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 orphan_inode_entry *orphan = NULL;
 315
 316         for (index = 0; index < orphan_blocks; index++)
 317                 grab_meta_page(sbi, start_blk + index);
 318
 319         index = 1;
 320         spin_lock(&sbi->orphan_inode_lock);
 321         head = &sbi->orphan_inode_list;
 322
 323         /* loop for each orphan inode entry and write them in Jornal block */
 324         list_for_each_entry(orphan, head, list) {
 325                 if (!page) {
 326                         page = find_get_page(META_MAPPING(sbi), start_blk++);
 327                         f2fs_bug_on(!page);
 328                         orphan_blk =
 329                                 (struct f2fs_orphan_block *)page_address(page);
 330                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 331                         f2fs_put_page(page, 0);
 332                 }
 333
 334                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 335
 336                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 337                         /*
 338                          * an orphan block is full of 1020 entries,
 339                          * then we need to flush current orphan blocks
 340                          * and bring another one in memory
 341                          */
 342                         orphan_blk->blk_addr = cpu_to_le16(index);
 343                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 344                         orphan_blk->entry_count = cpu_to_le32(nentries);
 345                         set_page_dirty(page);
 346                         f2fs_put_page(page, 1);
 347                         index++;
 348                         nentries = 0;
 349                         page = NULL;
 350                 }
 351         }
 352
 353         if (page) {
 354                 orphan_blk->blk_addr = cpu_to_le16(index);
 355                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 356                 orphan_blk->entry_count = cpu_to_le32(nentries);
 357                 set_page_dirty(page);
 358                 f2fs_put_page(page, 1);
 359         }
 360
 361         spin_unlock(&sbi->orphan_inode_lock);
 362 }
 363
 364 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 365                                 block_t cp_addr, unsigned long long *version)
 366 {
 367         struct page *cp_page_1, *cp_page_2 = NULL;
 368         unsigned long blk_size = sbi->blocksize;
 369         struct f2fs_checkpoint *cp_block;
 370         unsigned long long cur_version = 0, pre_version = 0;
 371         size_t crc_offset;
 372         __u32 crc = 0;
 373
 374         /* Read the 1st cp block in this CP pack */
 375         cp_page_1 = get_meta_page(sbi, cp_addr);
 376
 377         /* get the version number */
 378         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 379         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 380         if (crc_offset >= blk_size)
 381                 goto invalid_cp1;
 382
 383         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 384         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 385                 goto invalid_cp1;
 386
 387         pre_version = cur_cp_version(cp_block);
 388
 389         /* Read the 2nd cp block in this CP pack */
 390         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 391         cp_page_2 = get_meta_page(sbi, cp_addr);
 392
 393         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 394         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 395         if (crc_offset >= blk_size)
 396                 goto invalid_cp2;
 397
 398         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 399         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 400                 goto invalid_cp2;
 401
 402         cur_version = cur_cp_version(cp_block);
 403
 404         if (cur_version == pre_version) {
 405                 *version = cur_version;
 406                 f2fs_put_page(cp_page_2, 1);
 407                 return cp_page_1;
 408         }
 409 invalid_cp2:
 410         f2fs_put_page(cp_page_2, 1);
 411 invalid_cp1:
 412         f2fs_put_page(cp_page_1, 1);
 413         return NULL;
 414 }
 415
 416 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 417 {
 418         struct f2fs_checkpoint *cp_block;
 419         struct f2fs_super_block *fsb = sbi->raw_super;
 420         struct page *cp1, *cp2, *cur_page;
 421         unsigned long blk_size = sbi->blocksize;
 422         unsigned long long cp1_version = 0, cp2_version = 0;
 423         unsigned long long cp_start_blk_no;
 424
 425         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 426         if (!sbi->ckpt)
 427                 return -ENOMEM;
 428         /*
 429          * Finding out valid cp block involves read both
 430          * sets( cp pack1 and cp pack 2)
 431          */
 432         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 433         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 434
 435         /* The second checkpoint pack should start at the next segment */
 436         cp_start_blk_no += ((unsigned long long)1) <<
 437                                 le32_to_cpu(fsb->log_blocks_per_seg);
 438         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 439
 440         if (cp1 && cp2) {
 441                 if (ver_after(cp2_version, cp1_version))
 442                         cur_page = cp2;
 443                 else
 444                         cur_page = cp1;
 445         } else if (cp1) {
 446                 cur_page = cp1;
 447         } else if (cp2) {
 448                 cur_page = cp2;
 449         } else {
 450                 goto fail_no_cp;
 451         }
 452
 453         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 454         memcpy(sbi->ckpt, cp_block, blk_size);
 455
 456         f2fs_put_page(cp1, 1);
 457         f2fs_put_page(cp2, 1);
 458         return 0;
 459
 460 fail_no_cp:
 461         kfree(sbi->ckpt);
 462         return -EINVAL;
 463 }
 464
 465 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 466 {
 467         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 468         struct list_head *head = &sbi->dir_inode_list;
 469         struct list_head *this;
 470
 471         list_for_each(this, head) {
 472                 struct dir_inode_entry *entry;
 473                 entry = list_entry(this, struct dir_inode_entry, list);
 474                 if (unlikely(entry->inode == inode))
 475                         return -EEXIST;
 476         }
 477         list_add_tail(&new->list, head);
 478         stat_inc_dirty_dir(sbi);
 479         return 0;
 480 }
 481
 482 void set_dirty_dir_page(struct inode *inode, struct page *page)
 483 {
 484         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 485         struct dir_inode_entry *new;
 486
 487         if (!S_ISDIR(inode->i_mode))
 488                 return;
 489
 490         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 491         new->inode = inode;
 492         INIT_LIST_HEAD(&new->list);
 493
 494         spin_lock(&sbi->dir_inode_lock);
 495         if (__add_dirty_inode(inode, new))
 496                 kmem_cache_free(inode_entry_slab, new);
 497
 498         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 499         inode_inc_dirty_dents(inode);
 500         SetPagePrivate(page);
 501         spin_unlock(&sbi->dir_inode_lock);
 502 }
 503
 504 void add_dirty_dir_inode(struct inode *inode)
 505 {
 506         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 507         struct dir_inode_entry *new =
 508                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 509
 510         new->inode = inode;
 511         INIT_LIST_HEAD(&new->list);
 512
 513         spin_lock(&sbi->dir_inode_lock);
 514         if (__add_dirty_inode(inode, new))
 515                 kmem_cache_free(inode_entry_slab, new);
 516         spin_unlock(&sbi->dir_inode_lock);
 517 }
 518
 519 void remove_dirty_dir_inode(struct inode *inode)
 520 {
 521         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 522
 523         struct list_head *this, *head;
 524
 525         if (!S_ISDIR(inode->i_mode))
 526                 return;
 527
 528         spin_lock(&sbi->dir_inode_lock);
 529         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 530                 spin_unlock(&sbi->dir_inode_lock);
 531                 return;
 532         }
 533
 534         head = &sbi->dir_inode_list;
 535         list_for_each(this, head) {
 536                 struct dir_inode_entry *entry;
 537                 entry = list_entry(this, struct dir_inode_entry, list);
 538                 if (entry->inode == inode) {
 539                         list_del(&entry->list);
 540                         kmem_cache_free(inode_entry_slab, entry);
 541                         stat_dec_dirty_dir(sbi);
 542                         break;
 543                 }
 544         }
 545         spin_unlock(&sbi->dir_inode_lock);
 546
 547         /* Only from the recovery routine */
 548         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 549                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 550                 iput(inode);
 551         }
 552 }
 553
 554 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 555 {
 556
 557         struct list_head *this, *head;
 558         struct inode *inode = NULL;
 559
 560         spin_lock(&sbi->dir_inode_lock);
 561
 562         head = &sbi->dir_inode_list;
 563         list_for_each(this, head) {
 564                 struct dir_inode_entry *entry;
 565                 entry = list_entry(this, struct dir_inode_entry, list);
 566                 if (entry->inode->i_ino == ino) {
 567                         inode = entry->inode;
 568                         break;
 569                 }
 570         }
 571         spin_unlock(&sbi->dir_inode_lock);
 572         return inode;
 573 }
 574
 575 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 576 {
 577         struct list_head *head;
 578         struct dir_inode_entry *entry;
 579         struct inode *inode;
 580 retry:
 581         spin_lock(&sbi->dir_inode_lock);
 582
 583         head = &sbi->dir_inode_list;
 584         if (list_empty(head)) {
 585                 spin_unlock(&sbi->dir_inode_lock);
 586                 return;
 587         }
 588         entry = list_entry(head->next, struct dir_inode_entry, list);
 589         inode = igrab(entry->inode);
 590         spin_unlock(&sbi->dir_inode_lock);
 591         if (inode) {
 592                 filemap_flush(inode->i_mapping);
 593                 iput(inode);
 594         } else {
 595                 /*
 596                  * We should submit bio, since it exists several
 597                  * wribacking dentry pages in the freeing inode.
 598                  */
 599                 f2fs_submit_merged_bio(sbi, DATA, WRITE);
 600         }
 601         goto retry;
 602 }
 603
 604 /*
 605  * Freeze all the FS-operations for checkpoint.
 606  */
 607 static void block_operations(struct f2fs_sb_info *sbi)
 608 {
 609         struct writeback_control wbc = {
 610                 .sync_mode = WB_SYNC_ALL,
 611                 .nr_to_write = LONG_MAX,
 612                 .for_reclaim = 0,
 613         };
 614         struct blk_plug plug;
 615
 616         blk_start_plug(&plug);
 617
 618 retry_flush_dents:
 619         f2fs_lock_all(sbi);
 620         /* write all the dirty dentry pages */
 621         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 622                 f2fs_unlock_all(sbi);
 623                 sync_dirty_dir_inodes(sbi);
 624                 goto retry_flush_dents;
 625         }
 626
 627         /*
 628          * POR: we should ensure that there is no dirty node pages
 629          * until finishing nat/sit flush.
 630          */
 631 retry_flush_nodes:
 632         mutex_lock(&sbi->node_write);
 633
 634         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 635                 mutex_unlock(&sbi->node_write);
 636                 sync_node_pages(sbi, 0, &wbc);
 637                 goto retry_flush_nodes;
 638         }
 639         blk_finish_plug(&plug);
 640 }
 641
 642 static void unblock_operations(struct f2fs_sb_info *sbi)
 643 {
 644         mutex_unlock(&sbi->node_write);
 645         f2fs_unlock_all(sbi);
 646 }
 647
 648 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 649 {
 650         DEFINE_WAIT(wait);
 651
 652         for (;;) {
 653                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 654
 655                 if (!get_pages(sbi, F2FS_WRITEBACK))
 656                         break;
 657
 658                 io_schedule();
 659         }
 660         finish_wait(&sbi->cp_wait, &wait);
 661 }
 662
 663 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 664 {
 665         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 666         nid_t last_nid = 0;
 667         block_t start_blk;
 668         struct page *cp_page;
 669         unsigned int data_sum_blocks, orphan_blocks;
 670         __u32 crc32 = 0;
 671         void *kaddr;
 672         int i;
 673
 674         /* Flush all the NAT/SIT pages */
 675         while (get_pages(sbi, F2FS_DIRTY_META))
 676                 sync_meta_pages(sbi, META, LONG_MAX);
 677
 678         next_free_nid(sbi, &last_nid);
 679
 680         /*
 681          * modify checkpoint
 682          * version number is already updated
 683          */
 684         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 685         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 686         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 687         for (i = 0; i < 3; i++) {
 688                 ckpt->cur_node_segno[i] =
 689                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 690                 ckpt->cur_node_blkoff[i] =
 691                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 692                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 693                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 694         }
 695         for (i = 0; i < 3; i++) {
 696                 ckpt->cur_data_segno[i] =
 697                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 698                 ckpt->cur_data_blkoff[i] =
 699                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 700                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 701                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 702         }
 703
 704         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 705         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 706         ckpt->next_free_nid = cpu_to_le32(last_nid);
 707
 708         /* 2 cp  + n data seg summary + orphan inode blocks */
 709         data_sum_blocks = npages_for_summary_flush(sbi);
 710         if (data_sum_blocks < 3)
 711                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 712         else
 713                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 714
 715         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 716                                         / F2FS_ORPHANS_PER_BLOCK;
 717         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 718
 719         if (is_umount) {
 720                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 721                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 722                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 723         } else {
 724                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 725                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 726                         data_sum_blocks + orphan_blocks);
 727         }
 728
 729         if (sbi->n_orphans)
 730                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 731         else
 732                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 733
 734         /* update SIT/NAT bitmap */
 735         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 736         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 737
 738         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 739         *((__le32 *)((unsigned char *)ckpt +
 740                                 le32_to_cpu(ckpt->checksum_offset)))
 741                                 = cpu_to_le32(crc32);
 742
 743         start_blk = __start_cp_addr(sbi);
 744
 745         /* write out checkpoint buffer at block 0 */
 746         cp_page = grab_meta_page(sbi, start_blk++);
 747         kaddr = page_address(cp_page);
 748         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 749         set_page_dirty(cp_page);
 750         f2fs_put_page(cp_page, 1);
 751
 752         if (sbi->n_orphans) {
 753                 write_orphan_inodes(sbi, start_blk);
 754                 start_blk += orphan_blocks;
 755         }
 756
 757         write_data_summaries(sbi, start_blk);
 758         start_blk += data_sum_blocks;
 759         if (is_umount) {
 760                 write_node_summaries(sbi, start_blk);
 761                 start_blk += NR_CURSEG_NODE_TYPE;
 762         }
 763
 764         /* writeout checkpoint block */
 765         cp_page = grab_meta_page(sbi, start_blk);
 766         kaddr = page_address(cp_page);
 767         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 768         set_page_dirty(cp_page);
 769         f2fs_put_page(cp_page, 1);
 770
 771         /* wait for previous submitted node/meta pages writeback */
 772         wait_on_all_pages_writeback(sbi);
 773
 774         filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 775         filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 776
 777         /* update user_block_counts */
 778         sbi->last_valid_block_count = sbi->total_valid_block_count;
 779         sbi->alloc_valid_block_count = 0;
 780
 781         /* Here, we only have one bio having CP pack */
 782         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 783
 784         if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
 785                 clear_prefree_segments(sbi);
 786                 F2FS_RESET_SB_DIRT(sbi);
 787         }
 788 }
 789
 790 /*
 791  * We guarantee that this checkpoint procedure should not fail.
 792  */
 793 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 794 {
 795         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 796         unsigned long long ckpt_ver;
 797
 798         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 799
 800         mutex_lock(&sbi->cp_mutex);
 801         block_operations(sbi);
 802
 803         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 804
 805         f2fs_submit_merged_bio(sbi, DATA, WRITE);
 806         f2fs_submit_merged_bio(sbi, NODE, WRITE);
 807         f2fs_submit_merged_bio(sbi, META, WRITE);
 808
 809         /*
 810          * update checkpoint pack index
 811          * Increase the version number so that
 812          * SIT entries and seg summaries are written at correct place
 813          */
 814         ckpt_ver = cur_cp_version(ckpt);
 815         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 816
 817         /* write cached NAT/SIT entries to NAT/SIT area */
 818         flush_nat_entries(sbi);
 819         flush_sit_entries(sbi);
 820
 821         /* unlock all the fs_lock[] in do_checkpoint() */
 822         do_checkpoint(sbi, is_umount);
 823
 824         unblock_operations(sbi);
 825         mutex_unlock(&sbi->cp_mutex);
 826
 827         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 828 }
 829
 830 void init_orphan_info(struct f2fs_sb_info *sbi)
 831 {
 832         spin_lock_init(&sbi->orphan_inode_lock);
 833         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 834         sbi->n_orphans = 0;
 835         /*
 836          * considering 512 blocks in a segment 8 blocks are needed for cp
 837          * and log segment summaries. Remaining blocks are used to keep
 838          * orphan entries with the limitation one reserved segment
 839          * for cp pack we can have max 1020*504 orphan entries
 840          */
 841         sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 842                                 * F2FS_ORPHANS_PER_BLOCK;
 843 }
 844
 845 int __init create_checkpoint_caches(void)
 846 {
 847         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 848                         sizeof(struct orphan_inode_entry), NULL);
 849         if (!orphan_entry_slab)
 850                 return -ENOMEM;
 851         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 852                         sizeof(struct dir_inode_entry), NULL);
 853         if (!inode_entry_slab) {
 854                 kmem_cache_destroy(orphan_entry_slab);
 855                 return -ENOMEM;
 856         }
 857         return 0;
 858 }
 859
 860 void destroy_checkpoint_caches(void)
 861 {
 862         kmem_cache_destroy(orphan_entry_slab);
 863         kmem_cache_destroy(inode_entry_slab);
 864 }