xref: /illumos-gate/usr/src/uts/common/fs/fd/fdops.c (revision 941b7e9c849240e1a04a25b32722d30dae557dc3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2017 by Delphix. All rights reserved.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All rights reserved.  	*/
28 
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/cmn_err.h>
33 #include <sys/debug.h>
34 #include <sys/dirent.h>
35 #include <sys/errno.h>
36 #include <sys/file.h>
37 #include <sys/inline.h>
38 #include <sys/kmem.h>
39 #include <sys/pathname.h>
40 #include <sys/resource.h>
41 #include <sys/statvfs.h>
42 #include <sys/mount.h>
43 #include <sys/sysmacros.h>
44 #include <sys/systm.h>
45 #include <sys/uio.h>
46 #include <sys/vfs.h>
47 #include <sys/vfs_opreg.h>
48 #include <sys/vnode.h>
49 #include <sys/cred.h>
50 #include <sys/mntent.h>
51 #include <sys/mount.h>
52 #include <sys/user.h>
53 #include <sys/t_lock.h>
54 #include <sys/modctl.h>
55 #include <sys/policy.h>
56 #include <fs/fs_subr.h>
57 #include <sys/atomic.h>
58 #include <sys/mkdev.h>
59 
60 #define	round(r)	(((r)+sizeof (int)-1)&(~(sizeof (int)-1)))
61 #define	fdtoi(n)	((n)+100)
62 
63 #define	FDDIRSIZE 14
64 struct fddirect {
65 	short	d_ino;
66 	char	d_name[FDDIRSIZE];
67 };
68 
69 #define	FDROOTINO	2
70 #define	FDSDSIZE	sizeof (struct fddirect)
71 #define	FDNSIZE		10
72 
73 static int		fdfstype = 0;
74 static major_t		fdfsmaj;
75 static minor_t		fdfsmin;
76 static major_t		fdrmaj;
77 static kmutex_t		fd_minor_lock;
78 
79 static int fdget(vnode_t *, char *, vnode_t **);
80 
81 /* ARGSUSED */
82 static int
83 fdopen(vnode_t **vpp, int mode, cred_t *cr, caller_context_t *ct)
84 {
85 	if ((*vpp)->v_type != VDIR) {
86 		mutex_enter(&(*vpp)->v_lock);
87 		(*vpp)->v_flag |= VDUP;
88 		mutex_exit(&(*vpp)->v_lock);
89 	}
90 	return (0);
91 }
92 
93 /* ARGSUSED */
94 static int
95 fdclose(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
96     caller_context_t *ct)
97 {
98 	return (0);
99 }
100 
101 /* ARGSUSED */
102 static int
103 fdread(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
104 {
105 	static struct fddirect dotbuf[] = {
106 		{ FDROOTINO, "."  },
107 		{ FDROOTINO, ".." }
108 	};
109 	struct fddirect dirbuf;
110 	int i, n;
111 	int minfd, maxfd, modoff, error = 0;
112 	int nentries;
113 	rctl_qty_t fdno_ctl;
114 	int endoff;
115 
116 	if (vp->v_type != VDIR)
117 		return (ENOSYS);
118 
119 	mutex_enter(&curproc->p_lock);
120 	fdno_ctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_NOFILE],
121 	    curproc->p_rctls, curproc);
122 	nentries = MIN(P_FINFO(curproc)->fi_nfiles, (int)fdno_ctl);
123 	mutex_exit(&curproc->p_lock);
124 
125 	endoff = (nentries + 2) * FDSDSIZE;
126 
127 	/*
128 	 * Fake up ".", "..", and the /dev/fd directory entries.
129 	 */
130 	if (uiop->uio_loffset < (offset_t)0 ||
131 	    uiop->uio_loffset >= (offset_t)endoff ||
132 	    uiop->uio_resid <= 0)
133 		return (0);
134 	ASSERT(uiop->uio_loffset <= MAXOFF_T);
135 	if (uiop->uio_offset < 2*FDSDSIZE) {
136 		error = uiomove((caddr_t)dotbuf + uiop->uio_offset,
137 		    MIN(uiop->uio_resid, 2*FDSDSIZE - uiop->uio_offset),
138 		    UIO_READ, uiop);
139 		if (uiop->uio_resid <= 0 || error)
140 			return (error);
141 	}
142 	minfd = (uiop->uio_offset - 2*FDSDSIZE)/FDSDSIZE;
143 	maxfd = (uiop->uio_offset + uiop->uio_resid - 1)/FDSDSIZE;
144 	modoff = uiop->uio_offset % FDSDSIZE;
145 
146 	for (i = 0; i < FDDIRSIZE; i++)
147 		dirbuf.d_name[i] = '\0';
148 	for (i = minfd; i < MIN(maxfd, nentries); i++) {
149 		n = i;
150 		dirbuf.d_ino = fdtoi(n);
151 		numtos((ulong_t)n, dirbuf.d_name);
152 		error = uiomove((caddr_t)&dirbuf + modoff,
153 		    MIN(uiop->uio_resid, FDSDSIZE - modoff),
154 		    UIO_READ, uiop);
155 		if (uiop->uio_resid <= 0 || error)
156 			return (error);
157 		modoff = 0;
158 	}
159 
160 	return (error);
161 }
162 
163 /* ARGSUSED */
164 static int
165 fdgetattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
166     caller_context_t *ct)
167 {
168 	vfs_t *vfsp = vp->v_vfsp;
169 	timestruc_t now;
170 
171 	if (vp->v_type == VDIR) {
172 		vap->va_nlink = 2;
173 		vap->va_size = (u_offset_t)
174 		    ((P_FINFO(curproc)->fi_nfiles + 2) * FDSDSIZE);
175 		vap->va_mode = 0555;
176 		vap->va_nodeid = (ino64_t)FDROOTINO;
177 	} else {
178 		vap->va_nlink = 1;
179 		vap->va_size = (u_offset_t)0;
180 		vap->va_mode = 0666;
181 		vap->va_nodeid = (ino64_t)fdtoi(getminor(vp->v_rdev));
182 	}
183 	vap->va_type = vp->v_type;
184 	vap->va_rdev = vp->v_rdev;
185 	vap->va_blksize = vfsp->vfs_bsize;
186 	vap->va_nblocks = (fsblkcnt64_t)0;
187 	gethrestime(&now);
188 	vap->va_atime = vap->va_mtime = vap->va_ctime = now;
189 	vap->va_uid = 0;
190 	vap->va_gid = 0;
191 	vap->va_fsid = vfsp->vfs_dev;
192 	vap->va_seq = 0;
193 	return (0);
194 }
195 
196 /* ARGSUSED */
197 static int
198 fdaccess(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
199 {
200 	return (0);
201 }
202 
203 /* ARGSUSED */
204 static int
205 fdlookup(vnode_t *dp, char *comp, vnode_t **vpp, pathname_t *pnp, int flags,
206     vnode_t *rdir, cred_t *cr, caller_context_t *ct, int *direntflags,
207     pathname_t *realpnp)
208 {
209 	if (comp[0] == 0 || strcmp(comp, ".") == 0 || strcmp(comp, "..") == 0) {
210 		VN_HOLD(dp);
211 		*vpp = dp;
212 		return (0);
213 	}
214 	return (fdget(dp, comp, vpp));
215 }
216 
217 /* ARGSUSED */
218 static int
219 fdcreate(vnode_t *dvp, char *comp, vattr_t *vap, enum vcexcl excl, int mode,
220     vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
221     vsecattr_t *vsecp)
222 {
223 	return (fdget(dvp, comp, vpp));
224 }
225 
226 /* ARGSUSED */
227 static int
228 fdreaddir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp, caller_context_t *ct,
229     int flags)
230 {
231 	/* bp holds one dirent structure */
232 	u_offset_t bp[DIRENT64_RECLEN(FDNSIZE) / sizeof (u_offset_t)];
233 	struct dirent64 *dirent = (struct dirent64 *)bp;
234 	int reclen, nentries;
235 	rctl_qty_t fdno_ctl;
236 	int  n;
237 	int oresid;
238 	off_t off;
239 
240 	if (uiop->uio_offset < 0 || uiop->uio_resid <= 0 ||
241 	    (uiop->uio_offset % FDSDSIZE) != 0)
242 		return (ENOENT);
243 
244 	ASSERT(uiop->uio_loffset <= MAXOFF_T);
245 	oresid = uiop->uio_resid;
246 	bzero(bp, sizeof (bp));
247 
248 	mutex_enter(&curproc->p_lock);
249 	fdno_ctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_NOFILE],
250 	    curproc->p_rctls, curproc);
251 	nentries = MIN(P_FINFO(curproc)->fi_nfiles, (int)fdno_ctl);
252 	mutex_exit(&curproc->p_lock);
253 
254 	while (uiop->uio_resid > 0) {
255 		if ((off = uiop->uio_offset) == 0) {	/* "." */
256 			dirent->d_ino = (ino64_t)FDROOTINO;
257 			dirent->d_name[0] = '.';
258 			dirent->d_name[1] = '\0';
259 			reclen = DIRENT64_RECLEN(1);
260 		} else if (off == FDSDSIZE) {		/* ".." */
261 			dirent->d_ino = (ino64_t)FDROOTINO;
262 			dirent->d_name[0] = '.';
263 			dirent->d_name[1] = '.';
264 			dirent->d_name[2] = '\0';
265 			reclen = DIRENT64_RECLEN(2);
266 		} else {
267 			/*
268 			 * Return entries corresponding to the allowable
269 			 * number of file descriptors for this process.
270 			 */
271 			if ((n = (off-2*FDSDSIZE)/FDSDSIZE) >= nentries)
272 				break;
273 			dirent->d_ino = (ino64_t)fdtoi(n);
274 			numtos((ulong_t)n, dirent->d_name);
275 			reclen = DIRENT64_RECLEN(strlen(dirent->d_name));
276 		}
277 		dirent->d_off = (offset_t)(uiop->uio_offset + FDSDSIZE);
278 		dirent->d_reclen = (ushort_t)reclen;
279 
280 		if (reclen > uiop->uio_resid) {
281 			/*
282 			 * Error if no entries have been returned yet.
283 			 */
284 			if (uiop->uio_resid == oresid)
285 				return (EINVAL);
286 			break;
287 		}
288 		/*
289 		 * uiomove() updates both resid and offset by the same
290 		 * amount.  But we want offset to change in increments
291 		 * of FDSDSIZE, which is different from the number of bytes
292 		 * being returned to the user.  So we set uio_offset
293 		 * separately, ignoring what uiomove() does.
294 		 */
295 		if (uiomove((caddr_t)dirent, reclen, UIO_READ, uiop))
296 			return (EFAULT);
297 		uiop->uio_offset = off + FDSDSIZE;
298 	}
299 	if (eofp)
300 		*eofp = ((uiop->uio_offset-2*FDSDSIZE)/FDSDSIZE >= nentries);
301 	return (0);
302 }
303 
304 /* ARGSUSED */
305 static void
306 fdinactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
307 {
308 	mutex_enter(&vp->v_lock);
309 	ASSERT(vp->v_count >= 1);
310 	VN_RELE_LOCKED(vp);
311 	if (vp->v_count != 0) {
312 		mutex_exit(&vp->v_lock);
313 		return;
314 	}
315 	mutex_exit(&vp->v_lock);
316 	vn_invalid(vp);
317 	vn_free(vp);
318 }
319 
320 static struct vnodeops *fd_vnodeops;
321 
322 static const fs_operation_def_t fd_vnodeops_template[] = {
323 	VOPNAME_OPEN,		{ .vop_open = fdopen },
324 	VOPNAME_CLOSE,		{ .vop_close = fdclose },
325 	VOPNAME_READ,		{ .vop_read = fdread },
326 	VOPNAME_GETATTR,	{ .vop_getattr = fdgetattr },
327 	VOPNAME_ACCESS,		{ .vop_access = fdaccess },
328 	VOPNAME_LOOKUP,		{ .vop_lookup = fdlookup },
329 	VOPNAME_CREATE,		{ .vop_create = fdcreate },
330 	VOPNAME_READDIR,	{ .vop_readdir = fdreaddir },
331 	VOPNAME_INACTIVE,	{ .vop_inactive = fdinactive },
332 	VOPNAME_FRLOCK,		{ .error = fs_error },
333 	VOPNAME_POLL,		{ .error = fs_error },
334 	VOPNAME_DISPOSE,	{ .error = fs_error },
335 	NULL,			NULL
336 };
337 
338 static int
339 fdget(struct vnode *dvp, char *comp, struct vnode **vpp)
340 {
341 	int n = 0;
342 	struct vnode *vp;
343 
344 	while (*comp) {
345 		if (*comp < '0' || *comp > '9')
346 			return (ENOENT);
347 		n = 10 * n + *comp++ - '0';
348 	}
349 	vp = vn_alloc(KM_SLEEP);
350 	vp->v_type = VCHR;
351 	vp->v_vfsp = dvp->v_vfsp;
352 	vn_setops(vp, fd_vnodeops);
353 	vp->v_data = NULL;
354 	vp->v_flag = VNOMAP;
355 	vp->v_rdev = makedevice(fdrmaj, n);
356 	vn_exists(vp);
357 	*vpp = vp;
358 	return (0);
359 }
360 
361 /*
362  * fdfs is mounted on /dev/fd, however, there are two interesting
363  * possibilities - two threads racing to do the same mount (protected
364  * by vfs locking), and two threads mounting fdfs in different places.
365  */
366 /*ARGSUSED*/
367 static int
368 fdmount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
369 {
370 	struct vnode *vp;
371 
372 	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
373 		return (EPERM);
374 	if (mvp->v_type != VDIR)
375 		return (ENOTDIR);
376 
377 	mutex_enter(&mvp->v_lock);
378 	if ((uap->flags & MS_OVERLAY) == 0 &&
379 	    (mvp->v_count > 1 || (mvp->v_flag & VROOT))) {
380 		mutex_exit(&mvp->v_lock);
381 		return (EBUSY);
382 	}
383 	mutex_exit(&mvp->v_lock);
384 
385 	/*
386 	 * Having the resource be anything but "fd" doesn't make sense
387 	 */
388 	vfs_setresource(vfsp, "fd", 0);
389 
390 	vp = vn_alloc(KM_SLEEP);
391 	vp->v_vfsp = vfsp;
392 	vn_setops(vp, fd_vnodeops);
393 	vp->v_type = VDIR;
394 	vp->v_data = NULL;
395 	vp->v_flag |= VROOT;
396 	vfsp->vfs_fstype = fdfstype;
397 	vfsp->vfs_data = (char *)vp;
398 	mutex_enter(&fd_minor_lock);
399 	do {
400 		fdfsmin = (fdfsmin + 1) & L_MAXMIN32;
401 		vfsp->vfs_dev = makedevice(fdfsmaj, fdfsmin);
402 	} while (vfs_devismounted(vfsp->vfs_dev));
403 	mutex_exit(&fd_minor_lock);
404 	vfs_make_fsid(&vfsp->vfs_fsid, vfsp->vfs_dev, fdfstype);
405 	vfsp->vfs_bsize = 1024;
406 	return (0);
407 }
408 
409 /* ARGSUSED */
410 static int
411 fdunmount(vfs_t *vfsp, int flag, cred_t *cr)
412 {
413 	vnode_t *rvp;
414 
415 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
416 		return (EPERM);
417 
418 	/*
419 	 * forced unmount is not supported by this file system
420 	 * and thus, ENOTSUP, is being returned.
421 	 */
422 	if (flag & MS_FORCE)
423 		return (ENOTSUP);
424 
425 	rvp = (vnode_t *)vfsp->vfs_data;
426 	if (rvp->v_count > 1)
427 		return (EBUSY);
428 
429 	VN_RELE(rvp);
430 	return (0);
431 }
432 
433 /* ARGSUSED */
434 static int
435 fdroot(vfs_t *vfsp, vnode_t **vpp)
436 {
437 	vnode_t *vp = (vnode_t *)vfsp->vfs_data;
438 
439 	VN_HOLD(vp);
440 	*vpp = vp;
441 	return (0);
442 }
443 
444 /*
445  * No locking required because I held the root vnode before calling this
446  * function so the vfs won't disappear on me.  To be more explicit:
447  * fdvrootp->v_count will be greater than 1 so fdunmount will just return.
448  */
449 static int
450 fdstatvfs(struct vfs *vfsp, struct statvfs64 *sp)
451 {
452 	dev32_t d32;
453 	rctl_qty_t fdno_ctl;
454 
455 	mutex_enter(&curproc->p_lock);
456 	fdno_ctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_NOFILE],
457 	    curproc->p_rctls, curproc);
458 	mutex_exit(&curproc->p_lock);
459 
460 	bzero(sp, sizeof (*sp));
461 	sp->f_bsize = 1024;
462 	sp->f_frsize = 1024;
463 	sp->f_blocks = (fsblkcnt64_t)0;
464 	sp->f_bfree = (fsblkcnt64_t)0;
465 	sp->f_bavail = (fsblkcnt64_t)0;
466 	sp->f_files = (fsfilcnt64_t)
467 	    (MIN(P_FINFO(curproc)->fi_nfiles, fdno_ctl + 2));
468 	sp->f_ffree = (fsfilcnt64_t)0;
469 	sp->f_favail = (fsfilcnt64_t)0;
470 	(void) cmpldev(&d32, vfsp->vfs_dev);
471 	sp->f_fsid = d32;
472 	(void) strcpy(sp->f_basetype, vfssw[fdfstype].vsw_name);
473 	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
474 	sp->f_namemax = FDNSIZE;
475 	(void) strcpy(sp->f_fstr, "/dev/fd");
476 	(void) strcpy(&sp->f_fstr[8], "/dev/fd");
477 	return (0);
478 }
479 
480 int
481 fdinit(int fstype, char *name)
482 {
483 	static const fs_operation_def_t fd_vfsops_template[] = {
484 		VFSNAME_MOUNT,		{ .vfs_mount = fdmount },
485 		VFSNAME_UNMOUNT,	{ .vfs_unmount = fdunmount },
486 		VFSNAME_ROOT, 		{ .vfs_root = fdroot },
487 		VFSNAME_STATVFS,	{ .vfs_statvfs = fdstatvfs },
488 		NULL,			NULL
489 	};
490 	int error;
491 
492 	fdfstype = fstype;
493 	ASSERT(fdfstype != 0);
494 
495 	/*
496 	 * Associate VFS ops vector with this fstype.
497 	 */
498 	error = vfs_setfsops(fstype, fd_vfsops_template, NULL);
499 	if (error != 0) {
500 		cmn_err(CE_WARN, "fdinit: bad vnode ops template");
501 		return (error);
502 	}
503 
504 	error = vn_make_ops(name, fd_vnodeops_template, &fd_vnodeops);
505 	if (error != 0) {
506 		(void) vfs_freevfsops_by_type(fstype);
507 		cmn_err(CE_WARN, "fdinit: bad vnode ops template");
508 		return (error);
509 	}
510 
511 	/*
512 	 * Assign unique "device" numbers (reported by stat(2)).
513 	 */
514 	fdfsmaj = getudev();
515 	fdrmaj = getudev();
516 	if (fdfsmaj == (major_t)-1 || fdrmaj == (major_t)-1) {
517 		cmn_err(CE_WARN, "fdinit: can't get unique device numbers");
518 		if (fdfsmaj == (major_t)-1)
519 			fdfsmaj = 0;
520 		if (fdrmaj == (major_t)-1)
521 			fdrmaj = 0;
522 	}
523 	mutex_init(&fd_minor_lock, NULL, MUTEX_DEFAULT, NULL);
524 	return (0);
525 }
526 
527 /*
528  * FDFS Mount options table
529  */
530 static char *rw_cancel[] = { MNTOPT_RO, NULL };
531 
532 static mntopt_t mntopts[] = {
533 /*
534  *	option name		cancel option	default arg	flags
535  */
536 	{ MNTOPT_RW,		rw_cancel,	NULL,		MO_DEFAULT,
537 		(void *)MNTOPT_NOINTR },
538 	{ MNTOPT_IGNORE,	NULL,		NULL,		0,
539 		(void *)0 },
540 };
541 
542 static mntopts_t fdfs_mntopts = {
543 	sizeof (mntopts) / sizeof (mntopt_t),
544 	mntopts
545 };
546 
547 static vfsdef_t vfw = {
548 	VFSDEF_VERSION,
549 	"fd",
550 	fdinit,
551 	VSW_HASPROTO | VSW_ZMOUNT,
552 	&fdfs_mntopts
553 };
554 
555 static struct modlfs modlfs = {
556 	&mod_fsops,
557 	"filesystem for fd",
558 	&vfw
559 };
560 
561 static struct modlinkage modlinkage = {
562 	MODREV_1,
563 	&modlfs,
564 	NULL
565 };
566 
567 int
568 _init(void)
569 {
570 	return (mod_install(&modlinkage));
571 }
572 
573 int
574 _info(struct modinfo *modinfop)
575 {
576 	return (mod_info(&modlinkage, modinfop));
577 }
578