                         update_meta_page(sbi, nm_i->nat_bits +
1224                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1225
1226                 /* Flush all the NAT BITS pages */
1227                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1228                         sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1229                         if (unlikely(f2fs_cp_error(sbi)))
1230                                 return -EIO;
1231                 }
1232         }
1233
1234         /* need to wait for end_io results */
1235         wait_on_all_pages_writeback(sbi);
1236         if (unlikely(f2fs_cp_error(sbi)))
1237                 return -EIO;
1238
1239         /* write out checkpoint buffer at block 0 */
1240         update_meta_page(sbi, ckpt, start_blk++);
1241
1242         for (i = 1; i < 1 + cp_payload_blks; i++)
1243                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1244                                                         start_blk++);
1245
1246         if (orphan_num) {
1247                 write_orphan_inodes(sbi, start_blk);
1248                 start_blk += orphan_blocks;
1249         }
1250
1251         write_data_summaries(sbi, start_blk);
1252         start_blk += data_sum_blocks;
1253
1254         /* Record write statistics in the hot node summary */
1255         kbytes_written = sbi->kbytes_written;
1256         if (sb->s_bdev->bd_part)
1257                 kbytes_written += BD_PART_WRITTEN(sbi);
1258
1259         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1260
1261         if (__remain_node_summaries(cpc->reason)) {
1262                 write_node_summaries(sbi, start_blk);
1263                 start_blk += NR_CURSEG_NODE_TYPE;
1264         }
1265
1266         /* writeout checkpoint block */
1267         update_meta_page(sbi, ckpt, start_blk);
1268
1269         /* wait for previous submitted node/meta pages writeback */
1270         wait_on_all_pages_writeback(sbi);
1271
1272         if (unlikely(f2fs_cp_error(sbi)))
1273                 return -EIO;
1274
1275         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1276         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1277
1278         /* update user_block_counts */
1279         sbi->last_valid_block_count = sbi->total_valid_block_count;
1280         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1281
1282         /* Here, we only have one bio having CP pack */
1283         sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1284
1285         /* wait for previous submitted meta pages writeback */
1286         wait_on_all_pages_writeback(sbi);
1287
1288         release_ino_entry(sbi, false);
1289
1290         if (unlikely(f2fs_cp_error(sbi)))
1291                 return -EIO;
1292
1293         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1294         clear_sbi_flag(sbi, SBI_NEED_CP);
1295         __set_cp_next_pack(sbi);
1296
1297         /*
1298          * redirty superblock if metadata like node page or inode cache is
1299          * updated during writing checkpoint.
1300          */
1301         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1302                         get_pages(sbi, F2FS_DIRTY_IMETA))
1303                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1304
1305         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1306
1307         return 0;
1308 }
1309
1310 /*
1311  * We guarantee that this checkpoint procedure will not fail.
1312  */
1313 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1314 {
1315         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1316         unsigned long long ckpt_ver;
1317         int err = 0;
1318
1319         mutex_lock(&sbi->cp_mutex);
1320
1321         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1322                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1323                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1324                 goto out;
1325         if (unlikely(f2fs_cp_error(sbi))) {
1326                 err = -EIO;
1327                 goto out;
1328         }
1329         if (f2fs_readonly(sbi->sb)) {
1330                 err = -EROFS;
1331                 goto out;
1332         }
1333
1334         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1335
1336         err = block_operations(sbi);
1337         if (err)
1338                 goto out;
1339
1340         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1341
1342         f2fs_flush_merged_writes(sbi);
1343
1344         /* this is the case of multiple fstrims without any changes */
1345         if (cpc->reason & CP_DISCARD) {
1346                 if (!exist_trim_candidates(sbi, cpc)) {
1347                         unblock_operations(sbi);
1348                         goto out;
1349                 }
1350
1351                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1352                                 SIT_I(sbi)->dirty_sentries == 0 &&
1353                                 prefree_segments(sbi) == 0) {
1354                         flush_sit_entries(sbi, cpc);
1355                         clear_prefree_segments(sbi, cpc);
1356                         unblock_operations(sbi);
1357                         goto out;
1358                 }
1359         }
1360
1361         /*
1362          * update checkpoint pack index
1363          * Increase the version number so that
1364          * SIT entries and seg summaries are written at correct place
1365          */
1366         ckpt_ver = cur_cp_version(ckpt);
1367         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1368
1369         /* write cached NAT/SIT entries to NAT/SIT area */
1370         flush_nat_entries(sbi, cpc);
1371         flush_sit_entries(sbi, cpc);
1372
1373         /* unlock all the fs_lock[] in do_checkpoint() */
1374         err = do_checkpoint(sbi, cpc);
1375         if (err)
1376                 release_discard_addrs(sbi);
1377         else
1378                 clear_prefree_segments(sbi, cpc);
1379
1380         unblock_operations(sbi);
1381         stat_inc_cp_count(sbi->stat_info);
1382
1383         if (cpc->reason & CP_RECOVERY)
1384                 f2fs_msg(sbi->sb, KERN_NOTICE,
1385                         "checkpoint: version = %llx", ckpt_ver);
1386
1387         /* do checkpoint periodically */
1388         f2fs_update_time(sbi, CP_TIME);
1389         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1390 out:
1391         mutex_unlock(&sbi->cp_mutex);
1392         return err;
1393 }
1394
1395 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1396 {
1397         int i;
1398
1399         for (i = 0; i < MAX_INO_ENTRY; i++) {
1400                 struct inode_management *im = &sbi->im[i];
1401
1402                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1403                 spin_lock_init(&im->ino_lock);
1404                 INIT_LIST_HEAD(&im->ino_list);
1405                 im->ino_num = 0;
1406         }
1407
1408         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1409                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1410                                 F2FS_ORPHANS_PER_BLOCK;
1411 }
1412
1413 int __init create_checkpoint_caches(void)
1414 {
1415         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1416                         sizeof(struct ino_entry));
1417         if (!ino_entry_slab)
1418                 return -ENOMEM;
1419         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1420                         sizeof(struct inode_entry));
1421         if (!inode_entry_slab) {
1422                 kmem_cache_destroy(ino_entry_slab);
1423                 return -ENOMEM;
1424         }
1425         return 0;
1426 }
1427
1428 void destroy_checkpoint_caches(void)
1429 {
1430         kmem_cache_destroy(ino_entry_slab);
1431         kmem_cache_destroy(inode_entry_slab);
1432 }