xref: /titanic_41/usr/src/uts/common/fs/fd/fdops.c (revision 56d930ae56e5cfc2442f5214a7b2c47f08a2b920)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
29 /*	  All rights reserved.  	*/
30 
31 
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/cmn_err.h>
35 #include <sys/debug.h>
36 #include <sys/dirent.h>
37 #include <sys/errno.h>
38 #include <sys/file.h>
39 #include <sys/inline.h>
40 #include <sys/kmem.h>
41 #include <sys/pathname.h>
42 #include <sys/resource.h>
43 #include <sys/statvfs.h>
44 #include <sys/mount.h>
45 #include <sys/sysmacros.h>
46 #include <sys/systm.h>
47 #include <sys/uio.h>
48 #include <sys/vfs.h>
49 #include <sys/vfs_opreg.h>
50 #include <sys/vnode.h>
51 #include <sys/cred.h>
52 #include <sys/mntent.h>
53 #include <sys/mount.h>
54 #include <sys/user.h>
55 #include <sys/t_lock.h>
56 #include <sys/modctl.h>
57 #include <sys/policy.h>
58 #include <fs/fs_subr.h>
59 #include <sys/atomic.h>
60 #include <sys/mkdev.h>
61 
62 #define	round(r)	(((r)+sizeof (int)-1)&(~(sizeof (int)-1)))
63 #define	fdtoi(n)	((n)+100)
64 
65 #define	FDDIRSIZE 14
66 struct fddirect {
67 	short	d_ino;
68 	char	d_name[FDDIRSIZE];
69 };
70 
71 #define	FDROOTINO	2
72 #define	FDSDSIZE	sizeof (struct fddirect)
73 #define	FDNSIZE		10
74 
75 static int		fdfstype = 0;
76 static major_t		fdfsmaj;
77 static minor_t		fdfsmin;
78 static major_t		fdrmaj;
79 static kmutex_t		fd_minor_lock;
80 
81 static int fdget(vnode_t *, char *, vnode_t **);
82 
83 /* ARGSUSED */
84 static int
85 fdopen(vnode_t **vpp, int mode, cred_t *cr)
86 {
87 	if ((*vpp)->v_type != VDIR) {
88 		mutex_enter(&(*vpp)->v_lock);
89 		(*vpp)->v_flag |= VDUP;
90 		mutex_exit(&(*vpp)->v_lock);
91 	}
92 	return (0);
93 }
94 
95 /* ARGSUSED */
96 static int
97 fdclose(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
98 {
99 	return (0);
100 }
101 
102 /* ARGSUSED */
103 static int
104 fdread(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
105 {
106 	static struct fddirect dotbuf[] = {
107 		{ FDROOTINO, "."  },
108 		{ FDROOTINO, ".." }
109 	};
110 	struct fddirect dirbuf;
111 	int i, n;
112 	int minfd, maxfd, modoff, error = 0;
113 	int nentries;
114 	rctl_qty_t fdno_ctl;
115 	int endoff;
116 
117 	if (vp->v_type != VDIR)
118 		return (ENOSYS);
119 
120 	mutex_enter(&curproc->p_lock);
121 	fdno_ctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_NOFILE],
122 	    curproc->p_rctls, curproc);
123 	nentries = MIN(P_FINFO(curproc)->fi_nfiles, (int)fdno_ctl);
124 	mutex_exit(&curproc->p_lock);
125 
126 	endoff = (nentries + 2) * FDSDSIZE;
127 
128 	/*
129 	 * Fake up ".", "..", and the /dev/fd directory entries.
130 	 */
131 	if (uiop->uio_loffset < (offset_t)0 ||
132 	    uiop->uio_loffset >= (offset_t)endoff ||
133 	    uiop->uio_resid <= 0)
134 		return (0);
135 	ASSERT(uiop->uio_loffset <= MAXOFF_T);
136 	if (uiop->uio_offset < 2*FDSDSIZE) {
137 		error = uiomove((caddr_t)dotbuf + uiop->uio_offset,
138 		    MIN(uiop->uio_resid, 2*FDSDSIZE - uiop->uio_offset),
139 		    UIO_READ, uiop);
140 		if (uiop->uio_resid <= 0 || error)
141 			return (error);
142 	}
143 	minfd = (uiop->uio_offset - 2*FDSDSIZE)/FDSDSIZE;
144 	maxfd = (uiop->uio_offset + uiop->uio_resid - 1)/FDSDSIZE;
145 	modoff = uiop->uio_offset % FDSDSIZE;
146 
147 	for (i = 0; i < FDDIRSIZE; i++)
148 		dirbuf.d_name[i] = '\0';
149 	for (i = minfd; i < MIN(maxfd, nentries); i++) {
150 		n = i;
151 		dirbuf.d_ino = fdtoi(n);
152 		numtos((ulong_t)n, dirbuf.d_name);
153 		error = uiomove((caddr_t)&dirbuf + modoff,
154 		    MIN(uiop->uio_resid, FDSDSIZE - modoff),
155 		    UIO_READ, uiop);
156 		if (uiop->uio_resid <= 0 || error)
157 			return (error);
158 		modoff = 0;
159 	}
160 
161 	return (error);
162 }
163 
164 /* ARGSUSED */
165 static int
166 fdgetattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
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)
199 {
200 	return (0);
201 }
202 
203 /* ARGSUSED */
204 static int
205 fdlookup(vnode_t *dp, char *comp, vnode_t **vpp, pathname_t *pnp,
206 	int flags, vnode_t *rdir, cred_t *cr)
207 {
208 	if (comp[0] == 0 || strcmp(comp, ".") == 0 || strcmp(comp, "..") == 0) {
209 		VN_HOLD(dp);
210 		*vpp = dp;
211 		return (0);
212 	}
213 	return (fdget(dp, comp, vpp));
214 }
215 
216 /* ARGSUSED */
217 static int
218 fdcreate(vnode_t *dvp, char *comp, vattr_t *vap, enum vcexcl excl,
219 	int mode, vnode_t **vpp, cred_t *cr, int flag)
220 {
221 	return (fdget(dvp, comp, vpp));
222 }
223 
224 /* ARGSUSED */
225 static int
226 fdreaddir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp)
227 {
228 	/* bp holds one dirent structure */
229 	u_offset_t bp[DIRENT64_RECLEN(FDNSIZE) / sizeof (u_offset_t)];
230 	struct dirent64 *dirent = (struct dirent64 *)bp;
231 	int reclen, nentries;
232 	rctl_qty_t fdno_ctl;
233 	int  n;
234 	int oresid;
235 	off_t off;
236 
237 	if (uiop->uio_offset < 0 || uiop->uio_resid <= 0 ||
238 	    (uiop->uio_offset % FDSDSIZE) != 0)
239 		return (ENOENT);
240 
241 	ASSERT(uiop->uio_loffset <= MAXOFF_T);
242 	oresid = uiop->uio_resid;
243 	bzero(bp, sizeof (bp));
244 
245 	mutex_enter(&curproc->p_lock);
246 	fdno_ctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_NOFILE],
247 	    curproc->p_rctls, curproc);
248 	nentries = MIN(P_FINFO(curproc)->fi_nfiles, (int)fdno_ctl);
249 	mutex_exit(&curproc->p_lock);
250 
251 	while (uiop->uio_resid > 0) {
252 		if ((off = uiop->uio_offset) == 0) {	/* "." */
253 			dirent->d_ino = (ino64_t)FDROOTINO;
254 			dirent->d_name[0] = '.';
255 			dirent->d_name[1] = '\0';
256 			reclen = DIRENT64_RECLEN(1);
257 		} else if (off == FDSDSIZE) {		/* ".." */
258 			dirent->d_ino = (ino64_t)FDROOTINO;
259 			dirent->d_name[0] = '.';
260 			dirent->d_name[1] = '.';
261 			dirent->d_name[2] = '\0';
262 			reclen = DIRENT64_RECLEN(2);
263 		} else {
264 			/*
265 			 * Return entries corresponding to the allowable
266 			 * number of file descriptors for this process.
267 			 */
268 			if ((n = (off-2*FDSDSIZE)/FDSDSIZE) >= nentries)
269 				break;
270 			dirent->d_ino = (ino64_t)fdtoi(n);
271 			numtos((ulong_t)n, dirent->d_name);
272 			reclen = DIRENT64_RECLEN(strlen(dirent->d_name));
273 		}
274 		dirent->d_off = (offset_t)(uiop->uio_offset + FDSDSIZE);
275 		dirent->d_reclen = (ushort_t)reclen;
276 
277 		if (reclen > uiop->uio_resid) {
278 			/*
279 			 * Error if no entries have been returned yet.
280 			 */
281 			if (uiop->uio_resid == oresid)
282 				return (EINVAL);
283 			break;
284 		}
285 		/*
286 		 * uiomove() updates both resid and offset by the same
287 		 * amount.  But we want offset to change in increments
288 		 * of FDSDSIZE, which is different from the number of bytes
289 		 * being returned to the user.  So we set uio_offset
290 		 * separately, ignoring what uiomove() does.
291 		 */
292 		if (uiomove((caddr_t)dirent, reclen, UIO_READ, uiop))
293 			return (EFAULT);
294 		uiop->uio_offset = off + FDSDSIZE;
295 	}
296 	if (eofp)
297 		*eofp = ((uiop->uio_offset-2*FDSDSIZE)/FDSDSIZE >= nentries);
298 	return (0);
299 }
300 
301 /* ARGSUSED */
302 static void
303 fdinactive(vnode_t *vp, cred_t *cr)
304 {
305 	mutex_enter(&vp->v_lock);
306 	ASSERT(vp->v_count >= 1);
307 	if (--vp->v_count != 0) {
308 		mutex_exit(&vp->v_lock);
309 		return;
310 	}
311 	mutex_exit(&vp->v_lock);
312 	vn_invalid(vp);
313 	vn_free(vp);
314 }
315 
316 static struct vnodeops *fd_vnodeops;
317 
318 static const fs_operation_def_t fd_vnodeops_template[] = {
319 	VOPNAME_OPEN,		{ .vop_open = fdopen },
320 	VOPNAME_CLOSE,		{ .vop_close = fdclose },
321 	VOPNAME_READ,		{ .vop_read = fdread },
322 	VOPNAME_GETATTR,	{ .vop_getattr = fdgetattr },
323 	VOPNAME_ACCESS,		{ .vop_access = fdaccess },
324 	VOPNAME_LOOKUP,		{ .vop_lookup = fdlookup },
325 	VOPNAME_CREATE,		{ .vop_create = fdcreate },
326 	VOPNAME_READDIR,	{ .vop_readdir = fdreaddir },
327 	VOPNAME_INACTIVE,	{ .vop_inactive = fdinactive },
328 	VOPNAME_FRLOCK,		{ .error = fs_error },
329 	VOPNAME_POLL,		{ .error = fs_error },
330 	VOPNAME_DISPOSE,	{ .error = fs_error },
331 	NULL,			NULL
332 };
333 
334 static int
335 fdget(struct vnode *dvp, char *comp, struct vnode **vpp)
336 {
337 	int n = 0;
338 	struct vnode *vp;
339 
340 	while (*comp) {
341 		if (*comp < '0' || *comp > '9')
342 			return (ENOENT);
343 		n = 10 * n + *comp++ - '0';
344 	}
345 	vp = vn_alloc(KM_SLEEP);
346 	vp->v_type = VCHR;
347 	vp->v_vfsp = dvp->v_vfsp;
348 	vn_setops(vp, fd_vnodeops);
349 	vp->v_data = NULL;
350 	vp->v_flag = VNOMAP;
351 	vp->v_rdev = makedevice(fdrmaj, n);
352 	vn_exists(vp);
353 	*vpp = vp;
354 	return (0);
355 }
356 
357 /*
358  * fdfs is mounted on /dev/fd, however, there are two interesting
359  * possibilities - two threads racing to do the same mount (protected
360  * by vfs locking), and two threads mounting fdfs in different places.
361  */
362 /*ARGSUSED*/
363 static int
364 fdmount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
365 {
366 	struct vnode *vp;
367 
368 	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
369 		return (EPERM);
370 	if (mvp->v_type != VDIR)
371 		return (ENOTDIR);
372 
373 	mutex_enter(&mvp->v_lock);
374 	if ((uap->flags & MS_OVERLAY) == 0 &&
375 	    (mvp->v_count > 1 || (mvp->v_flag & VROOT))) {
376 		mutex_exit(&mvp->v_lock);
377 		return (EBUSY);
378 	}
379 	mutex_exit(&mvp->v_lock);
380 
381 	/*
382 	 * Having the resource be anything but "fd" doesn't make sense
383 	 */
384 	vfs_setresource(vfsp, "fd");
385 
386 	vp = vn_alloc(KM_SLEEP);
387 	vp->v_vfsp = vfsp;
388 	vn_setops(vp, fd_vnodeops);
389 	vp->v_type = VDIR;
390 	vp->v_data = NULL;
391 	vp->v_flag |= VROOT;
392 	vfsp->vfs_fstype = fdfstype;
393 	vfsp->vfs_data = (char *)vp;
394 	mutex_enter(&fd_minor_lock);
395 	do {
396 		fdfsmin = (fdfsmin + 1) & L_MAXMIN32;
397 		vfsp->vfs_dev = makedevice(fdfsmaj, fdfsmin);
398 	} while (vfs_devismounted(vfsp->vfs_dev));
399 	mutex_exit(&fd_minor_lock);
400 	vfs_make_fsid(&vfsp->vfs_fsid, vfsp->vfs_dev, fdfstype);
401 	vfsp->vfs_bsize = 1024;
402 	return (0);
403 }
404 
405 /* ARGSUSED */
406 static int
407 fdunmount(vfs_t *vfsp, int flag, cred_t *cr)
408 {
409 	vnode_t *rvp;
410 
411 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
412 		return (EPERM);
413 
414 	/*
415 	 * forced unmount is not supported by this file system
416 	 * and thus, ENOTSUP, is being returned.
417 	 */
418 	if (flag & MS_FORCE)
419 		return (ENOTSUP);
420 
421 	rvp = (vnode_t *)vfsp->vfs_data;
422 	if (rvp->v_count > 1)
423 		return (EBUSY);
424 
425 	VN_RELE(rvp);
426 	return (0);
427 }
428 
429 /* ARGSUSED */
430 static int
431 fdroot(vfs_t *vfsp, vnode_t **vpp)
432 {
433 	vnode_t *vp = (vnode_t *)vfsp->vfs_data;
434 
435 	VN_HOLD(vp);
436 	*vpp = vp;
437 	return (0);
438 }
439 
440 /*
441  * No locking required because I held the root vnode before calling this
442  * function so the vfs won't disappear on me.  To be more explicit:
443  * fdvrootp->v_count will be greater than 1 so fdunmount will just return.
444  */
445 static int
446 fdstatvfs(struct vfs *vfsp, struct statvfs64 *sp)
447 {
448 	dev32_t d32;
449 	rctl_qty_t fdno_ctl;
450 
451 	mutex_enter(&curproc->p_lock);
452 	fdno_ctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_NOFILE],
453 	    curproc->p_rctls, curproc);
454 	mutex_exit(&curproc->p_lock);
455 
456 	bzero(sp, sizeof (*sp));
457 	sp->f_bsize = 1024;
458 	sp->f_frsize = 1024;
459 	sp->f_blocks = (fsblkcnt64_t)0;
460 	sp->f_bfree = (fsblkcnt64_t)0;
461 	sp->f_bavail = (fsblkcnt64_t)0;
462 	sp->f_files = (fsfilcnt64_t)
463 	    (MIN(P_FINFO(curproc)->fi_nfiles, fdno_ctl + 2));
464 	sp->f_ffree = (fsfilcnt64_t)0;
465 	sp->f_favail = (fsfilcnt64_t)0;
466 	(void) cmpldev(&d32, vfsp->vfs_dev);
467 	sp->f_fsid = d32;
468 	(void) strcpy(sp->f_basetype, vfssw[fdfstype].vsw_name);
469 	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
470 	sp->f_namemax = FDNSIZE;
471 	(void) strcpy(sp->f_fstr, "/dev/fd");
472 	(void) strcpy(&sp->f_fstr[8], "/dev/fd");
473 	return (0);
474 }
475 
476 int
477 fdinit(int fstype, char *name)
478 {
479 	static const fs_operation_def_t fd_vfsops_template[] = {
480 		VFSNAME_MOUNT,		{ .vfs_mount = fdmount },
481 		VFSNAME_UNMOUNT,	{ .vfs_unmount = fdunmount },
482 		VFSNAME_ROOT, 		{ .vfs_root = fdroot },
483 		VFSNAME_STATVFS,	{ .vfs_statvfs = fdstatvfs },
484 		NULL,			NULL
485 	};
486 	int error;
487 
488 	fdfstype = fstype;
489 	ASSERT(fdfstype != 0);
490 
491 	/*
492 	 * Associate VFS ops vector with this fstype.
493 	 */
494 	error = vfs_setfsops(fstype, fd_vfsops_template, NULL);
495 	if (error != 0) {
496 		cmn_err(CE_WARN, "fdinit: bad vnode ops template");
497 		return (error);
498 	}
499 
500 	error = vn_make_ops(name, fd_vnodeops_template, &fd_vnodeops);
501 	if (error != 0) {
502 		(void) vfs_freevfsops_by_type(fstype);
503 		cmn_err(CE_WARN, "fdinit: bad vnode ops template");
504 		return (error);
505 	}
506 
507 	/*
508 	 * Assign unique "device" numbers (reported by stat(2)).
509 	 */
510 	fdfsmaj = getudev();
511 	fdrmaj = getudev();
512 	if (fdfsmaj == (major_t)-1 || fdrmaj == (major_t)-1) {
513 		cmn_err(CE_WARN, "fdinit: can't get unique device numbers");
514 		if (fdfsmaj == (major_t)-1)
515 			fdfsmaj = 0;
516 		if (fdrmaj == (major_t)-1)
517 			fdrmaj = 0;
518 	}
519 	mutex_init(&fd_minor_lock, NULL, MUTEX_DEFAULT, NULL);
520 	return (0);
521 }
522 
523 /*
524  * FDFS Mount options table
525  */
526 static char *rw_cancel[] = { MNTOPT_RO, NULL };
527 
528 static mntopt_t mntopts[] = {
529 /*
530  *	option name		cancel option	default arg	flags
531  */
532 	{ MNTOPT_RW,		rw_cancel,	NULL,		MO_DEFAULT,
533 		(void *)MNTOPT_NOINTR },
534 	{ MNTOPT_IGNORE,	NULL,		NULL,		0,
535 		(void *)0 },
536 };
537 
538 static mntopts_t fdfs_mntopts = {
539 	sizeof (mntopts) / sizeof (mntopt_t),
540 	mntopts
541 };
542 
543 static vfsdef_t vfw = {
544 	VFSDEF_VERSION,
545 	"fd",
546 	fdinit,
547 	VSW_HASPROTO,
548 	&fdfs_mntopts
549 };
550 
551 static struct modlfs modlfs = {
552 	&mod_fsops,
553 	"filesystem for fd",
554 	&vfw
555 };
556 
557 static struct modlinkage modlinkage = {
558 	MODREV_1,
559 	&modlfs,
560 	NULL
561 };
562 
563 int
564 _init(void)
565 {
566 	return (mod_install(&modlinkage));
567 }
568 
569 int
570 _info(struct modinfo *modinfop)
571 {
572 	return (mod_info(&modlinkage, modinfop));
573 }
574