1 /* 2 * fs/f2fs/checkpoint.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/bio.h> 13 #include <linux/mpage.h> 14 #include <linux/writeback.h> 15 #include <linux/blkdev.h> 16 #include <linux/f2fs_fs.h> 17 #include <linux/pagevec.h> 18 #include <linux/swap.h> 19 20 #include "f2fs.h" 21 #include "node.h" 22 #include "segment.h" 23 #include <trace/events/f2fs.h> 24 25 static struct kmem_cache *orphan_entry_slab; 26 static struct kmem_cache *inode_entry_slab; 27 28 /* 29 * We guarantee no failure on the returned page. 30 */ 31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) 32 { 33 struct address_space *mapping = sbi->meta_inode->i_mapping; 34 struct page *page = NULL; 35 repeat: 36 page = grab_cache_page(mapping, index); 37 if (!page) { 38 cond_resched(); 39 goto repeat; 40 } 41 42 /* We wait writeback only inside grab_meta_page() */ 43 wait_on_page_writeback(page); 44 SetPageUptodate(page); 45 return page; 46 } 47 48 /* 49 * We guarantee no failure on the returned page. 50 */ 51 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) 52 { 53 struct address_space *mapping = sbi->meta_inode->i_mapping; 54 struct page *page; 55 repeat: 56 page = grab_cache_page(mapping, index); 57 if (!page) { 58 cond_resched(); 59 goto repeat; 60 } 61 if (PageUptodate(page)) 62 goto out; 63 64 if (f2fs_submit_page_bio(sbi, page, index, 65 READ_SYNC | REQ_META | REQ_PRIO)) 66 goto repeat; 67 68 lock_page(page); 69 if (page->mapping != mapping) { 70 f2fs_put_page(page, 1); 71 goto repeat; 72 } 73 out: 74 mark_page_accessed(page); 75 return page; 76 } 77 78 static int f2fs_write_meta_page(struct page *page, 79 struct writeback_control *wbc) 80 { 81 struct inode *inode = page->mapping->host; 82 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 83 84 /* Should not write any meta pages, if any IO error was occurred */ 85 if (unlikely(sbi->por_doing || 86 is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG))) 87 goto redirty_out; 88 89 if (wbc->for_reclaim) 90 goto redirty_out; 91 92 wait_on_page_writeback(page); 93 94 write_meta_page(sbi, page); 95 dec_page_count(sbi, F2FS_DIRTY_META); 96 unlock_page(page); 97 return 0; 98 99 redirty_out: 100 dec_page_count(sbi, F2FS_DIRTY_META); 101 wbc->pages_skipped++; 102 set_page_dirty(page); 103 return AOP_WRITEPAGE_ACTIVATE; 104 } 105 106 static int f2fs_write_meta_pages(struct address_space *mapping, 107 struct writeback_control *wbc) 108 { 109 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 110 struct block_device *bdev = sbi->sb->s_bdev; 111 long written; 112 113 if (wbc->for_kupdate) 114 return 0; 115 116 if (get_pages(sbi, F2FS_DIRTY_META) == 0) 117 return 0; 118 119 /* if mounting is failed, skip writing node pages */ 120 mutex_lock(&sbi->cp_mutex); 121 written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev)); 122 mutex_unlock(&sbi->cp_mutex); 123 wbc->nr_to_write -= written; 124 return 0; 125 } 126 127 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, 128 long nr_to_write) 129 { 130 struct address_space *mapping = sbi->meta_inode->i_mapping; 131 pgoff_t index = 0, end = LONG_MAX; 132 struct pagevec pvec; 133 long nwritten = 0; 134 struct writeback_control wbc = { 135 .for_reclaim = 0, 136 }; 137 138 pagevec_init(&pvec, 0); 139 140 while (index <= end) { 141 int i, nr_pages; 142 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 143 PAGECACHE_TAG_DIRTY, 144 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 145 if (unlikely(nr_pages == 0)) 146 break; 147 148 for (i = 0; i < nr_pages; i++) { 149 struct page *page = pvec.pages[i]; 150 lock_page(page); 151 f2fs_bug_on(page->mapping != mapping); 152 f2fs_bug_on(!PageDirty(page)); 153 clear_page_dirty_for_io(page); 154 if (f2fs_write_meta_page(page, &wbc)) { 155 unlock_page(page); 156 break; 157 } 158 nwritten++; 159 if (unlikely(nwritten >= nr_to_write)) 160 break; 161 } 162 pagevec_release(&pvec); 163 cond_resched(); 164 } 165 166 if (nwritten) 167 f2fs_submit_merged_bio(sbi, type, nr_to_write == LONG_MAX, 168 WRITE); 169 170 return nwritten; 171 } 172 173 static int f2fs_set_meta_page_dirty(struct page *page) 174 { 175 struct address_space *mapping = page->mapping; 176 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); 177 178 trace_f2fs_set_page_dirty(page, META); 179 180 SetPageUptodate(page); 181 if (!PageDirty(page)) { 182 __set_page_dirty_nobuffers(page); 183 inc_page_count(sbi, F2FS_DIRTY_META); 184 return 1; 185 } 186 return 0; 187 } 188 189 const struct address_space_operations f2fs_meta_aops = { 190 .writepage = f2fs_write_meta_page, 191 .writepages = f2fs_write_meta_pages, 192 .set_page_dirty = f2fs_set_meta_page_dirty, 193 }; 194 195 int acquire_orphan_inode(struct f2fs_sb_info *sbi) 196 { 197 unsigned int max_orphans; 198 int err = 0; 199 200 /* 201 * considering 512 blocks in a segment 8 blocks are needed for cp 202 * and log segment summaries. Remaining blocks are used to keep 203 * orphan entries with the limitation one reserved segment 204 * for cp pack we can have max 1020*504 orphan entries 205 */ 206 max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE) 207 * F2FS_ORPHANS_PER_BLOCK; 208 mutex_lock(&sbi->orphan_inode_mutex); 209 if (unlikely(sbi->n_orphans >= max_orphans)) 210 err = -ENOSPC; 211 else 212 sbi->n_orphans++; 213 mutex_unlock(&sbi->orphan_inode_mutex); 214 return err; 215 } 216 217 void release_orphan_inode(struct f2fs_sb_info *sbi) 218 { 219 mutex_lock(&sbi->orphan_inode_mutex); 220 f2fs_bug_on(sbi->n_orphans == 0); 221 sbi->n_orphans--; 222 mutex_unlock(&sbi->orphan_inode_mutex); 223 } 224 225 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 226 { 227 struct list_head *head, *this; 228 struct orphan_inode_entry *new = NULL, *orphan = NULL; 229 230 mutex_lock(&sbi->orphan_inode_mutex); 231 head = &sbi->orphan_inode_list; 232 list_for_each(this, head) { 233 orphan = list_entry(this, struct orphan_inode_entry, list); 234 if (orphan->ino == ino) 235 goto out; 236 if (orphan->ino > ino) 237 break; 238 orphan = NULL; 239 } 240 241 new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); 242 new->ino = ino; 243 244 /* add new_oentry into list which is sorted by inode number */ 245 if (orphan) 246 list_add(&new->list, this->prev); 247 else 248 list_add_tail(&new->list, head); 249 out: 250 mutex_unlock(&sbi->orphan_inode_mutex); 251 } 252 253 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 254 { 255 struct list_head *head; 256 struct orphan_inode_entry *orphan; 257 258 mutex_lock(&sbi->orphan_inode_mutex); 259 head = &sbi->orphan_inode_list; 260 list_for_each_entry(orphan, head, list) { 261 if (orphan->ino == ino) { 262 list_del(&orphan->list); 263 kmem_cache_free(orphan_entry_slab, orphan); 264 f2fs_bug_on(sbi->n_orphans == 0); 265 sbi->n_orphans--; 266 break; 267 } 268 } 269 mutex_unlock(&sbi->orphan_inode_mutex); 270 } 271 272 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) 273 { 274 struct inode *inode = f2fs_iget(sbi->sb, ino); 275 f2fs_bug_on(IS_ERR(inode)); 276 clear_nlink(inode); 277 278 /* truncate all the data during iput */ 279 iput(inode); 280 } 281 282 void recover_orphan_inodes(struct f2fs_sb_info *sbi) 283 { 284 block_t start_blk, orphan_blkaddr, i, j; 285 286 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) 287 return; 288 289 sbi->por_doing = true; 290 start_blk = __start_cp_addr(sbi) + 1; 291 orphan_blkaddr = __start_sum_addr(sbi) - 1; 292 293 for (i = 0; i < orphan_blkaddr; i++) { 294 struct page *page = get_meta_page(sbi, start_blk + i); 295 struct f2fs_orphan_block *orphan_blk; 296 297 orphan_blk = (struct f2fs_orphan_block *)page_address(page); 298 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { 299 nid_t ino = le32_to_cpu(orphan_blk->ino[j]); 300 recover_orphan_inode(sbi, ino); 301 } 302 f2fs_put_page(page, 1); 303 } 304 /* clear Orphan Flag */ 305 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); 306 sbi->por_doing = false; 307 return; 308 } 309 310 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) 311 { 312 struct list_head *head; 313 struct f2fs_orphan_block *orphan_blk = NULL; 314 struct page *page = NULL; 315 unsigned int nentries = 0; 316 unsigned short index = 1; 317 unsigned short orphan_blocks; 318 struct orphan_inode_entry *orphan = NULL; 319 320 orphan_blocks = (unsigned short)((sbi->n_orphans + 321 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK); 322 323 mutex_lock(&sbi->orphan_inode_mutex); 324 head = &sbi->orphan_inode_list; 325 326 /* loop for each orphan inode entry and write them in Jornal block */ 327 list_for_each_entry(orphan, head, list) { 328 if (!page) { 329 page = grab_meta_page(sbi, start_blk); 330 orphan_blk = 331 (struct f2fs_orphan_block *)page_address(page); 332 memset(orphan_blk, 0, sizeof(*orphan_blk)); 333 } 334 335 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); 336 337 if (nentries == F2FS_ORPHANS_PER_BLOCK) { 338 /* 339 * an orphan block is full of 1020 entries, 340 * then we need to flush current orphan blocks 341 * and bring another one in memory 342 */ 343 orphan_blk->blk_addr = cpu_to_le16(index); 344 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 345 orphan_blk->entry_count = cpu_to_le32(nentries); 346 set_page_dirty(page); 347 f2fs_put_page(page, 1); 348 index++; 349 start_blk++; 350 nentries = 0; 351 page = NULL; 352 } 353 } 354 355 if (page) { 356 orphan_blk->blk_addr = cpu_to_le16(index); 357 orphan_blk->blk_count = cpu_to_le16(orphan_blocks); 358 orphan_blk->entry_count = cpu_to_le32(nentries); 359 set_page_dirty(page); 360 f2fs_put_page(page, 1); 361 } 362 363 mutex_unlock(&sbi->orphan_inode_mutex); 364 } 365 366 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, 367 block_t cp_addr, unsigned long long *version) 368 { 369 struct page *cp_page_1, *cp_page_2 = NULL; 370 unsigned long blk_size = sbi->blocksize; 371 struct f2fs_checkpoint *cp_block; 372 unsigned long long cur_version = 0, pre_version = 0; 373 size_t crc_offset; 374 __u32 crc = 0; 375 376 /* Read the 1st cp block in this CP pack */ 377 cp_page_1 = get_meta_page(sbi, cp_addr); 378 379 /* get the version number */ 380 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1); 381 crc_offset = le32_to_cpu(cp_block->checksum_offset); 382 if (crc_offset >= blk_size) 383 goto invalid_cp1; 384 385 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); 386 if (!f2fs_crc_valid(crc, cp_block, crc_offset)) 387 goto invalid_cp1; 388 389 pre_version = cur_cp_version(cp_block); 390 391 /* Read the 2nd cp block in this CP pack */ 392 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; 393 cp_page_2 = get_meta_page(sbi, cp_addr); 394 395 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2); 396 crc_offset = le32_to_cpu(cp_block->checksum_offset); 397 if (crc_offset >= blk_size) 398 goto invalid_cp2; 399 400 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); 401 if (!f2fs_crc_valid(crc, cp_block, crc_offset)) 402 goto invalid_cp2; 403 404 cur_version = cur_cp_version(cp_block); 405 406 if (cur_version == pre_version) { 407 *version = cur_version; 408 f2fs_put_page(cp_page_2, 1); 409 return cp_page_1; 410 } 411 invalid_cp2: 412 f2fs_put_page(cp_page_2, 1); 413 invalid_cp1: 414 f2fs_put_page(cp_page_1, 1); 415 return NULL; 416 } 417 418 int get_valid_checkpoint(struct f2fs_sb_info *sbi) 419 { 420 struct f2fs_checkpoint *cp_block; 421 struct f2fs_super_block *fsb = sbi->raw_super; 422 struct page *cp1, *cp2, *cur_page; 423 unsigned long blk_size = sbi->blocksize; 424 unsigned long long cp1_version = 0, cp2_version = 0; 425 unsigned long long cp_start_blk_no; 426 427 sbi->ckpt = kzalloc(blk_size, GFP_KERNEL); 428 if (!sbi->ckpt) 429 return -ENOMEM; 430 /* 431 * Finding out valid cp block involves read both 432 * sets( cp pack1 and cp pack 2) 433 */ 434 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); 435 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); 436 437 /* The second checkpoint pack should start at the next segment */ 438 cp_start_blk_no += ((unsigned long long)1) << 439 le32_to_cpu(fsb->log_blocks_per_seg); 440 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); 441 442 if (cp1 && cp2) { 443 if (ver_after(cp2_version, cp1_version)) 444 cur_page = cp2; 445 else 446 cur_page = cp1; 447 } else if (cp1) { 448 cur_page = cp1; 449 } else if (cp2) { 450 cur_page = cp2; 451 } else { 452 goto fail_no_cp; 453 } 454 455 cp_block = (struct f2fs_checkpoint *)page_address(cur_page); 456 memcpy(sbi->ckpt, cp_block, blk_size); 457 458 f2fs_put_page(cp1, 1); 459 f2fs_put_page(cp2, 1); 460 return 0; 461 462 fail_no_cp: 463 kfree(sbi->ckpt); 464 return -EINVAL; 465 } 466 467 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) 468 { 469 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 470 struct list_head *head = &sbi->dir_inode_list; 471 struct list_head *this; 472 473 list_for_each(this, head) { 474 struct dir_inode_entry *entry; 475 entry = list_entry(this, struct dir_inode_entry, list); 476 if (entry->inode == inode) 477 return -EEXIST; 478 } 479 list_add_tail(&new->list, head); 480 stat_inc_dirty_dir(sbi); 481 return 0; 482 } 483 484 void set_dirty_dir_page(struct inode *inode, struct page *page) 485 { 486 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 487 struct dir_inode_entry *new; 488 489 if (!S_ISDIR(inode->i_mode)) 490 return; 491 492 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); 493 new->inode = inode; 494 INIT_LIST_HEAD(&new->list); 495 496 spin_lock(&sbi->dir_inode_lock); 497 if (__add_dirty_inode(inode, new)) 498 kmem_cache_free(inode_entry_slab, new); 499 500 inc_page_count(sbi, F2FS_DIRTY_DENTS); 501 inode_inc_dirty_dents(inode); 502 SetPagePrivate(page); 503 spin_unlock(&sbi->dir_inode_lock); 504 } 505 506 void add_dirty_dir_inode(struct inode *inode) 507 { 508 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 509 struct dir_inode_entry *new = 510 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); 511 512 new->inode = inode; 513 INIT_LIST_HEAD(&new->list); 514 515 spin_lock(&sbi->dir_inode_lock); 516 if (__add_dirty_inode(inode, new)) 517 kmem_cache_free(inode_entry_slab, new); 518 spin_unlock(&sbi->dir_inode_lock); 519 } 520 521 void remove_dirty_dir_inode(struct inode *inode) 522 { 523 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 524 525 struct list_head *this, *head; 526 527 if (!S_ISDIR(inode->i_mode)) 528 return; 529 530 spin_lock(&sbi->dir_inode_lock); 531 if (atomic_read(&F2FS_I(inode)->dirty_dents)) { 532 spin_unlock(&sbi->dir_inode_lock); 533 return; 534 } 535 536 head = &sbi->dir_inode_list; 537 list_for_each(this, head) { 538 struct dir_inode_entry *entry; 539 entry = list_entry(this, struct dir_inode_entry, list); 540 if (entry->inode == inode) { 541 list_del(&entry->list); 542 kmem_cache_free(inode_entry_slab, entry); 543 stat_dec_dirty_dir(sbi); 544 break; 545 } 546 } 547 spin_unlock(&sbi->dir_inode_lock); 548 549 /* Only from the recovery routine */ 550 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { 551 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); 552 iput(inode); 553 } 554 } 555 556 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino) 557 { 558 559 struct list_head *this, *head; 560 struct inode *inode = NULL; 561 562 spin_lock(&sbi->dir_inode_lock); 563 564 head = &sbi->dir_inode_list; 565 list_for_each(this, head) { 566 struct dir_inode_entry *entry; 567 entry = list_entry(this, struct dir_inode_entry, list); 568 if (entry->inode->i_ino == ino) { 569 inode = entry->inode; 570 break; 571 } 572 } 573 spin_unlock(&sbi->dir_inode_lock); 574 return inode; 575 } 576 577 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) 578 { 579 struct list_head *head; 580 struct dir_inode_entry *entry; 581 struct inode *inode; 582 retry: 583 spin_lock(&sbi->dir_inode_lock); 584 585 head = &sbi->dir_inode_list; 586 if (list_empty(head)) { 587 spin_unlock(&sbi->dir_inode_lock); 588 return; 589 } 590 entry = list_entry(head->next, struct dir_inode_entry, list); 591 inode = igrab(entry->inode); 592 spin_unlock(&sbi->dir_inode_lock); 593 if (inode) { 594 filemap_flush(inode->i_mapping); 595 iput(inode); 596 } else { 597 /* 598 * We should submit bio, since it exists several 599 * wribacking dentry pages in the freeing inode. 600 */ 601 f2fs_submit_merged_bio(sbi, DATA, true, WRITE); 602 } 603 goto retry; 604 } 605 606 /* 607 * Freeze all the FS-operations for checkpoint. 608 */ 609 static void block_operations(struct f2fs_sb_info *sbi) 610 { 611 struct writeback_control wbc = { 612 .sync_mode = WB_SYNC_ALL, 613 .nr_to_write = LONG_MAX, 614 .for_reclaim = 0, 615 }; 616 struct blk_plug plug; 617 618 blk_start_plug(&plug); 619 620 retry_flush_dents: 621 f2fs_lock_all(sbi); 622 /* write all the dirty dentry pages */ 623 if (get_pages(sbi, F2FS_DIRTY_DENTS)) { 624 f2fs_unlock_all(sbi); 625 sync_dirty_dir_inodes(sbi); 626 goto retry_flush_dents; 627 } 628 629 /* 630 * POR: we should ensure that there is no dirty node pages 631 * until finishing nat/sit flush. 632 */ 633 retry_flush_nodes: 634 mutex_lock(&sbi->node_write); 635 636 if (get_pages(sbi, F2FS_DIRTY_NODES)) { 637 mutex_unlock(&sbi->node_write); 638 sync_node_pages(sbi, 0, &wbc); 639 goto retry_flush_nodes; 640 } 641 blk_finish_plug(&plug); 642 } 643 644 static void unblock_operations(struct f2fs_sb_info *sbi) 645 { 646 mutex_unlock(&sbi->node_write); 647 f2fs_unlock_all(sbi); 648 } 649 650 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi) 651 { 652 DEFINE_WAIT(wait); 653 654 for (;;) { 655 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); 656 657 if (!get_pages(sbi, F2FS_WRITEBACK)) 658 break; 659 660 io_schedule(); 661 } 662 finish_wait(&sbi->cp_wait, &wait); 663 } 664 665 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) 666 { 667 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 668 nid_t last_nid = 0; 669 block_t start_blk; 670 struct page *cp_page; 671 unsigned int data_sum_blocks, orphan_blocks; 672 __u32 crc32 = 0; 673 void *kaddr; 674 int i; 675 676 /* Flush all the NAT/SIT pages */ 677 while (get_pages(sbi, F2FS_DIRTY_META)) 678 sync_meta_pages(sbi, META, LONG_MAX); 679 680 next_free_nid(sbi, &last_nid); 681 682 /* 683 * modify checkpoint 684 * version number is already updated 685 */ 686 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); 687 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); 688 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); 689 for (i = 0; i < 3; i++) { 690 ckpt->cur_node_segno[i] = 691 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); 692 ckpt->cur_node_blkoff[i] = 693 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); 694 ckpt->alloc_type[i + CURSEG_HOT_NODE] = 695 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); 696 } 697 for (i = 0; i < 3; i++) { 698 ckpt->cur_data_segno[i] = 699 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); 700 ckpt->cur_data_blkoff[i] = 701 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); 702 ckpt->alloc_type[i + CURSEG_HOT_DATA] = 703 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); 704 } 705 706 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); 707 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); 708 ckpt->next_free_nid = cpu_to_le32(last_nid); 709 710 /* 2 cp + n data seg summary + orphan inode blocks */ 711 data_sum_blocks = npages_for_summary_flush(sbi); 712 if (data_sum_blocks < 3) 713 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 714 else 715 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); 716 717 orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) 718 / F2FS_ORPHANS_PER_BLOCK; 719 ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); 720 721 if (is_umount) { 722 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 723 ckpt->cp_pack_total_block_count = cpu_to_le32(2 + 724 data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); 725 } else { 726 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); 727 ckpt->cp_pack_total_block_count = cpu_to_le32(2 + 728 data_sum_blocks + orphan_blocks); 729 } 730 731 if (sbi->n_orphans) 732 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 733 else 734 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); 735 736 /* update SIT/NAT bitmap */ 737 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); 738 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); 739 740 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); 741 *((__le32 *)((unsigned char *)ckpt + 742 le32_to_cpu(ckpt->checksum_offset))) 743 = cpu_to_le32(crc32); 744 745 start_blk = __start_cp_addr(sbi); 746 747 /* write out checkpoint buffer at block 0 */ 748 cp_page = grab_meta_page(sbi, start_blk++); 749 kaddr = page_address(cp_page); 750 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); 751 set_page_dirty(cp_page); 752 f2fs_put_page(cp_page, 1); 753 754 if (sbi->n_orphans) { 755 write_orphan_inodes(sbi, start_blk); 756 start_blk += orphan_blocks; 757 } 758 759 write_data_summaries(sbi, start_blk); 760 start_blk += data_sum_blocks; 761 if (is_umount) { 762 write_node_summaries(sbi, start_blk); 763 start_blk += NR_CURSEG_NODE_TYPE; 764 } 765 766 /* writeout checkpoint block */ 767 cp_page = grab_meta_page(sbi, start_blk); 768 kaddr = page_address(cp_page); 769 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); 770 set_page_dirty(cp_page); 771 f2fs_put_page(cp_page, 1); 772 773 /* wait for previous submitted node/meta pages writeback */ 774 wait_on_all_pages_writeback(sbi); 775 776 filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX); 777 filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX); 778 779 /* update user_block_counts */ 780 sbi->last_valid_block_count = sbi->total_valid_block_count; 781 sbi->alloc_valid_block_count = 0; 782 783 /* Here, we only have one bio having CP pack */ 784 sync_meta_pages(sbi, META_FLUSH, LONG_MAX); 785 786 if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) { 787 clear_prefree_segments(sbi); 788 F2FS_RESET_SB_DIRT(sbi); 789 } 790 } 791 792 /* 793 * We guarantee that this checkpoint procedure should not fail. 794 */ 795 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) 796 { 797 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); 798 unsigned long long ckpt_ver; 799 800 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); 801 802 mutex_lock(&sbi->cp_mutex); 803 block_operations(sbi); 804 805 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); 806 807 f2fs_submit_merged_bio(sbi, DATA, true, WRITE); 808 f2fs_submit_merged_bio(sbi, NODE, true, WRITE); 809 f2fs_submit_merged_bio(sbi, META, true, WRITE); 810 811 /* 812 * update checkpoint pack index 813 * Increase the version number so that 814 * SIT entries and seg summaries are written at correct place 815 */ 816 ckpt_ver = cur_cp_version(ckpt); 817 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); 818 819 /* write cached NAT/SIT entries to NAT/SIT area */ 820 flush_nat_entries(sbi); 821 flush_sit_entries(sbi); 822 823 /* unlock all the fs_lock[] in do_checkpoint() */ 824 do_checkpoint(sbi, is_umount); 825 826 unblock_operations(sbi); 827 mutex_unlock(&sbi->cp_mutex); 828 829 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); 830 } 831 832 void init_orphan_info(struct f2fs_sb_info *sbi) 833 { 834 mutex_init(&sbi->orphan_inode_mutex); 835 INIT_LIST_HEAD(&sbi->orphan_inode_list); 836 sbi->n_orphans = 0; 837 } 838 839 int __init create_checkpoint_caches(void) 840 { 841 orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", 842 sizeof(struct orphan_inode_entry), NULL); 843 if (unlikely(!orphan_entry_slab)) 844 return -ENOMEM; 845 inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", 846 sizeof(struct dir_inode_entry), NULL); 847 if (unlikely(!inode_entry_slab)) { 848 kmem_cache_destroy(orphan_entry_slab); 849 return -ENOMEM; 850 } 851 return 0; 852 } 853 854 void destroy_checkpoint_caches(void) 855 { 856 kmem_cache_destroy(orphan_entry_slab); 857 kmem_cache_destroy(inode_entry_slab); 858 } 859