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
50 #define STATE_PENDING 1
53 struct posix_msg_tree_node {
54 struct rb_node rb_node;
55 struct list_head msg_list;
59 struct ext_wait_queue { /* queue of sleeping tasks */
60 struct task_struct *task;
61 struct list_head list;
62 struct msg_msg *msg; /* ptr of loaded message */
63 int state; /* one of STATE_* values */
66 struct mqueue_inode_info {
68 struct inode vfs_inode;
69 wait_queue_head_t wait_q;
71 struct rb_root msg_tree;
72 struct posix_msg_tree_node *node_cache;
75 struct sigevent notify;
76 struct pid* notify_owner;
77 struct user_namespace *notify_user_ns;
78 struct user_struct *user; /* user who created, for accounting */
79 struct sock *notify_sock;
80 struct sk_buff *notify_cookie;
82 /* for tasks waiting for free space and messages, respectively */
83 struct ext_wait_queue e_wait_q[2];
85 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
88 static const struct inode_operations mqueue_dir_inode_operations;
89 static const struct file_operations mqueue_file_operations;
90 static const struct super_operations mqueue_super_ops;
91 static void remove_notification(struct mqueue_inode_info *info);
93 static struct kmem_cache *mqueue_inode_cachep;
95 static struct ctl_table_header * mq_sysctl_table;
97 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
99 return container_of(inode, struct mqueue_inode_info, vfs_inode);
103 * This routine should be called with the mq_lock held.
105 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
107 return get_ipc_ns(inode->i_sb->s_fs_info);
110 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
112 struct ipc_namespace *ns;
115 ns = __get_ns_from_inode(inode);
116 spin_unlock(&mq_lock);
120 /* Auxiliary functions to manipulate messages' list */
121 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
123 struct rb_node **p, *parent = NULL;
124 struct posix_msg_tree_node *leaf;
126 p = &info->msg_tree.rb_node;
129 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
131 if (likely(leaf->priority == msg->m_type))
133 else if (msg->m_type < leaf->priority)
138 if (info->node_cache) {
139 leaf = info->node_cache;
140 info->node_cache = NULL;
142 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
145 INIT_LIST_HEAD(&leaf->msg_list);
147 leaf->priority = msg->m_type;
148 rb_link_node(&leaf->rb_node, parent, p);
149 rb_insert_color(&leaf->rb_node, &info->msg_tree);
151 info->attr.mq_curmsgs++;
152 info->qsize += msg->m_ts;
153 list_add_tail(&msg->m_list, &leaf->msg_list);
157 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
159 struct rb_node **p, *parent = NULL;
160 struct posix_msg_tree_node *leaf;
164 p = &info->msg_tree.rb_node;
168 * During insert, low priorities go to the left and high to the
169 * right. On receive, we want the highest priorities first, so
170 * walk all the way to the right.
175 if (info->attr.mq_curmsgs) {
176 pr_warn_once("Inconsistency in POSIX message queue, "
177 "no tree element, but supposedly messages "
179 info->attr.mq_curmsgs = 0;
183 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
184 if (unlikely(list_empty(&leaf->msg_list))) {
185 pr_warn_once("Inconsistency in POSIX message queue, "
186 "empty leaf node but we haven't implemented "
187 "lazy leaf delete!\n");
188 rb_erase(&leaf->rb_node, &info->msg_tree);
189 if (info->node_cache) {
192 info->node_cache = leaf;
196 msg = list_first_entry(&leaf->msg_list,
197 struct msg_msg, m_list);
198 list_del(&msg->m_list);
199 if (list_empty(&leaf->msg_list)) {
200 rb_erase(&leaf->rb_node, &info->msg_tree);
201 if (info->node_cache) {
204 info->node_cache = leaf;
208 info->attr.mq_curmsgs--;
209 info->qsize -= msg->m_ts;
213 static struct inode *mqueue_get_inode(struct super_block *sb,
214 struct ipc_namespace *ipc_ns, umode_t mode,
215 struct mq_attr *attr)
217 struct user_struct *u = current_user();
221 inode = new_inode(sb);
225 inode->i_ino = get_next_ino();
226 inode->i_mode = mode;
227 inode->i_uid = current_fsuid();
228 inode->i_gid = current_fsgid();
229 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
232 struct mqueue_inode_info *info;
233 unsigned long mq_bytes, mq_treesize;
235 inode->i_fop = &mqueue_file_operations;
236 inode->i_size = FILENT_SIZE;
237 /* mqueue specific info */
238 info = MQUEUE_I(inode);
239 spin_lock_init(&info->lock);
240 init_waitqueue_head(&info->wait_q);
241 INIT_LIST_HEAD(&info->e_wait_q[0].list);
242 INIT_LIST_HEAD(&info->e_wait_q[1].list);
243 info->notify_owner = NULL;
244 info->notify_user_ns = NULL;
246 info->user = NULL; /* set when all is ok */
247 info->msg_tree = RB_ROOT;
248 info->node_cache = NULL;
249 memset(&info->attr, 0, sizeof(info->attr));
250 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
251 ipc_ns->mq_msg_default);
252 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
253 ipc_ns->mq_msgsize_default);
255 info->attr.mq_maxmsg = attr->mq_maxmsg;
256 info->attr.mq_msgsize = attr->mq_msgsize;
259 * We used to allocate a static array of pointers and account
260 * the size of that array as well as one msg_msg struct per
261 * possible message into the queue size. That's no longer
262 * accurate as the queue is now an rbtree and will grow and
263 * shrink depending on usage patterns. We can, however, still
264 * account one msg_msg struct per message, but the nodes are
265 * allocated depending on priority usage, and most programs
266 * only use one, or a handful, of priorities. However, since
267 * this is pinned memory, we need to assume worst case, so
268 * that means the min(mq_maxmsg, max_priorities) * struct
269 * posix_msg_tree_node.
271 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
272 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
273 sizeof(struct posix_msg_tree_node);
275 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
276 info->attr.mq_msgsize);
279 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
280 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
281 spin_unlock(&mq_lock);
282 /* mqueue_evict_inode() releases info->messages */
286 u->mq_bytes += mq_bytes;
287 spin_unlock(&mq_lock);
290 info->user = get_uid(u);
291 } else if (S_ISDIR(mode)) {
293 /* Some things misbehave if size == 0 on a directory */
294 inode->i_size = 2 * DIRENT_SIZE;
295 inode->i_op = &mqueue_dir_inode_operations;
296 inode->i_fop = &simple_dir_operations;
306 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
309 struct ipc_namespace *ns = data;
311 sb->s_blocksize = PAGE_CACHE_SIZE;
312 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
313 sb->s_magic = MQUEUE_MAGIC;
314 sb->s_op = &mqueue_super_ops;
316 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
318 return PTR_ERR(inode);
320 sb->s_root = d_make_root(inode);
326 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
327 int flags, const char *dev_name,
330 if (!(flags & MS_KERNMOUNT)) {
331 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
332 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
333 * over the ipc namespace.
335 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
336 return ERR_PTR(-EPERM);
340 return mount_ns(fs_type, flags, data, 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;
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 = dentry->d_inode;
468 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
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;
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)
568 wq_add(info, sr, ewp);
571 set_current_state(TASK_INTERRUPTIBLE);
573 spin_unlock(&info->lock);
574 time = schedule_hrtimeout_range_clock(timeout, 0,
575 HRTIMER_MODE_ABS, CLOCK_REALTIME);
577 while (ewp->state == STATE_PENDING)
580 if (ewp->state == STATE_READY) {
584 spin_lock(&info->lock);
585 if (ewp->state == STATE_READY) {
589 if (signal_pending(current)) {
590 retval = -ERESTARTSYS;
598 list_del(&ewp->list);
600 spin_unlock(&info->lock);
606 * Returns waiting task that should be serviced first or NULL if none exists
608 static struct ext_wait_queue *wq_get_first_waiter(
609 struct mqueue_inode_info *info, int sr)
611 struct list_head *ptr;
613 ptr = info->e_wait_q[sr].list.prev;
614 if (ptr == &info->e_wait_q[sr].list)
616 return list_entry(ptr, struct ext_wait_queue, list);
620 static inline void set_cookie(struct sk_buff *skb, char code)
622 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
626 * The next function is only to split too long sys_mq_timedsend
628 static void __do_notify(struct mqueue_inode_info *info)
631 * invoked when there is registered process and there isn't process
632 * waiting synchronously for message AND state of queue changed from
633 * empty to not empty. Here we are sure that no one is waiting
635 if (info->notify_owner &&
636 info->attr.mq_curmsgs == 1) {
637 struct siginfo sig_i;
638 switch (info->notify.sigev_notify) {
644 sig_i.si_signo = info->notify.sigev_signo;
646 sig_i.si_code = SI_MESGQ;
647 sig_i.si_value = info->notify.sigev_value;
648 /* map current pid/uid into info->owner's namespaces */
650 sig_i.si_pid = task_tgid_nr_ns(current,
651 ns_of_pid(info->notify_owner));
652 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
655 kill_pid_info(info->notify.sigev_signo,
656 &sig_i, info->notify_owner);
659 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
660 netlink_sendskb(info->notify_sock, info->notify_cookie);
663 /* after notification unregisters process */
664 put_pid(info->notify_owner);
665 put_user_ns(info->notify_user_ns);
666 info->notify_owner = NULL;
667 info->notify_user_ns = NULL;
669 wake_up(&info->wait_q);
672 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
673 ktime_t *expires, struct timespec *ts)
675 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
677 if (!timespec_valid(ts))
680 *expires = timespec_to_ktime(*ts);
684 static void remove_notification(struct mqueue_inode_info *info)
686 if (info->notify_owner != NULL &&
687 info->notify.sigev_notify == SIGEV_THREAD) {
688 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
689 netlink_sendskb(info->notify_sock, info->notify_cookie);
691 put_pid(info->notify_owner);
692 put_user_ns(info->notify_user_ns);
693 info->notify_owner = NULL;
694 info->notify_user_ns = NULL;
697 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
700 unsigned long total_size;
702 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
704 if (capable(CAP_SYS_RESOURCE)) {
705 if (attr->mq_maxmsg > HARD_MSGMAX ||
706 attr->mq_msgsize > HARD_MSGSIZEMAX)
709 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
710 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
713 /* check for overflow */
714 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
716 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
717 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
718 sizeof(struct posix_msg_tree_node);
719 total_size = attr->mq_maxmsg * attr->mq_msgsize;
720 if (total_size + mq_treesize < total_size)
726 * Invoked when creating a new queue via sys_mq_open
728 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
729 struct path *path, int oflag, umode_t mode,
730 struct mq_attr *attr)
732 const struct cred *cred = current_cred();
736 ret = mq_attr_ok(ipc_ns, attr);
739 /* store for use during create */
740 path->dentry->d_fsdata = attr;
742 struct mq_attr def_attr;
744 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
745 ipc_ns->mq_msg_default);
746 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
747 ipc_ns->mq_msgsize_default);
748 ret = mq_attr_ok(ipc_ns, &def_attr);
753 mode &= ~current_umask();
754 ret = vfs_create(dir, path->dentry, mode, true);
755 path->dentry->d_fsdata = NULL;
758 return dentry_open(path, oflag, cred);
761 /* Opens existing queue */
762 static struct file *do_open(struct path *path, int oflag)
764 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
765 MAY_READ | MAY_WRITE };
767 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
768 return ERR_PTR(-EINVAL);
769 acc = oflag2acc[oflag & O_ACCMODE];
770 if (inode_permission(path->dentry->d_inode, acc))
771 return ERR_PTR(-EACCES);
772 return dentry_open(path, oflag, current_cred());
775 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
776 struct mq_attr __user *, u_attr)
780 struct filename *name;
783 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
784 struct vfsmount *mnt = ipc_ns->mq_mnt;
785 struct dentry *root = mnt->mnt_root;
788 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
791 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
793 if (IS_ERR(name = getname(u_name)))
794 return PTR_ERR(name);
796 fd = get_unused_fd_flags(O_CLOEXEC);
800 ro = mnt_want_write(mnt); /* we'll drop it in any case */
802 mutex_lock(&root->d_inode->i_mutex);
803 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
804 if (IS_ERR(path.dentry)) {
805 error = PTR_ERR(path.dentry);
808 path.mnt = mntget(mnt);
810 if (oflag & O_CREAT) {
811 if (path.dentry->d_inode) { /* entry already exists */
812 audit_inode(name, path.dentry, 0);
813 if (oflag & O_EXCL) {
817 filp = do_open(&path, oflag);
823 audit_inode_parent_hidden(name, root);
824 filp = do_create(ipc_ns, root->d_inode,
826 u_attr ? &attr : NULL);
829 if (!path.dentry->d_inode) {
833 audit_inode(name, path.dentry, 0);
834 filp = do_open(&path, oflag);
838 fd_install(fd, filp);
840 error = PTR_ERR(filp);
848 mutex_unlock(&root->d_inode->i_mutex);
856 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
859 struct filename *name;
860 struct dentry *dentry;
861 struct inode *inode = NULL;
862 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
863 struct vfsmount *mnt = ipc_ns->mq_mnt;
865 name = getname(u_name);
867 return PTR_ERR(name);
869 audit_inode_parent_hidden(name, mnt->mnt_root);
870 err = mnt_want_write(mnt);
873 mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
874 dentry = lookup_one_len(name->name, mnt->mnt_root,
876 if (IS_ERR(dentry)) {
877 err = PTR_ERR(dentry);
881 inode = dentry->d_inode;
886 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
891 mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
901 /* Pipelined send and receive functions.
903 * If a receiver finds no waiting message, then it registers itself in the
904 * list of waiting receivers. A sender checks that list before adding the new
905 * message into the message array. If there is a waiting receiver, then it
906 * bypasses the message array and directly hands the message over to the
908 * The receiver accepts the message and returns without grabbing the queue
909 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
910 * are necessary. The same algorithm is used for sysv semaphores, see
911 * ipc/sem.c for more details.
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 mqueue_inode_info *info,
920 struct msg_msg *message,
921 struct ext_wait_queue *receiver)
923 receiver->msg = message;
924 list_del(&receiver->list);
925 receiver->state = STATE_PENDING;
926 wake_up_process(receiver->task);
928 receiver->state = STATE_READY;
931 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
932 * gets its message and put to the queue (we have one free place for sure). */
933 static inline void pipelined_receive(struct mqueue_inode_info *info)
935 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
939 wake_up_interruptible(&info->wait_q);
942 if (msg_insert(sender->msg, info))
944 list_del(&sender->list);
945 sender->state = STATE_PENDING;
946 wake_up_process(sender->task);
948 sender->state = STATE_READY;
951 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
952 size_t, msg_len, unsigned int, msg_prio,
953 const struct timespec __user *, u_abs_timeout)
957 struct ext_wait_queue wait;
958 struct ext_wait_queue *receiver;
959 struct msg_msg *msg_ptr;
960 struct mqueue_inode_info *info;
961 ktime_t expires, *timeout = NULL;
963 struct posix_msg_tree_node *new_leaf = NULL;
967 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
973 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
976 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
979 if (unlikely(!f.file)) {
984 inode = file_inode(f.file);
985 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
989 info = MQUEUE_I(inode);
990 audit_inode(NULL, f.file->f_path.dentry, 0);
992 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
997 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1002 /* First try to allocate memory, before doing anything with
1003 * existing queues. */
1004 msg_ptr = load_msg(u_msg_ptr, msg_len);
1005 if (IS_ERR(msg_ptr)) {
1006 ret = PTR_ERR(msg_ptr);
1009 msg_ptr->m_ts = msg_len;
1010 msg_ptr->m_type = msg_prio;
1013 * msg_insert really wants us to have a valid, spare node struct so
1014 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1015 * fall back to that if necessary.
1017 if (!info->node_cache)
1018 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1020 spin_lock(&info->lock);
1022 if (!info->node_cache && new_leaf) {
1023 /* Save our speculative allocation into the cache */
1024 INIT_LIST_HEAD(&new_leaf->msg_list);
1025 info->node_cache = new_leaf;
1031 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1032 if (f.file->f_flags & O_NONBLOCK) {
1035 wait.task = current;
1036 wait.msg = (void *) msg_ptr;
1037 wait.state = STATE_NONE;
1038 ret = wq_sleep(info, SEND, timeout, &wait);
1040 * wq_sleep must be called with info->lock held, and
1041 * returns with the lock released
1046 receiver = wq_get_first_waiter(info, RECV);
1048 pipelined_send(info, msg_ptr, receiver);
1050 /* adds message to the queue */
1051 ret = msg_insert(msg_ptr, info);
1056 inode->i_atime = inode->i_mtime = inode->i_ctime =
1060 spin_unlock(&info->lock);
1070 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1071 size_t, msg_len, unsigned int __user *, u_msg_prio,
1072 const struct timespec __user *, u_abs_timeout)
1075 struct msg_msg *msg_ptr;
1077 struct inode *inode;
1078 struct mqueue_inode_info *info;
1079 struct ext_wait_queue wait;
1080 ktime_t expires, *timeout = NULL;
1082 struct posix_msg_tree_node *new_leaf = NULL;
1084 if (u_abs_timeout) {
1085 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1091 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1094 if (unlikely(!f.file)) {
1099 inode = file_inode(f.file);
1100 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1104 info = MQUEUE_I(inode);
1105 audit_inode(NULL, f.file->f_path.dentry, 0);
1107 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1112 /* checks if buffer is big enough */
1113 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1119 * msg_insert really wants us to have a valid, spare node struct so
1120 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1121 * fall back to that if necessary.
1123 if (!info->node_cache)
1124 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1126 spin_lock(&info->lock);
1128 if (!info->node_cache && new_leaf) {
1129 /* Save our speculative allocation into the cache */
1130 INIT_LIST_HEAD(&new_leaf->msg_list);
1131 info->node_cache = new_leaf;
1136 if (info->attr.mq_curmsgs == 0) {
1137 if (f.file->f_flags & O_NONBLOCK) {
1138 spin_unlock(&info->lock);
1141 wait.task = current;
1142 wait.state = STATE_NONE;
1143 ret = wq_sleep(info, RECV, timeout, &wait);
1147 msg_ptr = msg_get(info);
1149 inode->i_atime = inode->i_mtime = inode->i_ctime =
1152 /* There is now free space in queue. */
1153 pipelined_receive(info);
1154 spin_unlock(&info->lock);
1158 ret = msg_ptr->m_ts;
1160 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1161 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1173 * Notes: the case when user wants us to deregister (with NULL as pointer)
1174 * and he isn't currently owner of notification, will be silently discarded.
1175 * It isn't explicitly defined in the POSIX.
1177 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1178 const struct sigevent __user *, u_notification)
1183 struct inode *inode;
1184 struct sigevent notification;
1185 struct mqueue_inode_info *info;
1188 if (u_notification) {
1189 if (copy_from_user(¬ification, u_notification,
1190 sizeof(struct sigevent)))
1194 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1198 if (u_notification != NULL) {
1199 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1200 notification.sigev_notify != SIGEV_SIGNAL &&
1201 notification.sigev_notify != SIGEV_THREAD))
1203 if (notification.sigev_notify == SIGEV_SIGNAL &&
1204 !valid_signal(notification.sigev_signo)) {
1207 if (notification.sigev_notify == SIGEV_THREAD) {
1210 /* create the notify skb */
1211 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1216 if (copy_from_user(nc->data,
1217 notification.sigev_value.sival_ptr,
1218 NOTIFY_COOKIE_LEN)) {
1223 /* TODO: add a header? */
1224 skb_put(nc, NOTIFY_COOKIE_LEN);
1225 /* and attach it to the socket */
1227 f = fdget(notification.sigev_signo);
1232 sock = netlink_getsockbyfilp(f.file);
1235 ret = PTR_ERR(sock);
1240 timeo = MAX_SCHEDULE_TIMEOUT;
1241 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1258 inode = file_inode(f.file);
1259 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1263 info = MQUEUE_I(inode);
1266 spin_lock(&info->lock);
1267 if (u_notification == NULL) {
1268 if (info->notify_owner == task_tgid(current)) {
1269 remove_notification(info);
1270 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1272 } else if (info->notify_owner != NULL) {
1275 switch (notification.sigev_notify) {
1277 info->notify.sigev_notify = SIGEV_NONE;
1280 info->notify_sock = sock;
1281 info->notify_cookie = nc;
1284 info->notify.sigev_notify = SIGEV_THREAD;
1287 info->notify.sigev_signo = notification.sigev_signo;
1288 info->notify.sigev_value = notification.sigev_value;
1289 info->notify.sigev_notify = SIGEV_SIGNAL;
1293 info->notify_owner = get_pid(task_tgid(current));
1294 info->notify_user_ns = get_user_ns(current_user_ns());
1295 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1297 spin_unlock(&info->lock);
1302 netlink_detachskb(sock, nc);
1309 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1310 const struct mq_attr __user *, u_mqstat,
1311 struct mq_attr __user *, u_omqstat)
1314 struct mq_attr mqstat, omqstat;
1316 struct inode *inode;
1317 struct mqueue_inode_info *info;
1319 if (u_mqstat != NULL) {
1320 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1322 if (mqstat.mq_flags & (~O_NONBLOCK))
1332 inode = file_inode(f.file);
1333 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1337 info = MQUEUE_I(inode);
1339 spin_lock(&info->lock);
1341 omqstat = info->attr;
1342 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1344 audit_mq_getsetattr(mqdes, &mqstat);
1345 spin_lock(&f.file->f_lock);
1346 if (mqstat.mq_flags & O_NONBLOCK)
1347 f.file->f_flags |= O_NONBLOCK;
1349 f.file->f_flags &= ~O_NONBLOCK;
1350 spin_unlock(&f.file->f_lock);
1352 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1355 spin_unlock(&info->lock);
1358 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1359 sizeof(struct mq_attr)))
1368 static const struct inode_operations mqueue_dir_inode_operations = {
1369 .lookup = simple_lookup,
1370 .create = mqueue_create,
1371 .unlink = mqueue_unlink,
1374 static const struct file_operations mqueue_file_operations = {
1375 .flush = mqueue_flush_file,
1376 .poll = mqueue_poll_file,
1377 .read = mqueue_read_file,
1378 .llseek = default_llseek,
1381 static const struct super_operations mqueue_super_ops = {
1382 .alloc_inode = mqueue_alloc_inode,
1383 .destroy_inode = mqueue_destroy_inode,
1384 .evict_inode = mqueue_evict_inode,
1385 .statfs = simple_statfs,
1388 static struct file_system_type mqueue_fs_type = {
1390 .mount = mqueue_mount,
1391 .kill_sb = kill_litter_super,
1392 .fs_flags = FS_USERNS_MOUNT,
1395 int mq_init_ns(struct ipc_namespace *ns)
1397 ns->mq_queues_count = 0;
1398 ns->mq_queues_max = DFLT_QUEUESMAX;
1399 ns->mq_msg_max = DFLT_MSGMAX;
1400 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1401 ns->mq_msg_default = DFLT_MSG;
1402 ns->mq_msgsize_default = DFLT_MSGSIZE;
1404 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1405 if (IS_ERR(ns->mq_mnt)) {
1406 int err = PTR_ERR(ns->mq_mnt);
1413 void mq_clear_sbinfo(struct ipc_namespace *ns)
1415 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1418 void mq_put_mnt(struct ipc_namespace *ns)
1420 kern_unmount(ns->mq_mnt);
1423 static int __init init_mqueue_fs(void)
1427 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1428 sizeof(struct mqueue_inode_info), 0,
1429 SLAB_HWCACHE_ALIGN, init_once);
1430 if (mqueue_inode_cachep == NULL)
1433 /* ignore failures - they are not fatal */
1434 mq_sysctl_table = mq_register_sysctl_table();
1436 error = register_filesystem(&mqueue_fs_type);
1440 spin_lock_init(&mq_lock);
1442 error = mq_init_ns(&init_ipc_ns);
1444 goto out_filesystem;
1449 unregister_filesystem(&mqueue_fs_type);
1451 if (mq_sysctl_table)
1452 unregister_sysctl_table(mq_sysctl_table);
1453 kmem_cache_destroy(mqueue_inode_cachep);
1457 __initcall(init_mqueue_fs);