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