blob: d9ecd13393b0859ffa98034ec9cb77a3bd8f1912 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/include/linux/jbd.h
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 * 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_JBD_H
17#define _LINUX_JBD_H
18
Linus Torvalds1da177e2005-04-16 15:20:36 -070019/* Allow this file to be included directly into e2fsprogs */
20#ifndef __KERNEL__
21#include "jfs_compat.h"
22#define JFS_DEBUG
23#define jfs_debug jbd_debug
24#else
25
Andrew Mortonfe1dcbc2006-02-04 23:27:54 -080026#include <linux/types.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/buffer_head.h>
28#include <linux/journal-head.h>
29#include <linux/stddef.h>
Ingo Molnarfb1c8f92005-09-10 00:25:56 -070030#include <linux/bit_spinlock.h>
Arjan van de Ven2c68ee72006-03-23 03:00:35 -080031#include <linux/mutex.h>
Andrew Mortone3df1892006-03-25 03:06:53 -080032#include <linux/timer.h>
Peter Zijlstra34a3d1e2007-10-11 22:11:12 +020033#include <linux/lockdep.h>
Andrew Mortone3df1892006-03-25 03:06:53 -080034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/semaphore.h>
36#endif
37
38#define journal_oom_retry 1
39
40/*
41 * Define JBD_PARANIOD_IOFAIL to cause a kernel BUG() if ext3 finds
42 * certain classes of error which can occur due to failed IOs. Under
43 * normal use we want ext3 to continue after such errors, because
44 * hardware _can_ fail, but for debugging purposes when running tests on
45 * known-good hardware we may want to trap these errors.
46 */
47#undef JBD_PARANOID_IOFAIL
48
49/*
50 * The default maximum commit age, in seconds.
51 */
52#define JBD_DEFAULT_MAX_COMMIT_AGE 5
53
54#ifdef CONFIG_JBD_DEBUG
55/*
56 * Define JBD_EXPENSIVE_CHECKING to enable more expensive internal
57 * consistency checks. By default we don't do this unless
58 * CONFIG_JBD_DEBUG is on.
59 */
60#define JBD_EXPENSIVE_CHECKING
Jose R. Santosc2a91592007-10-18 23:39:22 -070061extern u8 journal_enable_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63#define jbd_debug(n, f, a...) \
64 do { \
65 if ((n) <= journal_enable_debug) { \
66 printk (KERN_DEBUG "(%s, %d): %s: ", \
67 __FILE__, __LINE__, __FUNCTION__); \
Dave Kleikampe9ad5622006-09-27 01:49:35 -070068 printk (f, ## a); \
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 } \
70 } while (0)
71#else
72#define jbd_debug(f, a...) /**/
73#endif
74
Mingming Caoc089d492007-10-16 18:38:25 -040075static inline void *jbd_alloc(size_t size, gfp_t flags)
76{
77 return (void *)__get_free_pages(flags, get_order(size));
78}
79
80static inline void jbd_free(void *ptr, size_t size)
81{
82 free_pages((unsigned long)ptr, get_order(size));
83};
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#define JFS_MIN_JOURNAL_BLOCKS 1024
86
87#ifdef __KERNEL__
88
89/**
90 * typedef handle_t - The handle_t type represents a single atomic update being performed by some process.
91 *
92 * All filesystem modifications made by the process go
93 * through this handle. Recursive operations (such as quota operations)
94 * are gathered into a single update.
95 *
96 * The buffer credits field is used to account for journaled buffers
97 * being modified by the running process. To ensure that there is
98 * enough log space for all outstanding operations, we need to limit the
99 * number of outstanding buffers possible at any time. When the
100 * operation completes, any buffer credits not used are credited back to
101 * the transaction, so that at all times we know how many buffers the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700102 * outstanding updates on a transaction might possibly touch.
103 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 * This is an opaque datatype.
105 **/
106typedef struct handle_s handle_t; /* Atomic operation type */
107
108
109/**
110 * typedef journal_t - The journal_t maintains all of the journaling state information for a single filesystem.
111 *
112 * journal_t is linked to from the fs superblock structure.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700113 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * We use the journal_t to keep track of all outstanding transaction
115 * activity on the filesystem, and to manage the state of the log
116 * writing process.
117 *
118 * This is an opaque datatype.
119 **/
120typedef struct journal_s journal_t; /* Journal control structure */
121#endif
122
123/*
124 * Internal structures used by the logging mechanism:
125 */
126
127#define JFS_MAGIC_NUMBER 0xc03b3998U /* The first 4 bytes of /dev/random! */
128
129/*
130 * On-disk structures
131 */
132
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700133/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * Descriptor block types:
135 */
136
137#define JFS_DESCRIPTOR_BLOCK 1
138#define JFS_COMMIT_BLOCK 2
139#define JFS_SUPERBLOCK_V1 3
140#define JFS_SUPERBLOCK_V2 4
141#define JFS_REVOKE_BLOCK 5
142
143/*
144 * Standard header for all descriptor blocks:
145 */
146typedef struct journal_header_s
147{
148 __be32 h_magic;
149 __be32 h_blocktype;
150 __be32 h_sequence;
151} journal_header_t;
152
153
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700154/*
155 * The block tag: used to describe a single buffer in the journal
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 */
157typedef struct journal_block_tag_s
158{
159 __be32 t_blocknr; /* The on-disk block number */
160 __be32 t_flags; /* See below */
161} journal_block_tag_t;
162
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700163/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 * The revoke descriptor: used on disk to describe a series of blocks to
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700165 * be revoked from the log
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 */
167typedef struct journal_revoke_header_s
168{
169 journal_header_t r_header;
170 __be32 r_count; /* Count of bytes used in the block */
171} journal_revoke_header_t;
172
173
174/* Definitions for the journal tag flags word: */
175#define JFS_FLAG_ESCAPE 1 /* on-disk block is escaped */
176#define JFS_FLAG_SAME_UUID 2 /* block has same uuid as previous */
177#define JFS_FLAG_DELETED 4 /* block deleted by this transaction */
178#define JFS_FLAG_LAST_TAG 8 /* last tag in this descriptor block */
179
180
181/*
182 * The journal superblock. All fields are in big-endian byte order.
183 */
184typedef struct journal_superblock_s
185{
186/* 0x0000 */
187 journal_header_t s_header;
188
189/* 0x000C */
190 /* Static information describing the journal */
191 __be32 s_blocksize; /* journal device blocksize */
192 __be32 s_maxlen; /* total blocks in journal file */
193 __be32 s_first; /* first block of log information */
194
195/* 0x0018 */
196 /* Dynamic information describing the current state of the log */
197 __be32 s_sequence; /* first commit ID expected in log */
198 __be32 s_start; /* blocknr of start of log */
199
200/* 0x0020 */
201 /* Error value, as set by journal_abort(). */
202 __be32 s_errno;
203
204/* 0x0024 */
205 /* Remaining fields are only valid in a version-2 superblock */
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700206 __be32 s_feature_compat; /* compatible feature set */
207 __be32 s_feature_incompat; /* incompatible feature set */
208 __be32 s_feature_ro_compat; /* readonly-compatible feature set */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209/* 0x0030 */
210 __u8 s_uuid[16]; /* 128-bit uuid for journal */
211
212/* 0x0040 */
213 __be32 s_nr_users; /* Nr of filesystems sharing log */
214
215 __be32 s_dynsuper; /* Blocknr of dynamic superblock copy*/
216
217/* 0x0048 */
218 __be32 s_max_transaction; /* Limit of journal blocks per trans.*/
219 __be32 s_max_trans_data; /* Limit of data blocks per trans. */
220
221/* 0x0050 */
222 __u32 s_padding[44];
223
224/* 0x0100 */
225 __u8 s_users[16*48]; /* ids of all fs'es sharing the log */
226/* 0x0400 */
227} journal_superblock_t;
228
229#define JFS_HAS_COMPAT_FEATURE(j,mask) \
230 ((j)->j_format_version >= 2 && \
231 ((j)->j_superblock->s_feature_compat & cpu_to_be32((mask))))
232#define JFS_HAS_RO_COMPAT_FEATURE(j,mask) \
233 ((j)->j_format_version >= 2 && \
234 ((j)->j_superblock->s_feature_ro_compat & cpu_to_be32((mask))))
235#define JFS_HAS_INCOMPAT_FEATURE(j,mask) \
236 ((j)->j_format_version >= 2 && \
237 ((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask))))
238
239#define JFS_FEATURE_INCOMPAT_REVOKE 0x00000001
240
241/* Features known to this kernel version: */
242#define JFS_KNOWN_COMPAT_FEATURES 0
243#define JFS_KNOWN_ROCOMPAT_FEATURES 0
244#define JFS_KNOWN_INCOMPAT_FEATURES JFS_FEATURE_INCOMPAT_REVOKE
245
246#ifdef __KERNEL__
247
248#include <linux/fs.h>
249#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Chris Snook1c099242007-10-18 23:39:21 -0700251#define J_ASSERT(assert) BUG_ON(!(assert))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253#if defined(CONFIG_BUFFER_DEBUG)
254void buffer_assertion_failure(struct buffer_head *bh);
255#define J_ASSERT_BH(bh, expr) \
256 do { \
257 if (!(expr)) \
258 buffer_assertion_failure(bh); \
259 J_ASSERT(expr); \
260 } while (0)
261#define J_ASSERT_JH(jh, expr) J_ASSERT_BH(jh2bh(jh), expr)
262#else
263#define J_ASSERT_BH(bh, expr) J_ASSERT(expr)
264#define J_ASSERT_JH(jh, expr) J_ASSERT(expr)
265#endif
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267#if defined(JBD_PARANOID_IOFAIL)
268#define J_EXPECT(expr, why...) J_ASSERT(expr)
269#define J_EXPECT_BH(bh, expr, why...) J_ASSERT_BH(bh, expr)
270#define J_EXPECT_JH(jh, expr, why...) J_ASSERT_JH(jh, expr)
271#else
272#define __journal_expect(expr, why...) \
273 ({ \
274 int val = (expr); \
275 if (!val) { \
276 printk(KERN_ERR \
277 "EXT3-fs unexpected failure: %s;\n",# expr); \
278 printk(KERN_ERR why "\n"); \
279 } \
280 val; \
281 })
282#define J_EXPECT(expr, why...) __journal_expect(expr, ## why)
283#define J_EXPECT_BH(bh, expr, why...) __journal_expect(expr, ## why)
284#define J_EXPECT_JH(jh, expr, why...) __journal_expect(expr, ## why)
285#endif
286
287enum jbd_state_bits {
288 BH_JBD /* Has an attached ext3 journal_head */
289 = BH_PrivateStart,
290 BH_JWrite, /* Being written to log (@@@ DEBUGGING) */
291 BH_Freed, /* Has been freed (truncated) */
292 BH_Revoked, /* Has been revoked from the log */
293 BH_RevokeValid, /* Revoked flag is valid */
294 BH_JBDDirty, /* Is dirty but journaled */
295 BH_State, /* Pins most journal_head state */
296 BH_JournalHead, /* Pins bh->b_private and jh->b_bh */
297 BH_Unshadow, /* Dummy bit, for BJ_Shadow wakeup filtering */
298};
299
300BUFFER_FNS(JBD, jbd)
301BUFFER_FNS(JWrite, jwrite)
302BUFFER_FNS(JBDDirty, jbddirty)
303TAS_BUFFER_FNS(JBDDirty, jbddirty)
304BUFFER_FNS(Revoked, revoked)
305TAS_BUFFER_FNS(Revoked, revoked)
306BUFFER_FNS(RevokeValid, revokevalid)
307TAS_BUFFER_FNS(RevokeValid, revokevalid)
308BUFFER_FNS(Freed, freed)
309
310static inline struct buffer_head *jh2bh(struct journal_head *jh)
311{
312 return jh->b_bh;
313}
314
315static inline struct journal_head *bh2jh(struct buffer_head *bh)
316{
317 return bh->b_private;
318}
319
320static inline void jbd_lock_bh_state(struct buffer_head *bh)
321{
322 bit_spin_lock(BH_State, &bh->b_state);
323}
324
325static inline int jbd_trylock_bh_state(struct buffer_head *bh)
326{
327 return bit_spin_trylock(BH_State, &bh->b_state);
328}
329
330static inline int jbd_is_locked_bh_state(struct buffer_head *bh)
331{
332 return bit_spin_is_locked(BH_State, &bh->b_state);
333}
334
335static inline void jbd_unlock_bh_state(struct buffer_head *bh)
336{
337 bit_spin_unlock(BH_State, &bh->b_state);
338}
339
340static inline void jbd_lock_bh_journal_head(struct buffer_head *bh)
341{
342 bit_spin_lock(BH_JournalHead, &bh->b_state);
343}
344
345static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh)
346{
347 bit_spin_unlock(BH_JournalHead, &bh->b_state);
348}
349
350struct jbd_revoke_table_s;
351
352/**
353 * struct handle_s - The handle_s type is the concrete type associated with
354 * handle_t.
355 * @h_transaction: Which compound transaction is this update a part of?
356 * @h_buffer_credits: Number of remaining buffers we are allowed to dirty.
357 * @h_ref: Reference count on this handle
358 * @h_err: Field for caller's use to track errors through large fs operations
359 * @h_sync: flag for sync-on-close
360 * @h_jdata: flag to force data journaling
361 * @h_aborted: flag indicating fatal error on handle
Randy Dunlape6716b82007-10-15 17:30:19 -0700362 * @h_lockdep_map: lockdep info for debugging lock problems
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 **/
364
365/* Docbook can't yet cope with the bit fields, but will leave the documentation
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700366 * in so it can be fixed later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 */
368
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700369struct handle_s
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
371 /* Which compound transaction is this update a part of? */
372 transaction_t *h_transaction;
373
374 /* Number of remaining buffers we are allowed to dirty: */
375 int h_buffer_credits;
376
377 /* Reference count on this handle */
378 int h_ref;
379
380 /* Field for caller's use to track errors through large fs */
381 /* operations */
382 int h_err;
383
384 /* Flags [no locking] */
385 unsigned int h_sync: 1; /* sync-on-close */
386 unsigned int h_jdata: 1; /* force data journaling */
387 unsigned int h_aborted: 1; /* fatal error on handle */
Peter Zijlstra34a3d1e2007-10-11 22:11:12 +0200388
389#ifdef CONFIG_DEBUG_LOCK_ALLOC
390 struct lockdep_map h_lockdep_map;
391#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392};
393
394
395/* The transaction_t type is the guts of the journaling mechanism. It
396 * tracks a compound transaction through its various states:
397 *
398 * RUNNING: accepting new updates
399 * LOCKED: Updates still running but we don't accept new ones
400 * RUNDOWN: Updates are tidying up but have finished requesting
401 * new buffers to modify (state not used for now)
402 * FLUSH: All updates complete, but we are still writing to disk
403 * COMMIT: All data on disk, writing commit record
404 * FINISHED: We still have to keep the transaction for checkpointing.
405 *
406 * The transaction keeps track of all of the buffers modified by a
407 * running transaction, and all of the buffers committed but not yet
408 * flushed to home for finished transactions.
409 */
410
411/*
412 * Lock ranking:
413 *
414 * j_list_lock
415 * ->jbd_lock_bh_journal_head() (This is "innermost")
416 *
417 * j_state_lock
418 * ->jbd_lock_bh_state()
419 *
420 * jbd_lock_bh_state()
421 * ->j_list_lock
422 *
423 * j_state_lock
424 * ->t_handle_lock
425 *
426 * j_state_lock
427 * ->j_list_lock (journal_unmap_buffer)
428 *
429 */
430
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700431struct transaction_s
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 /* Pointer to the journal for this transaction. [no locking] */
434 journal_t *t_journal;
435
436 /* Sequence number for this transaction [no locking] */
437 tid_t t_tid;
438
439 /*
440 * Transaction's current state
441 * [no locking - only kjournald alters this]
Jan Karad4beaf42007-12-04 23:45:27 -0800442 * [j_list_lock] guards transition of a transaction into T_FINISHED
443 * state and subsequent call of __journal_drop_transaction()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * FIXME: needs barriers
445 * KLUDGE: [use j_state_lock]
446 */
447 enum {
448 T_RUNNING,
449 T_LOCKED,
450 T_RUNDOWN,
451 T_FLUSH,
452 T_COMMIT,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700453 T_FINISHED
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 } t_state;
455
456 /*
457 * Where in the log does this transaction's commit start? [no locking]
458 */
459 unsigned long t_log_start;
460
461 /* Number of buffers on the t_buffers list [j_list_lock] */
462 int t_nr_buffers;
463
464 /*
465 * Doubly-linked circular list of all buffers reserved but not yet
466 * modified by this transaction [j_list_lock]
467 */
468 struct journal_head *t_reserved_list;
469
470 /*
471 * Doubly-linked circular list of all buffers under writeout during
472 * commit [j_list_lock]
473 */
474 struct journal_head *t_locked_list;
475
476 /*
477 * Doubly-linked circular list of all metadata buffers owned by this
478 * transaction [j_list_lock]
479 */
480 struct journal_head *t_buffers;
481
482 /*
483 * Doubly-linked circular list of all data buffers still to be
484 * flushed before this transaction can be committed [j_list_lock]
485 */
486 struct journal_head *t_sync_datalist;
487
488 /*
489 * Doubly-linked circular list of all forget buffers (superseded
490 * buffers which we can un-checkpoint once this transaction commits)
491 * [j_list_lock]
492 */
493 struct journal_head *t_forget;
494
495 /*
496 * Doubly-linked circular list of all buffers still to be flushed before
497 * this transaction can be checkpointed. [j_list_lock]
498 */
499 struct journal_head *t_checkpoint_list;
500
501 /*
Jan Kara78ce89c2006-06-23 02:06:05 -0700502 * Doubly-linked circular list of all buffers submitted for IO while
503 * checkpointing. [j_list_lock]
504 */
505 struct journal_head *t_checkpoint_io_list;
506
507 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 * Doubly-linked circular list of temporary buffers currently undergoing
509 * IO in the log [j_list_lock]
510 */
511 struct journal_head *t_iobuf_list;
512
513 /*
514 * Doubly-linked circular list of metadata buffers being shadowed by log
515 * IO. The IO buffers on the iobuf list and the shadow buffers on this
516 * list match each other one for one at all times. [j_list_lock]
517 */
518 struct journal_head *t_shadow_list;
519
520 /*
521 * Doubly-linked circular list of control buffers being written to the
522 * log. [j_list_lock]
523 */
524 struct journal_head *t_log_list;
525
526 /*
527 * Protects info related to handles
528 */
529 spinlock_t t_handle_lock;
530
531 /*
532 * Number of outstanding updates running on this transaction
533 * [t_handle_lock]
534 */
535 int t_updates;
536
537 /*
538 * Number of buffers reserved for use by all handles in this transaction
539 * handle but not yet modified. [t_handle_lock]
540 */
541 int t_outstanding_credits;
542
543 /*
544 * Forward and backward links for the circular list of all transactions
545 * awaiting checkpoint. [j_list_lock]
546 */
547 transaction_t *t_cpnext, *t_cpprev;
548
549 /*
550 * When will the transaction expire (become due for commit), in jiffies?
551 * [no locking]
552 */
553 unsigned long t_expires;
554
555 /*
556 * How many handles used this transaction? [t_handle_lock]
557 */
558 int t_handle_count;
559
560};
561
562/**
563 * struct journal_s - The journal_s type is the concrete type associated with
564 * journal_t.
565 * @j_flags: General journaling state flags
566 * @j_errno: Is there an outstanding uncleared error on the journal (from a
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700567 * prior abort)?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 * @j_sb_buffer: First part of superblock buffer
569 * @j_superblock: Second part of superblock buffer
570 * @j_format_version: Version of the superblock format
571 * @j_state_lock: Protect the various scalars in the journal
572 * @j_barrier_count: Number of processes waiting to create a barrier lock
573 * @j_barrier: The barrier lock itself
574 * @j_running_transaction: The current running transaction..
575 * @j_committing_transaction: the transaction we are pushing to disk
576 * @j_checkpoint_transactions: a linked circular list of all transactions
577 * waiting for checkpointing
578 * @j_wait_transaction_locked: Wait queue for waiting for a locked transaction
579 * to start committing, or for a barrier lock to be released
580 * @j_wait_logspace: Wait queue for waiting for checkpointing to complete
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700581 * @j_wait_done_commit: Wait queue for waiting for commit to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * @j_wait_checkpoint: Wait queue to trigger checkpointing
583 * @j_wait_commit: Wait queue to trigger commit
584 * @j_wait_updates: Wait queue to wait for updates to complete
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800585 * @j_checkpoint_mutex: Mutex for locking against concurrent checkpoints
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 * @j_head: Journal head - identifies the first unused block in the journal
587 * @j_tail: Journal tail - identifies the oldest still-used block in the
588 * journal.
589 * @j_free: Journal free - how many free blocks are there in the journal?
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700590 * @j_first: The block number of the first usable block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 * @j_last: The block number one beyond the last usable block
592 * @j_dev: Device where we store the journal
593 * @j_blocksize: blocksize for the location where we store the journal.
594 * @j_blk_offset: starting block offset for into the device where we store the
595 * journal
596 * @j_fs_dev: Device which holds the client fs. For internal journal this will
597 * be equal to j_dev
598 * @j_maxlen: Total maximum capacity of the journal region on disk.
599 * @j_list_lock: Protects the buffer lists and internal buffer state.
600 * @j_inode: Optional inode where we store the journal. If present, all journal
601 * block numbers are mapped into this inode via bmap().
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700602 * @j_tail_sequence: Sequence number of the oldest transaction in the log
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 * @j_transaction_sequence: Sequence number of the next transaction to grant
604 * @j_commit_sequence: Sequence number of the most recently committed
605 * transaction
606 * @j_commit_request: Sequence number of the most recent transaction wanting
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700607 * commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 * @j_uuid: Uuid of client object.
609 * @j_task: Pointer to the current commit thread for this journal
610 * @j_max_transaction_buffers: Maximum number of metadata buffers to allow in a
611 * single compound commit transaction
612 * @j_commit_interval: What is the maximum transaction lifetime before we begin
613 * a commit?
614 * @j_commit_timer: The timer used to wakeup the commit thread
615 * @j_revoke_lock: Protect the revoke table
616 * @j_revoke: The revoke table - maintains the list of revoked blocks in the
617 * current transaction.
618 * @j_revoke_table: alternate revoke tables for j_revoke
Randy Dunlap6c8bec62005-11-07 01:01:04 -0800619 * @j_wbuf: array of buffer_heads for journal_commit_transaction
620 * @j_wbufsize: maximum number of buffer_heads allowed in j_wbuf, the
621 * number that will fit in j_blocksize
Andrew Mortonfe1dcbc2006-02-04 23:27:54 -0800622 * @j_last_sync_writer: most recent pid which did a synchronous write
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 * @j_private: An opaque pointer to fs-private information.
624 */
625
626struct journal_s
627{
628 /* General journaling state flags [j_state_lock] */
629 unsigned long j_flags;
630
631 /*
632 * Is there an outstanding uncleared error on the journal (from a prior
633 * abort)? [j_state_lock]
634 */
635 int j_errno;
636
637 /* The superblock buffer */
638 struct buffer_head *j_sb_buffer;
639 journal_superblock_t *j_superblock;
640
641 /* Version of the superblock format */
642 int j_format_version;
643
644 /*
645 * Protect the various scalars in the journal
646 */
647 spinlock_t j_state_lock;
648
649 /*
650 * Number of processes waiting to create a barrier lock [j_state_lock]
651 */
652 int j_barrier_count;
653
654 /* The barrier lock itself */
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800655 struct mutex j_barrier;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 /*
658 * Transactions: The current running transaction...
659 * [j_state_lock] [caller holding open handle]
660 */
661 transaction_t *j_running_transaction;
662
663 /*
664 * the transaction we are pushing to disk
665 * [j_state_lock] [caller holding open handle]
666 */
667 transaction_t *j_committing_transaction;
668
669 /*
670 * ... and a linked circular list of all transactions waiting for
671 * checkpointing. [j_list_lock]
672 */
673 transaction_t *j_checkpoint_transactions;
674
675 /*
676 * Wait queue for waiting for a locked transaction to start committing,
677 * or for a barrier lock to be released
678 */
679 wait_queue_head_t j_wait_transaction_locked;
680
681 /* Wait queue for waiting for checkpointing to complete */
682 wait_queue_head_t j_wait_logspace;
683
684 /* Wait queue for waiting for commit to complete */
685 wait_queue_head_t j_wait_done_commit;
686
687 /* Wait queue to trigger checkpointing */
688 wait_queue_head_t j_wait_checkpoint;
689
690 /* Wait queue to trigger commit */
691 wait_queue_head_t j_wait_commit;
692
693 /* Wait queue to wait for updates to complete */
694 wait_queue_head_t j_wait_updates;
695
696 /* Semaphore for locking against concurrent checkpoints */
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700697 struct mutex j_checkpoint_mutex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699 /*
700 * Journal head: identifies the first unused block in the journal.
701 * [j_state_lock]
702 */
703 unsigned long j_head;
704
705 /*
706 * Journal tail: identifies the oldest still-used block in the journal.
707 * [j_state_lock]
708 */
709 unsigned long j_tail;
710
711 /*
712 * Journal free: how many free blocks are there in the journal?
713 * [j_state_lock]
714 */
715 unsigned long j_free;
716
717 /*
718 * Journal start and end: the block numbers of the first usable block
719 * and one beyond the last usable block in the journal. [j_state_lock]
720 */
721 unsigned long j_first;
722 unsigned long j_last;
723
724 /*
725 * Device, blocksize and starting block offset for the location where we
726 * store the journal.
727 */
728 struct block_device *j_dev;
729 int j_blocksize;
Eric Sandeen37ed3222006-09-27 01:49:31 -0700730 unsigned long j_blk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
732 /*
733 * Device which holds the client fs. For internal journal this will be
734 * equal to j_dev.
735 */
736 struct block_device *j_fs_dev;
737
738 /* Total maximum capacity of the journal region on disk. */
739 unsigned int j_maxlen;
740
741 /*
742 * Protects the buffer lists and internal buffer state.
743 */
744 spinlock_t j_list_lock;
745
746 /* Optional inode where we store the journal. If present, all */
747 /* journal block numbers are mapped into this inode via */
748 /* bmap(). */
749 struct inode *j_inode;
750
751 /*
752 * Sequence number of the oldest transaction in the log [j_state_lock]
753 */
754 tid_t j_tail_sequence;
755
756 /*
757 * Sequence number of the next transaction to grant [j_state_lock]
758 */
759 tid_t j_transaction_sequence;
760
761 /*
762 * Sequence number of the most recently committed transaction
763 * [j_state_lock].
764 */
765 tid_t j_commit_sequence;
766
767 /*
768 * Sequence number of the most recent transaction wanting commit
769 * [j_state_lock]
770 */
771 tid_t j_commit_request;
772
773 /*
774 * Journal uuid: identifies the object (filesystem, LVM volume etc)
775 * backed by this journal. This will eventually be replaced by an array
776 * of uuids, allowing us to index multiple devices within a single
777 * journal and to perform atomic updates across them.
778 */
779 __u8 j_uuid[16];
780
781 /* Pointer to the current commit thread for this journal */
782 struct task_struct *j_task;
783
784 /*
785 * Maximum number of metadata buffers to allow in a single compound
786 * commit transaction
787 */
788 int j_max_transaction_buffers;
789
790 /*
791 * What is the maximum transaction lifetime before we begin a commit?
792 */
793 unsigned long j_commit_interval;
794
795 /* The timer used to wakeup the commit thread: */
Andrew Mortone3df1892006-03-25 03:06:53 -0800796 struct timer_list j_commit_timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 /*
799 * The revoke table: maintains the list of revoked blocks in the
800 * current transaction. [j_revoke_lock]
801 */
802 spinlock_t j_revoke_lock;
803 struct jbd_revoke_table_s *j_revoke;
804 struct jbd_revoke_table_s *j_revoke_table[2];
805
806 /*
807 * array of bhs for journal_commit_transaction
808 */
809 struct buffer_head **j_wbuf;
810 int j_wbufsize;
811
Andrew Mortonfe1dcbc2006-02-04 23:27:54 -0800812 pid_t j_last_sync_writer;
813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 /*
815 * An opaque pointer to fs-private information. ext3 puts its
816 * superblock pointer here
817 */
818 void *j_private;
819};
820
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700821/*
822 * Journal flag definitions
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 */
824#define JFS_UNMOUNT 0x001 /* Journal thread is being destroyed */
825#define JFS_ABORT 0x002 /* Journaling has been aborted for errors. */
826#define JFS_ACK_ERR 0x004 /* The errno in the sb has been acked */
827#define JFS_FLUSHED 0x008 /* The journal superblock has been flushed */
828#define JFS_LOADED 0x010 /* The journal superblock has been loaded */
829#define JFS_BARRIER 0x020 /* Use IDE barriers */
830
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700831/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 * Function declarations for the journaling transaction and buffer
833 * management
834 */
835
836/* Filing buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837extern void journal_unfile_buffer(journal_t *, struct journal_head *);
838extern void __journal_unfile_buffer(struct journal_head *);
839extern void __journal_refile_buffer(struct journal_head *);
840extern void journal_refile_buffer(journal_t *, struct journal_head *);
841extern void __journal_file_buffer(struct journal_head *, transaction_t *, int);
842extern void __journal_free_buffer(struct journal_head *bh);
843extern void journal_file_buffer(struct journal_head *, transaction_t *, int);
844extern void __journal_clean_data_list(transaction_t *transaction);
845
846/* Log buffer allocation */
847extern struct journal_head * journal_get_descriptor_buffer(journal_t *);
848int journal_next_log_block(journal_t *, unsigned long *);
849
850/* Commit management */
851extern void journal_commit_transaction(journal_t *);
852
853/* Checkpoint list management */
854int __journal_clean_checkpoint_list(journal_t *journal);
Jan Kara78ce89c2006-06-23 02:06:05 -0700855int __journal_remove_checkpoint(struct journal_head *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856void __journal_insert_checkpoint(struct journal_head *, transaction_t *);
857
858/* Buffer IO */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700859extern int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860journal_write_metadata_buffer(transaction_t *transaction,
861 struct journal_head *jh_in,
862 struct journal_head **jh_out,
Eric Sandeen37ed3222006-09-27 01:49:31 -0700863 unsigned long blocknr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865/* Transaction locking */
866extern void __wait_on_journal (journal_t *);
867
868/*
869 * Journal locking.
870 *
871 * We need to lock the journal during transaction state changes so that nobody
872 * ever tries to take a handle on the running transaction while we are in the
873 * middle of moving it to the commit phase. j_state_lock does this.
874 *
875 * Note that the locking is completely interrupt unsafe. We never touch
876 * journal structures from interrupts.
877 */
878
879static inline handle_t *journal_current_handle(void)
880{
881 return current->journal_info;
882}
883
884/* The journaling code user interface:
885 *
886 * Create and destroy handles
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700887 * Register buffer modifications against the current transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 */
889
890extern handle_t *journal_start(journal_t *, int nblocks);
891extern int journal_restart (handle_t *, int nblocks);
892extern int journal_extend (handle_t *, int nblocks);
893extern int journal_get_write_access(handle_t *, struct buffer_head *);
894extern int journal_get_create_access (handle_t *, struct buffer_head *);
895extern int journal_get_undo_access(handle_t *, struct buffer_head *);
896extern int journal_dirty_data (handle_t *, struct buffer_head *);
897extern int journal_dirty_metadata (handle_t *, struct buffer_head *);
898extern void journal_release_buffer (handle_t *, struct buffer_head *);
899extern int journal_forget (handle_t *, struct buffer_head *);
900extern void journal_sync_buffer (struct buffer_head *);
NeilBrown2ff28e22006-03-26 01:37:18 -0800901extern void journal_invalidatepage(journal_t *,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 struct page *, unsigned long);
Al Viro27496a82005-10-21 03:20:48 -0400903extern int journal_try_to_free_buffers(journal_t *, struct page *, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904extern int journal_stop(handle_t *);
905extern int journal_flush (journal_t *);
906extern void journal_lock_updates (journal_t *);
907extern void journal_unlock_updates (journal_t *);
908
909extern journal_t * journal_init_dev(struct block_device *bdev,
910 struct block_device *fs_dev,
911 int start, int len, int bsize);
912extern journal_t * journal_init_inode (struct inode *);
913extern int journal_update_format (journal_t *);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700914extern int journal_check_used_features
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 (journal_t *, unsigned long, unsigned long, unsigned long);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700916extern int journal_check_available_features
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 (journal_t *, unsigned long, unsigned long, unsigned long);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700918extern int journal_set_features
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 (journal_t *, unsigned long, unsigned long, unsigned long);
920extern int journal_create (journal_t *);
921extern int journal_load (journal_t *journal);
922extern void journal_destroy (journal_t *);
923extern int journal_recover (journal_t *journal);
924extern int journal_wipe (journal_t *, int);
925extern int journal_skip_recovery (journal_t *);
926extern void journal_update_superblock (journal_t *, int);
927extern void __journal_abort_hard (journal_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928extern void journal_abort (journal_t *, int);
929extern int journal_errno (journal_t *);
930extern void journal_ack_err (journal_t *);
931extern int journal_clear_err (journal_t *);
932extern int journal_bmap(journal_t *, unsigned long, unsigned long *);
933extern int journal_force_commit(journal_t *);
934
935/*
936 * journal_head management
937 */
938struct journal_head *journal_add_journal_head(struct buffer_head *bh);
939struct journal_head *journal_grab_journal_head(struct buffer_head *bh);
940void journal_remove_journal_head(struct buffer_head *bh);
941void journal_put_journal_head(struct journal_head *jh);
942
943/*
944 * handle management
945 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800946extern struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Al Virodd0fc662005-10-07 07:46:04 +0100948static inline handle_t *jbd_alloc_handle(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
950 return kmem_cache_alloc(jbd_handle_cache, gfp_flags);
951}
952
953static inline void jbd_free_handle(handle_t *handle)
954{
955 kmem_cache_free(jbd_handle_cache, handle);
956}
957
958/* Primary revoke support */
959#define JOURNAL_REVOKE_DEFAULT_HASH 256
960extern int journal_init_revoke(journal_t *, int);
961extern void journal_destroy_revoke_caches(void);
962extern int journal_init_revoke_caches(void);
963
964extern void journal_destroy_revoke(journal_t *);
965extern int journal_revoke (handle_t *,
966 unsigned long, struct buffer_head *);
967extern int journal_cancel_revoke(handle_t *, struct journal_head *);
968extern void journal_write_revoke_records(journal_t *, transaction_t *);
969
970/* Recovery revoke support */
971extern int journal_set_revoke(journal_t *, unsigned long, tid_t);
972extern int journal_test_revoke(journal_t *, unsigned long, tid_t);
973extern void journal_clear_revoke(journal_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974extern void journal_switch_revoke_table(journal_t *journal);
975
976/*
977 * The log thread user interface:
978 *
979 * Request space in the current transaction, and force transaction commit
980 * transitions on demand.
981 */
982
983int __log_space_left(journal_t *); /* Called with journal locked */
984int log_start_commit(journal_t *journal, tid_t tid);
985int __log_start_commit(journal_t *journal, tid_t tid);
986int journal_start_commit(journal_t *journal, tid_t *tid);
987int journal_force_commit_nested(journal_t *journal);
988int log_wait_commit(journal_t *journal, tid_t tid);
989int log_do_checkpoint(journal_t *journal);
990
991void __log_wait_for_space(journal_t *journal);
992extern void __journal_drop_transaction(journal_t *, transaction_t *);
993extern int cleanup_journal_tail(journal_t *);
994
995/* Debugging code only: */
996
997#define jbd_ENOSYS() \
998do { \
999 printk (KERN_ERR "JBD unimplemented function %s\n", __FUNCTION__); \
1000 current->state = TASK_UNINTERRUPTIBLE; \
1001 schedule(); \
1002} while (1)
1003
1004/*
1005 * is_journal_abort
1006 *
1007 * Simple test wrapper function to test the JFS_ABORT state flag. This
1008 * bit, when set, indicates that we have had a fatal error somewhere,
1009 * either inside the journaling layer or indicated to us by the client
1010 * (eg. ext3), and that we and should not commit any further
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001011 * transactions.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 */
1013
1014static inline int is_journal_aborted(journal_t *journal)
1015{
1016 return journal->j_flags & JFS_ABORT;
1017}
1018
1019static inline int is_handle_aborted(handle_t *handle)
1020{
1021 if (handle->h_aborted)
1022 return 1;
1023 return is_journal_aborted(handle->h_transaction->t_journal);
1024}
1025
1026static inline void journal_abort_handle(handle_t *handle)
1027{
1028 handle->h_aborted = 1;
1029}
1030
1031#endif /* __KERNEL__ */
1032
1033/* Comparison functions for transaction IDs: perform comparisons using
1034 * modulo arithmetic so that they work over sequence number wraps. */
1035
1036static inline int tid_gt(tid_t x, tid_t y)
1037{
1038 int difference = (x - y);
1039 return (difference > 0);
1040}
1041
1042static inline int tid_geq(tid_t x, tid_t y)
1043{
1044 int difference = (x - y);
1045 return (difference >= 0);
1046}
1047
1048extern int journal_blocks_per_page(struct inode *inode);
1049
1050/*
1051 * Return the minimum number of blocks which must be free in the journal
1052 * before a new transaction may be started. Must be called under j_state_lock.
1053 */
1054static inline int jbd_space_needed(journal_t *journal)
1055{
1056 int nblocks = journal->j_max_transaction_buffers;
1057 if (journal->j_committing_transaction)
1058 nblocks += journal->j_committing_transaction->
1059 t_outstanding_credits;
1060 return nblocks;
1061}
1062
1063/*
1064 * Definitions which augment the buffer_head layer
1065 */
1066
1067/* journaling buffer types */
1068#define BJ_None 0 /* Not journaled */
1069#define BJ_SyncData 1 /* Normal data: flush before commit */
1070#define BJ_Metadata 2 /* Normal journaled metadata */
1071#define BJ_Forget 3 /* Buffer superseded by this transaction */
1072#define BJ_IO 4 /* Buffer is for temporary IO use */
1073#define BJ_Shadow 5 /* Buffer contents being shadowed to the log */
1074#define BJ_LogCtl 6 /* Buffer contains log descriptors */
1075#define BJ_Reserved 7 /* Buffer is reserved for access by journal */
1076#define BJ_Locked 8 /* Locked for I/O during commit */
1077#define BJ_Types 9
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079extern int jbd_blocks_per_page(struct inode *inode);
1080
1081#ifdef __KERNEL__
1082
1083#define buffer_trace_init(bh) do {} while (0)
1084#define print_buffer_fields(bh) do {} while (0)
1085#define print_buffer_trace(bh) do {} while (0)
1086#define BUFFER_TRACE(bh, info) do {} while (0)
1087#define BUFFER_TRACE2(bh, bh2, info) do {} while (0)
1088#define JBUFFER_TRACE(jh, info) do {} while (0)
1089
1090#endif /* __KERNEL__ */
1091
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092#endif /* _LINUX_JBD_H */