fs/f2fs/checkpoint.c

   1 /*
   2  * fs/f2fs/checkpoint.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/bio.h>
  13 #include <linux/mpage.h>
  14 #include <linux/writeback.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/f2fs_fs.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/swap.h>
  19
  20 #include "f2fs.h"
  21 #include "node.h"
  22 #include "segment.h"
  23 #include "trace.h"
  24 #include <trace/events/f2fs.h>
  25
  26 static struct kmem_cache *ino_entry_slab;
  27 struct kmem_cache *inode_entry_slab;
  28
  29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
  30 {
  31         set_ckpt_flags(sbi, CP_ERROR_FLAG);
  32         sbi->sb->s_flags |= MS_RDONLY;
  33         if (!end_io)
  34                 f2fs_flush_merged_writes(sbi);
  35 }
  36
  37 /*
  38  * We guarantee no failure on the returned page.
  39  */
  40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  41 {
  42         struct address_space *mapping = META_MAPPING(sbi);
  43         struct page *page = NULL;
  44 repeat:
  45         page = f2fs_grab_cache_page(mapping, index, false);
  46         if (!page) {
  47                 cond_resched();
  48                 goto repeat;
  49         }
  50         f2fs_wait_on_page_writeback(page, META, true);
  51         if (!PageUptodate(page))
  52                 SetPageUptodate(page);
  53         return page;
  54 }
  55
  56 /*
  57  * We guarantee no failure on the returned page.
  58  */
  59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  60                                                         bool is_meta)
  61 {
  62         struct address_space *mapping = META_MAPPING(sbi);
  63         struct page *page;
  64         struct f2fs_io_info fio = {
  65                 .sbi = sbi,
  66                 .type = META,
  67                 .op = REQ_OP_READ,
  68                 .op_flags = REQ_META | REQ_PRIO,
  69                 .old_blkaddr = index,
  70                 .new_blkaddr = index,
  71                 .encrypted_page = NULL,
  72         };
  73
  74         if (unlikely(!is_meta))
  75                 fio.op_flags &= ~REQ_META;
  76 repeat:
  77         page = f2fs_grab_cache_page(mapping, index, false);
  78         if (!page) {
  79                 cond_resched();
  80                 goto repeat;
  81         }
  82         if (PageUptodate(page))
  83                 goto out;
  84
  85         fio.page = page;
  86
  87         if (f2fs_submit_page_bio(&fio)) {
  88                 f2fs_put_page(page, 1);
  89                 goto repeat;
  90         }
  91
  92         lock_page(page);
  93         if (unlikely(page->mapping != mapping)) {
  94                 f2fs_put_page(page, 1);
  95                 goto repeat;
  96         }
  97
  98         /*
  99          * if there is any IO error when accessing device, make our filesystem
 100          * readonly and make sure do not write checkpoint with non-uptodate
 101          * meta page.
 102          */
 103         if (unlikely(!PageUptodate(page)))
 104                 f2fs_stop_checkpoint(sbi, false);
 105 out:
 106         return page;
 107 }
 108
 109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 110 {
 111         return __get_meta_page(sbi, index, true);
 112 }
 113
 114 /* for POR only */
 115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 116 {
 117         return __get_meta_page(sbi, index, false);
 118 }
 119
 120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
 121 {
 122         switch (type) {
 123         case META_NAT:
 124                 break;
 125         case META_SIT:
 126                 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 127                         return false;
 128                 break;
 129         case META_SSA:
 130                 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 131                         blkaddr < SM_I(sbi)->ssa_blkaddr))
 132                         return false;
 133                 break;
 134         case META_CP:
 135                 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 136                         blkaddr < __start_cp_addr(sbi)))
 137                         return false;
 138                 break;
 139         case META_POR:
 140                 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 141                         blkaddr < MAIN_BLKADDR(sbi)))
 142                         return false;
 143                 break;
 144         default:
 145                 BUG();
 146         }
 147
 148         return true;
 149 }
 150
 151 /*
 152  * Readahead CP/NAT/SIT/SSA pages
 153  */
 154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 155                                                         int type, bool sync)
 156 {
 157         struct page *page;
 158         block_t blkno = start;
 159         struct f2fs_io_info fio = {
 160                 .sbi = sbi,
 161                 .type = META,
 162                 .op = REQ_OP_READ,
 163                 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 164                 .encrypted_page = NULL,
 165         };
 166         struct blk_plug plug;
 167
 168         if (unlikely(type == META_POR))
 169                 fio.op_flags &= ~REQ_META;
 170
 171         blk_start_plug(&plug);
 172         for (; nrpages-- > 0; blkno++) {
 173
 174                 if (!is_valid_blkaddr(sbi, blkno, type))
 175                         goto out;
 176
 177                 switch (type) {
 178                 case META_NAT:
 179                         if (unlikely(blkno >=
 180                                         NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 181                                 blkno = 0;
 182                         /* get nat block addr */
 183                         fio.new_blkaddr = current_nat_addr(sbi,
 184                                         blkno * NAT_ENTRY_PER_BLOCK);
 185                         break;
 186                 case META_SIT:
 187                         /* get sit block addr */
 188                         fio.new_blkaddr = current_sit_addr(sbi,
 189                                         blkno * SIT_ENTRY_PER_BLOCK);
 190                         break;
 191                 case META_SSA:
 192                 case META_CP:
 193                 case META_POR:
 194                         fio.new_blkaddr = blkno;
 195                         break;
 196                 default:
 197                         BUG();
 198                 }
 199
 200                 page = f2fs_grab_cache_page(META_MAPPING(sbi),
 201                                                 fio.new_blkaddr, false);
 202                 if (!page)
 203                         continue;
 204                 if (PageUptodate(page)) {
 205                         f2fs_put_page(page, 1);
 206                         continue;
 207                 }
 208
 209                 fio.page = page;
 210                 f2fs_submit_page_bio(&fio);
 211                 f2fs_put_page(page, 0);
 212         }
 213 out:
 214         blk_finish_plug(&plug);
 215         return blkno - start;
 216 }
 217
 218 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
 219 {
 220         struct page *page;
 221         bool readahead = false;
 222
 223         page = find_get_page(META_MAPPING(sbi), index);
 224         if (!page || !PageUptodate(page))
 225                 readahead = true;
 226         f2fs_put_page(page, 0);
 227
 228         if (readahead)
 229                 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
 230 }
 231
 232 static int f2fs_write_meta_page(struct page *page,
 233                                 struct writeback_control *wbc)
 234 {
 235         struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 236
 237         trace_f2fs_writepage(page, META);
 238
 239         if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 240                 goto redirty_out;
 241         if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
 242                 goto redirty_out;
 243         if (unlikely(f2fs_cp_error(sbi)))
 244                 goto redirty_out;
 245
 246         write_meta_page(sbi, page);
 247         dec_page_count(sbi, F2FS_DIRTY_META);
 248
 249         if (wbc->for_reclaim)
 250                 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
 251                                                 0, page->index, META);
 252
 253         unlock_page(page);
 254
 255         if (unlikely(f2fs_cp_error(sbi)))
 256                 f2fs_submit_merged_write(sbi, META);
 257
 258         return 0;
 259
 260 redirty_out:
 261         redirty_page_for_writepage(wbc, page);
 262         return AOP_WRITEPAGE_ACTIVATE;
 263 }
 264
 265 static int f2fs_write_meta_pages(struct address_space *mapping,
 266                                 struct writeback_control *wbc)
 267 {
 268         struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 269         long diff, written;
 270
 271         /* collect a number of dirty meta pages and write together */
 272         if (wbc->for_kupdate ||
 273                 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
 274                 goto skip_write;
 275
 276         /* if locked failed, cp will flush dirty pages instead */
 277         if (!mutex_trylock(&sbi->cp_mutex))
 278                 goto skip_write;
 279
 280         trace_f2fs_writepages(mapping->host, wbc, META);
 281         diff = nr_pages_to_write(sbi, META, wbc);
 282         written = sync_meta_pages(sbi, META, wbc->nr_to_write);
 283         mutex_unlock(&sbi->cp_mutex);
 284         wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 285         return 0;
 286
 287 skip_write:
 288         wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 289         trace_f2fs_writepages(mapping->host, wbc, META);
 290         return 0;
 291 }
 292
 293 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 294                                                 long nr_to_write)
 295 {
 296         struct address_space *mapping = META_MAPPING(sbi);
 297         pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
 298         struct pagevec pvec;
 299         long nwritten = 0;
 300         struct writeback_control wbc = {
 301                 .for_reclaim = 0,
 302         };
 303         struct blk_plug plug;
 304
 305         pagevec_init(&pvec, 0);
 306
 307         blk_start_plug(&plug);
 308
 309         while (index <= end) {
 310                 int i, nr_pages;
 311                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 312                                 PAGECACHE_TAG_DIRTY,
 313                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 314                 if (unlikely(nr_pages == 0))
 315                         break;
 316
 317                 for (i = 0; i < nr_pages; i++) {
 318                         struct page *page = pvec.pages[i];
 319
 320                         if (prev == ULONG_MAX)
 321                                 prev = page->index - 1;
 322                         if (nr_to_write != LONG_MAX && page->index != prev + 1) {
 323                                 pagevec_release(&pvec);
 324                                 goto stop;
 325                         }
 326
 327                         lock_page(page);
 328
 329                         if (unlikely(page->mapping != mapping)) {
 330 continue_unlock:
 331                                 unlock_page(page);
 332                                 continue;
 333                         }
 334                         if (!PageDirty(page)) {
 335                                 /* someone wrote it for us */
 336                                 goto continue_unlock;
 337                         }
 338
 339                         f2fs_wait_on_page_writeback(page, META, true);
 340
 341                         BUG_ON(PageWriteback(page));
 342                         if (!clear_page_dirty_for_io(page))
 343                                 goto continue_unlock;
 344
 345                         if (mapping->a_ops->writepage(page, &wbc)) {
 346                                 unlock_page(page);
 347                                 break;
 348                         }
 349                         nwritten++;
 350                         prev = page->index;
 351                         if (unlikely(nwritten >= nr_to_write))
 352                                 break;
 353                 }
 354                 pagevec_release(&pvec);
 355                 cond_resched();
 356         }
 357 stop:
 358         if (nwritten)
 359                 f2fs_submit_merged_write(sbi, type);
 360
 361         blk_finish_plug(&plug);
 362
 363         return nwritten;
 364 }
 365
 366 static int f2fs_set_meta_page_dirty(struct page *page)
 367 {
 368         trace_f2fs_set_page_dirty(page, META);
 369
 370         if (!PageUptodate(page))
 371                 SetPageUptodate(page);
 372         if (!PageDirty(page)) {
 373                 f2fs_set_page_dirty_nobuffers(page);
 374                 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
 375                 SetPagePrivate(page);
 376                 f2fs_trace_pid(page);
 377                 return 1;
 378         }
 379         return 0;
 380 }
 381
 382 const struct address_space_operations f2fs_meta_aops = {
 383         .writepage      = f2fs_write_meta_page,
 384         .writepages     = f2fs_write_meta_pages,
 385         .set_page_dirty = f2fs_set_meta_page_dirty,
 386         .invalidatepage = f2fs_invalidate_page,
 387         .releasepage    = f2fs_release_page,
 388 #ifdef CONFIG_MIGRATION
 389         .migratepage    = f2fs_migrate_page,
 390 #endif
 391 };
 392
 393 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 394 {
 395         struct inode_management *im = &sbi->im[type];
 396         struct ino_entry *e, *tmp;
 397
 398         tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 399 retry:
 400         radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 401
 402         spin_lock(&im->ino_lock);
 403         e = radix_tree_lookup(&im->ino_root, ino);
 404         if (!e) {
 405                 e = tmp;
 406                 if (radix_tree_insert(&im->ino_root, ino, e)) {
 407                         spin_unlock(&im->ino_lock);
 408                         radix_tree_preload_end();
 409                         goto retry;
 410                 }
 411                 memset(e, 0, sizeof(struct ino_entry));
 412                 e->ino = ino;
 413
 414                 list_add_tail(&e->list, &im->ino_list);
 415                 if (type != ORPHAN_INO)
 416                         im->ino_num++;
 417         }
 418         spin_unlock(&im->ino_lock);
 419         radix_tree_preload_end();
 420
 421         if (e != tmp)
 422                 kmem_cache_free(ino_entry_slab, tmp);
 423 }
 424
 425 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 426 {
 427         struct inode_management *im = &sbi->im[type];
 428         struct ino_entry *e;
 429
 430         spin_lock(&im->ino_lock);
 431         e = radix_tree_lookup(&im->ino_root, ino);
 432         if (e) {
 433                 list_del(&e->list);
 434                 radix_tree_delete(&im->ino_root, ino);
 435                 im->ino_num--;
 436                 spin_unlock(&im->ino_lock);
 437                 kmem_cache_free(ino_entry_slab, e);
 438                 return;
 439         }
 440         spin_unlock(&im->ino_lock);
 441 }
 442
 443 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 444 {
 445         /* add new dirty ino entry into list */
 446         __add_ino_entry(sbi, ino, type);
 447 }
 448
 449 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 450 {
 451         /* remove dirty ino entry from list */
 452         __remove_ino_entry(sbi, ino, type);
 453 }
 454
 455 /* mode should be APPEND_INO or UPDATE_INO */
 456 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 457 {
 458         struct inode_management *im = &sbi->im[mode];
 459         struct ino_entry *e;
 460
 461         spin_lock(&im->ino_lock);
 462         e = radix_tree_lookup(&im->ino_root, ino);
 463         spin_unlock(&im->ino_lock);
 464         return e ? true : false;
 465 }
 466
 467 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 468 {
 469         struct ino_entry *e, *tmp;
 470         int i;
 471
 472         for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
 473                 struct inode_management *im = &sbi->im[i];
 474
 475                 spin_lock(&im->ino_lock);
 476                 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 477                         list_del(&e->list);
 478                         radix_tree_delete(&im->ino_root, e->ino);
 479                         kmem_cache_free(ino_entry_slab, e);
 480                         im->ino_num--;
 481                 }
 482                 spin_unlock(&im->ino_lock);
 483         }
 484 }
 485
 486 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 487 {
 488         struct inode_management *im = &sbi->im[ORPHAN_INO];
 489         int err = 0;
 490
 491         spin_lock(&im->ino_lock);
 492
 493 #ifdef CONFIG_F2FS_FAULT_INJECTION
 494         if (time_to_inject(sbi, FAULT_ORPHAN)) {
 495                 spin_unlock(&im->ino_lock);
 496                 f2fs_show_injection_info(FAULT_ORPHAN);
 497                 return -ENOSPC;
 498         }
 499 #endif
 500         if (unlikely(im->ino_num >= sbi->max_orphans))
 501                 err = -ENOSPC;
 502         else
 503                 im->ino_num++;
 504         spin_unlock(&im->ino_lock);
 505
 506         return err;
 507 }
 508
 509 void release_orphan_inode(struct f2fs_sb_info *sbi)
 510 {
 511         struct inode_management *im = &sbi->im[ORPHAN_INO];
 512
 513         spin_lock(&im->ino_lock);
 514         f2fs_bug_on(sbi, im->ino_num == 0);
 515         im->ino_num--;
 516         spin_unlock(&im->ino_lock);
 517 }
 518
 519 void add_orphan_inode(struct inode *inode)
 520 {
 521         /* add new orphan ino entry into list */
 522         __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
 523         update_inode_page(inode);
 524 }
 525
 526 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 527 {
 528         /* remove orphan entry from orphan list */
 529         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 530 }
 531
 532 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 533 {
 534         struct inode *inode;
 535         struct node_info ni;
 536         int err = acquire_orphan_inode(sbi);
 537
 538         if (err) {
 539                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 540                 f2fs_msg(sbi->sb, KERN_WARNING,
 541                                 "%s: orphan failed (ino=%x), run fsck to fix.",
 542                                 __func__, ino);
 543                 return err;
 544         }
 545
 546         __add_ino_entry(sbi, ino, ORPHAN_INO);
 547
 548         inode = f2fs_iget_retry(sbi->sb, ino);
 549         if (IS_ERR(inode)) {
 550                 /*
 551                  * there should be a bug that we can't find the entry
 552                  * to orphan inode.
 553                  */
 554                 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 555                 return PTR_ERR(inode);
 556         }
 557
 558         clear_nlink(inode);
 559
 560         /* truncate all the data during iput */
 561         iput(inode);
 562
 563         get_node_info(sbi, ino, &ni);
 564
 565         /* ENOMEM was fully retried in f2fs_evict_inode. */
 566         if (ni.blk_addr != NULL_ADDR) {
 567                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 568                 f2fs_msg(sbi->sb, KERN_WARNING,
 569                         "%s: orphan failed (ino=%x) by kernel, retry mount.",
 570                                 __func__, ino);
 571                 return -EIO;
 572         }
 573         __remove_ino_entry(sbi, ino, ORPHAN_INO);
 574         return 0;
 575 }
 576
 577 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 578 {
 579         block_t start_blk, orphan_blocks, i, j;
 580         int err;
 581
 582         if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 583                 return 0;
 584
 585         start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 586         orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 587
 588         ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 589
 590         for (i = 0; i < orphan_blocks; i++) {
 591                 struct page *page = get_meta_page(sbi, start_blk + i);
 592                 struct f2fs_orphan_block *orphan_blk;
 593
 594                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 595                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 596                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 597                         err = recover_orphan_inode(sbi, ino);
 598                         if (err) {
 599                                 f2fs_put_page(page, 1);
 600                                 return err;
 601                         }
 602                 }
 603                 f2fs_put_page(page, 1);
 604         }
 605         /* clear Orphan Flag */
 606         clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 607         return 0;
 608 }
 609
 610 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 611 {
 612         struct list_head *head;
 613         struct f2fs_orphan_block *orphan_blk = NULL;
 614         unsigned int nentries = 0;
 615         unsigned short index = 1;
 616         unsigned short orphan_blocks;
 617         struct page *page = NULL;
 618         struct ino_entry *orphan = NULL;
 619         struct inode_management *im = &sbi->im[ORPHAN_INO];
 620
 621         orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 622
 623         /*
 624          * we don't need to do spin_lock(&im->ino_lock) here, since all the
 625          * orphan inode operations are covered under f2fs_lock_op().
 626          * And, spin_lock should be avoided due to page operations below.
 627          */
 628         head = &im->ino_list;
 629
 630         /* loop for each orphan inode entry and write them in Jornal block */
 631         list_for_each_entry(orphan, head, list) {
 632                 if (!page) {
 633                         page = grab_meta_page(sbi, start_blk++);
 634                         orphan_blk =
 635                                 (struct f2fs_orphan_block *)page_address(page);
 636                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 637                 }
 638
 639                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 640
 641                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 642                         /*
 643                          * an orphan block is full of 1020 entries,
 644                          * then we need to flush current orphan blocks
 645                          * and bring another one in memory
 646                          */
 647                         orphan_blk->blk_addr = cpu_to_le16(index);
 648                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 649                         orphan_blk->entry_count = cpu_to_le32(nentries);
 650                         set_page_dirty(page);
 651                         f2fs_put_page(page, 1);
 652                         index++;
 653                         nentries = 0;
 654                         page = NULL;
 655                 }
 656         }
 657
 658         if (page) {
 659                 orphan_blk->blk_addr = cpu_to_le16(index);
 660                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 661                 orphan_blk->entry_count = cpu_to_le32(nentries);
 662                 set_page_dirty(page);
 663                 f2fs_put_page(page, 1);
 664         }
 665 }
 666
 667 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 668                 struct f2fs_checkpoint **cp_block, struct page **cp_page,
 669                 unsigned long long *version)
 670 {
 671         unsigned long blk_size = sbi->blocksize;
 672         size_t crc_offset = 0;
 673         __u32 crc = 0;
 674
 675         *cp_page = get_meta_page(sbi, cp_addr);
 676         *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 677
 678         crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 679         if (crc_offset > (blk_size - sizeof(__le32))) {
 680                 f2fs_msg(sbi->sb, KERN_WARNING,
 681                         "invalid crc_offset: %zu", crc_offset);
 682                 return -EINVAL;
 683         }
 684
 685         crc = cur_cp_crc(*cp_block);
 686         if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
 687                 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
 688                 return -EINVAL;
 689         }
 690
 691         *version = cur_cp_version(*cp_block);
 692         return 0;
 693 }
 694
 695 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 696                                 block_t cp_addr, unsigned long long *version)
 697 {
 698         struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 699         struct f2fs_checkpoint *cp_block = NULL;
 700         unsigned long long cur_version = 0, pre_version = 0;
 701         int err;
 702
 703         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 704                                         &cp_page_1, version);
 705         if (err)
 706                 goto invalid_cp1;
 707         pre_version = *version;
 708
 709         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 710         err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 711                                         &cp_page_2, version);
 712         if (err)
 713                 goto invalid_cp2;
 714         cur_version = *version;
 715
 716         if (cur_version == pre_version) {
 717                 *version = cur_version;
 718                 f2fs_put_page(cp_page_2, 1);
 719                 return cp_page_1;
 720         }
 721 invalid_cp2:
 722         f2fs_put_page(cp_page_2, 1);
 723 invalid_cp1:
 724         f2fs_put_page(cp_page_1, 1);
 725         return NULL;
 726 }
 727
 728 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 729 {
 730         struct f2fs_checkpoint *cp_block;
 731         struct f2fs_super_block *fsb = sbi->raw_super;
 732         struct page *cp1, *cp2, *cur_page;
 733         unsigned long blk_size = sbi->blocksize;
 734         unsigned long long cp1_version = 0, cp2_version = 0;
 735         unsigned long long cp_start_blk_no;
 736         unsigned int cp_blks = 1 + __cp_payload(sbi);
 737         block_t cp_blk_no;
 738         int i;
 739
 740         sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
 741         if (!sbi->ckpt)
 742                 return -ENOMEM;
 743         /*
 744          * Finding out valid cp block involves read both
 745          * sets( cp pack1 and cp pack 2)
 746          */
 747         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 748         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 749
 750         /* The second checkpoint pack should start at the next segment */
 751         cp_start_blk_no += ((unsigned long long)1) <<
 752                                 le32_to_cpu(fsb->log_blocks_per_seg);
 753         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 754
 755         if (cp1 && cp2) {
 756                 if (ver_after(cp2_version, cp1_version))
 757                         cur_page = cp2;
 758                 else
 759                         cur_page = cp1;
 760         } else if (cp1) {
 761                 cur_page = cp1;
 762         } else if (cp2) {
 763                 cur_page = cp2;
 764         } else {
 765                 goto fail_no_cp;
 766         }
 767
 768         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 769         memcpy(sbi->ckpt, cp_block, blk_size);
 770
 771         /* Sanity checking of checkpoint */
 772         if (sanity_check_ckpt(sbi))
 773                 goto free_fail_no_cp;
 774
 775         if (cur_page == cp1)
 776                 sbi->cur_cp_pack = 1;
 777         else
 778                 sbi->cur_cp_pack = 2;
 779
 780         if (cp_blks <= 1)
 781                 goto done;
 782
 783         cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 784         if (cur_page == cp2)
 785                 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
 786
 787         for (i = 1; i < cp_blks; i++) {
 788                 void *sit_bitmap_ptr;
 789                 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 790
 791                 cur_page = get_meta_page(sbi, cp_blk_no + i);
 792                 sit_bitmap_ptr = page_address(cur_page);
 793                 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 794                 f2fs_put_page(cur_page, 1);
 795         }
 796 done:
 797         f2fs_put_page(cp1, 1);
 798         f2fs_put_page(cp2, 1);
 799         return 0;
 800
 801 free_fail_no_cp:
 802         f2fs_put_page(cp1, 1);
 803         f2fs_put_page(cp2, 1);
 804 fail_no_cp:
 805         kfree(sbi->ckpt);
 806         return -EINVAL;
 807 }
 808
 809 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
 810 {
 811         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 812         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 813
 814         if (is_inode_flag_set(inode, flag))
 815                 return;
 816
 817         set_inode_flag(inode, flag);
 818         if (!f2fs_is_volatile_file(inode))
 819                 list_add_tail(&F2FS_I(inode)->dirty_list,
 820                                                 &sbi->inode_list[type]);
 821         stat_inc_dirty_inode(sbi, type);
 822 }
 823
 824 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
 825 {
 826         int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
 827
 828         if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
 829                 return;
 830
 831         list_del_init(&F2FS_I(inode)->dirty_list);
 832         clear_inode_flag(inode, flag);
 833         stat_dec_dirty_inode(F2FS_I_SB(inode), type);
 834 }
 835
 836 void update_dirty_page(struct inode *inode, struct page *page)
 837 {
 838         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 839         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 840
 841         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 842                         !S_ISLNK(inode->i_mode))
 843                 return;
 844
 845         spin_lock(&sbi->inode_lock[type]);
 846         if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
 847                 __add_dirty_inode(inode, type);
 848         inode_inc_dirty_pages(inode);
 849         spin_unlock(&sbi->inode_lock[type]);
 850
 851         SetPagePrivate(page);
 852         f2fs_trace_pid(page);
 853 }
 854
 855 void remove_dirty_inode(struct inode *inode)
 856 {
 857         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 858         enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
 859
 860         if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
 861                         !S_ISLNK(inode->i_mode))
 862                 return;
 863
 864         if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
 865                 return;
 866
 867         spin_lock(&sbi->inode_lock[type]);
 868         __remove_dirty_inode(inode, type);
 869         spin_unlock(&sbi->inode_lock[type]);
 870 }
 871
 872 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
 873 {
 874         struct list_head *head;
 875         struct inode *inode;
 876         struct f2fs_inode_info *fi;
 877         bool is_dir = (type == DIR_INODE);
 878
 879         trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
 880                                 get_pages(sbi, is_dir ?
 881                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 882 retry:
 883         if (unlikely(f2fs_cp_error(sbi)))
 884                 return -EIO;
 885
 886         spin_lock(&sbi->inode_lock[type]);
 887
 888         head = &sbi->inode_list[type];
 889         if (list_empty(head)) {
 890                 spin_unlock(&sbi->inode_lock[type]);
 891                 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
 892                                 get_pages(sbi, is_dir ?
 893                                 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
 894                 return 0;
 895         }
 896         fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
 897         inode = igrab(&fi->vfs_inode);
 898         spin_unlock(&sbi->inode_lock[type]);
 899         if (inode) {
 900                 filemap_fdatawrite(inode->i_mapping);
 901                 iput(inode);
 902         } else {
 903                 /*
 904                  * We should submit bio, since it exists several
 905                  * wribacking dentry pages in the freeing inode.
 906                  */
 907                 f2fs_submit_merged_write(sbi, DATA);
 908                 cond_resched();
 909         }
 910         goto retry;
 911 }
 912
 913 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
 914 {
 915         struct list_head *head = &sbi->inode_list[DIRTY_META];
 916         struct inode *inode;
 917         struct f2fs_inode_info *fi;
 918         s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
 919
 920         while (total--) {
 921                 if (unlikely(f2fs_cp_error(sbi)))
 922                         return -EIO;
 923
 924                 spin_lock(&sbi->inode_lock[DIRTY_META]);
 925                 if (list_empty(head)) {
 926                         spin_unlock(&sbi->inode_lock[DIRTY_META]);
 927                         return 0;
 928                 }
 929                 fi = list_first_entry(head, struct f2fs_inode_info,
 930                                                         gdirty_list);
 931                 inode = igrab(&fi->vfs_inode);
 932                 spin_unlock(&sbi->inode_lock[DIRTY_META]);
 933                 if (inode) {
 934                         sync_inode_metadata(inode, 0);
 935
 936                         /* it's on eviction */
 937                         if (is_inode_flag_set(inode, FI_DIRTY_INODE))
 938                                 update_inode_page(inode);
 939                         iput(inode);
 940                 }
 941         };
 942         return 0;
 943 }
 944
 945 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
 946 {
 947         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 948         struct f2fs_nm_info *nm_i = NM_I(sbi);
 949         nid_t last_nid = nm_i->next_scan_nid;
 950
 951         next_free_nid(sbi, &last_nid);
 952         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 953         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 954         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 955         ckpt->next_free_nid = cpu_to_le32(last_nid);
 956 }
 957
 958 /*
 959  * Freeze all the FS-operations for checkpoint.
 960  */
 961 static int block_operations(struct f2fs_sb_info *sbi)
 962 {
 963         struct writeback_control wbc = {
 964                 .sync_mode = WB_SYNC_ALL,
 965                 .nr_to_write = LONG_MAX,
 966                 .for_reclaim = 0,
 967         };
 968         struct blk_plug plug;
 969         int err = 0;
 970
 971         blk_start_plug(&plug);
 972
 973 retry_flush_dents:
 974         f2fs_lock_all(sbi);
 975         /* write all the dirty dentry pages */
 976         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 977                 f2fs_unlock_all(sbi);
 978                 err = sync_dirty_inodes(sbi, DIR_INODE);
 979                 if (err)
 980                         goto out;
 981                 cond_resched();
 982                 goto retry_flush_dents;
 983         }
 984
 985         /*
 986          * POR: we should ensure that there are no dirty node pages
 987          * until finishing nat/sit flush. inode->i_blocks can be updated.
 988          */
 989         down_write(&sbi->node_change);
 990
 991         if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
 992                 up_write(&sbi->node_change);
 993                 f2fs_unlock_all(sbi);
 994                 err = f2fs_sync_inode_meta(sbi);
 995                 if (err)
 996                         goto out;
 997                 cond_resched();
 998                 goto retry_flush_dents;
 999         }
1000
1001 retry_flush_nodes:
1002         down_write(&sbi->node_write);
1003
1004         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1005                 up_write(&sbi->node_write);
1006                 err = sync_node_pages(sbi, &wbc);
1007                 if (err) {
1008                         up_write(&sbi->node_change);
1009                         f2fs_unlock_all(sbi);
1010                         goto out;
1011                 }
1012                 cond_resched();
1013                 goto retry_flush_nodes;
1014         }
1015
1016         /*
1017          * sbi->node_change is used only for AIO write_begin path which produces
1018          * dirty node blocks and some checkpoint values by block allocation.
1019          */
1020         __prepare_cp_block(sbi);
1021         up_write(&sbi->node_change);
1022 out:
1023         blk_finish_plug(&plug);
1024         return err;
1025 }
1026
1027 static void unblock_operations(struct f2fs_sb_info *sbi)
1028 {
1029         up_write(&sbi->node_write);
1030         f2fs_unlock_all(sbi);
1031 }
1032
1033 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1034 {
1035         DEFINE_WAIT(wait);
1036
1037         for (;;) {
1038                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1039
1040                 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1041                         break;
1042
1043                 io_schedule_timeout(5*HZ);
1044         }
1045         finish_wait(&sbi->cp_wait, &wait);
1046 }
1047
1048 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1049 {
1050         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1051         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1052
1053         spin_lock(&sbi->cp_lock);
1054
1055         if ((cpc->reason & CP_UMOUNT) &&
1056                         le32_to_cpu(ckpt->cp_pack_total_block_count) >
1057                         sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1058                 disable_nat_bits(sbi, false);
1059
1060         if (cpc->reason & CP_TRIMMED)
1061                 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1062
1063         if (cpc->reason & CP_UMOUNT)
1064                 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1065         else
1066                 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1067
1068         if (cpc->reason & CP_FASTBOOT)
1069                 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1070         else
1071                 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1072
1073         if (orphan_num)
1074                 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1075         else
1076                 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1077
1078         if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1079                 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1080
1081         /* set this flag to activate crc|cp_ver for recovery */
1082         __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1083
1084         spin_unlock(&sbi->cp_lock);
1085 }
1086
1087 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1088 {
1089         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1090         struct f2fs_nm_info *nm_i = NM_I(sbi);
1091         unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1092         block_t start_blk;
1093         unsigned int data_sum_blocks, orphan_blocks;
1094         __u32 crc32 = 0;
1095         int i;
1096         int cp_payload_blks = __cp_payload(sbi);
1097         struct super_block *sb = sbi->sb;
1098         struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1099         u64 kbytes_written;
1100
1101         /* Flush all the NAT/SIT pages */
1102         while (get_pages(sbi, F2FS_DIRTY_META)) {
1103                 sync_meta_pages(sbi, META, LONG_MAX);
1104                 if (unlikely(f2fs_cp_error(sbi)))
1105                         return -EIO;
1106         }
1107
1108         /*
1109          * modify checkpoint
1110          * version number is already updated
1111          */
1112         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1113         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1114         for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1115                 ckpt->cur_node_segno[i] =
1116                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1117                 ckpt->cur_node_blkoff[i] =
1118                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1119                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1120                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1121         }
1122         for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1123                 ckpt->cur_data_segno[i] =
1124                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1125                 ckpt->cur_data_blkoff[i] =
1126                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1127                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1128                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1129         }
1130
1131         /* 2 cp  + n data seg summary + orphan inode blocks */
1132         data_sum_blocks = npages_for_summary_flush(sbi, false);
1133         spin_lock(&sbi->cp_lock);
1134         if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1135                 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1136         else
1137                 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1138         spin_unlock(&sbi->cp_lock);
1139
1140         orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1141         ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1142                         orphan_blocks);
1143
1144         if (__remain_node_summaries(cpc->reason))
1145                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1146                                 cp_payload_blks + data_sum_blocks +
1147                                 orphan_blocks + NR_CURSEG_NODE_TYPE);
1148         else
1149                 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1150                                 cp_payload_blks + data_sum_blocks +
1151                                 orphan_blocks);
1152
1153         /* update ckpt flag for checkpoint */
1154         update_ckpt_flags(sbi, cpc);
1155
1156         /* update SIT/NAT bitmap */
1157         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1158         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1159
1160         crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1161         *((__le32 *)((unsigned char *)ckpt +
1162                                 le32_to_cpu(ckpt->checksum_offset)))
1163                                 = cpu_to_le32(crc32);
1164
1165         start_blk = __start_cp_next_addr(sbi);
1166
1167         /* write nat bits */
1168         if (enabled_nat_bits(sbi, cpc)) {
1169                 __u64 cp_ver = cur_cp_version(ckpt);
1170                 block_t blk;
1171
1172                 cp_ver |= ((__u64)crc32 << 32);
1173                 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1174
1175                 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1176                 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1177                         update_meta_page(sbi, nm_i->nat_bits +
1178                                         (i << F2FS_BLKSIZE_BITS), blk + i);
1179
1180                 /* Flush all the NAT BITS pages */
1181                 while (get_pages(sbi, F2FS_DIRTY_META)) {
1182                         sync_meta_pages(sbi, META, LONG_MAX);
1183                         if (unlikely(f2fs_cp_error(sbi)))
1184                                 return -EIO;
1185                 }
1186         }
1187
1188         /* need to wait for end_io results */
1189         wait_on_all_pages_writeback(sbi);
1190         if (unlikely(f2fs_cp_error(sbi)))
1191                 return -EIO;
1192
1193         /* write out checkpoint buffer at block 0 */
1194         update_meta_page(sbi, ckpt, start_blk++);
1195
1196         for (i = 1; i < 1 + cp_payload_blks; i++)
1197                 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1198                                                         start_blk++);
1199
1200         if (orphan_num) {
1201                 write_orphan_inodes(sbi, start_blk);
1202                 start_blk += orphan_blocks;
1203         }
1204
1205         write_data_summaries(sbi, start_blk);
1206         start_blk += data_sum_blocks;
1207
1208         /* Record write statistics in the hot node summary */
1209         kbytes_written = sbi->kbytes_written;
1210         if (sb->s_bdev->bd_part)
1211                 kbytes_written += BD_PART_WRITTEN(sbi);
1212
1213         seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1214
1215         if (__remain_node_summaries(cpc->reason)) {
1216                 write_node_summaries(sbi, start_blk);
1217                 start_blk += NR_CURSEG_NODE_TYPE;
1218         }
1219
1220         /* writeout checkpoint block */
1221         update_meta_page(sbi, ckpt, start_blk);
1222
1223         /* wait for previous submitted node/meta pages writeback */
1224         wait_on_all_pages_writeback(sbi);
1225
1226         if (unlikely(f2fs_cp_error(sbi)))
1227                 return -EIO;
1228
1229         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1230         filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1231
1232         /* update user_block_counts */
1233         sbi->last_valid_block_count = sbi->total_valid_block_count;
1234         percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1235
1236         /* Here, we only have one bio having CP pack */
1237         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1238
1239         /* wait for previous submitted meta pages writeback */
1240         wait_on_all_pages_writeback(sbi);
1241
1242         release_ino_entry(sbi, false);
1243
1244         if (unlikely(f2fs_cp_error(sbi)))
1245                 return -EIO;
1246
1247         clear_sbi_flag(sbi, SBI_IS_DIRTY);
1248         clear_sbi_flag(sbi, SBI_NEED_CP);
1249         __set_cp_next_pack(sbi);
1250
1251         /*
1252          * redirty superblock if metadata like node page or inode cache is
1253          * updated during writing checkpoint.
1254          */
1255         if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1256                         get_pages(sbi, F2FS_DIRTY_IMETA))
1257                 set_sbi_flag(sbi, SBI_IS_DIRTY);
1258
1259         f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1260
1261         return 0;
1262 }
1263
1264 /*
1265  * We guarantee that this checkpoint procedure will not fail.
1266  */
1267 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1268 {
1269         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1270         unsigned long long ckpt_ver;
1271         int err = 0;
1272
1273         mutex_lock(&sbi->cp_mutex);
1274
1275         if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1276                 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1277                 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1278                 goto out;
1279         if (unlikely(f2fs_cp_error(sbi))) {
1280                 err = -EIO;
1281                 goto out;
1282         }
1283         if (f2fs_readonly(sbi->sb)) {
1284                 err = -EROFS;
1285                 goto out;
1286         }
1287
1288         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1289
1290         err = block_operations(sbi);
1291         if (err)
1292                 goto out;
1293
1294         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1295
1296         f2fs_flush_merged_writes(sbi);
1297
1298         /* this is the case of multiple fstrims without any changes */
1299         if (cpc->reason & CP_DISCARD) {
1300                 if (!exist_trim_candidates(sbi, cpc)) {
1301                         unblock_operations(sbi);
1302                         goto out;
1303                 }
1304
1305                 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1306                                 SIT_I(sbi)->dirty_sentries == 0 &&
1307                                 prefree_segments(sbi) == 0) {
1308                         flush_sit_entries(sbi, cpc);
1309                         clear_prefree_segments(sbi, cpc);
1310                         unblock_operations(sbi);
1311                         goto out;
1312                 }
1313         }
1314
1315         /*
1316          * update checkpoint pack index
1317          * Increase the version number so that
1318          * SIT entries and seg summaries are written at correct place
1319          */
1320         ckpt_ver = cur_cp_version(ckpt);
1321         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1322
1323         /* write cached NAT/SIT entries to NAT/SIT area */
1324         flush_nat_entries(sbi, cpc);
1325         flush_sit_entries(sbi, cpc);
1326
1327         /* unlock all the fs_lock[] in do_checkpoint() */
1328         err = do_checkpoint(sbi, cpc);
1329         if (err)
1330                 release_discard_addrs(sbi);
1331         else
1332                 clear_prefree_segments(sbi, cpc);
1333
1334         unblock_operations(sbi);
1335         stat_inc_cp_count(sbi->stat_info);
1336
1337         if (cpc->reason & CP_RECOVERY)
1338                 f2fs_msg(sbi->sb, KERN_NOTICE,
1339                         "checkpoint: version = %llx", ckpt_ver);
1340
1341         /* do checkpoint periodically */
1342         f2fs_update_time(sbi, CP_TIME);
1343         trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1344 out:
1345         mutex_unlock(&sbi->cp_mutex);
1346         return err;
1347 }
1348
1349 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1350 {
1351         int i;
1352
1353         for (i = 0; i < MAX_INO_ENTRY; i++) {
1354                 struct inode_management *im = &sbi->im[i];
1355
1356                 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1357                 spin_lock_init(&im->ino_lock);
1358                 INIT_LIST_HEAD(&im->ino_list);
1359                 im->ino_num = 0;
1360         }
1361
1362         sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1363                         NR_CURSEG_TYPE - __cp_payload(sbi)) *
1364                                 F2FS_ORPHANS_PER_BLOCK;
1365 }
1366
1367 int __init create_checkpoint_caches(void)
1368 {
1369         ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1370                         sizeof(struct ino_entry));
1371         if (!ino_entry_slab)
1372                 return -ENOMEM;
1373         inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1374                         sizeof(struct inode_entry));
1375         if (!inode_entry_slab) {
1376                 kmem_cache_destroy(ino_entry_slab);
1377                 return -ENOMEM;
1378         }
1379         return 0;
1380 }
1381
1382 void destroy_checkpoint_caches(void)
1383 {
1384         kmem_cache_destroy(ino_entry_slab);
1385         kmem_cache_destroy(inode_entry_slab);
1386 }