2 * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com> All rights reserved.
3 * cdevsw from kern/kern_conf.c Copyright (c) 1995 Terrence R. Lambert
4 * cdevsw from kern/kern_conf.c Copyright (c) 1995 Julian R. Elishcer,
6 * Copyright (c) 1982, 1986, 1991, 1993
7 * The Regents of the University of California. All rights reserved.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * $DragonFly: src/sys/kern/kern_device.c,v 1.27 2007/07/23 18:59:50 dillon Exp $
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/sysctl.h>
36 #include <sys/systm.h>
37 #include <sys/module.h>
38 #include <sys/malloc.h>
42 #include <sys/vnode.h>
43 #include <sys/queue.h>
44 #include <sys/device.h>
46 #include <sys/syslink_rpc.h>
48 #include <machine/stdarg.h>
49 #include <sys/thread2.h>
52 * system link descriptors identify the command in the
53 * arguments structure.
55 #define DDESCNAME(name) __CONCAT(__CONCAT(dev_,name),_desc)
57 #define DEVOP_DESC_INIT(name) \
58 struct syslink_desc DDESCNAME(name) = { \
59 __offsetof(struct dev_ops, __CONCAT(d_, name)), \
62 DEVOP_DESC_INIT(default);
63 DEVOP_DESC_INIT(open);
64 DEVOP_DESC_INIT(close);
65 DEVOP_DESC_INIT(read);
66 DEVOP_DESC_INIT(write);
67 DEVOP_DESC_INIT(ioctl);
68 DEVOP_DESC_INIT(dump);
69 DEVOP_DESC_INIT(psize);
70 DEVOP_DESC_INIT(poll);
71 DEVOP_DESC_INIT(mmap);
72 DEVOP_DESC_INIT(strategy);
73 DEVOP_DESC_INIT(kqfilter);
74 DEVOP_DESC_INIT(clone);
79 struct dev_ops dead_dev_ops;
81 struct dev_ops default_dev_ops = {
83 .d_default = NULL, /* must be NULL */
91 .d_strategy = nostrategy,
94 .d_kqfilter = nokqfilter,
98 /************************************************************************
99 * GENERAL DEVICE API FUNCTIONS *
100 ************************************************************************/
103 dev_dopen(cdev_t dev, int oflags, int devtype, struct ucred *cred)
105 struct dev_open_args ap;
107 ap.a_head.a_desc = &dev_open_desc;
108 ap.a_head.a_dev = dev;
109 ap.a_oflags = oflags;
110 ap.a_devtype = devtype;
112 return(dev->si_ops->d_open(&ap));
116 dev_dclose(cdev_t dev, int fflag, int devtype)
118 struct dev_close_args ap;
120 ap.a_head.a_desc = &dev_close_desc;
121 ap.a_head.a_dev = dev;
123 ap.a_devtype = devtype;
124 return(dev->si_ops->d_close(&ap));
128 dev_dread(cdev_t dev, struct uio *uio, int ioflag)
130 struct dev_read_args ap;
133 ap.a_head.a_desc = &dev_read_desc;
134 ap.a_head.a_dev = dev;
136 ap.a_ioflag = ioflag;
137 error = dev->si_ops->d_read(&ap);
139 dev->si_lastread = time_second;
144 dev_dwrite(cdev_t dev, struct uio *uio, int ioflag)
146 struct dev_write_args ap;
149 dev->si_lastwrite = time_second;
150 ap.a_head.a_desc = &dev_write_desc;
151 ap.a_head.a_dev = dev;
153 ap.a_ioflag = ioflag;
154 error = dev->si_ops->d_write(&ap);
159 dev_dioctl(cdev_t dev, u_long cmd, caddr_t data, int fflag, struct ucred *cred)
161 struct dev_ioctl_args ap;
163 ap.a_head.a_desc = &dev_ioctl_desc;
164 ap.a_head.a_dev = dev;
169 return(dev->si_ops->d_ioctl(&ap));
173 dev_dpoll(cdev_t dev, int events)
175 struct dev_poll_args ap;
178 ap.a_head.a_desc = &dev_poll_desc;
179 ap.a_head.a_dev = dev;
180 ap.a_events = events;
181 error = dev->si_ops->d_poll(&ap);
184 return (seltrue(dev, events));
188 dev_dmmap(cdev_t dev, vm_offset_t offset, int nprot)
190 struct dev_mmap_args ap;
193 ap.a_head.a_desc = &dev_mmap_desc;
194 ap.a_head.a_dev = dev;
195 ap.a_offset = offset;
197 error = dev->si_ops->d_mmap(&ap);
204 dev_dclone(cdev_t dev)
206 struct dev_clone_args ap;
208 ap.a_head.a_desc = &dev_clone_desc;
209 ap.a_head.a_dev = dev;
210 return (dev->si_ops->d_clone(&ap));
214 * Core device strategy call, used to issue I/O on a device. There are
215 * two versions, a non-chained version and a chained version. The chained
216 * version reuses a BIO set up by vn_strategy(). The only difference is
217 * that, for now, we do not push a new tracking structure when chaining
218 * from vn_strategy. XXX this will ultimately have to change.
221 dev_dstrategy(cdev_t dev, struct bio *bio)
223 struct dev_strategy_args ap;
224 struct bio_track *track;
226 ap.a_head.a_desc = &dev_strategy_desc;
227 ap.a_head.a_dev = dev;
230 KKASSERT(bio->bio_track == NULL);
231 KKASSERT(bio->bio_buf->b_cmd != BUF_CMD_DONE);
232 if (bio->bio_buf->b_cmd == BUF_CMD_READ)
233 track = &dev->si_track_read;
235 track = &dev->si_track_write;
236 atomic_add_int(&track->bk_active, 1);
237 bio->bio_track = track;
238 (void)dev->si_ops->d_strategy(&ap);
242 dev_dstrategy_chain(cdev_t dev, struct bio *bio)
244 struct dev_strategy_args ap;
246 KKASSERT(bio->bio_track != NULL);
247 ap.a_head.a_desc = &dev_strategy_desc;
248 ap.a_head.a_dev = dev;
250 (void)dev->si_ops->d_strategy(&ap);
254 * note: the disk layer is expected to set count, blkno, and secsize before
255 * forwarding the message.
258 dev_ddump(cdev_t dev)
260 struct dev_dump_args ap;
262 ap.a_head.a_desc = &dev_dump_desc;
263 ap.a_head.a_dev = dev;
267 return(dev->si_ops->d_dump(&ap));
271 dev_dpsize(cdev_t dev)
273 struct dev_psize_args ap;
276 ap.a_head.a_desc = &dev_psize_desc;
277 ap.a_head.a_dev = dev;
278 error = dev->si_ops->d_psize(&ap);
280 return (ap.a_result);
285 dev_dkqfilter(cdev_t dev, struct knote *kn)
287 struct dev_kqfilter_args ap;
290 ap.a_head.a_desc = &dev_kqfilter_desc;
291 ap.a_head.a_dev = dev;
293 error = dev->si_ops->d_kqfilter(&ap);
299 /************************************************************************
300 * DEVICE HELPER FUNCTIONS *
301 ************************************************************************/
304 dev_drefs(cdev_t dev)
306 return(dev->si_sysref.refcnt);
310 dev_dname(cdev_t dev)
312 return(dev->si_ops->head.name);
316 dev_dflags(cdev_t dev)
318 return(dev->si_ops->head.flags);
324 return(dev->si_ops->head.maj);
328 * Used when forwarding a request through layers. The caller adjusts
329 * ap->a_head.a_dev and then calls this function.
332 dev_doperate(struct dev_generic_args *ap)
334 int (*func)(struct dev_generic_args *);
336 func = *(void **)((char *)ap->a_dev->si_ops + ap->a_desc->sd_offset);
341 * Used by the console intercept code only. Issue an operation through
342 * a foreign ops structure allowing the ops structure associated
343 * with the device to remain intact.
346 dev_doperate_ops(struct dev_ops *ops, struct dev_generic_args *ap)
348 int (*func)(struct dev_generic_args *);
350 func = *(void **)((char *)ops + ap->a_desc->sd_offset);
355 * Convert a template dev_ops into the real thing by filling in
356 * uninitialized fields.
359 compile_dev_ops(struct dev_ops *ops)
363 for (offset = offsetof(struct dev_ops, dev_ops_first_field);
364 offset <= offsetof(struct dev_ops, dev_ops_last_field);
365 offset += sizeof(void *)
367 void **func_p = (void **)((char *)ops + offset);
368 void **def_p = (void **)((char *)&default_dev_ops + offset);
369 if (*func_p == NULL) {
371 *func_p = ops->d_default;
378 /************************************************************************
379 * MAJOR/MINOR SPACE FUNCTION *
380 ************************************************************************/
383 * This makes a dev_ops entry visible to userland (e.g /dev/<blah>).
385 * The kernel can overload a data space by making multiple dev_ops_add()
386 * calls, but only the most recent one in the list matching the mask/match
387 * will be visible to userland.
389 * make_dev() does not automatically call dev_ops_add() (nor do we want it
390 * to, since partition-managed disk devices are overloaded on top of the
393 * Disk devices typically register their major, e.g. 'ad0', and then call
394 * into the disk label management code which overloads its own onto e.g. 'ad0'
395 * to support all the various slice and partition combinations.
397 * The mask/match supplied in this call are a full 32 bits and the same
398 * mask and match must be specified in a later dev_ops_remove() call to
399 * match this add. However, the match value for the minor number should never
400 * have any bits set in the major number's bit range (8-15). The mask value
401 * may be conveniently specified as -1 without creating any major number
407 rb_dev_ops_compare(struct dev_ops_maj *a, struct dev_ops_maj *b)
411 else if (a->maj > b->maj)
416 RB_GENERATE2(dev_ops_rb_tree, dev_ops_maj, rbnode, rb_dev_ops_compare, int, maj);
418 struct dev_ops_rb_tree dev_ops_rbhead = RB_INITIALIZER(dev_ops_rbhead);
421 dev_ops_add(struct dev_ops *ops, u_int mask, u_int match)
423 static int next_maj = 256; /* first dynamic major number */
424 struct dev_ops_maj *rbmaj;
425 struct dev_ops_link *link;
427 compile_dev_ops(ops);
428 if (ops->head.maj < 0) {
429 while (dev_ops_rb_tree_RB_LOOKUP(&dev_ops_rbhead, next_maj) != NULL) {
433 ops->head.maj = next_maj;
435 rbmaj = dev_ops_rb_tree_RB_LOOKUP(&dev_ops_rbhead, ops->head.maj);
437 rbmaj = kmalloc(sizeof(*rbmaj), M_DEVBUF, M_INTWAIT | M_ZERO);
438 rbmaj->maj = ops->head.maj;
439 dev_ops_rb_tree_RB_INSERT(&dev_ops_rbhead, rbmaj);
441 for (link = rbmaj->link; link; link = link->next) {
443 * If we get an exact match we usurp the target, but we only print
444 * a warning message if a different device switch is installed.
446 if (link->mask == mask && link->match == match) {
447 if (link->ops != ops) {
448 kprintf("WARNING: \"%s\" (%p) is usurping \"%s\"'s"
451 link->ops->head.name, link->ops);
458 * XXX add additional warnings for overlaps
462 link = kmalloc(sizeof(struct dev_ops_link), M_DEVBUF, M_INTWAIT|M_ZERO);
466 link->next = rbmaj->link;
473 * Should only be used by udev2dev().
475 * If the minor number is -1, we match the first ops we find for this
476 * major. If the mask is not -1 then multiple minor numbers can match
479 * Note that this function will return NULL if the minor number is not within
480 * the bounds of the installed mask(s).
482 * The specified minor number should NOT include any major bits.
485 dev_ops_get(int x, int y)
487 struct dev_ops_maj *rbmaj;
488 struct dev_ops_link *link;
490 rbmaj = dev_ops_rb_tree_RB_LOOKUP(&dev_ops_rbhead, x);
493 for (link = rbmaj->link; link; link = link->next) {
494 if (y == -1 || (link->mask & y) == link->match)
501 * Take a cookie cutter to the major/minor device space for the passed
502 * device and generate a new dev_ops visible to userland which the caller
503 * can then modify. The original device is not modified but portions of
504 * its major/minor space will no longer be visible to userland.
507 dev_ops_add_override(cdev_t backing_dev, struct dev_ops *template,
508 u_int mask, u_int match)
511 struct dev_ops *backing_ops = backing_dev->si_ops;
513 ops = kmalloc(sizeof(struct dev_ops), M_DEVBUF, M_INTWAIT);
515 ops->head.name = backing_ops->head.name;
516 ops->head.maj = backing_ops->head.maj;
517 ops->head.flags |= backing_ops->head.flags & ~D_TRACKCLOSE;
518 compile_dev_ops(ops);
519 dev_ops_add(ops, mask, match);
525 dev_ops_remove_override(struct dev_ops *ops, u_int mask, u_int match)
527 dev_ops_remove(ops, mask, match);
528 if (ops->head.refs) {
529 kprintf("dev_ops_remove_override: %s still has %d refs!\n",
530 ops->head.name, ops->head.refs);
532 bzero(ops, sizeof(*ops));
533 kfree(ops, M_DEVBUF);
538 * Remove all matching dev_ops entries from the dev_ops_array[] major
539 * array so no new user opens can be performed, and destroy all devices
540 * installed in the hash table that are associated with this dev_ops. (see
541 * destroy_all_devs()).
543 * The mask and match should match a previous call to dev_ops_add*().
546 dev_ops_remove(struct dev_ops *ops, u_int mask, u_int match)
548 struct dev_ops_maj *rbmaj;
549 struct dev_ops_link *link;
550 struct dev_ops_link **plink;
552 if (ops != &dead_dev_ops)
553 destroy_all_devs(ops, mask, match);
555 rbmaj = dev_ops_rb_tree_RB_LOOKUP(&dev_ops_rbhead, ops->head.maj);
557 kprintf("double-remove of dev_ops %p for %s(%d)\n",
558 ops, ops->head.name, ops->head.maj);
561 for (plink = &rbmaj->link; (link = *plink) != NULL;
562 plink = &link->next) {
563 if (link->mask == mask && link->match == match) {
564 if (link->ops == ops)
566 kprintf("%s: ERROR: cannot remove dev_ops, "
567 "its major number %d was stolen by %s\n",
568 ops->head.name, ops->head.maj,
574 kprintf("%s(%d)[%08x/%08x]: WARNING: ops removed "
576 ops->head.name, ops->head.maj, mask, match);
579 --ops->head.refs; /* XXX ops_release() / record refs */
580 kfree(link, M_DEVBUF);
584 * Scrap the RB tree node for the major number if no ops are
585 * installed any longer.
587 if (rbmaj->link == NULL) {
588 dev_ops_rb_tree_RB_REMOVE(&dev_ops_rbhead, rbmaj);
589 kfree(rbmaj, M_DEVBUF);
594 * The same ops might be used with multiple devices, so don't
595 * complain if the ref count is non-zero.
597 if (ops->head.refs != 0) {
598 kprintf("%s(%d)[%08x/%08x]: Warning: dev_ops_remove() called "
599 "while %d device refs still exist!\n",
600 ops->head.name, ops->head.maj, mask, match,
604 kprintf("%s: ops removed\n", ops->head.name);
611 * dev_ops_scan() - Issue a callback for all installed dev_ops structures.
613 * The scan will terminate if a callback returns a negative number.
615 struct dev_ops_scan_info {
616 int (*callback)(struct dev_ops *, void *);
622 dev_ops_scan_callback(struct dev_ops_maj *rbmaj, void *arg)
624 struct dev_ops_scan_info *info = arg;
625 struct dev_ops_link *link;
629 for (link = rbmaj->link; link; link = link->next) {
630 r = info->callback(link->ops, info->arg);
639 dev_ops_scan(int (*callback)(struct dev_ops *, void *), void *arg)
641 struct dev_ops_scan_info info = { callback, arg };
643 return (dev_ops_rb_tree_RB_SCAN(&dev_ops_rbhead, NULL,
644 dev_ops_scan_callback, &info));
649 * Release a ops entry. When the ref count reaches zero, recurse
653 dev_ops_release(struct dev_ops *ops)
656 if (ops->head.refs == 0) {
662 dev_ops_intercept(cdev_t dev, struct dev_ops *iops)
664 struct dev_ops *oops = dev->si_ops;
666 compile_dev_ops(iops);
667 iops->head.maj = oops->head.maj;
668 iops->head.data = oops->head.data;
669 iops->head.flags = oops->head.flags;
671 dev->si_flags |= SI_INTERCEPTED;
677 dev_ops_restore(cdev_t dev, struct dev_ops *oops)
679 struct dev_ops *iops = dev->si_ops;
682 dev->si_flags &= ~SI_INTERCEPTED;
684 iops->head.data = NULL;
685 iops->head.flags = 0;
688 /************************************************************************
689 * DEFAULT DEV OPS FUNCTIONS *
690 ************************************************************************/
694 * Unsupported devswitch functions (e.g. for writing to read-only device).
695 * XXX may belong elsewhere.
699 noclone(struct dev_clone_args *ap)
702 return (0); /* allow the clone */
706 noopen(struct dev_open_args *ap)
712 noclose(struct dev_close_args *ap)
718 noread(struct dev_read_args *ap)
724 nowrite(struct dev_write_args *ap)
730 noioctl(struct dev_ioctl_args *ap)
736 nokqfilter(struct dev_kqfilter_args *ap)
742 nommap(struct dev_mmap_args *ap)
748 nopoll(struct dev_poll_args *ap)
755 nostrategy(struct dev_strategy_args *ap)
757 struct bio *bio = ap->a_bio;
759 bio->bio_buf->b_flags |= B_ERROR;
760 bio->bio_buf->b_error = EOPNOTSUPP;
766 nopsize(struct dev_psize_args *ap)
773 nodump(struct dev_dump_args *ap)
779 * XXX this is probably bogus. Any device that uses it isn't checking the
783 nullopen(struct dev_open_args *ap)
789 nullclose(struct dev_close_args *ap)