xref: /freebsd/sys/kern/kern_alq.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
1 /*-
2  * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
3  * Copyright (c) 2008-2009, Lawrence Stewart <lstewart@freebsd.org>
4  * Copyright (c) 2009-2010, The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * Portions of this software were developed at the Centre for Advanced
8  * Internet Architectures, Swinburne University of Technology, Melbourne,
9  * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice unmodified, this list of conditions, and the following
16  *    disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include "opt_mac.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/kthread.h>
42 #include <sys/lock.h>
43 #include <sys/mount.h>
44 #include <sys/mutex.h>
45 #include <sys/namei.h>
46 #include <sys/proc.h>
47 #include <sys/vnode.h>
48 #include <sys/alq.h>
49 #include <sys/malloc.h>
50 #include <sys/unistd.h>
51 #include <sys/fcntl.h>
52 #include <sys/eventhandler.h>
53 
54 #include <security/mac/mac_framework.h>
55 
56 /* Async. Logging Queue */
57 struct alq {
58 	char	*aq_entbuf;		/* Buffer for stored entries */
59 	int	aq_entmax;		/* Max entries */
60 	int	aq_entlen;		/* Entry length */
61 	int	aq_freebytes;		/* Bytes available in buffer */
62 	int	aq_buflen;		/* Total length of our buffer */
63 	int	aq_writehead;		/* Location for next write */
64 	int	aq_writetail;		/* Flush starts at this location */
65 	int	aq_wrapearly;		/* # bytes left blank at end of buf */
66 	int	aq_flags;		/* Queue flags */
67 	int	aq_waiters;		/* Num threads waiting for resources
68 					 * NB: Used as a wait channel so must
69 					 * not be first field in the alq struct
70 					 */
71 	struct	ale	aq_getpost;	/* ALE for use by get/post */
72 	struct mtx	aq_mtx;		/* Queue lock */
73 	struct vnode	*aq_vp;		/* Open vnode handle */
74 	struct ucred	*aq_cred;	/* Credentials of the opening thread */
75 	LIST_ENTRY(alq)	aq_act;		/* List of active queues */
76 	LIST_ENTRY(alq)	aq_link;	/* List of all queues */
77 };
78 
79 #define	AQ_WANTED	0x0001		/* Wakeup sleeper when io is done */
80 #define	AQ_ACTIVE	0x0002		/* on the active list */
81 #define	AQ_FLUSHING	0x0004		/* doing IO */
82 #define	AQ_SHUTDOWN	0x0008		/* Queue no longer valid */
83 #define	AQ_ORDERED	0x0010		/* Queue enforces ordered writes */
84 #define	AQ_LEGACY	0x0020		/* Legacy queue (fixed length writes) */
85 
86 #define	ALQ_LOCK(alq)	mtx_lock_spin(&(alq)->aq_mtx)
87 #define	ALQ_UNLOCK(alq)	mtx_unlock_spin(&(alq)->aq_mtx)
88 
89 #define HAS_PENDING_DATA(alq) ((alq)->aq_freebytes != (alq)->aq_buflen)
90 
91 static MALLOC_DEFINE(M_ALD, "ALD", "ALD");
92 
93 /*
94  * The ald_mtx protects the ald_queues list and the ald_active list.
95  */
96 static struct mtx ald_mtx;
97 static LIST_HEAD(, alq) ald_queues;
98 static LIST_HEAD(, alq) ald_active;
99 static int ald_shutingdown = 0;
100 struct thread *ald_thread;
101 static struct proc *ald_proc;
102 static eventhandler_tag alq_eventhandler_tag = NULL;
103 
104 #define	ALD_LOCK()	mtx_lock(&ald_mtx)
105 #define	ALD_UNLOCK()	mtx_unlock(&ald_mtx)
106 
107 /* Daemon functions */
108 static int ald_add(struct alq *);
109 static int ald_rem(struct alq *);
110 static void ald_startup(void *);
111 static void ald_daemon(void);
112 static void ald_shutdown(void *, int);
113 static void ald_activate(struct alq *);
114 static void ald_deactivate(struct alq *);
115 
116 /* Internal queue functions */
117 static void alq_shutdown(struct alq *);
118 static void alq_destroy(struct alq *);
119 static int alq_doio(struct alq *);
120 
121 
122 /*
123  * Add a new queue to the global list.  Fail if we're shutting down.
124  */
125 static int
126 ald_add(struct alq *alq)
127 {
128 	int error;
129 
130 	error = 0;
131 
132 	ALD_LOCK();
133 	if (ald_shutingdown) {
134 		error = EBUSY;
135 		goto done;
136 	}
137 	LIST_INSERT_HEAD(&ald_queues, alq, aq_link);
138 done:
139 	ALD_UNLOCK();
140 	return (error);
141 }
142 
143 /*
144  * Remove a queue from the global list unless we're shutting down.  If so,
145  * the ald will take care of cleaning up it's resources.
146  */
147 static int
148 ald_rem(struct alq *alq)
149 {
150 	int error;
151 
152 	error = 0;
153 
154 	ALD_LOCK();
155 	if (ald_shutingdown) {
156 		error = EBUSY;
157 		goto done;
158 	}
159 	LIST_REMOVE(alq, aq_link);
160 done:
161 	ALD_UNLOCK();
162 	return (error);
163 }
164 
165 /*
166  * Put a queue on the active list.  This will schedule it for writing.
167  */
168 static void
169 ald_activate(struct alq *alq)
170 {
171 	LIST_INSERT_HEAD(&ald_active, alq, aq_act);
172 	wakeup(&ald_active);
173 }
174 
175 static void
176 ald_deactivate(struct alq *alq)
177 {
178 	LIST_REMOVE(alq, aq_act);
179 	alq->aq_flags &= ~AQ_ACTIVE;
180 }
181 
182 static void
183 ald_startup(void *unused)
184 {
185 	mtx_init(&ald_mtx, "ALDmtx", NULL, MTX_DEF|MTX_QUIET);
186 	LIST_INIT(&ald_queues);
187 	LIST_INIT(&ald_active);
188 }
189 
190 static void
191 ald_daemon(void)
192 {
193 	int needwakeup;
194 	struct alq *alq;
195 
196 	ald_thread = FIRST_THREAD_IN_PROC(ald_proc);
197 
198 	alq_eventhandler_tag = EVENTHANDLER_REGISTER(shutdown_pre_sync,
199 	    ald_shutdown, NULL, SHUTDOWN_PRI_FIRST);
200 
201 	ALD_LOCK();
202 
203 	for (;;) {
204 		while ((alq = LIST_FIRST(&ald_active)) == NULL &&
205 		    !ald_shutingdown)
206 			mtx_sleep(&ald_active, &ald_mtx, PWAIT, "aldslp", 0);
207 
208 		/* Don't shutdown until all active ALQs are flushed. */
209 		if (ald_shutingdown && alq == NULL) {
210 			ALD_UNLOCK();
211 			break;
212 		}
213 
214 		ALQ_LOCK(alq);
215 		ald_deactivate(alq);
216 		ALD_UNLOCK();
217 		needwakeup = alq_doio(alq);
218 		ALQ_UNLOCK(alq);
219 		if (needwakeup)
220 			wakeup_one(alq);
221 		ALD_LOCK();
222 	}
223 
224 	kproc_exit(0);
225 }
226 
227 static void
228 ald_shutdown(void *arg, int howto)
229 {
230 	struct alq *alq;
231 
232 	EVENTHANDLER_DEREGISTER(shutdown_pre_sync, alq_eventhandler_tag);
233 
234 	ALD_LOCK();
235 
236 	/* Ensure no new queues can be created. */
237 	ald_shutingdown = 1;
238 
239 	/* Shutdown all ALQs prior to terminating the ald_daemon. */
240 	while ((alq = LIST_FIRST(&ald_queues)) != NULL) {
241 		LIST_REMOVE(alq, aq_link);
242 		ALD_UNLOCK();
243 		alq_shutdown(alq);
244 		ALD_LOCK();
245 	}
246 
247 	/* At this point, all ALQs are flushed and shutdown. */
248 
249 	/*
250 	 * Wake ald_daemon so that it exits. It won't be able to do
251 	 * anything until we mtx_sleep because we hold the ald_mtx.
252 	 */
253 	wakeup(&ald_active);
254 
255 	/* Wait for ald_daemon to exit. */
256 	mtx_sleep(ald_proc, &ald_mtx, PWAIT, "aldslp", 0);
257 
258 	ALD_UNLOCK();
259 }
260 
261 static void
262 alq_shutdown(struct alq *alq)
263 {
264 	ALQ_LOCK(alq);
265 
266 	/* Stop any new writers. */
267 	alq->aq_flags |= AQ_SHUTDOWN;
268 
269 	/*
270 	 * If the ALQ isn't active but has unwritten data (possible if
271 	 * the ALQ_NOACTIVATE flag has been used), explicitly activate the
272 	 * ALQ here so that the pending data gets flushed by the ald_daemon.
273 	 */
274 	if (!(alq->aq_flags & AQ_ACTIVE) && HAS_PENDING_DATA(alq)) {
275 		alq->aq_flags |= AQ_ACTIVE;
276 		ALQ_UNLOCK(alq);
277 		ALD_LOCK();
278 		ald_activate(alq);
279 		ALD_UNLOCK();
280 		ALQ_LOCK(alq);
281 	}
282 
283 	/* Drain IO */
284 	while (alq->aq_flags & AQ_ACTIVE) {
285 		alq->aq_flags |= AQ_WANTED;
286 		msleep_spin(alq, &alq->aq_mtx, "aldclose", 0);
287 	}
288 	ALQ_UNLOCK(alq);
289 
290 	vn_close(alq->aq_vp, FWRITE, alq->aq_cred,
291 	    curthread);
292 	crfree(alq->aq_cred);
293 }
294 
295 void
296 alq_destroy(struct alq *alq)
297 {
298 	/* Drain all pending IO. */
299 	alq_shutdown(alq);
300 
301 	mtx_destroy(&alq->aq_mtx);
302 	free(alq->aq_entbuf, M_ALD);
303 	free(alq, M_ALD);
304 }
305 
306 /*
307  * Flush all pending data to disk.  This operation will block.
308  */
309 static int
310 alq_doio(struct alq *alq)
311 {
312 	struct thread *td;
313 	struct mount *mp;
314 	struct vnode *vp;
315 	struct uio auio;
316 	struct iovec aiov[2];
317 	int totlen;
318 	int iov;
319 	int wrapearly;
320 
321 	KASSERT((HAS_PENDING_DATA(alq)), ("%s: queue empty!", __func__));
322 
323 	vp = alq->aq_vp;
324 	td = curthread;
325 	totlen = 0;
326 	iov = 1;
327 	wrapearly = alq->aq_wrapearly;
328 
329 	bzero(&aiov, sizeof(aiov));
330 	bzero(&auio, sizeof(auio));
331 
332 	/* Start the write from the location of our buffer tail pointer. */
333 	aiov[0].iov_base = alq->aq_entbuf + alq->aq_writetail;
334 
335 	if (alq->aq_writetail < alq->aq_writehead) {
336 		/* Buffer not wrapped. */
337 		totlen = aiov[0].iov_len = alq->aq_writehead - alq->aq_writetail;
338 	} else if (alq->aq_writehead == 0) {
339 		/* Buffer not wrapped (special case to avoid an empty iov). */
340 		totlen = aiov[0].iov_len = alq->aq_buflen - alq->aq_writetail -
341 		    wrapearly;
342 	} else {
343 		/*
344 		 * Buffer wrapped, requires 2 aiov entries:
345 		 * - first is from writetail to end of buffer
346 		 * - second is from start of buffer to writehead
347 		 */
348 		aiov[0].iov_len = alq->aq_buflen - alq->aq_writetail -
349 		    wrapearly;
350 		iov++;
351 		aiov[1].iov_base = alq->aq_entbuf;
352 		aiov[1].iov_len =  alq->aq_writehead;
353 		totlen = aiov[0].iov_len + aiov[1].iov_len;
354 	}
355 
356 	alq->aq_flags |= AQ_FLUSHING;
357 	ALQ_UNLOCK(alq);
358 
359 	auio.uio_iov = &aiov[0];
360 	auio.uio_offset = 0;
361 	auio.uio_segflg = UIO_SYSSPACE;
362 	auio.uio_rw = UIO_WRITE;
363 	auio.uio_iovcnt = iov;
364 	auio.uio_resid = totlen;
365 	auio.uio_td = td;
366 
367 	/*
368 	 * Do all of the junk required to write now.
369 	 */
370 	vn_start_write(vp, &mp, V_WAIT);
371 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
372 	/*
373 	 * XXX: VOP_WRITE error checks are ignored.
374 	 */
375 #ifdef MAC
376 	if (mac_vnode_check_write(alq->aq_cred, NOCRED, vp) == 0)
377 #endif
378 		VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, alq->aq_cred);
379 	VOP_UNLOCK(vp, 0);
380 	vn_finished_write(mp);
381 
382 	ALQ_LOCK(alq);
383 	alq->aq_flags &= ~AQ_FLUSHING;
384 
385 	/* Adjust writetail as required, taking into account wrapping. */
386 	alq->aq_writetail = (alq->aq_writetail + totlen + wrapearly) %
387 	    alq->aq_buflen;
388 	alq->aq_freebytes += totlen + wrapearly;
389 
390 	/*
391 	 * If we just flushed part of the buffer which wrapped, reset the
392 	 * wrapearly indicator.
393 	 */
394 	if (wrapearly)
395 		alq->aq_wrapearly = 0;
396 
397 	/*
398 	 * If we just flushed the buffer completely, reset indexes to 0 to
399 	 * minimise buffer wraps.
400 	 * This is also required to ensure alq_getn() can't wedge itself.
401 	 */
402 	if (!HAS_PENDING_DATA(alq))
403 		alq->aq_writehead = alq->aq_writetail = 0;
404 
405 	KASSERT((alq->aq_writetail >= 0 && alq->aq_writetail < alq->aq_buflen),
406 	    ("%s: aq_writetail < 0 || aq_writetail >= aq_buflen", __func__));
407 
408 	if (alq->aq_flags & AQ_WANTED) {
409 		alq->aq_flags &= ~AQ_WANTED;
410 		return (1);
411 	}
412 
413 	return(0);
414 }
415 
416 static struct kproc_desc ald_kp = {
417         "ALQ Daemon",
418         ald_daemon,
419         &ald_proc
420 };
421 
422 SYSINIT(aldthread, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &ald_kp);
423 SYSINIT(ald, SI_SUB_LOCK, SI_ORDER_ANY, ald_startup, NULL);
424 
425 
426 /* User visible queue functions */
427 
428 /*
429  * Create the queue data structure, allocate the buffer, and open the file.
430  */
431 
432 int
433 alq_open_flags(struct alq **alqp, const char *file, struct ucred *cred, int cmode,
434     int size, int flags)
435 {
436 	struct thread *td;
437 	struct nameidata nd;
438 	struct alq *alq;
439 	int oflags;
440 	int error;
441 
442 	KASSERT((size > 0), ("%s: size <= 0", __func__));
443 
444 	*alqp = NULL;
445 	td = curthread;
446 
447 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, file, td);
448 	oflags = FWRITE | O_NOFOLLOW | O_CREAT;
449 
450 	error = vn_open_cred(&nd, &oflags, cmode, 0, cred, NULL);
451 	if (error)
452 		return (error);
453 
454 	NDFREE(&nd, NDF_ONLY_PNBUF);
455 	/* We just unlock so we hold a reference */
456 	VOP_UNLOCK(nd.ni_vp, 0);
457 
458 	alq = malloc(sizeof(*alq), M_ALD, M_WAITOK|M_ZERO);
459 	alq->aq_vp = nd.ni_vp;
460 	alq->aq_cred = crhold(cred);
461 
462 	mtx_init(&alq->aq_mtx, "ALD Queue", NULL, MTX_SPIN|MTX_QUIET);
463 
464 	alq->aq_buflen = size;
465 	alq->aq_entmax = 0;
466 	alq->aq_entlen = 0;
467 
468 	alq->aq_freebytes = alq->aq_buflen;
469 	alq->aq_entbuf = malloc(alq->aq_buflen, M_ALD, M_WAITOK|M_ZERO);
470 	alq->aq_writehead = alq->aq_writetail = 0;
471 	if (flags & ALQ_ORDERED)
472 		alq->aq_flags |= AQ_ORDERED;
473 
474 	if ((error = ald_add(alq)) != 0) {
475 		alq_destroy(alq);
476 		return (error);
477 	}
478 
479 	*alqp = alq;
480 
481 	return (0);
482 }
483 
484 int
485 alq_open(struct alq **alqp, const char *file, struct ucred *cred, int cmode,
486     int size, int count)
487 {
488 	int ret;
489 
490 	KASSERT((count >= 0), ("%s: count < 0", __func__));
491 
492 	if (count > 0) {
493 		ret = alq_open_flags(alqp, file, cred, cmode, size*count, 0);
494 		(*alqp)->aq_flags |= AQ_LEGACY;
495 		(*alqp)->aq_entmax = count;
496 		(*alqp)->aq_entlen = size;
497 	} else
498 		ret = alq_open_flags(alqp, file, cred, cmode, size, 0);
499 
500 	return (ret);
501 }
502 
503 
504 /*
505  * Copy a new entry into the queue.  If the operation would block either
506  * wait or return an error depending on the value of waitok.
507  */
508 int
509 alq_writen(struct alq *alq, void *data, int len, int flags)
510 {
511 	int activate, copy, ret;
512 	void *waitchan;
513 
514 	KASSERT((len > 0 && len <= alq->aq_buflen),
515 	    ("%s: len <= 0 || len > aq_buflen", __func__));
516 
517 	activate = ret = 0;
518 	copy = len;
519 	waitchan = NULL;
520 
521 	ALQ_LOCK(alq);
522 
523 	/*
524 	 * Fail to perform the write and return EWOULDBLOCK if:
525 	 * - The message is larger than our underlying buffer.
526 	 * - The ALQ is being shutdown.
527 	 * - There is insufficient free space in our underlying buffer
528 	 *   to accept the message and the user can't wait for space.
529 	 * - There is insufficient free space in our underlying buffer
530 	 *   to accept the message and the alq is inactive due to prior
531 	 *   use of the ALQ_NOACTIVATE flag (which would lead to deadlock).
532 	 */
533 	if (len > alq->aq_buflen ||
534 	    alq->aq_flags & AQ_SHUTDOWN ||
535 	    (((flags & ALQ_NOWAIT) || (!(alq->aq_flags & AQ_ACTIVE) &&
536 	    HAS_PENDING_DATA(alq))) && alq->aq_freebytes < len)) {
537 		ALQ_UNLOCK(alq);
538 		return (EWOULDBLOCK);
539 	}
540 
541 	/*
542 	 * If we want ordered writes and there is already at least one thread
543 	 * waiting for resources to become available, sleep until we're woken.
544 	 */
545 	if (alq->aq_flags & AQ_ORDERED && alq->aq_waiters > 0) {
546 		KASSERT(!(flags & ALQ_NOWAIT),
547 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
548 		alq->aq_waiters++;
549 		msleep_spin(&alq->aq_waiters, &alq->aq_mtx, "alqwnord", 0);
550 		alq->aq_waiters--;
551 	}
552 
553 	/*
554 	 * (ALQ_WAITOK && aq_freebytes < len) or aq_freebytes >= len, either
555 	 * enter while loop and sleep until we have enough free bytes (former)
556 	 * or skip (latter). If AQ_ORDERED is set, only 1 thread at a time will
557 	 * be in this loop. Otherwise, multiple threads may be sleeping here
558 	 * competing for ALQ resources.
559 	 */
560 	while (alq->aq_freebytes < len && !(alq->aq_flags & AQ_SHUTDOWN)) {
561 		KASSERT(!(flags & ALQ_NOWAIT),
562 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
563 		alq->aq_flags |= AQ_WANTED;
564 		alq->aq_waiters++;
565 		if (waitchan)
566 			wakeup(waitchan);
567 		msleep_spin(alq, &alq->aq_mtx, "alqwnres", 0);
568 		alq->aq_waiters--;
569 
570 		/*
571 		 * If we're the first thread to wake after an AQ_WANTED wakeup
572 		 * but there isn't enough free space for us, we're going to loop
573 		 * and sleep again. If there are other threads waiting in this
574 		 * loop, schedule a wakeup so that they can see if the space
575 		 * they require is available.
576 		 */
577 		if (alq->aq_waiters > 0 && !(alq->aq_flags & AQ_ORDERED) &&
578 		    alq->aq_freebytes < len && !(alq->aq_flags & AQ_WANTED))
579 			waitchan = alq;
580 		else
581 			waitchan = NULL;
582 	}
583 
584 	/*
585 	 * If there are waiters, we need to signal the waiting threads after we
586 	 * complete our work. The alq ptr is used as a wait channel for threads
587 	 * requiring resources to be freed up. In the AQ_ORDERED case, threads
588 	 * are not allowed to concurrently compete for resources in the above
589 	 * while loop, so we use a different wait channel in this case.
590 	 */
591 	if (alq->aq_waiters > 0) {
592 		if (alq->aq_flags & AQ_ORDERED)
593 			waitchan = &alq->aq_waiters;
594 		else
595 			waitchan = alq;
596 	} else
597 		waitchan = NULL;
598 
599 	/* Bail if we're shutting down. */
600 	if (alq->aq_flags & AQ_SHUTDOWN) {
601 		ret = EWOULDBLOCK;
602 		goto unlock;
603 	}
604 
605 	/*
606 	 * If we need to wrap the buffer to accommodate the write,
607 	 * we'll need 2 calls to bcopy.
608 	 */
609 	if ((alq->aq_buflen - alq->aq_writehead) < len)
610 		copy = alq->aq_buflen - alq->aq_writehead;
611 
612 	/* Copy message (or part thereof if wrap required) to the buffer. */
613 	bcopy(data, alq->aq_entbuf + alq->aq_writehead, copy);
614 	alq->aq_writehead += copy;
615 
616 	if (alq->aq_writehead >= alq->aq_buflen) {
617 		KASSERT((alq->aq_writehead == alq->aq_buflen),
618 		    ("%s: alq->aq_writehead (%d) > alq->aq_buflen (%d)",
619 		    __func__,
620 		    alq->aq_writehead,
621 		    alq->aq_buflen));
622 		alq->aq_writehead = 0;
623 	}
624 
625 	if (copy != len) {
626 		/*
627 		 * Wrap the buffer by copying the remainder of our message
628 		 * to the start of the buffer and resetting aq_writehead.
629 		 */
630 		bcopy(((uint8_t *)data)+copy, alq->aq_entbuf, len - copy);
631 		alq->aq_writehead = len - copy;
632 	}
633 
634 	KASSERT((alq->aq_writehead >= 0 && alq->aq_writehead < alq->aq_buflen),
635 	    ("%s: aq_writehead < 0 || aq_writehead >= aq_buflen", __func__));
636 
637 	alq->aq_freebytes -= len;
638 
639 	if (!(alq->aq_flags & AQ_ACTIVE) && !(flags & ALQ_NOACTIVATE)) {
640 		alq->aq_flags |= AQ_ACTIVE;
641 		activate = 1;
642 	}
643 
644 	KASSERT((HAS_PENDING_DATA(alq)), ("%s: queue empty!", __func__));
645 
646 unlock:
647 	ALQ_UNLOCK(alq);
648 
649 	if (activate) {
650 		ALD_LOCK();
651 		ald_activate(alq);
652 		ALD_UNLOCK();
653 	}
654 
655 	/* NB: We rely on wakeup_one waking threads in a FIFO manner. */
656 	if (waitchan != NULL)
657 		wakeup_one(waitchan);
658 
659 	return (ret);
660 }
661 
662 int
663 alq_write(struct alq *alq, void *data, int flags)
664 {
665 	/* Should only be called in fixed length message (legacy) mode. */
666 	KASSERT((alq->aq_flags & AQ_LEGACY),
667 	    ("%s: fixed length write on variable length queue", __func__));
668 	return (alq_writen(alq, data, alq->aq_entlen, flags));
669 }
670 
671 /*
672  * Retrieve a pointer for the ALQ to write directly into, avoiding bcopy.
673  */
674 struct ale *
675 alq_getn(struct alq *alq, int len, int flags)
676 {
677 	int contigbytes;
678 	void *waitchan;
679 
680 	KASSERT((len > 0 && len <= alq->aq_buflen),
681 	    ("%s: len <= 0 || len > alq->aq_buflen", __func__));
682 
683 	waitchan = NULL;
684 
685 	ALQ_LOCK(alq);
686 
687 	/*
688 	 * Determine the number of free contiguous bytes.
689 	 * We ensure elsewhere that if aq_writehead == aq_writetail because
690 	 * the buffer is empty, they will both be set to 0 and therefore
691 	 * aq_freebytes == aq_buflen and is fully contiguous.
692 	 * If they are equal and the buffer is not empty, aq_freebytes will
693 	 * be 0 indicating the buffer is full.
694 	 */
695 	if (alq->aq_writehead <= alq->aq_writetail)
696 		contigbytes = alq->aq_freebytes;
697 	else {
698 		contigbytes = alq->aq_buflen - alq->aq_writehead;
699 
700 		if (contigbytes < len) {
701 			/*
702 			 * Insufficient space at end of buffer to handle a
703 			 * contiguous write. Wrap early if there's space at
704 			 * the beginning. This will leave a hole at the end
705 			 * of the buffer which we will have to skip over when
706 			 * flushing the buffer to disk.
707 			 */
708 			if (alq->aq_writetail >= len || flags & ALQ_WAITOK) {
709 				/* Keep track of # bytes left blank. */
710 				alq->aq_wrapearly = contigbytes;
711 				/* Do the wrap and adjust counters. */
712 				contigbytes = alq->aq_freebytes =
713 				    alq->aq_writetail;
714 				alq->aq_writehead = 0;
715 			}
716 		}
717 	}
718 
719 	/*
720 	 * Return a NULL ALE if:
721 	 * - The message is larger than our underlying buffer.
722 	 * - The ALQ is being shutdown.
723 	 * - There is insufficient free space in our underlying buffer
724 	 *   to accept the message and the user can't wait for space.
725 	 * - There is insufficient free space in our underlying buffer
726 	 *   to accept the message and the alq is inactive due to prior
727 	 *   use of the ALQ_NOACTIVATE flag (which would lead to deadlock).
728 	 */
729 	if (len > alq->aq_buflen ||
730 	    alq->aq_flags & AQ_SHUTDOWN ||
731 	    (((flags & ALQ_NOWAIT) || (!(alq->aq_flags & AQ_ACTIVE) &&
732 	    HAS_PENDING_DATA(alq))) && contigbytes < len)) {
733 		ALQ_UNLOCK(alq);
734 		return (NULL);
735 	}
736 
737 	/*
738 	 * If we want ordered writes and there is already at least one thread
739 	 * waiting for resources to become available, sleep until we're woken.
740 	 */
741 	if (alq->aq_flags & AQ_ORDERED && alq->aq_waiters > 0) {
742 		KASSERT(!(flags & ALQ_NOWAIT),
743 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
744 		alq->aq_waiters++;
745 		msleep_spin(&alq->aq_waiters, &alq->aq_mtx, "alqgnord", 0);
746 		alq->aq_waiters--;
747 	}
748 
749 	/*
750 	 * (ALQ_WAITOK && contigbytes < len) or contigbytes >= len, either enter
751 	 * while loop and sleep until we have enough contiguous free bytes
752 	 * (former) or skip (latter). If AQ_ORDERED is set, only 1 thread at a
753 	 * time will be in this loop. Otherwise, multiple threads may be
754 	 * sleeping here competing for ALQ resources.
755 	 */
756 	while (contigbytes < len && !(alq->aq_flags & AQ_SHUTDOWN)) {
757 		KASSERT(!(flags & ALQ_NOWAIT),
758 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
759 		alq->aq_flags |= AQ_WANTED;
760 		alq->aq_waiters++;
761 		if (waitchan)
762 			wakeup(waitchan);
763 		msleep_spin(alq, &alq->aq_mtx, "alqgnres", 0);
764 		alq->aq_waiters--;
765 
766 		if (alq->aq_writehead <= alq->aq_writetail)
767 			contigbytes = alq->aq_freebytes;
768 		else
769 			contigbytes = alq->aq_buflen - alq->aq_writehead;
770 
771 		/*
772 		 * If we're the first thread to wake after an AQ_WANTED wakeup
773 		 * but there isn't enough free space for us, we're going to loop
774 		 * and sleep again. If there are other threads waiting in this
775 		 * loop, schedule a wakeup so that they can see if the space
776 		 * they require is available.
777 		 */
778 		if (alq->aq_waiters > 0 && !(alq->aq_flags & AQ_ORDERED) &&
779 		    contigbytes < len && !(alq->aq_flags & AQ_WANTED))
780 			waitchan = alq;
781 		else
782 			waitchan = NULL;
783 	}
784 
785 	/*
786 	 * If there are waiters, we need to signal the waiting threads after we
787 	 * complete our work. The alq ptr is used as a wait channel for threads
788 	 * requiring resources to be freed up. In the AQ_ORDERED case, threads
789 	 * are not allowed to concurrently compete for resources in the above
790 	 * while loop, so we use a different wait channel in this case.
791 	 */
792 	if (alq->aq_waiters > 0) {
793 		if (alq->aq_flags & AQ_ORDERED)
794 			waitchan = &alq->aq_waiters;
795 		else
796 			waitchan = alq;
797 	} else
798 		waitchan = NULL;
799 
800 	/* Bail if we're shutting down. */
801 	if (alq->aq_flags & AQ_SHUTDOWN) {
802 		ALQ_UNLOCK(alq);
803 		if (waitchan != NULL)
804 			wakeup_one(waitchan);
805 		return (NULL);
806 	}
807 
808 	/*
809 	 * If we are here, we have a contiguous number of bytes >= len
810 	 * available in our buffer starting at aq_writehead.
811 	 */
812 	alq->aq_getpost.ae_data = alq->aq_entbuf + alq->aq_writehead;
813 	alq->aq_getpost.ae_bytesused = len;
814 
815 	return (&alq->aq_getpost);
816 }
817 
818 struct ale *
819 alq_get(struct alq *alq, int flags)
820 {
821 	/* Should only be called in fixed length message (legacy) mode. */
822 	KASSERT((alq->aq_flags & AQ_LEGACY),
823 	    ("%s: fixed length get on variable length queue", __func__));
824 	return (alq_getn(alq, alq->aq_entlen, flags));
825 }
826 
827 void
828 alq_post_flags(struct alq *alq, struct ale *ale, int flags)
829 {
830 	int activate;
831 	void *waitchan;
832 
833 	activate = 0;
834 
835 	if (ale->ae_bytesused > 0) {
836 		if (!(alq->aq_flags & AQ_ACTIVE) &&
837 		    !(flags & ALQ_NOACTIVATE)) {
838 			alq->aq_flags |= AQ_ACTIVE;
839 			activate = 1;
840 		}
841 
842 		alq->aq_writehead += ale->ae_bytesused;
843 		alq->aq_freebytes -= ale->ae_bytesused;
844 
845 		/* Wrap aq_writehead if we filled to the end of the buffer. */
846 		if (alq->aq_writehead == alq->aq_buflen)
847 			alq->aq_writehead = 0;
848 
849 		KASSERT((alq->aq_writehead >= 0 &&
850 		    alq->aq_writehead < alq->aq_buflen),
851 		    ("%s: aq_writehead < 0 || aq_writehead >= aq_buflen",
852 		    __func__));
853 
854 		KASSERT((HAS_PENDING_DATA(alq)), ("%s: queue empty!", __func__));
855 	}
856 
857 	/*
858 	 * If there are waiters, we need to signal the waiting threads after we
859 	 * complete our work. The alq ptr is used as a wait channel for threads
860 	 * requiring resources to be freed up. In the AQ_ORDERED case, threads
861 	 * are not allowed to concurrently compete for resources in the
862 	 * alq_getn() while loop, so we use a different wait channel in this case.
863 	 */
864 	if (alq->aq_waiters > 0) {
865 		if (alq->aq_flags & AQ_ORDERED)
866 			waitchan = &alq->aq_waiters;
867 		else
868 			waitchan = alq;
869 	} else
870 		waitchan = NULL;
871 
872 	ALQ_UNLOCK(alq);
873 
874 	if (activate) {
875 		ALD_LOCK();
876 		ald_activate(alq);
877 		ALD_UNLOCK();
878 	}
879 
880 	/* NB: We rely on wakeup_one waking threads in a FIFO manner. */
881 	if (waitchan != NULL)
882 		wakeup_one(waitchan);
883 }
884 
885 void
886 alq_flush(struct alq *alq)
887 {
888 	int needwakeup = 0;
889 
890 	ALD_LOCK();
891 	ALQ_LOCK(alq);
892 
893 	/*
894 	 * Pull the lever iff there is data to flush and we're
895 	 * not already in the middle of a flush operation.
896 	 */
897 	if (HAS_PENDING_DATA(alq) && !(alq->aq_flags & AQ_FLUSHING)) {
898 		if (alq->aq_flags & AQ_ACTIVE)
899 			ald_deactivate(alq);
900 
901 		ALD_UNLOCK();
902 		needwakeup = alq_doio(alq);
903 	} else
904 		ALD_UNLOCK();
905 
906 	ALQ_UNLOCK(alq);
907 
908 	if (needwakeup)
909 		wakeup_one(alq);
910 }
911 
912 /*
913  * Flush remaining data, close the file and free all resources.
914  */
915 void
916 alq_close(struct alq *alq)
917 {
918 	/* Only flush and destroy alq if not already shutting down. */
919 	if (ald_rem(alq) == 0)
920 		alq_destroy(alq);
921 }
922 
923 static int
924 alq_load_handler(module_t mod, int what, void *arg)
925 {
926 	int ret;
927 
928 	ret = 0;
929 
930 	switch (what) {
931 	case MOD_LOAD:
932 	case MOD_SHUTDOWN:
933 		break;
934 
935 	case MOD_QUIESCE:
936 		ALD_LOCK();
937 		/* Only allow unload if there are no open queues. */
938 		if (LIST_FIRST(&ald_queues) == NULL) {
939 			ald_shutingdown = 1;
940 			ALD_UNLOCK();
941 			ald_shutdown(NULL, 0);
942 			mtx_destroy(&ald_mtx);
943 		} else {
944 			ALD_UNLOCK();
945 			ret = EBUSY;
946 		}
947 		break;
948 
949 	case MOD_UNLOAD:
950 		/* If MOD_QUIESCE failed we must fail here too. */
951 		if (ald_shutingdown == 0)
952 			ret = EBUSY;
953 		break;
954 
955 	default:
956 		ret = EINVAL;
957 		break;
958 	}
959 
960 	return (ret);
961 }
962 
963 static moduledata_t alq_mod =
964 {
965 	"alq",
966 	alq_load_handler,
967 	NULL
968 };
969 
970 DECLARE_MODULE(alq, alq_mod, SI_SUB_SMP, SI_ORDER_ANY);
971 MODULE_VERSION(alq, 1);
972