xref: /freebsd/sys/kern/kern_devctl.c (revision 657729a89dd578d8cfc70d6616f5c65a48a8b33a)
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 __FBSDID("$FreeBSD$");
30 
31 #include "opt_bus.h"
32 #include "opt_ddb.h"
33 
34 #include <sys/param.h>
35 #include <sys/conf.h>
36 #include <sys/eventhandler.h>
37 #include <sys/filio.h>
38 #include <sys/lock.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/poll.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/condvar.h>
46 #include <sys/queue.h>
47 #include <machine/bus.h>
48 #include <sys/sbuf.h>
49 #include <sys/selinfo.h>
50 #include <sys/smp.h>
51 #include <sys/sysctl.h>
52 #include <sys/systm.h>
53 #include <sys/uio.h>
54 #include <sys/bus.h>
55 
56 #include <machine/cpu.h>
57 #include <machine/stdarg.h>
58 
59 #include <vm/uma.h>
60 #include <vm/vm.h>
61 
62 #include <ddb/ddb.h>
63 
64 STAILQ_HEAD(devq, dev_event_info);
65 
66 static struct dev_softc {
67 	int		inuse;
68 	int		nonblock;
69 	int		queued;
70 	int		async;
71 	struct mtx	mtx;
72 	struct cv	cv;
73 	struct selinfo	sel;
74 	struct devq	devq;
75 	struct sigio	*sigio;
76 	uma_zone_t	zone;
77 } devsoftc;
78 
79 /*
80  * This design allows only one reader for /dev/devctl.  This is not desirable
81  * in the long run, but will get a lot of hair out of this implementation.
82  * Maybe we should make this device a clonable device.
83  *
84  * Also note: we specifically do not attach a device to the device_t tree
85  * to avoid potential chicken and egg problems.  One could argue that all
86  * of this belongs to the root node.
87  */
88 
89 #define DEVCTL_DEFAULT_QUEUE_LEN 1000
90 static int sysctl_devctl_queue(SYSCTL_HANDLER_ARGS);
91 static int devctl_queue_length = DEVCTL_DEFAULT_QUEUE_LEN;
92 SYSCTL_PROC(_hw_bus, OID_AUTO, devctl_queue, CTLTYPE_INT | CTLFLAG_RWTUN |
93     CTLFLAG_MPSAFE, NULL, 0, sysctl_devctl_queue, "I", "devctl queue length");
94 
95 static void devctl_attach_handler(void *arg __unused, device_t dev);
96 static void devctl_detach_handler(void *arg __unused, device_t dev,
97     enum evhdev_detach state);
98 static void devctl_nomatch_handler(void *arg __unused, device_t dev);
99 
100 static d_open_t		devopen;
101 static d_close_t	devclose;
102 static d_read_t		devread;
103 static d_ioctl_t	devioctl;
104 static d_poll_t		devpoll;
105 static d_kqfilter_t	devkqfilter;
106 
107 #define DEVCTL_BUFFER (1024 - sizeof(void *))
108 struct dev_event_info {
109 	STAILQ_ENTRY(dev_event_info) dei_link;
110 	char dei_data[DEVCTL_BUFFER];
111 };
112 
113 
114 static struct cdevsw dev_cdevsw = {
115 	.d_version =	D_VERSION,
116 	.d_open =	devopen,
117 	.d_close =	devclose,
118 	.d_read =	devread,
119 	.d_ioctl =	devioctl,
120 	.d_poll =	devpoll,
121 	.d_kqfilter =	devkqfilter,
122 	.d_name =	"devctl",
123 };
124 
125 static void	filt_devctl_detach(struct knote *kn);
126 static int	filt_devctl_read(struct knote *kn, long hint);
127 
128 static struct filterops devctl_rfiltops = {
129 	.f_isfd = 1,
130 	.f_detach = filt_devctl_detach,
131 	.f_event = filt_devctl_read,
132 };
133 
134 static struct cdev *devctl_dev;
135 static void devaddq(const char *type, const char *what, device_t dev);
136 
137 static void
138 devctl_init(void)
139 {
140 	int reserve;
141 	uma_zone_t z;
142 
143 	devctl_dev = make_dev_credf(MAKEDEV_ETERNAL, &dev_cdevsw, 0, NULL,
144 	    UID_ROOT, GID_WHEEL, 0600, "devctl");
145 	mtx_init(&devsoftc.mtx, "dev mtx", "devd", MTX_DEF);
146 	cv_init(&devsoftc.cv, "dev cv");
147 	STAILQ_INIT(&devsoftc.devq);
148 	knlist_init_mtx(&devsoftc.sel.si_note, &devsoftc.mtx);
149 	if (devctl_queue_length > 0) {
150 		/*
151 		 * Allocate a zone for the messages. Preallocate 2% of these for
152 		 * a reserve. Allow only devctl_queue_length slabs to cap memory
153 		 * usage.  The reserve usually allows coverage of surges of
154 		 * events during memory shortages. Normally we won't have to
155 		 * re-use events from the queue, but will in extreme shortages.
156 		 */
157 		z = devsoftc.zone = uma_zcreate("DEVCTL",
158 		    sizeof(struct dev_event_info), NULL, NULL, NULL, NULL,
159 		    UMA_ALIGN_PTR, 0);
160 		reserve = max(devctl_queue_length / 50, 100);	/* 2% reserve */
161 		uma_zone_set_max(z, devctl_queue_length);
162 		uma_zone_set_maxcache(z, 0);
163 		uma_zone_reserve(z, reserve);
164 		uma_prealloc(z, reserve);
165 	}
166 	EVENTHANDLER_REGISTER(device_attach, devctl_attach_handler,
167 	    NULL, EVENTHANDLER_PRI_LAST);
168 	EVENTHANDLER_REGISTER(device_detach, devctl_detach_handler,
169 	    NULL, EVENTHANDLER_PRI_LAST);
170 	EVENTHANDLER_REGISTER(device_nomatch, devctl_nomatch_handler,
171 	    NULL, EVENTHANDLER_PRI_LAST);
172 }
173 SYSINIT(devctl_init, SI_SUB_DRIVERS, SI_ORDER_SECOND, devctl_init, NULL);
174 
175 /*
176  * A device was added to the tree.  We are called just after it successfully
177  * attaches (that is, probe and attach success for this device).  No call
178  * is made if a device is merely parented into the tree.  See devnomatch
179  * if probe fails.  If attach fails, no notification is sent (but maybe
180  * we should have a different message for this).
181  */
182 static void
183 devctl_attach_handler(void *arg __unused, device_t dev)
184 {
185 	devaddq("+", device_get_nameunit(dev), dev);
186 }
187 
188 /*
189  * A device was removed from the tree.  We are called just before this
190  * happens.
191  */
192 static void
193 devctl_detach_handler(void *arg __unused, device_t dev, enum evhdev_detach state)
194 {
195 	if (state == EVHDEV_DETACH_COMPLETE)
196 		devaddq("-", device_get_nameunit(dev), dev);
197 }
198 
199 /*
200  * Called when there's no match for this device.  This is only called
201  * the first time that no match happens, so we don't keep getting this
202  * message.  Should that prove to be undesirable, we can change it.
203  * This is called when all drivers that can attach to a given bus
204  * decline to accept this device.  Other errors may not be detected.
205  */
206 static void
207 devctl_nomatch_handler(void *arg __unused, device_t dev)
208 {
209 	devaddq("?", "", dev);
210 }
211 
212 static int
213 devopen(struct cdev *dev, int oflags, int devtype, struct thread *td)
214 {
215 	mtx_lock(&devsoftc.mtx);
216 	if (devsoftc.inuse) {
217 		mtx_unlock(&devsoftc.mtx);
218 		return (EBUSY);
219 	}
220 	/* move to init */
221 	devsoftc.inuse = 1;
222 	mtx_unlock(&devsoftc.mtx);
223 	return (0);
224 }
225 
226 static int
227 devclose(struct cdev *dev, int fflag, int devtype, struct thread *td)
228 {
229 	mtx_lock(&devsoftc.mtx);
230 	devsoftc.inuse = 0;
231 	devsoftc.nonblock = 0;
232 	devsoftc.async = 0;
233 	cv_broadcast(&devsoftc.cv);
234 	funsetown(&devsoftc.sigio);
235 	mtx_unlock(&devsoftc.mtx);
236 	return (0);
237 }
238 
239 /*
240  * The read channel for this device is used to report changes to
241  * userland in realtime.  We are required to free the data as well as
242  * the n1 object because we allocate them separately.  Also note that
243  * we return one record at a time.  If you try to read this device a
244  * character at a time, you will lose the rest of the data.  Listening
245  * programs are expected to cope.
246  */
247 static int
248 devread(struct cdev *dev, struct uio *uio, int ioflag)
249 {
250 	struct dev_event_info *n1;
251 	int rv;
252 
253 	mtx_lock(&devsoftc.mtx);
254 	while (STAILQ_EMPTY(&devsoftc.devq)) {
255 		if (devsoftc.nonblock) {
256 			mtx_unlock(&devsoftc.mtx);
257 			return (EAGAIN);
258 		}
259 		rv = cv_wait_sig(&devsoftc.cv, &devsoftc.mtx);
260 		if (rv) {
261 			/*
262 			 * Need to translate ERESTART to EINTR here? -- jake
263 			 */
264 			mtx_unlock(&devsoftc.mtx);
265 			return (rv);
266 		}
267 	}
268 	n1 = STAILQ_FIRST(&devsoftc.devq);
269 	STAILQ_REMOVE_HEAD(&devsoftc.devq, dei_link);
270 	devsoftc.queued--;
271 	mtx_unlock(&devsoftc.mtx);
272 	rv = uiomove(n1->dei_data, strlen(n1->dei_data), uio);
273 	uma_zfree(devsoftc.zone, n1);
274 	return (rv);
275 }
276 
277 static	int
278 devioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
279 {
280 	switch (cmd) {
281 	case FIONBIO:
282 		if (*(int*)data)
283 			devsoftc.nonblock = 1;
284 		else
285 			devsoftc.nonblock = 0;
286 		return (0);
287 	case FIOASYNC:
288 		if (*(int*)data)
289 			devsoftc.async = 1;
290 		else
291 			devsoftc.async = 0;
292 		return (0);
293 	case FIOSETOWN:
294 		return fsetown(*(int *)data, &devsoftc.sigio);
295 	case FIOGETOWN:
296 		*(int *)data = fgetown(&devsoftc.sigio);
297 		return (0);
298 
299 		/* (un)Support for other fcntl() calls. */
300 	case FIOCLEX:
301 	case FIONCLEX:
302 	case FIONREAD:
303 	default:
304 		break;
305 	}
306 	return (ENOTTY);
307 }
308 
309 static	int
310 devpoll(struct cdev *dev, int events, struct thread *td)
311 {
312 	int	revents = 0;
313 
314 	mtx_lock(&devsoftc.mtx);
315 	if (events & (POLLIN | POLLRDNORM)) {
316 		if (!STAILQ_EMPTY(&devsoftc.devq))
317 			revents = events & (POLLIN | POLLRDNORM);
318 		else
319 			selrecord(td, &devsoftc.sel);
320 	}
321 	mtx_unlock(&devsoftc.mtx);
322 
323 	return (revents);
324 }
325 
326 static int
327 devkqfilter(struct cdev *dev, struct knote *kn)
328 {
329 	int error;
330 
331 	if (kn->kn_filter == EVFILT_READ) {
332 		kn->kn_fop = &devctl_rfiltops;
333 		knlist_add(&devsoftc.sel.si_note, kn, 0);
334 		error = 0;
335 	} else
336 		error = EINVAL;
337 	return (error);
338 }
339 
340 static void
341 filt_devctl_detach(struct knote *kn)
342 {
343 	knlist_remove(&devsoftc.sel.si_note, kn, 0);
344 }
345 
346 static int
347 filt_devctl_read(struct knote *kn, long hint)
348 {
349 	kn->kn_data = devsoftc.queued;
350 	return (kn->kn_data != 0);
351 }
352 
353 /**
354  * @brief Return whether the userland process is running
355  */
356 bool
357 devctl_process_running(void)
358 {
359 	return (devsoftc.inuse == 1);
360 }
361 
362 static struct dev_event_info *
363 devctl_alloc_dei(void)
364 {
365 	struct dev_event_info *dei = NULL;
366 
367 	mtx_lock(&devsoftc.mtx);
368 	if (devctl_queue_length == 0)
369 		goto out;
370 	dei = uma_zalloc(devsoftc.zone, M_NOWAIT);
371 	if (dei == NULL)
372 		dei = uma_zalloc(devsoftc.zone, M_NOWAIT | M_USE_RESERVE);
373 	if (dei == NULL) {
374 		/*
375 		 * Guard against no items in the queue. Normally, this won't
376 		 * happen, but if lots of events happen all at once and there's
377 		 * a chance we're out of allocated space but none have yet been
378 		 * queued when we get here, leaving nothing to steal. This can
379 		 * also happen with error injection. Fail safe by returning
380 		 * NULL in that case..
381 		 */
382 		if (devsoftc.queued == 0)
383 			goto out;
384 		dei = STAILQ_FIRST(&devsoftc.devq);
385 		STAILQ_REMOVE_HEAD(&devsoftc.devq, dei_link);
386 		devsoftc.queued--;
387 	}
388 	MPASS(dei != NULL);
389 	*dei->dei_data = '\0';
390 out:
391 	mtx_unlock(&devsoftc.mtx);
392 	return (dei);
393 }
394 
395 static struct dev_event_info *
396 devctl_alloc_dei_sb(struct sbuf *sb)
397 {
398 	struct dev_event_info *dei;
399 
400 	dei = devctl_alloc_dei();
401 	if (dei != NULL)
402 		sbuf_new(sb, dei->dei_data, sizeof(dei->dei_data), SBUF_FIXEDLEN);
403 	return (dei);
404 }
405 
406 static void
407 devctl_free_dei(struct dev_event_info *dei)
408 {
409 	uma_zfree(devsoftc.zone, dei);
410 }
411 
412 static void
413 devctl_queue(struct dev_event_info *dei)
414 {
415 	mtx_lock(&devsoftc.mtx);
416 	STAILQ_INSERT_TAIL(&devsoftc.devq, dei, dei_link);
417 	devsoftc.queued++;
418 	cv_broadcast(&devsoftc.cv);
419 	KNOTE_LOCKED(&devsoftc.sel.si_note, 0);
420 	mtx_unlock(&devsoftc.mtx);
421 	selwakeup(&devsoftc.sel);
422 	if (devsoftc.async && devsoftc.sigio != NULL)
423 		pgsigio(&devsoftc.sigio, SIGIO, 0);
424 }
425 
426 /**
427  * @brief Send a 'notification' to userland, using standard ways
428  */
429 void
430 devctl_notify(const char *system, const char *subsystem, const char *type,
431     const char *data)
432 {
433 	struct dev_event_info *dei;
434 	struct sbuf sb;
435 
436 	if (system == NULL || subsystem == NULL || type == NULL)
437 		return;
438 	dei = devctl_alloc_dei_sb(&sb);
439 	if (dei == NULL)
440 		return;
441 	sbuf_cpy(&sb, "!system=");
442 	sbuf_cat(&sb, system);
443 	sbuf_cat(&sb, " subsystem=");
444 	sbuf_cat(&sb, subsystem);
445 	sbuf_cat(&sb, " type=");
446 	sbuf_cat(&sb, type);
447 	if (data != NULL) {
448 		sbuf_putc(&sb, ' ');
449 		sbuf_cat(&sb, data);
450 	}
451 	sbuf_putc(&sb, '\n');
452 	if (sbuf_finish(&sb) != 0)
453 		devctl_free_dei(dei);	/* overflow -> drop it */
454 	else
455 		devctl_queue(dei);
456 }
457 
458 /*
459  * Common routine that tries to make sending messages as easy as possible.
460  * We allocate memory for the data, copy strings into that, but do not
461  * free it unless there's an error.  The dequeue part of the driver should
462  * free the data.  We don't send data when the device is disabled.  We do
463  * send data, even when we have no listeners, because we wish to avoid
464  * races relating to startup and restart of listening applications.
465  *
466  * devaddq is designed to string together the type of event, with the
467  * object of that event, plus the plug and play info and location info
468  * for that event.  This is likely most useful for devices, but less
469  * useful for other consumers of this interface.  Those should use
470  * the devctl_notify() interface instead.
471  *
472  * Output:
473  *	${type}${what} at $(location dev) $(pnp-info dev) on $(parent dev)
474  */
475 static void
476 devaddq(const char *type, const char *what, device_t dev)
477 {
478 	struct dev_event_info *dei;
479 	const char *parstr;
480 	struct sbuf sb;
481 
482 	dei = devctl_alloc_dei_sb(&sb);
483 	if (dei == NULL)
484 		return;
485 	sbuf_cpy(&sb, type);
486 	sbuf_cat(&sb, what);
487 	sbuf_cat(&sb, " at ");
488 
489 	/* Add in the location */
490 	bus_child_location(dev, &sb);
491 	sbuf_putc(&sb, ' ');
492 
493 	/* Add in pnpinfo */
494 	bus_child_pnpinfo(dev, &sb);
495 
496 	/* Get the parent of this device, or / if high enough in the tree. */
497 	if (device_get_parent(dev) == NULL)
498 		parstr = ".";	/* Or '/' ? */
499 	else
500 		parstr = device_get_nameunit(device_get_parent(dev));
501 	sbuf_cat(&sb, " on ");
502 	sbuf_cat(&sb, parstr);
503 	sbuf_putc(&sb, '\n');
504 	if (sbuf_finish(&sb) != 0)
505 		goto bad;
506 	devctl_queue(dei);
507 	return;
508 bad:
509 	devctl_free_dei(dei);
510 }
511 
512 static int
513 sysctl_devctl_queue(SYSCTL_HANDLER_ARGS)
514 {
515 	int q, error;
516 
517 	q = devctl_queue_length;
518 	error = sysctl_handle_int(oidp, &q, 0, req);
519 	if (error || !req->newptr)
520 		return (error);
521 	if (q < 0)
522 		return (EINVAL);
523 
524 	/*
525 	 * When set as a tunable, we've not yet initialized the mutex.
526 	 * It is safe to just assign to devctl_queue_length and return
527 	 * as we're racing no one. We'll use whatever value set in
528 	 * devinit.
529 	 */
530 	if (!mtx_initialized(&devsoftc.mtx)) {
531 		devctl_queue_length = q;
532 		return (0);
533 	}
534 
535 	/*
536 	 * XXX It's hard to grow or shrink the UMA zone. Only allow
537 	 * disabling the queue size for the moment until underlying
538 	 * UMA issues can be sorted out.
539 	 */
540 	if (q != 0)
541 		return (EINVAL);
542 	if (q == devctl_queue_length)
543 		return (0);
544 	mtx_lock(&devsoftc.mtx);
545 	devctl_queue_length = 0;
546 	uma_zdestroy(devsoftc.zone);
547 	devsoftc.zone = 0;
548 	mtx_unlock(&devsoftc.mtx);
549 	return (0);
550 }
551 
552 /**
553  * @brief safely quotes strings that might have double quotes in them.
554  *
555  * The devctl protocol relies on quoted strings having matching quotes.
556  * This routine quotes any internal quotes so the resulting string
557  * is safe to pass to snprintf to construct, for example pnp info strings.
558  *
559  * @param sb	sbuf to place the characters into
560  * @param src	Original buffer.
561  */
562 void
563 devctl_safe_quote_sb(struct sbuf *sb, const char *src)
564 {
565 	while (*src != '\0') {
566 		if (*src == '"' || *src == '\\')
567 			sbuf_putc(sb, '\\');
568 		sbuf_putc(sb, *src++);
569 	}
570 }
571