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)
708 vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
712 error = getnewvnode("mqueue", mp, &mqfs_vnodeops, vpp);
715 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
716 error = insmntque(*vpp, mp);
721 vd = uma_zalloc(mvdata_zone, M_WAITOK);
725 TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
726 LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
728 switch (pn->mn_type) {
730 (*vpp)->v_vflag = VV_ROOT;
734 case mqfstype_parent:
735 (*vpp)->v_type = VDIR;
738 (*vpp)->v_type = VREG;
740 case mqfstype_symlink:
741 (*vpp)->v_type = VLNK;
744 KASSERT(0, ("mqfs_allocf called for null node\n"));
746 panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
752 * Search a directory entry
754 static struct mqfs_node *
755 mqfs_search(struct mqfs_node *pd, const char *name, int len)
757 struct mqfs_node *pn;
759 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
760 if (strncmp(pn->mn_name, name, len) == 0)
767 * Look up a file or directory.
770 mqfs_lookupx(struct vop_cachedlookup_args *ap)
772 struct componentname *cnp;
773 struct vnode *dvp, **vpp;
774 struct mqfs_node *pd;
775 struct mqfs_node *pn;
776 int nameiop, flags, error, namelen;
783 pname = cnp->cn_nameptr;
784 namelen = cnp->cn_namelen;
786 flags = cnp->cn_flags;
787 nameiop = cnp->cn_nameiop;
792 if (dvp->v_type != VDIR)
795 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
799 /* shortcut: check if the name is too long */
800 if (cnp->cn_namelen >= MQFS_NAMELEN)
804 if (namelen == 1 && pname[0] == '.') {
805 if ((flags & ISLASTCN) && nameiop != LOOKUP)
814 if (cnp->cn_flags & ISDOTDOT) {
815 if (dvp->v_vflag & VV_ROOT)
817 if ((flags & ISLASTCN) && nameiop != LOOKUP)
820 KASSERT(pd->mn_parent, ("non-root directory has no parent"));
822 error = mqfs_allocv(dvp->v_mount, vpp, pn);
823 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
828 pn = mqfs_search(pd, pname, namelen);
833 if (nameiop == DELETE && (flags & ISLASTCN)) {
834 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
845 error = mqfs_allocv(dvp->v_mount, vpp, pn);
846 if (error == 0 && cnp->cn_flags & MAKEENTRY)
847 cache_enter(dvp, *vpp, cnp);
853 /* will create a new entry in the directory ? */
854 if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
855 && (flags & ISLASTCN)) {
856 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
859 cnp->cn_flags |= SAVENAME;
860 return (EJUSTRETURN);
866 struct vop_lookup_args {
867 struct vop_generic_args a_gen;
869 struct vnode **a_vpp;
870 struct componentname *a_cnp;
875 * vnode lookup operation
878 mqfs_lookup(struct vop_cachedlookup_args *ap)
880 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
883 sx_xlock(&mqfs->mi_lock);
884 rc = mqfs_lookupx(ap);
885 sx_xunlock(&mqfs->mi_lock);
890 struct vop_create_args {
892 struct vnode **a_vpp;
893 struct componentname *a_cnp;
899 * vnode creation operation
902 mqfs_create(struct vop_create_args *ap)
904 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
905 struct componentname *cnp = ap->a_cnp;
906 struct mqfs_node *pd;
907 struct mqfs_node *pn;
911 pd = VTON(ap->a_dvp);
912 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
914 mq = mqueue_alloc(NULL);
917 sx_xlock(&mqfs->mi_lock);
920 pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
923 sx_xunlock(&mqfs->mi_lock);
927 if ((cnp->cn_flags & HASBUF) == 0)
928 panic("%s: no name", __func__);
930 pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
931 cnp->cn_cred, ap->a_vap->va_mode);
935 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
941 sx_xunlock(&mqfs->mi_lock);
951 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
953 struct mqfs_node *parent;
954 struct mqfs_vdata *vd;
957 sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
959 if (ucred->cr_uid != pn->mn_uid &&
960 (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
962 else if (!pn->mn_deleted) {
963 parent = pn->mn_parent;
964 pn->mn_parent = NULL;
966 LIST_REMOVE(pn, mn_sibling);
967 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
968 cache_purge(vd->mv_vnode);
970 taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
973 mqnode_release(parent);
980 struct vop_remove_args {
983 struct componentname *a_cnp;
988 * vnode removal operation
991 mqfs_remove(struct vop_remove_args *ap)
993 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
994 struct mqfs_node *pn;
997 if (ap->a_vp->v_type == VDIR)
1000 sx_xlock(&mqfs->mi_lock);
1001 error = do_unlink(pn, ap->a_cnp->cn_cred);
1002 sx_xunlock(&mqfs->mi_lock);
1007 struct vop_inactive_args {
1009 struct thread *a_td;
1014 mqfs_inactive(struct vop_inactive_args *ap)
1016 struct mqfs_node *pn = VTON(ap->a_vp);
1019 vrecycle(ap->a_vp, ap->a_td);
1024 struct vop_reclaim_args {
1025 struct vop_generic_args a_gen;
1027 struct thread *a_td;
1032 mqfs_reclaim(struct vop_reclaim_args *ap)
1034 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1035 struct vnode *vp = ap->a_vp;
1036 struct mqfs_node *pn;
1037 struct mqfs_vdata *vd;
1041 sx_xlock(&mqfs->mi_lock);
1043 LIST_REMOVE(vd, mv_link);
1044 uma_zfree(mvdata_zone, vd);
1046 sx_xunlock(&mqfs->mi_lock);
1051 struct vop_open_args {
1052 struct vop_generic_args a_gen;
1055 struct ucred *a_cred;
1056 struct thread *a_td;
1062 mqfs_open(struct vop_open_args *ap)
1068 struct vop_close_args {
1069 struct vop_generic_args a_gen;
1072 struct ucred *a_cred;
1073 struct thread *a_td;
1078 mqfs_close(struct vop_close_args *ap)
1084 struct vop_access_args {
1085 struct vop_generic_args a_gen;
1088 struct ucred *a_cred;
1089 struct thread *a_td;
1094 * Verify permissions
1097 mqfs_access(struct vop_access_args *ap)
1099 struct vnode *vp = ap->a_vp;
1103 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1106 error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1107 vattr.va_gid, ap->a_mode, ap->a_cred, NULL);
1112 struct vop_getattr_args {
1113 struct vop_generic_args a_gen;
1115 struct vattr *a_vap;
1116 struct ucred *a_cred;
1117 struct thread *a_td;
1122 * Get file attributes
1125 mqfs_getattr(struct vop_getattr_args *ap)
1127 struct vnode *vp = ap->a_vp;
1128 struct mqfs_node *pn = VTON(vp);
1129 struct vattr *vap = ap->a_vap;
1133 vap->va_type = vp->v_type;
1134 vap->va_mode = pn->mn_mode;
1136 vap->va_uid = pn->mn_uid;
1137 vap->va_gid = pn->mn_gid;
1138 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1139 vap->va_fileid = pn->mn_fileno;
1141 vap->va_blocksize = PAGE_SIZE;
1142 vap->va_bytes = vap->va_size = 0;
1143 vap->va_atime = pn->mn_atime;
1144 vap->va_mtime = pn->mn_mtime;
1145 vap->va_ctime = pn->mn_ctime;
1146 vap->va_birthtime = pn->mn_birth;
1151 vap->va_filerev = 0;
1152 vap->va_vaflags = 0;
1157 struct vop_setattr_args {
1158 struct vop_generic_args a_gen;
1160 struct vattr *a_vap;
1161 struct ucred *a_cred;
1162 struct thread *a_td;
1169 mqfs_setattr(struct vop_setattr_args *ap)
1171 struct mqfs_node *pn;
1180 if ((vap->va_type != VNON) ||
1181 (vap->va_nlink != VNOVAL) ||
1182 (vap->va_fsid != VNOVAL) ||
1183 (vap->va_fileid != VNOVAL) ||
1184 (vap->va_blocksize != VNOVAL) ||
1185 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1186 (vap->va_rdev != VNOVAL) ||
1187 ((int)vap->va_bytes != VNOVAL) ||
1188 (vap->va_gen != VNOVAL)) {
1195 if (vap->va_uid == (uid_t)VNOVAL)
1199 if (vap->va_gid == (gid_t)VNOVAL)
1204 if (uid != pn->mn_uid || gid != pn->mn_gid) {
1206 * To modify the ownership of a file, must possess VADMIN
1209 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)))
1213 * XXXRW: Why is there a privilege check here: shouldn't the
1214 * check in VOP_ACCESS() be enough? Also, are the group bits
1215 * below definitely right?
1217 if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1218 (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1219 (error = priv_check(ap->a_td, PRIV_MQ_ADMIN)) != 0)
1226 if (vap->va_mode != (mode_t)VNOVAL) {
1227 if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1228 (error = priv_check(ap->a_td, PRIV_MQ_ADMIN)))
1230 pn->mn_mode = vap->va_mode;
1234 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1235 /* See the comment in ufs_vnops::ufs_setattr(). */
1236 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1237 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1238 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1240 if (vap->va_atime.tv_sec != VNOVAL) {
1241 pn->mn_atime = vap->va_atime;
1243 if (vap->va_mtime.tv_sec != VNOVAL) {
1244 pn->mn_mtime = vap->va_mtime;
1249 vfs_timestamp(&pn->mn_ctime);
1255 struct vop_read_args {
1256 struct vop_generic_args a_gen;
1260 struct ucred *a_cred;
1268 mqfs_read(struct vop_read_args *ap)
1271 struct vnode *vp = ap->a_vp;
1272 struct uio *uio = ap->a_uio;
1273 struct mqfs_node *pn;
1277 if (vp->v_type != VREG)
1282 snprintf(buf, sizeof(buf),
1283 "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1288 buf[sizeof(buf)-1] = '\0';
1290 error = uiomove_frombuf(buf, len, uio);
1295 struct vop_readdir_args {
1296 struct vop_generic_args a_gen;
1299 struct ucred *a_cred;
1307 * Return directory entries.
1310 mqfs_readdir(struct vop_readdir_args *ap)
1313 struct mqfs_info *mi;
1314 struct mqfs_node *pd;
1315 struct mqfs_node *pn;
1316 struct dirent entry;
1318 int *tmp_ncookies = NULL;
1323 mi = VFSTOMQFS(vp->v_mount);
1327 if (vp->v_type != VDIR)
1330 if (uio->uio_offset < 0)
1333 if (ap->a_ncookies != NULL) {
1334 tmp_ncookies = ap->a_ncookies;
1335 *ap->a_ncookies = 0;
1336 ap->a_ncookies = NULL;
1342 sx_xlock(&mi->mi_lock);
1344 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1345 entry.d_reclen = sizeof(entry);
1347 mqfs_fileno_alloc(mi, pn);
1348 entry.d_fileno = pn->mn_fileno;
1349 for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1350 entry.d_name[i] = pn->mn_name[i];
1351 entry.d_name[i] = 0;
1353 switch (pn->mn_type) {
1357 case mqfstype_parent:
1358 entry.d_type = DT_DIR;
1361 entry.d_type = DT_REG;
1363 case mqfstype_symlink:
1364 entry.d_type = DT_LNK;
1367 panic("%s has unexpected node type: %d", pn->mn_name,
1370 if (entry.d_reclen > uio->uio_resid)
1372 if (offset >= uio->uio_offset) {
1373 error = vfs_read_dirent(ap, &entry, offset);
1377 offset += entry.d_reclen;
1379 sx_xunlock(&mi->mi_lock);
1381 uio->uio_offset = offset;
1383 if (tmp_ncookies != NULL)
1384 ap->a_ncookies = tmp_ncookies;
1392 struct vop_mkdir_args {
1393 struct vnode *a_dvp;
1394 struvt vnode **a_vpp;
1395 struvt componentname *a_cnp;
1396 struct vattr *a_vap;
1401 * Create a directory.
1404 mqfs_mkdir(struct vop_mkdir_args *ap)
1406 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1407 struct componentname *cnp = ap->a_cnp;
1408 struct mqfs_node *pd = VTON(ap->a_dvp);
1409 struct mqfs_node *pn;
1412 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1414 sx_xlock(&mqfs->mi_lock);
1417 pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1419 sx_xunlock(&mqfs->mi_lock);
1423 if ((cnp->cn_flags & HASBUF) == 0)
1424 panic("%s: no name", __func__);
1426 pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1427 ap->a_vap->cn_cred, ap->a_vap->va_mode);
1431 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1432 sx_xunlock(&mqfs->mi_lock);
1437 struct vop_rmdir_args {
1438 struct vnode *a_dvp;
1440 struct componentname *a_cnp;
1445 * Remove a directory.
1448 mqfs_rmdir(struct vop_rmdir_args *ap)
1450 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1451 struct mqfs_node *pn = VTON(ap->a_vp);
1452 struct mqfs_node *pt;
1454 if (pn->mn_type != mqfstype_dir)
1457 sx_xlock(&mqfs->mi_lock);
1458 if (pn->mn_deleted) {
1459 sx_xunlock(&mqfs->mi_lock);
1463 pt = LIST_FIRST(&pn->mn_children);
1464 pt = LIST_NEXT(pt, mn_sibling);
1465 pt = LIST_NEXT(pt, mn_sibling);
1467 sx_xunlock(&mqfs->mi_lock);
1471 pn->mn_parent = NULL;
1473 LIST_REMOVE(pn, mn_sibling);
1476 sx_xunlock(&mqfs->mi_lock);
1477 cache_purge(ap->a_vp);
1484 * Allocate a message queue
1486 static struct mqueue *
1487 mqueue_alloc(const struct mq_attr *attr)
1493 mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1494 TAILQ_INIT(&mq->mq_msgq);
1496 mq->mq_maxmsg = attr->mq_maxmsg;
1497 mq->mq_msgsize = attr->mq_msgsize;
1499 mq->mq_maxmsg = default_maxmsg;
1500 mq->mq_msgsize = default_msgsize;
1502 mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1503 knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1504 knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1505 atomic_add_int(&curmq, 1);
1510 * Destroy a message queue
1513 mqueue_free(struct mqueue *mq)
1515 struct mqueue_msg *msg;
1517 while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1518 TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1519 FREE(msg, M_MQUEUEDATA);
1522 mtx_destroy(&mq->mq_mutex);
1523 knlist_destroy(&mq->mq_rsel.si_note);
1524 knlist_destroy(&mq->mq_wsel.si_note);
1525 uma_zfree(mqueue_zone, mq);
1526 atomic_add_int(&curmq, -1);
1530 * Load a message from user space
1532 static struct mqueue_msg *
1533 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1535 struct mqueue_msg *msg;
1539 len = sizeof(struct mqueue_msg) + msg_size;
1540 MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1541 error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1544 FREE(msg, M_MQUEUEDATA);
1547 msg->msg_size = msg_size;
1548 msg->msg_prio = msg_prio;
1554 * Save a message to user space
1557 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1561 error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1563 if (error == 0 && msg_prio != NULL)
1564 error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1569 * Free a message's memory
1571 static __inline void
1572 mqueue_freemsg(struct mqueue_msg *msg)
1574 FREE(msg, M_MQUEUEDATA);
1578 * Send a message. if waitok is false, thread will not be
1579 * blocked if there is no data in queue, otherwise, absolute
1580 * time will be checked.
1583 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1584 size_t msg_len, unsigned msg_prio, int waitok,
1585 const struct timespec *abs_timeout)
1587 struct mqueue_msg *msg;
1588 struct timespec ets, ts, ts2;
1592 if (msg_prio >= MQ_PRIO_MAX)
1594 if (msg_len > mq->mq_msgsize)
1596 msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1600 /* O_NONBLOCK case */
1602 error = _mqueue_send(mq, msg, -1);
1608 /* we allow a null timeout (wait forever) */
1609 if (abs_timeout == NULL) {
1610 error = _mqueue_send(mq, msg, 0);
1616 /* send it before checking time */
1617 error = _mqueue_send(mq, msg, -1);
1621 if (error != EAGAIN)
1624 error = copyin(abs_timeout, &ets, sizeof(ets));
1627 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1634 timespecsub(&ts2, &ts);
1635 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1639 TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1640 error = _mqueue_send(mq, msg, tvtohz(&tv));
1641 if (error != ETIMEDOUT)
1647 mqueue_freemsg(msg);
1652 * Common routine to send a message
1655 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1657 struct mqueue_msg *msg2;
1660 mtx_lock(&mq->mq_mutex);
1661 while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1663 mtx_unlock(&mq->mq_mutex);
1667 error = msleep(&mq->mq_senders, &mq->mq_mutex,
1668 PCATCH, "mqsend", timo);
1670 if (error == EAGAIN)
1673 if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1674 mtx_unlock(&mq->mq_mutex);
1678 if (TAILQ_EMPTY(&mq->mq_msgq)) {
1679 TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1681 if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1682 TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1684 TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1685 if (msg2->msg_prio < msg->msg_prio)
1688 TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1692 mq->mq_totalbytes += msg->msg_size;
1693 if (mq->mq_receivers)
1694 wakeup_one(&mq->mq_receivers);
1695 else if (mq->mq_notifier != NULL)
1696 mqueue_send_notification(mq);
1697 if (mq->mq_flags & MQ_RSEL) {
1698 mq->mq_flags &= ~MQ_RSEL;
1699 selwakeup(&mq->mq_rsel);
1701 KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1702 mtx_unlock(&mq->mq_mutex);
1707 * Send realtime a signal to process which registered itself
1708 * successfully by mq_notify.
1711 mqueue_send_notification(struct mqueue *mq)
1713 struct mqueue_notifier *nt;
1716 mtx_assert(&mq->mq_mutex, MA_OWNED);
1717 nt = mq->mq_notifier;
1718 if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1721 if (!KSI_ONQ(&nt->nt_ksi))
1722 psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1725 mq->mq_notifier = NULL;
1729 * Get a message. if waitok is false, thread will not be
1730 * blocked if there is no data in queue, otherwise, absolute
1731 * time will be checked.
1734 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1735 size_t msg_len, unsigned *msg_prio, int waitok,
1736 const struct timespec *abs_timeout)
1738 struct mqueue_msg *msg;
1739 struct timespec ets, ts, ts2;
1743 if (msg_len < mq->mq_msgsize)
1746 /* O_NONBLOCK case */
1748 error = _mqueue_recv(mq, &msg, -1);
1754 /* we allow a null timeout (wait forever). */
1755 if (abs_timeout == NULL) {
1756 error = _mqueue_recv(mq, &msg, 0);
1762 /* try to get a message before checking time */
1763 error = _mqueue_recv(mq, &msg, -1);
1767 if (error != EAGAIN)
1770 error = copyin(abs_timeout, &ets, sizeof(ets));
1773 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1781 timespecsub(&ts2, &ts);
1782 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1786 TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1787 error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1790 if (error != ETIMEDOUT)
1795 error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1797 curthread->td_retval[0] = msg->msg_size;
1798 curthread->td_retval[1] = 0;
1800 mqueue_freemsg(msg);
1805 * Common routine to receive a message
1808 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1812 mtx_lock(&mq->mq_mutex);
1813 while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1815 mtx_unlock(&mq->mq_mutex);
1819 error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1820 PCATCH, "mqrecv", timo);
1822 if (error == EAGAIN)
1827 TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1829 mq->mq_totalbytes -= (*msg)->msg_size;
1831 wakeup_one(&mq->mq_senders);
1832 if (mq->mq_flags & MQ_WSEL) {
1833 mq->mq_flags &= ~MQ_WSEL;
1834 selwakeup(&mq->mq_wsel);
1836 KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1838 if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1839 !TAILQ_EMPTY(&mq->mq_msgq)) {
1840 mqueue_send_notification(mq);
1842 mtx_unlock(&mq->mq_mutex);
1846 static __inline struct mqueue_notifier *
1847 notifier_alloc(void)
1849 return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1852 static __inline void
1853 notifier_free(struct mqueue_notifier *p)
1855 uma_zfree(mqnoti_zone, p);
1858 static struct mqueue_notifier *
1859 notifier_search(struct proc *p, int fd)
1861 struct mqueue_notifier *nt;
1863 LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1864 if (nt->nt_ksi.ksi_mqd == fd)
1870 static __inline void
1871 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1873 LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1876 static __inline void
1877 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1879 LIST_REMOVE(nt, nt_link);
1884 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1886 struct mqueue_notifier *nt;
1888 mtx_assert(&mq->mq_mutex, MA_OWNED);
1890 nt = notifier_search(p, fd);
1892 if (mq->mq_notifier == nt)
1893 mq->mq_notifier = NULL;
1894 sigqueue_take(&nt->nt_ksi);
1895 notifier_delete(p, nt);
1901 * Syscall to open a message queue.
1904 kmq_open(struct thread *td, struct kmq_open_args *uap)
1906 char path[MQFS_NAMELEN + 1];
1907 struct mq_attr attr, *pattr;
1908 struct mqfs_node *pn;
1909 struct filedesc *fdp;
1912 int fd, error, len, flags, cmode;
1914 if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1917 fdp = td->td_proc->p_fd;
1918 flags = FFLAGS(uap->flags);
1919 cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1921 if ((flags & O_CREAT) && (uap->attr != NULL)) {
1922 error = copyin(uap->attr, &attr, sizeof(attr));
1925 if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1927 if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1933 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1938 * The first character of name must be a slash (/) character
1939 * and the remaining characters of name cannot include any slash
1943 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL)
1946 error = falloc(td, &fp, &fd);
1950 sx_xlock(&mqfs_data.mi_lock);
1951 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1953 if (!(flags & O_CREAT)) {
1956 mq = mqueue_alloc(pattr);
1960 pn = mqfs_create_file(mqfs_data.mi_root,
1961 path + 1, len - 1, td->td_ucred,
1974 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1983 error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1984 pn->mn_gid, acc_mode, td->td_ucred, NULL);
1989 sx_xunlock(&mqfs_data.mi_lock);
1990 fdclose(fdp, fp, fd, td);
1996 sx_xunlock(&mqfs_data.mi_lock);
1998 finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2001 FILEDESC_XLOCK(fdp);
2002 if (fdp->fd_ofiles[fd] == fp)
2003 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2004 FILEDESC_XUNLOCK(fdp);
2005 td->td_retval[0] = fd;
2011 * Syscall to unlink a message queue.
2014 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2016 char path[MQFS_NAMELEN+1];
2017 struct mqfs_node *pn;
2020 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2025 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL)
2028 sx_xlock(&mqfs_data.mi_lock);
2029 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2031 error = do_unlink(pn, td->td_ucred);
2034 sx_xunlock(&mqfs_data.mi_lock);
2038 typedef int (*_fgetf)(struct thread *, int, struct file **);
2041 * Get message queue by giving file slot
2044 _getmq(struct thread *td, int fd, _fgetf func,
2045 struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2047 struct mqfs_node *pn;
2050 error = func(td, fd, fpp);
2053 if (&mqueueops != (*fpp)->f_ops) {
2057 pn = (*fpp)->f_data;
2066 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2067 struct mqueue **pmq)
2069 return _getmq(td, fd, fget, fpp, ppn, pmq);
2073 getmq_read(struct thread *td, int fd, struct file **fpp,
2074 struct mqfs_node **ppn, struct mqueue **pmq)
2076 return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2080 getmq_write(struct thread *td, int fd, struct file **fpp,
2081 struct mqfs_node **ppn, struct mqueue **pmq)
2083 return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2087 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2091 struct mq_attr attr, oattr;
2096 error = copyin(uap->attr, &attr, sizeof(attr));
2099 if (attr.mq_flags & ~O_NONBLOCK)
2102 error = getmq(td, uap->mqd, &fp, NULL, &mq);
2105 oattr.mq_maxmsg = mq->mq_maxmsg;
2106 oattr.mq_msgsize = mq->mq_msgsize;
2107 oattr.mq_curmsgs = mq->mq_curmsgs;
2110 oflag = flag = fp->f_flag;
2111 flag &= ~O_NONBLOCK;
2112 flag |= (attr.mq_flags & O_NONBLOCK);
2113 } while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2116 oattr.mq_flags = (O_NONBLOCK & oflag);
2119 error = copyout(&oattr, uap->oattr, sizeof(oattr));
2124 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2131 error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2134 waitok = !(fp->f_flag & O_NONBLOCK);
2135 error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2136 uap->msg_prio, waitok, uap->abs_timeout);
2142 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2148 error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2151 waitok = !(fp->f_flag & O_NONBLOCK);
2152 error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2153 uap->msg_prio, waitok, uap->abs_timeout);
2159 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2162 struct filedesc *fdp;
2166 struct mqueue_notifier *nt, *newnt = NULL;
2170 fdp = td->td_proc->p_fd;
2172 error = copyin(uap->sigev, &ev, sizeof(ev));
2175 if (ev.sigev_notify != SIGEV_SIGNAL &&
2176 ev.sigev_notify != SIGEV_THREAD_ID &&
2177 ev.sigev_notify != SIGEV_NONE)
2179 if ((ev.sigev_notify == SIGEV_SIGNAL ||
2180 ev.sigev_notify == SIGEV_THREAD_ID) &&
2181 !_SIG_VALID(ev.sigev_signo))
2184 error = getmq(td, uap->mqd, &fp, NULL, &mq);
2188 FILEDESC_SLOCK(fdp);
2189 if (fget_locked(fdp, uap->mqd) != fp) {
2190 FILEDESC_SUNLOCK(fdp);
2194 mtx_lock(&mq->mq_mutex);
2195 FILEDESC_SUNLOCK(fdp);
2196 if (uap->sigev != NULL) {
2197 if (mq->mq_notifier != NULL) {
2201 nt = notifier_search(p, uap->mqd);
2203 if (newnt == NULL) {
2205 mtx_unlock(&mq->mq_mutex);
2206 newnt = notifier_alloc();
2212 sigqueue_take(&nt->nt_ksi);
2213 if (newnt != NULL) {
2214 notifier_free(newnt);
2220 ksiginfo_init(&nt->nt_ksi);
2221 nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2222 nt->nt_ksi.ksi_code = SI_MESGQ;
2224 nt->nt_ksi.ksi_mqd = uap->mqd;
2225 notifier_insert(p, nt);
2228 mq->mq_notifier = nt;
2231 * if there is no receivers and message queue
2232 * is not empty, we should send notification
2233 * as soon as possible.
2235 if (mq->mq_receivers == 0 &&
2236 !TAILQ_EMPTY(&mq->mq_msgq))
2237 mqueue_send_notification(mq);
2240 notifier_remove(p, mq, uap->mqd);
2242 mtx_unlock(&mq->mq_mutex);
2247 notifier_free(newnt);
2252 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2254 struct filedesc *fdp;
2257 fdp = td->td_proc->p_fd;
2258 FILEDESC_LOCK_ASSERT(fdp);
2260 if (fp->f_ops == &mqueueops) {
2262 mtx_lock(&mq->mq_mutex);
2263 notifier_remove(td->td_proc, mq, fd);
2265 /* have to wakeup thread in same process */
2266 if (mq->mq_flags & MQ_RSEL) {
2267 mq->mq_flags &= ~MQ_RSEL;
2268 selwakeup(&mq->mq_rsel);
2270 if (mq->mq_flags & MQ_WSEL) {
2271 mq->mq_flags &= ~MQ_WSEL;
2272 selwakeup(&mq->mq_wsel);
2274 mtx_unlock(&mq->mq_mutex);
2279 mq_proc_exit(void *arg __unused, struct proc *p)
2281 struct filedesc *fdp;
2287 FILEDESC_SLOCK(fdp);
2288 for (i = 0; i < fdp->fd_nfiles; ++i) {
2289 fp = fget_locked(fdp, i);
2290 if (fp != NULL && fp->f_ops == &mqueueops) {
2292 mtx_lock(&mq->mq_mutex);
2293 notifier_remove(p, FPTOMQ(fp), i);
2294 mtx_unlock(&mq->mq_mutex);
2297 FILEDESC_SUNLOCK(fdp);
2298 KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2302 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2303 int flags, struct thread *td)
2305 return (EOPNOTSUPP);
2309 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2310 int flags, struct thread *td)
2312 return (EOPNOTSUPP);
2316 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred,
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_truncate = mqf_truncate,
2441 .fo_ioctl = mqf_ioctl,
2442 .fo_poll = mqf_poll,
2443 .fo_kqfilter = mqf_kqfilter,
2444 .fo_stat = mqf_stat,
2445 .fo_close = mqf_close
2448 static struct vop_vector mqfs_vnodeops = {
2449 .vop_default = &default_vnodeops,
2450 .vop_access = mqfs_access,
2451 .vop_cachedlookup = mqfs_lookup,
2452 .vop_lookup = vfs_cache_lookup,
2453 .vop_reclaim = mqfs_reclaim,
2454 .vop_create = mqfs_create,
2455 .vop_remove = mqfs_remove,
2456 .vop_inactive = mqfs_inactive,
2457 .vop_open = mqfs_open,
2458 .vop_close = mqfs_close,
2459 .vop_getattr = mqfs_getattr,
2460 .vop_setattr = mqfs_setattr,
2461 .vop_read = mqfs_read,
2462 .vop_write = VOP_EOPNOTSUPP,
2463 .vop_readdir = mqfs_readdir,
2464 .vop_mkdir = VOP_EOPNOTSUPP,
2465 .vop_rmdir = VOP_EOPNOTSUPP
2468 static struct vfsops mqfs_vfsops = {
2469 .vfs_init = mqfs_init,
2470 .vfs_uninit = mqfs_uninit,
2471 .vfs_mount = mqfs_mount,
2472 .vfs_unmount = mqfs_unmount,
2473 .vfs_root = mqfs_root,
2474 .vfs_statfs = mqfs_statfs,
2477 SYSCALL_MODULE_HELPER(kmq_open);
2478 SYSCALL_MODULE_HELPER(kmq_setattr);
2479 SYSCALL_MODULE_HELPER(kmq_timedsend);
2480 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2481 SYSCALL_MODULE_HELPER(kmq_notify);
2482 SYSCALL_MODULE_HELPER(kmq_unlink);
2484 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2485 MODULE_VERSION(mqueuefs, 1);