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         if (!end_io)
  33                 f2fs_flush_merged_writes(sbi);
  34 }
  35
  36 /*
  37  * We guarantee no failure on the returned page.
  38  */
  39 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  40 {
  41         struct address_space *mapping = META_MAPPING(sbi);
  42         struct page *page = NULL;
  43 repeat:
  44         page = f2fs_grab_cache_page(mapping, index, false);
  45         if (!page) {
  46                 cond_resched();
  47                 goto repeat;
  48         }
  49         f2fs_wait_on_page_writeback(page, META, true);
  50         if (!PageUptodate(page))
  51                 SetPageUptodate(page);
  52         return page;
  53 }
  54
  55 /*
  56  * We guarantee no failure on the returned page.
  57  */
  58 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  59                                                         bool is_meta)
  60 {
  61         struct address_space *mapping = META_MAPPING(sbi);
  62         struct page *page;
  63         struct f2fs_io_info fio = {
  64                 .sbi = sbi,
  65                 .type = META,
  66                 .op = REQ_OP_READ,
  67                 .op_flags = REQ_META | REQ_PRIO,
  68                 .old_blkaddr = index,
  69                 .new_blkaddr = index,
  70                 .encrypted_page = NULL,
  71                 .is_meta = is_meta,
  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                 .is_meta = (type != META_POR),
 167         };
 168         struct blk_plug plug;
 169
 170         if (unlikely(type == META_POR))
 171                 fio.op_flags &= ~REQ_META;
 172
 173         blk_start_plug(&plug);
 174         for (; nrpages-- > 0; blkno++) {
 175
 176                 if (!is_valid_blkaddr(sbi, blkno, type))
 177                         goto out;
 178
 179                 switch (type) {
 180                 case META_NAT:
 181                         if (unlikely(blkno >=
 182                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 183                                 blkno = 0;
 184                         /* get nat block addr */
 185                         fio.new_blkaddr = current_nat_addr(sbi,
 186                                         blkno * NAT_ENTRY_PER_BLOCK);
 187                         break;
 188                 case META_SIT:
 189                         /* get sit block addr */
 190                         fio.new_blkaddr = current_sit_addr(sbi,
 191                                         blkno * SIT_ENTRY_PER_BLOCK);
 192                         break;
 193                 case META_SSA:
 194                 case META_CP:
 195                 case META_POR:
 196                         fio.new_blkaddr = blkno;
 197                         break;
 198                 default:
 199                         BUG();
 200                 }
 201
 202                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 203                                                 fio.new_blkaddr, false);
 204                 if (!page)
 205                         continue;
 206                 if (PageUptodate(page)) {
 207                         f2fs_put_page(page, 1);
 208                         continue;
 209                 }
 210
 211                 fio.page = page;
 212                 f2fs_submit_page_bio(&fio);
 213                 f2fs_put_page(page, 0);
 214         }
 215 out:
 216         blk_finish_plug(&plug);
 217         return blkno - start;
 218 }
 219
 220 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 221 {
 222         struct page *page;
 223         bool readahead = false;
 224
 225         page = find_get_page(META_MAPPING(sbi), index);
 226         if (!page || !PageUptodate(page))
 227                 readahead = true;
 228         f2fs_put_page(page, 0);
 229
 230         if (readahead)
 231                 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 232 }
 233
 234 static int __f2fs_write_meta_page(struct page *page,
 235                                 struct writeback_control *wbc,
 236                                 enum iostat_type io_type)
 237 {
 238         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 239
 240         trace_f2fs_writepage(page, META);
 241
 242         if (unlikely(f2fs_cp_error(sbi))) {
 243                 dec_page_count(sbi, F2FS_DIRTY_META);
 244                 unlock_page(page);
 245                 return 0;
 246         }
 247         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 248                 goto redirty_out;
 249         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 250                 goto redirty_out;
 251
 252         write_meta_page(sbi, page, io_type);
 253         dec_page_count(sbi, F2FS_DIRTY_META);
 254
 255         if (wbc->for_reclaim)
 256                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
 257                                                 0, page->index, META);
 258
 259         unlock_page(page);
 260
 261         if (unlikely(f2fs_cp_error(sbi)))
 262                 f2fs_submit_merged_write(sbi, META);
 263
 264         return 0;
 265
 266 redirty_out:
 267         redirty_page_for_writepage(wbc, page);
 268         return AOP_WRITEPAGE_ACTIVATE;
 269 }
 270
 271 static int f2fs_write_meta_page(struct page *page,
 272                                 struct writeback_control *wbc)
 273 {
 274         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 275 }
 276
 277 static int f2fs_write_meta_pages(struct address_space *mapping,
 278                                 struct writeback_control *wbc)
 279 {
 280         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 281         long diff, written;
 282
 283         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 284                 goto skip_write;
 285
 286         /* collect a number of dirty meta pages and write together */
 287         if (wbc->for_kupdate ||
 288                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 289                 goto skip_write;
 290
 291         /* if locked failed, cp will flush dirty pages instead */
 292         if (!mutex_trylock(&sbi->cp_mutex))
 293                 goto skip_write;
 294
 295         trace_f2fs_writepages(mapping->host, wbc, META);
 296         diff = nr_pages_to_write(sbi, META, wbc);
 297         written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 298         mutex_unlock(&sbi->cp_mutex);
 299         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 300         return 0;
 301
 302 skip_write:
 303         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 304         trace_f2fs_writepages(mapping->host, wbc, META);
 305         return 0;
 306 }
 307
 308 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 309                                 long nr_to_write, enum iostat_type io_type)
 310 {
 311         struct address_space *mapping = META_MAPPING(sbi);
 312         pgoff_t index = 0, prev = ULONG_MAX;
 313         struct pagevec pvec;
 314         long nwritten = 0;
 315         int nr_pages;
 316         struct writeback_control wbc = {
 317                 .for_reclaim = 0,
 318         };
 319         struct blk_plug plug;
 320
 321         pagevec_init(&pvec);
 322
 323         blk_start_plug(&plug);
 324
 325         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 326                                 PAGECACHE_TAG_DIRTY))) {
 327                 int i;
 328
 329                 for (i = 0; i < nr_pages; i++) {
 330                         struct page *page = pvec.pages[i];
 331
 332                         if (prev == ULONG_MAX)
 333                                 prev = page->index - 1;
 334                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 335                                 pagevec_release(&pvec);
 336                                 goto stop;
 337                         }
 338
 339                         lock_page(page);
 340
 341                         if (unlikely(page->mapping != mapping)) {
 342 continue_unlock:
 343                                 unlock_page(page);
 344                                 continue;
 345                         }
 346                         if (!PageDirty(page)) {
 347                                 /* someone wrote it for us */
 348                                 goto continue_unlock;
 349                         }
 350
 351                         f2fs_wait_on_page_writeback(page, META, true);
 352
 353                         BUG_ON(PageWriteback(page));
 354                         if (!clear_page_dirty_for_io(page))
 355                                 goto continue_unlock;
 356
 357                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 358                                 unlock_page(page);
 359                                 break;
 360                         }
 361                         nwritten++;
 362                         prev = page->index;
 363                         if (unlikely(nwritten >= nr_to_write))
 364                                 break;
 365                 }
 366                 pagevec_release(&pvec);
 367                 cond_resched();
 368         }
 369 stop:
 370         if (nwritten)
 371                 f2fs_submit_merged_write(sbi, type);
 372
 373         blk_finish_plug(&plug);
 374
 375         return nwritten;
 376 }
 377
 378 static int f2fs_set_meta_page_dirty(struct page *page)
 379 {
 380         trace_f2fs_set_page_dirty(page, META);
 381
 382         if (!PageUptodate(page))
 383                 SetPageUptodate(page);
 384         if (!PageDirty(page)) {
 385                 f2fs_set_page_dirty_nobuffers(page);
 386                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 387                 SetPagePrivate(page);
 388                 f2fs_trace_pid(page);
 389                 return 1;
 390         }
 391         return 0;
 392 }
 393
 394 const struct address_space_operations f2fs_meta_aops = {
 395         .writepage      = f2fs_write_meta_page,
 396         .writepages     = f2fs_write_meta_pages,
 397         .set_page_dirty = f2fs_set_meta_page_dirty,
 398         .invalidatepage = f2fs_invalidate_page,
 399         .releasepage    = f2fs_release_page,
 400 #ifdef CONFIG_MIGRATION
 401         .migratepage    = f2fs_migrate_page,
 402 #endif
 403 };
 404
 405 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 406                                                 unsigned int devidx, int type)
 407 {
 408         struct inode_management *im = &sbi->im[type];
 409         struct ino_entry *e, *tmp;
 410
 411         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 412
 413         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 414
 415         spin_lock(&im->ino_lock);
 416         e = radix_tree_lookup(&im->ino_root, ino);
 417         if (!e) {
 418                 e = tmp;
 419                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 420                         f2fs_bug_on(sbi, 1);
 421
 422                 memset(e, 0, sizeof(struct ino_entry));
 423                 e->ino = ino;
 424
 425                 list_add_tail(&e->list, &im->ino_list);
 426                 if (type != ORPHAN_INO)
 427                         im->ino_num++;
 428         }
 429
 430         if (type == FLUSH_INO)
 431                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 432
 433         spin_unlock(&im->ino_lock);
 434         radix_tree_preload_end();
 435
 436         if (e != tmp)
 437                 kmem_cache_free(ino_entry_slab, tmp);
 438 }
 439
 440 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 441 {
 442         struct inode_management *im = &sbi->im[type];
 443         struct ino_entry *e;
 444
 445         spin_lock(&im->ino_lock);
 446         e = radix_tree_lookup(&im->ino_root, ino);
 447         if (e) {
 448                 list_del(&e->list);
 449                 radix_tree_delete(&im->ino_root, ino);
 450                 im->ino_num--;
 451                 spin_unlock(&im->ino_lock);
 452                 kmem_cache_free(ino_entry_slab, e);
 453                 return;
 454         }
 455         spin_unlock(&im->ino_lock);
 456 }
 457
 458 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 459 {
 460         /* add new dirty ino entry into list */
 461         __add_ino_entry(sbi, ino, 0, type);
 462 }
 463
 464 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 465 {
 466         /* remove dirty ino entry from list */
 467         __remove_ino_entry(sbi, ino, type);
 468 }
 469
 470 /* mode should be APPEND_INO or UPDATE_INO */
 471 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 472 {
 473         struct inode_management *im = &sbi->im[mode];
 474         struct ino_entry *e;
 475
 476         spin_lock(&im->ino_lock);
 477         e = radix_tree_lookup(&im->ino_root, ino);
 478         spin_unlock(&im->ino_lock);
 479         return e ? true : false;
 480 }
 481
 482 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 483 {
 484         struct ino_entry *e, *tmp;
 485         int i;
 486
 487         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 488                 struct inode_management *im = &sbi->im[i];
 489
 490                 spin_lock(&im->ino_lock);
 491                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 492                         list_del(&e->list);
 493                         radix_tree_delete(&im->ino_root, e->ino);
 494                         kmem_cache_free(ino_entry_slab, e);
 495                         im->ino_num--;
 496                 }
 497                 spin_unlock(&im->ino_lock);
 498         }
 499 }
 500
 501 void set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 502                                         unsigned int devidx, int type)
 503 {
 504         __add_ino_entry(sbi, ino, devidx, type);
 505 }
 506
 507 bool is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 508                                         unsigned int devidx, int type)
 509 {
 510         struct inode_management *im = &sbi->im[type];
 511         struct ino_entry *e;
 512         bool is_dirty = false;
 513
 514         spin_lock(&im->ino_lock);
 515         e = radix_tree_lookup(&im->ino_root, ino);
 516         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 517                 is_dirty = true;
 518         spin_unlock(&im->ino_lock);
 519         return is_dirty;
 520 }
 521
 522 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 523 {
 524         struct inode_management *im = &sbi->im[ORPHAN_INO];
 525         int err = 0;
 526
 527         spin_lock(&im->ino_lock);
 528
 529 #ifdef CONFIG_F2FS_FAULT_INJECTION
 530         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 531                 spin_unlock(&im->ino_lock);
 532                 f2fs_show_injection_info(FAULT_ORPHAN);
 533                 return -ENOSPC;
 534         }
 535 #endif
 536         if (unlikely(im->ino_num >= sbi->max_orphans))
 537                 err = -ENOSPC;
 538         else
 539                 im->ino_num++;
 540         spin_unlock(&im->ino_lock);
 541
 542         return err;
 543 }
 544
 545 void release_orphan_inode(struct f2fs_sb_info *sbi)
 546 {
 547         struct inode_management *im = &sbi->im[ORPHAN_INO];
 548
 549         spin_lock(&im->ino_lock);
 550         f2fs_bug_on(sbi, im->ino_num == 0);
 551         im->ino_num--;
 552         spin_unlock(&im->ino_lock);
 553 }
 554
 555 void add_orphan_inode(struct inode *inode)
 556 {
 557         /* add new orphan ino entry into list */
 558         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 559         update_inode_page(inode);
 560 }
 561
 562 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 563 {
 564         /* remove orphan entry from orphan list */
 565         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 566 }
 567
 568 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 569 {
 570         struct inode *inode;
 571         struct node_info ni;
 572         int err = acquire_orphan_inode(sbi);
 573
 574         if (err)
 575                 goto err_out;
 576
 577         __add_ino_entry(sbi, ino, 0, ORPHAN_INO);
 578
 579         inode = f2fs_iget_retry(sbi->sb, ino);
 580         if (IS_ERR(inode)) {
 581                 /*
 582                  * there should be a bug that we can't find the entry
 583                  * to orphan inode.
 584                  */
 585                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 586                 return PTR_ERR(inode);
 587         }
 588
 589         err = dquot_initialize(inode);
 590         if (err)
 591                 goto err_out;
 592
 593         dquot_initialize(inode);
 594         clear_nlink(inode);
 595
 596         /* truncate all the data during iput */
 597         iput(inode);
 598
 599         get_node_info(sbi, ino, &ni);
 600
 601         /* ENOMEM was fully retried in f2fs_evict_inode. */
 602         if (ni.blk_addr != NULL_ADDR) {
 603                 err = -EIO;
 604                 goto err_out;
 605         }
 606         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 607         return 0;
 608
 609 err_out:
 610         set_sbi_flag(sbi, SBI_NEED_FSCK);
 611         f2fs_msg(sbi->sb, KERN_WARNING,
 612                         "%s: orphan failed (ino=%x), run fsck to fix.",
 613                         __func__, ino);
 614         return err;
 615 }
 616
 617 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 618 {
 619         block_t start_blk, orphan_blocks, i, j;
 620         unsigned int s_flags = sbi->sb->s_flags;
 621         int err = 0;
 622 #ifdef CONFIG_QUOTA
 623         int quota_enabled;
 624 #endif
 625
 626         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 627                 return 0;
 628
 629         if (s_flags & SB_RDONLY) {
 630                 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
 631                 sbi->sb->s_flags &= ~SB_RDONLY;
 632         }
 633
 634 #ifdef CONFIG_QUOTA
 635         /* Needed for iput() to work correctly and not trash data */
 636         sbi->sb->s_flags |= SB_ACTIVE;
 637
 638         /* Turn on quotas so that they are updated correctly */
 639         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 640 #endif
 641
 642         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 643         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 644
 645         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 646
 647         for (i = 0; i < orphan_blocks; i++) {
 648                 struct page *page = get_meta_page(sbi, start_blk + i);
 649                 struct f2fs_orphan_block *orphan_blk;
 650
 651                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 652                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 653                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 654                         err = recover_orphan_inode(sbi, ino);
 655                         if (err) {
 656                                 f2fs_put_page(page, 1);
 657                                 goto out;
 658                         }
 659                 }
 660                 f2fs_put_page(page, 1);
 661         }
 662         /* clear Orphan Flag */
 663         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 664 out:
 665 #ifdef CONFIG_QUOTA
 666         /* Turn quotas off */
 667         if (quota_enabled)
 668                 f2fs_quota_off_umount(sbi->sb);
 669 #endif
 670         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 671
 672         return err;
 673 }
 674
 675 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 676 {
 677         struct list_head *head;
 678         struct f2fs_orphan_block *orphan_blk = NULL;
 679         unsigned int nentries = 0;
 680         unsigned short index = 1;
 681         unsigned short orphan_blocks;
 682         struct page *page = NULL;
 683         struct ino_entry *orphan = NULL;
 684         struct inode_management *im = &sbi->im[ORPHAN_INO];
 685
 686         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 687
 688         /*
 689          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 690          * orphan inode operations are covered under f2fs_lock_op().
 691          * And, spin_lock should be avoided due to page operations below.
 692          */
 693         head = &im->ino_list;
 694
 695         /* loop for each orphan inode entry and write them in Jornal block */
 696         list_for_each_entry(orphan, head, list) {
 697                 if (!page) {
 698                         page = grab_meta_page(sbi, start_blk++);
 699                         orphan_blk =
 700                                 (struct f2fs_orphan_block *)page_address(page);
 701                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 702                 }
 703
 704                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 705
 706                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 707                         /*
 708                          * an orphan block is full of 1020 entries,
 709                          * then we need to flush current orphan blocks
 710                          * and bring another one in memory
 711                          */
 712                         orphan_blk->blk_addr = cpu_to_le16(index);
 713                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 714                         orphan_blk->entry_count = cpu_to_le32(nentries);
 715                         set_page_dirty(page);
 716                         f2fs_put_page(page, 1);
 717                         index++;
 718                         nentries = 0;
 719                         page = NULL;
 720                 }
 721         }
 722
 723         if (page) {
 724                 orphan_blk->blk_addr = cpu_to_le16(index);
 725                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 726                 orphan_blk->entry_count = cpu_to_le32(nentries);
 727                 set_page_dirty(page);
 728                 f2fs_put_page(page, 1);
 729         }
 730 }
 731
 732 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 733                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 734                 unsigned long long *version)
 735 {
 736         unsigned long blk_size = sbi->blocksize;
 737         size_t crc_offset = 0;
 738         __u32 crc = 0;
 739
 740         *cp_page = get_meta_page(sbi, cp_addr);
 741         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 742
 743         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 744         if (crc_offset > (blk_size - sizeof(__le32))) {
 745                 f2fs_msg(sbi->sb, KERN_WARNING,
 746                         "invalid crc_offset: %zu", crc_offset);
 747                 return -EINVAL;
 748         }
 749
 750         crc = cur_cp_crc(*cp_block);
 751         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 752                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 753                 return -EINVAL;
 754         }
 755
 756         *version = cur_cp_version(*cp_block);
 757         return 0;
 758 }
 759
 760 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 761                                 block_t cp_addr, unsigned long long *version)
 762 {
 763         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 764         struct f2fs_checkpoint *cp_block = NULL;
 765         unsigned long long cur_version = 0, pre_version = 0;
 766         int err;
 767
 768         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 769                                         &cp_page_1, version);
 770         if (err)
 771                 goto invalid_cp1;
 772         pre_version = *version;
 773
 774         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 775         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 776                                         &cp_page_2, version);
 777         if (err)
 778                 goto invalid_cp2;
 779         cur_version = *version;
 780
 781         if (cur_version == pre_version) {
 782                 *version = cur_version;
 783                 f2fs_put_page(cp_page_2, 1);
 784                 return cp_page_1;
 785         }
 786 invalid_cp2:
 787         f2fs_put_page(cp_page_2, 1);
 788 invalid_cp1:
 789         f2fs_put_page(cp_page_1, 1);
 790         return NULL;
 791 }
 792
 793 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 794 {
 795         struct f2fs_checkpoint *cp_block;
 796         struct f2fs_super_block *fsb = sbi->raw_super;
 797         struct page *cp1, *cp2, *cur_page;
 798         unsigned long blk_size = sbi->blocksize;
 799         unsigned long long cp1_version = 0, cp2_version = 0;
 800         unsigned long long cp_start_blk_no;
 801         unsigned int cp_blks = 1 + __cp_payload(sbi);
 802         block_t cp_blk_no;
 803         int i;
 804
 805         sbi->ckpt = f2fs_kzalloc(sbi, cp_blks * blk_size, GFP_KERNEL);
 806         if (!sbi->ckpt)
 807                 return -ENOMEM;
 808         /*
 809          * Finding out valid cp block involves read both
 810          * sets( cp pack1 and cp pack 2)
 811          */
 812         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 813         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 814
 815         /* The second checkpoint pack should start at the next segment */
 816         cp_start_blk_no += ((unsigned long long)1) <<
 817                                 le32_to_cpu(fsb->log_blocks_per_seg);
 818         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 819
 820         if (cp1 && cp2) {
 821                 if (ver_after(cp2_version, cp1_version))
 822                         cur_page = cp2;
 823                 else
 824                         cur_page = cp1;
 825         } else if (cp1) {
 826                 cur_page = cp1;
 827         } else if (cp2) {
 828                 cur_page = cp2;
 829         } else {
 830                 goto fail_no_cp;
 831         }
 832
 833         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 834         memcpy(sbi->ckpt, cp_block, blk_size);
 835
 836         /* Sanity checking of checkpoint */
 837         if (sanity_check_ckpt(sbi))
 838                 goto free_fail_no_cp;
 839
 840         if (cur_page == cp1)
 841                 sbi->cur_cp_pack = 1;
 842         else
 843                 sbi->cur_cp_pack = 2;
 844
 845         if (cp_blks <= 1)
 846                 goto done;
 847
 848         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 849         if (cur_page == cp2)
 850                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 851
 852         for (i = 1; i < cp_blks; i++) {
 853                 void *sit_bitmap_ptr;
 854                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 855
 856                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 857                 sit_bitmap_ptr = page_address(cur_page);
 858                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 859                 f2fs_put_page(cur_page, 1);
 860         }
 861 done:
 862         f2fs_put_page(cp1, 1);
 863         f2fs_put_page(cp2, 1);
 864         return 0;
 865
 866 free_fail_no_cp:
 867         f2fs_put_page(cp1, 1);
 868         f2fs_put_page(cp2, 1);
 869 fail_no_cp:
 870         kfree(sbi->ckpt);
 871         return -EINVAL;
 872 }
 873
 874 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 875 {
 876         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 877         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 878
 879         if (is_inode_flag_set(inode, flag))
 880                 return;
 881
 882         set_inode_flag(inode, flag);
 883         if (!f2fs_is_volatile_file(inode))
 884                 list_add_tail(&F2FS_I(inode)->dirty_list,
 885                                                 &sbi->inode_list[type]);
 886         stat_inc_dirty_inode(sbi, type);
 887 }
 888
 889 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 890 {
 891         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 892
 893         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 894                 return;
 895
 896         list_del_init(&F2FS_I(inode)->dirty_list);
 897         clear_inode_flag(inode, flag);
 898         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 899 }
 900
 901 void update_dirty_page(struct inode *inode, struct page *page)
 902 {
 903         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 904         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 905
 906         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 907                         !S_ISLNK(inode->i_mode))
 908                 return;
 909
 910         spin_lock(&sbi->inode_lock[type]);
 911         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 912                 __add_dirty_inode(inode, type);
 913         inode_inc_dirty_pages(inode);
 914         spin_unlock(&sbi->inode_lock[type]);
 915
 916         SetPagePrivate(page);
 917         f2fs_trace_pid(page);
 918 }
 919
 920 void remove_dirty_inode(struct inode *inode)
 921 {
 922         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 923         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 924
 925         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 926                         !S_ISLNK(inode->i_mode))
 927                 return;
 928
 929         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 930                 return;
 931
 932         spin_lock(&sbi->inode_lock[type]);
 933         __remove_dirty_inode(inode, type);
 934         spin_unlock(&sbi->inode_lock[type]);
 935 }
 936
 937 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 938 {
 939         struct list_head *head;
 940         struct inode *inode;
 941         struct f2fs_inode_info *fi;
 942         bool is_dir = (type == DIR_INODE);
 943         unsigned long ino = 0;
 944
 945         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 946                                 get_pages(sbi, is_dir ?
 947                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 948 retry:
 949         if (unlikely(f2fs_cp_error(sbi)))
 950                 return -EIO;
 951
 952         spin_lock(&sbi->inode_lock[type]);
 953
 954         head = &sbi->inode_list[type];
 955         if (list_empty(head)) {
 956                 spin_unlock(&sbi->inode_lock[type]);
 957                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 958                                 get_pages(sbi, is_dir ?
 959                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 960                 return 0;
 961         }
 962         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
 963         inode = igrab(&fi->vfs_inode);
 964         spin_unlock(&sbi->inode_lock[type]);
 965         if (inode) {
 966                 unsigned long cur_ino = inode->i_ino;
 967
 968                 if (is_dir)
 969                         F2FS_I(inode)->cp_task = current;
 970
 971                 filemap_fdatawrite(inode->i_mapping);
 972
 973                 if (is_dir)
 974                         F2FS_I(inode)->cp_task = NULL;
 975
 976                 iput(inode);
 977                 /* We need to give cpu to another writers. */
 978                 if (ino == cur_ino) {
 979                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 980                         cond_resched();
 981                 } else {
 982                         ino = cur_ino;
 983                 }
 984         } else {
 985                 /*
 986                  * We should submit bio, since it exists several
 987                  * wribacking dentry pages in the freeing inode.
 988                  */
 989                 f2fs_submit_merged_write(sbi, DATA);
 990                 cond_resched();
 991         }
 992         goto retry;
 993 }
 994
 995 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 996 {
 997         struct list_head *head = &sbi->inode_list[DIRTY_META];
 998         struct inode *inode;
 999         struct f2fs_inode_info *fi;
1000         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1001
1002         while (total--) {
1003                 if (unlikely(f2fs_cp_error(sbi)))
1004                         return -EIO;
1005
1006                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1007                 if (list_empty(head)) {
1008                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1009                         return 0;
1010                 }
1011                 fi = list_first_entry(head, struct f2fs_inode_info,
1012                                                         gdirty_list);
1013                 inode = igrab(&fi->vfs_inode);
1014                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1015                 if (inode) {
1016                         sync_inode_metadata(inode, 0);
1017
1018                         /* it's on eviction */
1019                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1020                                 update_inode_page(inode);
1021                         iput(inode);
1022                 }
1023         }
1024         return 0;
1025 }
1026
1027 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1028 {
1029         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1030         struct f2fs_nm_info *nm_i = NM_I(sbi);
1031         nid_t last_nid = nm_i->next_scan_nid;
1032
1033         next_free_nid(sbi, &last_nid);
1034         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1035         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1036         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1037         ckpt->next_free_nid = cpu_to_le32(last_nid);
1038 }
1039
1040 /*
1041  * Freeze all the FS-operations for checkpoint.
1042  */
1043 static int block_operations(struct f2fs_sb_info *sbi)
1044 {
1045         struct writeback_control wbc = {
1046                 .sync_mode = WB_SYNC_ALL,
1047                 .nr_to_write = LONG_MAX,
1048                 .for_reclaim = 0,
1049         };
1050         struct blk_plug plug;
1051         int err = 0;
1052
1053         blk_start_plug(&plug);
1054
1055 retry_flush_dents:
1056         f2fs_lock_all(sbi);
1057         /* write all the dirty dentry pages */
1058         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1059                 f2fs_unlock_all(sbi);
1060                 err = sync_dirty_inodes(sbi, DIR_INODE);
1061                 if (err)
1062                         goto out;
1063                 cond_resched();
1064                 goto retry_flush_dents;
1065         }
1066
1067         /*
1068          * POR: we should ensure that there are no dirty node pages
1069          * until finishing nat/sit flush. inode->i_blocks can be updated.
1070          */
1071         down_write(&sbi->node_change);
1072
1073         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1074                 up_write(&sbi->node_change);
1075                 f2fs_unlock_all(sbi);
1076                 err = f2fs_sync_inode_meta(sbi);
1077                 if (err)
1078                         goto out;
1079                 cond_resched();
1080                 goto retry_flush_dents;
1081         }
1082
1083 retry_flush_nodes:
1084         down_write(&sbi->node_write);
1085
1086         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1087                 up_write(&sbi->node_write);
1088                 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1089                 if (err) {
1090                         up_write(&sbi->node_change);
1091                         f2fs_unlock_all(sbi);
1092                         goto out;
1093                 }
1094                 cond_resched();
1095                 goto retry_flush_nodes;
1096         }
1097
1098         /*
1099          * sbi->node_change is used only for AIO write_begin path which produces
1100          * dirty node blocks and some checkpoint values by block allocation.
1101          */
1102         __prepare_cp_block(sbi);
1103         up_write(&sbi->node_change);
1104 out:
1105         blk_finish_plug(&plug);
1106         return err;
1107 }
1108
1109 static void unblock_operations(struct f2fs_sb_info *sbi)
1110 {
1111         up_write(&sbi->node_write);
1112         f2fs_unlock_all(sbi);
1113 }
1114
1115 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1116 {
1117         DEFINE_WAIT(wait);
1118
1119         for (;;) {
1120                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1121
1122                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1123                         break;
1124
1125                 io_schedule_timeout(5*HZ);
1126         }
1127         finish_wait(&sbi->cp_wait, &wait);
1128 }
1129
1130 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1131 {
1132         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1133         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1134         unsigned long flags;
1135
1136         spin_lock_irqsave(&sbi->cp_lock, flags);
1137
1138         if ((cpc->reason & CP_UMOUNT) &&
1139                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1140                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1141                 disable_nat_bits(sbi, false);
1142
1143         if (cpc->reason & CP_TRIMMED)
1144                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1145         else
1146                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1147
1148         if (cpc->reason & CP_UMOUNT)
1149                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1150         else
1151                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1152
1153         if (cpc->reason & CP_FASTBOOT)
1154                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1155         else
1156                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1157
1158         if (orphan_num)
1159                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1160         else
1161                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1162
1163         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1164                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1165
1166         /* set this flag to activate crc|cp_ver for recovery */
1167         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1168         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1169
1170         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1171 }
1172
1173 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1174         void *src, block_t blk_addr)
1175 {
1176         struct writeback_control wbc = {
1177                 .for_reclaim = 0,
1178         };
1179
1180         /*
1181          * pagevec_lookup_tag and lock_page again will take
1182          * some extra time. Therefore, update_meta_pages and
1183          * sync_meta_pages are combined in this function.
1184          */
1185         struct page *page = grab_meta_page(sbi, blk_addr);
1186         int err;
1187
1188         memcpy(page_address(page), src, PAGE_SIZE);
1189         set_page_dirty(page);
1190
1191         f2fs_wait_on_page_writeback(page, META, true);
1192         f2fs_bug_on(sbi, PageWriteback(page));
1193         if (unlikely(!clear_page_dirty_for_io(page)))
1194                 f2fs_bug_on(sbi, 1);
1195
1196         /* writeout cp pack 2 page */
1197         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1198         f2fs_bug_on(sbi, err);
1199
1200         f2fs_put_page(page, 0);
1201
1202         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1203         f2fs_submit_merged_write(sbi, META_FLUSH);
1204 }
1205
1206 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1207 {
1208         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1209         struct f2fs_nm_info *nm_i = NM_I(sbi);
1210         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1211         block_t start_blk;
1212         unsigned int data_sum_blocks, orphan_blocks;
1213         __u32 crc32 = 0;
1214         int i;
1215         int cp_payload_blks = __cp_payload(sbi);
1216         struct super_block *sb = sbi->sb;
1217         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1218         u64 kbytes_written;
1219         int err;
1220
1221         /* Flush all the NAT/SIT pages */
1222         while (get_pages(sbi, F2FS_DIRTY_META)) {
1223                 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1224                 if (unlikely(f2fs_cp_error(sbi)))
1225                         return -EIO;
1226         }
1227
1228         /*
1229          * modify checkpoint
1230          * version number is already updated
1231          */
1232         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1233         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1234         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1235                 ckpt->cur_node_segno[i] =
1236                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1237                 ckpt->cur_node_blkoff[i] =
1238                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1239                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1240                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1241         }
1242         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1243                 ckpt->cur_data_segno[i] =
1244                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1245                 ckpt->cur_data_blkoff[i] =
1246                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1247                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1248                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1249         }
1250
1251         /* 2 cp  + n data seg summary + orphan inode blocks */
1252         data_sum_blocks = npages_for_summary_flush(sbi, false);
1253         spin_lock_irqsave(&sbi->cp_lock, flags);
1254         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1255                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1256         else
1257                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1258         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1259
1260         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1261         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1262                         orphan_blocks);
1263
1264         if (__remain_node_summaries(cpc->reason))
1265                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1266                                 cp_payload_blks + data_sum_blocks +
1267                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1268         else
1269                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1270                                 cp_payload_blks + data_sum_blocks +
1271                                 orphan_blocks);
1272
1273         /* update ckpt flag for checkpoint */
1274         update_ckpt_flags(sbi, cpc);
1275
1276         /* update SIT/NAT bitmap */
1277         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1278         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1279
1280         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1281         *((__le32 *)((unsigned char *)ckpt +
1282                                 le32_to_cpu(ckpt->checksum_offset)))
1283                                 = cpu_to_le32(crc32);
1284
1285         start_blk = __start_cp_next_addr(sbi);
1286
1287         /* write nat bits */
1288         if (enabled_nat_bits(sbi, cpc)) {
1289                 __u64 cp_ver = cur_cp_version(ckpt);
1290                 block_t blk;
1291
1292                 cp_ver |= ((__u64)crc32 << 32);
1293                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1294
1295                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1296                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1297                         update_meta_page(sbi, nm_i->nat_bits +
1298                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1299
1300                 /* Flush all the NAT BITS pages */
1301                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1302                         sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1303                         if (unlikely(f2fs_cp_error(sbi)))
1304                                 return -EIO;
1305                 }
1306         }
1307
1308         /* write out checkpoint buffer at block 0 */
1309         update_meta_page(sbi, ckpt, start_blk++);
1310
1311         for (i = 1; i < 1 + cp_payload_blks; i++)
1312                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1313                                                         start_blk++);
1314
1315         if (orphan_num) {
1316                 write_orphan_inodes(sbi, start_blk);
1317                 start_blk += orphan_blocks;
1318         }
1319
1320         write_data_summaries(sbi, start_blk);
1321         start_blk += data_sum_blocks;
1322
1323         /* Record write statistics in the hot node summary */
1324         kbytes_written = sbi->kbytes_written;
1325         if (sb->s_bdev->bd_part)
1326                 kbytes_written += BD_PART_WRITTEN(sbi);
1327
1328         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1329
1330         if (__remain_node_summaries(cpc->reason)) {
1331                 write_node_summaries(sbi, start_blk);
1332                 start_blk += NR_CURSEG_NODE_TYPE;
1333         }
1334
1335         /* update user_block_counts */
1336         sbi->last_valid_block_count = sbi->total_valid_block_count;
1337         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1338
1339         /* Here, we have one bio having CP pack except cp pack 2 page */
1340         sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1341
1342         /* wait for previous submitted meta pages writeback */
1343         wait_on_all_pages_writeback(sbi);
1344
1345         if (unlikely(f2fs_cp_error(sbi)))
1346                 return -EIO;
1347
1348         /* flush all device cache */
1349         err = f2fs_flush_device_cache(sbi);
1350         if (err)
1351                 return err;
1352
1353         /* barrier and flush checkpoint cp pack 2 page if it can */
1354         commit_checkpoint(sbi, ckpt, start_blk);
1355         wait_on_all_pages_writeback(sbi);
1356
1357         release_ino_entry(sbi, false);
1358
1359         if (unlikely(f2fs_cp_error(sbi)))
1360                 return -EIO;
1361
1362         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1363         clear_sbi_flag(sbi, SBI_NEED_CP);
1364         __set_cp_next_pack(sbi);
1365
1366         /*
1367          * redirty superblock if metadata like node page or inode cache is
1368          * updated during writing checkpoint.
1369          */
1370         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1371                         get_pages(sbi, F2FS_DIRTY_IMETA))
1372                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1373
1374         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1375
1376         return 0;
1377 }
1378
1379 /*
1380  * We guarantee that this checkpoint procedure will not fail.
1381  */
1382 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1383 {
1384         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1385         unsigned long long ckpt_ver;
1386         int err = 0;
1387
1388         mutex_lock(&sbi->cp_mutex);
1389
1390         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1391                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1392                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1393                 goto out;
1394         if (unlikely(f2fs_cp_error(sbi))) {
1395                 err = -EIO;
1396                 goto out;
1397         }
1398         if (f2fs_readonly(sbi->sb)) {
1399                 err = -EROFS;
1400                 goto out;
1401         }
1402
1403         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1404
1405         err = block_operations(sbi);
1406         if (err)
1407                 goto out;
1408
1409         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1410
1411         f2fs_flush_merged_writes(sbi);
1412
1413         /* this is the case of multiple fstrims without any changes */
1414         if (cpc->reason & CP_DISCARD) {
1415                 if (!exist_trim_candidates(sbi, cpc)) {
1416                         unblock_operations(sbi);
1417                         goto out;
1418                 }
1419
1420                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1421                                 SIT_I(sbi)->dirty_sentries == 0 &&
1422                                 prefree_segments(sbi) == 0) {
1423                         flush_sit_entries(sbi, cpc);
1424                         clear_prefree_segments(sbi, cpc);
1425                         unblock_operations(sbi);
1426                         goto out;
1427                 }
1428         }
1429
1430         /*
1431          * update checkpoint pack index
1432          * Increase the version number so that
1433          * SIT entries and seg summaries are written at correct place
1434          */
1435         ckpt_ver = cur_cp_version(ckpt);
1436         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1437
1438         /* write cached NAT/SIT entries to NAT/SIT area */
1439         flush_nat_entries(sbi, cpc);
1440         flush_sit_entries(sbi, cpc);
1441
1442         /* unlock all the fs_lock[] in do_checkpoint() */
1443         err = do_checkpoint(sbi, cpc);
1444         if (err)
1445                 release_discard_addrs(sbi);
1446         else
1447                 clear_prefree_segments(sbi, cpc);
1448
1449         unblock_operations(sbi);
1450         stat_inc_cp_count(sbi->stat_info);
1451
1452         if (cpc->reason & CP_RECOVERY)
1453                 f2fs_msg(sbi->sb, KERN_NOTICE,
1454                         "checkpoint: version = %llx", ckpt_ver);
1455
1456         /* do checkpoint periodically */
1457         f2fs_update_time(sbi, CP_TIME);
1458         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1459 out:
1460         mutex_unlock(&sbi->cp_mutex);
1461         return err;
1462 }
1463
1464 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1465 {
1466         int i;
1467
1468         for (i = 0; i < MAX_INO_ENTRY; i++) {
1469                 struct inode_management *im = &sbi->im[i];
1470
1471                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1472                 spin_lock_init(&im->ino_lock);
1473                 INIT_LIST_HEAD(&im->ino_list);
1474                 im->ino_num = 0;
1475         }
1476
1477         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1478                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1479                                 F2FS_ORPHANS_PER_BLOCK;
1480 }
1481
1482 int __init create_checkpoint_caches(void)
1483 {
1484         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1485                         sizeof(struct ino_entry));
1486         if (!ino_entry_slab)
1487                 return -ENOMEM;
1488         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1489                         sizeof(struct inode_entry));
1490         if (!inode_entry_slab) {
1491                 kmem_cache_destroy(ino_entry_slab);
1492                 return -ENOMEM;
1493         }
1494         return 0;
1495 }
1496
1497 void destroy_checkpoint_caches(void)
1498 {
1499         kmem_cache_destroy(ino_entry_slab);
1500         kmem_cache_destroy(inode_entry_slab);
1501 }