1 /*
2 * linux/include/linux/jbd2.h
3 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>
5 *
6 * Copyright 1998-2000 Red Hat, Inc --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Definitions for transaction data structures for the buffer cache
13 * filesystem journaling support.
14 */
15
16 #ifndef _LINUX_JBD2_H
17 #define _LINUX_JBD2_H
18
19 /* Allow this file to be included directly into e2fsprogs */
20 #ifndef __KERNEL__
21 #include "jfs_compat.h"
22 #define JBD2_DEBUG
23 #define jfs_debug jbd_debug
24 #else
25
26 #include <linux/types.h>
27 #include <linux/buffer_head.h>
28 #include <linux/journal-head.h>
29 #include <linux/stddef.h>
30 #include <linux/bit_spinlock.h>
31 #include <linux/mutex.h>
32 #include <linux/timer.h>
33 #endif
34
35 #define journal_oom_retry 1
36
37 /*
38 * Define JBD2_PARANIOD_IOFAIL to cause a kernel BUG() if ext4 finds
39 * certain classes of error which can occur due to failed IOs. Under
40 * normal use we want ext4 to continue after such errors, because
41 * hardware _can_ fail, but for debugging purposes when running tests on
42 * known-good hardware we may want to trap these errors.
43 */
44 #undef JBD2_PARANOID_IOFAIL
45
46 /*
47 * The default maximum commit age, in seconds.
48 */
49 #define JBD2_DEFAULT_MAX_COMMIT_AGE 5
50
51 #ifdef CONFIG_JBD2_DEBUG
52 /*
53 * Define JBD2_EXPENSIVE_CHECKING to enable more expensive internal
54 * consistency checks. By default we don't do this unless
55 * CONFIG_JBD2_DEBUG is on.
56 */
57 #define JBD2_EXPENSIVE_CHECKING
58 extern u8 jbd2_journal_enable_debug;
59
60 #define jbd_debug(n, f, a...) \
61 do { \
62 if ((n) <= jbd2_journal_enable_debug) { \
63 printk (KERN_DEBUG "(%s, %d): %s: ", \
64 __FILE__, __LINE__, __func__); \
65 printk (f, ## a); \
66 } \
67 } while (0)
68 #else
69 #define jbd_debug(f, a...) /**/
70 #endif
71
jbd2_alloc(size_t size,gfp_t flags)72 static inline void *jbd2_alloc(size_t size, gfp_t flags)
73 {
74 return (void *)__get_free_pages(flags, get_order(size));
75 }
76
jbd2_free(void * ptr,size_t size)77 static inline void jbd2_free(void *ptr, size_t size)
78 {
79 free_pages((unsigned long)ptr, get_order(size));
80 };
81
82 #define JBD2_MIN_JOURNAL_BLOCKS 1024
83
84 #ifdef __KERNEL__
85
86 /**
87 * typedef handle_t - The handle_t type represents a single atomic update being performed by some process.
88 *
89 * All filesystem modifications made by the process go
90 * through this handle. Recursive operations (such as quota operations)
91 * are gathered into a single update.
92 *
93 * The buffer credits field is used to account for journaled buffers
94 * being modified by the running process. To ensure that there is
95 * enough log space for all outstanding operations, we need to limit the
96 * number of outstanding buffers possible at any time. When the
97 * operation completes, any buffer credits not used are credited back to
98 * the transaction, so that at all times we know how many buffers the
99 * outstanding updates on a transaction might possibly touch.
100 *
101 * This is an opaque datatype.
102 **/
103 typedef struct handle_s handle_t; /* Atomic operation type */
104
105
106 /**
107 * typedef journal_t - The journal_t maintains all of the journaling state information for a single filesystem.
108 *
109 * journal_t is linked to from the fs superblock structure.
110 *
111 * We use the journal_t to keep track of all outstanding transaction
112 * activity on the filesystem, and to manage the state of the log
113 * writing process.
114 *
115 * This is an opaque datatype.
116 **/
117 typedef struct journal_s journal_t; /* Journal control structure */
118 #endif
119
120 /*
121 * Internal structures used by the logging mechanism:
122 */
123
124 #define JBD2_MAGIC_NUMBER 0xc03b3998U /* The first 4 bytes of /dev/random! */
125
126 /*
127 * On-disk structures
128 */
129
130 /*
131 * Descriptor block types:
132 */
133
134 #define JBD2_DESCRIPTOR_BLOCK 1
135 #define JBD2_COMMIT_BLOCK 2
136 #define JBD2_SUPERBLOCK_V1 3
137 #define JBD2_SUPERBLOCK_V2 4
138 #define JBD2_REVOKE_BLOCK 5
139
140 /*
141 * Standard header for all descriptor blocks:
142 */
143 typedef struct journal_header_s
144 {
145 __be32 h_magic;
146 __be32 h_blocktype;
147 __be32 h_sequence;
148 } journal_header_t;
149
150 /*
151 * Checksum types.
152 */
153 #define JBD2_CRC32_CHKSUM 1
154 #define JBD2_MD5_CHKSUM 2
155 #define JBD2_SHA1_CHKSUM 3
156
157 #define JBD2_CRC32_CHKSUM_SIZE 4
158
159 #define JBD2_CHECKSUM_BYTES (32 / sizeof(u32))
160 /*
161 * Commit block header for storing transactional checksums:
162 */
163 struct commit_header {
164 __be32 h_magic;
165 __be32 h_blocktype;
166 __be32 h_sequence;
167 unsigned char h_chksum_type;
168 unsigned char h_chksum_size;
169 unsigned char h_padding[2];
170 __be32 h_chksum[JBD2_CHECKSUM_BYTES];
171 __be64 h_commit_sec;
172 __be32 h_commit_nsec;
173 };
174
175 /*
176 * The block tag: used to describe a single buffer in the journal.
177 * t_blocknr_high is only used if INCOMPAT_64BIT is set, so this
178 * raw struct shouldn't be used for pointer math or sizeof() - use
179 * journal_tag_bytes(journal) instead to compute this.
180 */
181 typedef struct journal_block_tag_s
182 {
183 __be32 t_blocknr; /* The on-disk block number */
184 __be32 t_flags; /* See below */
185 __be32 t_blocknr_high; /* most-significant high 32bits. */
186 } journal_block_tag_t;
187
188 #define JBD2_TAG_SIZE32 (offsetof(journal_block_tag_t, t_blocknr_high))
189 #define JBD2_TAG_SIZE64 (sizeof(journal_block_tag_t))
190
191 /*
192 * The revoke descriptor: used on disk to describe a series of blocks to
193 * be revoked from the log
194 */
195 typedef struct jbd2_journal_revoke_header_s
196 {
197 journal_header_t r_header;
198 __be32 r_count; /* Count of bytes used in the block */
199 } jbd2_journal_revoke_header_t;
200
201
202 /* Definitions for the journal tag flags word: */
203 #define JBD2_FLAG_ESCAPE 1 /* on-disk block is escaped */
204 #define JBD2_FLAG_SAME_UUID 2 /* block has same uuid as previous */
205 #define JBD2_FLAG_DELETED 4 /* block deleted by this transaction */
206 #define JBD2_FLAG_LAST_TAG 8 /* last tag in this descriptor block */
207
208
209 /*
210 * The journal superblock. All fields are in big-endian byte order.
211 */
212 typedef struct journal_superblock_s
213 {
214 /* 0x0000 */
215 journal_header_t s_header;
216
217 /* 0x000C */
218 /* Static information describing the journal */
219 __be32 s_blocksize; /* journal device blocksize */
220 __be32 s_maxlen; /* total blocks in journal file */
221 __be32 s_first; /* first block of log information */
222
223 /* 0x0018 */
224 /* Dynamic information describing the current state of the log */
225 __be32 s_sequence; /* first commit ID expected in log */
226 __be32 s_start; /* blocknr of start of log */
227
228 /* 0x0020 */
229 /* Error value, as set by jbd2_journal_abort(). */
230 __be32 s_errno;
231
232 /* 0x0024 */
233 /* Remaining fields are only valid in a version-2 superblock */
234 __be32 s_feature_compat; /* compatible feature set */
235 __be32 s_feature_incompat; /* incompatible feature set */
236 __be32 s_feature_ro_compat; /* readonly-compatible feature set */
237 /* 0x0030 */
238 __u8 s_uuid[16]; /* 128-bit uuid for journal */
239
240 /* 0x0040 */
241 __be32 s_nr_users; /* Nr of filesystems sharing log */
242
243 __be32 s_dynsuper; /* Blocknr of dynamic superblock copy*/
244
245 /* 0x0048 */
246 __be32 s_max_transaction; /* Limit of journal blocks per trans.*/
247 __be32 s_max_trans_data; /* Limit of data blocks per trans. */
248
249 /* 0x0050 */
250 __u32 s_padding[44];
251
252 /* 0x0100 */
253 __u8 s_users[16*48]; /* ids of all fs'es sharing the log */
254 /* 0x0400 */
255 } journal_superblock_t;
256
257 #define JBD2_HAS_COMPAT_FEATURE(j,mask) \
258 ((j)->j_format_version >= 2 && \
259 ((j)->j_superblock->s_feature_compat & cpu_to_be32((mask))))
260 #define JBD2_HAS_RO_COMPAT_FEATURE(j,mask) \
261 ((j)->j_format_version >= 2 && \
262 ((j)->j_superblock->s_feature_ro_compat & cpu_to_be32((mask))))
263 #define JBD2_HAS_INCOMPAT_FEATURE(j,mask) \
264 ((j)->j_format_version >= 2 && \
265 ((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask))))
266
267 #define JBD2_FEATURE_COMPAT_CHECKSUM 0x00000001
268
269 #define JBD2_FEATURE_INCOMPAT_REVOKE 0x00000001
270 #define JBD2_FEATURE_INCOMPAT_64BIT 0x00000002
271 #define JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT 0x00000004
272
273 /* Features known to this kernel version: */
274 #define JBD2_KNOWN_COMPAT_FEATURES JBD2_FEATURE_COMPAT_CHECKSUM
275 #define JBD2_KNOWN_ROCOMPAT_FEATURES 0
276 #define JBD2_KNOWN_INCOMPAT_FEATURES (JBD2_FEATURE_INCOMPAT_REVOKE | \
277 JBD2_FEATURE_INCOMPAT_64BIT | \
278 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)
279
280 #ifdef __KERNEL__
281
282 #include <linux/fs.h>
283 #include <linux/sched.h>
284
285 #define J_ASSERT(assert) BUG_ON(!(assert))
286
287 #if defined(CONFIG_BUFFER_DEBUG)
288 void buffer_assertion_failure(struct buffer_head *bh);
289 #define J_ASSERT_BH(bh, expr) \
290 do { \
291 if (!(expr)) \
292 buffer_assertion_failure(bh); \
293 J_ASSERT(expr); \
294 } while (0)
295 #define J_ASSERT_JH(jh, expr) J_ASSERT_BH(jh2bh(jh), expr)
296 #else
297 #define J_ASSERT_BH(bh, expr) J_ASSERT(expr)
298 #define J_ASSERT_JH(jh, expr) J_ASSERT(expr)
299 #endif
300
301 #if defined(JBD2_PARANOID_IOFAIL)
302 #define J_EXPECT(expr, why...) J_ASSERT(expr)
303 #define J_EXPECT_BH(bh, expr, why...) J_ASSERT_BH(bh, expr)
304 #define J_EXPECT_JH(jh, expr, why...) J_ASSERT_JH(jh, expr)
305 #else
306 #define __journal_expect(expr, why...) \
307 ({ \
308 int val = (expr); \
309 if (!val) { \
310 printk(KERN_ERR \
311 "JBD2 unexpected failure: %s: %s;\n", \
312 __func__, #expr); \
313 printk(KERN_ERR why "\n"); \
314 } \
315 val; \
316 })
317 #define J_EXPECT(expr, why...) __journal_expect(expr, ## why)
318 #define J_EXPECT_BH(bh, expr, why...) __journal_expect(expr, ## why)
319 #define J_EXPECT_JH(jh, expr, why...) __journal_expect(expr, ## why)
320 #endif
321
322 enum jbd_state_bits {
323 BH_JBD /* Has an attached ext3 journal_head */
324 = BH_PrivateStart,
325 BH_JWrite, /* Being written to log (@@@ DEBUGGING) */
326 BH_Freed, /* Has been freed (truncated) */
327 BH_Revoked, /* Has been revoked from the log */
328 BH_RevokeValid, /* Revoked flag is valid */
329 BH_JBDDirty, /* Is dirty but journaled */
330 BH_State, /* Pins most journal_head state */
331 BH_JournalHead, /* Pins bh->b_private and jh->b_bh */
332 BH_Unshadow, /* Dummy bit, for BJ_Shadow wakeup filtering */
333 BH_JBDPrivateStart, /* First bit available for private use by FS */
334 };
335
BUFFER_FNS(JBD,jbd)336 BUFFER_FNS(JBD, jbd)
337 BUFFER_FNS(JWrite, jwrite)
338 BUFFER_FNS(JBDDirty, jbddirty)
339 TAS_BUFFER_FNS(JBDDirty, jbddirty)
340 BUFFER_FNS(Revoked, revoked)
341 TAS_BUFFER_FNS(Revoked, revoked)
342 BUFFER_FNS(RevokeValid, revokevalid)
343 TAS_BUFFER_FNS(RevokeValid, revokevalid)
344 BUFFER_FNS(Freed, freed)
345
346 static inline struct buffer_head *jh2bh(struct journal_head *jh)
347 {
348 return jh->b_bh;
349 }
350
bh2jh(struct buffer_head * bh)351 static inline struct journal_head *bh2jh(struct buffer_head *bh)
352 {
353 return bh->b_private;
354 }
355
jbd_lock_bh_state(struct buffer_head * bh)356 static inline void jbd_lock_bh_state(struct buffer_head *bh)
357 {
358 bit_spin_lock(BH_State, &bh->b_state);
359 }
360
jbd_trylock_bh_state(struct buffer_head * bh)361 static inline int jbd_trylock_bh_state(struct buffer_head *bh)
362 {
363 return bit_spin_trylock(BH_State, &bh->b_state);
364 }
365
jbd_is_locked_bh_state(struct buffer_head * bh)366 static inline int jbd_is_locked_bh_state(struct buffer_head *bh)
367 {
368 return bit_spin_is_locked(BH_State, &bh->b_state);
369 }
370
jbd_unlock_bh_state(struct buffer_head * bh)371 static inline void jbd_unlock_bh_state(struct buffer_head *bh)
372 {
373 bit_spin_unlock(BH_State, &bh->b_state);
374 }
375
jbd_lock_bh_journal_head(struct buffer_head * bh)376 static inline void jbd_lock_bh_journal_head(struct buffer_head *bh)
377 {
378 bit_spin_lock(BH_JournalHead, &bh->b_state);
379 }
380
jbd_unlock_bh_journal_head(struct buffer_head * bh)381 static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh)
382 {
383 bit_spin_unlock(BH_JournalHead, &bh->b_state);
384 }
385
386 /* Flags in jbd_inode->i_flags */
387 #define __JI_COMMIT_RUNNING 0
388 /* Commit of the inode data in progress. We use this flag to protect us from
389 * concurrent deletion of inode. We cannot use reference to inode for this
390 * since we cannot afford doing last iput() on behalf of kjournald
391 */
392 #define JI_COMMIT_RUNNING (1 << __JI_COMMIT_RUNNING)
393
394 /**
395 * struct jbd_inode is the structure linking inodes in ordered mode
396 * present in a transaction so that we can sync them during commit.
397 */
398 struct jbd2_inode {
399 /* Which transaction does this inode belong to? Either the running
400 * transaction or the committing one. [j_list_lock] */
401 transaction_t *i_transaction;
402
403 /* Pointer to the running transaction modifying inode's data in case
404 * there is already a committing transaction touching it. [j_list_lock] */
405 transaction_t *i_next_transaction;
406
407 /* List of inodes in the i_transaction [j_list_lock] */
408 struct list_head i_list;
409
410 /* VFS inode this inode belongs to [constant during the lifetime
411 * of the structure] */
412 struct inode *i_vfs_inode;
413
414 /* Flags of inode [j_list_lock] */
415 unsigned int i_flags;
416 };
417
418 struct jbd2_revoke_table_s;
419
420 /**
421 * struct handle_s - The handle_s type is the concrete type associated with
422 * handle_t.
423 * @h_transaction: Which compound transaction is this update a part of?
424 * @h_buffer_credits: Number of remaining buffers we are allowed to dirty.
425 * @h_ref: Reference count on this handle
426 * @h_err: Field for caller's use to track errors through large fs operations
427 * @h_sync: flag for sync-on-close
428 * @h_jdata: flag to force data journaling
429 * @h_aborted: flag indicating fatal error on handle
430 **/
431
432 /* Docbook can't yet cope with the bit fields, but will leave the documentation
433 * in so it can be fixed later.
434 */
435
436 struct handle_s
437 {
438 /* Which compound transaction is this update a part of? */
439 transaction_t *h_transaction;
440
441 /* Number of remaining buffers we are allowed to dirty: */
442 int h_buffer_credits;
443
444 /* Reference count on this handle */
445 int h_ref;
446
447 /* Field for caller's use to track errors through large fs */
448 /* operations */
449 int h_err;
450
451 /* Flags [no locking] */
452 unsigned int h_sync: 1; /* sync-on-close */
453 unsigned int h_jdata: 1; /* force data journaling */
454 unsigned int h_aborted: 1; /* fatal error on handle */
455
456 #ifdef CONFIG_DEBUG_LOCK_ALLOC
457 struct lockdep_map h_lockdep_map;
458 #endif
459 };
460
461
462 /*
463 * Some stats for checkpoint phase
464 */
465 struct transaction_chp_stats_s {
466 unsigned long cs_chp_time;
467 unsigned long cs_forced_to_close;
468 unsigned long cs_written;
469 unsigned long cs_dropped;
470 };
471
472 /* The transaction_t type is the guts of the journaling mechanism. It
473 * tracks a compound transaction through its various states:
474 *
475 * RUNNING: accepting new updates
476 * LOCKED: Updates still running but we don't accept new ones
477 * RUNDOWN: Updates are tidying up but have finished requesting
478 * new buffers to modify (state not used for now)
479 * FLUSH: All updates complete, but we are still writing to disk
480 * COMMIT: All data on disk, writing commit record
481 * FINISHED: We still have to keep the transaction for checkpointing.
482 *
483 * The transaction keeps track of all of the buffers modified by a
484 * running transaction, and all of the buffers committed but not yet
485 * flushed to home for finished transactions.
486 */
487
488 /*
489 * Lock ranking:
490 *
491 * j_list_lock
492 * ->jbd_lock_bh_journal_head() (This is "innermost")
493 *
494 * j_state_lock
495 * ->jbd_lock_bh_state()
496 *
497 * jbd_lock_bh_state()
498 * ->j_list_lock
499 *
500 * j_state_lock
501 * ->t_handle_lock
502 *
503 * j_state_lock
504 * ->j_list_lock (journal_unmap_buffer)
505 *
506 */
507
508 struct transaction_s
509 {
510 /* Pointer to the journal for this transaction. [no locking] */
511 journal_t *t_journal;
512
513 /* Sequence number for this transaction [no locking] */
514 tid_t t_tid;
515
516 /*
517 * Transaction's current state
518 * [no locking - only kjournald2 alters this]
519 * [j_list_lock] guards transition of a transaction into T_FINISHED
520 * state and subsequent call of __jbd2_journal_drop_transaction()
521 * FIXME: needs barriers
522 * KLUDGE: [use j_state_lock]
523 */
524 enum {
525 T_RUNNING,
526 T_LOCKED,
527 T_RUNDOWN,
528 T_FLUSH,
529 T_COMMIT,
530 T_FINISHED
531 } t_state;
532
533 /*
534 * Where in the log does this transaction's commit start? [no locking]
535 */
536 unsigned long t_log_start;
537
538 /* Number of buffers on the t_buffers list [j_list_lock] */
539 int t_nr_buffers;
540
541 /*
542 * Doubly-linked circular list of all buffers reserved but not yet
543 * modified by this transaction [j_list_lock]
544 */
545 struct journal_head *t_reserved_list;
546
547 /*
548 * Doubly-linked circular list of all metadata buffers owned by this
549 * transaction [j_list_lock]
550 */
551 struct journal_head *t_buffers;
552
553 /*
554 * Doubly-linked circular list of all forget buffers (superseded
555 * buffers which we can un-checkpoint once this transaction commits)
556 * [j_list_lock]
557 */
558 struct journal_head *t_forget;
559
560 /*
561 * Doubly-linked circular list of all buffers still to be flushed before
562 * this transaction can be checkpointed. [j_list_lock]
563 */
564 struct journal_head *t_checkpoint_list;
565
566 /*
567 * Doubly-linked circular list of all buffers submitted for IO while
568 * checkpointing. [j_list_lock]
569 */
570 struct journal_head *t_checkpoint_io_list;
571
572 /*
573 * Doubly-linked circular list of temporary buffers currently undergoing
574 * IO in the log [j_list_lock]
575 */
576 struct journal_head *t_iobuf_list;
577
578 /*
579 * Doubly-linked circular list of metadata buffers being shadowed by log
580 * IO. The IO buffers on the iobuf list and the shadow buffers on this
581 * list match each other one for one at all times. [j_list_lock]
582 */
583 struct journal_head *t_shadow_list;
584
585 /*
586 * Doubly-linked circular list of control buffers being written to the
587 * log. [j_list_lock]
588 */
589 struct journal_head *t_log_list;
590
591 /*
592 * List of inodes whose data we've modified in data=ordered mode.
593 * [j_list_lock]
594 */
595 struct list_head t_inode_list;
596
597 /*
598 * Protects info related to handles
599 */
600 spinlock_t t_handle_lock;
601
602 /*
603 * Longest time some handle had to wait for running transaction
604 */
605 unsigned long t_max_wait;
606
607 /*
608 * When transaction started
609 */
610 unsigned long t_start;
611
612 /*
613 * Checkpointing stats [j_checkpoint_sem]
614 */
615 struct transaction_chp_stats_s t_chp_stats;
616
617 /*
618 * Number of outstanding updates running on this transaction
619 * [t_handle_lock]
620 */
621 int t_updates;
622
623 /*
624 * Number of buffers reserved for use by all handles in this transaction
625 * handle but not yet modified. [t_handle_lock]
626 */
627 int t_outstanding_credits;
628
629 /*
630 * Forward and backward links for the circular list of all transactions
631 * awaiting checkpoint. [j_list_lock]
632 */
633 transaction_t *t_cpnext, *t_cpprev;
634
635 /*
636 * When will the transaction expire (become due for commit), in jiffies?
637 * [no locking]
638 */
639 unsigned long t_expires;
640
641 /*
642 * When this transaction started, in nanoseconds [no locking]
643 */
644 ktime_t t_start_time;
645
646 /*
647 * How many handles used this transaction? [t_handle_lock]
648 */
649 int t_handle_count;
650
651 /*
652 * For use by the filesystem to store fs-specific data
653 * structures associated with the transaction
654 */
655 struct list_head t_private_list;
656 };
657
658 struct transaction_run_stats_s {
659 unsigned long rs_wait;
660 unsigned long rs_running;
661 unsigned long rs_locked;
662 unsigned long rs_flushing;
663 unsigned long rs_logging;
664
665 unsigned long rs_handle_count;
666 unsigned long rs_blocks;
667 unsigned long rs_blocks_logged;
668 };
669
670 struct transaction_stats_s {
671 int ts_type;
672 unsigned long ts_tid;
673 union {
674 struct transaction_run_stats_s run;
675 struct transaction_chp_stats_s chp;
676 } u;
677 };
678
679 #define JBD2_STATS_RUN 1
680 #define JBD2_STATS_CHECKPOINT 2
681
682 static inline unsigned long
jbd2_time_diff(unsigned long start,unsigned long end)683 jbd2_time_diff(unsigned long start, unsigned long end)
684 {
685 if (end >= start)
686 return end - start;
687
688 return end + (MAX_JIFFY_OFFSET - start);
689 }
690
691 #define JBD2_NR_BATCH 64
692
693 /**
694 * struct journal_s - The journal_s type is the concrete type associated with
695 * journal_t.
696 * @j_flags: General journaling state flags
697 * @j_errno: Is there an outstanding uncleared error on the journal (from a
698 * prior abort)?
699 * @j_sb_buffer: First part of superblock buffer
700 * @j_superblock: Second part of superblock buffer
701 * @j_format_version: Version of the superblock format
702 * @j_state_lock: Protect the various scalars in the journal
703 * @j_barrier_count: Number of processes waiting to create a barrier lock
704 * @j_barrier: The barrier lock itself
705 * @j_running_transaction: The current running transaction..
706 * @j_committing_transaction: the transaction we are pushing to disk
707 * @j_checkpoint_transactions: a linked circular list of all transactions
708 * waiting for checkpointing
709 * @j_wait_transaction_locked: Wait queue for waiting for a locked transaction
710 * to start committing, or for a barrier lock to be released
711 * @j_wait_logspace: Wait queue for waiting for checkpointing to complete
712 * @j_wait_done_commit: Wait queue for waiting for commit to complete
713 * @j_wait_checkpoint: Wait queue to trigger checkpointing
714 * @j_wait_commit: Wait queue to trigger commit
715 * @j_wait_updates: Wait queue to wait for updates to complete
716 * @j_checkpoint_mutex: Mutex for locking against concurrent checkpoints
717 * @j_head: Journal head - identifies the first unused block in the journal
718 * @j_tail: Journal tail - identifies the oldest still-used block in the
719 * journal.
720 * @j_free: Journal free - how many free blocks are there in the journal?
721 * @j_first: The block number of the first usable block
722 * @j_last: The block number one beyond the last usable block
723 * @j_dev: Device where we store the journal
724 * @j_blocksize: blocksize for the location where we store the journal.
725 * @j_blk_offset: starting block offset for into the device where we store the
726 * journal
727 * @j_fs_dev: Device which holds the client fs. For internal journal this will
728 * be equal to j_dev
729 * @j_maxlen: Total maximum capacity of the journal region on disk.
730 * @j_list_lock: Protects the buffer lists and internal buffer state.
731 * @j_inode: Optional inode where we store the journal. If present, all journal
732 * block numbers are mapped into this inode via bmap().
733 * @j_tail_sequence: Sequence number of the oldest transaction in the log
734 * @j_transaction_sequence: Sequence number of the next transaction to grant
735 * @j_commit_sequence: Sequence number of the most recently committed
736 * transaction
737 * @j_commit_request: Sequence number of the most recent transaction wanting
738 * commit
739 * @j_uuid: Uuid of client object.
740 * @j_task: Pointer to the current commit thread for this journal
741 * @j_max_transaction_buffers: Maximum number of metadata buffers to allow in a
742 * single compound commit transaction
743 * @j_commit_interval: What is the maximum transaction lifetime before we begin
744 * a commit?
745 * @j_commit_timer: The timer used to wakeup the commit thread
746 * @j_revoke_lock: Protect the revoke table
747 * @j_revoke: The revoke table - maintains the list of revoked blocks in the
748 * current transaction.
749 * @j_revoke_table: alternate revoke tables for j_revoke
750 * @j_wbuf: array of buffer_heads for jbd2_journal_commit_transaction
751 * @j_wbufsize: maximum number of buffer_heads allowed in j_wbuf, the
752 * number that will fit in j_blocksize
753 * @j_last_sync_writer: most recent pid which did a synchronous write
754 * @j_history: Buffer storing the transactions statistics history
755 * @j_history_max: Maximum number of transactions in the statistics history
756 * @j_history_cur: Current number of transactions in the statistics history
757 * @j_history_lock: Protect the transactions statistics history
758 * @j_proc_entry: procfs entry for the jbd statistics directory
759 * @j_stats: Overall statistics
760 * @j_private: An opaque pointer to fs-private information.
761 */
762
763 struct journal_s
764 {
765 /* General journaling state flags [j_state_lock] */
766 unsigned long j_flags;
767
768 /*
769 * Is there an outstanding uncleared error on the journal (from a prior
770 * abort)? [j_state_lock]
771 */
772 int j_errno;
773
774 /* The superblock buffer */
775 struct buffer_head *j_sb_buffer;
776 journal_superblock_t *j_superblock;
777
778 /* Version of the superblock format */
779 int j_format_version;
780
781 /*
782 * Protect the various scalars in the journal
783 */
784 spinlock_t j_state_lock;
785
786 /*
787 * Number of processes waiting to create a barrier lock [j_state_lock]
788 */
789 int j_barrier_count;
790
791 /* The barrier lock itself */
792 struct mutex j_barrier;
793
794 /*
795 * Transactions: The current running transaction...
796 * [j_state_lock] [caller holding open handle]
797 */
798 transaction_t *j_running_transaction;
799
800 /*
801 * the transaction we are pushing to disk
802 * [j_state_lock] [caller holding open handle]
803 */
804 transaction_t *j_committing_transaction;
805
806 /*
807 * ... and a linked circular list of all transactions waiting for
808 * checkpointing. [j_list_lock]
809 */
810 transaction_t *j_checkpoint_transactions;
811
812 /*
813 * Wait queue for waiting for a locked transaction to start committing,
814 * or for a barrier lock to be released
815 */
816 wait_queue_head_t j_wait_transaction_locked;
817
818 /* Wait queue for waiting for checkpointing to complete */
819 wait_queue_head_t j_wait_logspace;
820
821 /* Wait queue for waiting for commit to complete */
822 wait_queue_head_t j_wait_done_commit;
823
824 /* Wait queue to trigger checkpointing */
825 wait_queue_head_t j_wait_checkpoint;
826
827 /* Wait queue to trigger commit */
828 wait_queue_head_t j_wait_commit;
829
830 /* Wait queue to wait for updates to complete */
831 wait_queue_head_t j_wait_updates;
832
833 /* Semaphore for locking against concurrent checkpoints */
834 struct mutex j_checkpoint_mutex;
835
836 /*
837 * List of buffer heads used by the checkpoint routine. This
838 * was moved from jbd2_log_do_checkpoint() to reduce stack
839 * usage. Access to this array is controlled by the
840 * j_checkpoint_mutex. [j_checkpoint_mutex]
841 */
842 struct buffer_head *j_chkpt_bhs[JBD2_NR_BATCH];
843
844 /*
845 * Journal head: identifies the first unused block in the journal.
846 * [j_state_lock]
847 */
848 unsigned long j_head;
849
850 /*
851 * Journal tail: identifies the oldest still-used block in the journal.
852 * [j_state_lock]
853 */
854 unsigned long j_tail;
855
856 /*
857 * Journal free: how many free blocks are there in the journal?
858 * [j_state_lock]
859 */
860 unsigned long j_free;
861
862 /*
863 * Journal start and end: the block numbers of the first usable block
864 * and one beyond the last usable block in the journal. [j_state_lock]
865 */
866 unsigned long j_first;
867 unsigned long j_last;
868
869 /*
870 * Device, blocksize and starting block offset for the location where we
871 * store the journal.
872 */
873 struct block_device *j_dev;
874 int j_blocksize;
875 unsigned long long j_blk_offset;
876 char j_devname[BDEVNAME_SIZE+24];
877
878 /*
879 * Device which holds the client fs. For internal journal this will be
880 * equal to j_dev.
881 */
882 struct block_device *j_fs_dev;
883
884 /* Total maximum capacity of the journal region on disk. */
885 unsigned int j_maxlen;
886
887 /*
888 * Protects the buffer lists and internal buffer state.
889 */
890 spinlock_t j_list_lock;
891
892 /* Optional inode where we store the journal. If present, all */
893 /* journal block numbers are mapped into this inode via */
894 /* bmap(). */
895 struct inode *j_inode;
896
897 /*
898 * Sequence number of the oldest transaction in the log [j_state_lock]
899 */
900 tid_t j_tail_sequence;
901
902 /*
903 * Sequence number of the next transaction to grant [j_state_lock]
904 */
905 tid_t j_transaction_sequence;
906
907 /*
908 * Sequence number of the most recently committed transaction
909 * [j_state_lock].
910 */
911 tid_t j_commit_sequence;
912
913 /*
914 * Sequence number of the most recent transaction wanting commit
915 * [j_state_lock]
916 */
917 tid_t j_commit_request;
918
919 /*
920 * Journal uuid: identifies the object (filesystem, LVM volume etc)
921 * backed by this journal. This will eventually be replaced by an array
922 * of uuids, allowing us to index multiple devices within a single
923 * journal and to perform atomic updates across them.
924 */
925 __u8 j_uuid[16];
926
927 /* Pointer to the current commit thread for this journal */
928 struct task_struct *j_task;
929
930 /*
931 * Maximum number of metadata buffers to allow in a single compound
932 * commit transaction
933 */
934 int j_max_transaction_buffers;
935
936 /*
937 * What is the maximum transaction lifetime before we begin a commit?
938 */
939 unsigned long j_commit_interval;
940
941 /* The timer used to wakeup the commit thread: */
942 struct timer_list j_commit_timer;
943
944 /*
945 * The revoke table: maintains the list of revoked blocks in the
946 * current transaction. [j_revoke_lock]
947 */
948 spinlock_t j_revoke_lock;
949 struct jbd2_revoke_table_s *j_revoke;
950 struct jbd2_revoke_table_s *j_revoke_table[2];
951
952 /*
953 * array of bhs for jbd2_journal_commit_transaction
954 */
955 struct buffer_head **j_wbuf;
956 int j_wbufsize;
957
958 /*
959 * this is the pid of hte last person to run a synchronous operation
960 * through the journal
961 */
962 pid_t j_last_sync_writer;
963
964 /*
965 * the average amount of time in nanoseconds it takes to commit a
966 * transaction to disk. [j_state_lock]
967 */
968 u64 j_average_commit_time;
969
970 /*
971 * minimum and maximum times that we should wait for
972 * additional filesystem operations to get batched into a
973 * synchronous handle in microseconds
974 */
975 u32 j_min_batch_time;
976 u32 j_max_batch_time;
977
978 /* This function is called when a transaction is closed */
979 void (*j_commit_callback)(journal_t *,
980 transaction_t *);
981
982 /*
983 * Journal statistics
984 */
985 struct transaction_stats_s *j_history;
986 int j_history_max;
987 int j_history_cur;
988 /*
989 * Protect the transactions statistics history
990 */
991 spinlock_t j_history_lock;
992 struct proc_dir_entry *j_proc_entry;
993 struct transaction_stats_s j_stats;
994
995 /* Failed journal commit ID */
996 unsigned int j_failed_commit;
997
998 /*
999 * An opaque pointer to fs-private information. ext3 puts its
1000 * superblock pointer here
1001 */
1002 void *j_private;
1003 };
1004
1005 /*
1006 * Journal flag definitions
1007 */
1008 #define JBD2_UNMOUNT 0x001 /* Journal thread is being destroyed */
1009 #define JBD2_ABORT 0x002 /* Journaling has been aborted for errors. */
1010 #define JBD2_ACK_ERR 0x004 /* The errno in the sb has been acked */
1011 #define JBD2_FLUSHED 0x008 /* The journal superblock has been flushed */
1012 #define JBD2_LOADED 0x010 /* The journal superblock has been loaded */
1013 #define JBD2_BARRIER 0x020 /* Use IDE barriers */
1014 #define JBD2_ABORT_ON_SYNCDATA_ERR 0x040 /* Abort the journal on file
1015 * data write error in ordered
1016 * mode */
1017
1018 /*
1019 * Function declarations for the journaling transaction and buffer
1020 * management
1021 */
1022
1023 /* Filing buffers */
1024 extern void jbd2_journal_unfile_buffer(journal_t *, struct journal_head *);
1025 extern void __jbd2_journal_unfile_buffer(struct journal_head *);
1026 extern void __jbd2_journal_refile_buffer(struct journal_head *);
1027 extern void jbd2_journal_refile_buffer(journal_t *, struct journal_head *);
1028 extern void __jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int);
1029 extern void __journal_free_buffer(struct journal_head *bh);
1030 extern void jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int);
1031 extern void __journal_clean_data_list(transaction_t *transaction);
1032
1033 /* Log buffer allocation */
1034 extern struct journal_head * jbd2_journal_get_descriptor_buffer(journal_t *);
1035 int jbd2_journal_next_log_block(journal_t *, unsigned long long *);
1036
1037 /* Commit management */
1038 extern void jbd2_journal_commit_transaction(journal_t *);
1039
1040 /* Checkpoint list management */
1041 int __jbd2_journal_clean_checkpoint_list(journal_t *journal);
1042 int __jbd2_journal_remove_checkpoint(struct journal_head *);
1043 void __jbd2_journal_insert_checkpoint(struct journal_head *, transaction_t *);
1044
1045
1046 /*
1047 * Triggers
1048 */
1049
1050 struct jbd2_buffer_trigger_type {
1051 /*
1052 * Fired just before a buffer is written to the journal.
1053 * mapped_data is a mapped buffer that is the frozen data for
1054 * commit.
1055 */
1056 void (*t_commit)(struct jbd2_buffer_trigger_type *type,
1057 struct buffer_head *bh, void *mapped_data,
1058 size_t size);
1059
1060 /*
1061 * Fired during journal abort for dirty buffers that will not be
1062 * committed.
1063 */
1064 void (*t_abort)(struct jbd2_buffer_trigger_type *type,
1065 struct buffer_head *bh);
1066 };
1067
1068 extern void jbd2_buffer_commit_trigger(struct journal_head *jh,
1069 void *mapped_data,
1070 struct jbd2_buffer_trigger_type *triggers);
1071 extern void jbd2_buffer_abort_trigger(struct journal_head *jh,
1072 struct jbd2_buffer_trigger_type *triggers);
1073
1074 /* Buffer IO */
1075 extern int
1076 jbd2_journal_write_metadata_buffer(transaction_t *transaction,
1077 struct journal_head *jh_in,
1078 struct journal_head **jh_out,
1079 unsigned long long blocknr);
1080
1081 /* Transaction locking */
1082 extern void __wait_on_journal (journal_t *);
1083
1084 /*
1085 * Journal locking.
1086 *
1087 * We need to lock the journal during transaction state changes so that nobody
1088 * ever tries to take a handle on the running transaction while we are in the
1089 * middle of moving it to the commit phase. j_state_lock does this.
1090 *
1091 * Note that the locking is completely interrupt unsafe. We never touch
1092 * journal structures from interrupts.
1093 */
1094
journal_current_handle(void)1095 static inline handle_t *journal_current_handle(void)
1096 {
1097 return current->journal_info;
1098 }
1099
1100 /* The journaling code user interface:
1101 *
1102 * Create and destroy handles
1103 * Register buffer modifications against the current transaction.
1104 */
1105
1106 extern handle_t *jbd2_journal_start(journal_t *, int nblocks);
1107 extern int jbd2_journal_restart (handle_t *, int nblocks);
1108 extern int jbd2_journal_extend (handle_t *, int nblocks);
1109 extern int jbd2_journal_get_write_access(handle_t *, struct buffer_head *);
1110 extern int jbd2_journal_get_create_access (handle_t *, struct buffer_head *);
1111 extern int jbd2_journal_get_undo_access(handle_t *, struct buffer_head *);
1112 void jbd2_journal_set_triggers(struct buffer_head *,
1113 struct jbd2_buffer_trigger_type *type);
1114 extern int jbd2_journal_dirty_metadata (handle_t *, struct buffer_head *);
1115 extern void jbd2_journal_release_buffer (handle_t *, struct buffer_head *);
1116 extern int jbd2_journal_forget (handle_t *, struct buffer_head *);
1117 extern void journal_sync_buffer (struct buffer_head *);
1118 extern void jbd2_journal_invalidatepage(journal_t *,
1119 struct page *, unsigned long);
1120 extern int jbd2_journal_try_to_free_buffers(journal_t *, struct page *, gfp_t);
1121 extern int jbd2_journal_stop(handle_t *);
1122 extern int jbd2_journal_flush (journal_t *);
1123 extern void jbd2_journal_lock_updates (journal_t *);
1124 extern void jbd2_journal_unlock_updates (journal_t *);
1125
1126 extern journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1127 struct block_device *fs_dev,
1128 unsigned long long start, int len, int bsize);
1129 extern journal_t * jbd2_journal_init_inode (struct inode *);
1130 extern int jbd2_journal_update_format (journal_t *);
1131 extern int jbd2_journal_check_used_features
1132 (journal_t *, unsigned long, unsigned long, unsigned long);
1133 extern int jbd2_journal_check_available_features
1134 (journal_t *, unsigned long, unsigned long, unsigned long);
1135 extern int jbd2_journal_set_features
1136 (journal_t *, unsigned long, unsigned long, unsigned long);
1137 extern void jbd2_journal_clear_features
1138 (journal_t *, unsigned long, unsigned long, unsigned long);
1139 extern int jbd2_journal_load (journal_t *journal);
1140 extern int jbd2_journal_destroy (journal_t *);
1141 extern int jbd2_journal_recover (journal_t *journal);
1142 extern int jbd2_journal_wipe (journal_t *, int);
1143 extern int jbd2_journal_skip_recovery (journal_t *);
1144 extern void jbd2_journal_update_superblock (journal_t *, int);
1145 extern void __jbd2_journal_abort_hard (journal_t *);
1146 extern void jbd2_journal_abort (journal_t *, int);
1147 extern int jbd2_journal_errno (journal_t *);
1148 extern void jbd2_journal_ack_err (journal_t *);
1149 extern int jbd2_journal_clear_err (journal_t *);
1150 extern int jbd2_journal_bmap(journal_t *, unsigned long, unsigned long long *);
1151 extern int jbd2_journal_force_commit(journal_t *);
1152 extern int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *inode);
1153 extern int jbd2_journal_begin_ordered_truncate(journal_t *journal,
1154 struct jbd2_inode *inode, loff_t new_size);
1155 extern void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode);
1156 extern void jbd2_journal_release_jbd_inode(journal_t *journal, struct jbd2_inode *jinode);
1157
1158 /*
1159 * journal_head management
1160 */
1161 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh);
1162 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh);
1163 void jbd2_journal_remove_journal_head(struct buffer_head *bh);
1164 void jbd2_journal_put_journal_head(struct journal_head *jh);
1165
1166 /*
1167 * handle management
1168 */
1169 extern struct kmem_cache *jbd2_handle_cache;
1170
jbd2_alloc_handle(gfp_t gfp_flags)1171 static inline handle_t *jbd2_alloc_handle(gfp_t gfp_flags)
1172 {
1173 return kmem_cache_alloc(jbd2_handle_cache, gfp_flags);
1174 }
1175
jbd2_free_handle(handle_t * handle)1176 static inline void jbd2_free_handle(handle_t *handle)
1177 {
1178 kmem_cache_free(jbd2_handle_cache, handle);
1179 }
1180
1181 /* Primary revoke support */
1182 #define JOURNAL_REVOKE_DEFAULT_HASH 256
1183 extern int jbd2_journal_init_revoke(journal_t *, int);
1184 extern void jbd2_journal_destroy_revoke_caches(void);
1185 extern int jbd2_journal_init_revoke_caches(void);
1186
1187 extern void jbd2_journal_destroy_revoke(journal_t *);
1188 extern int jbd2_journal_revoke (handle_t *, unsigned long long, struct buffer_head *);
1189 extern int jbd2_journal_cancel_revoke(handle_t *, struct journal_head *);
1190 extern void jbd2_journal_write_revoke_records(journal_t *, transaction_t *);
1191
1192 /* Recovery revoke support */
1193 extern int jbd2_journal_set_revoke(journal_t *, unsigned long long, tid_t);
1194 extern int jbd2_journal_test_revoke(journal_t *, unsigned long long, tid_t);
1195 extern void jbd2_journal_clear_revoke(journal_t *);
1196 extern void jbd2_journal_switch_revoke_table(journal_t *journal);
1197
1198 /*
1199 * The log thread user interface:
1200 *
1201 * Request space in the current transaction, and force transaction commit
1202 * transitions on demand.
1203 */
1204
1205 int __jbd2_log_space_left(journal_t *); /* Called with journal locked */
1206 int jbd2_log_start_commit(journal_t *journal, tid_t tid);
1207 int __jbd2_log_start_commit(journal_t *journal, tid_t tid);
1208 int jbd2_journal_start_commit(journal_t *journal, tid_t *tid);
1209 int jbd2_journal_force_commit_nested(journal_t *journal);
1210 int jbd2_log_wait_commit(journal_t *journal, tid_t tid);
1211 int jbd2_log_do_checkpoint(journal_t *journal);
1212
1213 void __jbd2_log_wait_for_space(journal_t *journal);
1214 extern void __jbd2_journal_drop_transaction(journal_t *, transaction_t *);
1215 extern int jbd2_cleanup_journal_tail(journal_t *);
1216
1217 /* Debugging code only: */
1218
1219 #define jbd_ENOSYS() \
1220 do { \
1221 printk (KERN_ERR "JBD unimplemented function %s\n", __func__); \
1222 current->state = TASK_UNINTERRUPTIBLE; \
1223 schedule(); \
1224 } while (1)
1225
1226 /*
1227 * is_journal_abort
1228 *
1229 * Simple test wrapper function to test the JBD2_ABORT state flag. This
1230 * bit, when set, indicates that we have had a fatal error somewhere,
1231 * either inside the journaling layer or indicated to us by the client
1232 * (eg. ext3), and that we and should not commit any further
1233 * transactions.
1234 */
1235
is_journal_aborted(journal_t * journal)1236 static inline int is_journal_aborted(journal_t *journal)
1237 {
1238 return journal->j_flags & JBD2_ABORT;
1239 }
1240
is_handle_aborted(handle_t * handle)1241 static inline int is_handle_aborted(handle_t *handle)
1242 {
1243 if (handle->h_aborted)
1244 return 1;
1245 return is_journal_aborted(handle->h_transaction->t_journal);
1246 }
1247
jbd2_journal_abort_handle(handle_t * handle)1248 static inline void jbd2_journal_abort_handle(handle_t *handle)
1249 {
1250 handle->h_aborted = 1;
1251 }
1252
1253 #endif /* __KERNEL__ */
1254
1255 /* Comparison functions for transaction IDs: perform comparisons using
1256 * modulo arithmetic so that they work over sequence number wraps. */
1257
tid_gt(tid_t x,tid_t y)1258 static inline int tid_gt(tid_t x, tid_t y)
1259 {
1260 int difference = (x - y);
1261 return (difference > 0);
1262 }
1263
tid_geq(tid_t x,tid_t y)1264 static inline int tid_geq(tid_t x, tid_t y)
1265 {
1266 int difference = (x - y);
1267 return (difference >= 0);
1268 }
1269
1270 extern int jbd2_journal_blocks_per_page(struct inode *inode);
1271 extern size_t journal_tag_bytes(journal_t *journal);
1272
1273 /*
1274 * Return the minimum number of blocks which must be free in the journal
1275 * before a new transaction may be started. Must be called under j_state_lock.
1276 */
jbd_space_needed(journal_t * journal)1277 static inline int jbd_space_needed(journal_t *journal)
1278 {
1279 int nblocks = journal->j_max_transaction_buffers;
1280 if (journal->j_committing_transaction)
1281 nblocks += journal->j_committing_transaction->
1282 t_outstanding_credits;
1283 return nblocks;
1284 }
1285
1286 /*
1287 * Definitions which augment the buffer_head layer
1288 */
1289
1290 /* journaling buffer types */
1291 #define BJ_None 0 /* Not journaled */
1292 #define BJ_Metadata 1 /* Normal journaled metadata */
1293 #define BJ_Forget 2 /* Buffer superseded by this transaction */
1294 #define BJ_IO 3 /* Buffer is for temporary IO use */
1295 #define BJ_Shadow 4 /* Buffer contents being shadowed to the log */
1296 #define BJ_LogCtl 5 /* Buffer contains log descriptors */
1297 #define BJ_Reserved 6 /* Buffer is reserved for access by journal */
1298 #define BJ_Types 7
1299
1300 extern int jbd_blocks_per_page(struct inode *inode);
1301
1302 #ifdef __KERNEL__
1303
1304 #define buffer_trace_init(bh) do {} while (0)
1305 #define print_buffer_fields(bh) do {} while (0)
1306 #define print_buffer_trace(bh) do {} while (0)
1307 #define BUFFER_TRACE(bh, info) do {} while (0)
1308 #define BUFFER_TRACE2(bh, bh2, info) do {} while (0)
1309 #define JBUFFER_TRACE(jh, info) do {} while (0)
1310
1311 #endif /* __KERNEL__ */
1312
1313 #endif /* _LINUX_JBD2_H */
1314