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))
  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, *this, *next;
 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
 310         orphan_blocks = (unsigned short)((sbi->n_orphans +
 311                 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
 312
 313         mutex_lock(&sbi->orphan_inode_mutex);
 314         head = &sbi->orphan_inode_list;
 315
 316         /* loop for each orphan inode entry and write them in Jornal block */
 317         list_for_each_safe(this, next, head) {
 318                 struct orphan_inode_entry *orphan;
 319
 320                 orphan = list_entry(this, struct orphan_inode_entry, list);
 321
 322                 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 323                         /*
 324                          * an orphan block is full of 1020 entries,
 325                          * then we need to flush current orphan blocks
 326                          * and bring another one in memory
 327                          */
 328                         orphan_blk->blk_addr = cpu_to_le16(index);
 329                         orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 330                         orphan_blk->entry_count = cpu_to_le32(nentries);
 331                         set_page_dirty(page);
 332                         f2fs_put_page(page, 1);
 333                         index++;
 334                         start_blk++;
 335                         nentries = 0;
 336                         page = NULL;
 337                 }
 338                 if (page)
 339                         goto page_exist;
 340
 341                 page = grab_meta_page(sbi, start_blk);
 342                 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 343                 memset(orphan_blk, 0, sizeof(*orphan_blk));
 344 page_exist:
 345                 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 346         }
 347         if (!page)
 348                 goto end;
 349
 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 end:
 356         mutex_unlock(&sbi->orphan_inode_mutex);
 357 }
 358
 359 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 360                                 block_t cp_addr, unsigned long long *version)
 361 {
 362         struct page *cp_page_1, *cp_page_2 = NULL;
 363         unsigned long blk_size = sbi->blocksize;
 364         struct f2fs_checkpoint *cp_block;
 365         unsigned long long cur_version = 0, pre_version = 0;
 366         size_t crc_offset;
 367         __u32 crc = 0;
 368
 369         /* Read the 1st cp block in this CP pack */
 370         cp_page_1 = get_meta_page(sbi, cp_addr);
 371
 372         /* get the version number */
 373         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
 374         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 375         if (crc_offset >= blk_size)
 376                 goto invalid_cp1;
 377
 378         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 379         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 380                 goto invalid_cp1;
 381
 382         pre_version = cur_cp_version(cp_block);
 383
 384         /* Read the 2nd cp block in this CP pack */
 385         cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
 386         cp_page_2 = get_meta_page(sbi, cp_addr);
 387
 388         cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
 389         crc_offset = le32_to_cpu(cp_block->checksum_offset);
 390         if (crc_offset >= blk_size)
 391                 goto invalid_cp2;
 392
 393         crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
 394         if (!f2fs_crc_valid(crc, cp_block, crc_offset))
 395                 goto invalid_cp2;
 396
 397         cur_version = cur_cp_version(cp_block);
 398
 399         if (cur_version == pre_version) {
 400                 *version = cur_version;
 401                 f2fs_put_page(cp_page_2, 1);
 402                 return cp_page_1;
 403         }
 404 invalid_cp2:
 405         f2fs_put_page(cp_page_2, 1);
 406 invalid_cp1:
 407         f2fs_put_page(cp_page_1, 1);
 408         return NULL;
 409 }
 410
 411 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
 412 {
 413         struct f2fs_checkpoint *cp_block;
 414         struct f2fs_super_block *fsb = sbi->raw_super;
 415         struct page *cp1, *cp2, *cur_page;
 416         unsigned long blk_size = sbi->blocksize;
 417         unsigned long long cp1_version = 0, cp2_version = 0;
 418         unsigned long long cp_start_blk_no;
 419
 420         sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
 421         if (!sbi->ckpt)
 422                 return -ENOMEM;
 423         /*
 424          * Finding out valid cp block involves read both
 425          * sets( cp pack1 and cp pack 2)
 426          */
 427         cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 428         cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 429
 430         /* The second checkpoint pack should start at the next segment */
 431         cp_start_blk_no += 1 << 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         struct list_head *head = &sbi->dir_inode_list;
 517         struct list_head *this;
 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         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         struct list_head *head = &sbi->dir_inode_list;
 550         struct list_head *this;
 551         struct inode *inode = NULL;
 552
 553         spin_lock(&sbi->dir_inode_lock);
 554         list_for_each(this, head) {
 555                 struct dir_inode_entry *entry;
 556                 entry = list_entry(this, struct dir_inode_entry, list);
 557                 if (entry->inode->i_ino == ino) {
 558                         inode = entry->inode;
 559                         break;
 560                 }
 561         }
 562         spin_unlock(&sbi->dir_inode_lock);
 563         return inode;
 564 }
 565
 566 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 567 {
 568         struct list_head *head = &sbi->dir_inode_list;
 569         struct dir_inode_entry *entry;
 570         struct inode *inode;
 571 retry:
 572         spin_lock(&sbi->dir_inode_lock);
 573         if (list_empty(head)) {
 574                 spin_unlock(&sbi->dir_inode_lock);
 575                 return;
 576         }
 577         entry = list_entry(head->next, struct dir_inode_entry, list);
 578         inode = igrab(entry->inode);
 579         spin_unlock(&sbi->dir_inode_lock);
 580         if (inode) {
 581                 filemap_flush(inode->i_mapping);
 582                 iput(inode);
 583         } else {
 584                 /*
 585                  * We should submit bio, since it exists several
 586                  * wribacking dentry pages in the freeing inode.
 587                  */
 588                 f2fs_submit_bio(sbi, DATA, true);
 589         }
 590         goto retry;
 591 }
 592
 593 /*
 594  * Freeze all the FS-operations for checkpoint.
 595  */
 596 static void block_operations(struct f2fs_sb_info *sbi)
 597 {
 598         struct writeback_control wbc = {
 599                 .sync_mode = WB_SYNC_ALL,
 600                 .nr_to_write = LONG_MAX,
 601                 .for_reclaim = 0,
 602         };
 603         struct blk_plug plug;
 604
 605         blk_start_plug(&plug);
 606
 607 retry_flush_dents:
 608         f2fs_lock_all(sbi);
 609         /* write all the dirty dentry pages */
 610         if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 611                 f2fs_unlock_all(sbi);
 612                 sync_dirty_dir_inodes(sbi);
 613                 goto retry_flush_dents;
 614         }
 615
 616         /*
 617          * POR: we should ensure that there is no dirty node pages
 618          * until finishing nat/sit flush.
 619          */
 620 retry_flush_nodes:
 621         mutex_lock(&sbi->node_write);
 622
 623         if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 624                 mutex_unlock(&sbi->node_write);
 625                 sync_node_pages(sbi, 0, &wbc);
 626                 goto retry_flush_nodes;
 627         }
 628         blk_finish_plug(&plug);
 629 }
 630
 631 static void unblock_operations(struct f2fs_sb_info *sbi)
 632 {
 633         mutex_unlock(&sbi->node_write);
 634         f2fs_unlock_all(sbi);
 635 }
 636
 637 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
 638 {
 639         DEFINE_WAIT(wait);
 640
 641         for (;;) {
 642                 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 643
 644                 if (!get_pages(sbi, F2FS_WRITEBACK))
 645                         break;
 646
 647                 io_schedule();
 648         }
 649         finish_wait(&sbi->cp_wait, &wait);
 650 }
 651
 652 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 653 {
 654         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 655         nid_t last_nid = 0;
 656         block_t start_blk;
 657         struct page *cp_page;
 658         unsigned int data_sum_blocks, orphan_blocks;
 659         __u32 crc32 = 0;
 660         void *kaddr;
 661         int i;
 662
 663         /* Flush all the NAT/SIT pages */
 664         while (get_pages(sbi, F2FS_DIRTY_META))
 665                 sync_meta_pages(sbi, META, LONG_MAX);
 666
 667         next_free_nid(sbi, &last_nid);
 668
 669         /*
 670          * modify checkpoint
 671          * version number is already updated
 672          */
 673         ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
 674         ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
 675         ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
 676         for (i = 0; i < 3; i++) {
 677                 ckpt->cur_node_segno[i] =
 678                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
 679                 ckpt->cur_node_blkoff[i] =
 680                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
 681                 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
 682                                 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
 683         }
 684         for (i = 0; i < 3; i++) {
 685                 ckpt->cur_data_segno[i] =
 686                         cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
 687                 ckpt->cur_data_blkoff[i] =
 688                         cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
 689                 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
 690                                 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
 691         }
 692
 693         ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
 694         ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
 695         ckpt->next_free_nid = cpu_to_le32(last_nid);
 696
 697         /* 2 cp  + n data seg summary + orphan inode blocks */
 698         data_sum_blocks = npages_for_summary_flush(sbi);
 699         if (data_sum_blocks < 3)
 700                 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 701         else
 702                 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
 703
 704         orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
 705                                         / F2FS_ORPHANS_PER_BLOCK;
 706         ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
 707
 708         if (is_umount) {
 709                 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 710                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 711                         data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
 712         } else {
 713                 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
 714                 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
 715                         data_sum_blocks + orphan_blocks);
 716         }
 717
 718         if (sbi->n_orphans)
 719                 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 720         else
 721                 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
 722
 723         /* update SIT/NAT bitmap */
 724         get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
 725         get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
 726
 727         crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
 728         *((__le32 *)((unsigned char *)ckpt +
 729                                 le32_to_cpu(ckpt->checksum_offset)))
 730                                 = cpu_to_le32(crc32);
 731
 732         start_blk = __start_cp_addr(sbi);
 733
 734         /* write out checkpoint buffer at block 0 */
 735         cp_page = grab_meta_page(sbi, start_blk++);
 736         kaddr = page_address(cp_page);
 737         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 738         set_page_dirty(cp_page);
 739         f2fs_put_page(cp_page, 1);
 740
 741         if (sbi->n_orphans) {
 742                 write_orphan_inodes(sbi, start_blk);
 743                 start_blk += orphan_blocks;
 744         }
 745
 746         write_data_summaries(sbi, start_blk);
 747         start_blk += data_sum_blocks;
 748         if (is_umount) {
 749                 write_node_summaries(sbi, start_blk);
 750                 start_blk += NR_CURSEG_NODE_TYPE;
 751         }
 752
 753         /* writeout checkpoint block */
 754         cp_page = grab_meta_page(sbi, start_blk);
 755         kaddr = page_address(cp_page);
 756         memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
 757         set_page_dirty(cp_page);
 758         f2fs_put_page(cp_page, 1);
 759
 760         /* wait for previous submitted node/meta pages writeback */
 761         wait_on_all_pages_writeback(sbi);
 762
 763         filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
 764         filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
 765
 766         /* update user_block_counts */
 767         sbi->last_valid_block_count = sbi->total_valid_block_count;
 768         sbi->alloc_valid_block_count = 0;
 769
 770         /* Here, we only have one bio having CP pack */
 771         sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 772
 773         if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
 774                 clear_prefree_segments(sbi);
 775                 F2FS_RESET_SB_DIRT(sbi);
 776         }
 777 }
 778
 779 /*
 780  * We guarantee that this checkpoint procedure should not fail.
 781  */
 782 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 783 {
 784         struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 785         unsigned long long ckpt_ver;
 786
 787         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
 788
 789         mutex_lock(&sbi->cp_mutex);
 790         block_operations(sbi);
 791
 792         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 793
 794         f2fs_submit_bio(sbi, DATA, true);
 795         f2fs_submit_bio(sbi, NODE, true);
 796         f2fs_submit_bio(sbi, META, true);
 797
 798         /*
 799          * update checkpoint pack index
 800          * Increase the version number so that
 801          * SIT entries and seg summaries are written at correct place
 802          */
 803         ckpt_ver = cur_cp_version(ckpt);
 804         ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
 805
 806         /* write cached NAT/SIT entries to NAT/SIT area */
 807         flush_nat_entries(sbi);
 808         flush_sit_entries(sbi);
 809
 810         /* unlock all the fs_lock[] in do_checkpoint() */
 811         do_checkpoint(sbi, is_umount);
 812
 813         unblock_operations(sbi);
 814         mutex_unlock(&sbi->cp_mutex);
 815
 816         trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
 817 }
 818
 819 void init_orphan_info(struct f2fs_sb_info *sbi)
 820 {
 821         mutex_init(&sbi->orphan_inode_mutex);
 822         INIT_LIST_HEAD(&sbi->orphan_inode_list);
 823         sbi->n_orphans = 0;
 824 }
 825
 826 int __init create_checkpoint_caches(void)
 827 {
 828         orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
 829                         sizeof(struct orphan_inode_entry), NULL);
 830         if (unlikely(!orphan_entry_slab))
 831                 return -ENOMEM;
 832         inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
 833                         sizeof(struct dir_inode_entry), NULL);
 834         if (unlikely(!inode_entry_slab)) {
 835                 kmem_cache_destroy(orphan_entry_slab);
 836                 return -ENOMEM;
 837         }
 838         return 0;
 839 }
 840
 841 void destroy_checkpoint_caches(void)
 842 {
 843         kmem_cache_destroy(orphan_entry_slab);
 844         kmem_cache_destroy(inode_entry_slab);
 845 }