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_readpage(sbi, page, index, READ_SYNC | REQ_META | REQ_PRIO))
  65                 goto repeat;
  66
  67         lock_page(page);
  68         if (page->mapping != mapping) {
  69                 f2fs_put_page(page, 1);
  70                 goto repeat;
  71         }
  72 out:
  73         mark_page_accessed(page);
  74         return page;
  75 }
  76
  77 static int f2fs_write_meta_page(struct page *page,
  78                                 struct writeback_control *wbc)
  79 {
  80         struct inode *inode = page->mapping->host;
  81         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
  82
  83         /* Should not write any meta pages, if any IO error was occurred */
  84         if (wbc->for_reclaim || sbi->por_doing ||
  85                         is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)) {
  86                 dec_page_count(sbi, F2FS_DIRTY_META);
  87                 wbc->pages_skipped++;
  88                 set_page_dirty(page);
  89                 return AOP_WRITEPAGE_ACTIVATE;
  90         }
  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
 100 static int f2fs_write_meta_pages(struct address_space *mapping,
 101                                 struct writeback_control *wbc)
 102 {
 103         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 104         struct block_device *bdev = sbi->sb->s_bdev;
 105         long written;
 106
 107         if (wbc->for_kupdate)
 108                 return 0;
 109
 110         if (get_pages(sbi, F2FS_DIRTY_META) == 0)
 111                 return 0;
 112
 113         /* if mounting is failed, skip writing node pages */
 114         mutex_lock(&sbi->cp_mutex);
 115         written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev));
 116         mutex_unlock(&sbi->cp_mutex);
 117         wbc->nr_to_write -= written;
 118         return 0;
 119 }
 120
 121 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 122                                                 long nr_to_write)
 123 {
 124         struct address_space *mapping = sbi->meta_inode->i_mapping;
 125         pgoff_t index = 0, end = LONG_MAX;
 126         struct pagevec pvec;
 127         long nwritten = 0;
 128         struct writeback_control wbc = {
 129                 .for_reclaim = 0,
 130         };
 131
 132         pagevec_init(&pvec, 0);
 133
 134         while (index <= end) {
 135                 int i, nr_pages;
 136                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 137                                 PAGECACHE_TAG_DIRTY,
 138                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
 139                 if (nr_pages == 0)
 140                         break;
 141
 142                 for (i = 0; i < nr_pages; i++) {
 143                         struct page *page = pvec.pages[i];
 144                         lock_page(page);
 145                         f2fs_bug_on(page->mapping != mapping);
 146                         f2fs_bug_on(!PageDirty(page));
 147                         clear_page_dirty_for_io(page);
 148                         if (f2fs_write_meta_page(page, &wbc)) {
 149                                 unlock_page(page);
 150                                 break;
 151                         }
 152                         if (nwritten++ >= nr_to_write)
 153                                 break;
 154                 }
 155                 pagevec_release(&pvec);
 156                 cond_resched();
 157         }
 158
 159         if (nwritten)
 160                 f2fs_submit_bio(sbi, type, nr_to_write == LONG_MAX);
 161
 162         return nwritten;
 163 }
 164
 165 static int f2fs_set_meta_page_dirty(struct page *page)
 166 {
 167         struct address_space *mapping = page->mapping;
 168         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
 169
 170         trace_f2fs_set_page_dirty(page, META);
 171
 172         SetPageUptodate(page);
 173         if (!PageDirty(page)) {
 174                 __set_page_dirty_nobuffers(page);
 175                 inc_page_count(sbi, F2FS_DIRTY_META);
 176                 return 1;
 177         }
 178         return 0;
 179 }
 180
 181 const struct address_space_operations f2fs_meta_aops = {
 182         .writepage      = f2fs_write_meta_page,
 183         .writepages     = f2fs_write_meta_pages,
 184         .set_page_dirty = f2fs_set_meta_page_dirty,
 185 };
 186
 187 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
 188 {
 189         unsigned int max_orphans;
 190         int err = 0;
 191
 192         /*
 193          * considering 512 blocks in a segment 8 blocks are needed for cp
 194          * and log segment summaries. Remaining blocks are used to keep
 195          * orphan entries with the limitation one reserved segment
 196          * for cp pack we can have max 1020*504 orphan entries
 197          */
 198         max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
 199                                 * F2FS_ORPHANS_PER_BLOCK;
 200         mutex_lock(&sbi->orphan_inode_mutex);
 201         if (sbi->n_orphans >= max_orphans)
 202                 err = -ENOSPC;
 203         else
 204                 sbi->n_orphans++;
 205         mutex_unlock(&sbi->orphan_inode_mutex);
 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         struct page *page = NULL;
 307         unsigned int nentries = 0;
 308         unsigned short index = 1;
 309         unsigned short orphan_blocks;
 310         struct orphan_inode_entry *orphan = NULL;
 311
 312         orphan_blocks = (unsigned short)((sbi->n_orphans +
 313                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 314
 315         mutex_lock(&sbi->orphan_inode_mutex);
 316         head = &sbi->orphan_inode_list;
 317
 318         /* loop for each orphan inode entry and write them in Jornal block */
 319         list_for_each_entry(orphan, head, list) {
 320                 if (!page) {
 321                         page = grab_meta_page(sbi, start_blk);
 322                         orphan_blk =
 323                                 (struct f2fs_orphan_block *)page_address(page);
 324                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 325                 }
 326
 327                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 328
 329                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 330                         /*
 331                          * an orphan block is full of 1020 entries,
 332                          * then we need to flush current orphan blocks
 333                          * and bring another one in memory
 334                          */
 335                         orphan_blk->blk_addr = cpu_to_le16(index);
 336                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 337                         orphan_blk->entry_count = cpu_to_le32(nentries);
 338                         set_page_dirty(page);
 339                         f2fs_put_page(page, 1);
 340                         index++;
 341                         start_blk++;
 342                         nentries = 0;
 343                         page = NULL;
 344                 }
 345         }
 346
 347         if (page) {
 348                 orphan_blk->blk_addr = cpu_to_le16(index);
 349                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 350                 orphan_blk->entry_count = cpu_to_le32(nentries);
 351                 set_page_dirty(page);
 352                 f2fs_put_page(page, 1);
 353         }
 354
 355         mutex_unlock(&sbi->orphan_inode_mutex);
 356 }
 357
 358 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 359                                 block_t cp_addr, unsigned long long *version)
 360 {
 361         struct page *cp_page_1, *cp_page_2 = NULL;
 362         unsigned long blk_size = sbi->blocksize;
 363         struct f2fs_checkpoint *cp_block;
 364         unsigned long long cur_version = 0, pre_version = 0;
 365         size_t crc_offset;
 366         __u32 crc = 0;
 367
 368         /* Read the 1st cp block in this CP pack */
 369         cp_page_1 = get_meta_page(sbi, cp_addr);
 370
 371         /* get the version number */
 372         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 373         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 374         if (crc_offset >= blk_size)
 375                 goto invalid_cp1;
 376
 377         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 378         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 379                 goto invalid_cp1;
 380
 381         pre_version = cur_cp_version(cp_block);
 382
 383         /* Read the 2nd cp block in this CP pack */
 384         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 385         cp_page_2 = get_meta_page(sbi, cp_addr);
 386
 387         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 388         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 389         if (crc_offset >= blk_size)
 390                 goto invalid_cp2;
 391
 392         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 393         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 394                 goto invalid_cp2;
 395
 396         cur_version = cur_cp_version(cp_block);
 397
 398         if (cur_version == pre_version) {
 399                 *version = cur_version;
 400                 f2fs_put_page(cp_page_2, 1);
 401                 return cp_page_1;
 402         }
 403 invalid_cp2:
 404         f2fs_put_page(cp_page_2, 1);
 405 invalid_cp1:
 406         f2fs_put_page(cp_page_1, 1);
 407         return NULL;
 408 }
 409
 410 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 411 {
 412         struct f2fs_checkpoint *cp_block;
 413         struct f2fs_super_block *fsb = sbi->raw_super;
 414         struct page *cp1, *cp2, *cur_page;
 415         unsigned long blk_size = sbi->blocksize;
 416         unsigned long long cp1_version = 0, cp2_version = 0;
 417         unsigned long long cp_start_blk_no;
 418
 419         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 420         if (!sbi->ckpt)
 421                 return -ENOMEM;
 422         /*
 423          * Finding out valid cp block involves read both
 424          * sets( cp pack1 and cp pack 2)
 425          */
 426         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 427         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 428
 429         /* The second checkpoint pack should start at the next segment */
 430         cp_start_blk_no += ((unsigned long long)1) <<
 431                                 le32_to_cpu(fsb->log_blocks_per_seg);
 432         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 433
 434         if (cp1 && cp2) {
 435                 if (ver_after(cp2_version, cp1_version))
 436                         cur_page = cp2;
 437                 else
 438                         cur_page = cp1;
 439         } else if (cp1) {
 440                 cur_page = cp1;
 441         } else if (cp2) {
 442                 cur_page = cp2;
 443         } else {
 444                 goto fail_no_cp;
 445         }
 446
 447         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 448         memcpy(sbi->ckpt, cp_block, blk_size);
 449
 450         f2fs_put_page(cp1, 1);
 451         f2fs_put_page(cp2, 1);
 452         return 0;
 453
 454 fail_no_cp:
 455         kfree(sbi->ckpt);
 456         return -EINVAL;
 457 }
 458
 459 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 460 {
 461         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 462         struct list_head *head = &sbi->dir_inode_list;
 463         struct list_head *this;
 464
 465         list_for_each(this, head) {
 466                 struct dir_inode_entry *entry;
 467                 entry = list_entry(this, struct dir_inode_entry, list);
 468                 if (entry->inode == inode)
 469                         return -EEXIST;
 470         }
 471         list_add_tail(&new->list, head);
 472         stat_inc_dirty_dir(sbi);
 473         return 0;
 474 }
 475
 476 void set_dirty_dir_page(struct inode *inode, struct page *page)
 477 {
 478         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 479         struct dir_inode_entry *new;
 480
 481         if (!S_ISDIR(inode->i_mode))
 482                 return;
 483
 484         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 485         new->inode = inode;
 486         INIT_LIST_HEAD(&new->list);
 487
 488         spin_lock(&sbi->dir_inode_lock);
 489         if (__add_dirty_inode(inode, new))
 490                 kmem_cache_free(inode_entry_slab, new);
 491
 492         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 493         inode_inc_dirty_dents(inode);
 494         SetPagePrivate(page);
 495         spin_unlock(&sbi->dir_inode_lock);
 496 }
 497
 498 void add_dirty_dir_inode(struct inode *inode)
 499 {
 500         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 501         struct dir_inode_entry *new =
 502                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 503
 504         new->inode = inode;
 505         INIT_LIST_HEAD(&new->list);
 506
 507         spin_lock(&sbi->dir_inode_lock);
 508         if (__add_dirty_inode(inode, new))
 509                 kmem_cache_free(inode_entry_slab, new);
 510         spin_unlock(&sbi->dir_inode_lock);
 511 }
 512
 513 void remove_dirty_dir_inode(struct inode *inode)
 514 {
 515         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 516
 517         struct list_head *this, *head;
 518
 519         if (!S_ISDIR(inode->i_mode))
 520                 return;
 521
 522         spin_lock(&sbi->dir_inode_lock);
 523         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 524                 spin_unlock(&sbi->dir_inode_lock);
 525                 return;
 526         }
 527
 528         head = &sbi->dir_inode_list;
 529         list_for_each(this, head) {
 530                 struct dir_inode_entry *entry;
 531                 entry = list_entry(this, struct dir_inode_entry, list);
 532                 if (entry->inode == inode) {
 533                         list_del(&entry->list);
 534                         kmem_cache_free(inode_entry_slab, entry);
 535                         stat_dec_dirty_dir(sbi);
 536                         break;
 537                 }
 538         }
 539         spin_unlock(&sbi->dir_inode_lock);
 540
 541         /* Only from the recovery routine */
 542         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 543                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 544                 iput(inode);
 545         }
 546 }
 547
 548 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 549 {
 550
 551         struct list_head *this, *head;
 552         struct inode *inode = NULL;
 553
 554         spin_lock(&sbi->dir_inode_lock);
 555
 556         head = &sbi->dir_inode_list;
 557         list_for_each(this, head) {
 558                 struct dir_inode_entry *entry;
 559                 entry = list_entry(this, struct dir_inode_entry, list);
 560                 if (entry->inode->i_ino == ino) {
 561                         inode = entry->inode;
 562                         break;
 563                 }
 564         }
 565         spin_unlock(&sbi->dir_inode_lock);
 566         return inode;
 567 }
 568
 569 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 570 {
 571         struct list_head *head;
 572         struct dir_inode_entry *entry;
 573         struct inode *inode;
 574 retry:
 575         spin_lock(&sbi->dir_inode_lock);
 576
 577         head = &sbi->dir_inode_list;
 578         if (list_empty(head)) {
 579                 spin_unlock(&sbi->dir_inode_lock);
 580                 return;
 581         }
 582         entry = list_entry(head->next, struct dir_inode_entry, list);
 583         inode = igrab(entry->inode);
 584         spin_unlock(&sbi->dir_inode_lock);
 585         if (inode) {
 586                 filemap_flush(inode->i_mapping);
 587                 iput(inode);
 588         } else {
 589                 /*
 590                  * We should submit bio, since it exists several
 591                  * wribacking dentry pages in the freeing inode.
 592                  */
 593                 f2fs_submit_bio(sbi, DATA, true);
 594         }
 595         goto retry;
 596 }
 597
 598 /*
 599  * Freeze all the FS-operations for checkpoint.
 600  */
 601 static void block_operations(struct f2fs_sb_info *sbi)
 602 {
 603         struct writeback_control wbc = {
 604                 .sync_mode = WB_SYNC_ALL,
 605                 .nr_to_write = LONG_MAX,
 606                 .for_reclaim = 0,
 607         };
 608         struct blk_plug plug;
 609
 610         blk_start_plug(&plug);
 611
 612 retry_flush_dents:
 613         f2fs_lock_all(sbi);
 614         /* write all the dirty dentry pages */
 615         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 616                 f2fs_unlock_all(sbi);
 617                 sync_dirty_dir_inodes(sbi);
 618                 goto retry_flush_dents;
 619         }
 620
 621         /*
 622          * POR: we should ensure that there is no dirty node pages
 623          * until finishing nat/sit flush.
 624          */
 625 retry_flush_nodes:
 626         mutex_lock(&sbi->node_write);
 627
 628         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 629                 mutex_unlock(&sbi->node_write);
 630                 sync_node_pages(sbi, 0, &wbc);
 631                 goto retry_flush_nodes;
 632         }
 633         blk_finish_plug(&plug);
 634 }
 635
 636 static void unblock_operations(struct f2fs_sb_info *sbi)
 637 {
 638         mutex_unlock(&sbi->node_write);
 639         f2fs_unlock_all(sbi);
 640 }
 641
 642 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 643 {
 644         DEFINE_WAIT(wait);
 645
 646         for (;;) {
 647                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 648
 649                 if (!get_pages(sbi, F2FS_WRITEBACK))
 650                         break;
 651
 652                 io_schedule();
 653         }
 654         finish_wait(&sbi->cp_wait, &wait);
 655 }
 656
 657 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 658 {
 659         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 660         nid_t last_nid = 0;
 661         block_t start_blk;
 662         struct page *cp_page;
 663         unsigned int data_sum_blocks, orphan_blocks;
 664         __u32 crc32 = 0;
 665         void *kaddr;
 666         int i;
 667
 668         /* Flush all the NAT/SIT pages */
 669         while (get_pages(sbi, F2FS_DIRTY_META))
 670                 sync_meta_pages(sbi, META, LONG_MAX);
 671
 672         next_free_nid(sbi, &last_nid);
 673
 674         /*
 675          * modify checkpoint
 676          * version number is already updated
 677          */
 678         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 679         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 680         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 681         for (i = 0; i < 3; i++) {
 682                 ckpt->cur_node_segno[i] =
 683                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 684                 ckpt->cur_node_blkoff[i] =
 685                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 686                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 687                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 688         }
 689         for (i = 0; i < 3; i++) {
 690                 ckpt->cur_data_segno[i] =
 691                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 692                 ckpt->cur_data_blkoff[i] =
 693                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 694                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 695                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 696         }
 697
 698         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 699         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 700         ckpt->next_free_nid = cpu_to_le32(last_nid);
 701
 702         /* 2 cp  + n data seg summary + orphan inode blocks */
 703         data_sum_blocks = npages_for_summary_flush(sbi);
 704         if (data_sum_blocks < 3)
 705                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 706         else
 707                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 708
 709         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 710                                         / F2FS_ORPHANS_PER_BLOCK;
 711         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 712
 713         if (is_umount) {
 714                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 715                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 716                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 717         } else {
 718                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 719                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 720                         data_sum_blocks + orphan_blocks);
 721         }
 722
 723         if (sbi->n_orphans)
 724                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 725         else
 726                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 727
 728         /* update SIT/NAT bitmap */
 729         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 730         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 731
 732         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 733         *((__le32 *)((unsigned char *)ckpt +
 734                                 le32_to_cpu(ckpt->checksum_offset)))
 735                                 = cpu_to_le32(crc32);
 736
 737         start_blk = __start_cp_addr(sbi);
 738
 739         /* write out checkpoint buffer at block 0 */
 740         cp_page = grab_meta_page(sbi, start_blk++);
 741         kaddr = page_address(cp_page);
 742         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 743         set_page_dirty(cp_page);
 744         f2fs_put_page(cp_page, 1);
 745
 746         if (sbi->n_orphans) {
 747                 write_orphan_inodes(sbi, start_blk);
 748                 start_blk += orphan_blocks;
 749         }
 750
 751         write_data_summaries(sbi, start_blk);
 752         start_blk += data_sum_blocks;
 753         if (is_umount) {
 754                 write_node_summaries(sbi, start_blk);
 755                 start_blk += NR_CURSEG_NODE_TYPE;
 756         }
 757
 758         /* writeout checkpoint block */
 759         cp_page = grab_meta_page(sbi, start_blk);
 760         kaddr = page_address(cp_page);
 761         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 762         set_page_dirty(cp_page);
 763         f2fs_put_page(cp_page, 1);
 764
 765         /* wait for previous submitted node/meta pages writeback */
 766         wait_on_all_pages_writeback(sbi);
 767
 768         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 769         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 770
 771         /* update user_block_counts */
 772         sbi->last_valid_block_count = sbi->total_valid_block_count;
 773         sbi->alloc_valid_block_count = 0;
 774
 775         /* Here, we only have one bio having CP pack */
 776         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 777
 778         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 779                 clear_prefree_segments(sbi);
 780                 F2FS_RESET_SB_DIRT(sbi);
 781         }
 782 }
 783
 784 /*
 785  * We guarantee that this checkpoint procedure should not fail.
 786  */
 787 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 788 {
 789         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 790         unsigned long long ckpt_ver;
 791
 792         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 793
 794         mutex_lock(&sbi->cp_mutex);
 795         block_operations(sbi);
 796
 797         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 798
 799         f2fs_submit_bio(sbi, DATA, true);
 800         f2fs_submit_bio(sbi, NODE, true);
 801         f2fs_submit_bio(sbi, META, true);
 802
 803         /*
 804          * update checkpoint pack index
 805          * Increase the version number so that
 806          * SIT entries and seg summaries are written at correct place
 807          */
 808         ckpt_ver = cur_cp_version(ckpt);
 809         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 810
 811         /* write cached NAT/SIT entries to NAT/SIT area */
 812         flush_nat_entries(sbi);
 813         flush_sit_entries(sbi);
 814
 815         /* unlock all the fs_lock[] in do_checkpoint() */
 816         do_checkpoint(sbi, is_umount);
 817
 818         unblock_operations(sbi);
 819         mutex_unlock(&sbi->cp_mutex);
 820
 821         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 822 }
 823
 824 void init_orphan_info(struct f2fs_sb_info *sbi)
 825 {
 826         mutex_init(&sbi->orphan_inode_mutex);
 827         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 828         sbi->n_orphans = 0;
 829 }
 830
 831 int __init create_checkpoint_caches(void)
 832 {
 833         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 834                         sizeof(struct orphan_inode_entry), NULL);
 835         if (unlikely(!orphan_entry_slab))
 836                 return -ENOMEM;
 837         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 838                         sizeof(struct dir_inode_entry), NULL);
 839         if (unlikely(!inode_entry_slab)) {
 840                 kmem_cache_destroy(orphan_entry_slab);
 841                 return -ENOMEM;
 842         }
 843         return 0;
 844 }
 845
 846 void destroy_checkpoint_caches(void)
 847 {
 848         kmem_cache_destroy(orphan_entry_slab);
 849         kmem_cache_destroy(inode_entry_slab);
 850 }