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.
273 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
274 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
275 sizeof(struct posix_msg_tree_node);
277 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
278 info->attr.mq_msgsize);
281 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
282 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
283 spin_unlock(&mq_lock);
284 /* mqueue_evict_inode() releases info->messages */
288 u->mq_bytes += mq_bytes;
289 spin_unlock(&mq_lock);
292 info->user = get_uid(u);
293 } else if (S_ISDIR(mode)) {
295 /* Some things misbehave if size == 0 on a directory */
296 inode->i_size = 2 * DIRENT_SIZE;
297 inode->i_op = &mqueue_dir_inode_operations;
298 inode->i_fop = &simple_dir_operations;
308 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
311 struct ipc_namespace *ns = sb->s_fs_info;
313 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
314 sb->s_blocksize = PAGE_SIZE;
315 sb->s_blocksize_bits = PAGE_SHIFT;
316 sb->s_magic = MQUEUE_MAGIC;
317 sb->s_op = &mqueue_super_ops;
319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
321 return PTR_ERR(inode);
323 sb->s_root = d_make_root(inode);
329 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330 int flags, const char *dev_name,
333 struct ipc_namespace *ns;
334 if (flags & MS_KERNMOUNT) {
338 ns = current->nsproxy->ipc_ns;
340 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
343 static void init_once(void *foo)
345 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
347 inode_init_once(&p->vfs_inode);
350 static struct inode *mqueue_alloc_inode(struct super_block *sb)
352 struct mqueue_inode_info *ei;
354 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
357 return &ei->vfs_inode;
360 static void mqueue_i_callback(struct rcu_head *head)
362 struct inode *inode = container_of(head, struct inode, i_rcu);
363 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
366 static void mqueue_destroy_inode(struct inode *inode)
368 call_rcu(&inode->i_rcu, mqueue_i_callback);
371 static void mqueue_evict_inode(struct inode *inode)
373 struct mqueue_inode_info *info;
374 struct user_struct *user;
375 unsigned long mq_bytes, mq_treesize;
376 struct ipc_namespace *ipc_ns;
381 if (S_ISDIR(inode->i_mode))
384 ipc_ns = get_ns_from_inode(inode);
385 info = MQUEUE_I(inode);
386 spin_lock(&info->lock);
387 while ((msg = msg_get(info)) != NULL)
389 kfree(info->node_cache);
390 spin_unlock(&info->lock);
392 /* Total amount of bytes accounted for the mqueue */
393 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
394 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
395 sizeof(struct posix_msg_tree_node);
397 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
398 info->attr.mq_msgsize);
403 user->mq_bytes -= mq_bytes;
405 * get_ns_from_inode() ensures that the
406 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
407 * to which we now hold a reference, or it is NULL.
408 * We can't put it here under mq_lock, though.
411 ipc_ns->mq_queues_count--;
412 spin_unlock(&mq_lock);
419 static int mqueue_create(struct inode *dir, struct dentry *dentry,
420 umode_t mode, bool excl)
423 struct mq_attr *attr = dentry->d_fsdata;
425 struct ipc_namespace *ipc_ns;
428 ipc_ns = __get_ns_from_inode(dir);
434 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
435 !capable(CAP_SYS_RESOURCE)) {
439 ipc_ns->mq_queues_count++;
440 spin_unlock(&mq_lock);
442 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
444 error = PTR_ERR(inode);
446 ipc_ns->mq_queues_count--;
451 dir->i_size += DIRENT_SIZE;
452 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
454 d_instantiate(dentry, inode);
458 spin_unlock(&mq_lock);
464 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
466 struct inode *inode = d_inode(dentry);
468 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
469 dir->i_size -= DIRENT_SIZE;
476 * This is routine for system read from queue file.
477 * To avoid mess with doing here some sort of mq_receive we allow
478 * to read only queue size & notification info (the only values
479 * that are interesting from user point of view and aren't accessible
480 * through std routines)
482 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
483 size_t count, loff_t *off)
485 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
486 char buffer[FILENT_SIZE];
489 spin_lock(&info->lock);
490 snprintf(buffer, sizeof(buffer),
491 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
493 info->notify_owner ? info->notify.sigev_notify : 0,
494 (info->notify_owner &&
495 info->notify.sigev_notify == SIGEV_SIGNAL) ?
496 info->notify.sigev_signo : 0,
497 pid_vnr(info->notify_owner));
498 spin_unlock(&info->lock);
499 buffer[sizeof(buffer)-1] = '\0';
501 ret = simple_read_from_buffer(u_data, count, off, buffer,
506 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
510 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
512 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
514 spin_lock(&info->lock);
515 if (task_tgid(current) == info->notify_owner)
516 remove_notification(info);
518 spin_unlock(&info->lock);
522 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
524 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
527 poll_wait(filp, &info->wait_q, poll_tab);
529 spin_lock(&info->lock);
530 if (info->attr.mq_curmsgs)
531 retval = POLLIN | POLLRDNORM;
533 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
534 retval |= POLLOUT | POLLWRNORM;
535 spin_unlock(&info->lock);
540 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
541 static void wq_add(struct mqueue_inode_info *info, int sr,
542 struct ext_wait_queue *ewp)
544 struct ext_wait_queue *walk;
548 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
549 if (walk->task->static_prio <= current->static_prio) {
550 list_add_tail(&ewp->list, &walk->list);
554 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
558 * Puts current task to sleep. Caller must hold queue lock. After return
562 static int wq_sleep(struct mqueue_inode_info *info, int sr,
563 ktime_t *timeout, struct ext_wait_queue *ewp)
564 __releases(&info->lock)
569 wq_add(info, sr, ewp);
572 __set_current_state(TASK_INTERRUPTIBLE);
574 spin_unlock(&info->lock);
575 time = schedule_hrtimeout_range_clock(timeout, 0,
576 HRTIMER_MODE_ABS, CLOCK_REALTIME);
578 if (ewp->state == STATE_READY) {
582 spin_lock(&info->lock);
583 if (ewp->state == STATE_READY) {
587 if (signal_pending(current)) {
588 retval = -ERESTARTSYS;
596 list_del(&ewp->list);
598 spin_unlock(&info->lock);
604 * Returns waiting task that should be serviced first or NULL if none exists
606 static struct ext_wait_queue *wq_get_first_waiter(
607 struct mqueue_inode_info *info, int sr)
609 struct list_head *ptr;
611 ptr = info->e_wait_q[sr].list.prev;
612 if (ptr == &info->e_wait_q[sr].list)
614 return list_entry(ptr, struct ext_wait_queue, list);
618 static inline void set_cookie(struct sk_buff *skb, char code)
620 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
624 * The next function is only to split too long sys_mq_timedsend
626 static void __do_notify(struct mqueue_inode_info *info)
629 * invoked when there is registered process and there isn't process
630 * waiting synchronously for message AND state of queue changed from
631 * empty to not empty. Here we are sure that no one is waiting
633 if (info->notify_owner &&
634 info->attr.mq_curmsgs == 1) {
635 struct siginfo sig_i;
636 switch (info->notify.sigev_notify) {
642 sig_i.si_signo = info->notify.sigev_signo;
644 sig_i.si_code = SI_MESGQ;
645 sig_i.si_value = info->notify.sigev_value;
646 /* map current pid/uid into info->owner's namespaces */
648 sig_i.si_pid = task_tgid_nr_ns(current,
649 ns_of_pid(info->notify_owner));
650 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
653 kill_pid_info(info->notify.sigev_signo,
654 &sig_i, info->notify_owner);
657 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
658 netlink_sendskb(info->notify_sock, info->notify_cookie);
661 /* after notification unregisters process */
662 put_pid(info->notify_owner);
663 put_user_ns(info->notify_user_ns);
664 info->notify_owner = NULL;
665 info->notify_user_ns = NULL;
667 wake_up(&info->wait_q);
670 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
671 ktime_t *expires, struct timespec *ts)
673 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
675 if (!timespec_valid(ts))
678 *expires = timespec_to_ktime(*ts);
682 static void remove_notification(struct mqueue_inode_info *info)
684 if (info->notify_owner != NULL &&
685 info->notify.sigev_notify == SIGEV_THREAD) {
686 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
687 netlink_sendskb(info->notify_sock, info->notify_cookie);
689 put_pid(info->notify_owner);
690 put_user_ns(info->notify_user_ns);
691 info->notify_owner = NULL;
692 info->notify_user_ns = NULL;
695 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
698 unsigned long total_size;
700 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
702 if (capable(CAP_SYS_RESOURCE)) {
703 if (attr->mq_maxmsg > HARD_MSGMAX ||
704 attr->mq_msgsize > HARD_MSGSIZEMAX)
707 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
708 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
711 /* check for overflow */
712 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
714 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
715 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
716 sizeof(struct posix_msg_tree_node);
717 total_size = attr->mq_maxmsg * attr->mq_msgsize;
718 if (total_size + mq_treesize < total_size)
724 * Invoked when creating a new queue via sys_mq_open
726 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
727 struct path *path, int oflag, umode_t mode,
728 struct mq_attr *attr)
730 const struct cred *cred = current_cred();
734 ret = mq_attr_ok(ipc_ns, attr);
737 /* store for use during create */
738 path->dentry->d_fsdata = attr;
740 struct mq_attr def_attr;
742 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
743 ipc_ns->mq_msg_default);
744 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
745 ipc_ns->mq_msgsize_default);
746 ret = mq_attr_ok(ipc_ns, &def_attr);
751 mode &= ~current_umask();
752 ret = vfs_create(dir, path->dentry, mode, true);
753 path->dentry->d_fsdata = NULL;
756 return dentry_open(path, oflag, cred);
759 /* Opens existing queue */
760 static struct file *do_open(struct path *path, int oflag)
762 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
763 MAY_READ | MAY_WRITE };
765 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
766 return ERR_PTR(-EINVAL);
767 acc = oflag2acc[oflag & O_ACCMODE];
768 if (inode_permission(d_inode(path->dentry), acc))
769 return ERR_PTR(-EACCES);
770 return dentry_open(path, oflag, current_cred());
773 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
774 struct mq_attr __user *, u_attr)
778 struct filename *name;
781 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
782 struct vfsmount *mnt = ipc_ns->mq_mnt;
783 struct dentry *root = mnt->mnt_root;
786 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
789 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
791 if (IS_ERR(name = getname(u_name)))
792 return PTR_ERR(name);
794 fd = get_unused_fd_flags(O_CLOEXEC);
798 ro = mnt_want_write(mnt); /* we'll drop it in any case */
800 inode_lock(d_inode(root));
801 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
802 if (IS_ERR(path.dentry)) {
803 error = PTR_ERR(path.dentry);
806 path.mnt = mntget(mnt);
808 if (oflag & O_CREAT) {
809 if (d_really_is_positive(path.dentry)) { /* entry already exists */
810 audit_inode(name, path.dentry, 0);
811 if (oflag & O_EXCL) {
815 filp = do_open(&path, oflag);
821 audit_inode_parent_hidden(name, root);
822 filp = do_create(ipc_ns, d_inode(root),
824 u_attr ? &attr : NULL);
827 if (d_really_is_negative(path.dentry)) {
831 audit_inode(name, path.dentry, 0);
832 filp = do_open(&path, oflag);
836 fd_install(fd, filp);
838 error = PTR_ERR(filp);
846 inode_unlock(d_inode(root));
854 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
857 struct filename *name;
858 struct dentry *dentry;
859 struct inode *inode = NULL;
860 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
861 struct vfsmount *mnt = ipc_ns->mq_mnt;
863 name = getname(u_name);
865 return PTR_ERR(name);
867 audit_inode_parent_hidden(name, mnt->mnt_root);
868 err = mnt_want_write(mnt);
871 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
872 dentry = lookup_one_len(name->name, mnt->mnt_root,
874 if (IS_ERR(dentry)) {
875 err = PTR_ERR(dentry);
879 inode = d_inode(dentry);
884 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
889 inode_unlock(d_inode(mnt->mnt_root));
899 /* Pipelined send and receive functions.
901 * If a receiver finds no waiting message, then it registers itself in the
902 * list of waiting receivers. A sender checks that list before adding the new
903 * message into the message array. If there is a waiting receiver, then it
904 * bypasses the message array and directly hands the message over to the
905 * receiver. The receiver accepts the message and returns without grabbing the
908 * - Set pointer to message.
909 * - Queue the receiver task for later wakeup (without the info->lock).
910 * - Update its state to STATE_READY. Now the receiver can continue.
911 * - Wake up the process after the lock is dropped. Should the process wake up
912 * before this wakeup (due to a timeout or a signal) it will either see
913 * STATE_READY and continue or acquire the lock to check the state again.
915 * The same algorithm is used for senders.
918 /* pipelined_send() - send a message directly to the task waiting in
919 * sys_mq_timedreceive() (without inserting message into a queue).
921 static inline void pipelined_send(struct wake_q_head *wake_q,
922 struct mqueue_inode_info *info,
923 struct msg_msg *message,
924 struct ext_wait_queue *receiver)
926 receiver->msg = message;
927 list_del(&receiver->list);
928 wake_q_add(wake_q, receiver->task);
930 * Rely on the implicit cmpxchg barrier from wake_q_add such
931 * that we can ensure that updating receiver->state is the last
932 * write operation: As once set, the receiver can continue,
933 * and if we don't have the reference count from the wake_q,
934 * yet, at that point we can later have a use-after-free
935 * condition and bogus wakeup.
937 receiver->state = STATE_READY;
940 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
941 * gets its message and put to the queue (we have one free place for sure). */
942 static inline void pipelined_receive(struct wake_q_head *wake_q,
943 struct mqueue_inode_info *info)
945 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
949 wake_up_interruptible(&info->wait_q);
952 if (msg_insert(sender->msg, info))
955 list_del(&sender->list);
956 wake_q_add(wake_q, sender->task);
957 sender->state = STATE_READY;
960 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
961 size_t, msg_len, unsigned int, msg_prio,
962 const struct timespec __user *, u_abs_timeout)
966 struct ext_wait_queue wait;
967 struct ext_wait_queue *receiver;
968 struct msg_msg *msg_ptr;
969 struct mqueue_inode_info *info;
970 ktime_t expires, *timeout = NULL;
972 struct posix_msg_tree_node *new_leaf = NULL;
974 DEFINE_WAKE_Q(wake_q);
977 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
983 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
986 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
989 if (unlikely(!f.file)) {
994 inode = file_inode(f.file);
995 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
999 info = MQUEUE_I(inode);
1002 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1007 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1012 /* First try to allocate memory, before doing anything with
1013 * existing queues. */
1014 msg_ptr = load_msg(u_msg_ptr, msg_len);
1015 if (IS_ERR(msg_ptr)) {
1016 ret = PTR_ERR(msg_ptr);
1019 msg_ptr->m_ts = msg_len;
1020 msg_ptr->m_type = msg_prio;
1023 * msg_insert really wants us to have a valid, spare node struct so
1024 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1025 * fall back to that if necessary.
1027 if (!info->node_cache)
1028 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1030 spin_lock(&info->lock);
1032 if (!info->node_cache && new_leaf) {
1033 /* Save our speculative allocation into the cache */
1034 INIT_LIST_HEAD(&new_leaf->msg_list);
1035 info->node_cache = new_leaf;
1041 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1042 if (f.file->f_flags & O_NONBLOCK) {
1045 wait.task = current;
1046 wait.msg = (void *) msg_ptr;
1047 wait.state = STATE_NONE;
1048 ret = wq_sleep(info, SEND, timeout, &wait);
1050 * wq_sleep must be called with info->lock held, and
1051 * returns with the lock released
1056 receiver = wq_get_first_waiter(info, RECV);
1058 pipelined_send(&wake_q, info, msg_ptr, receiver);
1060 /* adds message to the queue */
1061 ret = msg_insert(msg_ptr, info);
1066 inode->i_atime = inode->i_mtime = inode->i_ctime =
1067 current_time(inode);
1070 spin_unlock(&info->lock);
1081 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1082 size_t, msg_len, unsigned int __user *, u_msg_prio,
1083 const struct timespec __user *, u_abs_timeout)
1086 struct msg_msg *msg_ptr;
1088 struct inode *inode;
1089 struct mqueue_inode_info *info;
1090 struct ext_wait_queue wait;
1091 ktime_t expires, *timeout = NULL;
1093 struct posix_msg_tree_node *new_leaf = NULL;
1095 if (u_abs_timeout) {
1096 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1102 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1105 if (unlikely(!f.file)) {
1110 inode = file_inode(f.file);
1111 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1115 info = MQUEUE_I(inode);
1118 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1123 /* checks if buffer is big enough */
1124 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1130 * msg_insert really wants us to have a valid, spare node struct so
1131 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1132 * fall back to that if necessary.
1134 if (!info->node_cache)
1135 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1137 spin_lock(&info->lock);
1139 if (!info->node_cache && new_leaf) {
1140 /* Save our speculative allocation into the cache */
1141 INIT_LIST_HEAD(&new_leaf->msg_list);
1142 info->node_cache = new_leaf;
1147 if (info->attr.mq_curmsgs == 0) {
1148 if (f.file->f_flags & O_NONBLOCK) {
1149 spin_unlock(&info->lock);
1152 wait.task = current;
1153 wait.state = STATE_NONE;
1154 ret = wq_sleep(info, RECV, timeout, &wait);
1158 DEFINE_WAKE_Q(wake_q);
1160 msg_ptr = msg_get(info);
1162 inode->i_atime = inode->i_mtime = inode->i_ctime =
1163 current_time(inode);
1165 /* There is now free space in queue. */
1166 pipelined_receive(&wake_q, info);
1167 spin_unlock(&info->lock);
1172 ret = msg_ptr->m_ts;
1174 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1175 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1187 * Notes: the case when user wants us to deregister (with NULL as pointer)
1188 * and he isn't currently owner of notification, will be silently discarded.
1189 * It isn't explicitly defined in the POSIX.
1191 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1192 const struct sigevent __user *, u_notification)
1197 struct inode *inode;
1198 struct sigevent notification;
1199 struct mqueue_inode_info *info;
1202 if (u_notification) {
1203 if (copy_from_user(¬ification, u_notification,
1204 sizeof(struct sigevent)))
1208 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1212 if (u_notification != NULL) {
1213 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1214 notification.sigev_notify != SIGEV_SIGNAL &&
1215 notification.sigev_notify != SIGEV_THREAD))
1217 if (notification.sigev_notify == SIGEV_SIGNAL &&
1218 !valid_signal(notification.sigev_signo)) {
1221 if (notification.sigev_notify == SIGEV_THREAD) {
1224 /* create the notify skb */
1225 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1230 if (copy_from_user(nc->data,
1231 notification.sigev_value.sival_ptr,
1232 NOTIFY_COOKIE_LEN)) {
1237 /* TODO: add a header? */
1238 skb_put(nc, NOTIFY_COOKIE_LEN);
1239 /* and attach it to the socket */
1241 f = fdget(notification.sigev_signo);
1246 sock = netlink_getsockbyfilp(f.file);
1249 ret = PTR_ERR(sock);
1254 timeo = MAX_SCHEDULE_TIMEOUT;
1255 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1272 inode = file_inode(f.file);
1273 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1277 info = MQUEUE_I(inode);
1280 spin_lock(&info->lock);
1281 if (u_notification == NULL) {
1282 if (info->notify_owner == task_tgid(current)) {
1283 remove_notification(info);
1284 inode->i_atime = inode->i_ctime = current_time(inode);
1286 } else if (info->notify_owner != NULL) {
1289 switch (notification.sigev_notify) {
1291 info->notify.sigev_notify = SIGEV_NONE;
1294 info->notify_sock = sock;
1295 info->notify_cookie = nc;
1298 info->notify.sigev_notify = SIGEV_THREAD;
1301 info->notify.sigev_signo = notification.sigev_signo;
1302 info->notify.sigev_value = notification.sigev_value;
1303 info->notify.sigev_notify = SIGEV_SIGNAL;
1307 info->notify_owner = get_pid(task_tgid(current));
1308 info->notify_user_ns = get_user_ns(current_user_ns());
1309 inode->i_atime = inode->i_ctime = current_time(inode);
1311 spin_unlock(&info->lock);
1316 netlink_detachskb(sock, nc);
1323 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1324 const struct mq_attr __user *, u_mqstat,
1325 struct mq_attr __user *, u_omqstat)
1328 struct mq_attr mqstat, omqstat;
1330 struct inode *inode;
1331 struct mqueue_inode_info *info;
1333 if (u_mqstat != NULL) {
1334 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1336 if (mqstat.mq_flags & (~O_NONBLOCK))
1346 inode = file_inode(f.file);
1347 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1351 info = MQUEUE_I(inode);
1353 spin_lock(&info->lock);
1355 omqstat = info->attr;
1356 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1358 audit_mq_getsetattr(mqdes, &mqstat);
1359 spin_lock(&f.file->f_lock);
1360 if (mqstat.mq_flags & O_NONBLOCK)
1361 f.file->f_flags |= O_NONBLOCK;
1363 f.file->f_flags &= ~O_NONBLOCK;
1364 spin_unlock(&f.file->f_lock);
1366 inode->i_atime = inode->i_ctime = current_time(inode);
1369 spin_unlock(&info->lock);
1372 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1373 sizeof(struct mq_attr)))
1382 static const struct inode_operations mqueue_dir_inode_operations = {
1383 .lookup = simple_lookup,
1384 .create = mqueue_create,
1385 .unlink = mqueue_unlink,
1388 static const struct file_operations mqueue_file_operations = {
1389 .flush = mqueue_flush_file,
1390 .poll = mqueue_poll_file,
1391 .read = mqueue_read_file,
1392 .llseek = default_llseek,
1395 static const struct super_operations mqueue_super_ops = {
1396 .alloc_inode = mqueue_alloc_inode,
1397 .destroy_inode = mqueue_destroy_inode,
1398 .evict_inode = mqueue_evict_inode,
1399 .statfs = simple_statfs,
1402 static struct file_system_type mqueue_fs_type = {
1404 .mount = mqueue_mount,
1405 .kill_sb = kill_litter_super,
1406 .fs_flags = FS_USERNS_MOUNT,
1409 int mq_init_ns(struct ipc_namespace *ns)
1411 ns->mq_queues_count = 0;
1412 ns->mq_queues_max = DFLT_QUEUESMAX;
1413 ns->mq_msg_max = DFLT_MSGMAX;
1414 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1415 ns->mq_msg_default = DFLT_MSG;
1416 ns->mq_msgsize_default = DFLT_MSGSIZE;
1418 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1419 if (IS_ERR(ns->mq_mnt)) {
1420 int err = PTR_ERR(ns->mq_mnt);
1427 void mq_clear_sbinfo(struct ipc_namespace *ns)
1429 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1432 void mq_put_mnt(struct ipc_namespace *ns)
1434 kern_unmount(ns->mq_mnt);
1437 static int __init init_mqueue_fs(void)
1441 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1442 sizeof(struct mqueue_inode_info), 0,
1443 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1444 if (mqueue_inode_cachep == NULL)
1447 /* ignore failures - they are not fatal */
1448 mq_sysctl_table = mq_register_sysctl_table();
1450 error = register_filesystem(&mqueue_fs_type);
1454 spin_lock_init(&mq_lock);
1456 error = mq_init_ns(&init_ipc_ns);
1458 goto out_filesystem;
1463 unregister_filesystem(&mqueue_fs_type);
1465 if (mq_sysctl_table)
1466 unregister_sysctl_table(mq_sysctl_table);
1467 kmem_cache_destroy(mqueue_inode_cachep);
1471 device_initcall(init_mqueue_fs);