xref: /netbsd-src/sys/miscfs/specfs/spec_vnops.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: spec_vnops.c,v 1.115 2008/01/25 16:21:04 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
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  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the NetBSD
18  *	Foundation, Inc. and its contributors.
19  * 4. Neither the name of The NetBSD Foundation nor the names of its
20  *    contributors may be used to endorse or promote products derived
21  *    from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
25  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
26  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
27  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /*
37  * Copyright (c) 1989, 1993
38  *	The Regents of the University of California.  All rights reserved.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the University nor the names of its contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  *
64  *	@(#)spec_vnops.c	8.15 (Berkeley) 7/14/95
65  */
66 
67 #include <sys/cdefs.h>
68 __KERNEL_RCSID(0, "$NetBSD: spec_vnops.c,v 1.115 2008/01/25 16:21:04 hannken Exp $");
69 
70 #include <sys/param.h>
71 #include <sys/proc.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/conf.h>
75 #include <sys/buf.h>
76 #include <sys/mount.h>
77 #include <sys/namei.h>
78 #include <sys/vnode.h>
79 #include <sys/stat.h>
80 #include <sys/errno.h>
81 #include <sys/ioctl.h>
82 #include <sys/poll.h>
83 #include <sys/file.h>
84 #include <sys/disklabel.h>
85 #include <sys/lockf.h>
86 #include <sys/tty.h>
87 #include <sys/kauth.h>
88 #include <sys/fstrans.h>
89 
90 #include <miscfs/genfs/genfs.h>
91 #include <miscfs/specfs/specdev.h>
92 
93 /* symbolic sleep message strings for devices */
94 const char	devopn[] = "devopn";
95 const char	devio[] = "devio";
96 const char	devwait[] = "devwait";
97 const char	devin[] = "devin";
98 const char	devout[] = "devout";
99 const char	devioc[] = "devioc";
100 const char	devcls[] = "devcls";
101 
102 vnode_t		*specfs_hash[SPECHSZ];
103 kmutex_t	specfs_lock;
104 
105 /*
106  * This vnode operations vector is used for special device nodes
107  * created from whole cloth by the kernel.  For the ops vector for
108  * vnodes built from special devices found in a filesystem, see (e.g)
109  * ffs_specop_entries[] in ffs_vnops.c or the equivalent for other
110  * filesystems.
111  */
112 
113 int (**spec_vnodeop_p)(void *);
114 const struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
115 	{ &vop_default_desc, vn_default_error },
116 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
117 	{ &vop_create_desc, spec_create },		/* create */
118 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
119 	{ &vop_open_desc, spec_open },			/* open */
120 	{ &vop_close_desc, spec_close },		/* close */
121 	{ &vop_access_desc, spec_access },		/* access */
122 	{ &vop_getattr_desc, spec_getattr },		/* getattr */
123 	{ &vop_setattr_desc, spec_setattr },		/* setattr */
124 	{ &vop_read_desc, spec_read },			/* read */
125 	{ &vop_write_desc, spec_write },		/* write */
126 	{ &vop_fcntl_desc, spec_fcntl },		/* fcntl */
127 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
128 	{ &vop_poll_desc, spec_poll },			/* poll */
129 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
130 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
131 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
132 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
133 	{ &vop_seek_desc, spec_seek },			/* seek */
134 	{ &vop_remove_desc, spec_remove },		/* remove */
135 	{ &vop_link_desc, spec_link },			/* link */
136 	{ &vop_rename_desc, spec_rename },		/* rename */
137 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
138 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
139 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
140 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
141 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
142 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
143 	{ &vop_inactive_desc, spec_inactive },		/* inactive */
144 	{ &vop_reclaim_desc, spec_reclaim },		/* reclaim */
145 	{ &vop_lock_desc, spec_lock },			/* lock */
146 	{ &vop_unlock_desc, spec_unlock },		/* unlock */
147 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
148 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
149 	{ &vop_print_desc, spec_print },		/* print */
150 	{ &vop_islocked_desc, spec_islocked },		/* islocked */
151 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
152 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
153 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
154 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
155 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
156 	{ NULL, NULL }
157 };
158 const struct vnodeopv_desc spec_vnodeop_opv_desc =
159 	{ &spec_vnodeop_p, spec_vnodeop_entries };
160 
161 /*
162  * Returns true if dev is /dev/mem or /dev/kmem.
163  */
164 int
165 iskmemdev(dev_t dev)
166 {
167 	/* mem_no is emitted by config(8) to generated devsw.c */
168 	extern const int mem_no;
169 
170 	/* minor 14 is /dev/io on i386 with COMPAT_10 */
171 	return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14));
172 }
173 
174 /*
175  * Initialize a vnode that represents a device.
176  */
177 void
178 spec_node_init(vnode_t *vp, dev_t rdev)
179 {
180 	specnode_t *sn;
181 	specdev_t *sd;
182 	vnode_t *vp2;
183 	vnode_t **vpp;
184 
185 	KASSERT(vp->v_type == VBLK || vp->v_type == VCHR);
186 	KASSERT(vp->v_specnode == NULL);
187 
188 	/*
189 	 * Search the hash table for this device.  If known, add a
190 	 * reference to the device structure.  If not known, create
191 	 * a new entry to represent the device.  In all cases add
192 	 * the vnode to the hash table.
193 	 */
194 	sn = kmem_alloc(sizeof(*sn), KM_SLEEP);
195 	if (sn == NULL) {
196 		/* XXX */
197 		panic("spec_node_init: unable to allocate memory");
198 	}
199 	sd = kmem_alloc(sizeof(*sd), KM_SLEEP);
200 	if (sd == NULL) {
201 		/* XXX */
202 		panic("spec_node_init: unable to allocate memory");
203 	}
204 	mutex_enter(&specfs_lock);
205 	vpp = &specfs_hash[SPECHASH(rdev)];
206 	for (vp2 = *vpp; vp2 != NULL; vp2 = vp2->v_specnext) {
207 		KASSERT(vp2->v_specnode != NULL);
208 		if (rdev == vp2->v_rdev && vp->v_type == vp2->v_type) {
209 			break;
210 		}
211 	}
212 	if (vp2 == NULL) {
213 		/* No existing record, create a new one. */
214 		sd->sd_rdev = rdev;
215 		sd->sd_mountpoint = NULL;
216 		sd->sd_lockf = NULL;
217 		sd->sd_refcnt = 1;
218 		sd->sd_opencnt = 0;
219 		sd->sd_bdevvp = NULL;
220 		sn->sn_dev = sd;
221 		sd = NULL;
222 	} else {
223 		/* Use the existing record. */
224 		sn->sn_dev = vp2->v_specnode->sn_dev;
225 		sn->sn_dev->sd_refcnt++;
226 	}
227 	/* Insert vnode into the hash chain. */
228 	sn->sn_opencnt = 0;
229 	sn->sn_rdev = rdev;
230 	sn->sn_gone = false;
231 	vp->v_specnode = sn;
232 	vp->v_specnext = *vpp;
233 	*vpp = vp;
234 	mutex_exit(&specfs_lock);
235 
236 	/* Free the record we allocated if unused. */
237 	if (sd != NULL) {
238 		kmem_free(sd, sizeof(*sd));
239 	}
240 }
241 
242 /*
243  * A vnode representing a special device is going away.  Close
244  * the device if the vnode holds it open.
245  */
246 void
247 spec_node_revoke(vnode_t *vp)
248 {
249 	specnode_t *sn;
250 	specdev_t *sd;
251 
252 	sn = vp->v_specnode;
253 	sd = sn->sn_dev;
254 
255 	KASSERT(vp->v_type == VBLK || vp->v_type == VCHR);
256 	KASSERT(vp->v_specnode != NULL);
257 	KASSERT((vp->v_iflag & VI_XLOCK) != 0);
258 	KASSERT(sn->sn_gone == false);
259 
260 	mutex_enter(&specfs_lock);
261 	KASSERT(sn->sn_opencnt <= sd->sd_opencnt);
262 	if (sn->sn_opencnt != 0) {
263 		sd->sd_opencnt -= (sn->sn_opencnt - 1);
264 		sn->sn_opencnt = 1;
265 		sn->sn_gone = true;
266 		mutex_exit(&specfs_lock);
267 
268 		VOP_CLOSE(vp, FNONBLOCK, NOCRED);
269 
270 		mutex_enter(&specfs_lock);
271 		KASSERT(sn->sn_opencnt == 0);
272 	}
273 	mutex_exit(&specfs_lock);
274 }
275 
276 /*
277  * A vnode representing a special device is being recycled.
278  * Destroy the specfs component.
279  */
280 void
281 spec_node_destroy(vnode_t *vp)
282 {
283 	specnode_t *sn;
284 	specdev_t *sd;
285 	vnode_t **vpp, *vp2;
286 	int refcnt;
287 
288 	sn = vp->v_specnode;
289 	sd = sn->sn_dev;
290 
291 	KASSERT(vp->v_type == VBLK || vp->v_type == VCHR);
292 	KASSERT(vp->v_specnode != NULL);
293 	KASSERT(sn->sn_opencnt == 0);
294 
295 	mutex_enter(&specfs_lock);
296 	/* Remove from the hash and destroy the node. */
297 	vpp = &specfs_hash[SPECHASH(vp->v_rdev)];
298 	for (vp2 = *vpp;; vp2 = vp2->v_specnext) {
299 		if (vp2 == NULL) {
300 			panic("spec_node_destroy: corrupt hash");
301 		}
302 		if (vp2 == vp) {
303 			KASSERT(vp == *vpp);
304 			*vpp = vp->v_specnext;
305 			break;
306 		}
307 		if (vp2->v_specnext == vp) {
308 			vp2->v_specnext = vp->v_specnext;
309 			break;
310 		}
311 	}
312 	sn = vp->v_specnode;
313 	vp->v_specnode = NULL;
314 	refcnt = sd->sd_refcnt--;
315 	KASSERT(refcnt > 0);
316 	mutex_exit(&specfs_lock);
317 
318 	/* If the device is no longer in use, destroy our record. */
319 	if (refcnt == 1) {
320 		KASSERT(sd->sd_opencnt == 0);
321 		KASSERT(sd->sd_bdevvp == NULL);
322 		kmem_free(sd, sizeof(*sd));
323 	}
324 	kmem_free(sn, sizeof(*sn));
325 }
326 
327 /*
328  * Trivial lookup routine that always fails.
329  */
330 int
331 spec_lookup(void *v)
332 {
333 	struct vop_lookup_args /* {
334 		struct vnode *a_dvp;
335 		struct vnode **a_vpp;
336 		struct componentname *a_cnp;
337 	} */ *ap = v;
338 
339 	*ap->a_vpp = NULL;
340 	return (ENOTDIR);
341 }
342 
343 /*
344  * Open a special file.
345  */
346 /* ARGSUSED */
347 int
348 spec_open(void *v)
349 {
350 	struct vop_open_args /* {
351 		struct vnode *a_vp;
352 		int  a_mode;
353 		kauth_cred_t a_cred;
354 	} */ *ap = v;
355 	struct lwp *l;
356 	struct vnode *vp;
357 	dev_t dev;
358 	int error;
359 	struct partinfo pi;
360 	enum kauth_device_req req;
361 	specnode_t *sn;
362 	specdev_t *sd;
363 
364 	l = curlwp;
365 	vp = ap->a_vp;
366 	dev = vp->v_rdev;
367 	sn = vp->v_specnode;
368 	sd = sn->sn_dev;
369 
370 	/*
371 	 * Don't allow open if fs is mounted -nodev.
372 	 */
373 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
374 		return (ENXIO);
375 
376 	switch (ap->a_mode & (FREAD | FWRITE)) {
377 	case FREAD | FWRITE:
378 		req = KAUTH_REQ_DEVICE_RAWIO_SPEC_RW;
379 		break;
380 	case FWRITE:
381 		req = KAUTH_REQ_DEVICE_RAWIO_SPEC_WRITE;
382 		break;
383 	default:
384 		req = KAUTH_REQ_DEVICE_RAWIO_SPEC_READ;
385 		break;
386 	}
387 
388 	switch (vp->v_type) {
389 	case VCHR:
390 		error = kauth_authorize_device_spec(ap->a_cred, req, vp);
391 		if (error != 0)
392 			return (error);
393 
394 		/*
395 		 * Character devices can accept opens from multiple
396 		 * vnodes.
397 		 */
398 		mutex_enter(&specfs_lock);
399 		if (sn->sn_gone) {
400 			mutex_exit(&specfs_lock);
401 			return (EBADF);
402 		}
403 		sd->sd_opencnt++;
404 		sn->sn_opencnt++;
405 		mutex_exit(&specfs_lock);
406 		if (cdev_type(dev) == D_TTY)
407 			vp->v_vflag |= VV_ISTTY;
408 		VOP_UNLOCK(vp, 0);
409 		error = cdev_open(dev, ap->a_mode, S_IFCHR, l);
410 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
411 		break;
412 
413 	case VBLK:
414 		error = kauth_authorize_device_spec(ap->a_cred, req, vp);
415 		if (error != 0)
416 			return (error);
417 
418 		/*
419 		 * For block devices, permit only one open.  The buffer
420 		 * cache cannot remain self-consistent with multiple
421 		 * vnodes holding a block device open.
422 		 */
423 		mutex_enter(&specfs_lock);
424 		if (sn->sn_gone) {
425 			mutex_exit(&specfs_lock);
426 			return (EBADF);
427 		}
428 		if (sd->sd_opencnt != 0) {
429 			mutex_exit(&specfs_lock);
430 			return EBUSY;
431 		}
432 		sn->sn_opencnt = 1;
433 		sd->sd_opencnt = 1;
434 		sd->sd_bdevvp = vp;
435 		mutex_exit(&specfs_lock);
436 
437 		error = bdev_open(dev, ap->a_mode, S_IFBLK, l);
438 		break;
439 
440 	case VNON:
441 	case VLNK:
442 	case VDIR:
443 	case VREG:
444 	case VBAD:
445 	case VFIFO:
446 	case VSOCK:
447 	default:
448 		return 0;
449 	}
450 
451 	mutex_enter(&specfs_lock);
452 	if (sn->sn_gone) {
453 		if (error == 0)
454 			error = EBADF;
455 	} else if (error != 0) {
456 		sd->sd_opencnt--;
457 		sn->sn_opencnt--;
458 		if (vp->v_type == VBLK)
459 			sd->sd_bdevvp = NULL;
460 
461 	}
462 	mutex_exit(&specfs_lock);
463 
464 	if (cdev_type(dev) != D_DISK || error != 0)
465 		return error;
466 
467 	if (vp->v_type == VCHR)
468 		error = cdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp);
469 	else
470 		error = bdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp);
471 	if (error == 0)
472 		uvm_vnp_setsize(vp,
473 		    (voff_t)pi.disklab->d_secsize * pi.part->p_size);
474 	return 0;
475 }
476 
477 /*
478  * Vnode op for read
479  */
480 /* ARGSUSED */
481 int
482 spec_read(void *v)
483 {
484 	struct vop_read_args /* {
485 		struct vnode *a_vp;
486 		struct uio *a_uio;
487 		int  a_ioflag;
488 		kauth_cred_t a_cred;
489 	} */ *ap = v;
490 	struct vnode *vp = ap->a_vp;
491 	struct uio *uio = ap->a_uio;
492  	struct lwp *l = curlwp;
493 	struct buf *bp;
494 	daddr_t bn;
495 	int bsize, bscale;
496 	struct partinfo dpart;
497 	int n, on;
498 	int error = 0;
499 
500 #ifdef DIAGNOSTIC
501 	if (uio->uio_rw != UIO_READ)
502 		panic("spec_read mode");
503 	if (&uio->uio_vmspace->vm_map != kernel_map &&
504 	    uio->uio_vmspace != curproc->p_vmspace)
505 		panic("spec_read proc");
506 #endif
507 	if (uio->uio_resid == 0)
508 		return (0);
509 
510 	switch (vp->v_type) {
511 
512 	case VCHR:
513 		VOP_UNLOCK(vp, 0);
514 		error = cdev_read(vp->v_rdev, uio, ap->a_ioflag);
515 		vn_lock(vp, LK_SHARED | LK_RETRY);
516 		return (error);
517 
518 	case VBLK:
519 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
520 		if (uio->uio_offset < 0)
521 			return (EINVAL);
522 		bsize = BLKDEV_IOSIZE;
523 		if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) {
524 			if (dpart.part->p_fstype == FS_BSDFFS &&
525 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
526 				bsize = dpart.part->p_frag *
527 				    dpart.part->p_fsize;
528 		}
529 		bscale = bsize >> DEV_BSHIFT;
530 		do {
531 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
532 			on = uio->uio_offset % bsize;
533 			n = min((unsigned)(bsize - on), uio->uio_resid);
534 			error = bread(vp, bn, bsize, NOCRED, &bp);
535 			n = min(n, bsize - bp->b_resid);
536 			if (error) {
537 				brelse(bp, 0);
538 				return (error);
539 			}
540 			error = uiomove((char *)bp->b_data + on, n, uio);
541 			brelse(bp, 0);
542 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
543 		return (error);
544 
545 	default:
546 		panic("spec_read type");
547 	}
548 	/* NOTREACHED */
549 }
550 
551 /*
552  * Vnode op for write
553  */
554 /* ARGSUSED */
555 int
556 spec_write(void *v)
557 {
558 	struct vop_write_args /* {
559 		struct vnode *a_vp;
560 		struct uio *a_uio;
561 		int  a_ioflag;
562 		kauth_cred_t a_cred;
563 	} */ *ap = v;
564 	struct vnode *vp = ap->a_vp;
565 	struct uio *uio = ap->a_uio;
566 	struct lwp *l = curlwp;
567 	struct buf *bp;
568 	daddr_t bn;
569 	int bsize, bscale;
570 	struct partinfo dpart;
571 	int n, on;
572 	int error = 0;
573 
574 #ifdef DIAGNOSTIC
575 	if (uio->uio_rw != UIO_WRITE)
576 		panic("spec_write mode");
577 	if (&uio->uio_vmspace->vm_map != kernel_map &&
578 	    uio->uio_vmspace != curproc->p_vmspace)
579 		panic("spec_write proc");
580 #endif
581 
582 	switch (vp->v_type) {
583 
584 	case VCHR:
585 		VOP_UNLOCK(vp, 0);
586 		error = cdev_write(vp->v_rdev, uio, ap->a_ioflag);
587 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
588 		return (error);
589 
590 	case VBLK:
591 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
592 		if (uio->uio_resid == 0)
593 			return (0);
594 		if (uio->uio_offset < 0)
595 			return (EINVAL);
596 		bsize = BLKDEV_IOSIZE;
597 		if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) {
598 			if (dpart.part->p_fstype == FS_BSDFFS &&
599 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
600 				bsize = dpart.part->p_frag *
601 				    dpart.part->p_fsize;
602 		}
603 		bscale = bsize >> DEV_BSHIFT;
604 		do {
605 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
606 			on = uio->uio_offset % bsize;
607 			n = min((unsigned)(bsize - on), uio->uio_resid);
608 			if (n == bsize)
609 				bp = getblk(vp, bn, bsize, 0, 0);
610 			else
611 				error = bread(vp, bn, bsize, NOCRED, &bp);
612 			if (error) {
613 				brelse(bp, 0);
614 				return (error);
615 			}
616 			n = min(n, bsize - bp->b_resid);
617 			error = uiomove((char *)bp->b_data + on, n, uio);
618 			if (error)
619 				brelse(bp, 0);
620 			else {
621 				if (n + on == bsize)
622 					bawrite(bp);
623 				else
624 					bdwrite(bp);
625 				error = bp->b_error;
626 			}
627 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
628 		return (error);
629 
630 	default:
631 		panic("spec_write type");
632 	}
633 	/* NOTREACHED */
634 }
635 
636 /*
637  * Device ioctl operation.
638  */
639 /* ARGSUSED */
640 int
641 spec_ioctl(void *v)
642 {
643 	struct vop_ioctl_args /* {
644 		struct vnode *a_vp;
645 		u_long a_command;
646 		void  *a_data;
647 		int  a_fflag;
648 		kauth_cred_t a_cred;
649 	} */ *ap = v;
650 	struct vnode *vp;
651 	dev_t dev;
652 
653 	/*
654 	 * Extract all the info we need from the vnode, taking care to
655 	 * avoid a race with VOP_REVOKE().
656 	 */
657 
658 	vp = ap->a_vp;
659 	dev = NODEV;
660 	mutex_enter(&vp->v_interlock);
661 	if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) {
662 		dev = vp->v_rdev;
663 	}
664 	mutex_exit(&vp->v_interlock);
665 	if (dev == NODEV) {
666 		return ENXIO;
667 	}
668 
669 	switch (vp->v_type) {
670 
671 	case VCHR:
672 		return cdev_ioctl(dev, ap->a_command, ap->a_data,
673 		    ap->a_fflag, curlwp);
674 
675 	case VBLK:
676 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
677 		return bdev_ioctl(dev, ap->a_command, ap->a_data,
678 		   ap->a_fflag, curlwp);
679 
680 	default:
681 		panic("spec_ioctl");
682 		/* NOTREACHED */
683 	}
684 }
685 
686 /* ARGSUSED */
687 int
688 spec_poll(void *v)
689 {
690 	struct vop_poll_args /* {
691 		struct vnode *a_vp;
692 		int a_events;
693 	} */ *ap = v;
694 	struct vnode *vp;
695 	dev_t dev;
696 
697 	/*
698 	 * Extract all the info we need from the vnode, taking care to
699 	 * avoid a race with VOP_REVOKE().
700 	 */
701 
702 	vp = ap->a_vp;
703 	dev = NODEV;
704 	mutex_enter(&vp->v_interlock);
705 	if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) {
706 		dev = vp->v_rdev;
707 	}
708 	mutex_exit(&vp->v_interlock);
709 	if (dev == NODEV) {
710 		return POLLERR;
711 	}
712 
713 	switch (vp->v_type) {
714 
715 	case VCHR:
716 		return cdev_poll(dev, ap->a_events, curlwp);
717 
718 	default:
719 		return (genfs_poll(v));
720 	}
721 }
722 
723 /* ARGSUSED */
724 int
725 spec_kqfilter(void *v)
726 {
727 	struct vop_kqfilter_args /* {
728 		struct vnode	*a_vp;
729 		struct proc	*a_kn;
730 	} */ *ap = v;
731 	dev_t dev;
732 
733 	switch (ap->a_vp->v_type) {
734 
735 	case VCHR:
736 		dev = ap->a_vp->v_rdev;
737 		return cdev_kqfilter(dev, ap->a_kn);
738 	default:
739 		/*
740 		 * Block devices don't support kqfilter, and refuse it
741 		 * for any other files (like those vflush()ed) too.
742 		 */
743 		return (EOPNOTSUPP);
744 	}
745 }
746 
747 /*
748  * Allow mapping of only D_DISK.  This is called only for VBLK.
749  */
750 int
751 spec_mmap(void *v)
752 {
753 	struct vop_mmap_args /* {
754 		struct vnode *a_vp;
755 		vm_prot_t a_prot;
756 		kauth_cred_t a_cred;
757 	} */ *ap = v;
758 	struct vnode *vp = ap->a_vp;
759 
760 	KASSERT(vp->v_type == VBLK);
761 	if (bdev_type(vp->v_rdev) != D_DISK)
762 		return EINVAL;
763 
764 	return 0;
765 }
766 
767 /*
768  * Synch buffers associated with a block device
769  */
770 /* ARGSUSED */
771 int
772 spec_fsync(void *v)
773 {
774 	struct vop_fsync_args /* {
775 		struct vnode *a_vp;
776 		kauth_cred_t a_cred;
777 		int  a_flags;
778 		off_t offlo;
779 		off_t offhi;
780 	} */ *ap = v;
781 	struct vnode *vp = ap->a_vp;
782 
783 	if (vp->v_type == VBLK) {
784 		vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0);
785 	}
786 	return (0);
787 }
788 
789 /*
790  * Just call the device strategy routine
791  */
792 int
793 spec_strategy(void *v)
794 {
795 	struct vop_strategy_args /* {
796 		struct vnode *a_vp;
797 		struct buf *a_bp;
798 	} */ *ap = v;
799 	struct vnode *vp = ap->a_vp;
800 	struct buf *bp = ap->a_bp;
801 	int error;
802 
803 	KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
804 
805 	error = 0;
806 	bp->b_dev = vp->v_rdev;
807 	if (!(bp->b_flags & B_READ) &&
808 	    (LIST_FIRST(&bp->b_dep)) != NULL && bioopsp)
809 		bioopsp->io_start(bp);
810 
811 	if (!(bp->b_flags & B_READ))
812 		error = fscow_run(bp, false);
813 
814 	if (error) {
815 		bp->b_error = error;
816 		biodone(bp);
817 		return (error);
818 	}
819 
820 	bdev_strategy(bp);
821 
822 	return (0);
823 }
824 
825 int
826 spec_inactive(void *v)
827 {
828 	struct vop_inactive_args /* {
829 		struct vnode *a_vp;
830 		struct proc *a_l;
831 	} */ *ap = v;
832 
833 	VOP_UNLOCK(ap->a_vp, 0);
834 	return (0);
835 }
836 
837 /*
838  * This is a noop, simply returning what one has been given.
839  */
840 int
841 spec_bmap(void *v)
842 {
843 	struct vop_bmap_args /* {
844 		struct vnode *a_vp;
845 		daddr_t  a_bn;
846 		struct vnode **a_vpp;
847 		daddr_t *a_bnp;
848 		int *a_runp;
849 	} */ *ap = v;
850 
851 	if (ap->a_vpp != NULL)
852 		*ap->a_vpp = ap->a_vp;
853 	if (ap->a_bnp != NULL)
854 		*ap->a_bnp = ap->a_bn;
855 	if (ap->a_runp != NULL)
856 		*ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1;
857 	return (0);
858 }
859 
860 /*
861  * Device close routine
862  */
863 /* ARGSUSED */
864 int
865 spec_close(void *v)
866 {
867 	struct vop_close_args /* {
868 		struct vnode *a_vp;
869 		int  a_fflag;
870 		kauth_cred_t a_cred;
871 	} */ *ap = v;
872 	struct vnode *vp = ap->a_vp;
873 	struct session *sess;
874 	dev_t dev = vp->v_rdev;
875 	int mode, error, flags, flags1, count;
876 	specnode_t *sn;
877 	specdev_t *sd;
878 
879 	flags = vp->v_iflag;
880 	sn = vp->v_specnode;
881 	sd = sn->sn_dev;
882 
883 	switch (vp->v_type) {
884 
885 	case VCHR:
886 		/*
887 		 * Hack: a tty device that is a controlling terminal
888 		 * has a reference from the session structure.  We
889 		 * cannot easily tell that a character device is a
890 		 * controlling terminal, unless it is the closing
891 		 * process' controlling terminal.  In that case, if the
892 		 * open count is 1 release the reference from the
893 		 * session.  Also, remove the link from the tty back to
894 		 * the session and pgrp.
895 		 *
896 		 * XXX V. fishy.
897 		 */
898 		mutex_enter(&proclist_lock);
899 		sess = curlwp->l_proc->p_session;
900 		if (sn->sn_opencnt == 1 && vp == sess->s_ttyvp) {
901 			mutex_spin_enter(&tty_lock);
902 			sess->s_ttyvp = NULL;
903 			if (sess->s_ttyp->t_session != NULL) {
904 				sess->s_ttyp->t_pgrp = NULL;
905 				sess->s_ttyp->t_session = NULL;
906 				mutex_spin_exit(&tty_lock);
907 				SESSRELE(sess);
908 				mutex_exit(&proclist_lock);
909 			} else {
910 				mutex_spin_exit(&tty_lock);
911 				if (sess->s_ttyp->t_pgrp != NULL)
912 					panic("spec_close: spurious pgrp ref");
913 				mutex_exit(&proclist_lock);
914 			}
915 			vrele(vp);
916 		} else
917 			mutex_exit(&proclist_lock);
918 
919 		/*
920 		 * If the vnode is locked, then we are in the midst
921 		 * of forcably closing the device, otherwise we only
922 		 * close on last reference.
923 		 */
924 		mode = S_IFCHR;
925 		break;
926 
927 	case VBLK:
928 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
929 		/*
930 		 * On last close of a block device (that isn't mounted)
931 		 * we must invalidate any in core blocks, so that
932 		 * we can, for instance, change floppy disks.
933 		 */
934 		error = vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 0, 0);
935 		if (error)
936 			return (error);
937 		/*
938 		 * We do not want to really close the device if it
939 		 * is still in use unless we are trying to close it
940 		 * forcibly. Since every use (buffer, vnode, swap, cmap)
941 		 * holds a reference to the vnode, and because we mark
942 		 * any other vnodes that alias this device, when the
943 		 * sum of the reference counts on all the aliased
944 		 * vnodes descends to one, we are on last close.
945 		 */
946 		mode = S_IFBLK;
947 		break;
948 
949 	default:
950 		panic("spec_close: not special");
951 	}
952 
953 	mutex_enter(&specfs_lock);
954 	sn->sn_opencnt--;
955 	count = --sd->sd_opencnt;
956 	if (vp->v_type == VBLK)
957 		sd->sd_bdevvp = NULL;
958 	mutex_exit(&specfs_lock);
959 
960 	if (count != 0)
961 		return 0;
962 
963 	flags1 = ap->a_fflag;
964 
965 	/*
966 	 * if VI_XLOCK is set, then we're going away soon, so make this
967 	 * non-blocking. Also ensures that we won't wedge in vn_lock below.
968 	 */
969 	if (flags & VI_XLOCK)
970 		flags1 |= FNONBLOCK;
971 
972 	/*
973 	 * If we're able to block, release the vnode lock & reacquire. We
974 	 * might end up sleeping for someone else who wants our queues. They
975 	 * won't get them if we hold the vnode locked. Also, if VI_XLOCK is
976 	 * set, don't release the lock as we won't be able to regain it.
977 	 */
978 	if (!(flags1 & FNONBLOCK))
979 		VOP_UNLOCK(vp, 0);
980 
981 	if (vp->v_type == VBLK)
982 		error = bdev_close(dev, flags1, mode, curlwp);
983 	else
984 		error = cdev_close(dev, flags1, mode, curlwp);
985 
986 	if (!(flags1 & FNONBLOCK))
987 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
988 
989 	return (error);
990 }
991 
992 /*
993  * Print out the contents of a special device vnode.
994  */
995 int
996 spec_print(void *v)
997 {
998 	struct vop_print_args /* {
999 		struct vnode *a_vp;
1000 	} */ *ap = v;
1001 
1002 	printf("dev %d, %d\n", major(ap->a_vp->v_rdev),
1003 	    minor(ap->a_vp->v_rdev));
1004 	return 0;
1005 }
1006 
1007 /*
1008  * Return POSIX pathconf information applicable to special devices.
1009  */
1010 int
1011 spec_pathconf(void *v)
1012 {
1013 	struct vop_pathconf_args /* {
1014 		struct vnode *a_vp;
1015 		int a_name;
1016 		register_t *a_retval;
1017 	} */ *ap = v;
1018 
1019 	switch (ap->a_name) {
1020 	case _PC_LINK_MAX:
1021 		*ap->a_retval = LINK_MAX;
1022 		return (0);
1023 	case _PC_MAX_CANON:
1024 		*ap->a_retval = MAX_CANON;
1025 		return (0);
1026 	case _PC_MAX_INPUT:
1027 		*ap->a_retval = MAX_INPUT;
1028 		return (0);
1029 	case _PC_PIPE_BUF:
1030 		*ap->a_retval = PIPE_BUF;
1031 		return (0);
1032 	case _PC_CHOWN_RESTRICTED:
1033 		*ap->a_retval = 1;
1034 		return (0);
1035 	case _PC_VDISABLE:
1036 		*ap->a_retval = _POSIX_VDISABLE;
1037 		return (0);
1038 	case _PC_SYNC_IO:
1039 		*ap->a_retval = 1;
1040 		return (0);
1041 	default:
1042 		return (EINVAL);
1043 	}
1044 	/* NOTREACHED */
1045 }
1046 
1047 /*
1048  * Advisory record locking support.
1049  */
1050 int
1051 spec_advlock(void *v)
1052 {
1053 	struct vop_advlock_args /* {
1054 		struct vnode *a_vp;
1055 		void *a_id;
1056 		int a_op;
1057 		struct flock *a_fl;
1058 		int a_flags;
1059 	} */ *ap = v;
1060 	struct vnode *vp = ap->a_vp;
1061 
1062 	return lf_advlock(ap, &vp->v_speclockf, (off_t)0);
1063 }
1064