xref: /freebsd-13-stable/sys/kern/uipc_mqueue.c (revision 6446348a4f37d49fe4c674ffe2d81a93b0dca0b5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2005 David Xu <davidxu@freebsd.org>
5  * Copyright (c) 2016-2017 Robert N. M. Watson
6  * All rights reserved.
7  *
8  * Portions of this software were developed by BAE Systems, the University of
9  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
10  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
11  * Computing (TC) research program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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  */
35 
36 /*
37  * POSIX message queue implementation.
38  *
39  * 1) A mqueue filesystem can be mounted, each message queue appears
40  *    in mounted directory, user can change queue's permission and
41  *    ownership, or remove a queue. Manually creating a file in the
42  *    directory causes a message queue to be created in the kernel with
43  *    default message queue attributes applied and same name used, this
44  *    method is not advocated since mq_open syscall allows user to specify
45  *    different attributes. Also the file system can be mounted multiple
46  *    times at different mount points but shows same contents.
47  *
48  * 2) Standard POSIX message queue API. The syscalls do not use vfs layer,
49  *    but directly operate on internal data structure, this allows user to
50  *    use the IPC facility without having to mount mqueue file system.
51  */
52 
53 #include "opt_capsicum.h"
54 
55 #include <sys/param.h>
56 #include <sys/kernel.h>
57 #include <sys/systm.h>
58 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/buf.h>
61 #include <sys/capsicum.h>
62 #include <sys/dirent.h>
63 #include <sys/event.h>
64 #include <sys/eventhandler.h>
65 #include <sys/fcntl.h>
66 #include <sys/file.h>
67 #include <sys/filedesc.h>
68 #include <sys/jail.h>
69 #include <sys/lock.h>
70 #include <sys/module.h>
71 #include <sys/mount.h>
72 #include <sys/mqueue.h>
73 #include <sys/mutex.h>
74 #include <sys/namei.h>
75 #include <sys/posix4.h>
76 #include <sys/poll.h>
77 #include <sys/priv.h>
78 #include <sys/proc.h>
79 #include <sys/queue.h>
80 #include <sys/sysproto.h>
81 #include <sys/stat.h>
82 #include <sys/syscall.h>
83 #include <sys/syscallsubr.h>
84 #include <sys/sysent.h>
85 #include <sys/sx.h>
86 #include <sys/sysctl.h>
87 #include <sys/taskqueue.h>
88 #include <sys/unistd.h>
89 #include <sys/user.h>
90 #include <sys/vnode.h>
91 #include <machine/atomic.h>
92 
93 #include <security/audit/audit.h>
94 
95 FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support");
96 
97 /*
98  * Limits and constants
99  */
100 #define	MQFS_NAMELEN		NAME_MAX
101 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
102 
103 /* node types */
104 typedef enum {
105 	mqfstype_none = 0,
106 	mqfstype_root,
107 	mqfstype_dir,
108 	mqfstype_this,
109 	mqfstype_parent,
110 	mqfstype_file,
111 	mqfstype_symlink,
112 } mqfs_type_t;
113 
114 struct mqfs_node;
115 
116 /*
117  * mqfs_info: describes a mqfs instance
118  */
119 struct mqfs_info {
120 	struct sx		mi_lock;
121 	struct mqfs_node	*mi_root;
122 	struct unrhdr		*mi_unrhdr;
123 };
124 
125 struct mqfs_vdata {
126 	LIST_ENTRY(mqfs_vdata)	mv_link;
127 	struct mqfs_node	*mv_node;
128 	struct vnode		*mv_vnode;
129 	struct task		mv_task;
130 };
131 
132 /*
133  * mqfs_node: describes a node (file or directory) within a mqfs
134  */
135 struct mqfs_node {
136 	char			mn_name[MQFS_NAMELEN+1];
137 	struct mqfs_info	*mn_info;
138 	struct mqfs_node	*mn_parent;
139 	LIST_HEAD(,mqfs_node)	mn_children;
140 	LIST_ENTRY(mqfs_node)	mn_sibling;
141 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
142 	const void		*mn_pr_root;
143 	int			mn_refcount;
144 	mqfs_type_t		mn_type;
145 	int			mn_deleted;
146 	uint32_t		mn_fileno;
147 	void			*mn_data;
148 	struct timespec		mn_birth;
149 	struct timespec		mn_ctime;
150 	struct timespec		mn_atime;
151 	struct timespec		mn_mtime;
152 	uid_t			mn_uid;
153 	gid_t			mn_gid;
154 	int			mn_mode;
155 };
156 
157 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
158 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
159 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
160 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
161 				(fp)->f_data)->mn_data))
162 
163 TAILQ_HEAD(msgq, mqueue_msg);
164 
165 struct mqueue;
166 
167 struct mqueue_notifier {
168 	LIST_ENTRY(mqueue_notifier)	nt_link;
169 	struct sigevent			nt_sigev;
170 	ksiginfo_t			nt_ksi;
171 	struct proc			*nt_proc;
172 };
173 
174 struct mqueue {
175 	struct mtx	mq_mutex;
176 	int		mq_flags;
177 	long		mq_maxmsg;
178 	long		mq_msgsize;
179 	long		mq_curmsgs;
180 	long		mq_totalbytes;
181 	struct msgq	mq_msgq;
182 	int		mq_receivers;
183 	int		mq_senders;
184 	struct selinfo	mq_rsel;
185 	struct selinfo	mq_wsel;
186 	struct mqueue_notifier	*mq_notifier;
187 };
188 
189 #define	MQ_RSEL		0x01
190 #define	MQ_WSEL		0x02
191 
192 struct mqueue_msg {
193 	TAILQ_ENTRY(mqueue_msg)	msg_link;
194 	unsigned int	msg_prio;
195 	unsigned int	msg_size;
196 	/* following real data... */
197 };
198 
199 static SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
200 	"POSIX real time message queue");
201 
202 static int	default_maxmsg  = 10;
203 static int	default_msgsize = 1024;
204 
205 static int	maxmsg = 100;
206 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
207     &maxmsg, 0, "Default maximum messages in queue");
208 static int	maxmsgsize = 16384;
209 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
210     &maxmsgsize, 0, "Default maximum message size");
211 static int	maxmq = 100;
212 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
213     &maxmq, 0, "maximum message queues");
214 static int	curmq = 0;
215 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
216     &curmq, 0, "current message queue number");
217 static int	unloadable = 0;
218 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
219 
220 static eventhandler_tag exit_tag;
221 
222 /* Only one instance per-system */
223 static struct mqfs_info		mqfs_data;
224 static uma_zone_t		mqnode_zone;
225 static uma_zone_t		mqueue_zone;
226 static uma_zone_t		mvdata_zone;
227 static uma_zone_t		mqnoti_zone;
228 static struct vop_vector	mqfs_vnodeops;
229 static struct fileops		mqueueops;
230 static unsigned			mqfs_osd_jail_slot;
231 
232 /*
233  * Directory structure construction and manipulation
234  */
235 #ifdef notyet
236 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
237 	const char *name, int namelen, struct ucred *cred, int mode);
238 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
239 	const char *name, int namelen, struct ucred *cred, int mode);
240 #endif
241 
242 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
243 	const char *name, int namelen, struct ucred *cred, int mode);
244 static int	mqfs_destroy(struct mqfs_node *mn);
245 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
246 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
247 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
248 static int	mqfs_prison_remove(void *obj, void *data);
249 
250 /*
251  * Message queue construction and maniplation
252  */
253 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
254 static void	mqueue_free(struct mqueue *mq);
255 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
256 			size_t msg_len, unsigned msg_prio, int waitok,
257 			const struct timespec *abs_timeout);
258 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
259 			size_t msg_len, unsigned *msg_prio, int waitok,
260 			const struct timespec *abs_timeout);
261 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
262 			int timo);
263 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
264 			int timo);
265 static void	mqueue_send_notification(struct mqueue *mq);
266 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
267 static void	mq_proc_exit(void *arg, struct proc *p);
268 
269 /*
270  * kqueue filters
271  */
272 static void	filt_mqdetach(struct knote *kn);
273 static int	filt_mqread(struct knote *kn, long hint);
274 static int	filt_mqwrite(struct knote *kn, long hint);
275 
276 struct filterops mq_rfiltops = {
277 	.f_isfd = 1,
278 	.f_detach = filt_mqdetach,
279 	.f_event = filt_mqread,
280 };
281 struct filterops mq_wfiltops = {
282 	.f_isfd = 1,
283 	.f_detach = filt_mqdetach,
284 	.f_event = filt_mqwrite,
285 };
286 
287 /*
288  * Initialize fileno bitmap
289  */
290 static void
mqfs_fileno_init(struct mqfs_info * mi)291 mqfs_fileno_init(struct mqfs_info *mi)
292 {
293 	struct unrhdr *up;
294 
295 	up = new_unrhdr(1, INT_MAX, NULL);
296 	mi->mi_unrhdr = up;
297 }
298 
299 /*
300  * Tear down fileno bitmap
301  */
302 static void
mqfs_fileno_uninit(struct mqfs_info * mi)303 mqfs_fileno_uninit(struct mqfs_info *mi)
304 {
305 	struct unrhdr *up;
306 
307 	up = mi->mi_unrhdr;
308 	mi->mi_unrhdr = NULL;
309 	delete_unrhdr(up);
310 }
311 
312 /*
313  * Allocate a file number
314  */
315 static void
mqfs_fileno_alloc(struct mqfs_info * mi,struct mqfs_node * mn)316 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
317 {
318 	/* make sure our parent has a file number */
319 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
320 		mqfs_fileno_alloc(mi, mn->mn_parent);
321 
322 	switch (mn->mn_type) {
323 	case mqfstype_root:
324 	case mqfstype_dir:
325 	case mqfstype_file:
326 	case mqfstype_symlink:
327 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
328 		break;
329 	case mqfstype_this:
330 		KASSERT(mn->mn_parent != NULL,
331 		    ("mqfstype_this node has no parent"));
332 		mn->mn_fileno = mn->mn_parent->mn_fileno;
333 		break;
334 	case mqfstype_parent:
335 		KASSERT(mn->mn_parent != NULL,
336 		    ("mqfstype_parent node has no parent"));
337 		if (mn->mn_parent == mi->mi_root) {
338 			mn->mn_fileno = mn->mn_parent->mn_fileno;
339 			break;
340 		}
341 		KASSERT(mn->mn_parent->mn_parent != NULL,
342 		    ("mqfstype_parent node has no grandparent"));
343 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
344 		break;
345 	default:
346 		KASSERT(0,
347 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
348 			mn->mn_type));
349 		break;
350 	}
351 }
352 
353 /*
354  * Release a file number
355  */
356 static void
mqfs_fileno_free(struct mqfs_info * mi,struct mqfs_node * mn)357 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
358 {
359 	switch (mn->mn_type) {
360 	case mqfstype_root:
361 	case mqfstype_dir:
362 	case mqfstype_file:
363 	case mqfstype_symlink:
364 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
365 		break;
366 	case mqfstype_this:
367 	case mqfstype_parent:
368 		/* ignore these, as they don't "own" their file number */
369 		break;
370 	default:
371 		KASSERT(0,
372 		    ("mqfs_fileno_free() called for unknown type node: %d",
373 			mn->mn_type));
374 		break;
375 	}
376 }
377 
378 static __inline struct mqfs_node *
mqnode_alloc(void)379 mqnode_alloc(void)
380 {
381 	return (uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO));
382 }
383 
384 static __inline void
mqnode_free(struct mqfs_node * node)385 mqnode_free(struct mqfs_node *node)
386 {
387 	uma_zfree(mqnode_zone, node);
388 }
389 
390 static __inline void
mqnode_addref(struct mqfs_node * node)391 mqnode_addref(struct mqfs_node *node)
392 {
393 	atomic_add_int(&node->mn_refcount, 1);
394 }
395 
396 static __inline void
mqnode_release(struct mqfs_node * node)397 mqnode_release(struct mqfs_node *node)
398 {
399 	struct mqfs_info *mqfs;
400 	int old, exp;
401 
402 	mqfs = node->mn_info;
403 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
404 	if (node->mn_type == mqfstype_dir ||
405 	    node->mn_type == mqfstype_root)
406 		exp = 3; /* include . and .. */
407 	else
408 		exp = 1;
409 	if (old == exp) {
410 		int locked = sx_xlocked(&mqfs->mi_lock);
411 		if (!locked)
412 			sx_xlock(&mqfs->mi_lock);
413 		mqfs_destroy(node);
414 		if (!locked)
415 			sx_xunlock(&mqfs->mi_lock);
416 	}
417 }
418 
419 /*
420  * Add a node to a directory
421  */
422 static int
mqfs_add_node(struct mqfs_node * parent,struct mqfs_node * node)423 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
424 {
425 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
426 	KASSERT(parent->mn_info != NULL,
427 	    ("%s(): parent has no mn_info", __func__));
428 	KASSERT(parent->mn_type == mqfstype_dir ||
429 	    parent->mn_type == mqfstype_root,
430 	    ("%s(): parent is not a directory", __func__));
431 
432 	node->mn_info = parent->mn_info;
433 	node->mn_parent = parent;
434 	LIST_INIT(&node->mn_children);
435 	LIST_INIT(&node->mn_vnodes);
436 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
437 	mqnode_addref(parent);
438 	return (0);
439 }
440 
441 static struct mqfs_node *
mqfs_create_node(const char * name,int namelen,struct ucred * cred,int mode,int nodetype)442 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
443 	int nodetype)
444 {
445 	struct mqfs_node *node;
446 
447 	node = mqnode_alloc();
448 	strncpy(node->mn_name, name, namelen);
449 	node->mn_pr_root = cred->cr_prison->pr_root;
450 	node->mn_type = nodetype;
451 	node->mn_refcount = 1;
452 	vfs_timestamp(&node->mn_birth);
453 	node->mn_ctime = node->mn_atime = node->mn_mtime =
454 	    node->mn_birth;
455 	node->mn_uid = cred->cr_uid;
456 	node->mn_gid = cred->cr_gid;
457 	node->mn_mode = mode;
458 	return (node);
459 }
460 
461 /*
462  * Create a file
463  */
464 static struct mqfs_node *
mqfs_create_file(struct mqfs_node * parent,const char * name,int namelen,struct ucred * cred,int mode)465 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
466 	struct ucred *cred, int mode)
467 {
468 	struct mqfs_node *node;
469 
470 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
471 	if (mqfs_add_node(parent, node) != 0) {
472 		mqnode_free(node);
473 		return (NULL);
474 	}
475 	return (node);
476 }
477 
478 /*
479  * Add . and .. to a directory
480  */
481 static int
mqfs_fixup_dir(struct mqfs_node * parent)482 mqfs_fixup_dir(struct mqfs_node *parent)
483 {
484 	struct mqfs_node *dir;
485 
486 	dir = mqnode_alloc();
487 	dir->mn_name[0] = '.';
488 	dir->mn_type = mqfstype_this;
489 	dir->mn_refcount = 1;
490 	if (mqfs_add_node(parent, dir) != 0) {
491 		mqnode_free(dir);
492 		return (-1);
493 	}
494 
495 	dir = mqnode_alloc();
496 	dir->mn_name[0] = dir->mn_name[1] = '.';
497 	dir->mn_type = mqfstype_parent;
498 	dir->mn_refcount = 1;
499 
500 	if (mqfs_add_node(parent, dir) != 0) {
501 		mqnode_free(dir);
502 		return (-1);
503 	}
504 
505 	return (0);
506 }
507 
508 #ifdef notyet
509 
510 /*
511  * Create a directory
512  */
513 static struct mqfs_node *
mqfs_create_dir(struct mqfs_node * parent,const char * name,int namelen,struct ucred * cred,int mode)514 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
515 	struct ucred *cred, int mode)
516 {
517 	struct mqfs_node *node;
518 
519 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
520 	if (mqfs_add_node(parent, node) != 0) {
521 		mqnode_free(node);
522 		return (NULL);
523 	}
524 
525 	if (mqfs_fixup_dir(node) != 0) {
526 		mqfs_destroy(node);
527 		return (NULL);
528 	}
529 	return (node);
530 }
531 
532 /*
533  * Create a symlink
534  */
535 static struct mqfs_node *
mqfs_create_link(struct mqfs_node * parent,const char * name,int namelen,struct ucred * cred,int mode)536 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
537 	struct ucred *cred, int mode)
538 {
539 	struct mqfs_node *node;
540 
541 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
542 	if (mqfs_add_node(parent, node) != 0) {
543 		mqnode_free(node);
544 		return (NULL);
545 	}
546 	return (node);
547 }
548 
549 #endif
550 
551 /*
552  * Destroy a node or a tree of nodes
553  */
554 static int
mqfs_destroy(struct mqfs_node * node)555 mqfs_destroy(struct mqfs_node *node)
556 {
557 	struct mqfs_node *parent;
558 
559 	KASSERT(node != NULL,
560 	    ("%s(): node is NULL", __func__));
561 	KASSERT(node->mn_info != NULL,
562 	    ("%s(): node has no mn_info", __func__));
563 
564 	/* destroy children */
565 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
566 		while (! LIST_EMPTY(&node->mn_children))
567 			mqfs_destroy(LIST_FIRST(&node->mn_children));
568 
569 	/* unlink from parent */
570 	if ((parent = node->mn_parent) != NULL) {
571 		KASSERT(parent->mn_info == node->mn_info,
572 		    ("%s(): parent has different mn_info", __func__));
573 		LIST_REMOVE(node, mn_sibling);
574 	}
575 
576 	if (node->mn_fileno != 0)
577 		mqfs_fileno_free(node->mn_info, node);
578 	if (node->mn_data != NULL)
579 		mqueue_free(node->mn_data);
580 	mqnode_free(node);
581 	return (0);
582 }
583 
584 /*
585  * Mount a mqfs instance
586  */
587 static int
mqfs_mount(struct mount * mp)588 mqfs_mount(struct mount *mp)
589 {
590 	struct statfs *sbp;
591 
592 	if (mp->mnt_flag & MNT_UPDATE)
593 		return (EOPNOTSUPP);
594 
595 	mp->mnt_data = &mqfs_data;
596 	MNT_ILOCK(mp);
597 	mp->mnt_flag |= MNT_LOCAL;
598 	MNT_IUNLOCK(mp);
599 	vfs_getnewfsid(mp);
600 
601 	sbp = &mp->mnt_stat;
602 	vfs_mountedfrom(mp, "mqueue");
603 	sbp->f_bsize = PAGE_SIZE;
604 	sbp->f_iosize = PAGE_SIZE;
605 	sbp->f_blocks = 1;
606 	sbp->f_bfree = 0;
607 	sbp->f_bavail = 0;
608 	sbp->f_files = 1;
609 	sbp->f_ffree = 0;
610 	return (0);
611 }
612 
613 /*
614  * Unmount a mqfs instance
615  */
616 static int
mqfs_unmount(struct mount * mp,int mntflags)617 mqfs_unmount(struct mount *mp, int mntflags)
618 {
619 	int error;
620 
621 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0,
622 	    curthread);
623 	return (error);
624 }
625 
626 /*
627  * Return a root vnode
628  */
629 static int
mqfs_root(struct mount * mp,int flags,struct vnode ** vpp)630 mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
631 {
632 	struct mqfs_info *mqfs;
633 	int ret;
634 
635 	mqfs = VFSTOMQFS(mp);
636 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
637 	return (ret);
638 }
639 
640 /*
641  * Return filesystem stats
642  */
643 static int
mqfs_statfs(struct mount * mp,struct statfs * sbp)644 mqfs_statfs(struct mount *mp, struct statfs *sbp)
645 {
646 	/* XXX update statistics */
647 	return (0);
648 }
649 
650 /*
651  * Initialize a mqfs instance
652  */
653 static int
mqfs_init(struct vfsconf * vfc)654 mqfs_init(struct vfsconf *vfc)
655 {
656 	struct mqfs_node *root;
657 	struct mqfs_info *mi;
658 	osd_method_t methods[PR_MAXMETHOD] = {
659 	    [PR_METHOD_REMOVE] = mqfs_prison_remove,
660 	};
661 
662 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
663 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
664 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
665 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
666 	mvdata_zone = uma_zcreate("mvdata",
667 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
668 		NULL, UMA_ALIGN_PTR, 0);
669 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
670 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
671 	mi = &mqfs_data;
672 	sx_init(&mi->mi_lock, "mqfs lock");
673 	/* set up the root diretory */
674 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
675 		mqfstype_root);
676 	root->mn_info = mi;
677 	LIST_INIT(&root->mn_children);
678 	LIST_INIT(&root->mn_vnodes);
679 	mi->mi_root = root;
680 	mqfs_fileno_init(mi);
681 	mqfs_fileno_alloc(mi, root);
682 	mqfs_fixup_dir(root);
683 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
684 	    EVENTHANDLER_PRI_ANY);
685 	mq_fdclose = mqueue_fdclose;
686 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
687 	mqfs_osd_jail_slot = osd_jail_register(NULL, methods);
688 	return (0);
689 }
690 
691 /*
692  * Destroy a mqfs instance
693  */
694 static int
mqfs_uninit(struct vfsconf * vfc)695 mqfs_uninit(struct vfsconf *vfc)
696 {
697 	struct mqfs_info *mi;
698 
699 	if (!unloadable)
700 		return (EOPNOTSUPP);
701 	osd_jail_deregister(mqfs_osd_jail_slot);
702 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
703 	mi = &mqfs_data;
704 	mqfs_destroy(mi->mi_root);
705 	mi->mi_root = NULL;
706 	mqfs_fileno_uninit(mi);
707 	sx_destroy(&mi->mi_lock);
708 	uma_zdestroy(mqnode_zone);
709 	uma_zdestroy(mqueue_zone);
710 	uma_zdestroy(mvdata_zone);
711 	uma_zdestroy(mqnoti_zone);
712 	return (0);
713 }
714 
715 /*
716  * task routine
717  */
718 static void
do_recycle(void * context,int pending __unused)719 do_recycle(void *context, int pending __unused)
720 {
721 	struct vnode *vp = (struct vnode *)context;
722 
723 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
724 	vrecycle(vp);
725 	VOP_UNLOCK(vp);
726 	vdrop(vp);
727 }
728 
729 /*
730  * Allocate a vnode
731  */
732 static int
mqfs_allocv(struct mount * mp,struct vnode ** vpp,struct mqfs_node * pn)733 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
734 {
735 	struct mqfs_vdata *vd;
736 	struct mqfs_info  *mqfs;
737 	struct vnode *newvpp;
738 	int error;
739 
740 	mqfs = pn->mn_info;
741 	*vpp = NULL;
742 	sx_xlock(&mqfs->mi_lock);
743 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
744 		if (vd->mv_vnode->v_mount == mp) {
745 			vhold(vd->mv_vnode);
746 			break;
747 		}
748 	}
749 
750 	if (vd != NULL) {
751 found:
752 		*vpp = vd->mv_vnode;
753 		sx_xunlock(&mqfs->mi_lock);
754 		error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE);
755 		vdrop(*vpp);
756 		return (error);
757 	}
758 	sx_xunlock(&mqfs->mi_lock);
759 
760 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
761 	if (error)
762 		return (error);
763 	vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
764 	error = insmntque(newvpp, mp);
765 	if (error != 0)
766 		return (error);
767 
768 	sx_xlock(&mqfs->mi_lock);
769 	/*
770 	 * Check if it has already been allocated
771 	 * while we were blocked.
772 	 */
773 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
774 		if (vd->mv_vnode->v_mount == mp) {
775 			vhold(vd->mv_vnode);
776 			sx_xunlock(&mqfs->mi_lock);
777 
778 			vgone(newvpp);
779 			vput(newvpp);
780 			goto found;
781 		}
782 	}
783 
784 	*vpp = newvpp;
785 
786 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
787 	(*vpp)->v_data = vd;
788 	vd->mv_vnode = *vpp;
789 	vd->mv_node = pn;
790 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
791 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
792 	mqnode_addref(pn);
793 	switch (pn->mn_type) {
794 	case mqfstype_root:
795 		(*vpp)->v_vflag = VV_ROOT;
796 		/* fall through */
797 	case mqfstype_dir:
798 	case mqfstype_this:
799 	case mqfstype_parent:
800 		(*vpp)->v_type = VDIR;
801 		break;
802 	case mqfstype_file:
803 		(*vpp)->v_type = VREG;
804 		break;
805 	case mqfstype_symlink:
806 		(*vpp)->v_type = VLNK;
807 		break;
808 	case mqfstype_none:
809 		KASSERT(0, ("mqfs_allocf called for null node\n"));
810 	default:
811 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
812 	}
813 	sx_xunlock(&mqfs->mi_lock);
814 	return (0);
815 }
816 
817 /*
818  * Search a directory entry
819  */
820 static struct mqfs_node *
mqfs_search(struct mqfs_node * pd,const char * name,int len,struct ucred * cred)821 mqfs_search(struct mqfs_node *pd, const char *name, int len, struct ucred *cred)
822 {
823 	struct mqfs_node *pn;
824 	const void *pr_root;
825 
826 	sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
827 	pr_root = cred->cr_prison->pr_root;
828 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
829 		/* Only match names within the same prison root directory */
830 		if ((pn->mn_pr_root == NULL || pn->mn_pr_root == pr_root) &&
831 		    strncmp(pn->mn_name, name, len) == 0 &&
832 		    pn->mn_name[len] == '\0')
833 			return (pn);
834 	}
835 	return (NULL);
836 }
837 
838 /*
839  * Look up a file or directory.
840  */
841 static int
mqfs_lookupx(struct vop_cachedlookup_args * ap)842 mqfs_lookupx(struct vop_cachedlookup_args *ap)
843 {
844 	struct componentname *cnp;
845 	struct vnode *dvp, **vpp;
846 	struct mqfs_node *pd;
847 	struct mqfs_node *pn;
848 	struct mqfs_info *mqfs;
849 	int nameiop, flags, error, namelen;
850 	char *pname;
851 	struct thread *td;
852 
853 	cnp = ap->a_cnp;
854 	vpp = ap->a_vpp;
855 	dvp = ap->a_dvp;
856 	pname = cnp->cn_nameptr;
857 	namelen = cnp->cn_namelen;
858 	td = cnp->cn_thread;
859 	flags = cnp->cn_flags;
860 	nameiop = cnp->cn_nameiop;
861 	pd = VTON(dvp);
862 	pn = NULL;
863 	mqfs = pd->mn_info;
864 	*vpp = NULLVP;
865 
866 	if (dvp->v_type != VDIR)
867 		return (ENOTDIR);
868 
869 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
870 	if (error)
871 		return (error);
872 
873 	/* shortcut: check if the name is too long */
874 	if (cnp->cn_namelen >= MQFS_NAMELEN)
875 		return (ENOENT);
876 
877 	/* self */
878 	if (namelen == 1 && pname[0] == '.') {
879 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
880 			return (EINVAL);
881 		pn = pd;
882 		*vpp = dvp;
883 		VREF(dvp);
884 		return (0);
885 	}
886 
887 	/* parent */
888 	if (cnp->cn_flags & ISDOTDOT) {
889 		if (dvp->v_vflag & VV_ROOT)
890 			return (EIO);
891 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
892 			return (EINVAL);
893 		VOP_UNLOCK(dvp);
894 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
895 		pn = pd->mn_parent;
896 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
897 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
898 		return (error);
899 	}
900 
901 	/* named node */
902 	sx_xlock(&mqfs->mi_lock);
903 	pn = mqfs_search(pd, pname, namelen, cnp->cn_cred);
904 	if (pn != NULL)
905 		mqnode_addref(pn);
906 	sx_xunlock(&mqfs->mi_lock);
907 
908 	/* found */
909 	if (pn != NULL) {
910 		/* DELETE */
911 		if (nameiop == DELETE && (flags & ISLASTCN)) {
912 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
913 			if (error) {
914 				mqnode_release(pn);
915 				return (error);
916 			}
917 			if (*vpp == dvp) {
918 				VREF(dvp);
919 				*vpp = dvp;
920 				mqnode_release(pn);
921 				return (0);
922 			}
923 		}
924 
925 		/* allocate vnode */
926 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
927 		mqnode_release(pn);
928 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
929 			cache_enter(dvp, *vpp, cnp);
930 		return (error);
931 	}
932 
933 	/* not found */
934 
935 	/* will create a new entry in the directory ? */
936 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
937 	    && (flags & ISLASTCN)) {
938 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
939 		if (error)
940 			return (error);
941 		cnp->cn_flags |= SAVENAME;
942 		return (EJUSTRETURN);
943 	}
944 	return (ENOENT);
945 }
946 
947 #if 0
948 struct vop_lookup_args {
949 	struct vop_generic_args a_gen;
950 	struct vnode *a_dvp;
951 	struct vnode **a_vpp;
952 	struct componentname *a_cnp;
953 };
954 #endif
955 
956 /*
957  * vnode lookup operation
958  */
959 static int
mqfs_lookup(struct vop_cachedlookup_args * ap)960 mqfs_lookup(struct vop_cachedlookup_args *ap)
961 {
962 	int rc;
963 
964 	rc = mqfs_lookupx(ap);
965 	return (rc);
966 }
967 
968 #if 0
969 struct vop_create_args {
970 	struct vnode *a_dvp;
971 	struct vnode **a_vpp;
972 	struct componentname *a_cnp;
973 	struct vattr *a_vap;
974 };
975 #endif
976 
977 /*
978  * vnode creation operation
979  */
980 static int
mqfs_create(struct vop_create_args * ap)981 mqfs_create(struct vop_create_args *ap)
982 {
983 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
984 	struct componentname *cnp = ap->a_cnp;
985 	struct mqfs_node *pd;
986 	struct mqfs_node *pn;
987 	struct mqueue *mq;
988 	int error;
989 
990 	pd = VTON(ap->a_dvp);
991 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
992 		return (ENOTDIR);
993 	mq = mqueue_alloc(NULL);
994 	if (mq == NULL)
995 		return (EAGAIN);
996 	sx_xlock(&mqfs->mi_lock);
997 	if ((cnp->cn_flags & HASBUF) == 0)
998 		panic("%s: no name", __func__);
999 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
1000 		cnp->cn_cred, ap->a_vap->va_mode);
1001 	if (pn == NULL) {
1002 		sx_xunlock(&mqfs->mi_lock);
1003 		error = ENOSPC;
1004 	} else {
1005 		mqnode_addref(pn);
1006 		sx_xunlock(&mqfs->mi_lock);
1007 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1008 		mqnode_release(pn);
1009 		if (error)
1010 			mqfs_destroy(pn);
1011 		else
1012 			pn->mn_data = mq;
1013 	}
1014 	if (error)
1015 		mqueue_free(mq);
1016 	return (error);
1017 }
1018 
1019 /*
1020  * Remove an entry
1021  */
1022 static int
do_unlink(struct mqfs_node * pn,struct ucred * ucred)1023 do_unlink(struct mqfs_node *pn, struct ucred *ucred)
1024 {
1025 	struct mqfs_node *parent;
1026 	struct mqfs_vdata *vd;
1027 	int error = 0;
1028 
1029 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
1030 
1031 	if (ucred->cr_uid != pn->mn_uid &&
1032 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN)) != 0)
1033 		error = EACCES;
1034 	else if (!pn->mn_deleted) {
1035 		parent = pn->mn_parent;
1036 		pn->mn_parent = NULL;
1037 		pn->mn_deleted = 1;
1038 		LIST_REMOVE(pn, mn_sibling);
1039 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
1040 			cache_purge(vd->mv_vnode);
1041 			vhold(vd->mv_vnode);
1042 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
1043 		}
1044 		mqnode_release(pn);
1045 		mqnode_release(parent);
1046 	} else
1047 		error = ENOENT;
1048 	return (error);
1049 }
1050 
1051 #if 0
1052 struct vop_remove_args {
1053 	struct vnode *a_dvp;
1054 	struct vnode *a_vp;
1055 	struct componentname *a_cnp;
1056 };
1057 #endif
1058 
1059 /*
1060  * vnode removal operation
1061  */
1062 static int
mqfs_remove(struct vop_remove_args * ap)1063 mqfs_remove(struct vop_remove_args *ap)
1064 {
1065 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1066 	struct mqfs_node *pn;
1067 	int error;
1068 
1069 	if (ap->a_vp->v_type == VDIR)
1070                 return (EPERM);
1071 	pn = VTON(ap->a_vp);
1072 	sx_xlock(&mqfs->mi_lock);
1073 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1074 	sx_xunlock(&mqfs->mi_lock);
1075 	return (error);
1076 }
1077 
1078 #if 0
1079 struct vop_inactive_args {
1080 	struct vnode *a_vp;
1081 	struct thread *a_td;
1082 };
1083 #endif
1084 
1085 static int
mqfs_inactive(struct vop_inactive_args * ap)1086 mqfs_inactive(struct vop_inactive_args *ap)
1087 {
1088 	struct mqfs_node *pn = VTON(ap->a_vp);
1089 
1090 	if (pn->mn_deleted)
1091 		vrecycle(ap->a_vp);
1092 	return (0);
1093 }
1094 
1095 #if 0
1096 struct vop_reclaim_args {
1097 	struct vop_generic_args a_gen;
1098 	struct vnode *a_vp;
1099 };
1100 #endif
1101 
1102 static int
mqfs_reclaim(struct vop_reclaim_args * ap)1103 mqfs_reclaim(struct vop_reclaim_args *ap)
1104 {
1105 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1106 	struct vnode *vp = ap->a_vp;
1107 	struct mqfs_node *pn;
1108 	struct mqfs_vdata *vd;
1109 
1110 	vd = vp->v_data;
1111 	pn = vd->mv_node;
1112 	sx_xlock(&mqfs->mi_lock);
1113 	vp->v_data = NULL;
1114 	LIST_REMOVE(vd, mv_link);
1115 	mqnode_release(pn);
1116 	sx_xunlock(&mqfs->mi_lock);
1117 	uma_zfree(mvdata_zone, vd);
1118 	return (0);
1119 }
1120 
1121 #if 0
1122 struct vop_open_args {
1123 	struct vop_generic_args a_gen;
1124 	struct vnode *a_vp;
1125 	int a_mode;
1126 	struct ucred *a_cred;
1127 	struct thread *a_td;
1128 	struct file *a_fp;
1129 };
1130 #endif
1131 
1132 static int
mqfs_open(struct vop_open_args * ap)1133 mqfs_open(struct vop_open_args *ap)
1134 {
1135 	return (0);
1136 }
1137 
1138 #if 0
1139 struct vop_close_args {
1140 	struct vop_generic_args a_gen;
1141 	struct vnode *a_vp;
1142 	int a_fflag;
1143 	struct ucred *a_cred;
1144 	struct thread *a_td;
1145 };
1146 #endif
1147 
1148 static int
mqfs_close(struct vop_close_args * ap)1149 mqfs_close(struct vop_close_args *ap)
1150 {
1151 	return (0);
1152 }
1153 
1154 #if 0
1155 struct vop_access_args {
1156 	struct vop_generic_args a_gen;
1157 	struct vnode *a_vp;
1158 	accmode_t a_accmode;
1159 	struct ucred *a_cred;
1160 	struct thread *a_td;
1161 };
1162 #endif
1163 
1164 /*
1165  * Verify permissions
1166  */
1167 static int
mqfs_access(struct vop_access_args * ap)1168 mqfs_access(struct vop_access_args *ap)
1169 {
1170 	struct vnode *vp = ap->a_vp;
1171 	struct vattr vattr;
1172 	int error;
1173 
1174 	error = VOP_GETATTR(vp, &vattr, ap->a_cred);
1175 	if (error)
1176 		return (error);
1177 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid, vattr.va_gid,
1178 	    ap->a_accmode, ap->a_cred);
1179 	return (error);
1180 }
1181 
1182 #if 0
1183 struct vop_getattr_args {
1184 	struct vop_generic_args a_gen;
1185 	struct vnode *a_vp;
1186 	struct vattr *a_vap;
1187 	struct ucred *a_cred;
1188 };
1189 #endif
1190 
1191 /*
1192  * Get file attributes
1193  */
1194 static int
mqfs_getattr(struct vop_getattr_args * ap)1195 mqfs_getattr(struct vop_getattr_args *ap)
1196 {
1197 	struct vnode *vp = ap->a_vp;
1198 	struct mqfs_node *pn = VTON(vp);
1199 	struct vattr *vap = ap->a_vap;
1200 	int error = 0;
1201 
1202 	vap->va_type = vp->v_type;
1203 	vap->va_mode = pn->mn_mode;
1204 	vap->va_nlink = 1;
1205 	vap->va_uid = pn->mn_uid;
1206 	vap->va_gid = pn->mn_gid;
1207 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1208 	vap->va_fileid = pn->mn_fileno;
1209 	vap->va_size = 0;
1210 	vap->va_blocksize = PAGE_SIZE;
1211 	vap->va_bytes = vap->va_size = 0;
1212 	vap->va_atime = pn->mn_atime;
1213 	vap->va_mtime = pn->mn_mtime;
1214 	vap->va_ctime = pn->mn_ctime;
1215 	vap->va_birthtime = pn->mn_birth;
1216 	vap->va_gen = 0;
1217 	vap->va_flags = 0;
1218 	vap->va_rdev = NODEV;
1219 	vap->va_bytes = 0;
1220 	vap->va_filerev = 0;
1221 	return (error);
1222 }
1223 
1224 #if 0
1225 struct vop_setattr_args {
1226 	struct vop_generic_args a_gen;
1227 	struct vnode *a_vp;
1228 	struct vattr *a_vap;
1229 	struct ucred *a_cred;
1230 };
1231 #endif
1232 /*
1233  * Set attributes
1234  */
1235 static int
mqfs_setattr(struct vop_setattr_args * ap)1236 mqfs_setattr(struct vop_setattr_args *ap)
1237 {
1238 	struct mqfs_node *pn;
1239 	struct vattr *vap;
1240 	struct vnode *vp;
1241 	struct thread *td;
1242 	int c, error;
1243 	uid_t uid;
1244 	gid_t gid;
1245 
1246 	td = curthread;
1247 	vap = ap->a_vap;
1248 	vp = ap->a_vp;
1249 	if (vap->va_type != VNON ||
1250 	    vap->va_nlink != VNOVAL ||
1251 	    vap->va_fsid != VNOVAL ||
1252 	    vap->va_fileid != VNOVAL ||
1253 	    vap->va_blocksize != VNOVAL ||
1254 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1255 	    vap->va_rdev != VNOVAL ||
1256 	    (int)vap->va_bytes != VNOVAL ||
1257 	    vap->va_gen != VNOVAL) {
1258 		return (EINVAL);
1259 	}
1260 
1261 	pn = VTON(vp);
1262 
1263 	error = c = 0;
1264 	if (vap->va_uid == (uid_t)VNOVAL)
1265 		uid = pn->mn_uid;
1266 	else
1267 		uid = vap->va_uid;
1268 	if (vap->va_gid == (gid_t)VNOVAL)
1269 		gid = pn->mn_gid;
1270 	else
1271 		gid = vap->va_gid;
1272 
1273 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1274 		/*
1275 		 * To modify the ownership of a file, must possess VADMIN
1276 		 * for that file.
1277 		 */
1278 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
1279 			return (error);
1280 
1281 		/*
1282 		 * XXXRW: Why is there a privilege check here: shouldn't the
1283 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1284 		 * below definitely right?
1285 		 */
1286 		if ((ap->a_cred->cr_uid != pn->mn_uid || uid != pn->mn_uid ||
1287 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1288 		    (error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
1289 			return (error);
1290 		pn->mn_uid = uid;
1291 		pn->mn_gid = gid;
1292 		c = 1;
1293 	}
1294 
1295 	if (vap->va_mode != (mode_t)VNOVAL) {
1296 		if (ap->a_cred->cr_uid != pn->mn_uid &&
1297 		    (error = priv_check(td, PRIV_MQ_ADMIN)))
1298 			return (error);
1299 		pn->mn_mode = vap->va_mode;
1300 		c = 1;
1301 	}
1302 
1303 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1304 		/* See the comment in ufs_vnops::ufs_setattr(). */
1305 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1306 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1307 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1308 			return (error);
1309 		if (vap->va_atime.tv_sec != VNOVAL) {
1310 			pn->mn_atime = vap->va_atime;
1311 		}
1312 		if (vap->va_mtime.tv_sec != VNOVAL) {
1313 			pn->mn_mtime = vap->va_mtime;
1314 		}
1315 		c = 1;
1316 	}
1317 	if (c) {
1318 		vfs_timestamp(&pn->mn_ctime);
1319 	}
1320 	return (0);
1321 }
1322 
1323 #if 0
1324 struct vop_read_args {
1325 	struct vop_generic_args a_gen;
1326 	struct vnode *a_vp;
1327 	struct uio *a_uio;
1328 	int a_ioflag;
1329 	struct ucred *a_cred;
1330 };
1331 #endif
1332 
1333 /*
1334  * Read from a file
1335  */
1336 static int
mqfs_read(struct vop_read_args * ap)1337 mqfs_read(struct vop_read_args *ap)
1338 {
1339 	char buf[80];
1340 	struct vnode *vp = ap->a_vp;
1341 	struct uio *uio = ap->a_uio;
1342 	struct mqueue *mq;
1343 	int len, error;
1344 
1345 	if (vp->v_type != VREG)
1346 		return (EINVAL);
1347 
1348 	mq = VTOMQ(vp);
1349 	snprintf(buf, sizeof(buf),
1350 	    "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1351 	    mq->mq_totalbytes,
1352 	    mq->mq_maxmsg,
1353 	    mq->mq_curmsgs,
1354 	    mq->mq_msgsize);
1355 	buf[sizeof(buf)-1] = '\0';
1356 	len = strlen(buf);
1357 	error = uiomove_frombuf(buf, len, uio);
1358 	return (error);
1359 }
1360 
1361 #if 0
1362 struct vop_readdir_args {
1363 	struct vop_generic_args a_gen;
1364 	struct vnode *a_vp;
1365 	struct uio *a_uio;
1366 	struct ucred *a_cred;
1367 	int *a_eofflag;
1368 	int *a_ncookies;
1369 	u_long **a_cookies;
1370 };
1371 #endif
1372 
1373 /*
1374  * Return directory entries.
1375  */
1376 static int
mqfs_readdir(struct vop_readdir_args * ap)1377 mqfs_readdir(struct vop_readdir_args *ap)
1378 {
1379 	struct vnode *vp;
1380 	struct mqfs_info *mi;
1381 	struct mqfs_node *pd;
1382 	struct mqfs_node *pn;
1383 	struct dirent entry;
1384 	struct uio *uio;
1385 	const void *pr_root;
1386 	int *tmp_ncookies = NULL;
1387 	off_t offset;
1388 	int error, i;
1389 
1390 	vp = ap->a_vp;
1391 	mi = VFSTOMQFS(vp->v_mount);
1392 	pd = VTON(vp);
1393 	uio = ap->a_uio;
1394 
1395 	if (vp->v_type != VDIR)
1396 		return (ENOTDIR);
1397 
1398 	if (uio->uio_offset < 0)
1399 		return (EINVAL);
1400 
1401 	if (ap->a_ncookies != NULL) {
1402 		tmp_ncookies = ap->a_ncookies;
1403 		*ap->a_ncookies = 0;
1404 		ap->a_ncookies = NULL;
1405         }
1406 
1407 	error = 0;
1408 	offset = 0;
1409 
1410 	pr_root = ap->a_cred->cr_prison->pr_root;
1411 	sx_xlock(&mi->mi_lock);
1412 
1413 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1414 		entry.d_reclen = sizeof(entry);
1415 
1416 		/*
1417 		 * Only show names within the same prison root directory
1418 		 * (or not associated with a prison, e.g. "." and "..").
1419 		 */
1420 		if (pn->mn_pr_root != NULL && pn->mn_pr_root != pr_root)
1421 			continue;
1422 		if (!pn->mn_fileno)
1423 			mqfs_fileno_alloc(mi, pn);
1424 		entry.d_fileno = pn->mn_fileno;
1425 		entry.d_off = offset + entry.d_reclen;
1426 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1427 			entry.d_name[i] = pn->mn_name[i];
1428 		entry.d_namlen = i;
1429 		switch (pn->mn_type) {
1430 		case mqfstype_root:
1431 		case mqfstype_dir:
1432 		case mqfstype_this:
1433 		case mqfstype_parent:
1434 			entry.d_type = DT_DIR;
1435 			break;
1436 		case mqfstype_file:
1437 			entry.d_type = DT_REG;
1438 			break;
1439 		case mqfstype_symlink:
1440 			entry.d_type = DT_LNK;
1441 			break;
1442 		default:
1443 			panic("%s has unexpected node type: %d", pn->mn_name,
1444 				pn->mn_type);
1445 		}
1446 		dirent_terminate(&entry);
1447 		if (entry.d_reclen > uio->uio_resid)
1448                         break;
1449 		if (offset >= uio->uio_offset) {
1450 			error = vfs_read_dirent(ap, &entry, offset);
1451                         if (error)
1452                                 break;
1453                 }
1454                 offset += entry.d_reclen;
1455 	}
1456 	sx_xunlock(&mi->mi_lock);
1457 
1458 	uio->uio_offset = offset;
1459 
1460 	if (tmp_ncookies != NULL)
1461 		ap->a_ncookies = tmp_ncookies;
1462 
1463 	return (error);
1464 }
1465 
1466 #ifdef notyet
1467 
1468 #if 0
1469 struct vop_mkdir_args {
1470 	struct vnode *a_dvp;
1471 	struvt vnode **a_vpp;
1472 	struvt componentname *a_cnp;
1473 	struct vattr *a_vap;
1474 };
1475 #endif
1476 
1477 /*
1478  * Create a directory.
1479  */
1480 static int
mqfs_mkdir(struct vop_mkdir_args * ap)1481 mqfs_mkdir(struct vop_mkdir_args *ap)
1482 {
1483 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1484 	struct componentname *cnp = ap->a_cnp;
1485 	struct mqfs_node *pd = VTON(ap->a_dvp);
1486 	struct mqfs_node *pn;
1487 	int error;
1488 
1489 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1490 		return (ENOTDIR);
1491 	sx_xlock(&mqfs->mi_lock);
1492 	if ((cnp->cn_flags & HASBUF) == 0)
1493 		panic("%s: no name", __func__);
1494 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1495 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1496 	if (pn != NULL)
1497 		mqnode_addref(pn);
1498 	sx_xunlock(&mqfs->mi_lock);
1499 	if (pn == NULL) {
1500 		error = ENOSPC;
1501 	} else {
1502 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1503 		mqnode_release(pn);
1504 	}
1505 	return (error);
1506 }
1507 
1508 #if 0
1509 struct vop_rmdir_args {
1510 	struct vnode *a_dvp;
1511 	struct vnode *a_vp;
1512 	struct componentname *a_cnp;
1513 };
1514 #endif
1515 
1516 /*
1517  * Remove a directory.
1518  */
1519 static int
mqfs_rmdir(struct vop_rmdir_args * ap)1520 mqfs_rmdir(struct vop_rmdir_args *ap)
1521 {
1522 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1523 	struct mqfs_node *pn = VTON(ap->a_vp);
1524 	struct mqfs_node *pt;
1525 
1526 	if (pn->mn_type != mqfstype_dir)
1527 		return (ENOTDIR);
1528 
1529 	sx_xlock(&mqfs->mi_lock);
1530 	if (pn->mn_deleted) {
1531 		sx_xunlock(&mqfs->mi_lock);
1532 		return (ENOENT);
1533 	}
1534 
1535 	pt = LIST_FIRST(&pn->mn_children);
1536 	pt = LIST_NEXT(pt, mn_sibling);
1537 	pt = LIST_NEXT(pt, mn_sibling);
1538 	if (pt != NULL) {
1539 		sx_xunlock(&mqfs->mi_lock);
1540 		return (ENOTEMPTY);
1541 	}
1542 	pt = pn->mn_parent;
1543 	pn->mn_parent = NULL;
1544 	pn->mn_deleted = 1;
1545 	LIST_REMOVE(pn, mn_sibling);
1546 	mqnode_release(pn);
1547 	mqnode_release(pt);
1548 	sx_xunlock(&mqfs->mi_lock);
1549 	cache_purge(ap->a_vp);
1550 	return (0);
1551 }
1552 
1553 #endif /* notyet */
1554 
1555 /*
1556  * See if this prison root is obsolete, and clean up associated queues if it is.
1557  */
1558 static int
mqfs_prison_remove(void * obj,void * data __unused)1559 mqfs_prison_remove(void *obj, void *data __unused)
1560 {
1561 	const struct prison *pr = obj;
1562 	struct prison *tpr;
1563 	struct mqfs_node *pn, *tpn;
1564 	struct vnode *pr_root;
1565 
1566 	pr_root = pr->pr_root;
1567 	if (pr->pr_parent->pr_root == pr_root)
1568 		return (0);
1569 	TAILQ_FOREACH(tpr, &allprison, pr_list) {
1570 		if (tpr != pr && tpr->pr_root == pr_root)
1571 			return (0);
1572 	}
1573 	/*
1574 	 * No jails are rooted in this directory anymore,
1575 	 * so no queues should be either.
1576 	 */
1577 	sx_xlock(&mqfs_data.mi_lock);
1578 	LIST_FOREACH_SAFE(pn, &mqfs_data.mi_root->mn_children,
1579 	    mn_sibling, tpn) {
1580 		if (pn->mn_pr_root == pr_root)
1581 			(void)do_unlink(pn, curthread->td_ucred);
1582 	}
1583 	sx_xunlock(&mqfs_data.mi_lock);
1584 	return (0);
1585 }
1586 
1587 /*
1588  * Allocate a message queue
1589  */
1590 static struct mqueue *
mqueue_alloc(const struct mq_attr * attr)1591 mqueue_alloc(const struct mq_attr *attr)
1592 {
1593 	struct mqueue *mq;
1594 
1595 	if (curmq >= maxmq)
1596 		return (NULL);
1597 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1598 	TAILQ_INIT(&mq->mq_msgq);
1599 	if (attr != NULL) {
1600 		mq->mq_maxmsg = attr->mq_maxmsg;
1601 		mq->mq_msgsize = attr->mq_msgsize;
1602 	} else {
1603 		mq->mq_maxmsg = default_maxmsg;
1604 		mq->mq_msgsize = default_msgsize;
1605 	}
1606 	mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
1607 	knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
1608 	knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
1609 	atomic_add_int(&curmq, 1);
1610 	return (mq);
1611 }
1612 
1613 /*
1614  * Destroy a message queue
1615  */
1616 static void
mqueue_free(struct mqueue * mq)1617 mqueue_free(struct mqueue *mq)
1618 {
1619 	struct mqueue_msg *msg;
1620 
1621 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1622 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1623 		free(msg, M_MQUEUEDATA);
1624 	}
1625 
1626 	mtx_destroy(&mq->mq_mutex);
1627 	seldrain(&mq->mq_rsel);
1628 	seldrain(&mq->mq_wsel);
1629 	knlist_destroy(&mq->mq_rsel.si_note);
1630 	knlist_destroy(&mq->mq_wsel.si_note);
1631 	uma_zfree(mqueue_zone, mq);
1632 	atomic_add_int(&curmq, -1);
1633 }
1634 
1635 /*
1636  * Load a message from user space
1637  */
1638 static struct mqueue_msg *
mqueue_loadmsg(const char * msg_ptr,size_t msg_size,int msg_prio)1639 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1640 {
1641 	struct mqueue_msg *msg;
1642 	size_t len;
1643 	int error;
1644 
1645 	len = sizeof(struct mqueue_msg) + msg_size;
1646 	msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1647 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1648 	    msg_size);
1649 	if (error) {
1650 		free(msg, M_MQUEUEDATA);
1651 		msg = NULL;
1652 	} else {
1653 		msg->msg_size = msg_size;
1654 		msg->msg_prio = msg_prio;
1655 	}
1656 	return (msg);
1657 }
1658 
1659 /*
1660  * Save a message to user space
1661  */
1662 static int
mqueue_savemsg(struct mqueue_msg * msg,char * msg_ptr,int * msg_prio)1663 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1664 {
1665 	int error;
1666 
1667 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1668 		msg->msg_size);
1669 	if (error == 0 && msg_prio != NULL)
1670 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1671 	return (error);
1672 }
1673 
1674 /*
1675  * Free a message's memory
1676  */
1677 static __inline void
mqueue_freemsg(struct mqueue_msg * msg)1678 mqueue_freemsg(struct mqueue_msg *msg)
1679 {
1680 	free(msg, M_MQUEUEDATA);
1681 }
1682 
1683 /*
1684  * Send a message. if waitok is false, thread will not be
1685  * blocked if there is no data in queue, otherwise, absolute
1686  * time will be checked.
1687  */
1688 int
mqueue_send(struct mqueue * mq,const char * msg_ptr,size_t msg_len,unsigned msg_prio,int waitok,const struct timespec * abs_timeout)1689 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1690 	size_t msg_len, unsigned msg_prio, int waitok,
1691 	const struct timespec *abs_timeout)
1692 {
1693 	struct mqueue_msg *msg;
1694 	struct timespec ts, ts2;
1695 	struct timeval tv;
1696 	int error;
1697 
1698 	if (msg_prio >= MQ_PRIO_MAX)
1699 		return (EINVAL);
1700 	if (msg_len > mq->mq_msgsize)
1701 		return (EMSGSIZE);
1702 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1703 	if (msg == NULL)
1704 		return (EFAULT);
1705 
1706 	/* O_NONBLOCK case */
1707 	if (!waitok) {
1708 		error = _mqueue_send(mq, msg, -1);
1709 		if (error)
1710 			goto bad;
1711 		return (0);
1712 	}
1713 
1714 	/* we allow a null timeout (wait forever) */
1715 	if (abs_timeout == NULL) {
1716 		error = _mqueue_send(mq, msg, 0);
1717 		if (error)
1718 			goto bad;
1719 		return (0);
1720 	}
1721 
1722 	/* send it before checking time */
1723 	error = _mqueue_send(mq, msg, -1);
1724 	if (error == 0)
1725 		return (0);
1726 
1727 	if (error != EAGAIN)
1728 		goto bad;
1729 
1730 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1731 		error = EINVAL;
1732 		goto bad;
1733 	}
1734 	for (;;) {
1735 		getnanotime(&ts);
1736 		timespecsub(abs_timeout, &ts, &ts2);
1737 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1738 			error = ETIMEDOUT;
1739 			break;
1740 		}
1741 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1742 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1743 		if (error != ETIMEDOUT)
1744 			break;
1745 	}
1746 	if (error == 0)
1747 		return (0);
1748 bad:
1749 	mqueue_freemsg(msg);
1750 	return (error);
1751 }
1752 
1753 /*
1754  * Common routine to send a message
1755  */
1756 static int
_mqueue_send(struct mqueue * mq,struct mqueue_msg * msg,int timo)1757 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1758 {
1759 	struct mqueue_msg *msg2;
1760 	int error = 0;
1761 
1762 	mtx_lock(&mq->mq_mutex);
1763 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1764 		if (timo < 0) {
1765 			mtx_unlock(&mq->mq_mutex);
1766 			return (EAGAIN);
1767 		}
1768 		mq->mq_senders++;
1769 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1770 			    PCATCH, "mqsend", timo);
1771 		mq->mq_senders--;
1772 		if (error == EAGAIN)
1773 			error = ETIMEDOUT;
1774 	}
1775 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1776 		mtx_unlock(&mq->mq_mutex);
1777 		return (error);
1778 	}
1779 	error = 0;
1780 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1781 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1782 	} else {
1783 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1784 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1785 		} else {
1786 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1787 				if (msg2->msg_prio < msg->msg_prio)
1788 					break;
1789 			}
1790 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1791 		}
1792 	}
1793 	mq->mq_curmsgs++;
1794 	mq->mq_totalbytes += msg->msg_size;
1795 	if (mq->mq_receivers)
1796 		wakeup_one(&mq->mq_receivers);
1797 	else if (mq->mq_notifier != NULL)
1798 		mqueue_send_notification(mq);
1799 	if (mq->mq_flags & MQ_RSEL) {
1800 		mq->mq_flags &= ~MQ_RSEL;
1801 		selwakeup(&mq->mq_rsel);
1802 	}
1803 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1804 	mtx_unlock(&mq->mq_mutex);
1805 	return (0);
1806 }
1807 
1808 /*
1809  * Send realtime a signal to process which registered itself
1810  * successfully by mq_notify.
1811  */
1812 static void
mqueue_send_notification(struct mqueue * mq)1813 mqueue_send_notification(struct mqueue *mq)
1814 {
1815 	struct mqueue_notifier *nt;
1816 	struct thread *td;
1817 	struct proc *p;
1818 	int error;
1819 
1820 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1821 	nt = mq->mq_notifier;
1822 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1823 		p = nt->nt_proc;
1824 		error = sigev_findtd(p, &nt->nt_sigev, &td);
1825 		if (error) {
1826 			mq->mq_notifier = NULL;
1827 			return;
1828 		}
1829 		if (!KSI_ONQ(&nt->nt_ksi)) {
1830 			ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
1831 			tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
1832 		}
1833 		PROC_UNLOCK(p);
1834 	}
1835 	mq->mq_notifier = NULL;
1836 }
1837 
1838 /*
1839  * Get a message. if waitok is false, thread will not be
1840  * blocked if there is no data in queue, otherwise, absolute
1841  * time will be checked.
1842  */
1843 int
mqueue_receive(struct mqueue * mq,char * msg_ptr,size_t msg_len,unsigned * msg_prio,int waitok,const struct timespec * abs_timeout)1844 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1845 	size_t msg_len, unsigned *msg_prio, int waitok,
1846 	const struct timespec *abs_timeout)
1847 {
1848 	struct mqueue_msg *msg;
1849 	struct timespec ts, ts2;
1850 	struct timeval tv;
1851 	int error;
1852 
1853 	if (msg_len < mq->mq_msgsize)
1854 		return (EMSGSIZE);
1855 
1856 	/* O_NONBLOCK case */
1857 	if (!waitok) {
1858 		error = _mqueue_recv(mq, &msg, -1);
1859 		if (error)
1860 			return (error);
1861 		goto received;
1862 	}
1863 
1864 	/* we allow a null timeout (wait forever). */
1865 	if (abs_timeout == NULL) {
1866 		error = _mqueue_recv(mq, &msg, 0);
1867 		if (error)
1868 			return (error);
1869 		goto received;
1870 	}
1871 
1872 	/* try to get a message before checking time */
1873 	error = _mqueue_recv(mq, &msg, -1);
1874 	if (error == 0)
1875 		goto received;
1876 
1877 	if (error != EAGAIN)
1878 		return (error);
1879 
1880 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1881 		error = EINVAL;
1882 		return (error);
1883 	}
1884 
1885 	for (;;) {
1886 		getnanotime(&ts);
1887 		timespecsub(abs_timeout, &ts, &ts2);
1888 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1889 			error = ETIMEDOUT;
1890 			return (error);
1891 		}
1892 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1893 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1894 		if (error == 0)
1895 			break;
1896 		if (error != ETIMEDOUT)
1897 			return (error);
1898 	}
1899 
1900 received:
1901 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1902 	if (error == 0) {
1903 		curthread->td_retval[0] = msg->msg_size;
1904 		curthread->td_retval[1] = 0;
1905 	}
1906 	mqueue_freemsg(msg);
1907 	return (error);
1908 }
1909 
1910 /*
1911  * Common routine to receive a message
1912  */
1913 static int
_mqueue_recv(struct mqueue * mq,struct mqueue_msg ** msg,int timo)1914 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1915 {
1916 	int error = 0;
1917 
1918 	mtx_lock(&mq->mq_mutex);
1919 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1920 		if (timo < 0) {
1921 			mtx_unlock(&mq->mq_mutex);
1922 			return (EAGAIN);
1923 		}
1924 		mq->mq_receivers++;
1925 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1926 			    PCATCH, "mqrecv", timo);
1927 		mq->mq_receivers--;
1928 		if (error == EAGAIN)
1929 			error = ETIMEDOUT;
1930 	}
1931 	if (*msg != NULL) {
1932 		error = 0;
1933 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1934 		mq->mq_curmsgs--;
1935 		mq->mq_totalbytes -= (*msg)->msg_size;
1936 		if (mq->mq_senders)
1937 			wakeup_one(&mq->mq_senders);
1938 		if (mq->mq_flags & MQ_WSEL) {
1939 			mq->mq_flags &= ~MQ_WSEL;
1940 			selwakeup(&mq->mq_wsel);
1941 		}
1942 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1943 	}
1944 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1945 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1946 		mqueue_send_notification(mq);
1947 	}
1948 	mtx_unlock(&mq->mq_mutex);
1949 	return (error);
1950 }
1951 
1952 static __inline struct mqueue_notifier *
notifier_alloc(void)1953 notifier_alloc(void)
1954 {
1955 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1956 }
1957 
1958 static __inline void
notifier_free(struct mqueue_notifier * p)1959 notifier_free(struct mqueue_notifier *p)
1960 {
1961 	uma_zfree(mqnoti_zone, p);
1962 }
1963 
1964 static struct mqueue_notifier *
notifier_search(struct proc * p,int fd)1965 notifier_search(struct proc *p, int fd)
1966 {
1967 	struct mqueue_notifier *nt;
1968 
1969 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1970 		if (nt->nt_ksi.ksi_mqd == fd)
1971 			break;
1972 	}
1973 	return (nt);
1974 }
1975 
1976 static __inline void
notifier_insert(struct proc * p,struct mqueue_notifier * nt)1977 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1978 {
1979 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1980 }
1981 
1982 static __inline void
notifier_delete(struct proc * p,struct mqueue_notifier * nt)1983 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1984 {
1985 	LIST_REMOVE(nt, nt_link);
1986 	notifier_free(nt);
1987 }
1988 
1989 static void
notifier_remove(struct proc * p,struct mqueue * mq,int fd)1990 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1991 {
1992 	struct mqueue_notifier *nt;
1993 
1994 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1995 	PROC_LOCK(p);
1996 	nt = notifier_search(p, fd);
1997 	if (nt != NULL) {
1998 		if (mq->mq_notifier == nt)
1999 			mq->mq_notifier = NULL;
2000 		sigqueue_take(&nt->nt_ksi);
2001 		notifier_delete(p, nt);
2002 	}
2003 	PROC_UNLOCK(p);
2004 }
2005 
2006 static int
kern_kmq_open(struct thread * td,const char * upath,int flags,mode_t mode,const struct mq_attr * attr)2007 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
2008     const struct mq_attr *attr)
2009 {
2010 	char path[MQFS_NAMELEN + 1];
2011 	struct mqfs_node *pn;
2012 	struct pwddesc *pdp;
2013 	struct file *fp;
2014 	struct mqueue *mq;
2015 	int fd, error, len, cmode;
2016 
2017 	AUDIT_ARG_FFLAGS(flags);
2018 	AUDIT_ARG_MODE(mode);
2019 
2020 	pdp = td->td_proc->p_pd;
2021 	cmode = ((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT;
2022 	mq = NULL;
2023 	if ((flags & O_CREAT) != 0 && attr != NULL) {
2024 		if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
2025 			return (EINVAL);
2026 		if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
2027 			return (EINVAL);
2028 	}
2029 
2030 	error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
2031         if (error)
2032 		return (error);
2033 
2034 	/*
2035 	 * The first character of name must be a slash  (/) character
2036 	 * and the remaining characters of name cannot include any slash
2037 	 * characters.
2038 	 */
2039 	len = strlen(path);
2040 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
2041 		return (EINVAL);
2042 	/*
2043 	 * "." and ".." are magic directories, populated on the fly, and cannot
2044 	 * be opened as queues.
2045 	 */
2046 	if (strcmp(path, "/.") == 0 || strcmp(path, "/..") == 0)
2047 		return (EINVAL);
2048 	AUDIT_ARG_UPATH1_CANON(path);
2049 
2050 	error = falloc(td, &fp, &fd, O_CLOEXEC);
2051 	if (error)
2052 		return (error);
2053 
2054 	sx_xlock(&mqfs_data.mi_lock);
2055 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred);
2056 	if (pn == NULL) {
2057 		if (!(flags & O_CREAT)) {
2058 			error = ENOENT;
2059 		} else {
2060 			mq = mqueue_alloc(attr);
2061 			if (mq == NULL) {
2062 				error = ENFILE;
2063 			} else {
2064 				pn = mqfs_create_file(mqfs_data.mi_root,
2065 				         path + 1, len - 1, td->td_ucred,
2066 					 cmode);
2067 				if (pn == NULL) {
2068 					error = ENOSPC;
2069 					mqueue_free(mq);
2070 				}
2071 			}
2072 		}
2073 
2074 		if (error == 0) {
2075 			pn->mn_data = mq;
2076 		}
2077 	} else {
2078 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
2079 			error = EEXIST;
2080 		} else {
2081 			accmode_t accmode = 0;
2082 
2083 			if (flags & FREAD)
2084 				accmode |= VREAD;
2085 			if (flags & FWRITE)
2086 				accmode |= VWRITE;
2087 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
2088 			    pn->mn_gid, accmode, td->td_ucred);
2089 		}
2090 	}
2091 
2092 	if (error) {
2093 		sx_xunlock(&mqfs_data.mi_lock);
2094 		fdclose(td, fp, fd);
2095 		fdrop(fp, td);
2096 		return (error);
2097 	}
2098 
2099 	mqnode_addref(pn);
2100 	sx_xunlock(&mqfs_data.mi_lock);
2101 
2102 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2103 	    &mqueueops);
2104 
2105 	td->td_retval[0] = fd;
2106 	fdrop(fp, td);
2107 	return (0);
2108 }
2109 
2110 /*
2111  * Syscall to open a message queue.
2112  */
2113 int
sys_kmq_open(struct thread * td,struct kmq_open_args * uap)2114 sys_kmq_open(struct thread *td, struct kmq_open_args *uap)
2115 {
2116 	struct mq_attr attr;
2117 	int flags, error;
2118 
2119 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2120 		return (EINVAL);
2121 	flags = FFLAGS(uap->flags);
2122 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2123 		error = copyin(uap->attr, &attr, sizeof(attr));
2124 		if (error)
2125 			return (error);
2126 	}
2127 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2128 	    uap->attr != NULL ? &attr : NULL));
2129 }
2130 
2131 /*
2132  * Syscall to unlink a message queue.
2133  */
2134 int
sys_kmq_unlink(struct thread * td,struct kmq_unlink_args * uap)2135 sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2136 {
2137 	char path[MQFS_NAMELEN+1];
2138 	struct mqfs_node *pn;
2139 	int error, len;
2140 
2141 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2142         if (error)
2143 		return (error);
2144 
2145 	len = strlen(path);
2146 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
2147 		return (EINVAL);
2148 	if (strcmp(path, "/.") == 0 || strcmp(path, "/..") == 0)
2149 		return (EINVAL);
2150 	AUDIT_ARG_UPATH1_CANON(path);
2151 
2152 	sx_xlock(&mqfs_data.mi_lock);
2153 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred);
2154 	if (pn != NULL)
2155 		error = do_unlink(pn, td->td_ucred);
2156 	else
2157 		error = ENOENT;
2158 	sx_xunlock(&mqfs_data.mi_lock);
2159 	return (error);
2160 }
2161 
2162 typedef int (*_fgetf)(struct thread *, int, cap_rights_t *, struct file **);
2163 
2164 /*
2165  * Get message queue by giving file slot
2166  */
2167 static int
_getmq(struct thread * td,int fd,cap_rights_t * rightsp,_fgetf func,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2168 _getmq(struct thread *td, int fd, cap_rights_t *rightsp, _fgetf func,
2169        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2170 {
2171 	struct mqfs_node *pn;
2172 	int error;
2173 
2174 	error = func(td, fd, rightsp, fpp);
2175 	if (error)
2176 		return (error);
2177 	if (&mqueueops != (*fpp)->f_ops) {
2178 		fdrop(*fpp, td);
2179 		return (EBADF);
2180 	}
2181 	pn = (*fpp)->f_data;
2182 	if (ppn)
2183 		*ppn = pn;
2184 	if (pmq)
2185 		*pmq = pn->mn_data;
2186 	return (0);
2187 }
2188 
2189 static __inline int
getmq(struct thread * td,int fd,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2190 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2191 	struct mqueue **pmq)
2192 {
2193 
2194 	return _getmq(td, fd, &cap_event_rights, fget,
2195 	    fpp, ppn, pmq);
2196 }
2197 
2198 static __inline int
getmq_read(struct thread * td,int fd,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2199 getmq_read(struct thread *td, int fd, struct file **fpp,
2200 	 struct mqfs_node **ppn, struct mqueue **pmq)
2201 {
2202 
2203 	return _getmq(td, fd, &cap_read_rights, fget_read,
2204 	    fpp, ppn, pmq);
2205 }
2206 
2207 static __inline int
getmq_write(struct thread * td,int fd,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2208 getmq_write(struct thread *td, int fd, struct file **fpp,
2209 	struct mqfs_node **ppn, struct mqueue **pmq)
2210 {
2211 
2212 	return _getmq(td, fd, &cap_write_rights, fget_write,
2213 	    fpp, ppn, pmq);
2214 }
2215 
2216 static int
kern_kmq_setattr(struct thread * td,int mqd,const struct mq_attr * attr,struct mq_attr * oattr)2217 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2218     struct mq_attr *oattr)
2219 {
2220 	struct mqueue *mq;
2221 	struct file *fp;
2222 	u_int oflag, flag;
2223 	int error;
2224 
2225 	AUDIT_ARG_FD(mqd);
2226 	if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2227 		return (EINVAL);
2228 	error = getmq(td, mqd, &fp, NULL, &mq);
2229 	if (error)
2230 		return (error);
2231 	oattr->mq_maxmsg  = mq->mq_maxmsg;
2232 	oattr->mq_msgsize = mq->mq_msgsize;
2233 	oattr->mq_curmsgs = mq->mq_curmsgs;
2234 	if (attr != NULL) {
2235 		do {
2236 			oflag = flag = fp->f_flag;
2237 			flag &= ~O_NONBLOCK;
2238 			flag |= (attr->mq_flags & O_NONBLOCK);
2239 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2240 	} else
2241 		oflag = fp->f_flag;
2242 	oattr->mq_flags = (O_NONBLOCK & oflag);
2243 	fdrop(fp, td);
2244 	return (error);
2245 }
2246 
2247 int
sys_kmq_setattr(struct thread * td,struct kmq_setattr_args * uap)2248 sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2249 {
2250 	struct mq_attr attr, oattr;
2251 	int error;
2252 
2253 	if (uap->attr != NULL) {
2254 		error = copyin(uap->attr, &attr, sizeof(attr));
2255 		if (error != 0)
2256 			return (error);
2257 	}
2258 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2259 	    &oattr);
2260 	if (error == 0 && uap->oattr != NULL) {
2261 		bzero(oattr.__reserved, sizeof(oattr.__reserved));
2262 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2263 	}
2264 	return (error);
2265 }
2266 
2267 int
sys_kmq_timedreceive(struct thread * td,struct kmq_timedreceive_args * uap)2268 sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2269 {
2270 	struct mqueue *mq;
2271 	struct file *fp;
2272 	struct timespec *abs_timeout, ets;
2273 	int error;
2274 	int waitok;
2275 
2276 	AUDIT_ARG_FD(uap->mqd);
2277 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2278 	if (error)
2279 		return (error);
2280 	if (uap->abs_timeout != NULL) {
2281 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2282 		if (error != 0)
2283 			goto out;
2284 		abs_timeout = &ets;
2285 	} else
2286 		abs_timeout = NULL;
2287 	waitok = !(fp->f_flag & O_NONBLOCK);
2288 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2289 		uap->msg_prio, waitok, abs_timeout);
2290 out:
2291 	fdrop(fp, td);
2292 	return (error);
2293 }
2294 
2295 int
sys_kmq_timedsend(struct thread * td,struct kmq_timedsend_args * uap)2296 sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2297 {
2298 	struct mqueue *mq;
2299 	struct file *fp;
2300 	struct timespec *abs_timeout, ets;
2301 	int error, waitok;
2302 
2303 	AUDIT_ARG_FD(uap->mqd);
2304 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2305 	if (error)
2306 		return (error);
2307 	if (uap->abs_timeout != NULL) {
2308 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2309 		if (error != 0)
2310 			goto out;
2311 		abs_timeout = &ets;
2312 	} else
2313 		abs_timeout = NULL;
2314 	waitok = !(fp->f_flag & O_NONBLOCK);
2315 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2316 		uap->msg_prio, waitok, abs_timeout);
2317 out:
2318 	fdrop(fp, td);
2319 	return (error);
2320 }
2321 
2322 static int
kern_kmq_notify(struct thread * td,int mqd,struct sigevent * sigev)2323 kern_kmq_notify(struct thread *td, int mqd, struct sigevent *sigev)
2324 {
2325 	struct filedesc *fdp;
2326 	struct proc *p;
2327 	struct mqueue *mq;
2328 	struct file *fp, *fp2;
2329 	struct mqueue_notifier *nt, *newnt = NULL;
2330 	int error;
2331 
2332 	AUDIT_ARG_FD(mqd);
2333 	if (sigev != NULL) {
2334 		if (sigev->sigev_notify != SIGEV_SIGNAL &&
2335 		    sigev->sigev_notify != SIGEV_THREAD_ID &&
2336 		    sigev->sigev_notify != SIGEV_NONE)
2337 			return (EINVAL);
2338 		if ((sigev->sigev_notify == SIGEV_SIGNAL ||
2339 		    sigev->sigev_notify == SIGEV_THREAD_ID) &&
2340 		    !_SIG_VALID(sigev->sigev_signo))
2341 			return (EINVAL);
2342 	}
2343 	p = td->td_proc;
2344 	fdp = td->td_proc->p_fd;
2345 	error = getmq(td, mqd, &fp, NULL, &mq);
2346 	if (error)
2347 		return (error);
2348 again:
2349 	FILEDESC_SLOCK(fdp);
2350 	fp2 = fget_locked(fdp, mqd);
2351 	if (fp2 == NULL) {
2352 		FILEDESC_SUNLOCK(fdp);
2353 		error = EBADF;
2354 		goto out;
2355 	}
2356 #ifdef CAPABILITIES
2357 	error = cap_check(cap_rights(fdp, mqd), &cap_event_rights);
2358 	if (error) {
2359 		FILEDESC_SUNLOCK(fdp);
2360 		goto out;
2361 	}
2362 #endif
2363 	if (fp2 != fp) {
2364 		FILEDESC_SUNLOCK(fdp);
2365 		error = EBADF;
2366 		goto out;
2367 	}
2368 	mtx_lock(&mq->mq_mutex);
2369 	FILEDESC_SUNLOCK(fdp);
2370 	if (sigev != NULL) {
2371 		if (mq->mq_notifier != NULL) {
2372 			error = EBUSY;
2373 		} else {
2374 			PROC_LOCK(p);
2375 			nt = notifier_search(p, mqd);
2376 			if (nt == NULL) {
2377 				if (newnt == NULL) {
2378 					PROC_UNLOCK(p);
2379 					mtx_unlock(&mq->mq_mutex);
2380 					newnt = notifier_alloc();
2381 					goto again;
2382 				}
2383 			}
2384 
2385 			if (nt != NULL) {
2386 				sigqueue_take(&nt->nt_ksi);
2387 				if (newnt != NULL) {
2388 					notifier_free(newnt);
2389 					newnt = NULL;
2390 				}
2391 			} else {
2392 				nt = newnt;
2393 				newnt = NULL;
2394 				ksiginfo_init(&nt->nt_ksi);
2395 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2396 				nt->nt_ksi.ksi_code = SI_MESGQ;
2397 				nt->nt_proc = p;
2398 				nt->nt_ksi.ksi_mqd = mqd;
2399 				notifier_insert(p, nt);
2400 			}
2401 			nt->nt_sigev = *sigev;
2402 			mq->mq_notifier = nt;
2403 			PROC_UNLOCK(p);
2404 			/*
2405 			 * if there is no receivers and message queue
2406 			 * is not empty, we should send notification
2407 			 * as soon as possible.
2408 			 */
2409 			if (mq->mq_receivers == 0 &&
2410 			    !TAILQ_EMPTY(&mq->mq_msgq))
2411 				mqueue_send_notification(mq);
2412 		}
2413 	} else {
2414 		notifier_remove(p, mq, mqd);
2415 	}
2416 	mtx_unlock(&mq->mq_mutex);
2417 
2418 out:
2419 	fdrop(fp, td);
2420 	if (newnt != NULL)
2421 		notifier_free(newnt);
2422 	return (error);
2423 }
2424 
2425 int
sys_kmq_notify(struct thread * td,struct kmq_notify_args * uap)2426 sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2427 {
2428 	struct sigevent ev, *evp;
2429 	int error;
2430 
2431 	if (uap->sigev == NULL) {
2432 		evp = NULL;
2433 	} else {
2434 		error = copyin(uap->sigev, &ev, sizeof(ev));
2435 		if (error != 0)
2436 			return (error);
2437 		evp = &ev;
2438 	}
2439 	return (kern_kmq_notify(td, uap->mqd, evp));
2440 }
2441 
2442 static void
mqueue_fdclose(struct thread * td,int fd,struct file * fp)2443 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2444 {
2445 	struct mqueue *mq;
2446 #ifdef INVARIANTS
2447 	struct filedesc *fdp;
2448 
2449 	fdp = td->td_proc->p_fd;
2450 	FILEDESC_LOCK_ASSERT(fdp);
2451 #endif
2452 
2453 	if (fp->f_ops == &mqueueops) {
2454 		mq = FPTOMQ(fp);
2455 		mtx_lock(&mq->mq_mutex);
2456 		notifier_remove(td->td_proc, mq, fd);
2457 
2458 		/* have to wakeup thread in same process */
2459 		if (mq->mq_flags & MQ_RSEL) {
2460 			mq->mq_flags &= ~MQ_RSEL;
2461 			selwakeup(&mq->mq_rsel);
2462 		}
2463 		if (mq->mq_flags & MQ_WSEL) {
2464 			mq->mq_flags &= ~MQ_WSEL;
2465 			selwakeup(&mq->mq_wsel);
2466 		}
2467 		mtx_unlock(&mq->mq_mutex);
2468 	}
2469 }
2470 
2471 static void
mq_proc_exit(void * arg __unused,struct proc * p)2472 mq_proc_exit(void *arg __unused, struct proc *p)
2473 {
2474 	struct filedesc *fdp;
2475 	struct file *fp;
2476 	struct mqueue *mq;
2477 	int i;
2478 
2479 	fdp = p->p_fd;
2480 	FILEDESC_SLOCK(fdp);
2481 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2482 		fp = fget_locked(fdp, i);
2483 		if (fp != NULL && fp->f_ops == &mqueueops) {
2484 			mq = FPTOMQ(fp);
2485 			mtx_lock(&mq->mq_mutex);
2486 			notifier_remove(p, FPTOMQ(fp), i);
2487 			mtx_unlock(&mq->mq_mutex);
2488 		}
2489 	}
2490 	FILEDESC_SUNLOCK(fdp);
2491 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2492 }
2493 
2494 static int
mqf_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)2495 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2496 	struct thread *td)
2497 {
2498 	struct mqueue *mq = FPTOMQ(fp);
2499 	int revents = 0;
2500 
2501 	mtx_lock(&mq->mq_mutex);
2502 	if (events & (POLLIN | POLLRDNORM)) {
2503 		if (mq->mq_curmsgs) {
2504 			revents |= events & (POLLIN | POLLRDNORM);
2505 		} else {
2506 			mq->mq_flags |= MQ_RSEL;
2507 			selrecord(td, &mq->mq_rsel);
2508  		}
2509 	}
2510 	if (events & POLLOUT) {
2511 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2512 			revents |= POLLOUT;
2513 		else {
2514 			mq->mq_flags |= MQ_WSEL;
2515 			selrecord(td, &mq->mq_wsel);
2516 		}
2517 	}
2518 	mtx_unlock(&mq->mq_mutex);
2519 	return (revents);
2520 }
2521 
2522 static int
mqf_close(struct file * fp,struct thread * td)2523 mqf_close(struct file *fp, struct thread *td)
2524 {
2525 	struct mqfs_node *pn;
2526 
2527 	fp->f_ops = &badfileops;
2528 	pn = fp->f_data;
2529 	fp->f_data = NULL;
2530 	sx_xlock(&mqfs_data.mi_lock);
2531 	mqnode_release(pn);
2532 	sx_xunlock(&mqfs_data.mi_lock);
2533 	return (0);
2534 }
2535 
2536 static int
mqf_stat(struct file * fp,struct stat * st,struct ucred * active_cred,struct thread * td)2537 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2538 	struct thread *td)
2539 {
2540 	struct mqfs_node *pn = fp->f_data;
2541 
2542 	bzero(st, sizeof *st);
2543 	sx_xlock(&mqfs_data.mi_lock);
2544 	st->st_atim = pn->mn_atime;
2545 	st->st_mtim = pn->mn_mtime;
2546 	st->st_ctim = pn->mn_ctime;
2547 	st->st_birthtim = pn->mn_birth;
2548 	st->st_uid = pn->mn_uid;
2549 	st->st_gid = pn->mn_gid;
2550 	st->st_mode = S_IFIFO | pn->mn_mode;
2551 	sx_xunlock(&mqfs_data.mi_lock);
2552 	return (0);
2553 }
2554 
2555 static int
mqf_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)2556 mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2557     struct thread *td)
2558 {
2559 	struct mqfs_node *pn;
2560 	int error;
2561 
2562 	error = 0;
2563 	pn = fp->f_data;
2564 	sx_xlock(&mqfs_data.mi_lock);
2565 	error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN,
2566 	    active_cred);
2567 	if (error != 0)
2568 		goto out;
2569 	pn->mn_mode = mode & ACCESSPERMS;
2570 out:
2571 	sx_xunlock(&mqfs_data.mi_lock);
2572 	return (error);
2573 }
2574 
2575 static int
mqf_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)2576 mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2577     struct thread *td)
2578 {
2579 	struct mqfs_node *pn;
2580 	int error;
2581 
2582 	error = 0;
2583 	pn = fp->f_data;
2584 	sx_xlock(&mqfs_data.mi_lock);
2585 	if (uid == (uid_t)-1)
2586 		uid = pn->mn_uid;
2587 	if (gid == (gid_t)-1)
2588 		gid = pn->mn_gid;
2589 	if (((uid != pn->mn_uid && uid != active_cred->cr_uid) ||
2590 	    (gid != pn->mn_gid && !groupmember(gid, active_cred))) &&
2591 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN)))
2592 		goto out;
2593 	pn->mn_uid = uid;
2594 	pn->mn_gid = gid;
2595 out:
2596 	sx_xunlock(&mqfs_data.mi_lock);
2597 	return (error);
2598 }
2599 
2600 static int
mqf_kqfilter(struct file * fp,struct knote * kn)2601 mqf_kqfilter(struct file *fp, struct knote *kn)
2602 {
2603 	struct mqueue *mq = FPTOMQ(fp);
2604 	int error = 0;
2605 
2606 	if (kn->kn_filter == EVFILT_READ) {
2607 		kn->kn_fop = &mq_rfiltops;
2608 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2609 	} else if (kn->kn_filter == EVFILT_WRITE) {
2610 		kn->kn_fop = &mq_wfiltops;
2611 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2612 	} else
2613 		error = EINVAL;
2614 	return (error);
2615 }
2616 
2617 static void
filt_mqdetach(struct knote * kn)2618 filt_mqdetach(struct knote *kn)
2619 {
2620 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2621 
2622 	if (kn->kn_filter == EVFILT_READ)
2623 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2624 	else if (kn->kn_filter == EVFILT_WRITE)
2625 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2626 	else
2627 		panic("filt_mqdetach");
2628 }
2629 
2630 static int
filt_mqread(struct knote * kn,long hint)2631 filt_mqread(struct knote *kn, long hint)
2632 {
2633 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2634 
2635 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2636 	return (mq->mq_curmsgs != 0);
2637 }
2638 
2639 static int
filt_mqwrite(struct knote * kn,long hint)2640 filt_mqwrite(struct knote *kn, long hint)
2641 {
2642 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2643 
2644 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2645 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2646 }
2647 
2648 static int
mqf_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)2649 mqf_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2650 {
2651 
2652 	kif->kf_type = KF_TYPE_MQUEUE;
2653 	return (0);
2654 }
2655 
2656 static struct fileops mqueueops = {
2657 	.fo_read		= invfo_rdwr,
2658 	.fo_write		= invfo_rdwr,
2659 	.fo_truncate		= invfo_truncate,
2660 	.fo_ioctl		= invfo_ioctl,
2661 	.fo_poll		= mqf_poll,
2662 	.fo_kqfilter		= mqf_kqfilter,
2663 	.fo_stat		= mqf_stat,
2664 	.fo_close		= mqf_close,
2665 	.fo_chmod		= mqf_chmod,
2666 	.fo_chown		= mqf_chown,
2667 	.fo_sendfile		= invfo_sendfile,
2668 	.fo_fill_kinfo		= mqf_fill_kinfo,
2669 	.fo_cmp			= file_kcmp_generic,
2670 	.fo_flags		= DFLAG_PASSABLE,
2671 };
2672 
2673 static struct vop_vector mqfs_vnodeops = {
2674 	.vop_default 		= &default_vnodeops,
2675 	.vop_access		= mqfs_access,
2676 	.vop_cachedlookup	= mqfs_lookup,
2677 	.vop_lookup		= vfs_cache_lookup,
2678 	.vop_reclaim		= mqfs_reclaim,
2679 	.vop_create		= mqfs_create,
2680 	.vop_remove		= mqfs_remove,
2681 	.vop_inactive		= mqfs_inactive,
2682 	.vop_open		= mqfs_open,
2683 	.vop_close		= mqfs_close,
2684 	.vop_getattr		= mqfs_getattr,
2685 	.vop_setattr		= mqfs_setattr,
2686 	.vop_read		= mqfs_read,
2687 	.vop_write		= VOP_EOPNOTSUPP,
2688 	.vop_readdir		= mqfs_readdir,
2689 	.vop_mkdir		= VOP_EOPNOTSUPP,
2690 	.vop_rmdir		= VOP_EOPNOTSUPP
2691 };
2692 VFS_VOP_VECTOR_REGISTER(mqfs_vnodeops);
2693 
2694 static struct vfsops mqfs_vfsops = {
2695 	.vfs_init 		= mqfs_init,
2696 	.vfs_uninit		= mqfs_uninit,
2697 	.vfs_mount		= mqfs_mount,
2698 	.vfs_unmount		= mqfs_unmount,
2699 	.vfs_root		= mqfs_root,
2700 	.vfs_statfs		= mqfs_statfs,
2701 };
2702 
2703 static struct vfsconf mqueuefs_vfsconf = {
2704 	.vfc_version = VFS_VERSION,
2705 	.vfc_name = "mqueuefs",
2706 	.vfc_vfsops = &mqfs_vfsops,
2707 	.vfc_typenum = -1,
2708 	.vfc_flags = VFCF_SYNTHETIC
2709 };
2710 
2711 static struct syscall_helper_data mq_syscalls[] = {
2712 	SYSCALL_INIT_HELPER(kmq_open),
2713 	SYSCALL_INIT_HELPER_F(kmq_setattr, SYF_CAPENABLED),
2714 	SYSCALL_INIT_HELPER_F(kmq_timedsend, SYF_CAPENABLED),
2715 	SYSCALL_INIT_HELPER_F(kmq_timedreceive, SYF_CAPENABLED),
2716 	SYSCALL_INIT_HELPER_F(kmq_notify, SYF_CAPENABLED),
2717 	SYSCALL_INIT_HELPER(kmq_unlink),
2718 	SYSCALL_INIT_LAST
2719 };
2720 
2721 #ifdef COMPAT_FREEBSD32
2722 #include <compat/freebsd32/freebsd32.h>
2723 #include <compat/freebsd32/freebsd32_proto.h>
2724 #include <compat/freebsd32/freebsd32_signal.h>
2725 #include <compat/freebsd32/freebsd32_syscall.h>
2726 #include <compat/freebsd32/freebsd32_util.h>
2727 
2728 static void
mq_attr_from32(const struct mq_attr32 * from,struct mq_attr * to)2729 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2730 {
2731 
2732 	to->mq_flags = from->mq_flags;
2733 	to->mq_maxmsg = from->mq_maxmsg;
2734 	to->mq_msgsize = from->mq_msgsize;
2735 	to->mq_curmsgs = from->mq_curmsgs;
2736 }
2737 
2738 static void
mq_attr_to32(const struct mq_attr * from,struct mq_attr32 * to)2739 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2740 {
2741 
2742 	to->mq_flags = from->mq_flags;
2743 	to->mq_maxmsg = from->mq_maxmsg;
2744 	to->mq_msgsize = from->mq_msgsize;
2745 	to->mq_curmsgs = from->mq_curmsgs;
2746 }
2747 
2748 int
freebsd32_kmq_open(struct thread * td,struct freebsd32_kmq_open_args * uap)2749 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2750 {
2751 	struct mq_attr attr;
2752 	struct mq_attr32 attr32;
2753 	int flags, error;
2754 
2755 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2756 		return (EINVAL);
2757 	flags = FFLAGS(uap->flags);
2758 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2759 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2760 		if (error)
2761 			return (error);
2762 		mq_attr_from32(&attr32, &attr);
2763 	}
2764 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2765 	    uap->attr != NULL ? &attr : NULL));
2766 }
2767 
2768 int
freebsd32_kmq_setattr(struct thread * td,struct freebsd32_kmq_setattr_args * uap)2769 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2770 {
2771 	struct mq_attr attr, oattr;
2772 	struct mq_attr32 attr32, oattr32;
2773 	int error;
2774 
2775 	if (uap->attr != NULL) {
2776 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2777 		if (error != 0)
2778 			return (error);
2779 		mq_attr_from32(&attr32, &attr);
2780 	}
2781 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2782 	    &oattr);
2783 	if (error == 0 && uap->oattr != NULL) {
2784 		mq_attr_to32(&oattr, &oattr32);
2785 		bzero(oattr32.__reserved, sizeof(oattr32.__reserved));
2786 		error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2787 	}
2788 	return (error);
2789 }
2790 
2791 int
freebsd32_kmq_timedsend(struct thread * td,struct freebsd32_kmq_timedsend_args * uap)2792 freebsd32_kmq_timedsend(struct thread *td,
2793     struct freebsd32_kmq_timedsend_args *uap)
2794 {
2795 	struct mqueue *mq;
2796 	struct file *fp;
2797 	struct timespec32 ets32;
2798 	struct timespec *abs_timeout, ets;
2799 	int error;
2800 	int waitok;
2801 
2802 	AUDIT_ARG_FD(uap->mqd);
2803 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2804 	if (error)
2805 		return (error);
2806 	if (uap->abs_timeout != NULL) {
2807 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2808 		if (error != 0)
2809 			goto out;
2810 		CP(ets32, ets, tv_sec);
2811 		CP(ets32, ets, tv_nsec);
2812 		abs_timeout = &ets;
2813 	} else
2814 		abs_timeout = NULL;
2815 	waitok = !(fp->f_flag & O_NONBLOCK);
2816 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2817 	    uap->msg_prio, waitok, abs_timeout);
2818 out:
2819 	fdrop(fp, td);
2820 	return (error);
2821 }
2822 
2823 int
freebsd32_kmq_timedreceive(struct thread * td,struct freebsd32_kmq_timedreceive_args * uap)2824 freebsd32_kmq_timedreceive(struct thread *td,
2825     struct freebsd32_kmq_timedreceive_args *uap)
2826 {
2827 	struct mqueue *mq;
2828 	struct file *fp;
2829 	struct timespec32 ets32;
2830 	struct timespec *abs_timeout, ets;
2831 	int error, waitok;
2832 
2833 	AUDIT_ARG_FD(uap->mqd);
2834 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2835 	if (error)
2836 		return (error);
2837 	if (uap->abs_timeout != NULL) {
2838 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2839 		if (error != 0)
2840 			goto out;
2841 		CP(ets32, ets, tv_sec);
2842 		CP(ets32, ets, tv_nsec);
2843 		abs_timeout = &ets;
2844 	} else
2845 		abs_timeout = NULL;
2846 	waitok = !(fp->f_flag & O_NONBLOCK);
2847 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2848 	    uap->msg_prio, waitok, abs_timeout);
2849 out:
2850 	fdrop(fp, td);
2851 	return (error);
2852 }
2853 
2854 int
freebsd32_kmq_notify(struct thread * td,struct freebsd32_kmq_notify_args * uap)2855 freebsd32_kmq_notify(struct thread *td, struct freebsd32_kmq_notify_args *uap)
2856 {
2857 	struct sigevent ev, *evp;
2858 	struct sigevent32 ev32;
2859 	int error;
2860 
2861 	if (uap->sigev == NULL) {
2862 		evp = NULL;
2863 	} else {
2864 		error = copyin(uap->sigev, &ev32, sizeof(ev32));
2865 		if (error != 0)
2866 			return (error);
2867 		error = convert_sigevent32(&ev32, &ev);
2868 		if (error != 0)
2869 			return (error);
2870 		evp = &ev;
2871 	}
2872 	return (kern_kmq_notify(td, uap->mqd, evp));
2873 }
2874 
2875 static struct syscall_helper_data mq32_syscalls[] = {
2876 	SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2877 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_setattr, SYF_CAPENABLED),
2878 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedsend, SYF_CAPENABLED),
2879 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedreceive, SYF_CAPENABLED),
2880 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_notify, SYF_CAPENABLED),
2881 	SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink),
2882 	SYSCALL_INIT_LAST
2883 };
2884 #endif
2885 
2886 static int
mqinit(void)2887 mqinit(void)
2888 {
2889 	int error;
2890 
2891 	error = syscall_helper_register(mq_syscalls, SY_THR_STATIC_KLD);
2892 	if (error != 0)
2893 		return (error);
2894 #ifdef COMPAT_FREEBSD32
2895 	error = syscall32_helper_register(mq32_syscalls, SY_THR_STATIC_KLD);
2896 	if (error != 0)
2897 		return (error);
2898 #endif
2899 	return (0);
2900 }
2901 
2902 static int
mqunload(void)2903 mqunload(void)
2904 {
2905 
2906 #ifdef COMPAT_FREEBSD32
2907 	syscall32_helper_unregister(mq32_syscalls);
2908 #endif
2909 	syscall_helper_unregister(mq_syscalls);
2910 	return (0);
2911 }
2912 
2913 static int
mq_modload(struct module * module,int cmd,void * arg)2914 mq_modload(struct module *module, int cmd, void *arg)
2915 {
2916 	int error = 0;
2917 
2918 	error = vfs_modevent(module, cmd, arg);
2919 	if (error != 0)
2920 		return (error);
2921 
2922 	switch (cmd) {
2923 	case MOD_LOAD:
2924 		error = mqinit();
2925 		if (error != 0)
2926 			mqunload();
2927 		break;
2928 	case MOD_UNLOAD:
2929 		error = mqunload();
2930 		break;
2931 	default:
2932 		break;
2933 	}
2934 	return (error);
2935 }
2936 
2937 static moduledata_t mqueuefs_mod = {
2938 	"mqueuefs",
2939 	mq_modload,
2940 	&mqueuefs_vfsconf
2941 };
2942 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2943 MODULE_VERSION(mqueuefs, 1);
2944