kern_alq.c revision 103830 
1/*
2 * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice unmodified, this list of conditions, and the following
10 *    disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 *
26 * $FreeBSD: head/sys/kern/kern_alq.c 103830 2002-09-23 05:20:00Z jeff $
27 *
28 */
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/kernel.h>
33#include <sys/kthread.h>
34#include <sys/lock.h>
35#include <sys/mutex.h>
36#include <sys/namei.h>
37#include <sys/proc.h>
38#include <sys/vnode.h>
39#include <sys/alq.h>
40#include <sys/malloc.h>
41#include <sys/unistd.h>
42#include <sys/fcntl.h>
43#include <sys/eventhandler.h>
44
45/* Async. Logging Queue */
46struct alq {
47	int	aq_entmax;		/* Max entries */
48	int	aq_entlen;		/* Entry length */
49	char	*aq_entbuf;		/* Buffer for stored entries */
50	int	aq_flags;		/* Queue flags */
51	struct mtx	aq_mtx;		/* Queue lock */
52	struct vnode	*aq_vp;		/* Open vnode handle */
53	struct ucred	*aq_cred;	/* Credentials of the opening thread */
54	struct ale	*aq_first;	/* First ent */
55	struct ale	*aq_entfree;	/* First free ent */
56	struct ale	*aq_entvalid;	/* First ent valid for writing */
57	LIST_ENTRY(alq)	aq_act;		/* List of active queues */
58	LIST_ENTRY(alq)	aq_link;	/* List of all queues */
59};
60
61#define	AQ_WANTED	0x0001		/* Wakeup sleeper when io is done */
62#define	AQ_ACTIVE	0x0002		/* on the active list */
63#define	AQ_FLUSHING	0x0004		/* doing IO */
64#define	AQ_SHUTDOWN	0x0008		/* Queue no longer valid */
65
66#define	ALQ_LOCK(alq)	mtx_lock_spin(&(alq)->aq_mtx)
67#define	ALQ_UNLOCK(alq)	mtx_unlock_spin(&(alq)->aq_mtx)
68
69static MALLOC_DEFINE(M_ALD, "ALD", "ALD");
70
71/*
72 * The ald_mtx protects the ald_queues list and the ald_active list.
73 */
74static struct mtx ald_mtx;
75static LIST_HEAD(, alq) ald_queues;
76static LIST_HEAD(, alq) ald_active;
77static struct proc *ald_thread;
78static int ald_shutingdown = 0;
79
80#define	ALD_LOCK()	mtx_lock(&ald_mtx)
81#define	ALD_UNLOCK()	mtx_unlock(&ald_mtx)
82
83/* Daemon functions */
84static int ald_add(struct alq *);
85static int ald_rem(struct alq *);
86static void ald_startup(void *);
87static void ald_daemon(void);
88static void ald_shutdown(void *, int);
89static void ald_activate(struct alq *);
90static void ald_deactivate(struct alq *);
91
92/* Internal queue functions */
93static void alq_shutdown(struct alq *);
94static int alq_doio(struct alq *);
95
96
97/*
98 * Add a new queue to the global list.  Fail if we're shutting down.
99 */
100static int
101ald_add(struct alq *alq)
102{
103	int error;
104
105	error = 0;
106
107	ALD_LOCK();
108	if (ald_shutingdown) {
109		error = EBUSY;
110		goto done;
111	}
112	LIST_INSERT_HEAD(&ald_queues, alq, aq_link);
113done:
114	ALD_UNLOCK();
115	return (error);
116}
117
118/*
119 * Remove a queue from the global list unless we're shutting down.  If so,
120 * the ald will take care of cleaning up it's resources.
121 */
122static int
123ald_rem(struct alq *alq)
124{
125	int error;
126
127	error = 0;
128
129	ALD_LOCK();
130	if (ald_shutingdown) {
131		error = EBUSY;
132		goto done;
133	}
134	LIST_REMOVE(alq, aq_link);
135done:
136	ALD_UNLOCK();
137	return (error);
138}
139
140/*
141 * Put a queue on the active list.  This will schedule it for writing.
142 */
143static void
144ald_activate(struct alq *alq)
145{
146	LIST_INSERT_HEAD(&ald_active, alq, aq_act);
147	wakeup(&ald_active);
148}
149
150static void
151ald_deactivate(struct alq *alq)
152{
153	LIST_REMOVE(alq, aq_act);
154	alq->aq_flags &= ~AQ_ACTIVE;
155}
156
157static void
158ald_startup(void *unused)
159{
160	mtx_init(&ald_mtx, "ALDmtx", NULL, MTX_DEF|MTX_QUIET);
161	LIST_INIT(&ald_queues);
162	LIST_INIT(&ald_active);
163}
164
165static void
166ald_daemon(void)
167{
168	int needwakeup;
169	struct alq *alq;
170
171	mtx_lock(&Giant);
172
173	EVENTHANDLER_REGISTER(shutdown_pre_sync, ald_shutdown, NULL,
174	    SHUTDOWN_PRI_FIRST);
175
176	ALD_LOCK();
177
178	for (;;) {
179		while ((alq = LIST_FIRST(&ald_active)) == NULL)
180			msleep(&ald_active, &ald_mtx, PWAIT, "aldslp", 0);
181
182		ALQ_LOCK(alq);
183		ald_deactivate(alq);
184		ALD_UNLOCK();
185		needwakeup = alq_doio(alq);
186		ALQ_UNLOCK(alq);
187		if (needwakeup)
188			wakeup(alq);
189		ALD_LOCK();
190	}
191}
192
193static void
194ald_shutdown(void *arg, int howto)
195{
196	struct alq *alq;
197
198	ALD_LOCK();
199	ald_shutingdown = 1;
200
201	while ((alq = LIST_FIRST(&ald_queues)) != NULL) {
202		LIST_REMOVE(alq, aq_link);
203		ALD_UNLOCK();
204		alq_shutdown(alq);
205		ALD_LOCK();
206	}
207	ALD_UNLOCK();
208}
209
210static void
211alq_shutdown(struct alq *alq)
212{
213	ALQ_LOCK(alq);
214
215	/* Stop any new writers. */
216	alq->aq_flags |= AQ_SHUTDOWN;
217
218	/* Drain IO */
219	while (alq->aq_flags & (AQ_FLUSHING|AQ_ACTIVE)) {
220		alq->aq_flags |= AQ_WANTED;
221		ALQ_UNLOCK(alq);
222		tsleep(alq, PWAIT, "aldclose", 0);
223		ALQ_LOCK(alq);
224	}
225	ALQ_UNLOCK(alq);
226
227	vn_close(alq->aq_vp, FREAD|FWRITE, alq->aq_cred,
228	    curthread);
229	crfree(alq->aq_cred);
230}
231
232/*
233 * Flush all pending data to disk.  This operation will block.
234 */
235static int
236alq_doio(struct alq *alq)
237{
238	struct thread *td;
239	struct mount *mp;
240	struct vnode *vp;
241	struct uio auio;
242	struct iovec aiov[2];
243	struct ale *ale;
244	struct ale *alstart;
245	int totlen;
246	int iov;
247
248	vp = alq->aq_vp;
249	td = curthread;
250	totlen = 0;
251	iov = 0;
252
253	alstart = ale = alq->aq_entvalid;
254	alq->aq_entvalid = NULL;
255
256	bzero(&aiov, sizeof(aiov));
257	bzero(&auio, sizeof(auio));
258
259	do {
260		if (aiov[iov].iov_base == NULL)
261			aiov[iov].iov_base = ale->ae_data;
262		aiov[iov].iov_len += alq->aq_entlen;
263		totlen += alq->aq_entlen;
264		/* Check to see if we're wrapping the buffer */
265		if (ale->ae_data + alq->aq_entlen != ale->ae_next->ae_data)
266			iov++;
267		ale->ae_flags &= ~AE_VALID;
268		ale = ale->ae_next;
269	} while (ale->ae_flags & AE_VALID);
270
271	alq->aq_flags |= AQ_FLUSHING;
272	ALQ_UNLOCK(alq);
273
274	if (iov == 2 || aiov[iov].iov_base == NULL)
275		iov--;
276
277	auio.uio_iov = &aiov[0];
278	auio.uio_offset = 0;
279	auio.uio_segflg = UIO_SYSSPACE;
280	auio.uio_rw = UIO_WRITE;
281	auio.uio_iovcnt = iov + 1;
282	auio.uio_resid = totlen;
283	auio.uio_td = td;
284
285	/*
286	 * Do all of the junk required to write now.
287	 */
288	vn_start_write(vp, &mp, V_WAIT);
289	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
290	VOP_LEASE(vp, td, alq->aq_cred, LEASE_WRITE);
291	/* XXX error ignored */
292	VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, alq->aq_cred);
293	VOP_UNLOCK(vp, 0, td);
294	vn_finished_write(mp);
295
296	ALQ_LOCK(alq);
297	alq->aq_flags &= ~AQ_FLUSHING;
298
299	if (alq->aq_entfree == NULL)
300		alq->aq_entfree = alstart;
301
302	if (alq->aq_flags & AQ_WANTED) {
303		alq->aq_flags &= ~AQ_WANTED;
304		return (1);
305	}
306
307	return(0);
308}
309
310static struct kproc_desc ald_kp = {
311        "ALQ Daemon",
312        ald_daemon,
313        &ald_thread
314};
315
316SYSINIT(aldthread, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &ald_kp)
317SYSINIT(ald, SI_SUB_LOCK, SI_ORDER_ANY, ald_startup, NULL)
318
319
320/* User visible queue functions */
321
322/*
323 * Create the queue data structure, allocate the buffer, and open the file.
324 */
325int
326alq_open(struct alq **alqp, const char *file, int size, int count)
327{
328	struct thread *td;
329	struct nameidata nd;
330	struct ale *ale;
331	struct ale *alp;
332	struct alq *alq;
333	char *bufp;
334	int flags;
335	int error;
336	int i;
337
338	*alqp = NULL;
339	td = curthread;
340
341	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, file, td);
342	flags = FREAD | FWRITE | O_NOFOLLOW | O_CREAT;
343
344	error = vn_open(&nd, &flags, 0);
345	if (error)
346		return (error);
347
348	NDFREE(&nd, NDF_ONLY_PNBUF);
349	/* We just unlock so we hold a reference */
350	VOP_UNLOCK(nd.ni_vp, 0, td);
351
352	alq = malloc(sizeof(*alq), M_ALD, M_WAITOK|M_ZERO);
353	alq->aq_entbuf = malloc(count * size, M_ALD, M_WAITOK|M_ZERO);
354	alq->aq_first = malloc(sizeof(*ale) * count, M_ALD, M_WAITOK|M_ZERO);
355	alq->aq_vp = nd.ni_vp;
356	alq->aq_cred = crhold(td->td_ucred);
357	alq->aq_entmax = count;
358	alq->aq_entlen = size;
359	alq->aq_entfree = alq->aq_first;
360
361	mtx_init(&alq->aq_mtx, "ALD Queue", NULL, MTX_SPIN|MTX_QUIET);
362
363	bufp = alq->aq_entbuf;
364	ale = alq->aq_first;
365	alp = NULL;
366
367	/* Match up entries with buffers */
368	for (i = 0; i < count; i++) {
369		if (alp)
370			alp->ae_next = ale;
371		ale->ae_data = bufp;
372		alp = ale;
373		ale++;
374		bufp += size;
375	}
376
377	alp->ae_next = alq->aq_first;
378
379	if ((error = ald_add(alq)) != 0)
380		return (error);
381	*alqp = alq;
382
383	return (0);
384}
385
386/*
387 * Copy a new entry into the queue.  If the operation would block either
388 * wait or return an error depending on the value of waitok.
389 */
390int
391alq_write(struct alq *alq, void *data, int waitok)
392{
393	struct ale *ale;
394
395	if ((ale = alq_get(alq, waitok)) == NULL)
396		return (EWOULDBLOCK);
397
398	bcopy(data, ale->ae_data, alq->aq_entlen);
399	alq_post(alq, ale);
400
401	return (0);
402}
403
404struct ale *
405alq_get(struct alq *alq, int waitok)
406{
407	struct ale *ale;
408	struct ale *aln;
409
410	ale = NULL;
411
412	ALQ_LOCK(alq);
413
414	/* Loop until we get an entry or we're shutting down */
415	while ((alq->aq_flags & AQ_SHUTDOWN) == 0 &&
416	    (ale = alq->aq_entfree) == NULL &&
417	    (waitok & ALQ_WAITOK)) {
418		alq->aq_flags |= AQ_WANTED;
419		ALQ_UNLOCK(alq);
420		tsleep(alq, PWAIT, "alqget", 0);
421		ALQ_LOCK(alq);
422	}
423
424	if (ale != NULL) {
425		aln = ale->ae_next;
426		if ((aln->ae_flags & AE_VALID) == 0)
427			alq->aq_entfree = aln;
428	} else
429		ALQ_UNLOCK(alq);
430
431
432	return (ale);
433}
434
435void
436alq_post(struct alq *alq, struct ale *ale)
437{
438	int activate;
439
440	ale->ae_flags |= AE_VALID;
441
442	if (alq->aq_entvalid == NULL)
443		alq->aq_entvalid = ale;
444
445	if ((alq->aq_flags & AQ_ACTIVE) == 0) {
446		alq->aq_flags |= AQ_ACTIVE;
447		activate = 1;
448	} else
449		activate = 0;
450
451	ALQ_UNLOCK(alq);
452	if (activate) {
453		ALD_LOCK();
454		ald_activate(alq);
455		ALD_UNLOCK();
456	}
457}
458
459void
460alq_flush(struct alq *alq)
461{
462	int needwakeup = 0;
463
464	ALD_LOCK();
465	ALQ_LOCK(alq);
466	if (alq->aq_flags & AQ_ACTIVE) {
467		ald_deactivate(alq);
468		ALD_UNLOCK();
469		needwakeup = alq_doio(alq);
470	} else
471		ALD_UNLOCK();
472	ALQ_UNLOCK(alq);
473
474	if (needwakeup)
475		wakeup(alq);
476}
477
478/*
479 * Flush remaining data, close the file and free all resources.
480 */
481void
482alq_close(struct alq *alq)
483{
484	/*
485	 * If we're already shuting down someone else will flush and close
486	 * the vnode.
487	 */
488	if (ald_rem(alq) != 0)
489		return;
490
491	/*
492	 * Drain all pending IO.
493	 */
494	alq_shutdown(alq);
495
496	mtx_destroy(&alq->aq_mtx);
497	free(alq->aq_first, M_ALD);
498	free(alq->aq_entbuf, M_ALD);
499	free(alq, M_ALD);
500}
501