xref: /freebsd/sys/kern/kern_alq.c (revision 43a5ec4eb41567cc92586503212743d89686d78f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
5  * Copyright (c) 2008-2009, Lawrence Stewart <lstewart@freebsd.org>
6  * Copyright (c) 2009-2010, The FreeBSD Foundation
7  * All rights reserved.
8  *
9  * Portions of this software were developed at the Centre for Advanced
10  * Internet Architectures, Swinburne University of Technology, Melbourne,
11  * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice unmodified, this list of conditions, and the following
18  *    disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include "opt_mac.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/kthread.h>
44 #include <sys/lock.h>
45 #include <sys/mount.h>
46 #include <sys/mutex.h>
47 #include <sys/namei.h>
48 #include <sys/proc.h>
49 #include <sys/vnode.h>
50 #include <sys/alq.h>
51 #include <sys/malloc.h>
52 #include <sys/unistd.h>
53 #include <sys/fcntl.h>
54 #include <sys/eventhandler.h>
55 
56 #include <security/mac/mac_framework.h>
57 
58 /* Async. Logging Queue */
59 struct alq {
60 	char	*aq_entbuf;		/* Buffer for stored entries */
61 	int	aq_entmax;		/* Max entries */
62 	int	aq_entlen;		/* Entry length */
63 	int	aq_freebytes;		/* Bytes available in buffer */
64 	int	aq_buflen;		/* Total length of our buffer */
65 	int	aq_writehead;		/* Location for next write */
66 	int	aq_writetail;		/* Flush starts at this location */
67 	int	aq_wrapearly;		/* # bytes left blank at end of buf */
68 	int	aq_flags;		/* Queue flags */
69 	int	aq_waiters;		/* Num threads waiting for resources
70 					 * NB: Used as a wait channel so must
71 					 * not be first field in the alq struct
72 					 */
73 	struct	ale	aq_getpost;	/* ALE for use by get/post */
74 	struct mtx	aq_mtx;		/* Queue lock */
75 	struct vnode	*aq_vp;		/* Open vnode handle */
76 	struct ucred	*aq_cred;	/* Credentials of the opening thread */
77 	LIST_ENTRY(alq)	aq_act;		/* List of active queues */
78 	LIST_ENTRY(alq)	aq_link;	/* List of all queues */
79 };
80 
81 #define	AQ_WANTED	0x0001		/* Wakeup sleeper when io is done */
82 #define	AQ_ACTIVE	0x0002		/* on the active list */
83 #define	AQ_FLUSHING	0x0004		/* doing IO */
84 #define	AQ_SHUTDOWN	0x0008		/* Queue no longer valid */
85 #define	AQ_ORDERED	0x0010		/* Queue enforces ordered writes */
86 #define	AQ_LEGACY	0x0020		/* Legacy queue (fixed length writes) */
87 
88 #define	ALQ_LOCK(alq)	mtx_lock_spin(&(alq)->aq_mtx)
89 #define	ALQ_UNLOCK(alq)	mtx_unlock_spin(&(alq)->aq_mtx)
90 
91 #define HAS_PENDING_DATA(alq) ((alq)->aq_freebytes != (alq)->aq_buflen)
92 
93 static MALLOC_DEFINE(M_ALD, "ALD", "ALD");
94 
95 /*
96  * The ald_mtx protects the ald_queues list and the ald_active list.
97  */
98 static struct mtx ald_mtx;
99 static LIST_HEAD(, alq) ald_queues;
100 static LIST_HEAD(, alq) ald_active;
101 static int ald_shutingdown = 0;
102 struct thread *ald_thread;
103 static struct proc *ald_proc;
104 static eventhandler_tag alq_eventhandler_tag = NULL;
105 
106 #define	ALD_LOCK()	mtx_lock(&ald_mtx)
107 #define	ALD_UNLOCK()	mtx_unlock(&ald_mtx)
108 
109 /* Daemon functions */
110 static int ald_add(struct alq *);
111 static int ald_rem(struct alq *);
112 static void ald_startup(void *);
113 static void ald_daemon(void);
114 static void ald_shutdown(void *, int);
115 static void ald_activate(struct alq *);
116 static void ald_deactivate(struct alq *);
117 
118 /* Internal queue functions */
119 static void alq_shutdown(struct alq *);
120 static void alq_destroy(struct alq *);
121 static int alq_doio(struct alq *);
122 
123 /*
124  * Add a new queue to the global list.  Fail if we're shutting down.
125  */
126 static int
127 ald_add(struct alq *alq)
128 {
129 	int error;
130 
131 	error = 0;
132 
133 	ALD_LOCK();
134 	if (ald_shutingdown) {
135 		error = EBUSY;
136 		goto done;
137 	}
138 	LIST_INSERT_HEAD(&ald_queues, alq, aq_link);
139 done:
140 	ALD_UNLOCK();
141 	return (error);
142 }
143 
144 /*
145  * Remove a queue from the global list unless we're shutting down.  If so,
146  * the ald will take care of cleaning up it's resources.
147  */
148 static int
149 ald_rem(struct alq *alq)
150 {
151 	int error;
152 
153 	error = 0;
154 
155 	ALD_LOCK();
156 	if (ald_shutingdown) {
157 		error = EBUSY;
158 		goto done;
159 	}
160 	LIST_REMOVE(alq, aq_link);
161 done:
162 	ALD_UNLOCK();
163 	return (error);
164 }
165 
166 /*
167  * Put a queue on the active list.  This will schedule it for writing.
168  */
169 static void
170 ald_activate(struct alq *alq)
171 {
172 	LIST_INSERT_HEAD(&ald_active, alq, aq_act);
173 	wakeup(&ald_active);
174 }
175 
176 static void
177 ald_deactivate(struct alq *alq)
178 {
179 	LIST_REMOVE(alq, aq_act);
180 	alq->aq_flags &= ~AQ_ACTIVE;
181 }
182 
183 static void
184 ald_startup(void *unused)
185 {
186 	mtx_init(&ald_mtx, "ALDmtx", NULL, MTX_DEF|MTX_QUIET);
187 	LIST_INIT(&ald_queues);
188 	LIST_INIT(&ald_active);
189 }
190 
191 static void
192 ald_daemon(void)
193 {
194 	int needwakeup;
195 	struct alq *alq;
196 
197 	ald_thread = FIRST_THREAD_IN_PROC(ald_proc);
198 
199 	alq_eventhandler_tag = EVENTHANDLER_REGISTER(shutdown_pre_sync,
200 	    ald_shutdown, NULL, SHUTDOWN_PRI_FIRST);
201 
202 	ALD_LOCK();
203 
204 	for (;;) {
205 		while ((alq = LIST_FIRST(&ald_active)) == NULL &&
206 		    !ald_shutingdown)
207 			mtx_sleep(&ald_active, &ald_mtx, PWAIT, "aldslp", 0);
208 
209 		/* Don't shutdown until all active ALQs are flushed. */
210 		if (ald_shutingdown && alq == NULL) {
211 			ALD_UNLOCK();
212 			break;
213 		}
214 
215 		ALQ_LOCK(alq);
216 		ald_deactivate(alq);
217 		ALD_UNLOCK();
218 		needwakeup = alq_doio(alq);
219 		ALQ_UNLOCK(alq);
220 		if (needwakeup)
221 			wakeup_one(alq);
222 		ALD_LOCK();
223 	}
224 
225 	kproc_exit(0);
226 }
227 
228 static void
229 ald_shutdown(void *arg, int howto)
230 {
231 	struct alq *alq;
232 
233 	ALD_LOCK();
234 
235 	/* Ensure no new queues can be created. */
236 	ald_shutingdown = 1;
237 
238 	/* Shutdown all ALQs prior to terminating the ald_daemon. */
239 	while ((alq = LIST_FIRST(&ald_queues)) != NULL) {
240 		LIST_REMOVE(alq, aq_link);
241 		ALD_UNLOCK();
242 		alq_shutdown(alq);
243 		ALD_LOCK();
244 	}
245 
246 	/* At this point, all ALQs are flushed and shutdown. */
247 
248 	/*
249 	 * Wake ald_daemon so that it exits. It won't be able to do
250 	 * anything until we mtx_sleep because we hold the ald_mtx.
251 	 */
252 	wakeup(&ald_active);
253 
254 	/* Wait for ald_daemon to exit. */
255 	mtx_sleep(ald_proc, &ald_mtx, PWAIT, "aldslp", 0);
256 
257 	ALD_UNLOCK();
258 }
259 
260 static void
261 alq_shutdown(struct alq *alq)
262 {
263 	ALQ_LOCK(alq);
264 
265 	/* Stop any new writers. */
266 	alq->aq_flags |= AQ_SHUTDOWN;
267 
268 	/*
269 	 * If the ALQ isn't active but has unwritten data (possible if
270 	 * the ALQ_NOACTIVATE flag has been used), explicitly activate the
271 	 * ALQ here so that the pending data gets flushed by the ald_daemon.
272 	 */
273 	if (!(alq->aq_flags & AQ_ACTIVE) && HAS_PENDING_DATA(alq)) {
274 		alq->aq_flags |= AQ_ACTIVE;
275 		ALQ_UNLOCK(alq);
276 		ALD_LOCK();
277 		ald_activate(alq);
278 		ALD_UNLOCK();
279 		ALQ_LOCK(alq);
280 	}
281 
282 	/* Drain IO */
283 	while (alq->aq_flags & AQ_ACTIVE) {
284 		alq->aq_flags |= AQ_WANTED;
285 		msleep_spin(alq, &alq->aq_mtx, "aldclose", 0);
286 	}
287 	ALQ_UNLOCK(alq);
288 
289 	vn_close(alq->aq_vp, FWRITE, alq->aq_cred,
290 	    curthread);
291 	crfree(alq->aq_cred);
292 }
293 
294 void
295 alq_destroy(struct alq *alq)
296 {
297 	/* Drain all pending IO. */
298 	alq_shutdown(alq);
299 
300 	mtx_destroy(&alq->aq_mtx);
301 	free(alq->aq_entbuf, M_ALD);
302 	free(alq, M_ALD);
303 }
304 
305 /*
306  * Flush all pending data to disk.  This operation will block.
307  */
308 static int
309 alq_doio(struct alq *alq)
310 {
311 	struct thread *td;
312 	struct mount *mp;
313 	struct vnode *vp;
314 	struct uio auio;
315 	struct iovec aiov[2];
316 	int totlen;
317 	int iov;
318 	int wrapearly;
319 
320 	KASSERT((HAS_PENDING_DATA(alq)), ("%s: queue empty!", __func__));
321 
322 	vp = alq->aq_vp;
323 	td = curthread;
324 	totlen = 0;
325 	iov = 1;
326 	wrapearly = alq->aq_wrapearly;
327 
328 	bzero(&aiov, sizeof(aiov));
329 	bzero(&auio, sizeof(auio));
330 
331 	/* Start the write from the location of our buffer tail pointer. */
332 	aiov[0].iov_base = alq->aq_entbuf + alq->aq_writetail;
333 
334 	if (alq->aq_writetail < alq->aq_writehead) {
335 		/* Buffer not wrapped. */
336 		totlen = aiov[0].iov_len = alq->aq_writehead - alq->aq_writetail;
337 	} else if (alq->aq_writehead == 0) {
338 		/* Buffer not wrapped (special case to avoid an empty iov). */
339 		totlen = aiov[0].iov_len = alq->aq_buflen - alq->aq_writetail -
340 		    wrapearly;
341 	} else {
342 		/*
343 		 * Buffer wrapped, requires 2 aiov entries:
344 		 * - first is from writetail to end of buffer
345 		 * - second is from start of buffer to writehead
346 		 */
347 		aiov[0].iov_len = alq->aq_buflen - alq->aq_writetail -
348 		    wrapearly;
349 		iov++;
350 		aiov[1].iov_base = alq->aq_entbuf;
351 		aiov[1].iov_len =  alq->aq_writehead;
352 		totlen = aiov[0].iov_len + aiov[1].iov_len;
353 	}
354 
355 	alq->aq_flags |= AQ_FLUSHING;
356 	ALQ_UNLOCK(alq);
357 
358 	auio.uio_iov = &aiov[0];
359 	auio.uio_offset = 0;
360 	auio.uio_segflg = UIO_SYSSPACE;
361 	auio.uio_rw = UIO_WRITE;
362 	auio.uio_iovcnt = iov;
363 	auio.uio_resid = totlen;
364 	auio.uio_td = td;
365 
366 	/*
367 	 * Do all of the junk required to write now.
368 	 */
369 	vn_start_write(vp, &mp, V_WAIT);
370 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
371 	/*
372 	 * XXX: VOP_WRITE error checks are ignored.
373 	 */
374 #ifdef MAC
375 	if (mac_vnode_check_write(alq->aq_cred, NOCRED, vp) == 0)
376 #endif
377 		VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, alq->aq_cred);
378 	VOP_UNLOCK(vp);
379 	vn_finished_write(mp);
380 
381 	ALQ_LOCK(alq);
382 	alq->aq_flags &= ~AQ_FLUSHING;
383 
384 	/* Adjust writetail as required, taking into account wrapping. */
385 	alq->aq_writetail = (alq->aq_writetail + totlen + wrapearly) %
386 	    alq->aq_buflen;
387 	alq->aq_freebytes += totlen + wrapearly;
388 
389 	/*
390 	 * If we just flushed part of the buffer which wrapped, reset the
391 	 * wrapearly indicator.
392 	 */
393 	if (wrapearly)
394 		alq->aq_wrapearly = 0;
395 
396 	/*
397 	 * If we just flushed the buffer completely, reset indexes to 0 to
398 	 * minimise buffer wraps.
399 	 * This is also required to ensure alq_getn() can't wedge itself.
400 	 */
401 	if (!HAS_PENDING_DATA(alq))
402 		alq->aq_writehead = alq->aq_writetail = 0;
403 
404 	KASSERT((alq->aq_writetail >= 0 && alq->aq_writetail < alq->aq_buflen),
405 	    ("%s: aq_writetail < 0 || aq_writetail >= aq_buflen", __func__));
406 
407 	if (alq->aq_flags & AQ_WANTED) {
408 		alq->aq_flags &= ~AQ_WANTED;
409 		return (1);
410 	}
411 
412 	return(0);
413 }
414 
415 static struct kproc_desc ald_kp = {
416         "ALQ Daemon",
417         ald_daemon,
418         &ald_proc
419 };
420 
421 SYSINIT(aldthread, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &ald_kp);
422 SYSINIT(ald, SI_SUB_LOCK, SI_ORDER_ANY, ald_startup, NULL);
423 
424 /* User visible queue functions */
425 
426 /*
427  * Create the queue data structure, allocate the buffer, and open the file.
428  */
429 
430 int
431 alq_open_flags(struct alq **alqp, const char *file, struct ucred *cred, int cmode,
432     int size, int flags)
433 {
434 	struct nameidata nd;
435 	struct alq *alq;
436 	int oflags;
437 	int error;
438 
439 	KASSERT((size > 0), ("%s: size <= 0", __func__));
440 
441 	*alqp = NULL;
442 
443 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, file);
444 	oflags = FWRITE | O_NOFOLLOW | O_CREAT;
445 
446 	error = vn_open_cred(&nd, &oflags, cmode, 0, cred, NULL);
447 	if (error)
448 		return (error);
449 
450 	NDFREE(&nd, NDF_ONLY_PNBUF);
451 	/* We just unlock so we hold a reference */
452 	VOP_UNLOCK(nd.ni_vp);
453 
454 	alq = malloc(sizeof(*alq), M_ALD, M_WAITOK|M_ZERO);
455 	alq->aq_vp = nd.ni_vp;
456 	alq->aq_cred = crhold(cred);
457 
458 	mtx_init(&alq->aq_mtx, "ALD Queue", NULL, MTX_SPIN|MTX_QUIET);
459 
460 	alq->aq_buflen = size;
461 	alq->aq_entmax = 0;
462 	alq->aq_entlen = 0;
463 
464 	alq->aq_freebytes = alq->aq_buflen;
465 	alq->aq_entbuf = malloc(alq->aq_buflen, M_ALD, M_WAITOK|M_ZERO);
466 	alq->aq_writehead = alq->aq_writetail = 0;
467 	if (flags & ALQ_ORDERED)
468 		alq->aq_flags |= AQ_ORDERED;
469 
470 	if ((error = ald_add(alq)) != 0) {
471 		alq_destroy(alq);
472 		return (error);
473 	}
474 
475 	*alqp = alq;
476 
477 	return (0);
478 }
479 
480 int
481 alq_open(struct alq **alqp, const char *file, struct ucred *cred, int cmode,
482     int size, int count)
483 {
484 	int ret;
485 
486 	KASSERT((count >= 0), ("%s: count < 0", __func__));
487 
488 	if (count > 0) {
489 		if ((ret = alq_open_flags(alqp, file, cred, cmode,
490 		    size*count, 0)) == 0) {
491 			(*alqp)->aq_flags |= AQ_LEGACY;
492 			(*alqp)->aq_entmax = count;
493 			(*alqp)->aq_entlen = size;
494 		}
495 	} else
496 		ret = alq_open_flags(alqp, file, cred, cmode, size, 0);
497 
498 	return (ret);
499 }
500 
501 /*
502  * Copy a new entry into the queue.  If the operation would block either
503  * wait or return an error depending on the value of waitok.
504  */
505 int
506 alq_writen(struct alq *alq, void *data, int len, int flags)
507 {
508 	int activate, copy, ret;
509 	void *waitchan;
510 
511 	KASSERT((len > 0 && len <= alq->aq_buflen),
512 	    ("%s: len <= 0 || len > aq_buflen", __func__));
513 
514 	activate = ret = 0;
515 	copy = len;
516 	waitchan = NULL;
517 
518 	ALQ_LOCK(alq);
519 
520 	/*
521 	 * Fail to perform the write and return EWOULDBLOCK if:
522 	 * - The message is larger than our underlying buffer.
523 	 * - The ALQ is being shutdown.
524 	 * - There is insufficient free space in our underlying buffer
525 	 *   to accept the message and the user can't wait for space.
526 	 * - There is insufficient free space in our underlying buffer
527 	 *   to accept the message and the alq is inactive due to prior
528 	 *   use of the ALQ_NOACTIVATE flag (which would lead to deadlock).
529 	 */
530 	if (len > alq->aq_buflen ||
531 	    alq->aq_flags & AQ_SHUTDOWN ||
532 	    (((flags & ALQ_NOWAIT) || (!(alq->aq_flags & AQ_ACTIVE) &&
533 	    HAS_PENDING_DATA(alq))) && alq->aq_freebytes < len)) {
534 		ALQ_UNLOCK(alq);
535 		return (EWOULDBLOCK);
536 	}
537 
538 	/*
539 	 * If we want ordered writes and there is already at least one thread
540 	 * waiting for resources to become available, sleep until we're woken.
541 	 */
542 	if (alq->aq_flags & AQ_ORDERED && alq->aq_waiters > 0) {
543 		KASSERT(!(flags & ALQ_NOWAIT),
544 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
545 		alq->aq_waiters++;
546 		msleep_spin(&alq->aq_waiters, &alq->aq_mtx, "alqwnord", 0);
547 		alq->aq_waiters--;
548 	}
549 
550 	/*
551 	 * (ALQ_WAITOK && aq_freebytes < len) or aq_freebytes >= len, either
552 	 * enter while loop and sleep until we have enough free bytes (former)
553 	 * or skip (latter). If AQ_ORDERED is set, only 1 thread at a time will
554 	 * be in this loop. Otherwise, multiple threads may be sleeping here
555 	 * competing for ALQ resources.
556 	 */
557 	while (alq->aq_freebytes < len && !(alq->aq_flags & AQ_SHUTDOWN)) {
558 		KASSERT(!(flags & ALQ_NOWAIT),
559 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
560 		alq->aq_flags |= AQ_WANTED;
561 		alq->aq_waiters++;
562 		if (waitchan)
563 			wakeup(waitchan);
564 		msleep_spin(alq, &alq->aq_mtx, "alqwnres", 0);
565 		alq->aq_waiters--;
566 
567 		/*
568 		 * If we're the first thread to wake after an AQ_WANTED wakeup
569 		 * but there isn't enough free space for us, we're going to loop
570 		 * and sleep again. If there are other threads waiting in this
571 		 * loop, schedule a wakeup so that they can see if the space
572 		 * they require is available.
573 		 */
574 		if (alq->aq_waiters > 0 && !(alq->aq_flags & AQ_ORDERED) &&
575 		    alq->aq_freebytes < len && !(alq->aq_flags & AQ_WANTED))
576 			waitchan = alq;
577 		else
578 			waitchan = NULL;
579 	}
580 
581 	/*
582 	 * If there are waiters, we need to signal the waiting threads after we
583 	 * complete our work. The alq ptr is used as a wait channel for threads
584 	 * requiring resources to be freed up. In the AQ_ORDERED case, threads
585 	 * are not allowed to concurrently compete for resources in the above
586 	 * while loop, so we use a different wait channel in this case.
587 	 */
588 	if (alq->aq_waiters > 0) {
589 		if (alq->aq_flags & AQ_ORDERED)
590 			waitchan = &alq->aq_waiters;
591 		else
592 			waitchan = alq;
593 	} else
594 		waitchan = NULL;
595 
596 	/* Bail if we're shutting down. */
597 	if (alq->aq_flags & AQ_SHUTDOWN) {
598 		ret = EWOULDBLOCK;
599 		goto unlock;
600 	}
601 
602 	/*
603 	 * If we need to wrap the buffer to accommodate the write,
604 	 * we'll need 2 calls to bcopy.
605 	 */
606 	if ((alq->aq_buflen - alq->aq_writehead) < len)
607 		copy = alq->aq_buflen - alq->aq_writehead;
608 
609 	/* Copy message (or part thereof if wrap required) to the buffer. */
610 	bcopy(data, alq->aq_entbuf + alq->aq_writehead, copy);
611 	alq->aq_writehead += copy;
612 
613 	if (alq->aq_writehead >= alq->aq_buflen) {
614 		KASSERT((alq->aq_writehead == alq->aq_buflen),
615 		    ("%s: alq->aq_writehead (%d) > alq->aq_buflen (%d)",
616 		    __func__,
617 		    alq->aq_writehead,
618 		    alq->aq_buflen));
619 		alq->aq_writehead = 0;
620 	}
621 
622 	if (copy != len) {
623 		/*
624 		 * Wrap the buffer by copying the remainder of our message
625 		 * to the start of the buffer and resetting aq_writehead.
626 		 */
627 		bcopy(((uint8_t *)data)+copy, alq->aq_entbuf, len - copy);
628 		alq->aq_writehead = len - copy;
629 	}
630 
631 	KASSERT((alq->aq_writehead >= 0 && alq->aq_writehead < alq->aq_buflen),
632 	    ("%s: aq_writehead < 0 || aq_writehead >= aq_buflen", __func__));
633 
634 	alq->aq_freebytes -= len;
635 
636 	if (!(alq->aq_flags & AQ_ACTIVE) && !(flags & ALQ_NOACTIVATE)) {
637 		alq->aq_flags |= AQ_ACTIVE;
638 		activate = 1;
639 	}
640 
641 	KASSERT((HAS_PENDING_DATA(alq)), ("%s: queue empty!", __func__));
642 
643 unlock:
644 	ALQ_UNLOCK(alq);
645 
646 	if (activate) {
647 		ALD_LOCK();
648 		ald_activate(alq);
649 		ALD_UNLOCK();
650 	}
651 
652 	/* NB: We rely on wakeup_one waking threads in a FIFO manner. */
653 	if (waitchan != NULL)
654 		wakeup_one(waitchan);
655 
656 	return (ret);
657 }
658 
659 int
660 alq_write(struct alq *alq, void *data, int flags)
661 {
662 	/* Should only be called in fixed length message (legacy) mode. */
663 	KASSERT((alq->aq_flags & AQ_LEGACY),
664 	    ("%s: fixed length write on variable length queue", __func__));
665 	return (alq_writen(alq, data, alq->aq_entlen, flags));
666 }
667 
668 /*
669  * Retrieve a pointer for the ALQ to write directly into, avoiding bcopy.
670  */
671 struct ale *
672 alq_getn(struct alq *alq, int len, int flags)
673 {
674 	int contigbytes;
675 	void *waitchan;
676 
677 	KASSERT((len > 0 && len <= alq->aq_buflen),
678 	    ("%s: len <= 0 || len > alq->aq_buflen", __func__));
679 
680 	waitchan = NULL;
681 
682 	ALQ_LOCK(alq);
683 
684 	/*
685 	 * Determine the number of free contiguous bytes.
686 	 * We ensure elsewhere that if aq_writehead == aq_writetail because
687 	 * the buffer is empty, they will both be set to 0 and therefore
688 	 * aq_freebytes == aq_buflen and is fully contiguous.
689 	 * If they are equal and the buffer is not empty, aq_freebytes will
690 	 * be 0 indicating the buffer is full.
691 	 */
692 	if (alq->aq_writehead <= alq->aq_writetail)
693 		contigbytes = alq->aq_freebytes;
694 	else {
695 		contigbytes = alq->aq_buflen - alq->aq_writehead;
696 
697 		if (contigbytes < len) {
698 			/*
699 			 * Insufficient space at end of buffer to handle a
700 			 * contiguous write. Wrap early if there's space at
701 			 * the beginning. This will leave a hole at the end
702 			 * of the buffer which we will have to skip over when
703 			 * flushing the buffer to disk.
704 			 */
705 			if (alq->aq_writetail >= len || flags & ALQ_WAITOK) {
706 				/* Keep track of # bytes left blank. */
707 				alq->aq_wrapearly = contigbytes;
708 				/* Do the wrap and adjust counters. */
709 				contigbytes = alq->aq_freebytes =
710 				    alq->aq_writetail;
711 				alq->aq_writehead = 0;
712 			}
713 		}
714 	}
715 
716 	/*
717 	 * Return a NULL ALE if:
718 	 * - The message is larger than our underlying buffer.
719 	 * - The ALQ is being shutdown.
720 	 * - There is insufficient free space in our underlying buffer
721 	 *   to accept the message and the user can't wait for space.
722 	 * - There is insufficient free space in our underlying buffer
723 	 *   to accept the message and the alq is inactive due to prior
724 	 *   use of the ALQ_NOACTIVATE flag (which would lead to deadlock).
725 	 */
726 	if (len > alq->aq_buflen ||
727 	    alq->aq_flags & AQ_SHUTDOWN ||
728 	    (((flags & ALQ_NOWAIT) || (!(alq->aq_flags & AQ_ACTIVE) &&
729 	    HAS_PENDING_DATA(alq))) && contigbytes < len)) {
730 		ALQ_UNLOCK(alq);
731 		return (NULL);
732 	}
733 
734 	/*
735 	 * If we want ordered writes and there is already at least one thread
736 	 * waiting for resources to become available, sleep until we're woken.
737 	 */
738 	if (alq->aq_flags & AQ_ORDERED && alq->aq_waiters > 0) {
739 		KASSERT(!(flags & ALQ_NOWAIT),
740 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
741 		alq->aq_waiters++;
742 		msleep_spin(&alq->aq_waiters, &alq->aq_mtx, "alqgnord", 0);
743 		alq->aq_waiters--;
744 	}
745 
746 	/*
747 	 * (ALQ_WAITOK && contigbytes < len) or contigbytes >= len, either enter
748 	 * while loop and sleep until we have enough contiguous free bytes
749 	 * (former) or skip (latter). If AQ_ORDERED is set, only 1 thread at a
750 	 * time will be in this loop. Otherwise, multiple threads may be
751 	 * sleeping here competing for ALQ resources.
752 	 */
753 	while (contigbytes < len && !(alq->aq_flags & AQ_SHUTDOWN)) {
754 		KASSERT(!(flags & ALQ_NOWAIT),
755 		    ("%s: ALQ_NOWAIT set but incorrectly ignored!", __func__));
756 		alq->aq_flags |= AQ_WANTED;
757 		alq->aq_waiters++;
758 		if (waitchan)
759 			wakeup(waitchan);
760 		msleep_spin(alq, &alq->aq_mtx, "alqgnres", 0);
761 		alq->aq_waiters--;
762 
763 		if (alq->aq_writehead <= alq->aq_writetail)
764 			contigbytes = alq->aq_freebytes;
765 		else
766 			contigbytes = alq->aq_buflen - alq->aq_writehead;
767 
768 		/*
769 		 * If we're the first thread to wake after an AQ_WANTED wakeup
770 		 * but there isn't enough free space for us, we're going to loop
771 		 * and sleep again. If there are other threads waiting in this
772 		 * loop, schedule a wakeup so that they can see if the space
773 		 * they require is available.
774 		 */
775 		if (alq->aq_waiters > 0 && !(alq->aq_flags & AQ_ORDERED) &&
776 		    contigbytes < len && !(alq->aq_flags & AQ_WANTED))
777 			waitchan = alq;
778 		else
779 			waitchan = NULL;
780 	}
781 
782 	/*
783 	 * If there are waiters, we need to signal the waiting threads after we
784 	 * complete our work. The alq ptr is used as a wait channel for threads
785 	 * requiring resources to be freed up. In the AQ_ORDERED case, threads
786 	 * are not allowed to concurrently compete for resources in the above
787 	 * while loop, so we use a different wait channel in this case.
788 	 */
789 	if (alq->aq_waiters > 0) {
790 		if (alq->aq_flags & AQ_ORDERED)
791 			waitchan = &alq->aq_waiters;
792 		else
793 			waitchan = alq;
794 	} else
795 		waitchan = NULL;
796 
797 	/* Bail if we're shutting down. */
798 	if (alq->aq_flags & AQ_SHUTDOWN) {
799 		ALQ_UNLOCK(alq);
800 		if (waitchan != NULL)
801 			wakeup_one(waitchan);
802 		return (NULL);
803 	}
804 
805 	/*
806 	 * If we are here, we have a contiguous number of bytes >= len
807 	 * available in our buffer starting at aq_writehead.
808 	 */
809 	alq->aq_getpost.ae_data = alq->aq_entbuf + alq->aq_writehead;
810 	alq->aq_getpost.ae_bytesused = len;
811 
812 	return (&alq->aq_getpost);
813 }
814 
815 struct ale *
816 alq_get(struct alq *alq, int flags)
817 {
818 	/* Should only be called in fixed length message (legacy) mode. */
819 	KASSERT((alq->aq_flags & AQ_LEGACY),
820 	    ("%s: fixed length get on variable length queue", __func__));
821 	return (alq_getn(alq, alq->aq_entlen, flags));
822 }
823 
824 void
825 alq_post_flags(struct alq *alq, struct ale *ale, int flags)
826 {
827 	int activate;
828 	void *waitchan;
829 
830 	activate = 0;
831 
832 	if (ale->ae_bytesused > 0) {
833 		if (!(alq->aq_flags & AQ_ACTIVE) &&
834 		    !(flags & ALQ_NOACTIVATE)) {
835 			alq->aq_flags |= AQ_ACTIVE;
836 			activate = 1;
837 		}
838 
839 		alq->aq_writehead += ale->ae_bytesused;
840 		alq->aq_freebytes -= ale->ae_bytesused;
841 
842 		/* Wrap aq_writehead if we filled to the end of the buffer. */
843 		if (alq->aq_writehead == alq->aq_buflen)
844 			alq->aq_writehead = 0;
845 
846 		KASSERT((alq->aq_writehead >= 0 &&
847 		    alq->aq_writehead < alq->aq_buflen),
848 		    ("%s: aq_writehead < 0 || aq_writehead >= aq_buflen",
849 		    __func__));
850 
851 		KASSERT((HAS_PENDING_DATA(alq)), ("%s: queue empty!", __func__));
852 	}
853 
854 	/*
855 	 * If there are waiters, we need to signal the waiting threads after we
856 	 * complete our work. The alq ptr is used as a wait channel for threads
857 	 * requiring resources to be freed up. In the AQ_ORDERED case, threads
858 	 * are not allowed to concurrently compete for resources in the
859 	 * alq_getn() while loop, so we use a different wait channel in this case.
860 	 */
861 	if (alq->aq_waiters > 0) {
862 		if (alq->aq_flags & AQ_ORDERED)
863 			waitchan = &alq->aq_waiters;
864 		else
865 			waitchan = alq;
866 	} else
867 		waitchan = NULL;
868 
869 	ALQ_UNLOCK(alq);
870 
871 	if (activate) {
872 		ALD_LOCK();
873 		ald_activate(alq);
874 		ALD_UNLOCK();
875 	}
876 
877 	/* NB: We rely on wakeup_one waking threads in a FIFO manner. */
878 	if (waitchan != NULL)
879 		wakeup_one(waitchan);
880 }
881 
882 void
883 alq_flush(struct alq *alq)
884 {
885 	int needwakeup = 0;
886 
887 	ALD_LOCK();
888 	ALQ_LOCK(alq);
889 
890 	/*
891 	 * Pull the lever iff there is data to flush and we're
892 	 * not already in the middle of a flush operation.
893 	 */
894 	if (HAS_PENDING_DATA(alq) && !(alq->aq_flags & AQ_FLUSHING)) {
895 		if (alq->aq_flags & AQ_ACTIVE)
896 			ald_deactivate(alq);
897 
898 		ALD_UNLOCK();
899 		needwakeup = alq_doio(alq);
900 	} else
901 		ALD_UNLOCK();
902 
903 	ALQ_UNLOCK(alq);
904 
905 	if (needwakeup)
906 		wakeup_one(alq);
907 }
908 
909 /*
910  * Flush remaining data, close the file and free all resources.
911  */
912 void
913 alq_close(struct alq *alq)
914 {
915 	/* Only flush and destroy alq if not already shutting down. */
916 	if (ald_rem(alq) == 0)
917 		alq_destroy(alq);
918 }
919 
920 static int
921 alq_load_handler(module_t mod, int what, void *arg)
922 {
923 	int ret;
924 
925 	ret = 0;
926 
927 	switch (what) {
928 	case MOD_LOAD:
929 	case MOD_SHUTDOWN:
930 		break;
931 
932 	case MOD_QUIESCE:
933 		ALD_LOCK();
934 		/* Only allow unload if there are no open queues. */
935 		if (LIST_FIRST(&ald_queues) == NULL) {
936 			ald_shutingdown = 1;
937 			ALD_UNLOCK();
938 			EVENTHANDLER_DEREGISTER(shutdown_pre_sync,
939 			    alq_eventhandler_tag);
940 			ald_shutdown(NULL, 0);
941 			mtx_destroy(&ald_mtx);
942 		} else {
943 			ALD_UNLOCK();
944 			ret = EBUSY;
945 		}
946 		break;
947 
948 	case MOD_UNLOAD:
949 		/* If MOD_QUIESCE failed we must fail here too. */
950 		if (ald_shutingdown == 0)
951 			ret = EBUSY;
952 		break;
953 
954 	default:
955 		ret = EINVAL;
956 		break;
957 	}
958 
959 	return (ret);
960 }
961 
962 static moduledata_t alq_mod =
963 {
964 	"alq",
965 	alq_load_handler,
966 	NULL
967 };
968 
969 DECLARE_MODULE(alq, alq_mod, SI_SUB_LAST, SI_ORDER_ANY);
970 MODULE_VERSION(alq, 1);
971