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 5 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*507 orphan entries
 197          */
 198         max_orphans = (sbi->blocks_per_seg - 5) * F2FS_ORPHANS_PER_BLOCK;
 199         mutex_lock(&sbi->orphan_inode_mutex);
 200         if (sbi->n_orphans >= max_orphans)
 201                 err = -ENOSPC;
 202         else
 203                 sbi->n_orphans++;
 204         mutex_unlock(&sbi->orphan_inode_mutex);
 205         return err;
 206 }
 207
 208 void release_orphan_inode(struct f2fs_sb_info *sbi)
 209 {
 210         mutex_lock(&sbi->orphan_inode_mutex);
 211         f2fs_bug_on(sbi->n_orphans == 0);
 212         sbi->n_orphans--;
 213         mutex_unlock(&sbi->orphan_inode_mutex);
 214 }
 215
 216 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 217 {
 218         struct list_head *head, *this;
 219         struct orphan_inode_entry *new = NULL, *orphan = NULL;
 220
 221         mutex_lock(&sbi->orphan_inode_mutex);
 222         head = &sbi->orphan_inode_list;
 223         list_for_each(this, head) {
 224                 orphan = list_entry(this, struct orphan_inode_entry, list);
 225                 if (orphan->ino == ino)
 226                         goto out;
 227                 if (orphan->ino > ino)
 228                         break;
 229                 orphan = NULL;
 230         }
 231
 232         new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
 233         new->ino = ino;
 234
 235         /* add new_oentry into list which is sorted by inode number */
 236         if (orphan)
 237                 list_add(&new->list, this->prev);
 238         else
 239                 list_add_tail(&new->list, head);
 240 out:
 241         mutex_unlock(&sbi->orphan_inode_mutex);
 242 }
 243
 244 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 245 {
 246         struct list_head *head;
 247         struct orphan_inode_entry *orphan;
 248
 249         mutex_lock(&sbi->orphan_inode_mutex);
 250         head = &sbi->orphan_inode_list;
 251         list_for_each_entry(orphan, head, list) {
 252                 if (orphan->ino == ino) {
 253                         list_del(&orphan->list);
 254                         kmem_cache_free(orphan_entry_slab, orphan);
 255                         f2fs_bug_on(sbi->n_orphans == 0);
 256                         sbi->n_orphans--;
 257                         break;
 258                 }
 259         }
 260         mutex_unlock(&sbi->orphan_inode_mutex);
 261 }
 262
 263 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 264 {
 265         struct inode *inode = f2fs_iget(sbi->sb, ino);
 266         f2fs_bug_on(IS_ERR(inode));
 267         clear_nlink(inode);
 268
 269         /* truncate all the data during iput */
 270         iput(inode);
 271 }
 272
 273 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
 274 {
 275         block_t start_blk, orphan_blkaddr, i, j;
 276
 277         if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
 278                 return 0;
 279
 280         sbi->por_doing = true;
 281         start_blk = __start_cp_addr(sbi) + 1;
 282         orphan_blkaddr = __start_sum_addr(sbi) - 1;
 283
 284         for (i = 0; i < orphan_blkaddr; i++) {
 285                 struct page *page = get_meta_page(sbi, start_blk + i);
 286                 struct f2fs_orphan_block *orphan_blk;
 287
 288                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 289                 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 290                         nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 291                         recover_orphan_inode(sbi, ino);
 292                 }
 293                 f2fs_put_page(page, 1);
 294         }
 295         /* clear Orphan Flag */
 296         clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
 297         sbi->por_doing = false;
 298         return 0;
 299 }
 300
 301 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 302 {
 303         struct list_head *head;
 304         struct f2fs_orphan_block *orphan_blk = NULL;
 305         struct page *page = NULL;
 306         unsigned int nentries = 0;
 307         unsigned short index = 1;
 308         unsigned short orphan_blocks;
 309         struct orphan_inode_entry *orphan = NULL;
 310
 311         orphan_blocks = (unsigned short)((sbi->n_orphans +
 312                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 313
 314         mutex_lock(&sbi->orphan_inode_mutex);
 315         head = &sbi->orphan_inode_list;
 316
 317         /* loop for each orphan inode entry and write them in Jornal block */
 318         list_for_each_entry(orphan, head, list) {
 319                 if (!page) {
 320                         page = grab_meta_page(sbi, start_blk);
 321                         orphan_blk =
 322                                 (struct f2fs_orphan_block *)page_address(page);
 323                         memset(orphan_blk, 0, sizeof(*orphan_blk));
 324                 }
 325
 326                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 327
 328                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 329                         /*
 330                          * an orphan block is full of 1020 entries,
 331                          * then we need to flush current orphan blocks
 332                          * and bring another one in memory
 333                          */
 334                         orphan_blk->blk_addr = cpu_to_le16(index);
 335                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 336                         orphan_blk->entry_count = cpu_to_le32(nentries);
 337                         set_page_dirty(page);
 338                         f2fs_put_page(page, 1);
 339                         index++;
 340                         start_blk++;
 341                         nentries = 0;
 342                         page = NULL;
 343                 }
 344         }
 345
 346         if (page) {
 347                 orphan_blk->blk_addr = cpu_to_le16(index);
 348                 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 349                 orphan_blk->entry_count = cpu_to_le32(nentries);
 350                 set_page_dirty(page);
 351                 f2fs_put_page(page, 1);
 352         }
 353
 354         mutex_unlock(&sbi->orphan_inode_mutex);
 355 }
 356
 357 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 358                                 block_t cp_addr, unsigned long long *version)
 359 {
 360         struct page *cp_page_1, *cp_page_2 = NULL;
 361         unsigned long blk_size = sbi->blocksize;
 362         struct f2fs_checkpoint *cp_block;
 363         unsigned long long cur_version = 0, pre_version = 0;
 364         size_t crc_offset;
 365         __u32 crc = 0;
 366
 367         /* Read the 1st cp block in this CP pack */
 368         cp_page_1 = get_meta_page(sbi, cp_addr);
 369
 370         /* get the version number */
 371         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 372         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 373         if (crc_offset >= blk_size)
 374                 goto invalid_cp1;
 375
 376         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 377         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 378                 goto invalid_cp1;
 379
 380         pre_version = cur_cp_version(cp_block);
 381
 382         /* Read the 2nd cp block in this CP pack */
 383         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 384         cp_page_2 = get_meta_page(sbi, cp_addr);
 385
 386         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 387         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 388         if (crc_offset >= blk_size)
 389                 goto invalid_cp2;
 390
 391         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 392         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 393                 goto invalid_cp2;
 394
 395         cur_version = cur_cp_version(cp_block);
 396
 397         if (cur_version == pre_version) {
 398                 *version = cur_version;
 399                 f2fs_put_page(cp_page_2, 1);
 400                 return cp_page_1;
 401         }
 402 invalid_cp2:
 403         f2fs_put_page(cp_page_2, 1);
 404 invalid_cp1:
 405         f2fs_put_page(cp_page_1, 1);
 406         return NULL;
 407 }
 408
 409 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 410 {
 411         struct f2fs_checkpoint *cp_block;
 412         struct f2fs_super_block *fsb = sbi->raw_super;
 413         struct page *cp1, *cp2, *cur_page;
 414         unsigned long blk_size = sbi->blocksize;
 415         unsigned long long cp1_version = 0, cp2_version = 0;
 416         unsigned long long cp_start_blk_no;
 417
 418         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 419         if (!sbi->ckpt)
 420                 return -ENOMEM;
 421         /*
 422          * Finding out valid cp block involves read both
 423          * sets( cp pack1 and cp pack 2)
 424          */
 425         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 426         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 427
 428         /* The second checkpoint pack should start at the next segment */
 429         cp_start_blk_no += ((unsigned long long)1) <<
 430                                 le32_to_cpu(fsb->log_blocks_per_seg);
 431         cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 432
 433         if (cp1 && cp2) {
 434                 if (ver_after(cp2_version, cp1_version))
 435                         cur_page = cp2;
 436                 else
 437                         cur_page = cp1;
 438         } else if (cp1) {
 439                 cur_page = cp1;
 440         } else if (cp2) {
 441                 cur_page = cp2;
 442         } else {
 443                 goto fail_no_cp;
 444         }
 445
 446         cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 447         memcpy(sbi->ckpt, cp_block, blk_size);
 448
 449         f2fs_put_page(cp1, 1);
 450         f2fs_put_page(cp2, 1);
 451         return 0;
 452
 453 fail_no_cp:
 454         kfree(sbi->ckpt);
 455         return -EINVAL;
 456 }
 457
 458 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
 459 {
 460         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 461         struct list_head *head = &sbi->dir_inode_list;
 462         struct list_head *this;
 463
 464         list_for_each(this, head) {
 465                 struct dir_inode_entry *entry;
 466                 entry = list_entry(this, struct dir_inode_entry, list);
 467                 if (entry->inode == inode)
 468                         return -EEXIST;
 469         }
 470         list_add_tail(&new->list, head);
 471         stat_inc_dirty_dir(sbi);
 472         return 0;
 473 }
 474
 475 void set_dirty_dir_page(struct inode *inode, struct page *page)
 476 {
 477         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 478         struct dir_inode_entry *new;
 479
 480         if (!S_ISDIR(inode->i_mode))
 481                 return;
 482
 483         new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 484         new->inode = inode;
 485         INIT_LIST_HEAD(&new->list);
 486
 487         spin_lock(&sbi->dir_inode_lock);
 488         if (__add_dirty_inode(inode, new))
 489                 kmem_cache_free(inode_entry_slab, new);
 490
 491         inc_page_count(sbi, F2FS_DIRTY_DENTS);
 492         inode_inc_dirty_dents(inode);
 493         SetPagePrivate(page);
 494         spin_unlock(&sbi->dir_inode_lock);
 495 }
 496
 497 void add_dirty_dir_inode(struct inode *inode)
 498 {
 499         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 500         struct dir_inode_entry *new =
 501                         f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
 502
 503         new->inode = inode;
 504         INIT_LIST_HEAD(&new->list);
 505
 506         spin_lock(&sbi->dir_inode_lock);
 507         if (__add_dirty_inode(inode, new))
 508                 kmem_cache_free(inode_entry_slab, new);
 509         spin_unlock(&sbi->dir_inode_lock);
 510 }
 511
 512 void remove_dirty_dir_inode(struct inode *inode)
 513 {
 514         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 515
 516         struct list_head *this, *head;
 517
 518         if (!S_ISDIR(inode->i_mode))
 519                 return;
 520
 521         spin_lock(&sbi->dir_inode_lock);
 522         if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
 523                 spin_unlock(&sbi->dir_inode_lock);
 524                 return;
 525         }
 526
 527         head = &sbi->dir_inode_list;
 528         list_for_each(this, head) {
 529                 struct dir_inode_entry *entry;
 530                 entry = list_entry(this, struct dir_inode_entry, list);
 531                 if (entry->inode == inode) {
 532                         list_del(&entry->list);
 533                         kmem_cache_free(inode_entry_slab, entry);
 534                         stat_dec_dirty_dir(sbi);
 535                         break;
 536                 }
 537         }
 538         spin_unlock(&sbi->dir_inode_lock);
 539
 540         /* Only from the recovery routine */
 541         if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
 542                 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
 543                 iput(inode);
 544         }
 545 }
 546
 547 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
 548 {
 549
 550         struct list_head *this, *head;
 551         struct inode *inode = NULL;
 552
 553         spin_lock(&sbi->dir_inode_lock);
 554
 555         head = &sbi->dir_inode_list;
 556         list_for_each(this, head) {
 557                 struct dir_inode_entry *entry;
 558                 entry = list_entry(this, struct dir_inode_entry, list);
 559                 if (entry->inode->i_ino == ino) {
 560                         inode = entry->inode;
 561                         break;
 562                 }
 563         }
 564         spin_unlock(&sbi->dir_inode_lock);
 565         return inode;
 566 }
 567
 568 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 569 {
 570         struct list_head *head;
 571         struct dir_inode_entry *entry;
 572         struct inode *inode;
 573 retry:
 574         spin_lock(&sbi->dir_inode_lock);
 575
 576         head = &sbi->dir_inode_list;
 577         if (list_empty(head)) {
 578                 spin_unlock(&sbi->dir_inode_lock);
 579                 return;
 580         }
 581         entry = list_entry(head->next, struct dir_inode_entry, list);
 582         inode = igrab(entry->inode);
 583         spin_unlock(&sbi->dir_inode_lock);
 584         if (inode) {
 585                 filemap_flush(inode->i_mapping);
 586                 iput(inode);
 587         } else {
 588                 /*
 589                  * We should submit bio, since it exists several
 590                  * wribacking dentry pages in the freeing inode.
 591                  */
 592                 f2fs_submit_bio(sbi, DATA, true);
 593         }
 594         goto retry;
 595 }
 596
 597 /*
 598  * Freeze all the FS-operations for checkpoint.
 599  */
 600 static void block_operations(struct f2fs_sb_info *sbi)
 601 {
 602         struct writeback_control wbc = {
 603                 .sync_mode = WB_SYNC_ALL,
 604                 .nr_to_write = LONG_MAX,
 605                 .for_reclaim = 0,
 606         };
 607         struct blk_plug plug;
 608
 609         blk_start_plug(&plug);
 610
 611 retry_flush_dents:
 612         f2fs_lock_all(sbi);
 613         /* write all the dirty dentry pages */
 614         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 615                 f2fs_unlock_all(sbi);
 616                 sync_dirty_dir_inodes(sbi);
 617                 goto retry_flush_dents;
 618         }
 619
 620         /*
 621          * POR: we should ensure that there is no dirty node pages
 622          * until finishing nat/sit flush.
 623          */
 624 retry_flush_nodes:
 625         mutex_lock(&sbi->node_write);
 626
 627         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 628                 mutex_unlock(&sbi->node_write);
 629                 sync_node_pages(sbi, 0, &wbc);
 630                 goto retry_flush_nodes;
 631         }
 632         blk_finish_plug(&plug);
 633 }
 634
 635 static void unblock_operations(struct f2fs_sb_info *sbi)
 636 {
 637         mutex_unlock(&sbi->node_write);
 638         f2fs_unlock_all(sbi);
 639 }
 640
 641 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 642 {
 643         DEFINE_WAIT(wait);
 644
 645         for (;;) {
 646                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 647
 648                 if (!get_pages(sbi, F2FS_WRITEBACK))
 649                         break;
 650
 651                 io_schedule();
 652         }
 653         finish_wait(&sbi->cp_wait, &wait);
 654 }
 655
 656 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 657 {
 658         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 659         nid_t last_nid = 0;
 660         block_t start_blk;
 661         struct page *cp_page;
 662         unsigned int data_sum_blocks, orphan_blocks;
 663         __u32 crc32 = 0;
 664         void *kaddr;
 665         int i;
 666
 667         /* Flush all the NAT/SIT pages */
 668         while (get_pages(sbi, F2FS_DIRTY_META))
 669                 sync_meta_pages(sbi, META, LONG_MAX);
 670
 671         next_free_nid(sbi, &last_nid);
 672
 673         /*
 674          * modify checkpoint
 675          * version number is already updated
 676          */
 677         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 678         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 679         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 680         for (i = 0; i < 3; i++) {
 681                 ckpt->cur_node_segno[i] =
 682                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 683                 ckpt->cur_node_blkoff[i] =
 684                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 685                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 686                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 687         }
 688         for (i = 0; i < 3; i++) {
 689                 ckpt->cur_data_segno[i] =
 690                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 691                 ckpt->cur_data_blkoff[i] =
 692                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 693                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 694                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 695         }
 696
 697         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 698         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 699         ckpt->next_free_nid = cpu_to_le32(last_nid);
 700
 701         /* 2 cp  + n data seg summary + orphan inode blocks */
 702         data_sum_blocks = npages_for_summary_flush(sbi);
 703         if (data_sum_blocks < 3)
 704                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 705         else
 706                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 707
 708         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 709                                         / F2FS_ORPHANS_PER_BLOCK;
 710         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 711
 712         if (is_umount) {
 713                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 714                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 715                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 716         } else {
 717                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 718                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 719                         data_sum_blocks + orphan_blocks);
 720         }
 721
 722         if (sbi->n_orphans)
 723                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 724         else
 725                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 726
 727         /* update SIT/NAT bitmap */
 728         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 729         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 730
 731         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 732         *((__le32 *)((unsigned char *)ckpt +
 733                                 le32_to_cpu(ckpt->checksum_offset)))
 734                                 = cpu_to_le32(crc32);
 735
 736         start_blk = __start_cp_addr(sbi);
 737
 738         /* write out checkpoint buffer at block 0 */
 739         cp_page = grab_meta_page(sbi, start_blk++);
 740         kaddr = page_address(cp_page);
 741         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 742         set_page_dirty(cp_page);
 743         f2fs_put_page(cp_page, 1);
 744
 745         if (sbi->n_orphans) {
 746                 write_orphan_inodes(sbi, start_blk);
 747                 start_blk += orphan_blocks;
 748         }
 749
 750         write_data_summaries(sbi, start_blk);
 751         start_blk += data_sum_blocks;
 752         if (is_umount) {
 753                 write_node_summaries(sbi, start_blk);
 754                 start_blk += NR_CURSEG_NODE_TYPE;
 755         }
 756
 757         /* writeout checkpoint block */
 758         cp_page = grab_meta_page(sbi, start_blk);
 759         kaddr = page_address(cp_page);
 760         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 761         set_page_dirty(cp_page);
 762         f2fs_put_page(cp_page, 1);
 763
 764         /* wait for previous submitted node/meta pages writeback */
 765         wait_on_all_pages_writeback(sbi);
 766
 767         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 768         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 769
 770         /* update user_block_counts */
 771         sbi->last_valid_block_count = sbi->total_valid_block_count;
 772         sbi->alloc_valid_block_count = 0;
 773
 774         /* Here, we only have one bio having CP pack */
 775         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 776
 777         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 778                 clear_prefree_segments(sbi);
 779                 F2FS_RESET_SB_DIRT(sbi);
 780         }
 781 }
 782
 783 /*
 784  * We guarantee that this checkpoint procedure should not fail.
 785  */
 786 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 787 {
 788         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 789         unsigned long long ckpt_ver;
 790
 791         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 792
 793         mutex_lock(&sbi->cp_mutex);
 794         block_operations(sbi);
 795
 796         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 797
 798         f2fs_submit_bio(sbi, DATA, true);
 799         f2fs_submit_bio(sbi, NODE, true);
 800         f2fs_submit_bio(sbi, META, true);
 801
 802         /*
 803          * update checkpoint pack index
 804          * Increase the version number so that
 805          * SIT entries and seg summaries are written at correct place
 806          */
 807         ckpt_ver = cur_cp_version(ckpt);
 808         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 809
 810         /* write cached NAT/SIT entries to NAT/SIT area */
 811         flush_nat_entries(sbi);
 812         flush_sit_entries(sbi);
 813
 814         /* unlock all the fs_lock[] in do_checkpoint() */
 815         do_checkpoint(sbi, is_umount);
 816
 817         unblock_operations(sbi);
 818         mutex_unlock(&sbi->cp_mutex);
 819
 820         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 821 }
 822
 823 void init_orphan_info(struct f2fs_sb_info *sbi)
 824 {
 825         mutex_init(&sbi->orphan_inode_mutex);
 826         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 827         sbi->n_orphans = 0;
 828 }
 829
 830 int __init create_checkpoint_caches(void)
 831 {
 832         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 833                         sizeof(struct orphan_inode_entry), NULL);
 834         if (unlikely(!orphan_entry_slab))
 835                 return -ENOMEM;
 836         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 837                         sizeof(struct dir_inode_entry), NULL);
 838         if (unlikely(!inode_entry_slab)) {
 839                 kmem_cache_destroy(orphan_entry_slab);
 840                 return -ENOMEM;
 841         }
 842         return 0;
 843 }
 844
 845 void destroy_checkpoint_caches(void)
 846 {
 847         kmem_cache_destroy(orphan_entry_slab);
 848         kmem_cache_destroy(inode_entry_slab);
 849 }