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);
146 info->qsize += sizeof(*leaf);
148 leaf->priority = msg->m_type;
149 rb_link_node(&leaf->rb_node, parent, p);
150 rb_insert_color(&leaf->rb_node, &info->msg_tree);
152 info->attr.mq_curmsgs++;
153 info->qsize += msg->m_ts;
154 list_add_tail(&msg->m_list, &leaf->msg_list);
158 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160 struct rb_node **p, *parent = NULL;
161 struct posix_msg_tree_node *leaf;
165 p = &info->msg_tree.rb_node;
169 * During insert, low priorities go to the left and high to the
170 * right. On receive, we want the highest priorities first, so
171 * walk all the way to the right.
176 if (info->attr.mq_curmsgs) {
177 pr_warn_once("Inconsistency in POSIX message queue, "
178 "no tree element, but supposedly messages "
180 info->attr.mq_curmsgs = 0;
184 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
185 if (unlikely(list_empty(&leaf->msg_list))) {
186 pr_warn_once("Inconsistency in POSIX message queue, "
187 "empty leaf node but we haven't implemented "
188 "lazy leaf delete!\n");
189 rb_erase(&leaf->rb_node, &info->msg_tree);
190 if (info->node_cache) {
191 info->qsize -= sizeof(*leaf);
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) {
204 info->qsize -= sizeof(*leaf);
207 info->node_cache = leaf;
211 info->attr.mq_curmsgs--;
212 info->qsize -= msg->m_ts;
216 static struct inode *mqueue_get_inode(struct super_block *sb,
217 struct ipc_namespace *ipc_ns, umode_t mode,
218 struct mq_attr *attr)
220 struct user_struct *u = current_user();
224 inode = new_inode(sb);
228 inode->i_ino = get_next_ino();
229 inode->i_mode = mode;
230 inode->i_uid = current_fsuid();
231 inode->i_gid = current_fsgid();
232 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
235 struct mqueue_inode_info *info;
236 unsigned long mq_bytes, mq_treesize;
238 inode->i_fop = &mqueue_file_operations;
239 inode->i_size = FILENT_SIZE;
240 /* mqueue specific info */
241 info = MQUEUE_I(inode);
242 spin_lock_init(&info->lock);
243 init_waitqueue_head(&info->wait_q);
244 INIT_LIST_HEAD(&info->e_wait_q[0].list);
245 INIT_LIST_HEAD(&info->e_wait_q[1].list);
246 info->notify_owner = NULL;
247 info->notify_user_ns = NULL;
249 info->user = NULL; /* set when all is ok */
250 info->msg_tree = RB_ROOT;
251 info->node_cache = NULL;
252 memset(&info->attr, 0, sizeof(info->attr));
253 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
254 ipc_ns->mq_msg_default);
255 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
256 ipc_ns->mq_msgsize_default);
258 info->attr.mq_maxmsg = attr->mq_maxmsg;
259 info->attr.mq_msgsize = attr->mq_msgsize;
262 * We used to allocate a static array of pointers and account
263 * the size of that array as well as one msg_msg struct per
264 * possible message into the queue size. That's no longer
265 * accurate as the queue is now an rbtree and will grow and
266 * shrink depending on usage patterns. We can, however, still
267 * account one msg_msg struct per message, but the nodes are
268 * allocated depending on priority usage, and most programs
269 * only use one, or a handful, of priorities. However, since
270 * this is pinned memory, we need to assume worst case, so
271 * that means the min(mq_maxmsg, max_priorities) * struct
272 * posix_msg_tree_node.
274 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
275 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
276 sizeof(struct posix_msg_tree_node);
278 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
279 info->attr.mq_msgsize);
282 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
283 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
284 spin_unlock(&mq_lock);
285 /* mqueue_evict_inode() releases info->messages */
289 u->mq_bytes += mq_bytes;
290 spin_unlock(&mq_lock);
293 info->user = get_uid(u);
294 } else if (S_ISDIR(mode)) {
296 /* Some things misbehave if size == 0 on a directory */
297 inode->i_size = 2 * DIRENT_SIZE;
298 inode->i_op = &mqueue_dir_inode_operations;
299 inode->i_fop = &simple_dir_operations;
309 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
312 struct ipc_namespace *ns = data;
314 sb->s_blocksize = PAGE_CACHE_SIZE;
315 sb->s_blocksize_bits = PAGE_CACHE_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 if (!(flags & MS_KERNMOUNT))
334 data = current->nsproxy->ipc_ns;
335 return mount_ns(fs_type, flags, data, mqueue_fill_super);
338 static void init_once(void *foo)
340 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
342 inode_init_once(&p->vfs_inode);
345 static struct inode *mqueue_alloc_inode(struct super_block *sb)
347 struct mqueue_inode_info *ei;
349 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
352 return &ei->vfs_inode;
355 static void mqueue_i_callback(struct rcu_head *head)
357 struct inode *inode = container_of(head, struct inode, i_rcu);
358 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
361 static void mqueue_destroy_inode(struct inode *inode)
363 call_rcu(&inode->i_rcu, mqueue_i_callback);
366 static void mqueue_evict_inode(struct inode *inode)
368 struct mqueue_inode_info *info;
369 struct user_struct *user;
370 unsigned long mq_bytes, mq_treesize;
371 struct ipc_namespace *ipc_ns;
376 if (S_ISDIR(inode->i_mode))
379 ipc_ns = get_ns_from_inode(inode);
380 info = MQUEUE_I(inode);
381 spin_lock(&info->lock);
382 while ((msg = msg_get(info)) != NULL)
384 kfree(info->node_cache);
385 spin_unlock(&info->lock);
387 /* Total amount of bytes accounted for the mqueue */
388 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
389 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
390 sizeof(struct posix_msg_tree_node);
392 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
393 info->attr.mq_msgsize);
398 user->mq_bytes -= mq_bytes;
400 * get_ns_from_inode() ensures that the
401 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
402 * to which we now hold a reference, or it is NULL.
403 * We can't put it here under mq_lock, though.
406 ipc_ns->mq_queues_count--;
407 spin_unlock(&mq_lock);
414 static int mqueue_create(struct inode *dir, struct dentry *dentry,
415 umode_t mode, bool excl)
418 struct mq_attr *attr = dentry->d_fsdata;
420 struct ipc_namespace *ipc_ns;
423 ipc_ns = __get_ns_from_inode(dir);
428 if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
429 (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
430 !capable(CAP_SYS_RESOURCE))) {
434 ipc_ns->mq_queues_count++;
435 spin_unlock(&mq_lock);
437 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
439 error = PTR_ERR(inode);
441 ipc_ns->mq_queues_count--;
446 dir->i_size += DIRENT_SIZE;
447 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
449 d_instantiate(dentry, inode);
453 spin_unlock(&mq_lock);
459 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
461 struct inode *inode = dentry->d_inode;
463 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
464 dir->i_size -= DIRENT_SIZE;
471 * This is routine for system read from queue file.
472 * To avoid mess with doing here some sort of mq_receive we allow
473 * to read only queue size & notification info (the only values
474 * that are interesting from user point of view and aren't accessible
475 * through std routines)
477 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
478 size_t count, loff_t *off)
480 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
481 char buffer[FILENT_SIZE];
484 spin_lock(&info->lock);
485 snprintf(buffer, sizeof(buffer),
486 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
488 info->notify_owner ? info->notify.sigev_notify : 0,
489 (info->notify_owner &&
490 info->notify.sigev_notify == SIGEV_SIGNAL) ?
491 info->notify.sigev_signo : 0,
492 pid_vnr(info->notify_owner));
493 spin_unlock(&info->lock);
494 buffer[sizeof(buffer)-1] = '\0';
496 ret = simple_read_from_buffer(u_data, count, off, buffer,
501 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
505 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
507 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
509 spin_lock(&info->lock);
510 if (task_tgid(current) == info->notify_owner)
511 remove_notification(info);
513 spin_unlock(&info->lock);
517 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
519 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
522 poll_wait(filp, &info->wait_q, poll_tab);
524 spin_lock(&info->lock);
525 if (info->attr.mq_curmsgs)
526 retval = POLLIN | POLLRDNORM;
528 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
529 retval |= POLLOUT | POLLWRNORM;
530 spin_unlock(&info->lock);
535 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
536 static void wq_add(struct mqueue_inode_info *info, int sr,
537 struct ext_wait_queue *ewp)
539 struct ext_wait_queue *walk;
543 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
544 if (walk->task->static_prio <= current->static_prio) {
545 list_add_tail(&ewp->list, &walk->list);
549 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
553 * Puts current task to sleep. Caller must hold queue lock. After return
557 static int wq_sleep(struct mqueue_inode_info *info, int sr,
558 ktime_t *timeout, struct ext_wait_queue *ewp)
563 wq_add(info, sr, ewp);
566 set_current_state(TASK_INTERRUPTIBLE);
568 spin_unlock(&info->lock);
569 time = schedule_hrtimeout_range_clock(timeout, 0,
570 HRTIMER_MODE_ABS, CLOCK_REALTIME);
572 while (ewp->state == STATE_PENDING)
575 if (ewp->state == STATE_READY) {
579 spin_lock(&info->lock);
580 if (ewp->state == STATE_READY) {
584 if (signal_pending(current)) {
585 retval = -ERESTARTSYS;
593 list_del(&ewp->list);
595 spin_unlock(&info->lock);
601 * Returns waiting task that should be serviced first or NULL if none exists
603 static struct ext_wait_queue *wq_get_first_waiter(
604 struct mqueue_inode_info *info, int sr)
606 struct list_head *ptr;
608 ptr = info->e_wait_q[sr].list.prev;
609 if (ptr == &info->e_wait_q[sr].list)
611 return list_entry(ptr, struct ext_wait_queue, list);
615 static inline void set_cookie(struct sk_buff *skb, char code)
617 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
621 * The next function is only to split too long sys_mq_timedsend
623 static void __do_notify(struct mqueue_inode_info *info)
626 * invoked when there is registered process and there isn't process
627 * waiting synchronously for message AND state of queue changed from
628 * empty to not empty. Here we are sure that no one is waiting
630 if (info->notify_owner &&
631 info->attr.mq_curmsgs == 1) {
632 struct siginfo sig_i;
633 switch (info->notify.sigev_notify) {
639 sig_i.si_signo = info->notify.sigev_signo;
641 sig_i.si_code = SI_MESGQ;
642 sig_i.si_value = info->notify.sigev_value;
643 /* map current pid/uid into info->owner's namespaces */
645 sig_i.si_pid = task_tgid_nr_ns(current,
646 ns_of_pid(info->notify_owner));
647 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
650 kill_pid_info(info->notify.sigev_signo,
651 &sig_i, info->notify_owner);
654 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
655 netlink_sendskb(info->notify_sock, info->notify_cookie);
658 /* after notification unregisters process */
659 put_pid(info->notify_owner);
660 put_user_ns(info->notify_user_ns);
661 info->notify_owner = NULL;
662 info->notify_user_ns = NULL;
664 wake_up(&info->wait_q);
667 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
668 ktime_t *expires, struct timespec *ts)
670 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
672 if (!timespec_valid(ts))
675 *expires = timespec_to_ktime(*ts);
679 static void remove_notification(struct mqueue_inode_info *info)
681 if (info->notify_owner != NULL &&
682 info->notify.sigev_notify == SIGEV_THREAD) {
683 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
684 netlink_sendskb(info->notify_sock, info->notify_cookie);
686 put_pid(info->notify_owner);
687 put_user_ns(info->notify_user_ns);
688 info->notify_owner = NULL;
689 info->notify_user_ns = NULL;
692 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
695 unsigned long total_size;
697 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
699 if (capable(CAP_SYS_RESOURCE)) {
700 if (attr->mq_maxmsg > HARD_MSGMAX ||
701 attr->mq_msgsize > HARD_MSGSIZEMAX)
704 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
705 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
708 /* check for overflow */
709 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
711 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
712 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
713 sizeof(struct posix_msg_tree_node);
714 total_size = attr->mq_maxmsg * attr->mq_msgsize;
715 if (total_size + mq_treesize < total_size)
721 * Invoked when creating a new queue via sys_mq_open
723 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
724 struct path *path, int oflag, umode_t mode,
725 struct mq_attr *attr)
727 const struct cred *cred = current_cred();
731 ret = mq_attr_ok(ipc_ns, attr);
734 /* store for use during create */
735 path->dentry->d_fsdata = attr;
737 struct mq_attr def_attr;
739 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
740 ipc_ns->mq_msg_default);
741 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
742 ipc_ns->mq_msgsize_default);
743 ret = mq_attr_ok(ipc_ns, &def_attr);
748 mode &= ~current_umask();
749 ret = vfs_create(dir, path->dentry, mode, true);
750 path->dentry->d_fsdata = NULL;
753 return dentry_open(path, oflag, cred);
756 /* Opens existing queue */
757 static struct file *do_open(struct path *path, int oflag)
759 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
760 MAY_READ | MAY_WRITE };
762 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
763 return ERR_PTR(-EINVAL);
764 acc = oflag2acc[oflag & O_ACCMODE];
765 if (inode_permission(path->dentry->d_inode, acc))
766 return ERR_PTR(-EACCES);
767 return dentry_open(path, oflag, current_cred());
770 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
771 struct mq_attr __user *, u_attr)
775 struct filename *name;
778 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
779 struct vfsmount *mnt = ipc_ns->mq_mnt;
780 struct dentry *root = mnt->mnt_root;
783 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
786 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
788 if (IS_ERR(name = getname(u_name)))
789 return PTR_ERR(name);
791 fd = get_unused_fd_flags(O_CLOEXEC);
795 ro = mnt_want_write(mnt); /* we'll drop it in any case */
797 mutex_lock(&root->d_inode->i_mutex);
798 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
799 if (IS_ERR(path.dentry)) {
800 error = PTR_ERR(path.dentry);
803 path.mnt = mntget(mnt);
805 if (oflag & O_CREAT) {
806 if (path.dentry->d_inode) { /* entry already exists */
807 audit_inode(name, path.dentry, 0);
808 if (oflag & O_EXCL) {
812 filp = do_open(&path, oflag);
818 filp = do_create(ipc_ns, root->d_inode,
820 u_attr ? &attr : NULL);
823 if (!path.dentry->d_inode) {
827 audit_inode(name, path.dentry, 0);
828 filp = do_open(&path, oflag);
832 fd_install(fd, filp);
834 error = PTR_ERR(filp);
842 mutex_unlock(&root->d_inode->i_mutex);
849 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
852 struct filename *name;
853 struct dentry *dentry;
854 struct inode *inode = NULL;
855 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
856 struct vfsmount *mnt = ipc_ns->mq_mnt;
858 name = getname(u_name);
860 return PTR_ERR(name);
862 err = mnt_want_write(mnt);
865 mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
866 dentry = lookup_one_len(name->name, mnt->mnt_root,
868 if (IS_ERR(dentry)) {
869 err = PTR_ERR(dentry);
873 inode = dentry->d_inode;
878 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
883 mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
893 /* Pipelined send and receive functions.
895 * If a receiver finds no waiting message, then it registers itself in the
896 * list of waiting receivers. A sender checks that list before adding the new
897 * message into the message array. If there is a waiting receiver, then it
898 * bypasses the message array and directly hands the message over to the
900 * The receiver accepts the message and returns without grabbing the queue
901 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
902 * are necessary. The same algorithm is used for sysv semaphores, see
903 * ipc/sem.c for more details.
905 * The same algorithm is used for senders.
908 /* pipelined_send() - send a message directly to the task waiting in
909 * sys_mq_timedreceive() (without inserting message into a queue).
911 static inline void pipelined_send(struct mqueue_inode_info *info,
912 struct msg_msg *message,
913 struct ext_wait_queue *receiver)
915 receiver->msg = message;
916 list_del(&receiver->list);
917 receiver->state = STATE_PENDING;
918 wake_up_process(receiver->task);
920 receiver->state = STATE_READY;
923 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
924 * gets its message and put to the queue (we have one free place for sure). */
925 static inline void pipelined_receive(struct mqueue_inode_info *info)
927 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
931 wake_up_interruptible(&info->wait_q);
934 if (msg_insert(sender->msg, info))
936 list_del(&sender->list);
937 sender->state = STATE_PENDING;
938 wake_up_process(sender->task);
940 sender->state = STATE_READY;
943 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
944 size_t, msg_len, unsigned int, msg_prio,
945 const struct timespec __user *, u_abs_timeout)
949 struct ext_wait_queue wait;
950 struct ext_wait_queue *receiver;
951 struct msg_msg *msg_ptr;
952 struct mqueue_inode_info *info;
953 ktime_t expires, *timeout = NULL;
955 struct posix_msg_tree_node *new_leaf = NULL;
959 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
965 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
968 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
971 if (unlikely(!f.file)) {
976 inode = file_inode(f.file);
977 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
981 info = MQUEUE_I(inode);
982 audit_inode(NULL, f.file->f_path.dentry, 0);
984 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
989 if (unlikely(msg_len > info->attr.mq_msgsize)) {
994 /* First try to allocate memory, before doing anything with
995 * existing queues. */
996 msg_ptr = load_msg(u_msg_ptr, msg_len);
997 if (IS_ERR(msg_ptr)) {
998 ret = PTR_ERR(msg_ptr);
1001 msg_ptr->m_ts = msg_len;
1002 msg_ptr->m_type = msg_prio;
1005 * msg_insert really wants us to have a valid, spare node struct so
1006 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1007 * fall back to that if necessary.
1009 if (!info->node_cache)
1010 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1012 spin_lock(&info->lock);
1014 if (!info->node_cache && new_leaf) {
1015 /* Save our speculative allocation into the cache */
1016 INIT_LIST_HEAD(&new_leaf->msg_list);
1017 info->node_cache = new_leaf;
1018 info->qsize += sizeof(*new_leaf);
1024 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1025 if (f.file->f_flags & O_NONBLOCK) {
1028 wait.task = current;
1029 wait.msg = (void *) msg_ptr;
1030 wait.state = STATE_NONE;
1031 ret = wq_sleep(info, SEND, timeout, &wait);
1033 * wq_sleep must be called with info->lock held, and
1034 * returns with the lock released
1039 receiver = wq_get_first_waiter(info, RECV);
1041 pipelined_send(info, msg_ptr, receiver);
1043 /* adds message to the queue */
1044 ret = msg_insert(msg_ptr, info);
1049 inode->i_atime = inode->i_mtime = inode->i_ctime =
1053 spin_unlock(&info->lock);
1063 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1064 size_t, msg_len, unsigned int __user *, u_msg_prio,
1065 const struct timespec __user *, u_abs_timeout)
1068 struct msg_msg *msg_ptr;
1070 struct inode *inode;
1071 struct mqueue_inode_info *info;
1072 struct ext_wait_queue wait;
1073 ktime_t expires, *timeout = NULL;
1075 struct posix_msg_tree_node *new_leaf = NULL;
1077 if (u_abs_timeout) {
1078 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1084 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1087 if (unlikely(!f.file)) {
1092 inode = file_inode(f.file);
1093 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1097 info = MQUEUE_I(inode);
1098 audit_inode(NULL, f.file->f_path.dentry, 0);
1100 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1105 /* checks if buffer is big enough */
1106 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1112 * msg_insert really wants us to have a valid, spare node struct so
1113 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1114 * fall back to that if necessary.
1116 if (!info->node_cache)
1117 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1119 spin_lock(&info->lock);
1121 if (!info->node_cache && new_leaf) {
1122 /* Save our speculative allocation into the cache */
1123 INIT_LIST_HEAD(&new_leaf->msg_list);
1124 info->node_cache = new_leaf;
1125 info->qsize += sizeof(*new_leaf);
1130 if (info->attr.mq_curmsgs == 0) {
1131 if (f.file->f_flags & O_NONBLOCK) {
1132 spin_unlock(&info->lock);
1135 wait.task = current;
1136 wait.state = STATE_NONE;
1137 ret = wq_sleep(info, RECV, timeout, &wait);
1141 msg_ptr = msg_get(info);
1143 inode->i_atime = inode->i_mtime = inode->i_ctime =
1146 /* There is now free space in queue. */
1147 pipelined_receive(info);
1148 spin_unlock(&info->lock);
1152 ret = msg_ptr->m_ts;
1154 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1155 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1167 * Notes: the case when user wants us to deregister (with NULL as pointer)
1168 * and he isn't currently owner of notification, will be silently discarded.
1169 * It isn't explicitly defined in the POSIX.
1171 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1172 const struct sigevent __user *, u_notification)
1177 struct inode *inode;
1178 struct sigevent notification;
1179 struct mqueue_inode_info *info;
1182 if (u_notification) {
1183 if (copy_from_user(¬ification, u_notification,
1184 sizeof(struct sigevent)))
1188 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1192 if (u_notification != NULL) {
1193 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1194 notification.sigev_notify != SIGEV_SIGNAL &&
1195 notification.sigev_notify != SIGEV_THREAD))
1197 if (notification.sigev_notify == SIGEV_SIGNAL &&
1198 !valid_signal(notification.sigev_signo)) {
1201 if (notification.sigev_notify == SIGEV_THREAD) {
1204 /* create the notify skb */
1205 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1210 if (copy_from_user(nc->data,
1211 notification.sigev_value.sival_ptr,
1212 NOTIFY_COOKIE_LEN)) {
1217 /* TODO: add a header? */
1218 skb_put(nc, NOTIFY_COOKIE_LEN);
1219 /* and attach it to the socket */
1221 f = fdget(notification.sigev_signo);
1226 sock = netlink_getsockbyfilp(f.file);
1229 ret = PTR_ERR(sock);
1234 timeo = MAX_SCHEDULE_TIMEOUT;
1235 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1252 inode = file_inode(f.file);
1253 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1257 info = MQUEUE_I(inode);
1260 spin_lock(&info->lock);
1261 if (u_notification == NULL) {
1262 if (info->notify_owner == task_tgid(current)) {
1263 remove_notification(info);
1264 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1266 } else if (info->notify_owner != NULL) {
1269 switch (notification.sigev_notify) {
1271 info->notify.sigev_notify = SIGEV_NONE;
1274 info->notify_sock = sock;
1275 info->notify_cookie = nc;
1278 info->notify.sigev_notify = SIGEV_THREAD;
1281 info->notify.sigev_signo = notification.sigev_signo;
1282 info->notify.sigev_value = notification.sigev_value;
1283 info->notify.sigev_notify = SIGEV_SIGNAL;
1287 info->notify_owner = get_pid(task_tgid(current));
1288 info->notify_user_ns = get_user_ns(current_user_ns());
1289 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1291 spin_unlock(&info->lock);
1296 netlink_detachskb(sock, nc);
1303 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1304 const struct mq_attr __user *, u_mqstat,
1305 struct mq_attr __user *, u_omqstat)
1308 struct mq_attr mqstat, omqstat;
1310 struct inode *inode;
1311 struct mqueue_inode_info *info;
1313 if (u_mqstat != NULL) {
1314 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1316 if (mqstat.mq_flags & (~O_NONBLOCK))
1326 inode = file_inode(f.file);
1327 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1331 info = MQUEUE_I(inode);
1333 spin_lock(&info->lock);
1335 omqstat = info->attr;
1336 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1338 audit_mq_getsetattr(mqdes, &mqstat);
1339 spin_lock(&f.file->f_lock);
1340 if (mqstat.mq_flags & O_NONBLOCK)
1341 f.file->f_flags |= O_NONBLOCK;
1343 f.file->f_flags &= ~O_NONBLOCK;
1344 spin_unlock(&f.file->f_lock);
1346 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1349 spin_unlock(&info->lock);
1352 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1353 sizeof(struct mq_attr)))
1362 static const struct inode_operations mqueue_dir_inode_operations = {
1363 .lookup = simple_lookup,
1364 .create = mqueue_create,
1365 .unlink = mqueue_unlink,
1368 static const struct file_operations mqueue_file_operations = {
1369 .flush = mqueue_flush_file,
1370 .poll = mqueue_poll_file,
1371 .read = mqueue_read_file,
1372 .llseek = default_llseek,
1375 static const struct super_operations mqueue_super_ops = {
1376 .alloc_inode = mqueue_alloc_inode,
1377 .destroy_inode = mqueue_destroy_inode,
1378 .evict_inode = mqueue_evict_inode,
1379 .statfs = simple_statfs,
1382 static struct file_system_type mqueue_fs_type = {
1384 .mount = mqueue_mount,
1385 .kill_sb = kill_litter_super,
1386 .fs_flags = FS_USERNS_MOUNT,
1389 int mq_init_ns(struct ipc_namespace *ns)
1391 ns->mq_queues_count = 0;
1392 ns->mq_queues_max = DFLT_QUEUESMAX;
1393 ns->mq_msg_max = DFLT_MSGMAX;
1394 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1395 ns->mq_msg_default = DFLT_MSG;
1396 ns->mq_msgsize_default = DFLT_MSGSIZE;
1398 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1399 if (IS_ERR(ns->mq_mnt)) {
1400 int err = PTR_ERR(ns->mq_mnt);
1407 void mq_clear_sbinfo(struct ipc_namespace *ns)
1409 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1412 void mq_put_mnt(struct ipc_namespace *ns)
1414 kern_unmount(ns->mq_mnt);
1417 static int __init init_mqueue_fs(void)
1421 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1422 sizeof(struct mqueue_inode_info), 0,
1423 SLAB_HWCACHE_ALIGN, init_once);
1424 if (mqueue_inode_cachep == NULL)
1427 /* ignore failures - they are not fatal */
1428 mq_sysctl_table = mq_register_sysctl_table();
1430 error = register_filesystem(&mqueue_fs_type);
1434 spin_lock_init(&mq_lock);
1436 error = mq_init_ns(&init_ipc_ns);
1438 goto out_filesystem;
1443 unregister_filesystem(&mqueue_fs_type);
1445 if (mq_sysctl_table)
1446 unregister_sysctl_table(mq_sysctl_table);
1447 kmem_cache_destroy(mqueue_inode_cachep);
1451 __initcall(init_mqueue_fs);