xref: /freebsd/sys/fs/fdescfs/fdesc_vnops.c (revision 7bd6fde3)
1 /*-
2  * Copyright (c) 1992, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software donated to Berkeley by
6  * Jan-Simon Pendry.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)fdesc_vnops.c	8.9 (Berkeley) 1/21/94
33  *
34  * $FreeBSD$
35  */
36 
37 /*
38  * /dev/fd Filesystem
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/conf.h>
44 #include <sys/dirent.h>
45 #include <sys/filedesc.h>
46 #include <sys/kernel.h>	/* boottime */
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/malloc.h>
50 #include <sys/file.h>	/* Must come after sys/malloc.h */
51 #include <sys/mount.h>
52 #include <sys/namei.h>
53 #include <sys/proc.h>
54 #include <sys/stat.h>
55 #include <sys/vnode.h>
56 
57 #include <fs/fdescfs/fdesc.h>
58 
59 #define FDL_WANT	0x01
60 #define FDL_LOCKED	0x02
61 static int fdcache_lock;
62 
63 #define	NFDCACHE 4
64 #define FD_NHASH(ix) \
65 	(&fdhashtbl[(ix) & fdhash])
66 static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl;
67 static u_long fdhash;
68 
69 static vop_getattr_t	fdesc_getattr;
70 static vop_inactive_t	fdesc_inactive;
71 static vop_lookup_t	fdesc_lookup;
72 static vop_open_t	fdesc_open;
73 static vop_readdir_t	fdesc_readdir;
74 static vop_reclaim_t	fdesc_reclaim;
75 static vop_setattr_t	fdesc_setattr;
76 
77 static struct vop_vector fdesc_vnodeops = {
78 	.vop_default =		&default_vnodeops,
79 
80 	.vop_access =		VOP_NULL,
81 	.vop_getattr =		fdesc_getattr,
82 	.vop_inactive =		fdesc_inactive,
83 	.vop_lookup =		fdesc_lookup,
84 	.vop_open =		fdesc_open,
85 	.vop_pathconf =		vop_stdpathconf,
86 	.vop_readdir =		fdesc_readdir,
87 	.vop_reclaim =		fdesc_reclaim,
88 	.vop_setattr =		fdesc_setattr,
89 };
90 
91 /*
92  * Initialise cache headers
93  */
94 int
95 fdesc_init(vfsp)
96 	struct vfsconf *vfsp;
97 {
98 
99 	fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash);
100 	return (0);
101 }
102 
103 int
104 fdesc_allocvp(ftype, ix, mp, vpp, td)
105 	fdntype ftype;
106 	int ix;
107 	struct mount *mp;
108 	struct vnode **vpp;
109 	struct thread *td;
110 {
111 	struct fdhashhead *fc;
112 	struct fdescnode *fd;
113 	int error = 0;
114 
115 	fc = FD_NHASH(ix);
116 loop:
117 	LIST_FOREACH(fd, fc, fd_hash) {
118 		if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) {
119 			if (vget(fd->fd_vnode, 0, td))
120 				goto loop;
121 			*vpp = fd->fd_vnode;
122 			return (error);
123 		}
124 	}
125 
126 	/*
127 	 * otherwise lock the array while we call getnewvnode
128 	 * since that can block.
129 	 */
130 	if (fdcache_lock & FDL_LOCKED) {
131 		fdcache_lock |= FDL_WANT;
132 		(void) tsleep( &fdcache_lock, PINOD, "fdalvp", 0);
133 		goto loop;
134 	}
135 	fdcache_lock |= FDL_LOCKED;
136 
137 	/*
138 	 * Do the MALLOC before the getnewvnode since doing so afterward
139 	 * might cause a bogus v_data pointer to get dereferenced
140 	 * elsewhere if MALLOC should block.
141 	 */
142 	MALLOC(fd, struct fdescnode *, sizeof(struct fdescnode), M_TEMP, M_WAITOK);
143 
144 	error = getnewvnode("fdesc", mp, &fdesc_vnodeops, vpp);
145 	if (error) {
146 		FREE(fd, M_TEMP);
147 		goto out;
148 	}
149 	(*vpp)->v_data = fd;
150 	fd->fd_vnode = *vpp;
151 	fd->fd_type = ftype;
152 	fd->fd_fd = -1;
153 	fd->fd_ix = ix;
154 	LIST_INSERT_HEAD(fc, fd, fd_hash);
155 
156 out:
157 	fdcache_lock &= ~FDL_LOCKED;
158 
159 	if (fdcache_lock & FDL_WANT) {
160 		fdcache_lock &= ~FDL_WANT;
161 		wakeup( &fdcache_lock);
162 	}
163 
164 	return (error);
165 }
166 
167 /*
168  * vp is the current namei directory
169  * ndp is the name to locate in that directory...
170  */
171 static int
172 fdesc_lookup(ap)
173 	struct vop_lookup_args /* {
174 		struct vnode * a_dvp;
175 		struct vnode ** a_vpp;
176 		struct componentname * a_cnp;
177 	} */ *ap;
178 {
179 	struct vnode **vpp = ap->a_vpp;
180 	struct vnode *dvp = ap->a_dvp;
181 	struct componentname *cnp = ap->a_cnp;
182 	char *pname = cnp->cn_nameptr;
183 	struct thread *td = cnp->cn_thread;
184 	struct file *fp;
185 	int nlen = cnp->cn_namelen;
186 	u_int fd;
187 	int error;
188 	struct vnode *fvp;
189 
190 	if ((cnp->cn_flags & ISLASTCN) &&
191 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
192 		error = EROFS;
193 		goto bad;
194 	}
195 
196 	if (cnp->cn_namelen == 1 && *pname == '.') {
197 		*vpp = dvp;
198 		VREF(dvp);
199 		return (0);
200 	}
201 
202 	if (VTOFDESC(dvp)->fd_type != Froot) {
203 		error = ENOTDIR;
204 		goto bad;
205 	}
206 
207 	fd = 0;
208 	/* the only time a leading 0 is acceptable is if it's "0" */
209 	if (*pname == '0' && nlen != 1) {
210 		error = ENOENT;
211 		goto bad;
212 	}
213 	while (nlen--) {
214 		if (*pname < '0' || *pname > '9') {
215 			error = ENOENT;
216 			goto bad;
217 		}
218 		fd = 10 * fd + *pname++ - '0';
219 	}
220 
221 	if ((error = fget(td, fd, &fp)) != 0)
222 		goto bad;
223 
224 	error = fdesc_allocvp(Fdesc, FD_DESC+fd, dvp->v_mount, &fvp, td);
225 	fdrop(fp, td);
226 	if (error)
227 		goto bad;
228 	VTOFDESC(fvp)->fd_fd = fd;
229 	if (fvp != dvp)
230 		vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, td);
231 	*vpp = fvp;
232 	return (0);
233 
234 bad:
235 	*vpp = NULL;
236 	return (error);
237 }
238 
239 static int
240 fdesc_open(ap)
241 	struct vop_open_args /* {
242 		struct vnode *a_vp;
243 		int  a_mode;
244 		struct ucred *a_cred;
245 		struct thread *a_td;
246 	} */ *ap;
247 {
248 	struct vnode *vp = ap->a_vp;
249 
250 	if (VTOFDESC(vp)->fd_type == Froot)
251 		return (0);
252 
253 	/*
254 	 * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the the file
255 	 * descriptor being sought for duplication. The error return ensures
256 	 * that the vnode for this device will be released by vn_open. Open
257 	 * will detect this special error and take the actions in dupfdopen.
258 	 * Other callers of vn_open or VOP_OPEN will simply report the
259 	 * error.
260 	 */
261 	ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd;	/* XXX */
262 	return (ENODEV);
263 }
264 
265 static int
266 fdesc_getattr(ap)
267 	struct vop_getattr_args /* {
268 		struct vnode *a_vp;
269 		struct vattr *a_vap;
270 		struct ucred *a_cred;
271 		struct thread *a_td;
272 	} */ *ap;
273 {
274 	struct vnode *vp = ap->a_vp;
275 	struct vattr *vap = ap->a_vap;
276 	struct file *fp;
277 	struct stat stb;
278 	u_int fd;
279 	int error = 0;
280 
281 	switch (VTOFDESC(vp)->fd_type) {
282 	case Froot:
283 		VATTR_NULL(vap);
284 
285 		vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
286 		vap->va_type = VDIR;
287 		vap->va_nlink = 2;
288 		vap->va_size = DEV_BSIZE;
289 		vap->va_fileid = VTOFDESC(vp)->fd_ix;
290 		vap->va_uid = 0;
291 		vap->va_gid = 0;
292 		vap->va_blocksize = DEV_BSIZE;
293 		vap->va_atime.tv_sec = boottime.tv_sec;
294 		vap->va_atime.tv_nsec = 0;
295 		vap->va_mtime = vap->va_atime;
296 		vap->va_ctime = vap->va_mtime;
297 		vap->va_gen = 0;
298 		vap->va_flags = 0;
299 		vap->va_rdev = 0;
300 		vap->va_bytes = 0;
301 		break;
302 
303 	case Fdesc:
304 		fd = VTOFDESC(vp)->fd_fd;
305 
306 		if ((error = fget(ap->a_td, fd, &fp)) != 0)
307 			return (error);
308 
309 		bzero(&stb, sizeof(stb));
310 		error = fo_stat(fp, &stb, ap->a_td->td_ucred, ap->a_td);
311 		fdrop(fp, ap->a_td);
312 		if (error == 0) {
313 			VATTR_NULL(vap);
314 			vap->va_type = IFTOVT(stb.st_mode);
315 			vap->va_mode = stb.st_mode;
316 #define FDRX (VREAD|VEXEC)
317 			if (vap->va_type == VDIR)
318 				vap->va_mode &= ~((FDRX)|(FDRX>>3)|(FDRX>>6));
319 #undef FDRX
320 			vap->va_nlink = 1;
321 			vap->va_flags = 0;
322 			vap->va_bytes = stb.st_blocks * stb.st_blksize;
323 			vap->va_fileid = VTOFDESC(vp)->fd_ix;
324 			vap->va_size = stb.st_size;
325 			vap->va_blocksize = stb.st_blksize;
326 			vap->va_rdev = stb.st_rdev;
327 
328 			/*
329 			 * If no time data is provided, use the current time.
330 			 */
331 			if (stb.st_atimespec.tv_sec == 0 &&
332 			    stb.st_atimespec.tv_nsec == 0)
333 				nanotime(&stb.st_atimespec);
334 
335 			if (stb.st_ctimespec.tv_sec == 0 &&
336 			    stb.st_ctimespec.tv_nsec == 0)
337 				nanotime(&stb.st_ctimespec);
338 
339 			if (stb.st_mtimespec.tv_sec == 0 &&
340 			    stb.st_mtimespec.tv_nsec == 0)
341 				nanotime(&stb.st_mtimespec);
342 
343 			vap->va_atime = stb.st_atimespec;
344 			vap->va_mtime = stb.st_mtimespec;
345 			vap->va_ctime = stb.st_ctimespec;
346 			vap->va_uid = stb.st_uid;
347 			vap->va_gid = stb.st_gid;
348 		}
349 		break;
350 
351 	default:
352 		panic("fdesc_getattr");
353 		break;
354 	}
355 
356 	if (error == 0)
357 		vp->v_type = vap->va_type;
358 	return (error);
359 }
360 
361 static int
362 fdesc_setattr(ap)
363 	struct vop_setattr_args /* {
364 		struct vnode *a_vp;
365 		struct vattr *a_vap;
366 		struct ucred *a_cred;
367 		struct thread *a_td;
368 	} */ *ap;
369 {
370 	struct vattr *vap = ap->a_vap;
371 	struct vnode *vp;
372 	struct mount *mp;
373 	struct file *fp;
374 	unsigned fd;
375 	int error;
376 
377 	/*
378 	 * Can't mess with the root vnode
379 	 */
380 	if (VTOFDESC(ap->a_vp)->fd_type == Froot)
381 		return (EACCES);
382 
383 	fd = VTOFDESC(ap->a_vp)->fd_fd;
384 
385 	/*
386 	 * Allow setattr where there is an underlying vnode.
387 	 */
388 	error = getvnode(ap->a_td->td_proc->p_fd, fd, &fp);
389 	if (error) {
390 		/*
391 		 * getvnode() returns EINVAL if the file descriptor is not
392 		 * backed by a vnode.  Silently drop all changes except
393 		 * chflags(2) in this case.
394 		 */
395 		if (error == EINVAL) {
396 			if (vap->va_flags != VNOVAL)
397 				error = EOPNOTSUPP;
398 			else
399 				error = 0;
400 		}
401 		return (error);
402 	}
403 	vp = fp->f_vnode;
404 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) {
405 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
406 		error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred, ap->a_td);
407 		VOP_UNLOCK(vp, 0, ap->a_td);
408 		vn_finished_write(mp);
409 	}
410 	fdrop(fp, ap->a_td);
411 	return (error);
412 }
413 
414 #define UIO_MX 16
415 
416 static int
417 fdesc_readdir(ap)
418 	struct vop_readdir_args /* {
419 		struct vnode *a_vp;
420 		struct uio *a_uio;
421 		struct ucred *a_cred;
422 		int *a_eofflag;
423 		u_long *a_cookies;
424 		int a_ncookies;
425 	} */ *ap;
426 {
427 	struct uio *uio = ap->a_uio;
428 	struct filedesc *fdp;
429 	struct dirent d;
430 	struct dirent *dp = &d;
431 	int error, i, off, fcnt;
432 
433 	/*
434 	 * We don't allow exporting fdesc mounts, and currently local
435 	 * requests do not need cookies.
436 	 */
437 	if (ap->a_ncookies)
438 		panic("fdesc_readdir: not hungry");
439 
440 	if (VTOFDESC(ap->a_vp)->fd_type != Froot)
441 		panic("fdesc_readdir: not dir");
442 
443 	off = (int)uio->uio_offset;
444 	if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 ||
445 	    uio->uio_resid < UIO_MX)
446 		return (EINVAL);
447 	i = (u_int)off / UIO_MX;
448 	fdp = uio->uio_td->td_proc->p_fd;
449 	error = 0;
450 
451 	fcnt = i - 2;		/* The first two nodes are `.' and `..' */
452 
453 	FILEDESC_LOCK_FAST(fdp);
454 	while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) {
455 		switch (i) {
456 		case 0:	/* `.' */
457 		case 1: /* `..' */
458 			bzero((caddr_t)dp, UIO_MX);
459 
460 			dp->d_fileno = i + FD_ROOT;
461 			dp->d_namlen = i + 1;
462 			dp->d_reclen = UIO_MX;
463 			bcopy("..", dp->d_name, dp->d_namlen);
464 			dp->d_name[i + 1] = '\0';
465 			dp->d_type = DT_DIR;
466 			break;
467 		default:
468 			if (fdp->fd_ofiles[fcnt] == NULL) {
469 				FILEDESC_UNLOCK_FAST(fdp);
470 				goto done;
471 			}
472 
473 			bzero((caddr_t) dp, UIO_MX);
474 			dp->d_namlen = sprintf(dp->d_name, "%d", fcnt);
475 			dp->d_reclen = UIO_MX;
476 			dp->d_type = DT_UNKNOWN;
477 			dp->d_fileno = i + FD_DESC;
478 			break;
479 		}
480 		/*
481 		 * And ship to userland
482 		 */
483 		FILEDESC_UNLOCK_FAST(fdp);
484 		error = uiomove(dp, UIO_MX, uio);
485 		if (error)
486 			goto done;
487 		FILEDESC_LOCK_FAST(fdp);
488 		i++;
489 		fcnt++;
490 	}
491 	FILEDESC_UNLOCK_FAST(fdp);
492 
493 done:
494 	uio->uio_offset = i * UIO_MX;
495 	return (error);
496 }
497 
498 static int
499 fdesc_inactive(ap)
500 	struct vop_inactive_args /* {
501 		struct vnode *a_vp;
502 		struct thread *a_td;
503 	} */ *ap;
504 {
505 	struct vnode *vp = ap->a_vp;
506 
507 	/*
508 	 * Clear out the v_type field to avoid
509 	 * nasty things happening in vgone().
510 	 */
511 	vp->v_type = VNON;
512 	return (0);
513 }
514 
515 static int
516 fdesc_reclaim(ap)
517 	struct vop_reclaim_args /* {
518 		struct vnode *a_vp;
519 	} */ *ap;
520 {
521 	struct vnode *vp = ap->a_vp;
522 	struct fdescnode *fd = VTOFDESC(vp);
523 
524 	LIST_REMOVE(fd, fd_hash);
525 	FREE(vp->v_data, M_TEMP);
526 	vp->v_data = 0;
527 
528 	return (0);
529 }
530