xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision 441d34b2441f59fde86fa4ef2d5d5cb7a6bfcb11)
1 /*
2  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Alex Hornung <ahornung@gmail.com>
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  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/time.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/fcntl.h>
40 #include <sys/proc.h>
41 #include <sys/priv.h>
42 #include <sys/signalvar.h>
43 #include <sys/vnode.h>
44 #include <sys/uio.h>
45 #include <sys/mount.h>
46 #include <sys/file.h>
47 #include <sys/fcntl.h>
48 #include <sys/namei.h>
49 #include <sys/dirent.h>
50 #include <sys/malloc.h>
51 #include <sys/stat.h>
52 #include <sys/reg.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_zone.h>
55 #include <vm/vm_object.h>
56 #include <sys/filio.h>
57 #include <sys/ttycom.h>
58 #include <sys/tty.h>
59 #include <sys/diskslice.h>
60 #include <sys/sysctl.h>
61 #include <sys/devfs.h>
62 #include <sys/pioctl.h>
63 
64 #include <machine/limits.h>
65 
66 #include <sys/buf2.h>
67 #include <sys/sysref2.h>
68 #include <sys/mplock2.h>
69 #include <vm/vm_page2.h>
70 
71 MALLOC_DECLARE(M_DEVFS);
72 #define DEVFS_BADOP	(void *)devfs_badop
73 
74 static int devfs_badop(struct vop_generic_args *);
75 static int devfs_access(struct vop_access_args *);
76 static int devfs_inactive(struct vop_inactive_args *);
77 static int devfs_reclaim(struct vop_reclaim_args *);
78 static int devfs_readdir(struct vop_readdir_args *);
79 static int devfs_getattr(struct vop_getattr_args *);
80 static int devfs_setattr(struct vop_setattr_args *);
81 static int devfs_readlink(struct vop_readlink_args *);
82 static int devfs_print(struct vop_print_args *);
83 
84 static int devfs_nresolve(struct vop_nresolve_args *);
85 static int devfs_nlookupdotdot(struct vop_nlookupdotdot_args *);
86 static int devfs_nmkdir(struct vop_nmkdir_args *);
87 static int devfs_nsymlink(struct vop_nsymlink_args *);
88 static int devfs_nrmdir(struct vop_nrmdir_args *);
89 static int devfs_nremove(struct vop_nremove_args *);
90 
91 static int devfs_spec_open(struct vop_open_args *);
92 static int devfs_spec_close(struct vop_close_args *);
93 static int devfs_spec_fsync(struct vop_fsync_args *);
94 
95 static int devfs_spec_read(struct vop_read_args *);
96 static int devfs_spec_write(struct vop_write_args *);
97 static int devfs_spec_ioctl(struct vop_ioctl_args *);
98 static int devfs_spec_poll(struct vop_poll_args *);
99 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
100 static int devfs_spec_strategy(struct vop_strategy_args *);
101 static void devfs_spec_strategy_done(struct bio *);
102 static int devfs_spec_freeblks(struct vop_freeblks_args *);
103 static int devfs_spec_bmap(struct vop_bmap_args *);
104 static int devfs_spec_advlock(struct vop_advlock_args *);
105 static void devfs_spec_getpages_iodone(struct bio *);
106 static int devfs_spec_getpages(struct vop_getpages_args *);
107 
108 
109 static int devfs_specf_close(struct file *);
110 static int devfs_specf_read(struct file *, struct uio *, struct ucred *, int);
111 static int devfs_specf_write(struct file *, struct uio *, struct ucred *, int);
112 static int devfs_specf_stat(struct file *, struct stat *, struct ucred *);
113 static int devfs_specf_kqfilter(struct file *, struct knote *);
114 static int devfs_specf_poll(struct file *, int, struct ucred *);
115 static int devfs_specf_ioctl(struct file *, u_long, caddr_t,
116 				struct ucred *, struct sysmsg *);
117 static __inline int sequential_heuristic(struct uio *, struct file *);
118 
119 extern struct lock devfs_lock;
120 
121 static int mpsafe_reads, mpsafe_writes, mplock_reads, mplock_writes;
122 
123 /*
124  * devfs vnode operations for regular files
125  */
126 struct vop_ops devfs_vnode_norm_vops = {
127 	.vop_default =		vop_defaultop,
128 	.vop_access =		devfs_access,
129 	.vop_advlock =		DEVFS_BADOP,
130 	.vop_bmap =			DEVFS_BADOP,
131 	.vop_close =		vop_stdclose,
132 	.vop_getattr =		devfs_getattr,
133 	.vop_inactive =		devfs_inactive,
134 	.vop_ncreate =		DEVFS_BADOP,
135 	.vop_nresolve =		devfs_nresolve,
136 	.vop_nlookupdotdot =	devfs_nlookupdotdot,
137 	.vop_nlink =		DEVFS_BADOP,
138 	.vop_nmkdir =		devfs_nmkdir,
139 	.vop_nmknod =		DEVFS_BADOP,
140 	.vop_nremove =		devfs_nremove,
141 	.vop_nrename =		DEVFS_BADOP,
142 	.vop_nrmdir =		devfs_nrmdir,
143 	.vop_nsymlink =		devfs_nsymlink,
144 	.vop_open =			vop_stdopen,
145 	.vop_pathconf =		vop_stdpathconf,
146 	.vop_print =		devfs_print,
147 	.vop_read =			DEVFS_BADOP,
148 	.vop_readdir =		devfs_readdir,
149 	.vop_readlink =		devfs_readlink,
150 	.vop_reclaim =		devfs_reclaim,
151 	.vop_setattr =		devfs_setattr,
152 	.vop_write =		DEVFS_BADOP,
153 	.vop_ioctl =		DEVFS_BADOP
154 };
155 
156 /*
157  * devfs vnode operations for character devices
158  */
159 struct vop_ops devfs_vnode_dev_vops = {
160 	.vop_default =		vop_defaultop,
161 	.vop_access =		devfs_access,
162 	.vop_advlock =		devfs_spec_advlock,
163 	.vop_bmap =			devfs_spec_bmap,
164 	.vop_close =		devfs_spec_close,
165 	.vop_freeblks =		devfs_spec_freeblks,
166 	.vop_fsync =		devfs_spec_fsync,
167 	.vop_getattr =		devfs_getattr,
168 	.vop_getpages =		devfs_spec_getpages,
169 	.vop_inactive =		devfs_inactive,
170 	.vop_open =			devfs_spec_open,
171 	.vop_pathconf =		vop_stdpathconf,
172 	.vop_print =		devfs_print,
173 	.vop_poll =			devfs_spec_poll,
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_reclaim,
179 	.vop_setattr =		devfs_setattr,
180 	.vop_strategy =		devfs_spec_strategy,
181 	.vop_write =		devfs_spec_write,
182 	.vop_ioctl =		devfs_spec_ioctl
183 };
184 
185 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
186 
187 struct fileops devfs_dev_fileops = {
188 	.fo_read = devfs_specf_read,
189 	.fo_write = devfs_specf_write,
190 	.fo_ioctl = devfs_specf_ioctl,
191 	.fo_poll = devfs_specf_poll,
192 	.fo_kqfilter = devfs_specf_kqfilter,
193 	.fo_stat = devfs_specf_stat,
194 	.fo_close = devfs_specf_close,
195 	.fo_shutdown = nofo_shutdown
196 };
197 
198 /*
199  * These two functions are possibly temporary hacks for
200  * devices (aka the pty code) which want to control the
201  * node attributes themselves.
202  *
203  * XXX we may ultimately desire to simply remove the uid/gid/mode
204  * from the node entirely.
205  */
206 static __inline void
207 node_sync_dev_get(struct devfs_node *node)
208 {
209 	cdev_t dev;
210 
211 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
212 		node->uid = dev->si_uid;
213 		node->gid = dev->si_gid;
214 		node->mode = dev->si_perms;
215 	}
216 }
217 
218 static __inline void
219 node_sync_dev_set(struct devfs_node *node)
220 {
221 	cdev_t dev;
222 
223 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
224 		dev->si_uid = node->uid;
225 		dev->si_gid = node->gid;
226 		dev->si_perms = node->mode;
227 	}
228 }
229 
230 /*
231  * generic entry point for unsupported operations
232  */
233 static int
234 devfs_badop(struct vop_generic_args *ap)
235 {
236 	return (EIO);
237 }
238 
239 
240 static int
241 devfs_access(struct vop_access_args *ap)
242 {
243 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
244 	int error;
245 
246 	if (!devfs_node_is_accessible(node))
247 		return ENOENT;
248 	node_sync_dev_get(node);
249 	error = vop_helper_access(ap, node->uid, node->gid,
250 				  node->mode, node->flags);
251 
252 	return error;
253 }
254 
255 
256 static int
257 devfs_inactive(struct vop_inactive_args *ap)
258 {
259 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
260 
261 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
262 		vrecycle(ap->a_vp);
263 	return 0;
264 }
265 
266 
267 static int
268 devfs_reclaim(struct vop_reclaim_args *ap)
269 {
270 	struct devfs_node *node;
271 	struct vnode *vp;
272 	int locked;
273 
274 	/*
275 	 * Check if it is locked already. if not, we acquire the devfs lock
276 	 */
277 	if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
278 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
279 		locked = 1;
280 	} else {
281 		locked = 0;
282 	}
283 
284 	/*
285 	 * Get rid of the devfs_node if it is no longer linked into the
286 	 * topology.
287 	 */
288 	vp = ap->a_vp;
289 	if ((node = DEVFS_NODE(vp)) != NULL) {
290 		node->v_node = NULL;
291 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
292 			devfs_freep(node);
293 	}
294 
295 	if (locked)
296 		lockmgr(&devfs_lock, LK_RELEASE);
297 
298 	/*
299 	 * v_rdev needs to be properly released using v_release_rdev
300 	 * Make sure v_data is NULL as well.
301 	 */
302 	vp->v_data = NULL;
303 	v_release_rdev(vp);
304 	return 0;
305 }
306 
307 
308 static int
309 devfs_readdir(struct vop_readdir_args *ap)
310 {
311 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
312 	struct devfs_node *node;
313 	int cookie_index;
314 	int ncookies;
315 	int error2;
316 	int error;
317 	int r;
318 	off_t *cookies;
319 	off_t saveoff;
320 
321 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
322 
323 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
324 		return (EINVAL);
325 	if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
326 		return (error);
327 
328 	if (!devfs_node_is_accessible(dnode)) {
329 		vn_unlock(ap->a_vp);
330 		return ENOENT;
331 	}
332 
333 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
334 
335 	saveoff = ap->a_uio->uio_offset;
336 
337 	if (ap->a_ncookies) {
338 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
339 		if (ncookies > 256)
340 			ncookies = 256;
341 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
342 		cookie_index = 0;
343 	} else {
344 		ncookies = -1;
345 		cookies = NULL;
346 		cookie_index = 0;
347 	}
348 
349 	nanotime(&dnode->atime);
350 
351 	if (saveoff == 0) {
352 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
353 				     DT_DIR, 1, ".");
354 		if (r)
355 			goto done;
356 		if (cookies)
357 			cookies[cookie_index] = saveoff;
358 		saveoff++;
359 		cookie_index++;
360 		if (cookie_index == ncookies)
361 			goto done;
362 	}
363 
364 	if (saveoff == 1) {
365 		if (dnode->parent) {
366 			r = vop_write_dirent(&error, ap->a_uio,
367 					     dnode->parent->d_dir.d_ino,
368 					     DT_DIR, 2, "..");
369 		} else {
370 			r = vop_write_dirent(&error, ap->a_uio,
371 					     dnode->d_dir.d_ino,
372 					     DT_DIR, 2, "..");
373 		}
374 		if (r)
375 			goto done;
376 		if (cookies)
377 			cookies[cookie_index] = saveoff;
378 		saveoff++;
379 		cookie_index++;
380 		if (cookie_index == ncookies)
381 			goto done;
382 	}
383 
384 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
385 		if ((node->flags & DEVFS_HIDDEN) ||
386 		    (node->flags & DEVFS_INVISIBLE)) {
387 			continue;
388 		}
389 
390 		/*
391 		 * If the node type is a valid devfs alias, then we make sure that the
392 		 * target isn't hidden. If it is, we don't show the link in the
393 		 * directory listing.
394 		 */
395 		if ((node->node_type == Plink) && (node->link_target != NULL) &&
396 			(node->link_target->flags & DEVFS_HIDDEN))
397 			continue;
398 
399 		if (node->cookie < saveoff)
400 			continue;
401 
402 		saveoff = node->cookie;
403 
404 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
405 					  node->d_dir.d_type,
406 					  node->d_dir.d_namlen,
407 					  node->d_dir.d_name);
408 
409 		if (error2)
410 			break;
411 
412 		saveoff++;
413 
414 		if (cookies)
415 			cookies[cookie_index] = node->cookie;
416 		++cookie_index;
417 		if (cookie_index == ncookies)
418 			break;
419 	}
420 
421 done:
422 	lockmgr(&devfs_lock, LK_RELEASE);
423 	vn_unlock(ap->a_vp);
424 
425 	ap->a_uio->uio_offset = saveoff;
426 	if (error && cookie_index == 0) {
427 		if (cookies) {
428 			kfree(cookies, M_TEMP);
429 			*ap->a_ncookies = 0;
430 			*ap->a_cookies = NULL;
431 		}
432 	} else {
433 		if (cookies) {
434 			*ap->a_ncookies = cookie_index;
435 			*ap->a_cookies = cookies;
436 		}
437 	}
438 	return (error);
439 }
440 
441 
442 static int
443 devfs_nresolve(struct vop_nresolve_args *ap)
444 {
445 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
446 	struct devfs_node *node, *found = NULL;
447 	struct namecache *ncp;
448 	struct vnode *vp = NULL;
449 	int error = 0;
450 	int len;
451 	int depth;
452 
453 	ncp = ap->a_nch->ncp;
454 	len = ncp->nc_nlen;
455 
456 	if (!devfs_node_is_accessible(dnode))
457 		return ENOENT;
458 
459 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
460 
461 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
462 		error = ENOENT;
463 		cache_setvp(ap->a_nch, NULL);
464 		goto out;
465 	}
466 
467 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
468 		if (len == node->d_dir.d_namlen) {
469 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
470 				found = node;
471 				break;
472 			}
473 		}
474 	}
475 
476 	if (found) {
477 		depth = 0;
478 		while ((found->node_type == Plink) && (found->link_target)) {
479 			if (depth >= 8) {
480 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
481 				break;
482 			}
483 
484 			found = found->link_target;
485 			++depth;
486 		}
487 
488 		if (!(found->flags & DEVFS_HIDDEN))
489 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
490 	}
491 
492 	if (vp == NULL) {
493 		error = ENOENT;
494 		cache_setvp(ap->a_nch, NULL);
495 		goto out;
496 
497 	}
498 	KKASSERT(vp);
499 	vn_unlock(vp);
500 	cache_setvp(ap->a_nch, vp);
501 	vrele(vp);
502 out:
503 	lockmgr(&devfs_lock, LK_RELEASE);
504 
505 	return error;
506 }
507 
508 
509 static int
510 devfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
511 {
512 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
513 
514 	*ap->a_vpp = NULL;
515 	if (!devfs_node_is_accessible(dnode))
516 		return ENOENT;
517 
518 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
519 	if (dnode->parent != NULL) {
520 		devfs_allocv(ap->a_vpp, dnode->parent);
521 		vn_unlock(*ap->a_vpp);
522 	}
523 	lockmgr(&devfs_lock, LK_RELEASE);
524 
525 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
526 }
527 
528 
529 static int
530 devfs_getattr(struct vop_getattr_args *ap)
531 {
532 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
533 	struct vattr *vap = ap->a_vap;
534 	struct partinfo pinfo;
535 	int error = 0;
536 
537 #if 0
538 	if (!devfs_node_is_accessible(node))
539 		return ENOENT;
540 #endif
541 	node_sync_dev_get(node);
542 
543 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
544 
545 	/* start by zeroing out the attributes */
546 	VATTR_NULL(vap);
547 
548 	/* next do all the common fields */
549 	vap->va_type = ap->a_vp->v_type;
550 	vap->va_mode = node->mode;
551 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
552 	vap->va_flags = 0;
553 	vap->va_blocksize = DEV_BSIZE;
554 	vap->va_bytes = vap->va_size = 0;
555 
556 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
557 
558 	vap->va_atime = node->atime;
559 	vap->va_mtime = node->mtime;
560 	vap->va_ctime = node->ctime;
561 
562 	vap->va_nlink = 1; /* number of references to file */
563 
564 	vap->va_uid = node->uid;
565 	vap->va_gid = node->gid;
566 
567 	vap->va_rmajor = 0;
568 	vap->va_rminor = 0;
569 
570 	if ((node->node_type == Pdev) && node->d_dev)  {
571 		reference_dev(node->d_dev);
572 		vap->va_rminor = node->d_dev->si_uminor;
573 		release_dev(node->d_dev);
574 	}
575 
576 	/* For a softlink the va_size is the length of the softlink */
577 	if (node->symlink_name != 0) {
578 		vap->va_bytes = vap->va_size = node->symlink_namelen;
579 	}
580 
581 	/*
582 	 * For a disk-type device, va_size is the size of the underlying
583 	 * device, so that lseek() works properly.
584 	 */
585 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
586 		bzero(&pinfo, sizeof(pinfo));
587 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
588 				   0, proc0.p_ucred, NULL);
589 		if ((error == 0) && (pinfo.media_blksize != 0)) {
590 			vap->va_size = pinfo.media_size;
591 		} else {
592 			vap->va_size = 0;
593 			error = 0;
594 		}
595 	}
596 
597 	lockmgr(&devfs_lock, LK_RELEASE);
598 
599 	return (error);
600 }
601 
602 
603 static int
604 devfs_setattr(struct vop_setattr_args *ap)
605 {
606 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
607 	struct vattr *vap;
608 	int error = 0;
609 
610 	if (!devfs_node_is_accessible(node))
611 		return ENOENT;
612 	node_sync_dev_get(node);
613 
614 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
615 
616 	vap = ap->a_vap;
617 
618 	if (vap->va_uid != (uid_t)VNOVAL) {
619 		if ((ap->a_cred->cr_uid != node->uid) &&
620 		    (!groupmember(node->gid, ap->a_cred))) {
621 			error = priv_check(curthread, PRIV_VFS_CHOWN);
622 			if (error)
623 				goto out;
624 		}
625 		node->uid = vap->va_uid;
626 	}
627 
628 	if (vap->va_gid != (uid_t)VNOVAL) {
629 		if ((ap->a_cred->cr_uid != node->uid) &&
630 		    (!groupmember(node->gid, ap->a_cred))) {
631 			error = priv_check(curthread, PRIV_VFS_CHOWN);
632 			if (error)
633 				goto out;
634 		}
635 		node->gid = vap->va_gid;
636 	}
637 
638 	if (vap->va_mode != (mode_t)VNOVAL) {
639 		if (ap->a_cred->cr_uid != node->uid) {
640 			error = priv_check(curthread, PRIV_VFS_ADMIN);
641 			if (error)
642 				goto out;
643 		}
644 		node->mode = vap->va_mode;
645 	}
646 
647 out:
648 	node_sync_dev_set(node);
649 	nanotime(&node->ctime);
650 	lockmgr(&devfs_lock, LK_RELEASE);
651 
652 	return error;
653 }
654 
655 
656 static int
657 devfs_readlink(struct vop_readlink_args *ap)
658 {
659 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
660 	int ret;
661 
662 	if (!devfs_node_is_accessible(node))
663 		return ENOENT;
664 
665 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
666 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
667 	lockmgr(&devfs_lock, LK_RELEASE);
668 
669 	return ret;
670 }
671 
672 
673 static int
674 devfs_print(struct vop_print_args *ap)
675 {
676 	return (0);
677 }
678 
679 static int
680 devfs_nmkdir(struct vop_nmkdir_args *ap)
681 {
682 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
683 	struct devfs_node *node;
684 
685 	if (!devfs_node_is_accessible(dnode))
686 		return ENOENT;
687 
688 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
689 		goto out;
690 
691 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
692 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir,
693 		      ap->a_nch->ncp->nc_name, dnode, NULL);
694 
695 	if (*ap->a_vpp) {
696 		node = DEVFS_NODE(*ap->a_vpp);
697 		node->flags |= DEVFS_USER_CREATED;
698 		cache_setunresolved(ap->a_nch);
699 		cache_setvp(ap->a_nch, *ap->a_vpp);
700 	}
701 	lockmgr(&devfs_lock, LK_RELEASE);
702 out:
703 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
704 }
705 
706 static int
707 devfs_nsymlink(struct vop_nsymlink_args *ap)
708 {
709 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
710 	struct devfs_node *node;
711 	size_t targetlen;
712 
713 	if (!devfs_node_is_accessible(dnode))
714 		return ENOENT;
715 
716 	ap->a_vap->va_type = VLNK;
717 
718 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
719 		goto out;
720 
721 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
722 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
723 		      ap->a_nch->ncp->nc_name, dnode, NULL);
724 
725 	targetlen = strlen(ap->a_target);
726 	if (*ap->a_vpp) {
727 		node = DEVFS_NODE(*ap->a_vpp);
728 		node->flags |= DEVFS_USER_CREATED;
729 		node->symlink_namelen = targetlen;
730 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
731 		memcpy(node->symlink_name, ap->a_target, targetlen);
732 		node->symlink_name[targetlen] = '\0';
733 		cache_setunresolved(ap->a_nch);
734 		cache_setvp(ap->a_nch, *ap->a_vpp);
735 	}
736 	lockmgr(&devfs_lock, LK_RELEASE);
737 out:
738 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
739 }
740 
741 static int
742 devfs_nrmdir(struct vop_nrmdir_args *ap)
743 {
744 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
745 	struct devfs_node *node;
746 	struct namecache *ncp;
747 	int error = ENOENT;
748 
749 	ncp = ap->a_nch->ncp;
750 
751 	if (!devfs_node_is_accessible(dnode))
752 		return ENOENT;
753 
754 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
755 
756 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
757 		goto out;
758 
759 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
760 		if (ncp->nc_nlen != node->d_dir.d_namlen)
761 			continue;
762 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
763 			continue;
764 
765 		/*
766 		 * only allow removal of user created dirs
767 		 */
768 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
769 			error = EPERM;
770 			goto out;
771 		} else if (node->node_type != Pdir) {
772 			error = ENOTDIR;
773 			goto out;
774 		} else if (node->nchildren > 2) {
775 			error = ENOTEMPTY;
776 			goto out;
777 		} else {
778 			if (node->v_node)
779 				cache_inval_vp(node->v_node, CINV_DESTROY);
780 			devfs_unlinkp(node);
781 			error = 0;
782 			break;
783 		}
784 	}
785 
786 	cache_setunresolved(ap->a_nch);
787 	cache_setvp(ap->a_nch, NULL);
788 
789 out:
790 	lockmgr(&devfs_lock, LK_RELEASE);
791 	return error;
792 }
793 
794 static int
795 devfs_nremove(struct vop_nremove_args *ap)
796 {
797 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
798 	struct devfs_node *node;
799 	struct namecache *ncp;
800 	int error = ENOENT;
801 
802 	ncp = ap->a_nch->ncp;
803 
804 	if (!devfs_node_is_accessible(dnode))
805 		return ENOENT;
806 
807 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
808 
809 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
810 		goto out;
811 
812 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
813 		if (ncp->nc_nlen != node->d_dir.d_namlen)
814 			continue;
815 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
816 			continue;
817 
818 		/*
819 		 * only allow removal of user created stuff (e.g. symlinks)
820 		 */
821 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
822 			error = EPERM;
823 			goto out;
824 		} else if (node->node_type == Pdir) {
825 			error = EISDIR;
826 			goto out;
827 		} else {
828 			if (node->v_node)
829 				cache_inval_vp(node->v_node, CINV_DESTROY);
830 			devfs_unlinkp(node);
831 			error = 0;
832 			break;
833 		}
834 	}
835 
836 	cache_setunresolved(ap->a_nch);
837 	cache_setvp(ap->a_nch, NULL);
838 
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 	if (node && ap->a_fp) {
866 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
867 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
868 
869 		ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
870 						ap->a_mode, ap->a_cred);
871 		if (ndev != NULL) {
872 			newnode = devfs_create_device_node(
873 					DEVFS_MNTDATA(vp->v_mount)->root_node,
874 					ndev, NULL, NULL);
875 			/* XXX: possibly destroy device if this happens */
876 
877 			if (newnode != NULL) {
878 				dev = ndev;
879 				devfs_link_dev(dev);
880 
881 				devfs_debug(DEVFS_DEBUG_DEBUG,
882 						"parent here is: %s, node is: |%s|\n",
883 						((node->parent->node_type == Proot) ?
884 						"ROOT!" : node->parent->d_dir.d_name),
885 						newnode->d_dir.d_name);
886 				devfs_debug(DEVFS_DEBUG_DEBUG,
887 						"test: %s\n",
888 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
889 
890 				/*
891 				 * orig_vp is set to the original vp if we cloned.
892 				 */
893 				/* node->flags |= DEVFS_CLONED; */
894 				devfs_allocv(&vp, newnode);
895 				orig_vp = ap->a_vp;
896 				ap->a_vp = vp;
897 			}
898 		}
899 		lockmgr(&devfs_lock, LK_RELEASE);
900 	}
901 
902 	devfs_debug(DEVFS_DEBUG_DEBUG,
903 		    "devfs_spec_open() called on %s! \n",
904 		    dev->si_name);
905 
906 	/*
907 	 * Make this field valid before any I/O in ->d_open
908 	 */
909 	if (!dev->si_iosize_max)
910 		dev->si_iosize_max = DFLTPHYS;
911 
912 	if (dev_dflags(dev) & D_TTY)
913 		vsetflags(vp, VISTTY);
914 
915 	vn_unlock(vp);
916 	error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
917 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
918 
919 	/*
920 	 * Clean up any cloned vp if we error out.
921 	 */
922 	if (error) {
923 		if (orig_vp) {
924 			vput(vp);
925 			ap->a_vp = orig_vp;
926 			/* orig_vp = NULL; */
927 		}
928 		return error;
929 	}
930 
931 	/*
932 	 * This checks if the disk device is going to be opened for writing.
933 	 * It will be only allowed in the cases where securelevel permits it
934 	 * and it's not mounted R/W.
935 	 */
936 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
937 	    (ap->a_cred != FSCRED)) {
938 
939 		/* Very secure mode. No open for writing allowed */
940 		if (securelevel >= 2)
941 			return EPERM;
942 
943 		/*
944 		 * If it is mounted R/W, do not allow to open for writing.
945 		 * In the case it's mounted read-only but securelevel
946 		 * is >= 1, then do not allow opening for writing either.
947 		 */
948 		if (vfs_mountedon(vp)) {
949 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
950 				return EBUSY;
951 			else if (securelevel >= 1)
952 				return EPERM;
953 		}
954 	}
955 
956 	if (dev_dflags(dev) & D_TTY) {
957 		if (dev->si_tty) {
958 			struct tty *tp;
959 			tp = dev->si_tty;
960 			if (!tp->t_stop) {
961 				devfs_debug(DEVFS_DEBUG_DEBUG,
962 					    "devfs: no t_stop\n");
963 				tp->t_stop = nottystop;
964 			}
965 		}
966 	}
967 
968 
969 	if (vn_isdisk(vp, NULL)) {
970 		if (!dev->si_bsize_phys)
971 			dev->si_bsize_phys = DEV_BSIZE;
972 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
973 	}
974 
975 	vop_stdopen(ap);
976 #if 0
977 	if (node)
978 		nanotime(&node->atime);
979 #endif
980 
981 	if (orig_vp)
982 		vn_unlock(vp);
983 
984 	/* Ugly pty magic, to make pty devices appear once they are opened */
985 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
986 		node->flags &= ~DEVFS_INVISIBLE;
987 
988 	if (ap->a_fp) {
989 		ap->a_fp->f_type = DTYPE_VNODE;
990 		ap->a_fp->f_flag = ap->a_mode & FMASK;
991 		ap->a_fp->f_ops = &devfs_dev_fileops;
992 		ap->a_fp->f_data = vp;
993 	}
994 
995 	return 0;
996 }
997 
998 
999 static int
1000 devfs_spec_close(struct vop_close_args *ap)
1001 {
1002 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
1003 	struct proc *p = curproc;
1004 	struct vnode *vp = ap->a_vp;
1005 	cdev_t dev = vp->v_rdev;
1006 	int error = 0;
1007 	int needrelock;
1008 
1009 	devfs_debug(DEVFS_DEBUG_DEBUG,
1010 		    "devfs_spec_close() called on %s! \n",
1011 		    dev->si_name);
1012 
1013 	/*
1014 	 * A couple of hacks for devices and tty devices.  The
1015 	 * vnode ref count cannot be used to figure out the
1016 	 * last close, but we can use v_opencount now that
1017 	 * revoke works properly.
1018 	 *
1019 	 * Detect the last close on a controlling terminal and clear
1020 	 * the session (half-close).
1021 	 */
1022 	if (dev)
1023 		reference_dev(dev);
1024 
1025 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1026 		p->p_session->s_ttyvp = NULL;
1027 		vrele(vp);
1028 	}
1029 
1030 	/*
1031 	 * Vnodes can be opened and closed multiple times.  Do not really
1032 	 * close the device unless (1) it is being closed forcibly,
1033 	 * (2) the device wants to track closes, or (3) this is the last
1034 	 * vnode doing its last close on the device.
1035 	 *
1036 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1037 	 * a closed device.  This might not occur now that our revoke is
1038 	 * fixed.
1039 	 */
1040 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1041 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1042 	    (dev_dflags(dev) & D_TRACKCLOSE) ||
1043 	    (vp->v_opencount == 1))) {
1044 		/*
1045 		 * Unlock around dev_dclose()
1046 		 */
1047 		needrelock = 0;
1048 		if (vn_islocked(vp)) {
1049 			needrelock = 1;
1050 			vn_unlock(vp);
1051 		}
1052 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1053 
1054 		/*
1055 		 * Ugly pty magic, to make pty devices disappear again once
1056 		 * they are closed
1057 		 */
1058 		if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
1059 			node->flags |= DEVFS_INVISIBLE;
1060 
1061 		if (needrelock)
1062 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1063 	} else {
1064 		error = 0;
1065 	}
1066 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1067 
1068 	/*
1069 	 * Track the actual opens and closes on the vnode.  The last close
1070 	 * disassociates the rdev.  If the rdev is already disassociated or
1071 	 * the opencount is already 0, the vnode might have been revoked
1072 	 * and no further opencount tracking occurs.
1073 	 */
1074 	if (dev)
1075 		release_dev(dev);
1076 	if (vp->v_opencount > 0)
1077 		vop_stdclose(ap);
1078 	return(error);
1079 
1080 }
1081 
1082 
1083 static int
1084 devfs_specf_close(struct file *fp)
1085 {
1086 	struct vnode *vp = (struct vnode *)fp->f_data;
1087 	int error;
1088 
1089 	get_mplock();
1090 	fp->f_ops = &badfileops;
1091 	error = vn_close(vp, fp->f_flag);
1092 	rel_mplock();
1093 
1094 	return (error);
1095 }
1096 
1097 
1098 /*
1099  * Device-optimized file table vnode read routine.
1100  *
1101  * This bypasses the VOP table and talks directly to the device.  Most
1102  * filesystems just route to specfs and can make this optimization.
1103  *
1104  * MPALMOSTSAFE - acquires mplock
1105  */
1106 static int
1107 devfs_specf_read(struct file *fp, struct uio *uio,
1108 		 struct ucred *cred, int flags)
1109 {
1110 	struct devfs_node *node;
1111 	struct vnode *vp;
1112 	int ioflag;
1113 	int error;
1114 	cdev_t dev;
1115 
1116 	KASSERT(uio->uio_td == curthread,
1117 		("uio_td %p is not td %p", uio->uio_td, curthread));
1118 
1119 	if (uio->uio_resid == 0)
1120 		return 0;
1121 
1122 	vp = (struct vnode *)fp->f_data;
1123 	if (vp == NULL || vp->v_type == VBAD)
1124 		return EBADF;
1125 
1126 	node = DEVFS_NODE(vp);
1127 
1128 	if ((dev = vp->v_rdev) == NULL)
1129 		return EBADF;
1130 
1131 	/* only acquire mplock for devices that require it */
1132 	if (!(dev_dflags(dev) & D_MPSAFE_READ)) {
1133 		atomic_add_int(&mplock_reads, 1);
1134 		get_mplock();
1135 	} else {
1136 		atomic_add_int(&mpsafe_reads, 1);
1137 	}
1138 
1139 	reference_dev(dev);
1140 
1141 	if ((flags & O_FOFFSET) == 0)
1142 		uio->uio_offset = fp->f_offset;
1143 
1144 	ioflag = 0;
1145 	if (flags & O_FBLOCKING) {
1146 		/* ioflag &= ~IO_NDELAY; */
1147 	} else if (flags & O_FNONBLOCKING) {
1148 		ioflag |= IO_NDELAY;
1149 	} else if (fp->f_flag & FNONBLOCK) {
1150 		ioflag |= IO_NDELAY;
1151 	}
1152 	if (flags & O_FBUFFERED) {
1153 		/* ioflag &= ~IO_DIRECT; */
1154 	} else if (flags & O_FUNBUFFERED) {
1155 		ioflag |= IO_DIRECT;
1156 	} else if (fp->f_flag & O_DIRECT) {
1157 		ioflag |= IO_DIRECT;
1158 	}
1159 	ioflag |= sequential_heuristic(uio, fp);
1160 
1161 	error = dev_dread(dev, uio, ioflag);
1162 
1163 	release_dev(dev);
1164 	if (node)
1165 		nanotime(&node->atime);
1166 	if ((flags & O_FOFFSET) == 0)
1167 		fp->f_offset = uio->uio_offset;
1168 	fp->f_nextoff = uio->uio_offset;
1169 
1170 	if (!(dev_dflags(dev) & D_MPSAFE_READ))
1171 		rel_mplock();
1172 
1173 	return (error);
1174 }
1175 
1176 
1177 static int
1178 devfs_specf_write(struct file *fp, struct uio *uio,
1179 		  struct ucred *cred, int flags)
1180 {
1181 	struct devfs_node *node;
1182 	struct vnode *vp;
1183 	int ioflag;
1184 	int error;
1185 	cdev_t dev;
1186 
1187 	KASSERT(uio->uio_td == curthread,
1188 		("uio_td %p is not p %p", uio->uio_td, curthread));
1189 
1190 	vp = (struct vnode *)fp->f_data;
1191 	if (vp == NULL || vp->v_type == VBAD)
1192 		return EBADF;
1193 
1194 	node = DEVFS_NODE(vp);
1195 
1196 	if (vp->v_type == VREG)
1197 		bwillwrite(uio->uio_resid);
1198 
1199 	vp = (struct vnode *)fp->f_data;
1200 
1201 	if ((dev = vp->v_rdev) == NULL)
1202 		return EBADF;
1203 
1204 	/* only acquire mplock for devices that require it */
1205 	if (!(dev_dflags(dev) & D_MPSAFE_WRITE)) {
1206 		atomic_add_int(&mplock_writes, 1);
1207 		get_mplock();
1208 	} else {
1209 		atomic_add_int(&mpsafe_writes, 1);
1210 	}
1211 
1212 	reference_dev(dev);
1213 
1214 	if ((flags & O_FOFFSET) == 0)
1215 		uio->uio_offset = fp->f_offset;
1216 
1217 	ioflag = IO_UNIT;
1218 	if (vp->v_type == VREG &&
1219 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1220 		ioflag |= IO_APPEND;
1221 	}
1222 
1223 	if (flags & O_FBLOCKING) {
1224 		/* ioflag &= ~IO_NDELAY; */
1225 	} else if (flags & O_FNONBLOCKING) {
1226 		ioflag |= IO_NDELAY;
1227 	} else if (fp->f_flag & FNONBLOCK) {
1228 		ioflag |= IO_NDELAY;
1229 	}
1230 	if (flags & O_FBUFFERED) {
1231 		/* ioflag &= ~IO_DIRECT; */
1232 	} else if (flags & O_FUNBUFFERED) {
1233 		ioflag |= IO_DIRECT;
1234 	} else if (fp->f_flag & O_DIRECT) {
1235 		ioflag |= IO_DIRECT;
1236 	}
1237 	if (flags & O_FASYNCWRITE) {
1238 		/* ioflag &= ~IO_SYNC; */
1239 	} else if (flags & O_FSYNCWRITE) {
1240 		ioflag |= IO_SYNC;
1241 	} else if (fp->f_flag & O_FSYNC) {
1242 		ioflag |= IO_SYNC;
1243 	}
1244 
1245 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1246 		ioflag |= IO_SYNC;
1247 	ioflag |= sequential_heuristic(uio, fp);
1248 
1249 	error = dev_dwrite(dev, uio, ioflag);
1250 
1251 	release_dev(dev);
1252 	if (node) {
1253 		nanotime(&node->atime);
1254 		nanotime(&node->mtime);
1255 	}
1256 
1257 	if ((flags & O_FOFFSET) == 0)
1258 		fp->f_offset = uio->uio_offset;
1259 	fp->f_nextoff = uio->uio_offset;
1260 
1261 	if (!(dev_dflags(dev) & D_MPSAFE_WRITE))
1262 		rel_mplock();
1263 	return (error);
1264 }
1265 
1266 
1267 static int
1268 devfs_specf_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1269 {
1270 	struct vnode *vp;
1271 	struct vattr vattr;
1272 	struct vattr *vap;
1273 	u_short mode;
1274 	cdev_t dev;
1275 	int error;
1276 
1277 	vp = (struct vnode *)fp->f_data;
1278 	if (vp == NULL || vp->v_type == VBAD)
1279 		return EBADF;
1280 
1281 	error = vn_stat(vp, sb, cred);
1282 	if (error)
1283 		return (error);
1284 
1285 	vap = &vattr;
1286 	error = VOP_GETATTR(vp, vap);
1287 	if (error)
1288 		return (error);
1289 
1290 	/*
1291 	 * Zero the spare stat fields
1292 	 */
1293 	sb->st_lspare = 0;
1294 	sb->st_qspare1 = 0;
1295 	sb->st_qspare2 = 0;
1296 
1297 	/*
1298 	 * Copy from vattr table ... or not in case it's a cloned device
1299 	 */
1300 	if (vap->va_fsid != VNOVAL)
1301 		sb->st_dev = vap->va_fsid;
1302 	else
1303 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1304 
1305 	sb->st_ino = vap->va_fileid;
1306 
1307 	mode = vap->va_mode;
1308 	mode |= S_IFCHR;
1309 	sb->st_mode = mode;
1310 
1311 	if (vap->va_nlink > (nlink_t)-1)
1312 		sb->st_nlink = (nlink_t)-1;
1313 	else
1314 		sb->st_nlink = vap->va_nlink;
1315 
1316 	sb->st_uid = vap->va_uid;
1317 	sb->st_gid = vap->va_gid;
1318 	sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1319 	sb->st_size = vap->va_bytes;
1320 	sb->st_atimespec = vap->va_atime;
1321 	sb->st_mtimespec = vap->va_mtime;
1322 	sb->st_ctimespec = vap->va_ctime;
1323 
1324 	/*
1325 	 * A VCHR and VBLK device may track the last access and last modified
1326 	 * time independantly of the filesystem.  This is particularly true
1327 	 * because device read and write calls may bypass the filesystem.
1328 	 */
1329 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1330 		dev = vp->v_rdev;
1331 		if (dev != NULL) {
1332 			if (dev->si_lastread) {
1333 				sb->st_atimespec.tv_sec = dev->si_lastread;
1334 				sb->st_atimespec.tv_nsec = 0;
1335 			}
1336 			if (dev->si_lastwrite) {
1337 				sb->st_atimespec.tv_sec = dev->si_lastwrite;
1338 				sb->st_atimespec.tv_nsec = 0;
1339 			}
1340 		}
1341 	}
1342 
1343         /*
1344 	 * According to www.opengroup.org, the meaning of st_blksize is
1345 	 *   "a filesystem-specific preferred I/O block size for this
1346 	 *    object.  In some filesystem types, this may vary from file
1347 	 *    to file"
1348 	 * Default to PAGE_SIZE after much discussion.
1349 	 */
1350 
1351 	sb->st_blksize = PAGE_SIZE;
1352 
1353 	sb->st_flags = vap->va_flags;
1354 
1355 	error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1356 	if (error)
1357 		sb->st_gen = 0;
1358 	else
1359 		sb->st_gen = (u_int32_t)vap->va_gen;
1360 
1361 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1362 
1363 	return (0);
1364 }
1365 
1366 
1367 static int
1368 devfs_specf_kqfilter(struct file *fp, struct knote *kn)
1369 {
1370 	struct vnode *vp;
1371 	int error;
1372 	cdev_t dev;
1373 
1374 	get_mplock();
1375 
1376 	vp = (struct vnode *)fp->f_data;
1377 	if (vp == NULL || vp->v_type == VBAD) {
1378 		error = EBADF;
1379 		goto done;
1380 	}
1381 	if ((dev = vp->v_rdev) == NULL) {
1382 		error = EBADF;
1383 		goto done;
1384 	}
1385 	reference_dev(dev);
1386 
1387 	error = dev_dkqfilter(dev, kn);
1388 
1389 	release_dev(dev);
1390 
1391 done:
1392 	rel_mplock();
1393 	return (error);
1394 }
1395 
1396 
1397 static int
1398 devfs_specf_poll(struct file *fp, int events, struct ucred *cred)
1399 {
1400 	struct devfs_node *node;
1401 	struct vnode *vp;
1402 	int error;
1403 	cdev_t dev;
1404 
1405 	get_mplock();
1406 
1407 	vp = (struct vnode *)fp->f_data;
1408 	if (vp == NULL || vp->v_type == VBAD) {
1409 		error = EBADF;
1410 		goto done;
1411 	}
1412 	node = DEVFS_NODE(vp);
1413 
1414 	if ((dev = vp->v_rdev) == NULL) {
1415 		error = EBADF;
1416 		goto done;
1417 	}
1418 	reference_dev(dev);
1419 	error = dev_dpoll(dev, events);
1420 
1421 	release_dev(dev);
1422 
1423 #if 0
1424 	if (node)
1425 		nanotime(&node->atime);
1426 #endif
1427 done:
1428 	rel_mplock();
1429 	return (error);
1430 }
1431 
1432 
1433 /*
1434  * MPALMOSTSAFE - acquires mplock
1435  */
1436 static int
1437 devfs_specf_ioctl(struct file *fp, u_long com, caddr_t data,
1438 		  struct ucred *ucred, struct sysmsg *msg)
1439 {
1440 	struct devfs_node *node;
1441 	struct vnode *vp;
1442 	struct vnode *ovp;
1443 	cdev_t	dev;
1444 	int error;
1445 	struct fiodname_args *name_args;
1446 	size_t namlen;
1447 	const char *name;
1448 
1449 	vp = ((struct vnode *)fp->f_data);
1450 
1451 	if ((dev = vp->v_rdev) == NULL)
1452 		return EBADF;		/* device was revoked */
1453 
1454 	reference_dev(dev);
1455 
1456 	node = DEVFS_NODE(vp);
1457 
1458 	devfs_debug(DEVFS_DEBUG_DEBUG,
1459 		    "devfs_specf_ioctl() called! for dev %s\n",
1460 		    dev->si_name);
1461 
1462 	if (com == FIODTYPE) {
1463 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1464 		error = 0;
1465 		goto out;
1466 	} else if (com == FIODNAME) {
1467 		name_args = (struct fiodname_args *)data;
1468 		name = dev->si_name;
1469 		namlen = strlen(name) + 1;
1470 
1471 		devfs_debug(DEVFS_DEBUG_DEBUG,
1472 			    "ioctl, got: FIODNAME for %s\n", name);
1473 
1474 		if (namlen <= name_args->len)
1475 			error = copyout(dev->si_name, name_args->name, namlen);
1476 		else
1477 			error = EINVAL;
1478 
1479 		devfs_debug(DEVFS_DEBUG_DEBUG,
1480 			    "ioctl stuff: error: %d\n", error);
1481 		goto out;
1482 	}
1483 
1484 	/* only acquire mplock for devices that require it */
1485 	if (!(dev_dflags(dev) & D_MPSAFE_IOCTL))
1486 		get_mplock();
1487 
1488 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
1489 
1490 #if 0
1491 	if (node) {
1492 		nanotime(&node->atime);
1493 		nanotime(&node->mtime);
1494 	}
1495 #endif
1496 
1497 	if (!(dev_dflags(dev) & D_MPSAFE_IOCTL))
1498 		rel_mplock();
1499 
1500 	if (com == TIOCSCTTY) {
1501 		devfs_debug(DEVFS_DEBUG_DEBUG,
1502 			    "devfs_specf_ioctl: got TIOCSCTTY on %s\n",
1503 			    dev->si_name);
1504 	}
1505 	if (error == 0 && com == TIOCSCTTY) {
1506 		struct proc *p = curthread->td_proc;
1507 		struct session *sess;
1508 
1509 		devfs_debug(DEVFS_DEBUG_DEBUG,
1510 			    "devfs_specf_ioctl: dealing with TIOCSCTTY on %s\n",
1511 			    dev->si_name);
1512 		if (p == NULL) {
1513 			error = ENOTTY;
1514 			goto out;
1515 		}
1516 		sess = p->p_session;
1517 
1518 		/*
1519 		 * Do nothing if reassigning same control tty
1520 		 */
1521 		if (sess->s_ttyvp == vp) {
1522 			error = 0;
1523 			goto out;
1524 		}
1525 
1526 		/*
1527 		 * Get rid of reference to old control tty
1528 		 */
1529 		ovp = sess->s_ttyvp;
1530 		vref(vp);
1531 		sess->s_ttyvp = vp;
1532 		if (ovp)
1533 			vrele(ovp);
1534 	}
1535 
1536 out:
1537 	release_dev(dev);
1538 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_specf_ioctl() finished! \n");
1539 	return (error);
1540 }
1541 
1542 
1543 static int
1544 devfs_spec_fsync(struct vop_fsync_args *ap)
1545 {
1546 	struct vnode *vp = ap->a_vp;
1547 	int error;
1548 
1549 	if (!vn_isdisk(vp, NULL))
1550 		return (0);
1551 
1552 	/*
1553 	 * Flush all dirty buffers associated with a block device.
1554 	 */
1555 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1556 	return (error);
1557 }
1558 
1559 static int
1560 devfs_spec_read(struct vop_read_args *ap)
1561 {
1562 	struct devfs_node *node;
1563 	struct vnode *vp;
1564 	struct uio *uio;
1565 	cdev_t dev;
1566 	int error;
1567 
1568 	vp = ap->a_vp;
1569 	dev = vp->v_rdev;
1570 	uio = ap->a_uio;
1571 	node = DEVFS_NODE(vp);
1572 
1573 	if (dev == NULL)		/* device was revoked */
1574 		return (EBADF);
1575 	if (uio->uio_resid == 0)
1576 		return (0);
1577 
1578 	vn_unlock(vp);
1579 	error = dev_dread(dev, uio, ap->a_ioflag);
1580 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1581 
1582 	if (node)
1583 		nanotime(&node->atime);
1584 
1585 	return (error);
1586 }
1587 
1588 /*
1589  * Vnode op for write
1590  *
1591  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1592  *	      struct ucred *a_cred)
1593  */
1594 static int
1595 devfs_spec_write(struct vop_write_args *ap)
1596 {
1597 	struct devfs_node *node;
1598 	struct vnode *vp;
1599 	struct uio *uio;
1600 	cdev_t dev;
1601 	int error;
1602 
1603 	vp = ap->a_vp;
1604 	dev = vp->v_rdev;
1605 	uio = ap->a_uio;
1606 	node = DEVFS_NODE(vp);
1607 
1608 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1609 
1610 	if (dev == NULL)		/* device was revoked */
1611 		return (EBADF);
1612 
1613 	vn_unlock(vp);
1614 	error = dev_dwrite(dev, uio, ap->a_ioflag);
1615 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1616 
1617 	if (node) {
1618 		nanotime(&node->atime);
1619 		nanotime(&node->mtime);
1620 	}
1621 
1622 	return (error);
1623 }
1624 
1625 /*
1626  * Device ioctl operation.
1627  *
1628  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1629  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1630  */
1631 static int
1632 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1633 {
1634 	struct vnode *vp = ap->a_vp;
1635 	struct devfs_node *node;
1636 	cdev_t dev;
1637 
1638 	if ((dev = vp->v_rdev) == NULL)
1639 		return (EBADF);		/* device was revoked */
1640 	node = DEVFS_NODE(vp);
1641 
1642 #if 0
1643 	if (node) {
1644 		nanotime(&node->atime);
1645 		nanotime(&node->mtime);
1646 	}
1647 #endif
1648 
1649 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1650 			   ap->a_cred, ap->a_sysmsg));
1651 }
1652 
1653 /*
1654  * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
1655  */
1656 /* ARGSUSED */
1657 static int
1658 devfs_spec_poll(struct vop_poll_args *ap)
1659 {
1660 	struct vnode *vp = ap->a_vp;
1661 	struct devfs_node *node;
1662 	cdev_t dev;
1663 
1664 	if ((dev = vp->v_rdev) == NULL)
1665 		return (EBADF);		/* device was revoked */
1666 	node = DEVFS_NODE(vp);
1667 
1668 #if 0
1669 	if (node)
1670 		nanotime(&node->atime);
1671 #endif
1672 
1673 	return (dev_dpoll(dev, ap->a_events));
1674 }
1675 
1676 /*
1677  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1678  */
1679 /* ARGSUSED */
1680 static int
1681 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1682 {
1683 	struct vnode *vp = ap->a_vp;
1684 	struct devfs_node *node;
1685 	cdev_t dev;
1686 
1687 	if ((dev = vp->v_rdev) == NULL)
1688 		return (EBADF);		/* device was revoked */
1689 	node = DEVFS_NODE(vp);
1690 
1691 #if 0
1692 	if (node)
1693 		nanotime(&node->atime);
1694 #endif
1695 
1696 	return (dev_dkqfilter(dev, ap->a_kn));
1697 }
1698 
1699 /*
1700  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1701  * calls are not limited to device DMA limits so we have to deal with the
1702  * case.
1703  *
1704  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1705  */
1706 static int
1707 devfs_spec_strategy(struct vop_strategy_args *ap)
1708 {
1709 	struct bio *bio = ap->a_bio;
1710 	struct buf *bp = bio->bio_buf;
1711 	struct buf *nbp;
1712 	struct vnode *vp;
1713 	struct mount *mp;
1714 	int chunksize;
1715 	int maxiosize;
1716 
1717 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1718 		buf_start(bp);
1719 
1720 	/*
1721 	 * Collect statistics on synchronous and asynchronous read
1722 	 * and write counts for disks that have associated filesystems.
1723 	 */
1724 	vp = ap->a_vp;
1725 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1726 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1727 		if (bp->b_cmd == BUF_CMD_READ) {
1728 			if (bp->b_flags & BIO_SYNC)
1729 				mp->mnt_stat.f_syncreads++;
1730 			else
1731 				mp->mnt_stat.f_asyncreads++;
1732 		} else {
1733 			if (bp->b_flags & BIO_SYNC)
1734 				mp->mnt_stat.f_syncwrites++;
1735 			else
1736 				mp->mnt_stat.f_asyncwrites++;
1737 		}
1738 	}
1739 
1740         /*
1741          * Device iosize limitations only apply to read and write.  Shortcut
1742          * the I/O if it fits.
1743          */
1744 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1745 		devfs_debug(DEVFS_DEBUG_DEBUG,
1746 			    "%s: si_iosize_max not set!\n",
1747 			    dev_dname(vp->v_rdev));
1748 		maxiosize = MAXPHYS;
1749 	}
1750 #if SPEC_CHAIN_DEBUG & 2
1751 	maxiosize = 4096;
1752 #endif
1753         if (bp->b_bcount <= maxiosize ||
1754             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1755                 dev_dstrategy_chain(vp->v_rdev, bio);
1756                 return (0);
1757         }
1758 
1759 	/*
1760 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1761 	 */
1762 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1763 	initbufbio(nbp);
1764 	buf_dep_init(nbp);
1765 	BUF_LOCKINIT(nbp);
1766 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1767 	BUF_KERNPROC(nbp);
1768 	nbp->b_vp = vp;
1769 	nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1770 	nbp->b_data = bp->b_data;
1771 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1772 	nbp->b_bio1.bio_offset = bio->bio_offset;
1773 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1774 
1775 	/*
1776 	 * Start the first transfer
1777 	 */
1778 	if (vn_isdisk(vp, NULL))
1779 		chunksize = vp->v_rdev->si_bsize_phys;
1780 	else
1781 		chunksize = DEV_BSIZE;
1782 	chunksize = maxiosize / chunksize * chunksize;
1783 #if SPEC_CHAIN_DEBUG & 1
1784 	devfs_debug(DEVFS_DEBUG_DEBUG,
1785 		    "spec_strategy chained I/O chunksize=%d\n",
1786 		    chunksize);
1787 #endif
1788 	nbp->b_cmd = bp->b_cmd;
1789 	nbp->b_bcount = chunksize;
1790 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1791 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1792 
1793 #if SPEC_CHAIN_DEBUG & 1
1794 	devfs_debug(DEVFS_DEBUG_DEBUG,
1795 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1796 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1797 #endif
1798 
1799 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1800 
1801 	if (DEVFS_NODE(vp)) {
1802 		nanotime(&DEVFS_NODE(vp)->atime);
1803 		nanotime(&DEVFS_NODE(vp)->mtime);
1804 	}
1805 
1806 	return (0);
1807 }
1808 
1809 /*
1810  * Chunked up transfer completion routine - chain transfers until done
1811  */
1812 static
1813 void
1814 devfs_spec_strategy_done(struct bio *nbio)
1815 {
1816 	struct buf *nbp = nbio->bio_buf;
1817 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1818 	struct buf *bp = bio->bio_buf;			/* original bp */
1819 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1820 	int boffset = nbp->b_data - bp->b_data;
1821 
1822 	if (nbp->b_flags & B_ERROR) {
1823 		/*
1824 		 * An error terminates the chain, propogate the error back
1825 		 * to the original bp
1826 		 */
1827 		bp->b_flags |= B_ERROR;
1828 		bp->b_error = nbp->b_error;
1829 		bp->b_resid = bp->b_bcount - boffset +
1830 			      (nbp->b_bcount - nbp->b_resid);
1831 #if SPEC_CHAIN_DEBUG & 1
1832 		devfs_debug(DEVFS_DEBUG_DEBUG,
1833 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1834 			    bp, bp->b_error, bp->b_bcount,
1835 			    bp->b_bcount - bp->b_resid);
1836 #endif
1837 		kfree(nbp, M_DEVBUF);
1838 		biodone(bio);
1839 	} else if (nbp->b_resid) {
1840 		/*
1841 		 * A short read or write terminates the chain
1842 		 */
1843 		bp->b_error = nbp->b_error;
1844 		bp->b_resid = bp->b_bcount - boffset +
1845 			      (nbp->b_bcount - nbp->b_resid);
1846 #if SPEC_CHAIN_DEBUG & 1
1847 		devfs_debug(DEVFS_DEBUG_DEBUG,
1848 			    "spec_strategy: chain %p short read(1) "
1849 			    "bcount %d/%d\n",
1850 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1851 #endif
1852 		kfree(nbp, M_DEVBUF);
1853 		biodone(bio);
1854 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1855 		/*
1856 		 * A short read or write can also occur by truncating b_bcount
1857 		 */
1858 #if SPEC_CHAIN_DEBUG & 1
1859 		devfs_debug(DEVFS_DEBUG_DEBUG,
1860 			    "spec_strategy: chain %p short read(2) "
1861 			    "bcount %d/%d\n",
1862 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1863 #endif
1864 		bp->b_error = 0;
1865 		bp->b_bcount = nbp->b_bcount + boffset;
1866 		bp->b_resid = nbp->b_resid;
1867 		kfree(nbp, M_DEVBUF);
1868 		biodone(bio);
1869 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1870 		/*
1871 		 * No more data terminates the chain
1872 		 */
1873 #if SPEC_CHAIN_DEBUG & 1
1874 		devfs_debug(DEVFS_DEBUG_DEBUG,
1875 			    "spec_strategy: chain %p finished bcount %d\n",
1876 			    bp, bp->b_bcount);
1877 #endif
1878 		bp->b_error = 0;
1879 		bp->b_resid = 0;
1880 		kfree(nbp, M_DEVBUF);
1881 		biodone(bio);
1882 	} else {
1883 		/*
1884 		 * Continue the chain
1885 		 */
1886 		boffset += nbp->b_bcount;
1887 		nbp->b_data = bp->b_data + boffset;
1888 		nbp->b_bcount = bp->b_bcount - boffset;
1889 		if (nbp->b_bcount > chunksize)
1890 			nbp->b_bcount = chunksize;
1891 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1892 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1893 
1894 #if SPEC_CHAIN_DEBUG & 1
1895 		devfs_debug(DEVFS_DEBUG_DEBUG,
1896 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1897 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
1898 #endif
1899 
1900 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1901 	}
1902 }
1903 
1904 /*
1905  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1906  */
1907 static int
1908 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1909 {
1910 	struct buf *bp;
1911 
1912 	/*
1913 	 * XXX: This assumes that strategy does the deed right away.
1914 	 * XXX: this may not be TRTTD.
1915 	 */
1916 	KKASSERT(ap->a_vp->v_rdev != NULL);
1917 	if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
1918 		return (0);
1919 	bp = geteblk(ap->a_length);
1920 	bp->b_cmd = BUF_CMD_FREEBLKS;
1921 	bp->b_bio1.bio_offset = ap->a_offset;
1922 	bp->b_bcount = ap->a_length;
1923 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1924 	return (0);
1925 }
1926 
1927 /*
1928  * Implement degenerate case where the block requested is the block
1929  * returned, and assume that the entire device is contiguous in regards
1930  * to the contiguous block range (runp and runb).
1931  *
1932  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1933  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
1934  */
1935 static int
1936 devfs_spec_bmap(struct vop_bmap_args *ap)
1937 {
1938 	if (ap->a_doffsetp != NULL)
1939 		*ap->a_doffsetp = ap->a_loffset;
1940 	if (ap->a_runp != NULL)
1941 		*ap->a_runp = MAXBSIZE;
1942 	if (ap->a_runb != NULL) {
1943 		if (ap->a_loffset < MAXBSIZE)
1944 			*ap->a_runb = (int)ap->a_loffset;
1945 		else
1946 			*ap->a_runb = MAXBSIZE;
1947 	}
1948 	return (0);
1949 }
1950 
1951 
1952 /*
1953  * Special device advisory byte-level locks.
1954  *
1955  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1956  *		struct flock *a_fl, int a_flags)
1957  */
1958 /* ARGSUSED */
1959 static int
1960 devfs_spec_advlock(struct vop_advlock_args *ap)
1961 {
1962 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1963 }
1964 
1965 static void
1966 devfs_spec_getpages_iodone(struct bio *bio)
1967 {
1968 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
1969 	wakeup(bio->bio_buf);
1970 }
1971 
1972 /*
1973  * spec_getpages() - get pages associated with device vnode.
1974  *
1975  * Note that spec_read and spec_write do not use the buffer cache, so we
1976  * must fully implement getpages here.
1977  */
1978 static int
1979 devfs_spec_getpages(struct vop_getpages_args *ap)
1980 {
1981 	vm_offset_t kva;
1982 	int error;
1983 	int i, pcount, size;
1984 	struct buf *bp;
1985 	vm_page_t m;
1986 	vm_ooffset_t offset;
1987 	int toff, nextoff, nread;
1988 	struct vnode *vp = ap->a_vp;
1989 	int blksiz;
1990 	int gotreqpage;
1991 
1992 	error = 0;
1993 	pcount = round_page(ap->a_count) / PAGE_SIZE;
1994 
1995 	/*
1996 	 * Calculate the offset of the transfer and do sanity check.
1997 	 */
1998 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1999 
2000 	/*
2001 	 * Round up physical size for real devices.  We cannot round using
2002 	 * v_mount's block size data because v_mount has nothing to do with
2003 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
2004 	 * size for the device itself.
2005 	 *
2006 	 * We can't use v_rdev->si_mountpoint because it only exists when the
2007 	 * block device is mounted.  However, we can use v_rdev.
2008 	 */
2009 	if (vn_isdisk(vp, NULL))
2010 		blksiz = vp->v_rdev->si_bsize_phys;
2011 	else
2012 		blksiz = DEV_BSIZE;
2013 
2014 	size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
2015 
2016 	bp = getpbuf(NULL);
2017 	kva = (vm_offset_t)bp->b_data;
2018 
2019 	/*
2020 	 * Map the pages to be read into the kva.
2021 	 */
2022 	pmap_qenter(kva, ap->a_m, pcount);
2023 
2024 	/* Build a minimal buffer header. */
2025 	bp->b_cmd = BUF_CMD_READ;
2026 	bp->b_bcount = size;
2027 	bp->b_resid = 0;
2028 	bp->b_runningbufspace = size;
2029 	if (size) {
2030 		runningbufspace += bp->b_runningbufspace;
2031 		++runningbufcount;
2032 	}
2033 
2034 	bp->b_bio1.bio_offset = offset;
2035 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2036 
2037 	mycpu->gd_cnt.v_vnodein++;
2038 	mycpu->gd_cnt.v_vnodepgsin += pcount;
2039 
2040 	/* Do the input. */
2041 	vn_strategy(ap->a_vp, &bp->b_bio1);
2042 
2043 	crit_enter();
2044 
2045 	/* We definitely need to be at splbio here. */
2046 	while (bp->b_cmd != BUF_CMD_DONE)
2047 		tsleep(bp, 0, "spread", 0);
2048 
2049 	crit_exit();
2050 
2051 	if (bp->b_flags & B_ERROR) {
2052 		if (bp->b_error)
2053 			error = bp->b_error;
2054 		else
2055 			error = EIO;
2056 	}
2057 
2058 	/*
2059 	 * If EOF is encountered we must zero-extend the result in order
2060 	 * to ensure that the page does not contain garabge.  When no
2061 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
2062 	 * b_resid is relative to b_bcount and should be 0, but some devices
2063 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
2064 	 */
2065 	nread = bp->b_bcount - bp->b_resid;
2066 	if (nread < ap->a_count)
2067 		bzero((caddr_t)kva + nread, ap->a_count - nread);
2068 	pmap_qremove(kva, pcount);
2069 
2070 	gotreqpage = 0;
2071 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2072 		nextoff = toff + PAGE_SIZE;
2073 		m = ap->a_m[i];
2074 
2075 		m->flags &= ~PG_ZERO;
2076 
2077 		/*
2078 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2079 		 *	 pmap modified bit.  pmap modified bit should have
2080 		 *	 already been cleared.
2081 		 */
2082 		if (nextoff <= nread) {
2083 			m->valid = VM_PAGE_BITS_ALL;
2084 			vm_page_undirty(m);
2085 		} else if (toff < nread) {
2086 			/*
2087 			 * Since this is a VM request, we have to supply the
2088 			 * unaligned offset to allow vm_page_set_valid()
2089 			 * to zero sub-DEV_BSIZE'd portions of the page.
2090 			 */
2091 			vm_page_set_valid(m, 0, nread - toff);
2092 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2093 		} else {
2094 			m->valid = 0;
2095 			vm_page_undirty(m);
2096 		}
2097 
2098 		if (i != ap->a_reqpage) {
2099 			/*
2100 			 * Just in case someone was asking for this page we
2101 			 * now tell them that it is ok to use.
2102 			 */
2103 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2104 				if (m->valid) {
2105 					if (m->flags & PG_WANTED) {
2106 						vm_page_activate(m);
2107 					} else {
2108 						vm_page_deactivate(m);
2109 					}
2110 					vm_page_wakeup(m);
2111 				} else {
2112 					vm_page_free(m);
2113 				}
2114 			} else {
2115 				vm_page_free(m);
2116 			}
2117 		} else if (m->valid) {
2118 			gotreqpage = 1;
2119 			/*
2120 			 * Since this is a VM request, we need to make the
2121 			 * entire page presentable by zeroing invalid sections.
2122 			 */
2123 			if (m->valid != VM_PAGE_BITS_ALL)
2124 			    vm_page_zero_invalid(m, FALSE);
2125 		}
2126 	}
2127 	if (!gotreqpage) {
2128 		m = ap->a_m[ap->a_reqpage];
2129 		devfs_debug(DEVFS_DEBUG_WARNING,
2130 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2131 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2132 		devfs_debug(DEVFS_DEBUG_WARNING,
2133 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2134 		    size, bp->b_resid, ap->a_count, m->valid);
2135 		devfs_debug(DEVFS_DEBUG_WARNING,
2136 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2137 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2138 		/*
2139 		 * Free the buffer header back to the swap buffer pool.
2140 		 */
2141 		relpbuf(bp, NULL);
2142 		return VM_PAGER_ERROR;
2143 	}
2144 	/*
2145 	 * Free the buffer header back to the swap buffer pool.
2146 	 */
2147 	relpbuf(bp, NULL);
2148 	if (DEVFS_NODE(ap->a_vp))
2149 		nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2150 	return VM_PAGER_OK;
2151 }
2152 
2153 static __inline
2154 int
2155 sequential_heuristic(struct uio *uio, struct file *fp)
2156 {
2157 	/*
2158 	 * Sequential heuristic - detect sequential operation
2159 	 */
2160 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2161 	    uio->uio_offset == fp->f_nextoff) {
2162 		/*
2163 		 * XXX we assume that the filesystem block size is
2164 		 * the default.  Not true, but still gives us a pretty
2165 		 * good indicator of how sequential the read operations
2166 		 * are.
2167 		 */
2168 		int tmpseq = fp->f_seqcount;
2169 
2170 		tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2171 		if (tmpseq > IO_SEQMAX)
2172 			tmpseq = IO_SEQMAX;
2173 		fp->f_seqcount = tmpseq;
2174 		return(fp->f_seqcount << IO_SEQSHIFT);
2175 	}
2176 
2177 	/*
2178 	 * Not sequential, quick draw-down of seqcount
2179 	 */
2180 	if (fp->f_seqcount > 1)
2181 		fp->f_seqcount = 1;
2182 	else
2183 		fp->f_seqcount = 0;
2184 	return(0);
2185 }
2186 
2187 extern SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "devfs");
2188 
2189 SYSCTL_INT(_vfs_devfs, OID_AUTO, mpsafe_writes, CTLFLAG_RD, &mpsafe_writes,
2190 		0, "mpsafe writes");
2191 SYSCTL_INT(_vfs_devfs, OID_AUTO, mplock_writes, CTLFLAG_RD, &mplock_writes,
2192 		0, "non-mpsafe writes");
2193 SYSCTL_INT(_vfs_devfs, OID_AUTO, mpsafe_reads, CTLFLAG_RD, &mpsafe_reads,
2194 		0, "mpsafe reads");
2195 SYSCTL_INT(_vfs_devfs, OID_AUTO, mplock_reads, CTLFLAG_RD, &mplock_reads,
2196 		0, "non-mpsafe reads");
2197