xref: /dflybsd-src/sys/kern/vfs_vnops.c (revision c9733229451fac5faa53b1a016b01866eae75a1c)
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_vnops.c	8.2 (Berkeley) 1/21/94
35  * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/uio.h>
41 #include <sys/fcntl.h>
42 #include <sys/file.h>
43 #include <sys/stat.h>
44 #include <sys/proc.h>
45 #include <sys/caps.h>
46 #include <sys/mount.h>
47 #include <sys/nlookup.h>
48 #include <sys/vnode.h>
49 #include <sys/buf.h>
50 #include <sys/filio.h>
51 #include <sys/ttycom.h>
52 #include <sys/conf.h>
53 #include <sys/sysctl.h>
54 #include <sys/syslog.h>
55 #include <sys/spinlock.h>
56 #include <sys/spinlock2.h>
57 
58 #include <sys/mplock2.h>
59 
60 static int vn_closefile (struct file *fp);
61 static int vn_ioctl (struct file *fp, u_long com, caddr_t data,
62 		struct ucred *cred, struct sysmsg *msg);
63 static int vn_read (struct file *fp, struct uio *uio,
64 		struct ucred *cred, int flags);
65 static int vn_kqfilter (struct file *fp, struct knote *kn);
66 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred);
67 static int vn_write (struct file *fp, struct uio *uio,
68 		struct ucred *cred, int flags);
69 static int vn_seek (struct file *fp, off_t offset, int whence, off_t *res);
70 
71 struct fileops vnode_fileops = {
72 	.fo_read = vn_read,
73 	.fo_write = vn_write,
74 	.fo_ioctl = vn_ioctl,
75 	.fo_kqfilter = vn_kqfilter,
76 	.fo_stat = vn_statfile,
77 	.fo_close = vn_closefile,
78 	.fo_shutdown = nofo_shutdown,
79 	.fo_seek = vn_seek
80 };
81 
82 /*
83  * Common code for vnode open operations.  Check permissions, and call
84  * the VOP_NOPEN or VOP_NCREATE routine.
85  *
86  * The caller is responsible for setting up nd with nlookup_init() and
87  * for cleaning it up with nlookup_done(), whether we return an error
88  * or not.
89  *
90  * On success nd->nl_open_vp will hold a referenced and, if requested,
91  * locked vnode.  A locked vnode is requested via NLC_LOCKVP.  If fp
92  * is non-NULL the vnode will be installed in the file pointer.
93  *
94  * NOTE: If the caller wishes the namecache entry to be operated with
95  *	 a shared lock it must use NLC_SHAREDLOCK.  If NLC_LOCKVP is set
96  *	 then the vnode lock will also be shared.
97  *
98  * NOTE: The vnode is referenced just once on return whether or not it
99  *	 is also installed in the file pointer.
100  */
101 int
102 vn_open(struct nlookupdata *nd, struct file **fpp, int fmode, int cmode)
103 {
104 	struct file *fp = fpp ? *fpp : NULL;
105 	struct vnode *vp;
106 	struct ucred *cred = nd->nl_cred;
107 	struct vattr vat;
108 	struct vattr *vap = &vat;
109 	int error;
110 	int vpexcl;
111 	u_int flags;
112 	uint64_t osize;
113 	struct mount *mp;
114 
115 	/*
116 	 * Certain combinations are illegal
117 	 */
118 	if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC)
119 		return(EACCES);
120 
121 	/*
122 	 * Lookup the path and create or obtain the vnode.  After a
123 	 * successful lookup a locked nd->nl_nch will be returned.
124 	 *
125 	 * The result of this section should be a locked vnode.
126 	 *
127 	 * XXX with only a little work we should be able to avoid locking
128 	 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set.
129 	 */
130 	nd->nl_flags |= NLC_OPEN;
131 	if (fmode & O_APPEND)
132 		nd->nl_flags |= NLC_APPEND;
133 	if (fmode & O_TRUNC)
134 		nd->nl_flags |= NLC_TRUNCATE;
135 	if (fmode & FREAD)
136 		nd->nl_flags |= NLC_READ;
137 	if (fmode & FWRITE)
138 		nd->nl_flags |= NLC_WRITE;
139 	if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
140 		nd->nl_flags |= NLC_FOLLOW;
141 
142 	if (fmode & O_CREAT) {
143 		/*
144 		 * CONDITIONAL CREATE FILE CASE
145 		 *
146 		 * Setting NLC_CREATE causes a negative hit to store
147 		 * the negative hit ncp and not return an error.  Then
148 		 * nc_error or nc_vp may be checked to see if the ncp
149 		 * represents a negative hit.  NLC_CREATE also requires
150 		 * write permission on the governing directory or EPERM
151 		 * is returned.
152 		 *
153 		 * If the file exists but is missing write permission,
154 		 * nlookup() returns EACCES. This has to be handled specially
155 		 * when combined with O_EXCL.
156 		 */
157 		nd->nl_flags |= NLC_CREATE;
158 		nd->nl_flags |= NLC_REFDVP;
159 		bwillinode(1);
160 		error = nlookup(nd);
161 		if (error == EACCES && nd->nl_nch.ncp->nc_vp != NULL &&
162 		    (fmode & O_EXCL) && !nd->nl_dir_error)
163 		{
164 			error = EEXIST;
165 		}
166 
167 		/*
168 		 * If no error and nd->nl_dvp is NULL, the nlookup represents
169 		 * a mount-point or cross-mount situation.  e.g.
170 		 * open("/var/cache", O_CREAT), where /var/cache is a
171 		 * mount point or a null-mount point.
172 		 */
173 		if (error == 0 && nd->nl_dvp == NULL)
174 			error = EINVAL;
175 	} else {
176 		/*
177 		 * NORMAL OPEN FILE CASE
178 		 */
179 		error = nlookup(nd);
180 	}
181 
182 	if (error)
183 		return (error);
184 
185 	/*
186 	 * split case to allow us to re-resolve and retry the ncp in case
187 	 * we get ESTALE.
188 	 *
189 	 * (error is 0 on entry / retry)
190 	 */
191 again:
192 	/*
193 	 * Checks for (likely) filesystem-modifying cases and allows
194 	 * the filesystem to stall the front-end.
195 	 */
196 	if ((fmode & (FWRITE | O_TRUNC)) ||
197 	    ((fmode & O_CREAT) && nd->nl_nch.ncp->nc_vp == NULL)) {
198 		error = ncp_writechk(&nd->nl_nch);
199 		if (error)
200 			return error;
201 	}
202 
203 	vpexcl = 1;
204 	if (fmode & O_CREAT) {
205 		if (nd->nl_nch.ncp->nc_vp == NULL) {
206 			VATTR_NULL(vap);
207 			vap->va_type = VREG;
208 			vap->va_mode = cmode;
209 			vap->va_fuseflags = fmode; /* FUSE */
210 			if (fmode & O_EXCL)
211 				vap->va_vaflags |= VA_EXCLUSIVE;
212 			error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp,
213 					    nd->nl_cred, vap);
214 			if (error)
215 				return (error);
216 			fmode &= ~O_TRUNC;
217 			/* locked vnode is returned */
218 		} else {
219 			if (fmode & O_EXCL) {
220 				error = EEXIST;
221 			} else {
222 				error = cache_vget(&nd->nl_nch, cred,
223 						    LK_EXCLUSIVE, &vp);
224 			}
225 			if (error)
226 				return (error);
227 			fmode &= ~O_CREAT;
228 		}
229 	} else {
230 		/*
231 		 * In most other cases a shared lock on the vnode is
232 		 * sufficient.  However, the O_RDWR case needs an
233 		 * exclusive lock if the vnode is executable.  The
234 		 * NLC_EXCLLOCK_IFEXEC and NCF_NOTX flags help resolve
235 		 * this.
236 		 *
237 		 * NOTE: If NCF_NOTX is not set, we do not know the
238 		 *	 the state of the 'x' bits and have to get
239 		 *	 an exclusive lock for the EXCLLOCK_IFEXEC case.
240 		 */
241 		if ((nd->nl_flags & NLC_SHAREDLOCK) &&
242 		    ((nd->nl_flags & NLC_EXCLLOCK_IFEXEC) == 0 ||
243 		     nd->nl_nch.ncp->nc_flag & NCF_NOTX)) {
244 			error = cache_vget(&nd->nl_nch, cred, LK_SHARED, &vp);
245 			vpexcl = 0;
246 		} else {
247 			error = cache_vget(&nd->nl_nch, cred,
248 					   LK_EXCLUSIVE, &vp);
249 		}
250 		if (error)
251 			return (error);
252 	}
253 
254 	/*
255 	 * We have a locked vnode and ncp now.  Note that the ncp will
256 	 * be cleaned up by the caller if nd->nl_nch is left intact.
257 	 */
258 	if (vp->v_type == VLNK) {
259 		error = EMLINK;
260 		goto bad;
261 	}
262 	if (vp->v_type == VSOCK) {
263 		error = EOPNOTSUPP;
264 		goto bad;
265 	}
266 	if (vp->v_type != VDIR && (fmode & O_DIRECTORY)) {
267 		error = ENOTDIR;
268 		goto bad;
269 	}
270 	if ((fmode & O_CREAT) == 0) {
271 		if (fmode & (FWRITE | O_TRUNC)) {
272 			if (vp->v_type == VDIR) {
273 				error = EISDIR;
274 				goto bad;
275 			}
276 
277 			/*
278 			 * Additional checks on vnode (does not substitute
279 			 * for ncp_writechk()).
280 			 */
281 			error = vn_writechk(vp);
282 			if (error) {
283 				/*
284 				 * Special stale handling, re-resolve the
285 				 * vnode.
286 				 */
287 				if (error == ESTALE) {
288 					u_int dummy_gen = 0;
289 
290 					vput(vp);
291 					vp = NULL;
292 					if (vpexcl == 0) {
293 						cache_unlock(&nd->nl_nch);
294 						cache_lock(&nd->nl_nch);
295 					}
296 					cache_setunresolved(&nd->nl_nch);
297 					error = cache_resolve(&nd->nl_nch,
298 							      &dummy_gen,
299 							      cred);
300 					if (error == 0)
301 						goto again;
302 				}
303 				goto bad;
304 			}
305 		}
306 	}
307 	if (fmode & O_TRUNC) {
308 		vn_unlock(vp);				/* XXX */
309 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
310 		osize = vp->v_filesize;
311 		VATTR_NULL(vap);
312 		vap->va_size = 0;
313 		error = VOP_SETATTR_FP(vp, vap, cred, fp);
314 		if (error)
315 			goto bad;
316 		error = VOP_GETATTR(vp, vap);
317 		if (error)
318 			goto bad;
319 		mp = vq_vptomp(vp);
320 		VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize);
321 	}
322 
323 	/*
324 	 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag.
325 	 * These particular bits a tracked all the way from the root.
326 	 *
327 	 * NOTE: Might not work properly on NFS servers due to the
328 	 * disconnected namecache.
329 	 */
330 	flags = nd->nl_nch.ncp->nc_flag;
331 	if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) &&
332 	    (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) {
333 		vsetflags(vp, VSWAPCACHE);
334 	} else {
335 		vclrflags(vp, VSWAPCACHE);
336 	}
337 
338 	/*
339 	 * Setup the fp so VOP_OPEN can override it.  No descriptor has been
340 	 * associated with the fp yet so we own it clean.
341 	 *
342 	 * f_nchandle inherits nl_nch.  This used to be necessary only for
343 	 * directories but now we do it unconditionally so f*() ops
344 	 * such as fchmod() can access the actual namespace that was
345 	 * used to open the file.
346 	 */
347 	if (fp) {
348 		if (nd->nl_flags & NLC_APPENDONLY)
349 			fmode |= FAPPENDONLY;
350 		fp->f_nchandle = nd->nl_nch;
351 		cache_zero(&nd->nl_nch);
352 		cache_unlock(&fp->f_nchandle);
353 	}
354 
355 	/*
356 	 * Get rid of nl_nch.  vn_open does not return it (it returns the
357 	 * vnode or the file pointer).
358 	 *
359 	 * NOTE: We can't leave nl_nch locked through the VOP_OPEN anyway
360 	 *	 since the VOP_OPEN may block, e.g. on /dev/ttyd0
361 	 *
362 	 * NOTE: The VOP_OPEN() can replace the *fpp we supply with its own
363 	 *	 (it will fdrop/fhold), and can also set the *fpp up however
364 	 *	 it wants, not necessarily using DTYPE_VNODE.
365 	 */
366 	if (nd->nl_nch.ncp)
367 		cache_put(&nd->nl_nch);
368 
369 	error = VOP_OPEN(vp, fmode, cred, fpp);
370 	fp = fpp ? *fpp : NULL;
371 
372 	if (error) {
373 		/*
374 		 * setting f_ops to &badfileops will prevent the descriptor
375 		 * code from trying to close and release the vnode, since
376 		 * the open failed we do not want to call close.
377 		 */
378 		if (fp) {
379 			fp->f_data = NULL;
380 			fp->f_ops = &badfileops;
381 		}
382 		goto bad;
383 	}
384 
385 #if 0
386 	/*
387 	 * Assert that VREG files have been setup for vmio.
388 	 */
389 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
390 		("vn_open: regular file was not VMIO enabled!"));
391 #endif
392 
393 	/*
394 	 * Return the vnode.  XXX needs some cleaning up.  The vnode is
395 	 * only returned in the fp == NULL case.
396 	 *
397 	 * NOTE: vnode stored in fp may be different
398 	 */
399 	if (fp == NULL) {
400 		nd->nl_open_vp = vp;
401 		nd->nl_vp_fmode = fmode;
402 		if ((nd->nl_flags & NLC_LOCKVP) == 0)
403 			vn_unlock(vp);
404 	} else {
405 		vput(vp);
406 	}
407 	return (0);
408 bad:
409 	if (vp)
410 		vput(vp);
411 	return (error);
412 }
413 
414 int
415 vn_opendisk(const char *devname, int fmode, struct vnode **vpp)
416 {
417 	struct vnode *vp;
418 	int error;
419 
420 	if (strncmp(devname, "/dev/", 5) == 0)
421 		devname += 5;
422 	if ((vp = getsynthvnode(devname)) == NULL) {
423 		error = ENODEV;
424 	} else {
425 		error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL);
426 		vn_unlock(vp);
427 		if (error) {
428 			vrele(vp);
429 			vp = NULL;
430 		}
431 	}
432 	*vpp = vp;
433 	return (error);
434 }
435 
436 /*
437  * Checks for special conditions on the vnode which might prevent writing
438  * after the vnode has (likely) been locked.  The vnode might or might not
439  * be locked as of this call, but will be at least referenced.
440  *
441  * Also re-checks the mount RDONLY flag that ncp_writechk() checked prior
442  * to the vnode being locked.
443  */
444 int
445 vn_writechk(struct vnode *vp)
446 {
447 	/*
448 	 * If there's shared text associated with
449 	 * the vnode, try to free it up once.  If
450 	 * we fail, we can't allow writing.
451 	 */
452 	if (vp->v_flag & VTEXT)
453 		return (ETXTBSY);
454 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_RDONLY))
455 		return (EROFS);
456 	return 0;
457 }
458 
459 /*
460  * Check whether the underlying mount is read-only.  The mount point
461  * referenced by the namecache may be different from the mount point
462  * used by the underlying vnode in the case of NULLFS, so a separate
463  * check is needed.
464  *
465  * Must be called PRIOR to any vnodes being locked.
466  */
467 int
468 ncp_writechk(struct nchandle *nch)
469 {
470 	struct mount *mp;
471 
472 	if ((mp = nch->mount) != NULL) {
473 		if (mp->mnt_flag & MNT_RDONLY)
474 			return (EROFS);
475 		if (mp->mnt_op->vfs_modifying != vfs_stdmodifying)
476 			VFS_MODIFYING(mp);
477 	}
478 	return(0);
479 }
480 
481 /*
482  * Vnode close call
483  *
484  * MPSAFE
485  */
486 int
487 vn_close(struct vnode *vp, int flags, struct file *fp)
488 {
489 	int error;
490 
491 	error = vn_lock(vp, LK_SHARED | LK_RETRY | LK_FAILRECLAIM);
492 	if (error == 0) {
493 		error = VOP_CLOSE(vp, flags, fp);
494 		vn_unlock(vp);
495 	}
496 	vrele(vp);
497 	return (error);
498 }
499 
500 /*
501  * Sequential heuristic.
502  *
503  * MPSAFE (f_seqcount and f_nextoff are allowed to race)
504  */
505 static __inline
506 int
507 sequential_heuristic(struct uio *uio, struct file *fp)
508 {
509 	/*
510 	 * Sequential heuristic - detect sequential operation
511 	 *
512 	 * NOTE: SMP: We allow f_seqcount updates to race.
513 	 */
514 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
515 	    uio->uio_offset == fp->f_nextoff) {
516 		int tmpseq = fp->f_seqcount;
517 
518 		tmpseq += howmany(uio->uio_resid, MAXBSIZE);
519 		if (tmpseq > IO_SEQMAX)
520 			tmpseq = IO_SEQMAX;
521 		fp->f_seqcount = tmpseq;
522 		return(fp->f_seqcount << IO_SEQSHIFT);
523 	}
524 
525 	/*
526 	 * Not sequential, quick draw-down of seqcount
527 	 *
528 	 * NOTE: SMP: We allow f_seqcount updates to race.
529 	 */
530 	if (fp->f_seqcount > 1)
531 		fp->f_seqcount = 1;
532 	else
533 		fp->f_seqcount = 0;
534 	return(0);
535 }
536 
537 /*
538  * get - lock and return the f_offset field.
539  * set - set and unlock the f_offset field.
540  *
541  * These routines serve the dual purpose of serializing access to the
542  * f_offset field (at least on x86) and guaranteeing operational integrity
543  * when multiple read()ers and write()ers are present on the same fp.
544  *
545  * MPSAFE
546  */
547 static __inline off_t
548 vn_get_fpf_offset(struct file *fp)
549 {
550 	u_int	flags;
551 	u_int	nflags;
552 
553 	/*
554 	 * Shortcut critical path.
555 	 */
556 	flags = fp->f_flag & ~FOFFSETLOCK;
557 	if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK))
558 		return(fp->f_offset);
559 
560 	/*
561 	 * The hard way
562 	 */
563 	for (;;) {
564 		flags = fp->f_flag;
565 		if (flags & FOFFSETLOCK) {
566 			nflags = flags | FOFFSETWAKE;
567 			tsleep_interlock(&fp->f_flag, 0);
568 			if (atomic_cmpset_int(&fp->f_flag, flags, nflags))
569 				tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0);
570 		} else {
571 			nflags = flags | FOFFSETLOCK;
572 			if (atomic_cmpset_int(&fp->f_flag, flags, nflags))
573 				break;
574 		}
575 	}
576 	return(fp->f_offset);
577 }
578 
579 /*
580  * MPSAFE
581  */
582 static __inline void
583 vn_set_fpf_offset(struct file *fp, off_t offset)
584 {
585 	u_int	flags;
586 	u_int	nflags;
587 
588 	/*
589 	 * We hold the lock so we can set the offset without interference.
590 	 */
591 	fp->f_offset = offset;
592 
593 	/*
594 	 * Normal release is already a reasonably critical path.
595 	 */
596 	for (;;) {
597 		flags = fp->f_flag;
598 		nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE);
599 		if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) {
600 			if (flags & FOFFSETWAKE)
601 				wakeup(&fp->f_flag);
602 			break;
603 		}
604 	}
605 }
606 
607 /*
608  * MPSAFE
609  */
610 static __inline off_t
611 vn_poll_fpf_offset(struct file *fp)
612 {
613 #if defined(__x86_64__)
614 	return(fp->f_offset);
615 #else
616 	off_t off = vn_get_fpf_offset(fp);
617 	vn_set_fpf_offset(fp, off);
618 	return(off);
619 #endif
620 }
621 
622 /*
623  * Package up an I/O request on a vnode into a uio and do it.
624  *
625  * MPSAFE
626  */
627 int
628 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len,
629 	off_t offset, enum uio_seg segflg, int ioflg,
630 	struct ucred *cred, int *aresid)
631 {
632 	struct uio auio;
633 	struct iovec aiov;
634 	int error;
635 
636 	if ((ioflg & IO_NODELOCKED) == 0)
637 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
638 	auio.uio_iov = &aiov;
639 	auio.uio_iovcnt = 1;
640 	aiov.iov_base = base;
641 	aiov.iov_len = len;
642 	auio.uio_resid = len;
643 	auio.uio_offset = offset;
644 	auio.uio_segflg = segflg;
645 	auio.uio_rw = rw;
646 	auio.uio_td = curthread;
647 	if (rw == UIO_READ) {
648 		error = VOP_READ(vp, &auio, ioflg, cred);
649 	} else {
650 		error = VOP_WRITE(vp, &auio, ioflg, cred);
651 	}
652 	if (aresid)
653 		*aresid = auio.uio_resid;
654 	else
655 		if (auio.uio_resid && error == 0)
656 			error = EIO;
657 	if ((ioflg & IO_NODELOCKED) == 0)
658 		vn_unlock(vp);
659 	return (error);
660 }
661 
662 /*
663  * Package up an I/O request on a vnode into a uio and do it.  The I/O
664  * request is split up into smaller chunks and we try to avoid saturating
665  * the buffer cache while potentially holding a vnode locked, so we
666  * check bwillwrite() before calling vn_rdwr().  We also call lwkt_user_yield()
667  * to give other processes a chance to lock the vnode (either other processes
668  * core'ing the same binary, or unrelated processes scanning the directory).
669  *
670  * MPSAFE
671  */
672 int
673 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len,
674 		 off_t offset, enum uio_seg segflg, int ioflg,
675 		 struct ucred *cred, int *aresid)
676 {
677 	int error = 0;
678 
679 	do {
680 		int chunk;
681 
682 		/*
683 		 * Force `offset' to a multiple of MAXBSIZE except possibly
684 		 * for the first chunk, so that filesystems only need to
685 		 * write full blocks except possibly for the first and last
686 		 * chunks.
687 		 */
688 		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
689 
690 		if (chunk > len)
691 			chunk = len;
692 		if (vp->v_type == VREG && (ioflg & IO_RECURSE) == 0) {
693 			switch(rw) {
694 			case UIO_READ:
695 				bwillread(chunk);
696 				break;
697 			case UIO_WRITE:
698 				bwillwrite(chunk);
699 				break;
700 			}
701 		}
702 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
703 				ioflg, cred, aresid);
704 		len -= chunk;	/* aresid calc already includes length */
705 		if (error)
706 			break;
707 		offset += chunk;
708 		base += chunk;
709 		lwkt_user_yield();
710 	} while (len);
711 	if (aresid)
712 		*aresid += len;
713 	return (error);
714 }
715 
716 /*
717  * File pointers can no longer get ripped up by revoke so
718  * we don't need to lock access to the vp.
719  *
720  * f_offset updates are not guaranteed against multiple readers
721  */
722 static int
723 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
724 {
725 	struct vnode *vp;
726 	int error, ioflag;
727 
728 	KASSERT(uio->uio_td == curthread,
729 		("uio_td %p is not td %p", uio->uio_td, curthread));
730 	vp = (struct vnode *)fp->f_data;
731 
732 	ioflag = 0;
733 	if (flags & O_FBLOCKING) {
734 		/* ioflag &= ~IO_NDELAY; */
735 	} else if (flags & O_FNONBLOCKING) {
736 		ioflag |= IO_NDELAY;
737 	} else if (fp->f_flag & FNONBLOCK) {
738 		ioflag |= IO_NDELAY;
739 	}
740 	if (fp->f_flag & O_DIRECT) {
741 		ioflag |= IO_DIRECT;
742 	}
743 	if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0)
744 		uio->uio_offset = vn_get_fpf_offset(fp);
745 	vn_lock(vp, LK_SHARED | LK_RETRY);
746 	ioflag |= sequential_heuristic(uio, fp);
747 
748 	error = VOP_READ_FP(vp, uio, ioflag, cred, fp);
749 	fp->f_nextoff = uio->uio_offset;
750 	vn_unlock(vp);
751 	if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0)
752 		vn_set_fpf_offset(fp, uio->uio_offset);
753 	return (error);
754 }
755 
756 /*
757  * MPSAFE
758  */
759 static int
760 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
761 {
762 	struct vnode *vp;
763 	int error, ioflag;
764 
765 	KASSERT(uio->uio_td == curthread,
766 		("uio_td %p is not p %p", uio->uio_td, curthread));
767 	vp = (struct vnode *)fp->f_data;
768 
769 	ioflag = IO_UNIT;
770 	if (vp->v_type == VREG &&
771 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
772 		ioflag |= IO_APPEND;
773 	}
774 
775 	if (flags & O_FBLOCKING) {
776 		/* ioflag &= ~IO_NDELAY; */
777 	} else if (flags & O_FNONBLOCKING) {
778 		ioflag |= IO_NDELAY;
779 	} else if (fp->f_flag & FNONBLOCK) {
780 		ioflag |= IO_NDELAY;
781 	}
782 	if (fp->f_flag & O_DIRECT) {
783 		ioflag |= IO_DIRECT;
784 	}
785 	if (flags & O_FASYNCWRITE) {
786 		/* ioflag &= ~IO_SYNC; */
787 	} else if (flags & O_FSYNCWRITE) {
788 		ioflag |= IO_SYNC;
789 	} else if (fp->f_flag & O_FSYNC) {
790 		ioflag |= IO_SYNC;
791 	}
792 
793 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
794 		ioflag |= IO_SYNC;
795 	if ((flags & O_FOFFSET) == 0)
796 		uio->uio_offset = vn_get_fpf_offset(fp);
797 	if (vp->v_mount)
798 		VFS_MODIFYING(vp->v_mount);
799 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
800 	ioflag |= sequential_heuristic(uio, fp);
801 	error = VOP_WRITE_FP(vp, uio, ioflag, cred, fp);
802 	fp->f_nextoff = uio->uio_offset;
803 	vn_unlock(vp);
804 	if ((flags & O_FOFFSET) == 0)
805 		vn_set_fpf_offset(fp, uio->uio_offset);
806 	return (error);
807 }
808 
809 /*
810  * MPSAFE
811  */
812 static int
813 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred)
814 {
815 	struct vnode *vp;
816 	int error;
817 
818 	vp = (struct vnode *)fp->f_data;
819 	error = vn_stat(vp, sb, cred);
820 	return (error);
821 }
822 
823 /*
824  * MPSAFE
825  */
826 int
827 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred)
828 {
829 	struct vattr vattr;
830 	struct vattr *vap;
831 	int error;
832 	u_short mode;
833 	cdev_t dev;
834 
835 	/*
836 	 * vp already has a ref and is validated, can call unlocked.
837 	 */
838 	vap = &vattr;
839 	error = VOP_GETATTR(vp, vap);
840 	if (error)
841 		return (error);
842 
843 	/*
844 	 * Zero the spare stat fields
845 	 */
846 	sb->st_lspare = 0;
847 	sb->st_qspare2 = 0;
848 
849 	/*
850 	 * Copy from vattr table
851 	 */
852 	if (vap->va_fsid != VNOVAL)
853 		sb->st_dev = vap->va_fsid;
854 	else
855 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
856 	sb->st_ino = vap->va_fileid;
857 	mode = vap->va_mode;
858 	switch (vap->va_type) {
859 	case VREG:
860 		mode |= S_IFREG;
861 		break;
862 	case VDATABASE:
863 		mode |= S_IFDB;
864 		break;
865 	case VDIR:
866 		mode |= S_IFDIR;
867 		break;
868 	case VBLK:
869 		mode |= S_IFBLK;
870 		break;
871 	case VCHR:
872 		mode |= S_IFCHR;
873 		break;
874 	case VLNK:
875 		mode |= S_IFLNK;
876 		/* This is a cosmetic change, symlinks do not have a mode. */
877 		if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
878 			sb->st_mode &= ~ACCESSPERMS;	/* 0000 */
879 		else
880 			sb->st_mode |= ACCESSPERMS;	/* 0777 */
881 		break;
882 	case VSOCK:
883 		mode |= S_IFSOCK;
884 		break;
885 	case VFIFO:
886 		mode |= S_IFIFO;
887 		break;
888 	default:
889 		return (EBADF);
890 	}
891 	sb->st_mode = mode;
892 	if (vap->va_nlink > (nlink_t)-1)
893 		sb->st_nlink = (nlink_t)-1;
894 	else
895 		sb->st_nlink = vap->va_nlink;
896 	sb->st_uid = vap->va_uid;
897 	sb->st_gid = vap->va_gid;
898 	sb->st_rdev = devid_from_dev(vp->v_rdev);
899 	sb->st_size = vap->va_size;
900 	sb->st_atimespec = vap->va_atime;
901 	sb->st_mtimespec = vap->va_mtime;
902 	sb->st_ctimespec = vap->va_ctime;
903 
904 	/*
905 	 * A VCHR and VBLK device may track the last access and last modified
906 	 * time independantly of the filesystem.  This is particularly true
907 	 * because device read and write calls may bypass the filesystem.
908 	 */
909 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
910 		dev = vp->v_rdev;
911 		if (dev != NULL) {
912 			if (dev->si_lastread) {
913 				sb->st_atimespec.tv_sec = time_second +
914 							  (dev->si_lastread -
915 							   time_uptime);
916 				sb->st_atimespec.tv_nsec = 0;
917 			}
918 			if (dev->si_lastwrite) {
919 				sb->st_mtimespec.tv_sec = time_second +
920 							  (dev->si_lastwrite -
921 							   time_uptime);
922 				sb->st_mtimespec.tv_nsec = 0;
923 			}
924 		}
925 	}
926 
927         /*
928 	 * According to www.opengroup.org, the meaning of st_blksize is
929 	 *   "a filesystem-specific preferred I/O block size for this
930 	 *    object.  In some filesystem types, this may vary from file
931 	 *    to file"
932 	 * Default to PAGE_SIZE after much discussion.
933 	 */
934 
935 	if (vap->va_type == VREG) {
936 		sb->st_blksize = vap->va_blocksize;
937 	} else if (vn_isdisk(vp, NULL)) {
938 		/*
939 		 * XXX this is broken.  If the device is not yet open (aka
940 		 * stat() call, aka v_rdev == NULL), how are we supposed
941 		 * to get a valid block size out of it?
942 		 */
943 		dev = vp->v_rdev;
944 
945 		sb->st_blksize = dev->si_bsize_best;
946 		if (sb->st_blksize < dev->si_bsize_phys)
947 			sb->st_blksize = dev->si_bsize_phys;
948 		if (sb->st_blksize < BLKDEV_IOSIZE)
949 			sb->st_blksize = BLKDEV_IOSIZE;
950 	} else {
951 		sb->st_blksize = PAGE_SIZE;
952 	}
953 
954 	sb->st_flags = vap->va_flags;
955 
956 	error = caps_priv_check(cred, SYSCAP_NOVFS_GENERATION);
957 	if (error)
958 		sb->st_gen = 0;
959 	else
960 		sb->st_gen = (u_int32_t)vap->va_gen;
961 
962 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
963 
964 	/*
965 	 * This is for ABI compatibility <= 5.7 (for ABI change made in
966 	 * 5.7 master).
967 	 */
968 	sb->__old_st_blksize = sb->st_blksize;
969 
970 	return (0);
971 }
972 
973 /*
974  * MPALMOSTSAFE - acquires mplock
975  */
976 static int
977 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred,
978 	 struct sysmsg *msg)
979 {
980 	struct vnode *vp = ((struct vnode *)fp->f_data);
981 	struct vnode *ovp;
982 	struct vattr vattr;
983 	int error;
984 	off_t size;
985 
986 	switch (vp->v_type) {
987 	case VREG:
988 	case VDIR:
989 		if (com == FIONREAD) {
990 			error = VOP_GETATTR(vp, &vattr);
991 			if (error)
992 				break;
993 			size = vattr.va_size;
994 			if ((vp->v_flag & VNOTSEEKABLE) == 0)
995 				size -= vn_poll_fpf_offset(fp);
996 			if (size > 0x7FFFFFFF)
997 				size = 0x7FFFFFFF;
998 			*(int *)data = size;
999 			error = 0;
1000 			break;
1001 		}
1002 		if (com == FIOASYNC) {				/* XXX */
1003 			error = 0;				/* XXX */
1004 			break;
1005 		}
1006 		/* fall into ... */
1007 	default:
1008 #if 0
1009 		return (ENOTTY);
1010 #endif
1011 	case VFIFO:
1012 	case VCHR:
1013 	case VBLK:
1014 		if (com == FIODTYPE) {
1015 			if (vp->v_type != VCHR && vp->v_type != VBLK) {
1016 				error = ENOTTY;
1017 				break;
1018 			}
1019 			*(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK;
1020 			error = 0;
1021 			break;
1022 		}
1023 		error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg);
1024 		if (error == 0 && com == TIOCSCTTY) {
1025 			struct proc *p = curthread->td_proc;
1026 			struct session *sess;
1027 
1028 			if (p == NULL) {
1029 				error = ENOTTY;
1030 				break;
1031 			}
1032 
1033 			get_mplock();
1034 			sess = p->p_session;
1035 			/* Do nothing if reassigning same control tty */
1036 			if (sess->s_ttyvp == vp) {
1037 				error = 0;
1038 				rel_mplock();
1039 				break;
1040 			}
1041 
1042 			/* Get rid of reference to old control tty */
1043 			ovp = sess->s_ttyvp;
1044 			vref(vp);
1045 			sess->s_ttyvp = vp;
1046 			if (ovp)
1047 				vrele(ovp);
1048 			rel_mplock();
1049 		}
1050 		break;
1051 	}
1052 	return (error);
1053 }
1054 
1055 /*
1056  * Obtain the requested vnode lock
1057  *
1058  *	LK_RETRY	Automatically retry on timeout
1059  *	LK_FAILRECLAIM	Fail if the vnode is being reclaimed
1060  *
1061  * Failures will occur if the vnode is undergoing recyclement, but not
1062  * all callers expect that the function will fail so the caller must pass
1063  * LK_FAILOK if it wants to process an error code.
1064  *
1065  * Errors can occur for other reasons if you pass in other LK_ flags,
1066  * regardless of whether you pass in LK_FAILRECLAIM
1067  */
1068 int
1069 vn_lock(struct vnode *vp, int flags)
1070 {
1071 	int error;
1072 
1073 	do {
1074 		error = lockmgr(&vp->v_lock, flags);
1075 		if (error == 0)
1076 			break;
1077 	} while (flags & LK_RETRY);
1078 
1079 	/*
1080 	 * Because we (had better!) have a ref on the vnode, once it
1081 	 * goes to VRECLAIMED state it will not be recycled until all
1082 	 * refs go away.  So we can just check the flag.
1083 	 */
1084 	if (error == 0 && (vp->v_flag & VRECLAIMED)) {
1085 		if (flags & LK_FAILRECLAIM) {
1086 			lockmgr(&vp->v_lock, LK_RELEASE);
1087 			error = ENOENT;
1088 		}
1089 	}
1090 	return (error);
1091 }
1092 
1093 int
1094 vn_relock(struct vnode *vp, int flags)
1095 {
1096 	int error;
1097 
1098 	do {
1099 		error = lockmgr(&vp->v_lock, flags);
1100 		if (error == 0)
1101 			break;
1102 	} while (flags & LK_RETRY);
1103 
1104 	return error;
1105 }
1106 
1107 #ifdef DEBUG_VN_UNLOCK
1108 
1109 void
1110 debug_vn_unlock(struct vnode *vp, const char *filename, int line)
1111 {
1112 	kprintf("vn_unlock from %s:%d\n", filename, line);
1113 	lockmgr(&vp->v_lock, LK_RELEASE);
1114 }
1115 
1116 #else
1117 
1118 void
1119 vn_unlock(struct vnode *vp)
1120 {
1121 	lockmgr(&vp->v_lock, LK_RELEASE);
1122 }
1123 
1124 #endif
1125 
1126 /*
1127  * MPSAFE
1128  */
1129 int
1130 vn_islocked(struct vnode *vp)
1131 {
1132 	return (lockstatus(&vp->v_lock, curthread));
1133 }
1134 
1135 /*
1136  * Return the lock status of a vnode and unlock the vnode
1137  * if we owned the lock.  This is not a boolean, if the
1138  * caller cares what the lock status is the caller must
1139  * check the various possible values.
1140  *
1141  * This only unlocks exclusive locks held by the caller,
1142  * it will NOT unlock shared locks (there is no way to
1143  * tell who the shared lock belongs to).
1144  *
1145  * MPSAFE
1146  */
1147 int
1148 vn_islocked_unlock(struct vnode *vp)
1149 {
1150 	int vpls;
1151 
1152 	vpls = lockstatus(&vp->v_lock, curthread);
1153 	if (vpls == LK_EXCLUSIVE)
1154 		lockmgr(&vp->v_lock, LK_RELEASE);
1155 	return(vpls);
1156 }
1157 
1158 /*
1159  * Restore a vnode lock that we previously released via
1160  * vn_islocked_unlock().  This is a NOP if we did not
1161  * own the original lock.
1162  *
1163  * MPSAFE
1164  */
1165 void
1166 vn_islocked_relock(struct vnode *vp, int vpls)
1167 {
1168 	int error;
1169 
1170 	if (vpls == LK_EXCLUSIVE)
1171 		error = lockmgr(&vp->v_lock, vpls);
1172 }
1173 
1174 /*
1175  * MPSAFE
1176  */
1177 static int
1178 vn_closefile(struct file *fp)
1179 {
1180 	int error;
1181 
1182 	fp->f_ops = &badfileops;
1183 	error = vn_close(((struct vnode *)fp->f_data), fp->f_flag, fp);
1184 	return (error);
1185 }
1186 
1187 /*
1188  * MPSAFE
1189  */
1190 static int
1191 vn_kqfilter(struct file *fp, struct knote *kn)
1192 {
1193 	int error;
1194 
1195 	error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn);
1196 	return (error);
1197 }
1198 
1199 int
1200 vn_seek(struct file *fp, off_t offset, int whence, off_t *res)
1201 {
1202 	/*
1203 	 * NOTE: devfs_dev_fileops uses exact same code
1204 	 */
1205 	struct vnode *vp;
1206 	struct vattr_lite lva;
1207 	off_t new_offset;
1208 	int error;
1209 
1210 	vp = (struct vnode *)fp->f_data;
1211 
1212 	switch (whence) {
1213 	case L_INCR:
1214 		spin_lock(&fp->f_spin);
1215 		new_offset = fp->f_offset + offset;
1216 		error = 0;
1217 		break;
1218 	case L_XTND:
1219 		error = VOP_GETATTR_LITE(vp, &lva);
1220 		spin_lock(&fp->f_spin);
1221 		new_offset = offset + lva.va_size;
1222 		break;
1223 	case L_SET:
1224 		new_offset = offset;
1225 		error = 0;
1226 		spin_lock(&fp->f_spin);
1227 		break;
1228 	default:
1229 		new_offset = 0;
1230 		error = EINVAL;
1231 		spin_lock(&fp->f_spin);
1232 		break;
1233 	}
1234 
1235 	/*
1236 	 * Validate the seek position.  Negative offsets are not allowed
1237 	 * for regular files or directories.
1238 	 *
1239 	 * Normally we would also not want to allow negative offsets for
1240 	 * character and block-special devices.  However kvm addresses
1241 	 * on 64 bit architectures might appear to be negative and must
1242 	 * be allowed.
1243 	 */
1244 	if (error == 0) {
1245 		if (new_offset < 0 &&
1246 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
1247 			error = EINVAL;
1248 		} else {
1249 			fp->f_offset = new_offset;
1250 		}
1251 	}
1252 	*res = fp->f_offset;
1253 	spin_unlock(&fp->f_spin);
1254 
1255 	return (error);
1256 }
1257