1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * linux/include/linux/jbd2.h
4 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>
6 *
7 * Copyright 1998-2000 Red Hat, Inc --- All Rights Reserved
8 *
9 * Definitions for transaction data structures for the buffer cache
10 * filesystem journaling support.
11 */
12
13 #ifndef _LINUX_JBD2_H
14 #define _LINUX_JBD2_H
15
16 /* Allow this file to be included directly into e2fsprogs */
17 #ifndef __KERNEL__
18 #include "jfs_compat.h"
19 #define JBD2_DEBUG
20 #else
21
22 #include <linux/types.h>
23 #include <linux/buffer_head.h>
24 #include <linux/journal-head.h>
25 #include <linux/stddef.h>
26 #include <linux/mutex.h>
27 #include <linux/timer.h>
28 #include <linux/slab.h>
29 #include <linux/bit_spinlock.h>
30 #include <linux/blkdev.h>
31 #include <linux/crc32c.h>
32 #endif
33
34 #define journal_oom_retry 1
35
36 /*
37 * Define JBD2_PARANIOD_IOFAIL to cause a kernel BUG() if ext4 finds
38 * certain classes of error which can occur due to failed IOs. Under
39 * normal use we want ext4 to continue after such errors, because
40 * hardware _can_ fail, but for debugging purposes when running tests on
41 * known-good hardware we may want to trap these errors.
42 */
43 #undef JBD2_PARANOID_IOFAIL
44
45 /*
46 * The default maximum commit age, in seconds.
47 */
48 #define JBD2_DEFAULT_MAX_COMMIT_AGE 5
49
50 #ifdef CONFIG_JBD2_DEBUG
51 /*
52 * Define JBD2_EXPENSIVE_CHECKING to enable more expensive internal
53 * consistency checks. By default we don't do this unless
54 * CONFIG_JBD2_DEBUG is on.
55 */
56 #define JBD2_EXPENSIVE_CHECKING
57 void __jbd2_debug(int level, const char *file, const char *func,
58 unsigned int line, const char *fmt, ...);
59
60 #define jbd2_debug(n, fmt, a...) \
61 __jbd2_debug((n), __FILE__, __func__, __LINE__, (fmt), ##a)
62 #else
63 #define jbd2_debug(n, fmt, a...) no_printk(fmt, ##a)
64 #endif
65
66 #define JBD2_MIN_JOURNAL_BLOCKS 1024
67 #define JBD2_DEFAULT_FAST_COMMIT_BLOCKS 256
68
69 #ifdef __KERNEL__
70
71 /**
72 * typedef handle_t - The handle_t type represents a single atomic update being performed by some process.
73 *
74 * All filesystem modifications made by the process go
75 * through this handle. Recursive operations (such as quota operations)
76 * are gathered into a single update.
77 *
78 * The buffer credits field is used to account for journaled buffers
79 * being modified by the running process. To ensure that there is
80 * enough log space for all outstanding operations, we need to limit the
81 * number of outstanding buffers possible at any time. When the
82 * operation completes, any buffer credits not used are credited back to
83 * the transaction, so that at all times we know how many buffers the
84 * outstanding updates on a transaction might possibly touch.
85 *
86 * This is an opaque datatype.
87 **/
88 typedef struct jbd2_journal_handle handle_t; /* Atomic operation type */
89
90
91 /**
92 * typedef journal_t - The journal_t maintains all of the journaling state information for a single filesystem.
93 *
94 * journal_t is linked to from the fs superblock structure.
95 *
96 * We use the journal_t to keep track of all outstanding transaction
97 * activity on the filesystem, and to manage the state of the log
98 * writing process.
99 *
100 * This is an opaque datatype.
101 **/
102 typedef struct journal_s journal_t; /* Journal control structure */
103 #endif
104
105 /*
106 * Internal structures used by the logging mechanism:
107 */
108
109 #define JBD2_MAGIC_NUMBER 0xc03b3998U /* The first 4 bytes of /dev/random! */
110
111 /*
112 * On-disk structures
113 */
114
115 /*
116 * Descriptor block types:
117 */
118
119 #define JBD2_DESCRIPTOR_BLOCK 1
120 #define JBD2_COMMIT_BLOCK 2
121 #define JBD2_SUPERBLOCK_V1 3
122 #define JBD2_SUPERBLOCK_V2 4
123 #define JBD2_REVOKE_BLOCK 5
124
125 /*
126 * Standard header for all descriptor blocks:
127 */
128 typedef struct journal_header_s
129 {
130 __be32 h_magic;
131 __be32 h_blocktype;
132 __be32 h_sequence;
133 } journal_header_t;
134
135 /*
136 * Checksum types.
137 */
138 #define JBD2_CRC32_CHKSUM 1
139 #define JBD2_MD5_CHKSUM 2
140 #define JBD2_SHA1_CHKSUM 3
141 #define JBD2_CRC32C_CHKSUM 4
142
143 #define JBD2_CRC32_CHKSUM_SIZE 4
144
145 #define JBD2_CHECKSUM_BYTES (32 / sizeof(u32))
146 /*
147 * Commit block header for storing transactional checksums:
148 *
149 * NOTE: If FEATURE_COMPAT_CHECKSUM (checksum v1) is set, the h_chksum*
150 * fields are used to store a checksum of the descriptor and data blocks.
151 *
152 * If FEATURE_INCOMPAT_CSUM_V2 (checksum v2) is set, then the h_chksum
153 * field is used to store crc32c(uuid+commit_block). Each journal metadata
154 * block gets its own checksum, and data block checksums are stored in
155 * journal_block_tag (in the descriptor). The other h_chksum* fields are
156 * not used.
157 *
158 * If FEATURE_INCOMPAT_CSUM_V3 is set, the descriptor block uses
159 * journal_block_tag3_t to store a full 32-bit checksum. Everything else
160 * is the same as v2.
161 *
162 * Checksum v1, v2, and v3 are mutually exclusive features.
163 */
164 struct commit_header {
165 __be32 h_magic;
166 __be32 h_blocktype;
167 __be32 h_sequence;
168 unsigned char h_chksum_type;
169 unsigned char h_chksum_size;
170 unsigned char h_padding[2];
171 __be32 h_chksum[JBD2_CHECKSUM_BYTES];
172 __be64 h_commit_sec;
173 __be32 h_commit_nsec;
174 };
175
176 /*
177 * The block tag: used to describe a single buffer in the journal.
178 * t_blocknr_high is only used if INCOMPAT_64BIT is set, so this
179 * raw struct shouldn't be used for pointer math or sizeof() - use
180 * journal_tag_bytes(journal) instead to compute this.
181 */
182 typedef struct journal_block_tag3_s
183 {
184 __be32 t_blocknr; /* The on-disk block number */
185 __be32 t_flags; /* See below */
186 __be32 t_blocknr_high; /* most-significant high 32bits. */
187 __be32 t_checksum; /* crc32c(uuid+seq+block) */
188 } journal_block_tag3_t;
189
190 typedef struct journal_block_tag_s
191 {
192 __be32 t_blocknr; /* The on-disk block number */
193 __be16 t_checksum; /* truncated crc32c(uuid+seq+block) */
194 __be16 t_flags; /* See below */
195 __be32 t_blocknr_high; /* most-significant high 32bits. */
196 } journal_block_tag_t;
197
198 /* Tail of descriptor or revoke block, for checksumming */
199 struct jbd2_journal_block_tail {
200 __be32 t_checksum; /* crc32c(uuid+descr_block) */
201 };
202
203 /*
204 * The revoke descriptor: used on disk to describe a series of blocks to
205 * be revoked from the log
206 */
207 typedef struct jbd2_journal_revoke_header_s
208 {
209 journal_header_t r_header;
210 __be32 r_count; /* Count of bytes used in the block */
211 } jbd2_journal_revoke_header_t;
212
213 /* Definitions for the journal tag flags word: */
214 #define JBD2_FLAG_ESCAPE 1 /* on-disk block is escaped */
215 #define JBD2_FLAG_SAME_UUID 2 /* block has same uuid as previous */
216 #define JBD2_FLAG_DELETED 4 /* block deleted by this transaction */
217 #define JBD2_FLAG_LAST_TAG 8 /* last tag in this descriptor block */
218
219
220 /*
221 * The journal superblock. All fields are in big-endian byte order.
222 */
223 typedef struct journal_superblock_s
224 {
225 /* 0x0000 */
226 journal_header_t s_header;
227
228 /* 0x000C */
229 /* Static information describing the journal */
230 __be32 s_blocksize; /* journal device blocksize */
231 __be32 s_maxlen; /* total blocks in journal file */
232 __be32 s_first; /* first block of log information */
233
234 /* 0x0018 */
235 /* Dynamic information describing the current state of the log */
236 __be32 s_sequence; /* first commit ID expected in log */
237 __be32 s_start; /* blocknr of start of log */
238
239 /* 0x0020 */
240 /* Error value, as set by jbd2_journal_abort(). */
241 __be32 s_errno;
242
243 /* 0x0024 */
244 /* Remaining fields are only valid in a version-2 superblock */
245 __be32 s_feature_compat; /* compatible feature set */
246 __be32 s_feature_incompat; /* incompatible feature set */
247 __be32 s_feature_ro_compat; /* readonly-compatible feature set */
248 /* 0x0030 */
249 __u8 s_uuid[16]; /* 128-bit uuid for journal */
250
251 /* 0x0040 */
252 __be32 s_nr_users; /* Nr of filesystems sharing log */
253
254 __be32 s_dynsuper; /* Blocknr of dynamic superblock copy*/
255
256 /* 0x0048 */
257 __be32 s_max_transaction; /* Limit of journal blocks per trans.*/
258 __be32 s_max_trans_data; /* Limit of data blocks per trans. */
259
260 /* 0x0050 */
261 __u8 s_checksum_type; /* checksum type */
262 __u8 s_padding2[3];
263 /* 0x0054 */
264 __be32 s_num_fc_blks; /* Number of fast commit blocks */
265 __be32 s_head; /* blocknr of head of log, only uptodate
266 * while the filesystem is clean */
267 /* 0x005C */
268 __u32 s_padding[40];
269 __be32 s_checksum; /* crc32c(superblock) */
270
271 /* 0x0100 */
272 __u8 s_users[16*48]; /* ids of all fs'es sharing the log */
273 /* 0x0400 */
274 } journal_superblock_t;
275
276 #define JBD2_FEATURE_COMPAT_CHECKSUM 0x00000001
277
278 #define JBD2_FEATURE_INCOMPAT_REVOKE 0x00000001
279 #define JBD2_FEATURE_INCOMPAT_64BIT 0x00000002
280 #define JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT 0x00000004
281 #define JBD2_FEATURE_INCOMPAT_CSUM_V2 0x00000008
282 #define JBD2_FEATURE_INCOMPAT_CSUM_V3 0x00000010
283 #define JBD2_FEATURE_INCOMPAT_FAST_COMMIT 0x00000020
284
285 /* See "journal feature predicate functions" below */
286
287 /* Features known to this kernel version: */
288 #define JBD2_KNOWN_COMPAT_FEATURES JBD2_FEATURE_COMPAT_CHECKSUM
289 #define JBD2_KNOWN_ROCOMPAT_FEATURES 0
290 #define JBD2_KNOWN_INCOMPAT_FEATURES (JBD2_FEATURE_INCOMPAT_REVOKE | \
291 JBD2_FEATURE_INCOMPAT_64BIT | \
292 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT | \
293 JBD2_FEATURE_INCOMPAT_CSUM_V2 | \
294 JBD2_FEATURE_INCOMPAT_CSUM_V3 | \
295 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)
296
297 #ifdef __KERNEL__
298
299 #include <linux/fs.h>
300 #include <linux/sched.h>
301
302 enum jbd_state_bits {
303 BH_JBD /* Has an attached ext3 journal_head */
304 = BH_PrivateStart,
305 BH_JWrite, /* Being written to log (@@@ DEBUGGING) */
306 BH_Freed, /* Has been freed (truncated) */
307 BH_Revoked, /* Has been revoked from the log */
308 BH_RevokeValid, /* Revoked flag is valid */
309 BH_JBDDirty, /* Is dirty but journaled */
310 BH_JournalHead, /* Pins bh->b_private and jh->b_bh */
311 BH_Shadow, /* IO on shadow buffer is running */
312 BH_Verified, /* Metadata block has been verified ok */
313 BH_JBDPrivateStart, /* First bit available for private use by FS */
314 };
315
BUFFER_FNS(JBD,jbd)316 BUFFER_FNS(JBD, jbd)
317 BUFFER_FNS(JWrite, jwrite)
318 BUFFER_FNS(JBDDirty, jbddirty)
319 TAS_BUFFER_FNS(JBDDirty, jbddirty)
320 BUFFER_FNS(Revoked, revoked)
321 TAS_BUFFER_FNS(Revoked, revoked)
322 BUFFER_FNS(RevokeValid, revokevalid)
323 TAS_BUFFER_FNS(RevokeValid, revokevalid)
324 BUFFER_FNS(Freed, freed)
325 BUFFER_FNS(Shadow, shadow)
326 BUFFER_FNS(Verified, verified)
327
328 static inline struct buffer_head *jh2bh(struct journal_head *jh)
329 {
330 return jh->b_bh;
331 }
332
bh2jh(struct buffer_head * bh)333 static inline struct journal_head *bh2jh(struct buffer_head *bh)
334 {
335 return bh->b_private;
336 }
337
jbd_lock_bh_journal_head(struct buffer_head * bh)338 static inline void jbd_lock_bh_journal_head(struct buffer_head *bh)
339 {
340 bit_spin_lock(BH_JournalHead, &bh->b_state);
341 }
342
jbd_unlock_bh_journal_head(struct buffer_head * bh)343 static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh)
344 {
345 bit_spin_unlock(BH_JournalHead, &bh->b_state);
346 }
347
348 #define J_ASSERT(assert) BUG_ON(!(assert))
349
350 #define J_ASSERT_BH(bh, expr) J_ASSERT(expr)
351 #define J_ASSERT_JH(jh, expr) J_ASSERT(expr)
352
353 #if defined(JBD2_PARANOID_IOFAIL)
354 #define J_EXPECT(expr, why...) J_ASSERT(expr)
355 #define J_EXPECT_BH(bh, expr, why...) J_ASSERT_BH(bh, expr)
356 #define J_EXPECT_JH(jh, expr, why...) J_ASSERT_JH(jh, expr)
357 #else
358 #define __journal_expect(expr, why...) \
359 ({ \
360 int val = (expr); \
361 if (!val) { \
362 printk(KERN_ERR \
363 "JBD2 unexpected failure: %s: %s;\n", \
364 __func__, #expr); \
365 printk(KERN_ERR why "\n"); \
366 } \
367 val; \
368 })
369 #define J_EXPECT(expr, why...) __journal_expect(expr, ## why)
370 #define J_EXPECT_BH(bh, expr, why...) __journal_expect(expr, ## why)
371 #define J_EXPECT_JH(jh, expr, why...) __journal_expect(expr, ## why)
372 #endif
373
374 /* Flags in jbd_inode->i_flags */
375 #define __JI_COMMIT_RUNNING 0
376 #define __JI_WRITE_DATA 1
377 #define __JI_WAIT_DATA 2
378
379 /*
380 * Commit of the inode data in progress. We use this flag to protect us from
381 * concurrent deletion of inode. We cannot use reference to inode for this
382 * since we cannot afford doing last iput() on behalf of kjournald
383 */
384 #define JI_COMMIT_RUNNING (1 << __JI_COMMIT_RUNNING)
385 /* Write allocated dirty buffers in this inode before commit */
386 #define JI_WRITE_DATA (1 << __JI_WRITE_DATA)
387 /* Wait for outstanding data writes for this inode before commit */
388 #define JI_WAIT_DATA (1 << __JI_WAIT_DATA)
389
390 /**
391 * struct jbd2_inode - The jbd_inode type is the structure linking inodes in
392 * ordered mode present in a transaction so that we can sync them during commit.
393 */
394 struct jbd2_inode {
395 /**
396 * @i_transaction:
397 *
398 * Which transaction does this inode belong to? Either the running
399 * transaction or the committing one. [j_list_lock]
400 */
401 transaction_t *i_transaction;
402
403 /**
404 * @i_next_transaction:
405 *
406 * Pointer to the running transaction modifying inode's data in case
407 * there is already a committing transaction touching it. [j_list_lock]
408 */
409 transaction_t *i_next_transaction;
410
411 /**
412 * @i_list: List of inodes in the i_transaction [j_list_lock]
413 */
414 struct list_head i_list;
415
416 /**
417 * @i_vfs_inode:
418 *
419 * VFS inode this inode belongs to [constant for lifetime of structure]
420 */
421 struct inode *i_vfs_inode;
422
423 /**
424 * @i_flags: Flags of inode [j_list_lock]
425 */
426 unsigned long i_flags;
427
428 /**
429 * @i_dirty_start_page:
430 *
431 * Dirty range start in PAGE_SIZE units.
432 *
433 * The dirty range is empty if @i_dirty_start_page is greater than or
434 * equal to @i_dirty_end_page.
435 *
436 * [j_list_lock]
437 */
438 pgoff_t i_dirty_start_page;
439
440 /**
441 * @i_dirty_end_page:
442 *
443 * Dirty range end in PAGE_SIZE units (exclusive).
444 *
445 * [j_list_lock]
446 */
447 pgoff_t i_dirty_end_page;
448 };
449
450 /*
451 * Lockless readers treat start_page >= end_page as an empty range.
452 * Writers publish a new non-empty range by storing i_dirty_end_page before
453 * i_dirty_start_page.
454 */
jbd2_jinode_get_dirty_range(const struct jbd2_inode * jinode,loff_t * start,loff_t * end)455 static inline bool jbd2_jinode_get_dirty_range(const struct jbd2_inode *jinode,
456 loff_t *start, loff_t *end)
457 {
458 pgoff_t start_page = READ_ONCE(jinode->i_dirty_start_page);
459 pgoff_t end_page = READ_ONCE(jinode->i_dirty_end_page);
460
461 if (start_page >= end_page)
462 return false;
463
464 *start = (loff_t)start_page << PAGE_SHIFT;
465 *end = ((loff_t)end_page << PAGE_SHIFT) - 1;
466 return true;
467 }
468
469 struct jbd2_revoke_table_s;
470
471 /**
472 * struct jbd2_journal_handle - The jbd2_journal_handle type is the concrete
473 * type associated with handle_t.
474 * @h_transaction: Which compound transaction is this update a part of?
475 * @h_journal: Which journal handle belongs to - used iff h_reserved set.
476 * @h_rsv_handle: Handle reserved for finishing the logical operation.
477 * @h_total_credits: Number of remaining buffers we are allowed to add to
478 * journal. These are dirty buffers and revoke descriptor blocks.
479 * @h_revoke_credits: Number of remaining revoke records available for handle
480 * @h_ref: Reference count on this handle.
481 * @h_err: Field for caller's use to track errors through large fs operations.
482 * @h_sync: Flag for sync-on-close.
483 * @h_reserved: Flag for handle for reserved credits.
484 * @h_aborted: Flag indicating fatal error on handle.
485 * @h_type: For handle statistics.
486 * @h_line_no: For handle statistics.
487 * @h_start_jiffies: Handle Start time.
488 * @h_requested_credits: Holds @h_total_credits after handle is started.
489 * @h_revoke_credits_requested: Holds @h_revoke_credits after handle is started.
490 * @saved_alloc_context: Saved context while transaction is open.
491 **/
492
493 /* Docbook can't yet cope with the bit fields, but will leave the documentation
494 * in so it can be fixed later.
495 */
496
497 struct jbd2_journal_handle
498 {
499 union {
500 transaction_t *h_transaction;
501 /* Which journal handle belongs to - used iff h_reserved set */
502 journal_t *h_journal;
503 };
504
505 handle_t *h_rsv_handle;
506 int h_total_credits;
507 int h_revoke_credits;
508 int h_revoke_credits_requested;
509 int h_ref;
510 int h_err;
511
512 /* Flags [no locking] */
513 unsigned char h_sync: 1;
514 unsigned char h_reserved: 1;
515 unsigned char h_aborted: 1;
516 unsigned char h_invalid: 1;
517 unsigned char h_type;
518 unsigned short h_line_no;
519
520 unsigned long h_start_jiffies;
521 unsigned int h_requested_credits;
522
523 unsigned int saved_alloc_context;
524 };
525
526
527 /*
528 * Some stats for checkpoint phase
529 */
530 struct transaction_chp_stats_s {
531 unsigned long cs_chp_time;
532 __u32 cs_forced_to_close;
533 __u32 cs_written;
534 __u32 cs_dropped;
535 };
536
537 /* The transaction_t type is the guts of the journaling mechanism. It
538 * tracks a compound transaction through its various states:
539 *
540 * RUNNING: accepting new updates
541 * LOCKED: Updates still running but we don't accept new ones
542 * RUNDOWN: Updates are tidying up but have finished requesting
543 * new buffers to modify (state not used for now)
544 * FLUSH: All updates complete, but we are still writing to disk
545 * COMMIT: All data on disk, writing commit record
546 * FINISHED: We still have to keep the transaction for checkpointing.
547 *
548 * The transaction keeps track of all of the buffers modified by a
549 * running transaction, and all of the buffers committed but not yet
550 * flushed to home for finished transactions.
551 * (Locking Documentation improved by LockDoc)
552 */
553
554 /*
555 * Lock ranking:
556 *
557 * j_list_lock
558 * ->jbd_lock_bh_journal_head() (This is "innermost")
559 *
560 * j_state_lock
561 * ->b_state_lock
562 *
563 * b_state_lock
564 * ->j_list_lock
565 *
566 * j_state_lock
567 * ->j_list_lock (journal_unmap_buffer)
568 *
569 */
570
571 struct transaction_s
572 {
573 /* Pointer to the journal for this transaction. [no locking] */
574 journal_t *t_journal;
575
576 /* Sequence number for this transaction [no locking] */
577 tid_t t_tid;
578
579 /*
580 * Transaction's current state
581 * [no locking - only kjournald2 alters this]
582 * [j_list_lock] guards transition of a transaction into T_FINISHED
583 * state and subsequent call of __jbd2_journal_drop_transaction()
584 * FIXME: needs barriers
585 * KLUDGE: [use j_state_lock]
586 */
587 enum {
588 T_RUNNING,
589 T_LOCKED,
590 T_SWITCH,
591 T_FLUSH,
592 T_COMMIT,
593 T_COMMIT_DFLUSH,
594 T_COMMIT_JFLUSH,
595 T_COMMIT_CALLBACK,
596 T_FINISHED
597 } t_state;
598
599 /*
600 * Where in the log does this transaction's commit start? [no locking]
601 */
602 unsigned long t_log_start;
603
604 /*
605 * Number of buffers on the t_buffers list [j_list_lock, no locks
606 * needed for jbd2 thread]
607 */
608 int t_nr_buffers;
609
610 /*
611 * Doubly-linked circular list of all buffers reserved but not yet
612 * modified by this transaction [j_list_lock, no locks needed fo
613 * jbd2 thread]
614 */
615 struct journal_head *t_reserved_list;
616
617 /*
618 * Doubly-linked circular list of all metadata buffers owned by this
619 * transaction [j_list_lock, no locks needed for jbd2 thread]
620 */
621 struct journal_head *t_buffers;
622
623 /*
624 * Doubly-linked circular list of all forget buffers (superseded
625 * buffers which we can un-checkpoint once this transaction commits)
626 * [j_list_lock]
627 */
628 struct journal_head *t_forget;
629
630 /*
631 * Doubly-linked circular list of all buffers still to be flushed before
632 * this transaction can be checkpointed. [j_list_lock]
633 */
634 struct journal_head *t_checkpoint_list;
635
636 /*
637 * Doubly-linked circular list of metadata buffers being
638 * shadowed by log IO. The IO buffers on the iobuf list and
639 * the shadow buffers on this list match each other one for
640 * one at all times. [j_list_lock, no locks needed for jbd2
641 * thread]
642 */
643 struct journal_head *t_shadow_list;
644
645 /*
646 * List of inodes associated with the transaction; e.g., ext4 uses
647 * this to track inodes in data=ordered and data=journal mode that
648 * need special handling on transaction commit; also used by ocfs2.
649 * [j_list_lock]
650 */
651 struct list_head t_inode_list;
652
653 /*
654 * Longest time some handle had to wait for running transaction
655 */
656 unsigned long t_max_wait;
657
658 /*
659 * When transaction started
660 */
661 unsigned long t_start;
662
663 /*
664 * When commit was requested [j_state_lock]
665 */
666 unsigned long t_requested;
667
668 /*
669 * Checkpointing stats [j_list_lock]
670 */
671 struct transaction_chp_stats_s t_chp_stats;
672
673 /*
674 * Number of outstanding updates running on this transaction
675 * [none]
676 */
677 atomic_t t_updates;
678
679 /*
680 * Number of blocks reserved for this transaction in the journal.
681 * This is including all credits reserved when starting transaction
682 * handles as well as all journal descriptor blocks needed for this
683 * transaction. [none]
684 */
685 atomic_t t_outstanding_credits;
686
687 /*
688 * Number of revoke records for this transaction added by already
689 * stopped handles. [none]
690 */
691 atomic_t t_outstanding_revokes;
692
693 /*
694 * How many handles used this transaction? [none]
695 */
696 atomic_t t_handle_count;
697
698 /*
699 * Forward and backward links for the circular list of all transactions
700 * awaiting checkpoint. [j_list_lock]
701 */
702 transaction_t *t_cpnext, *t_cpprev;
703
704 /*
705 * When will the transaction expire (become due for commit), in jiffies?
706 * [no locking]
707 */
708 unsigned long t_expires;
709
710 /*
711 * When this transaction started, in nanoseconds [no locking]
712 */
713 ktime_t t_start_time;
714
715 /*
716 * This transaction is being forced and some process is
717 * waiting for it to finish.
718 */
719 unsigned int t_synchronous_commit:1;
720
721 /* Disk flush needs to be sent to fs partition [no locking] */
722 int t_need_data_flush;
723 };
724
725 struct transaction_run_stats_s {
726 unsigned long rs_wait;
727 unsigned long rs_request_delay;
728 unsigned long rs_running;
729 unsigned long rs_locked;
730 unsigned long rs_flushing;
731 unsigned long rs_logging;
732
733 __u32 rs_handle_count;
734 __u32 rs_blocks;
735 __u32 rs_blocks_logged;
736 };
737
738 struct transaction_stats_s {
739 unsigned long ts_tid;
740 unsigned long ts_requested;
741 struct transaction_run_stats_s run;
742 };
743
744 static inline unsigned long
jbd2_time_diff(unsigned long start,unsigned long end)745 jbd2_time_diff(unsigned long start, unsigned long end)
746 {
747 if (end >= start)
748 return end - start;
749
750 return end + (MAX_JIFFY_OFFSET - start);
751 }
752
753 #define JBD2_NR_BATCH 64
754
755 enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
756
757 #define JBD2_FC_REPLAY_STOP 0
758 #define JBD2_FC_REPLAY_CONTINUE 1
759
760 /**
761 * struct journal_s - The journal_s type is the concrete type associated with
762 * journal_t.
763 */
764 struct journal_s
765 {
766 /**
767 * @j_flags: General journaling state flags [j_state_lock,
768 * no lock for quick racy checks]
769 */
770 unsigned long j_flags;
771
772 /**
773 * @j_errno:
774 *
775 * Is there an outstanding uncleared error on the journal (from a prior
776 * abort)? [j_state_lock]
777 */
778 int j_errno;
779
780 /**
781 * @j_abort_mutex: Lock the whole aborting procedure.
782 */
783 struct mutex j_abort_mutex;
784
785 /**
786 * @j_sb_buffer: The first part of the superblock buffer.
787 */
788 struct buffer_head *j_sb_buffer;
789
790 /**
791 * @j_superblock: The second part of the superblock buffer.
792 */
793 journal_superblock_t *j_superblock;
794
795 /**
796 * @j_state_lock: Protect the various scalars in the journal.
797 */
798 rwlock_t j_state_lock;
799
800 /**
801 * @j_barrier_count:
802 *
803 * Number of processes waiting to create a barrier lock [j_state_lock,
804 * no lock for quick racy checks]
805 */
806 int j_barrier_count;
807
808 /**
809 * @j_barrier: The barrier lock itself.
810 */
811 struct mutex j_barrier;
812
813 /**
814 * @j_running_transaction:
815 *
816 * Transactions: The current running transaction...
817 * [j_state_lock, no lock for quick racy checks] [caller holding
818 * open handle]
819 */
820 transaction_t *j_running_transaction;
821
822 /**
823 * @j_committing_transaction:
824 *
825 * the transaction we are pushing to disk
826 * [j_state_lock] [caller holding open handle]
827 */
828 transaction_t *j_committing_transaction;
829
830 /**
831 * @j_checkpoint_transactions:
832 *
833 * ... and a linked circular list of all transactions waiting for
834 * checkpointing. [j_list_lock]
835 */
836 transaction_t *j_checkpoint_transactions;
837
838 /**
839 * @j_wait_transaction_locked:
840 *
841 * Wait queue for waiting for a locked transaction to start committing,
842 * or for a barrier lock to be released.
843 */
844 wait_queue_head_t j_wait_transaction_locked;
845
846 /**
847 * @j_wait_done_commit: Wait queue for waiting for commit to complete.
848 */
849 wait_queue_head_t j_wait_done_commit;
850
851 /**
852 * @j_wait_commit: Wait queue to trigger commit.
853 */
854 wait_queue_head_t j_wait_commit;
855
856 /**
857 * @j_wait_updates: Wait queue to wait for updates to complete.
858 */
859 wait_queue_head_t j_wait_updates;
860
861 /**
862 * @j_wait_reserved:
863 *
864 * Wait queue to wait for reserved buffer credits to drop.
865 */
866 wait_queue_head_t j_wait_reserved;
867
868 /**
869 * @j_fc_wait:
870 *
871 * Wait queue to wait for completion of async fast commits.
872 */
873 wait_queue_head_t j_fc_wait;
874
875 /**
876 * @j_checkpoint_mutex:
877 *
878 * Semaphore for locking against concurrent checkpoints.
879 */
880 struct mutex j_checkpoint_mutex;
881
882 /**
883 * @j_chkpt_bhs:
884 *
885 * List of buffer heads used by the checkpoint routine. This
886 * was moved from jbd2_log_do_checkpoint() to reduce stack
887 * usage. Access to this array is controlled by the
888 * @j_checkpoint_mutex. [j_checkpoint_mutex]
889 */
890 struct buffer_head *j_chkpt_bhs[JBD2_NR_BATCH];
891
892 /**
893 * @j_shrinker:
894 *
895 * Journal head shrinker, reclaim buffer's journal head which
896 * has been written back.
897 */
898 struct shrinker *j_shrinker;
899
900 /**
901 * @j_checkpoint_jh_count:
902 *
903 * Number of journal buffers on the checkpoint list. [j_list_lock]
904 */
905 struct percpu_counter j_checkpoint_jh_count;
906
907 /**
908 * @j_shrink_transaction:
909 *
910 * Record next transaction will shrink on the checkpoint list.
911 * [j_list_lock]
912 */
913 transaction_t *j_shrink_transaction;
914
915 /**
916 * @j_head:
917 *
918 * Journal head: identifies the first unused block in the journal.
919 * [j_state_lock]
920 */
921 unsigned long j_head;
922
923 /**
924 * @j_tail:
925 *
926 * Journal tail: identifies the oldest still-used block in the journal.
927 * [j_state_lock]
928 */
929 unsigned long j_tail;
930
931 /**
932 * @j_free:
933 *
934 * Journal free: how many free blocks are there in the journal?
935 * [j_state_lock]
936 */
937 unsigned long j_free;
938
939 /**
940 * @j_first:
941 *
942 * The block number of the first usable block in the journal
943 * [j_state_lock].
944 */
945 unsigned long j_first;
946
947 /**
948 * @j_last:
949 *
950 * The block number one beyond the last usable block in the journal
951 * [j_state_lock].
952 */
953 unsigned long j_last;
954
955 /**
956 * @j_fc_first:
957 *
958 * The block number of the first fast commit block in the journal
959 * [j_state_lock].
960 */
961 unsigned long j_fc_first;
962
963 /**
964 * @j_fc_off:
965 *
966 * Number of fast commit blocks currently allocated. Accessed only
967 * during fast commit. Currently only process can do fast commit, so
968 * this field is not protected by any lock.
969 */
970 unsigned long j_fc_off;
971
972 /**
973 * @j_fc_last:
974 *
975 * The block number one beyond the last fast commit block in the journal
976 * [j_state_lock].
977 */
978 unsigned long j_fc_last;
979
980 /**
981 * @j_dev: Device where we store the journal.
982 */
983 struct block_device *j_dev;
984
985 /**
986 * @j_blocksize: Block size for the location where we store the journal.
987 */
988 int j_blocksize;
989
990 /**
991 * @j_blk_offset:
992 *
993 * Starting block offset into the device where we store the journal.
994 */
995 unsigned long long j_blk_offset;
996
997 /**
998 * @j_devname: Journal device name.
999 */
1000 char j_devname[BDEVNAME_SIZE+24];
1001
1002 /**
1003 * @j_fs_dev:
1004 *
1005 * Device which holds the client fs. For internal journal this will be
1006 * equal to j_dev.
1007 */
1008 struct block_device *j_fs_dev;
1009
1010 /**
1011 * @j_fs_dev_wb_err:
1012 *
1013 * Records the errseq of the client fs's backing block device.
1014 */
1015 errseq_t j_fs_dev_wb_err;
1016
1017 /**
1018 * @j_total_len: Total maximum capacity of the journal region on disk.
1019 */
1020 unsigned int j_total_len;
1021
1022 /**
1023 * @j_reserved_credits:
1024 *
1025 * Number of buffers reserved from the running transaction.
1026 */
1027 atomic_t j_reserved_credits;
1028
1029 /**
1030 * @j_list_lock: Protects the buffer lists and internal buffer state.
1031 */
1032 spinlock_t j_list_lock;
1033
1034 /**
1035 * @j_inode:
1036 *
1037 * Optional inode where we store the journal. If present, all
1038 * journal block numbers are mapped into this inode via bmap().
1039 */
1040 struct inode *j_inode;
1041
1042 /**
1043 * @j_tail_sequence:
1044 *
1045 * Sequence number of the oldest transaction in the log [j_state_lock]
1046 */
1047 tid_t j_tail_sequence;
1048
1049 /**
1050 * @j_transaction_sequence:
1051 *
1052 * Sequence number of the next transaction to grant [j_state_lock]
1053 */
1054 tid_t j_transaction_sequence;
1055
1056 /**
1057 * @j_commit_sequence:
1058 *
1059 * Sequence number of the most recently committed transaction
1060 * [j_state_lock, no lock for quick racy checks]
1061 */
1062 tid_t j_commit_sequence;
1063
1064 /**
1065 * @j_commit_request:
1066 *
1067 * Sequence number of the most recent transaction wanting commit
1068 * [j_state_lock, no lock for quick racy checks]
1069 */
1070 tid_t j_commit_request;
1071
1072 /**
1073 * @j_uuid:
1074 *
1075 * Journal uuid: identifies the object (filesystem, LVM volume etc)
1076 * backed by this journal. This will eventually be replaced by an array
1077 * of uuids, allowing us to index multiple devices within a single
1078 * journal and to perform atomic updates across them.
1079 */
1080 __u8 j_uuid[16];
1081
1082 /**
1083 * @j_task: Pointer to the current commit thread for this journal.
1084 */
1085 struct task_struct *j_task;
1086
1087 /**
1088 * @j_max_transaction_buffers:
1089 *
1090 * Maximum number of metadata buffers to allow in a single compound
1091 * commit transaction.
1092 */
1093 int j_max_transaction_buffers;
1094
1095 /**
1096 * @j_revoke_records_per_block:
1097 *
1098 * Number of revoke records that fit in one descriptor block.
1099 */
1100 int j_revoke_records_per_block;
1101
1102 /**
1103 * @j_transaction_overhead_buffers:
1104 *
1105 * Number of blocks each transaction needs for its own bookkeeping
1106 */
1107 int j_transaction_overhead_buffers;
1108
1109 /**
1110 * @j_commit_interval:
1111 *
1112 * What is the maximum transaction lifetime before we begin a commit?
1113 */
1114 unsigned long j_commit_interval;
1115
1116 /**
1117 * @j_commit_timer: The timer used to wakeup the commit thread.
1118 */
1119 struct timer_list j_commit_timer;
1120
1121 /**
1122 * @j_revoke_lock: Protect the revoke table.
1123 */
1124 spinlock_t j_revoke_lock;
1125
1126 /**
1127 * @j_revoke:
1128 *
1129 * The revoke table - maintains the list of revoked blocks in the
1130 * current transaction.
1131 */
1132 struct jbd2_revoke_table_s *j_revoke;
1133
1134 /**
1135 * @j_revoke_table: Alternate revoke tables for j_revoke.
1136 */
1137 struct jbd2_revoke_table_s *j_revoke_table[2];
1138
1139 /**
1140 * @j_wbuf: Array of bhs for jbd2_journal_commit_transaction.
1141 */
1142 struct buffer_head **j_wbuf;
1143
1144 /**
1145 * @j_fc_wbuf: Array of fast commit bhs for fast commit. Accessed only
1146 * during a fast commit. Currently only process can do fast commit, so
1147 * this field is not protected by any lock.
1148 */
1149 struct buffer_head **j_fc_wbuf;
1150
1151 /**
1152 * @j_wbufsize:
1153 *
1154 * Size of @j_wbuf array.
1155 */
1156 int j_wbufsize;
1157
1158 /**
1159 * @j_fc_wbufsize:
1160 *
1161 * Size of @j_fc_wbuf array.
1162 */
1163 int j_fc_wbufsize;
1164
1165 /**
1166 * @j_last_sync_writer:
1167 *
1168 * The pid of the last person to run a synchronous operation
1169 * through the journal.
1170 */
1171 pid_t j_last_sync_writer;
1172
1173 /**
1174 * @j_average_commit_time:
1175 *
1176 * The average amount of time in nanoseconds it takes to commit a
1177 * transaction to disk. [j_state_lock]
1178 */
1179 u64 j_average_commit_time;
1180
1181 /**
1182 * @j_min_batch_time:
1183 *
1184 * Minimum time that we should wait for additional filesystem operations
1185 * to get batched into a synchronous handle in microseconds.
1186 */
1187 u32 j_min_batch_time;
1188
1189 /**
1190 * @j_max_batch_time:
1191 *
1192 * Maximum time that we should wait for additional filesystem operations
1193 * to get batched into a synchronous handle in microseconds.
1194 */
1195 u32 j_max_batch_time;
1196
1197 /**
1198 * @j_commit_callback:
1199 *
1200 * This function is called when a transaction is closed.
1201 */
1202 void (*j_commit_callback)(journal_t *,
1203 transaction_t *);
1204
1205 /**
1206 * @j_submit_inode_data_buffers:
1207 *
1208 * This function is called for all inodes associated with the
1209 * committing transaction marked with JI_WRITE_DATA flag
1210 * before we start to write out the transaction to the journal.
1211 */
1212 int (*j_submit_inode_data_buffers)
1213 (struct jbd2_inode *);
1214
1215 /**
1216 * @j_finish_inode_data_buffers:
1217 *
1218 * This function is called for all inodes associated with the
1219 * committing transaction marked with JI_WAIT_DATA flag
1220 * after we have written the transaction to the journal
1221 * but before we write out the commit block.
1222 */
1223 int (*j_finish_inode_data_buffers)
1224 (struct jbd2_inode *);
1225
1226 /*
1227 * Journal statistics
1228 */
1229
1230 /**
1231 * @j_history_lock: Protect the transactions statistics history.
1232 */
1233 spinlock_t j_history_lock;
1234
1235 /**
1236 * @j_proc_entry: procfs entry for the jbd statistics directory.
1237 */
1238 struct proc_dir_entry *j_proc_entry;
1239
1240 /**
1241 * @j_stats: Overall statistics.
1242 */
1243 struct transaction_stats_s j_stats;
1244
1245 /**
1246 * @j_failed_commit: Failed journal commit ID.
1247 */
1248 unsigned int j_failed_commit;
1249
1250 /**
1251 * @j_private:
1252 *
1253 * An opaque pointer to fs-private information. ext3 puts its
1254 * superblock pointer here.
1255 */
1256 void *j_private;
1257
1258 /**
1259 * @j_csum_seed:
1260 *
1261 * Precomputed journal UUID checksum for seeding other checksums.
1262 */
1263 __u32 j_csum_seed;
1264
1265 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1266 /**
1267 * @j_trans_commit_map:
1268 *
1269 * Lockdep entity to track transaction commit dependencies. Handles
1270 * hold this "lock" for read, when we wait for commit, we acquire the
1271 * "lock" for writing. This matches the properties of jbd2 journalling
1272 * where the running transaction has to wait for all handles to be
1273 * dropped to commit that transaction and also acquiring a handle may
1274 * require transaction commit to finish.
1275 */
1276 struct lockdep_map j_trans_commit_map;
1277 #endif
1278 /**
1279 * @jbd2_trans_commit_key:
1280 *
1281 * "struct lock_class_key" for @j_trans_commit_map
1282 */
1283 struct lock_class_key jbd2_trans_commit_key;
1284
1285 /**
1286 * @j_fc_cleanup_callback:
1287 *
1288 * Clean-up after fast commit or full commit. JBD2 calls this function
1289 * after every commit operation.
1290 */
1291 void (*j_fc_cleanup_callback)(struct journal_s *journal, int full, tid_t tid);
1292
1293 /**
1294 * @j_fc_replay_callback:
1295 *
1296 * File-system specific function that performs replay of a fast
1297 * commit. JBD2 calls this function for each fast commit block found in
1298 * the journal. This function should return JBD2_FC_REPLAY_CONTINUE
1299 * to indicate that the block was processed correctly and more fast
1300 * commit replay should continue. Return value of JBD2_FC_REPLAY_STOP
1301 * indicates the end of replay (no more blocks remaining). A negative
1302 * return value indicates error.
1303 */
1304 int (*j_fc_replay_callback)(struct journal_s *journal,
1305 struct buffer_head *bh,
1306 enum passtype pass, int off,
1307 tid_t expected_commit_id);
1308
1309 /**
1310 * @j_bmap:
1311 *
1312 * Bmap function that should be used instead of the generic
1313 * VFS bmap function.
1314 */
1315 int (*j_bmap)(struct journal_s *journal, sector_t *block);
1316 };
1317
1318 #define jbd2_might_wait_for_commit(j) \
1319 do { \
1320 rwsem_acquire(&j->j_trans_commit_map, 0, 0, _THIS_IP_); \
1321 rwsem_release(&j->j_trans_commit_map, _THIS_IP_); \
1322 } while (0)
1323
1324 /*
1325 * We can support any known requested features iff the
1326 * superblock is not in version 1. Otherwise we fail to support any
1327 * extended sb features.
1328 */
jbd2_format_support_feature(journal_t * j)1329 static inline bool jbd2_format_support_feature(journal_t *j)
1330 {
1331 return j->j_superblock->s_header.h_blocktype !=
1332 cpu_to_be32(JBD2_SUPERBLOCK_V1);
1333 }
1334
1335 /* journal feature predicate functions */
1336 #define JBD2_FEATURE_COMPAT_FUNCS(name, flagname) \
1337 static inline bool jbd2_has_feature_##name(journal_t *j) \
1338 { \
1339 return (jbd2_format_support_feature(j) && \
1340 ((j)->j_superblock->s_feature_compat & \
1341 cpu_to_be32(JBD2_FEATURE_COMPAT_##flagname)) != 0); \
1342 } \
1343 static inline void jbd2_set_feature_##name(journal_t *j) \
1344 { \
1345 (j)->j_superblock->s_feature_compat |= \
1346 cpu_to_be32(JBD2_FEATURE_COMPAT_##flagname); \
1347 } \
1348 static inline void jbd2_clear_feature_##name(journal_t *j) \
1349 { \
1350 (j)->j_superblock->s_feature_compat &= \
1351 ~cpu_to_be32(JBD2_FEATURE_COMPAT_##flagname); \
1352 }
1353
1354 #define JBD2_FEATURE_RO_COMPAT_FUNCS(name, flagname) \
1355 static inline bool jbd2_has_feature_##name(journal_t *j) \
1356 { \
1357 return (jbd2_format_support_feature(j) && \
1358 ((j)->j_superblock->s_feature_ro_compat & \
1359 cpu_to_be32(JBD2_FEATURE_RO_COMPAT_##flagname)) != 0); \
1360 } \
1361 static inline void jbd2_set_feature_##name(journal_t *j) \
1362 { \
1363 (j)->j_superblock->s_feature_ro_compat |= \
1364 cpu_to_be32(JBD2_FEATURE_RO_COMPAT_##flagname); \
1365 } \
1366 static inline void jbd2_clear_feature_##name(journal_t *j) \
1367 { \
1368 (j)->j_superblock->s_feature_ro_compat &= \
1369 ~cpu_to_be32(JBD2_FEATURE_RO_COMPAT_##flagname); \
1370 }
1371
1372 #define JBD2_FEATURE_INCOMPAT_FUNCS(name, flagname) \
1373 static inline bool jbd2_has_feature_##name(journal_t *j) \
1374 { \
1375 return (jbd2_format_support_feature(j) && \
1376 ((j)->j_superblock->s_feature_incompat & \
1377 cpu_to_be32(JBD2_FEATURE_INCOMPAT_##flagname)) != 0); \
1378 } \
1379 static inline void jbd2_set_feature_##name(journal_t *j) \
1380 { \
1381 (j)->j_superblock->s_feature_incompat |= \
1382 cpu_to_be32(JBD2_FEATURE_INCOMPAT_##flagname); \
1383 } \
1384 static inline void jbd2_clear_feature_##name(journal_t *j) \
1385 { \
1386 (j)->j_superblock->s_feature_incompat &= \
1387 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_##flagname); \
1388 }
1389
1390 JBD2_FEATURE_COMPAT_FUNCS(checksum, CHECKSUM)
1391
1392 JBD2_FEATURE_INCOMPAT_FUNCS(revoke, REVOKE)
1393 JBD2_FEATURE_INCOMPAT_FUNCS(64bit, 64BIT)
1394 JBD2_FEATURE_INCOMPAT_FUNCS(async_commit, ASYNC_COMMIT)
1395 JBD2_FEATURE_INCOMPAT_FUNCS(csum2, CSUM_V2)
1396 JBD2_FEATURE_INCOMPAT_FUNCS(csum3, CSUM_V3)
1397 JBD2_FEATURE_INCOMPAT_FUNCS(fast_commit, FAST_COMMIT)
1398
1399 /* Journal high priority write IO operation flags */
1400 #define JBD2_JOURNAL_REQ_FLAGS (REQ_META | REQ_SYNC | REQ_IDLE)
1401
1402 /*
1403 * Journal flag definitions
1404 */
1405 #define JBD2_UNMOUNT 0x001 /* Journal thread is being destroyed */
1406 #define JBD2_ABORT 0x002 /* Journaling has been aborted for errors. */
1407 #define JBD2_ACK_ERR 0x004 /* The errno in the sb has been acked */
1408 #define JBD2_FLUSHED 0x008 /* The journal superblock has been flushed */
1409 #define JBD2_LOADED 0x010 /* The journal superblock has been loaded */
1410 #define JBD2_BARRIER 0x020 /* Use IDE barriers */
1411 #define JBD2_CYCLE_RECORD 0x080 /* Journal cycled record log on
1412 * clean and empty filesystem
1413 * logging area */
1414 #define JBD2_FAST_COMMIT_ONGOING 0x100 /* Fast commit is ongoing */
1415 #define JBD2_FULL_COMMIT_ONGOING 0x200 /* Full commit is ongoing */
1416 #define JBD2_JOURNAL_FLUSH_DISCARD 0x0001
1417 #define JBD2_JOURNAL_FLUSH_ZEROOUT 0x0002
1418 #define JBD2_JOURNAL_FLUSH_VALID (JBD2_JOURNAL_FLUSH_DISCARD | \
1419 JBD2_JOURNAL_FLUSH_ZEROOUT)
1420
1421 /*
1422 * Function declarations for the journaling transaction and buffer
1423 * management
1424 */
1425
1426 /* Filing buffers */
1427 extern bool __jbd2_journal_refile_buffer(struct journal_head *);
1428 extern void jbd2_journal_refile_buffer(journal_t *, struct journal_head *);
1429 extern void __jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int);
1430 extern void jbd2_journal_file_buffer(struct journal_head *, transaction_t *, int);
jbd2_file_log_bh(struct list_head * head,struct buffer_head * bh)1431 static inline void jbd2_file_log_bh(struct list_head *head, struct buffer_head *bh)
1432 {
1433 list_add_tail(&bh->b_assoc_buffers, head);
1434 }
jbd2_unfile_log_bh(struct buffer_head * bh)1435 static inline void jbd2_unfile_log_bh(struct buffer_head *bh)
1436 {
1437 list_del_init(&bh->b_assoc_buffers);
1438 }
1439
1440 /* Log buffer allocation */
1441 struct buffer_head *jbd2_journal_get_descriptor_buffer(transaction_t *, int);
1442 void jbd2_descriptor_block_csum_set(journal_t *, struct buffer_head *);
1443 int jbd2_journal_next_log_block(journal_t *, unsigned long long *);
1444 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1445 unsigned long *block);
1446 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block);
1447 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block);
1448
1449 /* Commit management */
1450 extern void jbd2_journal_commit_transaction(journal_t *);
1451
1452 /* Checkpoint list management */
1453 enum jbd2_shrink_type {JBD2_SHRINK_DESTROY, JBD2_SHRINK_BUSY_STOP, JBD2_SHRINK_BUSY_SKIP};
1454
1455 void __jbd2_journal_clean_checkpoint_list(journal_t *journal, enum jbd2_shrink_type type);
1456 unsigned long jbd2_journal_shrink_checkpoint_list(journal_t *journal, unsigned long *nr_to_scan);
1457 int __jbd2_journal_remove_checkpoint(struct journal_head *);
1458 int jbd2_journal_try_remove_checkpoint(struct journal_head *jh);
1459 void jbd2_journal_destroy_checkpoint(journal_t *journal);
1460 void __jbd2_journal_insert_checkpoint(struct journal_head *, transaction_t *);
1461
1462
1463 /*
1464 * Triggers
1465 */
1466
1467 struct jbd2_buffer_trigger_type {
1468 /*
1469 * Fired a the moment data to write to the journal are known to be
1470 * stable - so either at the moment b_frozen_data is created or just
1471 * before a buffer is written to the journal. mapped_data is a mapped
1472 * buffer that is the frozen data for commit.
1473 */
1474 void (*t_frozen)(struct jbd2_buffer_trigger_type *type,
1475 struct buffer_head *bh, void *mapped_data,
1476 size_t size);
1477
1478 /*
1479 * Fired during journal abort for dirty buffers that will not be
1480 * committed.
1481 */
1482 void (*t_abort)(struct jbd2_buffer_trigger_type *type,
1483 struct buffer_head *bh);
1484 };
1485
1486 extern void jbd2_buffer_frozen_trigger(struct journal_head *jh,
1487 void *mapped_data,
1488 struct jbd2_buffer_trigger_type *triggers);
1489 extern void jbd2_buffer_abort_trigger(struct journal_head *jh,
1490 struct jbd2_buffer_trigger_type *triggers);
1491
1492 /* Buffer IO */
1493 extern int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
1494 struct journal_head *jh_in,
1495 struct buffer_head **bh_out,
1496 sector_t blocknr);
1497
1498 /* Transaction cache support */
1499 extern void jbd2_journal_destroy_transaction_cache(void);
1500 extern int __init jbd2_journal_init_transaction_cache(void);
1501 extern void jbd2_journal_free_transaction(transaction_t *);
1502
1503 /*
1504 * Journal locking.
1505 *
1506 * We need to lock the journal during transaction state changes so that nobody
1507 * ever tries to take a handle on the running transaction while we are in the
1508 * middle of moving it to the commit phase. j_state_lock does this.
1509 *
1510 * Note that the locking is completely interrupt unsafe. We never touch
1511 * journal structures from interrupts.
1512 */
1513
journal_current_handle(void)1514 static inline handle_t *journal_current_handle(void)
1515 {
1516 return current->journal_info;
1517 }
1518
1519 /* The journaling code user interface:
1520 *
1521 * Create and destroy handles
1522 * Register buffer modifications against the current transaction.
1523 */
1524
1525 extern handle_t *jbd2_journal_start(journal_t *, int nblocks);
1526 extern handle_t *jbd2__journal_start(journal_t *, int blocks, int rsv_blocks,
1527 int revoke_records, gfp_t gfp_mask,
1528 unsigned int type, unsigned int line_no);
1529 extern int jbd2_journal_restart(handle_t *, int nblocks);
1530 extern int jbd2__journal_restart(handle_t *, int nblocks,
1531 int revoke_records, gfp_t gfp_mask);
1532 extern int jbd2_journal_start_reserved(handle_t *handle,
1533 unsigned int type, unsigned int line_no);
1534 extern void jbd2_journal_free_reserved(handle_t *handle);
1535 extern int jbd2_journal_extend(handle_t *handle, int nblocks,
1536 int revoke_records);
1537 extern int jbd2_journal_get_write_access(handle_t *, struct buffer_head *);
1538 extern int jbd2_journal_get_create_access (handle_t *, struct buffer_head *);
1539 extern int jbd2_journal_get_undo_access(handle_t *, struct buffer_head *);
1540 void jbd2_journal_set_triggers(struct buffer_head *,
1541 struct jbd2_buffer_trigger_type *type);
1542 extern int jbd2_journal_dirty_metadata (handle_t *, struct buffer_head *);
1543 extern int jbd2_journal_forget (handle_t *, struct buffer_head *);
1544 int jbd2_journal_invalidate_folio(journal_t *, struct folio *,
1545 size_t offset, size_t length);
1546 bool jbd2_journal_try_to_free_buffers(journal_t *journal, struct folio *folio);
1547 extern int jbd2_journal_stop(handle_t *);
1548 extern int jbd2_journal_flush(journal_t *journal, unsigned int flags);
1549 extern void jbd2_journal_lock_updates (journal_t *);
1550 extern void jbd2_journal_unlock_updates (journal_t *);
1551
1552 void jbd2_journal_wait_updates(journal_t *);
1553
1554 extern journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1555 struct block_device *fs_dev,
1556 unsigned long long start, int len, int bsize);
1557 extern journal_t * jbd2_journal_init_inode (struct inode *);
1558 extern int jbd2_journal_update_format (journal_t *);
1559 extern int jbd2_journal_check_used_features
1560 (journal_t *, unsigned long, unsigned long, unsigned long);
1561 extern int jbd2_journal_check_available_features
1562 (journal_t *, unsigned long, unsigned long, unsigned long);
1563 extern int jbd2_journal_set_features
1564 (journal_t *, unsigned long, unsigned long, unsigned long);
1565 extern void jbd2_journal_clear_features
1566 (journal_t *, unsigned long, unsigned long, unsigned long);
1567 extern int jbd2_journal_load (journal_t *journal);
1568 extern int jbd2_journal_destroy (journal_t *);
1569 extern int jbd2_journal_recover (journal_t *journal);
1570 extern int jbd2_journal_wipe (journal_t *, int);
1571 extern int jbd2_journal_skip_recovery (journal_t *);
1572 extern void jbd2_journal_update_sb_errno(journal_t *);
1573 extern int jbd2_journal_update_sb_log_tail (journal_t *, tid_t,
1574 unsigned long, blk_opf_t);
1575 extern void jbd2_journal_abort (journal_t *, int);
1576 extern int jbd2_journal_errno (journal_t *);
1577 extern void jbd2_journal_ack_err (journal_t *);
1578 extern int jbd2_journal_clear_err (journal_t *);
1579 extern int jbd2_journal_bmap(journal_t *, unsigned long, unsigned long long *);
1580 extern int jbd2_journal_force_commit(journal_t *);
1581 extern int jbd2_journal_force_commit_nested(journal_t *);
1582 extern int jbd2_journal_inode_ranged_write(handle_t *handle,
1583 struct jbd2_inode *inode, loff_t start_byte,
1584 loff_t length);
1585 extern int jbd2_journal_inode_ranged_wait(handle_t *handle,
1586 struct jbd2_inode *inode, loff_t start_byte,
1587 loff_t length);
1588 extern int jbd2_journal_finish_inode_data_buffers(
1589 struct jbd2_inode *jinode);
1590 extern int jbd2_journal_begin_ordered_truncate(journal_t *journal,
1591 struct jbd2_inode *inode, loff_t new_size);
1592 extern void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode);
1593 extern void jbd2_journal_release_jbd_inode(journal_t *journal, struct jbd2_inode *jinode);
1594
1595 /*
1596 * journal_head management
1597 */
1598 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh);
1599 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh);
1600 void jbd2_journal_put_journal_head(struct journal_head *jh);
1601
1602 /*
1603 * handle management
1604 */
1605 extern struct kmem_cache *jbd2_handle_cache;
1606
1607 /*
1608 * This specialized allocator has to be a macro for its allocations to be
1609 * accounted separately (to have a separate alloc_tag). The typecast is
1610 * intentional to enforce typesafety.
1611 */
1612 #define jbd2_alloc_handle(_gfp_flags) \
1613 ((handle_t *)kmem_cache_zalloc(jbd2_handle_cache, _gfp_flags))
1614
jbd2_free_handle(handle_t * handle)1615 static inline void jbd2_free_handle(handle_t *handle)
1616 {
1617 kmem_cache_free(jbd2_handle_cache, handle);
1618 }
1619
1620 /*
1621 * jbd2_inode management (optional, for those file systems that want to use
1622 * dynamically allocated jbd2_inode structures)
1623 */
1624 extern struct kmem_cache *jbd2_inode_cache;
1625
1626 /*
1627 * This specialized allocator has to be a macro for its allocations to be
1628 * accounted separately (to have a separate alloc_tag). The typecast is
1629 * intentional to enforce typesafety.
1630 */
1631 #define jbd2_alloc_inode(_gfp_flags) \
1632 ((struct jbd2_inode *)kmem_cache_alloc(jbd2_inode_cache, _gfp_flags))
1633
jbd2_free_inode(struct jbd2_inode * jinode)1634 static inline void jbd2_free_inode(struct jbd2_inode *jinode)
1635 {
1636 kmem_cache_free(jbd2_inode_cache, jinode);
1637 }
1638
1639 /* Primary revoke support */
1640 #define JOURNAL_REVOKE_DEFAULT_HASH 256
1641 extern int jbd2_journal_init_revoke(journal_t *, int);
1642 extern void jbd2_journal_destroy_revoke_record_cache(void);
1643 extern void jbd2_journal_destroy_revoke_table_cache(void);
1644 extern int __init jbd2_journal_init_revoke_record_cache(void);
1645 extern int __init jbd2_journal_init_revoke_table_cache(void);
1646 struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size);
1647 void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table);
1648
1649 extern void jbd2_journal_destroy_revoke(journal_t *);
1650 extern int jbd2_journal_revoke (handle_t *, unsigned long long, struct buffer_head *);
1651 extern void jbd2_journal_cancel_revoke(handle_t *, struct journal_head *);
1652 extern void jbd2_journal_write_revoke_records(transaction_t *transaction,
1653 struct list_head *log_bufs);
1654
1655 /* Recovery revoke support */
1656 extern int jbd2_journal_set_revoke(journal_t *, unsigned long long, tid_t);
1657 extern int jbd2_journal_test_revoke(journal_t *, unsigned long long, tid_t);
1658 extern void jbd2_journal_clear_revoke(journal_t *);
1659 extern void jbd2_journal_switch_revoke_table(journal_t *journal);
1660 extern void jbd2_clear_buffer_revoked_flags(journal_t *journal);
1661
1662 /*
1663 * The log thread user interface:
1664 *
1665 * Request space in the current transaction, and force transaction commit
1666 * transitions on demand.
1667 */
1668
1669 int jbd2_log_start_commit(journal_t *journal, tid_t tid);
1670 int jbd2_journal_start_commit(journal_t *journal, tid_t *tid);
1671 int jbd2_log_wait_commit(journal_t *journal, tid_t tid);
1672 int jbd2_transaction_committed(journal_t *journal, tid_t tid);
1673 int jbd2_complete_transaction(journal_t *journal, tid_t tid);
1674 int jbd2_log_do_checkpoint(journal_t *journal);
1675 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid);
1676
1677 void __jbd2_log_wait_for_space(journal_t *journal);
1678 extern void __jbd2_journal_drop_transaction(journal_t *, transaction_t *);
1679 extern int jbd2_cleanup_journal_tail(journal_t *);
1680
1681 /* Fast commit related APIs */
1682 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid);
1683 int jbd2_fc_end_commit(journal_t *journal);
1684 int jbd2_fc_end_commit_fallback(journal_t *journal);
1685 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out);
1686 int jbd2_submit_inode_data(journal_t *journal, struct jbd2_inode *jinode);
1687 int jbd2_wait_inode_data(journal_t *journal, struct jbd2_inode *jinode);
1688 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks);
1689 void jbd2_fc_release_bufs(journal_t *journal);
1690
1691 /*
1692 * is_journal_abort
1693 *
1694 * Simple test wrapper function to test the JBD2_ABORT state flag. This
1695 * bit, when set, indicates that we have had a fatal error somewhere,
1696 * either inside the journaling layer or indicated to us by the client
1697 * (eg. ext3), and that we and should not commit any further
1698 * transactions.
1699 */
1700
is_journal_aborted(journal_t * journal)1701 static inline int is_journal_aborted(journal_t *journal)
1702 {
1703 return journal->j_flags & JBD2_ABORT;
1704 }
1705
is_handle_aborted(handle_t * handle)1706 static inline int is_handle_aborted(handle_t *handle)
1707 {
1708 if (handle->h_aborted || !handle->h_transaction)
1709 return 1;
1710 return is_journal_aborted(handle->h_transaction->t_journal);
1711 }
1712
jbd2_journal_abort_handle(handle_t * handle)1713 static inline void jbd2_journal_abort_handle(handle_t *handle)
1714 {
1715 handle->h_aborted = 1;
1716 }
1717
jbd2_init_fs_dev_write_error(journal_t * journal)1718 static inline void jbd2_init_fs_dev_write_error(journal_t *journal)
1719 {
1720 struct address_space *mapping = journal->j_fs_dev->bd_mapping;
1721
1722 /*
1723 * Save the original wb_err value of client fs's bdev mapping which
1724 * could be used to detect the client fs's metadata async write error.
1725 */
1726 errseq_check_and_advance(&mapping->wb_err, &journal->j_fs_dev_wb_err);
1727 }
1728
jbd2_check_fs_dev_write_error(journal_t * journal)1729 static inline int jbd2_check_fs_dev_write_error(journal_t *journal)
1730 {
1731 struct address_space *mapping = journal->j_fs_dev->bd_mapping;
1732
1733 return errseq_check(&mapping->wb_err,
1734 READ_ONCE(journal->j_fs_dev_wb_err));
1735 }
1736
1737 #endif /* __KERNEL__ */
1738
1739 /* Comparison functions for transaction IDs: perform comparisons using
1740 * modulo arithmetic so that they work over sequence number wraps. */
1741
tid_gt(tid_t x,tid_t y)1742 static inline int tid_gt(tid_t x, tid_t y)
1743 {
1744 int difference = (x - y);
1745 return (difference > 0);
1746 }
1747
tid_geq(tid_t x,tid_t y)1748 static inline int tid_geq(tid_t x, tid_t y)
1749 {
1750 int difference = (x - y);
1751 return (difference >= 0);
1752 }
1753
1754 extern int jbd2_journal_blocks_per_folio(struct inode *inode);
1755 extern size_t journal_tag_bytes(journal_t *journal);
1756
jbd2_journal_has_csum_v2or3(journal_t * journal)1757 static inline int jbd2_journal_has_csum_v2or3(journal_t *journal)
1758 {
1759 return jbd2_has_feature_csum2(journal) ||
1760 jbd2_has_feature_csum3(journal);
1761 }
1762
jbd2_journal_get_num_fc_blks(journal_superblock_t * jsb)1763 static inline int jbd2_journal_get_num_fc_blks(journal_superblock_t *jsb)
1764 {
1765 int num_fc_blocks = be32_to_cpu(jsb->s_num_fc_blks);
1766
1767 return num_fc_blocks ? num_fc_blocks : JBD2_DEFAULT_FAST_COMMIT_BLOCKS;
1768 }
1769
1770 /*
1771 * Return number of free blocks in the log. Must be called under j_state_lock.
1772 */
jbd2_log_space_left(journal_t * journal)1773 static inline unsigned long jbd2_log_space_left(journal_t *journal)
1774 {
1775 /* Allow for rounding errors */
1776 long free = journal->j_free - 32;
1777
1778 if (journal->j_committing_transaction) {
1779 free -= atomic_read(&journal->
1780 j_committing_transaction->t_outstanding_credits);
1781 }
1782 return max_t(long, free, 0);
1783 }
1784
1785 /*
1786 * Definitions which augment the buffer_head layer
1787 */
1788
1789 /* journaling buffer types */
1790 #define BJ_None 0 /* Not journaled */
1791 #define BJ_Metadata 1 /* Normal journaled metadata */
1792 #define BJ_Forget 2 /* Buffer superseded by this transaction */
1793 #define BJ_Shadow 3 /* Buffer contents being shadowed to the log */
1794 #define BJ_Reserved 4 /* Buffer is reserved for access by journal */
1795 #define BJ_Types 5
1796
jbd2_chksum(u32 crc,const void * address,unsigned int length)1797 static inline u32 jbd2_chksum(u32 crc, const void *address, unsigned int length)
1798 {
1799 return crc32c(crc, address, length);
1800 }
1801
1802 /* Return most recent uncommitted transaction */
jbd2_get_latest_transaction(journal_t * journal)1803 static inline tid_t jbd2_get_latest_transaction(journal_t *journal)
1804 {
1805 tid_t tid;
1806
1807 read_lock(&journal->j_state_lock);
1808 tid = journal->j_commit_request;
1809 if (journal->j_running_transaction)
1810 tid = journal->j_running_transaction->t_tid;
1811 read_unlock(&journal->j_state_lock);
1812 return tid;
1813 }
1814
jbd2_handle_buffer_credits(handle_t * handle)1815 static inline int jbd2_handle_buffer_credits(handle_t *handle)
1816 {
1817 journal_t *journal;
1818
1819 if (!handle->h_reserved)
1820 journal = handle->h_transaction->t_journal;
1821 else
1822 journal = handle->h_journal;
1823
1824 return handle->h_total_credits -
1825 DIV_ROUND_UP(handle->h_revoke_credits_requested,
1826 journal->j_revoke_records_per_block);
1827 }
1828
1829 #ifdef __KERNEL__
1830
1831 #define buffer_trace_init(bh) do {} while (0)
1832 #define print_buffer_fields(bh) do {} while (0)
1833 #define print_buffer_trace(bh) do {} while (0)
1834 #define BUFFER_TRACE(bh, info) do {} while (0)
1835 #define BUFFER_TRACE2(bh, bh2, info) do {} while (0)
1836 #define JBUFFER_TRACE(jh, info) do {} while (0)
1837
1838 #endif /* __KERNEL__ */
1839
1840 #endif /* _LINUX_JBD2_H */
1841