xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision c9733229451fac5faa53b1a016b01866eae75a1c)
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/caps.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/dirent.h>
50 #include <sys/malloc.h>
51 #include <sys/stat.h>
52 #include <sys/reg.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_zone.h>
55 #include <vm/vm_object.h>
56 #include <sys/filio.h>
57 #include <sys/ttycom.h>
58 #include <sys/tty.h>
59 #include <sys/diskslice.h>
60 #include <sys/sysctl.h>
61 #include <sys/devfs.h>
62 #include <sys/pioctl.h>
63 #include <vfs/fifofs/fifo.h>
64 
65 #include <machine/limits.h>
66 
67 #include <sys/buf2.h>
68 #include <vm/vm_page2.h>
69 
70 #ifndef SPEC_CHAIN_DEBUG
71 #define SPEC_CHAIN_DEBUG 0
72 #endif
73 
74 MALLOC_DECLARE(M_DEVFS);
75 #define DEVFS_BADOP	(void *)devfs_vop_badop
76 
77 static int devfs_vop_badop(struct vop_generic_args *);
78 static int devfs_vop_access(struct vop_access_args *);
79 static int devfs_vop_inactive(struct vop_inactive_args *);
80 static int devfs_vop_reclaim(struct vop_reclaim_args *);
81 static int devfs_vop_readdir(struct vop_readdir_args *);
82 static int devfs_vop_getattr(struct vop_getattr_args *);
83 static int devfs_vop_setattr(struct vop_setattr_args *);
84 static int devfs_vop_readlink(struct vop_readlink_args *);
85 static int devfs_vop_print(struct vop_print_args *);
86 
87 static int devfs_vop_nresolve(struct vop_nresolve_args *);
88 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
89 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
90 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
91 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
92 static int devfs_vop_nremove(struct vop_nremove_args *);
93 
94 static int devfs_spec_open(struct vop_open_args *);
95 static int devfs_spec_close(struct vop_close_args *);
96 static int devfs_spec_fsync(struct vop_fsync_args *);
97 
98 static int devfs_spec_read(struct vop_read_args *);
99 static int devfs_spec_write(struct vop_write_args *);
100 static int devfs_spec_ioctl(struct vop_ioctl_args *);
101 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
102 static int devfs_spec_strategy(struct vop_strategy_args *);
103 static void devfs_spec_strategy_done(struct bio *);
104 static int devfs_spec_freeblks(struct vop_freeblks_args *);
105 static int devfs_spec_bmap(struct vop_bmap_args *);
106 static int devfs_spec_advlock(struct vop_advlock_args *);
107 static void devfs_spec_getpages_iodone(struct bio *);
108 static int devfs_spec_getpages(struct vop_getpages_args *);
109 
110 static int devfs_fo_close(struct file *);
111 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
112 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
113 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
114 static int devfs_fo_kqfilter(struct file *, struct knote *);
115 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
116 				struct ucred *, struct sysmsg *);
117 static int devfs_fo_seek(struct file *, off_t, int, off_t *);
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 	.fo_seek	= devfs_fo_seek
199 };
200 
201 /*
202  * These two functions are possibly temporary hacks for devices (aka
203  * the pty code) which want to control the node attributes themselves.
204  *
205  * XXX we may ultimately desire to simply remove the uid/gid/mode
206  * from the node entirely.
207  *
208  * MPSAFE - sorta.  Theoretically the overwrite can compete since they
209  *	    are loading from the same fields.
210  */
211 static __inline void
212 node_sync_dev_get(struct devfs_node *node)
213 {
214 	cdev_t dev;
215 
216 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
217 		node->uid = dev->si_uid;
218 		node->gid = dev->si_gid;
219 		node->mode = dev->si_perms;
220 	}
221 }
222 
223 static __inline void
224 node_sync_dev_set(struct devfs_node *node)
225 {
226 	cdev_t dev;
227 
228 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
229 		dev->si_uid = node->uid;
230 		dev->si_gid = node->gid;
231 		dev->si_perms = node->mode;
232 	}
233 }
234 
235 /*
236  * generic entry point for unsupported operations
237  */
238 static int
239 devfs_vop_badop(struct vop_generic_args *ap)
240 {
241 	return (EIO);
242 }
243 
244 
245 static int
246 devfs_vop_access(struct vop_access_args *ap)
247 {
248 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
249 	int error;
250 
251 	if (!devfs_node_is_accessible(node))
252 		return ENOENT;
253 	node_sync_dev_get(node);
254 	error = vop_helper_access(ap, node->uid, node->gid,
255 				  node->mode, node->flags);
256 
257 	return error;
258 }
259 
260 
261 static int
262 devfs_vop_inactive(struct vop_inactive_args *ap)
263 {
264 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
265 
266 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
267 		vrecycle(ap->a_vp);
268 	return 0;
269 }
270 
271 
272 static int
273 devfs_vop_reclaim(struct vop_reclaim_args *ap)
274 {
275 	struct devfs_node *node;
276 	struct vnode *vp;
277 	int locked;
278 
279 	/*
280 	 * Check if it is locked already. if not, we acquire the devfs lock
281 	 */
282 	if ((lockstatus(&devfs_lock, curthread)) != LK_EXCLUSIVE) {
283 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
284 		locked = 1;
285 	} else {
286 		locked = 0;
287 	}
288 
289 	/*
290 	 * Get rid of the devfs_node if it is no longer linked into the
291 	 * topology.  Interlocked by devfs_lock.  However, be careful
292 	 * interposing other operations between cleaning out v_data and
293 	 * devfs_freep() as the node is only protected by devfs_lock
294 	 * once the vnode is disassociated.
295 	 */
296 	vp = ap->a_vp;
297 	node = DEVFS_NODE(vp);
298 
299 	if (node) {
300 		if (node->v_node != vp) {
301 			kprintf("NODE->V_NODE MISMATCH VP=%p NODEVP=%p\n",
302 				vp, node->v_node);
303 		}
304 		vp->v_data = NULL;
305 		node->v_node = NULL;
306 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
307 			devfs_freep(node);
308 	}
309 	v_release_rdev(vp);
310 
311 	if (locked)
312 		lockmgr(&devfs_lock, LK_RELEASE);
313 
314 	/*
315 	 * v_rdev needs to be properly released using v_release_rdev
316 	 * Make sure v_data is NULL as well.
317 	 */
318 	return 0;
319 }
320 
321 
322 static int
323 devfs_vop_readdir(struct vop_readdir_args *ap)
324 {
325 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
326 	struct devfs_node *node;
327 	int cookie_index;
328 	int ncookies;
329 	int error2;
330 	int error;
331 	int r;
332 	off_t *cookies;
333 	off_t saveoff;
334 
335 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
336 
337 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
338 		return (EINVAL);
339 	error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
340 	if (error)
341 		return (error);
342 
343 	if (!devfs_node_is_accessible(dnode)) {
344 		vn_unlock(ap->a_vp);
345 		return ENOENT;
346 	}
347 
348 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
349 
350 	saveoff = ap->a_uio->uio_offset;
351 
352 	if (ap->a_ncookies) {
353 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
354 		if (ncookies > 256)
355 			ncookies = 256;
356 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
357 		cookie_index = 0;
358 	} else {
359 		ncookies = -1;
360 		cookies = NULL;
361 		cookie_index = 0;
362 	}
363 
364 	vfs_timestamp(&dnode->atime);
365 
366 	if (saveoff == 0) {
367 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
368 				     DT_DIR, 1, ".");
369 		if (r)
370 			goto done;
371 		if (cookies)
372 			cookies[cookie_index] = saveoff;
373 		saveoff++;
374 		cookie_index++;
375 		if (cookie_index == ncookies)
376 			goto done;
377 	}
378 
379 	if (saveoff == 1) {
380 		if (dnode->parent) {
381 			r = vop_write_dirent(&error, ap->a_uio,
382 					     dnode->parent->d_dir.d_ino,
383 					     DT_DIR, 2, "..");
384 		} else {
385 			r = vop_write_dirent(&error, ap->a_uio,
386 					     dnode->d_dir.d_ino,
387 					     DT_DIR, 2, "..");
388 		}
389 		if (r)
390 			goto done;
391 		if (cookies)
392 			cookies[cookie_index] = saveoff;
393 		saveoff++;
394 		cookie_index++;
395 		if (cookie_index == ncookies)
396 			goto done;
397 	}
398 
399 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
400 		if ((node->flags & DEVFS_HIDDEN) ||
401 		    (node->flags & DEVFS_INVISIBLE)) {
402 			continue;
403 		}
404 
405 		/*
406 		 * If the node type is a valid devfs alias, then we make
407 		 * sure that the target isn't hidden. If it is, we don't
408 		 * show the link in the directory listing.
409 		 */
410 		if ((node->node_type == Nlink) && (node->link_target != NULL) &&
411 			(node->link_target->flags & DEVFS_HIDDEN))
412 			continue;
413 
414 		if (node->cookie < saveoff)
415 			continue;
416 
417 		saveoff = node->cookie;
418 
419 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
420 					  node->d_dir.d_type,
421 					  node->d_dir.d_namlen,
422 					  node->d_dir.d_name);
423 
424 		if (error2)
425 			break;
426 
427 		saveoff++;
428 
429 		if (cookies)
430 			cookies[cookie_index] = node->cookie;
431 		++cookie_index;
432 		if (cookie_index == ncookies)
433 			break;
434 	}
435 
436 done:
437 	lockmgr(&devfs_lock, LK_RELEASE);
438 	vn_unlock(ap->a_vp);
439 
440 	ap->a_uio->uio_offset = saveoff;
441 	if (error && cookie_index == 0) {
442 		if (cookies) {
443 			kfree(cookies, M_TEMP);
444 			*ap->a_ncookies = 0;
445 			*ap->a_cookies = NULL;
446 		}
447 	} else {
448 		if (cookies) {
449 			*ap->a_ncookies = cookie_index;
450 			*ap->a_cookies = cookies;
451 		}
452 	}
453 	return (error);
454 }
455 
456 
457 static int
458 devfs_vop_nresolve(struct vop_nresolve_args *ap)
459 {
460 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
461 	struct devfs_node *node, *found = NULL;
462 	struct namecache *ncp;
463 	struct vnode *vp = NULL;
464 	int error = 0;
465 	int len;
466 	int depth;
467 
468 	ncp = ap->a_nch->ncp;
469 	len = ncp->nc_nlen;
470 
471 	if (!devfs_node_is_accessible(dnode))
472 		return ENOENT;
473 
474 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
475 
476 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) {
477 		error = ENOENT;
478 		cache_setvp(ap->a_nch, NULL);
479 		goto out;
480 	}
481 
482 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
483 		if (len == node->d_dir.d_namlen) {
484 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
485 				found = node;
486 				break;
487 			}
488 		}
489 	}
490 
491 	if (found) {
492 		depth = 0;
493 		while ((found->node_type == Nlink) && (found->link_target)) {
494 			if (depth >= 8) {
495 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
496 				break;
497 			}
498 
499 			found = found->link_target;
500 			++depth;
501 		}
502 
503 		if (!(found->flags & DEVFS_HIDDEN))
504 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
505 	}
506 
507 	if (vp == NULL) {
508 		error = ENOENT;
509 		cache_setvp(ap->a_nch, NULL);
510 		goto out;
511 
512 	}
513 	KKASSERT(vp);
514 	vn_unlock(vp);
515 	cache_setvp(ap->a_nch, vp);
516 	vrele(vp);
517 out:
518 	lockmgr(&devfs_lock, LK_RELEASE);
519 
520 	return error;
521 }
522 
523 
524 static int
525 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
526 {
527 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
528 
529 	*ap->a_vpp = NULL;
530 	if (!devfs_node_is_accessible(dnode))
531 		return ENOENT;
532 
533 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
534 	if (dnode->parent != NULL) {
535 		devfs_allocv(ap->a_vpp, dnode->parent);
536 		vn_unlock(*ap->a_vpp);
537 	}
538 	lockmgr(&devfs_lock, LK_RELEASE);
539 
540 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
541 }
542 
543 
544 /*
545  * getattr() - Does not need a lock since the vp is refd
546  */
547 static int
548 devfs_vop_getattr(struct vop_getattr_args *ap)
549 {
550 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
551 	struct vattr *vap = ap->a_vap;
552 	struct partinfo pinfo;
553 	int error = 0;
554 
555 #if 0
556 	if (!devfs_node_is_accessible(node))
557 		return ENOENT;
558 #endif
559 
560 	/*
561 	 * XXX This is a temporary hack to prevent crashes when the device is
562 	 * being destroyed (and so the underlying node will be gone) while
563 	 * a userland program is blocked in a read().
564 	 */
565 	if (node == NULL)
566 		return EIO;
567 
568 	node_sync_dev_get(node);
569 
570 	/* start by zeroing out the attributes */
571 	VATTR_NULL(vap);
572 
573 	/* next do all the common fields */
574 	vap->va_type = ap->a_vp->v_type;
575 	vap->va_mode = node->mode;
576 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
577 	vap->va_flags = 0;
578 	vap->va_blocksize = DEV_BSIZE;
579 	vap->va_bytes = vap->va_size = 0;
580 
581 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
582 
583 	vap->va_atime = node->atime;
584 	vap->va_mtime = node->mtime;
585 	vap->va_ctime = node->ctime;
586 
587 	vap->va_nlink = 1; /* number of references to file */
588 
589 	vap->va_uid = node->uid;
590 	vap->va_gid = node->gid;
591 
592 	vap->va_rmajor = 0;
593 	vap->va_rminor = 0;
594 
595 	if ((node->node_type == Ndev) && node->d_dev)  {
596 		reference_dev(node->d_dev);
597 		vap->va_rminor = node->d_dev->si_uminor;
598 		release_dev(node->d_dev);
599 	}
600 
601 	/* For a softlink the va_size is the length of the softlink */
602 	if (node->symlink_name != 0) {
603 		vap->va_bytes = vap->va_size = node->symlink_namelen;
604 	}
605 
606 	/*
607 	 * For a disk-type device, va_size is the size of the underlying
608 	 * device, so that lseek() works properly.
609 	 */
610 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
611 		bzero(&pinfo, sizeof(pinfo));
612 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
613 				   0, proc0.p_ucred, NULL, NULL);
614 		if ((error == 0) && (pinfo.media_blksize != 0)) {
615 			vap->va_size = pinfo.media_size;
616 		} else {
617 			vap->va_size = 0;
618 			error = 0;
619 		}
620 	}
621 
622 	return (error);
623 }
624 
625 static int
626 devfs_vop_setattr(struct vop_setattr_args *ap)
627 {
628 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
629 	struct vattr *vap;
630 	uid_t cur_uid;
631 	gid_t cur_gid;
632 	mode_t cur_mode;
633 	int error = 0;
634 
635 	if (!devfs_node_is_accessible(node))
636 		return ENOENT;
637 	node_sync_dev_get(node);
638 
639 	vap = ap->a_vap;
640 
641 	if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
642 		cur_uid = node->uid;
643 		cur_gid = node->gid;
644 		cur_mode = node->mode;
645 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
646 		    ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
647 		if (error)
648 			goto out;
649 
650 		if (node->uid != cur_uid || node->gid != cur_gid) {
651 			node->uid = cur_uid;
652 			node->gid = cur_gid;
653 			node->mode = cur_mode;
654 		}
655 	}
656 
657 	if (vap->va_mode != (mode_t)VNOVAL) {
658 		cur_mode = node->mode;
659 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
660 		    node->uid, node->gid, &cur_mode);
661 		if (error == 0 && node->mode != cur_mode) {
662 			node->mode = cur_mode;
663 		}
664 	}
665 
666 out:
667 	node_sync_dev_set(node);
668 	vfs_timestamp(&node->ctime);
669 
670 	return error;
671 }
672 
673 
674 static int
675 devfs_vop_readlink(struct vop_readlink_args *ap)
676 {
677 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
678 	int ret;
679 
680 	if (!devfs_node_is_accessible(node))
681 		return ENOENT;
682 
683 	lockmgr(&devfs_lock, LK_SHARED);
684 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
685 	lockmgr(&devfs_lock, LK_RELEASE);
686 
687 	return ret;
688 }
689 
690 
691 static int
692 devfs_vop_print(struct vop_print_args *ap)
693 {
694 	return (0);
695 }
696 
697 static int
698 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
699 {
700 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
701 	struct devfs_node *node;
702 
703 	if (!devfs_node_is_accessible(dnode))
704 		return ENOENT;
705 
706 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
707 		goto out;
708 
709 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
710 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir,
711 		      ap->a_nch->ncp->nc_name, dnode, NULL);
712 
713 	if (*ap->a_vpp) {
714 		node = DEVFS_NODE(*ap->a_vpp);
715 		node->flags |= DEVFS_USER_CREATED;
716 		cache_setunresolved(ap->a_nch);
717 		cache_setvp(ap->a_nch, *ap->a_vpp);
718 	}
719 	lockmgr(&devfs_lock, LK_RELEASE);
720 out:
721 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
722 }
723 
724 static int
725 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
726 {
727 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
728 	struct devfs_node *node;
729 	size_t targetlen;
730 
731 	if (!devfs_node_is_accessible(dnode))
732 		return ENOENT;
733 
734 	ap->a_vap->va_type = VLNK;
735 
736 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
737 		goto out;
738 
739 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
740 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink,
741 		      ap->a_nch->ncp->nc_name, dnode, NULL);
742 
743 	targetlen = strlen(ap->a_target);
744 	if (*ap->a_vpp) {
745 		node = DEVFS_NODE(*ap->a_vpp);
746 		node->flags |= DEVFS_USER_CREATED;
747 		node->symlink_namelen = targetlen;
748 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
749 		memcpy(node->symlink_name, ap->a_target, targetlen);
750 		node->symlink_name[targetlen] = '\0';
751 		cache_setunresolved(ap->a_nch);
752 		cache_setvp(ap->a_nch, *ap->a_vpp);
753 	}
754 	lockmgr(&devfs_lock, LK_RELEASE);
755 out:
756 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
757 }
758 
759 static int
760 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
761 {
762 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
763 	struct devfs_node *node;
764 	struct namecache *ncp;
765 	int error = ENOENT;
766 
767 	ncp = ap->a_nch->ncp;
768 
769 	if (!devfs_node_is_accessible(dnode))
770 		return ENOENT;
771 
772 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
773 
774 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
775 		goto out;
776 
777 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
778 		if (ncp->nc_nlen != node->d_dir.d_namlen)
779 			continue;
780 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
781 			continue;
782 
783 		/*
784 		 * only allow removal of user created dirs
785 		 */
786 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
787 			error = EPERM;
788 			goto out;
789 		} else if (node->node_type != Ndir) {
790 			error = ENOTDIR;
791 			goto out;
792 		} else if (node->nchildren > 2) {
793 			error = ENOTEMPTY;
794 			goto out;
795 		} else {
796 			if (node->v_node)
797 				cache_inval_vp(node->v_node, CINV_DESTROY);
798 			devfs_unlinkp(node);
799 			error = 0;
800 			break;
801 		}
802 	}
803 
804 	cache_unlink(ap->a_nch);
805 out:
806 	lockmgr(&devfs_lock, LK_RELEASE);
807 	return error;
808 }
809 
810 static int
811 devfs_vop_nremove(struct vop_nremove_args *ap)
812 {
813 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
814 	struct devfs_node *node;
815 	struct namecache *ncp;
816 	int error = ENOENT;
817 
818 	ncp = ap->a_nch->ncp;
819 
820 	if (!devfs_node_is_accessible(dnode))
821 		return ENOENT;
822 
823 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
824 
825 	if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
826 		goto out;
827 
828 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
829 		if (ncp->nc_nlen != node->d_dir.d_namlen)
830 			continue;
831 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
832 			continue;
833 
834 		/*
835 		 * only allow removal of user created stuff (e.g. symlinks)
836 		 */
837 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
838 			error = EPERM;
839 			goto out;
840 		} else if (node->node_type == Ndir) {
841 			error = EISDIR;
842 			goto out;
843 		} else {
844 			if (node->v_node)
845 				cache_inval_vp(node->v_node, CINV_DESTROY);
846 			devfs_unlinkp(node);
847 			error = 0;
848 			break;
849 		}
850 	}
851 
852 	cache_unlink(ap->a_nch);
853 out:
854 	lockmgr(&devfs_lock, LK_RELEASE);
855 	return error;
856 }
857 
858 
859 static int
860 devfs_spec_open(struct vop_open_args *ap)
861 {
862 	struct vnode *vp = ap->a_vp;
863 	struct vnode *orig_vp = NULL;
864 	struct devfs_node *node = DEVFS_NODE(vp);
865 	struct devfs_node *newnode;
866 	cdev_t dev, ndev = NULL;
867 	int error = 0;
868 
869 	if (node) {
870 		if (node->d_dev == NULL)
871 			return ENXIO;
872 		if (!devfs_node_is_accessible(node))
873 			return ENOENT;
874 	}
875 
876 	if ((dev = vp->v_rdev) == NULL)
877 		return ENXIO;
878 
879 	/*
880 	 * Simple devices that don't care.  Retain the shared lock.
881 	 */
882 	if (dev_dflags(dev) & D_QUICK) {
883 		vn_unlock(vp);
884 		error = dev_dopen(dev, ap->a_mode, S_IFCHR,
885 				  ap->a_cred, ap->a_fpp, vp);
886 		vn_lock(vp, LK_SHARED | LK_RETRY);
887 		if (error)
888 			return error;
889 		vop_stdopen(ap);
890 		goto skip;
891 	}
892 
893 	/*
894 	 * Slow code
895 	 */
896 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
897 	if (node && ap->a_fpp) {
898 		int exists;
899 
900 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
901 		lockmgr(&devfs_lock, LK_SHARED);
902 
903 		ndev = devfs_clone(dev, node->d_dir.d_name,
904 				   node->d_dir.d_namlen,
905 				   ap->a_mode, ap->a_cred);
906 		if (ndev != NULL) {
907 			lockmgr(&devfs_lock, LK_RELEASE);
908 			lockmgr(&devfs_lock, LK_EXCLUSIVE);
909 			newnode = devfs_create_device_node(
910 					DEVFS_MNTDATA(vp->v_mount)->root_node,
911 					ndev, &exists, NULL, NULL);
912 			/* XXX: possibly destroy device if this happens */
913 
914 			if (newnode != NULL) {
915 				dev = ndev;
916 				if (exists == 0)
917 					devfs_link_dev(dev);
918 
919 				devfs_debug(DEVFS_DEBUG_DEBUG,
920 						"parent here is: %s, node is: |%s|\n",
921 						((node->parent->node_type == Nroot) ?
922 						"ROOT!" : node->parent->d_dir.d_name),
923 						newnode->d_dir.d_name);
924 				devfs_debug(DEVFS_DEBUG_DEBUG,
925 						"test: %s\n",
926 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
927 
928 				/*
929 				 * orig_vp is set to the original vp if we
930 				 * cloned.
931 				 */
932 				/* node->flags |= DEVFS_CLONED; */
933 				devfs_allocv(&vp, newnode);
934 				orig_vp = ap->a_vp;
935 				ap->a_vp = vp;
936 			}
937 		}
938 		lockmgr(&devfs_lock, LK_RELEASE);
939 
940 		/*
941 		 * Synchronize devfs here to make sure that, if the cloned
942 		 * device creates other device nodes in addition to the
943 		 * cloned one, all of them are created by the time we return
944 		 * from opening the cloned one.
945 		 */
946 		if (ndev)
947 			devfs_config();
948 	}
949 
950 	devfs_debug(DEVFS_DEBUG_DEBUG,
951 		    "devfs_spec_open() called on %s! \n",
952 		    dev->si_name);
953 
954 	/*
955 	 * Make this field valid before any I/O in ->d_open
956 	 *
957 	 * NOTE: Shared vnode lock probably held, but its ok as long
958 	 *	 as assignments are consistent.
959 	 */
960 	if (!dev->si_iosize_max)
961 		/* XXX: old DFLTPHYS == 64KB dependency */
962 		dev->si_iosize_max = min(MAXPHYS,64*1024);
963 
964 	if (dev_dflags(dev) & D_TTY)
965 		vsetflags(vp, VISTTY);
966 
967 	/*
968 	 * Open the underlying device.
969 	 *
970 	 * NOTE: If the dev open returns EALREADY it has completed the open
971 	 *	 operation and is returning a fully initialized *a->a_fpp
972 	 *	 (which it may also have replaced).  This includes issuing
973 	 *	 any necessary VOP_OPEN().
974 	 *
975 	 *	 Also, the returned ap->a_fpp might not be DTYPE_VNODE and
976 	 *	 if it is might not be using the vp we supplied to it.
977 	 */
978 	vn_unlock(vp);
979 	error = dev_dopen(dev, ap->a_mode, S_IFCHR,
980 			  ap->a_cred, ap->a_fpp, vp);
981 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
982 
983 	if (__predict_false(error == EALREADY)) {
984 		if (orig_vp)
985 			vput(vp);
986 		return 0;
987 	}
988 
989 	/*
990 	 * Clean up any cloned vp if we error out.
991 	 */
992 	if (__predict_false(error != 0)) {
993 		if (orig_vp) {
994 			vput(vp);
995 			ap->a_vp = orig_vp;
996 			/* orig_vp = NULL; */
997 		}
998 		return error;
999 	}
1000 
1001 	/*
1002 	 * This checks if the disk device is going to be opened for writing.
1003 	 * It will be only allowed in the cases where securelevel permits it
1004 	 * and it's not mounted R/W.
1005 	 */
1006 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
1007 	    (ap->a_cred != FSCRED)) {
1008 
1009 		/* Very secure mode. No open for writing allowed */
1010 		if (securelevel >= 2)
1011 			return EPERM;
1012 
1013 		/*
1014 		 * If it is mounted R/W, do not allow to open for writing.
1015 		 * In the case it's mounted read-only but securelevel
1016 		 * is >= 1, then do not allow opening for writing either.
1017 		 */
1018 		if (vfs_mountedon(vp)) {
1019 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
1020 				return EBUSY;
1021 			else if (securelevel >= 1)
1022 				return EPERM;
1023 		}
1024 	}
1025 
1026 	/*
1027 	 * NOTE: vnode is still locked shared.  t_stop assignment should
1028 	 *	 remain consistent so we should be ok.
1029 	 */
1030 	if (dev_dflags(dev) & D_TTY) {
1031 		if (dev->si_tty) {
1032 			struct tty *tp;
1033 			tp = dev->si_tty;
1034 			if (!tp->t_stop) {
1035 				devfs_debug(DEVFS_DEBUG_DEBUG,
1036 					    "devfs: no t_stop\n");
1037 				tp->t_stop = nottystop;
1038 			}
1039 		}
1040 	}
1041 
1042 	/*
1043 	 * NOTE: vnode is still locked shared.  assignments should
1044 	 *	 remain consistent so we should be ok.  However,
1045 	 *	 upgrade to exclusive if we need a VM object.
1046 	 */
1047 	if (vn_isdisk(vp, NULL)) {
1048 		if (!dev->si_bsize_phys)
1049 			dev->si_bsize_phys = DEV_BSIZE;
1050 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
1051 	}
1052 
1053 	vop_stdopen(ap);
1054 #if 0
1055 	if (node)
1056 		vfs_timestamp(&node->atime);
1057 #endif
1058 	/*
1059 	 * If we replaced the vp the vop_stdopen() call will have loaded
1060 	 * it into fp->f_data and vref()d the vp, giving us two refs.  So
1061 	 * instead of just unlocking it here we have to vput() it.
1062 	 */
1063 	if (orig_vp)
1064 		vput(vp);
1065 
1066 	/* Ugly pty magic, to make pty devices appear once they are opened */
1067 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) {
1068 		if (node->flags & DEVFS_INVISIBLE)
1069 			node->flags &= ~DEVFS_INVISIBLE;
1070 	}
1071 
1072 skip:
1073 	if (ap->a_fpp) {
1074 		struct file *fp = *ap->a_fpp;
1075 
1076 		KKASSERT(fp->f_type == DTYPE_VNODE);
1077 		KKASSERT((fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1078 		fp->f_ops = &devfs_dev_fileops;
1079 		KKASSERT(fp->f_data == (void *)vp);
1080 	}
1081 
1082 	return 0;
1083 }
1084 
1085 static int
1086 devfs_spec_close(struct vop_close_args *ap)
1087 {
1088 	struct devfs_node *node;
1089 	struct proc *p = curproc;
1090 	struct vnode *vp = ap->a_vp;
1091 	cdev_t dev = vp->v_rdev;
1092 	int error = 0;
1093 	int needrelock;
1094 	int opencount;
1095 
1096 	/*
1097 	 * Devices flagged D_QUICK require no special handling.
1098 	 */
1099 	if (dev && dev_dflags(dev) & D_QUICK) {
1100 		opencount = vp->v_opencount;
1101 		if (opencount <= 1)
1102 			opencount = count_dev(dev);   /* XXX NOT SMP SAFE */
1103 		if (((vp->v_flag & VRECLAIMED) ||
1104 		    (dev_dflags(dev) & D_TRACKCLOSE) ||
1105 		    (opencount == 1))) {
1106 			vn_unlock(vp);
1107 			error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1108 			vn_lock(vp, LK_SHARED | LK_RETRY);
1109 		}
1110 		goto skip;
1111 	}
1112 
1113 	/*
1114 	 * We do special tests on the opencount so unfortunately we need
1115 	 * an exclusive lock.
1116 	 */
1117 	vn_lock(vp, LK_UPGRADE | LK_RETRY);
1118 
1119 	if (dev)
1120 		devfs_debug(DEVFS_DEBUG_DEBUG,
1121 			    "devfs_spec_close() called on %s! \n",
1122 			    dev->si_name);
1123 	else
1124 		devfs_debug(DEVFS_DEBUG_DEBUG,
1125 			    "devfs_spec_close() called, null vode!\n");
1126 
1127 	/*
1128 	 * A couple of hacks for devices and tty devices.  The
1129 	 * vnode ref count cannot be used to figure out the
1130 	 * last close, but we can use v_opencount now that
1131 	 * revoke works properly.
1132 	 *
1133 	 * Detect the last close on a controlling terminal and clear
1134 	 * the session (half-close).
1135 	 *
1136 	 * XXX opencount is not SMP safe.  The vnode is locked but there
1137 	 *     may be multiple vnodes referencing the same device.
1138 	 */
1139 	if (dev) {
1140 		/*
1141 		 * NOTE: Try to avoid global tokens when testing opencount
1142 		 * XXX hack, fixme. needs a struct lock and opencount in
1143 		 * struct cdev itself.
1144 		 */
1145 		reference_dev(dev);
1146 		opencount = vp->v_opencount;
1147 		if (opencount <= 1)
1148 			opencount = count_dev(dev);   /* XXX NOT SMP SAFE */
1149 	} else {
1150 		opencount = 0;
1151 	}
1152 
1153 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1154 		p->p_session->s_ttyvp = NULL;
1155 		vrele(vp);
1156 	}
1157 
1158 	/*
1159 	 * Vnodes can be opened and closed multiple times.  Do not really
1160 	 * close the device unless (1) it is being closed forcibly,
1161 	 * (2) the device wants to track closes, or (3) this is the last
1162 	 * vnode doing its last close on the device.
1163 	 *
1164 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1165 	 * a closed device.  This might not occur now that our revoke is
1166 	 * fixed.
1167 	 */
1168 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1169 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1170 		    (dev_dflags(dev) & D_TRACKCLOSE) ||
1171 		    (opencount == 1))) {
1172 		/*
1173 		 * Ugly pty magic, to make pty devices disappear again once
1174 		 * they are closed.
1175 		 */
1176 		node = DEVFS_NODE(ap->a_vp);
1177 		if (node && (node->flags & DEVFS_PTY))
1178 			node->flags |= DEVFS_INVISIBLE;
1179 
1180 		/*
1181 		 * Unlock around dev_dclose(), unless the vnode is
1182 		 * undergoing a vgone/reclaim (during umount).
1183 		 */
1184 		needrelock = 0;
1185 		if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1186 			needrelock = 1;
1187 			vn_unlock(vp);
1188 		}
1189 
1190 		/*
1191 		 * WARNING!  If the device destroys itself the devfs node
1192 		 *	     can disappear here.
1193 		 *
1194 		 * WARNING!  vn_lock() will fail if the vp is in a VRECLAIM,
1195 		 *	     which can occur during umount.
1196 		 */
1197 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1198 		/* node is now stale */
1199 
1200 		if (needrelock) {
1201 			if (vn_lock(vp, LK_EXCLUSIVE |
1202 					LK_RETRY |
1203 					LK_FAILRECLAIM) != 0) {
1204 				panic("devfs_spec_close: vnode %p "
1205 				      "unexpectedly could not be relocked",
1206 				      vp);
1207 			}
1208 		}
1209 	} else {
1210 		error = 0;
1211 	}
1212 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1213 
1214 	/*
1215 	 * Track the actual opens and closes on the vnode.  The last close
1216 	 * disassociates the rdev.  If the rdev is already disassociated or
1217 	 * the opencount is already 0, the vnode might have been revoked
1218 	 * and no further opencount tracking occurs.
1219 	 */
1220 	if (dev)
1221 		release_dev(dev);
1222 skip:
1223 	if (vp->v_opencount > 0)
1224 		vop_stdclose(ap);
1225 	return(error);
1226 
1227 }
1228 
1229 
1230 static int
1231 devfs_fo_close(struct file *fp)
1232 {
1233 	struct vnode *vp = (struct vnode *)fp->f_data;
1234 	int error;
1235 
1236 	fp->f_ops = &badfileops;
1237 	error = vn_close(vp, fp->f_flag, fp);
1238 	devfs_clear_cdevpriv(fp);
1239 
1240 	return (error);
1241 }
1242 
1243 
1244 /*
1245  * Device-optimized file table vnode read routine.
1246  *
1247  * This bypasses the VOP table and talks directly to the device.  Most
1248  * filesystems just route to specfs and can make this optimization.
1249  */
1250 static int
1251 devfs_fo_read(struct file *fp, struct uio *uio,
1252 		 struct ucred *cred, int flags)
1253 {
1254 	struct devfs_node *node;
1255 	struct vnode *vp;
1256 	int ioflag;
1257 	int error;
1258 	cdev_t dev;
1259 
1260 	KASSERT(uio->uio_td == curthread,
1261 		("uio_td %p is not td %p", uio->uio_td, curthread));
1262 
1263 	if (uio->uio_resid == 0)
1264 		return 0;
1265 
1266 	vp = (struct vnode *)fp->f_data;
1267 	if (vp == NULL || vp->v_type == VBAD)
1268 		return EBADF;
1269 
1270 	node = DEVFS_NODE(vp);
1271 
1272 	if ((dev = vp->v_rdev) == NULL)
1273 		return EBADF;
1274 
1275 	reference_dev(dev);
1276 
1277 	if ((flags & O_FOFFSET) == 0)
1278 		uio->uio_offset = fp->f_offset;
1279 
1280 	ioflag = 0;
1281 	if (flags & O_FBLOCKING) {
1282 		/* ioflag &= ~IO_NDELAY; */
1283 	} else if (flags & O_FNONBLOCKING) {
1284 		ioflag |= IO_NDELAY;
1285 	} else if (fp->f_flag & FNONBLOCK) {
1286 		ioflag |= IO_NDELAY;
1287 	}
1288 	if (fp->f_flag & O_DIRECT) {
1289 		ioflag |= IO_DIRECT;
1290 	}
1291 	ioflag |= sequential_heuristic(uio, fp);
1292 
1293 	error = dev_dread(dev, uio, ioflag, fp);
1294 
1295 	release_dev(dev);
1296 	if (node)
1297 		vfs_timestamp(&node->atime);
1298 	if ((flags & O_FOFFSET) == 0)
1299 		fp->f_offset = uio->uio_offset;
1300 	fp->f_nextoff = uio->uio_offset;
1301 
1302 	return (error);
1303 }
1304 
1305 
1306 static int
1307 devfs_fo_write(struct file *fp, struct uio *uio,
1308 		  struct ucred *cred, int flags)
1309 {
1310 	struct devfs_node *node;
1311 	struct vnode *vp;
1312 	int ioflag;
1313 	int error;
1314 	cdev_t dev;
1315 
1316 	KASSERT(uio->uio_td == curthread,
1317 		("uio_td %p is not p %p", uio->uio_td, curthread));
1318 
1319 	vp = (struct vnode *)fp->f_data;
1320 	if (vp == NULL || vp->v_type == VBAD)
1321 		return EBADF;
1322 
1323 	node = DEVFS_NODE(vp);
1324 
1325 	if (vp->v_type == VREG)
1326 		bwillwrite(uio->uio_resid);
1327 
1328 	vp = (struct vnode *)fp->f_data;
1329 
1330 	if ((dev = vp->v_rdev) == NULL)
1331 		return EBADF;
1332 
1333 	reference_dev(dev);
1334 
1335 	if ((flags & O_FOFFSET) == 0)
1336 		uio->uio_offset = fp->f_offset;
1337 
1338 	ioflag = IO_UNIT;
1339 	if (vp->v_type == VREG &&
1340 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1341 		ioflag |= IO_APPEND;
1342 	}
1343 
1344 	if (flags & O_FBLOCKING) {
1345 		/* ioflag &= ~IO_NDELAY; */
1346 	} else if (flags & O_FNONBLOCKING) {
1347 		ioflag |= IO_NDELAY;
1348 	} else if (fp->f_flag & FNONBLOCK) {
1349 		ioflag |= IO_NDELAY;
1350 	}
1351 	if (fp->f_flag & O_DIRECT) {
1352 		ioflag |= IO_DIRECT;
1353 	}
1354 	if (flags & O_FASYNCWRITE) {
1355 		/* ioflag &= ~IO_SYNC; */
1356 	} else if (flags & O_FSYNCWRITE) {
1357 		ioflag |= IO_SYNC;
1358 	} else if (fp->f_flag & O_FSYNC) {
1359 		ioflag |= IO_SYNC;
1360 	}
1361 
1362 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1363 		ioflag |= IO_SYNC;
1364 	ioflag |= sequential_heuristic(uio, fp);
1365 
1366 	error = dev_dwrite(dev, uio, ioflag, fp);
1367 
1368 	release_dev(dev);
1369 	if (node) {
1370 		vfs_timestamp(&node->atime);
1371 		vfs_timestamp(&node->mtime);
1372 	}
1373 
1374 	if ((flags & O_FOFFSET) == 0)
1375 		fp->f_offset = uio->uio_offset;
1376 	fp->f_nextoff = uio->uio_offset;
1377 
1378 	return (error);
1379 }
1380 
1381 
1382 static int
1383 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1384 {
1385 	struct vnode *vp;
1386 	struct vattr vattr;
1387 	struct vattr *vap;
1388 	u_short mode;
1389 	cdev_t dev;
1390 	int error;
1391 
1392 	vp = (struct vnode *)fp->f_data;
1393 	if (vp == NULL || vp->v_type == VBAD)
1394 		return EBADF;
1395 
1396 	error = vn_stat(vp, sb, cred);
1397 	if (error)
1398 		return (error);
1399 
1400 	vap = &vattr;
1401 	error = VOP_GETATTR(vp, vap);
1402 	if (error)
1403 		return (error);
1404 
1405 	/*
1406 	 * Zero the spare stat fields
1407 	 */
1408 	sb->st_lspare = 0;
1409 	sb->st_qspare2 = 0;
1410 
1411 	/*
1412 	 * Copy from vattr table ... or not in case it's a cloned device
1413 	 */
1414 	if (vap->va_fsid != VNOVAL)
1415 		sb->st_dev = vap->va_fsid;
1416 	else
1417 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1418 
1419 	sb->st_ino = vap->va_fileid;
1420 
1421 	mode = vap->va_mode;
1422 	mode |= S_IFCHR;
1423 	sb->st_mode = mode;
1424 
1425 	if (vap->va_nlink > (nlink_t)-1)
1426 		sb->st_nlink = (nlink_t)-1;
1427 	else
1428 		sb->st_nlink = vap->va_nlink;
1429 
1430 	sb->st_uid = vap->va_uid;
1431 	sb->st_gid = vap->va_gid;
1432 	sb->st_rdev = devid_from_dev(DEVFS_NODE(vp)->d_dev);
1433 	sb->st_size = vap->va_bytes;
1434 	sb->st_atimespec = vap->va_atime;
1435 	sb->st_mtimespec = vap->va_mtime;
1436 	sb->st_ctimespec = vap->va_ctime;
1437 
1438 	/*
1439 	 * A VCHR and VBLK device may track the last access and last modified
1440 	 * time independantly of the filesystem.  This is particularly true
1441 	 * because device read and write calls may bypass the filesystem.
1442 	 */
1443 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1444 		dev = vp->v_rdev;
1445 		if (dev != NULL) {
1446 			if (dev->si_lastread) {
1447 				sb->st_atimespec.tv_sec = time_second +
1448 							  (dev->si_lastread -
1449 							   time_uptime);
1450 				sb->st_atimespec.tv_nsec = 0;
1451 			}
1452 			if (dev->si_lastwrite) {
1453 				sb->st_mtimespec.tv_sec = time_second +
1454 							  (dev->si_lastwrite -
1455 							   time_uptime);
1456 				sb->st_mtimespec.tv_nsec = 0;
1457 			}
1458 		}
1459 	}
1460 
1461         /*
1462 	 * According to www.opengroup.org, the meaning of st_blksize is
1463 	 *   "a filesystem-specific preferred I/O block size for this
1464 	 *    object.  In some filesystem types, this may vary from file
1465 	 *    to file"
1466 	 * Default to PAGE_SIZE after much discussion.
1467 	 */
1468 
1469 	sb->st_blksize = PAGE_SIZE;
1470 
1471 	sb->st_flags = vap->va_flags;
1472 
1473 	error = caps_priv_check(cred, SYSCAP_NOVFS_GENERATION);
1474 	if (error)
1475 		sb->st_gen = 0;
1476 	else
1477 		sb->st_gen = (u_int32_t)vap->va_gen;
1478 
1479 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1480 
1481 	/*
1482 	 * This is for ABI compatibility <= 5.7 (for ABI change made in
1483 	 * 5.7 master).
1484 	 */
1485 	sb->__old_st_blksize = sb->st_blksize;
1486 
1487 	return (0);
1488 }
1489 
1490 
1491 static int
1492 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1493 {
1494 	struct vnode *vp;
1495 	int error;
1496 	cdev_t dev;
1497 
1498 	vp = (struct vnode *)fp->f_data;
1499 	if (vp == NULL || vp->v_type == VBAD) {
1500 		error = EBADF;
1501 		goto done;
1502 	}
1503 	if ((dev = vp->v_rdev) == NULL) {
1504 		error = EBADF;
1505 		goto done;
1506 	}
1507 	reference_dev(dev);
1508 
1509 	error = dev_dkqfilter(dev, kn, fp);
1510 
1511 	release_dev(dev);
1512 
1513 done:
1514 	return (error);
1515 }
1516 
1517 static int
1518 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1519 		  struct ucred *ucred, struct sysmsg *msg)
1520 {
1521 #if 0
1522 	struct devfs_node *node;
1523 #endif
1524 	struct vnode *vp;
1525 	struct vnode *ovp;
1526 	cdev_t	dev;
1527 	int error;
1528 	struct fiodname_args *name_args;
1529 	size_t namlen;
1530 	const char *name;
1531 
1532 	vp = ((struct vnode *)fp->f_data);
1533 
1534 	if ((dev = vp->v_rdev) == NULL)
1535 		return EBADF;		/* device was revoked */
1536 
1537 	reference_dev(dev);
1538 
1539 #if 0
1540 	node = DEVFS_NODE(vp);
1541 #endif
1542 
1543 	devfs_debug(DEVFS_DEBUG_DEBUG,
1544 		    "devfs_fo_ioctl() called! for dev %s\n",
1545 		    dev->si_name);
1546 
1547 	if (com == FIODTYPE) {
1548 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1549 		error = 0;
1550 		goto out;
1551 	} else if (com == FIODNAME) {
1552 		name_args = (struct fiodname_args *)data;
1553 		name = dev->si_name;
1554 		namlen = strlen(name) + 1;
1555 
1556 		devfs_debug(DEVFS_DEBUG_DEBUG,
1557 			    "ioctl, got: FIODNAME for %s\n", name);
1558 
1559 		if (namlen <= name_args->len)
1560 			error = copyout(dev->si_name, name_args->name, namlen);
1561 		else
1562 			error = EINVAL;
1563 
1564 		devfs_debug(DEVFS_DEBUG_DEBUG,
1565 			    "ioctl stuff: error: %d\n", error);
1566 		goto out;
1567 	}
1568 
1569 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg, fp);
1570 
1571 #if 0
1572 	if (node) {
1573 		vfs_timestamp(&node->atime);
1574 		vfs_timestamp(&node->mtime);
1575 	}
1576 #endif
1577 	if (com == TIOCSCTTY) {
1578 		devfs_debug(DEVFS_DEBUG_DEBUG,
1579 			    "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1580 			    dev->si_name);
1581 	}
1582 	if (error == 0 && com == TIOCSCTTY) {
1583 		struct proc *p = curthread->td_proc;
1584 		struct session *sess;
1585 
1586 		devfs_debug(DEVFS_DEBUG_DEBUG,
1587 			    "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1588 			    dev->si_name);
1589 		if (p == NULL) {
1590 			error = ENOTTY;
1591 			goto out;
1592 		}
1593 		sess = p->p_session;
1594 
1595 		/*
1596 		 * Do nothing if reassigning same control tty
1597 		 */
1598 		if (sess->s_ttyvp == vp) {
1599 			error = 0;
1600 			goto out;
1601 		}
1602 
1603 		/*
1604 		 * Get rid of reference to old control tty
1605 		 */
1606 		ovp = sess->s_ttyvp;
1607 		vref(vp);
1608 		sess->s_ttyvp = vp;
1609 		if (ovp)
1610 			vrele(ovp);
1611 	}
1612 
1613 out:
1614 	release_dev(dev);
1615 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1616 	return (error);
1617 }
1618 
1619 int
1620 devfs_fo_seek(struct file *fp, off_t offset, int whence, off_t *res)
1621 {
1622 	/*
1623 	 * NOTE: vnode_fileops uses exact same code
1624 	 */
1625 	struct vnode *vp;
1626 	struct vattr_lite lva;
1627 	off_t new_offset;
1628 	int error;
1629 
1630 	vp = (struct vnode *)fp->f_data;
1631 
1632 	switch (whence) {
1633 	case L_INCR:
1634 		spin_lock(&fp->f_spin);
1635 		new_offset = fp->f_offset + offset;
1636 		error = 0;
1637 		break;
1638 	case L_XTND:
1639 		error = VOP_GETATTR_LITE(vp, &lva);
1640 		spin_lock(&fp->f_spin);
1641 		new_offset = offset + lva.va_size;
1642 		break;
1643 	case L_SET:
1644 		new_offset = offset;
1645 		error = 0;
1646 		spin_lock(&fp->f_spin);
1647 		break;
1648 	default:
1649 		new_offset = 0;
1650 		error = EINVAL;
1651 		spin_lock(&fp->f_spin);
1652 		break;
1653 	}
1654 
1655 	/*
1656 	 * Validate the seek position.  Negative offsets are not allowed
1657 	 * for regular files or directories.
1658 	 *
1659 	 * Normally we would also not want to allow negative offsets for
1660 	 * character and block-special devices.  However kvm addresses
1661 	 * on 64 bit architectures might appear to be negative and must
1662 	 * be allowed.
1663 	 */
1664 	if (error == 0) {
1665 		if (new_offset < 0 &&
1666 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
1667 			error = EINVAL;
1668 		} else {
1669 			fp->f_offset = new_offset;
1670 		}
1671 	}
1672 	*res = fp->f_offset;
1673 	spin_unlock(&fp->f_spin);
1674 
1675 	return (error);
1676 }
1677 
1678 static int
1679 devfs_spec_fsync(struct vop_fsync_args *ap)
1680 {
1681 	struct vnode *vp = ap->a_vp;
1682 	int error;
1683 
1684 	if (!vn_isdisk(vp, NULL))
1685 		return (0);
1686 
1687 	/*
1688 	 * Flush all dirty buffers associated with a block device.
1689 	 */
1690 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1691 	return (error);
1692 }
1693 
1694 static int
1695 devfs_spec_read(struct vop_read_args *ap)
1696 {
1697 	struct devfs_node *node;
1698 	struct vnode *vp;
1699 	struct uio *uio;
1700 	cdev_t dev;
1701 	int error;
1702 
1703 	vp = ap->a_vp;
1704 	dev = vp->v_rdev;
1705 	uio = ap->a_uio;
1706 	node = DEVFS_NODE(vp);
1707 
1708 	if (dev == NULL)		/* device was revoked */
1709 		return (EBADF);
1710 	if (uio->uio_resid == 0)
1711 		return (0);
1712 
1713 	vn_unlock(vp);
1714 	error = dev_dread(dev, uio, ap->a_ioflag, NULL);
1715 	vn_lock(vp, LK_SHARED | LK_RETRY);
1716 
1717 	if (node)
1718 		vfs_timestamp(&node->atime);
1719 
1720 	return (error);
1721 }
1722 
1723 /*
1724  * Vnode op for write
1725  *
1726  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1727  *	      struct ucred *a_cred)
1728  */
1729 static int
1730 devfs_spec_write(struct vop_write_args *ap)
1731 {
1732 	struct devfs_node *node;
1733 	struct vnode *vp;
1734 	struct uio *uio;
1735 	cdev_t dev;
1736 	int error;
1737 
1738 	vp = ap->a_vp;
1739 	dev = vp->v_rdev;
1740 	uio = ap->a_uio;
1741 	node = DEVFS_NODE(vp);
1742 
1743 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1744 
1745 	if (dev == NULL)		/* device was revoked */
1746 		return (EBADF);
1747 
1748 	vn_unlock(vp);
1749 	error = dev_dwrite(dev, uio, ap->a_ioflag, NULL);
1750 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1751 
1752 	if (node) {
1753 		vfs_timestamp(&node->atime);
1754 		vfs_timestamp(&node->mtime);
1755 	}
1756 
1757 	return (error);
1758 }
1759 
1760 /*
1761  * Device ioctl operation.
1762  *
1763  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1764  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1765  */
1766 static int
1767 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1768 {
1769 	struct vnode *vp = ap->a_vp;
1770 #if 0
1771 	struct devfs_node *node;
1772 #endif
1773 	cdev_t dev;
1774 
1775 	if ((dev = vp->v_rdev) == NULL)
1776 		return (EBADF);		/* device was revoked */
1777 #if 0
1778 	node = DEVFS_NODE(vp);
1779 
1780 	if (node) {
1781 		vfs_timestamp(&node->atime);
1782 		vfs_timestamp(&node->mtime);
1783 	}
1784 #endif
1785 
1786 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1787 			   ap->a_cred, ap->a_sysmsg, NULL));
1788 }
1789 
1790 /*
1791  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1792  */
1793 /* ARGSUSED */
1794 static int
1795 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1796 {
1797 	struct vnode *vp = ap->a_vp;
1798 #if 0
1799 	struct devfs_node *node;
1800 #endif
1801 	cdev_t dev;
1802 
1803 	if ((dev = vp->v_rdev) == NULL)
1804 		return (EBADF);		/* device was revoked (EBADF) */
1805 #if 0
1806 	node = DEVFS_NODE(vp);
1807 
1808 	if (node)
1809 		vfs_timestamp(&node->atime);
1810 #endif
1811 
1812 	return (dev_dkqfilter(dev, ap->a_kn, NULL));
1813 }
1814 
1815 /*
1816  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1817  * calls are not limited to device DMA limits so we have to deal with the
1818  * case.
1819  *
1820  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1821  */
1822 static int
1823 devfs_spec_strategy(struct vop_strategy_args *ap)
1824 {
1825 	struct bio *bio = ap->a_bio;
1826 	struct buf *bp = bio->bio_buf;
1827 	struct buf *nbp;
1828 	struct vnode *vp;
1829 	struct mount *mp;
1830 	int chunksize;
1831 	int maxiosize;
1832 
1833 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1834 		buf_start(bp);
1835 
1836 	/*
1837 	 * Collect statistics on synchronous and asynchronous read
1838 	 * and write counts for disks that have associated filesystems.
1839 	 */
1840 	vp = ap->a_vp;
1841 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1842 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1843 		if (bp->b_cmd == BUF_CMD_READ) {
1844 			if (bp->b_flags & BIO_SYNC)
1845 				mp->mnt_stat.f_syncreads++;
1846 			else
1847 				mp->mnt_stat.f_asyncreads++;
1848 		} else {
1849 			if (bp->b_flags & BIO_SYNC)
1850 				mp->mnt_stat.f_syncwrites++;
1851 			else
1852 				mp->mnt_stat.f_asyncwrites++;
1853 		}
1854 	}
1855 
1856         /*
1857          * Device iosize limitations only apply to read and write.  Shortcut
1858          * the I/O if it fits.
1859          */
1860 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1861 		devfs_debug(DEVFS_DEBUG_DEBUG,
1862 			    "%s: si_iosize_max not set!\n",
1863 			    dev_dname(vp->v_rdev));
1864 		maxiosize = MAXPHYS;
1865 	}
1866 #if SPEC_CHAIN_DEBUG & 2
1867 	maxiosize = 4096;
1868 #endif
1869         if (bp->b_bcount <= maxiosize ||
1870             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1871                 dev_dstrategy_chain(vp->v_rdev, bio);
1872                 return (0);
1873         }
1874 
1875 	/*
1876 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1877 	 */
1878 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1879 	initbufbio(nbp);
1880 	buf_dep_init(nbp);
1881 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1882 	BUF_KERNPROC(nbp);
1883 	nbp->b_vp = vp;
1884 	nbp->b_flags = B_PAGING | B_KVABIO | (bp->b_flags & B_BNOCLIP);
1885 	nbp->b_cpumask = bp->b_cpumask;
1886 	nbp->b_data = bp->b_data;
1887 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1888 	nbp->b_bio1.bio_offset = bio->bio_offset;
1889 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1890 
1891 	/*
1892 	 * Start the first transfer
1893 	 */
1894 	if (vn_isdisk(vp, NULL))
1895 		chunksize = vp->v_rdev->si_bsize_phys;
1896 	else
1897 		chunksize = DEV_BSIZE;
1898 	chunksize = rounddown(maxiosize, chunksize);
1899 #if SPEC_CHAIN_DEBUG & 1
1900 	devfs_debug(DEVFS_DEBUG_DEBUG,
1901 		    "spec_strategy chained I/O chunksize=%d\n",
1902 		    chunksize);
1903 #endif
1904 	nbp->b_cmd = bp->b_cmd;
1905 	nbp->b_bcount = chunksize;
1906 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1907 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1908 
1909 #if SPEC_CHAIN_DEBUG & 1
1910 	devfs_debug(DEVFS_DEBUG_DEBUG,
1911 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1912 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1913 #endif
1914 
1915 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1916 
1917 	if (DEVFS_NODE(vp)) {
1918 		vfs_timestamp(&DEVFS_NODE(vp)->atime);
1919 		vfs_timestamp(&DEVFS_NODE(vp)->mtime);
1920 	}
1921 
1922 	return (0);
1923 }
1924 
1925 /*
1926  * Chunked up transfer completion routine - chain transfers until done
1927  *
1928  * NOTE: MPSAFE callback.
1929  */
1930 static
1931 void
1932 devfs_spec_strategy_done(struct bio *nbio)
1933 {
1934 	struct buf *nbp = nbio->bio_buf;
1935 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1936 	struct buf *bp = bio->bio_buf;			/* original bp */
1937 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1938 	int boffset = nbp->b_data - bp->b_data;
1939 
1940 	if (nbp->b_flags & B_ERROR) {
1941 		/*
1942 		 * An error terminates the chain, propogate the error back
1943 		 * to the original bp
1944 		 */
1945 		bp->b_flags |= B_ERROR;
1946 		bp->b_error = nbp->b_error;
1947 		bp->b_resid = bp->b_bcount - boffset +
1948 			      (nbp->b_bcount - nbp->b_resid);
1949 #if SPEC_CHAIN_DEBUG & 1
1950 		devfs_debug(DEVFS_DEBUG_DEBUG,
1951 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1952 			    bp, bp->b_error, bp->b_bcount,
1953 			    bp->b_bcount - bp->b_resid);
1954 #endif
1955 	} else if (nbp->b_resid) {
1956 		/*
1957 		 * A short read or write terminates the chain
1958 		 */
1959 		bp->b_error = nbp->b_error;
1960 		bp->b_resid = bp->b_bcount - boffset +
1961 			      (nbp->b_bcount - nbp->b_resid);
1962 #if SPEC_CHAIN_DEBUG & 1
1963 		devfs_debug(DEVFS_DEBUG_DEBUG,
1964 			    "spec_strategy: chain %p short read(1) "
1965 			    "bcount %d/%d\n",
1966 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1967 #endif
1968 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1969 		/*
1970 		 * A short read or write can also occur by truncating b_bcount
1971 		 */
1972 #if SPEC_CHAIN_DEBUG & 1
1973 		devfs_debug(DEVFS_DEBUG_DEBUG,
1974 			    "spec_strategy: chain %p short read(2) "
1975 			    "bcount %d/%d\n",
1976 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1977 #endif
1978 		bp->b_error = 0;
1979 		bp->b_bcount = nbp->b_bcount + boffset;
1980 		bp->b_resid = nbp->b_resid;
1981 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1982 		/*
1983 		 * No more data terminates the chain
1984 		 */
1985 #if SPEC_CHAIN_DEBUG & 1
1986 		devfs_debug(DEVFS_DEBUG_DEBUG,
1987 			    "spec_strategy: chain %p finished bcount %d\n",
1988 			    bp, bp->b_bcount);
1989 #endif
1990 		bp->b_error = 0;
1991 		bp->b_resid = 0;
1992 	} else {
1993 		/*
1994 		 * Continue the chain
1995 		 */
1996 		boffset += nbp->b_bcount;
1997 		nbp->b_data = bp->b_data + boffset;
1998 		nbp->b_bcount = bp->b_bcount - boffset;
1999 		if (nbp->b_bcount > chunksize)
2000 			nbp->b_bcount = chunksize;
2001 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
2002 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
2003 
2004 #if SPEC_CHAIN_DEBUG & 1
2005 		devfs_debug(DEVFS_DEBUG_DEBUG,
2006 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
2007 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
2008 #endif
2009 
2010 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
2011 		return;
2012 	}
2013 
2014 	/*
2015 	 * Fall through to here on termination.  biodone(bp) and
2016 	 * clean up and free nbp.
2017 	 */
2018 	biodone(bio);
2019 	BUF_UNLOCK(nbp);
2020 	uninitbufbio(nbp);
2021 	kfree(nbp, M_DEVBUF);
2022 }
2023 
2024 /*
2025  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
2026  */
2027 static int
2028 devfs_spec_freeblks(struct vop_freeblks_args *ap)
2029 {
2030 	struct buf *bp;
2031 
2032 	/*
2033 	 * Must be a synchronous operation
2034 	 */
2035 	KKASSERT(ap->a_vp->v_rdev != NULL);
2036 	if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
2037 		return (0);
2038 	bp = getpbuf(NULL);
2039 	bp->b_cmd = BUF_CMD_FREEBLKS;
2040 	bp->b_bio1.bio_flags |= BIO_SYNC;
2041 	bp->b_bio1.bio_offset = ap->a_offset;
2042 	bp->b_bio1.bio_done = biodone_sync;
2043 	bp->b_bcount = ap->a_length;
2044 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
2045 	biowait(&bp->b_bio1, "TRIM");
2046 	relpbuf(bp, NULL);
2047 
2048 	return (0);
2049 }
2050 
2051 /*
2052  * Implement degenerate case where the block requested is the block
2053  * returned, and assume that the entire device is contiguous in regards
2054  * to the contiguous block range (runp and runb).
2055  *
2056  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
2057  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
2058  */
2059 static int
2060 devfs_spec_bmap(struct vop_bmap_args *ap)
2061 {
2062 	if (ap->a_doffsetp != NULL)
2063 		*ap->a_doffsetp = ap->a_loffset;
2064 	if (ap->a_runp != NULL)
2065 		*ap->a_runp = MAXBSIZE;
2066 	if (ap->a_runb != NULL) {
2067 		if (ap->a_loffset < MAXBSIZE)
2068 			*ap->a_runb = (int)ap->a_loffset;
2069 		else
2070 			*ap->a_runb = MAXBSIZE;
2071 	}
2072 	return (0);
2073 }
2074 
2075 
2076 /*
2077  * Special device advisory byte-level locks.
2078  *
2079  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
2080  *		struct flock *a_fl, int a_flags)
2081  */
2082 /* ARGSUSED */
2083 static int
2084 devfs_spec_advlock(struct vop_advlock_args *ap)
2085 {
2086 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
2087 }
2088 
2089 /*
2090  * NOTE: MPSAFE callback.
2091  */
2092 static void
2093 devfs_spec_getpages_iodone(struct bio *bio)
2094 {
2095 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
2096 	wakeup(bio->bio_buf);
2097 }
2098 
2099 /*
2100  * spec_getpages() - get pages associated with device vnode.
2101  *
2102  * Note that spec_read and spec_write do not use the buffer cache, so we
2103  * must fully implement getpages here.
2104  */
2105 static int
2106 devfs_spec_getpages(struct vop_getpages_args *ap)
2107 {
2108 	vm_offset_t kva;
2109 	int error;
2110 	int i, pcount, size;
2111 	struct buf *bp;
2112 	vm_page_t m;
2113 	vm_ooffset_t offset;
2114 	int toff, nextoff, nread;
2115 	struct vnode *vp = ap->a_vp;
2116 	int blksiz;
2117 	int gotreqpage;
2118 
2119 	error = 0;
2120 	pcount = round_page(ap->a_count) / PAGE_SIZE;
2121 
2122 	/*
2123 	 * Calculate the offset of the transfer and do sanity check.
2124 	 */
2125 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
2126 
2127 	/*
2128 	 * Round up physical size for real devices.  We cannot round using
2129 	 * v_mount's block size data because v_mount has nothing to do with
2130 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
2131 	 * size for the device itself.
2132 	 *
2133 	 * We can't use v_rdev->si_mountpoint because it only exists when the
2134 	 * block device is mounted.  However, we can use v_rdev.
2135 	 */
2136 	if (vn_isdisk(vp, NULL))
2137 		blksiz = vp->v_rdev->si_bsize_phys;
2138 	else
2139 		blksiz = DEV_BSIZE;
2140 
2141 	size = roundup2(ap->a_count, blksiz);
2142 
2143 	bp = getpbuf_kva(NULL);
2144 	kva = (vm_offset_t)bp->b_data;
2145 
2146 	/*
2147 	 * Map the pages to be read into the kva.
2148 	 */
2149 	pmap_qenter_noinval(kva, ap->a_m, pcount);
2150 
2151 	/* Build a minimal buffer header. */
2152 	bp->b_cmd = BUF_CMD_READ;
2153 	bp->b_flags |= B_KVABIO;
2154 	bp->b_bcount = size;
2155 	bp->b_resid = 0;
2156 	bsetrunningbufspace(bp, size);
2157 
2158 	bp->b_bio1.bio_offset = offset;
2159 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2160 
2161 	mycpu->gd_cnt.v_vnodein++;
2162 	mycpu->gd_cnt.v_vnodepgsin += pcount;
2163 
2164 	/* Do the input. */
2165 	vn_strategy(ap->a_vp, &bp->b_bio1);
2166 
2167 	crit_enter();
2168 
2169 	/* We definitely need to be at splbio here. */
2170 	while (bp->b_cmd != BUF_CMD_DONE)
2171 		tsleep(bp, 0, "spread", 0);
2172 
2173 	crit_exit();
2174 
2175 	if (bp->b_flags & B_ERROR) {
2176 		if (bp->b_error)
2177 			error = bp->b_error;
2178 		else
2179 			error = EIO;
2180 	}
2181 
2182 	/*
2183 	 * If EOF is encountered we must zero-extend the result in order
2184 	 * to ensure that the page does not contain garabge.  When no
2185 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
2186 	 * b_resid is relative to b_bcount and should be 0, but some devices
2187 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
2188 	 */
2189 	nread = bp->b_bcount - bp->b_resid;
2190 	if (nread < ap->a_count) {
2191 		bkvasync(bp);
2192 		bzero((caddr_t)kva + nread, ap->a_count - nread);
2193 	}
2194 	pmap_qremove_noinval(kva, pcount);
2195 
2196 	gotreqpage = 0;
2197 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2198 		nextoff = toff + PAGE_SIZE;
2199 		m = ap->a_m[i];
2200 
2201 		/*
2202 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2203 		 *	 pmap modified bit.  pmap modified bit should have
2204 		 *	 already been cleared.
2205 		 */
2206 		if (nextoff <= nread) {
2207 			m->valid = VM_PAGE_BITS_ALL;
2208 			vm_page_undirty(m);
2209 		} else if (toff < nread) {
2210 			/*
2211 			 * Since this is a VM request, we have to supply the
2212 			 * unaligned offset to allow vm_page_set_valid()
2213 			 * to zero sub-DEV_BSIZE'd portions of the page.
2214 			 */
2215 			vm_page_set_valid(m, 0, nread - toff);
2216 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2217 		} else {
2218 			m->valid = 0;
2219 			vm_page_undirty(m);
2220 		}
2221 
2222 		if (i != ap->a_reqpage) {
2223 			/*
2224 			 * Just in case someone was asking for this page we
2225 			 * now tell them that it is ok to use.
2226 			 */
2227 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2228 				if (m->valid) {
2229 					if (m->flags & PG_REFERENCED) {
2230 						vm_page_activate(m);
2231 					} else {
2232 						vm_page_deactivate(m);
2233 					}
2234 					vm_page_wakeup(m);
2235 				} else {
2236 					vm_page_free(m);
2237 				}
2238 			} else {
2239 				vm_page_free(m);
2240 			}
2241 		} else if (m->valid) {
2242 			gotreqpage = 1;
2243 			/*
2244 			 * Since this is a VM request, we need to make the
2245 			 * entire page presentable by zeroing invalid sections.
2246 			 */
2247 			if (m->valid != VM_PAGE_BITS_ALL)
2248 			    vm_page_zero_invalid(m, FALSE);
2249 		}
2250 	}
2251 	if (!gotreqpage) {
2252 		m = ap->a_m[ap->a_reqpage];
2253 		devfs_debug(DEVFS_DEBUG_WARNING,
2254 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2255 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2256 		devfs_debug(DEVFS_DEBUG_WARNING,
2257 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2258 		    size, bp->b_resid, ap->a_count, m->valid);
2259 		devfs_debug(DEVFS_DEBUG_WARNING,
2260 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2261 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2262 		/*
2263 		 * Free the buffer header back to the swap buffer pool.
2264 		 */
2265 		relpbuf(bp, NULL);
2266 		return VM_PAGER_ERROR;
2267 	}
2268 	/*
2269 	 * Free the buffer header back to the swap buffer pool.
2270 	 */
2271 	relpbuf(bp, NULL);
2272 	if (DEVFS_NODE(ap->a_vp))
2273 		vfs_timestamp(&DEVFS_NODE(ap->a_vp)->mtime);
2274 	return VM_PAGER_OK;
2275 }
2276 
2277 static __inline
2278 int
2279 sequential_heuristic(struct uio *uio, struct file *fp)
2280 {
2281 	/*
2282 	 * Sequential heuristic - detect sequential operation
2283 	 */
2284 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2285 	    uio->uio_offset == fp->f_nextoff) {
2286 		/*
2287 		 * XXX we assume that the filesystem block size is
2288 		 * the default.  Not true, but still gives us a pretty
2289 		 * good indicator of how sequential the read operations
2290 		 * are.
2291 		 */
2292 		int tmpseq = fp->f_seqcount;
2293 
2294 		tmpseq += howmany(uio->uio_resid, MAXBSIZE);
2295 		if (tmpseq > IO_SEQMAX)
2296 			tmpseq = IO_SEQMAX;
2297 		fp->f_seqcount = tmpseq;
2298 		return(fp->f_seqcount << IO_SEQSHIFT);
2299 	}
2300 
2301 	/*
2302 	 * Not sequential, quick draw-down of seqcount
2303 	 */
2304 	if (fp->f_seqcount > 1)
2305 		fp->f_seqcount = 1;
2306 	else
2307 		fp->f_seqcount = 0;
2308 	return(0);
2309 }
2310