xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision 71c97a3cee892aedd0f1dc81caba4c8fbdf7fc00)
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 <vm/vm_page2.h>
70 
71 #ifndef SPEC_CHAIN_DEBUG
72 #define SPEC_CHAIN_DEBUG 0
73 #endif
74 
75 MALLOC_DECLARE(M_DEVFS);
76 #define DEVFS_BADOP	(void *)devfs_vop_badop
77 
78 static int devfs_vop_badop(struct vop_generic_args *);
79 static int devfs_vop_access(struct vop_access_args *);
80 static int devfs_vop_inactive(struct vop_inactive_args *);
81 static int devfs_vop_reclaim(struct vop_reclaim_args *);
82 static int devfs_vop_readdir(struct vop_readdir_args *);
83 static int devfs_vop_getattr(struct vop_getattr_args *);
84 static int devfs_vop_setattr(struct vop_setattr_args *);
85 static int devfs_vop_readlink(struct vop_readlink_args *);
86 static int devfs_vop_print(struct vop_print_args *);
87 
88 static int devfs_vop_nresolve(struct vop_nresolve_args *);
89 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
90 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
91 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
92 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
93 static int devfs_vop_nremove(struct vop_nremove_args *);
94 
95 static int devfs_spec_open(struct vop_open_args *);
96 static int devfs_spec_close(struct vop_close_args *);
97 static int devfs_spec_fsync(struct vop_fsync_args *);
98 
99 static int devfs_spec_read(struct vop_read_args *);
100 static int devfs_spec_write(struct vop_write_args *);
101 static int devfs_spec_ioctl(struct vop_ioctl_args *);
102 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
103 static int devfs_spec_strategy(struct vop_strategy_args *);
104 static void devfs_spec_strategy_done(struct bio *);
105 static int devfs_spec_freeblks(struct vop_freeblks_args *);
106 static int devfs_spec_bmap(struct vop_bmap_args *);
107 static int devfs_spec_advlock(struct vop_advlock_args *);
108 static void devfs_spec_getpages_iodone(struct bio *);
109 static int devfs_spec_getpages(struct vop_getpages_args *);
110 
111 static int devfs_fo_close(struct file *);
112 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
113 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
114 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
115 static int devfs_fo_kqfilter(struct file *, struct knote *);
116 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
117 				struct ucred *, struct sysmsg *);
118 static __inline int sequential_heuristic(struct uio *, struct file *);
119 
120 extern struct lock devfs_lock;
121 
122 /*
123  * devfs vnode operations for regular files.  All vnode ops are MPSAFE.
124  */
125 struct vop_ops devfs_vnode_norm_vops = {
126 	.vop_default =		vop_defaultop,
127 	.vop_access =		devfs_vop_access,
128 	.vop_advlock =		DEVFS_BADOP,
129 	.vop_bmap =		DEVFS_BADOP,
130 	.vop_close =		vop_stdclose,
131 	.vop_getattr =		devfs_vop_getattr,
132 	.vop_inactive =		devfs_vop_inactive,
133 	.vop_ncreate =		DEVFS_BADOP,
134 	.vop_nresolve =		devfs_vop_nresolve,
135 	.vop_nlookupdotdot =	devfs_vop_nlookupdotdot,
136 	.vop_nlink =		DEVFS_BADOP,
137 	.vop_nmkdir =		devfs_vop_nmkdir,
138 	.vop_nmknod =		DEVFS_BADOP,
139 	.vop_nremove =		devfs_vop_nremove,
140 	.vop_nrename =		DEVFS_BADOP,
141 	.vop_nrmdir =		devfs_vop_nrmdir,
142 	.vop_nsymlink =		devfs_vop_nsymlink,
143 	.vop_open =		vop_stdopen,
144 	.vop_pathconf =		vop_stdpathconf,
145 	.vop_print =		devfs_vop_print,
146 	.vop_read =		DEVFS_BADOP,
147 	.vop_readdir =		devfs_vop_readdir,
148 	.vop_readlink =		devfs_vop_readlink,
149 	.vop_reallocblks =	DEVFS_BADOP,
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_reallocblks =	DEVFS_BADOP,
178 	.vop_reclaim =		devfs_vop_reclaim,
179 	.vop_setattr =		devfs_vop_setattr,
180 	.vop_strategy =		devfs_spec_strategy,
181 	.vop_write =		devfs_spec_write,
182 	.vop_ioctl =		devfs_spec_ioctl
183 };
184 
185 /*
186  * devfs file pointer operations.  All fileops are MPSAFE.
187  */
188 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
189 
190 struct fileops devfs_dev_fileops = {
191 	.fo_read	= devfs_fo_read,
192 	.fo_write	= devfs_fo_write,
193 	.fo_ioctl	= devfs_fo_ioctl,
194 	.fo_kqfilter	= devfs_fo_kqfilter,
195 	.fo_stat	= devfs_fo_stat,
196 	.fo_close	= devfs_fo_close,
197 	.fo_shutdown	= nofo_shutdown
198 };
199 
200 /*
201  * These two functions are possibly temporary hacks for devices (aka
202  * the pty code) which want to control the node attributes themselves.
203  *
204  * XXX we may ultimately desire to simply remove the uid/gid/mode
205  * from the node entirely.
206  *
207  * MPSAFE - sorta.  Theoretically the overwrite can compete since they
208  *	    are loading from the same fields.
209  */
210 static __inline void
211 node_sync_dev_get(struct devfs_node *node)
212 {
213 	cdev_t dev;
214 
215 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
216 		node->uid = dev->si_uid;
217 		node->gid = dev->si_gid;
218 		node->mode = dev->si_perms;
219 	}
220 }
221 
222 static __inline void
223 node_sync_dev_set(struct devfs_node *node)
224 {
225 	cdev_t dev;
226 
227 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
228 		dev->si_uid = node->uid;
229 		dev->si_gid = node->gid;
230 		dev->si_perms = node->mode;
231 	}
232 }
233 
234 /*
235  * generic entry point for unsupported operations
236  */
237 static int
238 devfs_vop_badop(struct vop_generic_args *ap)
239 {
240 	return (EIO);
241 }
242 
243 
244 static int
245 devfs_vop_access(struct vop_access_args *ap)
246 {
247 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
248 	int error;
249 
250 	if (!devfs_node_is_accessible(node))
251 		return ENOENT;
252 	node_sync_dev_get(node);
253 	error = vop_helper_access(ap, node->uid, node->gid,
254 				  node->mode, node->flags);
255 
256 	return error;
257 }
258 
259 
260 static int
261 devfs_vop_inactive(struct vop_inactive_args *ap)
262 {
263 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
264 
265 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
266 		vrecycle(ap->a_vp);
267 	return 0;
268 }
269 
270 
271 static int
272 devfs_vop_reclaim(struct vop_reclaim_args *ap)
273 {
274 	struct devfs_node *node;
275 	struct vnode *vp;
276 	int locked;
277 
278 	/*
279 	 * Check if it is locked already. if not, we acquire the devfs lock
280 	 */
281 	if ((lockstatus(&devfs_lock, curthread)) != LK_EXCLUSIVE) {
282 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
283 		locked = 1;
284 	} else {
285 		locked = 0;
286 	}
287 
288 	/*
289 	 * Get rid of the devfs_node if it is no longer linked into the
290 	 * topology.  Interlocked by devfs_lock.  However, be careful
291 	 * interposing other operations between cleaning out v_data and
292 	 * devfs_freep() as the node is only protected by devfs_lock
293 	 * once the vnode is disassociated.
294 	 */
295 	vp = ap->a_vp;
296 	node = DEVFS_NODE(vp);
297 
298 	if (node) {
299 		if (node->v_node != vp) {
300 			kprintf("NODE->V_NODE MISMATCH VP=%p NODEVP=%p\n",
301 				vp, node->v_node);
302 		}
303 		vp->v_data = NULL;
304 		node->v_node = NULL;
305 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
306 			devfs_freep(node);
307 	}
308 	v_release_rdev(vp);
309 
310 	if (locked)
311 		lockmgr(&devfs_lock, LK_RELEASE);
312 
313 	/*
314 	 * v_rdev needs to be properly released using v_release_rdev
315 	 * Make sure v_data is NULL as well.
316 	 */
317 	return 0;
318 }
319 
320 
321 static int
322 devfs_vop_readdir(struct vop_readdir_args *ap)
323 {
324 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
325 	struct devfs_node *node;
326 	int cookie_index;
327 	int ncookies;
328 	int error2;
329 	int error;
330 	int r;
331 	off_t *cookies;
332 	off_t saveoff;
333 
334 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
335 
336 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
337 		return (EINVAL);
338 	error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
339 	if (error)
340 		return (error);
341 
342 	if (!devfs_node_is_accessible(dnode)) {
343 		vn_unlock(ap->a_vp);
344 		return ENOENT;
345 	}
346 
347 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
348 
349 	saveoff = ap->a_uio->uio_offset;
350 
351 	if (ap->a_ncookies) {
352 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
353 		if (ncookies > 256)
354 			ncookies = 256;
355 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
356 		cookie_index = 0;
357 	} else {
358 		ncookies = -1;
359 		cookies = NULL;
360 		cookie_index = 0;
361 	}
362 
363 	vfs_timestamp(&dnode->atime);
364 
365 	if (saveoff == 0) {
366 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
367 				     DT_DIR, 1, ".");
368 		if (r)
369 			goto done;
370 		if (cookies)
371 			cookies[cookie_index] = saveoff;
372 		saveoff++;
373 		cookie_index++;
374 		if (cookie_index == ncookies)
375 			goto done;
376 	}
377 
378 	if (saveoff == 1) {
379 		if (dnode->parent) {
380 			r = vop_write_dirent(&error, ap->a_uio,
381 					     dnode->parent->d_dir.d_ino,
382 					     DT_DIR, 2, "..");
383 		} else {
384 			r = vop_write_dirent(&error, ap->a_uio,
385 					     dnode->d_dir.d_ino,
386 					     DT_DIR, 2, "..");
387 		}
388 		if (r)
389 			goto done;
390 		if (cookies)
391 			cookies[cookie_index] = saveoff;
392 		saveoff++;
393 		cookie_index++;
394 		if (cookie_index == ncookies)
395 			goto done;
396 	}
397 
398 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
399 		if ((node->flags & DEVFS_HIDDEN) ||
400 		    (node->flags & DEVFS_INVISIBLE)) {
401 			continue;
402 		}
403 
404 		/*
405 		 * If the node type is a valid devfs alias, then we make
406 		 * sure that the target isn't hidden. If it is, we don't
407 		 * show the link in the directory listing.
408 		 */
409 		if ((node->node_type == Nlink) && (node->link_target != NULL) &&
410 			(node->link_target->flags & DEVFS_HIDDEN))
411 			continue;
412 
413 		if (node->cookie < saveoff)
414 			continue;
415 
416 		saveoff = node->cookie;
417 
418 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
419 					  node->d_dir.d_type,
420 					  node->d_dir.d_namlen,
421 					  node->d_dir.d_name);
422 
423 		if (error2)
424 			break;
425 
426 		saveoff++;
427 
428 		if (cookies)
429 			cookies[cookie_index] = node->cookie;
430 		++cookie_index;
431 		if (cookie_index == ncookies)
432 			break;
433 	}
434 
435 done:
436 	lockmgr(&devfs_lock, LK_RELEASE);
437 	vn_unlock(ap->a_vp);
438 
439 	ap->a_uio->uio_offset = saveoff;
440 	if (error && cookie_index == 0) {
441 		if (cookies) {
442 			kfree(cookies, M_TEMP);
443 			*ap->a_ncookies = 0;
444 			*ap->a_cookies = NULL;
445 		}
446 	} else {
447 		if (cookies) {
448 			*ap->a_ncookies = cookie_index;
449 			*ap->a_cookies = cookies;
450 		}
451 	}
452 	return (error);
453 }
454 
455 
456 static int
457 devfs_vop_nresolve(struct vop_nresolve_args *ap)
458 {
459 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
460 	struct devfs_node *node, *found = NULL;
461 	struct namecache *ncp;
462 	struct vnode *vp = NULL;
463 	int error = 0;
464 	int len;
465 	int depth;
466 
467 	ncp = ap->a_nch->ncp;
468 	len = ncp->nc_nlen;
469 
470 	if (!devfs_node_is_accessible(dnode))
471 		return ENOENT;
472 
473 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
474 
475 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) {
476 		error = ENOENT;
477 		cache_setvp(ap->a_nch, NULL);
478 		goto out;
479 	}
480 
481 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
482 		if (len == node->d_dir.d_namlen) {
483 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
484 				found = node;
485 				break;
486 			}
487 		}
488 	}
489 
490 	if (found) {
491 		depth = 0;
492 		while ((found->node_type == Nlink) && (found->link_target)) {
493 			if (depth >= 8) {
494 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
495 				break;
496 			}
497 
498 			found = found->link_target;
499 			++depth;
500 		}
501 
502 		if (!(found->flags & DEVFS_HIDDEN))
503 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
504 	}
505 
506 	if (vp == NULL) {
507 		error = ENOENT;
508 		cache_setvp(ap->a_nch, NULL);
509 		goto out;
510 
511 	}
512 	KKASSERT(vp);
513 	vn_unlock(vp);
514 	cache_setvp(ap->a_nch, vp);
515 	vrele(vp);
516 out:
517 	lockmgr(&devfs_lock, LK_RELEASE);
518 
519 	return error;
520 }
521 
522 
523 static int
524 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
525 {
526 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
527 
528 	*ap->a_vpp = NULL;
529 	if (!devfs_node_is_accessible(dnode))
530 		return ENOENT;
531 
532 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
533 	if (dnode->parent != NULL) {
534 		devfs_allocv(ap->a_vpp, dnode->parent);
535 		vn_unlock(*ap->a_vpp);
536 	}
537 	lockmgr(&devfs_lock, LK_RELEASE);
538 
539 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
540 }
541 
542 
543 /*
544  * getattr() - Does not need a lock since the vp is refd
545  */
546 static int
547 devfs_vop_getattr(struct vop_getattr_args *ap)
548 {
549 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
550 	struct vattr *vap = ap->a_vap;
551 	struct partinfo pinfo;
552 	int error = 0;
553 
554 #if 0
555 	if (!devfs_node_is_accessible(node))
556 		return ENOENT;
557 #endif
558 
559 	/*
560 	 * XXX This is a temporary hack to prevent crashes when the device is
561 	 * being destroyed (and so the underlying node will be gone) while
562 	 * a userland program is blocked in a read().
563 	 */
564 	if (node == NULL)
565 		return EIO;
566 
567 	node_sync_dev_get(node);
568 
569 	/* start by zeroing out the attributes */
570 	VATTR_NULL(vap);
571 
572 	/* next do all the common fields */
573 	vap->va_type = ap->a_vp->v_type;
574 	vap->va_mode = node->mode;
575 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
576 	vap->va_flags = 0;
577 	vap->va_blocksize = DEV_BSIZE;
578 	vap->va_bytes = vap->va_size = 0;
579 
580 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
581 
582 	vap->va_atime = node->atime;
583 	vap->va_mtime = node->mtime;
584 	vap->va_ctime = node->ctime;
585 
586 	vap->va_nlink = 1; /* number of references to file */
587 
588 	vap->va_uid = node->uid;
589 	vap->va_gid = node->gid;
590 
591 	vap->va_rmajor = 0;
592 	vap->va_rminor = 0;
593 
594 	if ((node->node_type == Ndev) && node->d_dev)  {
595 		reference_dev(node->d_dev);
596 		vap->va_rminor = node->d_dev->si_uminor;
597 		release_dev(node->d_dev);
598 	}
599 
600 	/* For a softlink the va_size is the length of the softlink */
601 	if (node->symlink_name != 0) {
602 		vap->va_bytes = vap->va_size = node->symlink_namelen;
603 	}
604 
605 	/*
606 	 * For a disk-type device, va_size is the size of the underlying
607 	 * device, so that lseek() works properly.
608 	 */
609 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
610 		bzero(&pinfo, sizeof(pinfo));
611 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
612 				   0, proc0.p_ucred, NULL, NULL);
613 		if ((error == 0) && (pinfo.media_blksize != 0)) {
614 			vap->va_size = pinfo.media_size;
615 		} else {
616 			vap->va_size = 0;
617 			error = 0;
618 		}
619 	}
620 
621 	return (error);
622 }
623 
624 static int
625 devfs_vop_setattr(struct vop_setattr_args *ap)
626 {
627 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
628 	struct vattr *vap;
629 	uid_t cur_uid;
630 	gid_t cur_gid;
631 	mode_t cur_mode;
632 	int error = 0;
633 
634 	if (!devfs_node_is_accessible(node))
635 		return ENOENT;
636 	node_sync_dev_get(node);
637 
638 	vap = ap->a_vap;
639 
640 	if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
641 		cur_uid = node->uid;
642 		cur_gid = node->gid;
643 		cur_mode = node->mode;
644 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
645 		    ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
646 		if (error)
647 			goto out;
648 
649 		if (node->uid != cur_uid || node->gid != cur_gid) {
650 			node->uid = cur_uid;
651 			node->gid = cur_gid;
652 			node->mode = cur_mode;
653 		}
654 	}
655 
656 	if (vap->va_mode != (mode_t)VNOVAL) {
657 		cur_mode = node->mode;
658 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
659 		    node->uid, node->gid, &cur_mode);
660 		if (error == 0 && node->mode != cur_mode) {
661 			node->mode = cur_mode;
662 		}
663 	}
664 
665 out:
666 	node_sync_dev_set(node);
667 	vfs_timestamp(&node->ctime);
668 
669 	return error;
670 }
671 
672 
673 static int
674 devfs_vop_readlink(struct vop_readlink_args *ap)
675 {
676 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
677 	int ret;
678 
679 	if (!devfs_node_is_accessible(node))
680 		return ENOENT;
681 
682 	lockmgr(&devfs_lock, LK_SHARED);
683 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
684 	lockmgr(&devfs_lock, LK_RELEASE);
685 
686 	return ret;
687 }
688 
689 
690 static int
691 devfs_vop_print(struct vop_print_args *ap)
692 {
693 	return (0);
694 }
695 
696 static int
697 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
698 {
699 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
700 	struct devfs_node *node;
701 
702 	if (!devfs_node_is_accessible(dnode))
703 		return ENOENT;
704 
705 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
706 		goto out;
707 
708 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
709 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir,
710 		      ap->a_nch->ncp->nc_name, dnode, NULL);
711 
712 	if (*ap->a_vpp) {
713 		node = DEVFS_NODE(*ap->a_vpp);
714 		node->flags |= DEVFS_USER_CREATED;
715 		cache_setunresolved(ap->a_nch);
716 		cache_setvp(ap->a_nch, *ap->a_vpp);
717 	}
718 	lockmgr(&devfs_lock, LK_RELEASE);
719 out:
720 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
721 }
722 
723 static int
724 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
725 {
726 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
727 	struct devfs_node *node;
728 	size_t targetlen;
729 
730 	if (!devfs_node_is_accessible(dnode))
731 		return ENOENT;
732 
733 	ap->a_vap->va_type = VLNK;
734 
735 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
736 		goto out;
737 
738 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
739 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink,
740 		      ap->a_nch->ncp->nc_name, dnode, NULL);
741 
742 	targetlen = strlen(ap->a_target);
743 	if (*ap->a_vpp) {
744 		node = DEVFS_NODE(*ap->a_vpp);
745 		node->flags |= DEVFS_USER_CREATED;
746 		node->symlink_namelen = targetlen;
747 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
748 		memcpy(node->symlink_name, ap->a_target, targetlen);
749 		node->symlink_name[targetlen] = '\0';
750 		cache_setunresolved(ap->a_nch);
751 		cache_setvp(ap->a_nch, *ap->a_vpp);
752 	}
753 	lockmgr(&devfs_lock, LK_RELEASE);
754 out:
755 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
756 }
757 
758 static int
759 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
760 {
761 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
762 	struct devfs_node *node;
763 	struct namecache *ncp;
764 	int error = ENOENT;
765 
766 	ncp = ap->a_nch->ncp;
767 
768 	if (!devfs_node_is_accessible(dnode))
769 		return ENOENT;
770 
771 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
772 
773 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
774 		goto out;
775 
776 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
777 		if (ncp->nc_nlen != node->d_dir.d_namlen)
778 			continue;
779 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
780 			continue;
781 
782 		/*
783 		 * only allow removal of user created dirs
784 		 */
785 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
786 			error = EPERM;
787 			goto out;
788 		} else if (node->node_type != Ndir) {
789 			error = ENOTDIR;
790 			goto out;
791 		} else if (node->nchildren > 2) {
792 			error = ENOTEMPTY;
793 			goto out;
794 		} else {
795 			if (node->v_node)
796 				cache_inval_vp(node->v_node, CINV_DESTROY);
797 			devfs_unlinkp(node);
798 			error = 0;
799 			break;
800 		}
801 	}
802 
803 	cache_unlink(ap->a_nch);
804 out:
805 	lockmgr(&devfs_lock, LK_RELEASE);
806 	return error;
807 }
808 
809 static int
810 devfs_vop_nremove(struct vop_nremove_args *ap)
811 {
812 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
813 	struct devfs_node *node;
814 	struct namecache *ncp;
815 	int error = ENOENT;
816 
817 	ncp = ap->a_nch->ncp;
818 
819 	if (!devfs_node_is_accessible(dnode))
820 		return ENOENT;
821 
822 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
823 
824 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
825 		goto out;
826 
827 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
828 		if (ncp->nc_nlen != node->d_dir.d_namlen)
829 			continue;
830 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
831 			continue;
832 
833 		/*
834 		 * only allow removal of user created stuff (e.g. symlinks)
835 		 */
836 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
837 			error = EPERM;
838 			goto out;
839 		} else if (node->node_type == Ndir) {
840 			error = EISDIR;
841 			goto out;
842 		} else {
843 			if (node->v_node)
844 				cache_inval_vp(node->v_node, CINV_DESTROY);
845 			devfs_unlinkp(node);
846 			error = 0;
847 			break;
848 		}
849 	}
850 
851 	cache_unlink(ap->a_nch);
852 out:
853 	lockmgr(&devfs_lock, LK_RELEASE);
854 	return error;
855 }
856 
857 
858 static int
859 devfs_spec_open(struct vop_open_args *ap)
860 {
861 	struct vnode *vp = ap->a_vp;
862 	struct vnode *orig_vp = NULL;
863 	struct devfs_node *node = DEVFS_NODE(vp);
864 	struct devfs_node *newnode;
865 	cdev_t dev, ndev = NULL;
866 	int error = 0;
867 
868 	if (node) {
869 		if (node->d_dev == NULL)
870 			return ENXIO;
871 		if (!devfs_node_is_accessible(node))
872 			return ENOENT;
873 	}
874 
875 	if ((dev = vp->v_rdev) == NULL)
876 		return ENXIO;
877 
878 	/*
879 	 * Simple devices that don't care.  Retain the shared lock.
880 	 */
881 	if (dev_dflags(dev) & D_QUICK) {
882 		vn_unlock(vp);
883 		error = dev_dopen(dev, ap->a_mode, S_IFCHR,
884 				  ap->a_cred, ap->a_fp, vp);
885 		vn_lock(vp, LK_SHARED | LK_RETRY);
886 		vop_stdopen(ap);
887 		goto skip;
888 	}
889 
890 	/*
891 	 * Slow code
892 	 */
893 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
894 	if (node && ap->a_fp) {
895 		int exists;
896 
897 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
898 		lockmgr(&devfs_lock, LK_SHARED);
899 
900 		ndev = devfs_clone(dev, node->d_dir.d_name,
901 				   node->d_dir.d_namlen,
902 				   ap->a_mode, ap->a_cred);
903 		if (ndev != NULL) {
904 			lockmgr(&devfs_lock, LK_RELEASE);
905 			lockmgr(&devfs_lock, LK_EXCLUSIVE);
906 			newnode = devfs_create_device_node(
907 					DEVFS_MNTDATA(vp->v_mount)->root_node,
908 					ndev, &exists, NULL, NULL);
909 			/* XXX: possibly destroy device if this happens */
910 
911 			if (newnode != NULL) {
912 				dev = ndev;
913 				if (exists == 0)
914 					devfs_link_dev(dev);
915 
916 				devfs_debug(DEVFS_DEBUG_DEBUG,
917 						"parent here is: %s, node is: |%s|\n",
918 						((node->parent->node_type == Nroot) ?
919 						"ROOT!" : node->parent->d_dir.d_name),
920 						newnode->d_dir.d_name);
921 				devfs_debug(DEVFS_DEBUG_DEBUG,
922 						"test: %s\n",
923 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
924 
925 				/*
926 				 * orig_vp is set to the original vp if we
927 				 * cloned.
928 				 */
929 				/* node->flags |= DEVFS_CLONED; */
930 				devfs_allocv(&vp, newnode);
931 				orig_vp = ap->a_vp;
932 				ap->a_vp = vp;
933 			}
934 		}
935 		lockmgr(&devfs_lock, LK_RELEASE);
936 
937 		/*
938 		 * Synchronize devfs here to make sure that, if the cloned
939 		 * device creates other device nodes in addition to the
940 		 * cloned one, all of them are created by the time we return
941 		 * from opening the cloned one.
942 		 */
943 		if (ndev)
944 			devfs_config();
945 	}
946 
947 	devfs_debug(DEVFS_DEBUG_DEBUG,
948 		    "devfs_spec_open() called on %s! \n",
949 		    dev->si_name);
950 
951 	/*
952 	 * Make this field valid before any I/O in ->d_open
953 	 *
954 	 * NOTE: Shared vnode lock probably held, but its ok as long
955 	 *	 as assignments are consistent.
956 	 */
957 	if (!dev->si_iosize_max)
958 		/* XXX: old DFLTPHYS == 64KB dependency */
959 		dev->si_iosize_max = min(MAXPHYS,64*1024);
960 
961 	if (dev_dflags(dev) & D_TTY)
962 		vsetflags(vp, VISTTY);
963 
964 	/*
965 	 * Open the underlying device
966 	 */
967 	vn_unlock(vp);
968 	error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred, ap->a_fp, vp);
969 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
970 
971 	/*
972 	 * Clean up any cloned vp if we error out.
973 	 */
974 	if (error) {
975 		if (orig_vp) {
976 			vput(vp);
977 			ap->a_vp = orig_vp;
978 			/* orig_vp = NULL; */
979 		}
980 		return error;
981 	}
982 
983 	/*
984 	 * This checks if the disk device is going to be opened for writing.
985 	 * It will be only allowed in the cases where securelevel permits it
986 	 * and it's not mounted R/W.
987 	 */
988 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
989 	    (ap->a_cred != FSCRED)) {
990 
991 		/* Very secure mode. No open for writing allowed */
992 		if (securelevel >= 2)
993 			return EPERM;
994 
995 		/*
996 		 * If it is mounted R/W, do not allow to open for writing.
997 		 * In the case it's mounted read-only but securelevel
998 		 * is >= 1, then do not allow opening for writing either.
999 		 */
1000 		if (vfs_mountedon(vp)) {
1001 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
1002 				return EBUSY;
1003 			else if (securelevel >= 1)
1004 				return EPERM;
1005 		}
1006 	}
1007 
1008 	/*
1009 	 * NOTE: vnode is still locked shared.  t_stop assignment should
1010 	 *	 remain consistent so we should be ok.
1011 	 */
1012 	if (dev_dflags(dev) & D_TTY) {
1013 		if (dev->si_tty) {
1014 			struct tty *tp;
1015 			tp = dev->si_tty;
1016 			if (!tp->t_stop) {
1017 				devfs_debug(DEVFS_DEBUG_DEBUG,
1018 					    "devfs: no t_stop\n");
1019 				tp->t_stop = nottystop;
1020 			}
1021 		}
1022 	}
1023 
1024 	/*
1025 	 * NOTE: vnode is still locked shared.  assignments should
1026 	 *	 remain consistent so we should be ok.  However,
1027 	 *	 upgrade to exclusive if we need a VM object.
1028 	 */
1029 	if (vn_isdisk(vp, NULL)) {
1030 		if (!dev->si_bsize_phys)
1031 			dev->si_bsize_phys = DEV_BSIZE;
1032 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
1033 	}
1034 
1035 	vop_stdopen(ap);
1036 #if 0
1037 	if (node)
1038 		vfs_timestamp(&node->atime);
1039 #endif
1040 	/*
1041 	 * If we replaced the vp the vop_stdopen() call will have loaded
1042 	 * it into fp->f_data and vref()d the vp, giving us two refs.  So
1043 	 * instead of just unlocking it here we have to vput() it.
1044 	 */
1045 	if (orig_vp)
1046 		vput(vp);
1047 
1048 	/* Ugly pty magic, to make pty devices appear once they are opened */
1049 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) {
1050 		if (node->flags & DEVFS_INVISIBLE)
1051 			node->flags &= ~DEVFS_INVISIBLE;
1052 	}
1053 
1054 skip:
1055 	if (ap->a_fp) {
1056 		KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
1057 		KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1058 		ap->a_fp->f_ops = &devfs_dev_fileops;
1059 		KKASSERT(ap->a_fp->f_data == (void *)vp);
1060 	}
1061 
1062 	return 0;
1063 }
1064 
1065 static int
1066 devfs_spec_close(struct vop_close_args *ap)
1067 {
1068 	struct devfs_node *node;
1069 	struct proc *p = curproc;
1070 	struct vnode *vp = ap->a_vp;
1071 	cdev_t dev = vp->v_rdev;
1072 	int error = 0;
1073 	int needrelock;
1074 	int opencount;
1075 
1076 	/*
1077 	 * Devices flagged D_QUICK require no special handling.
1078 	 */
1079 	if (dev && dev_dflags(dev) & D_QUICK) {
1080 		opencount = vp->v_opencount;
1081 		if (opencount <= 1)
1082 			opencount = count_dev(dev);   /* XXX NOT SMP SAFE */
1083 		if (((vp->v_flag & VRECLAIMED) ||
1084 		    (dev_dflags(dev) & D_TRACKCLOSE) ||
1085 		    (opencount == 1))) {
1086 			vn_unlock(vp);
1087 			error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1088 			vn_lock(vp, LK_SHARED | LK_RETRY);
1089 		}
1090 		goto skip;
1091 	}
1092 
1093 	/*
1094 	 * We do special tests on the opencount so unfortunately we need
1095 	 * an exclusive lock.
1096 	 */
1097 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
1098 
1099 	if (dev)
1100 		devfs_debug(DEVFS_DEBUG_DEBUG,
1101 			    "devfs_spec_close() called on %s! \n",
1102 			    dev->si_name);
1103 	else
1104 		devfs_debug(DEVFS_DEBUG_DEBUG,
1105 			    "devfs_spec_close() called, null vode!\n");
1106 
1107 	/*
1108 	 * A couple of hacks for devices and tty devices.  The
1109 	 * vnode ref count cannot be used to figure out the
1110 	 * last close, but we can use v_opencount now that
1111 	 * revoke works properly.
1112 	 *
1113 	 * Detect the last close on a controlling terminal and clear
1114 	 * the session (half-close).
1115 	 *
1116 	 * XXX opencount is not SMP safe.  The vnode is locked but there
1117 	 *     may be multiple vnodes referencing the same device.
1118 	 */
1119 	if (dev) {
1120 		/*
1121 		 * NOTE: Try to avoid global tokens when testing opencount
1122 		 * XXX hack, fixme. needs a struct lock and opencount in
1123 		 * struct cdev itself.
1124 		 */
1125 		reference_dev(dev);
1126 		opencount = vp->v_opencount;
1127 		if (opencount <= 1)
1128 			opencount = count_dev(dev);   /* XXX NOT SMP SAFE */
1129 	} else {
1130 		opencount = 0;
1131 	}
1132 
1133 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1134 		p->p_session->s_ttyvp = NULL;
1135 		vrele(vp);
1136 	}
1137 
1138 	/*
1139 	 * Vnodes can be opened and closed multiple times.  Do not really
1140 	 * close the device unless (1) it is being closed forcibly,
1141 	 * (2) the device wants to track closes, or (3) this is the last
1142 	 * vnode doing its last close on the device.
1143 	 *
1144 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1145 	 * a closed device.  This might not occur now that our revoke is
1146 	 * fixed.
1147 	 */
1148 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1149 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1150 		    (dev_dflags(dev) & D_TRACKCLOSE) ||
1151 		    (opencount == 1))) {
1152 		/*
1153 		 * Ugly pty magic, to make pty devices disappear again once
1154 		 * they are closed.
1155 		 */
1156 		node = DEVFS_NODE(ap->a_vp);
1157 		if (node && (node->flags & DEVFS_PTY))
1158 			node->flags |= DEVFS_INVISIBLE;
1159 
1160 		/*
1161 		 * Unlock around dev_dclose(), unless the vnode is
1162 		 * undergoing a vgone/reclaim (during umount).
1163 		 */
1164 		needrelock = 0;
1165 		if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1166 			needrelock = 1;
1167 			vn_unlock(vp);
1168 		}
1169 
1170 		/*
1171 		 * WARNING!  If the device destroys itself the devfs node
1172 		 *	     can disappear here.
1173 		 *
1174 		 * WARNING!  vn_lock() will fail if the vp is in a VRECLAIM,
1175 		 *	     which can occur during umount.
1176 		 */
1177 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1178 		/* node is now stale */
1179 
1180 		if (needrelock) {
1181 			if (vn_lock(vp, LK_EXCLUSIVE |
1182 					LK_RETRY |
1183 					LK_FAILRECLAIM) != 0) {
1184 				panic("devfs_spec_close: vnode %p "
1185 				      "unexpectedly could not be relocked",
1186 				      vp);
1187 			}
1188 		}
1189 	} else {
1190 		error = 0;
1191 	}
1192 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1193 
1194 	/*
1195 	 * Track the actual opens and closes on the vnode.  The last close
1196 	 * disassociates the rdev.  If the rdev is already disassociated or
1197 	 * the opencount is already 0, the vnode might have been revoked
1198 	 * and no further opencount tracking occurs.
1199 	 */
1200 	if (dev)
1201 		release_dev(dev);
1202 skip:
1203 	if (vp->v_opencount > 0)
1204 		vop_stdclose(ap);
1205 	return(error);
1206 
1207 }
1208 
1209 
1210 static int
1211 devfs_fo_close(struct file *fp)
1212 {
1213 	struct vnode *vp = (struct vnode *)fp->f_data;
1214 	int error;
1215 
1216 	fp->f_ops = &badfileops;
1217 	error = vn_close(vp, fp->f_flag, fp);
1218 	devfs_clear_cdevpriv(fp);
1219 
1220 	return (error);
1221 }
1222 
1223 
1224 /*
1225  * Device-optimized file table vnode read routine.
1226  *
1227  * This bypasses the VOP table and talks directly to the device.  Most
1228  * filesystems just route to specfs and can make this optimization.
1229  */
1230 static int
1231 devfs_fo_read(struct file *fp, struct uio *uio,
1232 		 struct ucred *cred, int flags)
1233 {
1234 	struct devfs_node *node;
1235 	struct vnode *vp;
1236 	int ioflag;
1237 	int error;
1238 	cdev_t dev;
1239 
1240 	KASSERT(uio->uio_td == curthread,
1241 		("uio_td %p is not td %p", uio->uio_td, curthread));
1242 
1243 	if (uio->uio_resid == 0)
1244 		return 0;
1245 
1246 	vp = (struct vnode *)fp->f_data;
1247 	if (vp == NULL || vp->v_type == VBAD)
1248 		return EBADF;
1249 
1250 	node = DEVFS_NODE(vp);
1251 
1252 	if ((dev = vp->v_rdev) == NULL)
1253 		return EBADF;
1254 
1255 	reference_dev(dev);
1256 
1257 	if ((flags & O_FOFFSET) == 0)
1258 		uio->uio_offset = fp->f_offset;
1259 
1260 	ioflag = 0;
1261 	if (flags & O_FBLOCKING) {
1262 		/* ioflag &= ~IO_NDELAY; */
1263 	} else if (flags & O_FNONBLOCKING) {
1264 		ioflag |= IO_NDELAY;
1265 	} else if (fp->f_flag & FNONBLOCK) {
1266 		ioflag |= IO_NDELAY;
1267 	}
1268 	if (fp->f_flag & O_DIRECT) {
1269 		ioflag |= IO_DIRECT;
1270 	}
1271 	ioflag |= sequential_heuristic(uio, fp);
1272 
1273 	error = dev_dread(dev, uio, ioflag, fp);
1274 
1275 	release_dev(dev);
1276 	if (node)
1277 		vfs_timestamp(&node->atime);
1278 	if ((flags & O_FOFFSET) == 0)
1279 		fp->f_offset = uio->uio_offset;
1280 	fp->f_nextoff = uio->uio_offset;
1281 
1282 	return (error);
1283 }
1284 
1285 
1286 static int
1287 devfs_fo_write(struct file *fp, struct uio *uio,
1288 		  struct ucred *cred, int flags)
1289 {
1290 	struct devfs_node *node;
1291 	struct vnode *vp;
1292 	int ioflag;
1293 	int error;
1294 	cdev_t dev;
1295 
1296 	KASSERT(uio->uio_td == curthread,
1297 		("uio_td %p is not p %p", uio->uio_td, curthread));
1298 
1299 	vp = (struct vnode *)fp->f_data;
1300 	if (vp == NULL || vp->v_type == VBAD)
1301 		return EBADF;
1302 
1303 	node = DEVFS_NODE(vp);
1304 
1305 	if (vp->v_type == VREG)
1306 		bwillwrite(uio->uio_resid);
1307 
1308 	vp = (struct vnode *)fp->f_data;
1309 
1310 	if ((dev = vp->v_rdev) == NULL)
1311 		return EBADF;
1312 
1313 	reference_dev(dev);
1314 
1315 	if ((flags & O_FOFFSET) == 0)
1316 		uio->uio_offset = fp->f_offset;
1317 
1318 	ioflag = IO_UNIT;
1319 	if (vp->v_type == VREG &&
1320 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1321 		ioflag |= IO_APPEND;
1322 	}
1323 
1324 	if (flags & O_FBLOCKING) {
1325 		/* ioflag &= ~IO_NDELAY; */
1326 	} else if (flags & O_FNONBLOCKING) {
1327 		ioflag |= IO_NDELAY;
1328 	} else if (fp->f_flag & FNONBLOCK) {
1329 		ioflag |= IO_NDELAY;
1330 	}
1331 	if (fp->f_flag & O_DIRECT) {
1332 		ioflag |= IO_DIRECT;
1333 	}
1334 	if (flags & O_FASYNCWRITE) {
1335 		/* ioflag &= ~IO_SYNC; */
1336 	} else if (flags & O_FSYNCWRITE) {
1337 		ioflag |= IO_SYNC;
1338 	} else if (fp->f_flag & O_FSYNC) {
1339 		ioflag |= IO_SYNC;
1340 	}
1341 
1342 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1343 		ioflag |= IO_SYNC;
1344 	ioflag |= sequential_heuristic(uio, fp);
1345 
1346 	error = dev_dwrite(dev, uio, ioflag, fp);
1347 
1348 	release_dev(dev);
1349 	if (node) {
1350 		vfs_timestamp(&node->atime);
1351 		vfs_timestamp(&node->mtime);
1352 	}
1353 
1354 	if ((flags & O_FOFFSET) == 0)
1355 		fp->f_offset = uio->uio_offset;
1356 	fp->f_nextoff = uio->uio_offset;
1357 
1358 	return (error);
1359 }
1360 
1361 
1362 static int
1363 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1364 {
1365 	struct vnode *vp;
1366 	struct vattr vattr;
1367 	struct vattr *vap;
1368 	u_short mode;
1369 	cdev_t dev;
1370 	int error;
1371 
1372 	vp = (struct vnode *)fp->f_data;
1373 	if (vp == NULL || vp->v_type == VBAD)
1374 		return EBADF;
1375 
1376 	error = vn_stat(vp, sb, cred);
1377 	if (error)
1378 		return (error);
1379 
1380 	vap = &vattr;
1381 	error = VOP_GETATTR(vp, vap);
1382 	if (error)
1383 		return (error);
1384 
1385 	/*
1386 	 * Zero the spare stat fields
1387 	 */
1388 	sb->st_lspare = 0;
1389 	sb->st_qspare2 = 0;
1390 
1391 	/*
1392 	 * Copy from vattr table ... or not in case it's a cloned device
1393 	 */
1394 	if (vap->va_fsid != VNOVAL)
1395 		sb->st_dev = vap->va_fsid;
1396 	else
1397 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1398 
1399 	sb->st_ino = vap->va_fileid;
1400 
1401 	mode = vap->va_mode;
1402 	mode |= S_IFCHR;
1403 	sb->st_mode = mode;
1404 
1405 	if (vap->va_nlink > (nlink_t)-1)
1406 		sb->st_nlink = (nlink_t)-1;
1407 	else
1408 		sb->st_nlink = vap->va_nlink;
1409 
1410 	sb->st_uid = vap->va_uid;
1411 	sb->st_gid = vap->va_gid;
1412 	sb->st_rdev = devid_from_dev(DEVFS_NODE(vp)->d_dev);
1413 	sb->st_size = vap->va_bytes;
1414 	sb->st_atimespec = vap->va_atime;
1415 	sb->st_mtimespec = vap->va_mtime;
1416 	sb->st_ctimespec = vap->va_ctime;
1417 
1418 	/*
1419 	 * A VCHR and VBLK device may track the last access and last modified
1420 	 * time independantly of the filesystem.  This is particularly true
1421 	 * because device read and write calls may bypass the filesystem.
1422 	 */
1423 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1424 		dev = vp->v_rdev;
1425 		if (dev != NULL) {
1426 			if (dev->si_lastread) {
1427 				sb->st_atimespec.tv_sec = time_second +
1428 							  (time_uptime -
1429 							   dev->si_lastread);
1430 				sb->st_atimespec.tv_nsec = 0;
1431 			}
1432 			if (dev->si_lastwrite) {
1433 				sb->st_atimespec.tv_sec = time_second +
1434 							  (time_uptime -
1435 							   dev->si_lastwrite);
1436 				sb->st_atimespec.tv_nsec = 0;
1437 			}
1438 		}
1439 	}
1440 
1441         /*
1442 	 * According to www.opengroup.org, the meaning of st_blksize is
1443 	 *   "a filesystem-specific preferred I/O block size for this
1444 	 *    object.  In some filesystem types, this may vary from file
1445 	 *    to file"
1446 	 * Default to PAGE_SIZE after much discussion.
1447 	 */
1448 
1449 	sb->st_blksize = PAGE_SIZE;
1450 
1451 	sb->st_flags = vap->va_flags;
1452 
1453 	error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1454 	if (error)
1455 		sb->st_gen = 0;
1456 	else
1457 		sb->st_gen = (u_int32_t)vap->va_gen;
1458 
1459 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1460 
1461 	/*
1462 	 * This is for ABI compatibility <= 5.7 (for ABI change made in
1463 	 * 5.7 master).
1464 	 */
1465 	sb->__old_st_blksize = sb->st_blksize;
1466 
1467 	return (0);
1468 }
1469 
1470 
1471 static int
1472 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1473 {
1474 	struct vnode *vp;
1475 	int error;
1476 	cdev_t dev;
1477 
1478 	vp = (struct vnode *)fp->f_data;
1479 	if (vp == NULL || vp->v_type == VBAD) {
1480 		error = EBADF;
1481 		goto done;
1482 	}
1483 	if ((dev = vp->v_rdev) == NULL) {
1484 		error = EBADF;
1485 		goto done;
1486 	}
1487 	reference_dev(dev);
1488 
1489 	error = dev_dkqfilter(dev, kn, fp);
1490 
1491 	release_dev(dev);
1492 
1493 done:
1494 	return (error);
1495 }
1496 
1497 static int
1498 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1499 		  struct ucred *ucred, struct sysmsg *msg)
1500 {
1501 #if 0
1502 	struct devfs_node *node;
1503 #endif
1504 	struct vnode *vp;
1505 	struct vnode *ovp;
1506 	cdev_t	dev;
1507 	int error;
1508 	struct fiodname_args *name_args;
1509 	size_t namlen;
1510 	const char *name;
1511 
1512 	vp = ((struct vnode *)fp->f_data);
1513 
1514 	if ((dev = vp->v_rdev) == NULL)
1515 		return EBADF;		/* device was revoked */
1516 
1517 	reference_dev(dev);
1518 
1519 #if 0
1520 	node = DEVFS_NODE(vp);
1521 #endif
1522 
1523 	devfs_debug(DEVFS_DEBUG_DEBUG,
1524 		    "devfs_fo_ioctl() called! for dev %s\n",
1525 		    dev->si_name);
1526 
1527 	if (com == FIODTYPE) {
1528 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1529 		error = 0;
1530 		goto out;
1531 	} else if (com == FIODNAME) {
1532 		name_args = (struct fiodname_args *)data;
1533 		name = dev->si_name;
1534 		namlen = strlen(name) + 1;
1535 
1536 		devfs_debug(DEVFS_DEBUG_DEBUG,
1537 			    "ioctl, got: FIODNAME for %s\n", name);
1538 
1539 		if (namlen <= name_args->len)
1540 			error = copyout(dev->si_name, name_args->name, namlen);
1541 		else
1542 			error = EINVAL;
1543 
1544 		devfs_debug(DEVFS_DEBUG_DEBUG,
1545 			    "ioctl stuff: error: %d\n", error);
1546 		goto out;
1547 	}
1548 
1549 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg, fp);
1550 
1551 #if 0
1552 	if (node) {
1553 		vfs_timestamp(&node->atime);
1554 		vfs_timestamp(&node->mtime);
1555 	}
1556 #endif
1557 	if (com == TIOCSCTTY) {
1558 		devfs_debug(DEVFS_DEBUG_DEBUG,
1559 			    "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1560 			    dev->si_name);
1561 	}
1562 	if (error == 0 && com == TIOCSCTTY) {
1563 		struct proc *p = curthread->td_proc;
1564 		struct session *sess;
1565 
1566 		devfs_debug(DEVFS_DEBUG_DEBUG,
1567 			    "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1568 			    dev->si_name);
1569 		if (p == NULL) {
1570 			error = ENOTTY;
1571 			goto out;
1572 		}
1573 		sess = p->p_session;
1574 
1575 		/*
1576 		 * Do nothing if reassigning same control tty
1577 		 */
1578 		if (sess->s_ttyvp == vp) {
1579 			error = 0;
1580 			goto out;
1581 		}
1582 
1583 		/*
1584 		 * Get rid of reference to old control tty
1585 		 */
1586 		ovp = sess->s_ttyvp;
1587 		vref(vp);
1588 		sess->s_ttyvp = vp;
1589 		if (ovp)
1590 			vrele(ovp);
1591 	}
1592 
1593 out:
1594 	release_dev(dev);
1595 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1596 	return (error);
1597 }
1598 
1599 
1600 static int
1601 devfs_spec_fsync(struct vop_fsync_args *ap)
1602 {
1603 	struct vnode *vp = ap->a_vp;
1604 	int error;
1605 
1606 	if (!vn_isdisk(vp, NULL))
1607 		return (0);
1608 
1609 	/*
1610 	 * Flush all dirty buffers associated with a block device.
1611 	 */
1612 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1613 	return (error);
1614 }
1615 
1616 static int
1617 devfs_spec_read(struct vop_read_args *ap)
1618 {
1619 	struct devfs_node *node;
1620 	struct vnode *vp;
1621 	struct uio *uio;
1622 	cdev_t dev;
1623 	int error;
1624 
1625 	vp = ap->a_vp;
1626 	dev = vp->v_rdev;
1627 	uio = ap->a_uio;
1628 	node = DEVFS_NODE(vp);
1629 
1630 	if (dev == NULL)		/* device was revoked */
1631 		return (EBADF);
1632 	if (uio->uio_resid == 0)
1633 		return (0);
1634 
1635 	vn_unlock(vp);
1636 	error = dev_dread(dev, uio, ap->a_ioflag, NULL);
1637 	vn_lock(vp, LK_SHARED | LK_RETRY);
1638 
1639 	if (node)
1640 		vfs_timestamp(&node->atime);
1641 
1642 	return (error);
1643 }
1644 
1645 /*
1646  * Vnode op for write
1647  *
1648  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1649  *	      struct ucred *a_cred)
1650  */
1651 static int
1652 devfs_spec_write(struct vop_write_args *ap)
1653 {
1654 	struct devfs_node *node;
1655 	struct vnode *vp;
1656 	struct uio *uio;
1657 	cdev_t dev;
1658 	int error;
1659 
1660 	vp = ap->a_vp;
1661 	dev = vp->v_rdev;
1662 	uio = ap->a_uio;
1663 	node = DEVFS_NODE(vp);
1664 
1665 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1666 
1667 	if (dev == NULL)		/* device was revoked */
1668 		return (EBADF);
1669 
1670 	vn_unlock(vp);
1671 	error = dev_dwrite(dev, uio, ap->a_ioflag, NULL);
1672 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1673 
1674 	if (node) {
1675 		vfs_timestamp(&node->atime);
1676 		vfs_timestamp(&node->mtime);
1677 	}
1678 
1679 	return (error);
1680 }
1681 
1682 /*
1683  * Device ioctl operation.
1684  *
1685  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1686  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1687  */
1688 static int
1689 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1690 {
1691 	struct vnode *vp = ap->a_vp;
1692 #if 0
1693 	struct devfs_node *node;
1694 #endif
1695 	cdev_t dev;
1696 
1697 	if ((dev = vp->v_rdev) == NULL)
1698 		return (EBADF);		/* device was revoked */
1699 #if 0
1700 	node = DEVFS_NODE(vp);
1701 
1702 	if (node) {
1703 		vfs_timestamp(&node->atime);
1704 		vfs_timestamp(&node->mtime);
1705 	}
1706 #endif
1707 
1708 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1709 			   ap->a_cred, ap->a_sysmsg, NULL));
1710 }
1711 
1712 /*
1713  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1714  */
1715 /* ARGSUSED */
1716 static int
1717 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1718 {
1719 	struct vnode *vp = ap->a_vp;
1720 #if 0
1721 	struct devfs_node *node;
1722 #endif
1723 	cdev_t dev;
1724 
1725 	if ((dev = vp->v_rdev) == NULL)
1726 		return (EBADF);		/* device was revoked (EBADF) */
1727 #if 0
1728 	node = DEVFS_NODE(vp);
1729 
1730 	if (node)
1731 		vfs_timestamp(&node->atime);
1732 #endif
1733 
1734 	return (dev_dkqfilter(dev, ap->a_kn, NULL));
1735 }
1736 
1737 /*
1738  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1739  * calls are not limited to device DMA limits so we have to deal with the
1740  * case.
1741  *
1742  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1743  */
1744 static int
1745 devfs_spec_strategy(struct vop_strategy_args *ap)
1746 {
1747 	struct bio *bio = ap->a_bio;
1748 	struct buf *bp = bio->bio_buf;
1749 	struct buf *nbp;
1750 	struct vnode *vp;
1751 	struct mount *mp;
1752 	int chunksize;
1753 	int maxiosize;
1754 
1755 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1756 		buf_start(bp);
1757 
1758 	/*
1759 	 * Collect statistics on synchronous and asynchronous read
1760 	 * and write counts for disks that have associated filesystems.
1761 	 */
1762 	vp = ap->a_vp;
1763 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1764 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1765 		if (bp->b_cmd == BUF_CMD_READ) {
1766 			if (bp->b_flags & BIO_SYNC)
1767 				mp->mnt_stat.f_syncreads++;
1768 			else
1769 				mp->mnt_stat.f_asyncreads++;
1770 		} else {
1771 			if (bp->b_flags & BIO_SYNC)
1772 				mp->mnt_stat.f_syncwrites++;
1773 			else
1774 				mp->mnt_stat.f_asyncwrites++;
1775 		}
1776 	}
1777 
1778         /*
1779          * Device iosize limitations only apply to read and write.  Shortcut
1780          * the I/O if it fits.
1781          */
1782 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1783 		devfs_debug(DEVFS_DEBUG_DEBUG,
1784 			    "%s: si_iosize_max not set!\n",
1785 			    dev_dname(vp->v_rdev));
1786 		maxiosize = MAXPHYS;
1787 	}
1788 #if SPEC_CHAIN_DEBUG & 2
1789 	maxiosize = 4096;
1790 #endif
1791         if (bp->b_bcount <= maxiosize ||
1792             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1793                 dev_dstrategy_chain(vp->v_rdev, bio);
1794                 return (0);
1795         }
1796 
1797 	/*
1798 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1799 	 */
1800 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1801 	initbufbio(nbp);
1802 	buf_dep_init(nbp);
1803 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1804 	BUF_KERNPROC(nbp);
1805 	nbp->b_vp = vp;
1806 	nbp->b_flags = B_PAGING | B_KVABIO | (bp->b_flags & B_BNOCLIP);
1807 	nbp->b_cpumask = bp->b_cpumask;
1808 	nbp->b_data = bp->b_data;
1809 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1810 	nbp->b_bio1.bio_offset = bio->bio_offset;
1811 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1812 
1813 	/*
1814 	 * Start the first transfer
1815 	 */
1816 	if (vn_isdisk(vp, NULL))
1817 		chunksize = vp->v_rdev->si_bsize_phys;
1818 	else
1819 		chunksize = DEV_BSIZE;
1820 	chunksize = rounddown(maxiosize, chunksize);
1821 #if SPEC_CHAIN_DEBUG & 1
1822 	devfs_debug(DEVFS_DEBUG_DEBUG,
1823 		    "spec_strategy chained I/O chunksize=%d\n",
1824 		    chunksize);
1825 #endif
1826 	nbp->b_cmd = bp->b_cmd;
1827 	nbp->b_bcount = chunksize;
1828 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1829 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1830 
1831 #if SPEC_CHAIN_DEBUG & 1
1832 	devfs_debug(DEVFS_DEBUG_DEBUG,
1833 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1834 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1835 #endif
1836 
1837 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1838 
1839 	if (DEVFS_NODE(vp)) {
1840 		vfs_timestamp(&DEVFS_NODE(vp)->atime);
1841 		vfs_timestamp(&DEVFS_NODE(vp)->mtime);
1842 	}
1843 
1844 	return (0);
1845 }
1846 
1847 /*
1848  * Chunked up transfer completion routine - chain transfers until done
1849  *
1850  * NOTE: MPSAFE callback.
1851  */
1852 static
1853 void
1854 devfs_spec_strategy_done(struct bio *nbio)
1855 {
1856 	struct buf *nbp = nbio->bio_buf;
1857 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1858 	struct buf *bp = bio->bio_buf;			/* original bp */
1859 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1860 	int boffset = nbp->b_data - bp->b_data;
1861 
1862 	if (nbp->b_flags & B_ERROR) {
1863 		/*
1864 		 * An error terminates the chain, propogate the error back
1865 		 * to the original bp
1866 		 */
1867 		bp->b_flags |= B_ERROR;
1868 		bp->b_error = nbp->b_error;
1869 		bp->b_resid = bp->b_bcount - boffset +
1870 			      (nbp->b_bcount - nbp->b_resid);
1871 #if SPEC_CHAIN_DEBUG & 1
1872 		devfs_debug(DEVFS_DEBUG_DEBUG,
1873 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1874 			    bp, bp->b_error, bp->b_bcount,
1875 			    bp->b_bcount - bp->b_resid);
1876 #endif
1877 	} else if (nbp->b_resid) {
1878 		/*
1879 		 * A short read or write terminates the chain
1880 		 */
1881 		bp->b_error = nbp->b_error;
1882 		bp->b_resid = bp->b_bcount - boffset +
1883 			      (nbp->b_bcount - nbp->b_resid);
1884 #if SPEC_CHAIN_DEBUG & 1
1885 		devfs_debug(DEVFS_DEBUG_DEBUG,
1886 			    "spec_strategy: chain %p short read(1) "
1887 			    "bcount %d/%d\n",
1888 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1889 #endif
1890 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1891 		/*
1892 		 * A short read or write can also occur by truncating b_bcount
1893 		 */
1894 #if SPEC_CHAIN_DEBUG & 1
1895 		devfs_debug(DEVFS_DEBUG_DEBUG,
1896 			    "spec_strategy: chain %p short read(2) "
1897 			    "bcount %d/%d\n",
1898 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1899 #endif
1900 		bp->b_error = 0;
1901 		bp->b_bcount = nbp->b_bcount + boffset;
1902 		bp->b_resid = nbp->b_resid;
1903 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1904 		/*
1905 		 * No more data terminates the chain
1906 		 */
1907 #if SPEC_CHAIN_DEBUG & 1
1908 		devfs_debug(DEVFS_DEBUG_DEBUG,
1909 			    "spec_strategy: chain %p finished bcount %d\n",
1910 			    bp, bp->b_bcount);
1911 #endif
1912 		bp->b_error = 0;
1913 		bp->b_resid = 0;
1914 	} else {
1915 		/*
1916 		 * Continue the chain
1917 		 */
1918 		boffset += nbp->b_bcount;
1919 		nbp->b_data = bp->b_data + boffset;
1920 		nbp->b_bcount = bp->b_bcount - boffset;
1921 		if (nbp->b_bcount > chunksize)
1922 			nbp->b_bcount = chunksize;
1923 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1924 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1925 
1926 #if SPEC_CHAIN_DEBUG & 1
1927 		devfs_debug(DEVFS_DEBUG_DEBUG,
1928 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1929 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
1930 #endif
1931 
1932 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1933 		return;
1934 	}
1935 
1936 	/*
1937 	 * Fall through to here on termination.  biodone(bp) and
1938 	 * clean up and free nbp.
1939 	 */
1940 	biodone(bio);
1941 	BUF_UNLOCK(nbp);
1942 	uninitbufbio(nbp);
1943 	kfree(nbp, M_DEVBUF);
1944 }
1945 
1946 /*
1947  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1948  */
1949 static int
1950 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1951 {
1952 	struct buf *bp;
1953 
1954 	/*
1955 	 * Must be a synchronous operation
1956 	 */
1957 	KKASSERT(ap->a_vp->v_rdev != NULL);
1958 	if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
1959 		return (0);
1960 	bp = getpbuf(NULL);
1961 	bp->b_cmd = BUF_CMD_FREEBLKS;
1962 	bp->b_bio1.bio_flags |= BIO_SYNC;
1963 	bp->b_bio1.bio_offset = ap->a_offset;
1964 	bp->b_bio1.bio_done = biodone_sync;
1965 	bp->b_bcount = ap->a_length;
1966 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1967 	biowait(&bp->b_bio1, "TRIM");
1968 	relpbuf(bp, NULL);
1969 
1970 	return (0);
1971 }
1972 
1973 /*
1974  * Implement degenerate case where the block requested is the block
1975  * returned, and assume that the entire device is contiguous in regards
1976  * to the contiguous block range (runp and runb).
1977  *
1978  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1979  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
1980  */
1981 static int
1982 devfs_spec_bmap(struct vop_bmap_args *ap)
1983 {
1984 	if (ap->a_doffsetp != NULL)
1985 		*ap->a_doffsetp = ap->a_loffset;
1986 	if (ap->a_runp != NULL)
1987 		*ap->a_runp = MAXBSIZE;
1988 	if (ap->a_runb != NULL) {
1989 		if (ap->a_loffset < MAXBSIZE)
1990 			*ap->a_runb = (int)ap->a_loffset;
1991 		else
1992 			*ap->a_runb = MAXBSIZE;
1993 	}
1994 	return (0);
1995 }
1996 
1997 
1998 /*
1999  * Special device advisory byte-level locks.
2000  *
2001  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
2002  *		struct flock *a_fl, int a_flags)
2003  */
2004 /* ARGSUSED */
2005 static int
2006 devfs_spec_advlock(struct vop_advlock_args *ap)
2007 {
2008 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
2009 }
2010 
2011 /*
2012  * NOTE: MPSAFE callback.
2013  */
2014 static void
2015 devfs_spec_getpages_iodone(struct bio *bio)
2016 {
2017 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
2018 	wakeup(bio->bio_buf);
2019 }
2020 
2021 /*
2022  * spec_getpages() - get pages associated with device vnode.
2023  *
2024  * Note that spec_read and spec_write do not use the buffer cache, so we
2025  * must fully implement getpages here.
2026  */
2027 static int
2028 devfs_spec_getpages(struct vop_getpages_args *ap)
2029 {
2030 	vm_offset_t kva;
2031 	int error;
2032 	int i, pcount, size;
2033 	struct buf *bp;
2034 	vm_page_t m;
2035 	vm_ooffset_t offset;
2036 	int toff, nextoff, nread;
2037 	struct vnode *vp = ap->a_vp;
2038 	int blksiz;
2039 	int gotreqpage;
2040 
2041 	error = 0;
2042 	pcount = round_page(ap->a_count) / PAGE_SIZE;
2043 
2044 	/*
2045 	 * Calculate the offset of the transfer and do sanity check.
2046 	 */
2047 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
2048 
2049 	/*
2050 	 * Round up physical size for real devices.  We cannot round using
2051 	 * v_mount's block size data because v_mount has nothing to do with
2052 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
2053 	 * size for the device itself.
2054 	 *
2055 	 * We can't use v_rdev->si_mountpoint because it only exists when the
2056 	 * block device is mounted.  However, we can use v_rdev.
2057 	 */
2058 	if (vn_isdisk(vp, NULL))
2059 		blksiz = vp->v_rdev->si_bsize_phys;
2060 	else
2061 		blksiz = DEV_BSIZE;
2062 
2063 	size = roundup2(ap->a_count, blksiz);
2064 
2065 	bp = getpbuf_kva(NULL);
2066 	kva = (vm_offset_t)bp->b_data;
2067 
2068 	/*
2069 	 * Map the pages to be read into the kva.
2070 	 */
2071 	pmap_qenter_noinval(kva, ap->a_m, pcount);
2072 
2073 	/* Build a minimal buffer header. */
2074 	bp->b_cmd = BUF_CMD_READ;
2075 	bp->b_flags |= B_KVABIO;
2076 	bp->b_bcount = size;
2077 	bp->b_resid = 0;
2078 	bsetrunningbufspace(bp, size);
2079 
2080 	bp->b_bio1.bio_offset = offset;
2081 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2082 
2083 	mycpu->gd_cnt.v_vnodein++;
2084 	mycpu->gd_cnt.v_vnodepgsin += pcount;
2085 
2086 	/* Do the input. */
2087 	vn_strategy(ap->a_vp, &bp->b_bio1);
2088 
2089 	crit_enter();
2090 
2091 	/* We definitely need to be at splbio here. */
2092 	while (bp->b_cmd != BUF_CMD_DONE)
2093 		tsleep(bp, 0, "spread", 0);
2094 
2095 	crit_exit();
2096 
2097 	if (bp->b_flags & B_ERROR) {
2098 		if (bp->b_error)
2099 			error = bp->b_error;
2100 		else
2101 			error = EIO;
2102 	}
2103 
2104 	/*
2105 	 * If EOF is encountered we must zero-extend the result in order
2106 	 * to ensure that the page does not contain garabge.  When no
2107 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
2108 	 * b_resid is relative to b_bcount and should be 0, but some devices
2109 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
2110 	 */
2111 	nread = bp->b_bcount - bp->b_resid;
2112 	if (nread < ap->a_count) {
2113 		bkvasync(bp);
2114 		bzero((caddr_t)kva + nread, ap->a_count - nread);
2115 	}
2116 	pmap_qremove_noinval(kva, pcount);
2117 
2118 	gotreqpage = 0;
2119 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2120 		nextoff = toff + PAGE_SIZE;
2121 		m = ap->a_m[i];
2122 
2123 		/*
2124 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2125 		 *	 pmap modified bit.  pmap modified bit should have
2126 		 *	 already been cleared.
2127 		 */
2128 		if (nextoff <= nread) {
2129 			m->valid = VM_PAGE_BITS_ALL;
2130 			vm_page_undirty(m);
2131 		} else if (toff < nread) {
2132 			/*
2133 			 * Since this is a VM request, we have to supply the
2134 			 * unaligned offset to allow vm_page_set_valid()
2135 			 * to zero sub-DEV_BSIZE'd portions of the page.
2136 			 */
2137 			vm_page_set_valid(m, 0, nread - toff);
2138 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2139 		} else {
2140 			m->valid = 0;
2141 			vm_page_undirty(m);
2142 		}
2143 
2144 		if (i != ap->a_reqpage) {
2145 			/*
2146 			 * Just in case someone was asking for this page we
2147 			 * now tell them that it is ok to use.
2148 			 */
2149 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2150 				if (m->valid) {
2151 					if (m->flags & PG_REFERENCED) {
2152 						vm_page_activate(m);
2153 					} else {
2154 						vm_page_deactivate(m);
2155 					}
2156 					vm_page_wakeup(m);
2157 				} else {
2158 					vm_page_free(m);
2159 				}
2160 			} else {
2161 				vm_page_free(m);
2162 			}
2163 		} else if (m->valid) {
2164 			gotreqpage = 1;
2165 			/*
2166 			 * Since this is a VM request, we need to make the
2167 			 * entire page presentable by zeroing invalid sections.
2168 			 */
2169 			if (m->valid != VM_PAGE_BITS_ALL)
2170 			    vm_page_zero_invalid(m, FALSE);
2171 		}
2172 	}
2173 	if (!gotreqpage) {
2174 		m = ap->a_m[ap->a_reqpage];
2175 		devfs_debug(DEVFS_DEBUG_WARNING,
2176 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2177 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2178 		devfs_debug(DEVFS_DEBUG_WARNING,
2179 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2180 		    size, bp->b_resid, ap->a_count, m->valid);
2181 		devfs_debug(DEVFS_DEBUG_WARNING,
2182 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2183 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2184 		/*
2185 		 * Free the buffer header back to the swap buffer pool.
2186 		 */
2187 		relpbuf(bp, NULL);
2188 		return VM_PAGER_ERROR;
2189 	}
2190 	/*
2191 	 * Free the buffer header back to the swap buffer pool.
2192 	 */
2193 	relpbuf(bp, NULL);
2194 	if (DEVFS_NODE(ap->a_vp))
2195 		vfs_timestamp(&DEVFS_NODE(ap->a_vp)->mtime);
2196 	return VM_PAGER_OK;
2197 }
2198 
2199 static __inline
2200 int
2201 sequential_heuristic(struct uio *uio, struct file *fp)
2202 {
2203 	/*
2204 	 * Sequential heuristic - detect sequential operation
2205 	 */
2206 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2207 	    uio->uio_offset == fp->f_nextoff) {
2208 		/*
2209 		 * XXX we assume that the filesystem block size is
2210 		 * the default.  Not true, but still gives us a pretty
2211 		 * good indicator of how sequential the read operations
2212 		 * are.
2213 		 */
2214 		int tmpseq = fp->f_seqcount;
2215 
2216 		tmpseq += (uio->uio_resid + MAXBSIZE - 1) / MAXBSIZE;
2217 		if (tmpseq > IO_SEQMAX)
2218 			tmpseq = IO_SEQMAX;
2219 		fp->f_seqcount = tmpseq;
2220 		return(fp->f_seqcount << IO_SEQSHIFT);
2221 	}
2222 
2223 	/*
2224 	 * Not sequential, quick draw-down of seqcount
2225 	 */
2226 	if (fp->f_seqcount > 1)
2227 		fp->f_seqcount = 1;
2228 	else
2229 		fp->f_seqcount = 0;
2230 	return(0);
2231 }
2232