xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision d4b8aec4bb44a374c3e91969c1a7df6569da7be3)
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/fcntl.h>
50 #include <sys/namei.h>
51 #include <sys/dirent.h>
52 #include <sys/malloc.h>
53 #include <sys/stat.h>
54 #include <sys/reg.h>
55 #include <vm/vm_pager.h>
56 #include <vm/vm_zone.h>
57 #include <vm/vm_object.h>
58 #include <sys/filio.h>
59 #include <sys/ttycom.h>
60 #include <sys/tty.h>
61 #include <sys/diskslice.h>
62 #include <sys/sysctl.h>
63 #include <sys/devfs.h>
64 #include <sys/pioctl.h>
65 #include <vfs/fifofs/fifo.h>
66 
67 #include <machine/limits.h>
68 
69 #include <sys/buf2.h>
70 #include <sys/sysref2.h>
71 #include <sys/mplock2.h>
72 #include <vm/vm_page2.h>
73 
74 MALLOC_DECLARE(M_DEVFS);
75 #define DEVFS_BADOP	(void *)devfs_vop_badop
76 
77 static int devfs_vop_badop(struct vop_generic_args *);
78 static int devfs_vop_access(struct vop_access_args *);
79 static int devfs_vop_inactive(struct vop_inactive_args *);
80 static int devfs_vop_reclaim(struct vop_reclaim_args *);
81 static int devfs_vop_readdir(struct vop_readdir_args *);
82 static int devfs_vop_getattr(struct vop_getattr_args *);
83 static int devfs_vop_setattr(struct vop_setattr_args *);
84 static int devfs_vop_readlink(struct vop_readlink_args *);
85 static int devfs_vop_print(struct vop_print_args *);
86 
87 static int devfs_vop_nresolve(struct vop_nresolve_args *);
88 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
89 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
90 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
91 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
92 static int devfs_vop_nremove(struct vop_nremove_args *);
93 
94 static int devfs_spec_open(struct vop_open_args *);
95 static int devfs_spec_close(struct vop_close_args *);
96 static int devfs_spec_fsync(struct vop_fsync_args *);
97 
98 static int devfs_spec_read(struct vop_read_args *);
99 static int devfs_spec_write(struct vop_write_args *);
100 static int devfs_spec_ioctl(struct vop_ioctl_args *);
101 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
102 static int devfs_spec_strategy(struct vop_strategy_args *);
103 static void devfs_spec_strategy_done(struct bio *);
104 static int devfs_spec_freeblks(struct vop_freeblks_args *);
105 static int devfs_spec_bmap(struct vop_bmap_args *);
106 static int devfs_spec_advlock(struct vop_advlock_args *);
107 static void devfs_spec_getpages_iodone(struct bio *);
108 static int devfs_spec_getpages(struct vop_getpages_args *);
109 
110 static int devfs_fo_close(struct file *);
111 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
112 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
113 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
114 static int devfs_fo_kqfilter(struct file *, struct knote *);
115 static int devfs_fo_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 /*
122  * devfs vnode operations for regular files.  All vnode ops are MPSAFE.
123  */
124 struct vop_ops devfs_vnode_norm_vops = {
125 	.vop_default =		vop_defaultop,
126 	.vop_access =		devfs_vop_access,
127 	.vop_advlock =		DEVFS_BADOP,
128 	.vop_bmap =		DEVFS_BADOP,
129 	.vop_close =		vop_stdclose,
130 	.vop_getattr =		devfs_vop_getattr,
131 	.vop_inactive =		devfs_vop_inactive,
132 	.vop_ncreate =		DEVFS_BADOP,
133 	.vop_nresolve =		devfs_vop_nresolve,
134 	.vop_nlookupdotdot =	devfs_vop_nlookupdotdot,
135 	.vop_nlink =		DEVFS_BADOP,
136 	.vop_nmkdir =		devfs_vop_nmkdir,
137 	.vop_nmknod =		DEVFS_BADOP,
138 	.vop_nremove =		devfs_vop_nremove,
139 	.vop_nrename =		DEVFS_BADOP,
140 	.vop_nrmdir =		devfs_vop_nrmdir,
141 	.vop_nsymlink =		devfs_vop_nsymlink,
142 	.vop_open =		vop_stdopen,
143 	.vop_pathconf =		vop_stdpathconf,
144 	.vop_print =		devfs_vop_print,
145 	.vop_read =		DEVFS_BADOP,
146 	.vop_readdir =		devfs_vop_readdir,
147 	.vop_readlink =		devfs_vop_readlink,
148 	.vop_reclaim =		devfs_vop_reclaim,
149 	.vop_setattr =		devfs_vop_setattr,
150 	.vop_write =		DEVFS_BADOP,
151 	.vop_ioctl =		DEVFS_BADOP
152 };
153 
154 /*
155  * devfs vnode operations for character devices.  All vnode ops are MPSAFE.
156  */
157 struct vop_ops devfs_vnode_dev_vops = {
158 	.vop_default =		vop_defaultop,
159 	.vop_access =		devfs_vop_access,
160 	.vop_advlock =		devfs_spec_advlock,
161 	.vop_bmap =		devfs_spec_bmap,
162 	.vop_close =		devfs_spec_close,
163 	.vop_freeblks =		devfs_spec_freeblks,
164 	.vop_fsync =		devfs_spec_fsync,
165 	.vop_getattr =		devfs_vop_getattr,
166 	.vop_getpages =		devfs_spec_getpages,
167 	.vop_inactive =		devfs_vop_inactive,
168 	.vop_open =		devfs_spec_open,
169 	.vop_pathconf =		vop_stdpathconf,
170 	.vop_print =		devfs_vop_print,
171 	.vop_kqfilter =		devfs_spec_kqfilter,
172 	.vop_read =		devfs_spec_read,
173 	.vop_readdir =		DEVFS_BADOP,
174 	.vop_readlink =		DEVFS_BADOP,
175 	.vop_reclaim =		devfs_vop_reclaim,
176 	.vop_setattr =		devfs_vop_setattr,
177 	.vop_strategy =		devfs_spec_strategy,
178 	.vop_write =		devfs_spec_write,
179 	.vop_ioctl =		devfs_spec_ioctl
180 };
181 
182 /*
183  * devfs file pointer operations.  All fileops are MPSAFE.
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_fo_read,
189 	.fo_write	= devfs_fo_write,
190 	.fo_ioctl	= devfs_fo_ioctl,
191 	.fo_kqfilter	= devfs_fo_kqfilter,
192 	.fo_stat	= devfs_fo_stat,
193 	.fo_close	= devfs_fo_close,
194 	.fo_shutdown	= nofo_shutdown
195 };
196 
197 /*
198  * These two functions are possibly temporary hacks for devices (aka
199  * the pty code) which want to control the node attributes themselves.
200  *
201  * XXX we may ultimately desire to simply remove the uid/gid/mode
202  * from the node entirely.
203  *
204  * MPSAFE - sorta.  Theoretically the overwrite can compete since they
205  *	    are loading from the same fields.
206  */
207 static __inline void
208 node_sync_dev_get(struct devfs_node *node)
209 {
210 	cdev_t dev;
211 
212 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
213 		node->uid = dev->si_uid;
214 		node->gid = dev->si_gid;
215 		node->mode = dev->si_perms;
216 	}
217 }
218 
219 static __inline void
220 node_sync_dev_set(struct devfs_node *node)
221 {
222 	cdev_t dev;
223 
224 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
225 		dev->si_uid = node->uid;
226 		dev->si_gid = node->gid;
227 		dev->si_perms = node->mode;
228 	}
229 }
230 
231 /*
232  * generic entry point for unsupported operations
233  */
234 static int
235 devfs_vop_badop(struct vop_generic_args *ap)
236 {
237 	return (EIO);
238 }
239 
240 
241 static int
242 devfs_vop_access(struct vop_access_args *ap)
243 {
244 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
245 	int error;
246 
247 	if (!devfs_node_is_accessible(node))
248 		return ENOENT;
249 	node_sync_dev_get(node);
250 	error = vop_helper_access(ap, node->uid, node->gid,
251 				  node->mode, node->flags);
252 
253 	return error;
254 }
255 
256 
257 static int
258 devfs_vop_inactive(struct vop_inactive_args *ap)
259 {
260 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
261 
262 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
263 		vrecycle(ap->a_vp);
264 	return 0;
265 }
266 
267 
268 static int
269 devfs_vop_reclaim(struct vop_reclaim_args *ap)
270 {
271 	struct devfs_node *node;
272 	struct vnode *vp;
273 	int locked;
274 
275 	/*
276 	 * Check if it is locked already. if not, we acquire the devfs lock
277 	 */
278 	if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
279 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
280 		locked = 1;
281 	} else {
282 		locked = 0;
283 	}
284 
285 	/*
286 	 * Get rid of the devfs_node if it is no longer linked into the
287 	 * topology.
288 	 */
289 	vp = ap->a_vp;
290 	if ((node = DEVFS_NODE(vp)) != NULL) {
291 		node->v_node = NULL;
292 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
293 			devfs_freep(node);
294 	}
295 
296 	if (locked)
297 		lockmgr(&devfs_lock, LK_RELEASE);
298 
299 	/*
300 	 * v_rdev needs to be properly released using v_release_rdev
301 	 * Make sure v_data is NULL as well.
302 	 */
303 	vp->v_data = NULL;
304 	v_release_rdev(vp);
305 	return 0;
306 }
307 
308 
309 static int
310 devfs_vop_readdir(struct vop_readdir_args *ap)
311 {
312 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
313 	struct devfs_node *node;
314 	int cookie_index;
315 	int ncookies;
316 	int error2;
317 	int error;
318 	int r;
319 	off_t *cookies;
320 	off_t saveoff;
321 
322 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
323 
324 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
325 		return (EINVAL);
326 	if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
327 		return (error);
328 
329 	if (!devfs_node_is_accessible(dnode)) {
330 		vn_unlock(ap->a_vp);
331 		return ENOENT;
332 	}
333 
334 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
335 
336 	saveoff = ap->a_uio->uio_offset;
337 
338 	if (ap->a_ncookies) {
339 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
340 		if (ncookies > 256)
341 			ncookies = 256;
342 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
343 		cookie_index = 0;
344 	} else {
345 		ncookies = -1;
346 		cookies = NULL;
347 		cookie_index = 0;
348 	}
349 
350 	nanotime(&dnode->atime);
351 
352 	if (saveoff == 0) {
353 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
354 				     DT_DIR, 1, ".");
355 		if (r)
356 			goto done;
357 		if (cookies)
358 			cookies[cookie_index] = saveoff;
359 		saveoff++;
360 		cookie_index++;
361 		if (cookie_index == ncookies)
362 			goto done;
363 	}
364 
365 	if (saveoff == 1) {
366 		if (dnode->parent) {
367 			r = vop_write_dirent(&error, ap->a_uio,
368 					     dnode->parent->d_dir.d_ino,
369 					     DT_DIR, 2, "..");
370 		} else {
371 			r = vop_write_dirent(&error, ap->a_uio,
372 					     dnode->d_dir.d_ino,
373 					     DT_DIR, 2, "..");
374 		}
375 		if (r)
376 			goto done;
377 		if (cookies)
378 			cookies[cookie_index] = saveoff;
379 		saveoff++;
380 		cookie_index++;
381 		if (cookie_index == ncookies)
382 			goto done;
383 	}
384 
385 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
386 		if ((node->flags & DEVFS_HIDDEN) ||
387 		    (node->flags & DEVFS_INVISIBLE)) {
388 			continue;
389 		}
390 
391 		/*
392 		 * If the node type is a valid devfs alias, then we make
393 		 * sure that the target isn't hidden. If it is, we don't
394 		 * show the link in the directory listing.
395 		 */
396 		if ((node->node_type == Plink) && (node->link_target != NULL) &&
397 			(node->link_target->flags & DEVFS_HIDDEN))
398 			continue;
399 
400 		if (node->cookie < saveoff)
401 			continue;
402 
403 		saveoff = node->cookie;
404 
405 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
406 					  node->d_dir.d_type,
407 					  node->d_dir.d_namlen,
408 					  node->d_dir.d_name);
409 
410 		if (error2)
411 			break;
412 
413 		saveoff++;
414 
415 		if (cookies)
416 			cookies[cookie_index] = node->cookie;
417 		++cookie_index;
418 		if (cookie_index == ncookies)
419 			break;
420 	}
421 
422 done:
423 	lockmgr(&devfs_lock, LK_RELEASE);
424 	vn_unlock(ap->a_vp);
425 
426 	ap->a_uio->uio_offset = saveoff;
427 	if (error && cookie_index == 0) {
428 		if (cookies) {
429 			kfree(cookies, M_TEMP);
430 			*ap->a_ncookies = 0;
431 			*ap->a_cookies = NULL;
432 		}
433 	} else {
434 		if (cookies) {
435 			*ap->a_ncookies = cookie_index;
436 			*ap->a_cookies = cookies;
437 		}
438 	}
439 	return (error);
440 }
441 
442 
443 static int
444 devfs_vop_nresolve(struct vop_nresolve_args *ap)
445 {
446 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
447 	struct devfs_node *node, *found = NULL;
448 	struct namecache *ncp;
449 	struct vnode *vp = NULL;
450 	int error = 0;
451 	int len;
452 	int depth;
453 
454 	ncp = ap->a_nch->ncp;
455 	len = ncp->nc_nlen;
456 
457 	if (!devfs_node_is_accessible(dnode))
458 		return ENOENT;
459 
460 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
461 
462 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
463 		error = ENOENT;
464 		cache_setvp(ap->a_nch, NULL);
465 		goto out;
466 	}
467 
468 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
469 		if (len == node->d_dir.d_namlen) {
470 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
471 				found = node;
472 				break;
473 			}
474 		}
475 	}
476 
477 	if (found) {
478 		depth = 0;
479 		while ((found->node_type == Plink) && (found->link_target)) {
480 			if (depth >= 8) {
481 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
482 				break;
483 			}
484 
485 			found = found->link_target;
486 			++depth;
487 		}
488 
489 		if (!(found->flags & DEVFS_HIDDEN))
490 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
491 	}
492 
493 	if (vp == NULL) {
494 		error = ENOENT;
495 		cache_setvp(ap->a_nch, NULL);
496 		goto out;
497 
498 	}
499 	KKASSERT(vp);
500 	vn_unlock(vp);
501 	cache_setvp(ap->a_nch, vp);
502 	vrele(vp);
503 out:
504 	lockmgr(&devfs_lock, LK_RELEASE);
505 
506 	return error;
507 }
508 
509 
510 static int
511 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
512 {
513 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
514 
515 	*ap->a_vpp = NULL;
516 	if (!devfs_node_is_accessible(dnode))
517 		return ENOENT;
518 
519 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
520 	if (dnode->parent != NULL) {
521 		devfs_allocv(ap->a_vpp, dnode->parent);
522 		vn_unlock(*ap->a_vpp);
523 	}
524 	lockmgr(&devfs_lock, LK_RELEASE);
525 
526 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
527 }
528 
529 
530 static int
531 devfs_vop_getattr(struct vop_getattr_args *ap)
532 {
533 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
534 	struct vattr *vap = ap->a_vap;
535 	struct partinfo pinfo;
536 	int error = 0;
537 
538 #if 0
539 	if (!devfs_node_is_accessible(node))
540 		return ENOENT;
541 #endif
542 	node_sync_dev_get(node);
543 
544 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
545 
546 	/* start by zeroing out the attributes */
547 	VATTR_NULL(vap);
548 
549 	/* next do all the common fields */
550 	vap->va_type = ap->a_vp->v_type;
551 	vap->va_mode = node->mode;
552 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
553 	vap->va_flags = 0;
554 	vap->va_blocksize = DEV_BSIZE;
555 	vap->va_bytes = vap->va_size = 0;
556 
557 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
558 
559 	vap->va_atime = node->atime;
560 	vap->va_mtime = node->mtime;
561 	vap->va_ctime = node->ctime;
562 
563 	vap->va_nlink = 1; /* number of references to file */
564 
565 	vap->va_uid = node->uid;
566 	vap->va_gid = node->gid;
567 
568 	vap->va_rmajor = 0;
569 	vap->va_rminor = 0;
570 
571 	if ((node->node_type == Pdev) && node->d_dev)  {
572 		reference_dev(node->d_dev);
573 		vap->va_rminor = node->d_dev->si_uminor;
574 		release_dev(node->d_dev);
575 	}
576 
577 	/* For a softlink the va_size is the length of the softlink */
578 	if (node->symlink_name != 0) {
579 		vap->va_bytes = vap->va_size = node->symlink_namelen;
580 	}
581 
582 	/*
583 	 * For a disk-type device, va_size is the size of the underlying
584 	 * device, so that lseek() works properly.
585 	 */
586 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
587 		bzero(&pinfo, sizeof(pinfo));
588 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
589 				   0, proc0.p_ucred, NULL);
590 		if ((error == 0) && (pinfo.media_blksize != 0)) {
591 			vap->va_size = pinfo.media_size;
592 		} else {
593 			vap->va_size = 0;
594 			error = 0;
595 		}
596 	}
597 
598 	lockmgr(&devfs_lock, LK_RELEASE);
599 
600 	return (error);
601 }
602 
603 
604 static int
605 devfs_vop_setattr(struct vop_setattr_args *ap)
606 {
607 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
608 	struct vattr *vap;
609 	int error = 0;
610 
611 	if (!devfs_node_is_accessible(node))
612 		return ENOENT;
613 	node_sync_dev_get(node);
614 
615 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
616 
617 	vap = ap->a_vap;
618 
619 	if (vap->va_uid != (uid_t)VNOVAL) {
620 		if ((ap->a_cred->cr_uid != node->uid) &&
621 		    (!groupmember(node->gid, ap->a_cred))) {
622 			error = priv_check(curthread, PRIV_VFS_CHOWN);
623 			if (error)
624 				goto out;
625 		}
626 		node->uid = vap->va_uid;
627 	}
628 
629 	if (vap->va_gid != (uid_t)VNOVAL) {
630 		if ((ap->a_cred->cr_uid != node->uid) &&
631 		    (!groupmember(node->gid, ap->a_cred))) {
632 			error = priv_check(curthread, PRIV_VFS_CHOWN);
633 			if (error)
634 				goto out;
635 		}
636 		node->gid = vap->va_gid;
637 	}
638 
639 	if (vap->va_mode != (mode_t)VNOVAL) {
640 		if (ap->a_cred->cr_uid != node->uid) {
641 			error = priv_check(curthread, PRIV_VFS_ADMIN);
642 			if (error)
643 				goto out;
644 		}
645 		node->mode = vap->va_mode;
646 	}
647 
648 out:
649 	node_sync_dev_set(node);
650 	nanotime(&node->ctime);
651 	lockmgr(&devfs_lock, LK_RELEASE);
652 
653 	return error;
654 }
655 
656 
657 static int
658 devfs_vop_readlink(struct vop_readlink_args *ap)
659 {
660 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
661 	int ret;
662 
663 	if (!devfs_node_is_accessible(node))
664 		return ENOENT;
665 
666 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
667 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
668 	lockmgr(&devfs_lock, LK_RELEASE);
669 
670 	return ret;
671 }
672 
673 
674 static int
675 devfs_vop_print(struct vop_print_args *ap)
676 {
677 	return (0);
678 }
679 
680 static int
681 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
682 {
683 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
684 	struct devfs_node *node;
685 
686 	if (!devfs_node_is_accessible(dnode))
687 		return ENOENT;
688 
689 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
690 		goto out;
691 
692 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
693 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir,
694 		      ap->a_nch->ncp->nc_name, dnode, NULL);
695 
696 	if (*ap->a_vpp) {
697 		node = DEVFS_NODE(*ap->a_vpp);
698 		node->flags |= DEVFS_USER_CREATED;
699 		cache_setunresolved(ap->a_nch);
700 		cache_setvp(ap->a_nch, *ap->a_vpp);
701 	}
702 	lockmgr(&devfs_lock, LK_RELEASE);
703 out:
704 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
705 }
706 
707 static int
708 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
709 {
710 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
711 	struct devfs_node *node;
712 	size_t targetlen;
713 
714 	if (!devfs_node_is_accessible(dnode))
715 		return ENOENT;
716 
717 	ap->a_vap->va_type = VLNK;
718 
719 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
720 		goto out;
721 
722 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
723 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
724 		      ap->a_nch->ncp->nc_name, dnode, NULL);
725 
726 	targetlen = strlen(ap->a_target);
727 	if (*ap->a_vpp) {
728 		node = DEVFS_NODE(*ap->a_vpp);
729 		node->flags |= DEVFS_USER_CREATED;
730 		node->symlink_namelen = targetlen;
731 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
732 		memcpy(node->symlink_name, ap->a_target, targetlen);
733 		node->symlink_name[targetlen] = '\0';
734 		cache_setunresolved(ap->a_nch);
735 		cache_setvp(ap->a_nch, *ap->a_vpp);
736 	}
737 	lockmgr(&devfs_lock, LK_RELEASE);
738 out:
739 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
740 }
741 
742 static int
743 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
744 {
745 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
746 	struct devfs_node *node;
747 	struct namecache *ncp;
748 	int error = ENOENT;
749 
750 	ncp = ap->a_nch->ncp;
751 
752 	if (!devfs_node_is_accessible(dnode))
753 		return ENOENT;
754 
755 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
756 
757 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
758 		goto out;
759 
760 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
761 		if (ncp->nc_nlen != node->d_dir.d_namlen)
762 			continue;
763 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
764 			continue;
765 
766 		/*
767 		 * only allow removal of user created dirs
768 		 */
769 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
770 			error = EPERM;
771 			goto out;
772 		} else if (node->node_type != Pdir) {
773 			error = ENOTDIR;
774 			goto out;
775 		} else if (node->nchildren > 2) {
776 			error = ENOTEMPTY;
777 			goto out;
778 		} else {
779 			if (node->v_node)
780 				cache_inval_vp(node->v_node, CINV_DESTROY);
781 			devfs_unlinkp(node);
782 			error = 0;
783 			break;
784 		}
785 	}
786 
787 	cache_setunresolved(ap->a_nch);
788 	cache_setvp(ap->a_nch, NULL);
789 
790 out:
791 	lockmgr(&devfs_lock, LK_RELEASE);
792 	return error;
793 }
794 
795 static int
796 devfs_vop_nremove(struct vop_nremove_args *ap)
797 {
798 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
799 	struct devfs_node *node;
800 	struct namecache *ncp;
801 	int error = ENOENT;
802 
803 	ncp = ap->a_nch->ncp;
804 
805 	if (!devfs_node_is_accessible(dnode))
806 		return ENOENT;
807 
808 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
809 
810 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
811 		goto out;
812 
813 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
814 		if (ncp->nc_nlen != node->d_dir.d_namlen)
815 			continue;
816 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
817 			continue;
818 
819 		/*
820 		 * only allow removal of user created stuff (e.g. symlinks)
821 		 */
822 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
823 			error = EPERM;
824 			goto out;
825 		} else if (node->node_type == Pdir) {
826 			error = EISDIR;
827 			goto out;
828 		} else {
829 			if (node->v_node)
830 				cache_inval_vp(node->v_node, CINV_DESTROY);
831 			devfs_unlinkp(node);
832 			error = 0;
833 			break;
834 		}
835 	}
836 
837 	cache_setunresolved(ap->a_nch);
838 	cache_setvp(ap->a_nch, NULL);
839 
840 out:
841 	lockmgr(&devfs_lock, LK_RELEASE);
842 	return error;
843 }
844 
845 
846 static int
847 devfs_spec_open(struct vop_open_args *ap)
848 {
849 	struct vnode *vp = ap->a_vp;
850 	struct vnode *orig_vp = NULL;
851 	struct devfs_node *node = DEVFS_NODE(vp);
852 	struct devfs_node *newnode;
853 	cdev_t dev, ndev = NULL;
854 	int error = 0;
855 
856 	if (node) {
857 		if (node->d_dev == NULL)
858 			return ENXIO;
859 		if (!devfs_node_is_accessible(node))
860 			return ENOENT;
861 	}
862 
863 	if ((dev = vp->v_rdev) == NULL)
864 		return ENXIO;
865 
866 	if (node && ap->a_fp) {
867 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
868 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
869 
870 		ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
871 						ap->a_mode, ap->a_cred);
872 		if (ndev != NULL) {
873 			newnode = devfs_create_device_node(
874 					DEVFS_MNTDATA(vp->v_mount)->root_node,
875 					ndev, NULL, NULL);
876 			/* XXX: possibly destroy device if this happens */
877 
878 			if (newnode != NULL) {
879 				dev = ndev;
880 				devfs_link_dev(dev);
881 
882 				devfs_debug(DEVFS_DEBUG_DEBUG,
883 						"parent here is: %s, node is: |%s|\n",
884 						((node->parent->node_type == Proot) ?
885 						"ROOT!" : node->parent->d_dir.d_name),
886 						newnode->d_dir.d_name);
887 				devfs_debug(DEVFS_DEBUG_DEBUG,
888 						"test: %s\n",
889 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
890 
891 				/*
892 				 * orig_vp is set to the original vp if we cloned.
893 				 */
894 				/* node->flags |= DEVFS_CLONED; */
895 				devfs_allocv(&vp, newnode);
896 				orig_vp = ap->a_vp;
897 				ap->a_vp = vp;
898 			}
899 		}
900 		lockmgr(&devfs_lock, LK_RELEASE);
901 	}
902 
903 	devfs_debug(DEVFS_DEBUG_DEBUG,
904 		    "devfs_spec_open() called on %s! \n",
905 		    dev->si_name);
906 
907 	/*
908 	 * Make this field valid before any I/O in ->d_open
909 	 */
910 	if (!dev->si_iosize_max)
911 		dev->si_iosize_max = DFLTPHYS;
912 
913 	if (dev_dflags(dev) & D_TTY)
914 		vsetflags(vp, VISTTY);
915 
916 	/*
917 	 * Open underlying device
918 	 */
919 	vn_unlock(vp);
920 	error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
921 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
922 
923 	/*
924 	 * Clean up any cloned vp if we error out.
925 	 */
926 	if (error) {
927 		if (orig_vp) {
928 			vput(vp);
929 			ap->a_vp = orig_vp;
930 			/* orig_vp = NULL; */
931 		}
932 		return error;
933 	}
934 
935 	/*
936 	 * This checks if the disk device is going to be opened for writing.
937 	 * It will be only allowed in the cases where securelevel permits it
938 	 * and it's not mounted R/W.
939 	 */
940 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
941 	    (ap->a_cred != FSCRED)) {
942 
943 		/* Very secure mode. No open for writing allowed */
944 		if (securelevel >= 2)
945 			return EPERM;
946 
947 		/*
948 		 * If it is mounted R/W, do not allow to open for writing.
949 		 * In the case it's mounted read-only but securelevel
950 		 * is >= 1, then do not allow opening for writing either.
951 		 */
952 		if (vfs_mountedon(vp)) {
953 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
954 				return EBUSY;
955 			else if (securelevel >= 1)
956 				return EPERM;
957 		}
958 	}
959 
960 	if (dev_dflags(dev) & D_TTY) {
961 		if (dev->si_tty) {
962 			struct tty *tp;
963 			tp = dev->si_tty;
964 			if (!tp->t_stop) {
965 				devfs_debug(DEVFS_DEBUG_DEBUG,
966 					    "devfs: no t_stop\n");
967 				tp->t_stop = nottystop;
968 			}
969 		}
970 	}
971 
972 
973 	if (vn_isdisk(vp, NULL)) {
974 		if (!dev->si_bsize_phys)
975 			dev->si_bsize_phys = DEV_BSIZE;
976 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
977 	}
978 
979 	vop_stdopen(ap);
980 #if 0
981 	if (node)
982 		nanotime(&node->atime);
983 #endif
984 
985 	/*
986 	 * If we replaced the vp the vop_stdopen() call will have loaded
987 	 * it into fp->f_data and vref()d the vp, giving us two refs.  So
988 	 * instead of just unlocking it here we have to vput() it.
989 	 */
990 	if (orig_vp)
991 		vput(vp);
992 
993 	/* Ugly pty magic, to make pty devices appear once they are opened */
994 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
995 		node->flags &= ~DEVFS_INVISIBLE;
996 
997 	if (ap->a_fp) {
998 		KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
999 		KKASSERT(ap->a_fp->f_flag == (ap->a_mode & FMASK));
1000 		ap->a_fp->f_ops = &devfs_dev_fileops;
1001 		KKASSERT(ap->a_fp->f_data == (void *)vp);
1002 	}
1003 
1004 	return 0;
1005 }
1006 
1007 
1008 static int
1009 devfs_spec_close(struct vop_close_args *ap)
1010 {
1011 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
1012 	struct proc *p = curproc;
1013 	struct vnode *vp = ap->a_vp;
1014 	cdev_t dev = vp->v_rdev;
1015 	int error = 0;
1016 	int needrelock;
1017 
1018 	devfs_debug(DEVFS_DEBUG_DEBUG,
1019 		    "devfs_spec_close() called on %s! \n",
1020 		    dev->si_name);
1021 
1022 	/*
1023 	 * A couple of hacks for devices and tty devices.  The
1024 	 * vnode ref count cannot be used to figure out the
1025 	 * last close, but we can use v_opencount now that
1026 	 * revoke works properly.
1027 	 *
1028 	 * Detect the last close on a controlling terminal and clear
1029 	 * the session (half-close).
1030 	 */
1031 	if (dev)
1032 		reference_dev(dev);
1033 
1034 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1035 		p->p_session->s_ttyvp = NULL;
1036 		vrele(vp);
1037 	}
1038 
1039 	/*
1040 	 * Vnodes can be opened and closed multiple times.  Do not really
1041 	 * close the device unless (1) it is being closed forcibly,
1042 	 * (2) the device wants to track closes, or (3) this is the last
1043 	 * vnode doing its last close on the device.
1044 	 *
1045 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1046 	 * a closed device.  This might not occur now that our revoke is
1047 	 * fixed.
1048 	 */
1049 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1050 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1051 	    (dev_dflags(dev) & D_TRACKCLOSE) ||
1052 	    (vp->v_opencount == 1))) {
1053 		/*
1054 		 * Unlock around dev_dclose()
1055 		 */
1056 		needrelock = 0;
1057 		if (vn_islocked(vp)) {
1058 			needrelock = 1;
1059 			vn_unlock(vp);
1060 		}
1061 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1062 
1063 		/*
1064 		 * Ugly pty magic, to make pty devices disappear again once
1065 		 * they are closed
1066 		 */
1067 		if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
1068 			node->flags |= DEVFS_INVISIBLE;
1069 
1070 		if (needrelock)
1071 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1072 	} else {
1073 		error = 0;
1074 	}
1075 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1076 
1077 	/*
1078 	 * Track the actual opens and closes on the vnode.  The last close
1079 	 * disassociates the rdev.  If the rdev is already disassociated or
1080 	 * the opencount is already 0, the vnode might have been revoked
1081 	 * and no further opencount tracking occurs.
1082 	 */
1083 	if (dev)
1084 		release_dev(dev);
1085 	if (vp->v_opencount > 0)
1086 		vop_stdclose(ap);
1087 	return(error);
1088 
1089 }
1090 
1091 
1092 static int
1093 devfs_fo_close(struct file *fp)
1094 {
1095 	struct vnode *vp = (struct vnode *)fp->f_data;
1096 	int error;
1097 
1098 	fp->f_ops = &badfileops;
1099 	error = vn_close(vp, fp->f_flag);
1100 
1101 	return (error);
1102 }
1103 
1104 
1105 /*
1106  * Device-optimized file table vnode read routine.
1107  *
1108  * This bypasses the VOP table and talks directly to the device.  Most
1109  * filesystems just route to specfs and can make this optimization.
1110  *
1111  * MPALMOSTSAFE - acquires mplock
1112  */
1113 static int
1114 devfs_fo_read(struct file *fp, struct uio *uio,
1115 		 struct ucred *cred, int flags)
1116 {
1117 	struct devfs_node *node;
1118 	struct vnode *vp;
1119 	int ioflag;
1120 	int error;
1121 	cdev_t dev;
1122 
1123 	KASSERT(uio->uio_td == curthread,
1124 		("uio_td %p is not td %p", uio->uio_td, curthread));
1125 
1126 	if (uio->uio_resid == 0)
1127 		return 0;
1128 
1129 	vp = (struct vnode *)fp->f_data;
1130 	if (vp == NULL || vp->v_type == VBAD)
1131 		return EBADF;
1132 
1133 	node = DEVFS_NODE(vp);
1134 
1135 	if ((dev = vp->v_rdev) == NULL)
1136 		return EBADF;
1137 
1138 	reference_dev(dev);
1139 
1140 	if ((flags & O_FOFFSET) == 0)
1141 		uio->uio_offset = fp->f_offset;
1142 
1143 	ioflag = 0;
1144 	if (flags & O_FBLOCKING) {
1145 		/* ioflag &= ~IO_NDELAY; */
1146 	} else if (flags & O_FNONBLOCKING) {
1147 		ioflag |= IO_NDELAY;
1148 	} else if (fp->f_flag & FNONBLOCK) {
1149 		ioflag |= IO_NDELAY;
1150 	}
1151 	if (flags & O_FBUFFERED) {
1152 		/* ioflag &= ~IO_DIRECT; */
1153 	} else if (flags & O_FUNBUFFERED) {
1154 		ioflag |= IO_DIRECT;
1155 	} else if (fp->f_flag & O_DIRECT) {
1156 		ioflag |= IO_DIRECT;
1157 	}
1158 	ioflag |= sequential_heuristic(uio, fp);
1159 
1160 	error = dev_dread(dev, uio, ioflag);
1161 
1162 	release_dev(dev);
1163 	if (node)
1164 		nanotime(&node->atime);
1165 	if ((flags & O_FOFFSET) == 0)
1166 		fp->f_offset = uio->uio_offset;
1167 	fp->f_nextoff = uio->uio_offset;
1168 
1169 	return (error);
1170 }
1171 
1172 
1173 static int
1174 devfs_fo_write(struct file *fp, struct uio *uio,
1175 		  struct ucred *cred, int flags)
1176 {
1177 	struct devfs_node *node;
1178 	struct vnode *vp;
1179 	int ioflag;
1180 	int error;
1181 	cdev_t dev;
1182 
1183 	KASSERT(uio->uio_td == curthread,
1184 		("uio_td %p is not p %p", uio->uio_td, curthread));
1185 
1186 	vp = (struct vnode *)fp->f_data;
1187 	if (vp == NULL || vp->v_type == VBAD)
1188 		return EBADF;
1189 
1190 	node = DEVFS_NODE(vp);
1191 
1192 	if (vp->v_type == VREG)
1193 		bwillwrite(uio->uio_resid);
1194 
1195 	vp = (struct vnode *)fp->f_data;
1196 
1197 	if ((dev = vp->v_rdev) == NULL)
1198 		return EBADF;
1199 
1200 	reference_dev(dev);
1201 
1202 	if ((flags & O_FOFFSET) == 0)
1203 		uio->uio_offset = fp->f_offset;
1204 
1205 	ioflag = IO_UNIT;
1206 	if (vp->v_type == VREG &&
1207 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1208 		ioflag |= IO_APPEND;
1209 	}
1210 
1211 	if (flags & O_FBLOCKING) {
1212 		/* ioflag &= ~IO_NDELAY; */
1213 	} else if (flags & O_FNONBLOCKING) {
1214 		ioflag |= IO_NDELAY;
1215 	} else if (fp->f_flag & FNONBLOCK) {
1216 		ioflag |= IO_NDELAY;
1217 	}
1218 	if (flags & O_FBUFFERED) {
1219 		/* ioflag &= ~IO_DIRECT; */
1220 	} else if (flags & O_FUNBUFFERED) {
1221 		ioflag |= IO_DIRECT;
1222 	} else if (fp->f_flag & O_DIRECT) {
1223 		ioflag |= IO_DIRECT;
1224 	}
1225 	if (flags & O_FASYNCWRITE) {
1226 		/* ioflag &= ~IO_SYNC; */
1227 	} else if (flags & O_FSYNCWRITE) {
1228 		ioflag |= IO_SYNC;
1229 	} else if (fp->f_flag & O_FSYNC) {
1230 		ioflag |= IO_SYNC;
1231 	}
1232 
1233 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1234 		ioflag |= IO_SYNC;
1235 	ioflag |= sequential_heuristic(uio, fp);
1236 
1237 	error = dev_dwrite(dev, uio, ioflag);
1238 
1239 	release_dev(dev);
1240 	if (node) {
1241 		nanotime(&node->atime);
1242 		nanotime(&node->mtime);
1243 	}
1244 
1245 	if ((flags & O_FOFFSET) == 0)
1246 		fp->f_offset = uio->uio_offset;
1247 	fp->f_nextoff = uio->uio_offset;
1248 
1249 	return (error);
1250 }
1251 
1252 
1253 static int
1254 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1255 {
1256 	struct vnode *vp;
1257 	struct vattr vattr;
1258 	struct vattr *vap;
1259 	u_short mode;
1260 	cdev_t dev;
1261 	int error;
1262 
1263 	vp = (struct vnode *)fp->f_data;
1264 	if (vp == NULL || vp->v_type == VBAD)
1265 		return EBADF;
1266 
1267 	error = vn_stat(vp, sb, cred);
1268 	if (error)
1269 		return (error);
1270 
1271 	vap = &vattr;
1272 	error = VOP_GETATTR(vp, vap);
1273 	if (error)
1274 		return (error);
1275 
1276 	/*
1277 	 * Zero the spare stat fields
1278 	 */
1279 	sb->st_lspare = 0;
1280 	sb->st_qspare1 = 0;
1281 	sb->st_qspare2 = 0;
1282 
1283 	/*
1284 	 * Copy from vattr table ... or not in case it's a cloned device
1285 	 */
1286 	if (vap->va_fsid != VNOVAL)
1287 		sb->st_dev = vap->va_fsid;
1288 	else
1289 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1290 
1291 	sb->st_ino = vap->va_fileid;
1292 
1293 	mode = vap->va_mode;
1294 	mode |= S_IFCHR;
1295 	sb->st_mode = mode;
1296 
1297 	if (vap->va_nlink > (nlink_t)-1)
1298 		sb->st_nlink = (nlink_t)-1;
1299 	else
1300 		sb->st_nlink = vap->va_nlink;
1301 
1302 	sb->st_uid = vap->va_uid;
1303 	sb->st_gid = vap->va_gid;
1304 	sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1305 	sb->st_size = vap->va_bytes;
1306 	sb->st_atimespec = vap->va_atime;
1307 	sb->st_mtimespec = vap->va_mtime;
1308 	sb->st_ctimespec = vap->va_ctime;
1309 
1310 	/*
1311 	 * A VCHR and VBLK device may track the last access and last modified
1312 	 * time independantly of the filesystem.  This is particularly true
1313 	 * because device read and write calls may bypass the filesystem.
1314 	 */
1315 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1316 		dev = vp->v_rdev;
1317 		if (dev != NULL) {
1318 			if (dev->si_lastread) {
1319 				sb->st_atimespec.tv_sec = dev->si_lastread;
1320 				sb->st_atimespec.tv_nsec = 0;
1321 			}
1322 			if (dev->si_lastwrite) {
1323 				sb->st_atimespec.tv_sec = dev->si_lastwrite;
1324 				sb->st_atimespec.tv_nsec = 0;
1325 			}
1326 		}
1327 	}
1328 
1329         /*
1330 	 * According to www.opengroup.org, the meaning of st_blksize is
1331 	 *   "a filesystem-specific preferred I/O block size for this
1332 	 *    object.  In some filesystem types, this may vary from file
1333 	 *    to file"
1334 	 * Default to PAGE_SIZE after much discussion.
1335 	 */
1336 
1337 	sb->st_blksize = PAGE_SIZE;
1338 
1339 	sb->st_flags = vap->va_flags;
1340 
1341 	error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1342 	if (error)
1343 		sb->st_gen = 0;
1344 	else
1345 		sb->st_gen = (u_int32_t)vap->va_gen;
1346 
1347 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1348 
1349 	return (0);
1350 }
1351 
1352 
1353 static int
1354 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1355 {
1356 	struct vnode *vp;
1357 	int error;
1358 	cdev_t dev;
1359 
1360 	vp = (struct vnode *)fp->f_data;
1361 	if (vp == NULL || vp->v_type == VBAD) {
1362 		error = EBADF;
1363 		goto done;
1364 	}
1365 	if ((dev = vp->v_rdev) == NULL) {
1366 		error = EBADF;
1367 		goto done;
1368 	}
1369 	reference_dev(dev);
1370 
1371 	error = dev_dkqfilter(dev, kn);
1372 
1373 	release_dev(dev);
1374 
1375 done:
1376 	return (error);
1377 }
1378 
1379 /*
1380  * MPALMOSTSAFE - acquires mplock
1381  */
1382 static int
1383 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1384 		  struct ucred *ucred, struct sysmsg *msg)
1385 {
1386 	struct devfs_node *node;
1387 	struct vnode *vp;
1388 	struct vnode *ovp;
1389 	cdev_t	dev;
1390 	int error;
1391 	struct fiodname_args *name_args;
1392 	size_t namlen;
1393 	const char *name;
1394 
1395 	vp = ((struct vnode *)fp->f_data);
1396 
1397 	if ((dev = vp->v_rdev) == NULL)
1398 		return EBADF;		/* device was revoked */
1399 
1400 	reference_dev(dev);
1401 
1402 	node = DEVFS_NODE(vp);
1403 
1404 	devfs_debug(DEVFS_DEBUG_DEBUG,
1405 		    "devfs_fo_ioctl() called! for dev %s\n",
1406 		    dev->si_name);
1407 
1408 	if (com == FIODTYPE) {
1409 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1410 		error = 0;
1411 		goto out;
1412 	} else if (com == FIODNAME) {
1413 		name_args = (struct fiodname_args *)data;
1414 		name = dev->si_name;
1415 		namlen = strlen(name) + 1;
1416 
1417 		devfs_debug(DEVFS_DEBUG_DEBUG,
1418 			    "ioctl, got: FIODNAME for %s\n", name);
1419 
1420 		if (namlen <= name_args->len)
1421 			error = copyout(dev->si_name, name_args->name, namlen);
1422 		else
1423 			error = EINVAL;
1424 
1425 		devfs_debug(DEVFS_DEBUG_DEBUG,
1426 			    "ioctl stuff: error: %d\n", error);
1427 		goto out;
1428 	}
1429 
1430 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
1431 
1432 #if 0
1433 	if (node) {
1434 		nanotime(&node->atime);
1435 		nanotime(&node->mtime);
1436 	}
1437 #endif
1438 	if (com == TIOCSCTTY) {
1439 		devfs_debug(DEVFS_DEBUG_DEBUG,
1440 			    "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1441 			    dev->si_name);
1442 	}
1443 	if (error == 0 && com == TIOCSCTTY) {
1444 		struct proc *p = curthread->td_proc;
1445 		struct session *sess;
1446 
1447 		devfs_debug(DEVFS_DEBUG_DEBUG,
1448 			    "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1449 			    dev->si_name);
1450 		if (p == NULL) {
1451 			error = ENOTTY;
1452 			goto out;
1453 		}
1454 		sess = p->p_session;
1455 
1456 		/*
1457 		 * Do nothing if reassigning same control tty
1458 		 */
1459 		if (sess->s_ttyvp == vp) {
1460 			error = 0;
1461 			goto out;
1462 		}
1463 
1464 		/*
1465 		 * Get rid of reference to old control tty
1466 		 */
1467 		ovp = sess->s_ttyvp;
1468 		vref(vp);
1469 		sess->s_ttyvp = vp;
1470 		if (ovp)
1471 			vrele(ovp);
1472 	}
1473 
1474 out:
1475 	release_dev(dev);
1476 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1477 	return (error);
1478 }
1479 
1480 
1481 static int
1482 devfs_spec_fsync(struct vop_fsync_args *ap)
1483 {
1484 	struct vnode *vp = ap->a_vp;
1485 	int error;
1486 
1487 	if (!vn_isdisk(vp, NULL))
1488 		return (0);
1489 
1490 	/*
1491 	 * Flush all dirty buffers associated with a block device.
1492 	 */
1493 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1494 	return (error);
1495 }
1496 
1497 static int
1498 devfs_spec_read(struct vop_read_args *ap)
1499 {
1500 	struct devfs_node *node;
1501 	struct vnode *vp;
1502 	struct uio *uio;
1503 	cdev_t dev;
1504 	int error;
1505 
1506 	vp = ap->a_vp;
1507 	dev = vp->v_rdev;
1508 	uio = ap->a_uio;
1509 	node = DEVFS_NODE(vp);
1510 
1511 	if (dev == NULL)		/* device was revoked */
1512 		return (EBADF);
1513 	if (uio->uio_resid == 0)
1514 		return (0);
1515 
1516 	vn_unlock(vp);
1517 	error = dev_dread(dev, uio, ap->a_ioflag);
1518 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1519 
1520 	if (node)
1521 		nanotime(&node->atime);
1522 
1523 	return (error);
1524 }
1525 
1526 /*
1527  * Vnode op for write
1528  *
1529  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1530  *	      struct ucred *a_cred)
1531  */
1532 static int
1533 devfs_spec_write(struct vop_write_args *ap)
1534 {
1535 	struct devfs_node *node;
1536 	struct vnode *vp;
1537 	struct uio *uio;
1538 	cdev_t dev;
1539 	int error;
1540 
1541 	vp = ap->a_vp;
1542 	dev = vp->v_rdev;
1543 	uio = ap->a_uio;
1544 	node = DEVFS_NODE(vp);
1545 
1546 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1547 
1548 	if (dev == NULL)		/* device was revoked */
1549 		return (EBADF);
1550 
1551 	vn_unlock(vp);
1552 	error = dev_dwrite(dev, uio, ap->a_ioflag);
1553 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1554 
1555 	if (node) {
1556 		nanotime(&node->atime);
1557 		nanotime(&node->mtime);
1558 	}
1559 
1560 	return (error);
1561 }
1562 
1563 /*
1564  * Device ioctl operation.
1565  *
1566  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1567  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1568  */
1569 static int
1570 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1571 {
1572 	struct vnode *vp = ap->a_vp;
1573 	struct devfs_node *node;
1574 	cdev_t dev;
1575 
1576 	if ((dev = vp->v_rdev) == NULL)
1577 		return (EBADF);		/* device was revoked */
1578 	node = DEVFS_NODE(vp);
1579 
1580 #if 0
1581 	if (node) {
1582 		nanotime(&node->atime);
1583 		nanotime(&node->mtime);
1584 	}
1585 #endif
1586 
1587 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1588 			   ap->a_cred, ap->a_sysmsg));
1589 }
1590 
1591 /*
1592  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1593  */
1594 /* ARGSUSED */
1595 static int
1596 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1597 {
1598 	struct vnode *vp = ap->a_vp;
1599 	struct devfs_node *node;
1600 	cdev_t dev;
1601 
1602 	if ((dev = vp->v_rdev) == NULL)
1603 		return (EBADF);		/* device was revoked (EBADF) */
1604 	node = DEVFS_NODE(vp);
1605 
1606 #if 0
1607 	if (node)
1608 		nanotime(&node->atime);
1609 #endif
1610 
1611 	return (dev_dkqfilter(dev, ap->a_kn));
1612 }
1613 
1614 /*
1615  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1616  * calls are not limited to device DMA limits so we have to deal with the
1617  * case.
1618  *
1619  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1620  */
1621 static int
1622 devfs_spec_strategy(struct vop_strategy_args *ap)
1623 {
1624 	struct bio *bio = ap->a_bio;
1625 	struct buf *bp = bio->bio_buf;
1626 	struct buf *nbp;
1627 	struct vnode *vp;
1628 	struct mount *mp;
1629 	int chunksize;
1630 	int maxiosize;
1631 
1632 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1633 		buf_start(bp);
1634 
1635 	/*
1636 	 * Collect statistics on synchronous and asynchronous read
1637 	 * and write counts for disks that have associated filesystems.
1638 	 */
1639 	vp = ap->a_vp;
1640 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1641 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1642 		if (bp->b_cmd == BUF_CMD_READ) {
1643 			if (bp->b_flags & BIO_SYNC)
1644 				mp->mnt_stat.f_syncreads++;
1645 			else
1646 				mp->mnt_stat.f_asyncreads++;
1647 		} else {
1648 			if (bp->b_flags & BIO_SYNC)
1649 				mp->mnt_stat.f_syncwrites++;
1650 			else
1651 				mp->mnt_stat.f_asyncwrites++;
1652 		}
1653 	}
1654 
1655         /*
1656          * Device iosize limitations only apply to read and write.  Shortcut
1657          * the I/O if it fits.
1658          */
1659 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1660 		devfs_debug(DEVFS_DEBUG_DEBUG,
1661 			    "%s: si_iosize_max not set!\n",
1662 			    dev_dname(vp->v_rdev));
1663 		maxiosize = MAXPHYS;
1664 	}
1665 #if SPEC_CHAIN_DEBUG & 2
1666 	maxiosize = 4096;
1667 #endif
1668         if (bp->b_bcount <= maxiosize ||
1669             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1670                 dev_dstrategy_chain(vp->v_rdev, bio);
1671                 return (0);
1672         }
1673 
1674 	/*
1675 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1676 	 */
1677 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1678 	initbufbio(nbp);
1679 	buf_dep_init(nbp);
1680 	BUF_LOCKINIT(nbp);
1681 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1682 	BUF_KERNPROC(nbp);
1683 	nbp->b_vp = vp;
1684 	nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1685 	nbp->b_data = bp->b_data;
1686 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1687 	nbp->b_bio1.bio_offset = bio->bio_offset;
1688 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1689 
1690 	/*
1691 	 * Start the first transfer
1692 	 */
1693 	if (vn_isdisk(vp, NULL))
1694 		chunksize = vp->v_rdev->si_bsize_phys;
1695 	else
1696 		chunksize = DEV_BSIZE;
1697 	chunksize = maxiosize / chunksize * chunksize;
1698 #if SPEC_CHAIN_DEBUG & 1
1699 	devfs_debug(DEVFS_DEBUG_DEBUG,
1700 		    "spec_strategy chained I/O chunksize=%d\n",
1701 		    chunksize);
1702 #endif
1703 	nbp->b_cmd = bp->b_cmd;
1704 	nbp->b_bcount = chunksize;
1705 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1706 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1707 
1708 #if SPEC_CHAIN_DEBUG & 1
1709 	devfs_debug(DEVFS_DEBUG_DEBUG,
1710 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1711 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1712 #endif
1713 
1714 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1715 
1716 	if (DEVFS_NODE(vp)) {
1717 		nanotime(&DEVFS_NODE(vp)->atime);
1718 		nanotime(&DEVFS_NODE(vp)->mtime);
1719 	}
1720 
1721 	return (0);
1722 }
1723 
1724 /*
1725  * Chunked up transfer completion routine - chain transfers until done
1726  */
1727 static
1728 void
1729 devfs_spec_strategy_done(struct bio *nbio)
1730 {
1731 	struct buf *nbp = nbio->bio_buf;
1732 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1733 	struct buf *bp = bio->bio_buf;			/* original bp */
1734 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1735 	int boffset = nbp->b_data - bp->b_data;
1736 
1737 	if (nbp->b_flags & B_ERROR) {
1738 		/*
1739 		 * An error terminates the chain, propogate the error back
1740 		 * to the original bp
1741 		 */
1742 		bp->b_flags |= B_ERROR;
1743 		bp->b_error = nbp->b_error;
1744 		bp->b_resid = bp->b_bcount - boffset +
1745 			      (nbp->b_bcount - nbp->b_resid);
1746 #if SPEC_CHAIN_DEBUG & 1
1747 		devfs_debug(DEVFS_DEBUG_DEBUG,
1748 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1749 			    bp, bp->b_error, bp->b_bcount,
1750 			    bp->b_bcount - bp->b_resid);
1751 #endif
1752 		kfree(nbp, M_DEVBUF);
1753 		biodone(bio);
1754 	} else if (nbp->b_resid) {
1755 		/*
1756 		 * A short read or write terminates the chain
1757 		 */
1758 		bp->b_error = nbp->b_error;
1759 		bp->b_resid = bp->b_bcount - boffset +
1760 			      (nbp->b_bcount - nbp->b_resid);
1761 #if SPEC_CHAIN_DEBUG & 1
1762 		devfs_debug(DEVFS_DEBUG_DEBUG,
1763 			    "spec_strategy: chain %p short read(1) "
1764 			    "bcount %d/%d\n",
1765 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1766 #endif
1767 		kfree(nbp, M_DEVBUF);
1768 		biodone(bio);
1769 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1770 		/*
1771 		 * A short read or write can also occur by truncating b_bcount
1772 		 */
1773 #if SPEC_CHAIN_DEBUG & 1
1774 		devfs_debug(DEVFS_DEBUG_DEBUG,
1775 			    "spec_strategy: chain %p short read(2) "
1776 			    "bcount %d/%d\n",
1777 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1778 #endif
1779 		bp->b_error = 0;
1780 		bp->b_bcount = nbp->b_bcount + boffset;
1781 		bp->b_resid = nbp->b_resid;
1782 		kfree(nbp, M_DEVBUF);
1783 		biodone(bio);
1784 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1785 		/*
1786 		 * No more data terminates the chain
1787 		 */
1788 #if SPEC_CHAIN_DEBUG & 1
1789 		devfs_debug(DEVFS_DEBUG_DEBUG,
1790 			    "spec_strategy: chain %p finished bcount %d\n",
1791 			    bp, bp->b_bcount);
1792 #endif
1793 		bp->b_error = 0;
1794 		bp->b_resid = 0;
1795 		kfree(nbp, M_DEVBUF);
1796 		biodone(bio);
1797 	} else {
1798 		/*
1799 		 * Continue the chain
1800 		 */
1801 		boffset += nbp->b_bcount;
1802 		nbp->b_data = bp->b_data + boffset;
1803 		nbp->b_bcount = bp->b_bcount - boffset;
1804 		if (nbp->b_bcount > chunksize)
1805 			nbp->b_bcount = chunksize;
1806 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1807 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1808 
1809 #if SPEC_CHAIN_DEBUG & 1
1810 		devfs_debug(DEVFS_DEBUG_DEBUG,
1811 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1812 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
1813 #endif
1814 
1815 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1816 	}
1817 }
1818 
1819 /*
1820  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1821  */
1822 static int
1823 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1824 {
1825 	struct buf *bp;
1826 
1827 	/*
1828 	 * XXX: This assumes that strategy does the deed right away.
1829 	 * XXX: this may not be TRTTD.
1830 	 */
1831 	KKASSERT(ap->a_vp->v_rdev != NULL);
1832 	if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
1833 		return (0);
1834 	bp = geteblk(ap->a_length);
1835 	bp->b_cmd = BUF_CMD_FREEBLKS;
1836 	bp->b_bio1.bio_offset = ap->a_offset;
1837 	bp->b_bcount = ap->a_length;
1838 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1839 	return (0);
1840 }
1841 
1842 /*
1843  * Implement degenerate case where the block requested is the block
1844  * returned, and assume that the entire device is contiguous in regards
1845  * to the contiguous block range (runp and runb).
1846  *
1847  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1848  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
1849  */
1850 static int
1851 devfs_spec_bmap(struct vop_bmap_args *ap)
1852 {
1853 	if (ap->a_doffsetp != NULL)
1854 		*ap->a_doffsetp = ap->a_loffset;
1855 	if (ap->a_runp != NULL)
1856 		*ap->a_runp = MAXBSIZE;
1857 	if (ap->a_runb != NULL) {
1858 		if (ap->a_loffset < MAXBSIZE)
1859 			*ap->a_runb = (int)ap->a_loffset;
1860 		else
1861 			*ap->a_runb = MAXBSIZE;
1862 	}
1863 	return (0);
1864 }
1865 
1866 
1867 /*
1868  * Special device advisory byte-level locks.
1869  *
1870  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1871  *		struct flock *a_fl, int a_flags)
1872  */
1873 /* ARGSUSED */
1874 static int
1875 devfs_spec_advlock(struct vop_advlock_args *ap)
1876 {
1877 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1878 }
1879 
1880 static void
1881 devfs_spec_getpages_iodone(struct bio *bio)
1882 {
1883 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
1884 	wakeup(bio->bio_buf);
1885 }
1886 
1887 /*
1888  * spec_getpages() - get pages associated with device vnode.
1889  *
1890  * Note that spec_read and spec_write do not use the buffer cache, so we
1891  * must fully implement getpages here.
1892  */
1893 static int
1894 devfs_spec_getpages(struct vop_getpages_args *ap)
1895 {
1896 	vm_offset_t kva;
1897 	int error;
1898 	int i, pcount, size;
1899 	struct buf *bp;
1900 	vm_page_t m;
1901 	vm_ooffset_t offset;
1902 	int toff, nextoff, nread;
1903 	struct vnode *vp = ap->a_vp;
1904 	int blksiz;
1905 	int gotreqpage;
1906 
1907 	error = 0;
1908 	pcount = round_page(ap->a_count) / PAGE_SIZE;
1909 
1910 	/*
1911 	 * Calculate the offset of the transfer and do sanity check.
1912 	 */
1913 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1914 
1915 	/*
1916 	 * Round up physical size for real devices.  We cannot round using
1917 	 * v_mount's block size data because v_mount has nothing to do with
1918 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
1919 	 * size for the device itself.
1920 	 *
1921 	 * We can't use v_rdev->si_mountpoint because it only exists when the
1922 	 * block device is mounted.  However, we can use v_rdev.
1923 	 */
1924 	if (vn_isdisk(vp, NULL))
1925 		blksiz = vp->v_rdev->si_bsize_phys;
1926 	else
1927 		blksiz = DEV_BSIZE;
1928 
1929 	size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
1930 
1931 	bp = getpbuf(NULL);
1932 	kva = (vm_offset_t)bp->b_data;
1933 
1934 	/*
1935 	 * Map the pages to be read into the kva.
1936 	 */
1937 	pmap_qenter(kva, ap->a_m, pcount);
1938 
1939 	/* Build a minimal buffer header. */
1940 	bp->b_cmd = BUF_CMD_READ;
1941 	bp->b_bcount = size;
1942 	bp->b_resid = 0;
1943 	bp->b_runningbufspace = size;
1944 	if (size) {
1945 		runningbufspace += bp->b_runningbufspace;
1946 		++runningbufcount;
1947 	}
1948 
1949 	bp->b_bio1.bio_offset = offset;
1950 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
1951 
1952 	mycpu->gd_cnt.v_vnodein++;
1953 	mycpu->gd_cnt.v_vnodepgsin += pcount;
1954 
1955 	/* Do the input. */
1956 	vn_strategy(ap->a_vp, &bp->b_bio1);
1957 
1958 	crit_enter();
1959 
1960 	/* We definitely need to be at splbio here. */
1961 	while (bp->b_cmd != BUF_CMD_DONE)
1962 		tsleep(bp, 0, "spread", 0);
1963 
1964 	crit_exit();
1965 
1966 	if (bp->b_flags & B_ERROR) {
1967 		if (bp->b_error)
1968 			error = bp->b_error;
1969 		else
1970 			error = EIO;
1971 	}
1972 
1973 	/*
1974 	 * If EOF is encountered we must zero-extend the result in order
1975 	 * to ensure that the page does not contain garabge.  When no
1976 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
1977 	 * b_resid is relative to b_bcount and should be 0, but some devices
1978 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
1979 	 */
1980 	nread = bp->b_bcount - bp->b_resid;
1981 	if (nread < ap->a_count)
1982 		bzero((caddr_t)kva + nread, ap->a_count - nread);
1983 	pmap_qremove(kva, pcount);
1984 
1985 	gotreqpage = 0;
1986 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
1987 		nextoff = toff + PAGE_SIZE;
1988 		m = ap->a_m[i];
1989 
1990 		m->flags &= ~PG_ZERO;
1991 
1992 		/*
1993 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
1994 		 *	 pmap modified bit.  pmap modified bit should have
1995 		 *	 already been cleared.
1996 		 */
1997 		if (nextoff <= nread) {
1998 			m->valid = VM_PAGE_BITS_ALL;
1999 			vm_page_undirty(m);
2000 		} else if (toff < nread) {
2001 			/*
2002 			 * Since this is a VM request, we have to supply the
2003 			 * unaligned offset to allow vm_page_set_valid()
2004 			 * to zero sub-DEV_BSIZE'd portions of the page.
2005 			 */
2006 			vm_page_set_valid(m, 0, nread - toff);
2007 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2008 		} else {
2009 			m->valid = 0;
2010 			vm_page_undirty(m);
2011 		}
2012 
2013 		if (i != ap->a_reqpage) {
2014 			/*
2015 			 * Just in case someone was asking for this page we
2016 			 * now tell them that it is ok to use.
2017 			 */
2018 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2019 				if (m->valid) {
2020 					if (m->flags & PG_WANTED) {
2021 						vm_page_activate(m);
2022 					} else {
2023 						vm_page_deactivate(m);
2024 					}
2025 					vm_page_wakeup(m);
2026 				} else {
2027 					vm_page_free(m);
2028 				}
2029 			} else {
2030 				vm_page_free(m);
2031 			}
2032 		} else if (m->valid) {
2033 			gotreqpage = 1;
2034 			/*
2035 			 * Since this is a VM request, we need to make the
2036 			 * entire page presentable by zeroing invalid sections.
2037 			 */
2038 			if (m->valid != VM_PAGE_BITS_ALL)
2039 			    vm_page_zero_invalid(m, FALSE);
2040 		}
2041 	}
2042 	if (!gotreqpage) {
2043 		m = ap->a_m[ap->a_reqpage];
2044 		devfs_debug(DEVFS_DEBUG_WARNING,
2045 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2046 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2047 		devfs_debug(DEVFS_DEBUG_WARNING,
2048 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2049 		    size, bp->b_resid, ap->a_count, m->valid);
2050 		devfs_debug(DEVFS_DEBUG_WARNING,
2051 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2052 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2053 		/*
2054 		 * Free the buffer header back to the swap buffer pool.
2055 		 */
2056 		relpbuf(bp, NULL);
2057 		return VM_PAGER_ERROR;
2058 	}
2059 	/*
2060 	 * Free the buffer header back to the swap buffer pool.
2061 	 */
2062 	relpbuf(bp, NULL);
2063 	if (DEVFS_NODE(ap->a_vp))
2064 		nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2065 	return VM_PAGER_OK;
2066 }
2067 
2068 static __inline
2069 int
2070 sequential_heuristic(struct uio *uio, struct file *fp)
2071 {
2072 	/*
2073 	 * Sequential heuristic - detect sequential operation
2074 	 */
2075 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2076 	    uio->uio_offset == fp->f_nextoff) {
2077 		/*
2078 		 * XXX we assume that the filesystem block size is
2079 		 * the default.  Not true, but still gives us a pretty
2080 		 * good indicator of how sequential the read operations
2081 		 * are.
2082 		 */
2083 		int tmpseq = fp->f_seqcount;
2084 
2085 		tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2086 		if (tmpseq > IO_SEQMAX)
2087 			tmpseq = IO_SEQMAX;
2088 		fp->f_seqcount = tmpseq;
2089 		return(fp->f_seqcount << IO_SEQSHIFT);
2090 	}
2091 
2092 	/*
2093 	 * Not sequential, quick draw-down of seqcount
2094 	 */
2095 	if (fp->f_seqcount > 1)
2096 		fp->f_seqcount = 1;
2097 	else
2098 		fp->f_seqcount = 0;
2099 	return(0);
2100 }
2101