xref: /freebsd/sys/fs/fdescfs/fdesc_vnops.c (revision dd41de95a84d979615a2ef11df6850622bf6184e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software donated to Berkeley by
8  * Jan-Simon Pendry.
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  *	@(#)fdesc_vnops.c	8.9 (Berkeley) 1/21/94
35  *
36  * $FreeBSD$
37  */
38 
39 /*
40  * /dev/fd Filesystem
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/capsicum.h>
46 #include <sys/conf.h>
47 #include <sys/dirent.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>	/* boottime */
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/malloc.h>
53 #include <sys/file.h>	/* Must come after sys/malloc.h */
54 #include <sys/mount.h>
55 #include <sys/namei.h>
56 #include <sys/proc.h>
57 #include <sys/stat.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/unistd.h>
60 #include <sys/vnode.h>
61 
62 #include <fs/fdescfs/fdesc.h>
63 
64 #define	NFDCACHE 4
65 #define FD_NHASH(ix) \
66 	(&fdhashtbl[(ix) & fdhash])
67 static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl;
68 static u_long fdhash;
69 
70 struct mtx fdesc_hashmtx;
71 
72 static vop_getattr_t	fdesc_getattr;
73 static vop_lookup_t	fdesc_lookup;
74 static vop_open_t	fdesc_open;
75 static vop_pathconf_t	fdesc_pathconf;
76 static vop_readdir_t	fdesc_readdir;
77 static vop_readlink_t	fdesc_readlink;
78 static vop_reclaim_t	fdesc_reclaim;
79 static vop_setattr_t	fdesc_setattr;
80 
81 static struct vop_vector fdesc_vnodeops = {
82 	.vop_default =		&default_vnodeops,
83 
84 	.vop_access =		VOP_NULL,
85 	.vop_getattr =		fdesc_getattr,
86 	.vop_lookup =		fdesc_lookup,
87 	.vop_open =		fdesc_open,
88 	.vop_pathconf =		fdesc_pathconf,
89 	.vop_readdir =		fdesc_readdir,
90 	.vop_readlink =		fdesc_readlink,
91 	.vop_reclaim =		fdesc_reclaim,
92 	.vop_setattr =		fdesc_setattr,
93 };
94 VFS_VOP_VECTOR_REGISTER(fdesc_vnodeops);
95 
96 static void fdesc_insmntque_dtr(struct vnode *, void *);
97 static void fdesc_remove_entry(struct fdescnode *);
98 
99 /*
100  * Initialise cache headers
101  */
102 int
103 fdesc_init(struct vfsconf *vfsp)
104 {
105 
106 	mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF);
107 	fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash);
108 	return (0);
109 }
110 
111 /*
112  * Uninit ready for unload.
113  */
114 int
115 fdesc_uninit(struct vfsconf *vfsp)
116 {
117 
118 	hashdestroy(fdhashtbl, M_CACHE, fdhash);
119 	mtx_destroy(&fdesc_hashmtx);
120 	return (0);
121 }
122 
123 /*
124  * If allocating vnode fails, call this.
125  */
126 static void
127 fdesc_insmntque_dtr(struct vnode *vp, void *arg)
128 {
129 
130 	vgone(vp);
131 	vput(vp);
132 }
133 
134 /*
135  * Remove an entry from the hash if it exists.
136  */
137 static void
138 fdesc_remove_entry(struct fdescnode *fd)
139 {
140 	struct fdhashhead *fc;
141 	struct fdescnode *fd2;
142 
143 	fc = FD_NHASH(fd->fd_ix);
144 	mtx_lock(&fdesc_hashmtx);
145 	LIST_FOREACH(fd2, fc, fd_hash) {
146 		if (fd == fd2) {
147 			LIST_REMOVE(fd, fd_hash);
148 			break;
149 		}
150 	}
151 	mtx_unlock(&fdesc_hashmtx);
152 }
153 
154 int
155 fdesc_allocvp(fdntype ftype, unsigned fd_fd, int ix, struct mount *mp,
156     struct vnode **vpp)
157 {
158 	struct fdescmount *fmp;
159 	struct fdhashhead *fc;
160 	struct fdescnode *fd, *fd2;
161 	struct vnode *vp, *vp2;
162 	struct thread *td;
163 	int error;
164 
165 	td = curthread;
166 	fc = FD_NHASH(ix);
167 loop:
168 	mtx_lock(&fdesc_hashmtx);
169 	/*
170 	 * If a forced unmount is progressing, we need to drop it. The flags are
171 	 * protected by the hashmtx.
172 	 */
173 	fmp = mp->mnt_data;
174 	if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) {
175 		mtx_unlock(&fdesc_hashmtx);
176 		return (-1);
177 	}
178 
179 	LIST_FOREACH(fd, fc, fd_hash) {
180 		if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) {
181 			/* Get reference to vnode in case it's being free'd */
182 			vp = fd->fd_vnode;
183 			VI_LOCK(vp);
184 			mtx_unlock(&fdesc_hashmtx);
185 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK))
186 				goto loop;
187 			*vpp = vp;
188 			return (0);
189 		}
190 	}
191 	mtx_unlock(&fdesc_hashmtx);
192 
193 	fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK);
194 
195 	error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp);
196 	if (error) {
197 		free(fd, M_TEMP);
198 		return (error);
199 	}
200 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
201 	vp->v_data = fd;
202 	fd->fd_vnode = vp;
203 	fd->fd_type = ftype;
204 	fd->fd_fd = fd_fd;
205 	fd->fd_ix = ix;
206 	if (ftype == Fdesc && fmp->flags & FMNT_LINRDLNKF)
207 		vp->v_vflag |= VV_READLINK;
208 	error = insmntque1(vp, mp, fdesc_insmntque_dtr, NULL);
209 	if (error != 0) {
210 		*vpp = NULLVP;
211 		return (error);
212 	}
213 
214 	/* Make sure that someone didn't beat us when inserting the vnode. */
215 	mtx_lock(&fdesc_hashmtx);
216 	/*
217 	 * If a forced unmount is progressing, we need to drop it. The flags are
218 	 * protected by the hashmtx.
219 	 */
220 	fmp = mp->mnt_data;
221 	if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) {
222 		mtx_unlock(&fdesc_hashmtx);
223 		vgone(vp);
224 		vput(vp);
225 		*vpp = NULLVP;
226 		return (-1);
227 	}
228 
229 	LIST_FOREACH(fd2, fc, fd_hash) {
230 		if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) {
231 			/* Get reference to vnode in case it's being free'd */
232 			vp2 = fd2->fd_vnode;
233 			VI_LOCK(vp2);
234 			mtx_unlock(&fdesc_hashmtx);
235 			error = vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK);
236 			/* Someone beat us, dec use count and wait for reclaim */
237 			vgone(vp);
238 			vput(vp);
239 			/* If we didn't get it, return no vnode. */
240 			if (error)
241 				vp2 = NULLVP;
242 			*vpp = vp2;
243 			return (error);
244 		}
245 	}
246 
247 	/* If we came here, we can insert it safely. */
248 	LIST_INSERT_HEAD(fc, fd, fd_hash);
249 	mtx_unlock(&fdesc_hashmtx);
250 	*vpp = vp;
251 	return (0);
252 }
253 
254 struct fdesc_get_ino_args {
255 	fdntype ftype;
256 	unsigned fd_fd;
257 	int ix;
258 	struct file *fp;
259 	struct thread *td;
260 };
261 
262 static int
263 fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
264     struct vnode **rvp)
265 {
266 	struct fdesc_get_ino_args *a;
267 	int error;
268 
269 	a = arg;
270 	error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp);
271 	fdrop(a->fp, a->td);
272 	return (error);
273 }
274 
275 /*
276  * vp is the current namei directory
277  * ndp is the name to locate in that directory...
278  */
279 static int
280 fdesc_lookup(struct vop_lookup_args *ap)
281 {
282 	struct vnode **vpp = ap->a_vpp;
283 	struct vnode *dvp = ap->a_dvp;
284 	struct componentname *cnp = ap->a_cnp;
285 	char *pname = cnp->cn_nameptr;
286 	struct thread *td = cnp->cn_thread;
287 	struct file *fp;
288 	struct fdesc_get_ino_args arg;
289 	int nlen = cnp->cn_namelen;
290 	u_int fd, fd1;
291 	int error;
292 	struct vnode *fvp;
293 
294 	if ((cnp->cn_flags & ISLASTCN) &&
295 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
296 		error = EROFS;
297 		goto bad;
298 	}
299 
300 	if (cnp->cn_namelen == 1 && *pname == '.') {
301 		*vpp = dvp;
302 		VREF(dvp);
303 		return (0);
304 	}
305 
306 	if (VTOFDESC(dvp)->fd_type != Froot) {
307 		error = ENOTDIR;
308 		goto bad;
309 	}
310 
311 	fd = 0;
312 	/* the only time a leading 0 is acceptable is if it's "0" */
313 	if (*pname == '0' && nlen != 1) {
314 		error = ENOENT;
315 		goto bad;
316 	}
317 	while (nlen--) {
318 		if (*pname < '0' || *pname > '9') {
319 			error = ENOENT;
320 			goto bad;
321 		}
322 		fd1 = 10 * fd + *pname++ - '0';
323 		if (fd1 < fd) {
324 			error = ENOENT;
325 			goto bad;
326 		}
327 		fd = fd1;
328 	}
329 
330 	/*
331 	 * No rights to check since 'fp' isn't actually used.
332 	 */
333 	if ((error = fget(td, fd, &cap_no_rights, &fp)) != 0)
334 		goto bad;
335 
336 	/* Check if we're looking up ourselves. */
337 	if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) {
338 		/*
339 		 * In case we're holding the last reference to the file, the dvp
340 		 * will be re-acquired.
341 		 */
342 		vhold(dvp);
343 		VOP_UNLOCK(dvp);
344 		fdrop(fp, td);
345 
346 		/* Re-aquire the lock afterwards. */
347 		vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE);
348 		vdrop(dvp);
349 		fvp = dvp;
350 		if (VN_IS_DOOMED(dvp))
351 			error = ENOENT;
352 	} else {
353 		/*
354 		 * Unlock our root node (dvp) when doing this, since we might
355 		 * deadlock since the vnode might be locked by another thread
356 		 * and the root vnode lock will be obtained afterwards (in case
357 		 * we're looking up the fd of the root vnode), which will be the
358 		 * opposite lock order. Vhold the root vnode first so we don't
359 		 * lose it.
360 		 */
361 		arg.ftype = Fdesc;
362 		arg.fd_fd = fd;
363 		arg.ix = FD_DESC + fd;
364 		arg.fp = fp;
365 		arg.td = td;
366 		error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg,
367 		    LK_EXCLUSIVE, &fvp);
368 	}
369 
370 	if (error)
371 		goto bad;
372 	*vpp = fvp;
373 	return (0);
374 
375 bad:
376 	*vpp = NULL;
377 	return (error);
378 }
379 
380 static int
381 fdesc_open(struct vop_open_args *ap)
382 {
383 	struct vnode *vp = ap->a_vp;
384 
385 	if (VTOFDESC(vp)->fd_type == Froot)
386 		return (0);
387 
388 	/*
389 	 * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file
390 	 * descriptor being sought for duplication. The error return ensures
391 	 * that the vnode for this device will be released by vn_open. Open
392 	 * will detect this special error and take the actions in dupfdopen.
393 	 * Other callers of vn_open or VOP_OPEN will simply report the
394 	 * error.
395 	 */
396 	ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd;	/* XXX */
397 	return (ENODEV);
398 }
399 
400 static int
401 fdesc_pathconf(struct vop_pathconf_args *ap)
402 {
403 	struct vnode *vp = ap->a_vp;
404 	int error;
405 
406 	switch (ap->a_name) {
407 	case _PC_NAME_MAX:
408 		*ap->a_retval = NAME_MAX;
409 		return (0);
410 	case _PC_LINK_MAX:
411 		if (VTOFDESC(vp)->fd_type == Froot)
412 			*ap->a_retval = 2;
413 		else
414 			*ap->a_retval = 1;
415 		return (0);
416 	default:
417 		if (VTOFDESC(vp)->fd_type == Froot)
418 			return (vop_stdpathconf(ap));
419 		vref(vp);
420 		VOP_UNLOCK(vp);
421 		error = kern_fpathconf(curthread, VTOFDESC(vp)->fd_fd,
422 		    ap->a_name, ap->a_retval);
423 		vn_lock(vp, LK_SHARED | LK_RETRY);
424 		vunref(vp);
425 		return (error);
426 	}
427 }
428 
429 static int
430 fdesc_getattr(struct vop_getattr_args *ap)
431 {
432 	struct vnode *vp = ap->a_vp;
433 	struct vattr *vap = ap->a_vap;
434 	struct timeval boottime;
435 
436 	getboottime(&boottime);
437 	vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
438 	vap->va_fileid = VTOFDESC(vp)->fd_ix;
439 	vap->va_uid = 0;
440 	vap->va_gid = 0;
441 	vap->va_blocksize = DEV_BSIZE;
442 	vap->va_atime.tv_sec = boottime.tv_sec;
443 	vap->va_atime.tv_nsec = 0;
444 	vap->va_mtime = vap->va_atime;
445 	vap->va_ctime = vap->va_mtime;
446 	vap->va_gen = 0;
447 	vap->va_flags = 0;
448 	vap->va_bytes = 0;
449 	vap->va_filerev = 0;
450 
451 	switch (VTOFDESC(vp)->fd_type) {
452 	case Froot:
453 		vap->va_type = VDIR;
454 		vap->va_nlink = 2;
455 		vap->va_size = DEV_BSIZE;
456 		vap->va_rdev = NODEV;
457 		break;
458 
459 	case Fdesc:
460 		vap->va_type = (vp->v_vflag & VV_READLINK) == 0 ? VCHR : VLNK;
461 		vap->va_nlink = 1;
462 		vap->va_size = 0;
463 		vap->va_rdev = makedev(0, vap->va_fileid);
464 		break;
465 
466 	default:
467 		panic("fdesc_getattr");
468 		break;
469 	}
470 
471 	vp->v_type = vap->va_type;
472 	return (0);
473 }
474 
475 static int
476 fdesc_setattr(struct vop_setattr_args *ap)
477 {
478 	struct vattr *vap = ap->a_vap;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	struct file *fp;
482 	struct thread *td = curthread;
483 	cap_rights_t rights;
484 	unsigned fd;
485 	int error;
486 
487 	/*
488 	 * Can't mess with the root vnode
489 	 */
490 	if (VTOFDESC(ap->a_vp)->fd_type == Froot)
491 		return (EACCES);
492 
493 	fd = VTOFDESC(ap->a_vp)->fd_fd;
494 
495 	/*
496 	 * Allow setattr where there is an underlying vnode.
497 	 */
498 	error = getvnode(td, fd,
499 	    cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp);
500 	if (error) {
501 		/*
502 		 * getvnode() returns EINVAL if the file descriptor is not
503 		 * backed by a vnode.  Silently drop all changes except
504 		 * chflags(2) in this case.
505 		 */
506 		if (error == EINVAL) {
507 			if (vap->va_flags != VNOVAL)
508 				error = EOPNOTSUPP;
509 			else
510 				error = 0;
511 		}
512 		return (error);
513 	}
514 	vp = fp->f_vnode;
515 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) {
516 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
517 		error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred);
518 		VOP_UNLOCK(vp);
519 		vn_finished_write(mp);
520 	}
521 	fdrop(fp, td);
522 	return (error);
523 }
524 
525 #define UIO_MX _GENERIC_DIRLEN(10) /* number of symbols in INT_MAX printout */
526 
527 static int
528 fdesc_readdir(struct vop_readdir_args *ap)
529 {
530 	struct fdescmount *fmp;
531 	struct uio *uio = ap->a_uio;
532 	struct filedesc *fdp;
533 	struct dirent d;
534 	struct dirent *dp = &d;
535 	int error, i, off, fcnt;
536 
537 	if (VTOFDESC(ap->a_vp)->fd_type != Froot)
538 		panic("fdesc_readdir: not dir");
539 
540 	fmp = VFSTOFDESC(ap->a_vp->v_mount);
541 	if (ap->a_ncookies != NULL)
542 		*ap->a_ncookies = 0;
543 
544 	off = (int)uio->uio_offset;
545 	if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 ||
546 	    uio->uio_resid < UIO_MX)
547 		return (EINVAL);
548 	i = (u_int)off / UIO_MX;
549 	fdp = uio->uio_td->td_proc->p_fd;
550 	error = 0;
551 
552 	fcnt = i - 2;		/* The first two nodes are `.' and `..' */
553 
554 	FILEDESC_SLOCK(fdp);
555 	while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) {
556 		bzero((caddr_t)dp, UIO_MX);
557 		switch (i) {
558 		case 0:	/* `.' */
559 		case 1: /* `..' */
560 			dp->d_fileno = i + FD_ROOT;
561 			dp->d_namlen = i + 1;
562 			dp->d_reclen = UIO_MX;
563 			bcopy("..", dp->d_name, dp->d_namlen);
564 			dp->d_type = DT_DIR;
565 			dirent_terminate(dp);
566 			break;
567 		default:
568 			if (fdp->fd_ofiles[fcnt].fde_file == NULL)
569 				break;
570 			dp->d_namlen = sprintf(dp->d_name, "%d", fcnt);
571 			dp->d_reclen = UIO_MX;
572 			dp->d_type = (fmp->flags & FMNT_LINRDLNKF) == 0 ?
573 			    DT_CHR : DT_LNK;
574 			dp->d_fileno = i + FD_DESC;
575 			dirent_terminate(dp);
576 			break;
577 		}
578 		/* NOTE: d_off is the offset of the *next* entry. */
579 		dp->d_off = UIO_MX * (i + 1);
580 		if (dp->d_namlen != 0) {
581 			/*
582 			 * And ship to userland
583 			 */
584 			FILEDESC_SUNLOCK(fdp);
585 			error = uiomove(dp, UIO_MX, uio);
586 			if (error)
587 				goto done;
588 			FILEDESC_SLOCK(fdp);
589 		}
590 		i++;
591 		fcnt++;
592 	}
593 	FILEDESC_SUNLOCK(fdp);
594 
595 done:
596 	uio->uio_offset = i * UIO_MX;
597 	return (error);
598 }
599 
600 static int
601 fdesc_reclaim(struct vop_reclaim_args *ap)
602 {
603 	struct vnode *vp;
604 	struct fdescnode *fd;
605 
606  	vp = ap->a_vp;
607  	fd = VTOFDESC(vp);
608 	fdesc_remove_entry(fd);
609 	free(vp->v_data, M_TEMP);
610 	vp->v_data = NULL;
611 	return (0);
612 }
613 
614 static int
615 fdesc_readlink(struct vop_readlink_args *va)
616 {
617 	struct vnode *vp, *vn;
618 	struct thread *td;
619 	struct uio *uio;
620 	struct file *fp;
621 	char *freepath, *fullpath;
622 	size_t pathlen;
623 	int lockflags, fd_fd;
624 	int error;
625 
626 	freepath = NULL;
627 	vn = va->a_vp;
628 	if (VTOFDESC(vn)->fd_type != Fdesc)
629 		panic("fdesc_readlink: not fdescfs link");
630 	fd_fd = ((struct fdescnode *)vn->v_data)->fd_fd;
631 	lockflags = VOP_ISLOCKED(vn);
632 	VOP_UNLOCK(vn);
633 
634 	td = curthread;
635 	error = fget_cap(td, fd_fd, &cap_no_rights, &fp, NULL);
636 	if (error != 0)
637 		goto out;
638 
639 	switch (fp->f_type) {
640 	case DTYPE_VNODE:
641 		vp = fp->f_vnode;
642 		error = vn_fullpath(vp, &fullpath, &freepath);
643 		break;
644 	default:
645 		fullpath = "anon_inode:[unknown]";
646 		break;
647 	}
648 	if (error == 0) {
649 		uio = va->a_uio;
650 		pathlen = strlen(fullpath);
651 		error = uiomove(fullpath, pathlen, uio);
652 	}
653 	if (freepath != NULL)
654 		free(freepath, M_TEMP);
655 	fdrop(fp, td);
656 
657 out:
658 	vn_lock(vn, lockflags | LK_RETRY);
659 	return (error);
660 }
661