xref: /freebsd/sys/fs/fdescfs/fdesc_vnops.c (revision 8657387683946d0c03e09fe77029edfe309eeb20)
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  * 3. 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_readlink_t	fdesc_readlink;
73 static vop_reclaim_t	fdesc_reclaim;
74 static vop_setattr_t	fdesc_setattr;
75 
76 static struct vop_vector fdesc_vnodeops = {
77 	.vop_default =		&default_vnodeops,
78 
79 	.vop_access =		VOP_NULL,
80 	.vop_getattr =		fdesc_getattr,
81 	.vop_lookup =		fdesc_lookup,
82 	.vop_open =		fdesc_open,
83 	.vop_pathconf =		vop_stdpathconf,
84 	.vop_readdir =		fdesc_readdir,
85 	.vop_readlink =		fdesc_readlink,
86 	.vop_reclaim =		fdesc_reclaim,
87 	.vop_setattr =		fdesc_setattr,
88 };
89 
90 static void fdesc_insmntque_dtr(struct vnode *, void *);
91 static void fdesc_remove_entry(struct fdescnode *);
92 
93 /*
94  * Initialise cache headers
95  */
96 int
97 fdesc_init(struct vfsconf *vfsp)
98 {
99 
100 	mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF);
101 	fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash);
102 	return (0);
103 }
104 
105 /*
106  * Uninit ready for unload.
107  */
108 int
109 fdesc_uninit(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(fdntype ftype, unsigned fd_fd, int ix, struct mount *mp,
150     struct vnode **vpp)
151 {
152 	struct fdescmount *fmp;
153 	struct fdhashhead *fc;
154 	struct fdescnode *fd, *fd2;
155 	struct vnode *vp, *vp2;
156 	struct thread *td;
157 	int error;
158 
159 	td = curthread;
160 	fc = FD_NHASH(ix);
161 loop:
162 	mtx_lock(&fdesc_hashmtx);
163 	/*
164 	 * If a forced unmount is progressing, we need to drop it. The flags are
165 	 * protected by the hashmtx.
166 	 */
167 	fmp = mp->mnt_data;
168 	if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) {
169 		mtx_unlock(&fdesc_hashmtx);
170 		return (-1);
171 	}
172 
173 	LIST_FOREACH(fd, fc, fd_hash) {
174 		if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) {
175 			/* Get reference to vnode in case it's being free'd */
176 			vp = fd->fd_vnode;
177 			VI_LOCK(vp);
178 			mtx_unlock(&fdesc_hashmtx);
179 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td))
180 				goto loop;
181 			*vpp = vp;
182 			return (0);
183 		}
184 	}
185 	mtx_unlock(&fdesc_hashmtx);
186 
187 	fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK);
188 
189 	error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp);
190 	if (error) {
191 		free(fd, M_TEMP);
192 		return (error);
193 	}
194 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
195 	vp->v_data = fd;
196 	fd->fd_vnode = vp;
197 	fd->fd_type = ftype;
198 	fd->fd_fd = fd_fd;
199 	fd->fd_ix = ix;
200 	if (ftype == Fdesc && fmp->flags & FMNT_LINRDLNKF)
201 		vp->v_vflag |= VV_READLINK;
202 	error = insmntque1(vp, mp, fdesc_insmntque_dtr, NULL);
203 	if (error != 0) {
204 		*vpp = NULLVP;
205 		return (error);
206 	}
207 
208 	/* Make sure that someone didn't beat us when inserting the vnode. */
209 	mtx_lock(&fdesc_hashmtx);
210 	/*
211 	 * If a forced unmount is progressing, we need to drop it. The flags are
212 	 * protected by the hashmtx.
213 	 */
214 	fmp = mp->mnt_data;
215 	if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) {
216 		mtx_unlock(&fdesc_hashmtx);
217 		vgone(vp);
218 		vput(vp);
219 		*vpp = NULLVP;
220 		return (-1);
221 	}
222 
223 	LIST_FOREACH(fd2, fc, fd_hash) {
224 		if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) {
225 			/* Get reference to vnode in case it's being free'd */
226 			vp2 = fd2->fd_vnode;
227 			VI_LOCK(vp2);
228 			mtx_unlock(&fdesc_hashmtx);
229 			error = vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, td);
230 			/* Someone beat us, dec use count and wait for reclaim */
231 			vgone(vp);
232 			vput(vp);
233 			/* If we didn't get it, return no vnode. */
234 			if (error)
235 				vp2 = NULLVP;
236 			*vpp = vp2;
237 			return (error);
238 		}
239 	}
240 
241 	/* If we came here, we can insert it safely. */
242 	LIST_INSERT_HEAD(fc, fd, fd_hash);
243 	mtx_unlock(&fdesc_hashmtx);
244 	*vpp = vp;
245 	return (0);
246 }
247 
248 struct fdesc_get_ino_args {
249 	fdntype ftype;
250 	unsigned fd_fd;
251 	int ix;
252 	struct file *fp;
253 	struct thread *td;
254 };
255 
256 static int
257 fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
258     struct vnode **rvp)
259 {
260 	struct fdesc_get_ino_args *a;
261 	int error;
262 
263 	a = arg;
264 	error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp);
265 	fdrop(a->fp, a->td);
266 	return (error);
267 }
268 
269 
270 /*
271  * vp is the current namei directory
272  * ndp is the name to locate in that directory...
273  */
274 static int
275 fdesc_lookup(struct vop_lookup_args *ap)
276 {
277 	struct vnode **vpp = ap->a_vpp;
278 	struct vnode *dvp = ap->a_dvp;
279 	struct componentname *cnp = ap->a_cnp;
280 	char *pname = cnp->cn_nameptr;
281 	struct thread *td = cnp->cn_thread;
282 	struct file *fp;
283 	struct fdesc_get_ino_args arg;
284 	cap_rights_t rights;
285 	int nlen = cnp->cn_namelen;
286 	u_int fd, fd1;
287 	int error;
288 	struct vnode *fvp;
289 
290 	if ((cnp->cn_flags & ISLASTCN) &&
291 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
292 		error = EROFS;
293 		goto bad;
294 	}
295 
296 	if (cnp->cn_namelen == 1 && *pname == '.') {
297 		*vpp = dvp;
298 		VREF(dvp);
299 		return (0);
300 	}
301 
302 	if (VTOFDESC(dvp)->fd_type != Froot) {
303 		error = ENOTDIR;
304 		goto bad;
305 	}
306 
307 	fd = 0;
308 	/* the only time a leading 0 is acceptable is if it's "0" */
309 	if (*pname == '0' && nlen != 1) {
310 		error = ENOENT;
311 		goto bad;
312 	}
313 	while (nlen--) {
314 		if (*pname < '0' || *pname > '9') {
315 			error = ENOENT;
316 			goto bad;
317 		}
318 		fd1 = 10 * fd + *pname++ - '0';
319 		if (fd1 < fd) {
320 			error = ENOENT;
321 			goto bad;
322 		}
323 		fd = fd1;
324 	}
325 
326 	/*
327 	 * No rights to check since 'fp' isn't actually used.
328 	 */
329 	if ((error = fget(td, fd, cap_rights_init(&rights), &fp)) != 0)
330 		goto bad;
331 
332 	/* Check if we're looking up ourselves. */
333 	if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) {
334 		/*
335 		 * In case we're holding the last reference to the file, the dvp
336 		 * will be re-acquired.
337 		 */
338 		vhold(dvp);
339 		VOP_UNLOCK(dvp, 0);
340 		fdrop(fp, td);
341 
342 		/* Re-aquire the lock afterwards. */
343 		vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE);
344 		vdrop(dvp);
345 		fvp = dvp;
346 		if ((dvp->v_iflag & VI_DOOMED) != 0)
347 			error = ENOENT;
348 	} else {
349 		/*
350 		 * Unlock our root node (dvp) when doing this, since we might
351 		 * deadlock since the vnode might be locked by another thread
352 		 * and the root vnode lock will be obtained afterwards (in case
353 		 * we're looking up the fd of the root vnode), which will be the
354 		 * opposite lock order. Vhold the root vnode first so we don't
355 		 * lose it.
356 		 */
357 		arg.ftype = Fdesc;
358 		arg.fd_fd = fd;
359 		arg.ix = FD_DESC + fd;
360 		arg.fp = fp;
361 		arg.td = td;
362 		error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg,
363 		    LK_EXCLUSIVE, &fvp);
364 	}
365 
366 	if (error)
367 		goto bad;
368 	*vpp = fvp;
369 	return (0);
370 
371 bad:
372 	*vpp = NULL;
373 	return (error);
374 }
375 
376 static int
377 fdesc_open(struct vop_open_args *ap)
378 {
379 	struct vnode *vp = ap->a_vp;
380 
381 	if (VTOFDESC(vp)->fd_type == Froot)
382 		return (0);
383 
384 	/*
385 	 * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file
386 	 * descriptor being sought for duplication. The error return ensures
387 	 * that the vnode for this device will be released by vn_open. Open
388 	 * will detect this special error and take the actions in dupfdopen.
389 	 * Other callers of vn_open or VOP_OPEN will simply report the
390 	 * error.
391 	 */
392 	ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd;	/* XXX */
393 	return (ENODEV);
394 }
395 
396 static int
397 fdesc_getattr(struct vop_getattr_args *ap)
398 {
399 	struct vnode *vp = ap->a_vp;
400 	struct vattr *vap = ap->a_vap;
401 	struct timeval boottime;
402 
403 	getboottime(&boottime);
404 	vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
405 	vap->va_fileid = VTOFDESC(vp)->fd_ix;
406 	vap->va_uid = 0;
407 	vap->va_gid = 0;
408 	vap->va_blocksize = DEV_BSIZE;
409 	vap->va_atime.tv_sec = boottime.tv_sec;
410 	vap->va_atime.tv_nsec = 0;
411 	vap->va_mtime = vap->va_atime;
412 	vap->va_ctime = vap->va_mtime;
413 	vap->va_gen = 0;
414 	vap->va_flags = 0;
415 	vap->va_bytes = 0;
416 	vap->va_filerev = 0;
417 
418 	switch (VTOFDESC(vp)->fd_type) {
419 	case Froot:
420 		vap->va_type = VDIR;
421 		vap->va_nlink = 2;
422 		vap->va_size = DEV_BSIZE;
423 		vap->va_rdev = NODEV;
424 		break;
425 
426 	case Fdesc:
427 		vap->va_type = (vp->v_vflag & VV_READLINK) == 0 ? VCHR : VLNK;
428 		vap->va_nlink = 1;
429 		vap->va_size = 0;
430 		vap->va_rdev = makedev(0, vap->va_fileid);
431 		break;
432 
433 	default:
434 		panic("fdesc_getattr");
435 		break;
436 	}
437 
438 	vp->v_type = vap->va_type;
439 	return (0);
440 }
441 
442 static int
443 fdesc_setattr(struct vop_setattr_args *ap)
444 {
445 	struct vattr *vap = ap->a_vap;
446 	struct vnode *vp;
447 	struct mount *mp;
448 	struct file *fp;
449 	struct thread *td = curthread;
450 	cap_rights_t rights;
451 	unsigned fd;
452 	int error;
453 
454 	/*
455 	 * Can't mess with the root vnode
456 	 */
457 	if (VTOFDESC(ap->a_vp)->fd_type == Froot)
458 		return (EACCES);
459 
460 	fd = VTOFDESC(ap->a_vp)->fd_fd;
461 
462 	/*
463 	 * Allow setattr where there is an underlying vnode.
464 	 */
465 	error = getvnode(td, fd,
466 	    cap_rights_init(&rights, CAP_EXTATTR_SET), &fp);
467 	if (error) {
468 		/*
469 		 * getvnode() returns EINVAL if the file descriptor is not
470 		 * backed by a vnode.  Silently drop all changes except
471 		 * chflags(2) in this case.
472 		 */
473 		if (error == EINVAL) {
474 			if (vap->va_flags != VNOVAL)
475 				error = EOPNOTSUPP;
476 			else
477 				error = 0;
478 		}
479 		return (error);
480 	}
481 	vp = fp->f_vnode;
482 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) {
483 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
484 		error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred);
485 		VOP_UNLOCK(vp, 0);
486 		vn_finished_write(mp);
487 	}
488 	fdrop(fp, td);
489 	return (error);
490 }
491 
492 #define UIO_MX _GENERIC_DIRLEN(10) /* number of symbols in INT_MAX printout */
493 
494 static int
495 fdesc_readdir(struct vop_readdir_args *ap)
496 {
497 	struct fdescmount *fmp;
498 	struct uio *uio = ap->a_uio;
499 	struct filedesc *fdp;
500 	struct dirent d;
501 	struct dirent *dp = &d;
502 	int error, i, off, fcnt;
503 
504 	if (VTOFDESC(ap->a_vp)->fd_type != Froot)
505 		panic("fdesc_readdir: not dir");
506 
507 	fmp = VFSTOFDESC(ap->a_vp->v_mount);
508 	if (ap->a_ncookies != NULL)
509 		*ap->a_ncookies = 0;
510 
511 	off = (int)uio->uio_offset;
512 	if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 ||
513 	    uio->uio_resid < UIO_MX)
514 		return (EINVAL);
515 	i = (u_int)off / UIO_MX;
516 	fdp = uio->uio_td->td_proc->p_fd;
517 	error = 0;
518 
519 	fcnt = i - 2;		/* The first two nodes are `.' and `..' */
520 
521 	FILEDESC_SLOCK(fdp);
522 	while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) {
523 		bzero((caddr_t)dp, UIO_MX);
524 		switch (i) {
525 		case 0:	/* `.' */
526 		case 1: /* `..' */
527 			dp->d_fileno = i + FD_ROOT;
528 			dp->d_namlen = i + 1;
529 			dp->d_reclen = UIO_MX;
530 			bcopy("..", dp->d_name, dp->d_namlen);
531 			dp->d_name[i + 1] = '\0';
532 			dp->d_type = DT_DIR;
533 			break;
534 		default:
535 			if (fdp->fd_ofiles[fcnt].fde_file == NULL)
536 				break;
537 			dp->d_namlen = sprintf(dp->d_name, "%d", fcnt);
538 			dp->d_reclen = UIO_MX;
539 			dp->d_type = (fmp->flags & FMNT_LINRDLNKF) == 0 ?
540 			    DT_CHR : DT_LNK;
541 			dp->d_fileno = i + FD_DESC;
542 			break;
543 		}
544 		if (dp->d_namlen != 0) {
545 			/*
546 			 * And ship to userland
547 			 */
548 			FILEDESC_SUNLOCK(fdp);
549 			error = uiomove(dp, UIO_MX, uio);
550 			if (error)
551 				goto done;
552 			FILEDESC_SLOCK(fdp);
553 		}
554 		i++;
555 		fcnt++;
556 	}
557 	FILEDESC_SUNLOCK(fdp);
558 
559 done:
560 	uio->uio_offset = i * UIO_MX;
561 	return (error);
562 }
563 
564 static int
565 fdesc_reclaim(struct vop_reclaim_args *ap)
566 {
567 	struct vnode *vp;
568 	struct fdescnode *fd;
569 
570  	vp = ap->a_vp;
571  	fd = VTOFDESC(vp);
572 	fdesc_remove_entry(fd);
573 	free(vp->v_data, M_TEMP);
574 	vp->v_data = NULL;
575 	return (0);
576 }
577 
578 static int
579 fdesc_readlink(struct vop_readlink_args *va)
580 {
581 	struct vnode *vp, *vn;
582 	cap_rights_t rights;
583 	struct thread *td;
584 	struct uio *uio;
585 	struct file *fp;
586 	char *freepath, *fullpath;
587 	size_t pathlen;
588 	int lockflags, fd_fd;
589 	int error;
590 
591 	freepath = NULL;
592 	vn = va->a_vp;
593 	if (VTOFDESC(vn)->fd_type != Fdesc)
594 		panic("fdesc_readlink: not fdescfs link");
595 	fd_fd = ((struct fdescnode *)vn->v_data)->fd_fd;
596 	lockflags = VOP_ISLOCKED(vn);
597 	VOP_UNLOCK(vn, 0);
598 
599 	td = curthread;
600 	error = fget_cap(td, fd_fd, cap_rights_init(&rights), &fp, NULL);
601 	if (error != 0)
602 		goto out;
603 
604 	switch (fp->f_type) {
605 	case DTYPE_VNODE:
606 		vp = fp->f_vnode;
607 		error = vn_fullpath(td, vp, &fullpath, &freepath);
608 		break;
609 	default:
610 		fullpath = "anon_inode:[unknown]";
611 		break;
612 	}
613 	if (error == 0) {
614 		uio = va->a_uio;
615 		pathlen = strlen(fullpath);
616 		error = uiomove(fullpath, pathlen, uio);
617 	}
618 	if (freepath != NULL)
619 		free(freepath, M_TEMP);
620 	fdrop(fp, td);
621 
622 out:
623 	vn_lock(vn, lockflags | LK_RETRY);
624 	return (error);
625 }
626