xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision e586f31ca9899b49a4fc156613d9ecd853defcec)
1 /*
2  * (MPSAFE)
3  *
4  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Alex Hornung <ahornung@gmail.com>
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in
17  *    the documentation and/or other materials provided with the
18  *    distribution.
19  * 3. Neither the name of The DragonFly Project nor the names of its
20  *    contributors may be used to endorse or promote products derived
21  *    from this software without specific, prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/time.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/fcntl.h>
42 #include <sys/proc.h>
43 #include <sys/priv.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
46 #include <sys/uio.h>
47 #include <sys/mount.h>
48 #include <sys/file.h>
49 #include <sys/namei.h>
50 #include <sys/dirent.h>
51 #include <sys/malloc.h>
52 #include <sys/stat.h>
53 #include <sys/reg.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_zone.h>
56 #include <vm/vm_object.h>
57 #include <sys/filio.h>
58 #include <sys/ttycom.h>
59 #include <sys/tty.h>
60 #include <sys/diskslice.h>
61 #include <sys/sysctl.h>
62 #include <sys/devfs.h>
63 #include <sys/pioctl.h>
64 #include <vfs/fifofs/fifo.h>
65 
66 #include <machine/limits.h>
67 
68 #include <sys/buf2.h>
69 #include <sys/sysref2.h>
70 #include <vm/vm_page2.h>
71 
72 #ifndef SPEC_CHAIN_DEBUG
73 #define SPEC_CHAIN_DEBUG 0
74 #endif
75 
76 MALLOC_DECLARE(M_DEVFS);
77 #define DEVFS_BADOP	(void *)devfs_vop_badop
78 
79 static int devfs_vop_badop(struct vop_generic_args *);
80 static int devfs_vop_access(struct vop_access_args *);
81 static int devfs_vop_inactive(struct vop_inactive_args *);
82 static int devfs_vop_reclaim(struct vop_reclaim_args *);
83 static int devfs_vop_readdir(struct vop_readdir_args *);
84 static int devfs_vop_getattr(struct vop_getattr_args *);
85 static int devfs_vop_setattr(struct vop_setattr_args *);
86 static int devfs_vop_readlink(struct vop_readlink_args *);
87 static int devfs_vop_print(struct vop_print_args *);
88 
89 static int devfs_vop_nresolve(struct vop_nresolve_args *);
90 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
91 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
92 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
93 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
94 static int devfs_vop_nremove(struct vop_nremove_args *);
95 
96 static int devfs_spec_open(struct vop_open_args *);
97 static int devfs_spec_close(struct vop_close_args *);
98 static int devfs_spec_fsync(struct vop_fsync_args *);
99 
100 static int devfs_spec_read(struct vop_read_args *);
101 static int devfs_spec_write(struct vop_write_args *);
102 static int devfs_spec_ioctl(struct vop_ioctl_args *);
103 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
104 static int devfs_spec_strategy(struct vop_strategy_args *);
105 static void devfs_spec_strategy_done(struct bio *);
106 static int devfs_spec_freeblks(struct vop_freeblks_args *);
107 static int devfs_spec_bmap(struct vop_bmap_args *);
108 static int devfs_spec_advlock(struct vop_advlock_args *);
109 static void devfs_spec_getpages_iodone(struct bio *);
110 static int devfs_spec_getpages(struct vop_getpages_args *);
111 
112 static int devfs_fo_close(struct file *);
113 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
114 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
115 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
116 static int devfs_fo_kqfilter(struct file *, struct knote *);
117 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
118 				struct ucred *, struct sysmsg *);
119 static __inline int sequential_heuristic(struct uio *, struct file *);
120 
121 extern struct lock devfs_lock;
122 
123 /*
124  * devfs vnode operations for regular files.  All vnode ops are MPSAFE.
125  */
126 struct vop_ops devfs_vnode_norm_vops = {
127 	.vop_default =		vop_defaultop,
128 	.vop_access =		devfs_vop_access,
129 	.vop_advlock =		DEVFS_BADOP,
130 	.vop_bmap =		DEVFS_BADOP,
131 	.vop_close =		vop_stdclose,
132 	.vop_getattr =		devfs_vop_getattr,
133 	.vop_inactive =		devfs_vop_inactive,
134 	.vop_ncreate =		DEVFS_BADOP,
135 	.vop_nresolve =		devfs_vop_nresolve,
136 	.vop_nlookupdotdot =	devfs_vop_nlookupdotdot,
137 	.vop_nlink =		DEVFS_BADOP,
138 	.vop_nmkdir =		devfs_vop_nmkdir,
139 	.vop_nmknod =		DEVFS_BADOP,
140 	.vop_nremove =		devfs_vop_nremove,
141 	.vop_nrename =		DEVFS_BADOP,
142 	.vop_nrmdir =		devfs_vop_nrmdir,
143 	.vop_nsymlink =		devfs_vop_nsymlink,
144 	.vop_open =		vop_stdopen,
145 	.vop_pathconf =		vop_stdpathconf,
146 	.vop_print =		devfs_vop_print,
147 	.vop_read =		DEVFS_BADOP,
148 	.vop_readdir =		devfs_vop_readdir,
149 	.vop_readlink =		devfs_vop_readlink,
150 	.vop_reclaim =		devfs_vop_reclaim,
151 	.vop_setattr =		devfs_vop_setattr,
152 	.vop_write =		DEVFS_BADOP,
153 	.vop_ioctl =		DEVFS_BADOP
154 };
155 
156 /*
157  * devfs vnode operations for character devices.  All vnode ops are MPSAFE.
158  */
159 struct vop_ops devfs_vnode_dev_vops = {
160 	.vop_default =		vop_defaultop,
161 	.vop_access =		devfs_vop_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_vop_getattr,
168 	.vop_getpages =		devfs_spec_getpages,
169 	.vop_inactive =		devfs_vop_inactive,
170 	.vop_open =		devfs_spec_open,
171 	.vop_pathconf =		vop_stdpathconf,
172 	.vop_print =		devfs_vop_print,
173 	.vop_kqfilter =		devfs_spec_kqfilter,
174 	.vop_read =		devfs_spec_read,
175 	.vop_readdir =		DEVFS_BADOP,
176 	.vop_readlink =		DEVFS_BADOP,
177 	.vop_reclaim =		devfs_vop_reclaim,
178 	.vop_setattr =		devfs_vop_setattr,
179 	.vop_strategy =		devfs_spec_strategy,
180 	.vop_write =		devfs_spec_write,
181 	.vop_ioctl =		devfs_spec_ioctl
182 };
183 
184 /*
185  * devfs file pointer operations.  All fileops are MPSAFE.
186  */
187 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
188 
189 struct fileops devfs_dev_fileops = {
190 	.fo_read	= devfs_fo_read,
191 	.fo_write	= devfs_fo_write,
192 	.fo_ioctl	= devfs_fo_ioctl,
193 	.fo_kqfilter	= devfs_fo_kqfilter,
194 	.fo_stat	= devfs_fo_stat,
195 	.fo_close	= devfs_fo_close,
196 	.fo_shutdown	= nofo_shutdown
197 };
198 
199 /*
200  * These two functions are possibly temporary hacks for devices (aka
201  * the pty code) which want to control the node attributes themselves.
202  *
203  * XXX we may ultimately desire to simply remove the uid/gid/mode
204  * from the node entirely.
205  *
206  * MPSAFE - sorta.  Theoretically the overwrite can compete since they
207  *	    are loading from the same fields.
208  */
209 static __inline void
210 node_sync_dev_get(struct devfs_node *node)
211 {
212 	cdev_t dev;
213 
214 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
215 		node->uid = dev->si_uid;
216 		node->gid = dev->si_gid;
217 		node->mode = dev->si_perms;
218 	}
219 }
220 
221 static __inline void
222 node_sync_dev_set(struct devfs_node *node)
223 {
224 	cdev_t dev;
225 
226 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
227 		dev->si_uid = node->uid;
228 		dev->si_gid = node->gid;
229 		dev->si_perms = node->mode;
230 	}
231 }
232 
233 /*
234  * generic entry point for unsupported operations
235  */
236 static int
237 devfs_vop_badop(struct vop_generic_args *ap)
238 {
239 	return (EIO);
240 }
241 
242 
243 static int
244 devfs_vop_access(struct vop_access_args *ap)
245 {
246 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
247 	int error;
248 
249 	if (!devfs_node_is_accessible(node))
250 		return ENOENT;
251 	node_sync_dev_get(node);
252 	error = vop_helper_access(ap, node->uid, node->gid,
253 				  node->mode, node->flags);
254 
255 	return error;
256 }
257 
258 
259 static int
260 devfs_vop_inactive(struct vop_inactive_args *ap)
261 {
262 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
263 
264 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
265 		vrecycle(ap->a_vp);
266 	return 0;
267 }
268 
269 
270 static int
271 devfs_vop_reclaim(struct vop_reclaim_args *ap)
272 {
273 	struct devfs_node *node;
274 	struct vnode *vp;
275 	int locked;
276 
277 	/*
278 	 * Check if it is locked already. if not, we acquire the devfs lock
279 	 */
280 	if ((lockstatus(&devfs_lock, curthread)) != LK_EXCLUSIVE) {
281 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
282 		locked = 1;
283 	} else {
284 		locked = 0;
285 	}
286 
287 	/*
288 	 * Get rid of the devfs_node if it is no longer linked into the
289 	 * topology.
290 	 */
291 	vp = ap->a_vp;
292 	if ((node = DEVFS_NODE(vp)) != NULL) {
293 		node->v_node = NULL;
294 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
295 			devfs_freep(node);
296 	}
297 
298 	if (locked)
299 		lockmgr(&devfs_lock, LK_RELEASE);
300 
301 	/*
302 	 * v_rdev needs to be properly released using v_release_rdev
303 	 * Make sure v_data is NULL as well.
304 	 */
305 	vp->v_data = NULL;
306 	v_release_rdev(vp);
307 	return 0;
308 }
309 
310 
311 static int
312 devfs_vop_readdir(struct vop_readdir_args *ap)
313 {
314 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
315 	struct devfs_node *node;
316 	int cookie_index;
317 	int ncookies;
318 	int error2;
319 	int error;
320 	int r;
321 	off_t *cookies;
322 	off_t saveoff;
323 
324 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
325 
326 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
327 		return (EINVAL);
328 	error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
329 	if (error)
330 		return (error);
331 
332 	if (!devfs_node_is_accessible(dnode)) {
333 		vn_unlock(ap->a_vp);
334 		return ENOENT;
335 	}
336 
337 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
338 
339 	saveoff = ap->a_uio->uio_offset;
340 
341 	if (ap->a_ncookies) {
342 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
343 		if (ncookies > 256)
344 			ncookies = 256;
345 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
346 		cookie_index = 0;
347 	} else {
348 		ncookies = -1;
349 		cookies = NULL;
350 		cookie_index = 0;
351 	}
352 
353 	nanotime(&dnode->atime);
354 
355 	if (saveoff == 0) {
356 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
357 				     DT_DIR, 1, ".");
358 		if (r)
359 			goto done;
360 		if (cookies)
361 			cookies[cookie_index] = saveoff;
362 		saveoff++;
363 		cookie_index++;
364 		if (cookie_index == ncookies)
365 			goto done;
366 	}
367 
368 	if (saveoff == 1) {
369 		if (dnode->parent) {
370 			r = vop_write_dirent(&error, ap->a_uio,
371 					     dnode->parent->d_dir.d_ino,
372 					     DT_DIR, 2, "..");
373 		} else {
374 			r = vop_write_dirent(&error, ap->a_uio,
375 					     dnode->d_dir.d_ino,
376 					     DT_DIR, 2, "..");
377 		}
378 		if (r)
379 			goto done;
380 		if (cookies)
381 			cookies[cookie_index] = saveoff;
382 		saveoff++;
383 		cookie_index++;
384 		if (cookie_index == ncookies)
385 			goto done;
386 	}
387 
388 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
389 		if ((node->flags & DEVFS_HIDDEN) ||
390 		    (node->flags & DEVFS_INVISIBLE)) {
391 			continue;
392 		}
393 
394 		/*
395 		 * If the node type is a valid devfs alias, then we make
396 		 * sure that the target isn't hidden. If it is, we don't
397 		 * show the link in the directory listing.
398 		 */
399 		if ((node->node_type == Nlink) && (node->link_target != NULL) &&
400 			(node->link_target->flags & DEVFS_HIDDEN))
401 			continue;
402 
403 		if (node->cookie < saveoff)
404 			continue;
405 
406 		saveoff = node->cookie;
407 
408 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
409 					  node->d_dir.d_type,
410 					  node->d_dir.d_namlen,
411 					  node->d_dir.d_name);
412 
413 		if (error2)
414 			break;
415 
416 		saveoff++;
417 
418 		if (cookies)
419 			cookies[cookie_index] = node->cookie;
420 		++cookie_index;
421 		if (cookie_index == ncookies)
422 			break;
423 	}
424 
425 done:
426 	lockmgr(&devfs_lock, LK_RELEASE);
427 	vn_unlock(ap->a_vp);
428 
429 	ap->a_uio->uio_offset = saveoff;
430 	if (error && cookie_index == 0) {
431 		if (cookies) {
432 			kfree(cookies, M_TEMP);
433 			*ap->a_ncookies = 0;
434 			*ap->a_cookies = NULL;
435 		}
436 	} else {
437 		if (cookies) {
438 			*ap->a_ncookies = cookie_index;
439 			*ap->a_cookies = cookies;
440 		}
441 	}
442 	return (error);
443 }
444 
445 
446 static int
447 devfs_vop_nresolve(struct vop_nresolve_args *ap)
448 {
449 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
450 	struct devfs_node *node, *found = NULL;
451 	struct namecache *ncp;
452 	struct vnode *vp = NULL;
453 	int error = 0;
454 	int len;
455 	int depth;
456 
457 	ncp = ap->a_nch->ncp;
458 	len = ncp->nc_nlen;
459 
460 	if (!devfs_node_is_accessible(dnode))
461 		return ENOENT;
462 
463 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
464 
465 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) {
466 		error = ENOENT;
467 		cache_setvp(ap->a_nch, NULL);
468 		goto out;
469 	}
470 
471 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
472 		if (len == node->d_dir.d_namlen) {
473 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
474 				found = node;
475 				break;
476 			}
477 		}
478 	}
479 
480 	if (found) {
481 		depth = 0;
482 		while ((found->node_type == Nlink) && (found->link_target)) {
483 			if (depth >= 8) {
484 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
485 				break;
486 			}
487 
488 			found = found->link_target;
489 			++depth;
490 		}
491 
492 		if (!(found->flags & DEVFS_HIDDEN))
493 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
494 	}
495 
496 	if (vp == NULL) {
497 		error = ENOENT;
498 		cache_setvp(ap->a_nch, NULL);
499 		goto out;
500 
501 	}
502 	KKASSERT(vp);
503 	vn_unlock(vp);
504 	cache_setvp(ap->a_nch, vp);
505 	vrele(vp);
506 out:
507 	lockmgr(&devfs_lock, LK_RELEASE);
508 
509 	return error;
510 }
511 
512 
513 static int
514 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
515 {
516 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
517 
518 	*ap->a_vpp = NULL;
519 	if (!devfs_node_is_accessible(dnode))
520 		return ENOENT;
521 
522 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
523 	if (dnode->parent != NULL) {
524 		devfs_allocv(ap->a_vpp, dnode->parent);
525 		vn_unlock(*ap->a_vpp);
526 	}
527 	lockmgr(&devfs_lock, LK_RELEASE);
528 
529 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
530 }
531 
532 
533 static int
534 devfs_vop_getattr(struct vop_getattr_args *ap)
535 {
536 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
537 	struct vattr *vap = ap->a_vap;
538 	struct partinfo pinfo;
539 	int error = 0;
540 
541 #if 0
542 	if (!devfs_node_is_accessible(node))
543 		return ENOENT;
544 #endif
545 	node_sync_dev_get(node);
546 
547 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
548 
549 	/* start by zeroing out the attributes */
550 	VATTR_NULL(vap);
551 
552 	/* next do all the common fields */
553 	vap->va_type = ap->a_vp->v_type;
554 	vap->va_mode = node->mode;
555 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
556 	vap->va_flags = 0;
557 	vap->va_blocksize = DEV_BSIZE;
558 	vap->va_bytes = vap->va_size = 0;
559 
560 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
561 
562 	vap->va_atime = node->atime;
563 	vap->va_mtime = node->mtime;
564 	vap->va_ctime = node->ctime;
565 
566 	vap->va_nlink = 1; /* number of references to file */
567 
568 	vap->va_uid = node->uid;
569 	vap->va_gid = node->gid;
570 
571 	vap->va_rmajor = 0;
572 	vap->va_rminor = 0;
573 
574 	if ((node->node_type == Ndev) && node->d_dev)  {
575 		reference_dev(node->d_dev);
576 		vap->va_rminor = node->d_dev->si_uminor;
577 		release_dev(node->d_dev);
578 	}
579 
580 	/* For a softlink the va_size is the length of the softlink */
581 	if (node->symlink_name != 0) {
582 		vap->va_bytes = vap->va_size = node->symlink_namelen;
583 	}
584 
585 	/*
586 	 * For a disk-type device, va_size is the size of the underlying
587 	 * device, so that lseek() works properly.
588 	 */
589 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
590 		bzero(&pinfo, sizeof(pinfo));
591 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
592 				   0, proc0.p_ucred, NULL, NULL);
593 		if ((error == 0) && (pinfo.media_blksize != 0)) {
594 			vap->va_size = pinfo.media_size;
595 		} else {
596 			vap->va_size = 0;
597 			error = 0;
598 		}
599 	}
600 
601 	lockmgr(&devfs_lock, LK_RELEASE);
602 
603 	return (error);
604 }
605 
606 
607 static int
608 devfs_vop_setattr(struct vop_setattr_args *ap)
609 {
610 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
611 	struct vattr *vap;
612 	uid_t cur_uid;
613 	gid_t cur_gid;
614 	mode_t cur_mode;
615 	int error = 0;
616 
617 	if (!devfs_node_is_accessible(node))
618 		return ENOENT;
619 	node_sync_dev_get(node);
620 
621 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
622 
623 	vap = ap->a_vap;
624 
625 	if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
626 		cur_uid = node->uid;
627 		cur_gid = node->gid;
628 		cur_mode = node->mode;
629 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
630 		    ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
631 		if (error)
632 			goto out;
633 
634 		if (node->uid != cur_uid || node->gid != cur_gid) {
635 			node->uid = cur_uid;
636 			node->gid = cur_gid;
637 			node->mode = cur_mode;
638 		}
639 	}
640 
641 	if (vap->va_mode != (mode_t)VNOVAL) {
642 		cur_mode = node->mode;
643 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
644 		    node->uid, node->gid, &cur_mode);
645 		if (error == 0 && node->mode != cur_mode) {
646 			node->mode = cur_mode;
647 		}
648 	}
649 
650 out:
651 	node_sync_dev_set(node);
652 	nanotime(&node->ctime);
653 	lockmgr(&devfs_lock, LK_RELEASE);
654 
655 	return error;
656 }
657 
658 
659 static int
660 devfs_vop_readlink(struct vop_readlink_args *ap)
661 {
662 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
663 	int ret;
664 
665 	if (!devfs_node_is_accessible(node))
666 		return ENOENT;
667 
668 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
669 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
670 	lockmgr(&devfs_lock, LK_RELEASE);
671 
672 	return ret;
673 }
674 
675 
676 static int
677 devfs_vop_print(struct vop_print_args *ap)
678 {
679 	return (0);
680 }
681 
682 static int
683 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
684 {
685 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
686 	struct devfs_node *node;
687 
688 	if (!devfs_node_is_accessible(dnode))
689 		return ENOENT;
690 
691 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
692 		goto out;
693 
694 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
695 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir,
696 		      ap->a_nch->ncp->nc_name, dnode, NULL);
697 
698 	if (*ap->a_vpp) {
699 		node = DEVFS_NODE(*ap->a_vpp);
700 		node->flags |= DEVFS_USER_CREATED;
701 		cache_setunresolved(ap->a_nch);
702 		cache_setvp(ap->a_nch, *ap->a_vpp);
703 	}
704 	lockmgr(&devfs_lock, LK_RELEASE);
705 out:
706 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
707 }
708 
709 static int
710 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
711 {
712 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
713 	struct devfs_node *node;
714 	size_t targetlen;
715 
716 	if (!devfs_node_is_accessible(dnode))
717 		return ENOENT;
718 
719 	ap->a_vap->va_type = VLNK;
720 
721 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
722 		goto out;
723 
724 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
725 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink,
726 		      ap->a_nch->ncp->nc_name, dnode, NULL);
727 
728 	targetlen = strlen(ap->a_target);
729 	if (*ap->a_vpp) {
730 		node = DEVFS_NODE(*ap->a_vpp);
731 		node->flags |= DEVFS_USER_CREATED;
732 		node->symlink_namelen = targetlen;
733 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
734 		memcpy(node->symlink_name, ap->a_target, targetlen);
735 		node->symlink_name[targetlen] = '\0';
736 		cache_setunresolved(ap->a_nch);
737 		cache_setvp(ap->a_nch, *ap->a_vpp);
738 	}
739 	lockmgr(&devfs_lock, LK_RELEASE);
740 out:
741 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
742 }
743 
744 static int
745 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
746 {
747 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
748 	struct devfs_node *node;
749 	struct namecache *ncp;
750 	int error = ENOENT;
751 
752 	ncp = ap->a_nch->ncp;
753 
754 	if (!devfs_node_is_accessible(dnode))
755 		return ENOENT;
756 
757 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
758 
759 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
760 		goto out;
761 
762 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
763 		if (ncp->nc_nlen != node->d_dir.d_namlen)
764 			continue;
765 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
766 			continue;
767 
768 		/*
769 		 * only allow removal of user created dirs
770 		 */
771 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
772 			error = EPERM;
773 			goto out;
774 		} else if (node->node_type != Ndir) {
775 			error = ENOTDIR;
776 			goto out;
777 		} else if (node->nchildren > 2) {
778 			error = ENOTEMPTY;
779 			goto out;
780 		} else {
781 			if (node->v_node)
782 				cache_inval_vp(node->v_node, CINV_DESTROY);
783 			devfs_unlinkp(node);
784 			error = 0;
785 			break;
786 		}
787 	}
788 
789 	cache_unlink(ap->a_nch);
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 != Nroot) && (dnode->node_type != Ndir))
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 == Ndir) {
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_unlink(ap->a_nch);
838 out:
839 	lockmgr(&devfs_lock, LK_RELEASE);
840 	return error;
841 }
842 
843 
844 static int
845 devfs_spec_open(struct vop_open_args *ap)
846 {
847 	struct vnode *vp = ap->a_vp;
848 	struct vnode *orig_vp = NULL;
849 	struct devfs_node *node = DEVFS_NODE(vp);
850 	struct devfs_node *newnode;
851 	cdev_t dev, ndev = NULL;
852 	int error = 0;
853 
854 	if (node) {
855 		if (node->d_dev == NULL)
856 			return ENXIO;
857 		if (!devfs_node_is_accessible(node))
858 			return ENOENT;
859 	}
860 
861 	if ((dev = vp->v_rdev) == NULL)
862 		return ENXIO;
863 
864 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
865 
866 	if (node && ap->a_fp) {
867 		int exists;
868 
869 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
870 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
871 
872 		ndev = devfs_clone(dev, node->d_dir.d_name,
873 				   node->d_dir.d_namlen,
874 				   ap->a_mode, ap->a_cred);
875 		if (ndev != NULL) {
876 			newnode = devfs_create_device_node(
877 					DEVFS_MNTDATA(vp->v_mount)->root_node,
878 					ndev, &exists, NULL, NULL);
879 			/* XXX: possibly destroy device if this happens */
880 
881 			if (newnode != NULL) {
882 				dev = ndev;
883 				if (exists == 0)
884 					devfs_link_dev(dev);
885 
886 				devfs_debug(DEVFS_DEBUG_DEBUG,
887 						"parent here is: %s, node is: |%s|\n",
888 						((node->parent->node_type == Nroot) ?
889 						"ROOT!" : node->parent->d_dir.d_name),
890 						newnode->d_dir.d_name);
891 				devfs_debug(DEVFS_DEBUG_DEBUG,
892 						"test: %s\n",
893 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
894 
895 				/*
896 				 * orig_vp is set to the original vp if we
897 				 * cloned.
898 				 */
899 				/* node->flags |= DEVFS_CLONED; */
900 				devfs_allocv(&vp, newnode);
901 				orig_vp = ap->a_vp;
902 				ap->a_vp = vp;
903 			}
904 		}
905 		lockmgr(&devfs_lock, LK_RELEASE);
906 		/*
907 		 * Synchronize devfs here to make sure that, if the cloned
908 		 * device creates other device nodes in addition to the
909 		 * cloned one, all of them are created by the time we return
910 		 * from opening the cloned one.
911 		 */
912 		if (ndev)
913 			devfs_config();
914 	}
915 
916 	devfs_debug(DEVFS_DEBUG_DEBUG,
917 		    "devfs_spec_open() called on %s! \n",
918 		    dev->si_name);
919 
920 	/*
921 	 * Make this field valid before any I/O in ->d_open
922 	 */
923 	if (!dev->si_iosize_max)
924 		/* XXX: old DFLTPHYS == 64KB dependency */
925 		dev->si_iosize_max = min(MAXPHYS,64*1024);
926 
927 	if (dev_dflags(dev) & D_TTY)
928 		vsetflags(vp, VISTTY);
929 
930 	/*
931 	 * Open underlying device
932 	 */
933 	vn_unlock(vp);
934 	error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred, ap->a_fp);
935 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
936 
937 	/*
938 	 * Clean up any cloned vp if we error out.
939 	 */
940 	if (error) {
941 		if (orig_vp) {
942 			vput(vp);
943 			ap->a_vp = orig_vp;
944 			/* orig_vp = NULL; */
945 		}
946 		return error;
947 	}
948 
949 	/*
950 	 * This checks if the disk device is going to be opened for writing.
951 	 * It will be only allowed in the cases where securelevel permits it
952 	 * and it's not mounted R/W.
953 	 */
954 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
955 	    (ap->a_cred != FSCRED)) {
956 
957 		/* Very secure mode. No open for writing allowed */
958 		if (securelevel >= 2)
959 			return EPERM;
960 
961 		/*
962 		 * If it is mounted R/W, do not allow to open for writing.
963 		 * In the case it's mounted read-only but securelevel
964 		 * is >= 1, then do not allow opening for writing either.
965 		 */
966 		if (vfs_mountedon(vp)) {
967 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
968 				return EBUSY;
969 			else if (securelevel >= 1)
970 				return EPERM;
971 		}
972 	}
973 
974 	if (dev_dflags(dev) & D_TTY) {
975 		if (dev->si_tty) {
976 			struct tty *tp;
977 			tp = dev->si_tty;
978 			if (!tp->t_stop) {
979 				devfs_debug(DEVFS_DEBUG_DEBUG,
980 					    "devfs: no t_stop\n");
981 				tp->t_stop = nottystop;
982 			}
983 		}
984 	}
985 
986 
987 	if (vn_isdisk(vp, NULL)) {
988 		if (!dev->si_bsize_phys)
989 			dev->si_bsize_phys = DEV_BSIZE;
990 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
991 	}
992 
993 	vop_stdopen(ap);
994 #if 0
995 	if (node)
996 		nanotime(&node->atime);
997 #endif
998 
999 	/*
1000 	 * If we replaced the vp the vop_stdopen() call will have loaded
1001 	 * it into fp->f_data and vref()d the vp, giving us two refs.  So
1002 	 * instead of just unlocking it here we have to vput() it.
1003 	 */
1004 	if (orig_vp)
1005 		vput(vp);
1006 
1007 	/* Ugly pty magic, to make pty devices appear once they are opened */
1008 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
1009 		node->flags &= ~DEVFS_INVISIBLE;
1010 
1011 	if (ap->a_fp) {
1012 		KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
1013 		KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1014 		ap->a_fp->f_ops = &devfs_dev_fileops;
1015 		KKASSERT(ap->a_fp->f_data == (void *)vp);
1016 	}
1017 
1018 	return 0;
1019 }
1020 
1021 static int
1022 devfs_spec_close(struct vop_close_args *ap)
1023 {
1024 	struct devfs_node *node;
1025 	struct proc *p = curproc;
1026 	struct vnode *vp = ap->a_vp;
1027 	cdev_t dev = vp->v_rdev;
1028 	int error = 0;
1029 	int needrelock;
1030 	int opencount;
1031 
1032 	/*
1033 	 * We do special tests on the opencount so unfortunately we need
1034 	 * an exclusive lock.
1035 	 */
1036 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
1037 
1038 	if (dev)
1039 		devfs_debug(DEVFS_DEBUG_DEBUG,
1040 			    "devfs_spec_close() called on %s! \n",
1041 			    dev->si_name);
1042 	else
1043 		devfs_debug(DEVFS_DEBUG_DEBUG,
1044 			    "devfs_spec_close() called, null vode!\n");
1045 
1046 	/*
1047 	 * A couple of hacks for devices and tty devices.  The
1048 	 * vnode ref count cannot be used to figure out the
1049 	 * last close, but we can use v_opencount now that
1050 	 * revoke works properly.
1051 	 *
1052 	 * Detect the last close on a controlling terminal and clear
1053 	 * the session (half-close).
1054 	 *
1055 	 * XXX opencount is not SMP safe.  The vnode is locked but there
1056 	 *     may be multiple vnodes referencing the same device.
1057 	 */
1058 	if (dev) {
1059 		/*
1060 		 * NOTE: Try to avoid global tokens when testing opencount
1061 		 * XXX hack, fixme. needs a struct lock and opencount in
1062 		 * struct cdev itself.
1063 		 */
1064 		reference_dev(dev);
1065 		opencount = vp->v_opencount;
1066 		if (opencount <= 1)
1067 			opencount = count_dev(dev);   /* XXX NOT SMP SAFE */
1068 	} else {
1069 		opencount = 0;
1070 	}
1071 
1072 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1073 		p->p_session->s_ttyvp = NULL;
1074 		vrele(vp);
1075 	}
1076 
1077 	/*
1078 	 * Vnodes can be opened and closed multiple times.  Do not really
1079 	 * close the device unless (1) it is being closed forcibly,
1080 	 * (2) the device wants to track closes, or (3) this is the last
1081 	 * vnode doing its last close on the device.
1082 	 *
1083 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1084 	 * a closed device.  This might not occur now that our revoke is
1085 	 * fixed.
1086 	 */
1087 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1088 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1089 	    (dev_dflags(dev) & D_TRACKCLOSE) ||
1090 	    (opencount == 1))) {
1091 		/*
1092 		 * Ugly pty magic, to make pty devices disappear again once
1093 		 * they are closed.
1094 		 */
1095 		node = DEVFS_NODE(ap->a_vp);
1096 		if (node && (node->flags & DEVFS_PTY))
1097 			node->flags |= DEVFS_INVISIBLE;
1098 
1099 		/*
1100 		 * Unlock around dev_dclose(), unless the vnode is
1101 		 * undergoing a vgone/reclaim (during umount).
1102 		 */
1103 		needrelock = 0;
1104 		if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1105 			needrelock = 1;
1106 			vn_unlock(vp);
1107 		}
1108 
1109 		/*
1110 		 * WARNING!  If the device destroys itself the devfs node
1111 		 *	     can disappear here.
1112 		 *
1113 		 * WARNING!  vn_lock() will fail if the vp is in a VRECLAIM,
1114 		 *	     which can occur during umount.
1115 		 */
1116 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1117 		/* node is now stale */
1118 
1119 		if (needrelock) {
1120 			if (vn_lock(vp, LK_EXCLUSIVE |
1121 					LK_RETRY |
1122 					LK_FAILRECLAIM) != 0) {
1123 				panic("devfs_spec_close: vnode %p "
1124 				      "unexpectedly could not be relocked",
1125 				      vp);
1126 			}
1127 		}
1128 	} else {
1129 		error = 0;
1130 	}
1131 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1132 
1133 	/*
1134 	 * Track the actual opens and closes on the vnode.  The last close
1135 	 * disassociates the rdev.  If the rdev is already disassociated or
1136 	 * the opencount is already 0, the vnode might have been revoked
1137 	 * and no further opencount tracking occurs.
1138 	 */
1139 	if (dev)
1140 		release_dev(dev);
1141 	if (vp->v_opencount > 0)
1142 		vop_stdclose(ap);
1143 	return(error);
1144 
1145 }
1146 
1147 
1148 static int
1149 devfs_fo_close(struct file *fp)
1150 {
1151 	struct vnode *vp = (struct vnode *)fp->f_data;
1152 	int error;
1153 
1154 	fp->f_ops = &badfileops;
1155 	error = vn_close(vp, fp->f_flag, fp);
1156 	devfs_clear_cdevpriv(fp);
1157 
1158 	return (error);
1159 }
1160 
1161 
1162 /*
1163  * Device-optimized file table vnode read routine.
1164  *
1165  * This bypasses the VOP table and talks directly to the device.  Most
1166  * filesystems just route to specfs and can make this optimization.
1167  */
1168 static int
1169 devfs_fo_read(struct file *fp, struct uio *uio,
1170 		 struct ucred *cred, int flags)
1171 {
1172 	struct devfs_node *node;
1173 	struct vnode *vp;
1174 	int ioflag;
1175 	int error;
1176 	cdev_t dev;
1177 
1178 	KASSERT(uio->uio_td == curthread,
1179 		("uio_td %p is not td %p", uio->uio_td, curthread));
1180 
1181 	if (uio->uio_resid == 0)
1182 		return 0;
1183 
1184 	vp = (struct vnode *)fp->f_data;
1185 	if (vp == NULL || vp->v_type == VBAD)
1186 		return EBADF;
1187 
1188 	node = DEVFS_NODE(vp);
1189 
1190 	if ((dev = vp->v_rdev) == NULL)
1191 		return EBADF;
1192 
1193 	reference_dev(dev);
1194 
1195 	if ((flags & O_FOFFSET) == 0)
1196 		uio->uio_offset = fp->f_offset;
1197 
1198 	ioflag = 0;
1199 	if (flags & O_FBLOCKING) {
1200 		/* ioflag &= ~IO_NDELAY; */
1201 	} else if (flags & O_FNONBLOCKING) {
1202 		ioflag |= IO_NDELAY;
1203 	} else if (fp->f_flag & FNONBLOCK) {
1204 		ioflag |= IO_NDELAY;
1205 	}
1206 	if (fp->f_flag & O_DIRECT) {
1207 		ioflag |= IO_DIRECT;
1208 	}
1209 	ioflag |= sequential_heuristic(uio, fp);
1210 
1211 	error = dev_dread(dev, uio, ioflag, fp);
1212 
1213 	release_dev(dev);
1214 	if (node)
1215 		nanotime(&node->atime);
1216 	if ((flags & O_FOFFSET) == 0)
1217 		fp->f_offset = uio->uio_offset;
1218 	fp->f_nextoff = uio->uio_offset;
1219 
1220 	return (error);
1221 }
1222 
1223 
1224 static int
1225 devfs_fo_write(struct file *fp, struct uio *uio,
1226 		  struct ucred *cred, int flags)
1227 {
1228 	struct devfs_node *node;
1229 	struct vnode *vp;
1230 	int ioflag;
1231 	int error;
1232 	cdev_t dev;
1233 
1234 	KASSERT(uio->uio_td == curthread,
1235 		("uio_td %p is not p %p", uio->uio_td, curthread));
1236 
1237 	vp = (struct vnode *)fp->f_data;
1238 	if (vp == NULL || vp->v_type == VBAD)
1239 		return EBADF;
1240 
1241 	node = DEVFS_NODE(vp);
1242 
1243 	if (vp->v_type == VREG)
1244 		bwillwrite(uio->uio_resid);
1245 
1246 	vp = (struct vnode *)fp->f_data;
1247 
1248 	if ((dev = vp->v_rdev) == NULL)
1249 		return EBADF;
1250 
1251 	reference_dev(dev);
1252 
1253 	if ((flags & O_FOFFSET) == 0)
1254 		uio->uio_offset = fp->f_offset;
1255 
1256 	ioflag = IO_UNIT;
1257 	if (vp->v_type == VREG &&
1258 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1259 		ioflag |= IO_APPEND;
1260 	}
1261 
1262 	if (flags & O_FBLOCKING) {
1263 		/* ioflag &= ~IO_NDELAY; */
1264 	} else if (flags & O_FNONBLOCKING) {
1265 		ioflag |= IO_NDELAY;
1266 	} else if (fp->f_flag & FNONBLOCK) {
1267 		ioflag |= IO_NDELAY;
1268 	}
1269 	if (fp->f_flag & O_DIRECT) {
1270 		ioflag |= IO_DIRECT;
1271 	}
1272 	if (flags & O_FASYNCWRITE) {
1273 		/* ioflag &= ~IO_SYNC; */
1274 	} else if (flags & O_FSYNCWRITE) {
1275 		ioflag |= IO_SYNC;
1276 	} else if (fp->f_flag & O_FSYNC) {
1277 		ioflag |= IO_SYNC;
1278 	}
1279 
1280 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1281 		ioflag |= IO_SYNC;
1282 	ioflag |= sequential_heuristic(uio, fp);
1283 
1284 	error = dev_dwrite(dev, uio, ioflag, fp);
1285 
1286 	release_dev(dev);
1287 	if (node) {
1288 		nanotime(&node->atime);
1289 		nanotime(&node->mtime);
1290 	}
1291 
1292 	if ((flags & O_FOFFSET) == 0)
1293 		fp->f_offset = uio->uio_offset;
1294 	fp->f_nextoff = uio->uio_offset;
1295 
1296 	return (error);
1297 }
1298 
1299 
1300 static int
1301 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1302 {
1303 	struct vnode *vp;
1304 	struct vattr vattr;
1305 	struct vattr *vap;
1306 	u_short mode;
1307 	cdev_t dev;
1308 	int error;
1309 
1310 	vp = (struct vnode *)fp->f_data;
1311 	if (vp == NULL || vp->v_type == VBAD)
1312 		return EBADF;
1313 
1314 	error = vn_stat(vp, sb, cred);
1315 	if (error)
1316 		return (error);
1317 
1318 	vap = &vattr;
1319 	error = VOP_GETATTR(vp, vap);
1320 	if (error)
1321 		return (error);
1322 
1323 	/*
1324 	 * Zero the spare stat fields
1325 	 */
1326 	sb->st_lspare = 0;
1327 	sb->st_qspare1 = 0;
1328 	sb->st_qspare2 = 0;
1329 
1330 	/*
1331 	 * Copy from vattr table ... or not in case it's a cloned device
1332 	 */
1333 	if (vap->va_fsid != VNOVAL)
1334 		sb->st_dev = vap->va_fsid;
1335 	else
1336 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1337 
1338 	sb->st_ino = vap->va_fileid;
1339 
1340 	mode = vap->va_mode;
1341 	mode |= S_IFCHR;
1342 	sb->st_mode = mode;
1343 
1344 	if (vap->va_nlink > (nlink_t)-1)
1345 		sb->st_nlink = (nlink_t)-1;
1346 	else
1347 		sb->st_nlink = vap->va_nlink;
1348 
1349 	sb->st_uid = vap->va_uid;
1350 	sb->st_gid = vap->va_gid;
1351 	sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1352 	sb->st_size = vap->va_bytes;
1353 	sb->st_atimespec = vap->va_atime;
1354 	sb->st_mtimespec = vap->va_mtime;
1355 	sb->st_ctimespec = vap->va_ctime;
1356 
1357 	/*
1358 	 * A VCHR and VBLK device may track the last access and last modified
1359 	 * time independantly of the filesystem.  This is particularly true
1360 	 * because device read and write calls may bypass the filesystem.
1361 	 */
1362 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1363 		dev = vp->v_rdev;
1364 		if (dev != NULL) {
1365 			if (dev->si_lastread) {
1366 				sb->st_atimespec.tv_sec = time_second +
1367 							  (time_uptime -
1368 							   dev->si_lastread);
1369 				sb->st_atimespec.tv_nsec = 0;
1370 			}
1371 			if (dev->si_lastwrite) {
1372 				sb->st_atimespec.tv_sec = time_second +
1373 							  (time_uptime -
1374 							   dev->si_lastwrite);
1375 				sb->st_atimespec.tv_nsec = 0;
1376 			}
1377 		}
1378 	}
1379 
1380         /*
1381 	 * According to www.opengroup.org, the meaning of st_blksize is
1382 	 *   "a filesystem-specific preferred I/O block size for this
1383 	 *    object.  In some filesystem types, this may vary from file
1384 	 *    to file"
1385 	 * Default to PAGE_SIZE after much discussion.
1386 	 */
1387 
1388 	sb->st_blksize = PAGE_SIZE;
1389 
1390 	sb->st_flags = vap->va_flags;
1391 
1392 	error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1393 	if (error)
1394 		sb->st_gen = 0;
1395 	else
1396 		sb->st_gen = (u_int32_t)vap->va_gen;
1397 
1398 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1399 
1400 	return (0);
1401 }
1402 
1403 
1404 static int
1405 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1406 {
1407 	struct vnode *vp;
1408 	int error;
1409 	cdev_t dev;
1410 
1411 	vp = (struct vnode *)fp->f_data;
1412 	if (vp == NULL || vp->v_type == VBAD) {
1413 		error = EBADF;
1414 		goto done;
1415 	}
1416 	if ((dev = vp->v_rdev) == NULL) {
1417 		error = EBADF;
1418 		goto done;
1419 	}
1420 	reference_dev(dev);
1421 
1422 	error = dev_dkqfilter(dev, kn, fp);
1423 
1424 	release_dev(dev);
1425 
1426 done:
1427 	return (error);
1428 }
1429 
1430 static int
1431 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1432 		  struct ucred *ucred, struct sysmsg *msg)
1433 {
1434 #if 0
1435 	struct devfs_node *node;
1436 #endif
1437 	struct vnode *vp;
1438 	struct vnode *ovp;
1439 	cdev_t	dev;
1440 	int error;
1441 	struct fiodname_args *name_args;
1442 	size_t namlen;
1443 	const char *name;
1444 
1445 	vp = ((struct vnode *)fp->f_data);
1446 
1447 	if ((dev = vp->v_rdev) == NULL)
1448 		return EBADF;		/* device was revoked */
1449 
1450 	reference_dev(dev);
1451 
1452 #if 0
1453 	node = DEVFS_NODE(vp);
1454 #endif
1455 
1456 	devfs_debug(DEVFS_DEBUG_DEBUG,
1457 		    "devfs_fo_ioctl() called! for dev %s\n",
1458 		    dev->si_name);
1459 
1460 	if (com == FIODTYPE) {
1461 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1462 		error = 0;
1463 		goto out;
1464 	} else if (com == FIODNAME) {
1465 		name_args = (struct fiodname_args *)data;
1466 		name = dev->si_name;
1467 		namlen = strlen(name) + 1;
1468 
1469 		devfs_debug(DEVFS_DEBUG_DEBUG,
1470 			    "ioctl, got: FIODNAME for %s\n", name);
1471 
1472 		if (namlen <= name_args->len)
1473 			error = copyout(dev->si_name, name_args->name, namlen);
1474 		else
1475 			error = EINVAL;
1476 
1477 		devfs_debug(DEVFS_DEBUG_DEBUG,
1478 			    "ioctl stuff: error: %d\n", error);
1479 		goto out;
1480 	}
1481 
1482 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg, fp);
1483 
1484 #if 0
1485 	if (node) {
1486 		nanotime(&node->atime);
1487 		nanotime(&node->mtime);
1488 	}
1489 #endif
1490 	if (com == TIOCSCTTY) {
1491 		devfs_debug(DEVFS_DEBUG_DEBUG,
1492 			    "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1493 			    dev->si_name);
1494 	}
1495 	if (error == 0 && com == TIOCSCTTY) {
1496 		struct proc *p = curthread->td_proc;
1497 		struct session *sess;
1498 
1499 		devfs_debug(DEVFS_DEBUG_DEBUG,
1500 			    "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1501 			    dev->si_name);
1502 		if (p == NULL) {
1503 			error = ENOTTY;
1504 			goto out;
1505 		}
1506 		sess = p->p_session;
1507 
1508 		/*
1509 		 * Do nothing if reassigning same control tty
1510 		 */
1511 		if (sess->s_ttyvp == vp) {
1512 			error = 0;
1513 			goto out;
1514 		}
1515 
1516 		/*
1517 		 * Get rid of reference to old control tty
1518 		 */
1519 		ovp = sess->s_ttyvp;
1520 		vref(vp);
1521 		sess->s_ttyvp = vp;
1522 		if (ovp)
1523 			vrele(ovp);
1524 	}
1525 
1526 out:
1527 	release_dev(dev);
1528 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1529 	return (error);
1530 }
1531 
1532 
1533 static int
1534 devfs_spec_fsync(struct vop_fsync_args *ap)
1535 {
1536 	struct vnode *vp = ap->a_vp;
1537 	int error;
1538 
1539 	if (!vn_isdisk(vp, NULL))
1540 		return (0);
1541 
1542 	/*
1543 	 * Flush all dirty buffers associated with a block device.
1544 	 */
1545 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1546 	return (error);
1547 }
1548 
1549 static int
1550 devfs_spec_read(struct vop_read_args *ap)
1551 {
1552 	struct devfs_node *node;
1553 	struct vnode *vp;
1554 	struct uio *uio;
1555 	cdev_t dev;
1556 	int error;
1557 
1558 	vp = ap->a_vp;
1559 	dev = vp->v_rdev;
1560 	uio = ap->a_uio;
1561 	node = DEVFS_NODE(vp);
1562 
1563 	if (dev == NULL)		/* device was revoked */
1564 		return (EBADF);
1565 	if (uio->uio_resid == 0)
1566 		return (0);
1567 
1568 	vn_unlock(vp);
1569 	error = dev_dread(dev, uio, ap->a_ioflag, NULL);
1570 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1571 
1572 	if (node)
1573 		nanotime(&node->atime);
1574 
1575 	return (error);
1576 }
1577 
1578 /*
1579  * Vnode op for write
1580  *
1581  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1582  *	      struct ucred *a_cred)
1583  */
1584 static int
1585 devfs_spec_write(struct vop_write_args *ap)
1586 {
1587 	struct devfs_node *node;
1588 	struct vnode *vp;
1589 	struct uio *uio;
1590 	cdev_t dev;
1591 	int error;
1592 
1593 	vp = ap->a_vp;
1594 	dev = vp->v_rdev;
1595 	uio = ap->a_uio;
1596 	node = DEVFS_NODE(vp);
1597 
1598 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1599 
1600 	if (dev == NULL)		/* device was revoked */
1601 		return (EBADF);
1602 
1603 	vn_unlock(vp);
1604 	error = dev_dwrite(dev, uio, ap->a_ioflag, NULL);
1605 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1606 
1607 	if (node) {
1608 		nanotime(&node->atime);
1609 		nanotime(&node->mtime);
1610 	}
1611 
1612 	return (error);
1613 }
1614 
1615 /*
1616  * Device ioctl operation.
1617  *
1618  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1619  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1620  */
1621 static int
1622 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1623 {
1624 	struct vnode *vp = ap->a_vp;
1625 #if 0
1626 	struct devfs_node *node;
1627 #endif
1628 	cdev_t dev;
1629 
1630 	if ((dev = vp->v_rdev) == NULL)
1631 		return (EBADF);		/* device was revoked */
1632 #if 0
1633 	node = DEVFS_NODE(vp);
1634 
1635 	if (node) {
1636 		nanotime(&node->atime);
1637 		nanotime(&node->mtime);
1638 	}
1639 #endif
1640 
1641 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1642 			   ap->a_cred, ap->a_sysmsg, NULL));
1643 }
1644 
1645 /*
1646  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1647  */
1648 /* ARGSUSED */
1649 static int
1650 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1651 {
1652 	struct vnode *vp = ap->a_vp;
1653 #if 0
1654 	struct devfs_node *node;
1655 #endif
1656 	cdev_t dev;
1657 
1658 	if ((dev = vp->v_rdev) == NULL)
1659 		return (EBADF);		/* device was revoked (EBADF) */
1660 #if 0
1661 	node = DEVFS_NODE(vp);
1662 
1663 	if (node)
1664 		nanotime(&node->atime);
1665 #endif
1666 
1667 	return (dev_dkqfilter(dev, ap->a_kn, NULL));
1668 }
1669 
1670 /*
1671  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1672  * calls are not limited to device DMA limits so we have to deal with the
1673  * case.
1674  *
1675  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1676  */
1677 static int
1678 devfs_spec_strategy(struct vop_strategy_args *ap)
1679 {
1680 	struct bio *bio = ap->a_bio;
1681 	struct buf *bp = bio->bio_buf;
1682 	struct buf *nbp;
1683 	struct vnode *vp;
1684 	struct mount *mp;
1685 	int chunksize;
1686 	int maxiosize;
1687 
1688 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1689 		buf_start(bp);
1690 
1691 	/*
1692 	 * Collect statistics on synchronous and asynchronous read
1693 	 * and write counts for disks that have associated filesystems.
1694 	 */
1695 	vp = ap->a_vp;
1696 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1697 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1698 		if (bp->b_cmd == BUF_CMD_READ) {
1699 			if (bp->b_flags & BIO_SYNC)
1700 				mp->mnt_stat.f_syncreads++;
1701 			else
1702 				mp->mnt_stat.f_asyncreads++;
1703 		} else {
1704 			if (bp->b_flags & BIO_SYNC)
1705 				mp->mnt_stat.f_syncwrites++;
1706 			else
1707 				mp->mnt_stat.f_asyncwrites++;
1708 		}
1709 	}
1710 
1711         /*
1712          * Device iosize limitations only apply to read and write.  Shortcut
1713          * the I/O if it fits.
1714          */
1715 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1716 		devfs_debug(DEVFS_DEBUG_DEBUG,
1717 			    "%s: si_iosize_max not set!\n",
1718 			    dev_dname(vp->v_rdev));
1719 		maxiosize = MAXPHYS;
1720 	}
1721 #if SPEC_CHAIN_DEBUG & 2
1722 	maxiosize = 4096;
1723 #endif
1724         if (bp->b_bcount <= maxiosize ||
1725             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1726                 dev_dstrategy_chain(vp->v_rdev, bio);
1727                 return (0);
1728         }
1729 
1730 	/*
1731 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1732 	 */
1733 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1734 	initbufbio(nbp);
1735 	buf_dep_init(nbp);
1736 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1737 	BUF_KERNPROC(nbp);
1738 	nbp->b_vp = vp;
1739 	nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1740 	nbp->b_data = bp->b_data;
1741 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1742 	nbp->b_bio1.bio_offset = bio->bio_offset;
1743 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1744 
1745 	/*
1746 	 * Start the first transfer
1747 	 */
1748 	if (vn_isdisk(vp, NULL))
1749 		chunksize = vp->v_rdev->si_bsize_phys;
1750 	else
1751 		chunksize = DEV_BSIZE;
1752 	chunksize = maxiosize / chunksize * chunksize;
1753 #if SPEC_CHAIN_DEBUG & 1
1754 	devfs_debug(DEVFS_DEBUG_DEBUG,
1755 		    "spec_strategy chained I/O chunksize=%d\n",
1756 		    chunksize);
1757 #endif
1758 	nbp->b_cmd = bp->b_cmd;
1759 	nbp->b_bcount = chunksize;
1760 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1761 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1762 
1763 #if SPEC_CHAIN_DEBUG & 1
1764 	devfs_debug(DEVFS_DEBUG_DEBUG,
1765 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1766 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1767 #endif
1768 
1769 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1770 
1771 	if (DEVFS_NODE(vp)) {
1772 		nanotime(&DEVFS_NODE(vp)->atime);
1773 		nanotime(&DEVFS_NODE(vp)->mtime);
1774 	}
1775 
1776 	return (0);
1777 }
1778 
1779 /*
1780  * Chunked up transfer completion routine - chain transfers until done
1781  *
1782  * NOTE: MPSAFE callback.
1783  */
1784 static
1785 void
1786 devfs_spec_strategy_done(struct bio *nbio)
1787 {
1788 	struct buf *nbp = nbio->bio_buf;
1789 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1790 	struct buf *bp = bio->bio_buf;			/* original bp */
1791 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1792 	int boffset = nbp->b_data - bp->b_data;
1793 
1794 	if (nbp->b_flags & B_ERROR) {
1795 		/*
1796 		 * An error terminates the chain, propogate the error back
1797 		 * to the original bp
1798 		 */
1799 		bp->b_flags |= B_ERROR;
1800 		bp->b_error = nbp->b_error;
1801 		bp->b_resid = bp->b_bcount - boffset +
1802 			      (nbp->b_bcount - nbp->b_resid);
1803 #if SPEC_CHAIN_DEBUG & 1
1804 		devfs_debug(DEVFS_DEBUG_DEBUG,
1805 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1806 			    bp, bp->b_error, bp->b_bcount,
1807 			    bp->b_bcount - bp->b_resid);
1808 #endif
1809 	} else if (nbp->b_resid) {
1810 		/*
1811 		 * A short read or write terminates the chain
1812 		 */
1813 		bp->b_error = nbp->b_error;
1814 		bp->b_resid = bp->b_bcount - boffset +
1815 			      (nbp->b_bcount - nbp->b_resid);
1816 #if SPEC_CHAIN_DEBUG & 1
1817 		devfs_debug(DEVFS_DEBUG_DEBUG,
1818 			    "spec_strategy: chain %p short read(1) "
1819 			    "bcount %d/%d\n",
1820 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1821 #endif
1822 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1823 		/*
1824 		 * A short read or write can also occur by truncating b_bcount
1825 		 */
1826 #if SPEC_CHAIN_DEBUG & 1
1827 		devfs_debug(DEVFS_DEBUG_DEBUG,
1828 			    "spec_strategy: chain %p short read(2) "
1829 			    "bcount %d/%d\n",
1830 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1831 #endif
1832 		bp->b_error = 0;
1833 		bp->b_bcount = nbp->b_bcount + boffset;
1834 		bp->b_resid = nbp->b_resid;
1835 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1836 		/*
1837 		 * No more data terminates the chain
1838 		 */
1839 #if SPEC_CHAIN_DEBUG & 1
1840 		devfs_debug(DEVFS_DEBUG_DEBUG,
1841 			    "spec_strategy: chain %p finished bcount %d\n",
1842 			    bp, bp->b_bcount);
1843 #endif
1844 		bp->b_error = 0;
1845 		bp->b_resid = 0;
1846 	} else {
1847 		/*
1848 		 * Continue the chain
1849 		 */
1850 		boffset += nbp->b_bcount;
1851 		nbp->b_data = bp->b_data + boffset;
1852 		nbp->b_bcount = bp->b_bcount - boffset;
1853 		if (nbp->b_bcount > chunksize)
1854 			nbp->b_bcount = chunksize;
1855 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1856 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1857 
1858 #if SPEC_CHAIN_DEBUG & 1
1859 		devfs_debug(DEVFS_DEBUG_DEBUG,
1860 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1861 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
1862 #endif
1863 
1864 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1865 		return;
1866 	}
1867 
1868 	/*
1869 	 * Fall through to here on termination.  biodone(bp) and
1870 	 * clean up and free nbp.
1871 	 */
1872 	biodone(bio);
1873 	BUF_UNLOCK(nbp);
1874 	uninitbufbio(nbp);
1875 	kfree(nbp, M_DEVBUF);
1876 }
1877 
1878 /*
1879  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1880  */
1881 static int
1882 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1883 {
1884 	struct buf *bp;
1885 
1886 	/*
1887 	 * Must be a synchronous operation
1888 	 */
1889 	KKASSERT(ap->a_vp->v_rdev != NULL);
1890 	if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
1891 		return (0);
1892 	bp = geteblk(ap->a_length);
1893 	bp->b_cmd = BUF_CMD_FREEBLKS;
1894 	bp->b_bio1.bio_flags |= BIO_SYNC;
1895 	bp->b_bio1.bio_offset = ap->a_offset;
1896 	bp->b_bio1.bio_done = biodone_sync;
1897 	bp->b_bcount = ap->a_length;
1898 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1899 	biowait(&bp->b_bio1, "TRIM");
1900 	brelse(bp);
1901 
1902 	return (0);
1903 }
1904 
1905 /*
1906  * Implement degenerate case where the block requested is the block
1907  * returned, and assume that the entire device is contiguous in regards
1908  * to the contiguous block range (runp and runb).
1909  *
1910  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1911  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
1912  */
1913 static int
1914 devfs_spec_bmap(struct vop_bmap_args *ap)
1915 {
1916 	if (ap->a_doffsetp != NULL)
1917 		*ap->a_doffsetp = ap->a_loffset;
1918 	if (ap->a_runp != NULL)
1919 		*ap->a_runp = MAXBSIZE;
1920 	if (ap->a_runb != NULL) {
1921 		if (ap->a_loffset < MAXBSIZE)
1922 			*ap->a_runb = (int)ap->a_loffset;
1923 		else
1924 			*ap->a_runb = MAXBSIZE;
1925 	}
1926 	return (0);
1927 }
1928 
1929 
1930 /*
1931  * Special device advisory byte-level locks.
1932  *
1933  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1934  *		struct flock *a_fl, int a_flags)
1935  */
1936 /* ARGSUSED */
1937 static int
1938 devfs_spec_advlock(struct vop_advlock_args *ap)
1939 {
1940 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1941 }
1942 
1943 /*
1944  * NOTE: MPSAFE callback.
1945  */
1946 static void
1947 devfs_spec_getpages_iodone(struct bio *bio)
1948 {
1949 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
1950 	wakeup(bio->bio_buf);
1951 }
1952 
1953 /*
1954  * spec_getpages() - get pages associated with device vnode.
1955  *
1956  * Note that spec_read and spec_write do not use the buffer cache, so we
1957  * must fully implement getpages here.
1958  */
1959 static int
1960 devfs_spec_getpages(struct vop_getpages_args *ap)
1961 {
1962 	vm_offset_t kva;
1963 	int error;
1964 	int i, pcount, size;
1965 	struct buf *bp;
1966 	vm_page_t m;
1967 	vm_ooffset_t offset;
1968 	int toff, nextoff, nread;
1969 	struct vnode *vp = ap->a_vp;
1970 	int blksiz;
1971 	int gotreqpage;
1972 
1973 	error = 0;
1974 	pcount = round_page(ap->a_count) / PAGE_SIZE;
1975 
1976 	/*
1977 	 * Calculate the offset of the transfer and do sanity check.
1978 	 */
1979 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1980 
1981 	/*
1982 	 * Round up physical size for real devices.  We cannot round using
1983 	 * v_mount's block size data because v_mount has nothing to do with
1984 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
1985 	 * size for the device itself.
1986 	 *
1987 	 * We can't use v_rdev->si_mountpoint because it only exists when the
1988 	 * block device is mounted.  However, we can use v_rdev.
1989 	 */
1990 	if (vn_isdisk(vp, NULL))
1991 		blksiz = vp->v_rdev->si_bsize_phys;
1992 	else
1993 		blksiz = DEV_BSIZE;
1994 
1995 	size = roundup2(ap->a_count, blksiz);
1996 
1997 	bp = getpbuf_kva(NULL);
1998 	kva = (vm_offset_t)bp->b_data;
1999 
2000 	/*
2001 	 * Map the pages to be read into the kva.
2002 	 */
2003 	pmap_qenter(kva, ap->a_m, pcount);
2004 
2005 	/* Build a minimal buffer header. */
2006 	bp->b_cmd = BUF_CMD_READ;
2007 	bp->b_bcount = size;
2008 	bp->b_resid = 0;
2009 	bsetrunningbufspace(bp, size);
2010 
2011 	bp->b_bio1.bio_offset = offset;
2012 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2013 
2014 	mycpu->gd_cnt.v_vnodein++;
2015 	mycpu->gd_cnt.v_vnodepgsin += pcount;
2016 
2017 	/* Do the input. */
2018 	vn_strategy(ap->a_vp, &bp->b_bio1);
2019 
2020 	crit_enter();
2021 
2022 	/* We definitely need to be at splbio here. */
2023 	while (bp->b_cmd != BUF_CMD_DONE)
2024 		tsleep(bp, 0, "spread", 0);
2025 
2026 	crit_exit();
2027 
2028 	if (bp->b_flags & B_ERROR) {
2029 		if (bp->b_error)
2030 			error = bp->b_error;
2031 		else
2032 			error = EIO;
2033 	}
2034 
2035 	/*
2036 	 * If EOF is encountered we must zero-extend the result in order
2037 	 * to ensure that the page does not contain garabge.  When no
2038 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
2039 	 * b_resid is relative to b_bcount and should be 0, but some devices
2040 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
2041 	 */
2042 	nread = bp->b_bcount - bp->b_resid;
2043 	if (nread < ap->a_count)
2044 		bzero((caddr_t)kva + nread, ap->a_count - nread);
2045 	pmap_qremove(kva, pcount);
2046 
2047 	gotreqpage = 0;
2048 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2049 		nextoff = toff + PAGE_SIZE;
2050 		m = ap->a_m[i];
2051 
2052 		/*
2053 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2054 		 *	 pmap modified bit.  pmap modified bit should have
2055 		 *	 already been cleared.
2056 		 */
2057 		if (nextoff <= nread) {
2058 			m->valid = VM_PAGE_BITS_ALL;
2059 			vm_page_undirty(m);
2060 		} else if (toff < nread) {
2061 			/*
2062 			 * Since this is a VM request, we have to supply the
2063 			 * unaligned offset to allow vm_page_set_valid()
2064 			 * to zero sub-DEV_BSIZE'd portions of the page.
2065 			 */
2066 			vm_page_set_valid(m, 0, nread - toff);
2067 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2068 		} else {
2069 			m->valid = 0;
2070 			vm_page_undirty(m);
2071 		}
2072 
2073 		if (i != ap->a_reqpage) {
2074 			/*
2075 			 * Just in case someone was asking for this page we
2076 			 * now tell them that it is ok to use.
2077 			 */
2078 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2079 				if (m->valid) {
2080 					if (m->flags & PG_REFERENCED) {
2081 						vm_page_activate(m);
2082 					} else {
2083 						vm_page_deactivate(m);
2084 					}
2085 					vm_page_wakeup(m);
2086 				} else {
2087 					vm_page_free(m);
2088 				}
2089 			} else {
2090 				vm_page_free(m);
2091 			}
2092 		} else if (m->valid) {
2093 			gotreqpage = 1;
2094 			/*
2095 			 * Since this is a VM request, we need to make the
2096 			 * entire page presentable by zeroing invalid sections.
2097 			 */
2098 			if (m->valid != VM_PAGE_BITS_ALL)
2099 			    vm_page_zero_invalid(m, FALSE);
2100 		}
2101 	}
2102 	if (!gotreqpage) {
2103 		m = ap->a_m[ap->a_reqpage];
2104 		devfs_debug(DEVFS_DEBUG_WARNING,
2105 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2106 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2107 		devfs_debug(DEVFS_DEBUG_WARNING,
2108 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2109 		    size, bp->b_resid, ap->a_count, m->valid);
2110 		devfs_debug(DEVFS_DEBUG_WARNING,
2111 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2112 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2113 		/*
2114 		 * Free the buffer header back to the swap buffer pool.
2115 		 */
2116 		relpbuf(bp, NULL);
2117 		return VM_PAGER_ERROR;
2118 	}
2119 	/*
2120 	 * Free the buffer header back to the swap buffer pool.
2121 	 */
2122 	relpbuf(bp, NULL);
2123 	if (DEVFS_NODE(ap->a_vp))
2124 		nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2125 	return VM_PAGER_OK;
2126 }
2127 
2128 static __inline
2129 int
2130 sequential_heuristic(struct uio *uio, struct file *fp)
2131 {
2132 	/*
2133 	 * Sequential heuristic - detect sequential operation
2134 	 */
2135 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2136 	    uio->uio_offset == fp->f_nextoff) {
2137 		/*
2138 		 * XXX we assume that the filesystem block size is
2139 		 * the default.  Not true, but still gives us a pretty
2140 		 * good indicator of how sequential the read operations
2141 		 * are.
2142 		 */
2143 		int tmpseq = fp->f_seqcount;
2144 
2145 		tmpseq += (uio->uio_resid + MAXBSIZE - 1) / MAXBSIZE;
2146 		if (tmpseq > IO_SEQMAX)
2147 			tmpseq = IO_SEQMAX;
2148 		fp->f_seqcount = tmpseq;
2149 		return(fp->f_seqcount << IO_SEQSHIFT);
2150 	}
2151 
2152 	/*
2153 	 * Not sequential, quick draw-down of seqcount
2154 	 */
2155 	if (fp->f_seqcount > 1)
2156 		fp->f_seqcount = 1;
2157 	else
2158 		fp->f_seqcount = 0;
2159 	return(0);
2160 }
2161