xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision 2a1ad637466621af45d5a17185b33f3dcaaa1b1c)
1 /*
2  * (MPSAFE)
3  *
4  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Alex Hornung <ahornung@gmail.com>
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in
17  *    the documentation and/or other materials provided with the
18  *    distribution.
19  * 3. Neither the name of The DragonFly Project 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 COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/time.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/fcntl.h>
42 #include <sys/proc.h>
43 #include <sys/priv.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
46 #include <sys/uio.h>
47 #include <sys/mount.h>
48 #include <sys/file.h>
49 #include <sys/namei.h>
50 #include <sys/dirent.h>
51 #include <sys/malloc.h>
52 #include <sys/stat.h>
53 #include <sys/reg.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_zone.h>
56 #include <vm/vm_object.h>
57 #include <sys/filio.h>
58 #include <sys/ttycom.h>
59 #include <sys/tty.h>
60 #include <sys/diskslice.h>
61 #include <sys/sysctl.h>
62 #include <sys/devfs.h>
63 #include <sys/pioctl.h>
64 #include <vfs/fifofs/fifo.h>
65 
66 #include <machine/limits.h>
67 
68 #include <sys/buf2.h>
69 #include <sys/sysref2.h>
70 #include <sys/mplock2.h>
71 #include <vm/vm_page2.h>
72 
73 #ifndef SPEC_CHAIN_DEBUG
74 #define SPEC_CHAIN_DEBUG 0
75 #endif
76 
77 MALLOC_DECLARE(M_DEVFS);
78 #define DEVFS_BADOP	(void *)devfs_vop_badop
79 
80 static int devfs_vop_badop(struct vop_generic_args *);
81 static int devfs_vop_access(struct vop_access_args *);
82 static int devfs_vop_inactive(struct vop_inactive_args *);
83 static int devfs_vop_reclaim(struct vop_reclaim_args *);
84 static int devfs_vop_readdir(struct vop_readdir_args *);
85 static int devfs_vop_getattr(struct vop_getattr_args *);
86 static int devfs_vop_setattr(struct vop_setattr_args *);
87 static int devfs_vop_readlink(struct vop_readlink_args *);
88 static int devfs_vop_print(struct vop_print_args *);
89 
90 static int devfs_vop_nresolve(struct vop_nresolve_args *);
91 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
92 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
93 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
94 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
95 static int devfs_vop_nremove(struct vop_nremove_args *);
96 
97 static int devfs_spec_open(struct vop_open_args *);
98 static int devfs_spec_close(struct vop_close_args *);
99 static int devfs_spec_fsync(struct vop_fsync_args *);
100 
101 static int devfs_spec_read(struct vop_read_args *);
102 static int devfs_spec_write(struct vop_write_args *);
103 static int devfs_spec_ioctl(struct vop_ioctl_args *);
104 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
105 static int devfs_spec_strategy(struct vop_strategy_args *);
106 static void devfs_spec_strategy_done(struct bio *);
107 static int devfs_spec_freeblks(struct vop_freeblks_args *);
108 static int devfs_spec_bmap(struct vop_bmap_args *);
109 static int devfs_spec_advlock(struct vop_advlock_args *);
110 static void devfs_spec_getpages_iodone(struct bio *);
111 static int devfs_spec_getpages(struct vop_getpages_args *);
112 
113 static int devfs_fo_close(struct file *);
114 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
115 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
116 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
117 static int devfs_fo_kqfilter(struct file *, struct knote *);
118 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
119 				struct ucred *, struct sysmsg *);
120 static __inline int sequential_heuristic(struct uio *, struct file *);
121 
122 extern struct lock devfs_lock;
123 
124 /*
125  * devfs vnode operations for regular files.  All vnode ops are MPSAFE.
126  */
127 struct vop_ops devfs_vnode_norm_vops = {
128 	.vop_default =		vop_defaultop,
129 	.vop_access =		devfs_vop_access,
130 	.vop_advlock =		DEVFS_BADOP,
131 	.vop_bmap =		DEVFS_BADOP,
132 	.vop_close =		vop_stdclose,
133 	.vop_getattr =		devfs_vop_getattr,
134 	.vop_inactive =		devfs_vop_inactive,
135 	.vop_ncreate =		DEVFS_BADOP,
136 	.vop_nresolve =		devfs_vop_nresolve,
137 	.vop_nlookupdotdot =	devfs_vop_nlookupdotdot,
138 	.vop_nlink =		DEVFS_BADOP,
139 	.vop_nmkdir =		devfs_vop_nmkdir,
140 	.vop_nmknod =		DEVFS_BADOP,
141 	.vop_nremove =		devfs_vop_nremove,
142 	.vop_nrename =		DEVFS_BADOP,
143 	.vop_nrmdir =		devfs_vop_nrmdir,
144 	.vop_nsymlink =		devfs_vop_nsymlink,
145 	.vop_open =		vop_stdopen,
146 	.vop_pathconf =		vop_stdpathconf,
147 	.vop_print =		devfs_vop_print,
148 	.vop_read =		DEVFS_BADOP,
149 	.vop_readdir =		devfs_vop_readdir,
150 	.vop_readlink =		devfs_vop_readlink,
151 	.vop_reclaim =		devfs_vop_reclaim,
152 	.vop_setattr =		devfs_vop_setattr,
153 	.vop_write =		DEVFS_BADOP,
154 	.vop_ioctl =		DEVFS_BADOP
155 };
156 
157 /*
158  * devfs vnode operations for character devices.  All vnode ops are MPSAFE.
159  */
160 struct vop_ops devfs_vnode_dev_vops = {
161 	.vop_default =		vop_defaultop,
162 	.vop_access =		devfs_vop_access,
163 	.vop_advlock =		devfs_spec_advlock,
164 	.vop_bmap =		devfs_spec_bmap,
165 	.vop_close =		devfs_spec_close,
166 	.vop_freeblks =		devfs_spec_freeblks,
167 	.vop_fsync =		devfs_spec_fsync,
168 	.vop_getattr =		devfs_vop_getattr,
169 	.vop_getpages =		devfs_spec_getpages,
170 	.vop_inactive =		devfs_vop_inactive,
171 	.vop_open =		devfs_spec_open,
172 	.vop_pathconf =		vop_stdpathconf,
173 	.vop_print =		devfs_vop_print,
174 	.vop_kqfilter =		devfs_spec_kqfilter,
175 	.vop_read =		devfs_spec_read,
176 	.vop_readdir =		DEVFS_BADOP,
177 	.vop_readlink =		DEVFS_BADOP,
178 	.vop_reclaim =		devfs_vop_reclaim,
179 	.vop_setattr =		devfs_vop_setattr,
180 	.vop_strategy =		devfs_spec_strategy,
181 	.vop_write =		devfs_spec_write,
182 	.vop_ioctl =		devfs_spec_ioctl
183 };
184 
185 /*
186  * devfs file pointer operations.  All fileops are MPSAFE.
187  */
188 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
189 
190 struct fileops devfs_dev_fileops = {
191 	.fo_read	= devfs_fo_read,
192 	.fo_write	= devfs_fo_write,
193 	.fo_ioctl	= devfs_fo_ioctl,
194 	.fo_kqfilter	= devfs_fo_kqfilter,
195 	.fo_stat	= devfs_fo_stat,
196 	.fo_close	= devfs_fo_close,
197 	.fo_shutdown	= nofo_shutdown
198 };
199 
200 /*
201  * These two functions are possibly temporary hacks for devices (aka
202  * the pty code) which want to control the node attributes themselves.
203  *
204  * XXX we may ultimately desire to simply remove the uid/gid/mode
205  * from the node entirely.
206  *
207  * MPSAFE - sorta.  Theoretically the overwrite can compete since they
208  *	    are loading from the same fields.
209  */
210 static __inline void
211 node_sync_dev_get(struct devfs_node *node)
212 {
213 	cdev_t dev;
214 
215 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
216 		node->uid = dev->si_uid;
217 		node->gid = dev->si_gid;
218 		node->mode = dev->si_perms;
219 	}
220 }
221 
222 static __inline void
223 node_sync_dev_set(struct devfs_node *node)
224 {
225 	cdev_t dev;
226 
227 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
228 		dev->si_uid = node->uid;
229 		dev->si_gid = node->gid;
230 		dev->si_perms = node->mode;
231 	}
232 }
233 
234 /*
235  * generic entry point for unsupported operations
236  */
237 static int
238 devfs_vop_badop(struct vop_generic_args *ap)
239 {
240 	return (EIO);
241 }
242 
243 
244 static int
245 devfs_vop_access(struct vop_access_args *ap)
246 {
247 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
248 	int error;
249 
250 	if (!devfs_node_is_accessible(node))
251 		return ENOENT;
252 	node_sync_dev_get(node);
253 	error = vop_helper_access(ap, node->uid, node->gid,
254 				  node->mode, node->flags);
255 
256 	return error;
257 }
258 
259 
260 static int
261 devfs_vop_inactive(struct vop_inactive_args *ap)
262 {
263 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
264 
265 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
266 		vrecycle(ap->a_vp);
267 	return 0;
268 }
269 
270 
271 static int
272 devfs_vop_reclaim(struct vop_reclaim_args *ap)
273 {
274 	struct devfs_node *node;
275 	struct vnode *vp;
276 	int locked;
277 
278 	/*
279 	 * Check if it is locked already. if not, we acquire the devfs lock
280 	 */
281 	if ((lockstatus(&devfs_lock, curthread)) != LK_EXCLUSIVE) {
282 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
283 		locked = 1;
284 	} else {
285 		locked = 0;
286 	}
287 
288 	/*
289 	 * Get rid of the devfs_node if it is no longer linked into the
290 	 * topology.
291 	 */
292 	vp = ap->a_vp;
293 	if ((node = DEVFS_NODE(vp)) != NULL) {
294 		node->v_node = NULL;
295 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
296 			devfs_freep(node);
297 	}
298 
299 	if (locked)
300 		lockmgr(&devfs_lock, LK_RELEASE);
301 
302 	/*
303 	 * v_rdev needs to be properly released using v_release_rdev
304 	 * Make sure v_data is NULL as well.
305 	 */
306 	vp->v_data = NULL;
307 	v_release_rdev(vp);
308 	return 0;
309 }
310 
311 
312 static int
313 devfs_vop_readdir(struct vop_readdir_args *ap)
314 {
315 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
316 	struct devfs_node *node;
317 	int cookie_index;
318 	int ncookies;
319 	int error2;
320 	int error;
321 	int r;
322 	off_t *cookies;
323 	off_t saveoff;
324 
325 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
326 
327 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
328 		return (EINVAL);
329 	error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
330 	if (error)
331 		return (error);
332 
333 	if (!devfs_node_is_accessible(dnode)) {
334 		vn_unlock(ap->a_vp);
335 		return ENOENT;
336 	}
337 
338 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
339 
340 	saveoff = ap->a_uio->uio_offset;
341 
342 	if (ap->a_ncookies) {
343 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
344 		if (ncookies > 256)
345 			ncookies = 256;
346 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
347 		cookie_index = 0;
348 	} else {
349 		ncookies = -1;
350 		cookies = NULL;
351 		cookie_index = 0;
352 	}
353 
354 	nanotime(&dnode->atime);
355 
356 	if (saveoff == 0) {
357 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
358 				     DT_DIR, 1, ".");
359 		if (r)
360 			goto done;
361 		if (cookies)
362 			cookies[cookie_index] = saveoff;
363 		saveoff++;
364 		cookie_index++;
365 		if (cookie_index == ncookies)
366 			goto done;
367 	}
368 
369 	if (saveoff == 1) {
370 		if (dnode->parent) {
371 			r = vop_write_dirent(&error, ap->a_uio,
372 					     dnode->parent->d_dir.d_ino,
373 					     DT_DIR, 2, "..");
374 		} else {
375 			r = vop_write_dirent(&error, ap->a_uio,
376 					     dnode->d_dir.d_ino,
377 					     DT_DIR, 2, "..");
378 		}
379 		if (r)
380 			goto done;
381 		if (cookies)
382 			cookies[cookie_index] = saveoff;
383 		saveoff++;
384 		cookie_index++;
385 		if (cookie_index == ncookies)
386 			goto done;
387 	}
388 
389 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
390 		if ((node->flags & DEVFS_HIDDEN) ||
391 		    (node->flags & DEVFS_INVISIBLE)) {
392 			continue;
393 		}
394 
395 		/*
396 		 * If the node type is a valid devfs alias, then we make
397 		 * sure that the target isn't hidden. If it is, we don't
398 		 * show the link in the directory listing.
399 		 */
400 		if ((node->node_type == Nlink) && (node->link_target != NULL) &&
401 			(node->link_target->flags & DEVFS_HIDDEN))
402 			continue;
403 
404 		if (node->cookie < saveoff)
405 			continue;
406 
407 		saveoff = node->cookie;
408 
409 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
410 					  node->d_dir.d_type,
411 					  node->d_dir.d_namlen,
412 					  node->d_dir.d_name);
413 
414 		if (error2)
415 			break;
416 
417 		saveoff++;
418 
419 		if (cookies)
420 			cookies[cookie_index] = node->cookie;
421 		++cookie_index;
422 		if (cookie_index == ncookies)
423 			break;
424 	}
425 
426 done:
427 	lockmgr(&devfs_lock, LK_RELEASE);
428 	vn_unlock(ap->a_vp);
429 
430 	ap->a_uio->uio_offset = saveoff;
431 	if (error && cookie_index == 0) {
432 		if (cookies) {
433 			kfree(cookies, M_TEMP);
434 			*ap->a_ncookies = 0;
435 			*ap->a_cookies = NULL;
436 		}
437 	} else {
438 		if (cookies) {
439 			*ap->a_ncookies = cookie_index;
440 			*ap->a_cookies = cookies;
441 		}
442 	}
443 	return (error);
444 }
445 
446 
447 static int
448 devfs_vop_nresolve(struct vop_nresolve_args *ap)
449 {
450 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
451 	struct devfs_node *node, *found = NULL;
452 	struct namecache *ncp;
453 	struct vnode *vp = NULL;
454 	int error = 0;
455 	int len;
456 	int depth;
457 
458 	ncp = ap->a_nch->ncp;
459 	len = ncp->nc_nlen;
460 
461 	if (!devfs_node_is_accessible(dnode))
462 		return ENOENT;
463 
464 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
465 
466 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) {
467 		error = ENOENT;
468 		cache_setvp(ap->a_nch, NULL);
469 		goto out;
470 	}
471 
472 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
473 		if (len == node->d_dir.d_namlen) {
474 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
475 				found = node;
476 				break;
477 			}
478 		}
479 	}
480 
481 	if (found) {
482 		depth = 0;
483 		while ((found->node_type == Nlink) && (found->link_target)) {
484 			if (depth >= 8) {
485 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
486 				break;
487 			}
488 
489 			found = found->link_target;
490 			++depth;
491 		}
492 
493 		if (!(found->flags & DEVFS_HIDDEN))
494 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
495 	}
496 
497 	if (vp == NULL) {
498 		error = ENOENT;
499 		cache_setvp(ap->a_nch, NULL);
500 		goto out;
501 
502 	}
503 	KKASSERT(vp);
504 	vn_unlock(vp);
505 	cache_setvp(ap->a_nch, vp);
506 	vrele(vp);
507 out:
508 	lockmgr(&devfs_lock, LK_RELEASE);
509 
510 	return error;
511 }
512 
513 
514 static int
515 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
516 {
517 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
518 
519 	*ap->a_vpp = NULL;
520 	if (!devfs_node_is_accessible(dnode))
521 		return ENOENT;
522 
523 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
524 	if (dnode->parent != NULL) {
525 		devfs_allocv(ap->a_vpp, dnode->parent);
526 		vn_unlock(*ap->a_vpp);
527 	}
528 	lockmgr(&devfs_lock, LK_RELEASE);
529 
530 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
531 }
532 
533 
534 static int
535 devfs_vop_getattr(struct vop_getattr_args *ap)
536 {
537 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
538 	struct vattr *vap = ap->a_vap;
539 	struct partinfo pinfo;
540 	int error = 0;
541 
542 #if 0
543 	if (!devfs_node_is_accessible(node))
544 		return ENOENT;
545 #endif
546 	node_sync_dev_get(node);
547 
548 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
549 
550 	/* start by zeroing out the attributes */
551 	VATTR_NULL(vap);
552 
553 	/* next do all the common fields */
554 	vap->va_type = ap->a_vp->v_type;
555 	vap->va_mode = node->mode;
556 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
557 	vap->va_flags = 0;
558 	vap->va_blocksize = DEV_BSIZE;
559 	vap->va_bytes = vap->va_size = 0;
560 
561 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
562 
563 	vap->va_atime = node->atime;
564 	vap->va_mtime = node->mtime;
565 	vap->va_ctime = node->ctime;
566 
567 	vap->va_nlink = 1; /* number of references to file */
568 
569 	vap->va_uid = node->uid;
570 	vap->va_gid = node->gid;
571 
572 	vap->va_rmajor = 0;
573 	vap->va_rminor = 0;
574 
575 	if ((node->node_type == Ndev) && node->d_dev)  {
576 		reference_dev(node->d_dev);
577 		vap->va_rminor = node->d_dev->si_uminor;
578 		release_dev(node->d_dev);
579 	}
580 
581 	/* For a softlink the va_size is the length of the softlink */
582 	if (node->symlink_name != 0) {
583 		vap->va_bytes = vap->va_size = node->symlink_namelen;
584 	}
585 
586 	/*
587 	 * For a disk-type device, va_size is the size of the underlying
588 	 * device, so that lseek() works properly.
589 	 */
590 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
591 		bzero(&pinfo, sizeof(pinfo));
592 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
593 				   0, proc0.p_ucred, NULL, NULL);
594 		if ((error == 0) && (pinfo.media_blksize != 0)) {
595 			vap->va_size = pinfo.media_size;
596 		} else {
597 			vap->va_size = 0;
598 			error = 0;
599 		}
600 	}
601 
602 	lockmgr(&devfs_lock, LK_RELEASE);
603 
604 	return (error);
605 }
606 
607 
608 static int
609 devfs_vop_setattr(struct vop_setattr_args *ap)
610 {
611 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
612 	struct vattr *vap;
613 	uid_t cur_uid;
614 	gid_t cur_gid;
615 	mode_t cur_mode;
616 	int error = 0;
617 
618 	if (!devfs_node_is_accessible(node))
619 		return ENOENT;
620 	node_sync_dev_get(node);
621 
622 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
623 
624 	vap = ap->a_vap;
625 
626 	if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
627 		cur_uid = node->uid;
628 		cur_gid = node->gid;
629 		cur_mode = node->mode;
630 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
631 		    ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
632 		if (error)
633 			goto out;
634 
635 		if (node->uid != cur_uid || node->gid != cur_gid) {
636 			node->uid = cur_uid;
637 			node->gid = cur_gid;
638 			node->mode = cur_mode;
639 		}
640 	}
641 
642 	if (vap->va_mode != (mode_t)VNOVAL) {
643 		cur_mode = node->mode;
644 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
645 		    node->uid, node->gid, &cur_mode);
646 		if (error == 0 && node->mode != cur_mode) {
647 			node->mode = cur_mode;
648 		}
649 	}
650 
651 out:
652 	node_sync_dev_set(node);
653 	nanotime(&node->ctime);
654 	lockmgr(&devfs_lock, LK_RELEASE);
655 
656 	return error;
657 }
658 
659 
660 static int
661 devfs_vop_readlink(struct vop_readlink_args *ap)
662 {
663 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
664 	int ret;
665 
666 	if (!devfs_node_is_accessible(node))
667 		return ENOENT;
668 
669 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
670 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
671 	lockmgr(&devfs_lock, LK_RELEASE);
672 
673 	return ret;
674 }
675 
676 
677 static int
678 devfs_vop_print(struct vop_print_args *ap)
679 {
680 	return (0);
681 }
682 
683 static int
684 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
685 {
686 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
687 	struct devfs_node *node;
688 
689 	if (!devfs_node_is_accessible(dnode))
690 		return ENOENT;
691 
692 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
693 		goto out;
694 
695 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
696 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir,
697 		      ap->a_nch->ncp->nc_name, dnode, NULL);
698 
699 	if (*ap->a_vpp) {
700 		node = DEVFS_NODE(*ap->a_vpp);
701 		node->flags |= DEVFS_USER_CREATED;
702 		cache_setunresolved(ap->a_nch);
703 		cache_setvp(ap->a_nch, *ap->a_vpp);
704 	}
705 	lockmgr(&devfs_lock, LK_RELEASE);
706 out:
707 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
708 }
709 
710 static int
711 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
712 {
713 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
714 	struct devfs_node *node;
715 	size_t targetlen;
716 
717 	if (!devfs_node_is_accessible(dnode))
718 		return ENOENT;
719 
720 	ap->a_vap->va_type = VLNK;
721 
722 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
723 		goto out;
724 
725 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
726 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink,
727 		      ap->a_nch->ncp->nc_name, dnode, NULL);
728 
729 	targetlen = strlen(ap->a_target);
730 	if (*ap->a_vpp) {
731 		node = DEVFS_NODE(*ap->a_vpp);
732 		node->flags |= DEVFS_USER_CREATED;
733 		node->symlink_namelen = targetlen;
734 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
735 		memcpy(node->symlink_name, ap->a_target, targetlen);
736 		node->symlink_name[targetlen] = '\0';
737 		cache_setunresolved(ap->a_nch);
738 		cache_setvp(ap->a_nch, *ap->a_vpp);
739 	}
740 	lockmgr(&devfs_lock, LK_RELEASE);
741 out:
742 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
743 }
744 
745 static int
746 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
747 {
748 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
749 	struct devfs_node *node;
750 	struct namecache *ncp;
751 	int error = ENOENT;
752 
753 	ncp = ap->a_nch->ncp;
754 
755 	if (!devfs_node_is_accessible(dnode))
756 		return ENOENT;
757 
758 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
759 
760 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
761 		goto out;
762 
763 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
764 		if (ncp->nc_nlen != node->d_dir.d_namlen)
765 			continue;
766 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
767 			continue;
768 
769 		/*
770 		 * only allow removal of user created dirs
771 		 */
772 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
773 			error = EPERM;
774 			goto out;
775 		} else if (node->node_type != Ndir) {
776 			error = ENOTDIR;
777 			goto out;
778 		} else if (node->nchildren > 2) {
779 			error = ENOTEMPTY;
780 			goto out;
781 		} else {
782 			if (node->v_node)
783 				cache_inval_vp(node->v_node, CINV_DESTROY);
784 			devfs_unlinkp(node);
785 			error = 0;
786 			break;
787 		}
788 	}
789 
790 	cache_unlink(ap->a_nch);
791 out:
792 	lockmgr(&devfs_lock, LK_RELEASE);
793 	return error;
794 }
795 
796 static int
797 devfs_vop_nremove(struct vop_nremove_args *ap)
798 {
799 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
800 	struct devfs_node *node;
801 	struct namecache *ncp;
802 	int error = ENOENT;
803 
804 	ncp = ap->a_nch->ncp;
805 
806 	if (!devfs_node_is_accessible(dnode))
807 		return ENOENT;
808 
809 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
810 
811 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
812 		goto out;
813 
814 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
815 		if (ncp->nc_nlen != node->d_dir.d_namlen)
816 			continue;
817 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
818 			continue;
819 
820 		/*
821 		 * only allow removal of user created stuff (e.g. symlinks)
822 		 */
823 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
824 			error = EPERM;
825 			goto out;
826 		} else if (node->node_type == Ndir) {
827 			error = EISDIR;
828 			goto out;
829 		} else {
830 			if (node->v_node)
831 				cache_inval_vp(node->v_node, CINV_DESTROY);
832 			devfs_unlinkp(node);
833 			error = 0;
834 			break;
835 		}
836 	}
837 
838 	cache_unlink(ap->a_nch);
839 out:
840 	lockmgr(&devfs_lock, LK_RELEASE);
841 	return error;
842 }
843 
844 
845 static int
846 devfs_spec_open(struct vop_open_args *ap)
847 {
848 	struct vnode *vp = ap->a_vp;
849 	struct vnode *orig_vp = NULL;
850 	struct devfs_node *node = DEVFS_NODE(vp);
851 	struct devfs_node *newnode;
852 	cdev_t dev, ndev = NULL;
853 	int error = 0;
854 
855 	if (node) {
856 		if (node->d_dev == NULL)
857 			return ENXIO;
858 		if (!devfs_node_is_accessible(node))
859 			return ENOENT;
860 	}
861 
862 	if ((dev = vp->v_rdev) == NULL)
863 		return ENXIO;
864 
865 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
866 
867 	if (node && ap->a_fp) {
868 		int exists;
869 
870 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
871 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
872 
873 		ndev = devfs_clone(dev, node->d_dir.d_name,
874 				   node->d_dir.d_namlen,
875 				   ap->a_mode, ap->a_cred);
876 		if (ndev != NULL) {
877 			newnode = devfs_create_device_node(
878 					DEVFS_MNTDATA(vp->v_mount)->root_node,
879 					ndev, &exists, NULL, NULL);
880 			/* XXX: possibly destroy device if this happens */
881 
882 			if (newnode != NULL) {
883 				dev = ndev;
884 				if (exists == 0)
885 					devfs_link_dev(dev);
886 
887 				devfs_debug(DEVFS_DEBUG_DEBUG,
888 						"parent here is: %s, node is: |%s|\n",
889 						((node->parent->node_type == Nroot) ?
890 						"ROOT!" : node->parent->d_dir.d_name),
891 						newnode->d_dir.d_name);
892 				devfs_debug(DEVFS_DEBUG_DEBUG,
893 						"test: %s\n",
894 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
895 
896 				/*
897 				 * orig_vp is set to the original vp if we
898 				 * cloned.
899 				 */
900 				/* node->flags |= DEVFS_CLONED; */
901 				devfs_allocv(&vp, newnode);
902 				orig_vp = ap->a_vp;
903 				ap->a_vp = vp;
904 			}
905 		}
906 		lockmgr(&devfs_lock, LK_RELEASE);
907 		/*
908 		 * Synchronize devfs here to make sure that, if the cloned
909 		 * device creates other device nodes in addition to the
910 		 * cloned one, all of them are created by the time we return
911 		 * from opening the cloned one.
912 		 */
913 		if (ndev)
914 			devfs_config();
915 	}
916 
917 	devfs_debug(DEVFS_DEBUG_DEBUG,
918 		    "devfs_spec_open() called on %s! \n",
919 		    dev->si_name);
920 
921 	/*
922 	 * Make this field valid before any I/O in ->d_open
923 	 */
924 	if (!dev->si_iosize_max)
925 		/* XXX: old DFLTPHYS == 64KB dependency */
926 		dev->si_iosize_max = min(MAXPHYS,64*1024);
927 
928 	if (dev_dflags(dev) & D_TTY)
929 		vsetflags(vp, VISTTY);
930 
931 	/*
932 	 * Open underlying device
933 	 */
934 	vn_unlock(vp);
935 	error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred, ap->a_fp);
936 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
937 
938 	/*
939 	 * Clean up any cloned vp if we error out.
940 	 */
941 	if (error) {
942 		if (orig_vp) {
943 			vput(vp);
944 			ap->a_vp = orig_vp;
945 			/* orig_vp = NULL; */
946 		}
947 		return error;
948 	}
949 
950 	/*
951 	 * This checks if the disk device is going to be opened for writing.
952 	 * It will be only allowed in the cases where securelevel permits it
953 	 * and it's not mounted R/W.
954 	 */
955 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
956 	    (ap->a_cred != FSCRED)) {
957 
958 		/* Very secure mode. No open for writing allowed */
959 		if (securelevel >= 2)
960 			return EPERM;
961 
962 		/*
963 		 * If it is mounted R/W, do not allow to open for writing.
964 		 * In the case it's mounted read-only but securelevel
965 		 * is >= 1, then do not allow opening for writing either.
966 		 */
967 		if (vfs_mountedon(vp)) {
968 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
969 				return EBUSY;
970 			else if (securelevel >= 1)
971 				return EPERM;
972 		}
973 	}
974 
975 	if (dev_dflags(dev) & D_TTY) {
976 		if (dev->si_tty) {
977 			struct tty *tp;
978 			tp = dev->si_tty;
979 			if (!tp->t_stop) {
980 				devfs_debug(DEVFS_DEBUG_DEBUG,
981 					    "devfs: no t_stop\n");
982 				tp->t_stop = nottystop;
983 			}
984 		}
985 	}
986 
987 
988 	if (vn_isdisk(vp, NULL)) {
989 		if (!dev->si_bsize_phys)
990 			dev->si_bsize_phys = DEV_BSIZE;
991 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
992 	}
993 
994 	vop_stdopen(ap);
995 #if 0
996 	if (node)
997 		nanotime(&node->atime);
998 #endif
999 
1000 	/*
1001 	 * If we replaced the vp the vop_stdopen() call will have loaded
1002 	 * it into fp->f_data and vref()d the vp, giving us two refs.  So
1003 	 * instead of just unlocking it here we have to vput() it.
1004 	 */
1005 	if (orig_vp)
1006 		vput(vp);
1007 
1008 	/* Ugly pty magic, to make pty devices appear once they are opened */
1009 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
1010 		node->flags &= ~DEVFS_INVISIBLE;
1011 
1012 	if (ap->a_fp) {
1013 		KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
1014 		KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1015 		ap->a_fp->f_ops = &devfs_dev_fileops;
1016 		KKASSERT(ap->a_fp->f_data == (void *)vp);
1017 	}
1018 
1019 	return 0;
1020 }
1021 
1022 static int
1023 devfs_spec_close(struct vop_close_args *ap)
1024 {
1025 	struct devfs_node *node;
1026 	struct proc *p = curproc;
1027 	struct vnode *vp = ap->a_vp;
1028 	cdev_t dev = vp->v_rdev;
1029 	int error = 0;
1030 	int needrelock;
1031 	int opencount;
1032 
1033 	/*
1034 	 * We do special tests on the opencount so unfortunately we need
1035 	 * an exclusive lock.
1036 	 */
1037 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
1038 
1039 	if (dev)
1040 		devfs_debug(DEVFS_DEBUG_DEBUG,
1041 			    "devfs_spec_close() called on %s! \n",
1042 			    dev->si_name);
1043 	else
1044 		devfs_debug(DEVFS_DEBUG_DEBUG,
1045 			    "devfs_spec_close() called, null vode!\n");
1046 
1047 	/*
1048 	 * A couple of hacks for devices and tty devices.  The
1049 	 * vnode ref count cannot be used to figure out the
1050 	 * last close, but we can use v_opencount now that
1051 	 * revoke works properly.
1052 	 *
1053 	 * Detect the last close on a controlling terminal and clear
1054 	 * the session (half-close).
1055 	 *
1056 	 * XXX opencount is not SMP safe.  The vnode is locked but there
1057 	 *     may be multiple vnodes referencing the same device.
1058 	 */
1059 	if (dev) {
1060 		/*
1061 		 * NOTE: Try to avoid global tokens when testing opencount
1062 		 * XXX hack, fixme. needs a struct lock and opencount in
1063 		 * struct cdev itself.
1064 		 */
1065 		reference_dev(dev);
1066 		opencount = vp->v_opencount;
1067 		if (opencount <= 1)
1068 			opencount = count_dev(dev);   /* XXX NOT SMP SAFE */
1069 	} else {
1070 		opencount = 0;
1071 	}
1072 
1073 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1074 		p->p_session->s_ttyvp = NULL;
1075 		vrele(vp);
1076 	}
1077 
1078 	/*
1079 	 * Vnodes can be opened and closed multiple times.  Do not really
1080 	 * close the device unless (1) it is being closed forcibly,
1081 	 * (2) the device wants to track closes, or (3) this is the last
1082 	 * vnode doing its last close on the device.
1083 	 *
1084 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1085 	 * a closed device.  This might not occur now that our revoke is
1086 	 * fixed.
1087 	 */
1088 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1089 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1090 	    (dev_dflags(dev) & D_TRACKCLOSE) ||
1091 	    (opencount == 1))) {
1092 		/*
1093 		 * Ugly pty magic, to make pty devices disappear again once
1094 		 * they are closed.
1095 		 */
1096 		node = DEVFS_NODE(ap->a_vp);
1097 		if (node && (node->flags & DEVFS_PTY))
1098 			node->flags |= DEVFS_INVISIBLE;
1099 
1100 		/*
1101 		 * Unlock around dev_dclose(), unless the vnode is
1102 		 * undergoing a vgone/reclaim (during umount).
1103 		 */
1104 		needrelock = 0;
1105 		if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1106 			needrelock = 1;
1107 			vn_unlock(vp);
1108 		}
1109 
1110 		/*
1111 		 * WARNING!  If the device destroys itself the devfs node
1112 		 *	     can disappear here.
1113 		 *
1114 		 * WARNING!  vn_lock() will fail if the vp is in a VRECLAIM,
1115 		 *	     which can occur during umount.
1116 		 */
1117 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1118 		/* node is now stale */
1119 
1120 		if (needrelock) {
1121 			if (vn_lock(vp, LK_EXCLUSIVE |
1122 					LK_RETRY |
1123 					LK_FAILRECLAIM) != 0) {
1124 				panic("devfs_spec_close: vnode %p "
1125 				      "unexpectedly could not be relocked",
1126 				      vp);
1127 			}
1128 		}
1129 	} else {
1130 		error = 0;
1131 	}
1132 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1133 
1134 	/*
1135 	 * Track the actual opens and closes on the vnode.  The last close
1136 	 * disassociates the rdev.  If the rdev is already disassociated or
1137 	 * the opencount is already 0, the vnode might have been revoked
1138 	 * and no further opencount tracking occurs.
1139 	 */
1140 	if (dev)
1141 		release_dev(dev);
1142 	if (vp->v_opencount > 0)
1143 		vop_stdclose(ap);
1144 	return(error);
1145 
1146 }
1147 
1148 
1149 static int
1150 devfs_fo_close(struct file *fp)
1151 {
1152 	struct vnode *vp = (struct vnode *)fp->f_data;
1153 	int error;
1154 
1155 	fp->f_ops = &badfileops;
1156 	error = vn_close(vp, fp->f_flag, fp);
1157 	devfs_clear_cdevpriv(fp);
1158 
1159 	return (error);
1160 }
1161 
1162 
1163 /*
1164  * Device-optimized file table vnode read routine.
1165  *
1166  * This bypasses the VOP table and talks directly to the device.  Most
1167  * filesystems just route to specfs and can make this optimization.
1168  *
1169  * MPALMOSTSAFE - acquires mplock
1170  */
1171 static int
1172 devfs_fo_read(struct file *fp, struct uio *uio,
1173 		 struct ucred *cred, int flags)
1174 {
1175 	struct devfs_node *node;
1176 	struct vnode *vp;
1177 	int ioflag;
1178 	int error;
1179 	cdev_t dev;
1180 
1181 	KASSERT(uio->uio_td == curthread,
1182 		("uio_td %p is not td %p", uio->uio_td, curthread));
1183 
1184 	if (uio->uio_resid == 0)
1185 		return 0;
1186 
1187 	vp = (struct vnode *)fp->f_data;
1188 	if (vp == NULL || vp->v_type == VBAD)
1189 		return EBADF;
1190 
1191 	node = DEVFS_NODE(vp);
1192 
1193 	if ((dev = vp->v_rdev) == NULL)
1194 		return EBADF;
1195 
1196 	reference_dev(dev);
1197 
1198 	if ((flags & O_FOFFSET) == 0)
1199 		uio->uio_offset = fp->f_offset;
1200 
1201 	ioflag = 0;
1202 	if (flags & O_FBLOCKING) {
1203 		/* ioflag &= ~IO_NDELAY; */
1204 	} else if (flags & O_FNONBLOCKING) {
1205 		ioflag |= IO_NDELAY;
1206 	} else if (fp->f_flag & FNONBLOCK) {
1207 		ioflag |= IO_NDELAY;
1208 	}
1209 	if (flags & O_FBUFFERED) {
1210 		/* ioflag &= ~IO_DIRECT; */
1211 	} else if (flags & O_FUNBUFFERED) {
1212 		ioflag |= IO_DIRECT;
1213 	} else if (fp->f_flag & O_DIRECT) {
1214 		ioflag |= IO_DIRECT;
1215 	}
1216 	ioflag |= sequential_heuristic(uio, fp);
1217 
1218 	error = dev_dread(dev, uio, ioflag, fp);
1219 
1220 	release_dev(dev);
1221 	if (node)
1222 		nanotime(&node->atime);
1223 	if ((flags & O_FOFFSET) == 0)
1224 		fp->f_offset = uio->uio_offset;
1225 	fp->f_nextoff = uio->uio_offset;
1226 
1227 	return (error);
1228 }
1229 
1230 
1231 static int
1232 devfs_fo_write(struct file *fp, struct uio *uio,
1233 		  struct ucred *cred, int flags)
1234 {
1235 	struct devfs_node *node;
1236 	struct vnode *vp;
1237 	int ioflag;
1238 	int error;
1239 	cdev_t dev;
1240 
1241 	KASSERT(uio->uio_td == curthread,
1242 		("uio_td %p is not p %p", uio->uio_td, curthread));
1243 
1244 	vp = (struct vnode *)fp->f_data;
1245 	if (vp == NULL || vp->v_type == VBAD)
1246 		return EBADF;
1247 
1248 	node = DEVFS_NODE(vp);
1249 
1250 	if (vp->v_type == VREG)
1251 		bwillwrite(uio->uio_resid);
1252 
1253 	vp = (struct vnode *)fp->f_data;
1254 
1255 	if ((dev = vp->v_rdev) == NULL)
1256 		return EBADF;
1257 
1258 	reference_dev(dev);
1259 
1260 	if ((flags & O_FOFFSET) == 0)
1261 		uio->uio_offset = fp->f_offset;
1262 
1263 	ioflag = IO_UNIT;
1264 	if (vp->v_type == VREG &&
1265 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1266 		ioflag |= IO_APPEND;
1267 	}
1268 
1269 	if (flags & O_FBLOCKING) {
1270 		/* ioflag &= ~IO_NDELAY; */
1271 	} else if (flags & O_FNONBLOCKING) {
1272 		ioflag |= IO_NDELAY;
1273 	} else if (fp->f_flag & FNONBLOCK) {
1274 		ioflag |= IO_NDELAY;
1275 	}
1276 	if (flags & O_FBUFFERED) {
1277 		/* ioflag &= ~IO_DIRECT; */
1278 	} else if (flags & O_FUNBUFFERED) {
1279 		ioflag |= IO_DIRECT;
1280 	} else if (fp->f_flag & O_DIRECT) {
1281 		ioflag |= IO_DIRECT;
1282 	}
1283 	if (flags & O_FASYNCWRITE) {
1284 		/* ioflag &= ~IO_SYNC; */
1285 	} else if (flags & O_FSYNCWRITE) {
1286 		ioflag |= IO_SYNC;
1287 	} else if (fp->f_flag & O_FSYNC) {
1288 		ioflag |= IO_SYNC;
1289 	}
1290 
1291 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1292 		ioflag |= IO_SYNC;
1293 	ioflag |= sequential_heuristic(uio, fp);
1294 
1295 	error = dev_dwrite(dev, uio, ioflag, fp);
1296 
1297 	release_dev(dev);
1298 	if (node) {
1299 		nanotime(&node->atime);
1300 		nanotime(&node->mtime);
1301 	}
1302 
1303 	if ((flags & O_FOFFSET) == 0)
1304 		fp->f_offset = uio->uio_offset;
1305 	fp->f_nextoff = uio->uio_offset;
1306 
1307 	return (error);
1308 }
1309 
1310 
1311 static int
1312 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1313 {
1314 	struct vnode *vp;
1315 	struct vattr vattr;
1316 	struct vattr *vap;
1317 	u_short mode;
1318 	cdev_t dev;
1319 	int error;
1320 
1321 	vp = (struct vnode *)fp->f_data;
1322 	if (vp == NULL || vp->v_type == VBAD)
1323 		return EBADF;
1324 
1325 	error = vn_stat(vp, sb, cred);
1326 	if (error)
1327 		return (error);
1328 
1329 	vap = &vattr;
1330 	error = VOP_GETATTR(vp, vap);
1331 	if (error)
1332 		return (error);
1333 
1334 	/*
1335 	 * Zero the spare stat fields
1336 	 */
1337 	sb->st_lspare = 0;
1338 	sb->st_qspare1 = 0;
1339 	sb->st_qspare2 = 0;
1340 
1341 	/*
1342 	 * Copy from vattr table ... or not in case it's a cloned device
1343 	 */
1344 	if (vap->va_fsid != VNOVAL)
1345 		sb->st_dev = vap->va_fsid;
1346 	else
1347 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1348 
1349 	sb->st_ino = vap->va_fileid;
1350 
1351 	mode = vap->va_mode;
1352 	mode |= S_IFCHR;
1353 	sb->st_mode = mode;
1354 
1355 	if (vap->va_nlink > (nlink_t)-1)
1356 		sb->st_nlink = (nlink_t)-1;
1357 	else
1358 		sb->st_nlink = vap->va_nlink;
1359 
1360 	sb->st_uid = vap->va_uid;
1361 	sb->st_gid = vap->va_gid;
1362 	sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1363 	sb->st_size = vap->va_bytes;
1364 	sb->st_atimespec = vap->va_atime;
1365 	sb->st_mtimespec = vap->va_mtime;
1366 	sb->st_ctimespec = vap->va_ctime;
1367 
1368 	/*
1369 	 * A VCHR and VBLK device may track the last access and last modified
1370 	 * time independantly of the filesystem.  This is particularly true
1371 	 * because device read and write calls may bypass the filesystem.
1372 	 */
1373 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1374 		dev = vp->v_rdev;
1375 		if (dev != NULL) {
1376 			if (dev->si_lastread) {
1377 				sb->st_atimespec.tv_sec = time_second +
1378 							  (time_uptime -
1379 							   dev->si_lastread);
1380 				sb->st_atimespec.tv_nsec = 0;
1381 			}
1382 			if (dev->si_lastwrite) {
1383 				sb->st_atimespec.tv_sec = time_second +
1384 							  (time_uptime -
1385 							   dev->si_lastwrite);
1386 				sb->st_atimespec.tv_nsec = 0;
1387 			}
1388 		}
1389 	}
1390 
1391         /*
1392 	 * According to www.opengroup.org, the meaning of st_blksize is
1393 	 *   "a filesystem-specific preferred I/O block size for this
1394 	 *    object.  In some filesystem types, this may vary from file
1395 	 *    to file"
1396 	 * Default to PAGE_SIZE after much discussion.
1397 	 */
1398 
1399 	sb->st_blksize = PAGE_SIZE;
1400 
1401 	sb->st_flags = vap->va_flags;
1402 
1403 	error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1404 	if (error)
1405 		sb->st_gen = 0;
1406 	else
1407 		sb->st_gen = (u_int32_t)vap->va_gen;
1408 
1409 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1410 
1411 	return (0);
1412 }
1413 
1414 
1415 static int
1416 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1417 {
1418 	struct vnode *vp;
1419 	int error;
1420 	cdev_t dev;
1421 
1422 	vp = (struct vnode *)fp->f_data;
1423 	if (vp == NULL || vp->v_type == VBAD) {
1424 		error = EBADF;
1425 		goto done;
1426 	}
1427 	if ((dev = vp->v_rdev) == NULL) {
1428 		error = EBADF;
1429 		goto done;
1430 	}
1431 	reference_dev(dev);
1432 
1433 	error = dev_dkqfilter(dev, kn, fp);
1434 
1435 	release_dev(dev);
1436 
1437 done:
1438 	return (error);
1439 }
1440 
1441 /*
1442  * MPALMOSTSAFE - acquires mplock
1443  */
1444 static int
1445 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1446 		  struct ucred *ucred, struct sysmsg *msg)
1447 {
1448 #if 0
1449 	struct devfs_node *node;
1450 #endif
1451 	struct vnode *vp;
1452 	struct vnode *ovp;
1453 	cdev_t	dev;
1454 	int error;
1455 	struct fiodname_args *name_args;
1456 	size_t namlen;
1457 	const char *name;
1458 
1459 	vp = ((struct vnode *)fp->f_data);
1460 
1461 	if ((dev = vp->v_rdev) == NULL)
1462 		return EBADF;		/* device was revoked */
1463 
1464 	reference_dev(dev);
1465 
1466 #if 0
1467 	node = DEVFS_NODE(vp);
1468 #endif
1469 
1470 	devfs_debug(DEVFS_DEBUG_DEBUG,
1471 		    "devfs_fo_ioctl() called! for dev %s\n",
1472 		    dev->si_name);
1473 
1474 	if (com == FIODTYPE) {
1475 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1476 		error = 0;
1477 		goto out;
1478 	} else if (com == FIODNAME) {
1479 		name_args = (struct fiodname_args *)data;
1480 		name = dev->si_name;
1481 		namlen = strlen(name) + 1;
1482 
1483 		devfs_debug(DEVFS_DEBUG_DEBUG,
1484 			    "ioctl, got: FIODNAME for %s\n", name);
1485 
1486 		if (namlen <= name_args->len)
1487 			error = copyout(dev->si_name, name_args->name, namlen);
1488 		else
1489 			error = EINVAL;
1490 
1491 		devfs_debug(DEVFS_DEBUG_DEBUG,
1492 			    "ioctl stuff: error: %d\n", error);
1493 		goto out;
1494 	}
1495 
1496 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg, fp);
1497 
1498 #if 0
1499 	if (node) {
1500 		nanotime(&node->atime);
1501 		nanotime(&node->mtime);
1502 	}
1503 #endif
1504 	if (com == TIOCSCTTY) {
1505 		devfs_debug(DEVFS_DEBUG_DEBUG,
1506 			    "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1507 			    dev->si_name);
1508 	}
1509 	if (error == 0 && com == TIOCSCTTY) {
1510 		struct proc *p = curthread->td_proc;
1511 		struct session *sess;
1512 
1513 		devfs_debug(DEVFS_DEBUG_DEBUG,
1514 			    "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1515 			    dev->si_name);
1516 		if (p == NULL) {
1517 			error = ENOTTY;
1518 			goto out;
1519 		}
1520 		sess = p->p_session;
1521 
1522 		/*
1523 		 * Do nothing if reassigning same control tty
1524 		 */
1525 		if (sess->s_ttyvp == vp) {
1526 			error = 0;
1527 			goto out;
1528 		}
1529 
1530 		/*
1531 		 * Get rid of reference to old control tty
1532 		 */
1533 		ovp = sess->s_ttyvp;
1534 		vref(vp);
1535 		sess->s_ttyvp = vp;
1536 		if (ovp)
1537 			vrele(ovp);
1538 	}
1539 
1540 out:
1541 	release_dev(dev);
1542 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1543 	return (error);
1544 }
1545 
1546 
1547 static int
1548 devfs_spec_fsync(struct vop_fsync_args *ap)
1549 {
1550 	struct vnode *vp = ap->a_vp;
1551 	int error;
1552 
1553 	if (!vn_isdisk(vp, NULL))
1554 		return (0);
1555 
1556 	/*
1557 	 * Flush all dirty buffers associated with a block device.
1558 	 */
1559 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1560 	return (error);
1561 }
1562 
1563 static int
1564 devfs_spec_read(struct vop_read_args *ap)
1565 {
1566 	struct devfs_node *node;
1567 	struct vnode *vp;
1568 	struct uio *uio;
1569 	cdev_t dev;
1570 	int error;
1571 
1572 	vp = ap->a_vp;
1573 	dev = vp->v_rdev;
1574 	uio = ap->a_uio;
1575 	node = DEVFS_NODE(vp);
1576 
1577 	if (dev == NULL)		/* device was revoked */
1578 		return (EBADF);
1579 	if (uio->uio_resid == 0)
1580 		return (0);
1581 
1582 	vn_unlock(vp);
1583 	error = dev_dread(dev, uio, ap->a_ioflag, NULL);
1584 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1585 
1586 	if (node)
1587 		nanotime(&node->atime);
1588 
1589 	return (error);
1590 }
1591 
1592 /*
1593  * Vnode op for write
1594  *
1595  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1596  *	      struct ucred *a_cred)
1597  */
1598 static int
1599 devfs_spec_write(struct vop_write_args *ap)
1600 {
1601 	struct devfs_node *node;
1602 	struct vnode *vp;
1603 	struct uio *uio;
1604 	cdev_t dev;
1605 	int error;
1606 
1607 	vp = ap->a_vp;
1608 	dev = vp->v_rdev;
1609 	uio = ap->a_uio;
1610 	node = DEVFS_NODE(vp);
1611 
1612 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1613 
1614 	if (dev == NULL)		/* device was revoked */
1615 		return (EBADF);
1616 
1617 	vn_unlock(vp);
1618 	error = dev_dwrite(dev, uio, ap->a_ioflag, NULL);
1619 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1620 
1621 	if (node) {
1622 		nanotime(&node->atime);
1623 		nanotime(&node->mtime);
1624 	}
1625 
1626 	return (error);
1627 }
1628 
1629 /*
1630  * Device ioctl operation.
1631  *
1632  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1633  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1634  */
1635 static int
1636 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1637 {
1638 	struct vnode *vp = ap->a_vp;
1639 #if 0
1640 	struct devfs_node *node;
1641 #endif
1642 	cdev_t dev;
1643 
1644 	if ((dev = vp->v_rdev) == NULL)
1645 		return (EBADF);		/* device was revoked */
1646 #if 0
1647 	node = DEVFS_NODE(vp);
1648 
1649 	if (node) {
1650 		nanotime(&node->atime);
1651 		nanotime(&node->mtime);
1652 	}
1653 #endif
1654 
1655 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1656 			   ap->a_cred, ap->a_sysmsg, NULL));
1657 }
1658 
1659 /*
1660  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1661  */
1662 /* ARGSUSED */
1663 static int
1664 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1665 {
1666 	struct vnode *vp = ap->a_vp;
1667 #if 0
1668 	struct devfs_node *node;
1669 #endif
1670 	cdev_t dev;
1671 
1672 	if ((dev = vp->v_rdev) == NULL)
1673 		return (EBADF);		/* device was revoked (EBADF) */
1674 #if 0
1675 	node = DEVFS_NODE(vp);
1676 
1677 	if (node)
1678 		nanotime(&node->atime);
1679 #endif
1680 
1681 	return (dev_dkqfilter(dev, ap->a_kn, NULL));
1682 }
1683 
1684 /*
1685  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1686  * calls are not limited to device DMA limits so we have to deal with the
1687  * case.
1688  *
1689  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1690  */
1691 static int
1692 devfs_spec_strategy(struct vop_strategy_args *ap)
1693 {
1694 	struct bio *bio = ap->a_bio;
1695 	struct buf *bp = bio->bio_buf;
1696 	struct buf *nbp;
1697 	struct vnode *vp;
1698 	struct mount *mp;
1699 	int chunksize;
1700 	int maxiosize;
1701 
1702 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1703 		buf_start(bp);
1704 
1705 	/*
1706 	 * Collect statistics on synchronous and asynchronous read
1707 	 * and write counts for disks that have associated filesystems.
1708 	 */
1709 	vp = ap->a_vp;
1710 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1711 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1712 		if (bp->b_cmd == BUF_CMD_READ) {
1713 			if (bp->b_flags & BIO_SYNC)
1714 				mp->mnt_stat.f_syncreads++;
1715 			else
1716 				mp->mnt_stat.f_asyncreads++;
1717 		} else {
1718 			if (bp->b_flags & BIO_SYNC)
1719 				mp->mnt_stat.f_syncwrites++;
1720 			else
1721 				mp->mnt_stat.f_asyncwrites++;
1722 		}
1723 	}
1724 
1725         /*
1726          * Device iosize limitations only apply to read and write.  Shortcut
1727          * the I/O if it fits.
1728          */
1729 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1730 		devfs_debug(DEVFS_DEBUG_DEBUG,
1731 			    "%s: si_iosize_max not set!\n",
1732 			    dev_dname(vp->v_rdev));
1733 		maxiosize = MAXPHYS;
1734 	}
1735 #if SPEC_CHAIN_DEBUG & 2
1736 	maxiosize = 4096;
1737 #endif
1738         if (bp->b_bcount <= maxiosize ||
1739             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1740                 dev_dstrategy_chain(vp->v_rdev, bio);
1741                 return (0);
1742         }
1743 
1744 	/*
1745 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1746 	 */
1747 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1748 	initbufbio(nbp);
1749 	buf_dep_init(nbp);
1750 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1751 	BUF_KERNPROC(nbp);
1752 	nbp->b_vp = vp;
1753 	nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1754 	nbp->b_data = bp->b_data;
1755 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1756 	nbp->b_bio1.bio_offset = bio->bio_offset;
1757 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1758 
1759 	/*
1760 	 * Start the first transfer
1761 	 */
1762 	if (vn_isdisk(vp, NULL))
1763 		chunksize = vp->v_rdev->si_bsize_phys;
1764 	else
1765 		chunksize = DEV_BSIZE;
1766 	chunksize = maxiosize / chunksize * chunksize;
1767 #if SPEC_CHAIN_DEBUG & 1
1768 	devfs_debug(DEVFS_DEBUG_DEBUG,
1769 		    "spec_strategy chained I/O chunksize=%d\n",
1770 		    chunksize);
1771 #endif
1772 	nbp->b_cmd = bp->b_cmd;
1773 	nbp->b_bcount = chunksize;
1774 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1775 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1776 
1777 #if SPEC_CHAIN_DEBUG & 1
1778 	devfs_debug(DEVFS_DEBUG_DEBUG,
1779 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1780 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1781 #endif
1782 
1783 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1784 
1785 	if (DEVFS_NODE(vp)) {
1786 		nanotime(&DEVFS_NODE(vp)->atime);
1787 		nanotime(&DEVFS_NODE(vp)->mtime);
1788 	}
1789 
1790 	return (0);
1791 }
1792 
1793 /*
1794  * Chunked up transfer completion routine - chain transfers until done
1795  *
1796  * NOTE: MPSAFE callback.
1797  */
1798 static
1799 void
1800 devfs_spec_strategy_done(struct bio *nbio)
1801 {
1802 	struct buf *nbp = nbio->bio_buf;
1803 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1804 	struct buf *bp = bio->bio_buf;			/* original bp */
1805 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1806 	int boffset = nbp->b_data - bp->b_data;
1807 
1808 	if (nbp->b_flags & B_ERROR) {
1809 		/*
1810 		 * An error terminates the chain, propogate the error back
1811 		 * to the original bp
1812 		 */
1813 		bp->b_flags |= B_ERROR;
1814 		bp->b_error = nbp->b_error;
1815 		bp->b_resid = bp->b_bcount - boffset +
1816 			      (nbp->b_bcount - nbp->b_resid);
1817 #if SPEC_CHAIN_DEBUG & 1
1818 		devfs_debug(DEVFS_DEBUG_DEBUG,
1819 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1820 			    bp, bp->b_error, bp->b_bcount,
1821 			    bp->b_bcount - bp->b_resid);
1822 #endif
1823 	} else if (nbp->b_resid) {
1824 		/*
1825 		 * A short read or write terminates the chain
1826 		 */
1827 		bp->b_error = nbp->b_error;
1828 		bp->b_resid = bp->b_bcount - boffset +
1829 			      (nbp->b_bcount - nbp->b_resid);
1830 #if SPEC_CHAIN_DEBUG & 1
1831 		devfs_debug(DEVFS_DEBUG_DEBUG,
1832 			    "spec_strategy: chain %p short read(1) "
1833 			    "bcount %d/%d\n",
1834 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1835 #endif
1836 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1837 		/*
1838 		 * A short read or write can also occur by truncating b_bcount
1839 		 */
1840 #if SPEC_CHAIN_DEBUG & 1
1841 		devfs_debug(DEVFS_DEBUG_DEBUG,
1842 			    "spec_strategy: chain %p short read(2) "
1843 			    "bcount %d/%d\n",
1844 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1845 #endif
1846 		bp->b_error = 0;
1847 		bp->b_bcount = nbp->b_bcount + boffset;
1848 		bp->b_resid = nbp->b_resid;
1849 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1850 		/*
1851 		 * No more data terminates the chain
1852 		 */
1853 #if SPEC_CHAIN_DEBUG & 1
1854 		devfs_debug(DEVFS_DEBUG_DEBUG,
1855 			    "spec_strategy: chain %p finished bcount %d\n",
1856 			    bp, bp->b_bcount);
1857 #endif
1858 		bp->b_error = 0;
1859 		bp->b_resid = 0;
1860 	} else {
1861 		/*
1862 		 * Continue the chain
1863 		 */
1864 		boffset += nbp->b_bcount;
1865 		nbp->b_data = bp->b_data + boffset;
1866 		nbp->b_bcount = bp->b_bcount - boffset;
1867 		if (nbp->b_bcount > chunksize)
1868 			nbp->b_bcount = chunksize;
1869 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1870 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1871 
1872 #if SPEC_CHAIN_DEBUG & 1
1873 		devfs_debug(DEVFS_DEBUG_DEBUG,
1874 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1875 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
1876 #endif
1877 
1878 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1879 		return;
1880 	}
1881 
1882 	/*
1883 	 * Fall through to here on termination.  biodone(bp) and
1884 	 * clean up and free nbp.
1885 	 */
1886 	biodone(bio);
1887 	BUF_UNLOCK(nbp);
1888 	uninitbufbio(nbp);
1889 	kfree(nbp, M_DEVBUF);
1890 }
1891 
1892 /*
1893  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1894  */
1895 static int
1896 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1897 {
1898 	struct buf *bp;
1899 
1900 	/*
1901 	 * XXX: This assumes that strategy does the deed right away.
1902 	 * XXX: this may not be TRTTD.
1903 	 */
1904 	KKASSERT(ap->a_vp->v_rdev != NULL);
1905 	if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
1906 		return (0);
1907 	bp = geteblk(ap->a_length);
1908 	bp->b_cmd = BUF_CMD_FREEBLKS;
1909 	bp->b_bio1.bio_offset = ap->a_offset;
1910 	bp->b_bcount = ap->a_length;
1911 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1912 	return (0);
1913 }
1914 
1915 /*
1916  * Implement degenerate case where the block requested is the block
1917  * returned, and assume that the entire device is contiguous in regards
1918  * to the contiguous block range (runp and runb).
1919  *
1920  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1921  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
1922  */
1923 static int
1924 devfs_spec_bmap(struct vop_bmap_args *ap)
1925 {
1926 	if (ap->a_doffsetp != NULL)
1927 		*ap->a_doffsetp = ap->a_loffset;
1928 	if (ap->a_runp != NULL)
1929 		*ap->a_runp = MAXBSIZE;
1930 	if (ap->a_runb != NULL) {
1931 		if (ap->a_loffset < MAXBSIZE)
1932 			*ap->a_runb = (int)ap->a_loffset;
1933 		else
1934 			*ap->a_runb = MAXBSIZE;
1935 	}
1936 	return (0);
1937 }
1938 
1939 
1940 /*
1941  * Special device advisory byte-level locks.
1942  *
1943  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1944  *		struct flock *a_fl, int a_flags)
1945  */
1946 /* ARGSUSED */
1947 static int
1948 devfs_spec_advlock(struct vop_advlock_args *ap)
1949 {
1950 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1951 }
1952 
1953 /*
1954  * NOTE: MPSAFE callback.
1955  */
1956 static void
1957 devfs_spec_getpages_iodone(struct bio *bio)
1958 {
1959 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
1960 	wakeup(bio->bio_buf);
1961 }
1962 
1963 /*
1964  * spec_getpages() - get pages associated with device vnode.
1965  *
1966  * Note that spec_read and spec_write do not use the buffer cache, so we
1967  * must fully implement getpages here.
1968  */
1969 static int
1970 devfs_spec_getpages(struct vop_getpages_args *ap)
1971 {
1972 	vm_offset_t kva;
1973 	int error;
1974 	int i, pcount, size;
1975 	struct buf *bp;
1976 	vm_page_t m;
1977 	vm_ooffset_t offset;
1978 	int toff, nextoff, nread;
1979 	struct vnode *vp = ap->a_vp;
1980 	int blksiz;
1981 	int gotreqpage;
1982 
1983 	error = 0;
1984 	pcount = round_page(ap->a_count) / PAGE_SIZE;
1985 
1986 	/*
1987 	 * Calculate the offset of the transfer and do sanity check.
1988 	 */
1989 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1990 
1991 	/*
1992 	 * Round up physical size for real devices.  We cannot round using
1993 	 * v_mount's block size data because v_mount has nothing to do with
1994 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
1995 	 * size for the device itself.
1996 	 *
1997 	 * We can't use v_rdev->si_mountpoint because it only exists when the
1998 	 * block device is mounted.  However, we can use v_rdev.
1999 	 */
2000 	if (vn_isdisk(vp, NULL))
2001 		blksiz = vp->v_rdev->si_bsize_phys;
2002 	else
2003 		blksiz = DEV_BSIZE;
2004 
2005 	size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
2006 
2007 	bp = getpbuf_kva(NULL);
2008 	kva = (vm_offset_t)bp->b_data;
2009 
2010 	/*
2011 	 * Map the pages to be read into the kva.
2012 	 */
2013 	pmap_qenter(kva, ap->a_m, pcount);
2014 
2015 	/* Build a minimal buffer header. */
2016 	bp->b_cmd = BUF_CMD_READ;
2017 	bp->b_bcount = size;
2018 	bp->b_resid = 0;
2019 	bsetrunningbufspace(bp, size);
2020 
2021 	bp->b_bio1.bio_offset = offset;
2022 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2023 
2024 	mycpu->gd_cnt.v_vnodein++;
2025 	mycpu->gd_cnt.v_vnodepgsin += pcount;
2026 
2027 	/* Do the input. */
2028 	vn_strategy(ap->a_vp, &bp->b_bio1);
2029 
2030 	crit_enter();
2031 
2032 	/* We definitely need to be at splbio here. */
2033 	while (bp->b_cmd != BUF_CMD_DONE)
2034 		tsleep(bp, 0, "spread", 0);
2035 
2036 	crit_exit();
2037 
2038 	if (bp->b_flags & B_ERROR) {
2039 		if (bp->b_error)
2040 			error = bp->b_error;
2041 		else
2042 			error = EIO;
2043 	}
2044 
2045 	/*
2046 	 * If EOF is encountered we must zero-extend the result in order
2047 	 * to ensure that the page does not contain garabge.  When no
2048 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
2049 	 * b_resid is relative to b_bcount and should be 0, but some devices
2050 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
2051 	 */
2052 	nread = bp->b_bcount - bp->b_resid;
2053 	if (nread < ap->a_count)
2054 		bzero((caddr_t)kva + nread, ap->a_count - nread);
2055 	pmap_qremove(kva, pcount);
2056 
2057 	gotreqpage = 0;
2058 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2059 		nextoff = toff + PAGE_SIZE;
2060 		m = ap->a_m[i];
2061 
2062 		m->flags &= ~PG_ZERO;
2063 
2064 		/*
2065 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2066 		 *	 pmap modified bit.  pmap modified bit should have
2067 		 *	 already been cleared.
2068 		 */
2069 		if (nextoff <= nread) {
2070 			m->valid = VM_PAGE_BITS_ALL;
2071 			vm_page_undirty(m);
2072 		} else if (toff < nread) {
2073 			/*
2074 			 * Since this is a VM request, we have to supply the
2075 			 * unaligned offset to allow vm_page_set_valid()
2076 			 * to zero sub-DEV_BSIZE'd portions of the page.
2077 			 */
2078 			vm_page_set_valid(m, 0, nread - toff);
2079 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2080 		} else {
2081 			m->valid = 0;
2082 			vm_page_undirty(m);
2083 		}
2084 
2085 		if (i != ap->a_reqpage) {
2086 			/*
2087 			 * Just in case someone was asking for this page we
2088 			 * now tell them that it is ok to use.
2089 			 */
2090 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2091 				if (m->valid) {
2092 					if (m->flags & PG_REFERENCED) {
2093 						vm_page_activate(m);
2094 					} else {
2095 						vm_page_deactivate(m);
2096 					}
2097 					vm_page_wakeup(m);
2098 				} else {
2099 					vm_page_free(m);
2100 				}
2101 			} else {
2102 				vm_page_free(m);
2103 			}
2104 		} else if (m->valid) {
2105 			gotreqpage = 1;
2106 			/*
2107 			 * Since this is a VM request, we need to make the
2108 			 * entire page presentable by zeroing invalid sections.
2109 			 */
2110 			if (m->valid != VM_PAGE_BITS_ALL)
2111 			    vm_page_zero_invalid(m, FALSE);
2112 		}
2113 	}
2114 	if (!gotreqpage) {
2115 		m = ap->a_m[ap->a_reqpage];
2116 		devfs_debug(DEVFS_DEBUG_WARNING,
2117 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2118 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2119 		devfs_debug(DEVFS_DEBUG_WARNING,
2120 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2121 		    size, bp->b_resid, ap->a_count, m->valid);
2122 		devfs_debug(DEVFS_DEBUG_WARNING,
2123 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2124 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2125 		/*
2126 		 * Free the buffer header back to the swap buffer pool.
2127 		 */
2128 		relpbuf(bp, NULL);
2129 		return VM_PAGER_ERROR;
2130 	}
2131 	/*
2132 	 * Free the buffer header back to the swap buffer pool.
2133 	 */
2134 	relpbuf(bp, NULL);
2135 	if (DEVFS_NODE(ap->a_vp))
2136 		nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2137 	return VM_PAGER_OK;
2138 }
2139 
2140 static __inline
2141 int
2142 sequential_heuristic(struct uio *uio, struct file *fp)
2143 {
2144 	/*
2145 	 * Sequential heuristic - detect sequential operation
2146 	 */
2147 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2148 	    uio->uio_offset == fp->f_nextoff) {
2149 		/*
2150 		 * XXX we assume that the filesystem block size is
2151 		 * the default.  Not true, but still gives us a pretty
2152 		 * good indicator of how sequential the read operations
2153 		 * are.
2154 		 */
2155 		int tmpseq = fp->f_seqcount;
2156 
2157 		tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2158 		if (tmpseq > IO_SEQMAX)
2159 			tmpseq = IO_SEQMAX;
2160 		fp->f_seqcount = tmpseq;
2161 		return(fp->f_seqcount << IO_SEQSHIFT);
2162 	}
2163 
2164 	/*
2165 	 * Not sequential, quick draw-down of seqcount
2166 	 */
2167 	if (fp->f_seqcount > 1)
2168 		fp->f_seqcount = 1;
2169 	else
2170 		fp->f_seqcount = 0;
2171 	return(0);
2172 }
2173