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
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/sysctl.h>
35 #include <sys/module.h>
36 #include <sys/malloc.h>
40 #include <sys/vnode.h>
41 #include <sys/queue.h>
42 #include <sys/device.h>
44 #include <sys/syslink_rpc.h>
46 #include <sys/dsched.h>
47 #include <sys/devfs.h>
49 #include <machine/stdarg.h>
51 #include <sys/mplock2.h>
54 * system link descriptors identify the command in the
55 * arguments structure.
57 #define DDESCNAME(name) __CONCAT(__CONCAT(dev_,name),_desc)
59 #define DEVOP_DESC_INIT(name) \
60 struct syslink_desc DDESCNAME(name) = { \
61 __offsetof(struct dev_ops, __CONCAT(d_, name)), \
64 DEVOP_DESC_INIT(default);
65 DEVOP_DESC_INIT(open);
66 DEVOP_DESC_INIT(close);
67 DEVOP_DESC_INIT(read);
68 DEVOP_DESC_INIT(write);
69 DEVOP_DESC_INIT(ioctl);
70 DEVOP_DESC_INIT(dump);
71 DEVOP_DESC_INIT(psize);
72 DEVOP_DESC_INIT(mmap);
73 DEVOP_DESC_INIT(mmap_single);
74 DEVOP_DESC_INIT(strategy);
75 DEVOP_DESC_INIT(kqfilter);
76 DEVOP_DESC_INIT(revoke);
77 DEVOP_DESC_INIT(clone);
82 struct dev_ops dead_dev_ops;
84 static d_open_t noopen;
85 static d_close_t noclose;
86 static d_read_t noread;
87 static d_write_t nowrite;
88 static d_ioctl_t noioctl;
89 static d_mmap_t nommap;
90 static d_mmap_single_t nommap_single;
91 static d_strategy_t nostrategy;
92 static d_dump_t nodump;
93 static d_psize_t nopsize;
94 static d_kqfilter_t nokqfilter;
95 static d_clone_t noclone;
96 static d_revoke_t norevoke;
98 struct dev_ops default_dev_ops = {
100 .d_default = NULL, /* must be NULL */
107 .d_mmap_single = nommap_single,
108 .d_strategy = nostrategy,
111 .d_kqfilter = nokqfilter,
112 .d_revoke = norevoke,
118 dev_needmplock(cdev_t dev)
120 return((dev->si_ops->head.flags & D_MPSAFE) == 0);
125 dev_nokvabio(cdev_t dev)
127 return((dev->si_ops->head.flags & D_KVABIO) == 0);
130 /************************************************************************
131 * GENERAL DEVICE API FUNCTIONS *
132 ************************************************************************
134 * The MPSAFEness of these depends on dev->si_ops->head.flags
137 dev_dopen(cdev_t dev, int oflags, int devtype, struct ucred *cred,
140 struct dev_open_args ap;
141 int needmplock = dev_needmplock(dev);
144 ap.a_head.a_desc = &dev_open_desc;
145 ap.a_head.a_dev = dev;
146 ap.a_oflags = oflags;
147 ap.a_devtype = devtype;
153 error = dev->si_ops->d_open(&ap);
160 dev_dclose(cdev_t dev, int fflag, int devtype, struct file *fp)
162 struct dev_close_args ap;
163 int needmplock = dev_needmplock(dev);
166 ap.a_head.a_desc = &dev_close_desc;
167 ap.a_head.a_dev = dev;
169 ap.a_devtype = devtype;
174 error = dev->si_ops->d_close(&ap);
181 dev_dread(cdev_t dev, struct uio *uio, int ioflag, struct file *fp)
183 struct dev_read_args ap;
184 int needmplock = dev_needmplock(dev);
187 ap.a_head.a_desc = &dev_read_desc;
188 ap.a_head.a_dev = dev;
190 ap.a_ioflag = ioflag;
195 error = dev->si_ops->d_read(&ap);
199 dev->si_lastread = time_uptime;
204 dev_dwrite(cdev_t dev, struct uio *uio, int ioflag, struct file *fp)
206 struct dev_write_args ap;
207 int needmplock = dev_needmplock(dev);
210 dev->si_lastwrite = time_uptime;
211 ap.a_head.a_desc = &dev_write_desc;
212 ap.a_head.a_dev = dev;
214 ap.a_ioflag = ioflag;
219 error = dev->si_ops->d_write(&ap);
226 dev_dioctl(cdev_t dev, u_long cmd, caddr_t data, int fflag, struct ucred *cred,
227 struct sysmsg *msg, struct file *fp)
229 struct dev_ioctl_args ap;
230 int needmplock = dev_needmplock(dev);
233 ap.a_head.a_desc = &dev_ioctl_desc;
234 ap.a_head.a_dev = dev;
244 error = dev->si_ops->d_ioctl(&ap);
251 dev_dmmap(cdev_t dev, vm_offset_t offset, int nprot, struct file *fp)
253 struct dev_mmap_args ap;
254 int needmplock = dev_needmplock(dev);
257 ap.a_head.a_desc = &dev_mmap_desc;
258 ap.a_head.a_dev = dev;
259 ap.a_offset = offset;
265 error = dev->si_ops->d_mmap(&ap);
275 dev_dmmap_single(cdev_t dev, vm_ooffset_t *offset, vm_size_t size,
276 struct vm_object **object, int nprot, struct file *fp)
278 struct dev_mmap_single_args ap;
279 int needmplock = dev_needmplock(dev);
282 ap.a_head.a_desc = &dev_mmap_single_desc;
283 ap.a_head.a_dev = dev;
284 ap.a_offset = offset;
286 ap.a_object = object;
292 error = dev->si_ops->d_mmap_single(&ap);
300 dev_dclone(cdev_t dev)
302 struct dev_clone_args ap;
303 int needmplock = dev_needmplock(dev);
306 ap.a_head.a_desc = &dev_clone_desc;
307 ap.a_head.a_dev = dev;
311 error = dev->si_ops->d_clone(&ap);
318 dev_drevoke(cdev_t dev)
320 struct dev_revoke_args ap;
321 int needmplock = dev_needmplock(dev);
324 ap.a_head.a_desc = &dev_revoke_desc;
325 ap.a_head.a_dev = dev;
329 error = dev->si_ops->d_revoke(&ap);
337 * Core device strategy call, used to issue I/O on a device. There are
338 * two versions, a non-chained version and a chained version. The chained
339 * version reuses a BIO set up by vn_strategy(). The only difference is
340 * that, for now, we do not push a new tracking structure when chaining
341 * from vn_strategy. XXX this will ultimately have to change.
344 dev_dstrategy(cdev_t dev, struct bio *bio)
346 struct dev_strategy_args ap;
347 struct bio_track *track;
348 struct buf *bp = bio->bio_buf;
349 int needmplock = dev_needmplock(dev);
352 * If the device doe snot support KVABIO and the buffer is using
353 * KVABIO, we must synchronize b_data to all cpus before dispatching.
355 if (dev_nokvabio(dev) && (bp->b_flags & B_KVABIO))
358 ap.a_head.a_desc = &dev_strategy_desc;
359 ap.a_head.a_dev = dev;
362 KKASSERT(bio->bio_track == NULL);
363 KKASSERT(bp->b_cmd != BUF_CMD_DONE);
364 if (bp->b_cmd == BUF_CMD_READ)
365 track = &dev->si_track_read;
367 track = &dev->si_track_write;
368 bio_track_ref(track);
369 bio->bio_track = track;
370 dsched_buf_enter(bp); /* might stack */
372 KKASSERT((bio->bio_flags & BIO_DONE) == 0);
375 (void)dev->si_ops->d_strategy(&ap);
381 dev_dstrategy_chain(cdev_t dev, struct bio *bio)
383 struct dev_strategy_args ap;
384 struct buf *bp = bio->bio_buf;
385 int needmplock = dev_needmplock(dev);
388 * If the device doe snot support KVABIO and the buffer is using
389 * KVABIO, we must synchronize b_data to all cpus before dispatching.
391 if (dev_nokvabio(dev) && (bp->b_flags & B_KVABIO))
394 ap.a_head.a_desc = &dev_strategy_desc;
395 ap.a_head.a_dev = dev;
398 KKASSERT(bio->bio_track != NULL);
399 KKASSERT((bio->bio_flags & BIO_DONE) == 0);
402 (void)dev->si_ops->d_strategy(&ap);
408 * note: the disk layer is expected to set count, blkno, and secsize before
409 * forwarding the message.
412 dev_ddump(cdev_t dev, void *virtual, vm_offset_t physical, off_t offset,
415 struct dev_dump_args ap;
416 int needmplock = dev_needmplock(dev);
419 ap.a_head.a_desc = &dev_dump_desc;
420 ap.a_head.a_dev = dev;
424 ap.a_virtual = virtual;
425 ap.a_physical = physical;
426 ap.a_offset = offset;
427 ap.a_length = length;
431 error = dev->si_ops->d_dump(&ap);
438 dev_dpsize(cdev_t dev)
440 struct dev_psize_args ap;
441 int needmplock = dev_needmplock(dev);
444 ap.a_head.a_desc = &dev_psize_desc;
445 ap.a_head.a_dev = dev;
449 error = dev->si_ops->d_psize(&ap);
454 return (ap.a_result);
459 * Pass-thru to the device kqfilter.
461 * NOTE: We explicitly preset a_result to 0 so d_kqfilter() functions
462 * which return 0 do not have to bother setting a_result.
465 dev_dkqfilter(cdev_t dev, struct knote *kn, struct file *fp)
467 struct dev_kqfilter_args ap;
468 int needmplock = dev_needmplock(dev);
471 ap.a_head.a_desc = &dev_kqfilter_desc;
472 ap.a_head.a_dev = dev;
479 error = dev->si_ops->d_kqfilter(&ap);
488 /************************************************************************
489 * DEVICE HELPER FUNCTIONS *
490 ************************************************************************/
496 dev_drefs(cdev_t dev)
498 return(dev->si_sysref.refcnt);
505 dev_dname(cdev_t dev)
507 return(dev->si_ops->head.name);
514 dev_dflags(cdev_t dev)
516 return(dev->si_ops->head.flags);
525 return(dev->si_ops->head.maj);
529 * Used when forwarding a request through layers. The caller adjusts
530 * ap->a_head.a_dev and then calls this function.
533 dev_doperate(struct dev_generic_args *ap)
535 int (*func)(struct dev_generic_args *);
536 int needmplock = dev_needmplock(ap->a_dev);
539 func = *(void **)((char *)ap->a_dev->si_ops + ap->a_desc->sd_offset);
551 * Used by the console intercept code only. Issue an operation through
552 * a foreign ops structure allowing the ops structure associated
553 * with the device to remain intact.
556 dev_doperate_ops(struct dev_ops *ops, struct dev_generic_args *ap)
558 int (*func)(struct dev_generic_args *);
559 int needmplock = ((ops->head.flags & D_MPSAFE) == 0);
562 func = *(void **)((char *)ops + ap->a_desc->sd_offset);
574 * Convert a template dev_ops into the real thing by filling in
575 * uninitialized fields.
578 compile_dev_ops(struct dev_ops *ops)
582 for (offset = offsetof(struct dev_ops, dev_ops_first_field);
583 offset <= offsetof(struct dev_ops, dev_ops_last_field);
584 offset += sizeof(void *)
586 void **func_p = (void **)((char *)ops + offset);
587 void **def_p = (void **)((char *)&default_dev_ops + offset);
588 if (*func_p == NULL) {
590 *func_p = ops->d_default;
597 /************************************************************************
598 * MAJOR/MINOR SPACE FUNCTION *
599 ************************************************************************/
602 * This makes a dev_ops entry visible to userland (e.g /dev/<blah>).
604 * Disk devices typically register their major, e.g. 'ad0', and then call
605 * into the disk label management code which overloads its own onto e.g. 'ad0'
606 * to support all the various slice and partition combinations.
608 * The mask/match supplied in this call are a full 32 bits and the same
609 * mask and match must be specified in a later dev_ops_remove() call to
610 * match this add. However, the match value for the minor number should never
611 * have any bits set in the major number's bit range (8-15). The mask value
612 * may be conveniently specified as -1 without creating any major number
618 rb_dev_ops_compare(struct dev_ops_maj *a, struct dev_ops_maj *b)
622 else if (a->maj > b->maj)
627 RB_GENERATE2(dev_ops_rb_tree, dev_ops_maj, rbnode, rb_dev_ops_compare, int, maj);
629 struct dev_ops_rb_tree dev_ops_rbhead = RB_INITIALIZER(dev_ops_rbhead);
632 dev_ops_remove_all(struct dev_ops *ops)
634 return devfs_destroy_dev_by_ops(ops, -1);
638 dev_ops_remove_minor(struct dev_ops *ops, int minor)
640 return devfs_destroy_dev_by_ops(ops, minor);
644 dev_ops_intercept(cdev_t dev, struct dev_ops *iops)
646 struct dev_ops *oops = dev->si_ops;
648 compile_dev_ops(iops);
649 iops->head.maj = oops->head.maj;
650 iops->head.data = oops->head.data;
651 iops->head.flags = oops->head.flags;
653 dev->si_flags |= SI_INTERCEPTED;
659 dev_ops_restore(cdev_t dev, struct dev_ops *oops)
661 struct dev_ops *iops = dev->si_ops;
664 dev->si_flags &= ~SI_INTERCEPTED;
666 iops->head.data = NULL;
667 iops->head.flags = 0;
670 /************************************************************************
671 * DEFAULT DEV OPS FUNCTIONS *
672 ************************************************************************/
676 * Unsupported devswitch functions (e.g. for writing to read-only device).
677 * XXX may belong elsewhere.
680 norevoke(struct dev_revoke_args *ap)
687 noclone(struct dev_clone_args *ap)
690 return (0); /* allow the clone */
694 noopen(struct dev_open_args *ap)
700 noclose(struct dev_close_args *ap)
706 noread(struct dev_read_args *ap)
712 nowrite(struct dev_write_args *ap)
718 noioctl(struct dev_ioctl_args *ap)
724 nokqfilter(struct dev_kqfilter_args *ap)
730 nommap(struct dev_mmap_args *ap)
736 nommap_single(struct dev_mmap_single_args *ap)
742 nostrategy(struct dev_strategy_args *ap)
744 struct bio *bio = ap->a_bio;
746 bio->bio_buf->b_flags |= B_ERROR;
747 bio->bio_buf->b_error = EOPNOTSUPP;
753 nopsize(struct dev_psize_args *ap)
760 nodump(struct dev_dump_args *ap)
766 * XXX this is probably bogus. Any device that uses it isn't checking the
770 nullopen(struct dev_open_args *ap)
776 nullclose(struct dev_close_args *ap)