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>
42 #define MQUEUE_MAGIC 0x19800202
43 #define DIRENT_SIZE 20
44 #define FILENT_SIZE 80
52 struct posix_msg_tree_node {
53 struct rb_node rb_node;
54 struct list_head msg_list;
58 struct ext_wait_queue { /* queue of sleeping tasks */
59 struct task_struct *task;
60 struct list_head list;
61 struct msg_msg *msg; /* ptr of loaded message */
62 int state; /* one of STATE_* values */
65 struct mqueue_inode_info {
67 struct inode vfs_inode;
68 wait_queue_head_t wait_q;
70 struct rb_root msg_tree;
71 struct posix_msg_tree_node *node_cache;
74 struct sigevent notify;
75 struct pid *notify_owner;
76 struct user_namespace *notify_user_ns;
77 struct user_struct *user; /* user who created, for accounting */
78 struct sock *notify_sock;
79 struct sk_buff *notify_cookie;
81 /* for tasks waiting for free space and messages, respectively */
82 struct ext_wait_queue e_wait_q[2];
84 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static const struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
92 static struct kmem_cache *mqueue_inode_cachep;
94 static struct ctl_table_header *mq_sysctl_table;
96 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98 return container_of(inode, struct mqueue_inode_info, vfs_inode);
102 * This routine should be called with the mq_lock held.
104 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
106 return get_ipc_ns(inode->i_sb->s_fs_info);
109 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
111 struct ipc_namespace *ns;
114 ns = __get_ns_from_inode(inode);
115 spin_unlock(&mq_lock);
119 /* Auxiliary functions to manipulate messages' list */
120 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122 struct rb_node **p, *parent = NULL;
123 struct posix_msg_tree_node *leaf;
125 p = &info->msg_tree.rb_node;
128 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130 if (likely(leaf->priority == msg->m_type))
132 else if (msg->m_type < leaf->priority)
137 if (info->node_cache) {
138 leaf = info->node_cache;
139 info->node_cache = NULL;
141 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
144 INIT_LIST_HEAD(&leaf->msg_list);
146 leaf->priority = msg->m_type;
147 rb_link_node(&leaf->rb_node, parent, p);
148 rb_insert_color(&leaf->rb_node, &info->msg_tree);
150 info->attr.mq_curmsgs++;
151 info->qsize += msg->m_ts;
152 list_add_tail(&msg->m_list, &leaf->msg_list);
156 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
158 struct rb_node **p, *parent = NULL;
159 struct posix_msg_tree_node *leaf;
163 p = &info->msg_tree.rb_node;
167 * During insert, low priorities go to the left and high to the
168 * right. On receive, we want the highest priorities first, so
169 * walk all the way to the right.
174 if (info->attr.mq_curmsgs) {
175 pr_warn_once("Inconsistency in POSIX message queue, "
176 "no tree element, but supposedly messages "
178 info->attr.mq_curmsgs = 0;
182 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
183 if (unlikely(list_empty(&leaf->msg_list))) {
184 pr_warn_once("Inconsistency in POSIX message queue, "
185 "empty leaf node but we haven't implemented "
186 "lazy leaf delete!\n");
187 rb_erase(&leaf->rb_node, &info->msg_tree);
188 if (info->node_cache) {
191 info->node_cache = leaf;
195 msg = list_first_entry(&leaf->msg_list,
196 struct msg_msg, m_list);
197 list_del(&msg->m_list);
198 if (list_empty(&leaf->msg_list)) {
199 rb_erase(&leaf->rb_node, &info->msg_tree);
200 if (info->node_cache) {
203 info->node_cache = leaf;
207 info->attr.mq_curmsgs--;
208 info->qsize -= msg->m_ts;
212 static struct inode *mqueue_get_inode(struct super_block *sb,
213 struct ipc_namespace *ipc_ns, umode_t mode,
214 struct mq_attr *attr)
216 struct user_struct *u = current_user();
220 inode = new_inode(sb);
224 inode->i_ino = get_next_ino();
225 inode->i_mode = mode;
226 inode->i_uid = current_fsuid();
227 inode->i_gid = current_fsgid();
228 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
231 struct mqueue_inode_info *info;
232 unsigned long mq_bytes, mq_treesize;
234 inode->i_fop = &mqueue_file_operations;
235 inode->i_size = FILENT_SIZE;
236 /* mqueue specific info */
237 info = MQUEUE_I(inode);
238 spin_lock_init(&info->lock);
239 init_waitqueue_head(&info->wait_q);
240 INIT_LIST_HEAD(&info->e_wait_q[0].list);
241 INIT_LIST_HEAD(&info->e_wait_q[1].list);
242 info->notify_owner = NULL;
243 info->notify_user_ns = NULL;
245 info->user = NULL; /* set when all is ok */
246 info->msg_tree = RB_ROOT;
247 info->node_cache = NULL;
248 memset(&info->attr, 0, sizeof(info->attr));
249 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
250 ipc_ns->mq_msg_default);
251 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
252 ipc_ns->mq_msgsize_default);
254 info->attr.mq_maxmsg = attr->mq_maxmsg;
255 info->attr.mq_msgsize = attr->mq_msgsize;
258 * We used to allocate a static array of pointers and account
259 * the size of that array as well as one msg_msg struct per
260 * possible message into the queue size. That's no longer
261 * accurate as the queue is now an rbtree and will grow and
262 * shrink depending on usage patterns. We can, however, still
263 * account one msg_msg struct per message, but the nodes are
264 * allocated depending on priority usage, and most programs
265 * only use one, or a handful, of priorities. However, since
266 * this is pinned memory, we need to assume worst case, so
267 * that means the min(mq_maxmsg, max_priorities) * struct
268 * posix_msg_tree_node.
270 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
271 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
272 sizeof(struct posix_msg_tree_node);
274 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
275 info->attr.mq_msgsize);
278 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
279 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
280 spin_unlock(&mq_lock);
281 /* mqueue_evict_inode() releases info->messages */
285 u->mq_bytes += mq_bytes;
286 spin_unlock(&mq_lock);
289 info->user = get_uid(u);
290 } else if (S_ISDIR(mode)) {
292 /* Some things misbehave if size == 0 on a directory */
293 inode->i_size = 2 * DIRENT_SIZE;
294 inode->i_op = &mqueue_dir_inode_operations;
295 inode->i_fop = &simple_dir_operations;
305 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
308 struct ipc_namespace *ns = data;
310 sb->s_blocksize = PAGE_CACHE_SIZE;
311 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
312 sb->s_magic = MQUEUE_MAGIC;
313 sb->s_op = &mqueue_super_ops;
315 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
317 return PTR_ERR(inode);
319 sb->s_root = d_make_root(inode);
325 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
326 int flags, const char *dev_name,
329 if (!(flags & MS_KERNMOUNT)) {
330 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
331 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
332 * over the ipc namespace.
334 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
335 return ERR_PTR(-EPERM);
339 return mount_ns(fs_type, flags, data, mqueue_fill_super);
342 static void init_once(void *foo)
344 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
346 inode_init_once(&p->vfs_inode);
349 static struct inode *mqueue_alloc_inode(struct super_block *sb)
351 struct mqueue_inode_info *ei;
353 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
356 return &ei->vfs_inode;
359 static void mqueue_i_callback(struct rcu_head *head)
361 struct inode *inode = container_of(head, struct inode, i_rcu);
362 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
365 static void mqueue_destroy_inode(struct inode *inode)
367 call_rcu(&inode->i_rcu, mqueue_i_callback);
370 static void mqueue_evict_inode(struct inode *inode)
372 struct mqueue_inode_info *info;
373 struct user_struct *user;
374 unsigned long mq_bytes, mq_treesize;
375 struct ipc_namespace *ipc_ns;
380 if (S_ISDIR(inode->i_mode))
383 ipc_ns = get_ns_from_inode(inode);
384 info = MQUEUE_I(inode);
385 spin_lock(&info->lock);
386 while ((msg = msg_get(info)) != NULL)
388 kfree(info->node_cache);
389 spin_unlock(&info->lock);
391 /* Total amount of bytes accounted for the mqueue */
392 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
393 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
394 sizeof(struct posix_msg_tree_node);
396 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
397 info->attr.mq_msgsize);
402 user->mq_bytes -= mq_bytes;
404 * get_ns_from_inode() ensures that the
405 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
406 * to which we now hold a reference, or it is NULL.
407 * We can't put it here under mq_lock, though.
410 ipc_ns->mq_queues_count--;
411 spin_unlock(&mq_lock);
418 static int mqueue_create(struct inode *dir, struct dentry *dentry,
419 umode_t mode, bool excl)
422 struct mq_attr *attr = dentry->d_fsdata;
424 struct ipc_namespace *ipc_ns;
427 ipc_ns = __get_ns_from_inode(dir);
433 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
434 !capable(CAP_SYS_RESOURCE)) {
438 ipc_ns->mq_queues_count++;
439 spin_unlock(&mq_lock);
441 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
443 error = PTR_ERR(inode);
445 ipc_ns->mq_queues_count--;
450 dir->i_size += DIRENT_SIZE;
451 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
453 d_instantiate(dentry, inode);
457 spin_unlock(&mq_lock);
463 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
465 struct inode *inode = d_inode(dentry);
467 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
468 dir->i_size -= DIRENT_SIZE;
475 * This is routine for system read from queue file.
476 * To avoid mess with doing here some sort of mq_receive we allow
477 * to read only queue size & notification info (the only values
478 * that are interesting from user point of view and aren't accessible
479 * through std routines)
481 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
482 size_t count, loff_t *off)
484 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
485 char buffer[FILENT_SIZE];
488 spin_lock(&info->lock);
489 snprintf(buffer, sizeof(buffer),
490 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
492 info->notify_owner ? info->notify.sigev_notify : 0,
493 (info->notify_owner &&
494 info->notify.sigev_notify == SIGEV_SIGNAL) ?
495 info->notify.sigev_signo : 0,
496 pid_vnr(info->notify_owner));
497 spin_unlock(&info->lock);
498 buffer[sizeof(buffer)-1] = '\0';
500 ret = simple_read_from_buffer(u_data, count, off, buffer,
505 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
509 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
511 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
513 spin_lock(&info->lock);
514 if (task_tgid(current) == info->notify_owner)
515 remove_notification(info);
517 spin_unlock(&info->lock);
521 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
523 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
526 poll_wait(filp, &info->wait_q, poll_tab);
528 spin_lock(&info->lock);
529 if (info->attr.mq_curmsgs)
530 retval = POLLIN | POLLRDNORM;
532 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
533 retval |= POLLOUT | POLLWRNORM;
534 spin_unlock(&info->lock);
539 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
540 static void wq_add(struct mqueue_inode_info *info, int sr,
541 struct ext_wait_queue *ewp)
543 struct ext_wait_queue *walk;
547 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
548 if (walk->task->static_prio <= current->static_prio) {
549 list_add_tail(&ewp->list, &walk->list);
553 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
557 * Puts current task to sleep. Caller must hold queue lock. After return
561 static int wq_sleep(struct mqueue_inode_info *info, int sr,
562 ktime_t *timeout, struct ext_wait_queue *ewp)
567 wq_add(info, sr, ewp);
570 __set_current_state(TASK_INTERRUPTIBLE);
572 spin_unlock(&info->lock);
573 time = schedule_hrtimeout_range_clock(timeout, 0,
574 HRTIMER_MODE_ABS, CLOCK_REALTIME);
576 if (ewp->state == STATE_READY) {
580 spin_lock(&info->lock);
581 if (ewp->state == STATE_READY) {
585 if (signal_pending(current)) {
586 retval = -ERESTARTSYS;
594 list_del(&ewp->list);
596 spin_unlock(&info->lock);
602 * Returns waiting task that should be serviced first or NULL if none exists
604 static struct ext_wait_queue *wq_get_first_waiter(
605 struct mqueue_inode_info *info, int sr)
607 struct list_head *ptr;
609 ptr = info->e_wait_q[sr].list.prev;
610 if (ptr == &info->e_wait_q[sr].list)
612 return list_entry(ptr, struct ext_wait_queue, list);
616 static inline void set_cookie(struct sk_buff *skb, char code)
618 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
622 * The next function is only to split too long sys_mq_timedsend
624 static void __do_notify(struct mqueue_inode_info *info)
627 * invoked when there is registered process and there isn't process
628 * waiting synchronously for message AND state of queue changed from
629 * empty to not empty. Here we are sure that no one is waiting
631 if (info->notify_owner &&
632 info->attr.mq_curmsgs == 1) {
633 struct siginfo sig_i;
634 switch (info->notify.sigev_notify) {
640 sig_i.si_signo = info->notify.sigev_signo;
642 sig_i.si_code = SI_MESGQ;
643 sig_i.si_value = info->notify.sigev_value;
644 /* map current pid/uid into info->owner's namespaces */
646 sig_i.si_pid = task_tgid_nr_ns(current,
647 ns_of_pid(info->notify_owner));
648 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
651 kill_pid_info(info->notify.sigev_signo,
652 &sig_i, info->notify_owner);
655 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
656 netlink_sendskb(info->notify_sock, info->notify_cookie);
659 /* after notification unregisters process */
660 put_pid(info->notify_owner);
661 put_user_ns(info->notify_user_ns);
662 info->notify_owner = NULL;
663 info->notify_user_ns = NULL;
665 wake_up(&info->wait_q);
668 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
669 ktime_t *expires, struct timespec *ts)
671 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
673 if (!timespec_valid(ts))
676 *expires = timespec_to_ktime(*ts);
680 static void remove_notification(struct mqueue_inode_info *info)
682 if (info->notify_owner != NULL &&
683 info->notify.sigev_notify == SIGEV_THREAD) {
684 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
685 netlink_sendskb(info->notify_sock, info->notify_cookie);
687 put_pid(info->notify_owner);
688 put_user_ns(info->notify_user_ns);
689 info->notify_owner = NULL;
690 info->notify_user_ns = NULL;
693 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
696 unsigned long total_size;
698 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
700 if (capable(CAP_SYS_RESOURCE)) {
701 if (attr->mq_maxmsg > HARD_MSGMAX ||
702 attr->mq_msgsize > HARD_MSGSIZEMAX)
705 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
706 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
709 /* check for overflow */
710 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
712 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
713 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
714 sizeof(struct posix_msg_tree_node);
715 total_size = attr->mq_maxmsg * attr->mq_msgsize;
716 if (total_size + mq_treesize < total_size)
722 * Invoked when creating a new queue via sys_mq_open
724 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
725 struct path *path, int oflag, umode_t mode,
726 struct mq_attr *attr)
728 const struct cred *cred = current_cred();
732 ret = mq_attr_ok(ipc_ns, attr);
735 /* store for use during create */
736 path->dentry->d_fsdata = attr;
738 struct mq_attr def_attr;
740 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
741 ipc_ns->mq_msg_default);
742 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
743 ipc_ns->mq_msgsize_default);
744 ret = mq_attr_ok(ipc_ns, &def_attr);
749 mode &= ~current_umask();
750 ret = vfs_create(dir, path->dentry, mode, true);
751 path->dentry->d_fsdata = NULL;
754 return dentry_open(path, oflag, cred);
757 /* Opens existing queue */
758 static struct file *do_open(struct path *path, int oflag)
760 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
761 MAY_READ | MAY_WRITE };
763 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
764 return ERR_PTR(-EINVAL);
765 acc = oflag2acc[oflag & O_ACCMODE];
766 if (inode_permission(d_inode(path->dentry), acc))
767 return ERR_PTR(-EACCES);
768 return dentry_open(path, oflag, current_cred());
771 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
772 struct mq_attr __user *, u_attr)
776 struct filename *name;
779 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
780 struct vfsmount *mnt = ipc_ns->mq_mnt;
781 struct dentry *root = mnt->mnt_root;
784 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
787 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
789 if (IS_ERR(name = getname(u_name)))
790 return PTR_ERR(name);
792 fd = get_unused_fd_flags(O_CLOEXEC);
796 ro = mnt_want_write(mnt); /* we'll drop it in any case */
798 mutex_lock(&d_inode(root)->i_mutex);
799 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
800 if (IS_ERR(path.dentry)) {
801 error = PTR_ERR(path.dentry);
804 path.mnt = mntget(mnt);
806 if (oflag & O_CREAT) {
807 if (d_really_is_positive(path.dentry)) { /* entry already exists */
808 audit_inode(name, path.dentry, 0);
809 if (oflag & O_EXCL) {
813 filp = do_open(&path, oflag);
819 audit_inode_parent_hidden(name, root);
820 filp = do_create(ipc_ns, d_inode(root),
822 u_attr ? &attr : NULL);
825 if (d_really_is_negative(path.dentry)) {
829 audit_inode(name, path.dentry, 0);
830 filp = do_open(&path, oflag);
834 fd_install(fd, filp);
836 error = PTR_ERR(filp);
844 mutex_unlock(&d_inode(root)->i_mutex);
852 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
855 struct filename *name;
856 struct dentry *dentry;
857 struct inode *inode = NULL;
858 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
859 struct vfsmount *mnt = ipc_ns->mq_mnt;
861 name = getname(u_name);
863 return PTR_ERR(name);
865 audit_inode_parent_hidden(name, mnt->mnt_root);
866 err = mnt_want_write(mnt);
869 mutex_lock_nested(&d_inode(mnt->mnt_root)->i_mutex, I_MUTEX_PARENT);
870 dentry = lookup_one_len(name->name, mnt->mnt_root,
872 if (IS_ERR(dentry)) {
873 err = PTR_ERR(dentry);
877 inode = d_inode(dentry);
882 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
887 mutex_unlock(&d_inode(mnt->mnt_root)->i_mutex);
897 /* Pipelined send and receive functions.
899 * If a receiver finds no waiting message, then it registers itself in the
900 * list of waiting receivers. A sender checks that list before adding the new
901 * message into the message array. If there is a waiting receiver, then it
902 * bypasses the message array and directly hands the message over to the
903 * receiver. The receiver accepts the message and returns without grabbing the
906 * - Set pointer to message.
907 * - Queue the receiver task for later wakeup (without the info->lock).
908 * - Update its state to STATE_READY. Now the receiver can continue.
909 * - Wake up the process after the lock is dropped. Should the process wake up
910 * before this wakeup (due to a timeout or a signal) it will either see
911 * STATE_READY and continue or acquire the lock to check the state again.
913 * The same algorithm is used for senders.
916 /* pipelined_send() - send a message directly to the task waiting in
917 * sys_mq_timedreceive() (without inserting message into a queue).
919 static inline void pipelined_send(struct wake_q_head *wake_q,
920 struct mqueue_inode_info *info,
921 struct msg_msg *message,
922 struct ext_wait_queue *receiver)
924 receiver->msg = message;
925 list_del(&receiver->list);
926 wake_q_add(wake_q, receiver->task);
928 * Rely on the implicit cmpxchg barrier from wake_q_add such
929 * that we can ensure that updating receiver->state is the last
930 * write operation: As once set, the receiver can continue,
931 * and if we don't have the reference count from the wake_q,
932 * yet, at that point we can later have a use-after-free
933 * condition and bogus wakeup.
935 receiver->state = STATE_READY;
938 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
939 * gets its message and put to the queue (we have one free place for sure). */
940 static inline void pipelined_receive(struct wake_q_head *wake_q,
941 struct mqueue_inode_info *info)
943 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
947 wake_up_interruptible(&info->wait_q);
950 if (msg_insert(sender->msg, info))
953 list_del(&sender->list);
954 wake_q_add(wake_q, sender->task);
955 sender->state = STATE_READY;
958 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
959 size_t, msg_len, unsigned int, msg_prio,
960 const struct timespec __user *, u_abs_timeout)
964 struct ext_wait_queue wait;
965 struct ext_wait_queue *receiver;
966 struct msg_msg *msg_ptr;
967 struct mqueue_inode_info *info;
968 ktime_t expires, *timeout = NULL;
970 struct posix_msg_tree_node *new_leaf = NULL;
975 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
981 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
984 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
987 if (unlikely(!f.file)) {
992 inode = file_inode(f.file);
993 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
997 info = MQUEUE_I(inode);
1000 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1005 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1010 /* First try to allocate memory, before doing anything with
1011 * existing queues. */
1012 msg_ptr = load_msg(u_msg_ptr, msg_len);
1013 if (IS_ERR(msg_ptr)) {
1014 ret = PTR_ERR(msg_ptr);
1017 msg_ptr->m_ts = msg_len;
1018 msg_ptr->m_type = msg_prio;
1021 * msg_insert really wants us to have a valid, spare node struct so
1022 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1023 * fall back to that if necessary.
1025 if (!info->node_cache)
1026 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1028 spin_lock(&info->lock);
1030 if (!info->node_cache && new_leaf) {
1031 /* Save our speculative allocation into the cache */
1032 INIT_LIST_HEAD(&new_leaf->msg_list);
1033 info->node_cache = new_leaf;
1039 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1040 if (f.file->f_flags & O_NONBLOCK) {
1043 wait.task = current;
1044 wait.msg = (void *) msg_ptr;
1045 wait.state = STATE_NONE;
1046 ret = wq_sleep(info, SEND, timeout, &wait);
1048 * wq_sleep must be called with info->lock held, and
1049 * returns with the lock released
1054 receiver = wq_get_first_waiter(info, RECV);
1056 pipelined_send(&wake_q, info, msg_ptr, receiver);
1058 /* adds message to the queue */
1059 ret = msg_insert(msg_ptr, info);
1064 inode->i_atime = inode->i_mtime = inode->i_ctime =
1068 spin_unlock(&info->lock);
1079 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1080 size_t, msg_len, unsigned int __user *, u_msg_prio,
1081 const struct timespec __user *, u_abs_timeout)
1084 struct msg_msg *msg_ptr;
1086 struct inode *inode;
1087 struct mqueue_inode_info *info;
1088 struct ext_wait_queue wait;
1089 ktime_t expires, *timeout = NULL;
1091 struct posix_msg_tree_node *new_leaf = NULL;
1093 if (u_abs_timeout) {
1094 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1100 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1103 if (unlikely(!f.file)) {
1108 inode = file_inode(f.file);
1109 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1113 info = MQUEUE_I(inode);
1116 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1121 /* checks if buffer is big enough */
1122 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1128 * msg_insert really wants us to have a valid, spare node struct so
1129 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1130 * fall back to that if necessary.
1132 if (!info->node_cache)
1133 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1135 spin_lock(&info->lock);
1137 if (!info->node_cache && new_leaf) {
1138 /* Save our speculative allocation into the cache */
1139 INIT_LIST_HEAD(&new_leaf->msg_list);
1140 info->node_cache = new_leaf;
1145 if (info->attr.mq_curmsgs == 0) {
1146 if (f.file->f_flags & O_NONBLOCK) {
1147 spin_unlock(&info->lock);
1150 wait.task = current;
1151 wait.state = STATE_NONE;
1152 ret = wq_sleep(info, RECV, timeout, &wait);
1158 msg_ptr = msg_get(info);
1160 inode->i_atime = inode->i_mtime = inode->i_ctime =
1163 /* There is now free space in queue. */
1164 pipelined_receive(&wake_q, info);
1165 spin_unlock(&info->lock);
1170 ret = msg_ptr->m_ts;
1172 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1173 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1185 * Notes: the case when user wants us to deregister (with NULL as pointer)
1186 * and he isn't currently owner of notification, will be silently discarded.
1187 * It isn't explicitly defined in the POSIX.
1189 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1190 const struct sigevent __user *, u_notification)
1195 struct inode *inode;
1196 struct sigevent notification;
1197 struct mqueue_inode_info *info;
1200 if (u_notification) {
1201 if (copy_from_user(¬ification, u_notification,
1202 sizeof(struct sigevent)))
1206 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1210 if (u_notification != NULL) {
1211 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1212 notification.sigev_notify != SIGEV_SIGNAL &&
1213 notification.sigev_notify != SIGEV_THREAD))
1215 if (notification.sigev_notify == SIGEV_SIGNAL &&
1216 !valid_signal(notification.sigev_signo)) {
1219 if (notification.sigev_notify == SIGEV_THREAD) {
1222 /* create the notify skb */
1223 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1228 if (copy_from_user(nc->data,
1229 notification.sigev_value.sival_ptr,
1230 NOTIFY_COOKIE_LEN)) {
1235 /* TODO: add a header? */
1236 skb_put(nc, NOTIFY_COOKIE_LEN);
1237 /* and attach it to the socket */
1239 f = fdget(notification.sigev_signo);
1244 sock = netlink_getsockbyfilp(f.file);
1247 ret = PTR_ERR(sock);
1252 timeo = MAX_SCHEDULE_TIMEOUT;
1253 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1270 inode = file_inode(f.file);
1271 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1275 info = MQUEUE_I(inode);
1278 spin_lock(&info->lock);
1279 if (u_notification == NULL) {
1280 if (info->notify_owner == task_tgid(current)) {
1281 remove_notification(info);
1282 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1284 } else if (info->notify_owner != NULL) {
1287 switch (notification.sigev_notify) {
1289 info->notify.sigev_notify = SIGEV_NONE;
1292 info->notify_sock = sock;
1293 info->notify_cookie = nc;
1296 info->notify.sigev_notify = SIGEV_THREAD;
1299 info->notify.sigev_signo = notification.sigev_signo;
1300 info->notify.sigev_value = notification.sigev_value;
1301 info->notify.sigev_notify = SIGEV_SIGNAL;
1305 info->notify_owner = get_pid(task_tgid(current));
1306 info->notify_user_ns = get_user_ns(current_user_ns());
1307 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1309 spin_unlock(&info->lock);
1314 netlink_detachskb(sock, nc);
1321 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1322 const struct mq_attr __user *, u_mqstat,
1323 struct mq_attr __user *, u_omqstat)
1326 struct mq_attr mqstat, omqstat;
1328 struct inode *inode;
1329 struct mqueue_inode_info *info;
1331 if (u_mqstat != NULL) {
1332 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1334 if (mqstat.mq_flags & (~O_NONBLOCK))
1344 inode = file_inode(f.file);
1345 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1349 info = MQUEUE_I(inode);
1351 spin_lock(&info->lock);
1353 omqstat = info->attr;
1354 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1356 audit_mq_getsetattr(mqdes, &mqstat);
1357 spin_lock(&f.file->f_lock);
1358 if (mqstat.mq_flags & O_NONBLOCK)
1359 f.file->f_flags |= O_NONBLOCK;
1361 f.file->f_flags &= ~O_NONBLOCK;
1362 spin_unlock(&f.file->f_lock);
1364 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1367 spin_unlock(&info->lock);
1370 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1371 sizeof(struct mq_attr)))
1380 static const struct inode_operations mqueue_dir_inode_operations = {
1381 .lookup = simple_lookup,
1382 .create = mqueue_create,
1383 .unlink = mqueue_unlink,
1386 static const struct file_operations mqueue_file_operations = {
1387 .flush = mqueue_flush_file,
1388 .poll = mqueue_poll_file,
1389 .read = mqueue_read_file,
1390 .llseek = default_llseek,
1393 static const struct super_operations mqueue_super_ops = {
1394 .alloc_inode = mqueue_alloc_inode,
1395 .destroy_inode = mqueue_destroy_inode,
1396 .evict_inode = mqueue_evict_inode,
1397 .statfs = simple_statfs,
1400 static struct file_system_type mqueue_fs_type = {
1402 .mount = mqueue_mount,
1403 .kill_sb = kill_litter_super,
1404 .fs_flags = FS_USERNS_MOUNT,
1407 int mq_init_ns(struct ipc_namespace *ns)
1409 ns->mq_queues_count = 0;
1410 ns->mq_queues_max = DFLT_QUEUESMAX;
1411 ns->mq_msg_max = DFLT_MSGMAX;
1412 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1413 ns->mq_msg_default = DFLT_MSG;
1414 ns->mq_msgsize_default = DFLT_MSGSIZE;
1416 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1417 if (IS_ERR(ns->mq_mnt)) {
1418 int err = PTR_ERR(ns->mq_mnt);
1425 void mq_clear_sbinfo(struct ipc_namespace *ns)
1427 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1430 void mq_put_mnt(struct ipc_namespace *ns)
1432 kern_unmount(ns->mq_mnt);
1435 static int __init init_mqueue_fs(void)
1439 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1440 sizeof(struct mqueue_inode_info), 0,
1441 SLAB_HWCACHE_ALIGN, init_once);
1442 if (mqueue_inode_cachep == NULL)
1445 /* ignore failures - they are not fatal */
1446 mq_sysctl_table = mq_register_sysctl_table();
1448 error = register_filesystem(&mqueue_fs_type);
1452 spin_lock_init(&mq_lock);
1454 error = mq_init_ns(&init_ipc_ns);
1456 goto out_filesystem;
1461 unregister_filesystem(&mqueue_fs_type);
1463 if (mq_sysctl_table)
1464 unregister_sysctl_table(mq_sysctl_table);
1465 kmem_cache_destroy(mqueue_inode_cachep);
1469 device_initcall(init_mqueue_fs);