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                         if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 222                                 goto out;
 223                         /* get sit block addr */
 224                         fio.new_blkaddr = current_sit_addr(sbi,
 225                                         blkno * SIT_ENTRY_PER_BLOCK);
 226                         break;
 227                 case META_SSA:
 228                 case META_CP:
 229                 case META_POR:
 230                         fio.new_blkaddr = blkno;
 231                         break;
 232                 default:
 233                         BUG();
 234                 }
 235
 236                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 237                                                 fio.new_blkaddr, false);
 238                 if (!page)
 239                         continue;
 240                 if (PageUptodate(page)) {
 241                         f2fs_put_page(page, 1);
 242                         continue;
 243                 }
 244
 245                 fio.page = page;
 246                 f2fs_submit_page_bio(&fio);
 247                 f2fs_put_page(page, 0);
 248         }
 249 out:
 250         blk_finish_plug(&plug);
 251         return blkno - start;
 252 }
 253
 254 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 255 {
 256         struct page *page;
 257         bool readahead = false;
 258
 259         page = find_get_page(META_MAPPING(sbi), index);
 260         if (!page || !PageUptodate(page))
 261                 readahead = true;
 262         f2fs_put_page(page, 0);
 263
 264         if (readahead)
 265                 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 266 }
 267
 268 static int __f2fs_write_meta_page(struct page *page,
 269                                 struct writeback_control *wbc,
 270                                 enum iostat_type io_type)
 271 {
 272         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 273
 274         trace_f2fs_writepage(page, META);
 275
 276         if (unlikely(f2fs_cp_error(sbi))) {
 277                 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
 278                         ClearPageUptodate(page);
 279                         dec_page_count(sbi, F2FS_DIRTY_META);
 280                         unlock_page(page);
 281                         return 0;
 282                 }
 283                 goto redirty_out;
 284         }
 285         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 286                 goto redirty_out;
 287         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 288                 goto redirty_out;
 289
 290         f2fs_do_write_meta_page(sbi, page, io_type);
 291         dec_page_count(sbi, F2FS_DIRTY_META);
 292
 293         if (wbc->for_reclaim)
 294                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
 295                                                 0, page->index, META);
 296
 297         unlock_page(page);
 298
 299         if (unlikely(f2fs_cp_error(sbi)))
 300                 f2fs_submit_merged_write(sbi, META);
 301
 302         return 0;
 303
 304 redirty_out:
 305         redirty_page_for_writepage(wbc, page);
 306         return AOP_WRITEPAGE_ACTIVATE;
 307 }
 308
 309 static int f2fs_write_meta_page(struct page *page,
 310                                 struct writeback_control *wbc)
 311 {
 312         return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 313 }
 314
 315 static int f2fs_write_meta_pages(struct address_space *mapping,
 316                                 struct writeback_control *wbc)
 317 {
 318         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 319         long diff, written;
 320
 321         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 322                 goto skip_write;
 323
 324         /* collect a number of dirty meta pages and write together */
 325         if (wbc->for_kupdate ||
 326                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 327                 goto skip_write;
 328
 329         /* if locked failed, cp will flush dirty pages instead */
 330         if (!mutex_trylock(&sbi->cp_mutex))
 331                 goto skip_write;
 332
 333         trace_f2fs_writepages(mapping->host, wbc, META);
 334         diff = nr_pages_to_write(sbi, META, wbc);
 335         written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 336         mutex_unlock(&sbi->cp_mutex);
 337         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 338         return 0;
 339
 340 skip_write:
 341         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 342         trace_f2fs_writepages(mapping->host, wbc, META);
 343         return 0;
 344 }
 345
 346 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 347                                 long nr_to_write, enum iostat_type io_type)
 348 {
 349         struct address_space *mapping = META_MAPPING(sbi);
 350         pgoff_t index = 0, prev = ULONG_MAX;
 351         struct pagevec pvec;
 352         long nwritten = 0;
 353         int nr_pages;
 354         struct writeback_control wbc = {
 355                 .for_reclaim = 0,
 356         };
 357         struct blk_plug plug;
 358
 359         pagevec_init(&pvec);
 360
 361         blk_start_plug(&plug);
 362
 363         while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 364                                 PAGECACHE_TAG_DIRTY))) {
 365                 int i;
 366
 367                 for (i = 0; i < nr_pages; i++) {
 368                         struct page *page = pvec.pages[i];
 369
 370                         if (prev == ULONG_MAX)
 371                                 prev = page->index - 1;
 372                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 373                                 pagevec_release(&pvec);
 374                                 goto stop;
 375                         }
 376
 377                         lock_page(page);
 378
 379                         if (unlikely(page->mapping != mapping)) {
 380 continue_unlock:
 381                                 unlock_page(page);
 382                                 continue;
 383                         }
 384                         if (!PageDirty(page)) {
 385                                 /* someone wrote it for us */
 386                                 goto continue_unlock;
 387                         }
 388
 389                         f2fs_wait_on_page_writeback(page, META, true);
 390
 391                         BUG_ON(PageWriteback(page));
 392                         if (!clear_page_dirty_for_io(page))
 393                                 goto continue_unlock;
 394
 395                         if (__f2fs_write_meta_page(page, &wbc, io_type)) {
 396                                 unlock_page(page);
 397                                 break;
 398                         }
 399                         nwritten++;
 400                         prev = page->index;
 401                         if (unlikely(nwritten >= nr_to_write))
 402                                 break;
 403                 }
 404                 pagevec_release(&pvec);
 405                 cond_resched();
 406         }
 407 stop:
 408         if (nwritten)
 409                 f2fs_submit_merged_write(sbi, type);
 410
 411         blk_finish_plug(&plug);
 412
 413         return nwritten;
 414 }
 415
 416 static int f2fs_set_meta_page_dirty(struct page *page)
 417 {
 418         trace_f2fs_set_page_dirty(page, META);
 419
 420         if (!PageUptodate(page))
 421                 SetPageUptodate(page);
 422         if (!PageDirty(page)) {
 423                 __set_page_dirty_nobuffers(page);
 424                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 425                 SetPagePrivate(page);
 426                 f2fs_trace_pid(page);
 427                 return 1;
 428         }
 429         return 0;
 430 }
 431
 432 const struct address_space_operations f2fs_meta_aops = {
 433         .writepage      = f2fs_write_meta_page,
 434         .writepages     = f2fs_write_meta_pages,
 435         .set_page_dirty = f2fs_set_meta_page_dirty,
 436         .invalidatepage = f2fs_invalidate_page,
 437         .releasepage    = f2fs_release_page,
 438 #ifdef CONFIG_MIGRATION
 439         .migratepage    = f2fs_migrate_page,
 440 #endif
 441 };
 442
 443 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 444                                                 unsigned int devidx, int type)
 445 {
 446         struct inode_management *im = &sbi->im[type];
 447         struct ino_entry *e, *tmp;
 448
 449         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 450
 451         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 452
 453         spin_lock(&im->ino_lock);
 454         e = radix_tree_lookup(&im->ino_root, ino);
 455         if (!e) {
 456                 e = tmp;
 457                 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 458                         f2fs_bug_on(sbi, 1);
 459
 460                 memset(e, 0, sizeof(struct ino_entry));
 461                 e->ino = ino;
 462
 463                 list_add_tail(&e->list, &im->ino_list);
 464                 if (type != ORPHAN_INO)
 465                         im->ino_num++;
 466         }
 467
 468         if (type == FLUSH_INO)
 469                 f2fs_set_bit(devidx, (char *)&e->dirty_device);
 470
 471         spin_unlock(&im->ino_lock);
 472         radix_tree_preload_end();
 473
 474         if (e != tmp)
 475                 kmem_cache_free(ino_entry_slab, tmp);
 476 }
 477
 478 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 479 {
 480         struct inode_management *im = &sbi->im[type];
 481         struct ino_entry *e;
 482
 483         spin_lock(&im->ino_lock);
 484         e = radix_tree_lookup(&im->ino_root, ino);
 485         if (e) {
 486                 list_del(&e->list);
 487                 radix_tree_delete(&im->ino_root, ino);
 488                 im->ino_num--;
 489                 spin_unlock(&im->ino_lock);
 490                 kmem_cache_free(ino_entry_slab, e);
 491                 return;
 492         }
 493         spin_unlock(&im->ino_lock);
 494 }
 495
 496 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 497 {
 498         /* add new dirty ino entry into list */
 499         __add_ino_entry(sbi, ino, 0, type);
 500 }
 501
 502 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 503 {
 504         /* remove dirty ino entry from list */
 505         __remove_ino_entry(sbi, ino, type);
 506 }
 507
 508 /* mode should be APPEND_INO or UPDATE_INO */
 509 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 510 {
 511         struct inode_management *im = &sbi->im[mode];
 512         struct ino_entry *e;
 513
 514         spin_lock(&im->ino_lock);
 515         e = radix_tree_lookup(&im->ino_root, ino);
 516         spin_unlock(&im->ino_lock);
 517         return e ? true : false;
 518 }
 519
 520 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 521 {
 522         struct ino_entry *e, *tmp;
 523         int i;
 524
 525         for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 526                 struct inode_management *im = &sbi->im[i];
 527
 528                 spin_lock(&im->ino_lock);
 529                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 530                         list_del(&e->list);
 531                         radix_tree_delete(&im->ino_root, e->ino);
 532                         kmem_cache_free(ino_entry_slab, e);
 533                         im->ino_num--;
 534                 }
 535                 spin_unlock(&im->ino_lock);
 536         }
 537 }
 538
 539 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 540                                         unsigned int devidx, int type)
 541 {
 542         __add_ino_entry(sbi, ino, devidx, type);
 543 }
 544
 545 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 546                                         unsigned int devidx, int type)
 547 {
 548         struct inode_management *im = &sbi->im[type];
 549         struct ino_entry *e;
 550         bool is_dirty = false;
 551
 552         spin_lock(&im->ino_lock);
 553         e = radix_tree_lookup(&im->ino_root, ino);
 554         if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 555                 is_dirty = true;
 556         spin_unlock(&im->ino_lock);
 557         return is_dirty;
 558 }
 559
 560 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 561 {
 562         struct inode_management *im = &sbi->im[ORPHAN_INO];
 563         int err = 0;
 564
 565         spin_lock(&im->ino_lock);
 566
 567         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 568                 spin_unlock(&im->ino_lock);
 569                 f2fs_show_injection_info(FAULT_ORPHAN);
 570                 return -ENOSPC;
 571         }
 572
 573         if (unlikely(im->ino_num >= sbi->max_orphans))
 574                 err = -ENOSPC;
 575         else
 576                 im->ino_num++;
 577         spin_unlock(&im->ino_lock);
 578
 579         return err;
 580 }
 581
 582 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 583 {
 584         struct inode_management *im = &sbi->im[ORPHAN_INO];
 585
 586         spin_lock(&im->ino_lock);
 587         f2fs_bug_on(sbi, im->ino_num == 0);
 588         im->ino_num--;
 589         spin_unlock(&im->ino_lock);
 590 }
 591
 592 void f2fs_add_orphan_inode(struct inode *inode)
 593 {
 594         /* add new orphan ino entry into list */
 595         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 596         f2fs_update_inode_page(inode);
 597 }
 598
 599 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 600 {
 601         /* remove orphan entry from orphan list */
 602         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 603 }
 604
 605 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 606 {
 607         struct inode *inode;
 608         struct node_info ni;
 609         int err;
 610
 611         inode = f2fs_iget_retry(sbi->sb, ino);
 612         if (IS_ERR(inode)) {
 613                 /*
 614                  * there should be a bug that we can't find the entry
 615                  * to orphan inode.
 616                  */
 617                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 618                 return PTR_ERR(inode);
 619         }
 620
 621         err = dquot_initialize(inode);
 622         if (err) {
 623                 iput(inode);
 624                 goto err_out;
 625         }
 626
 627         clear_nlink(inode);
 628
 629         /* truncate all the data during iput */
 630         iput(inode);
 631
 632         err = f2fs_get_node_info(sbi, ino, &ni);
 633         if (err)
 634                 goto err_out;
 635
 636         /* ENOMEM was fully retried in f2fs_evict_inode. */
 637         if (ni.blk_addr != NULL_ADDR) {
 638                 err = -EIO;
 639                 goto err_out;
 640         }
 641         return 0;
 642
 643 err_out:
 644         set_sbi_flag(sbi, SBI_NEED_FSCK);
 645         f2fs_msg(sbi->sb, KERN_WARNING,
 646                         "%s: orphan failed (ino=%x), run fsck to fix.",
 647                         __func__, ino);
 648         return err;
 649 }
 650
 651 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 652 {
 653         block_t start_blk, orphan_blocks, i, j;
 654         unsigned int s_flags = sbi->sb->s_flags;
 655         int err = 0;
 656 #ifdef CONFIG_QUOTA
 657         int quota_enabled;
 658 #endif
 659
 660         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 661                 return 0;
 662
 663         if (s_flags & SB_RDONLY) {
 664                 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
 665                 sbi->sb->s_flags &= ~SB_RDONLY;
 666         }
 667
 668 #ifdef CONFIG_QUOTA
 669         /* Needed for iput() to work correctly and not trash data */
 670         sbi->sb->s_flags |= SB_ACTIVE;
 671
 672         /*
 673          * Turn on quotas which were not enabled for read-only mounts if
 674          * filesystem has quota feature, so that they are updated correctly.
 675          */
 676         quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
 677 #endif
 678
 679         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 680         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 681
 682         f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 683
 684         for (i = 0; i < orphan_blocks; i++) {
 685                 struct page *page;
 686                 struct f2fs_orphan_block *orphan_blk;
 687
 688                 page = f2fs_get_meta_page(sbi, start_blk + i);
 689                 if (IS_ERR(page)) {
 690                         err = PTR_ERR(page);
 691                         goto out;
 692                 }
 693
 694                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 695                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 696                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 697                         err = recover_orphan_inode(sbi, ino);
 698                         if (err) {
 699                                 f2fs_put_page(page, 1);
 700                                 goto out;
 701                         }
 702                 }
 703                 f2fs_put_page(page, 1);
 704         }
 705         /* clear Orphan Flag */
 706         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 707 out:
 708         set_sbi_flag(sbi, SBI_IS_RECOVERED);
 709
 710 #ifdef CONFIG_QUOTA
 711         /* Turn quotas off */
 712         if (quota_enabled)
 713                 f2fs_quota_off_umount(sbi->sb);
 714 #endif
 715         sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
 716
 717         return err;
 718 }
 719
 720 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 721 {
 722         struct list_head *head;
 723         struct f2fs_orphan_block *orphan_blk = NULL;
 724         unsigned int nentries = 0;
 725         unsigned short index = 1;
 726         unsigned short orphan_blocks;
 727         struct page *page = NULL;
 728         struct ino_entry *orphan = NULL;
 729         struct inode_management *im = &sbi->im[ORPHAN_INO];
 730
 731         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 732
 733         /*
 734          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 735          * orphan inode operations are covered under f2fs_lock_op().
 736          * And, spin_lock should be avoided due to page operations below.
 737          */
 738         head = &im->ino_list;
 739
 740         /* loop for each orphan inode entry and write them in Jornal block */
 741         list_for_each_entry(orphan, head, list) {
 742                 if (!page) {
 743                         page = f2fs_grab_meta_page(sbi, start_blk++);
 744                         orphan_blk =
 745                                 (struct f2fs_orphan_block *)page_address(page);
 746                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 747                 }
 748
 749                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 750
 751                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 752                         /*
 753                          * an orphan block is full of 1020 entries,
 754                          * then we need to flush current orphan blocks
 755                          * and bring another one in memory
 756                          */
 757                         orphan_blk->blk_addr = cpu_to_le16(index);
 758                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 759                         orphan_blk->entry_count = cpu_to_le32(nentries);
 760                         set_page_dirty(page);
 761                         f2fs_put_page(page, 1);
 762                         index++;
 763                         nentries = 0;
 764                         page = NULL;
 765                 }
 766         }
 767
 768         if (page) {
 769                 orphan_blk->blk_addr = cpu_to_le16(index);
 770                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 771                 orphan_blk->entry_count = cpu_to_le32(nentries);
 772                 set_page_dirty(page);
 773                 f2fs_put_page(page, 1);
 774         }
 775 }
 776
 777 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 778                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 779                 unsigned long long *version)
 780 {
 781         unsigned long blk_size = sbi->blocksize;
 782         size_t crc_offset = 0;
 783         __u32 crc = 0;
 784
 785         *cp_page = f2fs_get_meta_page(sbi, cp_addr);
 786         if (IS_ERR(*cp_page))
 787                 return PTR_ERR(*cp_page);
 788
 789         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 790
 791         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 792         if (crc_offset > (blk_size - sizeof(__le32))) {
 793                 f2fs_put_page(*cp_page, 1);
 794                 f2fs_msg(sbi->sb, KERN_WARNING,
 795                         "invalid crc_offset: %zu", crc_offset);
 796                 return -EINVAL;
 797         }
 798
 799         crc = cur_cp_crc(*cp_block);
 800         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 801                 f2fs_put_page(*cp_page, 1);
 802                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 803                 return -EINVAL;
 804         }
 805
 806         *version = cur_cp_version(*cp_block);
 807         return 0;
 808 }
 809
 810 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 811                                 block_t cp_addr, unsigned long long *version)
 812 {
 813         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 814         struct f2fs_checkpoint *cp_block = NULL;
 815         unsigned long long cur_version = 0, pre_version = 0;
 816         int err;
 817
 818         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 819                                         &cp_page_1, version);
 820         if (err)
 821                 return NULL;
 822
 823         if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
 824                                         sbi->blocks_per_seg) {
 825                 f2fs_msg(sbi->sb, KERN_WARNING,
 826                         "invalid cp_pack_total_block_count:%u",
 827                         le32_to_cpu(cp_block->cp_pack_total_block_count));
 828                 goto invalid_cp;
 829         }
 830         pre_version = *version;
 831
 832         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 833         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 834                                         &cp_page_2, version);
 835         if (err)
 836                 goto invalid_cp;
 837         cur_version = *version;
 838
 839         if (cur_version == pre_version) {
 840                 *version = cur_version;
 841                 f2fs_put_page(cp_page_2, 1);
 842                 return cp_page_1;
 843         }
 844         f2fs_put_page(cp_page_2, 1);
 845 invalid_cp:
 846         f2fs_put_page(cp_page_1, 1);
 847         return NULL;
 848 }
 849
 850 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 851 {
 852         struct f2fs_checkpoint *cp_block;
 853         struct f2fs_super_block *fsb = sbi->raw_super;
 854         struct page *cp1, *cp2, *cur_page;
 855         unsigned long blk_size = sbi->blocksize;
 856         unsigned long long cp1_version = 0, cp2_version = 0;
 857         unsigned long long cp_start_blk_no;
 858         unsigned int cp_blks = 1 + __cp_payload(sbi);
 859         block_t cp_blk_no;
 860         int i;
 861         int err;
 862
 863         sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
 864                                  GFP_KERNEL);
 865         if (!sbi->ckpt)
 866                 return -ENOMEM;
 867         /*
 868          * Finding out valid cp block involves read both
 869          * sets( cp pack1 and cp pack 2)
 870          */
 871         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 872         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 873
 874         /* The second checkpoint pack should start at the next segment */
 875         cp_start_blk_no += ((unsigned long long)1) <<
 876                                 le32_to_cpu(fsb->log_blocks_per_seg);
 877         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 878
 879         if (cp1 && cp2) {
 880                 if (ver_after(cp2_version, cp1_version))
 881                         cur_page = cp2;
 882                 else
 883                         cur_page = cp1;
 884         } else if (cp1) {
 885                 cur_page = cp1;
 886         } else if (cp2) {
 887                 cur_page = cp2;
 888         } else {
 889                 err = -EFSCORRUPTED;
 890                 goto fail_no_cp;
 891         }
 892
 893         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 894         memcpy(sbi->ckpt, cp_block, blk_size);
 895
 896         if (cur_page == cp1)
 897                 sbi->cur_cp_pack = 1;
 898         else
 899                 sbi->cur_cp_pack = 2;
 900
 901         /* Sanity checking of checkpoint */
 902         if (f2fs_sanity_check_ckpt(sbi)) {
 903                 err = -EFSCORRUPTED;
 904                 goto free_fail_no_cp;
 905         }
 906
 907         if (cp_blks <= 1)
 908                 goto done;
 909
 910         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 911         if (cur_page == cp2)
 912                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 913
 914         for (i = 1; i < cp_blks; i++) {
 915                 void *sit_bitmap_ptr;
 916                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 917
 918                 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
 919                 if (IS_ERR(cur_page)) {
 920                         err = PTR_ERR(cur_page);
 921                         goto free_fail_no_cp;
 922                 }
 923                 sit_bitmap_ptr = page_address(cur_page);
 924                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 925                 f2fs_put_page(cur_page, 1);
 926         }
 927 done:
 928         f2fs_put_page(cp1, 1);
 929         f2fs_put_page(cp2, 1);
 930         return 0;
 931
 932 free_fail_no_cp:
 933         f2fs_put_page(cp1, 1);
 934         f2fs_put_page(cp2, 1);
 935 fail_no_cp:
 936         kfree(sbi->ckpt);
 937         return err;
 938 }
 939
 940 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 941 {
 942         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 943         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 944
 945         if (is_inode_flag_set(inode, flag))
 946                 return;
 947
 948         set_inode_flag(inode, flag);
 949         if (!f2fs_is_volatile_file(inode))
 950                 list_add_tail(&F2FS_I(inode)->dirty_list,
 951                                                 &sbi->inode_list[type]);
 952         stat_inc_dirty_inode(sbi, type);
 953 }
 954
 955 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 956 {
 957         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 958
 959         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 960                 return;
 961
 962         list_del_init(&F2FS_I(inode)->dirty_list);
 963         clear_inode_flag(inode, flag);
 964         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 965 }
 966
 967 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
 968 {
 969         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 970         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 971
 972         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 973                         !S_ISLNK(inode->i_mode))
 974                 return;
 975
 976         spin_lock(&sbi->inode_lock[type]);
 977         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 978                 __add_dirty_inode(inode, type);
 979         inode_inc_dirty_pages(inode);
 980         spin_unlock(&sbi->inode_lock[type]);
 981
 982         SetPagePrivate(page);
 983         f2fs_trace_pid(page);
 984 }
 985
 986 void f2fs_remove_dirty_inode(struct inode *inode)
 987 {
 988         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 989         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 990
 991         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 992                         !S_ISLNK(inode->i_mode))
 993                 return;
 994
 995         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 996                 return;
 997
 998         spin_lock(&sbi->inode_lock[type]);
 999         __remove_dirty_inode(inode, type);
1000         spin_unlock(&sbi->inode_lock[type]);
1001 }
1002
1003 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1004 {
1005         struct list_head *head;
1006         struct inode *inode;
1007         struct f2fs_inode_info *fi;
1008         bool is_dir = (type == DIR_INODE);
1009         unsigned long ino = 0;
1010
1011         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1012                                 get_pages(sbi, is_dir ?
1013                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1014 retry:
1015         if (unlikely(f2fs_cp_error(sbi))) {
1016                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1017                                 get_pages(sbi, is_dir ?
1018                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1019                 return -EIO;
1020         }
1021
1022         spin_lock(&sbi->inode_lock[type]);
1023
1024         head = &sbi->inode_list[type];
1025         if (list_empty(head)) {
1026                 spin_unlock(&sbi->inode_lock[type]);
1027                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1028                                 get_pages(sbi, is_dir ?
1029                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1030                 return 0;
1031         }
1032         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1033         inode = igrab(&fi->vfs_inode);
1034         spin_unlock(&sbi->inode_lock[type]);
1035         if (inode) {
1036                 unsigned long cur_ino = inode->i_ino;
1037
1038                 if (is_dir)
1039                         F2FS_I(inode)->cp_task = current;
1040
1041                 filemap_fdatawrite(inode->i_mapping);
1042
1043                 if (is_dir)
1044                         F2FS_I(inode)->cp_task = NULL;
1045
1046                 iput(inode);
1047                 /* We need to give cpu to another writers. */
1048                 if (ino == cur_ino)
1049                         cond_resched();
1050                 else
1051                         ino = cur_ino;
1052         } else {
1053                 /*
1054                  * We should submit bio, since it exists several
1055                  * wribacking dentry pages in the freeing inode.
1056                  */
1057                 f2fs_submit_merged_write(sbi, DATA);
1058                 cond_resched();
1059         }
1060         goto retry;
1061 }
1062
1063 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1064 {
1065         struct list_head *head = &sbi->inode_list[DIRTY_META];
1066         struct inode *inode;
1067         struct f2fs_inode_info *fi;
1068         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1069
1070         while (total--) {
1071                 if (unlikely(f2fs_cp_error(sbi)))
1072                         return -EIO;
1073
1074                 spin_lock(&sbi->inode_lock[DIRTY_META]);
1075                 if (list_empty(head)) {
1076                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
1077                         return 0;
1078                 }
1079                 fi = list_first_entry(head, struct f2fs_inode_info,
1080                                                         gdirty_list);
1081                 inode = igrab(&fi->vfs_inode);
1082                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1083                 if (inode) {
1084                         sync_inode_metadata(inode, 0);
1085
1086                         /* it's on eviction */
1087                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1088                                 f2fs_update_inode_page(inode);
1089                         iput(inode);
1090                 }
1091         }
1092         return 0;
1093 }
1094
1095 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1096 {
1097         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1098         struct f2fs_nm_info *nm_i = NM_I(sbi);
1099         nid_t last_nid = nm_i->next_scan_nid;
1100
1101         next_free_nid(sbi, &last_nid);
1102         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1103         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1104         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1105         ckpt->next_free_nid = cpu_to_le32(last_nid);
1106 }
1107
1108 /*
1109  * Freeze all the FS-operations for checkpoint.
1110  */
1111 static int block_operations(struct f2fs_sb_info *sbi)
1112 {
1113         struct writeback_control wbc = {
1114                 .sync_mode = WB_SYNC_ALL,
1115                 .nr_to_write = LONG_MAX,
1116                 .for_reclaim = 0,
1117         };
1118         struct blk_plug plug;
1119         int err = 0;
1120
1121         blk_start_plug(&plug);
1122
1123 retry_flush_dents:
1124         f2fs_lock_all(sbi);
1125         /* write all the dirty dentry pages */
1126         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1127                 f2fs_unlock_all(sbi);
1128                 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1129                 if (err)
1130                         goto out;
1131                 cond_resched();
1132                 goto retry_flush_dents;
1133         }
1134
1135         /*
1136          * POR: we should ensure that there are no dirty node pages
1137          * until finishing nat/sit flush. inode->i_blocks can be updated.
1138          */
1139         down_write(&sbi->node_change);
1140
1141         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1142                 up_write(&sbi->node_change);
1143                 f2fs_unlock_all(sbi);
1144                 err = f2fs_sync_inode_meta(sbi);
1145                 if (err)
1146                         goto out;
1147                 cond_resched();
1148                 goto retry_flush_dents;
1149         }
1150
1151 retry_flush_nodes:
1152         down_write(&sbi->node_write);
1153
1154         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1155                 up_write(&sbi->node_write);
1156                 atomic_inc(&sbi->wb_sync_req[NODE]);
1157                 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1158                 atomic_dec(&sbi->wb_sync_req[NODE]);
1159                 if (err) {
1160                         up_write(&sbi->node_change);
1161                         f2fs_unlock_all(sbi);
1162                         goto out;
1163                 }
1164                 cond_resched();
1165                 goto retry_flush_nodes;
1166         }
1167
1168         /*
1169          * sbi->node_change is used only for AIO write_begin path which produces
1170          * dirty node blocks and some checkpoint values by block allocation.
1171          */
1172         __prepare_cp_block(sbi);
1173         up_write(&sbi->node_change);
1174 out:
1175         blk_finish_plug(&plug);
1176         return err;
1177 }
1178
1179 static void unblock_operations(struct f2fs_sb_info *sbi)
1180 {
1181         up_write(&sbi->node_write);
1182         f2fs_unlock_all(sbi);
1183 }
1184
1185 void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1186 {
1187         DEFINE_WAIT(wait);
1188
1189         for (;;) {
1190                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1191
1192                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1193                         break;
1194
1195                 if (unlikely(f2fs_cp_error(sbi) &&
1196                         !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1197                         break;
1198
1199                 io_schedule_timeout(5*HZ);
1200         }
1201         finish_wait(&sbi->cp_wait, &wait);
1202 }
1203
1204 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1205 {
1206         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1207         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1208         unsigned long flags;
1209
1210         spin_lock_irqsave(&sbi->cp_lock, flags);
1211
1212         if ((cpc->reason & CP_UMOUNT) &&
1213                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1214                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1215                 disable_nat_bits(sbi, false);
1216
1217         if (cpc->reason & CP_TRIMMED)
1218                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1219         else
1220                 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1221
1222         if (cpc->reason & CP_UMOUNT)
1223                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1224         else
1225                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1226
1227         if (cpc->reason & CP_FASTBOOT)
1228                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1229         else
1230                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1231
1232         if (orphan_num)
1233                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1234         else
1235                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1236
1237         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1238                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1239
1240         /* set this flag to activate crc|cp_ver for recovery */
1241         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1242         __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1243
1244         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1245 }
1246
1247 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1248         void *src, block_t blk_addr)
1249 {
1250         struct writeback_control wbc = {
1251                 .for_reclaim = 0,
1252         };
1253
1254         /*
1255          * pagevec_lookup_tag and lock_page again will take
1256          * some extra time. Therefore, f2fs_update_meta_pages and
1257          * f2fs_sync_meta_pages are combined in this function.
1258          */
1259         struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1260         int err;
1261
1262         memcpy(page_address(page), src, PAGE_SIZE);
1263         set_page_dirty(page);
1264
1265         f2fs_wait_on_page_writeback(page, META, true);
1266         f2fs_bug_on(sbi, PageWriteback(page));
1267         if (unlikely(!clear_page_dirty_for_io(page)))
1268                 f2fs_bug_on(sbi, 1);
1269
1270         /* writeout cp pack 2 page */
1271         err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1272         if (unlikely(err && f2fs_cp_error(sbi))) {
1273                 f2fs_put_page(page, 1);
1274                 return;
1275         }
1276
1277         f2fs_bug_on(sbi, err);
1278         f2fs_put_page(page, 0);
1279
1280         /* submit checkpoint (with barrier if NOBARRIER is not set) */
1281         f2fs_submit_merged_write(sbi, META_FLUSH);
1282 }
1283
1284 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1285 {
1286         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1287         struct f2fs_nm_info *nm_i = NM_I(sbi);
1288         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1289         block_t start_blk;
1290         unsigned int data_sum_blocks, orphan_blocks;
1291         __u32 crc32 = 0;
1292         int i;
1293         int cp_payload_blks = __cp_payload(sbi);
1294         struct super_block *sb = sbi->sb;
1295         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1296         u64 kbytes_written;
1297         int err;
1298
1299         /* Flush all the NAT/SIT pages */
1300         while (get_pages(sbi, F2FS_DIRTY_META)) {
1301                 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1302                 if (unlikely(f2fs_cp_error(sbi)))
1303                         break;
1304         }
1305
1306         /*
1307          * modify checkpoint
1308          * version number is already updated
1309          */
1310         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1311         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1312         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1313                 ckpt->cur_node_segno[i] =
1314                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1315                 ckpt->cur_node_blkoff[i] =
1316                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1317                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1318                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1319         }
1320         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1321                 ckpt->cur_data_segno[i] =
1322                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1323                 ckpt->cur_data_blkoff[i] =
1324                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1325                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1326                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1327         }
1328
1329         /* 2 cp  + n data seg summary + orphan inode blocks */
1330         data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1331         spin_lock_irqsave(&sbi->cp_lock, flags);
1332         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1333                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1334         else
1335                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1336         spin_unlock_irqrestore(&sbi->cp_lock, flags);
1337
1338         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1339         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1340                         orphan_blocks);
1341
1342         if (__remain_node_summaries(cpc->reason))
1343                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1344                                 cp_payload_blks + data_sum_blocks +
1345                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1346         else
1347                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1348                                 cp_payload_blks + data_sum_blocks +
1349                                 orphan_blocks);
1350
1351         /* update ckpt flag for checkpoint */
1352         update_ckpt_flags(sbi, cpc);
1353
1354         /* update SIT/NAT bitmap */
1355         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1356         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1357
1358         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1359         *((__le32 *)((unsigned char *)ckpt +
1360                                 le32_to_cpu(ckpt->checksum_offset)))
1361                                 = cpu_to_le32(crc32);
1362
1363         start_blk = __start_cp_next_addr(sbi);
1364
1365         /* write nat bits */
1366         if (enabled_nat_bits(sbi, cpc)) {
1367                 __u64 cp_ver = cur_cp_version(ckpt);
1368                 block_t blk;
1369
1370                 cp_ver |= ((__u64)crc32 << 32);
1371                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1372
1373                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1374                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1375                         f2fs_update_meta_page(sbi, nm_i->nat_bits +
1376                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1377
1378                 /* Flush all the NAT BITS pages */
1379                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1380                         f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1381                                                         FS_CP_META_IO);
1382                         if (unlikely(f2fs_cp_error(sbi)))
1383                                 break;
1384                 }
1385         }
1386
1387         /* write out checkpoint buffer at block 0 */
1388         f2fs_update_meta_page(sbi, ckpt, start_blk++);
1389
1390         for (i = 1; i < 1 + cp_payload_blks; i++)
1391                 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1392                                                         start_blk++);
1393
1394         if (orphan_num) {
1395                 write_orphan_inodes(sbi, start_blk);
1396                 start_blk += orphan_blocks;
1397         }
1398
1399         f2fs_write_data_summaries(sbi, start_blk);
1400         start_blk += data_sum_blocks;
1401
1402         /* Record write statistics in the hot node summary */
1403         kbytes_written = sbi->kbytes_written;
1404         if (sb->s_bdev->bd_part)
1405                 kbytes_written += BD_PART_WRITTEN(sbi);
1406
1407         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1408
1409         if (__remain_node_summaries(cpc->reason)) {
1410                 f2fs_write_node_summaries(sbi, start_blk);
1411                 start_blk += NR_CURSEG_NODE_TYPE;
1412         }
1413
1414         /* update user_block_counts */
1415         sbi->last_valid_block_count = sbi->total_valid_block_count;
1416         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1417
1418         /* Here, we have one bio having CP pack except cp pack 2 page */
1419         f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1420
1421         /* wait for previous submitted meta pages writeback */
1422         f2fs_wait_on_all_pages_writeback(sbi);
1423
1424         /* flush all device cache */
1425         err = f2fs_flush_device_cache(sbi);
1426         if (err)
1427                 return err;
1428
1429         /* barrier and flush checkpoint cp pack 2 page if it can */
1430         commit_checkpoint(sbi, ckpt, start_blk);
1431         f2fs_wait_on_all_pages_writeback(sbi);
1432
1433         /*
1434          * invalidate intermediate page cache borrowed from meta inode
1435          * which are used for migration of encrypted inode's blocks.
1436          */
1437         if (f2fs_sb_has_encrypt(sbi->sb))
1438                 invalidate_mapping_pages(META_MAPPING(sbi),
1439                                 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1440
1441         f2fs_release_ino_entry(sbi, false);
1442
1443         f2fs_reset_fsync_node_info(sbi);
1444
1445         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1446         clear_sbi_flag(sbi, SBI_NEED_CP);
1447         __set_cp_next_pack(sbi);
1448
1449         /*
1450          * redirty superblock if metadata like node page or inode cache is
1451          * updated during writing checkpoint.
1452          */
1453         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1454                         get_pages(sbi, F2FS_DIRTY_IMETA))
1455                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1456
1457         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1458
1459         return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1460 }
1461
1462 /*
1463  * We guarantee that this checkpoint procedure will not fail.
1464  */
1465 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1466 {
1467         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1468         unsigned long long ckpt_ver;
1469         int err = 0;
1470
1471         mutex_lock(&sbi->cp_mutex);
1472
1473         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1474                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1475                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1476                 goto out;
1477         if (unlikely(f2fs_cp_error(sbi))) {
1478                 err = -EIO;
1479                 goto out;
1480         }
1481         if (f2fs_readonly(sbi->sb)) {
1482                 err = -EROFS;
1483                 goto out;
1484         }
1485
1486         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1487
1488         err = block_operations(sbi);
1489         if (err)
1490                 goto out;
1491
1492         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1493
1494         f2fs_flush_merged_writes(sbi);
1495
1496         /* this is the case of multiple fstrims without any changes */
1497         if (cpc->reason & CP_DISCARD) {
1498                 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1499                         unblock_operations(sbi);
1500                         goto out;
1501                 }
1502
1503                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1504                                 SIT_I(sbi)->dirty_sentries == 0 &&
1505                                 prefree_segments(sbi) == 0) {
1506                         f2fs_flush_sit_entries(sbi, cpc);
1507                         f2fs_clear_prefree_segments(sbi, cpc);
1508                         unblock_operations(sbi);
1509                         goto out;
1510                 }
1511         }
1512
1513         /*
1514          * update checkpoint pack index
1515          * Increase the version number so that
1516          * SIT entries and seg summaries are written at correct place
1517          */
1518         ckpt_ver = cur_cp_version(ckpt);
1519         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1520
1521         /* write cached NAT/SIT entries to NAT/SIT area */
1522         f2fs_flush_nat_entries(sbi, cpc);
1523         f2fs_flush_sit_entries(sbi, cpc);
1524
1525         /* unlock all the fs_lock[] in do_checkpoint() */
1526         err = do_checkpoint(sbi, cpc);
1527         if (err)
1528                 f2fs_release_discard_addrs(sbi);
1529         else
1530                 f2fs_clear_prefree_segments(sbi, cpc);
1531
1532         unblock_operations(sbi);
1533         stat_inc_cp_count(sbi->stat_info);
1534
1535         if (cpc->reason & CP_RECOVERY)
1536                 f2fs_msg(sbi->sb, KERN_NOTICE,
1537                         "checkpoint: version = %llx", ckpt_ver);
1538
1539         /* do checkpoint periodically */
1540         f2fs_update_time(sbi, CP_TIME);
1541         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1542 out:
1543         mutex_unlock(&sbi->cp_mutex);
1544         return err;
1545 }
1546
1547 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1548 {
1549         int i;
1550
1551         for (i = 0; i < MAX_INO_ENTRY; i++) {
1552                 struct inode_management *im = &sbi->im[i];
1553
1554                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1555                 spin_lock_init(&im->ino_lock);
1556                 INIT_LIST_HEAD(&im->ino_list);
1557                 im->ino_num = 0;
1558         }
1559
1560         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1561                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1562                                 F2FS_ORPHANS_PER_BLOCK;
1563 }
1564
1565 int __init f2fs_create_checkpoint_caches(void)
1566 {
1567         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1568                         sizeof(struct ino_entry));
1569         if (!ino_entry_slab)
1570                 return -ENOMEM;
1571         f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1572                         sizeof(struct inode_entry));
1573         if (!f2fs_inode_entry_slab) {
1574                 kmem_cache_destroy(ino_entry_slab);
1575                 return -ENOMEM;
1576         }
1577         return 0;
1578 }
1579
1580 void f2fs_destroy_checkpoint_caches(void)
1581 {
1582         kmem_cache_destroy(ino_entry_slab);
1583         kmem_cache_destroy(f2fs_inode_entry_slab);
1584 }