xref: /netbsd-src/sys/miscfs/specfs/spec_vnops.c (revision 466a16a118933bd295a8a104f095714fadf9cf68)
1 /*	$NetBSD: spec_vnops.c,v 1.119 2008/05/16 09:22:00 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  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * Copyright (c) 1989, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. Neither the name of the University nor the names of its contributors
42  *    may be used to endorse or promote products derived from this software
43  *    without specific prior written permission.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  *
57  *	@(#)spec_vnops.c	8.15 (Berkeley) 7/14/95
58  */
59 
60 #include <sys/cdefs.h>
61 __KERNEL_RCSID(0, "$NetBSD: spec_vnops.c,v 1.119 2008/05/16 09:22:00 hannken Exp $");
62 
63 #include <sys/param.h>
64 #include <sys/proc.h>
65 #include <sys/systm.h>
66 #include <sys/kernel.h>
67 #include <sys/conf.h>
68 #include <sys/buf.h>
69 #include <sys/mount.h>
70 #include <sys/namei.h>
71 #include <sys/vnode.h>
72 #include <sys/stat.h>
73 #include <sys/errno.h>
74 #include <sys/ioctl.h>
75 #include <sys/poll.h>
76 #include <sys/file.h>
77 #include <sys/disklabel.h>
78 #include <sys/lockf.h>
79 #include <sys/tty.h>
80 #include <sys/kauth.h>
81 #include <sys/fstrans.h>
82 
83 #include <miscfs/genfs/genfs.h>
84 #include <miscfs/specfs/specdev.h>
85 
86 /* symbolic sleep message strings for devices */
87 const char	devopn[] = "devopn";
88 const char	devio[] = "devio";
89 const char	devwait[] = "devwait";
90 const char	devin[] = "devin";
91 const char	devout[] = "devout";
92 const char	devioc[] = "devioc";
93 const char	devcls[] = "devcls";
94 
95 vnode_t		*specfs_hash[SPECHSZ];
96 kmutex_t	specfs_lock;
97 
98 /*
99  * This vnode operations vector is used for special device nodes
100  * created from whole cloth by the kernel.  For the ops vector for
101  * vnodes built from special devices found in a filesystem, see (e.g)
102  * ffs_specop_entries[] in ffs_vnops.c or the equivalent for other
103  * filesystems.
104  */
105 
106 int (**spec_vnodeop_p)(void *);
107 const struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
108 	{ &vop_default_desc, vn_default_error },
109 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
110 	{ &vop_create_desc, spec_create },		/* create */
111 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
112 	{ &vop_open_desc, spec_open },			/* open */
113 	{ &vop_close_desc, spec_close },		/* close */
114 	{ &vop_access_desc, spec_access },		/* access */
115 	{ &vop_getattr_desc, spec_getattr },		/* getattr */
116 	{ &vop_setattr_desc, spec_setattr },		/* setattr */
117 	{ &vop_read_desc, spec_read },			/* read */
118 	{ &vop_write_desc, spec_write },		/* write */
119 	{ &vop_fcntl_desc, spec_fcntl },		/* fcntl */
120 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
121 	{ &vop_poll_desc, spec_poll },			/* poll */
122 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
123 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
124 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
125 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
126 	{ &vop_seek_desc, spec_seek },			/* seek */
127 	{ &vop_remove_desc, spec_remove },		/* remove */
128 	{ &vop_link_desc, spec_link },			/* link */
129 	{ &vop_rename_desc, spec_rename },		/* rename */
130 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
131 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
132 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
133 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
134 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
135 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
136 	{ &vop_inactive_desc, spec_inactive },		/* inactive */
137 	{ &vop_reclaim_desc, spec_reclaim },		/* reclaim */
138 	{ &vop_lock_desc, spec_lock },			/* lock */
139 	{ &vop_unlock_desc, spec_unlock },		/* unlock */
140 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
141 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
142 	{ &vop_print_desc, spec_print },		/* print */
143 	{ &vop_islocked_desc, spec_islocked },		/* islocked */
144 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
145 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
146 	{ &vop_bwrite_desc, spec_bwrite },		/* bwrite */
147 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
148 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
149 	{ NULL, NULL }
150 };
151 const struct vnodeopv_desc spec_vnodeop_opv_desc =
152 	{ &spec_vnodeop_p, spec_vnodeop_entries };
153 
154 /*
155  * Returns true if dev is /dev/mem or /dev/kmem.
156  */
157 int
158 iskmemdev(dev_t dev)
159 {
160 	/* mem_no is emitted by config(8) to generated devsw.c */
161 	extern const int mem_no;
162 
163 	/* minor 14 is /dev/io on i386 with COMPAT_10 */
164 	return (major(dev) == mem_no && (minor(dev) < 2 || minor(dev) == 14));
165 }
166 
167 /*
168  * Initialize a vnode that represents a device.
169  */
170 void
171 spec_node_init(vnode_t *vp, dev_t rdev)
172 {
173 	specnode_t *sn;
174 	specdev_t *sd;
175 	vnode_t *vp2;
176 	vnode_t **vpp;
177 
178 	KASSERT(vp->v_type == VBLK || vp->v_type == VCHR);
179 	KASSERT(vp->v_specnode == NULL);
180 
181 	/*
182 	 * Search the hash table for this device.  If known, add a
183 	 * reference to the device structure.  If not known, create
184 	 * a new entry to represent the device.  In all cases add
185 	 * the vnode to the hash table.
186 	 */
187 	sn = kmem_alloc(sizeof(*sn), KM_SLEEP);
188 	if (sn == NULL) {
189 		/* XXX */
190 		panic("spec_node_init: unable to allocate memory");
191 	}
192 	sd = kmem_alloc(sizeof(*sd), KM_SLEEP);
193 	if (sd == NULL) {
194 		/* XXX */
195 		panic("spec_node_init: unable to allocate memory");
196 	}
197 	mutex_enter(&specfs_lock);
198 	vpp = &specfs_hash[SPECHASH(rdev)];
199 	for (vp2 = *vpp; vp2 != NULL; vp2 = vp2->v_specnext) {
200 		KASSERT(vp2->v_specnode != NULL);
201 		if (rdev == vp2->v_rdev && vp->v_type == vp2->v_type) {
202 			break;
203 		}
204 	}
205 	if (vp2 == NULL) {
206 		/* No existing record, create a new one. */
207 		sd->sd_rdev = rdev;
208 		sd->sd_mountpoint = NULL;
209 		sd->sd_lockf = NULL;
210 		sd->sd_refcnt = 1;
211 		sd->sd_opencnt = 0;
212 		sd->sd_bdevvp = NULL;
213 		sn->sn_dev = sd;
214 		sd = NULL;
215 	} else {
216 		/* Use the existing record. */
217 		sn->sn_dev = vp2->v_specnode->sn_dev;
218 		sn->sn_dev->sd_refcnt++;
219 	}
220 	/* Insert vnode into the hash chain. */
221 	sn->sn_opencnt = 0;
222 	sn->sn_rdev = rdev;
223 	sn->sn_gone = false;
224 	vp->v_specnode = sn;
225 	vp->v_specnext = *vpp;
226 	*vpp = vp;
227 	mutex_exit(&specfs_lock);
228 
229 	/* Free the record we allocated if unused. */
230 	if (sd != NULL) {
231 		kmem_free(sd, sizeof(*sd));
232 	}
233 }
234 
235 /*
236  * A vnode representing a special device is going away.  Close
237  * the device if the vnode holds it open.
238  */
239 void
240 spec_node_revoke(vnode_t *vp)
241 {
242 	specnode_t *sn;
243 	specdev_t *sd;
244 
245 	sn = vp->v_specnode;
246 	sd = sn->sn_dev;
247 
248 	KASSERT(vp->v_type == VBLK || vp->v_type == VCHR);
249 	KASSERT(vp->v_specnode != NULL);
250 	KASSERT((vp->v_iflag & VI_XLOCK) != 0);
251 	KASSERT(sn->sn_gone == false);
252 
253 	mutex_enter(&specfs_lock);
254 	KASSERT(sn->sn_opencnt <= sd->sd_opencnt);
255 	if (sn->sn_opencnt != 0) {
256 		sd->sd_opencnt -= (sn->sn_opencnt - 1);
257 		sn->sn_opencnt = 1;
258 		sn->sn_gone = true;
259 		mutex_exit(&specfs_lock);
260 
261 		VOP_CLOSE(vp, FNONBLOCK, NOCRED);
262 
263 		mutex_enter(&specfs_lock);
264 		KASSERT(sn->sn_opencnt == 0);
265 	}
266 	mutex_exit(&specfs_lock);
267 }
268 
269 /*
270  * A vnode representing a special device is being recycled.
271  * Destroy the specfs component.
272  */
273 void
274 spec_node_destroy(vnode_t *vp)
275 {
276 	specnode_t *sn;
277 	specdev_t *sd;
278 	vnode_t **vpp, *vp2;
279 	int refcnt;
280 
281 	sn = vp->v_specnode;
282 	sd = sn->sn_dev;
283 
284 	KASSERT(vp->v_type == VBLK || vp->v_type == VCHR);
285 	KASSERT(vp->v_specnode != NULL);
286 	KASSERT(sn->sn_opencnt == 0);
287 
288 	mutex_enter(&specfs_lock);
289 	/* Remove from the hash and destroy the node. */
290 	vpp = &specfs_hash[SPECHASH(vp->v_rdev)];
291 	for (vp2 = *vpp;; vp2 = vp2->v_specnext) {
292 		if (vp2 == NULL) {
293 			panic("spec_node_destroy: corrupt hash");
294 		}
295 		if (vp2 == vp) {
296 			KASSERT(vp == *vpp);
297 			*vpp = vp->v_specnext;
298 			break;
299 		}
300 		if (vp2->v_specnext == vp) {
301 			vp2->v_specnext = vp->v_specnext;
302 			break;
303 		}
304 	}
305 	sn = vp->v_specnode;
306 	vp->v_specnode = NULL;
307 	refcnt = sd->sd_refcnt--;
308 	KASSERT(refcnt > 0);
309 	mutex_exit(&specfs_lock);
310 
311 	/* If the device is no longer in use, destroy our record. */
312 	if (refcnt == 1) {
313 		KASSERT(sd->sd_opencnt == 0);
314 		KASSERT(sd->sd_bdevvp == NULL);
315 		kmem_free(sd, sizeof(*sd));
316 	}
317 	kmem_free(sn, sizeof(*sn));
318 }
319 
320 /*
321  * Trivial lookup routine that always fails.
322  */
323 int
324 spec_lookup(void *v)
325 {
326 	struct vop_lookup_args /* {
327 		struct vnode *a_dvp;
328 		struct vnode **a_vpp;
329 		struct componentname *a_cnp;
330 	} */ *ap = v;
331 
332 	*ap->a_vpp = NULL;
333 	return (ENOTDIR);
334 }
335 
336 /*
337  * Open a special file.
338  */
339 /* ARGSUSED */
340 int
341 spec_open(void *v)
342 {
343 	struct vop_open_args /* {
344 		struct vnode *a_vp;
345 		int  a_mode;
346 		kauth_cred_t a_cred;
347 	} */ *ap = v;
348 	struct lwp *l;
349 	struct vnode *vp;
350 	dev_t dev;
351 	int error;
352 	struct partinfo pi;
353 	enum kauth_device_req req;
354 	specnode_t *sn;
355 	specdev_t *sd;
356 
357 	l = curlwp;
358 	vp = ap->a_vp;
359 	dev = vp->v_rdev;
360 	sn = vp->v_specnode;
361 	sd = sn->sn_dev;
362 
363 	/*
364 	 * Don't allow open if fs is mounted -nodev.
365 	 */
366 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
367 		return (ENXIO);
368 
369 	switch (ap->a_mode & (FREAD | FWRITE)) {
370 	case FREAD | FWRITE:
371 		req = KAUTH_REQ_DEVICE_RAWIO_SPEC_RW;
372 		break;
373 	case FWRITE:
374 		req = KAUTH_REQ_DEVICE_RAWIO_SPEC_WRITE;
375 		break;
376 	default:
377 		req = KAUTH_REQ_DEVICE_RAWIO_SPEC_READ;
378 		break;
379 	}
380 
381 	switch (vp->v_type) {
382 	case VCHR:
383 		error = kauth_authorize_device_spec(ap->a_cred, req, vp);
384 		if (error != 0)
385 			return (error);
386 
387 		/*
388 		 * Character devices can accept opens from multiple
389 		 * vnodes.
390 		 */
391 		mutex_enter(&specfs_lock);
392 		if (sn->sn_gone) {
393 			mutex_exit(&specfs_lock);
394 			return (EBADF);
395 		}
396 		sd->sd_opencnt++;
397 		sn->sn_opencnt++;
398 		mutex_exit(&specfs_lock);
399 		if (cdev_type(dev) == D_TTY)
400 			vp->v_vflag |= VV_ISTTY;
401 		VOP_UNLOCK(vp, 0);
402 		error = cdev_open(dev, ap->a_mode, S_IFCHR, l);
403 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
404 		break;
405 
406 	case VBLK:
407 		error = kauth_authorize_device_spec(ap->a_cred, req, vp);
408 		if (error != 0)
409 			return (error);
410 
411 		/*
412 		 * For block devices, permit only one open.  The buffer
413 		 * cache cannot remain self-consistent with multiple
414 		 * vnodes holding a block device open.
415 		 */
416 		mutex_enter(&specfs_lock);
417 		if (sn->sn_gone) {
418 			mutex_exit(&specfs_lock);
419 			return (EBADF);
420 		}
421 		if (sd->sd_opencnt != 0) {
422 			mutex_exit(&specfs_lock);
423 			return EBUSY;
424 		}
425 		sn->sn_opencnt = 1;
426 		sd->sd_opencnt = 1;
427 		sd->sd_bdevvp = vp;
428 		mutex_exit(&specfs_lock);
429 
430 		error = bdev_open(dev, ap->a_mode, S_IFBLK, l);
431 		break;
432 
433 	case VNON:
434 	case VLNK:
435 	case VDIR:
436 	case VREG:
437 	case VBAD:
438 	case VFIFO:
439 	case VSOCK:
440 	default:
441 		return 0;
442 	}
443 
444 	mutex_enter(&specfs_lock);
445 	if (sn->sn_gone) {
446 		if (error == 0)
447 			error = EBADF;
448 	} else if (error != 0) {
449 		sd->sd_opencnt--;
450 		sn->sn_opencnt--;
451 		if (vp->v_type == VBLK)
452 			sd->sd_bdevvp = NULL;
453 
454 	}
455 	mutex_exit(&specfs_lock);
456 
457 	if (cdev_type(dev) != D_DISK || error != 0)
458 		return error;
459 
460 	if (vp->v_type == VCHR)
461 		error = cdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp);
462 	else
463 		error = bdev_ioctl(vp->v_rdev, DIOCGPART, &pi, FREAD, curlwp);
464 	if (error == 0)
465 		uvm_vnp_setsize(vp,
466 		    (voff_t)pi.disklab->d_secsize * pi.part->p_size);
467 	return 0;
468 }
469 
470 /*
471  * Vnode op for read
472  */
473 /* ARGSUSED */
474 int
475 spec_read(void *v)
476 {
477 	struct vop_read_args /* {
478 		struct vnode *a_vp;
479 		struct uio *a_uio;
480 		int  a_ioflag;
481 		kauth_cred_t a_cred;
482 	} */ *ap = v;
483 	struct vnode *vp = ap->a_vp;
484 	struct uio *uio = ap->a_uio;
485  	struct lwp *l = curlwp;
486 	struct buf *bp;
487 	daddr_t bn;
488 	int bsize, bscale;
489 	struct partinfo dpart;
490 	int n, on;
491 	int error = 0;
492 
493 #ifdef DIAGNOSTIC
494 	if (uio->uio_rw != UIO_READ)
495 		panic("spec_read mode");
496 	if (&uio->uio_vmspace->vm_map != kernel_map &&
497 	    uio->uio_vmspace != curproc->p_vmspace)
498 		panic("spec_read proc");
499 #endif
500 	if (uio->uio_resid == 0)
501 		return (0);
502 
503 	switch (vp->v_type) {
504 
505 	case VCHR:
506 		VOP_UNLOCK(vp, 0);
507 		error = cdev_read(vp->v_rdev, uio, ap->a_ioflag);
508 		vn_lock(vp, LK_SHARED | LK_RETRY);
509 		return (error);
510 
511 	case VBLK:
512 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
513 		if (uio->uio_offset < 0)
514 			return (EINVAL);
515 		bsize = BLKDEV_IOSIZE;
516 		if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) {
517 			if (dpart.part->p_fstype == FS_BSDFFS &&
518 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
519 				bsize = dpart.part->p_frag *
520 				    dpart.part->p_fsize;
521 		}
522 		bscale = bsize >> DEV_BSHIFT;
523 		do {
524 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
525 			on = uio->uio_offset % bsize;
526 			n = min((unsigned)(bsize - on), uio->uio_resid);
527 			error = bread(vp, bn, bsize, NOCRED, 0, &bp);
528 			n = min(n, bsize - bp->b_resid);
529 			if (error) {
530 				brelse(bp, 0);
531 				return (error);
532 			}
533 			error = uiomove((char *)bp->b_data + on, n, uio);
534 			brelse(bp, 0);
535 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
536 		return (error);
537 
538 	default:
539 		panic("spec_read type");
540 	}
541 	/* NOTREACHED */
542 }
543 
544 /*
545  * Vnode op for write
546  */
547 /* ARGSUSED */
548 int
549 spec_write(void *v)
550 {
551 	struct vop_write_args /* {
552 		struct vnode *a_vp;
553 		struct uio *a_uio;
554 		int  a_ioflag;
555 		kauth_cred_t a_cred;
556 	} */ *ap = v;
557 	struct vnode *vp = ap->a_vp;
558 	struct uio *uio = ap->a_uio;
559 	struct lwp *l = curlwp;
560 	struct buf *bp;
561 	daddr_t bn;
562 	int bsize, bscale;
563 	struct partinfo dpart;
564 	int n, on;
565 	int error = 0;
566 
567 #ifdef DIAGNOSTIC
568 	if (uio->uio_rw != UIO_WRITE)
569 		panic("spec_write mode");
570 	if (&uio->uio_vmspace->vm_map != kernel_map &&
571 	    uio->uio_vmspace != curproc->p_vmspace)
572 		panic("spec_write proc");
573 #endif
574 
575 	switch (vp->v_type) {
576 
577 	case VCHR:
578 		VOP_UNLOCK(vp, 0);
579 		error = cdev_write(vp->v_rdev, uio, ap->a_ioflag);
580 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
581 		return (error);
582 
583 	case VBLK:
584 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
585 		if (uio->uio_resid == 0)
586 			return (0);
587 		if (uio->uio_offset < 0)
588 			return (EINVAL);
589 		bsize = BLKDEV_IOSIZE;
590 		if (bdev_ioctl(vp->v_rdev, DIOCGPART, &dpart, FREAD, l) == 0) {
591 			if (dpart.part->p_fstype == FS_BSDFFS &&
592 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
593 				bsize = dpart.part->p_frag *
594 				    dpart.part->p_fsize;
595 		}
596 		bscale = bsize >> DEV_BSHIFT;
597 		do {
598 			bn = (uio->uio_offset >> DEV_BSHIFT) &~ (bscale - 1);
599 			on = uio->uio_offset % bsize;
600 			n = min((unsigned)(bsize - on), uio->uio_resid);
601 			if (n == bsize)
602 				bp = getblk(vp, bn, bsize, 0, 0);
603 			else
604 				error = bread(vp, bn, bsize, NOCRED,
605 				    B_MODIFY, &bp);
606 			if (error) {
607 				brelse(bp, 0);
608 				return (error);
609 			}
610 			n = min(n, bsize - bp->b_resid);
611 			error = uiomove((char *)bp->b_data + on, n, uio);
612 			if (error)
613 				brelse(bp, 0);
614 			else {
615 				if (n + on == bsize)
616 					bawrite(bp);
617 				else
618 					bdwrite(bp);
619 				error = bp->b_error;
620 			}
621 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
622 		return (error);
623 
624 	default:
625 		panic("spec_write type");
626 	}
627 	/* NOTREACHED */
628 }
629 
630 /*
631  * Device ioctl operation.
632  */
633 /* ARGSUSED */
634 int
635 spec_ioctl(void *v)
636 {
637 	struct vop_ioctl_args /* {
638 		struct vnode *a_vp;
639 		u_long a_command;
640 		void  *a_data;
641 		int  a_fflag;
642 		kauth_cred_t a_cred;
643 	} */ *ap = v;
644 	struct vnode *vp;
645 	dev_t dev;
646 
647 	/*
648 	 * Extract all the info we need from the vnode, taking care to
649 	 * avoid a race with VOP_REVOKE().
650 	 */
651 
652 	vp = ap->a_vp;
653 	dev = NODEV;
654 	mutex_enter(&vp->v_interlock);
655 	if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) {
656 		dev = vp->v_rdev;
657 	}
658 	mutex_exit(&vp->v_interlock);
659 	if (dev == NODEV) {
660 		return ENXIO;
661 	}
662 
663 	switch (vp->v_type) {
664 
665 	case VCHR:
666 		return cdev_ioctl(dev, ap->a_command, ap->a_data,
667 		    ap->a_fflag, curlwp);
668 
669 	case VBLK:
670 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
671 		return bdev_ioctl(dev, ap->a_command, ap->a_data,
672 		   ap->a_fflag, curlwp);
673 
674 	default:
675 		panic("spec_ioctl");
676 		/* NOTREACHED */
677 	}
678 }
679 
680 /* ARGSUSED */
681 int
682 spec_poll(void *v)
683 {
684 	struct vop_poll_args /* {
685 		struct vnode *a_vp;
686 		int a_events;
687 	} */ *ap = v;
688 	struct vnode *vp;
689 	dev_t dev;
690 
691 	/*
692 	 * Extract all the info we need from the vnode, taking care to
693 	 * avoid a race with VOP_REVOKE().
694 	 */
695 
696 	vp = ap->a_vp;
697 	dev = NODEV;
698 	mutex_enter(&vp->v_interlock);
699 	if ((vp->v_iflag & VI_XLOCK) == 0 && vp->v_specnode) {
700 		dev = vp->v_rdev;
701 	}
702 	mutex_exit(&vp->v_interlock);
703 	if (dev == NODEV) {
704 		return POLLERR;
705 	}
706 
707 	switch (vp->v_type) {
708 
709 	case VCHR:
710 		return cdev_poll(dev, ap->a_events, curlwp);
711 
712 	default:
713 		return (genfs_poll(v));
714 	}
715 }
716 
717 /* ARGSUSED */
718 int
719 spec_kqfilter(void *v)
720 {
721 	struct vop_kqfilter_args /* {
722 		struct vnode	*a_vp;
723 		struct proc	*a_kn;
724 	} */ *ap = v;
725 	dev_t dev;
726 
727 	switch (ap->a_vp->v_type) {
728 
729 	case VCHR:
730 		dev = ap->a_vp->v_rdev;
731 		return cdev_kqfilter(dev, ap->a_kn);
732 	default:
733 		/*
734 		 * Block devices don't support kqfilter, and refuse it
735 		 * for any other files (like those vflush()ed) too.
736 		 */
737 		return (EOPNOTSUPP);
738 	}
739 }
740 
741 /*
742  * Allow mapping of only D_DISK.  This is called only for VBLK.
743  */
744 int
745 spec_mmap(void *v)
746 {
747 	struct vop_mmap_args /* {
748 		struct vnode *a_vp;
749 		vm_prot_t a_prot;
750 		kauth_cred_t a_cred;
751 	} */ *ap = v;
752 	struct vnode *vp = ap->a_vp;
753 
754 	KASSERT(vp->v_type == VBLK);
755 	if (bdev_type(vp->v_rdev) != D_DISK)
756 		return EINVAL;
757 
758 	return 0;
759 }
760 
761 /*
762  * Synch buffers associated with a block device
763  */
764 /* ARGSUSED */
765 int
766 spec_fsync(void *v)
767 {
768 	struct vop_fsync_args /* {
769 		struct vnode *a_vp;
770 		kauth_cred_t a_cred;
771 		int  a_flags;
772 		off_t offlo;
773 		off_t offhi;
774 	} */ *ap = v;
775 	struct vnode *vp = ap->a_vp;
776 	struct mount *mp;
777 	int error;
778 
779 	if (vp->v_type == VBLK) {
780 		if ((mp = vp->v_specmountpoint) != NULL) {
781 			error = VFS_FSYNC(mp, vp, ap->a_flags | FSYNC_VFS);
782 			if (error != EOPNOTSUPP)
783 				return error;
784 		}
785 		vflushbuf(vp, (ap->a_flags & FSYNC_WAIT) != 0);
786 	}
787 	return (0);
788 }
789 
790 /*
791  * Just call the device strategy routine
792  */
793 int
794 spec_strategy(void *v)
795 {
796 	struct vop_strategy_args /* {
797 		struct vnode *a_vp;
798 		struct buf *a_bp;
799 	} */ *ap = v;
800 	struct vnode *vp = ap->a_vp;
801 	struct buf *bp = ap->a_bp;
802 	int error;
803 
804 	KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
805 
806 	error = 0;
807 	bp->b_dev = vp->v_rdev;
808 	if (!(bp->b_flags & B_READ) &&
809 	    (LIST_FIRST(&bp->b_dep)) != NULL && bioopsp)
810 		bioopsp->io_start(bp);
811 
812 	if (!(bp->b_flags & B_READ))
813 		error = fscow_run(bp, false);
814 
815 	if (error) {
816 		bp->b_error = error;
817 		biodone(bp);
818 		return (error);
819 	}
820 
821 	bdev_strategy(bp);
822 
823 	return (0);
824 }
825 
826 int
827 spec_inactive(void *v)
828 {
829 	struct vop_inactive_args /* {
830 		struct vnode *a_vp;
831 		struct proc *a_l;
832 	} */ *ap = v;
833 
834 	VOP_UNLOCK(ap->a_vp, 0);
835 	return (0);
836 }
837 
838 /*
839  * This is a noop, simply returning what one has been given.
840  */
841 int
842 spec_bmap(void *v)
843 {
844 	struct vop_bmap_args /* {
845 		struct vnode *a_vp;
846 		daddr_t  a_bn;
847 		struct vnode **a_vpp;
848 		daddr_t *a_bnp;
849 		int *a_runp;
850 	} */ *ap = v;
851 
852 	if (ap->a_vpp != NULL)
853 		*ap->a_vpp = ap->a_vp;
854 	if (ap->a_bnp != NULL)
855 		*ap->a_bnp = ap->a_bn;
856 	if (ap->a_runp != NULL)
857 		*ap->a_runp = (MAXBSIZE >> DEV_BSHIFT) - 1;
858 	return (0);
859 }
860 
861 /*
862  * Device close routine
863  */
864 /* ARGSUSED */
865 int
866 spec_close(void *v)
867 {
868 	struct vop_close_args /* {
869 		struct vnode *a_vp;
870 		int  a_fflag;
871 		kauth_cred_t a_cred;
872 	} */ *ap = v;
873 	struct vnode *vp = ap->a_vp;
874 	struct session *sess;
875 	dev_t dev = vp->v_rdev;
876 	int mode, error, flags, flags1, count;
877 	specnode_t *sn;
878 	specdev_t *sd;
879 
880 	flags = vp->v_iflag;
881 	sn = vp->v_specnode;
882 	sd = sn->sn_dev;
883 
884 	switch (vp->v_type) {
885 
886 	case VCHR:
887 		/*
888 		 * Hack: a tty device that is a controlling terminal
889 		 * has a reference from the session structure.  We
890 		 * cannot easily tell that a character device is a
891 		 * controlling terminal, unless it is the closing
892 		 * process' controlling terminal.  In that case, if the
893 		 * open count is 1 release the reference from the
894 		 * session.  Also, remove the link from the tty back to
895 		 * the session and pgrp.
896 		 *
897 		 * XXX V. fishy.
898 		 */
899 		mutex_enter(proc_lock);
900 		sess = curlwp->l_proc->p_session;
901 		if (sn->sn_opencnt == 1 && vp == sess->s_ttyvp) {
902 			mutex_spin_enter(&tty_lock);
903 			sess->s_ttyvp = NULL;
904 			if (sess->s_ttyp->t_session != NULL) {
905 				sess->s_ttyp->t_pgrp = NULL;
906 				sess->s_ttyp->t_session = NULL;
907 				mutex_spin_exit(&tty_lock);
908 				SESSRELE(sess);
909 				mutex_exit(proc_lock);
910 			} else {
911 				mutex_spin_exit(&tty_lock);
912 				if (sess->s_ttyp->t_pgrp != NULL)
913 					panic("spec_close: spurious pgrp ref");
914 				mutex_exit(proc_lock);
915 			}
916 			vrele(vp);
917 		} else
918 			mutex_exit(proc_lock);
919 
920 		/*
921 		 * If the vnode is locked, then we are in the midst
922 		 * of forcably closing the device, otherwise we only
923 		 * close on last reference.
924 		 */
925 		mode = S_IFCHR;
926 		break;
927 
928 	case VBLK:
929 		KASSERT(vp == vp->v_specnode->sn_dev->sd_bdevvp);
930 		/*
931 		 * On last close of a block device (that isn't mounted)
932 		 * we must invalidate any in core blocks, so that
933 		 * we can, for instance, change floppy disks.
934 		 */
935 		error = vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 0, 0);
936 		if (error)
937 			return (error);
938 		/*
939 		 * We do not want to really close the device if it
940 		 * is still in use unless we are trying to close it
941 		 * forcibly. Since every use (buffer, vnode, swap, cmap)
942 		 * holds a reference to the vnode, and because we mark
943 		 * any other vnodes that alias this device, when the
944 		 * sum of the reference counts on all the aliased
945 		 * vnodes descends to one, we are on last close.
946 		 */
947 		mode = S_IFBLK;
948 		break;
949 
950 	default:
951 		panic("spec_close: not special");
952 	}
953 
954 	mutex_enter(&specfs_lock);
955 	sn->sn_opencnt--;
956 	count = --sd->sd_opencnt;
957 	if (vp->v_type == VBLK)
958 		sd->sd_bdevvp = NULL;
959 	mutex_exit(&specfs_lock);
960 
961 	if (count != 0)
962 		return 0;
963 
964 	flags1 = ap->a_fflag;
965 
966 	/*
967 	 * if VI_XLOCK is set, then we're going away soon, so make this
968 	 * non-blocking. Also ensures that we won't wedge in vn_lock below.
969 	 */
970 	if (flags & VI_XLOCK)
971 		flags1 |= FNONBLOCK;
972 
973 	/*
974 	 * If we're able to block, release the vnode lock & reacquire. We
975 	 * might end up sleeping for someone else who wants our queues. They
976 	 * won't get them if we hold the vnode locked. Also, if VI_XLOCK is
977 	 * set, don't release the lock as we won't be able to regain it.
978 	 */
979 	if (!(flags1 & FNONBLOCK))
980 		VOP_UNLOCK(vp, 0);
981 
982 	if (vp->v_type == VBLK)
983 		error = bdev_close(dev, flags1, mode, curlwp);
984 	else
985 		error = cdev_close(dev, flags1, mode, curlwp);
986 
987 	if (!(flags1 & FNONBLOCK))
988 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
989 
990 	return (error);
991 }
992 
993 /*
994  * Print out the contents of a special device vnode.
995  */
996 int
997 spec_print(void *v)
998 {
999 	struct vop_print_args /* {
1000 		struct vnode *a_vp;
1001 	} */ *ap = v;
1002 
1003 	printf("dev %d, %d\n", major(ap->a_vp->v_rdev),
1004 	    minor(ap->a_vp->v_rdev));
1005 	return 0;
1006 }
1007 
1008 /*
1009  * Return POSIX pathconf information applicable to special devices.
1010  */
1011 int
1012 spec_pathconf(void *v)
1013 {
1014 	struct vop_pathconf_args /* {
1015 		struct vnode *a_vp;
1016 		int a_name;
1017 		register_t *a_retval;
1018 	} */ *ap = v;
1019 
1020 	switch (ap->a_name) {
1021 	case _PC_LINK_MAX:
1022 		*ap->a_retval = LINK_MAX;
1023 		return (0);
1024 	case _PC_MAX_CANON:
1025 		*ap->a_retval = MAX_CANON;
1026 		return (0);
1027 	case _PC_MAX_INPUT:
1028 		*ap->a_retval = MAX_INPUT;
1029 		return (0);
1030 	case _PC_PIPE_BUF:
1031 		*ap->a_retval = PIPE_BUF;
1032 		return (0);
1033 	case _PC_CHOWN_RESTRICTED:
1034 		*ap->a_retval = 1;
1035 		return (0);
1036 	case _PC_VDISABLE:
1037 		*ap->a_retval = _POSIX_VDISABLE;
1038 		return (0);
1039 	case _PC_SYNC_IO:
1040 		*ap->a_retval = 1;
1041 		return (0);
1042 	default:
1043 		return (EINVAL);
1044 	}
1045 	/* NOTREACHED */
1046 }
1047 
1048 /*
1049  * Advisory record locking support.
1050  */
1051 int
1052 spec_advlock(void *v)
1053 {
1054 	struct vop_advlock_args /* {
1055 		struct vnode *a_vp;
1056 		void *a_id;
1057 		int a_op;
1058 		struct flock *a_fl;
1059 		int a_flags;
1060 	} */ *ap = v;
1061 	struct vnode *vp = ap->a_vp;
1062 
1063 	return lf_advlock(ap, &vp->v_speclockf, (off_t)0);
1064 }
1065