2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2002-2020 M. Warner Losh <imp@FreeBSD.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
28 #include <sys/cdefs.h>
32 #include <sys/param.h>
34 #include <sys/eventhandler.h>
35 #include <sys/filio.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
43 #include <sys/condvar.h>
44 #include <sys/queue.h>
45 #include <machine/bus.h>
47 #include <sys/selinfo.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
54 #include <machine/cpu.h>
55 #include <machine/stdarg.h>
62 STAILQ_HEAD(devq, dev_event_info);
64 static struct dev_softc {
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.
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.
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");
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);
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;
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];
112 static struct cdevsw dev_cdevsw = {
113 .d_version = D_VERSION,
119 .d_kqfilter = devkqfilter,
123 static void filt_devctl_detach(struct knote *kn);
124 static int filt_devctl_read(struct knote *kn, long hint);
126 static struct filterops devctl_rfiltops = {
128 .f_detach = filt_devctl_detach,
129 .f_event = filt_devctl_read,
132 static struct cdev *devctl_dev;
133 static void devaddq(const char *type, const char *what, device_t dev);
135 static struct devctlbridge {
136 send_event_f *send_f;
137 } devctl_notify_hook = { .send_f = NULL };
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) {
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.
159 z = devsoftc.zone = uma_zcreate("DEVCTL",
160 sizeof(struct dev_event_info), NULL, NULL, NULL, NULL,
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);
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);
175 SYSINIT(devctl_init, SI_SUB_DRIVERS, SI_ORDER_SECOND, devctl_init, NULL);
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).
185 devctl_attach_handler(void *arg __unused, device_t dev)
187 devaddq("+", device_get_nameunit(dev), dev);
191 * A device was removed from the tree. We are called just before this
195 devctl_detach_handler(void *arg __unused, device_t dev, enum evhdev_detach state)
197 if (state == EVHDEV_DETACH_COMPLETE)
198 devaddq("-", device_get_nameunit(dev), dev);
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.
209 devctl_nomatch_handler(void *arg __unused, device_t dev)
211 devaddq("?", "", dev);
215 devopen(struct cdev *dev, int oflags, int devtype, struct thread *td)
217 mtx_lock(&devsoftc.mtx);
218 if (devsoftc.inuse) {
219 mtx_unlock(&devsoftc.mtx);
224 mtx_unlock(&devsoftc.mtx);
229 devclose(struct cdev *dev, int fflag, int devtype, struct thread *td)
231 mtx_lock(&devsoftc.mtx);
233 devsoftc.nonblock = 0;
235 cv_broadcast(&devsoftc.cv);
236 funsetown(&devsoftc.sigio);
237 mtx_unlock(&devsoftc.mtx);
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.
250 devread(struct cdev *dev, struct uio *uio, int ioflag)
252 struct dev_event_info *n1;
255 mtx_lock(&devsoftc.mtx);
256 while (STAILQ_EMPTY(&devsoftc.devq)) {
257 if (devsoftc.nonblock) {
258 mtx_unlock(&devsoftc.mtx);
261 rv = cv_wait_sig(&devsoftc.cv, &devsoftc.mtx);
264 * Need to translate ERESTART to EINTR here? -- jake
266 mtx_unlock(&devsoftc.mtx);
270 n1 = STAILQ_FIRST(&devsoftc.devq);
271 STAILQ_REMOVE_HEAD(&devsoftc.devq, dei_link);
273 mtx_unlock(&devsoftc.mtx);
274 rv = uiomove(n1->dei_data, strlen(n1->dei_data), uio);
275 uma_zfree(devsoftc.zone, n1);
280 devioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
285 devsoftc.nonblock = 1;
287 devsoftc.nonblock = 0;
296 return fsetown(*(int *)data, &devsoftc.sigio);
298 *(int *)data = fgetown(&devsoftc.sigio);
301 /* (un)Support for other fcntl() calls. */
312 devpoll(struct cdev *dev, int events, struct thread *td)
316 mtx_lock(&devsoftc.mtx);
317 if (events & (POLLIN | POLLRDNORM)) {
318 if (!STAILQ_EMPTY(&devsoftc.devq))
319 revents = events & (POLLIN | POLLRDNORM);
321 selrecord(td, &devsoftc.sel);
323 mtx_unlock(&devsoftc.mtx);
329 devkqfilter(struct cdev *dev, struct knote *kn)
333 if (kn->kn_filter == EVFILT_READ) {
334 kn->kn_fop = &devctl_rfiltops;
335 knlist_add(&devsoftc.sel.si_note, kn, 0);
343 filt_devctl_detach(struct knote *kn)
345 knlist_remove(&devsoftc.sel.si_note, kn, 0);
349 filt_devctl_read(struct knote *kn, long hint)
351 kn->kn_data = devsoftc.queued;
352 return (kn->kn_data != 0);
356 * @brief Return whether the userland process is running
359 devctl_process_running(void)
361 return (devsoftc.inuse == 1);
364 static struct dev_event_info *
365 devctl_alloc_dei(void)
367 struct dev_event_info *dei = NULL;
369 mtx_lock(&devsoftc.mtx);
370 if (devctl_queue_length == 0)
372 dei = uma_zalloc(devsoftc.zone, M_NOWAIT);
374 dei = uma_zalloc(devsoftc.zone, M_NOWAIT | M_USE_RESERVE);
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..
384 if (devsoftc.queued == 0)
386 dei = STAILQ_FIRST(&devsoftc.devq);
387 STAILQ_REMOVE_HEAD(&devsoftc.devq, dei_link);
391 *dei->dei_data = '\0';
393 mtx_unlock(&devsoftc.mtx);
397 static struct dev_event_info *
398 devctl_alloc_dei_sb(struct sbuf *sb)
400 struct dev_event_info *dei;
402 dei = devctl_alloc_dei();
404 sbuf_new(sb, dei->dei_data, sizeof(dei->dei_data), SBUF_FIXEDLEN);
409 devctl_free_dei(struct dev_event_info *dei)
411 uma_zfree(devsoftc.zone, dei);
415 devctl_queue(struct dev_event_info *dei)
417 mtx_lock(&devsoftc.mtx);
418 STAILQ_INSERT_TAIL(&devsoftc.devq, dei, dei_link);
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);
429 * @brief Send a 'notification' to userland, using standard ways
432 devctl_notify(const char *system, const char *subsystem, const char *type,
435 struct dev_event_info *dei;
438 if (system == NULL || subsystem == NULL || type == NULL)
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);
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=");
455 sbuf_putc(&sb, '\n');
456 if (sbuf_finish(&sb) != 0)
457 devctl_free_dei(dei); /* overflow -> drop it */
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.
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.
477 * ${type}${what} at $(location dev) $(pnp-info dev) on $(parent dev)
480 devaddq(const char *type, const char *what, device_t dev)
482 struct dev_event_info *dei;
487 dei = devctl_alloc_dei_sb(&sb);
492 sbuf_cat(&sb, " at ");
493 beginlen = sbuf_len(&sb);
495 /* Add in the location */
496 bus_child_location(dev, &sb);
500 bus_child_pnpinfo(dev, &sb);
502 /* Get the parent of this device, or / if high enough in the tree. */
503 if (device_get_parent(dev) == NULL)
504 parstr = "."; /* Or '/' ? */
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)
512 if (devctl_notify_hook.send_f != NULL) {
526 devctl_notify_hook.send_f("device",
527 what, t, sbuf_data(&sb) + beginlen);
532 devctl_free_dei(dei);
536 sysctl_devctl_queue(SYSCTL_HANDLER_ARGS)
540 q = devctl_queue_length;
541 error = sysctl_handle_int(oidp, &q, 0, req);
542 if (error || !req->newptr)
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
553 if (!mtx_initialized(&devsoftc.mtx)) {
554 devctl_queue_length = q;
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.
565 if (q == devctl_queue_length)
567 mtx_lock(&devsoftc.mtx);
568 devctl_queue_length = 0;
569 uma_zdestroy(devsoftc.zone);
571 mtx_unlock(&devsoftc.mtx);
576 * @brief safely quotes strings that might have double quotes in them.
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.
582 * @param sb sbuf to place the characters into
583 * @param src Original buffer.
586 devctl_safe_quote_sb(struct sbuf *sb, const char *src)
588 while (*src != '\0') {
589 if (*src == '"' || *src == '\\')
591 sbuf_putc(sb, *src++);
596 devctl_set_notify_hook(send_event_f *hook)
598 devctl_notify_hook.send_f = hook;
602 devctl_unset_notify_hook(void)
604 devctl_notify_hook.send_f = NULL;