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 unsigned long mq_bytes, mq_treesize;
393 struct ipc_namespace *ipc_ns;
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)
406 kfree(info->node_cache);
407 spin_unlock(&info->lock);
409 /* Total amount of bytes accounted for the mqueue */
410 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
411 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
412 sizeof(struct posix_msg_tree_node);
414 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
415 info->attr.mq_msgsize);
420 user->mq_bytes -= mq_bytes;
422 * get_ns_from_inode() ensures that the
423 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
424 * to which we now hold a reference, or it is NULL.
425 * We can't put it here under mq_lock, though.
428 ipc_ns->mq_queues_count--;
429 spin_unlock(&mq_lock);
436 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
438 struct inode *dir = dentry->d_parent->d_inode;
440 struct mq_attr *attr = arg;
442 struct ipc_namespace *ipc_ns;
445 ipc_ns = __get_ns_from_inode(dir);
451 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
452 !capable(CAP_SYS_RESOURCE)) {
456 ipc_ns->mq_queues_count++;
457 spin_unlock(&mq_lock);
459 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
461 error = PTR_ERR(inode);
463 ipc_ns->mq_queues_count--;
468 dir->i_size += DIRENT_SIZE;
469 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
471 d_instantiate(dentry, inode);
475 spin_unlock(&mq_lock);
481 static int mqueue_create(struct inode *dir, struct dentry *dentry,
482 umode_t mode, bool excl)
484 return mqueue_create_attr(dentry, mode, NULL);
487 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
489 struct inode *inode = d_inode(dentry);
491 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
492 dir->i_size -= DIRENT_SIZE;
499 * This is routine for system read from queue file.
500 * To avoid mess with doing here some sort of mq_receive we allow
501 * to read only queue size & notification info (the only values
502 * that are interesting from user point of view and aren't accessible
503 * through std routines)
505 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
506 size_t count, loff_t *off)
508 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
509 char buffer[FILENT_SIZE];
512 spin_lock(&info->lock);
513 snprintf(buffer, sizeof(buffer),
514 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
516 info->notify_owner ? info->notify.sigev_notify : 0,
517 (info->notify_owner &&
518 info->notify.sigev_notify == SIGEV_SIGNAL) ?
519 info->notify.sigev_signo : 0,
520 pid_vnr(info->notify_owner));
521 spin_unlock(&info->lock);
522 buffer[sizeof(buffer)-1] = '\0';
524 ret = simple_read_from_buffer(u_data, count, off, buffer,
529 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
533 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
535 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
537 spin_lock(&info->lock);
538 if (task_tgid(current) == info->notify_owner)
539 remove_notification(info);
541 spin_unlock(&info->lock);
545 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
547 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
550 poll_wait(filp, &info->wait_q, poll_tab);
552 spin_lock(&info->lock);
553 if (info->attr.mq_curmsgs)
554 retval = EPOLLIN | EPOLLRDNORM;
556 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
557 retval |= EPOLLOUT | EPOLLWRNORM;
558 spin_unlock(&info->lock);
563 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
564 static void wq_add(struct mqueue_inode_info *info, int sr,
565 struct ext_wait_queue *ewp)
567 struct ext_wait_queue *walk;
571 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
572 if (walk->task->prio <= current->prio) {
573 list_add_tail(&ewp->list, &walk->list);
577 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
581 * Puts current task to sleep. Caller must hold queue lock. After return
585 static int wq_sleep(struct mqueue_inode_info *info, int sr,
586 ktime_t *timeout, struct ext_wait_queue *ewp)
587 __releases(&info->lock)
592 wq_add(info, sr, ewp);
595 __set_current_state(TASK_INTERRUPTIBLE);
597 spin_unlock(&info->lock);
598 time = schedule_hrtimeout_range_clock(timeout, 0,
599 HRTIMER_MODE_ABS, CLOCK_REALTIME);
601 if (ewp->state == STATE_READY) {
605 spin_lock(&info->lock);
606 if (ewp->state == STATE_READY) {
610 if (signal_pending(current)) {
611 retval = -ERESTARTSYS;
619 list_del(&ewp->list);
621 spin_unlock(&info->lock);
627 * Returns waiting task that should be serviced first or NULL if none exists
629 static struct ext_wait_queue *wq_get_first_waiter(
630 struct mqueue_inode_info *info, int sr)
632 struct list_head *ptr;
634 ptr = info->e_wait_q[sr].list.prev;
635 if (ptr == &info->e_wait_q[sr].list)
637 return list_entry(ptr, struct ext_wait_queue, list);
641 static inline void set_cookie(struct sk_buff *skb, char code)
643 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
647 * The next function is only to split too long sys_mq_timedsend
649 static void __do_notify(struct mqueue_inode_info *info)
652 * invoked when there is registered process and there isn't process
653 * waiting synchronously for message AND state of queue changed from
654 * empty to not empty. Here we are sure that no one is waiting
656 if (info->notify_owner &&
657 info->attr.mq_curmsgs == 1) {
658 struct siginfo sig_i;
659 switch (info->notify.sigev_notify) {
665 clear_siginfo(&sig_i);
666 sig_i.si_signo = info->notify.sigev_signo;
668 sig_i.si_code = SI_MESGQ;
669 sig_i.si_value = info->notify.sigev_value;
670 /* map current pid/uid into info->owner's namespaces */
672 sig_i.si_pid = task_tgid_nr_ns(current,
673 ns_of_pid(info->notify_owner));
674 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
677 kill_pid_info(info->notify.sigev_signo,
678 &sig_i, info->notify_owner);
681 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
682 netlink_sendskb(info->notify_sock, info->notify_cookie);
685 /* after notification unregisters process */
686 put_pid(info->notify_owner);
687 put_user_ns(info->notify_user_ns);
688 info->notify_owner = NULL;
689 info->notify_user_ns = NULL;
691 wake_up(&info->wait_q);
694 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
695 struct timespec64 *ts)
697 if (get_timespec64(ts, u_abs_timeout))
699 if (!timespec64_valid(ts))
704 static void remove_notification(struct mqueue_inode_info *info)
706 if (info->notify_owner != NULL &&
707 info->notify.sigev_notify == SIGEV_THREAD) {
708 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
709 netlink_sendskb(info->notify_sock, info->notify_cookie);
711 put_pid(info->notify_owner);
712 put_user_ns(info->notify_user_ns);
713 info->notify_owner = NULL;
714 info->notify_user_ns = NULL;
717 static int prepare_open(struct dentry *dentry, int oflag, int ro,
718 umode_t mode, struct filename *name,
719 struct mq_attr *attr)
721 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
722 MAY_READ | MAY_WRITE };
725 if (d_really_is_negative(dentry)) {
726 if (!(oflag & O_CREAT))
730 audit_inode_parent_hidden(name, dentry->d_parent);
731 return vfs_mkobj(dentry, mode & ~current_umask(),
732 mqueue_create_attr, attr);
734 /* it already existed */
735 audit_inode(name, dentry, 0);
736 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
738 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
740 acc = oflag2acc[oflag & O_ACCMODE];
741 return inode_permission(d_inode(dentry), acc);
744 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
745 struct mq_attr *attr)
747 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
748 struct dentry *root = mnt->mnt_root;
749 struct filename *name;
754 audit_mq_open(oflag, mode, attr);
756 if (IS_ERR(name = getname(u_name)))
757 return PTR_ERR(name);
759 fd = get_unused_fd_flags(O_CLOEXEC);
763 ro = mnt_want_write(mnt); /* we'll drop it in any case */
764 inode_lock(d_inode(root));
765 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
766 if (IS_ERR(path.dentry)) {
767 error = PTR_ERR(path.dentry);
770 path.mnt = mntget(mnt);
771 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
773 struct file *file = dentry_open(&path, oflag, current_cred());
775 fd_install(fd, file);
777 error = PTR_ERR(file);
785 inode_unlock(d_inode(root));
793 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
794 struct mq_attr __user *, u_attr)
797 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
800 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
803 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
806 struct filename *name;
807 struct dentry *dentry;
808 struct inode *inode = NULL;
809 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
810 struct vfsmount *mnt = ipc_ns->mq_mnt;
812 name = getname(u_name);
814 return PTR_ERR(name);
816 audit_inode_parent_hidden(name, mnt->mnt_root);
817 err = mnt_want_write(mnt);
820 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
821 dentry = lookup_one_len(name->name, mnt->mnt_root,
823 if (IS_ERR(dentry)) {
824 err = PTR_ERR(dentry);
828 inode = d_inode(dentry);
833 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
838 inode_unlock(d_inode(mnt->mnt_root));
848 /* Pipelined send and receive functions.
850 * If a receiver finds no waiting message, then it registers itself in the
851 * list of waiting receivers. A sender checks that list before adding the new
852 * message into the message array. If there is a waiting receiver, then it
853 * bypasses the message array and directly hands the message over to the
854 * receiver. The receiver accepts the message and returns without grabbing the
857 * - Set pointer to message.
858 * - Queue the receiver task for later wakeup (without the info->lock).
859 * - Update its state to STATE_READY. Now the receiver can continue.
860 * - Wake up the process after the lock is dropped. Should the process wake up
861 * before this wakeup (due to a timeout or a signal) it will either see
862 * STATE_READY and continue or acquire the lock to check the state again.
864 * The same algorithm is used for senders.
867 /* pipelined_send() - send a message directly to the task waiting in
868 * sys_mq_timedreceive() (without inserting message into a queue).
870 static inline void pipelined_send(struct wake_q_head *wake_q,
871 struct mqueue_inode_info *info,
872 struct msg_msg *message,
873 struct ext_wait_queue *receiver)
875 receiver->msg = message;
876 list_del(&receiver->list);
877 wake_q_add(wake_q, receiver->task);
879 * Rely on the implicit cmpxchg barrier from wake_q_add such
880 * that we can ensure that updating receiver->state is the last
881 * write operation: As once set, the receiver can continue,
882 * and if we don't have the reference count from the wake_q,
883 * yet, at that point we can later have a use-after-free
884 * condition and bogus wakeup.
886 receiver->state = STATE_READY;
889 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
890 * gets its message and put to the queue (we have one free place for sure). */
891 static inline void pipelined_receive(struct wake_q_head *wake_q,
892 struct mqueue_inode_info *info)
894 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
898 wake_up_interruptible(&info->wait_q);
901 if (msg_insert(sender->msg, info))
904 list_del(&sender->list);
905 wake_q_add(wake_q, sender->task);
906 sender->state = STATE_READY;
909 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
910 size_t msg_len, unsigned int msg_prio,
911 struct timespec64 *ts)
915 struct ext_wait_queue wait;
916 struct ext_wait_queue *receiver;
917 struct msg_msg *msg_ptr;
918 struct mqueue_inode_info *info;
919 ktime_t expires, *timeout = NULL;
920 struct posix_msg_tree_node *new_leaf = NULL;
922 DEFINE_WAKE_Q(wake_q);
924 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
928 expires = timespec64_to_ktime(*ts);
932 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
935 if (unlikely(!f.file)) {
940 inode = file_inode(f.file);
941 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
945 info = MQUEUE_I(inode);
948 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
953 if (unlikely(msg_len > info->attr.mq_msgsize)) {
958 /* First try to allocate memory, before doing anything with
959 * existing queues. */
960 msg_ptr = load_msg(u_msg_ptr, msg_len);
961 if (IS_ERR(msg_ptr)) {
962 ret = PTR_ERR(msg_ptr);
965 msg_ptr->m_ts = msg_len;
966 msg_ptr->m_type = msg_prio;
969 * msg_insert really wants us to have a valid, spare node struct so
970 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
971 * fall back to that if necessary.
973 if (!info->node_cache)
974 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
976 spin_lock(&info->lock);
978 if (!info->node_cache && new_leaf) {
979 /* Save our speculative allocation into the cache */
980 INIT_LIST_HEAD(&new_leaf->msg_list);
981 info->node_cache = new_leaf;
987 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
988 if (f.file->f_flags & O_NONBLOCK) {
992 wait.msg = (void *) msg_ptr;
993 wait.state = STATE_NONE;
994 ret = wq_sleep(info, SEND, timeout, &wait);
996 * wq_sleep must be called with info->lock held, and
997 * returns with the lock released
1002 receiver = wq_get_first_waiter(info, RECV);
1004 pipelined_send(&wake_q, info, msg_ptr, receiver);
1006 /* adds message to the queue */
1007 ret = msg_insert(msg_ptr, info);
1012 inode->i_atime = inode->i_mtime = inode->i_ctime =
1013 current_time(inode);
1016 spin_unlock(&info->lock);
1027 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1028 size_t msg_len, unsigned int __user *u_msg_prio,
1029 struct timespec64 *ts)
1032 struct msg_msg *msg_ptr;
1034 struct inode *inode;
1035 struct mqueue_inode_info *info;
1036 struct ext_wait_queue wait;
1037 ktime_t expires, *timeout = NULL;
1038 struct posix_msg_tree_node *new_leaf = NULL;
1041 expires = timespec64_to_ktime(*ts);
1045 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1048 if (unlikely(!f.file)) {
1053 inode = file_inode(f.file);
1054 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1058 info = MQUEUE_I(inode);
1061 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1066 /* checks if buffer is big enough */
1067 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1073 * msg_insert really wants us to have a valid, spare node struct so
1074 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1075 * fall back to that if necessary.
1077 if (!info->node_cache)
1078 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1080 spin_lock(&info->lock);
1082 if (!info->node_cache && new_leaf) {
1083 /* Save our speculative allocation into the cache */
1084 INIT_LIST_HEAD(&new_leaf->msg_list);
1085 info->node_cache = new_leaf;
1090 if (info->attr.mq_curmsgs == 0) {
1091 if (f.file->f_flags & O_NONBLOCK) {
1092 spin_unlock(&info->lock);
1095 wait.task = current;
1096 wait.state = STATE_NONE;
1097 ret = wq_sleep(info, RECV, timeout, &wait);
1101 DEFINE_WAKE_Q(wake_q);
1103 msg_ptr = msg_get(info);
1105 inode->i_atime = inode->i_mtime = inode->i_ctime =
1106 current_time(inode);
1108 /* There is now free space in queue. */
1109 pipelined_receive(&wake_q, info);
1110 spin_unlock(&info->lock);
1115 ret = msg_ptr->m_ts;
1117 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1118 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1129 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1130 size_t, msg_len, unsigned int, msg_prio,
1131 const struct timespec __user *, u_abs_timeout)
1133 struct timespec64 ts, *p = NULL;
1134 if (u_abs_timeout) {
1135 int res = prepare_timeout(u_abs_timeout, &ts);
1140 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1143 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1144 size_t, msg_len, unsigned int __user *, u_msg_prio,
1145 const struct timespec __user *, u_abs_timeout)
1147 struct timespec64 ts, *p = NULL;
1148 if (u_abs_timeout) {
1149 int res = prepare_timeout(u_abs_timeout, &ts);
1154 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1158 * Notes: the case when user wants us to deregister (with NULL as pointer)
1159 * and he isn't currently owner of notification, will be silently discarded.
1160 * It isn't explicitly defined in the POSIX.
1162 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1167 struct inode *inode;
1168 struct mqueue_inode_info *info;
1171 audit_mq_notify(mqdes, notification);
1175 if (notification != NULL) {
1176 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1177 notification->sigev_notify != SIGEV_SIGNAL &&
1178 notification->sigev_notify != SIGEV_THREAD))
1180 if (notification->sigev_notify == SIGEV_SIGNAL &&
1181 !valid_signal(notification->sigev_signo)) {
1184 if (notification->sigev_notify == SIGEV_THREAD) {
1187 /* create the notify skb */
1188 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1193 if (copy_from_user(nc->data,
1194 notification->sigev_value.sival_ptr,
1195 NOTIFY_COOKIE_LEN)) {
1200 /* TODO: add a header? */
1201 skb_put(nc, NOTIFY_COOKIE_LEN);
1202 /* and attach it to the socket */
1204 f = fdget(notification->sigev_signo);
1209 sock = netlink_getsockbyfilp(f.file);
1212 ret = PTR_ERR(sock);
1217 timeo = MAX_SCHEDULE_TIMEOUT;
1218 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1237 inode = file_inode(f.file);
1238 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1242 info = MQUEUE_I(inode);
1245 spin_lock(&info->lock);
1246 if (notification == NULL) {
1247 if (info->notify_owner == task_tgid(current)) {
1248 remove_notification(info);
1249 inode->i_atime = inode->i_ctime = current_time(inode);
1251 } else if (info->notify_owner != NULL) {
1254 switch (notification->sigev_notify) {
1256 info->notify.sigev_notify = SIGEV_NONE;
1259 info->notify_sock = sock;
1260 info->notify_cookie = nc;
1263 info->notify.sigev_notify = SIGEV_THREAD;
1266 info->notify.sigev_signo = notification->sigev_signo;
1267 info->notify.sigev_value = notification->sigev_value;
1268 info->notify.sigev_notify = SIGEV_SIGNAL;
1272 info->notify_owner = get_pid(task_tgid(current));
1273 info->notify_user_ns = get_user_ns(current_user_ns());
1274 inode->i_atime = inode->i_ctime = current_time(inode);
1276 spin_unlock(&info->lock);
1281 netlink_detachskb(sock, nc);
1288 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1289 const struct sigevent __user *, u_notification)
1291 struct sigevent n, *p = NULL;
1292 if (u_notification) {
1293 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1297 return do_mq_notify(mqdes, p);
1300 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1303 struct inode *inode;
1304 struct mqueue_inode_info *info;
1306 if (new && (new->mq_flags & (~O_NONBLOCK)))
1313 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1318 inode = file_inode(f.file);
1319 info = MQUEUE_I(inode);
1321 spin_lock(&info->lock);
1325 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1328 audit_mq_getsetattr(mqdes, new);
1329 spin_lock(&f.file->f_lock);
1330 if (new->mq_flags & O_NONBLOCK)
1331 f.file->f_flags |= O_NONBLOCK;
1333 f.file->f_flags &= ~O_NONBLOCK;
1334 spin_unlock(&f.file->f_lock);
1336 inode->i_atime = inode->i_ctime = current_time(inode);
1339 spin_unlock(&info->lock);
1344 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1345 const struct mq_attr __user *, u_mqstat,
1346 struct mq_attr __user *, u_omqstat)
1349 struct mq_attr mqstat, omqstat;
1350 struct mq_attr *new = NULL, *old = NULL;
1354 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1360 ret = do_mq_getsetattr(mqdes, new, old);
1364 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1369 #ifdef CONFIG_COMPAT
1371 struct compat_mq_attr {
1372 compat_long_t mq_flags; /* message queue flags */
1373 compat_long_t mq_maxmsg; /* maximum number of messages */
1374 compat_long_t mq_msgsize; /* maximum message size */
1375 compat_long_t mq_curmsgs; /* number of messages currently queued */
1376 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1379 static inline int get_compat_mq_attr(struct mq_attr *attr,
1380 const struct compat_mq_attr __user *uattr)
1382 struct compat_mq_attr v;
1384 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1387 memset(attr, 0, sizeof(*attr));
1388 attr->mq_flags = v.mq_flags;
1389 attr->mq_maxmsg = v.mq_maxmsg;
1390 attr->mq_msgsize = v.mq_msgsize;
1391 attr->mq_curmsgs = v.mq_curmsgs;
1395 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1396 struct compat_mq_attr __user *uattr)
1398 struct compat_mq_attr v;
1400 memset(&v, 0, sizeof(v));
1401 v.mq_flags = attr->mq_flags;
1402 v.mq_maxmsg = attr->mq_maxmsg;
1403 v.mq_msgsize = attr->mq_msgsize;
1404 v.mq_curmsgs = attr->mq_curmsgs;
1405 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1410 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1411 int, oflag, compat_mode_t, mode,
1412 struct compat_mq_attr __user *, u_attr)
1414 struct mq_attr attr, *p = NULL;
1415 if (u_attr && oflag & O_CREAT) {
1417 if (get_compat_mq_attr(&attr, u_attr))
1420 return do_mq_open(u_name, oflag, mode, p);
1423 static int compat_prepare_timeout(const struct compat_timespec __user *p,
1424 struct timespec64 *ts)
1426 if (compat_get_timespec64(ts, p))
1428 if (!timespec64_valid(ts))
1433 COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1434 const char __user *, u_msg_ptr,
1435 compat_size_t, msg_len, unsigned int, msg_prio,
1436 const struct compat_timespec __user *, u_abs_timeout)
1438 struct timespec64 ts, *p = NULL;
1439 if (u_abs_timeout) {
1440 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1445 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1448 COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1449 char __user *, u_msg_ptr,
1450 compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1451 const struct compat_timespec __user *, u_abs_timeout)
1453 struct timespec64 ts, *p = NULL;
1454 if (u_abs_timeout) {
1455 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1460 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1463 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1464 const struct compat_sigevent __user *, u_notification)
1466 struct sigevent n, *p = NULL;
1467 if (u_notification) {
1468 if (get_compat_sigevent(&n, u_notification))
1470 if (n.sigev_notify == SIGEV_THREAD)
1471 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1474 return do_mq_notify(mqdes, p);
1477 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1478 const struct compat_mq_attr __user *, u_mqstat,
1479 struct compat_mq_attr __user *, u_omqstat)
1482 struct mq_attr mqstat, omqstat;
1483 struct mq_attr *new = NULL, *old = NULL;
1487 if (get_compat_mq_attr(new, u_mqstat))
1493 ret = do_mq_getsetattr(mqdes, new, old);
1497 if (put_compat_mq_attr(old, u_omqstat))
1503 static const struct inode_operations mqueue_dir_inode_operations = {
1504 .lookup = simple_lookup,
1505 .create = mqueue_create,
1506 .unlink = mqueue_unlink,
1509 static const struct file_operations mqueue_file_operations = {
1510 .flush = mqueue_flush_file,
1511 .poll = mqueue_poll_file,
1512 .read = mqueue_read_file,
1513 .llseek = default_llseek,
1516 static const struct super_operations mqueue_super_ops = {
1517 .alloc_inode = mqueue_alloc_inode,
1518 .destroy_inode = mqueue_destroy_inode,
1519 .evict_inode = mqueue_evict_inode,
1520 .statfs = simple_statfs,
1523 static struct file_system_type mqueue_fs_type = {
1525 .mount = mqueue_mount,
1526 .kill_sb = kill_litter_super,
1527 .fs_flags = FS_USERNS_MOUNT,
1530 int mq_init_ns(struct ipc_namespace *ns)
1532 ns->mq_queues_count = 0;
1533 ns->mq_queues_max = DFLT_QUEUESMAX;
1534 ns->mq_msg_max = DFLT_MSGMAX;
1535 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1536 ns->mq_msg_default = DFLT_MSG;
1537 ns->mq_msgsize_default = DFLT_MSGSIZE;
1539 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1540 if (IS_ERR(ns->mq_mnt)) {
1541 int err = PTR_ERR(ns->mq_mnt);
1548 void mq_clear_sbinfo(struct ipc_namespace *ns)
1550 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1553 void mq_put_mnt(struct ipc_namespace *ns)
1555 kern_unmount(ns->mq_mnt);
1558 static int __init init_mqueue_fs(void)
1562 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1563 sizeof(struct mqueue_inode_info), 0,
1564 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1565 if (mqueue_inode_cachep == NULL)
1568 /* ignore failures - they are not fatal */
1569 mq_sysctl_table = mq_register_sysctl_table();
1571 error = register_filesystem(&mqueue_fs_type);
1575 spin_lock_init(&mq_lock);
1577 error = mq_init_ns(&init_ipc_ns);
1579 goto out_filesystem;
1584 unregister_filesystem(&mqueue_fs_type);
1586 if (mq_sysctl_table)
1587 unregister_sysctl_table(mq_sysctl_table);
1588 kmem_cache_destroy(mqueue_inode_cachep);
1592 device_initcall(init_mqueue_fs);