2 * Copyright (c) 2005 David Xu <davidxu@freebsd.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 * POSIX message queue implementation.
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.
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.
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/limits.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/fcntl.h>
58 #include <sys/filedesc.h>
60 #include <sys/malloc.h>
61 #include <sys/module.h>
62 #include <sys/mount.h>
63 #include <sys/mqueue.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/posix4.h>
70 #include <sys/queue.h>
71 #include <sys/sysproto.h>
73 #include <sys/syscall.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
77 #include <sys/sysctl.h>
78 #include <sys/taskqueue.h>
79 #include <sys/unistd.h>
80 #include <sys/vnode.h>
81 #include <machine/atomic.h>
84 * Limits and constants
86 #define MQFS_NAMELEN NAME_MAX
87 #define MQFS_DELEN (8 + MQFS_NAMELEN)
103 * mqfs_info: describes a mqfs instance
107 struct mqfs_node *mi_root;
108 struct unrhdr *mi_unrhdr;
112 LIST_ENTRY(mqfs_vdata) mv_link;
113 struct mqfs_node *mv_node;
114 struct vnode *mv_vnode;
119 * mqfs_node: describes a node (file or directory) within a mqfs
122 char mn_name[MQFS_NAMELEN+1];
123 struct mqfs_info *mn_info;
124 struct mqfs_node *mn_parent;
125 LIST_HEAD(,mqfs_node) mn_children;
126 LIST_ENTRY(mqfs_node) mn_sibling;
127 LIST_HEAD(,mqfs_vdata) mn_vnodes;
133 struct timespec mn_birth;
134 struct timespec mn_ctime;
135 struct timespec mn_atime;
136 struct timespec mn_mtime;
142 #define VTON(vp) (((struct mqfs_vdata *)((vp)->v_data))->mv_node)
143 #define VTOMQ(vp) ((struct mqueue *)(VTON(vp)->mn_data))
144 #define VFSTOMQFS(m) ((struct mqfs_info *)((m)->mnt_data))
145 #define FPTOMQ(fp) ((struct mqueue *)(((struct mqfs_node *) \
146 (fp)->f_data)->mn_data))
148 TAILQ_HEAD(msgq, mqueue_msg);
152 struct mqueue_notifier {
153 LIST_ENTRY(mqueue_notifier) nt_link;
154 struct sigevent nt_sigev;
156 struct proc *nt_proc;
169 struct selinfo mq_rsel;
170 struct selinfo mq_wsel;
171 struct mqueue_notifier *mq_notifier;
178 TAILQ_ENTRY(mqueue_msg) msg_link;
179 unsigned int msg_prio;
180 unsigned int msg_size;
181 /* following real data... */
184 SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
185 "POSIX real time message queue");
187 static int default_maxmsg = 10;
188 static int default_msgsize = 1024;
190 static int maxmsg = 100;
191 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
192 &maxmsg, 0, "Default maximum messages in queue");
193 static int maxmsgsize = 16384;
194 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
195 &maxmsgsize, 0, "Default maximum message size");
196 static int maxmq = 100;
197 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
198 &maxmq, 0, "maximum message queues");
199 static int curmq = 0;
200 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
201 &curmq, 0, "current message queue number");
202 static int unloadable = 0;
203 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
205 static eventhandler_tag exit_tag;
207 /* Only one instance per-system */
208 static struct mqfs_info mqfs_data;
209 static uma_zone_t mqnode_zone;
210 static uma_zone_t mqueue_zone;
211 static uma_zone_t mvdata_zone;
212 static uma_zone_t mqnoti_zone;
213 static struct vop_vector mqfs_vnodeops;
214 static struct fileops mqueueops;
217 * Directory structure construction and manipulation
220 static struct mqfs_node *mqfs_create_dir(struct mqfs_node *parent,
221 const char *name, int namelen, struct ucred *cred, int mode);
222 static struct mqfs_node *mqfs_create_link(struct mqfs_node *parent,
223 const char *name, int namelen, struct ucred *cred, int mode);
226 static struct mqfs_node *mqfs_create_file(struct mqfs_node *parent,
227 const char *name, int namelen, struct ucred *cred, int mode);
228 static int mqfs_destroy(struct mqfs_node *mn);
229 static void mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
230 static void mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
231 static int mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
234 * Message queue construction and maniplation
236 static struct mqueue *mqueue_alloc(const struct mq_attr *attr);
237 static void mqueue_free(struct mqueue *mq);
238 static int mqueue_send(struct mqueue *mq, const char *msg_ptr,
239 size_t msg_len, unsigned msg_prio, int waitok,
240 const struct timespec *abs_timeout);
241 static int mqueue_receive(struct mqueue *mq, char *msg_ptr,
242 size_t msg_len, unsigned *msg_prio, int waitok,
243 const struct timespec *abs_timeout);
244 static int _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
246 static int _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
248 static void mqueue_send_notification(struct mqueue *mq);
249 static void mqueue_fdclose(struct thread *td, int fd, struct file *fp);
250 static void mq_proc_exit(void *arg, struct proc *p);
255 static void filt_mqdetach(struct knote *kn);
256 static int filt_mqread(struct knote *kn, long hint);
257 static int filt_mqwrite(struct knote *kn, long hint);
259 struct filterops mq_rfiltops =
260 { 1, NULL, filt_mqdetach, filt_mqread };
261 struct filterops mq_wfiltops =
262 { 1, NULL, filt_mqdetach, filt_mqwrite };
265 * Initialize fileno bitmap
268 mqfs_fileno_init(struct mqfs_info *mi)
272 up = new_unrhdr(1, INT_MAX, NULL);
277 * Tear down fileno bitmap
280 mqfs_fileno_uninit(struct mqfs_info *mi)
285 mi->mi_unrhdr = NULL;
290 * Allocate a file number
293 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
295 /* make sure our parent has a file number */
296 if (mn->mn_parent && !mn->mn_parent->mn_fileno)
297 mqfs_fileno_alloc(mi, mn->mn_parent);
299 switch (mn->mn_type) {
303 case mqfstype_symlink:
304 mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
307 KASSERT(mn->mn_parent != NULL,
308 ("mqfstype_this node has no parent"));
309 mn->mn_fileno = mn->mn_parent->mn_fileno;
311 case mqfstype_parent:
312 KASSERT(mn->mn_parent != NULL,
313 ("mqfstype_parent node has no parent"));
314 if (mn->mn_parent == mi->mi_root) {
315 mn->mn_fileno = mn->mn_parent->mn_fileno;
318 KASSERT(mn->mn_parent->mn_parent != NULL,
319 ("mqfstype_parent node has no grandparent"));
320 mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
324 ("mqfs_fileno_alloc() called for unknown type node: %d",
331 * Release a file number
334 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
336 switch (mn->mn_type) {
340 case mqfstype_symlink:
341 free_unr(mi->mi_unrhdr, mn->mn_fileno);
344 case mqfstype_parent:
345 /* ignore these, as they don't "own" their file number */
349 ("mqfs_fileno_free() called for unknown type node: %d",
355 static __inline struct mqfs_node *
358 return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
362 mqnode_free(struct mqfs_node *node)
364 uma_zfree(mqnode_zone, node);
368 mqnode_addref(struct mqfs_node *node)
370 atomic_fetchadd_int(&node->mn_refcount, 1);
374 mqnode_release(struct mqfs_node *node)
378 old = atomic_fetchadd_int(&node->mn_refcount, -1);
379 if (node->mn_type == mqfstype_dir ||
380 node->mn_type == mqfstype_root)
381 exp = 3; /* include . and .. */
389 * Add a node to a directory
392 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
394 KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
395 KASSERT(parent->mn_info != NULL,
396 ("%s(): parent has no mn_info", __func__));
397 KASSERT(parent->mn_type == mqfstype_dir ||
398 parent->mn_type == mqfstype_root,
399 ("%s(): parent is not a directory", __func__));
401 node->mn_info = parent->mn_info;
402 node->mn_parent = parent;
403 LIST_INIT(&node->mn_children);
404 LIST_INIT(&node->mn_vnodes);
405 LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
406 mqnode_addref(parent);
410 static struct mqfs_node *
411 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
414 struct mqfs_node *node;
416 node = mqnode_alloc();
417 strncpy(node->mn_name, name, namelen);
418 node->mn_type = nodetype;
419 node->mn_refcount = 1;
420 getnanotime(&node->mn_birth);
421 node->mn_ctime = node->mn_atime = node->mn_mtime
423 node->mn_uid = cred->cr_uid;
424 node->mn_gid = cred->cr_gid;
425 node->mn_mode = mode;
432 static struct mqfs_node *
433 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
434 struct ucred *cred, int mode)
436 struct mqfs_node *node;
438 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
439 if (mqfs_add_node(parent, node) != 0) {
447 * Add . and .. to a directory
450 mqfs_fixup_dir(struct mqfs_node *parent)
452 struct mqfs_node *dir;
454 dir = mqnode_alloc();
455 dir->mn_name[0] = '.';
456 dir->mn_type = mqfstype_this;
457 dir->mn_refcount = 1;
458 if (mqfs_add_node(parent, dir) != 0) {
463 dir = mqnode_alloc();
464 dir->mn_name[0] = dir->mn_name[1] = '.';
465 dir->mn_type = mqfstype_parent;
466 dir->mn_refcount = 1;
468 if (mqfs_add_node(parent, dir) != 0) {
481 static struct mqfs_node *
482 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
483 struct ucred *cred, int mode)
485 struct mqfs_node *node;
487 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
488 if (mqfs_add_node(parent, node) != 0) {
493 if (mqfs_fixup_dir(node) != 0) {
503 static struct mqfs_node *
504 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
505 struct ucred *cred, int mode)
507 struct mqfs_node *node;
509 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
510 if (mqfs_add_node(parent, node) != 0) {
520 * Destroy a node or a tree of nodes
523 mqfs_destroy(struct mqfs_node *node)
525 struct mqfs_node *parent;
527 KASSERT(node != NULL,
528 ("%s(): node is NULL", __func__));
529 KASSERT(node->mn_info != NULL,
530 ("%s(): node has no mn_info", __func__));
532 /* destroy children */
533 if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
534 while (! LIST_EMPTY(&node->mn_children))
535 mqfs_destroy(LIST_FIRST(&node->mn_children));
537 /* unlink from parent */
538 if ((parent = node->mn_parent) != NULL) {
539 KASSERT(parent->mn_info == node->mn_info,
540 ("%s(): parent has different mn_info", __func__));
541 LIST_REMOVE(node, mn_sibling);
544 if (node->mn_fileno != 0)
545 mqfs_fileno_free(node->mn_info, node);
546 if (node->mn_data != NULL)
547 mqueue_free(node->mn_data);
553 * Mount a mqfs instance
556 mqfs_mount(struct mount *mp, struct thread *td)
560 if (mp->mnt_flag & MNT_UPDATE)
563 mp->mnt_data = &mqfs_data;
565 mp->mnt_flag |= MNT_LOCAL;
566 mp->mnt_kern_flag |= MNTK_MPSAFE;
571 vfs_mountedfrom(mp, "mqueue");
572 sbp->f_bsize = PAGE_SIZE;
573 sbp->f_iosize = PAGE_SIZE;
583 * Unmount a mqfs instance
586 mqfs_unmount(struct mount *mp, int mntflags, struct thread *td)
590 error = vflush(mp, 0, (mntflags & MNT_FORCE) ? FORCECLOSE : 0, td);
595 * Return a root vnode
598 mqfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
600 struct mqfs_info *mqfs;
603 mqfs = VFSTOMQFS(mp);
604 sx_xlock(&mqfs->mi_lock);
605 ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
606 sx_xunlock(&mqfs->mi_lock);
611 * Return filesystem stats
614 mqfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
616 /* XXX update statistics */
621 * Initialize a mqfs instance
624 mqfs_init(struct vfsconf *vfc)
626 struct mqfs_node *root;
627 struct mqfs_info *mi;
629 mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
630 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
631 mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
632 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
633 mvdata_zone = uma_zcreate("mvdata",
634 sizeof(struct mqfs_vdata), NULL, NULL, NULL,
635 NULL, UMA_ALIGN_PTR, 0);
636 mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
637 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
639 sx_init(&mi->mi_lock, "mqfs lock");
640 /* set up the root diretory */
641 root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
644 LIST_INIT(&root->mn_children);
645 LIST_INIT(&root->mn_vnodes);
647 mqfs_fileno_init(mi);
648 mqfs_fileno_alloc(mi, root);
649 mqfs_fixup_dir(root);
650 exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
651 EVENTHANDLER_PRI_ANY);
652 mq_fdclose = mqueue_fdclose;
653 p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
658 * Destroy a mqfs instance
661 mqfs_uninit(struct vfsconf *vfc)
663 struct mqfs_info *mi;
667 EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
669 mqfs_destroy(mi->mi_root);
671 mqfs_fileno_uninit(mi);
672 sx_destroy(&mi->mi_lock);
673 uma_zdestroy(mqnode_zone);
674 uma_zdestroy(mqueue_zone);
675 uma_zdestroy(mvdata_zone);
676 uma_zdestroy(mqnoti_zone);
684 do_recycle(void *context, int pending __unused)
686 struct vnode *vp = (struct vnode *)context;
688 vrecycle(vp, curthread);
696 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
698 struct mqfs_vdata *vd;
701 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
702 if (vd->mv_vnode->v_mount == mp)
707 if (vget(vd->mv_vnode, 0, curthread) == 0) {
709 vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE,
713 /* XXX if this can happen, we're in trouble */
716 error = getnewvnode("mqueue", mp, &mqfs_vnodeops, vpp);
719 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, curthread);
720 error = insmntque(*vpp, mp);
725 vd = uma_zalloc(mvdata_zone, M_WAITOK);
729 TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
730 LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
732 switch (pn->mn_type) {
734 (*vpp)->v_vflag = VV_ROOT;
738 case mqfstype_parent:
739 (*vpp)->v_type = VDIR;
742 (*vpp)->v_type = VREG;
744 case mqfstype_symlink:
745 (*vpp)->v_type = VLNK;
748 KASSERT(0, ("mqfs_allocf called for null node\n"));
750 panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
756 * Search a directory entry
758 static struct mqfs_node *
759 mqfs_search(struct mqfs_node *pd, const char *name, int len)
761 struct mqfs_node *pn;
763 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
764 if (strncmp(pn->mn_name, name, len) == 0)
771 * Look up a file or directory.
774 mqfs_lookupx(struct vop_cachedlookup_args *ap)
776 struct componentname *cnp;
777 struct vnode *dvp, **vpp;
778 struct mqfs_node *pd;
779 struct mqfs_node *pn;
780 int nameiop, flags, error, namelen;
787 pname = cnp->cn_nameptr;
788 namelen = cnp->cn_namelen;
790 flags = cnp->cn_flags;
791 nameiop = cnp->cn_nameiop;
796 if (dvp->v_type != VDIR)
799 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
803 /* shortcut: check if the name is too long */
804 if (cnp->cn_namelen >= MQFS_NAMELEN)
808 if (namelen == 1 && pname[0] == '.') {
809 if ((flags & ISLASTCN) && nameiop != LOOKUP)
818 if (cnp->cn_flags & ISDOTDOT) {
819 if (dvp->v_vflag & VV_ROOT)
821 if ((flags & ISLASTCN) && nameiop != LOOKUP)
823 VOP_UNLOCK(dvp, 0, cnp->cn_thread);
824 KASSERT(pd->mn_parent, ("non-root directory has no parent"));
826 error = mqfs_allocv(dvp->v_mount, vpp, pn);
827 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
832 pn = mqfs_search(pd, pname, namelen);
837 if (nameiop == DELETE && (flags & ISLASTCN)) {
838 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
849 error = mqfs_allocv(dvp->v_mount, vpp, pn);
850 if (error == 0 && cnp->cn_flags & MAKEENTRY)
851 cache_enter(dvp, *vpp, cnp);
857 /* will create a new entry in the directory ? */
858 if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
859 && (flags & ISLASTCN)) {
860 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
863 cnp->cn_flags |= SAVENAME;
864 return (EJUSTRETURN);
870 struct vop_lookup_args {
871 struct vop_generic_args a_gen;
873 struct vnode **a_vpp;
874 struct componentname *a_cnp;
879 * vnode lookup operation
882 mqfs_lookup(struct vop_cachedlookup_args *ap)
884 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
887 sx_xlock(&mqfs->mi_lock);
888 rc = mqfs_lookupx(ap);
889 sx_xunlock(&mqfs->mi_lock);
894 struct vop_create_args {
896 struct vnode **a_vpp;
897 struct componentname *a_cnp;
903 * vnode creation operation
906 mqfs_create(struct vop_create_args *ap)
908 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
909 struct componentname *cnp = ap->a_cnp;
910 struct mqfs_node *pd;
911 struct mqfs_node *pn;
915 pd = VTON(ap->a_dvp);
916 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
918 mq = mqueue_alloc(NULL);
921 sx_xlock(&mqfs->mi_lock);
924 pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
927 sx_xunlock(&mqfs->mi_lock);
931 if ((cnp->cn_flags & HASBUF) == 0)
932 panic("%s: no name", __func__);
934 pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
935 cnp->cn_cred, ap->a_vap->va_mode);
939 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
945 sx_xunlock(&mqfs->mi_lock);
955 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
957 struct mqfs_node *parent;
958 struct mqfs_vdata *vd;
961 sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
963 if (ucred->cr_uid != pn->mn_uid &&
964 (error = priv_check_cred(ucred, PRIV_MQ_ADMIN,
965 SUSER_ALLOWJAIL)) != 0)
967 else if (!pn->mn_deleted) {
968 parent = pn->mn_parent;
969 pn->mn_parent = NULL;
971 LIST_REMOVE(pn, mn_sibling);
972 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
973 cache_purge(vd->mv_vnode);
975 taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
978 mqnode_release(parent);
985 struct vop_remove_args {
988 struct componentname *a_cnp;
993 * vnode removal operation
996 mqfs_remove(struct vop_remove_args *ap)
998 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
999 struct mqfs_node *pn;
1002 if (ap->a_vp->v_type == VDIR)
1004 pn = VTON(ap->a_vp);
1005 sx_xlock(&mqfs->mi_lock);
1006 error = do_unlink(pn, ap->a_cnp->cn_cred);
1007 sx_xunlock(&mqfs->mi_lock);
1012 struct vop_inactive_args {
1014 struct thread *a_td;
1019 mqfs_inactive(struct vop_inactive_args *ap)
1021 struct mqfs_node *pn = VTON(ap->a_vp);
1024 vrecycle(ap->a_vp, ap->a_td);
1029 struct vop_reclaim_args {
1030 struct vop_generic_args a_gen;
1032 struct thread *a_td;
1037 mqfs_reclaim(struct vop_reclaim_args *ap)
1039 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1040 struct vnode *vp = ap->a_vp;
1041 struct mqfs_node *pn;
1042 struct mqfs_vdata *vd;
1046 sx_xlock(&mqfs->mi_lock);
1048 LIST_REMOVE(vd, mv_link);
1049 uma_zfree(mvdata_zone, vd);
1051 sx_xunlock(&mqfs->mi_lock);
1056 struct vop_open_args {
1057 struct vop_generic_args a_gen;
1060 struct ucred *a_cred;
1061 struct thread *a_td;
1067 mqfs_open(struct vop_open_args *ap)
1073 struct vop_close_args {
1074 struct vop_generic_args a_gen;
1077 struct ucred *a_cred;
1078 struct thread *a_td;
1083 mqfs_close(struct vop_close_args *ap)
1089 struct vop_access_args {
1090 struct vop_generic_args a_gen;
1093 struct ucred *a_cred;
1094 struct thread *a_td;
1099 * Verify permissions
1102 mqfs_access(struct vop_access_args *ap)
1104 struct vnode *vp = ap->a_vp;
1108 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1111 error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1112 vattr.va_gid, ap->a_mode, ap->a_cred, NULL);
1117 struct vop_getattr_args {
1118 struct vop_generic_args a_gen;
1120 struct vattr *a_vap;
1121 struct ucred *a_cred;
1122 struct thread *a_td;
1127 * Get file attributes
1130 mqfs_getattr(struct vop_getattr_args *ap)
1132 struct vnode *vp = ap->a_vp;
1133 struct mqfs_node *pn = VTON(vp);
1134 struct vattr *vap = ap->a_vap;
1138 vap->va_type = vp->v_type;
1139 vap->va_mode = pn->mn_mode;
1141 vap->va_uid = pn->mn_uid;
1142 vap->va_gid = pn->mn_gid;
1143 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1144 vap->va_fileid = pn->mn_fileno;
1146 vap->va_blocksize = PAGE_SIZE;
1147 vap->va_bytes = vap->va_size = 0;
1148 vap->va_atime = pn->mn_atime;
1149 vap->va_mtime = pn->mn_mtime;
1150 vap->va_ctime = pn->mn_ctime;
1151 vap->va_birthtime = pn->mn_birth;
1156 vap->va_filerev = 0;
1157 vap->va_vaflags = 0;
1162 struct vop_setattr_args {
1163 struct vop_generic_args a_gen;
1165 struct vattr *a_vap;
1166 struct ucred *a_cred;
1167 struct thread *a_td;
1174 mqfs_setattr(struct vop_setattr_args *ap)
1176 struct mqfs_node *pn;
1185 if ((vap->va_type != VNON) ||
1186 (vap->va_nlink != VNOVAL) ||
1187 (vap->va_fsid != VNOVAL) ||
1188 (vap->va_fileid != VNOVAL) ||
1189 (vap->va_blocksize != VNOVAL) ||
1190 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1191 (vap->va_rdev != VNOVAL) ||
1192 ((int)vap->va_bytes != VNOVAL) ||
1193 (vap->va_gen != VNOVAL)) {
1200 if (vap->va_uid == (uid_t)VNOVAL)
1204 if (vap->va_gid == (gid_t)VNOVAL)
1209 if (uid != pn->mn_uid || gid != pn->mn_gid) {
1211 * To modify the ownership of a file, must possess VADMIN
1214 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)))
1218 * XXXRW: Why is there a privilege check here: shouldn't the
1219 * check in VOP_ACCESS() be enough? Also, are the group bits
1220 * below definitely right?
1222 if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1223 (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1224 (error = priv_check_cred(ap->a_td->td_ucred,
1225 PRIV_MQ_ADMIN, SUSER_ALLOWJAIL)) != 0)
1232 if (vap->va_mode != (mode_t)VNOVAL) {
1233 if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1234 (error = priv_check_cred(ap->a_td->td_ucred,
1235 PRIV_MQ_ADMIN, SUSER_ALLOWJAIL)))
1237 pn->mn_mode = vap->va_mode;
1241 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1242 /* See the comment in ufs_vnops::ufs_setattr(). */
1243 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1244 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1245 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1247 if (vap->va_atime.tv_sec != VNOVAL) {
1248 pn->mn_atime = vap->va_atime;
1250 if (vap->va_mtime.tv_sec != VNOVAL) {
1251 pn->mn_mtime = vap->va_mtime;
1256 vfs_timestamp(&pn->mn_ctime);
1262 struct vop_read_args {
1263 struct vop_generic_args a_gen;
1267 struct ucred *a_cred;
1275 mqfs_read(struct vop_read_args *ap)
1278 struct vnode *vp = ap->a_vp;
1279 struct uio *uio = ap->a_uio;
1280 struct mqfs_node *pn;
1284 if (vp->v_type != VREG)
1289 snprintf(buf, sizeof(buf),
1290 "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1295 buf[sizeof(buf)-1] = '\0';
1297 error = uiomove_frombuf(buf, len, uio);
1302 struct vop_readdir_args {
1303 struct vop_generic_args a_gen;
1306 struct ucred *a_cred;
1314 * Return directory entries.
1317 mqfs_readdir(struct vop_readdir_args *ap)
1320 struct mqfs_info *mi;
1321 struct mqfs_node *pd;
1322 struct mqfs_node *pn;
1323 struct dirent entry;
1325 int *tmp_ncookies = NULL;
1330 mi = VFSTOMQFS(vp->v_mount);
1334 if (vp->v_type != VDIR)
1337 if (uio->uio_offset < 0)
1340 if (ap->a_ncookies != NULL) {
1341 tmp_ncookies = ap->a_ncookies;
1342 *ap->a_ncookies = 0;
1343 ap->a_ncookies = NULL;
1349 sx_xlock(&mi->mi_lock);
1351 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1352 entry.d_reclen = sizeof(entry);
1354 mqfs_fileno_alloc(mi, pn);
1355 entry.d_fileno = pn->mn_fileno;
1356 for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1357 entry.d_name[i] = pn->mn_name[i];
1358 entry.d_name[i] = 0;
1360 switch (pn->mn_type) {
1364 case mqfstype_parent:
1365 entry.d_type = DT_DIR;
1368 entry.d_type = DT_REG;
1370 case mqfstype_symlink:
1371 entry.d_type = DT_LNK;
1374 panic("%s has unexpected node type: %d", pn->mn_name,
1377 if (entry.d_reclen > uio->uio_resid)
1379 if (offset >= uio->uio_offset) {
1380 error = vfs_read_dirent(ap, &entry, offset);
1384 offset += entry.d_reclen;
1386 sx_xunlock(&mi->mi_lock);
1388 uio->uio_offset = offset;
1390 if (tmp_ncookies != NULL)
1391 ap->a_ncookies = tmp_ncookies;
1399 struct vop_mkdir_args {
1400 struct vnode *a_dvp;
1401 struvt vnode **a_vpp;
1402 struvt componentname *a_cnp;
1403 struct vattr *a_vap;
1408 * Create a directory.
1411 mqfs_mkdir(struct vop_mkdir_args *ap)
1413 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1414 struct componentname *cnp = ap->a_cnp;
1415 struct mqfs_node *pd = VTON(ap->a_dvp);
1416 struct mqfs_node *pn;
1419 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1421 sx_xlock(&mqfs->mi_lock);
1424 pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1426 sx_xunlock(&mqfs->mi_lock);
1430 if ((cnp->cn_flags & HASBUF) == 0)
1431 panic("%s: no name", __func__);
1433 pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1434 ap->a_vap->cn_cred, ap->a_vap->va_mode);
1438 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1439 sx_xunlock(&mqfs->mi_lock);
1444 struct vop_rmdir_args {
1445 struct vnode *a_dvp;
1447 struct componentname *a_cnp;
1452 * Remove a directory.
1455 mqfs_rmdir(struct vop_rmdir_args *ap)
1457 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1458 struct mqfs_node *pn = VTON(ap->a_vp);
1459 struct mqfs_node *pt;
1461 if (pn->mn_type != mqfstype_dir)
1464 sx_xlock(&mqfs->mi_lock);
1465 if (pn->mn_deleted) {
1466 sx_xunlock(&mqfs->mi_lock);
1470 pt = LIST_FIRST(&pn->mn_children);
1471 pt = LIST_NEXT(pt, mn_sibling);
1472 pt = LIST_NEXT(pt, mn_sibling);
1474 sx_xunlock(&mqfs->mi_lock);
1478 pn->mn_parent = NULL;
1480 LIST_REMOVE(pn, mn_sibling);
1483 sx_xunlock(&mqfs->mi_lock);
1484 cache_purge(ap->a_vp);
1491 * Allocate a message queue
1493 static struct mqueue *
1494 mqueue_alloc(const struct mq_attr *attr)
1500 mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1501 TAILQ_INIT(&mq->mq_msgq);
1503 mq->mq_maxmsg = attr->mq_maxmsg;
1504 mq->mq_msgsize = attr->mq_msgsize;
1506 mq->mq_maxmsg = default_maxmsg;
1507 mq->mq_msgsize = default_msgsize;
1509 mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1510 knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1511 knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1512 atomic_add_int(&curmq, 1);
1517 * Destroy a message queue
1520 mqueue_free(struct mqueue *mq)
1522 struct mqueue_msg *msg;
1524 while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1525 TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1526 FREE(msg, M_MQUEUEDATA);
1529 mtx_destroy(&mq->mq_mutex);
1530 knlist_destroy(&mq->mq_rsel.si_note);
1531 knlist_destroy(&mq->mq_wsel.si_note);
1532 uma_zfree(mqueue_zone, mq);
1533 atomic_add_int(&curmq, -1);
1537 * Load a message from user space
1539 static struct mqueue_msg *
1540 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1542 struct mqueue_msg *msg;
1546 len = sizeof(struct mqueue_msg) + msg_size;
1547 MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1548 error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1551 FREE(msg, M_MQUEUEDATA);
1554 msg->msg_size = msg_size;
1555 msg->msg_prio = msg_prio;
1561 * Save a message to user space
1564 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1568 error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1570 if (error == 0 && msg_prio != NULL)
1571 error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1576 * Free a message's memory
1578 static __inline void
1579 mqueue_freemsg(struct mqueue_msg *msg)
1581 FREE(msg, M_MQUEUEDATA);
1585 * Send a message. if waitok is false, thread will not be
1586 * blocked if there is no data in queue, otherwise, absolute
1587 * time will be checked.
1590 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1591 size_t msg_len, unsigned msg_prio, int waitok,
1592 const struct timespec *abs_timeout)
1594 struct mqueue_msg *msg;
1595 struct timespec ets, ts, ts2;
1599 if (msg_prio >= MQ_PRIO_MAX)
1601 if (msg_len > mq->mq_msgsize)
1603 msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1607 /* O_NONBLOCK case */
1609 error = _mqueue_send(mq, msg, -1);
1615 /* we allow a null timeout (wait forever) */
1616 if (abs_timeout == NULL) {
1617 error = _mqueue_send(mq, msg, 0);
1623 /* send it before checking time */
1624 error = _mqueue_send(mq, msg, -1);
1628 if (error != EAGAIN)
1631 error = copyin(abs_timeout, &ets, sizeof(ets));
1634 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1641 timespecsub(&ts2, &ts);
1642 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1646 TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1647 error = _mqueue_send(mq, msg, tvtohz(&tv));
1648 if (error != ETIMEDOUT)
1654 mqueue_freemsg(msg);
1659 * Common routine to send a message
1662 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1664 struct mqueue_msg *msg2;
1667 mtx_lock(&mq->mq_mutex);
1668 while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1670 mtx_unlock(&mq->mq_mutex);
1674 error = msleep(&mq->mq_senders, &mq->mq_mutex,
1675 PCATCH, "mqsend", timo);
1677 if (error == EAGAIN)
1680 if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1681 mtx_unlock(&mq->mq_mutex);
1685 if (TAILQ_EMPTY(&mq->mq_msgq)) {
1686 TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1688 if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1689 TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1691 TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1692 if (msg2->msg_prio < msg->msg_prio)
1695 TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1699 mq->mq_totalbytes += msg->msg_size;
1700 if (mq->mq_receivers)
1701 wakeup_one(&mq->mq_receivers);
1702 else if (mq->mq_notifier != NULL)
1703 mqueue_send_notification(mq);
1704 if (mq->mq_flags & MQ_RSEL) {
1705 mq->mq_flags &= ~MQ_RSEL;
1706 selwakeup(&mq->mq_rsel);
1708 KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1709 mtx_unlock(&mq->mq_mutex);
1714 * Send realtime a signal to process which registered itself
1715 * successfully by mq_notify.
1718 mqueue_send_notification(struct mqueue *mq)
1720 struct mqueue_notifier *nt;
1723 mtx_assert(&mq->mq_mutex, MA_OWNED);
1724 nt = mq->mq_notifier;
1725 if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1728 if (!KSI_ONQ(&nt->nt_ksi))
1729 psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1732 mq->mq_notifier = NULL;
1736 * Get a message. if waitok is false, thread will not be
1737 * blocked if there is no data in queue, otherwise, absolute
1738 * time will be checked.
1741 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1742 size_t msg_len, unsigned *msg_prio, int waitok,
1743 const struct timespec *abs_timeout)
1745 struct mqueue_msg *msg;
1746 struct timespec ets, ts, ts2;
1750 if (msg_len < mq->mq_msgsize)
1753 /* O_NONBLOCK case */
1755 error = _mqueue_recv(mq, &msg, -1);
1761 /* we allow a null timeout (wait forever). */
1762 if (abs_timeout == NULL) {
1763 error = _mqueue_recv(mq, &msg, 0);
1769 /* try to get a message before checking time */
1770 error = _mqueue_recv(mq, &msg, -1);
1774 if (error != EAGAIN)
1777 error = copyin(abs_timeout, &ets, sizeof(ets));
1780 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1788 timespecsub(&ts2, &ts);
1789 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1793 TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1794 error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1797 if (error != ETIMEDOUT)
1802 error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1804 curthread->td_retval[0] = msg->msg_size;
1805 curthread->td_retval[1] = 0;
1807 mqueue_freemsg(msg);
1812 * Common routine to receive a message
1815 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1819 mtx_lock(&mq->mq_mutex);
1820 while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1822 mtx_unlock(&mq->mq_mutex);
1826 error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1827 PCATCH, "mqrecv", timo);
1829 if (error == EAGAIN)
1834 TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1836 mq->mq_totalbytes -= (*msg)->msg_size;
1838 wakeup_one(&mq->mq_senders);
1839 if (mq->mq_flags & MQ_WSEL) {
1840 mq->mq_flags &= ~MQ_WSEL;
1841 selwakeup(&mq->mq_wsel);
1843 KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1845 if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1846 !TAILQ_EMPTY(&mq->mq_msgq)) {
1847 mqueue_send_notification(mq);
1849 mtx_unlock(&mq->mq_mutex);
1853 static __inline struct mqueue_notifier *
1854 notifier_alloc(void)
1856 return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1859 static __inline void
1860 notifier_free(struct mqueue_notifier *p)
1862 uma_zfree(mqnoti_zone, p);
1865 static struct mqueue_notifier *
1866 notifier_search(struct proc *p, int fd)
1868 struct mqueue_notifier *nt;
1870 LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1871 if (nt->nt_ksi.ksi_mqd == fd)
1877 static __inline void
1878 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1880 LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1883 static __inline void
1884 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1886 LIST_REMOVE(nt, nt_link);
1891 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1893 struct mqueue_notifier *nt;
1895 mtx_assert(&mq->mq_mutex, MA_OWNED);
1897 nt = notifier_search(p, fd);
1899 if (mq->mq_notifier == nt)
1900 mq->mq_notifier = NULL;
1901 sigqueue_take(&nt->nt_ksi);
1902 notifier_delete(p, nt);
1908 * Syscall to open a message queue.
1911 kmq_open(struct thread *td, struct kmq_open_args *uap)
1913 char path[MQFS_NAMELEN + 1];
1914 struct mq_attr attr, *pattr;
1915 struct mqfs_node *pn;
1916 struct filedesc *fdp;
1919 int fd, error, len, flags, cmode;
1921 if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1924 fdp = td->td_proc->p_fd;
1925 flags = FFLAGS(uap->flags);
1926 cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1928 if ((flags & O_CREAT) && (uap->attr != NULL)) {
1929 error = copyin(uap->attr, &attr, sizeof(attr));
1932 if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1934 if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1940 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1945 * The first character of name must be a slash (/) character
1946 * and the remaining characters of name cannot include any slash
1950 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL)
1953 error = falloc(td, &fp, &fd);
1957 sx_xlock(&mqfs_data.mi_lock);
1958 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1960 if (!(flags & O_CREAT)) {
1963 mq = mqueue_alloc(pattr);
1967 pn = mqfs_create_file(mqfs_data.mi_root,
1968 path + 1, len - 1, td->td_ucred,
1981 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1990 error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1991 pn->mn_gid, acc_mode, td->td_ucred, NULL);
1996 sx_xunlock(&mqfs_data.mi_lock);
1997 fdclose(fdp, fp, fd, td);
2003 sx_xunlock(&mqfs_data.mi_lock);
2006 fp->f_flag = (flags & (FREAD | FWRITE | O_NONBLOCK));
2007 fp->f_type = DTYPE_MQUEUE;
2009 fp->f_ops = &mqueueops;
2012 FILEDESC_XLOCK(fdp);
2013 if (fdp->fd_ofiles[fd] == fp)
2014 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2015 FILEDESC_XUNLOCK(fdp);
2016 td->td_retval[0] = fd;
2022 * Syscall to unlink a message queue.
2025 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2027 char path[MQFS_NAMELEN+1];
2028 struct mqfs_node *pn;
2031 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2036 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL)
2039 sx_xlock(&mqfs_data.mi_lock);
2040 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2042 error = do_unlink(pn, td->td_ucred);
2045 sx_xunlock(&mqfs_data.mi_lock);
2049 typedef int (*_fgetf)(struct thread *, int, struct file **);
2052 * Get message queue by giving file slot
2055 _getmq(struct thread *td, int fd, _fgetf func,
2056 struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2058 struct mqfs_node *pn;
2061 error = func(td, fd, fpp);
2064 if (&mqueueops != (*fpp)->f_ops) {
2068 pn = (*fpp)->f_data;
2077 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2078 struct mqueue **pmq)
2080 return _getmq(td, fd, fget, fpp, ppn, pmq);
2084 getmq_read(struct thread *td, int fd, struct file **fpp,
2085 struct mqfs_node **ppn, struct mqueue **pmq)
2087 return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2091 getmq_write(struct thread *td, int fd, struct file **fpp,
2092 struct mqfs_node **ppn, struct mqueue **pmq)
2094 return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2098 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2102 struct mq_attr attr, oattr;
2106 error = copyin(uap->attr, &attr, sizeof(attr));
2109 if (attr.mq_flags & ~O_NONBLOCK)
2112 error = getmq(td, uap->mqd, &fp, NULL, &mq);
2115 oattr.mq_maxmsg = mq->mq_maxmsg;
2116 oattr.mq_msgsize = mq->mq_msgsize;
2117 oattr.mq_curmsgs = mq->mq_curmsgs;
2119 oattr.mq_flags = (O_NONBLOCK & fp->f_flag);
2121 fp->f_flag &= ~O_NONBLOCK;
2122 fp->f_flag |= (attr.mq_flags & O_NONBLOCK);
2127 error = copyout(&oattr, uap->oattr, sizeof(oattr));
2132 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2139 error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2142 waitok = !(fp->f_flag & O_NONBLOCK);
2143 error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2144 uap->msg_prio, waitok, uap->abs_timeout);
2150 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2156 error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2159 waitok = !(fp->f_flag & O_NONBLOCK);
2160 error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2161 uap->msg_prio, waitok, uap->abs_timeout);
2167 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2170 struct filedesc *fdp;
2174 struct mqueue_notifier *nt, *newnt = NULL;
2178 fdp = td->td_proc->p_fd;
2180 error = copyin(uap->sigev, &ev, sizeof(ev));
2183 if (ev.sigev_notify != SIGEV_SIGNAL &&
2184 ev.sigev_notify != SIGEV_THREAD_ID &&
2185 ev.sigev_notify != SIGEV_NONE)
2187 if ((ev.sigev_notify == SIGEV_SIGNAL ||
2188 ev.sigev_notify == SIGEV_THREAD_ID) &&
2189 !_SIG_VALID(ev.sigev_signo))
2192 error = getmq(td, uap->mqd, &fp, NULL, &mq);
2196 FILEDESC_SLOCK(fdp);
2197 if (fget_locked(fdp, uap->mqd) != fp) {
2198 FILEDESC_SUNLOCK(fdp);
2202 mtx_lock(&mq->mq_mutex);
2203 FILEDESC_SUNLOCK(fdp);
2204 if (uap->sigev != NULL) {
2205 if (mq->mq_notifier != NULL) {
2209 nt = notifier_search(p, uap->mqd);
2211 if (newnt == NULL) {
2213 mtx_unlock(&mq->mq_mutex);
2214 newnt = notifier_alloc();
2220 sigqueue_take(&nt->nt_ksi);
2221 if (newnt != NULL) {
2222 notifier_free(newnt);
2228 ksiginfo_init(&nt->nt_ksi);
2229 nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2230 nt->nt_ksi.ksi_code = SI_MESGQ;
2232 nt->nt_ksi.ksi_mqd = uap->mqd;
2233 notifier_insert(p, nt);
2236 mq->mq_notifier = nt;
2239 * if there is no receivers and message queue
2240 * is not empty, we should send notification
2241 * as soon as possible.
2243 if (mq->mq_receivers == 0 &&
2244 !TAILQ_EMPTY(&mq->mq_msgq))
2245 mqueue_send_notification(mq);
2248 notifier_remove(p, mq, uap->mqd);
2250 mtx_unlock(&mq->mq_mutex);
2255 notifier_free(newnt);
2260 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2262 struct filedesc *fdp;
2265 fdp = td->td_proc->p_fd;
2266 FILEDESC_LOCK_ASSERT(fdp);
2268 if (fp->f_ops == &mqueueops) {
2270 mtx_lock(&mq->mq_mutex);
2271 notifier_remove(td->td_proc, mq, fd);
2273 /* have to wakeup thread in same process */
2274 if (mq->mq_flags & MQ_RSEL) {
2275 mq->mq_flags &= ~MQ_RSEL;
2276 selwakeup(&mq->mq_rsel);
2278 if (mq->mq_flags & MQ_WSEL) {
2279 mq->mq_flags &= ~MQ_WSEL;
2280 selwakeup(&mq->mq_wsel);
2282 mtx_unlock(&mq->mq_mutex);
2287 mq_proc_exit(void *arg __unused, struct proc *p)
2289 struct filedesc *fdp;
2295 FILEDESC_SLOCK(fdp);
2296 for (i = 0; i < fdp->fd_nfiles; ++i) {
2297 fp = fget_locked(fdp, i);
2298 if (fp != NULL && fp->f_ops == &mqueueops) {
2300 mtx_lock(&mq->mq_mutex);
2301 notifier_remove(p, FPTOMQ(fp), i);
2302 mtx_unlock(&mq->mq_mutex);
2305 FILEDESC_SUNLOCK(fdp);
2306 KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2310 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2311 int flags, struct thread *td)
2313 return (EOPNOTSUPP);
2317 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2318 int flags, struct thread *td)
2320 return (EOPNOTSUPP);
2324 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2325 struct ucred *active_cred, struct thread *td)
2331 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2334 struct mqueue *mq = FPTOMQ(fp);
2337 mtx_lock(&mq->mq_mutex);
2338 if (events & (POLLIN | POLLRDNORM)) {
2339 if (mq->mq_curmsgs) {
2340 revents |= events & (POLLIN | POLLRDNORM);
2342 mq->mq_flags |= MQ_RSEL;
2343 selrecord(td, &mq->mq_rsel);
2346 if (events & POLLOUT) {
2347 if (mq->mq_curmsgs < mq->mq_maxmsg)
2350 mq->mq_flags |= MQ_WSEL;
2351 selrecord(td, &mq->mq_wsel);
2354 mtx_unlock(&mq->mq_mutex);
2359 mqf_close(struct file *fp, struct thread *td)
2361 struct mqfs_node *pn;
2363 fp->f_ops = &badfileops;
2366 sx_xlock(&mqfs_data.mi_lock);
2368 sx_xunlock(&mqfs_data.mi_lock);
2373 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2376 struct mqfs_node *pn = fp->f_data;
2378 bzero(st, sizeof *st);
2379 st->st_atimespec = pn->mn_atime;
2380 st->st_mtimespec = pn->mn_mtime;
2381 st->st_ctimespec = pn->mn_ctime;
2382 st->st_birthtimespec = pn->mn_birth;
2383 st->st_uid = pn->mn_uid;
2384 st->st_gid = pn->mn_gid;
2385 st->st_mode = S_IFIFO | pn->mn_mode;
2390 mqf_kqfilter(struct file *fp, struct knote *kn)
2392 struct mqueue *mq = FPTOMQ(fp);
2395 if (kn->kn_filter == EVFILT_READ) {
2396 kn->kn_fop = &mq_rfiltops;
2397 knlist_add(&mq->mq_rsel.si_note, kn, 0);
2398 } else if (kn->kn_filter == EVFILT_WRITE) {
2399 kn->kn_fop = &mq_wfiltops;
2400 knlist_add(&mq->mq_wsel.si_note, kn, 0);
2407 filt_mqdetach(struct knote *kn)
2409 struct mqueue *mq = FPTOMQ(kn->kn_fp);
2411 if (kn->kn_filter == EVFILT_READ)
2412 knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2413 else if (kn->kn_filter == EVFILT_WRITE)
2414 knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2416 panic("filt_mqdetach");
2420 filt_mqread(struct knote *kn, long hint)
2422 struct mqueue *mq = FPTOMQ(kn->kn_fp);
2424 mtx_assert(&mq->mq_mutex, MA_OWNED);
2425 return (mq->mq_curmsgs != 0);
2429 filt_mqwrite(struct knote *kn, long hint)
2431 struct mqueue *mq = FPTOMQ(kn->kn_fp);
2433 mtx_assert(&mq->mq_mutex, MA_OWNED);
2434 return (mq->mq_curmsgs < mq->mq_maxmsg);
2437 static struct fileops mqueueops = {
2438 .fo_read = mqf_read,
2439 .fo_write = mqf_write,
2440 .fo_ioctl = mqf_ioctl,
2441 .fo_poll = mqf_poll,
2442 .fo_kqfilter = mqf_kqfilter,
2443 .fo_stat = mqf_stat,
2444 .fo_close = mqf_close
2447 static struct vop_vector mqfs_vnodeops = {
2448 .vop_default = &default_vnodeops,
2449 .vop_access = mqfs_access,
2450 .vop_cachedlookup = mqfs_lookup,
2451 .vop_lookup = vfs_cache_lookup,
2452 .vop_reclaim = mqfs_reclaim,
2453 .vop_create = mqfs_create,
2454 .vop_remove = mqfs_remove,
2455 .vop_inactive = mqfs_inactive,
2456 .vop_open = mqfs_open,
2457 .vop_close = mqfs_close,
2458 .vop_getattr = mqfs_getattr,
2459 .vop_setattr = mqfs_setattr,
2460 .vop_read = mqfs_read,
2461 .vop_write = VOP_EOPNOTSUPP,
2462 .vop_readdir = mqfs_readdir,
2463 .vop_mkdir = VOP_EOPNOTSUPP,
2464 .vop_rmdir = VOP_EOPNOTSUPP
2467 static struct vfsops mqfs_vfsops = {
2468 .vfs_init = mqfs_init,
2469 .vfs_uninit = mqfs_uninit,
2470 .vfs_mount = mqfs_mount,
2471 .vfs_unmount = mqfs_unmount,
2472 .vfs_root = mqfs_root,
2473 .vfs_statfs = mqfs_statfs,
2476 SYSCALL_MODULE_HELPER(kmq_open);
2477 SYSCALL_MODULE_HELPER(kmq_setattr);
2478 SYSCALL_MODULE_HELPER(kmq_timedsend);
2479 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2480 SYSCALL_MODULE_HELPER(kmq_notify);
2481 SYSCALL_MODULE_HELPER(kmq_unlink);
2483 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2484 MODULE_VERSION(mqueuefs, 1);