xfs_log_priv.h revision 2ce82b72
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#ifndef	__XFS_LOG_PRIV_H__
7#define __XFS_LOG_PRIV_H__
8
9struct xfs_buf;
10struct xlog;
11struct xlog_ticket;
12struct xfs_mount;
13
14/*
15 * get client id from packed copy.
16 *
17 * this hack is here because the xlog_pack code copies four bytes
18 * of xlog_op_header containing the fields oh_clientid, oh_flags
19 * and oh_res2 into the packed copy.
20 *
21 * later on this four byte chunk is treated as an int and the
22 * client id is pulled out.
23 *
24 * this has endian issues, of course.
25 */
26static inline uint xlog_get_client_id(__be32 i)
27{
28	return be32_to_cpu(i) >> 24;
29}
30
31/*
32 * In core log state
33 */
34enum xlog_iclog_state {
35	XLOG_STATE_ACTIVE,	/* Current IC log being written to */
36	XLOG_STATE_WANT_SYNC,	/* Want to sync this iclog; no more writes */
37	XLOG_STATE_SYNCING,	/* This IC log is syncing */
38	XLOG_STATE_DONE_SYNC,	/* Done syncing to disk */
39	XLOG_STATE_CALLBACK,	/* Callback functions now */
40	XLOG_STATE_DIRTY,	/* Dirty IC log, not ready for ACTIVE status */
41};
42
43#define XLOG_STATE_STRINGS \
44	{ XLOG_STATE_ACTIVE,	"XLOG_STATE_ACTIVE" }, \
45	{ XLOG_STATE_WANT_SYNC,	"XLOG_STATE_WANT_SYNC" }, \
46	{ XLOG_STATE_SYNCING,	"XLOG_STATE_SYNCING" }, \
47	{ XLOG_STATE_DONE_SYNC,	"XLOG_STATE_DONE_SYNC" }, \
48	{ XLOG_STATE_CALLBACK,	"XLOG_STATE_CALLBACK" }, \
49	{ XLOG_STATE_DIRTY,	"XLOG_STATE_DIRTY" }
50
51/*
52 * In core log flags
53 */
54#define XLOG_ICL_NEED_FLUSH	(1 << 0)	/* iclog needs REQ_PREFLUSH */
55#define XLOG_ICL_NEED_FUA	(1 << 1)	/* iclog needs REQ_FUA */
56
57#define XLOG_ICL_STRINGS \
58	{ XLOG_ICL_NEED_FLUSH,	"XLOG_ICL_NEED_FLUSH" }, \
59	{ XLOG_ICL_NEED_FUA,	"XLOG_ICL_NEED_FUA" }
60
61
62/*
63 * Log ticket flags
64 */
65#define XLOG_TIC_PERM_RESERV	0x1	/* permanent reservation */
66
67#define XLOG_TIC_FLAGS \
68	{ XLOG_TIC_PERM_RESERV,	"XLOG_TIC_PERM_RESERV" }
69
70/*
71 * Below are states for covering allocation transactions.
72 * By covering, we mean changing the h_tail_lsn in the last on-disk
73 * log write such that no allocation transactions will be re-done during
74 * recovery after a system crash. Recovery starts at the last on-disk
75 * log write.
76 *
77 * These states are used to insert dummy log entries to cover
78 * space allocation transactions which can undo non-transactional changes
79 * after a crash. Writes to a file with space
80 * already allocated do not result in any transactions. Allocations
81 * might include space beyond the EOF. So if we just push the EOF a
82 * little, the last transaction for the file could contain the wrong
83 * size. If there is no file system activity, after an allocation
84 * transaction, and the system crashes, the allocation transaction
85 * will get replayed and the file will be truncated. This could
86 * be hours/days/... after the allocation occurred.
87 *
88 * The fix for this is to do two dummy transactions when the
89 * system is idle. We need two dummy transaction because the h_tail_lsn
90 * in the log record header needs to point beyond the last possible
91 * non-dummy transaction. The first dummy changes the h_tail_lsn to
92 * the first transaction before the dummy. The second dummy causes
93 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
94 *
95 * These dummy transactions get committed when everything
96 * is idle (after there has been some activity).
97 *
98 * There are 5 states used to control this.
99 *
100 *  IDLE -- no logging has been done on the file system or
101 *		we are done covering previous transactions.
102 *  NEED -- logging has occurred and we need a dummy transaction
103 *		when the log becomes idle.
104 *  DONE -- we were in the NEED state and have committed a dummy
105 *		transaction.
106 *  NEED2 -- we detected that a dummy transaction has gone to the
107 *		on disk log with no other transactions.
108 *  DONE2 -- we committed a dummy transaction when in the NEED2 state.
109 *
110 * There are two places where we switch states:
111 *
112 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
113 *	We commit the dummy transaction and switch to DONE or DONE2,
114 *	respectively. In all other states, we don't do anything.
115 *
116 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
117 *
118 *	No matter what state we are in, if this isn't the dummy
119 *	transaction going out, the next state is NEED.
120 *	So, if we aren't in the DONE or DONE2 states, the next state
121 *	is NEED. We can't be finishing a write of the dummy record
122 *	unless it was committed and the state switched to DONE or DONE2.
123 *
124 *	If we are in the DONE state and this was a write of the
125 *		dummy transaction, we move to NEED2.
126 *
127 *	If we are in the DONE2 state and this was a write of the
128 *		dummy transaction, we move to IDLE.
129 *
130 *
131 * Writing only one dummy transaction can get appended to
132 * one file space allocation. When this happens, the log recovery
133 * code replays the space allocation and a file could be truncated.
134 * This is why we have the NEED2 and DONE2 states before going idle.
135 */
136
137#define XLOG_STATE_COVER_IDLE	0
138#define XLOG_STATE_COVER_NEED	1
139#define XLOG_STATE_COVER_DONE	2
140#define XLOG_STATE_COVER_NEED2	3
141#define XLOG_STATE_COVER_DONE2	4
142
143#define XLOG_COVER_OPS		5
144
145/* Ticket reservation region accounting */
146#define XLOG_TIC_LEN_MAX	15
147
148/*
149 * Reservation region
150 * As would be stored in xfs_log_iovec but without the i_addr which
151 * we don't care about.
152 */
153typedef struct xlog_res {
154	uint	r_len;	/* region length		:4 */
155	uint	r_type;	/* region's transaction type	:4 */
156} xlog_res_t;
157
158typedef struct xlog_ticket {
159	struct list_head   t_queue;	 /* reserve/write queue */
160	struct task_struct *t_task;	 /* task that owns this ticket */
161	xlog_tid_t	   t_tid;	 /* transaction identifier	 : 4  */
162	atomic_t	   t_ref;	 /* ticket reference count       : 4  */
163	int		   t_curr_res;	 /* current reservation in bytes : 4  */
164	int		   t_unit_res;	 /* unit reservation in bytes    : 4  */
165	char		   t_ocnt;	 /* original count		 : 1  */
166	char		   t_cnt;	 /* current count		 : 1  */
167	char		   t_clientid;	 /* who does this belong to;	 : 1  */
168	char		   t_flags;	 /* properties of reservation	 : 1  */
169
170        /* reservation array fields */
171	uint		   t_res_num;                    /* num in array : 4 */
172	uint		   t_res_num_ophdrs;		 /* num op hdrs  : 4 */
173	uint		   t_res_arr_sum;		 /* array sum    : 4 */
174	uint		   t_res_o_flow;		 /* sum overflow : 4 */
175	xlog_res_t	   t_res_arr[XLOG_TIC_LEN_MAX];  /* array of res : 8 * 15 */
176} xlog_ticket_t;
177
178/*
179 * - A log record header is 512 bytes.  There is plenty of room to grow the
180 *	xlog_rec_header_t into the reserved space.
181 * - ic_data follows, so a write to disk can start at the beginning of
182 *	the iclog.
183 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
184 * - ic_next is the pointer to the next iclog in the ring.
185 * - ic_log is a pointer back to the global log structure.
186 * - ic_size is the full size of the log buffer, minus the cycle headers.
187 * - ic_offset is the current number of bytes written to in this iclog.
188 * - ic_refcnt is bumped when someone is writing to the log.
189 * - ic_state is the state of the iclog.
190 *
191 * Because of cacheline contention on large machines, we need to separate
192 * various resources onto different cachelines. To start with, make the
193 * structure cacheline aligned. The following fields can be contended on
194 * by independent processes:
195 *
196 *	- ic_callbacks
197 *	- ic_refcnt
198 *	- fields protected by the global l_icloglock
199 *
200 * so we need to ensure that these fields are located in separate cachelines.
201 * We'll put all the read-only and l_icloglock fields in the first cacheline,
202 * and move everything else out to subsequent cachelines.
203 */
204typedef struct xlog_in_core {
205	wait_queue_head_t	ic_force_wait;
206	wait_queue_head_t	ic_write_wait;
207	struct xlog_in_core	*ic_next;
208	struct xlog_in_core	*ic_prev;
209	struct xlog		*ic_log;
210	u32			ic_size;
211	u32			ic_offset;
212	enum xlog_iclog_state	ic_state;
213	unsigned int		ic_flags;
214	char			*ic_datap;	/* pointer to iclog data */
215	struct list_head	ic_callbacks;
216
217	/* reference counts need their own cacheline */
218	atomic_t		ic_refcnt ____cacheline_aligned_in_smp;
219	xlog_in_core_2_t	*ic_data;
220#define ic_header	ic_data->hic_header
221#ifdef DEBUG
222	bool			ic_fail_crc : 1;
223#endif
224	struct semaphore	ic_sema;
225	struct work_struct	ic_end_io_work;
226	struct bio		ic_bio;
227	struct bio_vec		ic_bvec[];
228} xlog_in_core_t;
229
230/*
231 * The CIL context is used to aggregate per-transaction details as well be
232 * passed to the iclog for checkpoint post-commit processing.  After being
233 * passed to the iclog, another context needs to be allocated for tracking the
234 * next set of transactions to be aggregated into a checkpoint.
235 */
236struct xfs_cil;
237
238struct xfs_cil_ctx {
239	struct xfs_cil		*cil;
240	xfs_csn_t		sequence;	/* chkpt sequence # */
241	xfs_lsn_t		start_lsn;	/* first LSN of chkpt commit */
242	xfs_lsn_t		commit_lsn;	/* chkpt commit record lsn */
243	struct xlog_ticket	*ticket;	/* chkpt ticket */
244	int			nvecs;		/* number of regions */
245	int			space_used;	/* aggregate size of regions */
246	struct list_head	busy_extents;	/* busy extents in chkpt */
247	struct xfs_log_vec	*lv_chain;	/* logvecs being pushed */
248	struct list_head	iclog_entry;
249	struct list_head	committing;	/* ctx committing list */
250	struct work_struct	discard_endio_work;
251};
252
253/*
254 * Committed Item List structure
255 *
256 * This structure is used to track log items that have been committed but not
257 * yet written into the log. It is used only when the delayed logging mount
258 * option is enabled.
259 *
260 * This structure tracks the list of committing checkpoint contexts so
261 * we can avoid the problem of having to hold out new transactions during a
262 * flush until we have a the commit record LSN of the checkpoint. We can
263 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
264 * sequence match and extract the commit LSN directly from there. If the
265 * checkpoint is still in the process of committing, we can block waiting for
266 * the commit LSN to be determined as well. This should make synchronous
267 * operations almost as efficient as the old logging methods.
268 */
269struct xfs_cil {
270	struct xlog		*xc_log;
271	struct list_head	xc_cil;
272	spinlock_t		xc_cil_lock;
273
274	struct rw_semaphore	xc_ctx_lock ____cacheline_aligned_in_smp;
275	struct xfs_cil_ctx	*xc_ctx;
276
277	spinlock_t		xc_push_lock ____cacheline_aligned_in_smp;
278	xfs_csn_t		xc_push_seq;
279	struct list_head	xc_committing;
280	wait_queue_head_t	xc_commit_wait;
281	xfs_csn_t		xc_current_sequence;
282	struct work_struct	xc_push_work;
283	wait_queue_head_t	xc_push_wait;	/* background push throttle */
284} ____cacheline_aligned_in_smp;
285
286/*
287 * The amount of log space we allow the CIL to aggregate is difficult to size.
288 * Whatever we choose, we have to make sure we can get a reservation for the
289 * log space effectively, that it is large enough to capture sufficient
290 * relogging to reduce log buffer IO significantly, but it is not too large for
291 * the log or induces too much latency when writing out through the iclogs. We
292 * track both space consumed and the number of vectors in the checkpoint
293 * context, so we need to decide which to use for limiting.
294 *
295 * Every log buffer we write out during a push needs a header reserved, which
296 * is at least one sector and more for v2 logs. Hence we need a reservation of
297 * at least 512 bytes per 32k of log space just for the LR headers. That means
298 * 16KB of reservation per megabyte of delayed logging space we will consume,
299 * plus various headers.  The number of headers will vary based on the num of
300 * io vectors, so limiting on a specific number of vectors is going to result
301 * in transactions of varying size. IOWs, it is more consistent to track and
302 * limit space consumed in the log rather than by the number of objects being
303 * logged in order to prevent checkpoint ticket overruns.
304 *
305 * Further, use of static reservations through the log grant mechanism is
306 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
307 * grant) and a significant deadlock potential because regranting write space
308 * can block on log pushes. Hence if we have to regrant log space during a log
309 * push, we can deadlock.
310 *
311 * However, we can avoid this by use of a dynamic "reservation stealing"
312 * technique during transaction commit whereby unused reservation space in the
313 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
314 * space needed by the checkpoint transaction. This means that we never need to
315 * specifically reserve space for the CIL checkpoint transaction, nor do we
316 * need to regrant space once the checkpoint completes. This also means the
317 * checkpoint transaction ticket is specific to the checkpoint context, rather
318 * than the CIL itself.
319 *
320 * With dynamic reservations, we can effectively make up arbitrary limits for
321 * the checkpoint size so long as they don't violate any other size rules.
322 * Recovery imposes a rule that no transaction exceed half the log, so we are
323 * limited by that.  Furthermore, the log transaction reservation subsystem
324 * tries to keep 25% of the log free, so we need to keep below that limit or we
325 * risk running out of free log space to start any new transactions.
326 *
327 * In order to keep background CIL push efficient, we only need to ensure the
328 * CIL is large enough to maintain sufficient in-memory relogging to avoid
329 * repeated physical writes of frequently modified metadata. If we allow the CIL
330 * to grow to a substantial fraction of the log, then we may be pinning hundreds
331 * of megabytes of metadata in memory until the CIL flushes. This can cause
332 * issues when we are running low on memory - pinned memory cannot be reclaimed,
333 * and the CIL consumes a lot of memory. Hence we need to set an upper physical
334 * size limit for the CIL that limits the maximum amount of memory pinned by the
335 * CIL but does not limit performance by reducing relogging efficiency
336 * significantly.
337 *
338 * As such, the CIL push threshold ends up being the smaller of two thresholds:
339 * - a threshold large enough that it allows CIL to be pushed and progress to be
340 *   made without excessive blocking of incoming transaction commits. This is
341 *   defined to be 12.5% of the log space - half the 25% push threshold of the
342 *   AIL.
343 * - small enough that it doesn't pin excessive amounts of memory but maintains
344 *   close to peak relogging efficiency. This is defined to be 16x the iclog
345 *   buffer window (32MB) as measurements have shown this to be roughly the
346 *   point of diminishing performance increases under highly concurrent
347 *   modification workloads.
348 *
349 * To prevent the CIL from overflowing upper commit size bounds, we introduce a
350 * new threshold at which we block committing transactions until the background
351 * CIL commit commences and switches to a new context. While this is not a hard
352 * limit, it forces the process committing a transaction to the CIL to block and
353 * yeild the CPU, giving the CIL push work a chance to be scheduled and start
354 * work. This prevents a process running lots of transactions from overfilling
355 * the CIL because it is not yielding the CPU. We set the blocking limit at
356 * twice the background push space threshold so we keep in line with the AIL
357 * push thresholds.
358 *
359 * Note: this is not a -hard- limit as blocking is applied after the transaction
360 * is inserted into the CIL and the push has been triggered. It is largely a
361 * throttling mechanism that allows the CIL push to be scheduled and run. A hard
362 * limit will be difficult to implement without introducing global serialisation
363 * in the CIL commit fast path, and it's not at all clear that we actually need
364 * such hard limits given the ~7 years we've run without a hard limit before
365 * finding the first situation where a checkpoint size overflow actually
366 * occurred. Hence the simple throttle, and an ASSERT check to tell us that
367 * we've overrun the max size.
368 */
369#define XLOG_CIL_SPACE_LIMIT(log)	\
370	min_t(int, (log)->l_logsize >> 3, BBTOB(XLOG_TOTAL_REC_SHIFT(log)) << 4)
371
372#define XLOG_CIL_BLOCKING_SPACE_LIMIT(log)	\
373	(XLOG_CIL_SPACE_LIMIT(log) * 2)
374
375/*
376 * ticket grant locks, queues and accounting have their own cachlines
377 * as these are quite hot and can be operated on concurrently.
378 */
379struct xlog_grant_head {
380	spinlock_t		lock ____cacheline_aligned_in_smp;
381	struct list_head	waiters;
382	atomic64_t		grant;
383};
384
385/*
386 * The reservation head lsn is not made up of a cycle number and block number.
387 * Instead, it uses a cycle number and byte number.  Logs don't expect to
388 * overflow 31 bits worth of byte offset, so using a byte number will mean
389 * that round off problems won't occur when releasing partial reservations.
390 */
391struct xlog {
392	/* The following fields don't need locking */
393	struct xfs_mount	*l_mp;	        /* mount point */
394	struct xfs_ail		*l_ailp;	/* AIL log is working with */
395	struct xfs_cil		*l_cilp;	/* CIL log is working with */
396	struct xfs_buftarg	*l_targ;        /* buftarg of log */
397	struct workqueue_struct	*l_ioend_workqueue; /* for I/O completions */
398	struct delayed_work	l_work;		/* background flush work */
399	long			l_opstate;	/* operational state */
400	uint			l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
401	struct list_head	*l_buf_cancel_table;
402	int			l_iclog_hsize;  /* size of iclog header */
403	int			l_iclog_heads;  /* # of iclog header sectors */
404	uint			l_sectBBsize;   /* sector size in BBs (2^n) */
405	int			l_iclog_size;	/* size of log in bytes */
406	int			l_iclog_bufs;	/* number of iclog buffers */
407	xfs_daddr_t		l_logBBstart;   /* start block of log */
408	int			l_logsize;      /* size of log in bytes */
409	int			l_logBBsize;    /* size of log in BB chunks */
410
411	/* The following block of fields are changed while holding icloglock */
412	wait_queue_head_t	l_flush_wait ____cacheline_aligned_in_smp;
413						/* waiting for iclog flush */
414	int			l_covered_state;/* state of "covering disk
415						 * log entries" */
416	xlog_in_core_t		*l_iclog;       /* head log queue	*/
417	spinlock_t		l_icloglock;    /* grab to change iclog state */
418	int			l_curr_cycle;   /* Cycle number of log writes */
419	int			l_prev_cycle;   /* Cycle number before last
420						 * block increment */
421	int			l_curr_block;   /* current logical log block */
422	int			l_prev_block;   /* previous logical log block */
423
424	/*
425	 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
426	 * read without needing to hold specific locks. To avoid operations
427	 * contending with other hot objects, place each of them on a separate
428	 * cacheline.
429	 */
430	/* lsn of last LR on disk */
431	atomic64_t		l_last_sync_lsn ____cacheline_aligned_in_smp;
432	/* lsn of 1st LR with unflushed * buffers */
433	atomic64_t		l_tail_lsn ____cacheline_aligned_in_smp;
434
435	struct xlog_grant_head	l_reserve_head;
436	struct xlog_grant_head	l_write_head;
437
438	struct xfs_kobj		l_kobj;
439
440	/* The following field are used for debugging; need to hold icloglock */
441#ifdef DEBUG
442	void			*l_iclog_bak[XLOG_MAX_ICLOGS];
443#endif
444	/* log recovery lsn tracking (for buffer submission */
445	xfs_lsn_t		l_recovery_lsn;
446
447	uint32_t		l_iclog_roundoff;/* padding roundoff */
448
449	/* Users of log incompat features should take a read lock. */
450	struct rw_semaphore	l_incompat_users;
451};
452
453#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
454	((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE))
455
456/*
457 * Bits for operational state
458 */
459#define XLOG_ACTIVE_RECOVERY	0	/* in the middle of recovery */
460#define XLOG_RECOVERY_NEEDED	1	/* log was recovered */
461#define XLOG_IO_ERROR		2	/* log hit an I/O error, and being
462				   shutdown */
463#define XLOG_TAIL_WARN		3	/* log tail verify warning issued */
464
465static inline bool
466xlog_recovery_needed(struct xlog *log)
467{
468	return test_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate);
469}
470
471static inline bool
472xlog_in_recovery(struct xlog *log)
473{
474	return test_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);
475}
476
477static inline bool
478xlog_is_shutdown(struct xlog *log)
479{
480	return test_bit(XLOG_IO_ERROR, &log->l_opstate);
481}
482
483/* common routines */
484extern int
485xlog_recover(
486	struct xlog		*log);
487extern int
488xlog_recover_finish(
489	struct xlog		*log);
490extern void
491xlog_recover_cancel(struct xlog *);
492
493extern __le32	 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
494			    char *dp, int size);
495
496extern kmem_zone_t *xfs_log_ticket_zone;
497struct xlog_ticket *
498xlog_ticket_alloc(
499	struct xlog	*log,
500	int		unit_bytes,
501	int		count,
502	char		client,
503	bool		permanent);
504
505static inline void
506xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
507{
508	*ptr += bytes;
509	*len -= bytes;
510	*off += bytes;
511}
512
513void	xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
514void	xlog_print_trans(struct xfs_trans *);
515int	xlog_write(struct xlog *log, struct xfs_log_vec *log_vector,
516		struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
517		struct xlog_in_core **commit_iclog, uint optype);
518void	xfs_log_ticket_ungrant(struct xlog *log, struct xlog_ticket *ticket);
519void	xfs_log_ticket_regrant(struct xlog *log, struct xlog_ticket *ticket);
520
521int xlog_state_release_iclog(struct xlog *log, struct xlog_in_core *iclog,
522		xfs_lsn_t log_tail_lsn);
523
524/*
525 * When we crack an atomic LSN, we sample it first so that the value will not
526 * change while we are cracking it into the component values. This means we
527 * will always get consistent component values to work from. This should always
528 * be used to sample and crack LSNs that are stored and updated in atomic
529 * variables.
530 */
531static inline void
532xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
533{
534	xfs_lsn_t val = atomic64_read(lsn);
535
536	*cycle = CYCLE_LSN(val);
537	*block = BLOCK_LSN(val);
538}
539
540/*
541 * Calculate and assign a value to an atomic LSN variable from component pieces.
542 */
543static inline void
544xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
545{
546	atomic64_set(lsn, xlog_assign_lsn(cycle, block));
547}
548
549/*
550 * When we crack the grant head, we sample it first so that the value will not
551 * change while we are cracking it into the component values. This means we
552 * will always get consistent component values to work from.
553 */
554static inline void
555xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
556{
557	*cycle = val >> 32;
558	*space = val & 0xffffffff;
559}
560
561static inline void
562xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
563{
564	xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
565}
566
567static inline int64_t
568xlog_assign_grant_head_val(int cycle, int space)
569{
570	return ((int64_t)cycle << 32) | space;
571}
572
573static inline void
574xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
575{
576	atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
577}
578
579/*
580 * Committed Item List interfaces
581 */
582int	xlog_cil_init(struct xlog *log);
583void	xlog_cil_init_post_recovery(struct xlog *log);
584void	xlog_cil_destroy(struct xlog *log);
585bool	xlog_cil_empty(struct xlog *log);
586void	xlog_cil_commit(struct xlog *log, struct xfs_trans *tp,
587			xfs_csn_t *commit_seq, bool regrant);
588
589/*
590 * CIL force routines
591 */
592xfs_lsn_t xlog_cil_force_seq(struct xlog *log, xfs_csn_t sequence);
593
594static inline void
595xlog_cil_force(struct xlog *log)
596{
597	xlog_cil_force_seq(log, log->l_cilp->xc_current_sequence);
598}
599
600/*
601 * Wrapper function for waiting on a wait queue serialised against wakeups
602 * by a spinlock. This matches the semantics of all the wait queues used in the
603 * log code.
604 */
605static inline void
606xlog_wait(
607	struct wait_queue_head	*wq,
608	struct spinlock		*lock)
609		__releases(lock)
610{
611	DECLARE_WAITQUEUE(wait, current);
612
613	add_wait_queue_exclusive(wq, &wait);
614	__set_current_state(TASK_UNINTERRUPTIBLE);
615	spin_unlock(lock);
616	schedule();
617	remove_wait_queue(wq, &wait);
618}
619
620int xlog_wait_on_iclog(struct xlog_in_core *iclog);
621
622/*
623 * The LSN is valid so long as it is behind the current LSN. If it isn't, this
624 * means that the next log record that includes this metadata could have a
625 * smaller LSN. In turn, this means that the modification in the log would not
626 * replay.
627 */
628static inline bool
629xlog_valid_lsn(
630	struct xlog	*log,
631	xfs_lsn_t	lsn)
632{
633	int		cur_cycle;
634	int		cur_block;
635	bool		valid = true;
636
637	/*
638	 * First, sample the current lsn without locking to avoid added
639	 * contention from metadata I/O. The current cycle and block are updated
640	 * (in xlog_state_switch_iclogs()) and read here in a particular order
641	 * to avoid false negatives (e.g., thinking the metadata LSN is valid
642	 * when it is not).
643	 *
644	 * The current block is always rewound before the cycle is bumped in
645	 * xlog_state_switch_iclogs() to ensure the current LSN is never seen in
646	 * a transiently forward state. Instead, we can see the LSN in a
647	 * transiently behind state if we happen to race with a cycle wrap.
648	 */
649	cur_cycle = READ_ONCE(log->l_curr_cycle);
650	smp_rmb();
651	cur_block = READ_ONCE(log->l_curr_block);
652
653	if ((CYCLE_LSN(lsn) > cur_cycle) ||
654	    (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) {
655		/*
656		 * If the metadata LSN appears invalid, it's possible the check
657		 * above raced with a wrap to the next log cycle. Grab the lock
658		 * to check for sure.
659		 */
660		spin_lock(&log->l_icloglock);
661		cur_cycle = log->l_curr_cycle;
662		cur_block = log->l_curr_block;
663		spin_unlock(&log->l_icloglock);
664
665		if ((CYCLE_LSN(lsn) > cur_cycle) ||
666		    (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block))
667			valid = false;
668	}
669
670	return valid;
671}
672
673#endif	/* __XFS_LOG_PRIV_H__ */
674