xref: /freebsd/sys/kern/kern_devctl.c (revision 031beb4e239bfce798af17f5fe8dba8bcaf13d99)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2002-2020 M. Warner Losh <imp@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 #include <sys/cdefs.h>
29 #include "opt_bus.h"
30 #include "opt_ddb.h"
31 
32 #include <sys/param.h>
33 #include <sys/conf.h>
34 #include <sys/eventhandler.h>
35 #include <sys/filio.h>
36 #include <sys/lock.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/poll.h>
41 #include <sys/priv.h>
42 #include <sys/proc.h>
43 #include <sys/condvar.h>
44 #include <sys/queue.h>
45 #include <machine/bus.h>
46 #include <sys/sbuf.h>
47 #include <sys/selinfo.h>
48 #include <sys/smp.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/uio.h>
52 #include <sys/bus.h>
53 
54 #include <machine/cpu.h>
55 #include <machine/stdarg.h>
56 
57 #include <vm/uma.h>
58 #include <vm/vm.h>
59 
60 #include <ddb/ddb.h>
61 
62 STAILQ_HEAD(devq, dev_event_info);
63 
64 static struct dev_softc {
65 	int		inuse;
66 	int		nonblock;
67 	int		queued;
68 	int		async;
69 	struct mtx	mtx;
70 	struct cv	cv;
71 	struct selinfo	sel;
72 	struct devq	devq;
73 	struct sigio	*sigio;
74 	uma_zone_t	zone;
75 } devsoftc;
76 
77 /*
78  * This design allows only one reader for /dev/devctl.  This is not desirable
79  * in the long run, but will get a lot of hair out of this implementation.
80  * Maybe we should make this device a clonable device.
81  *
82  * Also note: we specifically do not attach a device to the device_t tree
83  * to avoid potential chicken and egg problems.  One could argue that all
84  * of this belongs to the root node.
85  */
86 
87 #define DEVCTL_DEFAULT_QUEUE_LEN 1000
88 static int sysctl_devctl_queue(SYSCTL_HANDLER_ARGS);
89 static int devctl_queue_length = DEVCTL_DEFAULT_QUEUE_LEN;
90 SYSCTL_PROC(_hw_bus, OID_AUTO, devctl_queue, CTLTYPE_INT | CTLFLAG_RWTUN |
91     CTLFLAG_MPSAFE, NULL, 0, sysctl_devctl_queue, "I", "devctl queue length");
92 
93 static void devctl_attach_handler(void *arg __unused, device_t dev);
94 static void devctl_detach_handler(void *arg __unused, device_t dev,
95     enum evhdev_detach state);
96 static void devctl_nomatch_handler(void *arg __unused, device_t dev);
97 
98 static d_open_t		devopen;
99 static d_close_t	devclose;
100 static d_read_t		devread;
101 static d_ioctl_t	devioctl;
102 static d_poll_t		devpoll;
103 static d_kqfilter_t	devkqfilter;
104 
105 #define DEVCTL_BUFFER (1024 - sizeof(void *))
106 struct dev_event_info {
107 	STAILQ_ENTRY(dev_event_info) dei_link;
108 	char dei_data[DEVCTL_BUFFER];
109 };
110 
111 
112 static struct cdevsw dev_cdevsw = {
113 	.d_version =	D_VERSION,
114 	.d_open =	devopen,
115 	.d_close =	devclose,
116 	.d_read =	devread,
117 	.d_ioctl =	devioctl,
118 	.d_poll =	devpoll,
119 	.d_kqfilter =	devkqfilter,
120 	.d_name =	"devctl",
121 };
122 
123 static void	filt_devctl_detach(struct knote *kn);
124 static int	filt_devctl_read(struct knote *kn, long hint);
125 
126 static struct filterops devctl_rfiltops = {
127 	.f_isfd = 1,
128 	.f_detach = filt_devctl_detach,
129 	.f_event = filt_devctl_read,
130 };
131 
132 static struct cdev *devctl_dev;
133 static void devaddq(const char *type, const char *what, device_t dev);
134 
135 static struct devctlbridge {
136 	send_event_f *send_f;
137 } devctl_notify_hook = { .send_f = NULL };
138 
139 static void
140 devctl_init(void)
141 {
142 	int reserve;
143 	uma_zone_t z;
144 
145 	devctl_dev = make_dev_credf(MAKEDEV_ETERNAL, &dev_cdevsw, 0, NULL,
146 	    UID_ROOT, GID_WHEEL, 0600, "devctl");
147 	mtx_init(&devsoftc.mtx, "dev mtx", "devd", MTX_DEF);
148 	cv_init(&devsoftc.cv, "dev cv");
149 	STAILQ_INIT(&devsoftc.devq);
150 	knlist_init_mtx(&devsoftc.sel.si_note, &devsoftc.mtx);
151 	if (devctl_queue_length > 0) {
152 		/*
153 		 * Allocate a zone for the messages. Preallocate 2% of these for
154 		 * a reserve. Allow only devctl_queue_length slabs to cap memory
155 		 * usage.  The reserve usually allows coverage of surges of
156 		 * events during memory shortages. Normally we won't have to
157 		 * re-use events from the queue, but will in extreme shortages.
158 		 */
159 		z = devsoftc.zone = uma_zcreate("DEVCTL",
160 		    sizeof(struct dev_event_info), NULL, NULL, NULL, NULL,
161 		    UMA_ALIGN_PTR, 0);
162 		reserve = max(devctl_queue_length / 50, 100);	/* 2% reserve */
163 		uma_zone_set_max(z, devctl_queue_length);
164 		uma_zone_set_maxcache(z, 0);
165 		uma_zone_reserve(z, reserve);
166 		uma_prealloc(z, reserve);
167 	}
168 	EVENTHANDLER_REGISTER(device_attach, devctl_attach_handler,
169 	    NULL, EVENTHANDLER_PRI_LAST);
170 	EVENTHANDLER_REGISTER(device_detach, devctl_detach_handler,
171 	    NULL, EVENTHANDLER_PRI_LAST);
172 	EVENTHANDLER_REGISTER(device_nomatch, devctl_nomatch_handler,
173 	    NULL, EVENTHANDLER_PRI_LAST);
174 }
175 SYSINIT(devctl_init, SI_SUB_DRIVERS, SI_ORDER_SECOND, devctl_init, NULL);
176 
177 /*
178  * A device was added to the tree.  We are called just after it successfully
179  * attaches (that is, probe and attach success for this device).  No call
180  * is made if a device is merely parented into the tree.  See devnomatch
181  * if probe fails.  If attach fails, no notification is sent (but maybe
182  * we should have a different message for this).
183  */
184 static void
185 devctl_attach_handler(void *arg __unused, device_t dev)
186 {
187 	devaddq("+", device_get_nameunit(dev), dev);
188 }
189 
190 /*
191  * A device was removed from the tree.  We are called just before this
192  * happens.
193  */
194 static void
195 devctl_detach_handler(void *arg __unused, device_t dev, enum evhdev_detach state)
196 {
197 	if (state == EVHDEV_DETACH_COMPLETE)
198 		devaddq("-", device_get_nameunit(dev), dev);
199 }
200 
201 /*
202  * Called when there's no match for this device.  This is only called
203  * the first time that no match happens, so we don't keep getting this
204  * message.  Should that prove to be undesirable, we can change it.
205  * This is called when all drivers that can attach to a given bus
206  * decline to accept this device.  Other errors may not be detected.
207  */
208 static void
209 devctl_nomatch_handler(void *arg __unused, device_t dev)
210 {
211 	devaddq("?", "", dev);
212 }
213 
214 static int
215 devopen(struct cdev *dev, int oflags, int devtype, struct thread *td)
216 {
217 	mtx_lock(&devsoftc.mtx);
218 	if (devsoftc.inuse) {
219 		mtx_unlock(&devsoftc.mtx);
220 		return (EBUSY);
221 	}
222 	/* move to init */
223 	devsoftc.inuse = 1;
224 	mtx_unlock(&devsoftc.mtx);
225 	return (0);
226 }
227 
228 static int
229 devclose(struct cdev *dev, int fflag, int devtype, struct thread *td)
230 {
231 	mtx_lock(&devsoftc.mtx);
232 	devsoftc.inuse = 0;
233 	devsoftc.nonblock = 0;
234 	devsoftc.async = 0;
235 	cv_broadcast(&devsoftc.cv);
236 	funsetown(&devsoftc.sigio);
237 	mtx_unlock(&devsoftc.mtx);
238 	return (0);
239 }
240 
241 /*
242  * The read channel for this device is used to report changes to
243  * userland in realtime.  We are required to free the data as well as
244  * the n1 object because we allocate them separately.  Also note that
245  * we return one record at a time.  If you try to read this device a
246  * character at a time, you will lose the rest of the data.  Listening
247  * programs are expected to cope.
248  */
249 static int
250 devread(struct cdev *dev, struct uio *uio, int ioflag)
251 {
252 	struct dev_event_info *n1;
253 	int rv;
254 
255 	mtx_lock(&devsoftc.mtx);
256 	while (STAILQ_EMPTY(&devsoftc.devq)) {
257 		if (devsoftc.nonblock) {
258 			mtx_unlock(&devsoftc.mtx);
259 			return (EAGAIN);
260 		}
261 		rv = cv_wait_sig(&devsoftc.cv, &devsoftc.mtx);
262 		if (rv) {
263 			/*
264 			 * Need to translate ERESTART to EINTR here? -- jake
265 			 */
266 			mtx_unlock(&devsoftc.mtx);
267 			return (rv);
268 		}
269 	}
270 	n1 = STAILQ_FIRST(&devsoftc.devq);
271 	STAILQ_REMOVE_HEAD(&devsoftc.devq, dei_link);
272 	devsoftc.queued--;
273 	mtx_unlock(&devsoftc.mtx);
274 	rv = uiomove(n1->dei_data, strlen(n1->dei_data), uio);
275 	uma_zfree(devsoftc.zone, n1);
276 	return (rv);
277 }
278 
279 static	int
280 devioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
281 {
282 	switch (cmd) {
283 	case FIONBIO:
284 		if (*(int*)data)
285 			devsoftc.nonblock = 1;
286 		else
287 			devsoftc.nonblock = 0;
288 		return (0);
289 	case FIOASYNC:
290 		if (*(int*)data)
291 			devsoftc.async = 1;
292 		else
293 			devsoftc.async = 0;
294 		return (0);
295 	case FIOSETOWN:
296 		return fsetown(*(int *)data, &devsoftc.sigio);
297 	case FIOGETOWN:
298 		*(int *)data = fgetown(&devsoftc.sigio);
299 		return (0);
300 
301 		/* (un)Support for other fcntl() calls. */
302 	case FIOCLEX:
303 	case FIONCLEX:
304 	case FIONREAD:
305 	default:
306 		break;
307 	}
308 	return (ENOTTY);
309 }
310 
311 static	int
312 devpoll(struct cdev *dev, int events, struct thread *td)
313 {
314 	int	revents = 0;
315 
316 	mtx_lock(&devsoftc.mtx);
317 	if (events & (POLLIN | POLLRDNORM)) {
318 		if (!STAILQ_EMPTY(&devsoftc.devq))
319 			revents = events & (POLLIN | POLLRDNORM);
320 		else
321 			selrecord(td, &devsoftc.sel);
322 	}
323 	mtx_unlock(&devsoftc.mtx);
324 
325 	return (revents);
326 }
327 
328 static int
329 devkqfilter(struct cdev *dev, struct knote *kn)
330 {
331 	int error;
332 
333 	if (kn->kn_filter == EVFILT_READ) {
334 		kn->kn_fop = &devctl_rfiltops;
335 		knlist_add(&devsoftc.sel.si_note, kn, 0);
336 		error = 0;
337 	} else
338 		error = EINVAL;
339 	return (error);
340 }
341 
342 static void
343 filt_devctl_detach(struct knote *kn)
344 {
345 	knlist_remove(&devsoftc.sel.si_note, kn, 0);
346 }
347 
348 static int
349 filt_devctl_read(struct knote *kn, long hint)
350 {
351 	kn->kn_data = devsoftc.queued;
352 	return (kn->kn_data != 0);
353 }
354 
355 /**
356  * @brief Return whether the userland process is running
357  */
358 bool
359 devctl_process_running(void)
360 {
361 	return (devsoftc.inuse == 1);
362 }
363 
364 static struct dev_event_info *
365 devctl_alloc_dei(void)
366 {
367 	struct dev_event_info *dei = NULL;
368 
369 	mtx_lock(&devsoftc.mtx);
370 	if (devctl_queue_length == 0)
371 		goto out;
372 	dei = uma_zalloc(devsoftc.zone, M_NOWAIT);
373 	if (dei == NULL)
374 		dei = uma_zalloc(devsoftc.zone, M_NOWAIT | M_USE_RESERVE);
375 	if (dei == NULL) {
376 		/*
377 		 * Guard against no items in the queue. Normally, this won't
378 		 * happen, but if lots of events happen all at once and there's
379 		 * a chance we're out of allocated space but none have yet been
380 		 * queued when we get here, leaving nothing to steal. This can
381 		 * also happen with error injection. Fail safe by returning
382 		 * NULL in that case..
383 		 */
384 		if (devsoftc.queued == 0)
385 			goto out;
386 		dei = STAILQ_FIRST(&devsoftc.devq);
387 		STAILQ_REMOVE_HEAD(&devsoftc.devq, dei_link);
388 		devsoftc.queued--;
389 	}
390 	MPASS(dei != NULL);
391 	*dei->dei_data = '\0';
392 out:
393 	mtx_unlock(&devsoftc.mtx);
394 	return (dei);
395 }
396 
397 static struct dev_event_info *
398 devctl_alloc_dei_sb(struct sbuf *sb)
399 {
400 	struct dev_event_info *dei;
401 
402 	dei = devctl_alloc_dei();
403 	if (dei != NULL)
404 		sbuf_new(sb, dei->dei_data, sizeof(dei->dei_data), SBUF_FIXEDLEN);
405 	return (dei);
406 }
407 
408 static void
409 devctl_free_dei(struct dev_event_info *dei)
410 {
411 	uma_zfree(devsoftc.zone, dei);
412 }
413 
414 static void
415 devctl_queue(struct dev_event_info *dei)
416 {
417 	mtx_lock(&devsoftc.mtx);
418 	STAILQ_INSERT_TAIL(&devsoftc.devq, dei, dei_link);
419 	devsoftc.queued++;
420 	cv_broadcast(&devsoftc.cv);
421 	KNOTE_LOCKED(&devsoftc.sel.si_note, 0);
422 	mtx_unlock(&devsoftc.mtx);
423 	selwakeup(&devsoftc.sel);
424 	if (devsoftc.async && devsoftc.sigio != NULL)
425 		pgsigio(&devsoftc.sigio, SIGIO, 0);
426 }
427 
428 /**
429  * @brief Send a 'notification' to userland, using standard ways
430  */
431 void
432 devctl_notify(const char *system, const char *subsystem, const char *type,
433     const char *data)
434 {
435 	struct dev_event_info *dei;
436 	struct sbuf sb;
437 
438 	if (system == NULL || subsystem == NULL || type == NULL)
439 		return;
440 	if (devctl_notify_hook.send_f != NULL)
441 		devctl_notify_hook.send_f(system, subsystem, type, data);
442 	dei = devctl_alloc_dei_sb(&sb);
443 	if (dei == NULL)
444 		return;
445 	sbuf_cpy(&sb, "!system=");
446 	sbuf_cat(&sb, system);
447 	sbuf_cat(&sb, " subsystem=");
448 	sbuf_cat(&sb, subsystem);
449 	sbuf_cat(&sb, " type=");
450 	sbuf_cat(&sb, type);
451 	if (data != NULL) {
452 		sbuf_putc(&sb, ' ');
453 		sbuf_cat(&sb, data);
454 	}
455 	sbuf_putc(&sb, '\n');
456 	if (sbuf_finish(&sb) != 0)
457 		devctl_free_dei(dei);	/* overflow -> drop it */
458 	else
459 		devctl_queue(dei);
460 }
461 
462 /*
463  * Common routine that tries to make sending messages as easy as possible.
464  * We allocate memory for the data, copy strings into that, but do not
465  * free it unless there's an error.  The dequeue part of the driver should
466  * free the data.  We don't send data when the device is disabled.  We do
467  * send data, even when we have no listeners, because we wish to avoid
468  * races relating to startup and restart of listening applications.
469  *
470  * devaddq is designed to string together the type of event, with the
471  * object of that event, plus the plug and play info and location info
472  * for that event.  This is likely most useful for devices, but less
473  * useful for other consumers of this interface.  Those should use
474  * the devctl_notify() interface instead.
475  *
476  * Output:
477  *	${type}${what} at $(location dev) $(pnp-info dev) on $(parent dev)
478  */
479 static void
480 devaddq(const char *type, const char *what, device_t dev)
481 {
482 	struct dev_event_info *dei;
483 	const char *parstr;
484 	struct sbuf sb;
485 	size_t beginlen;
486 
487 	dei = devctl_alloc_dei_sb(&sb);
488 	if (dei == NULL)
489 		return;
490 	sbuf_cpy(&sb, type);
491 	sbuf_cat(&sb, what);
492 	sbuf_cat(&sb, " at ");
493 	beginlen = sbuf_len(&sb);
494 
495 	/* Add in the location */
496 	bus_child_location(dev, &sb);
497 	sbuf_putc(&sb, ' ');
498 
499 	/* Add in pnpinfo */
500 	bus_child_pnpinfo(dev, &sb);
501 
502 	/* Get the parent of this device, or / if high enough in the tree. */
503 	if (device_get_parent(dev) == NULL)
504 		parstr = ".";	/* Or '/' ? */
505 	else
506 		parstr = device_get_nameunit(device_get_parent(dev));
507 	sbuf_cat(&sb, " on ");
508 	sbuf_cat(&sb, parstr);
509 	sbuf_putc(&sb, '\n');
510 	if (sbuf_finish(&sb) != 0)
511 		goto bad;
512 	if (devctl_notify_hook.send_f != NULL) {
513 		const char *t;
514 
515 		switch (*type) {
516 		case '+':
517 			t = "ATTACH";
518 			break;
519 		case '-':
520 			t = "DETACH";
521 			break;
522 		default:
523 			t = "NOMATCH";
524 			break;
525 		}
526 		devctl_notify_hook.send_f("device",
527 		    what, t, sbuf_data(&sb) + beginlen);
528 	}
529 	devctl_queue(dei);
530 	return;
531 bad:
532 	devctl_free_dei(dei);
533 }
534 
535 static int
536 sysctl_devctl_queue(SYSCTL_HANDLER_ARGS)
537 {
538 	int q, error;
539 
540 	q = devctl_queue_length;
541 	error = sysctl_handle_int(oidp, &q, 0, req);
542 	if (error || !req->newptr)
543 		return (error);
544 	if (q < 0)
545 		return (EINVAL);
546 
547 	/*
548 	 * When set as a tunable, we've not yet initialized the mutex.
549 	 * It is safe to just assign to devctl_queue_length and return
550 	 * as we're racing no one. We'll use whatever value set in
551 	 * devinit.
552 	 */
553 	if (!mtx_initialized(&devsoftc.mtx)) {
554 		devctl_queue_length = q;
555 		return (0);
556 	}
557 
558 	/*
559 	 * XXX It's hard to grow or shrink the UMA zone. Only allow
560 	 * disabling the queue size for the moment until underlying
561 	 * UMA issues can be sorted out.
562 	 */
563 	if (q != 0)
564 		return (EINVAL);
565 	if (q == devctl_queue_length)
566 		return (0);
567 	mtx_lock(&devsoftc.mtx);
568 	devctl_queue_length = 0;
569 	uma_zdestroy(devsoftc.zone);
570 	devsoftc.zone = 0;
571 	mtx_unlock(&devsoftc.mtx);
572 	return (0);
573 }
574 
575 /**
576  * @brief safely quotes strings that might have double quotes in them.
577  *
578  * The devctl protocol relies on quoted strings having matching quotes.
579  * This routine quotes any internal quotes so the resulting string
580  * is safe to pass to snprintf to construct, for example pnp info strings.
581  *
582  * @param sb	sbuf to place the characters into
583  * @param src	Original buffer.
584  */
585 void
586 devctl_safe_quote_sb(struct sbuf *sb, const char *src)
587 {
588 	while (*src != '\0') {
589 		if (*src == '"' || *src == '\\')
590 			sbuf_putc(sb, '\\');
591 		sbuf_putc(sb, *src++);
592 	}
593 }
594 
595 void
596 devctl_set_notify_hook(send_event_f *hook)
597 {
598 	devctl_notify_hook.send_f = hook;
599 }
600 
601 void
602 devctl_unset_notify_hook(void)
603 {
604 	devctl_notify_hook.send_f = NULL;
605 }
606 
607