xref: /freebsd/sys/kern/kern_alq.c (revision 35a04710d7286aa9538917fd7f8e417dbee95b82)
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 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_mac.h"
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/kthread.h>
36 #include <sys/lock.h>
37 #include <sys/mount.h>
38 #include <sys/mutex.h>
39 #include <sys/namei.h>
40 #include <sys/proc.h>
41 #include <sys/vnode.h>
42 #include <sys/alq.h>
43 #include <sys/malloc.h>
44 #include <sys/unistd.h>
45 #include <sys/fcntl.h>
46 #include <sys/eventhandler.h>
47 
48 #include <security/mac/mac_framework.h>
49 
50 /* Async. Logging Queue */
51 struct alq {
52 	int	aq_entmax;		/* Max entries */
53 	int	aq_entlen;		/* Entry length */
54 	char	*aq_entbuf;		/* Buffer for stored entries */
55 	int	aq_flags;		/* Queue flags */
56 	struct mtx	aq_mtx;		/* Queue lock */
57 	struct vnode	*aq_vp;		/* Open vnode handle */
58 	struct ucred	*aq_cred;	/* Credentials of the opening thread */
59 	struct ale	*aq_first;	/* First ent */
60 	struct ale	*aq_entfree;	/* First free ent */
61 	struct ale	*aq_entvalid;	/* First ent valid for writing */
62 	LIST_ENTRY(alq)	aq_act;		/* List of active queues */
63 	LIST_ENTRY(alq)	aq_link;	/* List of all queues */
64 };
65 
66 #define	AQ_WANTED	0x0001		/* Wakeup sleeper when io is done */
67 #define	AQ_ACTIVE	0x0002		/* on the active list */
68 #define	AQ_FLUSHING	0x0004		/* doing IO */
69 #define	AQ_SHUTDOWN	0x0008		/* Queue no longer valid */
70 
71 #define	ALQ_LOCK(alq)	mtx_lock_spin(&(alq)->aq_mtx)
72 #define	ALQ_UNLOCK(alq)	mtx_unlock_spin(&(alq)->aq_mtx)
73 
74 static MALLOC_DEFINE(M_ALD, "ALD", "ALD");
75 
76 /*
77  * The ald_mtx protects the ald_queues list and the ald_active list.
78  */
79 static struct mtx ald_mtx;
80 static LIST_HEAD(, alq) ald_queues;
81 static LIST_HEAD(, alq) ald_active;
82 static int ald_shutingdown = 0;
83 struct thread *ald_thread;
84 static struct proc *ald_proc;
85 
86 #define	ALD_LOCK()	mtx_lock(&ald_mtx)
87 #define	ALD_UNLOCK()	mtx_unlock(&ald_mtx)
88 
89 /* Daemon functions */
90 static int ald_add(struct alq *);
91 static int ald_rem(struct alq *);
92 static void ald_startup(void *);
93 static void ald_daemon(void);
94 static void ald_shutdown(void *, int);
95 static void ald_activate(struct alq *);
96 static void ald_deactivate(struct alq *);
97 
98 /* Internal queue functions */
99 static void alq_shutdown(struct alq *);
100 static int alq_doio(struct alq *);
101 
102 
103 /*
104  * Add a new queue to the global list.  Fail if we're shutting down.
105  */
106 static int
107 ald_add(struct alq *alq)
108 {
109 	int error;
110 
111 	error = 0;
112 
113 	ALD_LOCK();
114 	if (ald_shutingdown) {
115 		error = EBUSY;
116 		goto done;
117 	}
118 	LIST_INSERT_HEAD(&ald_queues, alq, aq_link);
119 done:
120 	ALD_UNLOCK();
121 	return (error);
122 }
123 
124 /*
125  * Remove a queue from the global list unless we're shutting down.  If so,
126  * the ald will take care of cleaning up it's resources.
127  */
128 static int
129 ald_rem(struct alq *alq)
130 {
131 	int error;
132 
133 	error = 0;
134 
135 	ALD_LOCK();
136 	if (ald_shutingdown) {
137 		error = EBUSY;
138 		goto done;
139 	}
140 	LIST_REMOVE(alq, aq_link);
141 done:
142 	ALD_UNLOCK();
143 	return (error);
144 }
145 
146 /*
147  * Put a queue on the active list.  This will schedule it for writing.
148  */
149 static void
150 ald_activate(struct alq *alq)
151 {
152 	LIST_INSERT_HEAD(&ald_active, alq, aq_act);
153 	wakeup(&ald_active);
154 }
155 
156 static void
157 ald_deactivate(struct alq *alq)
158 {
159 	LIST_REMOVE(alq, aq_act);
160 	alq->aq_flags &= ~AQ_ACTIVE;
161 }
162 
163 static void
164 ald_startup(void *unused)
165 {
166 	mtx_init(&ald_mtx, "ALDmtx", NULL, MTX_DEF|MTX_QUIET);
167 	LIST_INIT(&ald_queues);
168 	LIST_INIT(&ald_active);
169 }
170 
171 static void
172 ald_daemon(void)
173 {
174 	int needwakeup;
175 	struct alq *alq;
176 
177 	ald_thread = FIRST_THREAD_IN_PROC(ald_proc);
178 
179 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ald_shutdown, NULL,
180 	    SHUTDOWN_PRI_FIRST);
181 
182 	ALD_LOCK();
183 
184 	for (;;) {
185 		while ((alq = LIST_FIRST(&ald_active)) == NULL)
186 			msleep(&ald_active, &ald_mtx, PWAIT, "aldslp", 0);
187 
188 		ALQ_LOCK(alq);
189 		ald_deactivate(alq);
190 		ALD_UNLOCK();
191 		needwakeup = alq_doio(alq);
192 		ALQ_UNLOCK(alq);
193 		if (needwakeup)
194 			wakeup(alq);
195 		ALD_LOCK();
196 	}
197 }
198 
199 static void
200 ald_shutdown(void *arg, int howto)
201 {
202 	struct alq *alq;
203 
204 	ALD_LOCK();
205 	ald_shutingdown = 1;
206 
207 	while ((alq = LIST_FIRST(&ald_queues)) != NULL) {
208 		LIST_REMOVE(alq, aq_link);
209 		ALD_UNLOCK();
210 		alq_shutdown(alq);
211 		ALD_LOCK();
212 	}
213 	ALD_UNLOCK();
214 }
215 
216 static void
217 alq_shutdown(struct alq *alq)
218 {
219 	ALQ_LOCK(alq);
220 
221 	/* Stop any new writers. */
222 	alq->aq_flags |= AQ_SHUTDOWN;
223 
224 	/* Drain IO */
225 	while (alq->aq_flags & (AQ_FLUSHING|AQ_ACTIVE)) {
226 		alq->aq_flags |= AQ_WANTED;
227 		ALQ_UNLOCK(alq);
228 		tsleep(alq, PWAIT, "aldclose", 0);
229 		ALQ_LOCK(alq);
230 	}
231 	ALQ_UNLOCK(alq);
232 
233 	vn_close(alq->aq_vp, FWRITE, alq->aq_cred,
234 	    curthread);
235 	crfree(alq->aq_cred);
236 }
237 
238 /*
239  * Flush all pending data to disk.  This operation will block.
240  */
241 static int
242 alq_doio(struct alq *alq)
243 {
244 	struct thread *td;
245 	struct mount *mp;
246 	struct vnode *vp;
247 	struct uio auio;
248 	struct iovec aiov[2];
249 	struct ale *ale;
250 	struct ale *alstart;
251 	int totlen;
252 	int iov;
253 	int vfslocked;
254 
255 	vp = alq->aq_vp;
256 	td = curthread;
257 	totlen = 0;
258 	iov = 0;
259 
260 	alstart = ale = alq->aq_entvalid;
261 	alq->aq_entvalid = NULL;
262 
263 	bzero(&aiov, sizeof(aiov));
264 	bzero(&auio, sizeof(auio));
265 
266 	do {
267 		if (aiov[iov].iov_base == NULL)
268 			aiov[iov].iov_base = ale->ae_data;
269 		aiov[iov].iov_len += alq->aq_entlen;
270 		totlen += alq->aq_entlen;
271 		/* Check to see if we're wrapping the buffer */
272 		if (ale->ae_data + alq->aq_entlen != ale->ae_next->ae_data)
273 			iov++;
274 		ale->ae_flags &= ~AE_VALID;
275 		ale = ale->ae_next;
276 	} while (ale->ae_flags & AE_VALID);
277 
278 	alq->aq_flags |= AQ_FLUSHING;
279 	ALQ_UNLOCK(alq);
280 
281 	if (iov == 2 || aiov[iov].iov_base == NULL)
282 		iov--;
283 
284 	auio.uio_iov = &aiov[0];
285 	auio.uio_offset = 0;
286 	auio.uio_segflg = UIO_SYSSPACE;
287 	auio.uio_rw = UIO_WRITE;
288 	auio.uio_iovcnt = iov + 1;
289 	auio.uio_resid = totlen;
290 	auio.uio_td = td;
291 
292 	/*
293 	 * Do all of the junk required to write now.
294 	 */
295 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
296 	vn_start_write(vp, &mp, V_WAIT);
297 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
298 	VOP_LEASE(vp, td, alq->aq_cred, LEASE_WRITE);
299 	/*
300 	 * XXX: VOP_WRITE error checks are ignored.
301 	 */
302 #ifdef MAC
303 	if (mac_vnode_check_write(alq->aq_cred, NOCRED, vp) == 0)
304 #endif
305 		VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, alq->aq_cred);
306 	VOP_UNLOCK(vp, 0, td);
307 	vn_finished_write(mp);
308 	VFS_UNLOCK_GIANT(vfslocked);
309 
310 	ALQ_LOCK(alq);
311 	alq->aq_flags &= ~AQ_FLUSHING;
312 
313 	if (alq->aq_entfree == NULL)
314 		alq->aq_entfree = alstart;
315 
316 	if (alq->aq_flags & AQ_WANTED) {
317 		alq->aq_flags &= ~AQ_WANTED;
318 		return (1);
319 	}
320 
321 	return(0);
322 }
323 
324 static struct kproc_desc ald_kp = {
325         "ALQ Daemon",
326         ald_daemon,
327         &ald_proc
328 };
329 
330 SYSINIT(aldthread, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &ald_kp)
331 SYSINIT(ald, SI_SUB_LOCK, SI_ORDER_ANY, ald_startup, NULL)
332 
333 
334 /* User visible queue functions */
335 
336 /*
337  * Create the queue data structure, allocate the buffer, and open the file.
338  */
339 int
340 alq_open(struct alq **alqp, const char *file, struct ucred *cred, int cmode,
341     int size, int count)
342 {
343 	struct thread *td;
344 	struct nameidata nd;
345 	struct ale *ale;
346 	struct ale *alp;
347 	struct alq *alq;
348 	char *bufp;
349 	int flags;
350 	int error;
351 	int i, vfslocked;
352 
353 	*alqp = NULL;
354 	td = curthread;
355 
356 	NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, file, td);
357 	flags = FWRITE | O_NOFOLLOW | O_CREAT;
358 
359 	error = vn_open_cred(&nd, &flags, cmode, cred, NULL);
360 	if (error)
361 		return (error);
362 
363 	vfslocked = NDHASGIANT(&nd);
364 	NDFREE(&nd, NDF_ONLY_PNBUF);
365 	/* We just unlock so we hold a reference */
366 	VOP_UNLOCK(nd.ni_vp, 0, td);
367 	VFS_UNLOCK_GIANT(vfslocked);
368 
369 	alq = malloc(sizeof(*alq), M_ALD, M_WAITOK|M_ZERO);
370 	alq->aq_entbuf = malloc(count * size, M_ALD, M_WAITOK|M_ZERO);
371 	alq->aq_first = malloc(sizeof(*ale) * count, M_ALD, M_WAITOK|M_ZERO);
372 	alq->aq_vp = nd.ni_vp;
373 	alq->aq_cred = crhold(cred);
374 	alq->aq_entmax = count;
375 	alq->aq_entlen = size;
376 	alq->aq_entfree = alq->aq_first;
377 
378 	mtx_init(&alq->aq_mtx, "ALD Queue", NULL, MTX_SPIN|MTX_QUIET);
379 
380 	bufp = alq->aq_entbuf;
381 	ale = alq->aq_first;
382 	alp = NULL;
383 
384 	/* Match up entries with buffers */
385 	for (i = 0; i < count; i++) {
386 		if (alp)
387 			alp->ae_next = ale;
388 		ale->ae_data = bufp;
389 		alp = ale;
390 		ale++;
391 		bufp += size;
392 	}
393 
394 	alp->ae_next = alq->aq_first;
395 
396 	if ((error = ald_add(alq)) != 0)
397 		return (error);
398 	*alqp = alq;
399 
400 	return (0);
401 }
402 
403 /*
404  * Copy a new entry into the queue.  If the operation would block either
405  * wait or return an error depending on the value of waitok.
406  */
407 int
408 alq_write(struct alq *alq, void *data, int waitok)
409 {
410 	struct ale *ale;
411 
412 	if ((ale = alq_get(alq, waitok)) == NULL)
413 		return (EWOULDBLOCK);
414 
415 	bcopy(data, ale->ae_data, alq->aq_entlen);
416 	alq_post(alq, ale);
417 
418 	return (0);
419 }
420 
421 struct ale *
422 alq_get(struct alq *alq, int waitok)
423 {
424 	struct ale *ale;
425 	struct ale *aln;
426 
427 	ale = NULL;
428 
429 	ALQ_LOCK(alq);
430 
431 	/* Loop until we get an entry or we're shutting down */
432 	while ((alq->aq_flags & AQ_SHUTDOWN) == 0 &&
433 	    (ale = alq->aq_entfree) == NULL &&
434 	    (waitok & ALQ_WAITOK)) {
435 		alq->aq_flags |= AQ_WANTED;
436 		ALQ_UNLOCK(alq);
437 		tsleep(alq, PWAIT, "alqget", 0);
438 		ALQ_LOCK(alq);
439 	}
440 
441 	if (ale != NULL) {
442 		aln = ale->ae_next;
443 		if ((aln->ae_flags & AE_VALID) == 0)
444 			alq->aq_entfree = aln;
445 		else
446 			alq->aq_entfree = NULL;
447 	} else
448 		ALQ_UNLOCK(alq);
449 
450 
451 	return (ale);
452 }
453 
454 void
455 alq_post(struct alq *alq, struct ale *ale)
456 {
457 	int activate;
458 
459 	ale->ae_flags |= AE_VALID;
460 
461 	if (alq->aq_entvalid == NULL)
462 		alq->aq_entvalid = ale;
463 
464 	if ((alq->aq_flags & AQ_ACTIVE) == 0) {
465 		alq->aq_flags |= AQ_ACTIVE;
466 		activate = 1;
467 	} else
468 		activate = 0;
469 
470 	ALQ_UNLOCK(alq);
471 	if (activate) {
472 		ALD_LOCK();
473 		ald_activate(alq);
474 		ALD_UNLOCK();
475 	}
476 }
477 
478 void
479 alq_flush(struct alq *alq)
480 {
481 	int needwakeup = 0;
482 
483 	ALD_LOCK();
484 	ALQ_LOCK(alq);
485 	if (alq->aq_flags & AQ_ACTIVE) {
486 		ald_deactivate(alq);
487 		ALD_UNLOCK();
488 		needwakeup = alq_doio(alq);
489 	} else
490 		ALD_UNLOCK();
491 	ALQ_UNLOCK(alq);
492 
493 	if (needwakeup)
494 		wakeup(alq);
495 }
496 
497 /*
498  * Flush remaining data, close the file and free all resources.
499  */
500 void
501 alq_close(struct alq *alq)
502 {
503 	/*
504 	 * If we're already shuting down someone else will flush and close
505 	 * the vnode.
506 	 */
507 	if (ald_rem(alq) != 0)
508 		return;
509 
510 	/*
511 	 * Drain all pending IO.
512 	 */
513 	alq_shutdown(alq);
514 
515 	mtx_destroy(&alq->aq_mtx);
516 	free(alq->aq_first, M_ALD);
517 	free(alq->aq_entbuf, M_ALD);
518 	free(alq, M_ALD);
519 }
520