xref: /dflybsd-src/sys/kern/vfs_vnops.c (revision d54592ee9e96c920b951af2e00cd72c0081ccae3)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
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 the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
39  * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $
40  * $DragonFly: src/sys/kern/vfs_vnops.c,v 1.58 2008/06/28 17:59:49 dillon Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/fcntl.h>
46 #include <sys/file.h>
47 #include <sys/stat.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/nlookup.h>
51 #include <sys/vnode.h>
52 #include <sys/buf.h>
53 #include <sys/filio.h>
54 #include <sys/ttycom.h>
55 #include <sys/conf.h>
56 #include <sys/syslog.h>
57 
58 static int vn_closefile (struct file *fp);
59 static int vn_ioctl (struct file *fp, u_long com, caddr_t data,
60 		struct ucred *cred);
61 static int vn_read (struct file *fp, struct uio *uio,
62 		struct ucred *cred, int flags);
63 static int svn_read (struct file *fp, struct uio *uio,
64 		struct ucred *cred, int flags);
65 static int vn_poll (struct file *fp, int events, struct ucred *cred);
66 static int vn_kqfilter (struct file *fp, struct knote *kn);
67 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred);
68 static int vn_write (struct file *fp, struct uio *uio,
69 		struct ucred *cred, int flags);
70 static int svn_write (struct file *fp, struct uio *uio,
71 		struct ucred *cred, int flags);
72 
73 struct fileops vnode_fileops = {
74 	.fo_read = vn_read,
75 	.fo_write = vn_write,
76 	.fo_ioctl = vn_ioctl,
77 	.fo_poll = vn_poll,
78 	.fo_kqfilter = vn_kqfilter,
79 	.fo_stat = vn_statfile,
80 	.fo_close = vn_closefile,
81 	.fo_shutdown = nofo_shutdown
82 };
83 
84 struct fileops specvnode_fileops = {
85 	.fo_read = svn_read,
86 	.fo_write = svn_write,
87 	.fo_ioctl = vn_ioctl,
88 	.fo_poll = vn_poll,
89 	.fo_kqfilter = vn_kqfilter,
90 	.fo_stat = vn_statfile,
91 	.fo_close = vn_closefile,
92 	.fo_shutdown = nofo_shutdown
93 };
94 
95 /*
96  * Shortcut the device read/write.  This avoids a lot of vnode junk.
97  * Basically the specfs vnops for read and write take the locked vnode,
98  * unlock it (because we can't hold the vnode locked while reading or writing
99  * a device which may block indefinitely), issues the device operation, then
100  * relock the vnode before returning, plus other junk.  This bypasses all
101  * of that and just does the device operation.
102  */
103 void
104 vn_setspecops(struct file *fp)
105 {
106 	if (vfs_fastdev && fp->f_ops == &vnode_fileops) {
107 		fp->f_ops = &specvnode_fileops;
108 	}
109 }
110 
111 /*
112  * Common code for vnode open operations.  Check permissions, and call
113  * the VOP_NOPEN or VOP_NCREATE routine.
114  *
115  * The caller is responsible for setting up nd with nlookup_init() and
116  * for cleaning it up with nlookup_done(), whether we return an error
117  * or not.
118  *
119  * On success nd->nl_open_vp will hold a referenced and, if requested,
120  * locked vnode.  A locked vnode is requested via NLC_LOCKVP.  If fp
121  * is non-NULL the vnode will be installed in the file pointer.
122  *
123  * NOTE: The vnode is referenced just once on return whether or not it
124  * is also installed in the file pointer.
125  */
126 int
127 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode)
128 {
129 	struct vnode *vp;
130 	struct ucred *cred = nd->nl_cred;
131 	struct vattr vat;
132 	struct vattr *vap = &vat;
133 	int mode, error;
134 
135 	/*
136 	 * Lookup the path and create or obtain the vnode.  After a
137 	 * successful lookup a locked nd->nl_nch will be returned.
138 	 *
139 	 * The result of this section should be a locked vnode.
140 	 *
141 	 * XXX with only a little work we should be able to avoid locking
142 	 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set.
143 	 */
144 	if (fmode & O_CREAT) {
145 		/*
146 		 * CONDITIONAL CREATE FILE CASE
147 		 *
148 		 * Setting NLC_CREATE causes a negative hit to store
149 		 * the negative hit ncp and not return an error.  Then
150 		 * nc_error or nc_vp may be checked to see if the ncp
151 		 * represents a negative hit.  NLC_CREATE also requires
152 		 * write permission on the governing directory or EPERM
153 		 * is returned.
154 		 */
155 		if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
156 			nd->nl_flags |= NLC_FOLLOW;
157 		nd->nl_flags |= NLC_CREATE;
158 		nd->nl_flags |= NLC_REFDVP;
159 		bwillinode(1);
160 		error = nlookup(nd);
161 	} else {
162 		/*
163 		 * NORMAL OPEN FILE CASE
164 		 */
165 		error = nlookup(nd);
166 	}
167 
168 	if (error)
169 		return (error);
170 
171 	/*
172 	 * split case to allow us to re-resolve and retry the ncp in case
173 	 * we get ESTALE.
174 	 */
175 again:
176 	if (fmode & O_CREAT) {
177 		if (nd->nl_nch.ncp->nc_vp == NULL) {
178 			if ((error = ncp_writechk(&nd->nl_nch)) != 0)
179 				return (error);
180 			VATTR_NULL(vap);
181 			vap->va_type = VREG;
182 			vap->va_mode = cmode;
183 			if (fmode & O_EXCL)
184 				vap->va_vaflags |= VA_EXCLUSIVE;
185 			error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp,
186 					    nd->nl_cred, vap);
187 			if (error)
188 				return (error);
189 			fmode &= ~O_TRUNC;
190 			/* locked vnode is returned */
191 		} else {
192 			if (fmode & O_EXCL) {
193 				error = EEXIST;
194 			} else {
195 				error = cache_vget(&nd->nl_nch, cred,
196 						    LK_EXCLUSIVE, &vp);
197 			}
198 			if (error)
199 				return (error);
200 			fmode &= ~O_CREAT;
201 		}
202 	} else {
203 		error = cache_vget(&nd->nl_nch, cred, LK_EXCLUSIVE, &vp);
204 		if (error)
205 			return (error);
206 	}
207 
208 	/*
209 	 * We have a locked vnode and ncp now.  Note that the ncp will
210 	 * be cleaned up by the caller if nd->nl_nch is left intact.
211 	 */
212 	if (vp->v_type == VLNK) {
213 		error = EMLINK;
214 		goto bad;
215 	}
216 	if (vp->v_type == VSOCK) {
217 		error = EOPNOTSUPP;
218 		goto bad;
219 	}
220 	if ((fmode & O_CREAT) == 0) {
221 		mode = 0;
222 		if (fmode & (FWRITE | O_TRUNC)) {
223 			if (vp->v_type == VDIR) {
224 				error = EISDIR;
225 				goto bad;
226 			}
227 			error = vn_writechk(vp, &nd->nl_nch);
228 			if (error) {
229 				/*
230 				 * Special stale handling, re-resolve the
231 				 * vnode.
232 				 */
233 				if (error == ESTALE) {
234 					vput(vp);
235 					vp = NULL;
236 					cache_setunresolved(&nd->nl_nch);
237 					error = cache_resolve(&nd->nl_nch, cred);
238 					if (error == 0)
239 						goto again;
240 				}
241 				goto bad;
242 			}
243 			mode |= VWRITE;
244 		}
245 		if (fmode & FREAD)
246 			mode |= VREAD;
247 		if (mode) {
248 		        error = VOP_ACCESS(vp, mode, cred);
249 			if (error) {
250 				/*
251 				 * Special stale handling, re-resolve the
252 				 * vnode.
253 				 */
254 				if (error == ESTALE) {
255 					vput(vp);
256 					vp = NULL;
257 					cache_setunresolved(&nd->nl_nch);
258 					error = cache_resolve(&nd->nl_nch, cred);
259 					if (error == 0)
260 						goto again;
261 				}
262 				goto bad;
263 			}
264 		}
265 	}
266 	if (fmode & O_TRUNC) {
267 		vn_unlock(vp);				/* XXX */
268 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
269 		VATTR_NULL(vap);
270 		vap->va_size = 0;
271 		error = VOP_SETATTR(vp, vap, cred);
272 		if (error)
273 			goto bad;
274 	}
275 
276 	/*
277 	 * Setup the fp so VOP_OPEN can override it.  No descriptor has been
278 	 * associated with the fp yet so we own it clean.
279 	 *
280 	 * f_nchandle inherits nl_nch.  This used to be necessary only for
281 	 * directories but now we do it unconditionally so f*() ops
282 	 * such as fchmod() can access the actual namespace that was
283 	 * used to open the file.
284 	 */
285 	if (fp) {
286 		fp->f_nchandle = nd->nl_nch;
287 		cache_zero(&nd->nl_nch);
288 		cache_unlock(&fp->f_nchandle);
289 	}
290 
291 	/*
292 	 * Get rid of nl_nch.  vn_open does not return it (it returns the
293 	 * vnode or the file pointer).  Note: we can't leave nl_nch locked
294 	 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g.
295 	 * on /dev/ttyd0
296 	 */
297 	if (nd->nl_nch.ncp)
298 		cache_put(&nd->nl_nch);
299 
300 	error = VOP_OPEN(vp, fmode, cred, fp);
301 	if (error) {
302 		/*
303 		 * setting f_ops to &badfileops will prevent the descriptor
304 		 * code from trying to close and release the vnode, since
305 		 * the open failed we do not want to call close.
306 		 */
307 		if (fp) {
308 			fp->f_data = NULL;
309 			fp->f_ops = &badfileops;
310 		}
311 		goto bad;
312 	}
313 
314 #if 0
315 	/*
316 	 * Assert that VREG files have been setup for vmio.
317 	 */
318 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
319 		("vn_open: regular file was not VMIO enabled!"));
320 #endif
321 
322 	/*
323 	 * Return the vnode.  XXX needs some cleaning up.  The vnode is
324 	 * only returned in the fp == NULL case.
325 	 */
326 	if (fp == NULL) {
327 		nd->nl_open_vp = vp;
328 		nd->nl_vp_fmode = fmode;
329 		if ((nd->nl_flags & NLC_LOCKVP) == 0)
330 			vn_unlock(vp);
331 	} else {
332 		vput(vp);
333 	}
334 	return (0);
335 bad:
336 	if (vp)
337 		vput(vp);
338 	return (error);
339 }
340 
341 int
342 vn_opendisk(const char *devname, int fmode, struct vnode **vpp)
343 {
344 	struct vnode *vp;
345 	int error;
346 
347 	if (strncmp(devname, "/dev/", 5) == 0)
348 		devname += 5;
349 	if ((vp = getsynthvnode(devname)) == NULL) {
350 		error = ENODEV;
351 	} else {
352 		error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL);
353 		vn_unlock(vp);
354 		if (error) {
355 			vrele(vp);
356 			vp = NULL;
357 		}
358 	}
359 	*vpp = vp;
360 	return (error);
361 }
362 
363 /*
364  * Check for write permissions on the specified vnode.  nch may be NULL.
365  */
366 int
367 vn_writechk(struct vnode *vp, struct nchandle *nch)
368 {
369 	/*
370 	 * If there's shared text associated with
371 	 * the vnode, try to free it up once.  If
372 	 * we fail, we can't allow writing.
373 	 */
374 	if (vp->v_flag & VTEXT)
375 		return (ETXTBSY);
376 
377 	/*
378 	 * If the vnode represents a regular file, check the mount
379 	 * point via the nch.  This may be a different mount point
380 	 * then the one embedded in the vnode (e.g. nullfs).
381 	 *
382 	 * We can still write to non-regular files (e.g. devices)
383 	 * via read-only mounts.
384 	 */
385 	if (nch && nch->ncp && vp->v_type == VREG)
386 		return (ncp_writechk(nch));
387 	return (0);
388 }
389 
390 /*
391  * Check whether the underlying mount is read-only.  The mount point
392  * referenced by the namecache may be different from the mount point
393  * used by the underlying vnode in the case of NULLFS, so a separate
394  * check is needed.
395  */
396 int
397 ncp_writechk(struct nchandle *nch)
398 {
399 	if (nch->mount && (nch->mount->mnt_flag & MNT_RDONLY))
400 		return (EROFS);
401 	return(0);
402 }
403 
404 /*
405  * Vnode close call
406  */
407 int
408 vn_close(struct vnode *vp, int flags)
409 {
410 	int error;
411 
412 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) == 0) {
413 		error = VOP_CLOSE(vp, flags);
414 		vn_unlock(vp);
415 	}
416 	vrele(vp);
417 	return (error);
418 }
419 
420 static __inline
421 int
422 sequential_heuristic(struct uio *uio, struct file *fp)
423 {
424 	/*
425 	 * Sequential heuristic - detect sequential operation
426 	 */
427 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
428 	    uio->uio_offset == fp->f_nextoff) {
429 		int tmpseq = fp->f_seqcount;
430 		/*
431 		 * XXX we assume that the filesystem block size is
432 		 * the default.  Not true, but still gives us a pretty
433 		 * good indicator of how sequential the read operations
434 		 * are.
435 		 */
436 		tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
437 		if (tmpseq > IO_SEQMAX)
438 			tmpseq = IO_SEQMAX;
439 		fp->f_seqcount = tmpseq;
440 		return(fp->f_seqcount << IO_SEQSHIFT);
441 	}
442 
443 	/*
444 	 * Not sequential, quick draw-down of seqcount
445 	 */
446 	if (fp->f_seqcount > 1)
447 		fp->f_seqcount = 1;
448 	else
449 		fp->f_seqcount = 0;
450 	return(0);
451 }
452 
453 /*
454  * Package up an I/O request on a vnode into a uio and do it.
455  */
456 int
457 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len,
458 	off_t offset, enum uio_seg segflg, int ioflg,
459 	struct ucred *cred, int *aresid)
460 {
461 	struct uio auio;
462 	struct iovec aiov;
463 	struct ccms_lock ccms_lock;
464 	int error;
465 
466 	if ((ioflg & IO_NODELOCKED) == 0)
467 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
468 	auio.uio_iov = &aiov;
469 	auio.uio_iovcnt = 1;
470 	aiov.iov_base = base;
471 	aiov.iov_len = len;
472 	auio.uio_resid = len;
473 	auio.uio_offset = offset;
474 	auio.uio_segflg = segflg;
475 	auio.uio_rw = rw;
476 	auio.uio_td = curthread;
477 	ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, &auio);
478 	if (rw == UIO_READ) {
479 		error = VOP_READ(vp, &auio, ioflg, cred);
480 	} else {
481 		error = VOP_WRITE(vp, &auio, ioflg, cred);
482 	}
483 	ccms_lock_put(&vp->v_ccms, &ccms_lock);
484 	if (aresid)
485 		*aresid = auio.uio_resid;
486 	else
487 		if (auio.uio_resid && error == 0)
488 			error = EIO;
489 	if ((ioflg & IO_NODELOCKED) == 0)
490 		vn_unlock(vp);
491 	return (error);
492 }
493 
494 /*
495  * Package up an I/O request on a vnode into a uio and do it.  The I/O
496  * request is split up into smaller chunks and we try to avoid saturating
497  * the buffer cache while potentially holding a vnode locked, so we
498  * check bwillwrite() before calling vn_rdwr().  We also call uio_yield()
499  * to give other processes a chance to lock the vnode (either other processes
500  * core'ing the same binary, or unrelated processes scanning the directory).
501  */
502 int
503 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len,
504 		 off_t offset, enum uio_seg segflg, int ioflg,
505 		 struct ucred *cred, int *aresid)
506 {
507 	int error = 0;
508 
509 	do {
510 		int chunk;
511 
512 		/*
513 		 * Force `offset' to a multiple of MAXBSIZE except possibly
514 		 * for the first chunk, so that filesystems only need to
515 		 * write full blocks except possibly for the first and last
516 		 * chunks.
517 		 */
518 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
519 
520 		if (chunk > len)
521 			chunk = len;
522 		if (vp->v_type == VREG) {
523 			switch(rw) {
524 			case UIO_READ:
525 				bwillread(chunk);
526 				break;
527 			case UIO_WRITE:
528 				bwillwrite(chunk);
529 				break;
530 			}
531 		}
532 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
533 			    ioflg, cred, aresid);
534 		len -= chunk;	/* aresid calc already includes length */
535 		if (error)
536 			break;
537 		offset += chunk;
538 		base += chunk;
539 		uio_yield();
540 	} while (len);
541 	if (aresid)
542 		*aresid += len;
543 	return (error);
544 }
545 
546 /*
547  * MPALMOSTSAFE - acquires mplock
548  */
549 static int
550 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
551 {
552 	struct ccms_lock ccms_lock;
553 	struct vnode *vp;
554 	int error, ioflag;
555 
556 	get_mplock();
557 	KASSERT(uio->uio_td == curthread,
558 		("uio_td %p is not td %p", uio->uio_td, curthread));
559 	vp = (struct vnode *)fp->f_data;
560 
561 	ioflag = 0;
562 	if (flags & O_FBLOCKING) {
563 		/* ioflag &= ~IO_NDELAY; */
564 	} else if (flags & O_FNONBLOCKING) {
565 		ioflag |= IO_NDELAY;
566 	} else if (fp->f_flag & FNONBLOCK) {
567 		ioflag |= IO_NDELAY;
568 	}
569 	if (flags & O_FBUFFERED) {
570 		/* ioflag &= ~IO_DIRECT; */
571 	} else if (flags & O_FUNBUFFERED) {
572 		ioflag |= IO_DIRECT;
573 	} else if (fp->f_flag & O_DIRECT) {
574 		ioflag |= IO_DIRECT;
575 	}
576 	vn_lock(vp, LK_SHARED | LK_RETRY);
577 	if ((flags & O_FOFFSET) == 0)
578 		uio->uio_offset = fp->f_offset;
579 	ioflag |= sequential_heuristic(uio, fp);
580 
581 	ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio);
582 	error = VOP_READ(vp, uio, ioflag, cred);
583 	ccms_lock_put(&vp->v_ccms, &ccms_lock);
584 	if ((flags & O_FOFFSET) == 0)
585 		fp->f_offset = uio->uio_offset;
586 	fp->f_nextoff = uio->uio_offset;
587 	vn_unlock(vp);
588 	rel_mplock();
589 	return (error);
590 }
591 
592 /*
593  * Device-optimized file table vnode read routine.
594  *
595  * This bypasses the VOP table and talks directly to the device.  Most
596  * filesystems just route to specfs and can make this optimization.
597  *
598  * MPALMOSTSAFE - acquires mplock
599  */
600 static int
601 svn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
602 {
603 	struct vnode *vp;
604 	int ioflag;
605 	int error;
606 	cdev_t dev;
607 
608 	get_mplock();
609 	KASSERT(uio->uio_td == curthread,
610 		("uio_td %p is not td %p", uio->uio_td, curthread));
611 
612 	vp = (struct vnode *)fp->f_data;
613 	if (vp == NULL || vp->v_type == VBAD) {
614 		error = EBADF;
615 		goto done;
616 	}
617 
618 	if ((dev = vp->v_rdev) == NULL) {
619 		error = EBADF;
620 		goto done;
621 	}
622 	reference_dev(dev);
623 
624 	if (uio->uio_resid == 0) {
625 		error = 0;
626 		goto done;
627 	}
628 	if ((flags & O_FOFFSET) == 0)
629 		uio->uio_offset = fp->f_offset;
630 
631 	ioflag = 0;
632 	if (flags & O_FBLOCKING) {
633 		/* ioflag &= ~IO_NDELAY; */
634 	} else if (flags & O_FNONBLOCKING) {
635 		ioflag |= IO_NDELAY;
636 	} else if (fp->f_flag & FNONBLOCK) {
637 		ioflag |= IO_NDELAY;
638 	}
639 	if (flags & O_FBUFFERED) {
640 		/* ioflag &= ~IO_DIRECT; */
641 	} else if (flags & O_FUNBUFFERED) {
642 		ioflag |= IO_DIRECT;
643 	} else if (fp->f_flag & O_DIRECT) {
644 		ioflag |= IO_DIRECT;
645 	}
646 	ioflag |= sequential_heuristic(uio, fp);
647 
648 	error = dev_dread(dev, uio, ioflag);
649 
650 	release_dev(dev);
651 	if ((flags & O_FOFFSET) == 0)
652 		fp->f_offset = uio->uio_offset;
653 	fp->f_nextoff = uio->uio_offset;
654 done:
655 	rel_mplock();
656 	return (error);
657 }
658 
659 /*
660  * MPALMOSTSAFE - acquires mplock
661  */
662 static int
663 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
664 {
665 	struct ccms_lock ccms_lock;
666 	struct vnode *vp;
667 	int error, ioflag;
668 
669 	get_mplock();
670 	KASSERT(uio->uio_td == curthread,
671 		("uio_td %p is not p %p", uio->uio_td, curthread));
672 	vp = (struct vnode *)fp->f_data;
673 #if 0
674 	/* VOP_WRITE should handle this now */
675 	if (vp->v_type == VREG || vp->v_type == VDATABASE)
676 		bwillwrite();
677 #endif
678 	vp = (struct vnode *)fp->f_data;	/* XXX needed? */
679 
680 	ioflag = IO_UNIT;
681 	if (vp->v_type == VREG &&
682 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
683 		ioflag |= IO_APPEND;
684 	}
685 
686 	if (flags & O_FBLOCKING) {
687 		/* ioflag &= ~IO_NDELAY; */
688 	} else if (flags & O_FNONBLOCKING) {
689 		ioflag |= IO_NDELAY;
690 	} else if (fp->f_flag & FNONBLOCK) {
691 		ioflag |= IO_NDELAY;
692 	}
693 	if (flags & O_FBUFFERED) {
694 		/* ioflag &= ~IO_DIRECT; */
695 	} else if (flags & O_FUNBUFFERED) {
696 		ioflag |= IO_DIRECT;
697 	} else if (fp->f_flag & O_DIRECT) {
698 		ioflag |= IO_DIRECT;
699 	}
700 	if (flags & O_FASYNCWRITE) {
701 		/* ioflag &= ~IO_SYNC; */
702 	} else if (flags & O_FSYNCWRITE) {
703 		ioflag |= IO_SYNC;
704 	} else if (fp->f_flag & O_FSYNC) {
705 		ioflag |= IO_SYNC;
706 	}
707 
708 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
709 		ioflag |= IO_SYNC;
710 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
711 	if ((flags & O_FOFFSET) == 0)
712 		uio->uio_offset = fp->f_offset;
713 	ioflag |= sequential_heuristic(uio, fp);
714 	ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio);
715 	error = VOP_WRITE(vp, uio, ioflag, cred);
716 	ccms_lock_put(&vp->v_ccms, &ccms_lock);
717 	if ((flags & O_FOFFSET) == 0)
718 		fp->f_offset = uio->uio_offset;
719 	fp->f_nextoff = uio->uio_offset;
720 	vn_unlock(vp);
721 	rel_mplock();
722 	return (error);
723 }
724 
725 /*
726  * Device-optimized file table vnode write routine.
727  *
728  * This bypasses the VOP table and talks directly to the device.  Most
729  * filesystems just route to specfs and can make this optimization.
730  *
731  * MPALMOSTSAFE - acquires mplock
732  */
733 static int
734 svn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
735 {
736 	struct vnode *vp;
737 	int ioflag;
738 	int error;
739 	cdev_t dev;
740 
741 	get_mplock();
742 	KASSERT(uio->uio_td == curthread,
743 		("uio_td %p is not p %p", uio->uio_td, curthread));
744 
745 	vp = (struct vnode *)fp->f_data;
746 	if (vp == NULL || vp->v_type == VBAD) {
747 		error = EBADF;
748 		goto done;
749 	}
750 	if (vp->v_type == VREG)
751 		bwillwrite(uio->uio_resid);
752 	vp = (struct vnode *)fp->f_data;	/* XXX needed? */
753 
754 	if ((dev = vp->v_rdev) == NULL) {
755 		error = EBADF;
756 		goto done;
757 	}
758 	reference_dev(dev);
759 
760 	if ((flags & O_FOFFSET) == 0)
761 		uio->uio_offset = fp->f_offset;
762 
763 	ioflag = IO_UNIT;
764 	if (vp->v_type == VREG &&
765 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
766 		ioflag |= IO_APPEND;
767 	}
768 
769 	if (flags & O_FBLOCKING) {
770 		/* ioflag &= ~IO_NDELAY; */
771 	} else if (flags & O_FNONBLOCKING) {
772 		ioflag |= IO_NDELAY;
773 	} else if (fp->f_flag & FNONBLOCK) {
774 		ioflag |= IO_NDELAY;
775 	}
776 	if (flags & O_FBUFFERED) {
777 		/* ioflag &= ~IO_DIRECT; */
778 	} else if (flags & O_FUNBUFFERED) {
779 		ioflag |= IO_DIRECT;
780 	} else if (fp->f_flag & O_DIRECT) {
781 		ioflag |= IO_DIRECT;
782 	}
783 	if (flags & O_FASYNCWRITE) {
784 		/* ioflag &= ~IO_SYNC; */
785 	} else if (flags & O_FSYNCWRITE) {
786 		ioflag |= IO_SYNC;
787 	} else if (fp->f_flag & O_FSYNC) {
788 		ioflag |= IO_SYNC;
789 	}
790 
791 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
792 		ioflag |= IO_SYNC;
793 	ioflag |= sequential_heuristic(uio, fp);
794 
795 	error = dev_dwrite(dev, uio, ioflag);
796 
797 	release_dev(dev);
798 	if ((flags & O_FOFFSET) == 0)
799 		fp->f_offset = uio->uio_offset;
800 	fp->f_nextoff = uio->uio_offset;
801 done:
802 	rel_mplock();
803 	return (error);
804 }
805 
806 /*
807  * MPALMOSTSAFE - acquires mplock
808  */
809 static int
810 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred)
811 {
812 	struct vnode *vp;
813 	int error;
814 
815 	get_mplock();
816 	vp = (struct vnode *)fp->f_data;
817 	error = vn_stat(vp, sb, cred);
818 	rel_mplock();
819 	return (error);
820 }
821 
822 int
823 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred)
824 {
825 	struct vattr vattr;
826 	struct vattr *vap;
827 	int error;
828 	u_short mode;
829 	cdev_t dev;
830 
831 	vap = &vattr;
832 	error = VOP_GETATTR(vp, vap);
833 	if (error)
834 		return (error);
835 
836 	/*
837 	 * Zero the spare stat fields
838 	 */
839 	sb->st_lspare = 0;
840 	sb->st_qspare = 0;
841 
842 	/*
843 	 * Copy from vattr table
844 	 */
845 	if (vap->va_fsid != VNOVAL)
846 		sb->st_dev = vap->va_fsid;
847 	else
848 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
849 	sb->st_ino = vap->va_fileid;
850 	mode = vap->va_mode;
851 	switch (vap->va_type) {
852 	case VREG:
853 		mode |= S_IFREG;
854 		break;
855 	case VDATABASE:
856 		mode |= S_IFDB;
857 		break;
858 	case VDIR:
859 		mode |= S_IFDIR;
860 		break;
861 	case VBLK:
862 		mode |= S_IFBLK;
863 		break;
864 	case VCHR:
865 		mode |= S_IFCHR;
866 		break;
867 	case VLNK:
868 		mode |= S_IFLNK;
869 		/* This is a cosmetic change, symlinks do not have a mode. */
870 		if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
871 			sb->st_mode &= ~ACCESSPERMS;	/* 0000 */
872 		else
873 			sb->st_mode |= ACCESSPERMS;	/* 0777 */
874 		break;
875 	case VSOCK:
876 		mode |= S_IFSOCK;
877 		break;
878 	case VFIFO:
879 		mode |= S_IFIFO;
880 		break;
881 	default:
882 		return (EBADF);
883 	};
884 	sb->st_mode = mode;
885 	if (vap->va_nlink > (nlink_t)-1)
886 		sb->st_nlink = (nlink_t)-1;
887 	else
888 		sb->st_nlink = vap->va_nlink;
889 	sb->st_uid = vap->va_uid;
890 	sb->st_gid = vap->va_gid;
891 	sb->st_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
892 	sb->st_size = vap->va_size;
893 	sb->st_atimespec = vap->va_atime;
894 	sb->st_mtimespec = vap->va_mtime;
895 	sb->st_ctimespec = vap->va_ctime;
896 
897 	/*
898 	 * A VCHR and VBLK device may track the last access and last modified
899 	 * time independantly of the filesystem.  This is particularly true
900 	 * because device read and write calls may bypass the filesystem.
901 	 */
902 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
903 		if ((dev = vp->v_rdev) != NULL) {
904 			if (dev->si_lastread) {
905 				sb->st_atimespec.tv_sec = dev->si_lastread;
906 				sb->st_atimespec.tv_nsec = 0;
907 			}
908 			if (dev->si_lastwrite) {
909 				sb->st_atimespec.tv_sec = dev->si_lastwrite;
910 				sb->st_atimespec.tv_nsec = 0;
911 			}
912 		}
913 	}
914 
915         /*
916 	 * According to www.opengroup.org, the meaning of st_blksize is
917 	 *   "a filesystem-specific preferred I/O block size for this
918 	 *    object.  In some filesystem types, this may vary from file
919 	 *    to file"
920 	 * Default to PAGE_SIZE after much discussion.
921 	 */
922 
923 	if (vap->va_type == VREG) {
924 		sb->st_blksize = vap->va_blocksize;
925 	} else if (vn_isdisk(vp, NULL)) {
926 		/*
927 		 * XXX this is broken.  If the device is not yet open (aka
928 		 * stat() call, aka v_rdev == NULL), how are we supposed
929 		 * to get a valid block size out of it?
930 		 */
931 		cdev_t dev;
932 
933 		if ((dev = vp->v_rdev) == NULL) {
934 			if (vp->v_type == VCHR)
935 				dev = get_dev(vp->v_umajor, vp->v_uminor);
936 		}
937 		sb->st_blksize = dev->si_bsize_best;
938 		if (sb->st_blksize < dev->si_bsize_phys)
939 			sb->st_blksize = dev->si_bsize_phys;
940 		if (sb->st_blksize < BLKDEV_IOSIZE)
941 			sb->st_blksize = BLKDEV_IOSIZE;
942 	} else {
943 		sb->st_blksize = PAGE_SIZE;
944 	}
945 
946 	sb->st_flags = vap->va_flags;
947 	if (suser_cred(cred, 0))
948 		sb->st_gen = 0;
949 	else
950 		sb->st_gen = (u_int32_t)vap->va_gen;
951 
952 #if (S_BLKSIZE == 512)
953 	/* Optimize this case */
954 	sb->st_blocks = vap->va_bytes >> 9;
955 #else
956 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
957 #endif
958 	sb->st_fsmid = vap->va_fsmid;
959 	return (0);
960 }
961 
962 /*
963  * MPALMOSTSAFE - acquires mplock
964  */
965 static int
966 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred)
967 {
968 	struct vnode *vp = ((struct vnode *)fp->f_data);
969 	struct vnode *ovp;
970 	struct vattr vattr;
971 	int error;
972 
973 	get_mplock();
974 
975 	switch (vp->v_type) {
976 	case VREG:
977 	case VDIR:
978 		if (com == FIONREAD) {
979 			if ((error = VOP_GETATTR(vp, &vattr)) != 0)
980 				break;
981 			*(int *)data = vattr.va_size - fp->f_offset;
982 			error = 0;
983 			break;
984 		}
985 		if (com == FIOASYNC) {				/* XXX */
986 			error = 0;				/* XXX */
987 			break;
988 		}
989 		/* fall into ... */
990 	default:
991 #if 0
992 		return (ENOTTY);
993 #endif
994 	case VFIFO:
995 	case VCHR:
996 	case VBLK:
997 		if (com == FIODTYPE) {
998 			if (vp->v_type != VCHR && vp->v_type != VBLK) {
999 				error = ENOTTY;
1000 				break;
1001 			}
1002 			*(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK;
1003 			error = 0;
1004 			break;
1005 		}
1006 		error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred);
1007 		if (error == 0 && com == TIOCSCTTY) {
1008 			struct proc *p = curthread->td_proc;
1009 			struct session *sess;
1010 
1011 			if (p == NULL) {
1012 				error = ENOTTY;
1013 				break;
1014 			}
1015 
1016 			sess = p->p_session;
1017 			/* Do nothing if reassigning same control tty */
1018 			if (sess->s_ttyvp == vp) {
1019 				error = 0;
1020 				break;
1021 			}
1022 
1023 			/* Get rid of reference to old control tty */
1024 			ovp = sess->s_ttyvp;
1025 			vref(vp);
1026 			sess->s_ttyvp = vp;
1027 			if (ovp)
1028 				vrele(ovp);
1029 		}
1030 		break;
1031 	}
1032 	rel_mplock();
1033 	return (error);
1034 }
1035 
1036 /*
1037  * MPALMOSTSAFE - acquires mplock
1038  */
1039 static int
1040 vn_poll(struct file *fp, int events, struct ucred *cred)
1041 {
1042 	int error;
1043 
1044 	get_mplock();
1045 	error = VOP_POLL(((struct vnode *)fp->f_data), events, cred);
1046 	rel_mplock();
1047 	return (error);
1048 }
1049 
1050 /*
1051  * Check that the vnode is still valid, and if so
1052  * acquire requested lock.
1053  */
1054 int
1055 #ifndef	DEBUG_LOCKS
1056 vn_lock(struct vnode *vp, int flags)
1057 #else
1058 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line)
1059 #endif
1060 {
1061 	int error;
1062 
1063 	do {
1064 #ifdef	DEBUG_LOCKS
1065 		vp->filename = filename;
1066 		vp->line = line;
1067 		error = debuglockmgr(&vp->v_lock, flags,
1068 				     "vn_lock", filename, line);
1069 #else
1070 		error = lockmgr(&vp->v_lock, flags);
1071 #endif
1072 		if (error == 0)
1073 			break;
1074 	} while (flags & LK_RETRY);
1075 
1076 	/*
1077 	 * Because we (had better!) have a ref on the vnode, once it
1078 	 * goes to VRECLAIMED state it will not be recycled until all
1079 	 * refs go away.  So we can just check the flag.
1080 	 */
1081 	if (error == 0 && (vp->v_flag & VRECLAIMED)) {
1082 		lockmgr(&vp->v_lock, LK_RELEASE);
1083 		error = ENOENT;
1084 	}
1085 	return (error);
1086 }
1087 
1088 void
1089 vn_unlock(struct vnode *vp)
1090 {
1091 	lockmgr(&vp->v_lock, LK_RELEASE);
1092 }
1093 
1094 int
1095 vn_islocked(struct vnode *vp)
1096 {
1097 	return (lockstatus(&vp->v_lock, curthread));
1098 }
1099 
1100 /*
1101  * MPALMOSTSAFE - acquires mplock
1102  */
1103 static int
1104 vn_closefile(struct file *fp)
1105 {
1106 	int error;
1107 
1108 	get_mplock();
1109 	fp->f_ops = &badfileops;
1110 	error = vn_close(((struct vnode *)fp->f_data), fp->f_flag);
1111 	rel_mplock();
1112 	return(error);
1113 }
1114 
1115 /*
1116  * MPALMOSTSAFE - acquires mplock
1117  */
1118 static int
1119 vn_kqfilter(struct file *fp, struct knote *kn)
1120 {
1121 	int error;
1122 
1123 	get_mplock();
1124 	error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn);
1125 	rel_mplock();
1126 	return (error);
1127 }
1128