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/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/vmalloc.h>
28 #include <linux/netlink.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/signal.h>
32 #include <linux/mutex.h>
33 #include <linux/nsproxy.h>
34 #include <linux/pid.h>
35 #include <linux/ipc_namespace.h>
36 #include <linux/user_namespace.h>
37 #include <linux/slab.h>
38 #include <linux/sched/wake_q.h>
39 #include <linux/sched/signal.h>
40 #include <linux/sched/user.h>
45 #define MQUEUE_MAGIC 0x19800202
46 #define DIRENT_SIZE 20
47 #define FILENT_SIZE 80
55 struct posix_msg_tree_node {
56 struct rb_node rb_node;
57 struct list_head msg_list;
61 struct ext_wait_queue { /* queue of sleeping tasks */
62 struct task_struct *task;
63 struct list_head list;
64 struct msg_msg *msg; /* ptr of loaded message */
65 int state; /* one of STATE_* values */
68 struct mqueue_inode_info {
70 struct inode vfs_inode;
71 wait_queue_head_t wait_q;
73 struct rb_root msg_tree;
74 struct posix_msg_tree_node *node_cache;
77 struct sigevent notify;
78 struct pid *notify_owner;
79 struct user_namespace *notify_user_ns;
80 struct user_struct *user; /* user who created, for accounting */
81 struct sock *notify_sock;
82 struct sk_buff *notify_cookie;
84 /* for tasks waiting for free space and messages, respectively */
85 struct ext_wait_queue e_wait_q[2];
87 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
90 static const struct inode_operations mqueue_dir_inode_operations;
91 static const struct file_operations mqueue_file_operations;
92 static const struct super_operations mqueue_super_ops;
93 static void remove_notification(struct mqueue_inode_info *info);
95 static struct kmem_cache *mqueue_inode_cachep;
97 static struct ctl_table_header *mq_sysctl_table;
99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
105 * This routine should be called with the mq_lock held.
107 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
109 return get_ipc_ns(inode->i_sb->s_fs_info);
112 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
114 struct ipc_namespace *ns;
117 ns = __get_ns_from_inode(inode);
118 spin_unlock(&mq_lock);
122 /* Auxiliary functions to manipulate messages' list */
123 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
125 struct rb_node **p, *parent = NULL;
126 struct posix_msg_tree_node *leaf;
128 p = &info->msg_tree.rb_node;
131 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
133 if (likely(leaf->priority == msg->m_type))
135 else if (msg->m_type < leaf->priority)
140 if (info->node_cache) {
141 leaf = info->node_cache;
142 info->node_cache = NULL;
144 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
147 INIT_LIST_HEAD(&leaf->msg_list);
149 leaf->priority = msg->m_type;
150 rb_link_node(&leaf->rb_node, parent, p);
151 rb_insert_color(&leaf->rb_node, &info->msg_tree);
153 info->attr.mq_curmsgs++;
154 info->qsize += msg->m_ts;
155 list_add_tail(&msg->m_list, &leaf->msg_list);
159 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
161 struct rb_node **p, *parent = NULL;
162 struct posix_msg_tree_node *leaf;
166 p = &info->msg_tree.rb_node;
170 * During insert, low priorities go to the left and high to the
171 * right. On receive, we want the highest priorities first, so
172 * walk all the way to the right.
177 if (info->attr.mq_curmsgs) {
178 pr_warn_once("Inconsistency in POSIX message queue, "
179 "no tree element, but supposedly messages "
181 info->attr.mq_curmsgs = 0;
185 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186 if (unlikely(list_empty(&leaf->msg_list))) {
187 pr_warn_once("Inconsistency in POSIX message queue, "
188 "empty leaf node but we haven't implemented "
189 "lazy leaf delete!\n");
190 rb_erase(&leaf->rb_node, &info->msg_tree);
191 if (info->node_cache) {
194 info->node_cache = leaf;
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
203 if (info->node_cache) {
206 info->node_cache = leaf;
210 info->attr.mq_curmsgs--;
211 info->qsize -= msg->m_ts;
215 static struct inode *mqueue_get_inode(struct super_block *sb,
216 struct ipc_namespace *ipc_ns, umode_t mode,
217 struct mq_attr *attr)
219 struct user_struct *u = current_user();
223 inode = new_inode(sb);
227 inode->i_ino = get_next_ino();
228 inode->i_mode = mode;
229 inode->i_uid = current_fsuid();
230 inode->i_gid = current_fsgid();
231 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
234 struct mqueue_inode_info *info;
235 unsigned long mq_bytes, mq_treesize;
237 inode->i_fop = &mqueue_file_operations;
238 inode->i_size = FILENT_SIZE;
239 /* mqueue specific info */
240 info = MQUEUE_I(inode);
241 spin_lock_init(&info->lock);
242 init_waitqueue_head(&info->wait_q);
243 INIT_LIST_HEAD(&info->e_wait_q[0].list);
244 INIT_LIST_HEAD(&info->e_wait_q[1].list);
245 info->notify_owner = NULL;
246 info->notify_user_ns = NULL;
248 info->user = NULL; /* set when all is ok */
249 info->msg_tree = RB_ROOT;
250 info->node_cache = NULL;
251 memset(&info->attr, 0, sizeof(info->attr));
252 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253 ipc_ns->mq_msg_default);
254 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255 ipc_ns->mq_msgsize_default);
257 info->attr.mq_maxmsg = attr->mq_maxmsg;
258 info->attr.mq_msgsize = attr->mq_msgsize;
261 * We used to allocate a static array of pointers and account
262 * the size of that array as well as one msg_msg struct per
263 * possible message into the queue size. That's no longer
264 * accurate as the queue is now an rbtree and will grow and
265 * shrink depending on usage patterns. We can, however, still
266 * account one msg_msg struct per message, but the nodes are
267 * allocated depending on priority usage, and most programs
268 * only use one, or a handful, of priorities. However, since
269 * this is pinned memory, we need to assume worst case, so
270 * that means the min(mq_maxmsg, max_priorities) * struct
271 * posix_msg_tree_node.
275 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
277 if (capable(CAP_SYS_RESOURCE)) {
278 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
279 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
282 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
283 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
287 /* check for overflow */
288 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
290 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
291 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
292 sizeof(struct posix_msg_tree_node);
293 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
294 if (mq_bytes + mq_treesize < mq_bytes)
296 mq_bytes += mq_treesize;
298 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
299 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
300 spin_unlock(&mq_lock);
301 /* mqueue_evict_inode() releases info->messages */
305 u->mq_bytes += mq_bytes;
306 spin_unlock(&mq_lock);
309 info->user = get_uid(u);
310 } else if (S_ISDIR(mode)) {
312 /* Some things misbehave if size == 0 on a directory */
313 inode->i_size = 2 * DIRENT_SIZE;
314 inode->i_op = &mqueue_dir_inode_operations;
315 inode->i_fop = &simple_dir_operations;
325 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
328 struct ipc_namespace *ns = sb->s_fs_info;
330 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
331 sb->s_blocksize = PAGE_SIZE;
332 sb->s_blocksize_bits = PAGE_SHIFT;
333 sb->s_magic = MQUEUE_MAGIC;
334 sb->s_op = &mqueue_super_ops;
336 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
338 return PTR_ERR(inode);
340 sb->s_root = d_make_root(inode);
346 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
347 int flags, const char *dev_name,
350 struct ipc_namespace *ns;
351 if (flags & SB_KERNMOUNT) {
355 ns = current->nsproxy->ipc_ns;
357 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
360 static void init_once(void *foo)
362 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
364 inode_init_once(&p->vfs_inode);
367 static struct inode *mqueue_alloc_inode(struct super_block *sb)
369 struct mqueue_inode_info *ei;
371 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
374 return &ei->vfs_inode;
377 static void mqueue_i_callback(struct rcu_head *head)
379 struct inode *inode = container_of(head, struct inode, i_rcu);
380 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
383 static void mqueue_destroy_inode(struct inode *inode)
385 call_rcu(&inode->i_rcu, mqueue_i_callback);
388 static void mqueue_evict_inode(struct inode *inode)
390 struct mqueue_inode_info *info;
391 struct user_struct *user;
392 struct ipc_namespace *ipc_ns;
393 struct msg_msg *msg, *nmsg;
398 if (S_ISDIR(inode->i_mode))
401 ipc_ns = get_ns_from_inode(inode);
402 info = MQUEUE_I(inode);
403 spin_lock(&info->lock);
404 while ((msg = msg_get(info)) != NULL)
405 list_add_tail(&msg->m_list, &tmp_msg);
406 kfree(info->node_cache);
407 spin_unlock(&info->lock);
409 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
410 list_del(&msg->m_list);
416 unsigned long mq_bytes, mq_treesize;
418 /* Total amount of bytes accounted for the mqueue */
419 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
420 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
421 sizeof(struct posix_msg_tree_node);
423 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
424 info->attr.mq_msgsize);
427 user->mq_bytes -= mq_bytes;
429 * get_ns_from_inode() ensures that the
430 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
431 * to which we now hold a reference, or it is NULL.
432 * We can't put it here under mq_lock, though.
435 ipc_ns->mq_queues_count--;
436 spin_unlock(&mq_lock);
443 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
445 struct inode *dir = dentry->d_parent->d_inode;
447 struct mq_attr *attr = arg;
449 struct ipc_namespace *ipc_ns;
452 ipc_ns = __get_ns_from_inode(dir);
458 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
459 !capable(CAP_SYS_RESOURCE)) {
463 ipc_ns->mq_queues_count++;
464 spin_unlock(&mq_lock);
466 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
468 error = PTR_ERR(inode);
470 ipc_ns->mq_queues_count--;
475 dir->i_size += DIRENT_SIZE;
476 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
478 d_instantiate(dentry, inode);
482 spin_unlock(&mq_lock);
488 static int mqueue_create(struct inode *dir, struct dentry *dentry,
489 umode_t mode, bool excl)
491 return mqueue_create_attr(dentry, mode, NULL);
494 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
496 struct inode *inode = d_inode(dentry);
498 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
499 dir->i_size -= DIRENT_SIZE;
506 * This is routine for system read from queue file.
507 * To avoid mess with doing here some sort of mq_receive we allow
508 * to read only queue size & notification info (the only values
509 * that are interesting from user point of view and aren't accessible
510 * through std routines)
512 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
513 size_t count, loff_t *off)
515 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
516 char buffer[FILENT_SIZE];
519 spin_lock(&info->lock);
520 snprintf(buffer, sizeof(buffer),
521 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
523 info->notify_owner ? info->notify.sigev_notify : 0,
524 (info->notify_owner &&
525 info->notify.sigev_notify == SIGEV_SIGNAL) ?
526 info->notify.sigev_signo : 0,
527 pid_vnr(info->notify_owner));
528 spin_unlock(&info->lock);
529 buffer[sizeof(buffer)-1] = '\0';
531 ret = simple_read_from_buffer(u_data, count, off, buffer,
536 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
540 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
542 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
544 spin_lock(&info->lock);
545 if (task_tgid(current) == info->notify_owner)
546 remove_notification(info);
548 spin_unlock(&info->lock);
552 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
554 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
557 poll_wait(filp, &info->wait_q, poll_tab);
559 spin_lock(&info->lock);
560 if (info->attr.mq_curmsgs)
561 retval = EPOLLIN | EPOLLRDNORM;
563 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
564 retval |= EPOLLOUT | EPOLLWRNORM;
565 spin_unlock(&info->lock);
570 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
571 static void wq_add(struct mqueue_inode_info *info, int sr,
572 struct ext_wait_queue *ewp)
574 struct ext_wait_queue *walk;
578 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
579 if (walk->task->prio <= current->prio) {
580 list_add_tail(&ewp->list, &walk->list);
584 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
588 * Puts current task to sleep. Caller must hold queue lock. After return
592 static int wq_sleep(struct mqueue_inode_info *info, int sr,
593 ktime_t *timeout, struct ext_wait_queue *ewp)
594 __releases(&info->lock)
599 wq_add(info, sr, ewp);
602 __set_current_state(TASK_INTERRUPTIBLE);
604 spin_unlock(&info->lock);
605 time = schedule_hrtimeout_range_clock(timeout, 0,
606 HRTIMER_MODE_ABS, CLOCK_REALTIME);
608 if (ewp->state == STATE_READY) {
612 spin_lock(&info->lock);
613 if (ewp->state == STATE_READY) {
617 if (signal_pending(current)) {
618 retval = -ERESTARTSYS;
626 list_del(&ewp->list);
628 spin_unlock(&info->lock);
634 * Returns waiting task that should be serviced first or NULL if none exists
636 static struct ext_wait_queue *wq_get_first_waiter(
637 struct mqueue_inode_info *info, int sr)
639 struct list_head *ptr;
641 ptr = info->e_wait_q[sr].list.prev;
642 if (ptr == &info->e_wait_q[sr].list)
644 return list_entry(ptr, struct ext_wait_queue, list);
648 static inline void set_cookie(struct sk_buff *skb, char code)
650 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
654 * The next function is only to split too long sys_mq_timedsend
656 static void __do_notify(struct mqueue_inode_info *info)
659 * invoked when there is registered process and there isn't process
660 * waiting synchronously for message AND state of queue changed from
661 * empty to not empty. Here we are sure that no one is waiting
663 if (info->notify_owner &&
664 info->attr.mq_curmsgs == 1) {
665 struct siginfo sig_i;
666 switch (info->notify.sigev_notify) {
672 clear_siginfo(&sig_i);
673 sig_i.si_signo = info->notify.sigev_signo;
675 sig_i.si_code = SI_MESGQ;
676 sig_i.si_value = info->notify.sigev_value;
677 /* map current pid/uid into info->owner's namespaces */
679 sig_i.si_pid = task_tgid_nr_ns(current,
680 ns_of_pid(info->notify_owner));
681 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
684 kill_pid_info(info->notify.sigev_signo,
685 &sig_i, info->notify_owner);
688 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
689 netlink_sendskb(info->notify_sock, info->notify_cookie);
692 /* after notification unregisters process */
693 put_pid(info->notify_owner);
694 put_user_ns(info->notify_user_ns);
695 info->notify_owner = NULL;
696 info->notify_user_ns = NULL;
698 wake_up(&info->wait_q);
701 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
702 struct timespec64 *ts)
704 if (get_timespec64(ts, u_abs_timeout))
706 if (!timespec64_valid(ts))
711 static void remove_notification(struct mqueue_inode_info *info)
713 if (info->notify_owner != NULL &&
714 info->notify.sigev_notify == SIGEV_THREAD) {
715 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
716 netlink_sendskb(info->notify_sock, info->notify_cookie);
718 put_pid(info->notify_owner);
719 put_user_ns(info->notify_user_ns);
720 info->notify_owner = NULL;
721 info->notify_user_ns = NULL;
724 static int prepare_open(struct dentry *dentry, int oflag, int ro,
725 umode_t mode, struct filename *name,
726 struct mq_attr *attr)
728 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
729 MAY_READ | MAY_WRITE };
732 if (d_really_is_negative(dentry)) {
733 if (!(oflag & O_CREAT))
737 audit_inode_parent_hidden(name, dentry->d_parent);
738 return vfs_mkobj(dentry, mode & ~current_umask(),
739 mqueue_create_attr, attr);
741 /* it already existed */
742 audit_inode(name, dentry, 0);
743 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
745 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
747 acc = oflag2acc[oflag & O_ACCMODE];
748 return inode_permission(d_inode(dentry), acc);
751 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
752 struct mq_attr *attr)
754 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
755 struct dentry *root = mnt->mnt_root;
756 struct filename *name;
761 audit_mq_open(oflag, mode, attr);
763 if (IS_ERR(name = getname(u_name)))
764 return PTR_ERR(name);
766 fd = get_unused_fd_flags(O_CLOEXEC);
770 ro = mnt_want_write(mnt); /* we'll drop it in any case */
771 inode_lock(d_inode(root));
772 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
773 if (IS_ERR(path.dentry)) {
774 error = PTR_ERR(path.dentry);
777 path.mnt = mntget(mnt);
778 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
780 struct file *file = dentry_open(&path, oflag, current_cred());
782 fd_install(fd, file);
784 error = PTR_ERR(file);
792 inode_unlock(d_inode(root));
800 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
801 struct mq_attr __user *, u_attr)
804 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
807 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
810 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
813 struct filename *name;
814 struct dentry *dentry;
815 struct inode *inode = NULL;
816 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
817 struct vfsmount *mnt = ipc_ns->mq_mnt;
819 name = getname(u_name);
821 return PTR_ERR(name);
823 audit_inode_parent_hidden(name, mnt->mnt_root);
824 err = mnt_want_write(mnt);
827 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
828 dentry = lookup_one_len(name->name, mnt->mnt_root,
830 if (IS_ERR(dentry)) {
831 err = PTR_ERR(dentry);
835 inode = d_inode(dentry);
840 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
845 inode_unlock(d_inode(mnt->mnt_root));
855 /* Pipelined send and receive functions.
857 * If a receiver finds no waiting message, then it registers itself in the
858 * list of waiting receivers. A sender checks that list before adding the new
859 * message into the message array. If there is a waiting receiver, then it
860 * bypasses the message array and directly hands the message over to the
861 * receiver. The receiver accepts the message and returns without grabbing the
864 * - Set pointer to message.
865 * - Queue the receiver task for later wakeup (without the info->lock).
866 * - Update its state to STATE_READY. Now the receiver can continue.
867 * - Wake up the process after the lock is dropped. Should the process wake up
868 * before this wakeup (due to a timeout or a signal) it will either see
869 * STATE_READY and continue or acquire the lock to check the state again.
871 * The same algorithm is used for senders.
874 /* pipelined_send() - send a message directly to the task waiting in
875 * sys_mq_timedreceive() (without inserting message into a queue).
877 static inline void pipelined_send(struct wake_q_head *wake_q,
878 struct mqueue_inode_info *info,
879 struct msg_msg *message,
880 struct ext_wait_queue *receiver)
882 receiver->msg = message;
883 list_del(&receiver->list);
884 wake_q_add(wake_q, receiver->task);
886 * Rely on the implicit cmpxchg barrier from wake_q_add such
887 * that we can ensure that updating receiver->state is the last
888 * write operation: As once set, the receiver can continue,
889 * and if we don't have the reference count from the wake_q,
890 * yet, at that point we can later have a use-after-free
891 * condition and bogus wakeup.
893 receiver->state = STATE_READY;
896 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
897 * gets its message and put to the queue (we have one free place for sure). */
898 static inline void pipelined_receive(struct wake_q_head *wake_q,
899 struct mqueue_inode_info *info)
901 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
905 wake_up_interruptible(&info->wait_q);
908 if (msg_insert(sender->msg, info))
911 list_del(&sender->list);
912 wake_q_add(wake_q, sender->task);
913 sender->state = STATE_READY;
916 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
917 size_t msg_len, unsigned int msg_prio,
918 struct timespec64 *ts)
922 struct ext_wait_queue wait;
923 struct ext_wait_queue *receiver;
924 struct msg_msg *msg_ptr;
925 struct mqueue_inode_info *info;
926 ktime_t expires, *timeout = NULL;
927 struct posix_msg_tree_node *new_leaf = NULL;
929 DEFINE_WAKE_Q(wake_q);
931 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
935 expires = timespec64_to_ktime(*ts);
939 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
942 if (unlikely(!f.file)) {
947 inode = file_inode(f.file);
948 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
952 info = MQUEUE_I(inode);
955 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
960 if (unlikely(msg_len > info->attr.mq_msgsize)) {
965 /* First try to allocate memory, before doing anything with
966 * existing queues. */
967 msg_ptr = load_msg(u_msg_ptr, msg_len);
968 if (IS_ERR(msg_ptr)) {
969 ret = PTR_ERR(msg_ptr);
972 msg_ptr->m_ts = msg_len;
973 msg_ptr->m_type = msg_prio;
976 * msg_insert really wants us to have a valid, spare node struct so
977 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
978 * fall back to that if necessary.
980 if (!info->node_cache)
981 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
983 spin_lock(&info->lock);
985 if (!info->node_cache && new_leaf) {
986 /* Save our speculative allocation into the cache */
987 INIT_LIST_HEAD(&new_leaf->msg_list);
988 info->node_cache = new_leaf;
994 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
995 if (f.file->f_flags & O_NONBLOCK) {
999 wait.msg = (void *) msg_ptr;
1000 wait.state = STATE_NONE;
1001 ret = wq_sleep(info, SEND, timeout, &wait);
1003 * wq_sleep must be called with info->lock held, and
1004 * returns with the lock released
1009 receiver = wq_get_first_waiter(info, RECV);
1011 pipelined_send(&wake_q, info, msg_ptr, receiver);
1013 /* adds message to the queue */
1014 ret = msg_insert(msg_ptr, info);
1019 inode->i_atime = inode->i_mtime = inode->i_ctime =
1020 current_time(inode);
1023 spin_unlock(&info->lock);
1034 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1035 size_t msg_len, unsigned int __user *u_msg_prio,
1036 struct timespec64 *ts)
1039 struct msg_msg *msg_ptr;
1041 struct inode *inode;
1042 struct mqueue_inode_info *info;
1043 struct ext_wait_queue wait;
1044 ktime_t expires, *timeout = NULL;
1045 struct posix_msg_tree_node *new_leaf = NULL;
1048 expires = timespec64_to_ktime(*ts);
1052 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1055 if (unlikely(!f.file)) {
1060 inode = file_inode(f.file);
1061 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1065 info = MQUEUE_I(inode);
1068 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1073 /* checks if buffer is big enough */
1074 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1080 * msg_insert really wants us to have a valid, spare node struct so
1081 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1082 * fall back to that if necessary.
1084 if (!info->node_cache)
1085 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1087 spin_lock(&info->lock);
1089 if (!info->node_cache && new_leaf) {
1090 /* Save our speculative allocation into the cache */
1091 INIT_LIST_HEAD(&new_leaf->msg_list);
1092 info->node_cache = new_leaf;
1097 if (info->attr.mq_curmsgs == 0) {
1098 if (f.file->f_flags & O_NONBLOCK) {
1099 spin_unlock(&info->lock);
1102 wait.task = current;
1103 wait.state = STATE_NONE;
1104 ret = wq_sleep(info, RECV, timeout, &wait);
1108 DEFINE_WAKE_Q(wake_q);
1110 msg_ptr = msg_get(info);
1112 inode->i_atime = inode->i_mtime = inode->i_ctime =
1113 current_time(inode);
1115 /* There is now free space in queue. */
1116 pipelined_receive(&wake_q, info);
1117 spin_unlock(&info->lock);
1122 ret = msg_ptr->m_ts;
1124 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1125 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1136 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1137 size_t, msg_len, unsigned int, msg_prio,
1138 const struct __kernel_timespec __user *, u_abs_timeout)
1140 struct timespec64 ts, *p = NULL;
1141 if (u_abs_timeout) {
1142 int res = prepare_timeout(u_abs_timeout, &ts);
1147 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1150 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1151 size_t, msg_len, unsigned int __user *, u_msg_prio,
1152 const struct __kernel_timespec __user *, u_abs_timeout)
1154 struct timespec64 ts, *p = NULL;
1155 if (u_abs_timeout) {
1156 int res = prepare_timeout(u_abs_timeout, &ts);
1161 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1165 * Notes: the case when user wants us to deregister (with NULL as pointer)
1166 * and he isn't currently owner of notification, will be silently discarded.
1167 * It isn't explicitly defined in the POSIX.
1169 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1174 struct inode *inode;
1175 struct mqueue_inode_info *info;
1178 audit_mq_notify(mqdes, notification);
1182 if (notification != NULL) {
1183 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1184 notification->sigev_notify != SIGEV_SIGNAL &&
1185 notification->sigev_notify != SIGEV_THREAD))
1187 if (notification->sigev_notify == SIGEV_SIGNAL &&
1188 !valid_signal(notification->sigev_signo)) {
1191 if (notification->sigev_notify == SIGEV_THREAD) {
1194 /* create the notify skb */
1195 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1200 if (copy_from_user(nc->data,
1201 notification->sigev_value.sival_ptr,
1202 NOTIFY_COOKIE_LEN)) {
1207 /* TODO: add a header? */
1208 skb_put(nc, NOTIFY_COOKIE_LEN);
1209 /* and attach it to the socket */
1211 f = fdget(notification->sigev_signo);
1216 sock = netlink_getsockbyfilp(f.file);
1219 ret = PTR_ERR(sock);
1224 timeo = MAX_SCHEDULE_TIMEOUT;
1225 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1244 inode = file_inode(f.file);
1245 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1249 info = MQUEUE_I(inode);
1252 spin_lock(&info->lock);
1253 if (notification == NULL) {
1254 if (info->notify_owner == task_tgid(current)) {
1255 remove_notification(info);
1256 inode->i_atime = inode->i_ctime = current_time(inode);
1258 } else if (info->notify_owner != NULL) {
1261 switch (notification->sigev_notify) {
1263 info->notify.sigev_notify = SIGEV_NONE;
1266 info->notify_sock = sock;
1267 info->notify_cookie = nc;
1270 info->notify.sigev_notify = SIGEV_THREAD;
1273 info->notify.sigev_signo = notification->sigev_signo;
1274 info->notify.sigev_value = notification->sigev_value;
1275 info->notify.sigev_notify = SIGEV_SIGNAL;
1279 info->notify_owner = get_pid(task_tgid(current));
1280 info->notify_user_ns = get_user_ns(current_user_ns());
1281 inode->i_atime = inode->i_ctime = current_time(inode);
1283 spin_unlock(&info->lock);
1288 netlink_detachskb(sock, nc);
1295 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1296 const struct sigevent __user *, u_notification)
1298 struct sigevent n, *p = NULL;
1299 if (u_notification) {
1300 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1304 return do_mq_notify(mqdes, p);
1307 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1310 struct inode *inode;
1311 struct mqueue_inode_info *info;
1313 if (new && (new->mq_flags & (~O_NONBLOCK)))
1320 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1325 inode = file_inode(f.file);
1326 info = MQUEUE_I(inode);
1328 spin_lock(&info->lock);
1332 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1335 audit_mq_getsetattr(mqdes, new);
1336 spin_lock(&f.file->f_lock);
1337 if (new->mq_flags & O_NONBLOCK)
1338 f.file->f_flags |= O_NONBLOCK;
1340 f.file->f_flags &= ~O_NONBLOCK;
1341 spin_unlock(&f.file->f_lock);
1343 inode->i_atime = inode->i_ctime = current_time(inode);
1346 spin_unlock(&info->lock);
1351 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1352 const struct mq_attr __user *, u_mqstat,
1353 struct mq_attr __user *, u_omqstat)
1356 struct mq_attr mqstat, omqstat;
1357 struct mq_attr *new = NULL, *old = NULL;
1361 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1367 ret = do_mq_getsetattr(mqdes, new, old);
1371 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1376 #ifdef CONFIG_COMPAT
1378 struct compat_mq_attr {
1379 compat_long_t mq_flags; /* message queue flags */
1380 compat_long_t mq_maxmsg; /* maximum number of messages */
1381 compat_long_t mq_msgsize; /* maximum message size */
1382 compat_long_t mq_curmsgs; /* number of messages currently queued */
1383 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1386 static inline int get_compat_mq_attr(struct mq_attr *attr,
1387 const struct compat_mq_attr __user *uattr)
1389 struct compat_mq_attr v;
1391 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1394 memset(attr, 0, sizeof(*attr));
1395 attr->mq_flags = v.mq_flags;
1396 attr->mq_maxmsg = v.mq_maxmsg;
1397 attr->mq_msgsize = v.mq_msgsize;
1398 attr->mq_curmsgs = v.mq_curmsgs;
1402 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1403 struct compat_mq_attr __user *uattr)
1405 struct compat_mq_attr v;
1407 memset(&v, 0, sizeof(v));
1408 v.mq_flags = attr->mq_flags;
1409 v.mq_maxmsg = attr->mq_maxmsg;
1410 v.mq_msgsize = attr->mq_msgsize;
1411 v.mq_curmsgs = attr->mq_curmsgs;
1412 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1417 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1418 int, oflag, compat_mode_t, mode,
1419 struct compat_mq_attr __user *, u_attr)
1421 struct mq_attr attr, *p = NULL;
1422 if (u_attr && oflag & O_CREAT) {
1424 if (get_compat_mq_attr(&attr, u_attr))
1427 return do_mq_open(u_name, oflag, mode, p);
1430 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1431 const struct compat_sigevent __user *, u_notification)
1433 struct sigevent n, *p = NULL;
1434 if (u_notification) {
1435 if (get_compat_sigevent(&n, u_notification))
1437 if (n.sigev_notify == SIGEV_THREAD)
1438 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1441 return do_mq_notify(mqdes, p);
1444 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1445 const struct compat_mq_attr __user *, u_mqstat,
1446 struct compat_mq_attr __user *, u_omqstat)
1449 struct mq_attr mqstat, omqstat;
1450 struct mq_attr *new = NULL, *old = NULL;
1454 if (get_compat_mq_attr(new, u_mqstat))
1460 ret = do_mq_getsetattr(mqdes, new, old);
1464 if (put_compat_mq_attr(old, u_omqstat))
1470 #ifdef CONFIG_COMPAT_32BIT_TIME
1471 static int compat_prepare_timeout(const struct compat_timespec __user *p,
1472 struct timespec64 *ts)
1474 if (compat_get_timespec64(ts, p))
1476 if (!timespec64_valid(ts))
1481 COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1482 const char __user *, u_msg_ptr,
1483 compat_size_t, msg_len, unsigned int, msg_prio,
1484 const struct compat_timespec __user *, u_abs_timeout)
1486 struct timespec64 ts, *p = NULL;
1487 if (u_abs_timeout) {
1488 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1493 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1496 COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1497 char __user *, u_msg_ptr,
1498 compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1499 const struct compat_timespec __user *, u_abs_timeout)
1501 struct timespec64 ts, *p = NULL;
1502 if (u_abs_timeout) {
1503 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1508 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1512 static const struct inode_operations mqueue_dir_inode_operations = {
1513 .lookup = simple_lookup,
1514 .create = mqueue_create,
1515 .unlink = mqueue_unlink,
1518 static const struct file_operations mqueue_file_operations = {
1519 .flush = mqueue_flush_file,
1520 .poll = mqueue_poll_file,
1521 .read = mqueue_read_file,
1522 .llseek = default_llseek,
1525 static const struct super_operations mqueue_super_ops = {
1526 .alloc_inode = mqueue_alloc_inode,
1527 .destroy_inode = mqueue_destroy_inode,
1528 .evict_inode = mqueue_evict_inode,
1529 .statfs = simple_statfs,
1532 static struct file_system_type mqueue_fs_type = {
1534 .mount = mqueue_mount,
1535 .kill_sb = kill_litter_super,
1536 .fs_flags = FS_USERNS_MOUNT,
1539 int mq_init_ns(struct ipc_namespace *ns)
1541 ns->mq_queues_count = 0;
1542 ns->mq_queues_max = DFLT_QUEUESMAX;
1543 ns->mq_msg_max = DFLT_MSGMAX;
1544 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1545 ns->mq_msg_default = DFLT_MSG;
1546 ns->mq_msgsize_default = DFLT_MSGSIZE;
1548 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1549 if (IS_ERR(ns->mq_mnt)) {
1550 int err = PTR_ERR(ns->mq_mnt);
1557 void mq_clear_sbinfo(struct ipc_namespace *ns)
1559 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1562 void mq_put_mnt(struct ipc_namespace *ns)
1564 kern_unmount(ns->mq_mnt);
1567 static int __init init_mqueue_fs(void)
1571 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1572 sizeof(struct mqueue_inode_info), 0,
1573 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1574 if (mqueue_inode_cachep == NULL)
1577 /* ignore failures - they are not fatal */
1578 mq_sysctl_table = mq_register_sysctl_table();
1580 error = register_filesystem(&mqueue_fs_type);
1584 spin_lock_init(&mq_lock);
1586 error = mq_init_ns(&init_ipc_ns);
1588 goto out_filesystem;
1593 unregister_filesystem(&mqueue_fs_type);
1595 if (mq_sysctl_table)
1596 unregister_sysctl_table(mq_sysctl_table);
1597 kmem_cache_destroy(mqueue_inode_cachep);
1601 device_initcall(init_mqueue_fs);