2 * POSIX message queues filesystem for Linux.
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
5 * Michal Wronski (michal.wronski@gmail.com)
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
13 * This file is released under the GPL.
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/fs_context.h>
22 #include <linux/namei.h>
23 #include <linux/sysctl.h>
24 #include <linux/poll.h>
25 #include <linux/mqueue.h>
26 #include <linux/msg.h>
27 #include <linux/skbuff.h>
28 #include <linux/vmalloc.h>
29 #include <linux/netlink.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
32 #include <linux/signal.h>
33 #include <linux/mutex.h>
34 #include <linux/nsproxy.h>
35 #include <linux/pid.h>
36 #include <linux/ipc_namespace.h>
37 #include <linux/user_namespace.h>
38 #include <linux/slab.h>
39 #include <linux/sched/wake_q.h>
40 #include <linux/sched/signal.h>
41 #include <linux/sched/user.h>
46 struct mqueue_fs_context {
47 struct ipc_namespace *ipc_ns;
50 #define MQUEUE_MAGIC 0x19800202
51 #define DIRENT_SIZE 20
52 #define FILENT_SIZE 80
60 struct posix_msg_tree_node {
61 struct rb_node rb_node;
62 struct list_head msg_list;
66 struct ext_wait_queue { /* queue of sleeping tasks */
67 struct task_struct *task;
68 struct list_head list;
69 struct msg_msg *msg; /* ptr of loaded message */
70 int state; /* one of STATE_* values */
73 struct mqueue_inode_info {
75 struct inode vfs_inode;
76 wait_queue_head_t wait_q;
78 struct rb_root msg_tree;
79 struct rb_node *msg_tree_rightmost;
80 struct posix_msg_tree_node *node_cache;
83 struct sigevent notify;
84 struct pid *notify_owner;
85 u32 notify_self_exec_id;
86 struct user_namespace *notify_user_ns;
87 struct user_struct *user; /* user who created, for accounting */
88 struct sock *notify_sock;
89 struct sk_buff *notify_cookie;
91 /* for tasks waiting for free space and messages, respectively */
92 struct ext_wait_queue e_wait_q[2];
94 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
97 static struct file_system_type mqueue_fs_type;
98 static const struct inode_operations mqueue_dir_inode_operations;
99 static const struct file_operations mqueue_file_operations;
100 static const struct super_operations mqueue_super_ops;
101 static const struct fs_context_operations mqueue_fs_context_ops;
102 static void remove_notification(struct mqueue_inode_info *info);
104 static struct kmem_cache *mqueue_inode_cachep;
106 static struct ctl_table_header *mq_sysctl_table;
108 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
110 return container_of(inode, struct mqueue_inode_info, vfs_inode);
114 * This routine should be called with the mq_lock held.
116 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
118 return get_ipc_ns(inode->i_sb->s_fs_info);
121 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
123 struct ipc_namespace *ns;
126 ns = __get_ns_from_inode(inode);
127 spin_unlock(&mq_lock);
131 /* Auxiliary functions to manipulate messages' list */
132 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
134 struct rb_node **p, *parent = NULL;
135 struct posix_msg_tree_node *leaf;
136 bool rightmost = true;
138 p = &info->msg_tree.rb_node;
141 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
143 if (likely(leaf->priority == msg->m_type))
145 else if (msg->m_type < leaf->priority) {
151 if (info->node_cache) {
152 leaf = info->node_cache;
153 info->node_cache = NULL;
155 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
158 INIT_LIST_HEAD(&leaf->msg_list);
160 leaf->priority = msg->m_type;
163 info->msg_tree_rightmost = &leaf->rb_node;
165 rb_link_node(&leaf->rb_node, parent, p);
166 rb_insert_color(&leaf->rb_node, &info->msg_tree);
168 info->attr.mq_curmsgs++;
169 info->qsize += msg->m_ts;
170 list_add_tail(&msg->m_list, &leaf->msg_list);
174 static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
175 struct mqueue_inode_info *info)
177 struct rb_node *node = &leaf->rb_node;
179 if (info->msg_tree_rightmost == node)
180 info->msg_tree_rightmost = rb_prev(node);
182 rb_erase(node, &info->msg_tree);
183 if (info->node_cache) {
186 info->node_cache = leaf;
190 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
192 struct rb_node *parent = NULL;
193 struct posix_msg_tree_node *leaf;
198 * During insert, low priorities go to the left and high to the
199 * right. On receive, we want the highest priorities first, so
200 * walk all the way to the right.
202 parent = info->msg_tree_rightmost;
204 if (info->attr.mq_curmsgs) {
205 pr_warn_once("Inconsistency in POSIX message queue, "
206 "no tree element, but supposedly messages "
208 info->attr.mq_curmsgs = 0;
212 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
213 if (unlikely(list_empty(&leaf->msg_list))) {
214 pr_warn_once("Inconsistency in POSIX message queue, "
215 "empty leaf node but we haven't implemented "
216 "lazy leaf delete!\n");
217 msg_tree_erase(leaf, info);
220 msg = list_first_entry(&leaf->msg_list,
221 struct msg_msg, m_list);
222 list_del(&msg->m_list);
223 if (list_empty(&leaf->msg_list)) {
224 msg_tree_erase(leaf, info);
227 info->attr.mq_curmsgs--;
228 info->qsize -= msg->m_ts;
232 static struct inode *mqueue_get_inode(struct super_block *sb,
233 struct ipc_namespace *ipc_ns, umode_t mode,
234 struct mq_attr *attr)
236 struct user_struct *u = current_user();
240 inode = new_inode(sb);
244 inode->i_ino = get_next_ino();
245 inode->i_mode = mode;
246 inode->i_uid = current_fsuid();
247 inode->i_gid = current_fsgid();
248 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
251 struct mqueue_inode_info *info;
252 unsigned long mq_bytes, mq_treesize;
254 inode->i_fop = &mqueue_file_operations;
255 inode->i_size = FILENT_SIZE;
256 /* mqueue specific info */
257 info = MQUEUE_I(inode);
258 spin_lock_init(&info->lock);
259 init_waitqueue_head(&info->wait_q);
260 INIT_LIST_HEAD(&info->e_wait_q[0].list);
261 INIT_LIST_HEAD(&info->e_wait_q[1].list);
262 info->notify_owner = NULL;
263 info->notify_user_ns = NULL;
265 info->user = NULL; /* set when all is ok */
266 info->msg_tree = RB_ROOT;
267 info->msg_tree_rightmost = NULL;
268 info->node_cache = NULL;
269 memset(&info->attr, 0, sizeof(info->attr));
270 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
271 ipc_ns->mq_msg_default);
272 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
273 ipc_ns->mq_msgsize_default);
275 info->attr.mq_maxmsg = attr->mq_maxmsg;
276 info->attr.mq_msgsize = attr->mq_msgsize;
279 * We used to allocate a static array of pointers and account
280 * the size of that array as well as one msg_msg struct per
281 * possible message into the queue size. That's no longer
282 * accurate as the queue is now an rbtree and will grow and
283 * shrink depending on usage patterns. We can, however, still
284 * account one msg_msg struct per message, but the nodes are
285 * allocated depending on priority usage, and most programs
286 * only use one, or a handful, of priorities. However, since
287 * this is pinned memory, we need to assume worst case, so
288 * that means the min(mq_maxmsg, max_priorities) * struct
289 * posix_msg_tree_node.
293 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
295 if (capable(CAP_SYS_RESOURCE)) {
296 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
297 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
300 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
301 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
305 /* check for overflow */
306 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
308 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
309 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
310 sizeof(struct posix_msg_tree_node);
311 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
312 if (mq_bytes + mq_treesize < mq_bytes)
314 mq_bytes += mq_treesize;
316 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
317 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
318 spin_unlock(&mq_lock);
319 /* mqueue_evict_inode() releases info->messages */
323 u->mq_bytes += mq_bytes;
324 spin_unlock(&mq_lock);
327 info->user = get_uid(u);
328 } else if (S_ISDIR(mode)) {
330 /* Some things misbehave if size == 0 on a directory */
331 inode->i_size = 2 * DIRENT_SIZE;
332 inode->i_op = &mqueue_dir_inode_operations;
333 inode->i_fop = &simple_dir_operations;
343 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
346 struct ipc_namespace *ns = sb->s_fs_info;
348 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
349 sb->s_blocksize = PAGE_SIZE;
350 sb->s_blocksize_bits = PAGE_SHIFT;
351 sb->s_magic = MQUEUE_MAGIC;
352 sb->s_op = &mqueue_super_ops;
354 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
356 return PTR_ERR(inode);
358 sb->s_root = d_make_root(inode);
364 static int mqueue_get_tree(struct fs_context *fc)
366 struct mqueue_fs_context *ctx = fc->fs_private;
368 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
371 static void mqueue_fs_context_free(struct fs_context *fc)
373 struct mqueue_fs_context *ctx = fc->fs_private;
375 put_ipc_ns(ctx->ipc_ns);
379 static int mqueue_init_fs_context(struct fs_context *fc)
381 struct mqueue_fs_context *ctx;
383 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
387 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
388 put_user_ns(fc->user_ns);
389 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
390 fc->fs_private = ctx;
391 fc->ops = &mqueue_fs_context_ops;
395 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
397 struct mqueue_fs_context *ctx;
398 struct fs_context *fc;
399 struct vfsmount *mnt;
401 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
405 ctx = fc->fs_private;
406 put_ipc_ns(ctx->ipc_ns);
407 ctx->ipc_ns = get_ipc_ns(ns);
408 put_user_ns(fc->user_ns);
409 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
416 static void init_once(void *foo)
418 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
420 inode_init_once(&p->vfs_inode);
423 static struct inode *mqueue_alloc_inode(struct super_block *sb)
425 struct mqueue_inode_info *ei;
427 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
430 return &ei->vfs_inode;
433 static void mqueue_free_inode(struct inode *inode)
435 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
438 static void mqueue_evict_inode(struct inode *inode)
440 struct mqueue_inode_info *info;
441 struct user_struct *user;
442 struct ipc_namespace *ipc_ns;
443 struct msg_msg *msg, *nmsg;
448 if (S_ISDIR(inode->i_mode))
451 ipc_ns = get_ns_from_inode(inode);
452 info = MQUEUE_I(inode);
453 spin_lock(&info->lock);
454 while ((msg = msg_get(info)) != NULL)
455 list_add_tail(&msg->m_list, &tmp_msg);
456 kfree(info->node_cache);
457 spin_unlock(&info->lock);
459 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
460 list_del(&msg->m_list);
466 unsigned long mq_bytes, mq_treesize;
468 /* Total amount of bytes accounted for the mqueue */
469 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
470 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
471 sizeof(struct posix_msg_tree_node);
473 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
474 info->attr.mq_msgsize);
477 user->mq_bytes -= mq_bytes;
479 * get_ns_from_inode() ensures that the
480 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
481 * to which we now hold a reference, or it is NULL.
482 * We can't put it here under mq_lock, though.
485 ipc_ns->mq_queues_count--;
486 spin_unlock(&mq_lock);
493 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
495 struct inode *dir = dentry->d_parent->d_inode;
497 struct mq_attr *attr = arg;
499 struct ipc_namespace *ipc_ns;
502 ipc_ns = __get_ns_from_inode(dir);
508 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
509 !capable(CAP_SYS_RESOURCE)) {
513 ipc_ns->mq_queues_count++;
514 spin_unlock(&mq_lock);
516 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
518 error = PTR_ERR(inode);
520 ipc_ns->mq_queues_count--;
525 dir->i_size += DIRENT_SIZE;
526 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
528 d_instantiate(dentry, inode);
532 spin_unlock(&mq_lock);
538 static int mqueue_create(struct inode *dir, struct dentry *dentry,
539 umode_t mode, bool excl)
541 return mqueue_create_attr(dentry, mode, NULL);
544 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
546 struct inode *inode = d_inode(dentry);
548 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
549 dir->i_size -= DIRENT_SIZE;
556 * This is routine for system read from queue file.
557 * To avoid mess with doing here some sort of mq_receive we allow
558 * to read only queue size & notification info (the only values
559 * that are interesting from user point of view and aren't accessible
560 * through std routines)
562 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
563 size_t count, loff_t *off)
565 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
566 char buffer[FILENT_SIZE];
569 spin_lock(&info->lock);
570 snprintf(buffer, sizeof(buffer),
571 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
573 info->notify_owner ? info->notify.sigev_notify : 0,
574 (info->notify_owner &&
575 info->notify.sigev_notify == SIGEV_SIGNAL) ?
576 info->notify.sigev_signo : 0,
577 pid_vnr(info->notify_owner));
578 spin_unlock(&info->lock);
579 buffer[sizeof(buffer)-1] = '\0';
581 ret = simple_read_from_buffer(u_data, count, off, buffer,
586 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
590 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
592 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
594 spin_lock(&info->lock);
595 if (task_tgid(current) == info->notify_owner)
596 remove_notification(info);
598 spin_unlock(&info->lock);
602 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
604 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
607 poll_wait(filp, &info->wait_q, poll_tab);
609 spin_lock(&info->lock);
610 if (info->attr.mq_curmsgs)
611 retval = EPOLLIN | EPOLLRDNORM;
613 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
614 retval |= EPOLLOUT | EPOLLWRNORM;
615 spin_unlock(&info->lock);
620 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
621 static void wq_add(struct mqueue_inode_info *info, int sr,
622 struct ext_wait_queue *ewp)
624 struct ext_wait_queue *walk;
626 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
627 if (walk->task->prio <= current->prio) {
628 list_add_tail(&ewp->list, &walk->list);
632 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
636 * Puts current task to sleep. Caller must hold queue lock. After return
640 static int wq_sleep(struct mqueue_inode_info *info, int sr,
641 ktime_t *timeout, struct ext_wait_queue *ewp)
642 __releases(&info->lock)
647 wq_add(info, sr, ewp);
650 __set_current_state(TASK_INTERRUPTIBLE);
652 spin_unlock(&info->lock);
653 time = schedule_hrtimeout_range_clock(timeout, 0,
654 HRTIMER_MODE_ABS, CLOCK_REALTIME);
656 if (ewp->state == STATE_READY) {
660 spin_lock(&info->lock);
661 if (ewp->state == STATE_READY) {
665 if (signal_pending(current)) {
666 retval = -ERESTARTSYS;
674 list_del(&ewp->list);
676 spin_unlock(&info->lock);
682 * Returns waiting task that should be serviced first or NULL if none exists
684 static struct ext_wait_queue *wq_get_first_waiter(
685 struct mqueue_inode_info *info, int sr)
687 struct list_head *ptr;
689 ptr = info->e_wait_q[sr].list.prev;
690 if (ptr == &info->e_wait_q[sr].list)
692 return list_entry(ptr, struct ext_wait_queue, list);
696 static inline void set_cookie(struct sk_buff *skb, char code)
698 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
702 * The next function is only to split too long sys_mq_timedsend
704 static void __do_notify(struct mqueue_inode_info *info)
707 * invoked when there is registered process and there isn't process
708 * waiting synchronously for message AND state of queue changed from
709 * empty to not empty. Here we are sure that no one is waiting
711 if (info->notify_owner &&
712 info->attr.mq_curmsgs == 1) {
713 switch (info->notify.sigev_notify) {
717 struct kernel_siginfo sig_i;
718 struct task_struct *task;
720 /* do_mq_notify() accepts sigev_signo == 0, why?? */
721 if (!info->notify.sigev_signo)
724 clear_siginfo(&sig_i);
725 sig_i.si_signo = info->notify.sigev_signo;
727 sig_i.si_code = SI_MESGQ;
728 sig_i.si_value = info->notify.sigev_value;
730 /* map current pid/uid into info->owner's namespaces */
731 sig_i.si_pid = task_tgid_nr_ns(current,
732 ns_of_pid(info->notify_owner));
733 sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
736 * We can't use kill_pid_info(), this signal should
737 * bypass check_kill_permission(). It is from kernel
738 * but si_fromuser() can't know this.
739 * We do check the self_exec_id, to avoid sending
740 * signals to programs that don't expect them.
742 task = pid_task(info->notify_owner, PIDTYPE_TGID);
743 if (task && task->self_exec_id ==
744 info->notify_self_exec_id) {
745 do_send_sig_info(info->notify.sigev_signo,
746 &sig_i, task, PIDTYPE_TGID);
752 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
753 netlink_sendskb(info->notify_sock, info->notify_cookie);
756 /* after notification unregisters process */
757 put_pid(info->notify_owner);
758 put_user_ns(info->notify_user_ns);
759 info->notify_owner = NULL;
760 info->notify_user_ns = NULL;
762 wake_up(&info->wait_q);
765 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
766 struct timespec64 *ts)
768 if (get_timespec64(ts, u_abs_timeout))
770 if (!timespec64_valid(ts))
775 static void remove_notification(struct mqueue_inode_info *info)
777 if (info->notify_owner != NULL &&
778 info->notify.sigev_notify == SIGEV_THREAD) {
779 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
780 netlink_sendskb(info->notify_sock, info->notify_cookie);
782 put_pid(info->notify_owner);
783 put_user_ns(info->notify_user_ns);
784 info->notify_owner = NULL;
785 info->notify_user_ns = NULL;
788 static int prepare_open(struct dentry *dentry, int oflag, int ro,
789 umode_t mode, struct filename *name,
790 struct mq_attr *attr)
792 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
793 MAY_READ | MAY_WRITE };
796 if (d_really_is_negative(dentry)) {
797 if (!(oflag & O_CREAT))
801 audit_inode_parent_hidden(name, dentry->d_parent);
802 return vfs_mkobj(dentry, mode & ~current_umask(),
803 mqueue_create_attr, attr);
805 /* it already existed */
806 audit_inode(name, dentry, 0);
807 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
809 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
811 acc = oflag2acc[oflag & O_ACCMODE];
812 return inode_permission(d_inode(dentry), acc);
815 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
816 struct mq_attr *attr)
818 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
819 struct dentry *root = mnt->mnt_root;
820 struct filename *name;
825 audit_mq_open(oflag, mode, attr);
827 if (IS_ERR(name = getname(u_name)))
828 return PTR_ERR(name);
830 fd = get_unused_fd_flags(O_CLOEXEC);
834 ro = mnt_want_write(mnt); /* we'll drop it in any case */
835 inode_lock(d_inode(root));
836 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
837 if (IS_ERR(path.dentry)) {
838 error = PTR_ERR(path.dentry);
841 path.mnt = mntget(mnt);
842 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
844 struct file *file = dentry_open(&path, oflag, current_cred());
846 fd_install(fd, file);
848 error = PTR_ERR(file);
856 inode_unlock(d_inode(root));
864 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
865 struct mq_attr __user *, u_attr)
868 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
871 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
874 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
877 struct filename *name;
878 struct dentry *dentry;
879 struct inode *inode = NULL;
880 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
881 struct vfsmount *mnt = ipc_ns->mq_mnt;
883 name = getname(u_name);
885 return PTR_ERR(name);
887 audit_inode_parent_hidden(name, mnt->mnt_root);
888 err = mnt_want_write(mnt);
891 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
892 dentry = lookup_one_len(name->name, mnt->mnt_root,
894 if (IS_ERR(dentry)) {
895 err = PTR_ERR(dentry);
899 inode = d_inode(dentry);
904 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
909 inode_unlock(d_inode(mnt->mnt_root));
919 /* Pipelined send and receive functions.
921 * If a receiver finds no waiting message, then it registers itself in the
922 * list of waiting receivers. A sender checks that list before adding the new
923 * message into the message array. If there is a waiting receiver, then it
924 * bypasses the message array and directly hands the message over to the
925 * receiver. The receiver accepts the message and returns without grabbing the
928 * - Set pointer to message.
929 * - Queue the receiver task for later wakeup (without the info->lock).
930 * - Update its state to STATE_READY. Now the receiver can continue.
931 * - Wake up the process after the lock is dropped. Should the process wake up
932 * before this wakeup (due to a timeout or a signal) it will either see
933 * STATE_READY and continue or acquire the lock to check the state again.
935 * The same algorithm is used for senders.
938 /* pipelined_send() - send a message directly to the task waiting in
939 * sys_mq_timedreceive() (without inserting message into a queue).
941 static inline void pipelined_send(struct wake_q_head *wake_q,
942 struct mqueue_inode_info *info,
943 struct msg_msg *message,
944 struct ext_wait_queue *receiver)
946 receiver->msg = message;
947 list_del(&receiver->list);
948 wake_q_add(wake_q, receiver->task);
950 * Rely on the implicit cmpxchg barrier from wake_q_add such
951 * that we can ensure that updating receiver->state is the last
952 * write operation: As once set, the receiver can continue,
953 * and if we don't have the reference count from the wake_q,
954 * yet, at that point we can later have a use-after-free
955 * condition and bogus wakeup.
957 receiver->state = STATE_READY;
960 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
961 * gets its message and put to the queue (we have one free place for sure). */
962 static inline void pipelined_receive(struct wake_q_head *wake_q,
963 struct mqueue_inode_info *info)
965 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
969 wake_up_interruptible(&info->wait_q);
972 if (msg_insert(sender->msg, info))
975 list_del(&sender->list);
976 wake_q_add(wake_q, sender->task);
977 sender->state = STATE_READY;
980 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
981 size_t msg_len, unsigned int msg_prio,
982 struct timespec64 *ts)
986 struct ext_wait_queue wait;
987 struct ext_wait_queue *receiver;
988 struct msg_msg *msg_ptr;
989 struct mqueue_inode_info *info;
990 ktime_t expires, *timeout = NULL;
991 struct posix_msg_tree_node *new_leaf = NULL;
993 DEFINE_WAKE_Q(wake_q);
995 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
999 expires = timespec64_to_ktime(*ts);
1003 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1006 if (unlikely(!f.file)) {
1011 inode = file_inode(f.file);
1012 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1016 info = MQUEUE_I(inode);
1019 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1024 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1029 /* First try to allocate memory, before doing anything with
1030 * existing queues. */
1031 msg_ptr = load_msg(u_msg_ptr, msg_len);
1032 if (IS_ERR(msg_ptr)) {
1033 ret = PTR_ERR(msg_ptr);
1036 msg_ptr->m_ts = msg_len;
1037 msg_ptr->m_type = msg_prio;
1040 * msg_insert really wants us to have a valid, spare node struct so
1041 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1042 * fall back to that if necessary.
1044 if (!info->node_cache)
1045 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1047 spin_lock(&info->lock);
1049 if (!info->node_cache && new_leaf) {
1050 /* Save our speculative allocation into the cache */
1051 INIT_LIST_HEAD(&new_leaf->msg_list);
1052 info->node_cache = new_leaf;
1058 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1059 if (f.file->f_flags & O_NONBLOCK) {
1062 wait.task = current;
1063 wait.msg = (void *) msg_ptr;
1064 wait.state = STATE_NONE;
1065 ret = wq_sleep(info, SEND, timeout, &wait);
1067 * wq_sleep must be called with info->lock held, and
1068 * returns with the lock released
1073 receiver = wq_get_first_waiter(info, RECV);
1075 pipelined_send(&wake_q, info, msg_ptr, receiver);
1077 /* adds message to the queue */
1078 ret = msg_insert(msg_ptr, info);
1083 inode->i_atime = inode->i_mtime = inode->i_ctime =
1084 current_time(inode);
1087 spin_unlock(&info->lock);
1098 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1099 size_t msg_len, unsigned int __user *u_msg_prio,
1100 struct timespec64 *ts)
1103 struct msg_msg *msg_ptr;
1105 struct inode *inode;
1106 struct mqueue_inode_info *info;
1107 struct ext_wait_queue wait;
1108 ktime_t expires, *timeout = NULL;
1109 struct posix_msg_tree_node *new_leaf = NULL;
1112 expires = timespec64_to_ktime(*ts);
1116 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1119 if (unlikely(!f.file)) {
1124 inode = file_inode(f.file);
1125 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1129 info = MQUEUE_I(inode);
1132 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1137 /* checks if buffer is big enough */
1138 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1144 * msg_insert really wants us to have a valid, spare node struct so
1145 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1146 * fall back to that if necessary.
1148 if (!info->node_cache)
1149 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1151 spin_lock(&info->lock);
1153 if (!info->node_cache && new_leaf) {
1154 /* Save our speculative allocation into the cache */
1155 INIT_LIST_HEAD(&new_leaf->msg_list);
1156 info->node_cache = new_leaf;
1161 if (info->attr.mq_curmsgs == 0) {
1162 if (f.file->f_flags & O_NONBLOCK) {
1163 spin_unlock(&info->lock);
1166 wait.task = current;
1167 wait.state = STATE_NONE;
1168 ret = wq_sleep(info, RECV, timeout, &wait);
1172 DEFINE_WAKE_Q(wake_q);
1174 msg_ptr = msg_get(info);
1176 inode->i_atime = inode->i_mtime = inode->i_ctime =
1177 current_time(inode);
1179 /* There is now free space in queue. */
1180 pipelined_receive(&wake_q, info);
1181 spin_unlock(&info->lock);
1186 ret = msg_ptr->m_ts;
1188 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1189 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1200 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1201 size_t, msg_len, unsigned int, msg_prio,
1202 const struct __kernel_timespec __user *, u_abs_timeout)
1204 struct timespec64 ts, *p = NULL;
1205 if (u_abs_timeout) {
1206 int res = prepare_timeout(u_abs_timeout, &ts);
1211 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1214 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1215 size_t, msg_len, unsigned int __user *, u_msg_prio,
1216 const struct __kernel_timespec __user *, u_abs_timeout)
1218 struct timespec64 ts, *p = NULL;
1219 if (u_abs_timeout) {
1220 int res = prepare_timeout(u_abs_timeout, &ts);
1225 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1229 * Notes: the case when user wants us to deregister (with NULL as pointer)
1230 * and he isn't currently owner of notification, will be silently discarded.
1231 * It isn't explicitly defined in the POSIX.
1233 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1238 struct inode *inode;
1239 struct mqueue_inode_info *info;
1242 audit_mq_notify(mqdes, notification);
1246 if (notification != NULL) {
1247 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1248 notification->sigev_notify != SIGEV_SIGNAL &&
1249 notification->sigev_notify != SIGEV_THREAD))
1251 if (notification->sigev_notify == SIGEV_SIGNAL &&
1252 !valid_signal(notification->sigev_signo)) {
1255 if (notification->sigev_notify == SIGEV_THREAD) {
1258 /* create the notify skb */
1259 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1263 if (copy_from_user(nc->data,
1264 notification->sigev_value.sival_ptr,
1265 NOTIFY_COOKIE_LEN)) {
1270 /* TODO: add a header? */
1271 skb_put(nc, NOTIFY_COOKIE_LEN);
1272 /* and attach it to the socket */
1274 f = fdget(notification->sigev_signo);
1279 sock = netlink_getsockbyfilp(f.file);
1282 ret = PTR_ERR(sock);
1286 timeo = MAX_SCHEDULE_TIMEOUT;
1287 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1303 inode = file_inode(f.file);
1304 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1308 info = MQUEUE_I(inode);
1311 spin_lock(&info->lock);
1312 if (notification == NULL) {
1313 if (info->notify_owner == task_tgid(current)) {
1314 remove_notification(info);
1315 inode->i_atime = inode->i_ctime = current_time(inode);
1317 } else if (info->notify_owner != NULL) {
1320 switch (notification->sigev_notify) {
1322 info->notify.sigev_notify = SIGEV_NONE;
1325 info->notify_sock = sock;
1326 info->notify_cookie = nc;
1329 info->notify.sigev_notify = SIGEV_THREAD;
1332 info->notify.sigev_signo = notification->sigev_signo;
1333 info->notify.sigev_value = notification->sigev_value;
1334 info->notify.sigev_notify = SIGEV_SIGNAL;
1335 info->notify_self_exec_id = current->self_exec_id;
1339 info->notify_owner = get_pid(task_tgid(current));
1340 info->notify_user_ns = get_user_ns(current_user_ns());
1341 inode->i_atime = inode->i_ctime = current_time(inode);
1343 spin_unlock(&info->lock);
1348 netlink_detachskb(sock, nc);
1356 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1357 const struct sigevent __user *, u_notification)
1359 struct sigevent n, *p = NULL;
1360 if (u_notification) {
1361 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1365 return do_mq_notify(mqdes, p);
1368 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1371 struct inode *inode;
1372 struct mqueue_inode_info *info;
1374 if (new && (new->mq_flags & (~O_NONBLOCK)))
1381 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1386 inode = file_inode(f.file);
1387 info = MQUEUE_I(inode);
1389 spin_lock(&info->lock);
1393 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1396 audit_mq_getsetattr(mqdes, new);
1397 spin_lock(&f.file->f_lock);
1398 if (new->mq_flags & O_NONBLOCK)
1399 f.file->f_flags |= O_NONBLOCK;
1401 f.file->f_flags &= ~O_NONBLOCK;
1402 spin_unlock(&f.file->f_lock);
1404 inode->i_atime = inode->i_ctime = current_time(inode);
1407 spin_unlock(&info->lock);
1412 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1413 const struct mq_attr __user *, u_mqstat,
1414 struct mq_attr __user *, u_omqstat)
1417 struct mq_attr mqstat, omqstat;
1418 struct mq_attr *new = NULL, *old = NULL;
1422 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1428 ret = do_mq_getsetattr(mqdes, new, old);
1432 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1437 #ifdef CONFIG_COMPAT
1439 struct compat_mq_attr {
1440 compat_long_t mq_flags; /* message queue flags */
1441 compat_long_t mq_maxmsg; /* maximum number of messages */
1442 compat_long_t mq_msgsize; /* maximum message size */
1443 compat_long_t mq_curmsgs; /* number of messages currently queued */
1444 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1447 static inline int get_compat_mq_attr(struct mq_attr *attr,
1448 const struct compat_mq_attr __user *uattr)
1450 struct compat_mq_attr v;
1452 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1455 memset(attr, 0, sizeof(*attr));
1456 attr->mq_flags = v.mq_flags;
1457 attr->mq_maxmsg = v.mq_maxmsg;
1458 attr->mq_msgsize = v.mq_msgsize;
1459 attr->mq_curmsgs = v.mq_curmsgs;
1463 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1464 struct compat_mq_attr __user *uattr)
1466 struct compat_mq_attr v;
1468 memset(&v, 0, sizeof(v));
1469 v.mq_flags = attr->mq_flags;
1470 v.mq_maxmsg = attr->mq_maxmsg;
1471 v.mq_msgsize = attr->mq_msgsize;
1472 v.mq_curmsgs = attr->mq_curmsgs;
1473 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1478 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1479 int, oflag, compat_mode_t, mode,
1480 struct compat_mq_attr __user *, u_attr)
1482 struct mq_attr attr, *p = NULL;
1483 if (u_attr && oflag & O_CREAT) {
1485 if (get_compat_mq_attr(&attr, u_attr))
1488 return do_mq_open(u_name, oflag, mode, p);
1491 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1492 const struct compat_sigevent __user *, u_notification)
1494 struct sigevent n, *p = NULL;
1495 if (u_notification) {
1496 if (get_compat_sigevent(&n, u_notification))
1498 if (n.sigev_notify == SIGEV_THREAD)
1499 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1502 return do_mq_notify(mqdes, p);
1505 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1506 const struct compat_mq_attr __user *, u_mqstat,
1507 struct compat_mq_attr __user *, u_omqstat)
1510 struct mq_attr mqstat, omqstat;
1511 struct mq_attr *new = NULL, *old = NULL;
1515 if (get_compat_mq_attr(new, u_mqstat))
1521 ret = do_mq_getsetattr(mqdes, new, old);
1525 if (put_compat_mq_attr(old, u_omqstat))
1531 #ifdef CONFIG_COMPAT_32BIT_TIME
1532 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1533 struct timespec64 *ts)
1535 if (get_old_timespec32(ts, p))
1537 if (!timespec64_valid(ts))
1542 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1543 const char __user *, u_msg_ptr,
1544 unsigned int, msg_len, unsigned int, msg_prio,
1545 const struct old_timespec32 __user *, u_abs_timeout)
1547 struct timespec64 ts, *p = NULL;
1548 if (u_abs_timeout) {
1549 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1554 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1557 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1558 char __user *, u_msg_ptr,
1559 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1560 const struct old_timespec32 __user *, u_abs_timeout)
1562 struct timespec64 ts, *p = NULL;
1563 if (u_abs_timeout) {
1564 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1569 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1573 static const struct inode_operations mqueue_dir_inode_operations = {
1574 .lookup = simple_lookup,
1575 .create = mqueue_create,
1576 .unlink = mqueue_unlink,
1579 static const struct file_operations mqueue_file_operations = {
1580 .flush = mqueue_flush_file,
1581 .poll = mqueue_poll_file,
1582 .read = mqueue_read_file,
1583 .llseek = default_llseek,
1586 static const struct super_operations mqueue_super_ops = {
1587 .alloc_inode = mqueue_alloc_inode,
1588 .free_inode = mqueue_free_inode,
1589 .evict_inode = mqueue_evict_inode,
1590 .statfs = simple_statfs,
1593 static const struct fs_context_operations mqueue_fs_context_ops = {
1594 .free = mqueue_fs_context_free,
1595 .get_tree = mqueue_get_tree,
1598 static struct file_system_type mqueue_fs_type = {
1600 .init_fs_context = mqueue_init_fs_context,
1601 .kill_sb = kill_litter_super,
1602 .fs_flags = FS_USERNS_MOUNT,
1605 int mq_init_ns(struct ipc_namespace *ns)
1609 ns->mq_queues_count = 0;
1610 ns->mq_queues_max = DFLT_QUEUESMAX;
1611 ns->mq_msg_max = DFLT_MSGMAX;
1612 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1613 ns->mq_msg_default = DFLT_MSG;
1614 ns->mq_msgsize_default = DFLT_MSGSIZE;
1616 m = mq_create_mount(ns);
1623 void mq_clear_sbinfo(struct ipc_namespace *ns)
1625 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1628 void mq_put_mnt(struct ipc_namespace *ns)
1630 kern_unmount(ns->mq_mnt);
1633 static int __init init_mqueue_fs(void)
1637 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1638 sizeof(struct mqueue_inode_info), 0,
1639 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1640 if (mqueue_inode_cachep == NULL)
1643 /* ignore failures - they are not fatal */
1644 mq_sysctl_table = mq_register_sysctl_table();
1646 error = register_filesystem(&mqueue_fs_type);
1650 spin_lock_init(&mq_lock);
1652 error = mq_init_ns(&init_ipc_ns);
1654 goto out_filesystem;
1659 unregister_filesystem(&mqueue_fs_type);
1661 if (mq_sysctl_table)
1662 unregister_sysctl_table(mq_sysctl_table);
1663 kmem_cache_destroy(mqueue_inode_cachep);
1667 device_initcall(init_mqueue_fs);