xref: /freebsd/sys/kern/uipc_mqueue.c (revision 0d224af399a66f00a5b33e5512fc018062cabf1d)
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 SYSCTL_INT(_kern_mqueue, OID_AUTO, default_maxmsg, CTLFLAG_RD,
204     &default_maxmsg, 0, "Default maximum messages in queue");
205 static int	default_msgsize = 1024;
206 SYSCTL_INT(_kern_mqueue, OID_AUTO, default_msgsize, CTLFLAG_RD,
207     &default_msgsize, 0, "Default maximum message size");
208 
209 static int	maxmsg = 100;
210 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
211     &maxmsg, 0, "maximum messages in queue");
212 static int	maxmsgsize = 16384;
213 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
214     &maxmsgsize, 0, "maximum message size");
215 static int	maxmq = 100;
216 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
217     &maxmq, 0, "maximum message queues");
218 static int	curmq = 0;
219 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
220     &curmq, 0, "current message queue number");
221 static int	unloadable = 0;
222 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
223 
224 static eventhandler_tag exit_tag;
225 
226 /* Only one instance per-system */
227 static struct mqfs_info		mqfs_data;
228 static uma_zone_t		mqnode_zone;
229 static uma_zone_t		mqueue_zone;
230 static uma_zone_t		mvdata_zone;
231 static uma_zone_t		mqnoti_zone;
232 static struct vop_vector	mqfs_vnodeops;
233 static const struct fileops	mqueueops;
234 static unsigned			mqfs_osd_jail_slot;
235 
236 /*
237  * Directory structure construction and manipulation
238  */
239 #ifdef notyet
240 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
241 	const char *name, int namelen, struct ucred *cred, int mode);
242 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
243 	const char *name, int namelen, struct ucred *cred, int mode);
244 #endif
245 
246 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
247 	const char *name, int namelen, struct ucred *cred, int mode);
248 static int	mqfs_destroy(struct mqfs_node *mn);
249 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
250 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
251 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
252 static int	mqfs_prison_remove(void *obj, void *data);
253 
254 /*
255  * Message queue construction and maniplation
256  */
257 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
258 static void	mqueue_free(struct mqueue *mq);
259 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
260 			size_t msg_len, unsigned msg_prio, int waitok,
261 			const struct timespec *abs_timeout);
262 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
263 			size_t msg_len, unsigned *msg_prio, int waitok,
264 			const struct timespec *abs_timeout);
265 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
266 			int timo);
267 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
268 			int timo);
269 static void	mqueue_send_notification(struct mqueue *mq);
270 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
271 static void	mq_proc_exit(void *arg, struct proc *p);
272 
273 /*
274  * kqueue filters
275  */
276 static void	filt_mqdetach(struct knote *kn);
277 static int	filt_mqread(struct knote *kn, long hint);
278 static int	filt_mqwrite(struct knote *kn, long hint);
279 
280 static const struct filterops mq_rfiltops = {
281 	.f_isfd = 1,
282 	.f_detach = filt_mqdetach,
283 	.f_event = filt_mqread,
284 };
285 static const struct filterops mq_wfiltops = {
286 	.f_isfd = 1,
287 	.f_detach = filt_mqdetach,
288 	.f_event = filt_mqwrite,
289 };
290 
291 /*
292  * Initialize fileno bitmap
293  */
294 static void
mqfs_fileno_init(struct mqfs_info * mi)295 mqfs_fileno_init(struct mqfs_info *mi)
296 {
297 	struct unrhdr *up;
298 
299 	up = new_unrhdr(1, INT_MAX, NULL);
300 	mi->mi_unrhdr = up;
301 }
302 
303 /*
304  * Tear down fileno bitmap
305  */
306 static void
mqfs_fileno_uninit(struct mqfs_info * mi)307 mqfs_fileno_uninit(struct mqfs_info *mi)
308 {
309 	struct unrhdr *up;
310 
311 	up = mi->mi_unrhdr;
312 	mi->mi_unrhdr = NULL;
313 	delete_unrhdr(up);
314 }
315 
316 /*
317  * Allocate a file number
318  */
319 static void
mqfs_fileno_alloc(struct mqfs_info * mi,struct mqfs_node * mn)320 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
321 {
322 	/* make sure our parent has a file number */
323 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
324 		mqfs_fileno_alloc(mi, mn->mn_parent);
325 
326 	switch (mn->mn_type) {
327 	case mqfstype_root:
328 	case mqfstype_dir:
329 	case mqfstype_file:
330 	case mqfstype_symlink:
331 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
332 		break;
333 	case mqfstype_this:
334 		KASSERT(mn->mn_parent != NULL,
335 		    ("mqfstype_this node has no parent"));
336 		mn->mn_fileno = mn->mn_parent->mn_fileno;
337 		break;
338 	case mqfstype_parent:
339 		KASSERT(mn->mn_parent != NULL,
340 		    ("mqfstype_parent node has no parent"));
341 		if (mn->mn_parent == mi->mi_root) {
342 			mn->mn_fileno = mn->mn_parent->mn_fileno;
343 			break;
344 		}
345 		KASSERT(mn->mn_parent->mn_parent != NULL,
346 		    ("mqfstype_parent node has no grandparent"));
347 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
348 		break;
349 	default:
350 		KASSERT(0,
351 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
352 			mn->mn_type));
353 		break;
354 	}
355 }
356 
357 /*
358  * Release a file number
359  */
360 static void
mqfs_fileno_free(struct mqfs_info * mi,struct mqfs_node * mn)361 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
362 {
363 	switch (mn->mn_type) {
364 	case mqfstype_root:
365 	case mqfstype_dir:
366 	case mqfstype_file:
367 	case mqfstype_symlink:
368 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
369 		break;
370 	case mqfstype_this:
371 	case mqfstype_parent:
372 		/* ignore these, as they don't "own" their file number */
373 		break;
374 	default:
375 		KASSERT(0,
376 		    ("mqfs_fileno_free() called for unknown type node: %d",
377 			mn->mn_type));
378 		break;
379 	}
380 }
381 
382 static __inline struct mqfs_node *
mqnode_alloc(void)383 mqnode_alloc(void)
384 {
385 	return (uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO));
386 }
387 
388 static __inline void
mqnode_free(struct mqfs_node * node)389 mqnode_free(struct mqfs_node *node)
390 {
391 	uma_zfree(mqnode_zone, node);
392 }
393 
394 static __inline void
mqnode_addref(struct mqfs_node * node)395 mqnode_addref(struct mqfs_node *node)
396 {
397 	atomic_add_int(&node->mn_refcount, 1);
398 }
399 
400 static __inline void
mqnode_release(struct mqfs_node * node)401 mqnode_release(struct mqfs_node *node)
402 {
403 	struct mqfs_info *mqfs;
404 	int old, exp;
405 
406 	mqfs = node->mn_info;
407 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
408 	if (node->mn_type == mqfstype_dir ||
409 	    node->mn_type == mqfstype_root)
410 		exp = 3; /* include . and .. */
411 	else
412 		exp = 1;
413 	if (old == exp) {
414 		int locked = sx_xlocked(&mqfs->mi_lock);
415 		if (!locked)
416 			sx_xlock(&mqfs->mi_lock);
417 		mqfs_destroy(node);
418 		if (!locked)
419 			sx_xunlock(&mqfs->mi_lock);
420 	}
421 }
422 
423 /*
424  * Add a node to a directory
425  */
426 static int
mqfs_add_node(struct mqfs_node * parent,struct mqfs_node * node)427 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
428 {
429 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
430 	KASSERT(parent->mn_info != NULL,
431 	    ("%s(): parent has no mn_info", __func__));
432 	KASSERT(parent->mn_type == mqfstype_dir ||
433 	    parent->mn_type == mqfstype_root,
434 	    ("%s(): parent is not a directory", __func__));
435 
436 	node->mn_info = parent->mn_info;
437 	node->mn_parent = parent;
438 	LIST_INIT(&node->mn_children);
439 	LIST_INIT(&node->mn_vnodes);
440 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
441 	mqnode_addref(parent);
442 	return (0);
443 }
444 
445 static struct mqfs_node *
mqfs_create_node(const char * name,int namelen,struct ucred * cred,int mode,int nodetype)446 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
447 	int nodetype)
448 {
449 	struct mqfs_node *node;
450 
451 	node = mqnode_alloc();
452 	strncpy(node->mn_name, name, namelen);
453 	node->mn_pr_root = cred->cr_prison->pr_root;
454 	node->mn_type = nodetype;
455 	node->mn_refcount = 1;
456 	vfs_timestamp(&node->mn_birth);
457 	node->mn_ctime = node->mn_atime = node->mn_mtime =
458 	    node->mn_birth;
459 	node->mn_uid = cred->cr_uid;
460 	node->mn_gid = cred->cr_gid;
461 	node->mn_mode = mode;
462 	return (node);
463 }
464 
465 /*
466  * Create a file
467  */
468 static struct mqfs_node *
mqfs_create_file(struct mqfs_node * parent,const char * name,int namelen,struct ucred * cred,int mode)469 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
470 	struct ucred *cred, int mode)
471 {
472 	struct mqfs_node *node;
473 
474 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
475 	if (mqfs_add_node(parent, node) != 0) {
476 		mqnode_free(node);
477 		return (NULL);
478 	}
479 	return (node);
480 }
481 
482 /*
483  * Add . and .. to a directory
484  */
485 static int
mqfs_fixup_dir(struct mqfs_node * parent)486 mqfs_fixup_dir(struct mqfs_node *parent)
487 {
488 	struct mqfs_node *dir;
489 
490 	dir = mqnode_alloc();
491 	dir->mn_name[0] = '.';
492 	dir->mn_type = mqfstype_this;
493 	dir->mn_refcount = 1;
494 	if (mqfs_add_node(parent, dir) != 0) {
495 		mqnode_free(dir);
496 		return (-1);
497 	}
498 
499 	dir = mqnode_alloc();
500 	dir->mn_name[0] = dir->mn_name[1] = '.';
501 	dir->mn_type = mqfstype_parent;
502 	dir->mn_refcount = 1;
503 
504 	if (mqfs_add_node(parent, dir) != 0) {
505 		mqnode_free(dir);
506 		return (-1);
507 	}
508 
509 	return (0);
510 }
511 
512 #ifdef notyet
513 
514 /*
515  * Create a directory
516  */
517 static struct mqfs_node *
mqfs_create_dir(struct mqfs_node * parent,const char * name,int namelen,struct ucred * cred,int mode)518 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
519 	struct ucred *cred, int mode)
520 {
521 	struct mqfs_node *node;
522 
523 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
524 	if (mqfs_add_node(parent, node) != 0) {
525 		mqnode_free(node);
526 		return (NULL);
527 	}
528 
529 	if (mqfs_fixup_dir(node) != 0) {
530 		mqfs_destroy(node);
531 		return (NULL);
532 	}
533 	return (node);
534 }
535 
536 /*
537  * Create a symlink
538  */
539 static struct mqfs_node *
mqfs_create_link(struct mqfs_node * parent,const char * name,int namelen,struct ucred * cred,int mode)540 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
541 	struct ucred *cred, int mode)
542 {
543 	struct mqfs_node *node;
544 
545 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
546 	if (mqfs_add_node(parent, node) != 0) {
547 		mqnode_free(node);
548 		return (NULL);
549 	}
550 	return (node);
551 }
552 
553 #endif
554 
555 /*
556  * Destroy a node or a tree of nodes
557  */
558 static int
mqfs_destroy(struct mqfs_node * node)559 mqfs_destroy(struct mqfs_node *node)
560 {
561 	struct mqfs_node *parent;
562 
563 	KASSERT(node != NULL,
564 	    ("%s(): node is NULL", __func__));
565 	KASSERT(node->mn_info != NULL,
566 	    ("%s(): node has no mn_info", __func__));
567 
568 	/* destroy children */
569 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
570 		while (! LIST_EMPTY(&node->mn_children))
571 			mqfs_destroy(LIST_FIRST(&node->mn_children));
572 
573 	/* unlink from parent */
574 	if ((parent = node->mn_parent) != NULL) {
575 		KASSERT(parent->mn_info == node->mn_info,
576 		    ("%s(): parent has different mn_info", __func__));
577 		LIST_REMOVE(node, mn_sibling);
578 	}
579 
580 	if (node->mn_fileno != 0)
581 		mqfs_fileno_free(node->mn_info, node);
582 	if (node->mn_data != NULL)
583 		mqueue_free(node->mn_data);
584 	mqnode_free(node);
585 	return (0);
586 }
587 
588 /*
589  * Mount a mqfs instance
590  */
591 static int
mqfs_mount(struct mount * mp)592 mqfs_mount(struct mount *mp)
593 {
594 	struct statfs *sbp;
595 
596 	if (mp->mnt_flag & MNT_UPDATE)
597 		return (EOPNOTSUPP);
598 
599 	mp->mnt_data = &mqfs_data;
600 	MNT_ILOCK(mp);
601 	mp->mnt_flag |= MNT_LOCAL;
602 	MNT_IUNLOCK(mp);
603 	vfs_getnewfsid(mp);
604 
605 	sbp = &mp->mnt_stat;
606 	vfs_mountedfrom(mp, "mqueue");
607 	sbp->f_bsize = PAGE_SIZE;
608 	sbp->f_iosize = PAGE_SIZE;
609 	sbp->f_blocks = 1;
610 	sbp->f_bfree = 1;
611 	sbp->f_bavail = 0;
612 	sbp->f_files = 0;
613 	sbp->f_ffree = 0;
614 	return (0);
615 }
616 
617 /*
618  * Unmount a mqfs instance
619  */
620 static int
mqfs_unmount(struct mount * mp,int mntflags)621 mqfs_unmount(struct mount *mp, int mntflags)
622 {
623 	int error;
624 
625 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0,
626 	    curthread);
627 	return (error);
628 }
629 
630 /*
631  * Return a root vnode
632  */
633 static int
mqfs_root(struct mount * mp,int flags,struct vnode ** vpp)634 mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
635 {
636 	struct mqfs_info *mqfs;
637 	int ret;
638 
639 	mqfs = VFSTOMQFS(mp);
640 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
641 	return (ret);
642 }
643 
644 /*
645  * Return filesystem stats
646  */
647 static int
mqfs_statfs(struct mount * mp,struct statfs * sbp)648 mqfs_statfs(struct mount *mp, struct statfs *sbp)
649 {
650 	/* XXX update statistics */
651 	return (0);
652 }
653 
654 /*
655  * Initialize a mqfs instance
656  */
657 static int
mqfs_init(struct vfsconf * vfc)658 mqfs_init(struct vfsconf *vfc)
659 {
660 	struct mqfs_node *root;
661 	struct mqfs_info *mi;
662 	osd_method_t methods[PR_MAXMETHOD] = {
663 	    [PR_METHOD_REMOVE] = mqfs_prison_remove,
664 	};
665 
666 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
667 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
668 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
669 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
670 	mvdata_zone = uma_zcreate("mvdata",
671 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
672 		NULL, UMA_ALIGN_PTR, 0);
673 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
674 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
675 	mi = &mqfs_data;
676 	sx_init(&mi->mi_lock, "mqfs lock");
677 	/* set up the root diretory */
678 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
679 		mqfstype_root);
680 	root->mn_info = mi;
681 	LIST_INIT(&root->mn_children);
682 	LIST_INIT(&root->mn_vnodes);
683 	mi->mi_root = root;
684 	mqfs_fileno_init(mi);
685 	mqfs_fileno_alloc(mi, root);
686 	mqfs_fixup_dir(root);
687 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
688 	    EVENTHANDLER_PRI_ANY);
689 	mq_fdclose = mqueue_fdclose;
690 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
691 	mqfs_osd_jail_slot = osd_jail_register(NULL, methods);
692 	return (0);
693 }
694 
695 /*
696  * Destroy a mqfs instance
697  */
698 static int
mqfs_uninit(struct vfsconf * vfc)699 mqfs_uninit(struct vfsconf *vfc)
700 {
701 	struct mqfs_info *mi;
702 
703 	if (!unloadable)
704 		return (EOPNOTSUPP);
705 	osd_jail_deregister(mqfs_osd_jail_slot);
706 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
707 	mi = &mqfs_data;
708 	mqfs_destroy(mi->mi_root);
709 	mi->mi_root = NULL;
710 	mqfs_fileno_uninit(mi);
711 	sx_destroy(&mi->mi_lock);
712 	uma_zdestroy(mqnode_zone);
713 	uma_zdestroy(mqueue_zone);
714 	uma_zdestroy(mvdata_zone);
715 	uma_zdestroy(mqnoti_zone);
716 	return (0);
717 }
718 
719 /*
720  * task routine
721  */
722 static void
do_recycle(void * context,int pending __unused)723 do_recycle(void *context, int pending __unused)
724 {
725 	struct vnode *vp = (struct vnode *)context;
726 
727 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
728 	vrecycle(vp);
729 	VOP_UNLOCK(vp);
730 	vdrop(vp);
731 }
732 
733 /*
734  * Allocate a vnode
735  */
736 static int
mqfs_allocv(struct mount * mp,struct vnode ** vpp,struct mqfs_node * pn)737 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
738 {
739 	struct mqfs_vdata *vd;
740 	struct mqfs_info  *mqfs;
741 	struct vnode *newvpp;
742 	int error;
743 
744 	mqfs = pn->mn_info;
745 	*vpp = NULL;
746 	sx_xlock(&mqfs->mi_lock);
747 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
748 		if (vd->mv_vnode->v_mount == mp) {
749 			vhold(vd->mv_vnode);
750 			break;
751 		}
752 	}
753 
754 	if (vd != NULL) {
755 found:
756 		*vpp = vd->mv_vnode;
757 		sx_xunlock(&mqfs->mi_lock);
758 		error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE);
759 		vdrop(*vpp);
760 		return (error);
761 	}
762 	sx_xunlock(&mqfs->mi_lock);
763 
764 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
765 	if (error)
766 		return (error);
767 	vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
768 	error = insmntque(newvpp, mp);
769 	if (error != 0)
770 		return (error);
771 
772 	sx_xlock(&mqfs->mi_lock);
773 	/*
774 	 * Check if it has already been allocated
775 	 * while we were blocked.
776 	 */
777 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
778 		if (vd->mv_vnode->v_mount == mp) {
779 			vhold(vd->mv_vnode);
780 			sx_xunlock(&mqfs->mi_lock);
781 
782 			vgone(newvpp);
783 			vput(newvpp);
784 			goto found;
785 		}
786 	}
787 
788 	*vpp = newvpp;
789 
790 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
791 	(*vpp)->v_data = vd;
792 	vd->mv_vnode = *vpp;
793 	vd->mv_node = pn;
794 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
795 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
796 	mqnode_addref(pn);
797 	switch (pn->mn_type) {
798 	case mqfstype_root:
799 		(*vpp)->v_vflag = VV_ROOT;
800 		/* fall through */
801 	case mqfstype_dir:
802 	case mqfstype_this:
803 	case mqfstype_parent:
804 		(*vpp)->v_type = VDIR;
805 		break;
806 	case mqfstype_file:
807 		(*vpp)->v_type = VREG;
808 		break;
809 	case mqfstype_symlink:
810 		(*vpp)->v_type = VLNK;
811 		break;
812 	case mqfstype_none:
813 		KASSERT(0, ("mqfs_allocf called for null node\n"));
814 	default:
815 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
816 	}
817 	sx_xunlock(&mqfs->mi_lock);
818 	vn_set_state(*vpp, VSTATE_CONSTRUCTED);
819 	return (0);
820 }
821 
822 /*
823  * Search a directory entry
824  */
825 static struct mqfs_node *
mqfs_search(struct mqfs_node * pd,const char * name,int len,struct ucred * cred)826 mqfs_search(struct mqfs_node *pd, const char *name, int len, struct ucred *cred)
827 {
828 	struct mqfs_node *pn;
829 	const void *pr_root;
830 
831 	sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
832 	pr_root = cred->cr_prison->pr_root;
833 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
834 		/* Only match names within the same prison root directory */
835 		if ((pn->mn_pr_root == NULL || pn->mn_pr_root == pr_root) &&
836 		    strncmp(pn->mn_name, name, len) == 0 &&
837 		    pn->mn_name[len] == '\0')
838 			return (pn);
839 	}
840 	return (NULL);
841 }
842 
843 /*
844  * Look up a file or directory.
845  */
846 static int
mqfs_lookupx(struct vop_cachedlookup_args * ap)847 mqfs_lookupx(struct vop_cachedlookup_args *ap)
848 {
849 	struct componentname *cnp;
850 	struct vnode *dvp, **vpp;
851 	struct mqfs_node *pd;
852 	struct mqfs_node *pn;
853 	struct mqfs_info *mqfs;
854 	uint64_t flags;
855 	int nameiop, error, namelen;
856 	char *pname;
857 	struct thread *td;
858 
859 	td = curthread;
860 	cnp = ap->a_cnp;
861 	vpp = ap->a_vpp;
862 	dvp = ap->a_dvp;
863 	pname = cnp->cn_nameptr;
864 	namelen = cnp->cn_namelen;
865 	flags = cnp->cn_flags;
866 	nameiop = cnp->cn_nameiop;
867 	pd = VTON(dvp);
868 	pn = NULL;
869 	mqfs = pd->mn_info;
870 	*vpp = NULLVP;
871 
872 	if (dvp->v_type != VDIR)
873 		return (ENOTDIR);
874 
875 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
876 	if (error)
877 		return (error);
878 
879 	/* shortcut: check if the name is too long */
880 	if (cnp->cn_namelen >= MQFS_NAMELEN)
881 		return (ENOENT);
882 
883 	/* self */
884 	if (namelen == 1 && pname[0] == '.') {
885 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
886 			return (EINVAL);
887 		pn = pd;
888 		*vpp = dvp;
889 		VREF(dvp);
890 		return (0);
891 	}
892 
893 	/* parent */
894 	if (cnp->cn_flags & ISDOTDOT) {
895 		if (dvp->v_vflag & VV_ROOT)
896 			return (EIO);
897 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
898 			return (EINVAL);
899 		VOP_UNLOCK(dvp);
900 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
901 		pn = pd->mn_parent;
902 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
903 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
904 		return (error);
905 	}
906 
907 	/* named node */
908 	sx_xlock(&mqfs->mi_lock);
909 	pn = mqfs_search(pd, pname, namelen, cnp->cn_cred);
910 	if (pn != NULL)
911 		mqnode_addref(pn);
912 	sx_xunlock(&mqfs->mi_lock);
913 
914 	/* found */
915 	if (pn != NULL) {
916 		/* DELETE */
917 		if (nameiop == DELETE && (flags & ISLASTCN)) {
918 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
919 			if (error) {
920 				mqnode_release(pn);
921 				return (error);
922 			}
923 			if (*vpp == dvp) {
924 				VREF(dvp);
925 				*vpp = dvp;
926 				mqnode_release(pn);
927 				return (0);
928 			}
929 		}
930 
931 		/* allocate vnode */
932 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
933 		mqnode_release(pn);
934 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
935 			cache_enter(dvp, *vpp, cnp);
936 		return (error);
937 	}
938 
939 	/* not found */
940 
941 	/* will create a new entry in the directory ? */
942 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
943 	    && (flags & ISLASTCN)) {
944 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
945 		if (error)
946 			return (error);
947 		return (EJUSTRETURN);
948 	}
949 	return (ENOENT);
950 }
951 
952 #if 0
953 struct vop_lookup_args {
954 	struct vop_generic_args a_gen;
955 	struct vnode *a_dvp;
956 	struct vnode **a_vpp;
957 	struct componentname *a_cnp;
958 };
959 #endif
960 
961 /*
962  * vnode lookup operation
963  */
964 static int
mqfs_lookup(struct vop_cachedlookup_args * ap)965 mqfs_lookup(struct vop_cachedlookup_args *ap)
966 {
967 	int rc;
968 
969 	rc = mqfs_lookupx(ap);
970 	return (rc);
971 }
972 
973 #if 0
974 struct vop_create_args {
975 	struct vnode *a_dvp;
976 	struct vnode **a_vpp;
977 	struct componentname *a_cnp;
978 	struct vattr *a_vap;
979 };
980 #endif
981 
982 /*
983  * vnode creation operation
984  */
985 static int
mqfs_create(struct vop_create_args * ap)986 mqfs_create(struct vop_create_args *ap)
987 {
988 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
989 	struct componentname *cnp = ap->a_cnp;
990 	struct mqfs_node *pd;
991 	struct mqfs_node *pn;
992 	struct mqueue *mq;
993 	int error;
994 
995 	pd = VTON(ap->a_dvp);
996 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
997 		return (ENOTDIR);
998 	mq = mqueue_alloc(NULL);
999 	if (mq == NULL)
1000 		return (EAGAIN);
1001 	sx_xlock(&mqfs->mi_lock);
1002 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
1003 		cnp->cn_cred, ap->a_vap->va_mode);
1004 	if (pn == NULL) {
1005 		sx_xunlock(&mqfs->mi_lock);
1006 		error = ENOSPC;
1007 	} else {
1008 		mqnode_addref(pn);
1009 		sx_xunlock(&mqfs->mi_lock);
1010 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1011 		mqnode_release(pn);
1012 		if (error)
1013 			mqfs_destroy(pn);
1014 		else
1015 			pn->mn_data = mq;
1016 	}
1017 	if (error)
1018 		mqueue_free(mq);
1019 	return (error);
1020 }
1021 
1022 /*
1023  * Remove an entry
1024  */
1025 static int
do_unlink(struct mqfs_node * pn,struct ucred * ucred)1026 do_unlink(struct mqfs_node *pn, struct ucred *ucred)
1027 {
1028 	struct mqfs_node *parent;
1029 	struct mqfs_vdata *vd;
1030 	int error = 0;
1031 
1032 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
1033 
1034 	if (ucred->cr_uid != pn->mn_uid &&
1035 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN)) != 0)
1036 		error = EACCES;
1037 	else if (!pn->mn_deleted) {
1038 		parent = pn->mn_parent;
1039 		pn->mn_parent = NULL;
1040 		pn->mn_deleted = 1;
1041 		LIST_REMOVE(pn, mn_sibling);
1042 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
1043 			cache_purge(vd->mv_vnode);
1044 			vhold(vd->mv_vnode);
1045 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
1046 		}
1047 		mqnode_release(pn);
1048 		mqnode_release(parent);
1049 	} else
1050 		error = ENOENT;
1051 	return (error);
1052 }
1053 
1054 #if 0
1055 struct vop_remove_args {
1056 	struct vnode *a_dvp;
1057 	struct vnode *a_vp;
1058 	struct componentname *a_cnp;
1059 };
1060 #endif
1061 
1062 /*
1063  * vnode removal operation
1064  */
1065 static int
mqfs_remove(struct vop_remove_args * ap)1066 mqfs_remove(struct vop_remove_args *ap)
1067 {
1068 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1069 	struct mqfs_node *pn;
1070 	int error;
1071 
1072 	if (ap->a_vp->v_type == VDIR)
1073                 return (EPERM);
1074 	pn = VTON(ap->a_vp);
1075 	sx_xlock(&mqfs->mi_lock);
1076 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1077 	sx_xunlock(&mqfs->mi_lock);
1078 	return (error);
1079 }
1080 
1081 #if 0
1082 struct vop_inactive_args {
1083 	struct vnode *a_vp;
1084 	struct thread *a_td;
1085 };
1086 #endif
1087 
1088 static int
mqfs_inactive(struct vop_inactive_args * ap)1089 mqfs_inactive(struct vop_inactive_args *ap)
1090 {
1091 	struct mqfs_node *pn = VTON(ap->a_vp);
1092 
1093 	if (pn->mn_deleted)
1094 		vrecycle(ap->a_vp);
1095 	return (0);
1096 }
1097 
1098 #if 0
1099 struct vop_reclaim_args {
1100 	struct vop_generic_args a_gen;
1101 	struct vnode *a_vp;
1102 };
1103 #endif
1104 
1105 static int
mqfs_reclaim(struct vop_reclaim_args * ap)1106 mqfs_reclaim(struct vop_reclaim_args *ap)
1107 {
1108 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1109 	struct vnode *vp = ap->a_vp;
1110 	struct mqfs_node *pn;
1111 	struct mqfs_vdata *vd;
1112 
1113 	vd = vp->v_data;
1114 	pn = vd->mv_node;
1115 	sx_xlock(&mqfs->mi_lock);
1116 	vp->v_data = NULL;
1117 	LIST_REMOVE(vd, mv_link);
1118 	mqnode_release(pn);
1119 	sx_xunlock(&mqfs->mi_lock);
1120 	uma_zfree(mvdata_zone, vd);
1121 	return (0);
1122 }
1123 
1124 #if 0
1125 struct vop_open_args {
1126 	struct vop_generic_args a_gen;
1127 	struct vnode *a_vp;
1128 	int a_mode;
1129 	struct ucred *a_cred;
1130 	struct thread *a_td;
1131 	struct file *a_fp;
1132 };
1133 #endif
1134 
1135 static int
mqfs_open(struct vop_open_args * ap)1136 mqfs_open(struct vop_open_args *ap)
1137 {
1138 	return (0);
1139 }
1140 
1141 #if 0
1142 struct vop_close_args {
1143 	struct vop_generic_args a_gen;
1144 	struct vnode *a_vp;
1145 	int a_fflag;
1146 	struct ucred *a_cred;
1147 	struct thread *a_td;
1148 };
1149 #endif
1150 
1151 static int
mqfs_close(struct vop_close_args * ap)1152 mqfs_close(struct vop_close_args *ap)
1153 {
1154 	return (0);
1155 }
1156 
1157 #if 0
1158 struct vop_access_args {
1159 	struct vop_generic_args a_gen;
1160 	struct vnode *a_vp;
1161 	accmode_t a_accmode;
1162 	struct ucred *a_cred;
1163 	struct thread *a_td;
1164 };
1165 #endif
1166 
1167 /*
1168  * Verify permissions
1169  */
1170 static int
mqfs_access(struct vop_access_args * ap)1171 mqfs_access(struct vop_access_args *ap)
1172 {
1173 	struct vnode *vp = ap->a_vp;
1174 	struct vattr vattr;
1175 	int error;
1176 
1177 	error = VOP_GETATTR(vp, &vattr, ap->a_cred);
1178 	if (error)
1179 		return (error);
1180 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid, vattr.va_gid,
1181 	    ap->a_accmode, ap->a_cred);
1182 	return (error);
1183 }
1184 
1185 #if 0
1186 struct vop_getattr_args {
1187 	struct vop_generic_args a_gen;
1188 	struct vnode *a_vp;
1189 	struct vattr *a_vap;
1190 	struct ucred *a_cred;
1191 };
1192 #endif
1193 
1194 /*
1195  * Get file attributes
1196  */
1197 static int
mqfs_getattr(struct vop_getattr_args * ap)1198 mqfs_getattr(struct vop_getattr_args *ap)
1199 {
1200 	struct vnode *vp = ap->a_vp;
1201 	struct mqfs_node *pn = VTON(vp);
1202 	struct vattr *vap = ap->a_vap;
1203 	int error = 0;
1204 
1205 	vap->va_type = vp->v_type;
1206 	vap->va_mode = pn->mn_mode;
1207 	vap->va_nlink = 1;
1208 	vap->va_uid = pn->mn_uid;
1209 	vap->va_gid = pn->mn_gid;
1210 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1211 	vap->va_fileid = pn->mn_fileno;
1212 	vap->va_size = 0;
1213 	vap->va_blocksize = PAGE_SIZE;
1214 	vap->va_bytes = vap->va_size = 0;
1215 	vap->va_atime = pn->mn_atime;
1216 	vap->va_mtime = pn->mn_mtime;
1217 	vap->va_ctime = pn->mn_ctime;
1218 	vap->va_birthtime = pn->mn_birth;
1219 	vap->va_gen = 0;
1220 	vap->va_flags = 0;
1221 	vap->va_rdev = NODEV;
1222 	vap->va_bytes = 0;
1223 	vap->va_filerev = 0;
1224 	return (error);
1225 }
1226 
1227 #if 0
1228 struct vop_setattr_args {
1229 	struct vop_generic_args a_gen;
1230 	struct vnode *a_vp;
1231 	struct vattr *a_vap;
1232 	struct ucred *a_cred;
1233 };
1234 #endif
1235 /*
1236  * Set attributes
1237  */
1238 static int
mqfs_setattr(struct vop_setattr_args * ap)1239 mqfs_setattr(struct vop_setattr_args *ap)
1240 {
1241 	struct mqfs_node *pn;
1242 	struct vattr *vap;
1243 	struct vnode *vp;
1244 	struct thread *td;
1245 	int c, error;
1246 	uid_t uid;
1247 	gid_t gid;
1248 
1249 	td = curthread;
1250 	vap = ap->a_vap;
1251 	vp = ap->a_vp;
1252 	if (vap->va_type != VNON ||
1253 	    vap->va_nlink != VNOVAL ||
1254 	    vap->va_fsid != VNOVAL ||
1255 	    vap->va_fileid != VNOVAL ||
1256 	    vap->va_blocksize != VNOVAL ||
1257 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1258 	    vap->va_rdev != VNOVAL ||
1259 	    (int)vap->va_bytes != VNOVAL ||
1260 	    vap->va_gen != VNOVAL) {
1261 		return (EINVAL);
1262 	}
1263 
1264 	pn = VTON(vp);
1265 
1266 	error = c = 0;
1267 	if (vap->va_uid == (uid_t)VNOVAL)
1268 		uid = pn->mn_uid;
1269 	else
1270 		uid = vap->va_uid;
1271 	if (vap->va_gid == (gid_t)VNOVAL)
1272 		gid = pn->mn_gid;
1273 	else
1274 		gid = vap->va_gid;
1275 
1276 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1277 		/*
1278 		 * To modify the ownership of a file, must possess VADMIN
1279 		 * for that file.
1280 		 */
1281 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
1282 			return (error);
1283 
1284 		/*
1285 		 * XXXRW: Why is there a privilege check here: shouldn't the
1286 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1287 		 * below definitely right?
1288 		 */
1289 		if ((ap->a_cred->cr_uid != pn->mn_uid || uid != pn->mn_uid ||
1290 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1291 		    (error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
1292 			return (error);
1293 		pn->mn_uid = uid;
1294 		pn->mn_gid = gid;
1295 		c = 1;
1296 	}
1297 
1298 	if (vap->va_mode != (mode_t)VNOVAL) {
1299 		if (ap->a_cred->cr_uid != pn->mn_uid &&
1300 		    (error = priv_check(td, PRIV_MQ_ADMIN)))
1301 			return (error);
1302 		pn->mn_mode = vap->va_mode;
1303 		c = 1;
1304 	}
1305 
1306 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1307 		/* See the comment in ufs_vnops::ufs_setattr(). */
1308 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1309 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1310 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1311 			return (error);
1312 		if (vap->va_atime.tv_sec != VNOVAL) {
1313 			pn->mn_atime = vap->va_atime;
1314 		}
1315 		if (vap->va_mtime.tv_sec != VNOVAL) {
1316 			pn->mn_mtime = vap->va_mtime;
1317 		}
1318 		c = 1;
1319 	}
1320 	if (c) {
1321 		vfs_timestamp(&pn->mn_ctime);
1322 	}
1323 	return (0);
1324 }
1325 
1326 #if 0
1327 struct vop_read_args {
1328 	struct vop_generic_args a_gen;
1329 	struct vnode *a_vp;
1330 	struct uio *a_uio;
1331 	int a_ioflag;
1332 	struct ucred *a_cred;
1333 };
1334 #endif
1335 
1336 /*
1337  * Read from a file
1338  */
1339 static int
mqfs_read(struct vop_read_args * ap)1340 mqfs_read(struct vop_read_args *ap)
1341 {
1342 	char buf[80];
1343 	struct vnode *vp = ap->a_vp;
1344 	struct uio *uio = ap->a_uio;
1345 	struct mqueue *mq;
1346 	int len, error;
1347 
1348 	if (vp->v_type != VREG)
1349 		return (EINVAL);
1350 
1351 	mq = VTOMQ(vp);
1352 	snprintf(buf, sizeof(buf),
1353 	    "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1354 	    mq->mq_totalbytes,
1355 	    mq->mq_maxmsg,
1356 	    mq->mq_curmsgs,
1357 	    mq->mq_msgsize);
1358 	buf[sizeof(buf)-1] = '\0';
1359 	len = strlen(buf);
1360 	error = uiomove_frombuf(buf, len, uio);
1361 	return (error);
1362 }
1363 
1364 #if 0
1365 struct vop_readdir_args {
1366 	struct vop_generic_args a_gen;
1367 	struct vnode *a_vp;
1368 	struct uio *a_uio;
1369 	struct ucred *a_cred;
1370 	int *a_eofflag;
1371 	int *a_ncookies;
1372 	uint64_t **a_cookies;
1373 };
1374 #endif
1375 
1376 /*
1377  * Return directory entries.
1378  */
1379 static int
mqfs_readdir(struct vop_readdir_args * ap)1380 mqfs_readdir(struct vop_readdir_args *ap)
1381 {
1382 	struct vnode *vp;
1383 	struct mqfs_info *mi;
1384 	struct mqfs_node *pd;
1385 	struct mqfs_node *pn;
1386 	struct dirent entry;
1387 	struct uio *uio;
1388 	const void *pr_root;
1389 	int *tmp_ncookies = NULL;
1390 	off_t offset;
1391 	int error, i;
1392 
1393 	vp = ap->a_vp;
1394 	mi = VFSTOMQFS(vp->v_mount);
1395 	pd = VTON(vp);
1396 	uio = ap->a_uio;
1397 
1398 	if (vp->v_type != VDIR)
1399 		return (ENOTDIR);
1400 
1401 	if (uio->uio_offset < 0)
1402 		return (EINVAL);
1403 
1404 	if (ap->a_ncookies != NULL) {
1405 		tmp_ncookies = ap->a_ncookies;
1406 		*ap->a_ncookies = 0;
1407 		ap->a_ncookies = NULL;
1408         }
1409 
1410 	error = 0;
1411 	offset = 0;
1412 
1413 	pr_root = ap->a_cred->cr_prison->pr_root;
1414 	sx_xlock(&mi->mi_lock);
1415 
1416 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1417 		entry.d_reclen = sizeof(entry);
1418 
1419 		/*
1420 		 * Only show names within the same prison root directory
1421 		 * (or not associated with a prison, e.g. "." and "..").
1422 		 */
1423 		if (pn->mn_pr_root != NULL && pn->mn_pr_root != pr_root)
1424 			continue;
1425 		if (!pn->mn_fileno)
1426 			mqfs_fileno_alloc(mi, pn);
1427 		entry.d_fileno = pn->mn_fileno;
1428 		entry.d_off = offset + entry.d_reclen;
1429 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1430 			entry.d_name[i] = pn->mn_name[i];
1431 		entry.d_namlen = i;
1432 		switch (pn->mn_type) {
1433 		case mqfstype_root:
1434 		case mqfstype_dir:
1435 		case mqfstype_this:
1436 		case mqfstype_parent:
1437 			entry.d_type = DT_DIR;
1438 			break;
1439 		case mqfstype_file:
1440 			entry.d_type = DT_REG;
1441 			break;
1442 		case mqfstype_symlink:
1443 			entry.d_type = DT_LNK;
1444 			break;
1445 		default:
1446 			panic("%s has unexpected node type: %d", pn->mn_name,
1447 				pn->mn_type);
1448 		}
1449 		dirent_terminate(&entry);
1450 		if (entry.d_reclen > uio->uio_resid)
1451                         break;
1452 		if (offset >= uio->uio_offset) {
1453 			error = vfs_read_dirent(ap, &entry, offset);
1454                         if (error)
1455                                 break;
1456                 }
1457                 offset += entry.d_reclen;
1458 	}
1459 	sx_xunlock(&mi->mi_lock);
1460 
1461 	uio->uio_offset = offset;
1462 
1463 	if (tmp_ncookies != NULL)
1464 		ap->a_ncookies = tmp_ncookies;
1465 
1466 	return (error);
1467 }
1468 
1469 #ifdef notyet
1470 
1471 #if 0
1472 struct vop_mkdir_args {
1473 	struct vnode *a_dvp;
1474 	struvt vnode **a_vpp;
1475 	struvt componentname *a_cnp;
1476 	struct vattr *a_vap;
1477 };
1478 #endif
1479 
1480 /*
1481  * Create a directory.
1482  */
1483 static int
mqfs_mkdir(struct vop_mkdir_args * ap)1484 mqfs_mkdir(struct vop_mkdir_args *ap)
1485 {
1486 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1487 	struct componentname *cnp = ap->a_cnp;
1488 	struct mqfs_node *pd = VTON(ap->a_dvp);
1489 	struct mqfs_node *pn;
1490 	int error;
1491 
1492 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1493 		return (ENOTDIR);
1494 	sx_xlock(&mqfs->mi_lock);
1495 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1496 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1497 	if (pn != NULL)
1498 		mqnode_addref(pn);
1499 	sx_xunlock(&mqfs->mi_lock);
1500 	if (pn == NULL) {
1501 		error = ENOSPC;
1502 	} else {
1503 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1504 		mqnode_release(pn);
1505 	}
1506 	return (error);
1507 }
1508 
1509 #if 0
1510 struct vop_rmdir_args {
1511 	struct vnode *a_dvp;
1512 	struct vnode *a_vp;
1513 	struct componentname *a_cnp;
1514 };
1515 #endif
1516 
1517 /*
1518  * Remove a directory.
1519  */
1520 static int
mqfs_rmdir(struct vop_rmdir_args * ap)1521 mqfs_rmdir(struct vop_rmdir_args *ap)
1522 {
1523 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1524 	struct mqfs_node *pn = VTON(ap->a_vp);
1525 	struct mqfs_node *pt;
1526 
1527 	if (pn->mn_type != mqfstype_dir)
1528 		return (ENOTDIR);
1529 
1530 	sx_xlock(&mqfs->mi_lock);
1531 	if (pn->mn_deleted) {
1532 		sx_xunlock(&mqfs->mi_lock);
1533 		return (ENOENT);
1534 	}
1535 
1536 	pt = LIST_FIRST(&pn->mn_children);
1537 	pt = LIST_NEXT(pt, mn_sibling);
1538 	pt = LIST_NEXT(pt, mn_sibling);
1539 	if (pt != NULL) {
1540 		sx_xunlock(&mqfs->mi_lock);
1541 		return (ENOTEMPTY);
1542 	}
1543 	pt = pn->mn_parent;
1544 	pn->mn_parent = NULL;
1545 	pn->mn_deleted = 1;
1546 	LIST_REMOVE(pn, mn_sibling);
1547 	mqnode_release(pn);
1548 	mqnode_release(pt);
1549 	sx_xunlock(&mqfs->mi_lock);
1550 	cache_purge(ap->a_vp);
1551 	return (0);
1552 }
1553 
1554 #endif /* notyet */
1555 
1556 /*
1557  * See if this prison root is obsolete, and clean up associated queues if it is.
1558  */
1559 static int
mqfs_prison_remove(void * obj,void * data __unused)1560 mqfs_prison_remove(void *obj, void *data __unused)
1561 {
1562 	const struct prison *pr = obj;
1563 	struct prison *tpr;
1564 	struct mqfs_node *pn, *tpn;
1565 	struct vnode *pr_root;
1566 
1567 	pr_root = pr->pr_root;
1568 	if (pr->pr_parent->pr_root == pr_root)
1569 		return (0);
1570 	TAILQ_FOREACH(tpr, &allprison, pr_list) {
1571 		if (tpr != pr && tpr->pr_root == pr_root)
1572 			return (0);
1573 	}
1574 	/*
1575 	 * No jails are rooted in this directory anymore,
1576 	 * so no queues should be either.
1577 	 */
1578 	sx_xlock(&mqfs_data.mi_lock);
1579 	LIST_FOREACH_SAFE(pn, &mqfs_data.mi_root->mn_children,
1580 	    mn_sibling, tpn) {
1581 		if (pn->mn_pr_root == pr_root)
1582 			(void)do_unlink(pn, curthread->td_ucred);
1583 	}
1584 	sx_xunlock(&mqfs_data.mi_lock);
1585 	return (0);
1586 }
1587 
1588 /*
1589  * Allocate a message queue
1590  */
1591 static struct mqueue *
mqueue_alloc(const struct mq_attr * attr)1592 mqueue_alloc(const struct mq_attr *attr)
1593 {
1594 	struct mqueue *mq;
1595 
1596 	if (curmq >= maxmq)
1597 		return (NULL);
1598 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1599 	TAILQ_INIT(&mq->mq_msgq);
1600 	if (attr != NULL) {
1601 		mq->mq_maxmsg = attr->mq_maxmsg;
1602 		mq->mq_msgsize = attr->mq_msgsize;
1603 	} else {
1604 		mq->mq_maxmsg = default_maxmsg;
1605 		mq->mq_msgsize = default_msgsize;
1606 	}
1607 	mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
1608 	knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
1609 	knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
1610 	atomic_add_int(&curmq, 1);
1611 	return (mq);
1612 }
1613 
1614 /*
1615  * Destroy a message queue
1616  */
1617 static void
mqueue_free(struct mqueue * mq)1618 mqueue_free(struct mqueue *mq)
1619 {
1620 	struct mqueue_msg *msg;
1621 
1622 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1623 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1624 		free(msg, M_MQUEUEDATA);
1625 	}
1626 
1627 	mtx_destroy(&mq->mq_mutex);
1628 	seldrain(&mq->mq_rsel);
1629 	seldrain(&mq->mq_wsel);
1630 	knlist_destroy(&mq->mq_rsel.si_note);
1631 	knlist_destroy(&mq->mq_wsel.si_note);
1632 	uma_zfree(mqueue_zone, mq);
1633 	atomic_add_int(&curmq, -1);
1634 }
1635 
1636 /*
1637  * Load a message from user space
1638  */
1639 static struct mqueue_msg *
mqueue_loadmsg(const char * msg_ptr,size_t msg_size,int msg_prio)1640 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1641 {
1642 	struct mqueue_msg *msg;
1643 	size_t len;
1644 	int error;
1645 
1646 	len = sizeof(struct mqueue_msg) + msg_size;
1647 	msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1648 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1649 	    msg_size);
1650 	if (error) {
1651 		free(msg, M_MQUEUEDATA);
1652 		msg = NULL;
1653 	} else {
1654 		msg->msg_size = msg_size;
1655 		msg->msg_prio = msg_prio;
1656 	}
1657 	return (msg);
1658 }
1659 
1660 /*
1661  * Save a message to user space
1662  */
1663 static int
mqueue_savemsg(struct mqueue_msg * msg,char * msg_ptr,int * msg_prio)1664 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1665 {
1666 	int error;
1667 
1668 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1669 		msg->msg_size);
1670 	if (error == 0 && msg_prio != NULL)
1671 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1672 	return (error);
1673 }
1674 
1675 /*
1676  * Free a message's memory
1677  */
1678 static __inline void
mqueue_freemsg(struct mqueue_msg * msg)1679 mqueue_freemsg(struct mqueue_msg *msg)
1680 {
1681 	free(msg, M_MQUEUEDATA);
1682 }
1683 
1684 /*
1685  * Send a message. if waitok is false, thread will not be
1686  * blocked if there is no data in queue, otherwise, absolute
1687  * time will be checked.
1688  */
1689 int
mqueue_send(struct mqueue * mq,const char * msg_ptr,size_t msg_len,unsigned msg_prio,int waitok,const struct timespec * abs_timeout)1690 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1691 	size_t msg_len, unsigned msg_prio, int waitok,
1692 	const struct timespec *abs_timeout)
1693 {
1694 	struct mqueue_msg *msg;
1695 	struct timespec ts, ts2;
1696 	struct timeval tv;
1697 	int error;
1698 
1699 	if (msg_prio >= MQ_PRIO_MAX)
1700 		return (EINVAL);
1701 	if (msg_len > mq->mq_msgsize)
1702 		return (EMSGSIZE);
1703 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1704 	if (msg == NULL)
1705 		return (EFAULT);
1706 
1707 	/* O_NONBLOCK case */
1708 	if (!waitok) {
1709 		error = _mqueue_send(mq, msg, -1);
1710 		if (error)
1711 			goto bad;
1712 		return (0);
1713 	}
1714 
1715 	/* we allow a null timeout (wait forever) */
1716 	if (abs_timeout == NULL) {
1717 		error = _mqueue_send(mq, msg, 0);
1718 		if (error)
1719 			goto bad;
1720 		return (0);
1721 	}
1722 
1723 	/* send it before checking time */
1724 	error = _mqueue_send(mq, msg, -1);
1725 	if (error == 0)
1726 		return (0);
1727 
1728 	if (error != EAGAIN)
1729 		goto bad;
1730 
1731 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1732 		error = EINVAL;
1733 		goto bad;
1734 	}
1735 	for (;;) {
1736 		getnanotime(&ts);
1737 		timespecsub(abs_timeout, &ts, &ts2);
1738 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1739 			error = ETIMEDOUT;
1740 			break;
1741 		}
1742 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1743 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1744 		if (error != ETIMEDOUT)
1745 			break;
1746 	}
1747 	if (error == 0)
1748 		return (0);
1749 bad:
1750 	mqueue_freemsg(msg);
1751 	return (error);
1752 }
1753 
1754 /*
1755  * Common routine to send a message
1756  */
1757 static int
_mqueue_send(struct mqueue * mq,struct mqueue_msg * msg,int timo)1758 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1759 {
1760 	struct mqueue_msg *msg2;
1761 	int error = 0;
1762 
1763 	mtx_lock(&mq->mq_mutex);
1764 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1765 		if (timo < 0) {
1766 			mtx_unlock(&mq->mq_mutex);
1767 			return (EAGAIN);
1768 		}
1769 		mq->mq_senders++;
1770 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1771 			    PCATCH, "mqsend", timo);
1772 		mq->mq_senders--;
1773 		if (error == EAGAIN)
1774 			error = ETIMEDOUT;
1775 	}
1776 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1777 		mtx_unlock(&mq->mq_mutex);
1778 		return (error);
1779 	}
1780 	error = 0;
1781 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1782 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1783 	} else {
1784 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1785 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1786 		} else {
1787 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1788 				if (msg2->msg_prio < msg->msg_prio)
1789 					break;
1790 			}
1791 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1792 		}
1793 	}
1794 	mq->mq_curmsgs++;
1795 	mq->mq_totalbytes += msg->msg_size;
1796 	if (mq->mq_receivers)
1797 		wakeup_one(&mq->mq_receivers);
1798 	else if (mq->mq_notifier != NULL)
1799 		mqueue_send_notification(mq);
1800 	if (mq->mq_flags & MQ_RSEL) {
1801 		mq->mq_flags &= ~MQ_RSEL;
1802 		selwakeup(&mq->mq_rsel);
1803 	}
1804 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1805 	mtx_unlock(&mq->mq_mutex);
1806 	return (0);
1807 }
1808 
1809 /*
1810  * Send realtime a signal to process which registered itself
1811  * successfully by mq_notify.
1812  */
1813 static void
mqueue_send_notification(struct mqueue * mq)1814 mqueue_send_notification(struct mqueue *mq)
1815 {
1816 	struct mqueue_notifier *nt;
1817 	struct thread *td;
1818 	struct proc *p;
1819 	int error;
1820 
1821 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1822 	nt = mq->mq_notifier;
1823 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1824 		p = nt->nt_proc;
1825 		error = sigev_findtd(p, &nt->nt_sigev, &td);
1826 		if (error) {
1827 			mq->mq_notifier = NULL;
1828 			return;
1829 		}
1830 		if (!KSI_ONQ(&nt->nt_ksi)) {
1831 			ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
1832 			tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
1833 		}
1834 		PROC_UNLOCK(p);
1835 	}
1836 	mq->mq_notifier = NULL;
1837 }
1838 
1839 /*
1840  * Get a message. if waitok is false, thread will not be
1841  * blocked if there is no data in queue, otherwise, absolute
1842  * time will be checked.
1843  */
1844 int
mqueue_receive(struct mqueue * mq,char * msg_ptr,size_t msg_len,unsigned * msg_prio,int waitok,const struct timespec * abs_timeout)1845 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1846 	size_t msg_len, unsigned *msg_prio, int waitok,
1847 	const struct timespec *abs_timeout)
1848 {
1849 	struct mqueue_msg *msg;
1850 	struct timespec ts, ts2;
1851 	struct timeval tv;
1852 	int error;
1853 
1854 	if (msg_len < mq->mq_msgsize)
1855 		return (EMSGSIZE);
1856 
1857 	/* O_NONBLOCK case */
1858 	if (!waitok) {
1859 		error = _mqueue_recv(mq, &msg, -1);
1860 		if (error)
1861 			return (error);
1862 		goto received;
1863 	}
1864 
1865 	/* we allow a null timeout (wait forever). */
1866 	if (abs_timeout == NULL) {
1867 		error = _mqueue_recv(mq, &msg, 0);
1868 		if (error)
1869 			return (error);
1870 		goto received;
1871 	}
1872 
1873 	/* try to get a message before checking time */
1874 	error = _mqueue_recv(mq, &msg, -1);
1875 	if (error == 0)
1876 		goto received;
1877 
1878 	if (error != EAGAIN)
1879 		return (error);
1880 
1881 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1882 		error = EINVAL;
1883 		return (error);
1884 	}
1885 
1886 	for (;;) {
1887 		getnanotime(&ts);
1888 		timespecsub(abs_timeout, &ts, &ts2);
1889 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1890 			error = ETIMEDOUT;
1891 			return (error);
1892 		}
1893 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1894 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1895 		if (error == 0)
1896 			break;
1897 		if (error != ETIMEDOUT)
1898 			return (error);
1899 	}
1900 
1901 received:
1902 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1903 	if (error == 0) {
1904 		curthread->td_retval[0] = msg->msg_size;
1905 		curthread->td_retval[1] = 0;
1906 	}
1907 	mqueue_freemsg(msg);
1908 	return (error);
1909 }
1910 
1911 /*
1912  * Common routine to receive a message
1913  */
1914 static int
_mqueue_recv(struct mqueue * mq,struct mqueue_msg ** msg,int timo)1915 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1916 {
1917 	int error = 0;
1918 
1919 	mtx_lock(&mq->mq_mutex);
1920 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1921 		if (timo < 0) {
1922 			mtx_unlock(&mq->mq_mutex);
1923 			return (EAGAIN);
1924 		}
1925 		mq->mq_receivers++;
1926 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1927 			    PCATCH, "mqrecv", timo);
1928 		mq->mq_receivers--;
1929 		if (error == EAGAIN)
1930 			error = ETIMEDOUT;
1931 	}
1932 	if (*msg != NULL) {
1933 		error = 0;
1934 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1935 		mq->mq_curmsgs--;
1936 		mq->mq_totalbytes -= (*msg)->msg_size;
1937 		if (mq->mq_senders)
1938 			wakeup_one(&mq->mq_senders);
1939 		if (mq->mq_flags & MQ_WSEL) {
1940 			mq->mq_flags &= ~MQ_WSEL;
1941 			selwakeup(&mq->mq_wsel);
1942 		}
1943 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1944 	}
1945 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1946 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1947 		mqueue_send_notification(mq);
1948 	}
1949 	mtx_unlock(&mq->mq_mutex);
1950 	return (error);
1951 }
1952 
1953 static __inline struct mqueue_notifier *
notifier_alloc(void)1954 notifier_alloc(void)
1955 {
1956 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1957 }
1958 
1959 static __inline void
notifier_free(struct mqueue_notifier * p)1960 notifier_free(struct mqueue_notifier *p)
1961 {
1962 	uma_zfree(mqnoti_zone, p);
1963 }
1964 
1965 static struct mqueue_notifier *
notifier_search(struct proc * p,int fd)1966 notifier_search(struct proc *p, int fd)
1967 {
1968 	struct mqueue_notifier *nt;
1969 
1970 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1971 		if (nt->nt_ksi.ksi_mqd == fd)
1972 			break;
1973 	}
1974 	return (nt);
1975 }
1976 
1977 static __inline void
notifier_insert(struct proc * p,struct mqueue_notifier * nt)1978 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1979 {
1980 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1981 }
1982 
1983 static __inline void
notifier_delete(struct proc * p,struct mqueue_notifier * nt)1984 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1985 {
1986 	LIST_REMOVE(nt, nt_link);
1987 	notifier_free(nt);
1988 }
1989 
1990 static void
notifier_remove(struct proc * p,struct mqueue * mq,int fd)1991 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1992 {
1993 	struct mqueue_notifier *nt;
1994 
1995 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1996 	PROC_LOCK(p);
1997 	nt = notifier_search(p, fd);
1998 	if (nt != NULL) {
1999 		if (mq->mq_notifier == nt)
2000 			mq->mq_notifier = NULL;
2001 		sigqueue_take(&nt->nt_ksi);
2002 		notifier_delete(p, nt);
2003 	}
2004 	PROC_UNLOCK(p);
2005 }
2006 
2007 int
kern_kmq_open(struct thread * td,const char * upath,int flags,mode_t mode,const struct mq_attr * attr)2008 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
2009     const struct mq_attr *attr)
2010 {
2011 	char *path, pathbuf[MQFS_NAMELEN + 1];
2012 	struct mqfs_node *pn;
2013 	struct pwddesc *pdp;
2014 	struct file *fp;
2015 	struct mqueue *mq;
2016 	int fd, error, len, cmode;
2017 
2018 	AUDIT_ARG_FFLAGS(flags);
2019 	AUDIT_ARG_MODE(mode);
2020 
2021 	pdp = td->td_proc->p_pd;
2022 	cmode = ((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT;
2023 	mq = NULL;
2024 	if ((flags & O_CREAT) != 0 && attr != NULL) {
2025 		if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
2026 			return (EINVAL);
2027 		if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
2028 			return (EINVAL);
2029 	}
2030 
2031 	path = pathbuf;
2032 	error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
2033         if (error)
2034 		return (error);
2035 
2036 	/*
2037 	 * The first character of name may be a slash (/) character
2038 	 * and the remaining characters of name cannot include any slash
2039 	 * characters.
2040 	 */
2041 	len = strlen(path);
2042 	if (len < 2 || strchr(path + 1, '/') != NULL)
2043 		return (EINVAL);
2044 	if (path[0] == '/') {
2045 		path++;
2046 		len--;
2047 	}
2048 	/*
2049 	 * "." and ".." are magic directories, populated on the fly, and cannot
2050 	 * be opened as queues.
2051 	 */
2052 	if (strcmp(path, ".") == 0 || strcmp(path, "..") == 0)
2053 		return (EINVAL);
2054 	AUDIT_ARG_UPATH1_CANON(pathbuf);
2055 
2056 	error = falloc(td, &fp, &fd, O_CLOEXEC);
2057 	if (error)
2058 		return (error);
2059 
2060 	sx_xlock(&mqfs_data.mi_lock);
2061 	pn = mqfs_search(mqfs_data.mi_root, path, len, td->td_ucred);
2062 	if (pn == NULL) {
2063 		if (!(flags & O_CREAT)) {
2064 			error = ENOENT;
2065 		} else {
2066 			mq = mqueue_alloc(attr);
2067 			if (mq == NULL) {
2068 				error = ENFILE;
2069 			} else {
2070 				pn = mqfs_create_file(mqfs_data.mi_root,
2071 				         path, len, td->td_ucred,
2072 					 cmode);
2073 				if (pn == NULL) {
2074 					error = ENOSPC;
2075 					mqueue_free(mq);
2076 				}
2077 			}
2078 		}
2079 
2080 		if (error == 0) {
2081 			pn->mn_data = mq;
2082 		}
2083 	} else {
2084 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
2085 			error = EEXIST;
2086 		} else {
2087 			accmode_t accmode = 0;
2088 
2089 			if (flags & FREAD)
2090 				accmode |= VREAD;
2091 			if (flags & FWRITE)
2092 				accmode |= VWRITE;
2093 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
2094 			    pn->mn_gid, accmode, td->td_ucred);
2095 		}
2096 	}
2097 
2098 	if (error) {
2099 		sx_xunlock(&mqfs_data.mi_lock);
2100 		fdclose(td, fp, fd);
2101 		fdrop(fp, td);
2102 		return (error);
2103 	}
2104 
2105 	mqnode_addref(pn);
2106 	sx_xunlock(&mqfs_data.mi_lock);
2107 
2108 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2109 	    &mqueueops);
2110 
2111 	td->td_retval[0] = fd;
2112 	fdrop(fp, td);
2113 	return (0);
2114 }
2115 
2116 /*
2117  * Syscall to open a message queue.
2118  */
2119 int
sys_kmq_open(struct thread * td,struct kmq_open_args * uap)2120 sys_kmq_open(struct thread *td, struct kmq_open_args *uap)
2121 {
2122 	struct mq_attr attr;
2123 	int flags, error;
2124 
2125 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2126 		return (EINVAL);
2127 	flags = FFLAGS(uap->flags);
2128 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2129 		error = copyin(uap->attr, &attr, sizeof(attr));
2130 		if (error)
2131 			return (error);
2132 	}
2133 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2134 	    uap->attr != NULL ? &attr : NULL));
2135 }
2136 
2137 /*
2138  * Syscall to unlink a message queue.
2139  */
2140 int
sys_kmq_unlink(struct thread * td,struct kmq_unlink_args * uap)2141 sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2142 {
2143 	char *path, pathbuf[MQFS_NAMELEN + 1];
2144 	struct mqfs_node *pn;
2145 	int error, len;
2146 
2147 	path = pathbuf;
2148 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2149         if (error)
2150 		return (error);
2151 
2152 	len = strlen(path);
2153 	if (len < 2 || strchr(path + 1, '/') != NULL)
2154 		return (EINVAL);
2155 	if (path[0] == '/') {
2156 		path++;
2157 		len--;
2158 	}
2159 	if (strcmp(path, ".") == 0 || strcmp(path, "..") == 0)
2160 		return (EINVAL);
2161 	AUDIT_ARG_UPATH1_CANON(pathbuf);
2162 
2163 	sx_xlock(&mqfs_data.mi_lock);
2164 	pn = mqfs_search(mqfs_data.mi_root, path, len, td->td_ucred);
2165 	if (pn != NULL)
2166 		error = do_unlink(pn, td->td_ucred);
2167 	else
2168 		error = ENOENT;
2169 	sx_xunlock(&mqfs_data.mi_lock);
2170 	return (error);
2171 }
2172 
2173 typedef int (*_fgetf)(struct thread *, int, const cap_rights_t *,
2174     struct file **);
2175 
2176 /*
2177  * Get message queue by giving file slot
2178  */
2179 static int
_getmq(struct thread * td,int fd,const cap_rights_t * rightsp,_fgetf func,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2180 _getmq(struct thread *td, int fd, const cap_rights_t *rightsp, _fgetf func,
2181        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2182 {
2183 	struct mqfs_node *pn;
2184 	int error;
2185 
2186 	error = func(td, fd, rightsp, fpp);
2187 	if (error)
2188 		return (error);
2189 	if (&mqueueops != (*fpp)->f_ops) {
2190 		fdrop(*fpp, td);
2191 		return (EBADF);
2192 	}
2193 	pn = (*fpp)->f_data;
2194 	if (ppn)
2195 		*ppn = pn;
2196 	if (pmq)
2197 		*pmq = pn->mn_data;
2198 	return (0);
2199 }
2200 
2201 static __inline int
getmq(struct thread * td,int fd,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2202 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2203 	struct mqueue **pmq)
2204 {
2205 
2206 	return _getmq(td, fd, &cap_event_rights, fget,
2207 	    fpp, ppn, pmq);
2208 }
2209 
2210 static __inline int
getmq_read(struct thread * td,int fd,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2211 getmq_read(struct thread *td, int fd, struct file **fpp,
2212 	 struct mqfs_node **ppn, struct mqueue **pmq)
2213 {
2214 
2215 	return _getmq(td, fd, &cap_read_rights, fget_read,
2216 	    fpp, ppn, pmq);
2217 }
2218 
2219 static __inline int
getmq_write(struct thread * td,int fd,struct file ** fpp,struct mqfs_node ** ppn,struct mqueue ** pmq)2220 getmq_write(struct thread *td, int fd, struct file **fpp,
2221 	struct mqfs_node **ppn, struct mqueue **pmq)
2222 {
2223 
2224 	return _getmq(td, fd, &cap_write_rights, fget_write,
2225 	    fpp, ppn, pmq);
2226 }
2227 
2228 int
kern_kmq_setattr(struct thread * td,int mqd,const struct mq_attr * attr,struct mq_attr * oattr)2229 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2230     struct mq_attr *oattr)
2231 {
2232 	struct mqueue *mq;
2233 	struct file *fp;
2234 	u_int oflag, flag;
2235 	int error;
2236 
2237 	AUDIT_ARG_FD(mqd);
2238 	if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2239 		return (EINVAL);
2240 	error = getmq(td, mqd, &fp, NULL, &mq);
2241 	if (error)
2242 		return (error);
2243 	oattr->mq_maxmsg  = mq->mq_maxmsg;
2244 	oattr->mq_msgsize = mq->mq_msgsize;
2245 	oattr->mq_curmsgs = mq->mq_curmsgs;
2246 	if (attr != NULL) {
2247 		do {
2248 			oflag = flag = fp->f_flag;
2249 			flag &= ~O_NONBLOCK;
2250 			flag |= (attr->mq_flags & O_NONBLOCK);
2251 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2252 	} else
2253 		oflag = fp->f_flag;
2254 	oattr->mq_flags = (O_NONBLOCK & oflag);
2255 	fdrop(fp, td);
2256 	return (error);
2257 }
2258 
2259 int
sys_kmq_setattr(struct thread * td,struct kmq_setattr_args * uap)2260 sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2261 {
2262 	struct mq_attr attr, oattr;
2263 	int error;
2264 
2265 	if (uap->attr != NULL) {
2266 		error = copyin(uap->attr, &attr, sizeof(attr));
2267 		if (error != 0)
2268 			return (error);
2269 	}
2270 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2271 	    &oattr);
2272 	if (error == 0 && uap->oattr != NULL) {
2273 		bzero(oattr.__reserved, sizeof(oattr.__reserved));
2274 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2275 	}
2276 	return (error);
2277 }
2278 
2279 int
kern_kmq_timedreceive(struct thread * td,int mqd,char * msg_ptr,size_t msg_len,unsigned int * msg_prio,const struct timespec * abs_timeout)2280 kern_kmq_timedreceive(struct thread *td, int mqd, char *msg_ptr,
2281 	size_t msg_len, unsigned int *msg_prio, const struct timespec *abs_timeout)
2282 {
2283 	struct mqueue *mq;
2284 	struct file *fp;
2285 	int error, waitok;
2286 
2287 	AUDIT_ARG_FD(mqd);
2288 	error = getmq_read(td, mqd, &fp, NULL, &mq);
2289 	if (error != 0)
2290 		return (error);
2291 	waitok = (fp->f_flag & O_NONBLOCK) == 0;
2292 	error = mqueue_receive(mq, msg_ptr, msg_len, msg_prio, waitok,
2293 	    abs_timeout);
2294 	fdrop(fp, td);
2295 	return (error);
2296 }
2297 
2298 int
sys_kmq_timedreceive(struct thread * td,struct kmq_timedreceive_args * uap)2299 sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2300 {
2301 	struct timespec *abs_timeout, ets;
2302 	int error;
2303 
2304 	if (uap->abs_timeout != NULL) {
2305 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2306 		if (error != 0)
2307 			return (error);
2308 		abs_timeout = &ets;
2309 	} else
2310 		abs_timeout = NULL;
2311 
2312 	return (kern_kmq_timedreceive(td, uap->mqd, uap->msg_ptr, uap->msg_len,
2313 		uap->msg_prio, abs_timeout));
2314 }
2315 
2316 int
kern_kmq_timedsend(struct thread * td,int mqd,const char * msg_ptr,size_t msg_len,unsigned int msg_prio,const struct timespec * abs_timeout)2317 kern_kmq_timedsend(struct thread *td, int mqd, const char *msg_ptr,
2318 	size_t msg_len, unsigned int msg_prio, const struct timespec *abs_timeout)
2319 {
2320 	struct mqueue *mq;
2321 	struct file *fp;
2322 	int error, waitok;
2323 
2324 	AUDIT_ARG_FD(mqd);
2325 	error = getmq_write(td, mqd, &fp, NULL, &mq);
2326 	if (error != 0)
2327 		return (error);
2328 	waitok = (fp->f_flag & O_NONBLOCK) == 0;
2329 	error = mqueue_send(mq, msg_ptr, msg_len, msg_prio, waitok,
2330 		abs_timeout);
2331 	fdrop(fp, td);
2332 	return (error);
2333 }
2334 
2335 int
sys_kmq_timedsend(struct thread * td,struct kmq_timedsend_args * uap)2336 sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2337 {
2338 	struct timespec *abs_timeout, ets;
2339 	int error;
2340 
2341 	if (uap->abs_timeout != NULL) {
2342 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2343 		if (error != 0)
2344 			return (error);
2345 		abs_timeout = &ets;
2346 	} else
2347 		abs_timeout = NULL;
2348 
2349 	return (kern_kmq_timedsend(td, uap->mqd, uap->msg_ptr, uap->msg_len,
2350 		uap->msg_prio, abs_timeout));
2351 }
2352 
2353 int
kern_kmq_notify(struct thread * td,int mqd,struct sigevent * sigev)2354 kern_kmq_notify(struct thread *td, int mqd, struct sigevent *sigev)
2355 {
2356 	struct filedesc *fdp;
2357 	struct proc *p;
2358 	struct mqueue *mq;
2359 	struct file *fp, *fp2;
2360 	struct mqueue_notifier *nt, *newnt = NULL;
2361 	int error;
2362 
2363 	AUDIT_ARG_FD(mqd);
2364 	if (sigev != NULL) {
2365 		if (sigev->sigev_notify != SIGEV_SIGNAL &&
2366 		    sigev->sigev_notify != SIGEV_THREAD_ID &&
2367 		    sigev->sigev_notify != SIGEV_NONE)
2368 			return (EINVAL);
2369 		if ((sigev->sigev_notify == SIGEV_SIGNAL ||
2370 		    sigev->sigev_notify == SIGEV_THREAD_ID) &&
2371 		    !_SIG_VALID(sigev->sigev_signo))
2372 			return (EINVAL);
2373 	}
2374 	p = td->td_proc;
2375 	fdp = td->td_proc->p_fd;
2376 	error = getmq(td, mqd, &fp, NULL, &mq);
2377 	if (error)
2378 		return (error);
2379 again:
2380 	FILEDESC_SLOCK(fdp);
2381 	fp2 = fget_noref(fdp, mqd);
2382 	if (fp2 == NULL) {
2383 		FILEDESC_SUNLOCK(fdp);
2384 		error = EBADF;
2385 		goto out;
2386 	}
2387 #ifdef CAPABILITIES
2388 	error = cap_check(cap_rights(fdp, mqd), &cap_event_rights);
2389 	if (error) {
2390 		FILEDESC_SUNLOCK(fdp);
2391 		goto out;
2392 	}
2393 #endif
2394 	if (fp2 != fp) {
2395 		FILEDESC_SUNLOCK(fdp);
2396 		error = EBADF;
2397 		goto out;
2398 	}
2399 	mtx_lock(&mq->mq_mutex);
2400 	FILEDESC_SUNLOCK(fdp);
2401 	if (sigev != NULL) {
2402 		if (mq->mq_notifier != NULL) {
2403 			error = EBUSY;
2404 		} else {
2405 			PROC_LOCK(p);
2406 			nt = notifier_search(p, mqd);
2407 			if (nt == NULL) {
2408 				if (newnt == NULL) {
2409 					PROC_UNLOCK(p);
2410 					mtx_unlock(&mq->mq_mutex);
2411 					newnt = notifier_alloc();
2412 					goto again;
2413 				}
2414 			}
2415 
2416 			if (nt != NULL) {
2417 				sigqueue_take(&nt->nt_ksi);
2418 				if (newnt != NULL) {
2419 					notifier_free(newnt);
2420 					newnt = NULL;
2421 				}
2422 			} else {
2423 				nt = newnt;
2424 				newnt = NULL;
2425 				ksiginfo_init(&nt->nt_ksi);
2426 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2427 				nt->nt_ksi.ksi_code = SI_MESGQ;
2428 				nt->nt_proc = p;
2429 				nt->nt_ksi.ksi_mqd = mqd;
2430 				notifier_insert(p, nt);
2431 			}
2432 			nt->nt_sigev = *sigev;
2433 			mq->mq_notifier = nt;
2434 			PROC_UNLOCK(p);
2435 			/*
2436 			 * if there is no receivers and message queue
2437 			 * is not empty, we should send notification
2438 			 * as soon as possible.
2439 			 */
2440 			if (mq->mq_receivers == 0 &&
2441 			    !TAILQ_EMPTY(&mq->mq_msgq))
2442 				mqueue_send_notification(mq);
2443 		}
2444 	} else {
2445 		notifier_remove(p, mq, mqd);
2446 	}
2447 	mtx_unlock(&mq->mq_mutex);
2448 
2449 out:
2450 	fdrop(fp, td);
2451 	if (newnt != NULL)
2452 		notifier_free(newnt);
2453 	return (error);
2454 }
2455 
2456 int
sys_kmq_notify(struct thread * td,struct kmq_notify_args * uap)2457 sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2458 {
2459 	struct sigevent ev, *evp;
2460 	int error;
2461 
2462 	if (uap->sigev == NULL) {
2463 		evp = NULL;
2464 	} else {
2465 		error = copyin(uap->sigev, &ev, sizeof(ev));
2466 		if (error != 0)
2467 			return (error);
2468 		evp = &ev;
2469 	}
2470 	return (kern_kmq_notify(td, uap->mqd, evp));
2471 }
2472 
2473 static void
mqueue_fdclose(struct thread * td,int fd,struct file * fp)2474 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2475 {
2476 	struct mqueue *mq;
2477 #ifdef INVARIANTS
2478 	struct filedesc *fdp;
2479 
2480 	fdp = td->td_proc->p_fd;
2481 	FILEDESC_LOCK_ASSERT(fdp);
2482 #endif
2483 
2484 	if (fp->f_ops == &mqueueops) {
2485 		mq = FPTOMQ(fp);
2486 		mtx_lock(&mq->mq_mutex);
2487 		notifier_remove(td->td_proc, mq, fd);
2488 
2489 		/* have to wakeup thread in same process */
2490 		if (mq->mq_flags & MQ_RSEL) {
2491 			mq->mq_flags &= ~MQ_RSEL;
2492 			selwakeup(&mq->mq_rsel);
2493 		}
2494 		if (mq->mq_flags & MQ_WSEL) {
2495 			mq->mq_flags &= ~MQ_WSEL;
2496 			selwakeup(&mq->mq_wsel);
2497 		}
2498 		mtx_unlock(&mq->mq_mutex);
2499 	}
2500 }
2501 
2502 static void
mq_proc_exit(void * arg __unused,struct proc * p)2503 mq_proc_exit(void *arg __unused, struct proc *p)
2504 {
2505 	struct filedesc *fdp;
2506 	struct file *fp;
2507 	struct mqueue *mq;
2508 	int i;
2509 
2510 	fdp = p->p_fd;
2511 	FILEDESC_SLOCK(fdp);
2512 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2513 		fp = fget_noref(fdp, i);
2514 		if (fp != NULL && fp->f_ops == &mqueueops) {
2515 			mq = FPTOMQ(fp);
2516 			mtx_lock(&mq->mq_mutex);
2517 			notifier_remove(p, FPTOMQ(fp), i);
2518 			mtx_unlock(&mq->mq_mutex);
2519 		}
2520 	}
2521 	FILEDESC_SUNLOCK(fdp);
2522 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2523 }
2524 
2525 static int
mqf_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)2526 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2527 	struct thread *td)
2528 {
2529 	struct mqueue *mq = FPTOMQ(fp);
2530 	int revents = 0;
2531 
2532 	mtx_lock(&mq->mq_mutex);
2533 	if (events & (POLLIN | POLLRDNORM)) {
2534 		if (mq->mq_curmsgs) {
2535 			revents |= events & (POLLIN | POLLRDNORM);
2536 		} else {
2537 			mq->mq_flags |= MQ_RSEL;
2538 			selrecord(td, &mq->mq_rsel);
2539  		}
2540 	}
2541 	if (events & POLLOUT) {
2542 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2543 			revents |= POLLOUT;
2544 		else {
2545 			mq->mq_flags |= MQ_WSEL;
2546 			selrecord(td, &mq->mq_wsel);
2547 		}
2548 	}
2549 	mtx_unlock(&mq->mq_mutex);
2550 	return (revents);
2551 }
2552 
2553 static int
mqf_close(struct file * fp,struct thread * td)2554 mqf_close(struct file *fp, struct thread *td)
2555 {
2556 	struct mqfs_node *pn;
2557 
2558 	fp->f_ops = &badfileops;
2559 	pn = fp->f_data;
2560 	fp->f_data = NULL;
2561 	sx_xlock(&mqfs_data.mi_lock);
2562 	mqnode_release(pn);
2563 	sx_xunlock(&mqfs_data.mi_lock);
2564 	return (0);
2565 }
2566 
2567 static int
mqf_stat(struct file * fp,struct stat * st,struct ucred * active_cred)2568 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2569 {
2570 	struct mqfs_node *pn = fp->f_data;
2571 
2572 	bzero(st, sizeof *st);
2573 	sx_xlock(&mqfs_data.mi_lock);
2574 	st->st_atim = pn->mn_atime;
2575 	st->st_mtim = pn->mn_mtime;
2576 	st->st_ctim = pn->mn_ctime;
2577 	st->st_birthtim = pn->mn_birth;
2578 	st->st_uid = pn->mn_uid;
2579 	st->st_gid = pn->mn_gid;
2580 	st->st_mode = S_IFIFO | pn->mn_mode;
2581 	sx_xunlock(&mqfs_data.mi_lock);
2582 	return (0);
2583 }
2584 
2585 static int
mqf_chmod(struct file * fp,mode_t mode,struct ucred * active_cred,struct thread * td)2586 mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2587     struct thread *td)
2588 {
2589 	struct mqfs_node *pn;
2590 	int error;
2591 
2592 	error = 0;
2593 	pn = fp->f_data;
2594 	sx_xlock(&mqfs_data.mi_lock);
2595 	error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN,
2596 	    active_cred);
2597 	if (error != 0)
2598 		goto out;
2599 	pn->mn_mode = mode & ACCESSPERMS;
2600 out:
2601 	sx_xunlock(&mqfs_data.mi_lock);
2602 	return (error);
2603 }
2604 
2605 static int
mqf_chown(struct file * fp,uid_t uid,gid_t gid,struct ucred * active_cred,struct thread * td)2606 mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2607     struct thread *td)
2608 {
2609 	struct mqfs_node *pn;
2610 	int error;
2611 
2612 	error = 0;
2613 	pn = fp->f_data;
2614 	sx_xlock(&mqfs_data.mi_lock);
2615 	if (uid == (uid_t)-1)
2616 		uid = pn->mn_uid;
2617 	if (gid == (gid_t)-1)
2618 		gid = pn->mn_gid;
2619 	if (((uid != pn->mn_uid && uid != active_cred->cr_uid) ||
2620 	    (gid != pn->mn_gid && !groupmember(gid, active_cred))) &&
2621 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN)))
2622 		goto out;
2623 	pn->mn_uid = uid;
2624 	pn->mn_gid = gid;
2625 out:
2626 	sx_xunlock(&mqfs_data.mi_lock);
2627 	return (error);
2628 }
2629 
2630 static int
mqf_kqfilter(struct file * fp,struct knote * kn)2631 mqf_kqfilter(struct file *fp, struct knote *kn)
2632 {
2633 	struct mqueue *mq = FPTOMQ(fp);
2634 	int error = 0;
2635 
2636 	if (kn->kn_filter == EVFILT_READ) {
2637 		kn->kn_fop = &mq_rfiltops;
2638 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2639 	} else if (kn->kn_filter == EVFILT_WRITE) {
2640 		kn->kn_fop = &mq_wfiltops;
2641 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2642 	} else
2643 		error = EINVAL;
2644 	return (error);
2645 }
2646 
2647 static void
filt_mqdetach(struct knote * kn)2648 filt_mqdetach(struct knote *kn)
2649 {
2650 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2651 
2652 	if (kn->kn_filter == EVFILT_READ)
2653 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2654 	else if (kn->kn_filter == EVFILT_WRITE)
2655 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2656 	else
2657 		panic("filt_mqdetach");
2658 }
2659 
2660 static int
filt_mqread(struct knote * kn,long hint)2661 filt_mqread(struct knote *kn, long hint)
2662 {
2663 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2664 
2665 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2666 	return (mq->mq_curmsgs != 0);
2667 }
2668 
2669 static int
filt_mqwrite(struct knote * kn,long hint)2670 filt_mqwrite(struct knote *kn, long hint)
2671 {
2672 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2673 
2674 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2675 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2676 }
2677 
2678 static int
mqf_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)2679 mqf_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2680 {
2681 
2682 	kif->kf_type = KF_TYPE_MQUEUE;
2683 	return (0);
2684 }
2685 
2686 static const struct fileops mqueueops = {
2687 	.fo_read		= invfo_rdwr,
2688 	.fo_write		= invfo_rdwr,
2689 	.fo_truncate		= invfo_truncate,
2690 	.fo_ioctl		= invfo_ioctl,
2691 	.fo_poll		= mqf_poll,
2692 	.fo_kqfilter		= mqf_kqfilter,
2693 	.fo_stat		= mqf_stat,
2694 	.fo_close		= mqf_close,
2695 	.fo_chmod		= mqf_chmod,
2696 	.fo_chown		= mqf_chown,
2697 	.fo_sendfile		= invfo_sendfile,
2698 	.fo_fill_kinfo		= mqf_fill_kinfo,
2699 	.fo_cmp			= file_kcmp_generic,
2700 	.fo_flags		= DFLAG_PASSABLE,
2701 };
2702 
2703 static struct vop_vector mqfs_vnodeops = {
2704 	.vop_default 		= &default_vnodeops,
2705 	.vop_access		= mqfs_access,
2706 	.vop_cachedlookup	= mqfs_lookup,
2707 	.vop_lookup		= vfs_cache_lookup,
2708 	.vop_reclaim		= mqfs_reclaim,
2709 	.vop_create		= mqfs_create,
2710 	.vop_remove		= mqfs_remove,
2711 	.vop_inactive		= mqfs_inactive,
2712 	.vop_open		= mqfs_open,
2713 	.vop_close		= mqfs_close,
2714 	.vop_getattr		= mqfs_getattr,
2715 	.vop_setattr		= mqfs_setattr,
2716 	.vop_read		= mqfs_read,
2717 	.vop_write		= VOP_EOPNOTSUPP,
2718 	.vop_readdir		= mqfs_readdir,
2719 	.vop_mkdir		= VOP_EOPNOTSUPP,
2720 	.vop_rmdir		= VOP_EOPNOTSUPP
2721 };
2722 VFS_VOP_VECTOR_REGISTER(mqfs_vnodeops);
2723 
2724 static struct vfsops mqfs_vfsops = {
2725 	.vfs_init 		= mqfs_init,
2726 	.vfs_uninit		= mqfs_uninit,
2727 	.vfs_mount		= mqfs_mount,
2728 	.vfs_unmount		= mqfs_unmount,
2729 	.vfs_root		= mqfs_root,
2730 	.vfs_statfs		= mqfs_statfs,
2731 };
2732 
2733 static struct vfsconf mqueuefs_vfsconf = {
2734 	.vfc_version = VFS_VERSION,
2735 	.vfc_name = "mqueuefs",
2736 	.vfc_vfsops = &mqfs_vfsops,
2737 	.vfc_typenum = -1,
2738 	.vfc_flags = VFCF_SYNTHETIC
2739 };
2740 
2741 static struct syscall_helper_data mq_syscalls[] = {
2742 	SYSCALL_INIT_HELPER(kmq_open),
2743 	SYSCALL_INIT_HELPER_F(kmq_setattr, SYF_CAPENABLED),
2744 	SYSCALL_INIT_HELPER_F(kmq_timedsend, SYF_CAPENABLED),
2745 	SYSCALL_INIT_HELPER_F(kmq_timedreceive, SYF_CAPENABLED),
2746 	SYSCALL_INIT_HELPER_F(kmq_notify, SYF_CAPENABLED),
2747 	SYSCALL_INIT_HELPER(kmq_unlink),
2748 	SYSCALL_INIT_LAST
2749 };
2750 
2751 #ifdef COMPAT_FREEBSD32
2752 #include <compat/freebsd32/freebsd32.h>
2753 #include <compat/freebsd32/freebsd32_proto.h>
2754 #include <compat/freebsd32/freebsd32_signal.h>
2755 #include <compat/freebsd32/freebsd32_syscall.h>
2756 #include <compat/freebsd32/freebsd32_util.h>
2757 
2758 static void
mq_attr_from32(const struct mq_attr32 * from,struct mq_attr * to)2759 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2760 {
2761 
2762 	to->mq_flags = from->mq_flags;
2763 	to->mq_maxmsg = from->mq_maxmsg;
2764 	to->mq_msgsize = from->mq_msgsize;
2765 	to->mq_curmsgs = from->mq_curmsgs;
2766 }
2767 
2768 static void
mq_attr_to32(const struct mq_attr * from,struct mq_attr32 * to)2769 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2770 {
2771 
2772 	to->mq_flags = from->mq_flags;
2773 	to->mq_maxmsg = from->mq_maxmsg;
2774 	to->mq_msgsize = from->mq_msgsize;
2775 	to->mq_curmsgs = from->mq_curmsgs;
2776 }
2777 
2778 int
freebsd32_kmq_open(struct thread * td,struct freebsd32_kmq_open_args * uap)2779 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2780 {
2781 	struct mq_attr attr;
2782 	struct mq_attr32 attr32;
2783 	int flags, error;
2784 
2785 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2786 		return (EINVAL);
2787 	flags = FFLAGS(uap->flags);
2788 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2789 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2790 		if (error)
2791 			return (error);
2792 		mq_attr_from32(&attr32, &attr);
2793 	}
2794 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2795 	    uap->attr != NULL ? &attr : NULL));
2796 }
2797 
2798 int
freebsd32_kmq_setattr(struct thread * td,struct freebsd32_kmq_setattr_args * uap)2799 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2800 {
2801 	struct mq_attr attr, oattr;
2802 	struct mq_attr32 attr32, oattr32;
2803 	int error;
2804 
2805 	if (uap->attr != NULL) {
2806 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2807 		if (error != 0)
2808 			return (error);
2809 		mq_attr_from32(&attr32, &attr);
2810 	}
2811 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2812 	    &oattr);
2813 	if (error == 0 && uap->oattr != NULL) {
2814 		mq_attr_to32(&oattr, &oattr32);
2815 		bzero(oattr32.__reserved, sizeof(oattr32.__reserved));
2816 		error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2817 	}
2818 	return (error);
2819 }
2820 
2821 int
freebsd32_kmq_timedsend(struct thread * td,struct freebsd32_kmq_timedsend_args * uap)2822 freebsd32_kmq_timedsend(struct thread *td,
2823     struct freebsd32_kmq_timedsend_args *uap)
2824 {
2825 	struct timespec32 ets32;
2826 	struct timespec *abs_timeout, ets;
2827 	int error;
2828 
2829 	if (uap->abs_timeout != NULL) {
2830 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2831 		if (error != 0)
2832 			return (error);
2833 		CP(ets32, ets, tv_sec);
2834 		CP(ets32, ets, tv_nsec);
2835 		abs_timeout = &ets;
2836 	} else
2837 		abs_timeout = NULL;
2838 
2839 	return (kern_kmq_timedsend(td, uap->mqd, uap->msg_ptr, uap->msg_len,
2840 		uap->msg_prio, abs_timeout));
2841 }
2842 
2843 int
freebsd32_kmq_timedreceive(struct thread * td,struct freebsd32_kmq_timedreceive_args * uap)2844 freebsd32_kmq_timedreceive(struct thread *td,
2845     struct freebsd32_kmq_timedreceive_args *uap)
2846 {
2847 	struct timespec32 ets32;
2848 	struct timespec *abs_timeout, ets;
2849 	int error;
2850 
2851 	if (uap->abs_timeout != NULL) {
2852 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2853 		if (error != 0)
2854 			return (error);
2855 		CP(ets32, ets, tv_sec);
2856 		CP(ets32, ets, tv_nsec);
2857 		abs_timeout = &ets;
2858 	} else
2859 		abs_timeout = NULL;
2860 
2861 	return (kern_kmq_timedreceive(td, uap->mqd, uap->msg_ptr, uap->msg_len,
2862 		uap->msg_prio, abs_timeout));
2863 }
2864 
2865 int
freebsd32_kmq_notify(struct thread * td,struct freebsd32_kmq_notify_args * uap)2866 freebsd32_kmq_notify(struct thread *td, struct freebsd32_kmq_notify_args *uap)
2867 {
2868 	struct sigevent ev, *evp;
2869 	struct sigevent32 ev32;
2870 	int error;
2871 
2872 	if (uap->sigev == NULL) {
2873 		evp = NULL;
2874 	} else {
2875 		error = copyin(uap->sigev, &ev32, sizeof(ev32));
2876 		if (error != 0)
2877 			return (error);
2878 		error = convert_sigevent32(&ev32, &ev);
2879 		if (error != 0)
2880 			return (error);
2881 		evp = &ev;
2882 	}
2883 	return (kern_kmq_notify(td, uap->mqd, evp));
2884 }
2885 
2886 static struct syscall_helper_data mq32_syscalls[] = {
2887 	SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2888 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_setattr, SYF_CAPENABLED),
2889 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedsend, SYF_CAPENABLED),
2890 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedreceive, SYF_CAPENABLED),
2891 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_notify, SYF_CAPENABLED),
2892 	SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink),
2893 	SYSCALL_INIT_LAST
2894 };
2895 #endif
2896 
2897 static int
mqinit(void)2898 mqinit(void)
2899 {
2900 	int error;
2901 
2902 	error = syscall_helper_register(mq_syscalls, SY_THR_STATIC_KLD);
2903 	if (error != 0)
2904 		return (error);
2905 #ifdef COMPAT_FREEBSD32
2906 	error = syscall32_helper_register(mq32_syscalls, SY_THR_STATIC_KLD);
2907 	if (error != 0)
2908 		return (error);
2909 #endif
2910 	return (0);
2911 }
2912 
2913 static int
mqunload(void)2914 mqunload(void)
2915 {
2916 
2917 #ifdef COMPAT_FREEBSD32
2918 	syscall32_helper_unregister(mq32_syscalls);
2919 #endif
2920 	syscall_helper_unregister(mq_syscalls);
2921 	return (0);
2922 }
2923 
2924 static int
mq_modload(struct module * module,int cmd,void * arg)2925 mq_modload(struct module *module, int cmd, void *arg)
2926 {
2927 	int error = 0;
2928 
2929 	error = vfs_modevent(module, cmd, arg);
2930 	if (error != 0)
2931 		return (error);
2932 
2933 	switch (cmd) {
2934 	case MOD_LOAD:
2935 		error = mqinit();
2936 		if (error != 0)
2937 			mqunload();
2938 		break;
2939 	case MOD_UNLOAD:
2940 		error = mqunload();
2941 		break;
2942 	default:
2943 		break;
2944 	}
2945 	return (error);
2946 }
2947 
2948 static moduledata_t mqueuefs_mod = {
2949 	"mqueuefs",
2950 	mq_modload,
2951 	&mqueuefs_vfsconf
2952 };
2953 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2954 MODULE_VERSION(mqueuefs, 1);
2955