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 *f2fs_inode_entry_slab;
  28
  29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         f2fs_build_fault_attr(sbi, 0, 0);
  32         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  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 *f2fs_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                 .is_meta = is_meta,
  73         };
  74         int err;
  75
  76         if (unlikely(!is_meta))
  77                 fio.op_flags &= ~REQ_META;
  78 repeat:
  79         page = f2fs_grab_cache_page(mapping, index, false);
  80         if (!page) {
  81                 cond_resched();
  82                 goto repeat;
  83         }
  84         if (PageUptodate(page))
  85                 goto out;
  86
  87         fio.page = page;
  88
  89         err = f2fs_submit_page_bio(&fio);
  90         if (err) {
  91                 f2fs_put_page(page, 1);
  92                 return ERR_PTR(err);
  93         }
  94
  95         lock_page(page);
  96         if (unlikely(page->mapping != mapping)) {
  97                 f2fs_put_page(page, 1);
  98                 goto repeat;
  99         }
 100
 101         if (unlikely(!PageUptodate(page))) {
 102                 f2fs_put_page(page, 1);
 103                 return ERR_PTR(-EIO);
 104         }
 105 out:
 106         return page;
 107 }
 108
 109 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 110 {
 111         return __get_meta_page(sbi, index, true);
 112 }
 113
 114 struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
 115 {
 116         struct page *page;
 117         int count = 0;
 118
 119 retry:
 120         page = __get_meta_page(sbi, index, true);
 121         if (IS_ERR(page)) {
 122                 if (PTR_ERR(page) == -EIO &&
 123                                 ++count <= DEFAULT_RETRY_IO_COUNT)
 124                         goto retry;
 125
 126                 f2fs_stop_checkpoint(sbi, false);
 127                 f2fs_bug_on(sbi, 1);
 128         }
 129
 130         return page;
 131 }
 132
 133 /* for POR only */
 134 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 135 {
 136         return __get_meta_page(sbi, index, false);
 137 }
 138
 139 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 140                                         block_t blkaddr, int type)
 141 {
 142         switch (type) {
 143         case META_NAT:
 144                 break;
 145         case META_SIT:
 146                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 147                         return false;
 148                 break;
 149         case META_SSA:
 150                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 151                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 152                         return false;
 153                 break;
 154         case META_CP:
 155                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 156                         blkaddr < __start_cp_addr(sbi)))
 157                         return false;
 158                 break;
 159         case META_POR:
 160         case DATA_GENERIC:
 161                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 162                         blkaddr < MAIN_BLKADDR(sbi))) {
 163                         if (type == DATA_GENERIC) {
 164                                 f2fs_msg(sbi->sb, KERN_WARNING,
 165                                         "access invalid blkaddr:%u", blkaddr);
 166                                 WARN_ON(1);
 167                         }
 168                         return false;
 169                 }
 170                 break;
 171         case META_GENERIC:
 172                 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 173                         blkaddr >= MAIN_BLKADDR(sbi)))
 174                         return false;
 175                 break;
 176         default:
 177                 BUG();
 178         }
 179
 180         return true;
 181 }
 182
 183 /*
 184  * Readahead CP/NAT/SIT/SSA pages
 185  */
 186 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 187                                                         int type, bool sync)
 188 {
 189         struct page *page;
 190         block_t blkno = start;
 191         struct f2fs_io_info fio = {
 192                 .sbi = sbi,
 193                 .type = META,
 194                 .op = REQ_OP_READ,
 195                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 196                 .encrypted_page = NULL,
 197                 .in_list = false,
 198                 .is_meta = (type != META_POR),
 199         };
 200         struct blk_plug plug;
 201
 202         if (unlikely(type == META_POR))
 203                 fio.op_flags &= ~REQ_META;
 204
 205         blk_start_plug(&plug);
 206         for (; nrpages-- > 0; blkno++) {
 207
 208                 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 209                         goto out;
 210
 211                 switch (type) {
 212                 case META_NAT:
 213                         if (unlikely(blkno >=
 214                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 215                                 blkno = 0;
 216                         /* get nat block addr */
 217                         fio.new_blkaddr = current_nat_addr(sbi,
 218                                         blkno * NAT_ENTRY_PER_BLOCK);
 219                         break;
 220                 case META_SIT:
 221                         /* get sit block addr */
 222                         fio.new_blkaddr = current_sit_addr(sbi,
 223                                         blkno * SIT_ENTRY_PER_BLOCK);
 224                         break;
 225                 case META_SSA:
 226                 case META_CP:
 227                 case META_POR:
 228                         fio.new_blkaddr = blkno;
 229                         break;
 230                 default:
 231                         BUG();
 232                 }
 233
 234                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 235                                                 fio.new_blkaddr, false);
 236                 if (!page)
 237                         continue;
 238                 if (PageUptodate(page)) {
 239                         f2fs_put_page(page, 1);
 240                         continue;
 241                 }
 242
 243                 fio.page = page;
 244                 f2fs_submit_page_bio(&fio);
 245                 f2fs_put_page(page, 0);
 246         }
 247 out:
 248         blk_finish_plug(&plug);
 249         return blkno - start;
 250 }
 251
 252 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 253 {
 254         struct page *page;
 255         bool readahead = false;
 256
 257         page = find_get_page(META_MAPPING(sbi), index);
 258         if (!page || !PageUptodate(page))
 259                 readahead = true;
 260         f2fs_put_page(page, 0);
 261
 262         if (readahead)
 263                 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 264 }
 265
 266 static int __f2fs_write_meta_page(struct page *page,
 267                                 struct writeback_control *wbc,
 268                                 enum iostat_type io_type)
 269 {
 270         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 271
 272         trace_f2fs_writepage(page, META);
 273
 274         if (unlikely(f2fs_cp_error(sbi)))
 275                 goto redirty_out;
 276         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 277                 goto redirty_out;
 278         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 279                 goto redirty_out;
 280
 281         f2fs_do_write_meta_page(sbi, page, io_type);
 282         dec_page_count(sbi, F2FS_DIRTY_META);
 283
 284         if (wbc->for_reclaim)
 285                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
 286                                                 0, page->index, META);
 287
 288         unlock_page(page);
 289
 290         if (unlikely(f2fs_cp_error(sbi)))
 291                 f2fs_submit_merged_write(sbi, META);
 292
 293         return 0;
 294
 295 redirty_out:
 296         redirty_page_for_writepage(wbc, page);
 297         return AOP_WRITEPAGE_ACTIVATE;
 298 }
 299
 300 static int f2fs_write_meta_page(struct page *page,
 301                                 struct writeback_control *wbc)
 302 {
 303         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 304 }
 305
 306 static int f2fs_write_meta_pages(struct address_space *mapping,
 307                                 struct writeback_control *wbc)
 308 {
 309         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 310         long diff, written;
 311
 312         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 313                 goto skip_write;
 314
 315         /* collect a number of dirty meta pages and write together */
 316         if (wbc->for_kupdate ||
 317                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 318                 goto skip_write;
 319
 320         /* if locked failed, cp will flush dirty pages instead */
 321         if (!mutex_trylock(&sbi->cp_mutex))
 322                 goto skip_write;
 323
 324         trace_f2fs_writepages(mapping->host, wbc, META);
 325         diff = nr_pages_to_write(sbi, META, wbc);
 326         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 327         mutex_unlock(&sbi->cp_mutex);
 328         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 329         return 0;
 330
 331 skip_write:
 332         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 333         trace_f2fs_writepages(mapping->host, wbc, META);
 334         return 0;
 335 }
 336
 337 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 338                                 long nr_to_write, enum iostat_type io_type)
 339 {
 340         struct address_space *mapping = META_MAPPING(sbi);
 341         pgoff_t index = 0, prev = ULONG_MAX;
 342         struct pagevec pvec;
 343         long nwritten = 0;
 344         int nr_pages;
 345         struct writeback_control wbc = {
 346                 .for_reclaim = 0,
 347         };
 348         struct blk_plug plug;
 349
 350         pagevec_init(&pvec);
 351
 352         blk_start_plug(&plug);
 353
 354         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 355                                 PAGECACHE_TAG_DIRTY))) {
 356                 int i;
 357
 358                 for (i = 0; i < nr_pages; i++) {
 359                         struct page *page = pvec.pages[i];
 360
 361                         if (prev == ULONG_MAX)
 362                                 prev = page->index - 1;
 363                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 364                                 pagevec_release(&pvec);
 365                                 goto stop;
 366                         }
 367
 368                         lock_page(page);
 369
 370                         if (unlikely(page->mapping != mapping)) {
 371 continue_unlock:
 372                                 unlock_page(page);
 373                                 continue;
 374                         }
 375                         if (!PageDirty(page)) {
 376                                 /* someone wrote it for us */
 377                                 goto continue_unlock;
 378                         }
 379
 380                         f2fs_wait_on_page_writeback(page, META, true);
 381
 382                         BUG_ON(PageWriteback(page));
 383                         if (!clear_page_dirty_for_io(page))
 384                                 goto continue_unlock;
 385
 386                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 387                                 unlock_page(page);
 388                                 break;
 389                         }
 390                         nwritten++;
 391                         prev = page->index;
 392                         if (unlikely(nwritten >= nr_to_write))
 393                                 break;
 394                 }
 395                 pagevec_release(&pvec);
 396                 cond_resched();
 397         }
 398 stop:
 399         if (nwritten)
 400                 f2fs_submit_merged_write(sbi, type);
 401
 402         blk_finish_plug(&plug);
 403
 404         return nwritten;
 405 }
 406
 407 static int f2fs_set_meta_page_dirty(struct page *page)
 408 {
 409         trace_f2fs_set_page_dirty(page, META);
 410
 411         if (!PageUptodate(page))
 412                 SetPageUptodate(page);
 413         if (!PageDirty(page)) {
 414                 __set_page_dirty_nobuffers(page);
 415                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 416                 SetPagePrivate(page);
 417                 f2fs_trace_pid(page);
 418                 return 1;
 419         }
 420         return 0;
 421 }
 422
 423 const struct address_space_operations f2fs_meta_aops = {
 424         .writepage      = f2fs_write_meta_page,
 425         .writepages     = f2fs_write_meta_pages,
 426         .set_page_dirty = f2fs_set_meta_page_dirty,
 427         .invalidatepage = f2fs_invalidate_page,
 428         .releasepage    = f2fs_release_page,
 429 #ifdef CONFIG_MIGRATION
 430         .migratepage    = f2fs_migrate_page,
 431 #endif
 432 };
 433
 434 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 435                                                 unsigned int devidx, int type)
 436 {
 437         struct inode_management *im = &sbi->im[type];
 438         struct ino_entry *e, *tmp;
 439
 440         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 441
 442         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 443
 444         spin_lock(&im->ino_lock);
 445         e = radix_tree_lookup(&im->ino_root, ino);
 446         if (!e) {
 447                 e = tmp;
 448                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 449                         f2fs_bug_on(sbi, 1);
 450
 451                 memset(e, 0, sizeof(struct ino_entry));
 452                 e->ino = ino;
 453
 454                 list_add_tail(&e->list, &im->ino_list);
 455                 if (type != ORPHAN_INO)
 456                         im->ino_num++;
 457         }
 458
 459         if (type == FLUSH_INO)
 460                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 461
 462         spin_unlock(&im->ino_lock);
 463         radix_tree_preload_end();
 464
 465         if (e != tmp)
 466                 kmem_cache_free(ino_entry_slab, tmp);
 467 }
 468
 469 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 470 {
 471         struct inode_management *im = &sbi->im[type];
 472         struct ino_entry *e;
 473
 474         spin_lock(&im->ino_lock);
 475         e = radix_tree_lookup(&im->ino_root, ino);
 476         if (e) {
 477                 list_del(&e->list);
 478                 radix_tree_delete(&im->ino_root, ino);
 479                 im->ino_num--;
 480                 spin_unlock(&im->ino_lock);
 481                 kmem_cache_free(ino_entry_slab, e);
 482                 return;
 483         }
 484         spin_unlock(&im->ino_lock);
 485 }
 486
 487 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 488 {
 489         /* add new dirty ino entry into list */
 490         __add_ino_entry(sbi, ino, 0, type);
 491 }
 492
 493 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 494 {
 495         /* remove dirty ino entry from list */
 496         __remove_ino_entry(sbi, ino, type);
 497 }
 498
 499 /* mode should be APPEND_INO or UPDATE_INO */
 500 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 501 {
 502         struct inode_management *im = &sbi->im[mode];
 503         struct ino_entry *e;
 504
 505         spin_lock(&im->ino_lock);
 506         e = radix_tree_lookup(&im->ino_root, ino);
 507         spin_unlock(&im->ino_lock);
 508         return e ? true : false;
 509 }
 510
 511 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 512 {
 513         struct ino_entry *e, *tmp;
 514         int i;
 515
 516         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 517                 struct inode_management *im = &sbi->im[i];
 518
 519                 spin_lock(&im->ino_lock);
 520                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 521                         list_del(&e->list);
 522                         radix_tree_delete(&im->ino_root, e->ino);
 523                         kmem_cache_free(ino_entry_slab, e);
 524                         im->ino_num--;
 525                 }
 526                 spin_unlock(&im->ino_lock);
 527         }
 528 }
 529
 530 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 531                                         unsigned int devidx, int type)
 532 {
 533         __add_ino_entry(sbi, ino, devidx, type);
 534 }
 535
 536 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 537                                         unsigned int devidx, int type)
 538 {
 539         struct inode_management *im = &sbi->im[type];
 540         struct ino_entry *e;
 541         bool is_dirty = false;
 542
 543         spin_lock(&im->ino_lock);
 544         e = radix_tree_lookup(&im->ino_root, ino);
 545         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 546                 is_dirty = true;
 547         spin_unlock(&im->ino_lock);
 548         return is_dirty;
 549 }
 550
 551 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 552 {
 553         struct inode_management *im = &sbi->im[ORPHAN_INO];
 554         int err = 0;
 555
 556         spin_lock(&im->ino_lock);
 557
 558         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 559                 spin_unlock(&im->ino_lock);
 560                 f2fs_show_injection_info(FAULT_ORPHAN);
 561                 return -ENOSPC;
 562         }
 563
 564         if (unlikely(im->ino_num >= sbi->max_orphans))
 565                 err = -ENOSPC;
 566         else
 567                 im->ino_num++;
 568         spin_unlock(&im->ino_lock);
 569
 570         return err;
 571 }
 572
 573 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 574 {
 575         struct inode_management *im = &sbi->im[ORPHAN_INO];
 576
 577         spin_lock(&im->ino_lock);
 578         f2fs_bug_on(sbi, im->ino_num == 0);
 579         im->ino_num--;
 580         spin_unlock(&im->ino_lock);
 581 }
 582
 583 void f2fs_add_orphan_inode(struct inode *inode)
 584 {
 585         /* add new orphan ino entry into list */
 586         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 587         f2fs_update_inode_page(inode);
 588 }
 589
 590 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 591 {
 592         /* remove orphan entry from orphan list */
 593         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 594 }
 595
 596 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 597 {
 598         struct inode *inode;
 599         struct node_info ni;
 600         int err;
 601
 602         inode = f2fs_iget_retry(sbi->sb, ino);
 603         if (IS_ERR(inode)) {
 604                 /*
 605                  * there should be a bug that we can't find the entry
 606                  * to orphan inode.
 607                  */
 608                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 609                 return PTR_ERR(inode);
 610         }
 611
 612         err = dquot_initialize(inode);
 613         if (err) {
 614                 iput(inode);
 615                 goto err_out;
 616         }
 617
 618         clear_nlink(inode);
 619
 620         /* truncate all the data during iput */
 621         iput(inode);
 622
 623         err = f2fs_get_node_info(sbi, ino, &ni);
 624         if (err)
 625                 goto err_out;
 626
 627         /* ENOMEM was fully retried in f2fs_evict_inode. */
 628         if (ni.blk_addr != NULL_ADDR) {
 629                 err = -EIO;
 630                 goto err_out;
 631         }
 632         return 0;
 633
 634 err_out:
 635         set_sbi_flag(sbi, SBI_NEED_FSCK);
 636         f2fs_msg(sbi->sb, KERN_WARNING,
 637                         "%s: orphan failed (ino=%x), run fsck to fix.",
 638                         __func__, ino);
 639         return err;
 640 }
 641
 642 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 643 {
 644         block_t start_blk, orphan_blocks, i, j;
 645         unsigned int s_flags = sbi->sb->s_flags;
 646         int err = 0;
 647 #ifdef CONFIG_QUOTA
 648         int quota_enabled;
 649 #endif
 650
 651         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 652                 return 0;
 653
 654         if (s_flags & SB_RDONLY) {
 655                 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
 656                 sbi->sb->s_flags &= ~SB_RDONLY;
 657         }
 658
 659 #ifdef CONFIG_QUOTA
 660         /* Needed for iput() to work correctly and not trash data */
 661         sbi->sb->s_flags |= SB_ACTIVE;
 662
 663         /*
 664          * Turn on quotas which were not enabled for read-only mounts if
 665          * filesystem has quota feature, so that they are updated correctly.
 666          */
 667         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 668 #endif
 669
 670         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 671         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 672
 673         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 674
 675         for (i = 0; i < orphan_blocks; i++) {
 676                 struct page *page;
 677                 struct f2fs_orphan_block *orphan_blk;
 678
 679                 page = f2fs_get_meta_page(sbi, start_blk + i);
 680                 if (IS_ERR(page)) {
 681                         err = PTR_ERR(page);
 682                         goto out;
 683                 }
 684
 685                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 686                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 687                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 688                         err = recover_orphan_inode(sbi, ino);
 689                         if (err) {
 690                                 f2fs_put_page(page, 1);
 691                                 goto out;
 692                         }
 693                 }
 694                 f2fs_put_page(page, 1);
 695         }
 696         /* clear Orphan Flag */
 697         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 698 out:
 699         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 700
 701 #ifdef CONFIG_QUOTA
 702         /* Turn quotas off */
 703         if (quota_enabled)
 704                 f2fs_quota_off_umount(sbi->sb);
 705 #endif
 706         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 707
 708         return err;
 709 }
 710
 711 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 712 {
 713         struct list_head *head;
 714         struct f2fs_orphan_block *orphan_blk = NULL;
 715         unsigned int nentries = 0;
 716         unsigned short index = 1;
 717         unsigned short orphan_blocks;
 718         struct page *page = NULL;
 719         struct ino_entry *orphan = NULL;
 720         struct inode_management *im = &sbi->im[ORPHAN_INO];
 721
 722         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 723
 724         /*
 725          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 726          * orphan inode operations are covered under f2fs_lock_op().
 727          * And, spin_lock should be avoided due to page operations below.
 728          */
 729         head = &im->ino_list;
 730
 731         /* loop for each orphan inode entry and write them in Jornal block */
 732         list_for_each_entry(orphan, head, list) {
 733                 if (!page) {
 734                         page = f2fs_grab_meta_page(sbi, start_blk++);
 735                         orphan_blk =
 736                                 (struct f2fs_orphan_block *)page_address(page);
 737                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 738                 }
 739
 740                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 741
 742                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 743                         /*
 744                          * an orphan block is full of 1020 entries,
 745                          * then we need to flush current orphan blocks
 746                          * and bring another one in memory
 747                          */
 748                         orphan_blk->blk_addr = cpu_to_le16(index);
 749                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 750                         orphan_blk->entry_count = cpu_to_le32(nentries);
 751                         set_page_dirty(page);
 752                         f2fs_put_page(page, 1);
 753                         index++;
 754                         nentries = 0;
 755                         page = NULL;
 756                 }
 757         }
 758
 759         if (page) {
 760                 orphan_blk->blk_addr = cpu_to_le16(index);
 761                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 762                 orphan_blk->entry_count = cpu_to_le32(nentries);
 763                 set_page_dirty(page);
 764                 f2fs_put_page(page, 1);
 765         }
 766 }
 767
 768 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 769                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 770                 unsigned long long *version)
 771 {
 772         unsigned long blk_size = sbi->blocksize;
 773         size_t crc_offset = 0;
 774         __u32 crc = 0;
 775
 776         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 777         if (IS_ERR(*cp_page))
 778                 return PTR_ERR(*cp_page);
 779
 780         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 781
 782         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 783         if (crc_offset > (blk_size - sizeof(__le32))) {
 784                 f2fs_put_page(*cp_page, 1);
 785                 f2fs_msg(sbi->sb, KERN_WARNING,
 786                         "invalid crc_offset: %zu", crc_offset);
 787                 return -EINVAL;
 788         }
 789
 790         crc = cur_cp_crc(*cp_block);
 791         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 792                 f2fs_put_page(*cp_page, 1);
 793                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 794                 return -EINVAL;
 795         }
 796
 797         *version = cur_cp_version(*cp_block);
 798         return 0;
 799 }
 800
 801 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 802                                 block_t cp_addr, unsigned long long *version)
 803 {
 804         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 805         struct f2fs_checkpoint *cp_block = NULL;
 806         unsigned long long cur_version = 0, pre_version = 0;
 807         int err;
 808
 809         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 810                                         &cp_page_1, version);
 811         if (err)
 812                 return NULL;
 813
 814         if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
 815                                         sbi->blocks_per_seg) {
 816                 f2fs_msg(sbi->sb, KERN_WARNING,
 817                         "invalid cp_pack_total_block_count:%u",
 818                         le32_to_cpu(cp_block->cp_pack_total_block_count));
 819                 goto invalid_cp;
 820         }
 821         pre_version = *version;
 822
 823         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 824         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 825                                         &cp_page_2, version);
 826         if (err)
 827                 goto invalid_cp;
 828         cur_version = *version;
 829
 830         if (cur_version == pre_version) {
 831                 *version = cur_version;
 832                 f2fs_put_page(cp_page_2, 1);
 833                 return cp_page_1;
 834         }
 835         f2fs_put_page(cp_page_2, 1);
 836 invalid_cp:
 837         f2fs_put_page(cp_page_1, 1);
 838         return NULL;
 839 }
 840
 841 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 842 {
 843         struct f2fs_checkpoint *cp_block;
 844         struct f2fs_super_block *fsb = sbi->raw_super;
 845         struct page *cp1, *cp2, *cur_page;
 846         unsigned long blk_size = sbi->blocksize;
 847         unsigned long long cp1_version = 0, cp2_version = 0;
 848         unsigned long long cp_start_blk_no;
 849         unsigned int cp_blks = 1 + __cp_payload(sbi);
 850         block_t cp_blk_no;
 851         int i;
 852         int err;
 853
 854         sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
 855                                  GFP_KERNEL);
 856         if (!sbi->ckpt)
 857                 return -ENOMEM;
 858         /*
 859          * Finding out valid cp block involves read both
 860          * sets( cp pack1 and cp pack 2)
 861          */
 862         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 863         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 864
 865         /* The second checkpoint pack should start at the next segment */
 866         cp_start_blk_no += ((unsigned long long)1) <<
 867                                 le32_to_cpu(fsb->log_blocks_per_seg);
 868         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 869
 870         if (cp1 && cp2) {
 871                 if (ver_after(cp2_version, cp1_version))
 872                         cur_page = cp2;
 873                 else
 874                         cur_page = cp1;
 875         } else if (cp1) {
 876                 cur_page = cp1;
 877         } else if (cp2) {
 878                 cur_page = cp2;
 879         } else {
 880                 err = -EFSCORRUPTED;
 881                 goto fail_no_cp;
 882         }
 883
 884         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 885         memcpy(sbi->ckpt, cp_block, blk_size);
 886
 887         if (cur_page == cp1)
 888                 sbi->cur_cp_pack = 1;
 889         else
 890                 sbi->cur_cp_pack = 2;
 891
 892         /* Sanity checking of checkpoint */
 893         if (f2fs_sanity_check_ckpt(sbi)) {
 894                 err = -EFSCORRUPTED;
 895                 goto free_fail_no_cp;
 896         }
 897
 898         if (cp_blks <= 1)
 899                 goto done;
 900
 901         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 902         if (cur_page == cp2)
 903                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 904
 905         for (i = 1; i < cp_blks; i++) {
 906                 void *sit_bitmap_ptr;
 907                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 908
 909                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 910                 if (IS_ERR(cur_page)) {
 911                         err = PTR_ERR(cur_page);
 912                         goto free_fail_no_cp;
 913                 }
 914                 sit_bitmap_ptr = page_address(cur_page);
 915                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 916                 f2fs_put_page(cur_page, 1);
 917         }
 918 done:
 919         f2fs_put_page(cp1, 1);
 920         f2fs_put_page(cp2, 1);
 921         return 0;
 922
 923 free_fail_no_cp:
 924         f2fs_put_page(cp1, 1);
 925         f2fs_put_page(cp2, 1);
 926 fail_no_cp:
 927         kfree(sbi->ckpt);
 928         return err;
 929 }
 930
 931 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 932 {
 933         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 934         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 935
 936         if (is_inode_flag_set(inode, flag))
 937                 return;
 938
 939         set_inode_flag(inode, flag);
 940         if (!f2fs_is_volatile_file(inode))
 941                 list_add_tail(&F2FS_I(inode)->dirty_list,
 942                                                 &sbi->inode_list[type]);
 943         stat_inc_dirty_inode(sbi, type);
 944 }
 945
 946 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 947 {
 948         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 949
 950         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 951                 return;
 952
 953         list_del_init(&F2FS_I(inode)->dirty_list);
 954         clear_inode_flag(inode, flag);
 955         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 956 }
 957
 958 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
 959 {
 960         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 961         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 962
 963         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 964                         !S_ISLNK(inode->i_mode))
 965                 return;
 966
 967         spin_lock(&sbi->inode_lock[type]);
 968         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 969                 __add_dirty_inode(inode, type);
 970         inode_inc_dirty_pages(inode);
 971         spin_unlock(&sbi->inode_lock[type]);
 972
 973         SetPagePrivate(page);
 974         f2fs_trace_pid(page);
 975 }
 976
 977 void f2fs_remove_dirty_inode(struct inode *inode)
 978 {
 979         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 980         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 981
 982         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 983                         !S_ISLNK(inode->i_mode))
 984                 return;
 985
 986         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 987                 return;
 988
 989         spin_lock(&sbi->inode_lock[type]);
 990         __remove_dirty_inode(inode, type);
 991         spin_unlock(&sbi->inode_lock[type]);
 992 }
 993
 994 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 995 {
 996         struct list_head *head;
 997         struct inode *inode;
 998         struct f2fs_inode_info *fi;
 999         bool is_dir = (type == DIR_INODE);
1000         unsigned long ino = 0;
1001
1002         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1003                                 get_pages(sbi, is_dir ?
1004                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1005 retry:
1006         if (unlikely(f2fs_cp_error(sbi)))
1007                 return -EIO;
1008
1009         spin_lock(&sbi->inode_lock[type]);
1010
1011         head = &sbi->inode_list[type];
1012         if (list_empty(head)) {
1013                 spin_unlock(&sbi->inode_lock[type]);
1014                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1015                                 get_pages(sbi, is_dir ?
1016                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1017                 return 0;
1018         }
1019         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1020         inode = igrab(&fi->vfs_inode);
1021         spin_unlock(&sbi->inode_lock[type]);
1022         if (inode) {
1023                 unsigned long cur_ino = inode->i_ino;
1024
1025                 if (is_dir)
1026                         F2FS_I(inode)->cp_task = current;
1027
1028                 filemap_fdatawrite(inode->i_mapping);
1029
1030                 if (is_dir)
1031                         F2FS_I(inode)->cp_task = NULL;
1032
1033                 iput(inode);
1034                 /* We need to give cpu to another writers. */
1035                 if (ino == cur_ino)
1036                         cond_resched();
1037                 else
1038                         ino = cur_ino;
1039         } else {
1040                 /*
1041                  * We should submit bio, since it exists several
1042                  * wribacking dentry pages in the freeing inode.
1043                  */
1044                 f2fs_submit_merged_write(sbi, DATA);
1045                 cond_resched();
1046         }
1047         goto retry;
1048 }
1049
1050 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1051 {
1052         struct list_head *head = &sbi->inode_list[DIRTY_META];
1053         struct inode *inode;
1054         struct f2fs_inode_info *fi;
1055         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1056
1057         while (total--) {
1058                 if (unlikely(f2fs_cp_error(sbi)))
1059                         return -EIO;
1060
1061                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1062                 if (list_empty(head)) {
1063                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1064                         return 0;
1065                 }
1066                 fi = list_first_entry(head, struct f2fs_inode_info,
1067                                                         gdirty_list);
1068                 inode = igrab(&fi->vfs_inode);
1069                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1070                 if (inode) {
1071                         sync_inode_metadata(inode, 0);
1072
1073                         /* it's on eviction */
1074                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1075                                 f2fs_update_inode_page(inode);
1076                         iput(inode);
1077                 }
1078         }
1079         return 0;
1080 }
1081
1082 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1083 {
1084         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1085         struct f2fs_nm_info *nm_i = NM_I(sbi);
1086         nid_t last_nid = nm_i->next_scan_nid;
1087
1088         next_free_nid(sbi, &last_nid);
1089         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1090         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1091         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1092         ckpt->next_free_nid = cpu_to_le32(last_nid);
1093 }
1094
1095 /*
1096  * Freeze all the FS-operations for checkpoint.
1097  */
1098 static int block_operations(struct f2fs_sb_info *sbi)
1099 {
1100         struct writeback_control wbc = {
1101                 .sync_mode = WB_SYNC_ALL,
1102                 .nr_to_write = LONG_MAX,
1103                 .for_reclaim = 0,
1104         };
1105         struct blk_plug plug;
1106         int err = 0;
1107
1108         blk_start_plug(&plug);
1109
1110 retry_flush_dents:
1111         f2fs_lock_all(sbi);
1112         /* write all the dirty dentry pages */
1113         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1114                 f2fs_unlock_all(sbi);
1115                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1116                 if (err)
1117                         goto out;
1118                 cond_resched();
1119                 goto retry_flush_dents;
1120         }
1121
1122         /*
1123          * POR: we should ensure that there are no dirty node pages
1124          * until finishing nat/sit flush. inode->i_blocks can be updated.
1125          */
1126         down_write(&sbi->node_change);
1127
1128         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1129                 up_write(&sbi->node_change);
1130                 f2fs_unlock_all(sbi);
1131                 err = f2fs_sync_inode_meta(sbi);
1132                 if (err)
1133                         goto out;
1134                 cond_resched();
1135                 goto retry_flush_dents;
1136         }
1137
1138 retry_flush_nodes:
1139         down_write(&sbi->node_write);
1140
1141         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1142                 up_write(&sbi->node_write);
1143                 atomic_inc(&sbi->wb_sync_req[NODE]);
1144                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1145                 atomic_dec(&sbi->wb_sync_req[NODE]);
1146                 if (err) {
1147                         up_write(&sbi->node_change);
1148                         f2fs_unlock_all(sbi);
1149                         goto out;
1150                 }
1151                 cond_resched();
1152                 goto retry_flush_nodes;
1153         }
1154
1155         /*
1156          * sbi->node_change is used only for AIO write_begin path which produces
1157          * dirty node blocks and some checkpoint values by block allocation.
1158          */
1159         __prepare_cp_block(sbi);
1160         up_write(&sbi->node_change);
1161 out:
1162         blk_finish_plug(&plug);
1163         return err;
1164 }
1165
1166 static void unblock_operations(struct f2fs_sb_info *sbi)
1167 {
1168         up_write(&sbi->node_write);
1169         f2fs_unlock_all(sbi);
1170 }
1171
1172 void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1173 {
1174         DEFINE_WAIT(wait);
1175
1176         for (;;) {
1177                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1178
1179                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1180                         break;
1181
1182                 if (unlikely(f2fs_cp_error(sbi)))
1183                         break;
1184
1185                 io_schedule_timeout(5*HZ);
1186         }
1187         finish_wait(&sbi->cp_wait, &wait);
1188 }
1189
1190 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1191 {
1192         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1193         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1194         unsigned long flags;
1195
1196         spin_lock_irqsave(&sbi->cp_lock, flags);
1197
1198         if ((cpc->reason & CP_UMOUNT) &&
1199                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1200                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1201                 disable_nat_bits(sbi, false);
1202
1203         if (cpc->reason & CP_TRIMMED)
1204                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1205         else
1206                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1207
1208         if (cpc->reason & CP_UMOUNT)
1209                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1210         else
1211                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1212
1213         if (cpc->reason & CP_FASTBOOT)
1214                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1215         else
1216                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1217
1218         if (orphan_num)
1219                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1220         else
1221                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1222
1223         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1224                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1225
1226         /* set this flag to activate crc|cp_ver for recovery */
1227         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1228         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1229
1230         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1231 }
1232
1233 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1234         void *src, block_t blk_addr)
1235 {
1236         struct writeback_control wbc = {
1237                 .for_reclaim = 0,
1238         };
1239
1240         /*
1241          * pagevec_lookup_tag and lock_page again will take
1242          * some extra time. Therefore, f2fs_update_meta_pages and
1243          * f2fs_sync_meta_pages are combined in this function.
1244          */
1245         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1246         int err;
1247
1248         memcpy(page_address(page), src, PAGE_SIZE);
1249         set_page_dirty(page);
1250
1251         f2fs_wait_on_page_writeback(page, META, true);
1252         f2fs_bug_on(sbi, PageWriteback(page));
1253         if (unlikely(!clear_page_dirty_for_io(page)))
1254                 f2fs_bug_on(sbi, 1);
1255
1256         /* writeout cp pack 2 page */
1257         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1258         if (unlikely(err && f2fs_cp_error(sbi))) {
1259                 f2fs_put_page(page, 1);
1260                 return;
1261         }
1262
1263         f2fs_bug_on(sbi, err);
1264         f2fs_put_page(page, 0);
1265
1266         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1267         f2fs_submit_merged_write(sbi, META_FLUSH);
1268 }
1269
1270 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1271 {
1272         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1273         struct f2fs_nm_info *nm_i = NM_I(sbi);
1274         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1275         block_t start_blk;
1276         unsigned int data_sum_blocks, orphan_blocks;
1277         __u32 crc32 = 0;
1278         int i;
1279         int cp_payload_blks = __cp_payload(sbi);
1280         struct super_block *sb = sbi->sb;
1281         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1282         u64 kbytes_written;
1283         int err;
1284
1285         /* Flush all the NAT/SIT pages */
1286         while (get_pages(sbi, F2FS_DIRTY_META)) {
1287                 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1288                 if (unlikely(f2fs_cp_error(sbi)))
1289                         break;
1290         }
1291
1292         /*
1293          * modify checkpoint
1294          * version number is already updated
1295          */
1296         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1297         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1298         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1299                 ckpt->cur_node_segno[i] =
1300                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1301                 ckpt->cur_node_blkoff[i] =
1302                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1303                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1304                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1305         }
1306         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1307                 ckpt->cur_data_segno[i] =
1308                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1309                 ckpt->cur_data_blkoff[i] =
1310                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1311                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1312                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1313         }
1314
1315         /* 2 cp  + n data seg summary + orphan inode blocks */
1316         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1317         spin_lock_irqsave(&sbi->cp_lock, flags);
1318         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1319                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1320         else
1321                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1322         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1323
1324         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1325         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1326                         orphan_blocks);
1327
1328         if (__remain_node_summaries(cpc->reason))
1329                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1330                                 cp_payload_blks + data_sum_blocks +
1331                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1332         else
1333                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1334                                 cp_payload_blks + data_sum_blocks +
1335                                 orphan_blocks);
1336
1337         /* update ckpt flag for checkpoint */
1338         update_ckpt_flags(sbi, cpc);
1339
1340         /* update SIT/NAT bitmap */
1341         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1342         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1343
1344         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1345         *((__le32 *)((unsigned char *)ckpt +
1346                                 le32_to_cpu(ckpt->checksum_offset)))
1347                                 = cpu_to_le32(crc32);
1348
1349         start_blk = __start_cp_next_addr(sbi);
1350
1351         /* write nat bits */
1352         if (enabled_nat_bits(sbi, cpc)) {
1353                 __u64 cp_ver = cur_cp_version(ckpt);
1354                 block_t blk;
1355
1356                 cp_ver |= ((__u64)crc32 << 32);
1357                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1358
1359                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1360                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1361                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1362                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1363
1364                 /* Flush all the NAT BITS pages */
1365                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1366                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1367                                                         FS_CP_META_IO);
1368                         if (unlikely(f2fs_cp_error(sbi)))
1369                                 break;
1370                 }
1371         }
1372
1373         /* write out checkpoint buffer at block 0 */
1374         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1375
1376         for (i = 1; i < 1 + cp_payload_blks; i++)
1377                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1378                                                         start_blk++);
1379
1380         if (orphan_num) {
1381                 write_orphan_inodes(sbi, start_blk);
1382                 start_blk += orphan_blocks;
1383         }
1384
1385         f2fs_write_data_summaries(sbi, start_blk);
1386         start_blk += data_sum_blocks;
1387
1388         /* Record write statistics in the hot node summary */
1389         kbytes_written = sbi->kbytes_written;
1390         if (sb->s_bdev->bd_part)
1391                 kbytes_written += BD_PART_WRITTEN(sbi);
1392
1393         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1394
1395         if (__remain_node_summaries(cpc->reason)) {
1396                 f2fs_write_node_summaries(sbi, start_blk);
1397                 start_blk += NR_CURSEG_NODE_TYPE;
1398         }
1399
1400         /* update user_block_counts */
1401         sbi->last_valid_block_count = sbi->total_valid_block_count;
1402         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1403
1404         /* Here, we have one bio having CP pack except cp pack 2 page */
1405         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1406
1407         /* wait for previous submitted meta pages writeback */
1408         f2fs_wait_on_all_pages_writeback(sbi);
1409
1410         /* flush all device cache */
1411         err = f2fs_flush_device_cache(sbi);
1412         if (err)
1413                 return err;
1414
1415         /* barrier and flush checkpoint cp pack 2 page if it can */
1416         commit_checkpoint(sbi, ckpt, start_blk);
1417         f2fs_wait_on_all_pages_writeback(sbi);
1418
1419         /*
1420          * invalidate intermediate page cache borrowed from meta inode
1421          * which are used for migration of encrypted inode's blocks.
1422          */
1423         if (f2fs_sb_has_encrypt(sbi->sb))
1424                 invalidate_mapping_pages(META_MAPPING(sbi),
1425                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1426
1427         f2fs_release_ino_entry(sbi, false);
1428
1429         f2fs_reset_fsync_node_info(sbi);
1430
1431         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1432         clear_sbi_flag(sbi, SBI_NEED_CP);
1433         __set_cp_next_pack(sbi);
1434
1435         /*
1436          * redirty superblock if metadata like node page or inode cache is
1437          * updated during writing checkpoint.
1438          */
1439         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1440                         get_pages(sbi, F2FS_DIRTY_IMETA))
1441                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1442
1443         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1444
1445         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1446 }
1447
1448 /*
1449  * We guarantee that this checkpoint procedure will not fail.
1450  */
1451 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1452 {
1453         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1454         unsigned long long ckpt_ver;
1455         int err = 0;
1456
1457         mutex_lock(&sbi->cp_mutex);
1458
1459         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1460                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1461                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1462                 goto out;
1463         if (unlikely(f2fs_cp_error(sbi))) {
1464                 err = -EIO;
1465                 goto out;
1466         }
1467         if (f2fs_readonly(sbi->sb)) {
1468                 err = -EROFS;
1469                 goto out;
1470         }
1471
1472         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1473
1474         err = block_operations(sbi);
1475         if (err)
1476                 goto out;
1477
1478         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1479
1480         f2fs_flush_merged_writes(sbi);
1481
1482         /* this is the case of multiple fstrims without any changes */
1483         if (cpc->reason & CP_DISCARD) {
1484                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1485                         unblock_operations(sbi);
1486                         goto out;
1487                 }
1488
1489                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1490                                 SIT_I(sbi)->dirty_sentries == 0 &&
1491                                 prefree_segments(sbi) == 0) {
1492                         f2fs_flush_sit_entries(sbi, cpc);
1493                         f2fs_clear_prefree_segments(sbi, cpc);
1494                         unblock_operations(sbi);
1495                         goto out;
1496                 }
1497         }
1498
1499         /*
1500          * update checkpoint pack index
1501          * Increase the version number so that
1502          * SIT entries and seg summaries are written at correct place
1503          */
1504         ckpt_ver = cur_cp_version(ckpt);
1505         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1506
1507         /* write cached NAT/SIT entries to NAT/SIT area */
1508         f2fs_flush_nat_entries(sbi, cpc);
1509         f2fs_flush_sit_entries(sbi, cpc);
1510
1511         /* unlock all the fs_lock[] in do_checkpoint() */
1512         err = do_checkpoint(sbi, cpc);
1513         if (err)
1514                 f2fs_release_discard_addrs(sbi);
1515         else
1516                 f2fs_clear_prefree_segments(sbi, cpc);
1517
1518         unblock_operations(sbi);
1519         stat_inc_cp_count(sbi->stat_info);
1520
1521         if (cpc->reason & CP_RECOVERY)
1522                 f2fs_msg(sbi->sb, KERN_NOTICE,
1523                         "checkpoint: version = %llx", ckpt_ver);
1524
1525         /* do checkpoint periodically */
1526         f2fs_update_time(sbi, CP_TIME);
1527         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1528 out:
1529         mutex_unlock(&sbi->cp_mutex);
1530         return err;
1531 }
1532
1533 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1534 {
1535         int i;
1536
1537         for (i = 0; i < MAX_INO_ENTRY; i++) {
1538                 struct inode_management *im = &sbi->im[i];
1539
1540                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1541                 spin_lock_init(&im->ino_lock);
1542                 INIT_LIST_HEAD(&im->ino_list);
1543                 im->ino_num = 0;
1544         }
1545
1546         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1547                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1548                                 F2FS_ORPHANS_PER_BLOCK;
1549 }
1550
1551 int __init f2fs_create_checkpoint_caches(void)
1552 {
1553         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1554                         sizeof(struct ino_entry));
1555         if (!ino_entry_slab)
1556                 return -ENOMEM;
1557         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1558                         sizeof(struct inode_entry));
1559         if (!f2fs_inode_entry_slab) {
1560                 kmem_cache_destroy(ino_entry_slab);
1561                 return -ENOMEM;
1562         }
1563         return 0;
1564 }
1565
1566 void f2fs_destroy_checkpoint_caches(void)
1567 {
1568         kmem_cache_destroy(ino_entry_slab);
1569         kmem_cache_destroy(f2fs_inode_entry_slab);
1570 }