Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
logfs.h
Go to the documentation of this file.
1 /*
2  * fs/logfs/logfs.h
3  *
4  * As should be obvious for Linux kernel code, license is GPLv2
5  *
6  * Copyright (c) 2005-2008 Joern Engel <[email protected]>
7  *
8  * Private header for logfs.
9  */
10 #ifndef FS_LOGFS_LOGFS_H
11 #define FS_LOGFS_LOGFS_H
12 
13 #undef __CHECK_ENDIAN__
14 #define __CHECK_ENDIAN__
15 
16 #include <linux/btree.h>
17 #include <linux/crc32.h>
18 #include <linux/fs.h>
19 #include <linux/kernel.h>
20 #include <linux/mempool.h>
21 #include <linux/pagemap.h>
22 #include <linux/mtd/mtd.h>
23 #include "logfs_abi.h"
24 
25 #define LOGFS_DEBUG_SUPER (0x0001)
26 #define LOGFS_DEBUG_SEGMENT (0x0002)
27 #define LOGFS_DEBUG_JOURNAL (0x0004)
28 #define LOGFS_DEBUG_DIR (0x0008)
29 #define LOGFS_DEBUG_FILE (0x0010)
30 #define LOGFS_DEBUG_INODE (0x0020)
31 #define LOGFS_DEBUG_READWRITE (0x0040)
32 #define LOGFS_DEBUG_GC (0x0080)
33 #define LOGFS_DEBUG_GC_NOISY (0x0100)
34 #define LOGFS_DEBUG_ALIASES (0x0200)
35 #define LOGFS_DEBUG_BLOCKMOVE (0x0400)
36 #define LOGFS_DEBUG_ALL (0xffffffff)
37 
38 #define LOGFS_DEBUG (0x01)
39 /*
40  * To enable specific log messages, simply define LOGFS_DEBUG to match any
41  * or all of the above.
42  */
43 #ifndef LOGFS_DEBUG
44 #define LOGFS_DEBUG (0)
45 #endif
46 
47 #define log_cond(cond, fmt, arg...) do { \
48  if (cond) \
49  printk(KERN_DEBUG fmt, ##arg); \
50 } while (0)
51 
52 #define log_super(fmt, arg...) \
53  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_SUPER, fmt, ##arg)
54 #define log_segment(fmt, arg...) \
55  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_SEGMENT, fmt, ##arg)
56 #define log_journal(fmt, arg...) \
57  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_JOURNAL, fmt, ##arg)
58 #define log_dir(fmt, arg...) \
59  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_DIR, fmt, ##arg)
60 #define log_file(fmt, arg...) \
61  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_FILE, fmt, ##arg)
62 #define log_inode(fmt, arg...) \
63  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_INODE, fmt, ##arg)
64 #define log_readwrite(fmt, arg...) \
65  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_READWRITE, fmt, ##arg)
66 #define log_gc(fmt, arg...) \
67  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_GC, fmt, ##arg)
68 #define log_gc_noisy(fmt, arg...) \
69  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_GC_NOISY, fmt, ##arg)
70 #define log_aliases(fmt, arg...) \
71  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_ALIASES, fmt, ##arg)
72 #define log_blockmove(fmt, arg...) \
73  log_cond(LOGFS_DEBUG & LOGFS_DEBUG_BLOCKMOVE, fmt, ##arg)
74 
75 #define PG_pre_locked PG_owner_priv_1
76 #define PagePreLocked(page) test_bit(PG_pre_locked, &(page)->flags)
77 #define SetPagePreLocked(page) set_bit(PG_pre_locked, &(page)->flags)
78 #define ClearPagePreLocked(page) clear_bit(PG_pre_locked, &(page)->flags)
79 
80 /* FIXME: This should really be somewhere in the 64bit area. */
81 #define LOGFS_LINK_MAX (1<<30)
82 
83 /* Read-only filesystem */
84 #define LOGFS_SB_FLAG_RO 0x0001
85 #define LOGFS_SB_FLAG_DIRTY 0x0002
86 #define LOGFS_SB_FLAG_OBJ_ALIAS 0x0004
87 #define LOGFS_SB_FLAG_SHUTDOWN 0x0008
88 
89 /* Write Control Flags */
90 #define WF_LOCK 0x01 /* take write lock */
91 #define WF_WRITE 0x02 /* write block */
92 #define WF_DELETE 0x04 /* delete old block */
93 
96 
97 #define LEVEL(level) ((__force level_t)(level))
98 #define GC_LEVEL(gc_level) ((__force gc_level_t)(gc_level))
99 
100 #define SUBLEVEL(level) ( (void)((level) == LEVEL(1)), \
101  (__force level_t)((__force u8)(level) - 1) )
102 
115 struct logfs_area { /* a segment open for writing */
116  struct super_block *a_sb;
121  const struct logfs_area_ops *a_ops;
124 };
125 
134  void (*get_free_segment)(struct logfs_area *area);
135  void (*get_erase_count)(struct logfs_area *area);
136  int (*erase_segment)(struct logfs_area *area);
137 };
138 
139 struct logfs_super; /* forward */
151  struct page *(*find_first_sb)(struct super_block *sb, u64 *ofs);
152  struct page *(*find_last_sb)(struct super_block *sb, u64 *ofs);
153  int (*write_sb)(struct super_block *sb, struct page *page);
154  int (*readpage)(void *_sb, struct page *page);
155  void (*writeseg)(struct super_block *sb, u64 ofs, size_t len);
156  int (*erase)(struct super_block *sb, loff_t ofs, size_t len,
157  int ensure_write);
158  int (*can_write_buf)(struct super_block *sb, u64 ofs);
159  void (*sync)(struct super_block *sb);
161 };
162 
167  struct rb_root rb_tree;
168  int count;
169  int maxcount;
171 };
172 
188 struct gc_candidate {
189  struct rb_node rb_node;
195 };
196 
206  int used;
211 };
212 
214  CREATE_1 = 1,
223 };
224 
237 };
238 
249 struct logfs_shadow {
254  int old_len;
255  int new_len;
257 };
258 
266 struct shadow_tree {
267  struct btree_head64 new;
268  struct btree_head64 old;
269  struct btree_head32 segment_map;
271 };
272 
274  struct list_head list;
276  int child_no;
277 };
278 
291 #define BLOCK_INDIRECT 1 /* Indirect block */
292 #define BLOCK_INODE 2 /* Inode */
293 struct logfs_block_ops;
294 struct logfs_block {
297  struct super_block *sb;
301  struct page *page;
302  struct inode *inode;
306  int full;
307  int partial;
309 };
310 
311 typedef int write_alias_t(struct super_block *sb, u64 ino, u64 bix,
312  level_t level, int child_no, __be64 val);
317  struct logfs_block *block,
318  write_alias_t *write_one_alias);
319 };
320 
321 #define MAX_JOURNAL_ENTRIES 256
322 
323 struct logfs_super {
324  struct mtd_info *s_mtd; /* underlying device */
325  struct block_device *s_bdev; /* underlying device */
326  const struct logfs_device_ops *s_devops;/* device access */
327  struct inode *s_master_inode; /* inode file */
328  struct inode *s_segfile_inode; /* segment file */
329  struct inode *s_mapping_inode; /* device mapping */
330  atomic_t s_pending_writes; /* outstanting bios */
331  long s_flags;
332  mempool_t *s_btree_pool; /* for btree nodes */
333  mempool_t *s_alias_pool; /* aliases in segment.c */
339  struct page *s_erase_page; /* for dev_bdev.c */
340  /* alias.c fields */
341  struct btree_head32 s_segment_alias; /* remapped segments */
343  struct list_head s_object_alias; /* remapped objects */
344  struct btree_head128 s_object_alias_tree; /* remapped objects */
346  /* dir.c fields */
347  struct mutex s_dirop_mutex; /* for creat/unlink/rename */
348  u64 s_victim_ino; /* used for atomic dir-ops */
349  u64 s_rename_dir; /* source directory ino */
350  u64 s_rename_pos; /* position of source dd */
351  /* gc.c fields */
352  long s_segsize; /* size of a segment */
353  int s_segshift; /* log2 of segment size */
354  long s_segmask; /* 1 << s_segshift - 1 */
355  long s_no_segs; /* segments on device */
356  long s_no_journal_segs; /* segments used for journal */
357  long s_no_blocks; /* blocks per segment */
358  long s_writesize; /* minimum write size */
359  int s_writeshift; /* log2 of write size */
360  u64 s_size; /* filesystem size */
361  struct logfs_area *s_area[LOGFS_NO_AREAS]; /* open segment array */
362  u64 s_gec; /* global erase count */
363  u64 s_wl_gec_ostore; /* time of last wl event */
364  u64 s_wl_gec_journal; /* time of last wl event */
365  u64 s_sweeper; /* current sweeper pos */
366  u8 s_ifile_levels; /* max level of ifile */
367  u8 s_iblock_levels; /* max level of regular files */
368  u8 s_data_levels; /* # of segments to leaf block*/
369  u8 s_total_levels; /* sum of above three */
370  struct btree_head32 s_cand_tree; /* all candidates */
371  struct candidate_list s_free_list; /* 100% free segments */
372  struct candidate_list s_reserve_list; /* Bad segment reserve */
373  struct candidate_list s_low_list[LOGFS_NO_AREAS];/* good candidates */
374  struct candidate_list s_ec_list; /* wear level candidates */
375  struct btree_head32 s_reserved_segments;/* sb, journal, bad, etc. */
376  /* inode.c fields */
377  u64 s_last_ino; /* highest ino used */
379  u32 s_generation; /* i_generation for new files */
380  struct list_head s_freeing_list; /* inodes being freed */
381  /* journal.c fields */
383  void *s_je; /* journal entry to compress */
384  void *s_compressed_je; /* block to write to journal */
385  u32 s_journal_seg[LOGFS_JOURNAL_SEGS]; /* journal segments */
386  u32 s_journal_ec[LOGFS_JOURNAL_SEGS]; /* journal erasecounts */
388  struct logfs_area *s_journal_area; /* open journal segment */
390  int s_no_je;
391 
392  int s_sum_index; /* for the 12 summaries */
394  int s_je_fill; /* index of current je */
395  /* readwrite.c fields */
398  mempool_t *s_block_pool; /* struct logfs_block pool */
399  mempool_t *s_shadow_pool; /* struct logfs_shadow pool */
400  struct list_head s_writeback_list; /* writeback pages */
401  /*
402  * Space accounting:
403  * - s_used_bytes specifies space used to store valid data objects.
404  * - s_dirty_used_bytes is space used to store non-committed data
405  * objects. Those objects have already been written themselves,
406  * but they don't become valid until all indirect blocks up to the
407  * journal have been written as well.
408  * - s_dirty_free_bytes is space used to store the old copy of a
409  * replaced object, as long as the replacement is non-committed.
410  * In other words, it is the amount of space freed when all dirty
411  * blocks are written back.
412  * - s_free_bytes is the amount of free space available for any
413  * purpose.
414  * - s_root_reserve is the amount of free space available only to
415  * the root user. Non-privileged users can no longer write once
416  * this watermark has been reached.
417  * - s_speed_reserve is space which remains unused to speed up
418  * garbage collection performance.
419  * - s_dirty_pages is the space reserved for currently dirty pages.
420  * It is a pessimistic estimate, so some/most will get freed on
421  * page writeback.
422  *
423  * s_used_bytes + s_free_bytes + s_speed_reserve = total usable size
424  */
432  /* Bad block handling:
433  * - s_bad_seg_reserve is a number of segments usually kept
434  * free. When encountering bad blocks, the affected segment's data
435  * is _temporarily_ moved to a reserved segment.
436  * - s_bad_segments is the number of known bad segments.
437  */
440 };
441 
452 struct logfs_inode {
453  struct inode vfs_inode;
461 };
462 
463 #define journal_for_each(__i) for (__i = 0; __i < LOGFS_JOURNAL_SEGS; __i++)
464 #define for_each_area(__i) for (__i = 0; __i < LOGFS_NO_AREAS; __i++)
465 #define for_each_area_down(__i) for (__i = LOGFS_NO_AREAS - 1; __i >= 0; __i--)
466 
467 /* compr.c */
468 int logfs_compress(void *in, void *out, size_t inlen, size_t outlen);
469 int logfs_uncompress(void *in, void *out, size_t inlen, size_t outlen);
470 int __init logfs_compr_init(void);
471 void logfs_compr_exit(void);
472 
473 /* dev_bdev.c */
474 #ifdef CONFIG_BLOCK
475 int logfs_get_sb_bdev(struct logfs_super *s,
476  struct file_system_type *type,
477  const char *devname);
478 #else
479 static inline int logfs_get_sb_bdev(struct logfs_super *s,
480  struct file_system_type *type,
481  const char *devname)
482 {
483  return -ENODEV;
484 }
485 #endif
486 
487 /* dev_mtd.c */
488 #ifdef CONFIG_MTD
489 int logfs_get_sb_mtd(struct logfs_super *s, int mtdnr);
490 #else
491 static inline int logfs_get_sb_mtd(struct logfs_super *s, int mtdnr)
492 {
493  return -ENODEV;
494 }
495 #endif
496 
497 /* dir.c */
498 extern const struct inode_operations logfs_symlink_iops;
499 extern const struct inode_operations logfs_dir_iops;
500 extern const struct file_operations logfs_dir_fops;
501 int logfs_replay_journal(struct super_block *sb);
502 
503 /* file.c */
504 extern const struct inode_operations logfs_reg_iops;
505 extern const struct file_operations logfs_reg_fops;
506 extern const struct address_space_operations logfs_reg_aops;
507 int logfs_readpage(struct file *file, struct page *page);
508 long logfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
509 int logfs_fsync(struct file *file, loff_t start, loff_t end, int datasync);
510 
511 /* gc.c */
512 u32 get_best_cand(struct super_block *sb, struct candidate_list *list, u32 *ec);
513 void logfs_gc_pass(struct super_block *sb);
514 int logfs_check_areas(struct super_block *sb);
515 int logfs_init_gc(struct super_block *sb);
516 void logfs_cleanup_gc(struct super_block *sb);
517 
518 /* inode.c */
519 extern const struct super_operations logfs_super_operations;
520 struct inode *logfs_iget(struct super_block *sb, ino_t ino);
521 struct inode *logfs_safe_iget(struct super_block *sb, ino_t ino, int *cookie);
522 void logfs_safe_iput(struct inode *inode, int cookie);
523 struct inode *logfs_new_inode(struct inode *dir, umode_t mode);
524 struct inode *logfs_new_meta_inode(struct super_block *sb, u64 ino);
525 struct inode *logfs_read_meta_inode(struct super_block *sb, u64 ino);
526 int logfs_init_inode_cache(void);
527 void logfs_destroy_inode_cache(void);
528 void logfs_set_blocks(struct inode *inode, u64 no);
529 /* these logically belong into inode.c but actually reside in readwrite.c */
530 int logfs_read_inode(struct inode *inode);
531 int __logfs_write_inode(struct inode *inode, struct page *, long flags);
532 void logfs_evict_inode(struct inode *inode);
533 
534 /* journal.c */
535 void logfs_write_anchor(struct super_block *sb);
536 int logfs_init_journal(struct super_block *sb);
537 void logfs_cleanup_journal(struct super_block *sb);
539  level_t level, int child_no, __be64 val);
541 
542 /* readwrite.c */
545 int logfs_inode_write(struct inode *inode, const void *buf, size_t count,
546  loff_t bix, long flags, struct shadow_tree *shadow_tree);
547 int logfs_readpage_nolock(struct page *page);
548 int logfs_write_buf(struct inode *inode, struct page *page, long flags);
549 int logfs_delete(struct inode *inode, pgoff_t index,
550  struct shadow_tree *shadow_tree);
551 int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs,
552  gc_level_t gc_level, long flags);
553 int logfs_is_valid_block(struct super_block *sb, u64 ofs, u64 ino, u64 bix,
555 int logfs_truncate(struct inode *inode, u64 size);
558 int logfs_open_segfile(struct super_block *sb);
559 int logfs_init_rw(struct super_block *sb);
560 void logfs_cleanup_rw(struct super_block *sb);
561 void logfs_add_transaction(struct inode *inode, struct logfs_transaction *ta);
562 void logfs_del_transaction(struct inode *inode, struct logfs_transaction *ta);
563 void logfs_write_block(struct logfs_block *block, long flags);
566  struct logfs_segment_entry *se);
567 void logfs_set_segment_used(struct super_block *sb, u64 ofs, int increment);
572 struct logfs_block *__alloc_block(struct super_block *sb,
573  u64 ino, u64 bix, level_t level);
574 void __free_block(struct super_block *sb, struct logfs_block *block);
575 void btree_write_block(struct logfs_block *block);
576 void initialize_block_counters(struct page *page, struct logfs_block *block,
577  __be64 *array, int page_is_empty);
578 int logfs_exist_block(struct inode *inode, u64 bix);
579 int get_page_reserve(struct inode *inode, struct page *page);
580 void logfs_get_wblocks(struct super_block *sb, struct page *page, int lock);
581 void logfs_put_wblocks(struct super_block *sb, struct page *page, int lock);
582 extern struct logfs_block_ops indirect_block_ops;
583 
584 /* segment.c */
585 int logfs_erase_segment(struct super_block *sb, u32 ofs, int ensure_erase);
586 int wbuf_read(struct super_block *sb, u64 ofs, size_t len, void *buf);
587 int logfs_segment_read(struct inode *inode, struct page *page, u64 ofs, u64 bix,
588  level_t level);
589 int logfs_segment_write(struct inode *inode, struct page *page,
590  struct logfs_shadow *shadow);
591 int logfs_segment_delete(struct inode *inode, struct logfs_shadow *shadow);
593  struct logfs_obj_alias *oa, int count);
594 void move_page_to_btree(struct page *page);
595 int logfs_init_mapping(struct super_block *sb);
596 void logfs_sync_area(struct logfs_area *area);
597 void logfs_sync_segments(struct super_block *sb);
598 void freeseg(struct super_block *sb, u32 segno);
599 void free_areas(struct super_block *sb);
600 
601 /* area handling */
602 int logfs_init_areas(struct super_block *sb);
603 void logfs_cleanup_areas(struct super_block *sb);
604 int logfs_open_area(struct logfs_area *area, size_t bytes);
605 int __logfs_buf_write(struct logfs_area *area, u64 ofs, void *buf, size_t len,
606  int use_filler);
607 
608 static inline int logfs_buf_write(struct logfs_area *area, u64 ofs,
609  void *buf, size_t len)
610 {
611  return __logfs_buf_write(area, ofs, buf, len, 0);
612 }
613 
614 static inline int logfs_buf_recover(struct logfs_area *area, u64 ofs,
615  void *buf, size_t len)
616 {
617  return __logfs_buf_write(area, ofs, buf, len, 1);
618 }
619 
620 /* super.c */
622 void emergency_read_end(struct page *page);
623 void logfs_crash_dump(struct super_block *sb);
624 int logfs_statfs(struct dentry *dentry, struct kstatfs *stats);
625 int logfs_check_ds(struct logfs_disk_super *ds);
626 int logfs_write_sb(struct super_block *sb);
627 
628 static inline struct logfs_super *logfs_super(struct super_block *sb)
629 {
630  return sb->s_fs_info;
631 }
632 
633 static inline struct logfs_inode *logfs_inode(struct inode *inode)
634 {
635  return container_of(inode, struct logfs_inode, vfs_inode);
636 }
637 
638 static inline void logfs_set_ro(struct super_block *sb)
639 {
641 }
642 
643 #define LOGFS_BUG(sb) do { \
644  struct super_block *__sb = sb; \
645  logfs_crash_dump(__sb); \
646  logfs_super(__sb)->s_flags |= LOGFS_SB_FLAG_RO; \
647  BUG(); \
648 } while (0)
649 
650 #define LOGFS_BUG_ON(condition, sb) \
651  do { if (unlikely(condition)) LOGFS_BUG((sb)); } while (0)
652 
653 static inline __be32 logfs_crc32(void *data, size_t len, size_t skip)
654 {
655  return cpu_to_be32(crc32(~0, data+skip, len-skip));
656 }
657 
658 static inline u8 logfs_type(struct inode *inode)
659 {
660  return (inode->i_mode >> 12) & 15;
661 }
662 
663 static inline pgoff_t logfs_index(struct super_block *sb, u64 pos)
664 {
665  return pos >> sb->s_blocksize_bits;
666 }
667 
668 static inline u64 dev_ofs(struct super_block *sb, u32 segno, u32 ofs)
669 {
670  return ((u64)segno << logfs_super(sb)->s_segshift) + ofs;
671 }
672 
673 static inline u32 seg_no(struct super_block *sb, u64 ofs)
674 {
675  return ofs >> logfs_super(sb)->s_segshift;
676 }
677 
678 static inline u32 seg_ofs(struct super_block *sb, u64 ofs)
679 {
680  return ofs & logfs_super(sb)->s_segmask;
681 }
682 
683 static inline u64 seg_align(struct super_block *sb, u64 ofs)
684 {
685  return ofs & ~logfs_super(sb)->s_segmask;
686 }
687 
688 static inline struct logfs_block *logfs_block(struct page *page)
689 {
690  return (void *)page->private;
691 }
692 
693 static inline level_t shrink_level(gc_level_t __level)
694 {
695  u8 level = (__force u8)__level;
696 
697  if (level >= LOGFS_MAX_LEVELS)
698  level -= LOGFS_MAX_LEVELS;
699  return (__force level_t)level;
700 }
701 
702 static inline gc_level_t expand_level(u64 ino, level_t __level)
703 {
704  u8 level = (__force u8)__level;
705 
706  if (ino == LOGFS_INO_MASTER) {
707  /* ifile has separate areas */
708  level += LOGFS_MAX_LEVELS;
709  }
710  return (__force gc_level_t)level;
711 }
712 
713 static inline int logfs_block_shift(struct super_block *sb, level_t level)
714 {
715  level = shrink_level((__force gc_level_t)level);
716  return (__force int)level * (sb->s_blocksize_bits - 3);
717 }
718 
719 static inline u64 logfs_block_mask(struct super_block *sb, level_t level)
720 {
721  return ~0ull << logfs_block_shift(sb, level);
722 }
723 
724 static inline struct logfs_area *get_area(struct super_block *sb,
726 {
727  return logfs_super(sb)->s_area[(__force u8)gc_level];
728 }
729 
730 static inline void logfs_mempool_destroy(mempool_t *pool)
731 {
732  if (pool)
733  mempool_destroy(pool);
734 }
735 
736 #endif