2 * POSIX message queues filesystem for Linux.
8 * Lockless receive & send, fd based notify:
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 posix_msg_tree_node *node_cache;
82 struct sigevent notify;
83 struct pid *notify_owner;
84 struct user_namespace *notify_user_ns;
85 struct user_struct *user; /* user who created, for accounting */
86 struct sock *notify_sock;
87 struct sk_buff *notify_cookie;
89 /* for tasks waiting for free space and messages, respectively */
90 struct ext_wait_queue e_wait_q[2];
92 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
95 static struct file_system_type mqueue_fs_type;
96 static const struct inode_operations mqueue_dir_inode_operations;
97 static const struct file_operations mqueue_file_operations;
98 static const struct super_operations mqueue_super_ops;
99 static const struct fs_context_operations mqueue_fs_context_ops;
100 static void remove_notification(struct mqueue_inode_info *info);
102 static struct kmem_cache *mqueue_inode_cachep;
104 static struct ctl_table_header *mq_sysctl_table;
106 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
108 return container_of(inode, struct mqueue_inode_info, vfs_inode);
112 * This routine should be called with the mq_lock held.
114 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
116 return get_ipc_ns(inode->i_sb->s_fs_info);
119 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
121 struct ipc_namespace *ns;
124 ns = __get_ns_from_inode(inode);
125 spin_unlock(&mq_lock);
129 /* Auxiliary functions to manipulate messages' list */
130 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
132 struct rb_node **p, *parent = NULL;
133 struct posix_msg_tree_node *leaf;
135 p = &info->msg_tree.rb_node;
138 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
140 if (likely(leaf->priority == msg->m_type))
142 else if (msg->m_type < leaf->priority)
147 if (info->node_cache) {
148 leaf = info->node_cache;
149 info->node_cache = NULL;
151 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
154 INIT_LIST_HEAD(&leaf->msg_list);
156 leaf->priority = msg->m_type;
157 rb_link_node(&leaf->rb_node, parent, p);
158 rb_insert_color(&leaf->rb_node, &info->msg_tree);
160 info->attr.mq_curmsgs++;
161 info->qsize += msg->m_ts;
162 list_add_tail(&msg->m_list, &leaf->msg_list);
166 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
168 struct rb_node **p, *parent = NULL;
169 struct posix_msg_tree_node *leaf;
173 p = &info->msg_tree.rb_node;
177 * During insert, low priorities go to the left and high to the
178 * right. On receive, we want the highest priorities first, so
179 * walk all the way to the right.
184 if (info->attr.mq_curmsgs) {
185 pr_warn_once("Inconsistency in POSIX message queue, "
186 "no tree element, but supposedly messages "
188 info->attr.mq_curmsgs = 0;
192 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
193 if (unlikely(list_empty(&leaf->msg_list))) {
194 pr_warn_once("Inconsistency in POSIX message queue, "
195 "empty leaf node but we haven't implemented "
196 "lazy leaf delete!\n");
197 rb_erase(&leaf->rb_node, &info->msg_tree);
198 if (info->node_cache) {
201 info->node_cache = leaf;
205 msg = list_first_entry(&leaf->msg_list,
206 struct msg_msg, m_list);
207 list_del(&msg->m_list);
208 if (list_empty(&leaf->msg_list)) {
209 rb_erase(&leaf->rb_node, &info->msg_tree);
210 if (info->node_cache) {
213 info->node_cache = leaf;
217 info->attr.mq_curmsgs--;
218 info->qsize -= msg->m_ts;
222 static struct inode *mqueue_get_inode(struct super_block *sb,
223 struct ipc_namespace *ipc_ns, umode_t mode,
224 struct mq_attr *attr)
226 struct user_struct *u = current_user();
230 inode = new_inode(sb);
234 inode->i_ino = get_next_ino();
235 inode->i_mode = mode;
236 inode->i_uid = current_fsuid();
237 inode->i_gid = current_fsgid();
238 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
241 struct mqueue_inode_info *info;
242 unsigned long mq_bytes, mq_treesize;
244 inode->i_fop = &mqueue_file_operations;
245 inode->i_size = FILENT_SIZE;
246 /* mqueue specific info */
247 info = MQUEUE_I(inode);
248 spin_lock_init(&info->lock);
249 init_waitqueue_head(&info->wait_q);
250 INIT_LIST_HEAD(&info->e_wait_q[0].list);
251 INIT_LIST_HEAD(&info->e_wait_q[1].list);
252 info->notify_owner = NULL;
253 info->notify_user_ns = NULL;
255 info->user = NULL; /* set when all is ok */
256 info->msg_tree = RB_ROOT;
257 info->node_cache = NULL;
258 memset(&info->attr, 0, sizeof(info->attr));
259 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
260 ipc_ns->mq_msg_default);
261 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
262 ipc_ns->mq_msgsize_default);
264 info->attr.mq_maxmsg = attr->mq_maxmsg;
265 info->attr.mq_msgsize = attr->mq_msgsize;
268 * We used to allocate a static array of pointers and account
269 * the size of that array as well as one msg_msg struct per
270 * possible message into the queue size. That's no longer
271 * accurate as the queue is now an rbtree and will grow and
272 * shrink depending on usage patterns. We can, however, still
273 * account one msg_msg struct per message, but the nodes are
274 * allocated depending on priority usage, and most programs
275 * only use one, or a handful, of priorities. However, since
276 * this is pinned memory, we need to assume worst case, so
277 * that means the min(mq_maxmsg, max_priorities) * struct
278 * posix_msg_tree_node.
282 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
284 if (capable(CAP_SYS_RESOURCE)) {
285 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
286 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
289 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
290 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
294 /* check for overflow */
295 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
297 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
298 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
299 sizeof(struct posix_msg_tree_node);
300 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
301 if (mq_bytes + mq_treesize < mq_bytes)
303 mq_bytes += mq_treesize;
305 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
306 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
307 spin_unlock(&mq_lock);
308 /* mqueue_evict_inode() releases info->messages */
312 u->mq_bytes += mq_bytes;
313 spin_unlock(&mq_lock);
316 info->user = get_uid(u);
317 } else if (S_ISDIR(mode)) {
319 /* Some things misbehave if size == 0 on a directory */
320 inode->i_size = 2 * DIRENT_SIZE;
321 inode->i_op = &mqueue_dir_inode_operations;
322 inode->i_fop = &simple_dir_operations;
332 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
335 struct ipc_namespace *ns = sb->s_fs_info;
337 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
338 sb->s_blocksize = PAGE_SIZE;
339 sb->s_blocksize_bits = PAGE_SHIFT;
340 sb->s_magic = MQUEUE_MAGIC;
341 sb->s_op = &mqueue_super_ops;
343 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
345 return PTR_ERR(inode);
347 sb->s_root = d_make_root(inode);
353 static int mqueue_get_tree(struct fs_context *fc)
355 struct mqueue_fs_context *ctx = fc->fs_private;
357 put_user_ns(fc->user_ns);
358 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
359 fc->s_fs_info = ctx->ipc_ns;
360 return vfs_get_super(fc, vfs_get_keyed_super, mqueue_fill_super);
363 static void mqueue_fs_context_free(struct fs_context *fc)
365 struct mqueue_fs_context *ctx = fc->fs_private;
368 put_ipc_ns(ctx->ipc_ns);
372 static int mqueue_init_fs_context(struct fs_context *fc)
374 struct mqueue_fs_context *ctx;
376 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
380 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
381 fc->fs_private = ctx;
382 fc->ops = &mqueue_fs_context_ops;
386 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
388 struct mqueue_fs_context *ctx;
389 struct fs_context *fc;
390 struct vfsmount *mnt;
392 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
396 ctx = fc->fs_private;
397 put_ipc_ns(ctx->ipc_ns);
398 ctx->ipc_ns = get_ipc_ns(ns);
405 static void init_once(void *foo)
407 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
409 inode_init_once(&p->vfs_inode);
412 static struct inode *mqueue_alloc_inode(struct super_block *sb)
414 struct mqueue_inode_info *ei;
416 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
419 return &ei->vfs_inode;
422 static void mqueue_free_inode(struct inode *inode)
424 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
427 static void mqueue_evict_inode(struct inode *inode)
429 struct mqueue_inode_info *info;
430 struct user_struct *user;
431 unsigned long mq_bytes, mq_treesize;
432 struct ipc_namespace *ipc_ns;
437 if (S_ISDIR(inode->i_mode))
440 ipc_ns = get_ns_from_inode(inode);
441 info = MQUEUE_I(inode);
442 spin_lock(&info->lock);
443 while ((msg = msg_get(info)) != NULL)
445 kfree(info->node_cache);
446 spin_unlock(&info->lock);
448 /* Total amount of bytes accounted for the mqueue */
449 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
450 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
451 sizeof(struct posix_msg_tree_node);
453 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
454 info->attr.mq_msgsize);
459 user->mq_bytes -= mq_bytes;
461 * get_ns_from_inode() ensures that the
462 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
463 * to which we now hold a reference, or it is NULL.
464 * We can't put it here under mq_lock, though.
467 ipc_ns->mq_queues_count--;
468 spin_unlock(&mq_lock);
475 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
477 struct inode *dir = dentry->d_parent->d_inode;
479 struct mq_attr *attr = arg;
481 struct ipc_namespace *ipc_ns;
484 ipc_ns = __get_ns_from_inode(dir);
490 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
491 !capable(CAP_SYS_RESOURCE)) {
495 ipc_ns->mq_queues_count++;
496 spin_unlock(&mq_lock);
498 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
500 error = PTR_ERR(inode);
502 ipc_ns->mq_queues_count--;
507 dir->i_size += DIRENT_SIZE;
508 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
510 d_instantiate(dentry, inode);
514 spin_unlock(&mq_lock);
520 static int mqueue_create(struct inode *dir, struct dentry *dentry,
521 umode_t mode, bool excl)
523 return mqueue_create_attr(dentry, mode, NULL);
526 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
528 struct inode *inode = d_inode(dentry);
530 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
531 dir->i_size -= DIRENT_SIZE;
538 * This is routine for system read from queue file.
539 * To avoid mess with doing here some sort of mq_receive we allow
540 * to read only queue size & notification info (the only values
541 * that are interesting from user point of view and aren't accessible
542 * through std routines)
544 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
545 size_t count, loff_t *off)
547 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
548 char buffer[FILENT_SIZE];
551 spin_lock(&info->lock);
552 snprintf(buffer, sizeof(buffer),
553 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
555 info->notify_owner ? info->notify.sigev_notify : 0,
556 (info->notify_owner &&
557 info->notify.sigev_notify == SIGEV_SIGNAL) ?
558 info->notify.sigev_signo : 0,
559 pid_vnr(info->notify_owner));
560 spin_unlock(&info->lock);
561 buffer[sizeof(buffer)-1] = '\0';
563 ret = simple_read_from_buffer(u_data, count, off, buffer,
568 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
572 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
574 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
576 spin_lock(&info->lock);
577 if (task_tgid(current) == info->notify_owner)
578 remove_notification(info);
580 spin_unlock(&info->lock);
584 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
586 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
589 poll_wait(filp, &info->wait_q, poll_tab);
591 spin_lock(&info->lock);
592 if (info->attr.mq_curmsgs)
593 retval = EPOLLIN | EPOLLRDNORM;
595 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
596 retval |= EPOLLOUT | EPOLLWRNORM;
597 spin_unlock(&info->lock);
602 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
603 static void wq_add(struct mqueue_inode_info *info, int sr,
604 struct ext_wait_queue *ewp)
606 struct ext_wait_queue *walk;
610 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
611 if (walk->task->prio <= current->prio) {
612 list_add_tail(&ewp->list, &walk->list);
616 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
620 * Puts current task to sleep. Caller must hold queue lock. After return
624 static int wq_sleep(struct mqueue_inode_info *info, int sr,
625 ktime_t *timeout, struct ext_wait_queue *ewp)
626 __releases(&info->lock)
631 wq_add(info, sr, ewp);
634 __set_current_state(TASK_INTERRUPTIBLE);
636 spin_unlock(&info->lock);
637 time = schedule_hrtimeout_range_clock(timeout, 0,
638 HRTIMER_MODE_ABS, CLOCK_REALTIME);
640 if (ewp->state == STATE_READY) {
644 spin_lock(&info->lock);
645 if (ewp->state == STATE_READY) {
649 if (signal_pending(current)) {
650 retval = -ERESTARTSYS;
658 list_del(&ewp->list);
660 spin_unlock(&info->lock);
666 * Returns waiting task that should be serviced first or NULL if none exists
668 static struct ext_wait_queue *wq_get_first_waiter(
669 struct mqueue_inode_info *info, int sr)
671 struct list_head *ptr;
673 ptr = info->e_wait_q[sr].list.prev;
674 if (ptr == &info->e_wait_q[sr].list)
676 return list_entry(ptr, struct ext_wait_queue, list);
680 static inline void set_cookie(struct sk_buff *skb, char code)
682 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
686 * The next function is only to split too long sys_mq_timedsend
688 static void __do_notify(struct mqueue_inode_info *info)
691 * invoked when there is registered process and there isn't process
692 * waiting synchronously for message AND state of queue changed from
693 * empty to not empty. Here we are sure that no one is waiting
695 if (info->notify_owner &&
696 info->attr.mq_curmsgs == 1) {
697 struct kernel_siginfo sig_i;
698 switch (info->notify.sigev_notify) {
704 clear_siginfo(&sig_i);
705 sig_i.si_signo = info->notify.sigev_signo;
707 sig_i.si_code = SI_MESGQ;
708 sig_i.si_value = info->notify.sigev_value;
709 /* map current pid/uid into info->owner's namespaces */
711 sig_i.si_pid = task_tgid_nr_ns(current,
712 ns_of_pid(info->notify_owner));
713 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
716 kill_pid_info(info->notify.sigev_signo,
717 &sig_i, info->notify_owner);
720 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
721 netlink_sendskb(info->notify_sock, info->notify_cookie);
724 /* after notification unregisters process */
725 put_pid(info->notify_owner);
726 put_user_ns(info->notify_user_ns);
727 info->notify_owner = NULL;
728 info->notify_user_ns = NULL;
730 wake_up(&info->wait_q);
733 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
734 struct timespec64 *ts)
736 if (get_timespec64(ts, u_abs_timeout))
738 if (!timespec64_valid(ts))
743 static void remove_notification(struct mqueue_inode_info *info)
745 if (info->notify_owner != NULL &&
746 info->notify.sigev_notify == SIGEV_THREAD) {
747 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
748 netlink_sendskb(info->notify_sock, info->notify_cookie);
750 put_pid(info->notify_owner);
751 put_user_ns(info->notify_user_ns);
752 info->notify_owner = NULL;
753 info->notify_user_ns = NULL;
756 static int prepare_open(struct dentry *dentry, int oflag, int ro,
757 umode_t mode, struct filename *name,
758 struct mq_attr *attr)
760 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
761 MAY_READ | MAY_WRITE };
764 if (d_really_is_negative(dentry)) {
765 if (!(oflag & O_CREAT))
769 audit_inode_parent_hidden(name, dentry->d_parent);
770 return vfs_mkobj(dentry, mode & ~current_umask(),
771 mqueue_create_attr, attr);
773 /* it already existed */
774 audit_inode(name, dentry, 0);
775 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
777 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
779 acc = oflag2acc[oflag & O_ACCMODE];
780 return inode_permission(d_inode(dentry), acc);
783 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
784 struct mq_attr *attr)
786 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
787 struct dentry *root = mnt->mnt_root;
788 struct filename *name;
793 audit_mq_open(oflag, mode, attr);
795 if (IS_ERR(name = getname(u_name)))
796 return PTR_ERR(name);
798 fd = get_unused_fd_flags(O_CLOEXEC);
802 ro = mnt_want_write(mnt); /* we'll drop it in any case */
803 inode_lock(d_inode(root));
804 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
805 if (IS_ERR(path.dentry)) {
806 error = PTR_ERR(path.dentry);
809 path.mnt = mntget(mnt);
810 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
812 struct file *file = dentry_open(&path, oflag, current_cred());
814 fd_install(fd, file);
816 error = PTR_ERR(file);
824 inode_unlock(d_inode(root));
832 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
833 struct mq_attr __user *, u_attr)
836 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
839 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
842 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
845 struct filename *name;
846 struct dentry *dentry;
847 struct inode *inode = NULL;
848 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
849 struct vfsmount *mnt = ipc_ns->mq_mnt;
851 name = getname(u_name);
853 return PTR_ERR(name);
855 audit_inode_parent_hidden(name, mnt->mnt_root);
856 err = mnt_want_write(mnt);
859 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
860 dentry = lookup_one_len(name->name, mnt->mnt_root,
862 if (IS_ERR(dentry)) {
863 err = PTR_ERR(dentry);
867 inode = d_inode(dentry);
872 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
877 inode_unlock(d_inode(mnt->mnt_root));
887 /* Pipelined send and receive functions.
889 * If a receiver finds no waiting message, then it registers itself in the
890 * list of waiting receivers. A sender checks that list before adding the new
891 * message into the message array. If there is a waiting receiver, then it
892 * bypasses the message array and directly hands the message over to the
893 * receiver. The receiver accepts the message and returns without grabbing the
896 * - Set pointer to message.
897 * - Queue the receiver task for later wakeup (without the info->lock).
898 * - Update its state to STATE_READY. Now the receiver can continue.
899 * - Wake up the process after the lock is dropped. Should the process wake up
900 * before this wakeup (due to a timeout or a signal) it will either see
901 * STATE_READY and continue or acquire the lock to check the state again.
903 * The same algorithm is used for senders.
906 /* pipelined_send() - send a message directly to the task waiting in
907 * sys_mq_timedreceive() (without inserting message into a queue).
909 static inline void pipelined_send(struct wake_q_head *wake_q,
910 struct mqueue_inode_info *info,
911 struct msg_msg *message,
912 struct ext_wait_queue *receiver)
914 receiver->msg = message;
915 list_del(&receiver->list);
916 wake_q_add(wake_q, receiver->task);
918 * Rely on the implicit cmpxchg barrier from wake_q_add such
919 * that we can ensure that updating receiver->state is the last
920 * write operation: As once set, the receiver can continue,
921 * and if we don't have the reference count from the wake_q,
922 * yet, at that point we can later have a use-after-free
923 * condition and bogus wakeup.
925 receiver->state = STATE_READY;
928 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
929 * gets its message and put to the queue (we have one free place for sure). */
930 static inline void pipelined_receive(struct wake_q_head *wake_q,
931 struct mqueue_inode_info *info)
933 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
937 wake_up_interruptible(&info->wait_q);
940 if (msg_insert(sender->msg, info))
943 list_del(&sender->list);
944 wake_q_add(wake_q, sender->task);
945 sender->state = STATE_READY;
948 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
949 size_t msg_len, unsigned int msg_prio,
950 struct timespec64 *ts)
954 struct ext_wait_queue wait;
955 struct ext_wait_queue *receiver;
956 struct msg_msg *msg_ptr;
957 struct mqueue_inode_info *info;
958 ktime_t expires, *timeout = NULL;
959 struct posix_msg_tree_node *new_leaf = NULL;
961 DEFINE_WAKE_Q(wake_q);
963 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
967 expires = timespec64_to_ktime(*ts);
971 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
974 if (unlikely(!f.file)) {
979 inode = file_inode(f.file);
980 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
984 info = MQUEUE_I(inode);
987 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
992 if (unlikely(msg_len > info->attr.mq_msgsize)) {
997 /* First try to allocate memory, before doing anything with
998 * existing queues. */
999 msg_ptr = load_msg(u_msg_ptr, msg_len);
1000 if (IS_ERR(msg_ptr)) {
1001 ret = PTR_ERR(msg_ptr);
1004 msg_ptr->m_ts = msg_len;
1005 msg_ptr->m_type = msg_prio;
1008 * msg_insert really wants us to have a valid, spare node struct so
1009 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1010 * fall back to that if necessary.
1012 if (!info->node_cache)
1013 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1015 spin_lock(&info->lock);
1017 if (!info->node_cache && new_leaf) {
1018 /* Save our speculative allocation into the cache */
1019 INIT_LIST_HEAD(&new_leaf->msg_list);
1020 info->node_cache = new_leaf;
1026 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1027 if (f.file->f_flags & O_NONBLOCK) {
1030 wait.task = current;
1031 wait.msg = (void *) msg_ptr;
1032 wait.state = STATE_NONE;
1033 ret = wq_sleep(info, SEND, timeout, &wait);
1035 * wq_sleep must be called with info->lock held, and
1036 * returns with the lock released
1041 receiver = wq_get_first_waiter(info, RECV);
1043 pipelined_send(&wake_q, info, msg_ptr, receiver);
1045 /* adds message to the queue */
1046 ret = msg_insert(msg_ptr, info);
1051 inode->i_atime = inode->i_mtime = inode->i_ctime =
1052 current_time(inode);
1055 spin_unlock(&info->lock);
1066 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1067 size_t msg_len, unsigned int __user *u_msg_prio,
1068 struct timespec64 *ts)
1071 struct msg_msg *msg_ptr;
1073 struct inode *inode;
1074 struct mqueue_inode_info *info;
1075 struct ext_wait_queue wait;
1076 ktime_t expires, *timeout = NULL;
1077 struct posix_msg_tree_node *new_leaf = NULL;
1080 expires = timespec64_to_ktime(*ts);
1084 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1087 if (unlikely(!f.file)) {
1092 inode = file_inode(f.file);
1093 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1097 info = MQUEUE_I(inode);
1100 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1105 /* checks if buffer is big enough */
1106 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1112 * msg_insert really wants us to have a valid, spare node struct so
1113 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1114 * fall back to that if necessary.
1116 if (!info->node_cache)
1117 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1119 spin_lock(&info->lock);
1121 if (!info->node_cache && new_leaf) {
1122 /* Save our speculative allocation into the cache */
1123 INIT_LIST_HEAD(&new_leaf->msg_list);
1124 info->node_cache = new_leaf;
1129 if (info->attr.mq_curmsgs == 0) {
1130 if (f.file->f_flags & O_NONBLOCK) {
1131 spin_unlock(&info->lock);
1134 wait.task = current;
1135 wait.state = STATE_NONE;
1136 ret = wq_sleep(info, RECV, timeout, &wait);
1140 DEFINE_WAKE_Q(wake_q);
1142 msg_ptr = msg_get(info);
1144 inode->i_atime = inode->i_mtime = inode->i_ctime =
1145 current_time(inode);
1147 /* There is now free space in queue. */
1148 pipelined_receive(&wake_q, info);
1149 spin_unlock(&info->lock);
1154 ret = msg_ptr->m_ts;
1156 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1157 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1168 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1169 size_t, msg_len, unsigned int, msg_prio,
1170 const struct __kernel_timespec __user *, u_abs_timeout)
1172 struct timespec64 ts, *p = NULL;
1173 if (u_abs_timeout) {
1174 int res = prepare_timeout(u_abs_timeout, &ts);
1179 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1182 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1183 size_t, msg_len, unsigned int __user *, u_msg_prio,
1184 const struct __kernel_timespec __user *, u_abs_timeout)
1186 struct timespec64 ts, *p = NULL;
1187 if (u_abs_timeout) {
1188 int res = prepare_timeout(u_abs_timeout, &ts);
1193 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1197 * Notes: the case when user wants us to deregister (with NULL as pointer)
1198 * and he isn't currently owner of notification, will be silently discarded.
1199 * It isn't explicitly defined in the POSIX.
1201 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1206 struct inode *inode;
1207 struct mqueue_inode_info *info;
1210 audit_mq_notify(mqdes, notification);
1214 if (notification != NULL) {
1215 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1216 notification->sigev_notify != SIGEV_SIGNAL &&
1217 notification->sigev_notify != SIGEV_THREAD))
1219 if (notification->sigev_notify == SIGEV_SIGNAL &&
1220 !valid_signal(notification->sigev_signo)) {
1223 if (notification->sigev_notify == SIGEV_THREAD) {
1226 /* create the notify skb */
1227 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1232 if (copy_from_user(nc->data,
1233 notification->sigev_value.sival_ptr,
1234 NOTIFY_COOKIE_LEN)) {
1239 /* TODO: add a header? */
1240 skb_put(nc, NOTIFY_COOKIE_LEN);
1241 /* and attach it to the socket */
1243 f = fdget(notification->sigev_signo);
1248 sock = netlink_getsockbyfilp(f.file);
1251 ret = PTR_ERR(sock);
1256 timeo = MAX_SCHEDULE_TIMEOUT;
1257 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1276 inode = file_inode(f.file);
1277 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1281 info = MQUEUE_I(inode);
1284 spin_lock(&info->lock);
1285 if (notification == NULL) {
1286 if (info->notify_owner == task_tgid(current)) {
1287 remove_notification(info);
1288 inode->i_atime = inode->i_ctime = current_time(inode);
1290 } else if (info->notify_owner != NULL) {
1293 switch (notification->sigev_notify) {
1295 info->notify.sigev_notify = SIGEV_NONE;
1298 info->notify_sock = sock;
1299 info->notify_cookie = nc;
1302 info->notify.sigev_notify = SIGEV_THREAD;
1305 info->notify.sigev_signo = notification->sigev_signo;
1306 info->notify.sigev_value = notification->sigev_value;
1307 info->notify.sigev_notify = SIGEV_SIGNAL;
1311 info->notify_owner = get_pid(task_tgid(current));
1312 info->notify_user_ns = get_user_ns(current_user_ns());
1313 inode->i_atime = inode->i_ctime = current_time(inode);
1315 spin_unlock(&info->lock);
1320 netlink_detachskb(sock, nc);
1327 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1328 const struct sigevent __user *, u_notification)
1330 struct sigevent n, *p = NULL;
1331 if (u_notification) {
1332 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1336 return do_mq_notify(mqdes, p);
1339 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1342 struct inode *inode;
1343 struct mqueue_inode_info *info;
1345 if (new && (new->mq_flags & (~O_NONBLOCK)))
1352 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1357 inode = file_inode(f.file);
1358 info = MQUEUE_I(inode);
1360 spin_lock(&info->lock);
1364 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1367 audit_mq_getsetattr(mqdes, new);
1368 spin_lock(&f.file->f_lock);
1369 if (new->mq_flags & O_NONBLOCK)
1370 f.file->f_flags |= O_NONBLOCK;
1372 f.file->f_flags &= ~O_NONBLOCK;
1373 spin_unlock(&f.file->f_lock);
1375 inode->i_atime = inode->i_ctime = current_time(inode);
1378 spin_unlock(&info->lock);
1383 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1384 const struct mq_attr __user *, u_mqstat,
1385 struct mq_attr __user *, u_omqstat)
1388 struct mq_attr mqstat, omqstat;
1389 struct mq_attr *new = NULL, *old = NULL;
1393 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1399 ret = do_mq_getsetattr(mqdes, new, old);
1403 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1408 #ifdef CONFIG_COMPAT
1410 struct compat_mq_attr {
1411 compat_long_t mq_flags; /* message queue flags */
1412 compat_long_t mq_maxmsg; /* maximum number of messages */
1413 compat_long_t mq_msgsize; /* maximum message size */
1414 compat_long_t mq_curmsgs; /* number of messages currently queued */
1415 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1418 static inline int get_compat_mq_attr(struct mq_attr *attr,
1419 const struct compat_mq_attr __user *uattr)
1421 struct compat_mq_attr v;
1423 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1426 memset(attr, 0, sizeof(*attr));
1427 attr->mq_flags = v.mq_flags;
1428 attr->mq_maxmsg = v.mq_maxmsg;
1429 attr->mq_msgsize = v.mq_msgsize;
1430 attr->mq_curmsgs = v.mq_curmsgs;
1434 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1435 struct compat_mq_attr __user *uattr)
1437 struct compat_mq_attr v;
1439 memset(&v, 0, sizeof(v));
1440 v.mq_flags = attr->mq_flags;
1441 v.mq_maxmsg = attr->mq_maxmsg;
1442 v.mq_msgsize = attr->mq_msgsize;
1443 v.mq_curmsgs = attr->mq_curmsgs;
1444 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1449 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1450 int, oflag, compat_mode_t, mode,
1451 struct compat_mq_attr __user *, u_attr)
1453 struct mq_attr attr, *p = NULL;
1454 if (u_attr && oflag & O_CREAT) {
1456 if (get_compat_mq_attr(&attr, u_attr))
1459 return do_mq_open(u_name, oflag, mode, p);
1462 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1463 const struct compat_sigevent __user *, u_notification)
1465 struct sigevent n, *p = NULL;
1466 if (u_notification) {
1467 if (get_compat_sigevent(&n, u_notification))
1469 if (n.sigev_notify == SIGEV_THREAD)
1470 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1473 return do_mq_notify(mqdes, p);
1476 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1477 const struct compat_mq_attr __user *, u_mqstat,
1478 struct compat_mq_attr __user *, u_omqstat)
1481 struct mq_attr mqstat, omqstat;
1482 struct mq_attr *new = NULL, *old = NULL;
1486 if (get_compat_mq_attr(new, u_mqstat))
1492 ret = do_mq_getsetattr(mqdes, new, old);
1496 if (put_compat_mq_attr(old, u_omqstat))
1502 #ifdef CONFIG_COMPAT_32BIT_TIME
1503 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1504 struct timespec64 *ts)
1506 if (get_old_timespec32(ts, p))
1508 if (!timespec64_valid(ts))
1513 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1514 const char __user *, u_msg_ptr,
1515 unsigned int, msg_len, unsigned int, msg_prio,
1516 const struct old_timespec32 __user *, u_abs_timeout)
1518 struct timespec64 ts, *p = NULL;
1519 if (u_abs_timeout) {
1520 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1525 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1528 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1529 char __user *, u_msg_ptr,
1530 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1531 const struct old_timespec32 __user *, u_abs_timeout)
1533 struct timespec64 ts, *p = NULL;
1534 if (u_abs_timeout) {
1535 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1540 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1544 static const struct inode_operations mqueue_dir_inode_operations = {
1545 .lookup = simple_lookup,
1546 .create = mqueue_create,
1547 .unlink = mqueue_unlink,
1550 static const struct file_operations mqueue_file_operations = {
1551 .flush = mqueue_flush_file,
1552 .poll = mqueue_poll_file,
1553 .read = mqueue_read_file,
1554 .llseek = default_llseek,
1557 static const struct super_operations mqueue_super_ops = {
1558 .alloc_inode = mqueue_alloc_inode,
1559 .free_inode = mqueue_free_inode,
1560 .evict_inode = mqueue_evict_inode,
1561 .statfs = simple_statfs,
1564 static const struct fs_context_operations mqueue_fs_context_ops = {
1565 .free = mqueue_fs_context_free,
1566 .get_tree = mqueue_get_tree,
1569 static struct file_system_type mqueue_fs_type = {
1571 .init_fs_context = mqueue_init_fs_context,
1572 .kill_sb = kill_litter_super,
1573 .fs_flags = FS_USERNS_MOUNT,
1576 int mq_init_ns(struct ipc_namespace *ns)
1580 ns->mq_queues_count = 0;
1581 ns->mq_queues_max = DFLT_QUEUESMAX;
1582 ns->mq_msg_max = DFLT_MSGMAX;
1583 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1584 ns->mq_msg_default = DFLT_MSG;
1585 ns->mq_msgsize_default = DFLT_MSGSIZE;
1587 m = mq_create_mount(ns);
1594 void mq_clear_sbinfo(struct ipc_namespace *ns)
1596 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1599 void mq_put_mnt(struct ipc_namespace *ns)
1601 kern_unmount(ns->mq_mnt);
1604 static int __init init_mqueue_fs(void)
1608 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1609 sizeof(struct mqueue_inode_info), 0,
1610 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1611 if (mqueue_inode_cachep == NULL)
1614 /* ignore failures - they are not fatal */
1615 mq_sysctl_table = mq_register_sysctl_table();
1617 error = register_filesystem(&mqueue_fs_type);
1621 spin_lock_init(&mq_lock);
1623 error = mq_init_ns(&init_ipc_ns);
1625 goto out_filesystem;
1630 unregister_filesystem(&mqueue_fs_type);
1632 if (mq_sysctl_table)
1633 unregister_sysctl_table(mq_sysctl_table);
1634 kmem_cache_destroy(mqueue_inode_cachep);
1638 device_initcall(init_mqueue_fs);