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/events/f2fs.h>
  24
  25 static struct kmem_cache *orphan_entry_slab;
  26 static struct kmem_cache *inode_entry_slab;
  27
  28 /*
  29  * We guarantee no failure on the returned page.
  30  */
  31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  32 {
  33         struct address_space *mapping = META_MAPPING(sbi);
  34         struct page *page = NULL;
  35 repeat:
  36         page = grab_cache_page(mapping, index);
  37         if (!page) {
  38                 cond_resched();
  39                 goto repeat;
  40         }
  41
  42         /* We wait writeback only inside grab_meta_page() */
  43         wait_on_page_writeback(page);
  44         SetPageUptodate(page);
  45         return page;
  46 }
  47
  48 /*
  49  * We guarantee no failure on the returned page.
  50  */
  51 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  52 {
  53         struct address_space *mapping = META_MAPPING(sbi);
  54         struct page *page;
  55 repeat:
  56         page = grab_cache_page(mapping, index);
  57         if (!page) {
  58                 cond_resched();
  59                 goto repeat;
  60         }
  61         if (PageUptodate(page))
  62                 goto out;
  63
  64         if (f2fs_submit_page_bio(sbi, page, index,
  65                                 READ_SYNC | REQ_META | REQ_PRIO))
  66                 goto repeat;
  67
  68         lock_page(page);
  69         if (unlikely(page->mapping != mapping)) {
  70                 f2fs_put_page(page, 1);
  71                 goto repeat;
  72         }
  73 out:
  74         mark_page_accessed(page);
  75         return page;
  76 }
  77
  78 static int f2fs_write_meta_page(struct page *page,
  79                                 struct writeback_control *wbc)
  80 {
  81         struct inode *inode = page->mapping->host;
  82         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
  83
  84         /* Should not write any meta pages, if any IO error was occurred */
  85         if (unlikely(sbi->por_doing ||
  86                         is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
  87                 goto redirty_out;
  88
  89         if (wbc->for_reclaim)
  90                 goto redirty_out;
  91
  92         wait_on_page_writeback(page);
  93
  94         write_meta_page(sbi, page);
  95         dec_page_count(sbi, F2FS_DIRTY_META);
  96         unlock_page(page);
  97         return 0;
  98
  99 redirty_out:
 100         dec_page_count(sbi, F2FS_DIRTY_META);
 101         wbc->pages_skipped++;
 102         set_page_dirty(page);
 103         return AOP_WRITEPAGE_ACTIVATE;
 104 }
 105
 106 static int f2fs_write_meta_pages(struct address_space *mapping,
 107                                 struct writeback_control *wbc)
 108 {
 109         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 110         int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
 111         long written;
 112
 113         if (wbc->for_kupdate)
 114                 return 0;
 115
 116         /* collect a number of dirty meta pages and write together */
 117         if (get_pages(sbi, F2FS_DIRTY_META) < nrpages)
 118                 return 0;
 119
 120         /* if mounting is failed, skip writing node pages */
 121         mutex_lock(&sbi->cp_mutex);
 122         written = sync_meta_pages(sbi, META, nrpages);
 123         mutex_unlock(&sbi->cp_mutex);
 124         wbc->nr_to_write -= written;
 125         return 0;
 126 }
 127
 128 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 129                                                 long nr_to_write)
 130 {
 131         struct address_space *mapping = META_MAPPING(sbi);
 132         pgoff_t index = 0, end = LONG_MAX;
 133         struct pagevec pvec;
 134         long nwritten = 0;
 135         struct writeback_control wbc = {
 136                 .for_reclaim = 0,
 137         };
 138
 139         pagevec_init(&pvec, 0);
 140
 141         while (index <= end) {
 142                 int i, nr_pages;
 143                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 144                                 PAGECACHE_TAG_DIRTY,
 145                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 146                 if (unlikely(nr_pages == 0))
 147                         break;
 148
 149                 for (i = 0; i < nr_pages; i++) {
 150                         struct page *page = pvec.pages[i];
 151                         lock_page(page);
 152                         f2fs_bug_on(page->mapping != mapping);
 153                         f2fs_bug_on(!PageDirty(page));
 154                         clear_page_dirty_for_io(page);
 155                         if (f2fs_write_meta_page(page, &wbc)) {
 156                                 unlock_page(page);
 157                                 break;
 158                         }
 159                         nwritten++;
 160                         if (unlikely(nwritten >= nr_to_write))
 161                                 break;
 162                 }
 163                 pagevec_release(&pvec);
 164                 cond_resched();
 165         }
 166
 167         if (nwritten)
 168                 f2fs_submit_merged_bio(sbi, type, WRITE);
 169
 170         return nwritten;
 171 }
 172
 173 static int f2fs_set_meta_page_dirty(struct page *page)
 174 {
 175         struct address_space *mapping = page->mapping;
 176         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 177
 178         trace_f2fs_set_page_dirty(page, META);
 179
 180         SetPageUptodate(page);
 181         if (!PageDirty(page)) {
 182                 __set_page_dirty_nobuffers(page);
 183                 inc_page_count(sbi, F2FS_DIRTY_META);
 184                 return 1;
 185         }
 186         return 0;
 187 }
 188
 189 const struct address_space_operations f2fs_meta_aops = {
 190         .writepage      = f2fs_write_meta_page,
 191         .writepages     = f2fs_write_meta_pages,
 192         .set_page_dirty = f2fs_set_meta_page_dirty,
 193 };
 194
 195 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 196 {
 197         int err = 0;
 198
 199         spin_lock(&sbi->orphan_inode_lock);
 200         if (unlikely(sbi->n_orphans >= sbi->max_orphans))
 201                 err = -ENOSPC;
 202         else
 203                 sbi->n_orphans++;
 204         spin_unlock(&sbi->orphan_inode_lock);
 205
 206         return err;
 207 }
 208
 209 void release_orphan_inode(struct f2fs_sb_info *sbi)
 210 {
 211         spin_lock(&sbi->orphan_inode_lock);
 212         f2fs_bug_on(sbi->n_orphans == 0);
 213         sbi->n_orphans--;
 214         spin_unlock(&sbi->orphan_inode_lock);
 215 }
 216
 217 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 218 {
 219         struct list_head *head, *this;
 220         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 221
 222         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 223         new->ino = ino;
 224
 225         spin_lock(&sbi->orphan_inode_lock);
 226         head = &sbi->orphan_inode_list;
 227         list_for_each(this, head) {
 228                 orphan = list_entry(this, struct orphan_inode_entry, list);
 229                 if (orphan->ino == ino) {
 230                         spin_unlock(&sbi->orphan_inode_lock);
 231                         kmem_cache_free(orphan_entry_slab, new);
 232                         return;
 233                 }
 234
 235                 if (orphan->ino > ino)
 236                         break;
 237                 orphan = NULL;
 238         }
 239
 240         /* add new_oentry into list which is sorted by inode number */
 241         if (orphan)
 242                 list_add(&new->list, this->prev);
 243         else
 244                 list_add_tail(&new->list, head);
 245         spin_unlock(&sbi->orphan_inode_lock);
 246 }
 247
 248 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 249 {
 250         struct list_head *head;
 251         struct orphan_inode_entry *orphan;
 252
 253         spin_lock(&sbi->orphan_inode_lock);
 254         head = &sbi->orphan_inode_list;
 255         list_for_each_entry(orphan, head, list) {
 256                 if (orphan->ino == ino) {
 257                         list_del(&orphan->list);
 258                         kmem_cache_free(orphan_entry_slab, orphan);
 259                         f2fs_bug_on(sbi->n_orphans == 0);
 260                         sbi->n_orphans--;
 261                         break;
 262                 }
 263         }
 264         spin_unlock(&sbi->orphan_inode_lock);
 265 }
 266
 267 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 268 {
 269         struct inode *inode = f2fs_iget(sbi->sb, ino);
 270         f2fs_bug_on(IS_ERR(inode));
 271         clear_nlink(inode);
 272
 273         /* truncate all the data during iput */
 274         iput(inode);
 275 }
 276
 277 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
 278 {
 279         block_t start_blk, orphan_blkaddr, i, j;
 280
 281         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 282                 return;
 283
 284         sbi->por_doing = true;
 285         start_blk = __start_cp_addr(sbi) + 1;
 286         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 287
 288         for (i = 0; i < orphan_blkaddr; i++) {
 289                 struct page *page = get_meta_page(sbi, start_blk + i);
 290                 struct f2fs_orphan_block *orphan_blk;
 291
 292                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 293                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 294                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 295                         recover_orphan_inode(sbi, ino);
 296                 }
 297                 f2fs_put_page(page, 1);
 298         }
 299         /* clear Orphan Flag */
 300         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 301         sbi->por_doing = false;
 302         return;
 303 }
 304
 305 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 306 {
 307         struct list_head *head;
 308         struct f2fs_orphan_block *orphan_blk = NULL;
 309         unsigned int nentries = 0;
 310         unsigned short index;
 311         unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
 312                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 313         struct page *page = NULL;
 314         struct page *pages[orphan_blocks];
 315         struct orphan_inode_entry *orphan = NULL;
 316
 317         for (index = 0; index < orphan_blocks; index++)
 318                 pages[index] = grab_meta_page(sbi, start_blk + index);
 319
 320         index = 1;
 321         spin_lock(&sbi->orphan_inode_lock);
 322         head = &sbi->orphan_inode_list;
 323
 324         /* loop for each orphan inode entry and write them in Jornal block */
 325         list_for_each_entry(orphan, head, list) {
 326                 if (!page) {
 327                         page = pages[index - 1];
 328                         orphan_blk =
 329                                 (struct f2fs_orphan_block *)page_address(page);
 330                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 331                 }
 332
 333                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 334
 335                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 336                         /*
 337                          * an orphan block is full of 1020 entries,
 338                          * then we need to flush current orphan blocks
 339                          * and bring another one in memory
 340                          */
 341                         orphan_blk->blk_addr = cpu_to_le16(index);
 342                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 343                         orphan_blk->entry_count = cpu_to_le32(nentries);
 344                         set_page_dirty(page);
 345                         f2fs_put_page(page, 1);
 346                         index++;
 347                         nentries = 0;
 348                         page = NULL;
 349                 }
 350         }
 351
 352         if (page) {
 353                 orphan_blk->blk_addr = cpu_to_le16(index);
 354                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 355                 orphan_blk->entry_count = cpu_to_le32(nentries);
 356                 set_page_dirty(page);
 357                 f2fs_put_page(page, 1);
 358         }
 359
 360         spin_unlock(&sbi->orphan_inode_lock);
 361 }
 362
 363 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 364                                 block_t cp_addr, unsigned long long *version)
 365 {
 366         struct page *cp_page_1, *cp_page_2 = NULL;
 367         unsigned long blk_size = sbi->blocksize;
 368         struct f2fs_checkpoint *cp_block;
 369         unsigned long long cur_version = 0, pre_version = 0;
 370         size_t crc_offset;
 371         __u32 crc = 0;
 372
 373         /* Read the 1st cp block in this CP pack */
 374         cp_page_1 = get_meta_page(sbi, cp_addr);
 375
 376         /* get the version number */
 377         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 378         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 379         if (crc_offset >= blk_size)
 380                 goto invalid_cp1;
 381
 382         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 383         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 384                 goto invalid_cp1;
 385
 386         pre_version = cur_cp_version(cp_block);
 387
 388         /* Read the 2nd cp block in this CP pack */
 389         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 390         cp_page_2 = get_meta_page(sbi, cp_addr);
 391
 392         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 393         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 394         if (crc_offset >= blk_size)
 395                 goto invalid_cp2;
 396
 397         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 398         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 399                 goto invalid_cp2;
 400
 401         cur_version = cur_cp_version(cp_block);
 402
 403         if (cur_version == pre_version) {
 404                 *version = cur_version;
 405                 f2fs_put_page(cp_page_2, 1);
 406                 return cp_page_1;
 407         }
 408 invalid_cp2:
 409         f2fs_put_page(cp_page_2, 1);
 410 invalid_cp1:
 411         f2fs_put_page(cp_page_1, 1);
 412         return NULL;
 413 }
 414
 415 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 416 {
 417         struct f2fs_checkpoint *cp_block;
 418         struct f2fs_super_block *fsb = sbi->raw_super;
 419         struct page *cp1, *cp2, *cur_page;
 420         unsigned long blk_size = sbi->blocksize;
 421         unsigned long long cp1_version = 0, cp2_version = 0;
 422         unsigned long long cp_start_blk_no;
 423
 424         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 425         if (!sbi->ckpt)
 426                 return -ENOMEM;
 427         /*
 428          * Finding out valid cp block involves read both
 429          * sets( cp pack1 and cp pack 2)
 430          */
 431         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 432         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 433
 434         /* The second checkpoint pack should start at the next segment */
 435         cp_start_blk_no += ((unsigned long long)1) <<
 436                                 le32_to_cpu(fsb->log_blocks_per_seg);
 437         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 438
 439         if (cp1 && cp2) {
 440                 if (ver_after(cp2_version, cp1_version))
 441                         cur_page = cp2;
 442                 else
 443                         cur_page = cp1;
 444         } else if (cp1) {
 445                 cur_page = cp1;
 446         } else if (cp2) {
 447                 cur_page = cp2;
 448         } else {
 449                 goto fail_no_cp;
 450         }
 451
 452         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 453         memcpy(sbi->ckpt, cp_block, blk_size);
 454
 455         f2fs_put_page(cp1, 1);
 456         f2fs_put_page(cp2, 1);
 457         return 0;
 458
 459 fail_no_cp:
 460         kfree(sbi->ckpt);
 461         return -EINVAL;
 462 }
 463
 464 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 465 {
 466         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 467         struct list_head *head = &sbi->dir_inode_list;
 468         struct list_head *this;
 469
 470         list_for_each(this, head) {
 471                 struct dir_inode_entry *entry;
 472                 entry = list_entry(this, struct dir_inode_entry, list);
 473                 if (unlikely(entry->inode == inode))
 474                         return -EEXIST;
 475         }
 476         list_add_tail(&new->list, head);
 477         stat_inc_dirty_dir(sbi);
 478         return 0;
 479 }
 480
 481 void set_dirty_dir_page(struct inode *inode, struct page *page)
 482 {
 483         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 484         struct dir_inode_entry *new;
 485
 486         if (!S_ISDIR(inode->i_mode))
 487                 return;
 488
 489         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 490         new->inode = inode;
 491         INIT_LIST_HEAD(&new->list);
 492
 493         spin_lock(&sbi->dir_inode_lock);
 494         if (__add_dirty_inode(inode, new))
 495                 kmem_cache_free(inode_entry_slab, new);
 496
 497         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 498         inode_inc_dirty_dents(inode);
 499         SetPagePrivate(page);
 500         spin_unlock(&sbi->dir_inode_lock);
 501 }
 502
 503 void add_dirty_dir_inode(struct inode *inode)
 504 {
 505         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 506         struct dir_inode_entry *new =
 507                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 508
 509         new->inode = inode;
 510         INIT_LIST_HEAD(&new->list);
 511
 512         spin_lock(&sbi->dir_inode_lock);
 513         if (__add_dirty_inode(inode, new))
 514                 kmem_cache_free(inode_entry_slab, new);
 515         spin_unlock(&sbi->dir_inode_lock);
 516 }
 517
 518 void remove_dirty_dir_inode(struct inode *inode)
 519 {
 520         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 521
 522         struct list_head *this, *head;
 523
 524         if (!S_ISDIR(inode->i_mode))
 525                 return;
 526
 527         spin_lock(&sbi->dir_inode_lock);
 528         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 529                 spin_unlock(&sbi->dir_inode_lock);
 530                 return;
 531         }
 532
 533         head = &sbi->dir_inode_list;
 534         list_for_each(this, head) {
 535                 struct dir_inode_entry *entry;
 536                 entry = list_entry(this, struct dir_inode_entry, list);
 537                 if (entry->inode == inode) {
 538                         list_del(&entry->list);
 539                         kmem_cache_free(inode_entry_slab, entry);
 540                         stat_dec_dirty_dir(sbi);
 541                         break;
 542                 }
 543         }
 544         spin_unlock(&sbi->dir_inode_lock);
 545
 546         /* Only from the recovery routine */
 547         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 548                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 549                 iput(inode);
 550         }
 551 }
 552
 553 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 554 {
 555
 556         struct list_head *this, *head;
 557         struct inode *inode = NULL;
 558
 559         spin_lock(&sbi->dir_inode_lock);
 560
 561         head = &sbi->dir_inode_list;
 562         list_for_each(this, head) {
 563                 struct dir_inode_entry *entry;
 564                 entry = list_entry(this, struct dir_inode_entry, list);
 565                 if (entry->inode->i_ino == ino) {
 566                         inode = entry->inode;
 567                         break;
 568                 }
 569         }
 570         spin_unlock(&sbi->dir_inode_lock);
 571         return inode;
 572 }
 573
 574 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 575 {
 576         struct list_head *head;
 577         struct dir_inode_entry *entry;
 578         struct inode *inode;
 579 retry:
 580         spin_lock(&sbi->dir_inode_lock);
 581
 582         head = &sbi->dir_inode_list;
 583         if (list_empty(head)) {
 584                 spin_unlock(&sbi->dir_inode_lock);
 585                 return;
 586         }
 587         entry = list_entry(head->next, struct dir_inode_entry, list);
 588         inode = igrab(entry->inode);
 589         spin_unlock(&sbi->dir_inode_lock);
 590         if (inode) {
 591                 filemap_flush(inode->i_mapping);
 592                 iput(inode);
 593         } else {
 594                 /*
 595                  * We should submit bio, since it exists several
 596                  * wribacking dentry pages in the freeing inode.
 597                  */
 598                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 599         }
 600         goto retry;
 601 }
 602
 603 /*
 604  * Freeze all the FS-operations for checkpoint.
 605  */
 606 static void block_operations(struct f2fs_sb_info *sbi)
 607 {
 608         struct writeback_control wbc = {
 609                 .sync_mode = WB_SYNC_ALL,
 610                 .nr_to_write = LONG_MAX,
 611                 .for_reclaim = 0,
 612         };
 613         struct blk_plug plug;
 614
 615         blk_start_plug(&plug);
 616
 617 retry_flush_dents:
 618         f2fs_lock_all(sbi);
 619         /* write all the dirty dentry pages */
 620         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 621                 f2fs_unlock_all(sbi);
 622                 sync_dirty_dir_inodes(sbi);
 623                 goto retry_flush_dents;
 624         }
 625
 626         /*
 627          * POR: we should ensure that there is no dirty node pages
 628          * until finishing nat/sit flush.
 629          */
 630 retry_flush_nodes:
 631         mutex_lock(&sbi->node_write);
 632
 633         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 634                 mutex_unlock(&sbi->node_write);
 635                 sync_node_pages(sbi, 0, &wbc);
 636                 goto retry_flush_nodes;
 637         }
 638         blk_finish_plug(&plug);
 639 }
 640
 641 static void unblock_operations(struct f2fs_sb_info *sbi)
 642 {
 643         mutex_unlock(&sbi->node_write);
 644         f2fs_unlock_all(sbi);
 645 }
 646
 647 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 648 {
 649         DEFINE_WAIT(wait);
 650
 651         for (;;) {
 652                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 653
 654                 if (!get_pages(sbi, F2FS_WRITEBACK))
 655                         break;
 656
 657                 io_schedule();
 658         }
 659         finish_wait(&sbi->cp_wait, &wait);
 660 }
 661
 662 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 663 {
 664         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 665         nid_t last_nid = 0;
 666         block_t start_blk;
 667         struct page *cp_page;
 668         unsigned int data_sum_blocks, orphan_blocks;
 669         __u32 crc32 = 0;
 670         void *kaddr;
 671         int i;
 672
 673         /* Flush all the NAT/SIT pages */
 674         while (get_pages(sbi, F2FS_DIRTY_META))
 675                 sync_meta_pages(sbi, META, LONG_MAX);
 676
 677         next_free_nid(sbi, &last_nid);
 678
 679         /*
 680          * modify checkpoint
 681          * version number is already updated
 682          */
 683         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 684         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 685         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 686         for (i = 0; i < 3; i++) {
 687                 ckpt->cur_node_segno[i] =
 688                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 689                 ckpt->cur_node_blkoff[i] =
 690                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 691                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 692                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 693         }
 694         for (i = 0; i < 3; i++) {
 695                 ckpt->cur_data_segno[i] =
 696                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 697                 ckpt->cur_data_blkoff[i] =
 698                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 699                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 700                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 701         }
 702
 703         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 704         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 705         ckpt->next_free_nid = cpu_to_le32(last_nid);
 706
 707         /* 2 cp  + n data seg summary + orphan inode blocks */
 708         data_sum_blocks = npages_for_summary_flush(sbi);
 709         if (data_sum_blocks < 3)
 710                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 711         else
 712                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 713
 714         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 715                                         / F2FS_ORPHANS_PER_BLOCK;
 716         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 717
 718         if (is_umount) {
 719                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 720                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 721                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 722         } else {
 723                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 724                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 725                         data_sum_blocks + orphan_blocks);
 726         }
 727
 728         if (sbi->n_orphans)
 729                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 730         else
 731                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 732
 733         /* update SIT/NAT bitmap */
 734         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 735         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 736
 737         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 738         *((__le32 *)((unsigned char *)ckpt +
 739                                 le32_to_cpu(ckpt->checksum_offset)))
 740                                 = cpu_to_le32(crc32);
 741
 742         start_blk = __start_cp_addr(sbi);
 743
 744         /* write out checkpoint buffer at block 0 */
 745         cp_page = grab_meta_page(sbi, start_blk++);
 746         kaddr = page_address(cp_page);
 747         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 748         set_page_dirty(cp_page);
 749         f2fs_put_page(cp_page, 1);
 750
 751         if (sbi->n_orphans) {
 752                 write_orphan_inodes(sbi, start_blk);
 753                 start_blk += orphan_blocks;
 754         }
 755
 756         write_data_summaries(sbi, start_blk);
 757         start_blk += data_sum_blocks;
 758         if (is_umount) {
 759                 write_node_summaries(sbi, start_blk);
 760                 start_blk += NR_CURSEG_NODE_TYPE;
 761         }
 762
 763         /* writeout checkpoint block */
 764         cp_page = grab_meta_page(sbi, start_blk);
 765         kaddr = page_address(cp_page);
 766         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 767         set_page_dirty(cp_page);
 768         f2fs_put_page(cp_page, 1);
 769
 770         /* wait for previous submitted node/meta pages writeback */
 771         wait_on_all_pages_writeback(sbi);
 772
 773         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 774         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 775
 776         /* update user_block_counts */
 777         sbi->last_valid_block_count = sbi->total_valid_block_count;
 778         sbi->alloc_valid_block_count = 0;
 779
 780         /* Here, we only have one bio having CP pack */
 781         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 782
 783         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 784                 clear_prefree_segments(sbi);
 785                 F2FS_RESET_SB_DIRT(sbi);
 786         }
 787 }
 788
 789 /*
 790  * We guarantee that this checkpoint procedure should not fail.
 791  */
 792 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 793 {
 794         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 795         unsigned long long ckpt_ver;
 796
 797         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 798
 799         mutex_lock(&sbi->cp_mutex);
 800         block_operations(sbi);
 801
 802         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 803
 804         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 805         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 806         f2fs_submit_merged_bio(sbi, META, WRITE);
 807
 808         /*
 809          * update checkpoint pack index
 810          * Increase the version number so that
 811          * SIT entries and seg summaries are written at correct place
 812          */
 813         ckpt_ver = cur_cp_version(ckpt);
 814         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 815
 816         /* write cached NAT/SIT entries to NAT/SIT area */
 817         flush_nat_entries(sbi);
 818         flush_sit_entries(sbi);
 819
 820         /* unlock all the fs_lock[] in do_checkpoint() */
 821         do_checkpoint(sbi, is_umount);
 822
 823         unblock_operations(sbi);
 824         mutex_unlock(&sbi->cp_mutex);
 825
 826         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 827 }
 828
 829 void init_orphan_info(struct f2fs_sb_info *sbi)
 830 {
 831         spin_lock_init(&sbi->orphan_inode_lock);
 832         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 833         sbi->n_orphans = 0;
 834         /*
 835          * considering 512 blocks in a segment 8 blocks are needed for cp
 836          * and log segment summaries. Remaining blocks are used to keep
 837          * orphan entries with the limitation one reserved segment
 838          * for cp pack we can have max 1020*504 orphan entries
 839          */
 840         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 841                                 * F2FS_ORPHANS_PER_BLOCK;
 842 }
 843
 844 int __init create_checkpoint_caches(void)
 845 {
 846         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 847                         sizeof(struct orphan_inode_entry), NULL);
 848         if (!orphan_entry_slab)
 849                 return -ENOMEM;
 850         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 851                         sizeof(struct dir_inode_entry), NULL);
 852         if (!inode_entry_slab) {
 853                 kmem_cache_destroy(orphan_entry_slab);
 854                 return -ENOMEM;
 855         }
 856         return 0;
 857 }
 858
 859 void destroy_checkpoint_caches(void)
 860 {
 861         kmem_cache_destroy(orphan_entry_slab);
 862         kmem_cache_destroy(inode_entry_slab);
 863 }