1 /*-
2 * Copyright (c) 2005 David Xu <davidxu@freebsd.org>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 */
27
28 /*
29 * POSIX message queue implementation.
30 *
31 * 1) A mqueue filesystem can be mounted, each message queue appears
32 * in mounted directory, user can change queue's permission and
33 * ownership, or remove a queue. Manually creating a file in the
34 * directory causes a message queue to be created in the kernel with
35 * default message queue attributes applied and same name used, this
36 * method is not advocated since mq_open syscall allows user to specify
37 * different attributes. Also the file system can be mounted multiple
38 * times at different mount points but shows same contents.
39 *
40 * 2) Standard POSIX message queue API. The syscalls do not use vfs layer,
41 * but directly operate on internal data structure, this allows user to
42 * use the IPC facility without having to mount mqueue file system.
43 */
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD: releng/8.2/sys/kern/uipc_mqueue.c 206351 2010-04-07 14:49:35Z kib $");
47
48 #include "opt_compat.h"
49
50 #include <sys/param.h>
51 #include <sys/kernel.h>
52 #include <sys/systm.h>
53 #include <sys/limits.h>
54 #include <sys/buf.h>
55 #include <sys/dirent.h>
56 #include <sys/event.h>
57 #include <sys/eventhandler.h>
58 #include <sys/fcntl.h>
59 #include <sys/file.h>
60 #include <sys/filedesc.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/module.h>
64 #include <sys/mount.h>
65 #include <sys/mqueue.h>
66 #include <sys/mutex.h>
67 #include <sys/namei.h>
68 #include <sys/posix4.h>
69 #include <sys/poll.h>
70 #include <sys/priv.h>
71 #include <sys/proc.h>
72 #include <sys/queue.h>
73 #include <sys/sysproto.h>
74 #include <sys/stat.h>
75 #include <sys/syscall.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysent.h>
78 #include <sys/sx.h>
79 #include <sys/sysctl.h>
80 #include <sys/taskqueue.h>
81 #include <sys/unistd.h>
82 #include <sys/vnode.h>
83 #include <machine/atomic.h>
84
85 /*
86 * Limits and constants
87 */
88 #define MQFS_NAMELEN NAME_MAX
89 #define MQFS_DELEN (8 + MQFS_NAMELEN)
90
91 /* node types */
92 typedef enum {
93 mqfstype_none = 0,
94 mqfstype_root,
95 mqfstype_dir,
96 mqfstype_this,
97 mqfstype_parent,
98 mqfstype_file,
99 mqfstype_symlink,
100 } mqfs_type_t;
101
102 struct mqfs_node;
103
104 /*
105 * mqfs_info: describes a mqfs instance
106 */
107 struct mqfs_info {
108 struct sx mi_lock;
109 struct mqfs_node *mi_root;
110 struct unrhdr *mi_unrhdr;
111 };
112
113 struct mqfs_vdata {
114 LIST_ENTRY(mqfs_vdata) mv_link;
115 struct mqfs_node *mv_node;
116 struct vnode *mv_vnode;
117 struct task mv_task;
118 };
119
120 /*
121 * mqfs_node: describes a node (file or directory) within a mqfs
122 */
123 struct mqfs_node {
124 char mn_name[MQFS_NAMELEN+1];
125 struct mqfs_info *mn_info;
126 struct mqfs_node *mn_parent;
127 LIST_HEAD(,mqfs_node) mn_children;
128 LIST_ENTRY(mqfs_node) mn_sibling;
129 LIST_HEAD(,mqfs_vdata) mn_vnodes;
130 int mn_refcount;
131 mqfs_type_t mn_type;
132 int mn_deleted;
133 u_int32_t mn_fileno;
134 void *mn_data;
135 struct timespec mn_birth;
136 struct timespec mn_ctime;
137 struct timespec mn_atime;
138 struct timespec mn_mtime;
139 uid_t mn_uid;
140 gid_t mn_gid;
141 int mn_mode;
142 };
143
144 #define VTON(vp) (((struct mqfs_vdata *)((vp)->v_data))->mv_node)
145 #define VTOMQ(vp) ((struct mqueue *)(VTON(vp)->mn_data))
146 #define VFSTOMQFS(m) ((struct mqfs_info *)((m)->mnt_data))
147 #define FPTOMQ(fp) ((struct mqueue *)(((struct mqfs_node *) \
148 (fp)->f_data)->mn_data))
149
150 TAILQ_HEAD(msgq, mqueue_msg);
151
152 struct mqueue;
153
154 struct mqueue_notifier {
155 LIST_ENTRY(mqueue_notifier) nt_link;
156 struct sigevent nt_sigev;
157 ksiginfo_t nt_ksi;
158 struct proc *nt_proc;
159 };
160
161 struct mqueue {
162 struct mtx mq_mutex;
163 int mq_flags;
164 long mq_maxmsg;
165 long mq_msgsize;
166 long mq_curmsgs;
167 long mq_totalbytes;
168 struct msgq mq_msgq;
169 int mq_receivers;
170 int mq_senders;
171 struct selinfo mq_rsel;
172 struct selinfo mq_wsel;
173 struct mqueue_notifier *mq_notifier;
174 };
175
176 #define MQ_RSEL 0x01
177 #define MQ_WSEL 0x02
178
179 struct mqueue_msg {
180 TAILQ_ENTRY(mqueue_msg) msg_link;
181 unsigned int msg_prio;
182 unsigned int msg_size;
183 /* following real data... */
184 };
185
186 SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
187 "POSIX real time message queue");
188
189 static int default_maxmsg = 10;
190 static int default_msgsize = 1024;
191
192 static int maxmsg = 100;
193 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
194 &maxmsg, 0, "Default maximum messages in queue");
195 static int maxmsgsize = 16384;
196 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
197 &maxmsgsize, 0, "Default maximum message size");
198 static int maxmq = 100;
199 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
200 &maxmq, 0, "maximum message queues");
201 static int curmq = 0;
202 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
203 &curmq, 0, "current message queue number");
204 static int unloadable = 0;
205 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
206
207 static eventhandler_tag exit_tag;
208
209 /* Only one instance per-system */
210 static struct mqfs_info mqfs_data;
211 static uma_zone_t mqnode_zone;
212 static uma_zone_t mqueue_zone;
213 static uma_zone_t mvdata_zone;
214 static uma_zone_t mqnoti_zone;
215 static struct vop_vector mqfs_vnodeops;
216 static struct fileops mqueueops;
217
218 /*
219 * Directory structure construction and manipulation
220 */
221 #ifdef notyet
222 static struct mqfs_node *mqfs_create_dir(struct mqfs_node *parent,
223 const char *name, int namelen, struct ucred *cred, int mode);
224 static struct mqfs_node *mqfs_create_link(struct mqfs_node *parent,
225 const char *name, int namelen, struct ucred *cred, int mode);
226 #endif
227
228 static struct mqfs_node *mqfs_create_file(struct mqfs_node *parent,
229 const char *name, int namelen, struct ucred *cred, int mode);
230 static int mqfs_destroy(struct mqfs_node *mn);
231 static void mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
232 static void mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
233 static int mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
234
235 /*
236 * Message queue construction and maniplation
237 */
238 static struct mqueue *mqueue_alloc(const struct mq_attr *attr);
239 static void mqueue_free(struct mqueue *mq);
240 static int mqueue_send(struct mqueue *mq, const char *msg_ptr,
241 size_t msg_len, unsigned msg_prio, int waitok,
242 const struct timespec *abs_timeout);
243 static int mqueue_receive(struct mqueue *mq, char *msg_ptr,
244 size_t msg_len, unsigned *msg_prio, int waitok,
245 const struct timespec *abs_timeout);
246 static int _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
247 int timo);
248 static int _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
249 int timo);
250 static void mqueue_send_notification(struct mqueue *mq);
251 static void mqueue_fdclose(struct thread *td, int fd, struct file *fp);
252 static void mq_proc_exit(void *arg, struct proc *p);
253
254 /*
255 * kqueue filters
256 */
257 static void filt_mqdetach(struct knote *kn);
258 static int filt_mqread(struct knote *kn, long hint);
259 static int filt_mqwrite(struct knote *kn, long hint);
260
261 struct filterops mq_rfiltops =
262 { 1, NULL, filt_mqdetach, filt_mqread };
263 struct filterops mq_wfiltops =
264 { 1, NULL, filt_mqdetach, filt_mqwrite };
265
266 /*
267 * Initialize fileno bitmap
268 */
269 static void
270 mqfs_fileno_init(struct mqfs_info *mi)
271 {
272 struct unrhdr *up;
273
274 up = new_unrhdr(1, INT_MAX, NULL);
275 mi->mi_unrhdr = up;
276 }
277
278 /*
279 * Tear down fileno bitmap
280 */
281 static void
282 mqfs_fileno_uninit(struct mqfs_info *mi)
283 {
284 struct unrhdr *up;
285
286 up = mi->mi_unrhdr;
287 mi->mi_unrhdr = NULL;
288 delete_unrhdr(up);
289 }
290
291 /*
292 * Allocate a file number
293 */
294 static void
295 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
296 {
297 /* make sure our parent has a file number */
298 if (mn->mn_parent && !mn->mn_parent->mn_fileno)
299 mqfs_fileno_alloc(mi, mn->mn_parent);
300
301 switch (mn->mn_type) {
302 case mqfstype_root:
303 case mqfstype_dir:
304 case mqfstype_file:
305 case mqfstype_symlink:
306 mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
307 break;
308 case mqfstype_this:
309 KASSERT(mn->mn_parent != NULL,
310 ("mqfstype_this node has no parent"));
311 mn->mn_fileno = mn->mn_parent->mn_fileno;
312 break;
313 case mqfstype_parent:
314 KASSERT(mn->mn_parent != NULL,
315 ("mqfstype_parent node has no parent"));
316 if (mn->mn_parent == mi->mi_root) {
317 mn->mn_fileno = mn->mn_parent->mn_fileno;
318 break;
319 }
320 KASSERT(mn->mn_parent->mn_parent != NULL,
321 ("mqfstype_parent node has no grandparent"));
322 mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
323 break;
324 default:
325 KASSERT(0,
326 ("mqfs_fileno_alloc() called for unknown type node: %d",
327 mn->mn_type));
328 break;
329 }
330 }
331
332 /*
333 * Release a file number
334 */
335 static void
336 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
337 {
338 switch (mn->mn_type) {
339 case mqfstype_root:
340 case mqfstype_dir:
341 case mqfstype_file:
342 case mqfstype_symlink:
343 free_unr(mi->mi_unrhdr, mn->mn_fileno);
344 break;
345 case mqfstype_this:
346 case mqfstype_parent:
347 /* ignore these, as they don't "own" their file number */
348 break;
349 default:
350 KASSERT(0,
351 ("mqfs_fileno_free() called for unknown type node: %d",
352 mn->mn_type));
353 break;
354 }
355 }
356
357 static __inline struct mqfs_node *
358 mqnode_alloc(void)
359 {
360 return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
361 }
362
363 static __inline void
364 mqnode_free(struct mqfs_node *node)
365 {
366 uma_zfree(mqnode_zone, node);
367 }
368
369 static __inline void
370 mqnode_addref(struct mqfs_node *node)
371 {
372 atomic_fetchadd_int(&node->mn_refcount, 1);
373 }
374
375 static __inline void
376 mqnode_release(struct mqfs_node *node)
377 {
378 struct mqfs_info *mqfs;
379 int old, exp;
380
381 mqfs = node->mn_info;
382 old = atomic_fetchadd_int(&node->mn_refcount, -1);
383 if (node->mn_type == mqfstype_dir ||
384 node->mn_type == mqfstype_root)
385 exp = 3; /* include . and .. */
386 else
387 exp = 1;
388 if (old == exp) {
389 int locked = sx_xlocked(&mqfs->mi_lock);
390 if (!locked)
391 sx_xlock(&mqfs->mi_lock);
392 mqfs_destroy(node);
393 if (!locked)
394 sx_xunlock(&mqfs->mi_lock);
395 }
396 }
397
398 /*
399 * Add a node to a directory
400 */
401 static int
402 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
403 {
404 KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
405 KASSERT(parent->mn_info != NULL,
406 ("%s(): parent has no mn_info", __func__));
407 KASSERT(parent->mn_type == mqfstype_dir ||
408 parent->mn_type == mqfstype_root,
409 ("%s(): parent is not a directory", __func__));
410
411 node->mn_info = parent->mn_info;
412 node->mn_parent = parent;
413 LIST_INIT(&node->mn_children);
414 LIST_INIT(&node->mn_vnodes);
415 LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
416 mqnode_addref(parent);
417 return (0);
418 }
419
420 static struct mqfs_node *
421 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
422 int nodetype)
423 {
424 struct mqfs_node *node;
425
426 node = mqnode_alloc();
427 strncpy(node->mn_name, name, namelen);
428 node->mn_type = nodetype;
429 node->mn_refcount = 1;
430 vfs_timestamp(&node->mn_birth);
431 node->mn_ctime = node->mn_atime = node->mn_mtime
432 = node->mn_birth;
433 node->mn_uid = cred->cr_uid;
434 node->mn_gid = cred->cr_gid;
435 node->mn_mode = mode;
436 return (node);
437 }
438
439 /*
440 * Create a file
441 */
442 static struct mqfs_node *
443 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
444 struct ucred *cred, int mode)
445 {
446 struct mqfs_node *node;
447
448 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
449 if (mqfs_add_node(parent, node) != 0) {
450 mqnode_free(node);
451 return (NULL);
452 }
453 return (node);
454 }
455
456 /*
457 * Add . and .. to a directory
458 */
459 static int
460 mqfs_fixup_dir(struct mqfs_node *parent)
461 {
462 struct mqfs_node *dir;
463
464 dir = mqnode_alloc();
465 dir->mn_name[0] = '.';
466 dir->mn_type = mqfstype_this;
467 dir->mn_refcount = 1;
468 if (mqfs_add_node(parent, dir) != 0) {
469 mqnode_free(dir);
470 return (-1);
471 }
472
473 dir = mqnode_alloc();
474 dir->mn_name[0] = dir->mn_name[1] = '.';
475 dir->mn_type = mqfstype_parent;
476 dir->mn_refcount = 1;
477
478 if (mqfs_add_node(parent, dir) != 0) {
479 mqnode_free(dir);
480 return (-1);
481 }
482
483 return (0);
484 }
485
486 #ifdef notyet
487
488 /*
489 * Create a directory
490 */
491 static struct mqfs_node *
492 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
493 struct ucred *cred, int mode)
494 {
495 struct mqfs_node *node;
496
497 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
498 if (mqfs_add_node(parent, node) != 0) {
499 mqnode_free(node);
500 return (NULL);
501 }
502
503 if (mqfs_fixup_dir(node) != 0) {
504 mqfs_destroy(node);
505 return (NULL);
506 }
507 return (node);
508 }
509
510 /*
511 * Create a symlink
512 */
513 static struct mqfs_node *
514 mqfs_create_link(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_symlink);
520 if (mqfs_add_node(parent, node) != 0) {
521 mqnode_free(node);
522 return (NULL);
523 }
524 return (node);
525 }
526
527 #endif
528
529 /*
530 * Destroy a node or a tree of nodes
531 */
532 static int
533 mqfs_destroy(struct mqfs_node *node)
534 {
535 struct mqfs_node *parent;
536
537 KASSERT(node != NULL,
538 ("%s(): node is NULL", __func__));
539 KASSERT(node->mn_info != NULL,
540 ("%s(): node has no mn_info", __func__));
541
542 /* destroy children */
543 if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
544 while (! LIST_EMPTY(&node->mn_children))
545 mqfs_destroy(LIST_FIRST(&node->mn_children));
546
547 /* unlink from parent */
548 if ((parent = node->mn_parent) != NULL) {
549 KASSERT(parent->mn_info == node->mn_info,
550 ("%s(): parent has different mn_info", __func__));
551 LIST_REMOVE(node, mn_sibling);
552 }
553
554 if (node->mn_fileno != 0)
555 mqfs_fileno_free(node->mn_info, node);
556 if (node->mn_data != NULL)
557 mqueue_free(node->mn_data);
558 mqnode_free(node);
559 return (0);
560 }
561
562 /*
563 * Mount a mqfs instance
564 */
565 static int
566 mqfs_mount(struct mount *mp)
567 {
568 struct statfs *sbp;
569
570 if (mp->mnt_flag & MNT_UPDATE)
571 return (EOPNOTSUPP);
572
573 mp->mnt_data = &mqfs_data;
574 MNT_ILOCK(mp);
575 mp->mnt_flag |= MNT_LOCAL;
576 mp->mnt_kern_flag |= MNTK_MPSAFE;
577 MNT_IUNLOCK(mp);
578 vfs_getnewfsid(mp);
579
580 sbp = &mp->mnt_stat;
581 vfs_mountedfrom(mp, "mqueue");
582 sbp->f_bsize = PAGE_SIZE;
583 sbp->f_iosize = PAGE_SIZE;
584 sbp->f_blocks = 1;
585 sbp->f_bfree = 0;
586 sbp->f_bavail = 0;
587 sbp->f_files = 1;
588 sbp->f_ffree = 0;
589 return (0);
590 }
591
592 /*
593 * Unmount a mqfs instance
594 */
595 static int
596 mqfs_unmount(struct mount *mp, int mntflags)
597 {
598 int error;
599
600 error = vflush(mp, 0, (mntflags & MNT_FORCE) ? FORCECLOSE : 0,
601 curthread);
602 return (error);
603 }
604
605 /*
606 * Return a root vnode
607 */
608 static int
609 mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
610 {
611 struct mqfs_info *mqfs;
612 int ret;
613
614 mqfs = VFSTOMQFS(mp);
615 ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
616 return (ret);
617 }
618
619 /*
620 * Return filesystem stats
621 */
622 static int
623 mqfs_statfs(struct mount *mp, struct statfs *sbp)
624 {
625 /* XXX update statistics */
626 return (0);
627 }
628
629 /*
630 * Initialize a mqfs instance
631 */
632 static int
633 mqfs_init(struct vfsconf *vfc)
634 {
635 struct mqfs_node *root;
636 struct mqfs_info *mi;
637
638 mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
639 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
640 mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
641 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
642 mvdata_zone = uma_zcreate("mvdata",
643 sizeof(struct mqfs_vdata), NULL, NULL, NULL,
644 NULL, UMA_ALIGN_PTR, 0);
645 mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
646 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
647 mi = &mqfs_data;
648 sx_init(&mi->mi_lock, "mqfs lock");
649 /* set up the root diretory */
650 root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
651 mqfstype_root);
652 root->mn_info = mi;
653 LIST_INIT(&root->mn_children);
654 LIST_INIT(&root->mn_vnodes);
655 mi->mi_root = root;
656 mqfs_fileno_init(mi);
657 mqfs_fileno_alloc(mi, root);
658 mqfs_fixup_dir(root);
659 exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
660 EVENTHANDLER_PRI_ANY);
661 mq_fdclose = mqueue_fdclose;
662 p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
663 return (0);
664 }
665
666 /*
667 * Destroy a mqfs instance
668 */
669 static int
670 mqfs_uninit(struct vfsconf *vfc)
671 {
672 struct mqfs_info *mi;
673
674 if (!unloadable)
675 return (EOPNOTSUPP);
676 EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
677 mi = &mqfs_data;
678 mqfs_destroy(mi->mi_root);
679 mi->mi_root = NULL;
680 mqfs_fileno_uninit(mi);
681 sx_destroy(&mi->mi_lock);
682 uma_zdestroy(mqnode_zone);
683 uma_zdestroy(mqueue_zone);
684 uma_zdestroy(mvdata_zone);
685 uma_zdestroy(mqnoti_zone);
686 return (0);
687 }
688
689 /*
690 * task routine
691 */
692 static void
693 do_recycle(void *context, int pending __unused)
694 {
695 struct vnode *vp = (struct vnode *)context;
696
697 vrecycle(vp, curthread);
698 vdrop(vp);
699 }
700
701 /*
702 * Allocate a vnode
703 */
704 static int
705 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
706 {
707 struct mqfs_vdata *vd;
708 struct mqfs_info *mqfs;
709 struct vnode *newvpp;
710 int error;
711
712 mqfs = pn->mn_info;
713 *vpp = NULL;
714 sx_xlock(&mqfs->mi_lock);
715 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
716 if (vd->mv_vnode->v_mount == mp) {
717 vhold(vd->mv_vnode);
718 break;
719 }
720 }
721
722 if (vd != NULL) {
723 found:
724 *vpp = vd->mv_vnode;
725 sx_xunlock(&mqfs->mi_lock);
726 error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
727 vdrop(*vpp);
728 return (error);
729 }
730 sx_xunlock(&mqfs->mi_lock);
731
732 error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
733 if (error)
734 return (error);
735 vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
736 error = insmntque(newvpp, mp);
737 if (error != 0)
738 return (error);
739
740 sx_xlock(&mqfs->mi_lock);
741 /*
742 * Check if it has already been allocated
743 * while we were blocked.
744 */
745 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
746 if (vd->mv_vnode->v_mount == mp) {
747 vhold(vd->mv_vnode);
748 sx_xunlock(&mqfs->mi_lock);
749
750 vgone(newvpp);
751 vput(newvpp);
752 goto found;
753 }
754 }
755
756 *vpp = newvpp;
757
758 vd = uma_zalloc(mvdata_zone, M_WAITOK);
759 (*vpp)->v_data = vd;
760 vd->mv_vnode = *vpp;
761 vd->mv_node = pn;
762 TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
763 LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
764 mqnode_addref(pn);
765 switch (pn->mn_type) {
766 case mqfstype_root:
767 (*vpp)->v_vflag = VV_ROOT;
768 /* fall through */
769 case mqfstype_dir:
770 case mqfstype_this:
771 case mqfstype_parent:
772 (*vpp)->v_type = VDIR;
773 break;
774 case mqfstype_file:
775 (*vpp)->v_type = VREG;
776 break;
777 case mqfstype_symlink:
778 (*vpp)->v_type = VLNK;
779 break;
780 case mqfstype_none:
781 KASSERT(0, ("mqfs_allocf called for null node\n"));
782 default:
783 panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
784 }
785 sx_xunlock(&mqfs->mi_lock);
786 return (0);
787 }
788
789 /*
790 * Search a directory entry
791 */
792 static struct mqfs_node *
793 mqfs_search(struct mqfs_node *pd, const char *name, int len)
794 {
795 struct mqfs_node *pn;
796
797 sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
798 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
799 if (strncmp(pn->mn_name, name, len) == 0 &&
800 pn->mn_name[len] == '\0')
801 return (pn);
802 }
803 return (NULL);
804 }
805
806 /*
807 * Look up a file or directory.
808 */
809 static int
810 mqfs_lookupx(struct vop_cachedlookup_args *ap)
811 {
812 struct componentname *cnp;
813 struct vnode *dvp, **vpp;
814 struct mqfs_node *pd;
815 struct mqfs_node *pn;
816 struct mqfs_info *mqfs;
817 int nameiop, flags, error, namelen;
818 char *pname;
819 struct thread *td;
820
821 cnp = ap->a_cnp;
822 vpp = ap->a_vpp;
823 dvp = ap->a_dvp;
824 pname = cnp->cn_nameptr;
825 namelen = cnp->cn_namelen;
826 td = cnp->cn_thread;
827 flags = cnp->cn_flags;
828 nameiop = cnp->cn_nameiop;
829 pd = VTON(dvp);
830 pn = NULL;
831 mqfs = pd->mn_info;
832 *vpp = NULLVP;
833
834 if (dvp->v_type != VDIR)
835 return (ENOTDIR);
836
837 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
838 if (error)
839 return (error);
840
841 /* shortcut: check if the name is too long */
842 if (cnp->cn_namelen >= MQFS_NAMELEN)
843 return (ENOENT);
844
845 /* self */
846 if (namelen == 1 && pname[0] == '.') {
847 if ((flags & ISLASTCN) && nameiop != LOOKUP)
848 return (EINVAL);
849 pn = pd;
850 *vpp = dvp;
851 VREF(dvp);
852 return (0);
853 }
854
855 /* parent */
856 if (cnp->cn_flags & ISDOTDOT) {
857 if (dvp->v_vflag & VV_ROOT)
858 return (EIO);
859 if ((flags & ISLASTCN) && nameiop != LOOKUP)
860 return (EINVAL);
861 VOP_UNLOCK(dvp, 0);
862 KASSERT(pd->mn_parent, ("non-root directory has no parent"));
863 pn = pd->mn_parent;
864 error = mqfs_allocv(dvp->v_mount, vpp, pn);
865 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
866 return (error);
867 }
868
869 /* named node */
870 sx_xlock(&mqfs->mi_lock);
871 pn = mqfs_search(pd, pname, namelen);
872 if (pn != NULL)
873 mqnode_addref(pn);
874 sx_xunlock(&mqfs->mi_lock);
875
876 /* found */
877 if (pn != NULL) {
878 /* DELETE */
879 if (nameiop == DELETE && (flags & ISLASTCN)) {
880 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
881 if (error) {
882 mqnode_release(pn);
883 return (error);
884 }
885 if (*vpp == dvp) {
886 VREF(dvp);
887 *vpp = dvp;
888 mqnode_release(pn);
889 return (0);
890 }
891 }
892
893 /* allocate vnode */
894 error = mqfs_allocv(dvp->v_mount, vpp, pn);
895 mqnode_release(pn);
896 if (error == 0 && cnp->cn_flags & MAKEENTRY)
897 cache_enter(dvp, *vpp, cnp);
898 return (error);
899 }
900
901 /* not found */
902
903 /* will create a new entry in the directory ? */
904 if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
905 && (flags & ISLASTCN)) {
906 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
907 if (error)
908 return (error);
909 cnp->cn_flags |= SAVENAME;
910 return (EJUSTRETURN);
911 }
912 return (ENOENT);
913 }
914
915 #if 0
916 struct vop_lookup_args {
917 struct vop_generic_args a_gen;
918 struct vnode *a_dvp;
919 struct vnode **a_vpp;
920 struct componentname *a_cnp;
921 };
922 #endif
923
924 /*
925 * vnode lookup operation
926 */
927 static int
928 mqfs_lookup(struct vop_cachedlookup_args *ap)
929 {
930 int rc;
931
932 rc = mqfs_lookupx(ap);
933 return (rc);
934 }
935
936 #if 0
937 struct vop_create_args {
938 struct vnode *a_dvp;
939 struct vnode **a_vpp;
940 struct componentname *a_cnp;
941 struct vattr *a_vap;
942 };
943 #endif
944
945 /*
946 * vnode creation operation
947 */
948 static int
949 mqfs_create(struct vop_create_args *ap)
950 {
951 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
952 struct componentname *cnp = ap->a_cnp;
953 struct mqfs_node *pd;
954 struct mqfs_node *pn;
955 struct mqueue *mq;
956 int error;
957
958 pd = VTON(ap->a_dvp);
959 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
960 return (ENOTDIR);
961 mq = mqueue_alloc(NULL);
962 if (mq == NULL)
963 return (EAGAIN);
964 sx_xlock(&mqfs->mi_lock);
965 if ((cnp->cn_flags & HASBUF) == 0)
966 panic("%s: no name", __func__);
967 pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
968 cnp->cn_cred, ap->a_vap->va_mode);
969 if (pn == NULL) {
970 sx_xunlock(&mqfs->mi_lock);
971 error = ENOSPC;
972 } else {
973 mqnode_addref(pn);
974 sx_xunlock(&mqfs->mi_lock);
975 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
976 mqnode_release(pn);
977 if (error)
978 mqfs_destroy(pn);
979 else
980 pn->mn_data = mq;
981 }
982 if (error)
983 mqueue_free(mq);
984 return (error);
985 }
986
987 /*
988 * Remove an entry
989 */
990 static
991 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
992 {
993 struct mqfs_node *parent;
994 struct mqfs_vdata *vd;
995 int error = 0;
996
997 sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
998
999 if (ucred->cr_uid != pn->mn_uid &&
1000 (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
1001 error = EACCES;
1002 else if (!pn->mn_deleted) {
1003 parent = pn->mn_parent;
1004 pn->mn_parent = NULL;
1005 pn->mn_deleted = 1;
1006 LIST_REMOVE(pn, mn_sibling);
1007 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
1008 cache_purge(vd->mv_vnode);
1009 vhold(vd->mv_vnode);
1010 taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
1011 }
1012 mqnode_release(pn);
1013 mqnode_release(parent);
1014 } else
1015 error = ENOENT;
1016 return (error);
1017 }
1018
1019 #if 0
1020 struct vop_remove_args {
1021 struct vnode *a_dvp;
1022 struct vnode *a_vp;
1023 struct componentname *a_cnp;
1024 };
1025 #endif
1026
1027 /*
1028 * vnode removal operation
1029 */
1030 static int
1031 mqfs_remove(struct vop_remove_args *ap)
1032 {
1033 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1034 struct mqfs_node *pn;
1035 int error;
1036
1037 if (ap->a_vp->v_type == VDIR)
1038 return (EPERM);
1039 pn = VTON(ap->a_vp);
1040 sx_xlock(&mqfs->mi_lock);
1041 error = do_unlink(pn, ap->a_cnp->cn_cred);
1042 sx_xunlock(&mqfs->mi_lock);
1043 return (error);
1044 }
1045
1046 #if 0
1047 struct vop_inactive_args {
1048 struct vnode *a_vp;
1049 struct thread *a_td;
1050 };
1051 #endif
1052
1053 static int
1054 mqfs_inactive(struct vop_inactive_args *ap)
1055 {
1056 struct mqfs_node *pn = VTON(ap->a_vp);
1057
1058 if (pn->mn_deleted)
1059 vrecycle(ap->a_vp, ap->a_td);
1060 return (0);
1061 }
1062
1063 #if 0
1064 struct vop_reclaim_args {
1065 struct vop_generic_args a_gen;
1066 struct vnode *a_vp;
1067 struct thread *a_td;
1068 };
1069 #endif
1070
1071 static int
1072 mqfs_reclaim(struct vop_reclaim_args *ap)
1073 {
1074 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1075 struct vnode *vp = ap->a_vp;
1076 struct mqfs_node *pn;
1077 struct mqfs_vdata *vd;
1078
1079 vd = vp->v_data;
1080 pn = vd->mv_node;
1081 sx_xlock(&mqfs->mi_lock);
1082 vp->v_data = NULL;
1083 LIST_REMOVE(vd, mv_link);
1084 uma_zfree(mvdata_zone, vd);
1085 mqnode_release(pn);
1086 sx_xunlock(&mqfs->mi_lock);
1087 return (0);
1088 }
1089
1090 #if 0
1091 struct vop_open_args {
1092 struct vop_generic_args a_gen;
1093 struct vnode *a_vp;
1094 int a_mode;
1095 struct ucred *a_cred;
1096 struct thread *a_td;
1097 struct file *a_fp;
1098 };
1099 #endif
1100
1101 static int
1102 mqfs_open(struct vop_open_args *ap)
1103 {
1104 return (0);
1105 }
1106
1107 #if 0
1108 struct vop_close_args {
1109 struct vop_generic_args a_gen;
1110 struct vnode *a_vp;
1111 int a_fflag;
1112 struct ucred *a_cred;
1113 struct thread *a_td;
1114 };
1115 #endif
1116
1117 static int
1118 mqfs_close(struct vop_close_args *ap)
1119 {
1120 return (0);
1121 }
1122
1123 #if 0
1124 struct vop_access_args {
1125 struct vop_generic_args a_gen;
1126 struct vnode *a_vp;
1127 accmode_t a_accmode;
1128 struct ucred *a_cred;
1129 struct thread *a_td;
1130 };
1131 #endif
1132
1133 /*
1134 * Verify permissions
1135 */
1136 static int
1137 mqfs_access(struct vop_access_args *ap)
1138 {
1139 struct vnode *vp = ap->a_vp;
1140 struct vattr vattr;
1141 int error;
1142
1143 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
1144 if (error)
1145 return (error);
1146 error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1147 vattr.va_gid, ap->a_accmode, ap->a_cred, NULL);
1148 return (error);
1149 }
1150
1151 #if 0
1152 struct vop_getattr_args {
1153 struct vop_generic_args a_gen;
1154 struct vnode *a_vp;
1155 struct vattr *a_vap;
1156 struct ucred *a_cred;
1157 };
1158 #endif
1159
1160 /*
1161 * Get file attributes
1162 */
1163 static int
1164 mqfs_getattr(struct vop_getattr_args *ap)
1165 {
1166 struct vnode *vp = ap->a_vp;
1167 struct mqfs_node *pn = VTON(vp);
1168 struct vattr *vap = ap->a_vap;
1169 int error = 0;
1170
1171 vap->va_type = vp->v_type;
1172 vap->va_mode = pn->mn_mode;
1173 vap->va_nlink = 1;
1174 vap->va_uid = pn->mn_uid;
1175 vap->va_gid = pn->mn_gid;
1176 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1177 vap->va_fileid = pn->mn_fileno;
1178 vap->va_size = 0;
1179 vap->va_blocksize = PAGE_SIZE;
1180 vap->va_bytes = vap->va_size = 0;
1181 vap->va_atime = pn->mn_atime;
1182 vap->va_mtime = pn->mn_mtime;
1183 vap->va_ctime = pn->mn_ctime;
1184 vap->va_birthtime = pn->mn_birth;
1185 vap->va_gen = 0;
1186 vap->va_flags = 0;
1187 vap->va_rdev = NODEV;
1188 vap->va_bytes = 0;
1189 vap->va_filerev = 0;
1190 return (error);
1191 }
1192
1193 #if 0
1194 struct vop_setattr_args {
1195 struct vop_generic_args a_gen;
1196 struct vnode *a_vp;
1197 struct vattr *a_vap;
1198 struct ucred *a_cred;
1199 };
1200 #endif
1201 /*
1202 * Set attributes
1203 */
1204 static int
1205 mqfs_setattr(struct vop_setattr_args *ap)
1206 {
1207 struct mqfs_node *pn;
1208 struct vattr *vap;
1209 struct vnode *vp;
1210 struct thread *td;
1211 int c, error;
1212 uid_t uid;
1213 gid_t gid;
1214
1215 td = curthread;
1216 vap = ap->a_vap;
1217 vp = ap->a_vp;
1218 if ((vap->va_type != VNON) ||
1219 (vap->va_nlink != VNOVAL) ||
1220 (vap->va_fsid != VNOVAL) ||
1221 (vap->va_fileid != VNOVAL) ||
1222 (vap->va_blocksize != VNOVAL) ||
1223 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1224 (vap->va_rdev != VNOVAL) ||
1225 ((int)vap->va_bytes != VNOVAL) ||
1226 (vap->va_gen != VNOVAL)) {
1227 return (EINVAL);
1228 }
1229
1230 pn = VTON(vp);
1231
1232 error = c = 0;
1233 if (vap->va_uid == (uid_t)VNOVAL)
1234 uid = pn->mn_uid;
1235 else
1236 uid = vap->va_uid;
1237 if (vap->va_gid == (gid_t)VNOVAL)
1238 gid = pn->mn_gid;
1239 else
1240 gid = vap->va_gid;
1241
1242 if (uid != pn->mn_uid || gid != pn->mn_gid) {
1243 /*
1244 * To modify the ownership of a file, must possess VADMIN
1245 * for that file.
1246 */
1247 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
1248 return (error);
1249
1250 /*
1251 * XXXRW: Why is there a privilege check here: shouldn't the
1252 * check in VOP_ACCESS() be enough? Also, are the group bits
1253 * below definitely right?
1254 */
1255 if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1256 (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1257 (error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
1258 return (error);
1259 pn->mn_uid = uid;
1260 pn->mn_gid = gid;
1261 c = 1;
1262 }
1263
1264 if (vap->va_mode != (mode_t)VNOVAL) {
1265 if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1266 (error = priv_check(td, PRIV_MQ_ADMIN)))
1267 return (error);
1268 pn->mn_mode = vap->va_mode;
1269 c = 1;
1270 }
1271
1272 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1273 /* See the comment in ufs_vnops::ufs_setattr(). */
1274 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1275 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1276 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1277 return (error);
1278 if (vap->va_atime.tv_sec != VNOVAL) {
1279 pn->mn_atime = vap->va_atime;
1280 }
1281 if (vap->va_mtime.tv_sec != VNOVAL) {
1282 pn->mn_mtime = vap->va_mtime;
1283 }
1284 c = 1;
1285 }
1286 if (c) {
1287 vfs_timestamp(&pn->mn_ctime);
1288 }
1289 return (0);
1290 }
1291
1292 #if 0
1293 struct vop_read_args {
1294 struct vop_generic_args a_gen;
1295 struct vnode *a_vp;
1296 struct uio *a_uio;
1297 int a_ioflag;
1298 struct ucred *a_cred;
1299 };
1300 #endif
1301
1302 /*
1303 * Read from a file
1304 */
1305 static int
1306 mqfs_read(struct vop_read_args *ap)
1307 {
1308 char buf[80];
1309 struct vnode *vp = ap->a_vp;
1310 struct uio *uio = ap->a_uio;
1311 struct mqfs_node *pn;
1312 struct mqueue *mq;
1313 int len, error;
1314
1315 if (vp->v_type != VREG)
1316 return (EINVAL);
1317
1318 pn = VTON(vp);
1319 mq = VTOMQ(vp);
1320 snprintf(buf, sizeof(buf),
1321 "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1322 mq->mq_totalbytes,
1323 mq->mq_maxmsg,
1324 mq->mq_curmsgs,
1325 mq->mq_msgsize);
1326 buf[sizeof(buf)-1] = '\0';
1327 len = strlen(buf);
1328 error = uiomove_frombuf(buf, len, uio);
1329 return (error);
1330 }
1331
1332 #if 0
1333 struct vop_readdir_args {
1334 struct vop_generic_args a_gen;
1335 struct vnode *a_vp;
1336 struct uio *a_uio;
1337 struct ucred *a_cred;
1338 int *a_eofflag;
1339 int *a_ncookies;
1340 u_long **a_cookies;
1341 };
1342 #endif
1343
1344 /*
1345 * Return directory entries.
1346 */
1347 static int
1348 mqfs_readdir(struct vop_readdir_args *ap)
1349 {
1350 struct vnode *vp;
1351 struct mqfs_info *mi;
1352 struct mqfs_node *pd;
1353 struct mqfs_node *pn;
1354 struct dirent entry;
1355 struct uio *uio;
1356 int *tmp_ncookies = NULL;
1357 off_t offset;
1358 int error, i;
1359
1360 vp = ap->a_vp;
1361 mi = VFSTOMQFS(vp->v_mount);
1362 pd = VTON(vp);
1363 uio = ap->a_uio;
1364
1365 if (vp->v_type != VDIR)
1366 return (ENOTDIR);
1367
1368 if (uio->uio_offset < 0)
1369 return (EINVAL);
1370
1371 if (ap->a_ncookies != NULL) {
1372 tmp_ncookies = ap->a_ncookies;
1373 *ap->a_ncookies = 0;
1374 ap->a_ncookies = NULL;
1375 }
1376
1377 error = 0;
1378 offset = 0;
1379
1380 sx_xlock(&mi->mi_lock);
1381
1382 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1383 entry.d_reclen = sizeof(entry);
1384 if (!pn->mn_fileno)
1385 mqfs_fileno_alloc(mi, pn);
1386 entry.d_fileno = pn->mn_fileno;
1387 for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1388 entry.d_name[i] = pn->mn_name[i];
1389 entry.d_name[i] = 0;
1390 entry.d_namlen = i;
1391 switch (pn->mn_type) {
1392 case mqfstype_root:
1393 case mqfstype_dir:
1394 case mqfstype_this:
1395 case mqfstype_parent:
1396 entry.d_type = DT_DIR;
1397 break;
1398 case mqfstype_file:
1399 entry.d_type = DT_REG;
1400 break;
1401 case mqfstype_symlink:
1402 entry.d_type = DT_LNK;
1403 break;
1404 default:
1405 panic("%s has unexpected node type: %d", pn->mn_name,
1406 pn->mn_type);
1407 }
1408 if (entry.d_reclen > uio->uio_resid)
1409 break;
1410 if (offset >= uio->uio_offset) {
1411 error = vfs_read_dirent(ap, &entry, offset);
1412 if (error)
1413 break;
1414 }
1415 offset += entry.d_reclen;
1416 }
1417 sx_xunlock(&mi->mi_lock);
1418
1419 uio->uio_offset = offset;
1420
1421 if (tmp_ncookies != NULL)
1422 ap->a_ncookies = tmp_ncookies;
1423
1424 return (error);
1425 }
1426
1427 #ifdef notyet
1428
1429 #if 0
1430 struct vop_mkdir_args {
1431 struct vnode *a_dvp;
1432 struvt vnode **a_vpp;
1433 struvt componentname *a_cnp;
1434 struct vattr *a_vap;
1435 };
1436 #endif
1437
1438 /*
1439 * Create a directory.
1440 */
1441 static int
1442 mqfs_mkdir(struct vop_mkdir_args *ap)
1443 {
1444 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1445 struct componentname *cnp = ap->a_cnp;
1446 struct mqfs_node *pd = VTON(ap->a_dvp);
1447 struct mqfs_node *pn;
1448 int error;
1449
1450 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1451 return (ENOTDIR);
1452 sx_xlock(&mqfs->mi_lock);
1453 if ((cnp->cn_flags & HASBUF) == 0)
1454 panic("%s: no name", __func__);
1455 pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1456 ap->a_vap->cn_cred, ap->a_vap->va_mode);
1457 if (pn != NULL)
1458 mqnode_addref(pn);
1459 sx_xunlock(&mqfs->mi_lock);
1460 if (pn == NULL) {
1461 error = ENOSPC;
1462 } else {
1463 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1464 mqnode_release(pn);
1465 }
1466 return (error);
1467 }
1468
1469 #if 0
1470 struct vop_rmdir_args {
1471 struct vnode *a_dvp;
1472 struct vnode *a_vp;
1473 struct componentname *a_cnp;
1474 };
1475 #endif
1476
1477 /*
1478 * Remove a directory.
1479 */
1480 static int
1481 mqfs_rmdir(struct vop_rmdir_args *ap)
1482 {
1483 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1484 struct mqfs_node *pn = VTON(ap->a_vp);
1485 struct mqfs_node *pt;
1486
1487 if (pn->mn_type != mqfstype_dir)
1488 return (ENOTDIR);
1489
1490 sx_xlock(&mqfs->mi_lock);
1491 if (pn->mn_deleted) {
1492 sx_xunlock(&mqfs->mi_lock);
1493 return (ENOENT);
1494 }
1495
1496 pt = LIST_FIRST(&pn->mn_children);
1497 pt = LIST_NEXT(pt, mn_sibling);
1498 pt = LIST_NEXT(pt, mn_sibling);
1499 if (pt != NULL) {
1500 sx_xunlock(&mqfs->mi_lock);
1501 return (ENOTEMPTY);
1502 }
1503 pt = pn->mn_parent;
1504 pn->mn_parent = NULL;
1505 pn->mn_deleted = 1;
1506 LIST_REMOVE(pn, mn_sibling);
1507 mqnode_release(pn);
1508 mqnode_release(pt);
1509 sx_xunlock(&mqfs->mi_lock);
1510 cache_purge(ap->a_vp);
1511 return (0);
1512 }
1513
1514 #endif /* notyet */
1515
1516 /*
1517 * Allocate a message queue
1518 */
1519 static struct mqueue *
1520 mqueue_alloc(const struct mq_attr *attr)
1521 {
1522 struct mqueue *mq;
1523
1524 if (curmq >= maxmq)
1525 return (NULL);
1526 mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1527 TAILQ_INIT(&mq->mq_msgq);
1528 if (attr != NULL) {
1529 mq->mq_maxmsg = attr->mq_maxmsg;
1530 mq->mq_msgsize = attr->mq_msgsize;
1531 } else {
1532 mq->mq_maxmsg = default_maxmsg;
1533 mq->mq_msgsize = default_msgsize;
1534 }
1535 mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
1536 knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
1537 knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
1538 atomic_add_int(&curmq, 1);
1539 return (mq);
1540 }
1541
1542 /*
1543 * Destroy a message queue
1544 */
1545 static void
1546 mqueue_free(struct mqueue *mq)
1547 {
1548 struct mqueue_msg *msg;
1549
1550 while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1551 TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1552 free(msg, M_MQUEUEDATA);
1553 }
1554
1555 mtx_destroy(&mq->mq_mutex);
1556 knlist_destroy(&mq->mq_rsel.si_note);
1557 knlist_destroy(&mq->mq_wsel.si_note);
1558 uma_zfree(mqueue_zone, mq);
1559 atomic_add_int(&curmq, -1);
1560 }
1561
1562 /*
1563 * Load a message from user space
1564 */
1565 static struct mqueue_msg *
1566 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1567 {
1568 struct mqueue_msg *msg;
1569 size_t len;
1570 int error;
1571
1572 len = sizeof(struct mqueue_msg) + msg_size;
1573 msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1574 error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1575 msg_size);
1576 if (error) {
1577 free(msg, M_MQUEUEDATA);
1578 msg = NULL;
1579 } else {
1580 msg->msg_size = msg_size;
1581 msg->msg_prio = msg_prio;
1582 }
1583 return (msg);
1584 }
1585
1586 /*
1587 * Save a message to user space
1588 */
1589 static int
1590 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1591 {
1592 int error;
1593
1594 error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1595 msg->msg_size);
1596 if (error == 0 && msg_prio != NULL)
1597 error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1598 return (error);
1599 }
1600
1601 /*
1602 * Free a message's memory
1603 */
1604 static __inline void
1605 mqueue_freemsg(struct mqueue_msg *msg)
1606 {
1607 free(msg, M_MQUEUEDATA);
1608 }
1609
1610 /*
1611 * Send a message. if waitok is false, thread will not be
1612 * blocked if there is no data in queue, otherwise, absolute
1613 * time will be checked.
1614 */
1615 int
1616 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1617 size_t msg_len, unsigned msg_prio, int waitok,
1618 const struct timespec *abs_timeout)
1619 {
1620 struct mqueue_msg *msg;
1621 struct timespec ts, ts2;
1622 struct timeval tv;
1623 int error;
1624
1625 if (msg_prio >= MQ_PRIO_MAX)
1626 return (EINVAL);
1627 if (msg_len > mq->mq_msgsize)
1628 return (EMSGSIZE);
1629 msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1630 if (msg == NULL)
1631 return (EFAULT);
1632
1633 /* O_NONBLOCK case */
1634 if (!waitok) {
1635 error = _mqueue_send(mq, msg, -1);
1636 if (error)
1637 goto bad;
1638 return (0);
1639 }
1640
1641 /* we allow a null timeout (wait forever) */
1642 if (abs_timeout == NULL) {
1643 error = _mqueue_send(mq, msg, 0);
1644 if (error)
1645 goto bad;
1646 return (0);
1647 }
1648
1649 /* send it before checking time */
1650 error = _mqueue_send(mq, msg, -1);
1651 if (error == 0)
1652 return (0);
1653
1654 if (error != EAGAIN)
1655 goto bad;
1656
1657 if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1658 error = EINVAL;
1659 goto bad;
1660 }
1661 for (;;) {
1662 ts2 = *abs_timeout;
1663 getnanotime(&ts);
1664 timespecsub(&ts2, &ts);
1665 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1666 error = ETIMEDOUT;
1667 break;
1668 }
1669 TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1670 error = _mqueue_send(mq, msg, tvtohz(&tv));
1671 if (error != ETIMEDOUT)
1672 break;
1673 }
1674 if (error == 0)
1675 return (0);
1676 bad:
1677 mqueue_freemsg(msg);
1678 return (error);
1679 }
1680
1681 /*
1682 * Common routine to send a message
1683 */
1684 static int
1685 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1686 {
1687 struct mqueue_msg *msg2;
1688 int error = 0;
1689
1690 mtx_lock(&mq->mq_mutex);
1691 while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1692 if (timo < 0) {
1693 mtx_unlock(&mq->mq_mutex);
1694 return (EAGAIN);
1695 }
1696 mq->mq_senders++;
1697 error = msleep(&mq->mq_senders, &mq->mq_mutex,
1698 PCATCH, "mqsend", timo);
1699 mq->mq_senders--;
1700 if (error == EAGAIN)
1701 error = ETIMEDOUT;
1702 }
1703 if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1704 mtx_unlock(&mq->mq_mutex);
1705 return (error);
1706 }
1707 error = 0;
1708 if (TAILQ_EMPTY(&mq->mq_msgq)) {
1709 TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1710 } else {
1711 if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1712 TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1713 } else {
1714 TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1715 if (msg2->msg_prio < msg->msg_prio)
1716 break;
1717 }
1718 TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1719 }
1720 }
1721 mq->mq_curmsgs++;
1722 mq->mq_totalbytes += msg->msg_size;
1723 if (mq->mq_receivers)
1724 wakeup_one(&mq->mq_receivers);
1725 else if (mq->mq_notifier != NULL)
1726 mqueue_send_notification(mq);
1727 if (mq->mq_flags & MQ_RSEL) {
1728 mq->mq_flags &= ~MQ_RSEL;
1729 selwakeup(&mq->mq_rsel);
1730 }
1731 KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1732 mtx_unlock(&mq->mq_mutex);
1733 return (0);
1734 }
1735
1736 /*
1737 * Send realtime a signal to process which registered itself
1738 * successfully by mq_notify.
1739 */
1740 static void
1741 mqueue_send_notification(struct mqueue *mq)
1742 {
1743 struct mqueue_notifier *nt;
1744 struct proc *p;
1745
1746 mtx_assert(&mq->mq_mutex, MA_OWNED);
1747 nt = mq->mq_notifier;
1748 if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1749 p = nt->nt_proc;
1750 PROC_LOCK(p);
1751 if (!KSI_ONQ(&nt->nt_ksi))
1752 psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1753 PROC_UNLOCK(p);
1754 }
1755 mq->mq_notifier = NULL;
1756 }
1757
1758 /*
1759 * Get a message. if waitok is false, thread will not be
1760 * blocked if there is no data in queue, otherwise, absolute
1761 * time will be checked.
1762 */
1763 int
1764 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1765 size_t msg_len, unsigned *msg_prio, int waitok,
1766 const struct timespec *abs_timeout)
1767 {
1768 struct mqueue_msg *msg;
1769 struct timespec ts, ts2;
1770 struct timeval tv;
1771 int error;
1772
1773 if (msg_len < mq->mq_msgsize)
1774 return (EMSGSIZE);
1775
1776 /* O_NONBLOCK case */
1777 if (!waitok) {
1778 error = _mqueue_recv(mq, &msg, -1);
1779 if (error)
1780 return (error);
1781 goto received;
1782 }
1783
1784 /* we allow a null timeout (wait forever). */
1785 if (abs_timeout == NULL) {
1786 error = _mqueue_recv(mq, &msg, 0);
1787 if (error)
1788 return (error);
1789 goto received;
1790 }
1791
1792 /* try to get a message before checking time */
1793 error = _mqueue_recv(mq, &msg, -1);
1794 if (error == 0)
1795 goto received;
1796
1797 if (error != EAGAIN)
1798 return (error);
1799
1800 if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1801 error = EINVAL;
1802 return (error);
1803 }
1804
1805 for (;;) {
1806 ts2 = *abs_timeout;
1807 getnanotime(&ts);
1808 timespecsub(&ts2, &ts);
1809 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1810 error = ETIMEDOUT;
1811 return (error);
1812 }
1813 TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1814 error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1815 if (error == 0)
1816 break;
1817 if (error != ETIMEDOUT)
1818 return (error);
1819 }
1820
1821 received:
1822 error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1823 if (error == 0) {
1824 curthread->td_retval[0] = msg->msg_size;
1825 curthread->td_retval[1] = 0;
1826 }
1827 mqueue_freemsg(msg);
1828 return (error);
1829 }
1830
1831 /*
1832 * Common routine to receive a message
1833 */
1834 static int
1835 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1836 {
1837 int error = 0;
1838
1839 mtx_lock(&mq->mq_mutex);
1840 while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1841 if (timo < 0) {
1842 mtx_unlock(&mq->mq_mutex);
1843 return (EAGAIN);
1844 }
1845 mq->mq_receivers++;
1846 error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1847 PCATCH, "mqrecv", timo);
1848 mq->mq_receivers--;
1849 if (error == EAGAIN)
1850 error = ETIMEDOUT;
1851 }
1852 if (*msg != NULL) {
1853 error = 0;
1854 TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1855 mq->mq_curmsgs--;
1856 mq->mq_totalbytes -= (*msg)->msg_size;
1857 if (mq->mq_senders)
1858 wakeup_one(&mq->mq_senders);
1859 if (mq->mq_flags & MQ_WSEL) {
1860 mq->mq_flags &= ~MQ_WSEL;
1861 selwakeup(&mq->mq_wsel);
1862 }
1863 KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1864 }
1865 if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1866 !TAILQ_EMPTY(&mq->mq_msgq)) {
1867 mqueue_send_notification(mq);
1868 }
1869 mtx_unlock(&mq->mq_mutex);
1870 return (error);
1871 }
1872
1873 static __inline struct mqueue_notifier *
1874 notifier_alloc(void)
1875 {
1876 return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1877 }
1878
1879 static __inline void
1880 notifier_free(struct mqueue_notifier *p)
1881 {
1882 uma_zfree(mqnoti_zone, p);
1883 }
1884
1885 static struct mqueue_notifier *
1886 notifier_search(struct proc *p, int fd)
1887 {
1888 struct mqueue_notifier *nt;
1889
1890 LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1891 if (nt->nt_ksi.ksi_mqd == fd)
1892 break;
1893 }
1894 return (nt);
1895 }
1896
1897 static __inline void
1898 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1899 {
1900 LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1901 }
1902
1903 static __inline void
1904 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1905 {
1906 LIST_REMOVE(nt, nt_link);
1907 notifier_free(nt);
1908 }
1909
1910 static void
1911 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1912 {
1913 struct mqueue_notifier *nt;
1914
1915 mtx_assert(&mq->mq_mutex, MA_OWNED);
1916 PROC_LOCK(p);
1917 nt = notifier_search(p, fd);
1918 if (nt != NULL) {
1919 if (mq->mq_notifier == nt)
1920 mq->mq_notifier = NULL;
1921 sigqueue_take(&nt->nt_ksi);
1922 notifier_delete(p, nt);
1923 }
1924 PROC_UNLOCK(p);
1925 }
1926
1927 static int
1928 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
1929 const struct mq_attr *attr)
1930 {
1931 char path[MQFS_NAMELEN + 1];
1932 struct mqfs_node *pn;
1933 struct filedesc *fdp;
1934 struct file *fp;
1935 struct mqueue *mq;
1936 int fd, error, len, cmode;
1937
1938 fdp = td->td_proc->p_fd;
1939 cmode = (((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1940 mq = NULL;
1941 if ((flags & O_CREAT) != 0 && attr != NULL) {
1942 if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
1943 return (EINVAL);
1944 if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
1945 return (EINVAL);
1946 }
1947
1948 error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
1949 if (error)
1950 return (error);
1951
1952 /*
1953 * The first character of name must be a slash (/) character
1954 * and the remaining characters of name cannot include any slash
1955 * characters.
1956 */
1957 len = strlen(path);
1958 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL)
1959 return (EINVAL);
1960
1961 error = falloc(td, &fp, &fd);
1962 if (error)
1963 return (error);
1964
1965 sx_xlock(&mqfs_data.mi_lock);
1966 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1967 if (pn == NULL) {
1968 if (!(flags & O_CREAT)) {
1969 error = ENOENT;
1970 } else {
1971 mq = mqueue_alloc(attr);
1972 if (mq == NULL) {
1973 error = ENFILE;
1974 } else {
1975 pn = mqfs_create_file(mqfs_data.mi_root,
1976 path + 1, len - 1, td->td_ucred,
1977 cmode);
1978 if (pn == NULL) {
1979 error = ENOSPC;
1980 mqueue_free(mq);
1981 }
1982 }
1983 }
1984
1985 if (error == 0) {
1986 pn->mn_data = mq;
1987 }
1988 } else {
1989 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1990 error = EEXIST;
1991 } else {
1992 accmode_t accmode = 0;
1993
1994 if (flags & FREAD)
1995 accmode |= VREAD;
1996 if (flags & FWRITE)
1997 accmode |= VWRITE;
1998 error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1999 pn->mn_gid, accmode, td->td_ucred, NULL);
2000 }
2001 }
2002
2003 if (error) {
2004 sx_xunlock(&mqfs_data.mi_lock);
2005 fdclose(fdp, fp, fd, td);
2006 fdrop(fp, td);
2007 return (error);
2008 }
2009
2010 mqnode_addref(pn);
2011 sx_xunlock(&mqfs_data.mi_lock);
2012
2013 finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2014 &mqueueops);
2015
2016 FILEDESC_XLOCK(fdp);
2017 if (fdp->fd_ofiles[fd] == fp)
2018 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2019 FILEDESC_XUNLOCK(fdp);
2020 td->td_retval[0] = fd;
2021 fdrop(fp, td);
2022 return (0);
2023 }
2024
2025 /*
2026 * Syscall to open a message queue.
2027 */
2028 int
2029 kmq_open(struct thread *td, struct kmq_open_args *uap)
2030 {
2031 struct mq_attr attr;
2032 int flags, error;
2033
2034 if ((uap->flags & O_ACCMODE) == O_ACCMODE)
2035 return (EINVAL);
2036 flags = FFLAGS(uap->flags);
2037 if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2038 error = copyin(uap->attr, &attr, sizeof(attr));
2039 if (error)
2040 return (error);
2041 }
2042 return (kern_kmq_open(td, uap->path, flags, uap->mode,
2043 uap->attr != NULL ? &attr : NULL));
2044 }
2045
2046 /*
2047 * Syscall to unlink a message queue.
2048 */
2049 int
2050 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2051 {
2052 char path[MQFS_NAMELEN+1];
2053 struct mqfs_node *pn;
2054 int error, len;
2055
2056 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2057 if (error)
2058 return (error);
2059
2060 len = strlen(path);
2061 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL)
2062 return (EINVAL);
2063
2064 sx_xlock(&mqfs_data.mi_lock);
2065 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2066 if (pn != NULL)
2067 error = do_unlink(pn, td->td_ucred);
2068 else
2069 error = ENOENT;
2070 sx_xunlock(&mqfs_data.mi_lock);
2071 return (error);
2072 }
2073
2074 typedef int (*_fgetf)(struct thread *, int, struct file **);
2075
2076 /*
2077 * Get message queue by giving file slot
2078 */
2079 static int
2080 _getmq(struct thread *td, int fd, _fgetf func,
2081 struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2082 {
2083 struct mqfs_node *pn;
2084 int error;
2085
2086 error = func(td, fd, fpp);
2087 if (error)
2088 return (error);
2089 if (&mqueueops != (*fpp)->f_ops) {
2090 fdrop(*fpp, td);
2091 return (EBADF);
2092 }
2093 pn = (*fpp)->f_data;
2094 if (ppn)
2095 *ppn = pn;
2096 if (pmq)
2097 *pmq = pn->mn_data;
2098 return (0);
2099 }
2100
2101 static __inline int
2102 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2103 struct mqueue **pmq)
2104 {
2105 return _getmq(td, fd, fget, fpp, ppn, pmq);
2106 }
2107
2108 static __inline int
2109 getmq_read(struct thread *td, int fd, struct file **fpp,
2110 struct mqfs_node **ppn, struct mqueue **pmq)
2111 {
2112 return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2113 }
2114
2115 static __inline int
2116 getmq_write(struct thread *td, int fd, struct file **fpp,
2117 struct mqfs_node **ppn, struct mqueue **pmq)
2118 {
2119 return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2120 }
2121
2122 static int
2123 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2124 struct mq_attr *oattr)
2125 {
2126 struct mqueue *mq;
2127 struct file *fp;
2128 u_int oflag, flag;
2129 int error;
2130
2131 if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2132 return (EINVAL);
2133 error = getmq(td, mqd, &fp, NULL, &mq);
2134 if (error)
2135 return (error);
2136 oattr->mq_maxmsg = mq->mq_maxmsg;
2137 oattr->mq_msgsize = mq->mq_msgsize;
2138 oattr->mq_curmsgs = mq->mq_curmsgs;
2139 if (attr != NULL) {
2140 do {
2141 oflag = flag = fp->f_flag;
2142 flag &= ~O_NONBLOCK;
2143 flag |= (attr->mq_flags & O_NONBLOCK);
2144 } while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2145 } else
2146 oflag = fp->f_flag;
2147 oattr->mq_flags = (O_NONBLOCK & oflag);
2148 fdrop(fp, td);
2149 return (error);
2150 }
2151
2152 int
2153 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2154 {
2155 struct mq_attr attr, oattr;
2156 int error;
2157
2158 if (uap->attr != NULL) {
2159 error = copyin(uap->attr, &attr, sizeof(attr));
2160 if (error != 0)
2161 return (error);
2162 }
2163 error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2164 &oattr);
2165 if (error != 0)
2166 return (error);
2167 if (uap->oattr != NULL)
2168 error = copyout(&oattr, uap->oattr, sizeof(oattr));
2169 return (error);
2170 }
2171
2172 int
2173 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2174 {
2175 struct mqueue *mq;
2176 struct file *fp;
2177 struct timespec *abs_timeout, ets;
2178 int error;
2179 int waitok;
2180
2181 error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2182 if (error)
2183 return (error);
2184 if (uap->abs_timeout != NULL) {
2185 error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2186 if (error != 0)
2187 return (error);
2188 abs_timeout = &ets;
2189 } else
2190 abs_timeout = NULL;
2191 waitok = !(fp->f_flag & O_NONBLOCK);
2192 error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2193 uap->msg_prio, waitok, abs_timeout);
2194 fdrop(fp, td);
2195 return (error);
2196 }
2197
2198 int
2199 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2200 {
2201 struct mqueue *mq;
2202 struct file *fp;
2203 struct timespec *abs_timeout, ets;
2204 int error, waitok;
2205
2206 error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2207 if (error)
2208 return (error);
2209 if (uap->abs_timeout != NULL) {
2210 error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2211 if (error != 0)
2212 return (error);
2213 abs_timeout = &ets;
2214 } else
2215 abs_timeout = NULL;
2216 waitok = !(fp->f_flag & O_NONBLOCK);
2217 error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2218 uap->msg_prio, waitok, abs_timeout);
2219 fdrop(fp, td);
2220 return (error);
2221 }
2222
2223 int
2224 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2225 {
2226 struct sigevent ev;
2227 struct filedesc *fdp;
2228 struct proc *p;
2229 struct mqueue *mq;
2230 struct file *fp;
2231 struct mqueue_notifier *nt, *newnt = NULL;
2232 int error;
2233
2234 p = td->td_proc;
2235 fdp = td->td_proc->p_fd;
2236 if (uap->sigev) {
2237 error = copyin(uap->sigev, &ev, sizeof(ev));
2238 if (error)
2239 return (error);
2240 if (ev.sigev_notify != SIGEV_SIGNAL &&
2241 ev.sigev_notify != SIGEV_THREAD_ID &&
2242 ev.sigev_notify != SIGEV_NONE)
2243 return (EINVAL);
2244 if ((ev.sigev_notify == SIGEV_SIGNAL ||
2245 ev.sigev_notify == SIGEV_THREAD_ID) &&
2246 !_SIG_VALID(ev.sigev_signo))
2247 return (EINVAL);
2248 }
2249 error = getmq(td, uap->mqd, &fp, NULL, &mq);
2250 if (error)
2251 return (error);
2252 again:
2253 FILEDESC_SLOCK(fdp);
2254 if (fget_locked(fdp, uap->mqd) != fp) {
2255 FILEDESC_SUNLOCK(fdp);
2256 error = EBADF;
2257 goto out;
2258 }
2259 mtx_lock(&mq->mq_mutex);
2260 FILEDESC_SUNLOCK(fdp);
2261 if (uap->sigev != NULL) {
2262 if (mq->mq_notifier != NULL) {
2263 error = EBUSY;
2264 } else {
2265 PROC_LOCK(p);
2266 nt = notifier_search(p, uap->mqd);
2267 if (nt == NULL) {
2268 if (newnt == NULL) {
2269 PROC_UNLOCK(p);
2270 mtx_unlock(&mq->mq_mutex);
2271 newnt = notifier_alloc();
2272 goto again;
2273 }
2274 }
2275
2276 if (nt != NULL) {
2277 sigqueue_take(&nt->nt_ksi);
2278 if (newnt != NULL) {
2279 notifier_free(newnt);
2280 newnt = NULL;
2281 }
2282 } else {
2283 nt = newnt;
2284 newnt = NULL;
2285 ksiginfo_init(&nt->nt_ksi);
2286 nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2287 nt->nt_ksi.ksi_code = SI_MESGQ;
2288 nt->nt_proc = p;
2289 nt->nt_ksi.ksi_mqd = uap->mqd;
2290 notifier_insert(p, nt);
2291 }
2292 nt->nt_sigev = ev;
2293 mq->mq_notifier = nt;
2294 PROC_UNLOCK(p);
2295 /*
2296 * if there is no receivers and message queue
2297 * is not empty, we should send notification
2298 * as soon as possible.
2299 */
2300 if (mq->mq_receivers == 0 &&
2301 !TAILQ_EMPTY(&mq->mq_msgq))
2302 mqueue_send_notification(mq);
2303 }
2304 } else {
2305 notifier_remove(p, mq, uap->mqd);
2306 }
2307 mtx_unlock(&mq->mq_mutex);
2308
2309 out:
2310 fdrop(fp, td);
2311 if (newnt != NULL)
2312 notifier_free(newnt);
2313 return (error);
2314 }
2315
2316 static void
2317 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2318 {
2319 struct filedesc *fdp;
2320 struct mqueue *mq;
2321
2322 fdp = td->td_proc->p_fd;
2323 FILEDESC_LOCK_ASSERT(fdp);
2324
2325 if (fp->f_ops == &mqueueops) {
2326 mq = FPTOMQ(fp);
2327 mtx_lock(&mq->mq_mutex);
2328 notifier_remove(td->td_proc, mq, fd);
2329
2330 /* have to wakeup thread in same process */
2331 if (mq->mq_flags & MQ_RSEL) {
2332 mq->mq_flags &= ~MQ_RSEL;
2333 selwakeup(&mq->mq_rsel);
2334 }
2335 if (mq->mq_flags & MQ_WSEL) {
2336 mq->mq_flags &= ~MQ_WSEL;
2337 selwakeup(&mq->mq_wsel);
2338 }
2339 mtx_unlock(&mq->mq_mutex);
2340 }
2341 }
2342
2343 static void
2344 mq_proc_exit(void *arg __unused, struct proc *p)
2345 {
2346 struct filedesc *fdp;
2347 struct file *fp;
2348 struct mqueue *mq;
2349 int i;
2350
2351 fdp = p->p_fd;
2352 FILEDESC_SLOCK(fdp);
2353 for (i = 0; i < fdp->fd_nfiles; ++i) {
2354 fp = fget_locked(fdp, i);
2355 if (fp != NULL && fp->f_ops == &mqueueops) {
2356 mq = FPTOMQ(fp);
2357 mtx_lock(&mq->mq_mutex);
2358 notifier_remove(p, FPTOMQ(fp), i);
2359 mtx_unlock(&mq->mq_mutex);
2360 }
2361 }
2362 FILEDESC_SUNLOCK(fdp);
2363 KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2364 }
2365
2366 static int
2367 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2368 int flags, struct thread *td)
2369 {
2370 return (EOPNOTSUPP);
2371 }
2372
2373 static int
2374 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2375 int flags, struct thread *td)
2376 {
2377 return (EOPNOTSUPP);
2378 }
2379
2380 static int
2381 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred,
2382 struct thread *td)
2383 {
2384
2385 return (EINVAL);
2386 }
2387
2388 static int
2389 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2390 struct ucred *active_cred, struct thread *td)
2391 {
2392 return (ENOTTY);
2393 }
2394
2395 static int
2396 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2397 struct thread *td)
2398 {
2399 struct mqueue *mq = FPTOMQ(fp);
2400 int revents = 0;
2401
2402 mtx_lock(&mq->mq_mutex);
2403 if (events & (POLLIN | POLLRDNORM)) {
2404 if (mq->mq_curmsgs) {
2405 revents |= events & (POLLIN | POLLRDNORM);
2406 } else {
2407 mq->mq_flags |= MQ_RSEL;
2408 selrecord(td, &mq->mq_rsel);
2409 }
2410 }
2411 if (events & POLLOUT) {
2412 if (mq->mq_curmsgs < mq->mq_maxmsg)
2413 revents |= POLLOUT;
2414 else {
2415 mq->mq_flags |= MQ_WSEL;
2416 selrecord(td, &mq->mq_wsel);
2417 }
2418 }
2419 mtx_unlock(&mq->mq_mutex);
2420 return (revents);
2421 }
2422
2423 static int
2424 mqf_close(struct file *fp, struct thread *td)
2425 {
2426 struct mqfs_node *pn;
2427
2428 fp->f_ops = &badfileops;
2429 pn = fp->f_data;
2430 fp->f_data = NULL;
2431 sx_xlock(&mqfs_data.mi_lock);
2432 mqnode_release(pn);
2433 sx_xunlock(&mqfs_data.mi_lock);
2434 return (0);
2435 }
2436
2437 static int
2438 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2439 struct thread *td)
2440 {
2441 struct mqfs_node *pn = fp->f_data;
2442
2443 bzero(st, sizeof *st);
2444 st->st_atimespec = pn->mn_atime;
2445 st->st_mtimespec = pn->mn_mtime;
2446 st->st_ctimespec = pn->mn_ctime;
2447 st->st_birthtimespec = pn->mn_birth;
2448 st->st_uid = pn->mn_uid;
2449 st->st_gid = pn->mn_gid;
2450 st->st_mode = S_IFIFO | pn->mn_mode;
2451 return (0);
2452 }
2453
2454 static int
2455 mqf_kqfilter(struct file *fp, struct knote *kn)
2456 {
2457 struct mqueue *mq = FPTOMQ(fp);
2458 int error = 0;
2459
2460 if (kn->kn_filter == EVFILT_READ) {
2461 kn->kn_fop = &mq_rfiltops;
2462 knlist_add(&mq->mq_rsel.si_note, kn, 0);
2463 } else if (kn->kn_filter == EVFILT_WRITE) {
2464 kn->kn_fop = &mq_wfiltops;
2465 knlist_add(&mq->mq_wsel.si_note, kn, 0);
2466 } else
2467 error = EINVAL;
2468 return (error);
2469 }
2470
2471 static void
2472 filt_mqdetach(struct knote *kn)
2473 {
2474 struct mqueue *mq = FPTOMQ(kn->kn_fp);
2475
2476 if (kn->kn_filter == EVFILT_READ)
2477 knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2478 else if (kn->kn_filter == EVFILT_WRITE)
2479 knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2480 else
2481 panic("filt_mqdetach");
2482 }
2483
2484 static int
2485 filt_mqread(struct knote *kn, long hint)
2486 {
2487 struct mqueue *mq = FPTOMQ(kn->kn_fp);
2488
2489 mtx_assert(&mq->mq_mutex, MA_OWNED);
2490 return (mq->mq_curmsgs != 0);
2491 }
2492
2493 static int
2494 filt_mqwrite(struct knote *kn, long hint)
2495 {
2496 struct mqueue *mq = FPTOMQ(kn->kn_fp);
2497
2498 mtx_assert(&mq->mq_mutex, MA_OWNED);
2499 return (mq->mq_curmsgs < mq->mq_maxmsg);
2500 }
2501
2502 static struct fileops mqueueops = {
2503 .fo_read = mqf_read,
2504 .fo_write = mqf_write,
2505 .fo_truncate = mqf_truncate,
2506 .fo_ioctl = mqf_ioctl,
2507 .fo_poll = mqf_poll,
2508 .fo_kqfilter = mqf_kqfilter,
2509 .fo_stat = mqf_stat,
2510 .fo_close = mqf_close
2511 };
2512
2513 static struct vop_vector mqfs_vnodeops = {
2514 .vop_default = &default_vnodeops,
2515 .vop_access = mqfs_access,
2516 .vop_cachedlookup = mqfs_lookup,
2517 .vop_lookup = vfs_cache_lookup,
2518 .vop_reclaim = mqfs_reclaim,
2519 .vop_create = mqfs_create,
2520 .vop_remove = mqfs_remove,
2521 .vop_inactive = mqfs_inactive,
2522 .vop_open = mqfs_open,
2523 .vop_close = mqfs_close,
2524 .vop_getattr = mqfs_getattr,
2525 .vop_setattr = mqfs_setattr,
2526 .vop_read = mqfs_read,
2527 .vop_write = VOP_EOPNOTSUPP,
2528 .vop_readdir = mqfs_readdir,
2529 .vop_mkdir = VOP_EOPNOTSUPP,
2530 .vop_rmdir = VOP_EOPNOTSUPP
2531 };
2532
2533 static struct vfsops mqfs_vfsops = {
2534 .vfs_init = mqfs_init,
2535 .vfs_uninit = mqfs_uninit,
2536 .vfs_mount = mqfs_mount,
2537 .vfs_unmount = mqfs_unmount,
2538 .vfs_root = mqfs_root,
2539 .vfs_statfs = mqfs_statfs,
2540 };
2541
2542 static struct vfsconf mqueuefs_vfsconf = {
2543 .vfc_version = VFS_VERSION,
2544 .vfc_name = "mqueuefs",
2545 .vfc_vfsops = &mqfs_vfsops,
2546 .vfc_typenum = -1,
2547 .vfc_flags = VFCF_SYNTHETIC
2548 };
2549
2550 static struct syscall_helper_data mq_syscalls[] = {
2551 SYSCALL_INIT_HELPER(kmq_open),
2552 SYSCALL_INIT_HELPER(kmq_setattr),
2553 SYSCALL_INIT_HELPER(kmq_timedsend),
2554 SYSCALL_INIT_HELPER(kmq_timedreceive),
2555 SYSCALL_INIT_HELPER(kmq_notify),
2556 SYSCALL_INIT_HELPER(kmq_unlink),
2557 SYSCALL_INIT_LAST
2558 };
2559
2560 #ifdef COMPAT_FREEBSD32
2561 #include <compat/freebsd32/freebsd32.h>
2562 #include <compat/freebsd32/freebsd32_proto.h>
2563 #include <compat/freebsd32/freebsd32_syscall.h>
2564 #include <compat/freebsd32/freebsd32_util.h>
2565
2566 static void
2567 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2568 {
2569
2570 to->mq_flags = from->mq_flags;
2571 to->mq_maxmsg = from->mq_maxmsg;
2572 to->mq_msgsize = from->mq_msgsize;
2573 to->mq_curmsgs = from->mq_curmsgs;
2574 }
2575
2576 static void
2577 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2578 {
2579
2580 to->mq_flags = from->mq_flags;
2581 to->mq_maxmsg = from->mq_maxmsg;
2582 to->mq_msgsize = from->mq_msgsize;
2583 to->mq_curmsgs = from->mq_curmsgs;
2584 }
2585
2586 int
2587 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2588 {
2589 struct mq_attr attr;
2590 struct mq_attr32 attr32;
2591 int flags, error;
2592
2593 if ((uap->flags & O_ACCMODE) == O_ACCMODE)
2594 return (EINVAL);
2595 flags = FFLAGS(uap->flags);
2596 if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2597 error = copyin(uap->attr, &attr32, sizeof(attr32));
2598 if (error)
2599 return (error);
2600 mq_attr_from32(&attr32, &attr);
2601 }
2602 return (kern_kmq_open(td, uap->path, flags, uap->mode,
2603 uap->attr != NULL ? &attr : NULL));
2604 }
2605
2606 int
2607 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2608 {
2609 struct mq_attr attr, oattr;
2610 struct mq_attr32 attr32, oattr32;
2611 int error;
2612
2613 if (uap->attr != NULL) {
2614 error = copyin(uap->attr, &attr32, sizeof(attr32));
2615 if (error != 0)
2616 return (error);
2617 mq_attr_from32(&attr32, &attr);
2618 }
2619 error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2620 &oattr);
2621 if (error != 0)
2622 return (error);
2623 if (uap->oattr != NULL) {
2624 mq_attr_to32(&oattr, &oattr32);
2625 error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2626 }
2627 return (error);
2628 }
2629
2630 int
2631 freebsd32_kmq_timedsend(struct thread *td,
2632 struct freebsd32_kmq_timedsend_args *uap)
2633 {
2634 struct mqueue *mq;
2635 struct file *fp;
2636 struct timespec32 ets32;
2637 struct timespec *abs_timeout, ets;
2638 int error;
2639 int waitok;
2640
2641 error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2642 if (error)
2643 return (error);
2644 if (uap->abs_timeout != NULL) {
2645 error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2646 if (error != 0)
2647 return (error);
2648 CP(ets32, ets, tv_sec);
2649 CP(ets32, ets, tv_nsec);
2650 abs_timeout = &ets;
2651 } else
2652 abs_timeout = NULL;
2653 waitok = !(fp->f_flag & O_NONBLOCK);
2654 error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2655 uap->msg_prio, waitok, abs_timeout);
2656 fdrop(fp, td);
2657 return (error);
2658 }
2659
2660 int
2661 freebsd32_kmq_timedreceive(struct thread *td,
2662 struct freebsd32_kmq_timedreceive_args *uap)
2663 {
2664 struct mqueue *mq;
2665 struct file *fp;
2666 struct timespec32 ets32;
2667 struct timespec *abs_timeout, ets;
2668 int error, waitok;
2669
2670 error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2671 if (error)
2672 return (error);
2673 if (uap->abs_timeout != NULL) {
2674 error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2675 if (error != 0)
2676 return (error);
2677 CP(ets32, ets, tv_sec);
2678 CP(ets32, ets, tv_nsec);
2679 abs_timeout = &ets;
2680 } else
2681 abs_timeout = NULL;
2682 waitok = !(fp->f_flag & O_NONBLOCK);
2683 error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2684 uap->msg_prio, waitok, abs_timeout);
2685 fdrop(fp, td);
2686 return (error);
2687 }
2688
2689 static struct syscall_helper_data mq32_syscalls[] = {
2690 SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2691 SYSCALL32_INIT_HELPER(freebsd32_kmq_setattr),
2692 SYSCALL32_INIT_HELPER(freebsd32_kmq_timedsend),
2693 SYSCALL32_INIT_HELPER(freebsd32_kmq_timedreceive),
2694 SYSCALL32_INIT_HELPER(kmq_notify),
2695 SYSCALL32_INIT_HELPER(kmq_unlink),
2696 SYSCALL_INIT_LAST
2697 };
2698 #endif
2699
2700 static int
2701 mqinit(void)
2702 {
2703 int error;
2704
2705 error = syscall_helper_register(mq_syscalls);
2706 if (error != 0)
2707 return (error);
2708 #ifdef COMPAT_FREEBSD32
2709 error = syscall32_helper_register(mq32_syscalls);
2710 if (error != 0)
2711 return (error);
2712 #endif
2713 return (0);
2714 }
2715
2716 static int
2717 mqunload(void)
2718 {
2719
2720 #ifdef COMPAT_FREEBSD32
2721 syscall32_helper_unregister(mq32_syscalls);
2722 #endif
2723 syscall_helper_unregister(mq_syscalls);
2724 return (0);
2725 }
2726
2727 static int
2728 mq_modload(struct module *module, int cmd, void *arg)
2729 {
2730 int error = 0;
2731
2732 error = vfs_modevent(module, cmd, arg);
2733 if (error != 0)
2734 return (error);
2735
2736 switch (cmd) {
2737 case MOD_LOAD:
2738 error = mqinit();
2739 if (error != 0)
2740 mqunload();
2741 break;
2742 case MOD_UNLOAD:
2743 error = mqunload();
2744 break;
2745 default:
2746 break;
2747 }
2748 return (error);
2749 }
2750
2751 static moduledata_t mqueuefs_mod = {
2752 "mqueuefs",
2753 mq_modload,
2754 &mqueuefs_vfsconf
2755 };
2756 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2757 MODULE_VERSION(mqueuefs, 1);
Cache object: 0be3e6b55863ee6ee8485244f005ad0f
|