xref: /freebsd/sys/fs/fdescfs/fdesc_vnops.c (revision 4232f826683298e85d469c0ef17259cecd13b2c7)
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/capsicum.h>
44 #include <sys/conf.h>
45 #include <sys/dirent.h>
46 #include <sys/filedesc.h>
47 #include <sys/kernel.h>	/* boottime */
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/malloc.h>
51 #include <sys/file.h>	/* Must come after sys/malloc.h */
52 #include <sys/mount.h>
53 #include <sys/namei.h>
54 #include <sys/proc.h>
55 #include <sys/stat.h>
56 #include <sys/vnode.h>
57 
58 #include <fs/fdescfs/fdesc.h>
59 
60 #define	NFDCACHE 4
61 #define FD_NHASH(ix) \
62 	(&fdhashtbl[(ix) & fdhash])
63 static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl;
64 static u_long fdhash;
65 
66 struct mtx fdesc_hashmtx;
67 
68 static vop_getattr_t	fdesc_getattr;
69 static vop_lookup_t	fdesc_lookup;
70 static vop_open_t	fdesc_open;
71 static vop_readdir_t	fdesc_readdir;
72 static vop_reclaim_t	fdesc_reclaim;
73 static vop_setattr_t	fdesc_setattr;
74 
75 static struct vop_vector fdesc_vnodeops = {
76 	.vop_default =		&default_vnodeops,
77 
78 	.vop_access =		VOP_NULL,
79 	.vop_getattr =		fdesc_getattr,
80 	.vop_lookup =		fdesc_lookup,
81 	.vop_open =		fdesc_open,
82 	.vop_pathconf =		vop_stdpathconf,
83 	.vop_readdir =		fdesc_readdir,
84 	.vop_reclaim =		fdesc_reclaim,
85 	.vop_setattr =		fdesc_setattr,
86 };
87 
88 static void fdesc_insmntque_dtr(struct vnode *, void *);
89 static void fdesc_remove_entry(struct fdescnode *);
90 
91 /*
92  * Initialise cache headers
93  */
94 int
95 fdesc_init(vfsp)
96 	struct vfsconf *vfsp;
97 {
98 
99 	mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF);
100 	fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash);
101 	return (0);
102 }
103 
104 /*
105  * Uninit ready for unload.
106  */
107 int
108 fdesc_uninit(vfsp)
109 	struct vfsconf *vfsp;
110 {
111 
112 	hashdestroy(fdhashtbl, M_CACHE, fdhash);
113 	mtx_destroy(&fdesc_hashmtx);
114 	return (0);
115 }
116 
117 /*
118  * If allocating vnode fails, call this.
119  */
120 static void
121 fdesc_insmntque_dtr(struct vnode *vp, void *arg)
122 {
123 
124 	vgone(vp);
125 	vput(vp);
126 }
127 
128 /*
129  * Remove an entry from the hash if it exists.
130  */
131 static void
132 fdesc_remove_entry(struct fdescnode *fd)
133 {
134 	struct fdhashhead *fc;
135 	struct fdescnode *fd2;
136 
137 	fc = FD_NHASH(fd->fd_ix);
138 	mtx_lock(&fdesc_hashmtx);
139 	LIST_FOREACH(fd2, fc, fd_hash) {
140 		if (fd == fd2) {
141 			LIST_REMOVE(fd, fd_hash);
142 			break;
143 		}
144 	}
145 	mtx_unlock(&fdesc_hashmtx);
146 }
147 
148 int
149 fdesc_allocvp(ftype, fd_fd, ix, mp, vpp)
150 	fdntype ftype;
151 	unsigned fd_fd;
152 	int ix;
153 	struct mount *mp;
154 	struct vnode **vpp;
155 {
156 	struct fdescmount *fmp;
157 	struct fdhashhead *fc;
158 	struct fdescnode *fd, *fd2;
159 	struct vnode *vp, *vp2;
160 	struct thread *td;
161 	int error = 0;
162 
163 	td = curthread;
164 	fc = FD_NHASH(ix);
165 loop:
166 	mtx_lock(&fdesc_hashmtx);
167 	/*
168 	 * If a forced unmount is progressing, we need to drop it. The flags are
169 	 * protected by the hashmtx.
170 	 */
171 	fmp = (struct fdescmount *)mp->mnt_data;
172 	if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) {
173 		mtx_unlock(&fdesc_hashmtx);
174 		return (-1);
175 	}
176 
177 	LIST_FOREACH(fd, fc, fd_hash) {
178 		if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) {
179 			/* Get reference to vnode in case it's being free'd */
180 			vp = fd->fd_vnode;
181 			VI_LOCK(vp);
182 			mtx_unlock(&fdesc_hashmtx);
183 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td))
184 				goto loop;
185 			*vpp = vp;
186 			return (0);
187 		}
188 	}
189 	mtx_unlock(&fdesc_hashmtx);
190 
191 	fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK);
192 
193 	error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp);
194 	if (error) {
195 		free(fd, M_TEMP);
196 		return (error);
197 	}
198 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
199 	vp->v_data = fd;
200 	fd->fd_vnode = vp;
201 	fd->fd_type = ftype;
202 	fd->fd_fd = fd_fd;
203 	fd->fd_ix = ix;
204 	error = insmntque1(vp, mp, fdesc_insmntque_dtr, NULL);
205 	if (error != 0) {
206 		*vpp = NULLVP;
207 		return (error);
208 	}
209 
210 	/* Make sure that someone didn't beat us when inserting the vnode. */
211 	mtx_lock(&fdesc_hashmtx);
212 	/*
213 	 * If a forced unmount is progressing, we need to drop it. The flags are
214 	 * protected by the hashmtx.
215 	 */
216 	fmp = (struct fdescmount *)mp->mnt_data;
217 	if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) {
218 		mtx_unlock(&fdesc_hashmtx);
219 		vgone(vp);
220 		vput(vp);
221 		*vpp = NULLVP;
222 		return (-1);
223 	}
224 
225 	LIST_FOREACH(fd2, fc, fd_hash) {
226 		if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) {
227 			/* Get reference to vnode in case it's being free'd */
228 			vp2 = fd2->fd_vnode;
229 			VI_LOCK(vp2);
230 			mtx_unlock(&fdesc_hashmtx);
231 			error = vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, td);
232 			/* Someone beat us, dec use count and wait for reclaim */
233 			vgone(vp);
234 			vput(vp);
235 			/* If we didn't get it, return no vnode. */
236 			if (error)
237 				vp2 = NULLVP;
238 			*vpp = vp2;
239 			return (error);
240 		}
241 	}
242 
243 	/* If we came here, we can insert it safely. */
244 	LIST_INSERT_HEAD(fc, fd, fd_hash);
245 	mtx_unlock(&fdesc_hashmtx);
246 	*vpp = vp;
247 	return (0);
248 }
249 
250 struct fdesc_get_ino_args {
251 	fdntype ftype;
252 	unsigned fd_fd;
253 	int ix;
254 	struct file *fp;
255 	struct thread *td;
256 };
257 
258 static int
259 fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
260     struct vnode **rvp)
261 {
262 	struct fdesc_get_ino_args *a;
263 	int error;
264 
265 	a = arg;
266 	error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp);
267 	fdrop(a->fp, a->td);
268 	return (error);
269 }
270 
271 
272 /*
273  * vp is the current namei directory
274  * ndp is the name to locate in that directory...
275  */
276 static int
277 fdesc_lookup(ap)
278 	struct vop_lookup_args /* {
279 		struct vnode * a_dvp;
280 		struct vnode ** a_vpp;
281 		struct componentname * a_cnp;
282 	} */ *ap;
283 {
284 	struct vnode **vpp = ap->a_vpp;
285 	struct vnode *dvp = ap->a_dvp;
286 	struct componentname *cnp = ap->a_cnp;
287 	char *pname = cnp->cn_nameptr;
288 	struct thread *td = cnp->cn_thread;
289 	struct file *fp;
290 	struct fdesc_get_ino_args arg;
291 	int nlen = cnp->cn_namelen;
292 	u_int fd, fd1;
293 	int error;
294 	struct vnode *fvp;
295 
296 	if ((cnp->cn_flags & ISLASTCN) &&
297 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
298 		error = EROFS;
299 		goto bad;
300 	}
301 
302 	if (cnp->cn_namelen == 1 && *pname == '.') {
303 		*vpp = dvp;
304 		VREF(dvp);
305 		return (0);
306 	}
307 
308 	if (VTOFDESC(dvp)->fd_type != Froot) {
309 		error = ENOTDIR;
310 		goto bad;
311 	}
312 
313 	fd = 0;
314 	/* the only time a leading 0 is acceptable is if it's "0" */
315 	if (*pname == '0' && nlen != 1) {
316 		error = ENOENT;
317 		goto bad;
318 	}
319 	while (nlen--) {
320 		if (*pname < '0' || *pname > '9') {
321 			error = ENOENT;
322 			goto bad;
323 		}
324 		fd1 = 10 * fd + *pname++ - '0';
325 		if (fd1 < fd) {
326 			error = ENOENT;
327 			goto bad;
328 		}
329 		fd = fd1;
330 	}
331 
332 	/*
333 	 * No rights to check since 'fp' isn't actually used.
334 	 */
335 	if ((error = fget(td, fd, NULL, &fp)) != 0)
336 		goto bad;
337 
338 	/* Check if we're looking up ourselves. */
339 	if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) {
340 		/*
341 		 * In case we're holding the last reference to the file, the dvp
342 		 * will be re-acquired.
343 		 */
344 		vhold(dvp);
345 		VOP_UNLOCK(dvp, 0);
346 		fdrop(fp, td);
347 
348 		/* Re-aquire the lock afterwards. */
349 		vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE);
350 		vdrop(dvp);
351 		fvp = dvp;
352 		if ((dvp->v_iflag & VI_DOOMED) != 0)
353 			error = ENOENT;
354 	} else {
355 		/*
356 		 * Unlock our root node (dvp) when doing this, since we might
357 		 * deadlock since the vnode might be locked by another thread
358 		 * and the root vnode lock will be obtained afterwards (in case
359 		 * we're looking up the fd of the root vnode), which will be the
360 		 * opposite lock order. Vhold the root vnode first so we don't
361 		 * lose it.
362 		 */
363 		arg.ftype = Fdesc;
364 		arg.fd_fd = fd;
365 		arg.ix = FD_DESC + fd;
366 		arg.fp = fp;
367 		arg.td = td;
368 		error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg,
369 		    LK_EXCLUSIVE, &fvp);
370 	}
371 
372 	if (error)
373 		goto bad;
374 	*vpp = fvp;
375 	return (0);
376 
377 bad:
378 	*vpp = NULL;
379 	return (error);
380 }
381 
382 static int
383 fdesc_open(ap)
384 	struct vop_open_args /* {
385 		struct vnode *a_vp;
386 		int  a_mode;
387 		struct ucred *a_cred;
388 		struct thread *a_td;
389 	} */ *ap;
390 {
391 	struct vnode *vp = ap->a_vp;
392 
393 	if (VTOFDESC(vp)->fd_type == Froot)
394 		return (0);
395 
396 	/*
397 	 * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file
398 	 * descriptor being sought for duplication. The error return ensures
399 	 * that the vnode for this device will be released by vn_open. Open
400 	 * will detect this special error and take the actions in dupfdopen.
401 	 * Other callers of vn_open or VOP_OPEN will simply report the
402 	 * error.
403 	 */
404 	ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd;	/* XXX */
405 	return (ENODEV);
406 }
407 
408 static int
409 fdesc_getattr(ap)
410 	struct vop_getattr_args /* {
411 		struct vnode *a_vp;
412 		struct vattr *a_vap;
413 		struct ucred *a_cred;
414 	} */ *ap;
415 {
416 	struct vnode *vp = ap->a_vp;
417 	struct vattr *vap = ap->a_vap;
418 
419 	vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
420 	vap->va_fileid = VTOFDESC(vp)->fd_ix;
421 	vap->va_uid = 0;
422 	vap->va_gid = 0;
423 	vap->va_blocksize = DEV_BSIZE;
424 	vap->va_atime.tv_sec = boottime.tv_sec;
425 	vap->va_atime.tv_nsec = 0;
426 	vap->va_mtime = vap->va_atime;
427 	vap->va_ctime = vap->va_mtime;
428 	vap->va_gen = 0;
429 	vap->va_flags = 0;
430 	vap->va_bytes = 0;
431 	vap->va_filerev = 0;
432 
433 	switch (VTOFDESC(vp)->fd_type) {
434 	case Froot:
435 		vap->va_type = VDIR;
436 		vap->va_nlink = 2;
437 		vap->va_size = DEV_BSIZE;
438 		vap->va_rdev = NODEV;
439 		break;
440 
441 	case Fdesc:
442 		vap->va_type = VCHR;
443 		vap->va_nlink = 1;
444 		vap->va_size = 0;
445 		vap->va_rdev = makedev(0, vap->va_fileid);
446 		break;
447 
448 	default:
449 		panic("fdesc_getattr");
450 		break;
451 	}
452 
453 	vp->v_type = vap->va_type;
454 	return (0);
455 }
456 
457 static int
458 fdesc_setattr(ap)
459 	struct vop_setattr_args /* {
460 		struct vnode *a_vp;
461 		struct vattr *a_vap;
462 		struct ucred *a_cred;
463 	} */ *ap;
464 {
465 	struct vattr *vap = ap->a_vap;
466 	struct vnode *vp;
467 	struct mount *mp;
468 	struct file *fp;
469 	struct thread *td = curthread;
470 	cap_rights_t rights;
471 	unsigned fd;
472 	int error;
473 
474 	/*
475 	 * Can't mess with the root vnode
476 	 */
477 	if (VTOFDESC(ap->a_vp)->fd_type == Froot)
478 		return (EACCES);
479 
480 	fd = VTOFDESC(ap->a_vp)->fd_fd;
481 
482 	/*
483 	 * Allow setattr where there is an underlying vnode.
484 	 */
485 	error = getvnode(td->td_proc->p_fd, fd,
486 	    cap_rights_init(&rights, CAP_EXTATTR_SET), &fp);
487 	if (error) {
488 		/*
489 		 * getvnode() returns EINVAL if the file descriptor is not
490 		 * backed by a vnode.  Silently drop all changes except
491 		 * chflags(2) in this case.
492 		 */
493 		if (error == EINVAL) {
494 			if (vap->va_flags != VNOVAL)
495 				error = EOPNOTSUPP;
496 			else
497 				error = 0;
498 		}
499 		return (error);
500 	}
501 	vp = fp->f_vnode;
502 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) {
503 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
504 		error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred);
505 		VOP_UNLOCK(vp, 0);
506 		vn_finished_write(mp);
507 	}
508 	fdrop(fp, td);
509 	return (error);
510 }
511 
512 #define UIO_MX 16
513 
514 static int
515 fdesc_readdir(ap)
516 	struct vop_readdir_args /* {
517 		struct vnode *a_vp;
518 		struct uio *a_uio;
519 		struct ucred *a_cred;
520 		int *a_eofflag;
521 		u_long *a_cookies;
522 		int a_ncookies;
523 	} */ *ap;
524 {
525 	struct uio *uio = ap->a_uio;
526 	struct filedesc *fdp;
527 	struct dirent d;
528 	struct dirent *dp = &d;
529 	int error, i, off, fcnt;
530 
531 	if (VTOFDESC(ap->a_vp)->fd_type != Froot)
532 		panic("fdesc_readdir: not dir");
533 
534 	if (ap->a_ncookies != NULL)
535 		*ap->a_ncookies = 0;
536 
537 	off = (int)uio->uio_offset;
538 	if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 ||
539 	    uio->uio_resid < UIO_MX)
540 		return (EINVAL);
541 	i = (u_int)off / UIO_MX;
542 	fdp = uio->uio_td->td_proc->p_fd;
543 	error = 0;
544 
545 	fcnt = i - 2;		/* The first two nodes are `.' and `..' */
546 
547 	FILEDESC_SLOCK(fdp);
548 	while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) {
549 		bzero((caddr_t)dp, UIO_MX);
550 		switch (i) {
551 		case 0:	/* `.' */
552 		case 1: /* `..' */
553 			dp->d_fileno = i + FD_ROOT;
554 			dp->d_namlen = i + 1;
555 			dp->d_reclen = UIO_MX;
556 			bcopy("..", dp->d_name, dp->d_namlen);
557 			dp->d_name[i + 1] = '\0';
558 			dp->d_type = DT_DIR;
559 			break;
560 		default:
561 			if (fdp->fd_ofiles[fcnt].fde_file == NULL)
562 				break;
563 			dp->d_namlen = sprintf(dp->d_name, "%d", fcnt);
564 			dp->d_reclen = UIO_MX;
565 			dp->d_type = DT_CHR;
566 			dp->d_fileno = i + FD_DESC;
567 			break;
568 		}
569 		if (dp->d_namlen != 0) {
570 			/*
571 			 * And ship to userland
572 			 */
573 			FILEDESC_SUNLOCK(fdp);
574 			error = uiomove(dp, UIO_MX, uio);
575 			if (error)
576 				goto done;
577 			FILEDESC_SLOCK(fdp);
578 		}
579 		i++;
580 		fcnt++;
581 	}
582 	FILEDESC_SUNLOCK(fdp);
583 
584 done:
585 	uio->uio_offset = i * UIO_MX;
586 	return (error);
587 }
588 
589 static int
590 fdesc_reclaim(ap)
591 	struct vop_reclaim_args /* {
592 		struct vnode *a_vp;
593 	} */ *ap;
594 {
595 	struct vnode *vp;
596 	struct fdescnode *fd;
597 
598  	vp = ap->a_vp;
599  	fd = VTOFDESC(vp);
600 	fdesc_remove_entry(fd);
601 	free(vp->v_data, M_TEMP);
602 	vp->v_data = NULL;
603 	return (0);
604 }
605