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 #include <linux/slab.h>
34 #endif
35
36 #define journal_oom_retry 1
37
38 /*
39 * Define JBD2_PARANIOD_IOFAIL to cause a kernel BUG() if ext4 finds
40 * certain classes of error which can occur due to failed IOs. Under
41 * normal use we want ext4 to continue after such errors, because
42 * hardware _can_ fail, but for debugging purposes when running tests on
43 * known-good hardware we may want to trap these errors.
44 */
45 #undef JBD2_PARANOID_IOFAIL
46
47 /*
48 * The default maximum commit age, in seconds.
49 */
50 #define JBD2_DEFAULT_MAX_COMMIT_AGE 5
51
52 #ifdef CONFIG_JBD2_DEBUG
53 /*
54 * Define JBD2_EXPENSIVE_CHECKING to enable more expensive internal
55 * consistency checks. By default we don't do this unless
56 * CONFIG_JBD2_DEBUG is on.
57 */
58 #define JBD2_EXPENSIVE_CHECKING
59 extern u8 jbd2_journal_enable_debug;
60
61 #define jbd_debug(n, f, a...) \
62 do { \
63 if ((n) <= jbd2_journal_enable_debug) { \
64 printk (KERN_DEBUG "(%s, %d): %s: ", \
65 __FILE__, __LINE__, __func__); \
66 printk (f, ## a); \
67 } \
68 } while (0)
69 #else
70 #define jbd_debug(f, a...) /**/
71 #endif
72
73 extern void *jbd2_alloc(size_t size, gfp_t flags);
74 extern void jbd2_free(void *ptr, size_t size);
75
76 #define JBD2_MIN_JOURNAL_BLOCKS 1024
77
78 #ifdef __KERNEL__
79
80 /**
81 * typedef handle_t - The handle_t type represents a single atomic update being performed by some process.
82 *
83 * All filesystem modifications made by the process go
84 * through this handle. Recursive operations (such as quota operations)
85 * are gathered into a single update.
86 *
87 * The buffer credits field is used to account for journaled buffers
88 * being modified by the running process. To ensure that there is
89 * enough log space for all outstanding operations, we need to limit the
90 * number of outstanding buffers possible at any time. When the
91 * operation completes, any buffer credits not used are credited back to
92 * the transaction, so that at all times we know how many buffers the
93 * outstanding updates on a transaction might possibly touch.
94 *
95 * This is an opaque datatype.
96 **/
97 typedef struct jbd2_journal_handle handle_t; /* Atomic operation type */
98
99
100 /**
101 * typedef journal_t - The journal_t maintains all of the journaling state information for a single filesystem.
102 *
103 * journal_t is linked to from the fs superblock structure.
104 *
105 * We use the journal_t to keep track of all outstanding transaction
106 * activity on the filesystem, and to manage the state of the log
107 * writing process.
108 *
109 * This is an opaque datatype.
110 **/
111 typedef struct journal_s journal_t; /* Journal control structure */
112 #endif
113
114 /*
115 * Internal structures used by the logging mechanism:
116 */
117
118 #define JBD2_MAGIC_NUMBER 0xc03b3998U /* The first 4 bytes of /dev/random! */
119
120 /*
121 * On-disk structures
122 */
123
124 /*
125 * Descriptor block types:
126 */
127
128 #define JBD2_DESCRIPTOR_BLOCK 1
129 #define JBD2_COMMIT_BLOCK 2
130 #define JBD2_SUPERBLOCK_V1 3
131 #define JBD2_SUPERBLOCK_V2 4
132 #define JBD2_REVOKE_BLOCK 5
133
134 /*
135 * Standard header for all descriptor blocks:
136 */
137 typedef struct journal_header_s
138 {
139 __be32 h_magic;
140 __be32 h_blocktype;
141 __be32 h_sequence;
142 } journal_header_t;
143
144 /*
145 * Checksum types.
146 */
147 #define JBD2_CRC32_CHKSUM 1
148 #define JBD2_MD5_CHKSUM 2
149 #define JBD2_SHA1_CHKSUM 3
150
151 #define JBD2_CRC32_CHKSUM_SIZE 4
152
153 #define JBD2_CHECKSUM_BYTES (32 / sizeof(u32))
154 /*
155 * Commit block header for storing transactional checksums:
156 */
157 struct commit_header {
158 __be32 h_magic;
159 __be32 h_blocktype;
160 __be32 h_sequence;
161 unsigned char h_chksum_type;
162 unsigned char h_chksum_size;
163 unsigned char h_padding[2];
164 __be32 h_chksum[JBD2_CHECKSUM_BYTES];
165 __be64 h_commit_sec;
166 __be32 h_commit_nsec;
167 };
168
169 /*
170 * The block tag: used to describe a single buffer in the journal.
171 * t_blocknr_high is only used if INCOMPAT_64BIT is set, so this
172 * raw struct shouldn't be used for pointer math or sizeof() - use
173 * journal_tag_bytes(journal) instead to compute this.
174 */
175 typedef struct journal_block_tag_s
176 {
177 __be32 t_blocknr; /* The on-disk block number */
178 __be32 t_flags; /* See below */
179 __be32 t_blocknr_high; /* most-significant high 32bits. */
180 } journal_block_tag_t;
181
182 #define JBD2_TAG_SIZE32 (offsetof(journal_block_tag_t, t_blocknr_high))
183 #define JBD2_TAG_SIZE64 (sizeof(journal_block_tag_t))
184
185 /*
186 * The revoke descriptor: used on disk to describe a series of blocks to
187 * be revoked from the log
188 */
189 typedef struct jbd2_journal_revoke_header_s
190 {
191 journal_header_t r_header;
192 __be32 r_count; /* Count of bytes used in the block */
193 } jbd2_journal_revoke_header_t;
194
195
196 /* Definitions for the journal tag flags word: */
197 #define JBD2_FLAG_ESCAPE 1 /* on-disk block is escaped */
198 #define JBD2_FLAG_SAME_UUID 2 /* block has same uuid as previous */
199 #define JBD2_FLAG_DELETED 4 /* block deleted by this transaction */
200 #define JBD2_FLAG_LAST_TAG 8 /* last tag in this descriptor block */
201
202
203 /*
204 * The journal superblock. All fields are in big-endian byte order.
205 */
206 typedef struct journal_superblock_s
207 {
208 /* 0x0000 */
209 journal_header_t s_header;
210
211 /* 0x000C */
212 /* Static information describing the journal */
213 __be32 s_blocksize; /* journal device blocksize */
214 __be32 s_maxlen; /* total blocks in journal file */
215 __be32 s_first; /* first block of log information */
216
217 /* 0x0018 */
218 /* Dynamic information describing the current state of the log */
219 __be32 s_sequence; /* first commit ID expected in log */
220 __be32 s_start; /* blocknr of start of log */
221
222 /* 0x0020 */
223 /* Error value, as set by jbd2_journal_abort(). */
224 __be32 s_errno;
225
226 /* 0x0024 */
227 /* Remaining fields are only valid in a version-2 superblock */
228 __be32 s_feature_compat; /* compatible feature set */
229 __be32 s_feature_incompat; /* incompatible feature set */
230 __be32 s_feature_ro_compat; /* readonly-compatible feature set */
231 /* 0x0030 */
232 __u8 s_uuid[16]; /* 128-bit uuid for journal */
233
234 /* 0x0040 */
235 __be32 s_nr_users; /* Nr of filesystems sharing log */
236
237 __be32 s_dynsuper; /* Blocknr of dynamic superblock copy*/
238
239 /* 0x0048 */
240 __be32 s_max_transaction; /* Limit of journal blocks per trans.*/
241 __be32 s_max_trans_data; /* Limit of data blocks per trans. */
242
243 /* 0x0050 */
244 __u32 s_padding[44];
245
246 /* 0x0100 */
247 __u8 s_users[16*48]; /* ids of all fs'es sharing the log */
248 /* 0x0400 */
249 } journal_superblock_t;
250
251 #define JBD2_HAS_COMPAT_FEATURE(j,mask) \
252 ((j)->j_format_version >= 2 && \
253 ((j)->j_superblock->s_feature_compat & cpu_to_be32((mask))))
254 #define JBD2_HAS_RO_COMPAT_FEATURE(j,mask) \
255 ((j)->j_format_version >= 2 && \
256 ((j)->j_superblock->s_feature_ro_compat & cpu_to_be32((mask))))
257 #define JBD2_HAS_INCOMPAT_FEATURE(j,mask) \
258 ((j)->j_format_version >= 2 && \
259 ((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask))))
260
261 #define JBD2_FEATURE_COMPAT_CHECKSUM 0x00000001
262
263 #define JBD2_FEATURE_INCOMPAT_REVOKE 0x00000001
264 #define JBD2_FEATURE_INCOMPAT_64BIT 0x00000002
265 #define JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT 0x00000004
266
267 /* Features known to this kernel version: */
268 #define JBD2_KNOWN_COMPAT_FEATURES JBD2_FEATURE_COMPAT_CHECKSUM
269 #define JBD2_KNOWN_ROCOMPAT_FEATURES 0
270 #define JBD2_KNOWN_INCOMPAT_FEATURES (JBD2_FEATURE_INCOMPAT_REVOKE | \
271 JBD2_FEATURE_INCOMPAT_64BIT | \
272 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)
273
274 #ifdef __KERNEL__
275
276 #include <linux/fs.h>
277 #include <linux/sched.h>
278 #include <linux/jbd_common.h>
279
280 #define J_ASSERT(assert) BUG_ON(!(assert))
281
282 #define J_ASSERT_BH(bh, expr) J_ASSERT(expr)
283 #define J_ASSERT_JH(jh, expr) J_ASSERT(expr)
284
285 #if defined(JBD2_PARANOID_IOFAIL)
286 #define J_EXPECT(expr, why...) J_ASSERT(expr)
287 #define J_EXPECT_BH(bh, expr, why...) J_ASSERT_BH(bh, expr)
288 #define J_EXPECT_JH(jh, expr, why...) J_ASSERT_JH(jh, expr)
289 #else
290 #define __journal_expect(expr, why...) \
291 ({ \
292 int val = (expr); \
293 if (!val) { \
294 printk(KERN_ERR \
295 "JBD2 unexpected failure: %s: %s;\n", \
296 __func__, #expr); \
297 printk(KERN_ERR why "\n"); \
298 } \
299 val; \
300 })
301 #define J_EXPECT(expr, why...) __journal_expect(expr, ## why)
302 #define J_EXPECT_BH(bh, expr, why...) __journal_expect(expr, ## why)
303 #define J_EXPECT_JH(jh, expr, why...) __journal_expect(expr, ## why)
304 #endif
305
306 /* Flags in jbd_inode->i_flags */
307 #define __JI_COMMIT_RUNNING 0
308 /* Commit of the inode data in progress. We use this flag to protect us from
309 * concurrent deletion of inode. We cannot use reference to inode for this
310 * since we cannot afford doing last iput() on behalf of kjournald
311 */
312 #define JI_COMMIT_RUNNING (1 << __JI_COMMIT_RUNNING)
313
314 /**
315 * struct jbd_inode is the structure linking inodes in ordered mode
316 * present in a transaction so that we can sync them during commit.
317 */
318 struct jbd2_inode {
319 /* Which transaction does this inode belong to? Either the running
320 * transaction or the committing one. [j_list_lock] */
321 transaction_t *i_transaction;
322
323 /* Pointer to the running transaction modifying inode's data in case
324 * there is already a committing transaction touching it. [j_list_lock] */
325 transaction_t *i_next_transaction;
326
327 /* List of inodes in the i_transaction [j_list_lock] */
328 struct list_head i_list;
329
330 /* VFS inode this inode belongs to [constant during the lifetime
331 * of the structure] */
332 struct inode *i_vfs_inode;
333
334 /* Flags of inode [j_list_lock] */
335 unsigned long i_flags;
336 };
337
338 struct jbd2_revoke_table_s;
339
340 /**
341 * struct handle_s - The handle_s type is the concrete type associated with
342 * handle_t.
343 * @h_transaction: Which compound transaction is this update a part of?
344 * @h_buffer_credits: Number of remaining buffers we are allowed to dirty.
345 * @h_ref: Reference count on this handle
346 * @h_err: Field for caller's use to track errors through large fs operations
347 * @h_sync: flag for sync-on-close
348 * @h_jdata: flag to force data journaling
349 * @h_aborted: flag indicating fatal error on handle
350 **/
351
352 /* Docbook can't yet cope with the bit fields, but will leave the documentation
353 * in so it can be fixed later.
354 */
355
356 struct jbd2_journal_handle
357 {
358 /* Which compound transaction is this update a part of? */
359 transaction_t *h_transaction;
360
361 /* Number of remaining buffers we are allowed to dirty: */
362 int h_buffer_credits;
363
364 /* Reference count on this handle */
365 int h_ref;
366
367 /* Field for caller's use to track errors through large fs */
368 /* operations */
369 int h_err;
370
371 /* Flags [no locking] */
372 unsigned int h_sync:1; /* sync-on-close */
373 unsigned int h_jdata:1; /* force data journaling */
374 unsigned int h_aborted:1; /* fatal error on handle */
375 unsigned int h_cowing:1; /* COWing block to snapshot */
376
377 /* Number of buffers requested by user:
378 * (before adding the COW credits factor) */
379 unsigned int h_base_credits:14;
380
381 /* Number of buffers the user is allowed to dirty:
382 * (counts only buffers dirtied when !h_cowing) */
383 unsigned int h_user_credits:14;
384
385
386 #ifdef CONFIG_DEBUG_LOCK_ALLOC
387 struct lockdep_map h_lockdep_map;
388 #endif
389
390 #ifdef CONFIG_JBD2_DEBUG
391 /* COW debugging counters: */
392 unsigned int h_cow_moved; /* blocks moved to snapshot */
393 unsigned int h_cow_copied; /* blocks copied to snapshot */
394 unsigned int h_cow_ok_jh; /* blocks already COWed during current
395 transaction */
396 unsigned int h_cow_ok_bitmap; /* blocks not set in COW bitmap */
397 unsigned int h_cow_ok_mapped;/* blocks already mapped in snapshot */
398 unsigned int h_cow_bitmaps; /* COW bitmaps created */
399 unsigned int h_cow_excluded; /* blocks set in exclude bitmap */
400 #endif
401 };
402
403
404 /*
405 * Some stats for checkpoint phase
406 */
407 struct transaction_chp_stats_s {
408 unsigned long cs_chp_time;
409 __u32 cs_forced_to_close;
410 __u32 cs_written;
411 __u32 cs_dropped;
412 };
413
414 /* The transaction_t type is the guts of the journaling mechanism. It
415 * tracks a compound transaction through its various states:
416 *
417 * RUNNING: accepting new updates
418 * LOCKED: Updates still running but we don't accept new ones
419 * RUNDOWN: Updates are tidying up but have finished requesting
420 * new buffers to modify (state not used for now)
421 * FLUSH: All updates complete, but we are still writing to disk
422 * COMMIT: All data on disk, writing commit record
423 * FINISHED: We still have to keep the transaction for checkpointing.
424 *
425 * The transaction keeps track of all of the buffers modified by a
426 * running transaction, and all of the buffers committed but not yet
427 * flushed to home for finished transactions.
428 */
429
430 /*
431 * Lock ranking:
432 *
433 * j_list_lock
434 * ->jbd_lock_bh_journal_head() (This is "innermost")
435 *
436 * j_state_lock
437 * ->jbd_lock_bh_state()
438 *
439 * jbd_lock_bh_state()
440 * ->j_list_lock
441 *
442 * j_state_lock
443 * ->t_handle_lock
444 *
445 * j_state_lock
446 * ->j_list_lock (journal_unmap_buffer)
447 *
448 */
449
450 struct transaction_s
451 {
452 /* Pointer to the journal for this transaction. [no locking] */
453 journal_t *t_journal;
454
455 /* Sequence number for this transaction [no locking] */
456 tid_t t_tid;
457
458 /*
459 * Transaction's current state
460 * [no locking - only kjournald2 alters this]
461 * [j_list_lock] guards transition of a transaction into T_FINISHED
462 * state and subsequent call of __jbd2_journal_drop_transaction()
463 * FIXME: needs barriers
464 * KLUDGE: [use j_state_lock]
465 */
466 enum {
467 T_RUNNING,
468 T_LOCKED,
469 T_FLUSH,
470 T_COMMIT,
471 T_COMMIT_DFLUSH,
472 T_COMMIT_JFLUSH,
473 T_COMMIT_CALLBACK,
474 T_FINISHED
475 } t_state;
476
477 /*
478 * Where in the log does this transaction's commit start? [no locking]
479 */
480 unsigned long t_log_start;
481
482 /* Number of buffers on the t_buffers list [j_list_lock] */
483 int t_nr_buffers;
484
485 /*
486 * Doubly-linked circular list of all buffers reserved but not yet
487 * modified by this transaction [j_list_lock]
488 */
489 struct journal_head *t_reserved_list;
490
491 /*
492 * Doubly-linked circular list of all metadata buffers owned by this
493 * transaction [j_list_lock]
494 */
495 struct journal_head *t_buffers;
496
497 /*
498 * Doubly-linked circular list of all forget buffers (superseded
499 * buffers which we can un-checkpoint once this transaction commits)
500 * [j_list_lock]
501 */
502 struct journal_head *t_forget;
503
504 /*
505 * Doubly-linked circular list of all buffers still to be flushed before
506 * this transaction can be checkpointed. [j_list_lock]
507 */
508 struct journal_head *t_checkpoint_list;
509
510 /*
511 * Doubly-linked circular list of all buffers submitted for IO while
512 * checkpointing. [j_list_lock]
513 */
514 struct journal_head *t_checkpoint_io_list;
515
516 /*
517 * Doubly-linked circular list of temporary buffers currently undergoing
518 * IO in the log [j_list_lock]
519 */
520 struct journal_head *t_iobuf_list;
521
522 /*
523 * Doubly-linked circular list of metadata buffers being shadowed by log
524 * IO. The IO buffers on the iobuf list and the shadow buffers on this
525 * list match each other one for one at all times. [j_list_lock]
526 */
527 struct journal_head *t_shadow_list;
528
529 /*
530 * Doubly-linked circular list of control buffers being written to the
531 * log. [j_list_lock]
532 */
533 struct journal_head *t_log_list;
534
535 /*
536 * List of inodes whose data we've modified in data=ordered mode.
537 * [j_list_lock]
538 */
539 struct list_head t_inode_list;
540
541 /*
542 * Protects info related to handles
543 */
544 spinlock_t t_handle_lock;
545
546 /*
547 * Longest time some handle had to wait for running transaction
548 */
549 unsigned long t_max_wait;
550
551 /*
552 * When transaction started
553 */
554 unsigned long t_start;
555
556 /*
557 * Checkpointing stats [j_checkpoint_sem]
558 */
559 struct transaction_chp_stats_s t_chp_stats;
560
561 /*
562 * Number of outstanding updates running on this transaction
563 * [t_handle_lock]
564 */
565 atomic_t t_updates;
566
567 /*
568 * Number of buffers reserved for use by all handles in this transaction
569 * handle but not yet modified. [t_handle_lock]
570 */
571 atomic_t t_outstanding_credits;
572
573 /*
574 * Forward and backward links for the circular list of all transactions
575 * awaiting checkpoint. [j_list_lock]
576 */
577 transaction_t *t_cpnext, *t_cpprev;
578
579 /*
580 * When will the transaction expire (become due for commit), in jiffies?
581 * [no locking]
582 */
583 unsigned long t_expires;
584
585 /*
586 * When this transaction started, in nanoseconds [no locking]
587 */
588 ktime_t t_start_time;
589
590 /*
591 * How many handles used this transaction? [t_handle_lock]
592 */
593 atomic_t t_handle_count;
594
595 /*
596 * This transaction is being forced and some process is
597 * waiting for it to finish.
598 */
599 unsigned int t_synchronous_commit:1;
600
601 /* Disk flush needs to be sent to fs partition [no locking] */
602 int t_need_data_flush;
603
604 /*
605 * For use by the filesystem to store fs-specific data
606 * structures associated with the transaction
607 */
608 struct list_head t_private_list;
609 };
610
611 struct transaction_run_stats_s {
612 unsigned long rs_wait;
613 unsigned long rs_running;
614 unsigned long rs_locked;
615 unsigned long rs_flushing;
616 unsigned long rs_logging;
617
618 __u32 rs_handle_count;
619 __u32 rs_blocks;
620 __u32 rs_blocks_logged;
621 };
622
623 struct transaction_stats_s {
624 unsigned long ts_tid;
625 struct transaction_run_stats_s run;
626 };
627
628 static inline unsigned long
jbd2_time_diff(unsigned long start,unsigned long end)629 jbd2_time_diff(unsigned long start, unsigned long end)
630 {
631 if (end >= start)
632 return end - start;
633
634 return end + (MAX_JIFFY_OFFSET - start);
635 }
636
637 #define JBD2_NR_BATCH 64
638
639 /**
640 * struct journal_s - The journal_s type is the concrete type associated with
641 * journal_t.
642 * @j_flags: General journaling state flags
643 * @j_errno: Is there an outstanding uncleared error on the journal (from a
644 * prior abort)?
645 * @j_sb_buffer: First part of superblock buffer
646 * @j_superblock: Second part of superblock buffer
647 * @j_format_version: Version of the superblock format
648 * @j_state_lock: Protect the various scalars in the journal
649 * @j_barrier_count: Number of processes waiting to create a barrier lock
650 * @j_barrier: The barrier lock itself
651 * @j_running_transaction: The current running transaction..
652 * @j_committing_transaction: the transaction we are pushing to disk
653 * @j_checkpoint_transactions: a linked circular list of all transactions
654 * waiting for checkpointing
655 * @j_wait_transaction_locked: Wait queue for waiting for a locked transaction
656 * to start committing, or for a barrier lock to be released
657 * @j_wait_logspace: Wait queue for waiting for checkpointing to complete
658 * @j_wait_done_commit: Wait queue for waiting for commit to complete
659 * @j_wait_checkpoint: Wait queue to trigger checkpointing
660 * @j_wait_commit: Wait queue to trigger commit
661 * @j_wait_updates: Wait queue to wait for updates to complete
662 * @j_checkpoint_mutex: Mutex for locking against concurrent checkpoints
663 * @j_head: Journal head - identifies the first unused block in the journal
664 * @j_tail: Journal tail - identifies the oldest still-used block in the
665 * journal.
666 * @j_free: Journal free - how many free blocks are there in the journal?
667 * @j_first: The block number of the first usable block
668 * @j_last: The block number one beyond the last usable block
669 * @j_dev: Device where we store the journal
670 * @j_blocksize: blocksize for the location where we store the journal.
671 * @j_blk_offset: starting block offset for into the device where we store the
672 * journal
673 * @j_fs_dev: Device which holds the client fs. For internal journal this will
674 * be equal to j_dev
675 * @j_maxlen: Total maximum capacity of the journal region on disk.
676 * @j_list_lock: Protects the buffer lists and internal buffer state.
677 * @j_inode: Optional inode where we store the journal. If present, all journal
678 * block numbers are mapped into this inode via bmap().
679 * @j_tail_sequence: Sequence number of the oldest transaction in the log
680 * @j_transaction_sequence: Sequence number of the next transaction to grant
681 * @j_commit_sequence: Sequence number of the most recently committed
682 * transaction
683 * @j_commit_request: Sequence number of the most recent transaction wanting
684 * commit
685 * @j_uuid: Uuid of client object.
686 * @j_task: Pointer to the current commit thread for this journal
687 * @j_max_transaction_buffers: Maximum number of metadata buffers to allow in a
688 * single compound commit transaction
689 * @j_commit_interval: What is the maximum transaction lifetime before we begin
690 * a commit?
691 * @j_commit_timer: The timer used to wakeup the commit thread
692 * @j_revoke_lock: Protect the revoke table
693 * @j_revoke: The revoke table - maintains the list of revoked blocks in the
694 * current transaction.
695 * @j_revoke_table: alternate revoke tables for j_revoke
696 * @j_wbuf: array of buffer_heads for jbd2_journal_commit_transaction
697 * @j_wbufsize: maximum number of buffer_heads allowed in j_wbuf, the
698 * number that will fit in j_blocksize
699 * @j_last_sync_writer: most recent pid which did a synchronous write
700 * @j_history: Buffer storing the transactions statistics history
701 * @j_history_max: Maximum number of transactions in the statistics history
702 * @j_history_cur: Current number of transactions in the statistics history
703 * @j_history_lock: Protect the transactions statistics history
704 * @j_proc_entry: procfs entry for the jbd statistics directory
705 * @j_stats: Overall statistics
706 * @j_private: An opaque pointer to fs-private information.
707 */
708
709 struct journal_s
710 {
711 /* General journaling state flags [j_state_lock] */
712 unsigned long j_flags;
713
714 /*
715 * Is there an outstanding uncleared error on the journal (from a prior
716 * abort)? [j_state_lock]
717 */
718 int j_errno;
719
720 /* The superblock buffer */
721 struct buffer_head *j_sb_buffer;
722 journal_superblock_t *j_superblock;
723
724 /* Version of the superblock format */
725 int j_format_version;
726
727 /*
728 * Protect the various scalars in the journal
729 */
730 rwlock_t j_state_lock;
731
732 /*
733 * Number of processes waiting to create a barrier lock [j_state_lock]
734 */
735 int j_barrier_count;
736
737 /* The barrier lock itself */
738 struct mutex j_barrier;
739
740 /*
741 * Transactions: The current running transaction...
742 * [j_state_lock] [caller holding open handle]
743 */
744 transaction_t *j_running_transaction;
745
746 /*
747 * the transaction we are pushing to disk
748 * [j_state_lock] [caller holding open handle]
749 */
750 transaction_t *j_committing_transaction;
751
752 /*
753 * ... and a linked circular list of all transactions waiting for
754 * checkpointing. [j_list_lock]
755 */
756 transaction_t *j_checkpoint_transactions;
757
758 /*
759 * Wait queue for waiting for a locked transaction to start committing,
760 * or for a barrier lock to be released
761 */
762 wait_queue_head_t j_wait_transaction_locked;
763
764 /* Wait queue for waiting for checkpointing to complete */
765 wait_queue_head_t j_wait_logspace;
766
767 /* Wait queue for waiting for commit to complete */
768 wait_queue_head_t j_wait_done_commit;
769
770 /* Wait queue to trigger checkpointing */
771 wait_queue_head_t j_wait_checkpoint;
772
773 /* Wait queue to trigger commit */
774 wait_queue_head_t j_wait_commit;
775
776 /* Wait queue to wait for updates to complete */
777 wait_queue_head_t j_wait_updates;
778
779 /* Semaphore for locking against concurrent checkpoints */
780 struct mutex j_checkpoint_mutex;
781
782 /*
783 * List of buffer heads used by the checkpoint routine. This
784 * was moved from jbd2_log_do_checkpoint() to reduce stack
785 * usage. Access to this array is controlled by the
786 * j_checkpoint_mutex. [j_checkpoint_mutex]
787 */
788 struct buffer_head *j_chkpt_bhs[JBD2_NR_BATCH];
789
790 /*
791 * Journal head: identifies the first unused block in the journal.
792 * [j_state_lock]
793 */
794 unsigned long j_head;
795
796 /*
797 * Journal tail: identifies the oldest still-used block in the journal.
798 * [j_state_lock]
799 */
800 unsigned long j_tail;
801
802 /*
803 * Journal free: how many free blocks are there in the journal?
804 * [j_state_lock]
805 */
806 unsigned long j_free;
807
808 /*
809 * Journal start and end: the block numbers of the first usable block
810 * and one beyond the last usable block in the journal. [j_state_lock]
811 */
812 unsigned long j_first;
813 unsigned long j_last;
814
815 /*
816 * Device, blocksize and starting block offset for the location where we
817 * store the journal.
818 */
819 struct block_device *j_dev;
820 int j_blocksize;
821 unsigned long long j_blk_offset;
822 char j_devname[BDEVNAME_SIZE+24];
823
824 /*
825 * Device which holds the client fs. For internal journal this will be
826 * equal to j_dev.
827 */
828 struct block_device *j_fs_dev;
829
830 /* Total maximum capacity of the journal region on disk. */
831 unsigned int j_maxlen;
832
833 /*
834 * Protects the buffer lists and internal buffer state.
835 */
836 spinlock_t j_list_lock;
837
838 /* Optional inode where we store the journal. If present, all */
839 /* journal block numbers are mapped into this inode via */
840 /* bmap(). */
841 struct inode *j_inode;
842
843 /*
844 * Sequence number of the oldest transaction in the log [j_state_lock]
845 */
846 tid_t j_tail_sequence;
847
848 /*
849 * Sequence number of the next transaction to grant [j_state_lock]
850 */
851 tid_t j_transaction_sequence;
852
853 /*
854 * Sequence number of the most recently committed transaction
855 * [j_state_lock].
856 */
857 tid_t j_commit_sequence;
858
859 /*
860 * Sequence number of the most recent transaction wanting commit
861 * [j_state_lock]
862 */
863 tid_t j_commit_request;
864
865 /*
866 * Journal uuid: identifies the object (filesystem, LVM volume etc)
867 * backed by this journal. This will eventually be replaced by an array
868 * of uuids, allowing us to index multiple devices within a single
869 * journal and to perform atomic updates across them.
870 */
871 __u8 j_uuid[16];
872
873 /* Pointer to the current commit thread for this journal */
874 struct task_struct *j_task;
875
876 /*
877 * Maximum number of metadata buffers to allow in a single compound
878 * commit transaction
879 */
880 int j_max_transaction_buffers;
881
882 /*
883 * What is the maximum transaction lifetime before we begin a commit?
884 */
885 unsigned long j_commit_interval;
886
887 /* The timer used to wakeup the commit thread: */
888 struct timer_list j_commit_timer;
889
890 /*
891 * The revoke table: maintains the list of revoked blocks in the
892 * current transaction. [j_revoke_lock]
893 */
894 spinlock_t j_revoke_lock;
895 struct jbd2_revoke_table_s *j_revoke;
896 struct jbd2_revoke_table_s *j_revoke_table[2];
897
898 /*
899 * array of bhs for jbd2_journal_commit_transaction
900 */
901 struct buffer_head **j_wbuf;
902 int j_wbufsize;
903
904 /*
905 * this is the pid of hte last person to run a synchronous operation
906 * through the journal
907 */
908 pid_t j_last_sync_writer;
909
910 /*
911 * the average amount of time in nanoseconds it takes to commit a
912 * transaction to disk. [j_state_lock]
913 */
914 u64 j_average_commit_time;
915
916 /*
917 * minimum and maximum times that we should wait for
918 * additional filesystem operations to get batched into a
919 * synchronous handle in microseconds
920 */
921 u32 j_min_batch_time;
922 u32 j_max_batch_time;
923
924 /* This function is called when a transaction is closed */
925 void (*j_commit_callback)(journal_t *,
926 transaction_t *);
927
928 /*
929 * Journal statistics
930 */
931 spinlock_t j_history_lock;
932 struct proc_dir_entry *j_proc_entry;
933 struct transaction_stats_s j_stats;
934
935 /* Failed journal commit ID */
936 unsigned int j_failed_commit;
937
938 /*
939 * An opaque pointer to fs-private information. ext3 puts its
940 * superblock pointer here
941 */
942 void *j_private;
943 };
944
945 /*
946 * Journal flag definitions
947 */
948 #define JBD2_UNMOUNT 0x001 /* Journal thread is being destroyed */
949 #define JBD2_ABORT 0x002 /* Journaling has been aborted for errors. */
950 #define JBD2_ACK_ERR 0x004 /* The errno in the sb has been acked */
951 #define JBD2_FLUSHED 0x008 /* The journal superblock has been flushed */
952 #define JBD2_LOADED 0x010 /* The journal superblock has been loaded */
953 #define JBD2_BARRIER 0x020 /* Use IDE barriers */
954 #define JBD2_ABORT_ON_SYNCDATA_ERR 0x040 /* Abort the journal on file
955 * data write error in ordered
956 * mode */
957
958 /*
959 * Function declarations for the journaling transaction and buffer
960 * management
961 */
962
963 /* Filing buffers */
964 extern void jbd2_journal_unfile_buffer(journal_t *, struct journal_head *);
965 extern void __jbd2_journal_refile_buffer(struct journal_head *);
966 extern void jbd2_journal_refile_buffer(journal_t *, struct journal_head *);
967 extern void __jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int);
968 extern void __journal_free_buffer(struct journal_head *bh);
969 extern void jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int);
970 extern void __journal_clean_data_list(transaction_t *transaction);
971
972 /* Log buffer allocation */
973 extern struct journal_head * jbd2_journal_get_descriptor_buffer(journal_t *);
974 int jbd2_journal_next_log_block(journal_t *, unsigned long long *);
975 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
976 unsigned long *block);
977 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block);
978 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block);
979
980 /* Commit management */
981 extern void jbd2_journal_commit_transaction(journal_t *);
982
983 /* Checkpoint list management */
984 int __jbd2_journal_clean_checkpoint_list(journal_t *journal);
985 int __jbd2_journal_remove_checkpoint(struct journal_head *);
986 void __jbd2_journal_insert_checkpoint(struct journal_head *, transaction_t *);
987
988
989 /*
990 * Triggers
991 */
992
993 struct jbd2_buffer_trigger_type {
994 /*
995 * Fired a the moment data to write to the journal are known to be
996 * stable - so either at the moment b_frozen_data is created or just
997 * before a buffer is written to the journal. mapped_data is a mapped
998 * buffer that is the frozen data for commit.
999 */
1000 void (*t_frozen)(struct jbd2_buffer_trigger_type *type,
1001 struct buffer_head *bh, void *mapped_data,
1002 size_t size);
1003
1004 /*
1005 * Fired during journal abort for dirty buffers that will not be
1006 * committed.
1007 */
1008 void (*t_abort)(struct jbd2_buffer_trigger_type *type,
1009 struct buffer_head *bh);
1010 };
1011
1012 extern void jbd2_buffer_frozen_trigger(struct journal_head *jh,
1013 void *mapped_data,
1014 struct jbd2_buffer_trigger_type *triggers);
1015 extern void jbd2_buffer_abort_trigger(struct journal_head *jh,
1016 struct jbd2_buffer_trigger_type *triggers);
1017
1018 /* Buffer IO */
1019 extern int
1020 jbd2_journal_write_metadata_buffer(transaction_t *transaction,
1021 struct journal_head *jh_in,
1022 struct journal_head **jh_out,
1023 unsigned long long blocknr);
1024
1025 /* Transaction locking */
1026 extern void __wait_on_journal (journal_t *);
1027
1028 /* Transaction cache support */
1029 extern void jbd2_journal_destroy_transaction_cache(void);
1030 extern int jbd2_journal_init_transaction_cache(void);
1031 extern void jbd2_journal_free_transaction(transaction_t *);
1032
1033 /*
1034 * Journal locking.
1035 *
1036 * We need to lock the journal during transaction state changes so that nobody
1037 * ever tries to take a handle on the running transaction while we are in the
1038 * middle of moving it to the commit phase. j_state_lock does this.
1039 *
1040 * Note that the locking is completely interrupt unsafe. We never touch
1041 * journal structures from interrupts.
1042 */
1043
journal_current_handle(void)1044 static inline handle_t *journal_current_handle(void)
1045 {
1046 return current->journal_info;
1047 }
1048
1049 /* The journaling code user interface:
1050 *
1051 * Create and destroy handles
1052 * Register buffer modifications against the current transaction.
1053 */
1054
1055 extern handle_t *jbd2_journal_start(journal_t *, int nblocks);
1056 extern handle_t *jbd2__journal_start(journal_t *, int nblocks, gfp_t gfp_mask);
1057 extern int jbd2_journal_restart(handle_t *, int nblocks);
1058 extern int jbd2__journal_restart(handle_t *, int nblocks, gfp_t gfp_mask);
1059 extern int jbd2_journal_extend (handle_t *, int nblocks);
1060 extern int jbd2_journal_get_write_access(handle_t *, struct buffer_head *);
1061 extern int jbd2_journal_get_create_access (handle_t *, struct buffer_head *);
1062 extern int jbd2_journal_get_undo_access(handle_t *, struct buffer_head *);
1063 void jbd2_journal_set_triggers(struct buffer_head *,
1064 struct jbd2_buffer_trigger_type *type);
1065 extern int jbd2_journal_dirty_metadata (handle_t *, struct buffer_head *);
1066 extern void jbd2_journal_release_buffer (handle_t *, struct buffer_head *);
1067 extern int jbd2_journal_forget (handle_t *, struct buffer_head *);
1068 extern void journal_sync_buffer (struct buffer_head *);
1069 extern void jbd2_journal_invalidatepage(journal_t *,
1070 struct page *, unsigned long);
1071 extern int jbd2_journal_try_to_free_buffers(journal_t *, struct page *, gfp_t);
1072 extern int jbd2_journal_stop(handle_t *);
1073 extern int jbd2_journal_flush (journal_t *);
1074 extern void jbd2_journal_lock_updates (journal_t *);
1075 extern void jbd2_journal_unlock_updates (journal_t *);
1076
1077 extern journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1078 struct block_device *fs_dev,
1079 unsigned long long start, int len, int bsize);
1080 extern journal_t * jbd2_journal_init_inode (struct inode *);
1081 extern int jbd2_journal_update_format (journal_t *);
1082 extern int jbd2_journal_check_used_features
1083 (journal_t *, unsigned long, unsigned long, unsigned long);
1084 extern int jbd2_journal_check_available_features
1085 (journal_t *, unsigned long, unsigned long, unsigned long);
1086 extern int jbd2_journal_set_features
1087 (journal_t *, unsigned long, unsigned long, unsigned long);
1088 extern void jbd2_journal_clear_features
1089 (journal_t *, unsigned long, unsigned long, unsigned long);
1090 extern int jbd2_journal_load (journal_t *journal);
1091 extern int jbd2_journal_destroy (journal_t *);
1092 extern int jbd2_journal_recover (journal_t *journal);
1093 extern int jbd2_journal_wipe (journal_t *, int);
1094 extern int jbd2_journal_skip_recovery (journal_t *);
1095 extern void jbd2_journal_update_sb_errno(journal_t *);
1096 extern void jbd2_journal_update_sb_log_tail (journal_t *, tid_t,
1097 unsigned long, int);
1098 extern void __jbd2_journal_abort_hard (journal_t *);
1099 extern void jbd2_journal_abort (journal_t *, int);
1100 extern int jbd2_journal_errno (journal_t *);
1101 extern void jbd2_journal_ack_err (journal_t *);
1102 extern int jbd2_journal_clear_err (journal_t *);
1103 extern int jbd2_journal_bmap(journal_t *, unsigned long, unsigned long long *);
1104 extern int jbd2_journal_force_commit(journal_t *);
1105 extern int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *inode);
1106 extern int jbd2_journal_begin_ordered_truncate(journal_t *journal,
1107 struct jbd2_inode *inode, loff_t new_size);
1108 extern void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode);
1109 extern void jbd2_journal_release_jbd_inode(journal_t *journal, struct jbd2_inode *jinode);
1110
1111 /*
1112 * journal_head management
1113 */
1114 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh);
1115 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh);
1116 void jbd2_journal_put_journal_head(struct journal_head *jh);
1117
1118 /*
1119 * handle management
1120 */
1121 extern struct kmem_cache *jbd2_handle_cache;
1122
jbd2_alloc_handle(gfp_t gfp_flags)1123 static inline handle_t *jbd2_alloc_handle(gfp_t gfp_flags)
1124 {
1125 return kmem_cache_alloc(jbd2_handle_cache, gfp_flags);
1126 }
1127
jbd2_free_handle(handle_t * handle)1128 static inline void jbd2_free_handle(handle_t *handle)
1129 {
1130 kmem_cache_free(jbd2_handle_cache, handle);
1131 }
1132
1133 /*
1134 * jbd2_inode management (optional, for those file systems that want to use
1135 * dynamically allocated jbd2_inode structures)
1136 */
1137 extern struct kmem_cache *jbd2_inode_cache;
1138
jbd2_alloc_inode(gfp_t gfp_flags)1139 static inline struct jbd2_inode *jbd2_alloc_inode(gfp_t gfp_flags)
1140 {
1141 return kmem_cache_alloc(jbd2_inode_cache, gfp_flags);
1142 }
1143
jbd2_free_inode(struct jbd2_inode * jinode)1144 static inline void jbd2_free_inode(struct jbd2_inode *jinode)
1145 {
1146 kmem_cache_free(jbd2_inode_cache, jinode);
1147 }
1148
1149 /* Primary revoke support */
1150 #define JOURNAL_REVOKE_DEFAULT_HASH 256
1151 extern int jbd2_journal_init_revoke(journal_t *, int);
1152 extern void jbd2_journal_destroy_revoke_caches(void);
1153 extern int jbd2_journal_init_revoke_caches(void);
1154
1155 extern void jbd2_journal_destroy_revoke(journal_t *);
1156 extern int jbd2_journal_revoke (handle_t *, unsigned long long, struct buffer_head *);
1157 extern int jbd2_journal_cancel_revoke(handle_t *, struct journal_head *);
1158 extern void jbd2_journal_write_revoke_records(journal_t *,
1159 transaction_t *, int);
1160
1161 /* Recovery revoke support */
1162 extern int jbd2_journal_set_revoke(journal_t *, unsigned long long, tid_t);
1163 extern int jbd2_journal_test_revoke(journal_t *, unsigned long long, tid_t);
1164 extern void jbd2_journal_clear_revoke(journal_t *);
1165 extern void jbd2_journal_switch_revoke_table(journal_t *journal);
1166 extern void jbd2_clear_buffer_revoked_flags(journal_t *journal);
1167
1168 /*
1169 * The log thread user interface:
1170 *
1171 * Request space in the current transaction, and force transaction commit
1172 * transitions on demand.
1173 */
1174
1175 int __jbd2_log_space_left(journal_t *); /* Called with journal locked */
1176 int jbd2_log_start_commit(journal_t *journal, tid_t tid);
1177 int __jbd2_log_start_commit(journal_t *journal, tid_t tid);
1178 int jbd2_journal_start_commit(journal_t *journal, tid_t *tid);
1179 int jbd2_journal_force_commit_nested(journal_t *journal);
1180 int jbd2_log_wait_commit(journal_t *journal, tid_t tid);
1181 int jbd2_complete_transaction(journal_t *journal, tid_t tid);
1182 int jbd2_log_do_checkpoint(journal_t *journal);
1183 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid);
1184
1185 void __jbd2_log_wait_for_space(journal_t *journal);
1186 extern void __jbd2_journal_drop_transaction(journal_t *, transaction_t *);
1187 extern int jbd2_cleanup_journal_tail(journal_t *);
1188
1189 /* Debugging code only: */
1190
1191 #define jbd_ENOSYS() \
1192 do { \
1193 printk (KERN_ERR "JBD unimplemented function %s\n", __func__); \
1194 current->state = TASK_UNINTERRUPTIBLE; \
1195 schedule(); \
1196 } while (1)
1197
1198 /*
1199 * is_journal_abort
1200 *
1201 * Simple test wrapper function to test the JBD2_ABORT state flag. This
1202 * bit, when set, indicates that we have had a fatal error somewhere,
1203 * either inside the journaling layer or indicated to us by the client
1204 * (eg. ext3), and that we and should not commit any further
1205 * transactions.
1206 */
1207
is_journal_aborted(journal_t * journal)1208 static inline int is_journal_aborted(journal_t *journal)
1209 {
1210 return journal->j_flags & JBD2_ABORT;
1211 }
1212
is_handle_aborted(handle_t * handle)1213 static inline int is_handle_aborted(handle_t *handle)
1214 {
1215 if (handle->h_aborted)
1216 return 1;
1217 return is_journal_aborted(handle->h_transaction->t_journal);
1218 }
1219
jbd2_journal_abort_handle(handle_t * handle)1220 static inline void jbd2_journal_abort_handle(handle_t *handle)
1221 {
1222 handle->h_aborted = 1;
1223 }
1224
1225 #endif /* __KERNEL__ */
1226
1227 /* Comparison functions for transaction IDs: perform comparisons using
1228 * modulo arithmetic so that they work over sequence number wraps. */
1229
tid_gt(tid_t x,tid_t y)1230 static inline int tid_gt(tid_t x, tid_t y)
1231 {
1232 int difference = (x - y);
1233 return (difference > 0);
1234 }
1235
tid_geq(tid_t x,tid_t y)1236 static inline int tid_geq(tid_t x, tid_t y)
1237 {
1238 int difference = (x - y);
1239 return (difference >= 0);
1240 }
1241
1242 extern int jbd2_journal_blocks_per_page(struct inode *inode);
1243 extern size_t journal_tag_bytes(journal_t *journal);
1244
1245 /*
1246 * Return the minimum number of blocks which must be free in the journal
1247 * before a new transaction may be started. Must be called under j_state_lock.
1248 */
jbd_space_needed(journal_t * journal)1249 static inline int jbd_space_needed(journal_t *journal)
1250 {
1251 int nblocks = journal->j_max_transaction_buffers;
1252 if (journal->j_committing_transaction)
1253 nblocks += atomic_read(&journal->j_committing_transaction->
1254 t_outstanding_credits);
1255 return nblocks;
1256 }
1257
1258 /*
1259 * Definitions which augment the buffer_head layer
1260 */
1261
1262 /* journaling buffer types */
1263 #define BJ_None 0 /* Not journaled */
1264 #define BJ_Metadata 1 /* Normal journaled metadata */
1265 #define BJ_Forget 2 /* Buffer superseded by this transaction */
1266 #define BJ_IO 3 /* Buffer is for temporary IO use */
1267 #define BJ_Shadow 4 /* Buffer contents being shadowed to the log */
1268 #define BJ_LogCtl 5 /* Buffer contains log descriptors */
1269 #define BJ_Reserved 6 /* Buffer is reserved for access by journal */
1270 #define BJ_Types 7
1271
1272 extern int jbd_blocks_per_page(struct inode *inode);
1273
1274 #ifdef __KERNEL__
1275
1276 #define buffer_trace_init(bh) do {} while (0)
1277 #define print_buffer_fields(bh) do {} while (0)
1278 #define print_buffer_trace(bh) do {} while (0)
1279 #define BUFFER_TRACE(bh, info) do {} while (0)
1280 #define BUFFER_TRACE2(bh, bh2, info) do {} while (0)
1281 #define JBUFFER_TRACE(jh, info) do {} while (0)
1282
1283 #endif /* __KERNEL__ */
1284
1285 #endif /* _LINUX_JBD2_H */
1286