2 * POSIX message queues filesystem for Linux.
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
5 * Michal Wronski (michal.wronski@gmail.com)
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
13 * This file is released under the GPL.
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/vmalloc.h>
28 #include <linux/netlink.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/signal.h>
32 #include <linux/mutex.h>
33 #include <linux/nsproxy.h>
34 #include <linux/pid.h>
35 #include <linux/ipc_namespace.h>
36 #include <linux/user_namespace.h>
37 #include <linux/slab.h>
38 #include <linux/sched/wake_q.h>
39 #include <linux/sched/signal.h>
40 #include <linux/sched/user.h>
45 #define MQUEUE_MAGIC 0x19800202
46 #define DIRENT_SIZE 20
47 #define FILENT_SIZE 80
55 struct posix_msg_tree_node {
56 struct rb_node rb_node;
57 struct list_head msg_list;
61 struct ext_wait_queue { /* queue of sleeping tasks */
62 struct task_struct *task;
63 struct list_head list;
64 struct msg_msg *msg; /* ptr of loaded message */
65 int state; /* one of STATE_* values */
68 struct mqueue_inode_info {
70 struct inode vfs_inode;
71 wait_queue_head_t wait_q;
73 struct rb_root msg_tree;
74 struct posix_msg_tree_node *node_cache;
77 struct sigevent notify;
78 struct pid *notify_owner;
79 struct user_namespace *notify_user_ns;
80 struct user_struct *user; /* user who created, for accounting */
81 struct sock *notify_sock;
82 struct sk_buff *notify_cookie;
84 /* for tasks waiting for free space and messages, respectively */
85 struct ext_wait_queue e_wait_q[2];
87 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
90 static const struct inode_operations mqueue_dir_inode_operations;
91 static const struct file_operations mqueue_file_operations;
92 static const struct super_operations mqueue_super_ops;
93 static void remove_notification(struct mqueue_inode_info *info);
95 static struct kmem_cache *mqueue_inode_cachep;
97 static struct ctl_table_header *mq_sysctl_table;
99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
105 * This routine should be called with the mq_lock held.
107 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
109 return get_ipc_ns(inode->i_sb->s_fs_info);
112 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
114 struct ipc_namespace *ns;
117 ns = __get_ns_from_inode(inode);
118 spin_unlock(&mq_lock);
122 /* Auxiliary functions to manipulate messages' list */
123 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
125 struct rb_node **p, *parent = NULL;
126 struct posix_msg_tree_node *leaf;
128 p = &info->msg_tree.rb_node;
131 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
133 if (likely(leaf->priority == msg->m_type))
135 else if (msg->m_type < leaf->priority)
140 if (info->node_cache) {
141 leaf = info->node_cache;
142 info->node_cache = NULL;
144 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
147 INIT_LIST_HEAD(&leaf->msg_list);
149 leaf->priority = msg->m_type;
150 rb_link_node(&leaf->rb_node, parent, p);
151 rb_insert_color(&leaf->rb_node, &info->msg_tree);
153 info->attr.mq_curmsgs++;
154 info->qsize += msg->m_ts;
155 list_add_tail(&msg->m_list, &leaf->msg_list);
159 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
161 struct rb_node **p, *parent = NULL;
162 struct posix_msg_tree_node *leaf;
166 p = &info->msg_tree.rb_node;
170 * During insert, low priorities go to the left and high to the
171 * right. On receive, we want the highest priorities first, so
172 * walk all the way to the right.
177 if (info->attr.mq_curmsgs) {
178 pr_warn_once("Inconsistency in POSIX message queue, "
179 "no tree element, but supposedly messages "
181 info->attr.mq_curmsgs = 0;
185 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186 if (unlikely(list_empty(&leaf->msg_list))) {
187 pr_warn_once("Inconsistency in POSIX message queue, "
188 "empty leaf node but we haven't implemented "
189 "lazy leaf delete!\n");
190 rb_erase(&leaf->rb_node, &info->msg_tree);
191 if (info->node_cache) {
194 info->node_cache = leaf;
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
203 if (info->node_cache) {
206 info->node_cache = leaf;
210 info->attr.mq_curmsgs--;
211 info->qsize -= msg->m_ts;
215 static struct inode *mqueue_get_inode(struct super_block *sb,
216 struct ipc_namespace *ipc_ns, umode_t mode,
217 struct mq_attr *attr)
219 struct user_struct *u = current_user();
223 inode = new_inode(sb);
227 inode->i_ino = get_next_ino();
228 inode->i_mode = mode;
229 inode->i_uid = current_fsuid();
230 inode->i_gid = current_fsgid();
231 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
234 struct mqueue_inode_info *info;
235 unsigned long mq_bytes, mq_treesize;
237 inode->i_fop = &mqueue_file_operations;
238 inode->i_size = FILENT_SIZE;
239 /* mqueue specific info */
240 info = MQUEUE_I(inode);
241 spin_lock_init(&info->lock);
242 init_waitqueue_head(&info->wait_q);
243 INIT_LIST_HEAD(&info->e_wait_q[0].list);
244 INIT_LIST_HEAD(&info->e_wait_q[1].list);
245 info->notify_owner = NULL;
246 info->notify_user_ns = NULL;
248 info->user = NULL; /* set when all is ok */
249 info->msg_tree = RB_ROOT;
250 info->node_cache = NULL;
251 memset(&info->attr, 0, sizeof(info->attr));
252 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253 ipc_ns->mq_msg_default);
254 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255 ipc_ns->mq_msgsize_default);
257 info->attr.mq_maxmsg = attr->mq_maxmsg;
258 info->attr.mq_msgsize = attr->mq_msgsize;
261 * We used to allocate a static array of pointers and account
262 * the size of that array as well as one msg_msg struct per
263 * possible message into the queue size. That's no longer
264 * accurate as the queue is now an rbtree and will grow and
265 * shrink depending on usage patterns. We can, however, still
266 * account one msg_msg struct per message, but the nodes are
267 * allocated depending on priority usage, and most programs
268 * only use one, or a handful, of priorities. However, since
269 * this is pinned memory, we need to assume worst case, so
270 * that means the min(mq_maxmsg, max_priorities) * struct
271 * posix_msg_tree_node.
273 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
274 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
275 sizeof(struct posix_msg_tree_node);
277 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
278 info->attr.mq_msgsize);
281 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
282 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
283 spin_unlock(&mq_lock);
284 /* mqueue_evict_inode() releases info->messages */
288 u->mq_bytes += mq_bytes;
289 spin_unlock(&mq_lock);
292 info->user = get_uid(u);
293 } else if (S_ISDIR(mode)) {
295 /* Some things misbehave if size == 0 on a directory */
296 inode->i_size = 2 * DIRENT_SIZE;
297 inode->i_op = &mqueue_dir_inode_operations;
298 inode->i_fop = &simple_dir_operations;
308 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
311 struct ipc_namespace *ns = sb->s_fs_info;
313 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
314 sb->s_blocksize = PAGE_SIZE;
315 sb->s_blocksize_bits = PAGE_SHIFT;
316 sb->s_magic = MQUEUE_MAGIC;
317 sb->s_op = &mqueue_super_ops;
319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
321 return PTR_ERR(inode);
323 sb->s_root = d_make_root(inode);
329 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330 int flags, const char *dev_name,
333 struct ipc_namespace *ns;
334 if (flags & MS_KERNMOUNT) {
338 ns = current->nsproxy->ipc_ns;
340 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
343 static void init_once(void *foo)
345 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
347 inode_init_once(&p->vfs_inode);
350 static struct inode *mqueue_alloc_inode(struct super_block *sb)
352 struct mqueue_inode_info *ei;
354 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
357 return &ei->vfs_inode;
360 static void mqueue_i_callback(struct rcu_head *head)
362 struct inode *inode = container_of(head, struct inode, i_rcu);
363 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
366 static void mqueue_destroy_inode(struct inode *inode)
368 call_rcu(&inode->i_rcu, mqueue_i_callback);
371 static void mqueue_evict_inode(struct inode *inode)
373 struct mqueue_inode_info *info;
374 struct user_struct *user;
375 struct ipc_namespace *ipc_ns;
376 struct msg_msg *msg, *nmsg;
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)
388 list_add_tail(&msg->m_list, &tmp_msg);
389 kfree(info->node_cache);
390 spin_unlock(&info->lock);
392 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
393 list_del(&msg->m_list);
399 unsigned long mq_bytes, mq_treesize;
401 /* Total amount of bytes accounted for the mqueue */
402 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
403 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
404 sizeof(struct posix_msg_tree_node);
406 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
407 info->attr.mq_msgsize);
410 user->mq_bytes -= mq_bytes;
412 * get_ns_from_inode() ensures that the
413 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
414 * to which we now hold a reference, or it is NULL.
415 * We can't put it here under mq_lock, though.
418 ipc_ns->mq_queues_count--;
419 spin_unlock(&mq_lock);
426 static int mqueue_create(struct inode *dir, struct dentry *dentry,
427 umode_t mode, bool excl)
430 struct mq_attr *attr = dentry->d_fsdata;
432 struct ipc_namespace *ipc_ns;
435 ipc_ns = __get_ns_from_inode(dir);
441 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
442 !capable(CAP_SYS_RESOURCE)) {
446 ipc_ns->mq_queues_count++;
447 spin_unlock(&mq_lock);
449 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
451 error = PTR_ERR(inode);
453 ipc_ns->mq_queues_count--;
458 dir->i_size += DIRENT_SIZE;
459 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
461 d_instantiate(dentry, inode);
465 spin_unlock(&mq_lock);
471 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
473 struct inode *inode = d_inode(dentry);
475 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
476 dir->i_size -= DIRENT_SIZE;
483 * This is routine for system read from queue file.
484 * To avoid mess with doing here some sort of mq_receive we allow
485 * to read only queue size & notification info (the only values
486 * that are interesting from user point of view and aren't accessible
487 * through std routines)
489 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
490 size_t count, loff_t *off)
492 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
493 char buffer[FILENT_SIZE];
496 spin_lock(&info->lock);
497 snprintf(buffer, sizeof(buffer),
498 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
500 info->notify_owner ? info->notify.sigev_notify : 0,
501 (info->notify_owner &&
502 info->notify.sigev_notify == SIGEV_SIGNAL) ?
503 info->notify.sigev_signo : 0,
504 pid_vnr(info->notify_owner));
505 spin_unlock(&info->lock);
506 buffer[sizeof(buffer)-1] = '\0';
508 ret = simple_read_from_buffer(u_data, count, off, buffer,
513 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
517 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
519 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
521 spin_lock(&info->lock);
522 if (task_tgid(current) == info->notify_owner)
523 remove_notification(info);
525 spin_unlock(&info->lock);
529 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
531 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
534 poll_wait(filp, &info->wait_q, poll_tab);
536 spin_lock(&info->lock);
537 if (info->attr.mq_curmsgs)
538 retval = POLLIN | POLLRDNORM;
540 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
541 retval |= POLLOUT | POLLWRNORM;
542 spin_unlock(&info->lock);
547 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
548 static void wq_add(struct mqueue_inode_info *info, int sr,
549 struct ext_wait_queue *ewp)
551 struct ext_wait_queue *walk;
555 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
556 if (walk->task->static_prio <= current->static_prio) {
557 list_add_tail(&ewp->list, &walk->list);
561 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
565 * Puts current task to sleep. Caller must hold queue lock. After return
569 static int wq_sleep(struct mqueue_inode_info *info, int sr,
570 ktime_t *timeout, struct ext_wait_queue *ewp)
571 __releases(&info->lock)
576 wq_add(info, sr, ewp);
579 __set_current_state(TASK_INTERRUPTIBLE);
581 spin_unlock(&info->lock);
582 time = schedule_hrtimeout_range_clock(timeout, 0,
583 HRTIMER_MODE_ABS, CLOCK_REALTIME);
585 if (ewp->state == STATE_READY) {
589 spin_lock(&info->lock);
590 if (ewp->state == STATE_READY) {
594 if (signal_pending(current)) {
595 retval = -ERESTARTSYS;
603 list_del(&ewp->list);
605 spin_unlock(&info->lock);
611 * Returns waiting task that should be serviced first or NULL if none exists
613 static struct ext_wait_queue *wq_get_first_waiter(
614 struct mqueue_inode_info *info, int sr)
616 struct list_head *ptr;
618 ptr = info->e_wait_q[sr].list.prev;
619 if (ptr == &info->e_wait_q[sr].list)
621 return list_entry(ptr, struct ext_wait_queue, list);
625 static inline void set_cookie(struct sk_buff *skb, char code)
627 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
631 * The next function is only to split too long sys_mq_timedsend
633 static void __do_notify(struct mqueue_inode_info *info)
636 * invoked when there is registered process and there isn't process
637 * waiting synchronously for message AND state of queue changed from
638 * empty to not empty. Here we are sure that no one is waiting
640 if (info->notify_owner &&
641 info->attr.mq_curmsgs == 1) {
642 struct siginfo sig_i;
643 switch (info->notify.sigev_notify) {
649 sig_i.si_signo = info->notify.sigev_signo;
651 sig_i.si_code = SI_MESGQ;
652 sig_i.si_value = info->notify.sigev_value;
653 /* map current pid/uid into info->owner's namespaces */
655 sig_i.si_pid = task_tgid_nr_ns(current,
656 ns_of_pid(info->notify_owner));
657 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
660 kill_pid_info(info->notify.sigev_signo,
661 &sig_i, info->notify_owner);
664 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
665 netlink_sendskb(info->notify_sock, info->notify_cookie);
668 /* after notification unregisters process */
669 put_pid(info->notify_owner);
670 put_user_ns(info->notify_user_ns);
671 info->notify_owner = NULL;
672 info->notify_user_ns = NULL;
674 wake_up(&info->wait_q);
677 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
678 struct timespec64 *ts)
680 if (get_timespec64(ts, u_abs_timeout))
682 if (!timespec64_valid(ts))
687 static void remove_notification(struct mqueue_inode_info *info)
689 if (info->notify_owner != NULL &&
690 info->notify.sigev_notify == SIGEV_THREAD) {
691 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
692 netlink_sendskb(info->notify_sock, info->notify_cookie);
694 put_pid(info->notify_owner);
695 put_user_ns(info->notify_user_ns);
696 info->notify_owner = NULL;
697 info->notify_user_ns = NULL;
700 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
703 unsigned long total_size;
705 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
707 if (capable(CAP_SYS_RESOURCE)) {
708 if (attr->mq_maxmsg > HARD_MSGMAX ||
709 attr->mq_msgsize > HARD_MSGSIZEMAX)
712 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
713 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
716 /* check for overflow */
717 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
719 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
720 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
721 sizeof(struct posix_msg_tree_node);
722 total_size = attr->mq_maxmsg * attr->mq_msgsize;
723 if (total_size + mq_treesize < total_size)
729 * Invoked when creating a new queue via sys_mq_open
731 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
732 struct path *path, int oflag, umode_t mode,
733 struct mq_attr *attr)
735 const struct cred *cred = current_cred();
739 ret = mq_attr_ok(ipc_ns, attr);
742 /* store for use during create */
743 path->dentry->d_fsdata = attr;
745 struct mq_attr def_attr;
747 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
748 ipc_ns->mq_msg_default);
749 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
750 ipc_ns->mq_msgsize_default);
751 ret = mq_attr_ok(ipc_ns, &def_attr);
756 mode &= ~current_umask();
757 ret = vfs_create(dir, path->dentry, mode, true);
758 path->dentry->d_fsdata = NULL;
761 return dentry_open(path, oflag, cred);
764 /* Opens existing queue */
765 static struct file *do_open(struct path *path, int oflag)
767 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
768 MAY_READ | MAY_WRITE };
770 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
771 return ERR_PTR(-EINVAL);
772 acc = oflag2acc[oflag & O_ACCMODE];
773 if (inode_permission(d_inode(path->dentry), acc))
774 return ERR_PTR(-EACCES);
775 return dentry_open(path, oflag, current_cred());
778 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
779 struct mq_attr *attr)
783 struct filename *name;
785 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
786 struct vfsmount *mnt = ipc_ns->mq_mnt;
787 struct dentry *root = mnt->mnt_root;
790 audit_mq_open(oflag, mode, attr);
792 if (IS_ERR(name = getname(u_name)))
793 return PTR_ERR(name);
795 fd = get_unused_fd_flags(O_CLOEXEC);
799 ro = mnt_want_write(mnt); /* we'll drop it in any case */
801 inode_lock(d_inode(root));
802 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
803 if (IS_ERR(path.dentry)) {
804 error = PTR_ERR(path.dentry);
807 path.mnt = mntget(mnt);
809 if (oflag & O_CREAT) {
810 if (d_really_is_positive(path.dentry)) { /* entry already exists */
811 audit_inode(name, path.dentry, 0);
812 if (oflag & O_EXCL) {
816 filp = do_open(&path, oflag);
822 audit_inode_parent_hidden(name, root);
823 filp = do_create(ipc_ns, d_inode(root), &path,
827 if (d_really_is_negative(path.dentry)) {
831 audit_inode(name, path.dentry, 0);
832 filp = do_open(&path, oflag);
836 fd_install(fd, filp);
838 error = PTR_ERR(filp);
846 inode_unlock(d_inode(root));
854 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
855 struct mq_attr __user *, u_attr)
858 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
861 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
864 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
867 struct filename *name;
868 struct dentry *dentry;
869 struct inode *inode = NULL;
870 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
871 struct vfsmount *mnt = ipc_ns->mq_mnt;
873 name = getname(u_name);
875 return PTR_ERR(name);
877 audit_inode_parent_hidden(name, mnt->mnt_root);
878 err = mnt_want_write(mnt);
881 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
882 dentry = lookup_one_len(name->name, mnt->mnt_root,
884 if (IS_ERR(dentry)) {
885 err = PTR_ERR(dentry);
889 inode = d_inode(dentry);
894 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
899 inode_unlock(d_inode(mnt->mnt_root));
909 /* Pipelined send and receive functions.
911 * If a receiver finds no waiting message, then it registers itself in the
912 * list of waiting receivers. A sender checks that list before adding the new
913 * message into the message array. If there is a waiting receiver, then it
914 * bypasses the message array and directly hands the message over to the
915 * receiver. The receiver accepts the message and returns without grabbing the
918 * - Set pointer to message.
919 * - Queue the receiver task for later wakeup (without the info->lock).
920 * - Update its state to STATE_READY. Now the receiver can continue.
921 * - Wake up the process after the lock is dropped. Should the process wake up
922 * before this wakeup (due to a timeout or a signal) it will either see
923 * STATE_READY and continue or acquire the lock to check the state again.
925 * The same algorithm is used for senders.
928 /* pipelined_send() - send a message directly to the task waiting in
929 * sys_mq_timedreceive() (without inserting message into a queue).
931 static inline void pipelined_send(struct wake_q_head *wake_q,
932 struct mqueue_inode_info *info,
933 struct msg_msg *message,
934 struct ext_wait_queue *receiver)
936 receiver->msg = message;
937 list_del(&receiver->list);
938 wake_q_add(wake_q, receiver->task);
940 * Rely on the implicit cmpxchg barrier from wake_q_add such
941 * that we can ensure that updating receiver->state is the last
942 * write operation: As once set, the receiver can continue,
943 * and if we don't have the reference count from the wake_q,
944 * yet, at that point we can later have a use-after-free
945 * condition and bogus wakeup.
947 receiver->state = STATE_READY;
950 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
951 * gets its message and put to the queue (we have one free place for sure). */
952 static inline void pipelined_receive(struct wake_q_head *wake_q,
953 struct mqueue_inode_info *info)
955 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
959 wake_up_interruptible(&info->wait_q);
962 if (msg_insert(sender->msg, info))
965 list_del(&sender->list);
966 wake_q_add(wake_q, sender->task);
967 sender->state = STATE_READY;
970 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
971 size_t msg_len, unsigned int msg_prio,
972 struct timespec64 *ts)
976 struct ext_wait_queue wait;
977 struct ext_wait_queue *receiver;
978 struct msg_msg *msg_ptr;
979 struct mqueue_inode_info *info;
980 ktime_t expires, *timeout = NULL;
981 struct posix_msg_tree_node *new_leaf = NULL;
983 DEFINE_WAKE_Q(wake_q);
985 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
989 expires = timespec64_to_ktime(*ts);
993 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
996 if (unlikely(!f.file)) {
1001 inode = file_inode(f.file);
1002 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1006 info = MQUEUE_I(inode);
1009 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1014 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1019 /* First try to allocate memory, before doing anything with
1020 * existing queues. */
1021 msg_ptr = load_msg(u_msg_ptr, msg_len);
1022 if (IS_ERR(msg_ptr)) {
1023 ret = PTR_ERR(msg_ptr);
1026 msg_ptr->m_ts = msg_len;
1027 msg_ptr->m_type = msg_prio;
1030 * msg_insert really wants us to have a valid, spare node struct so
1031 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1032 * fall back to that if necessary.
1034 if (!info->node_cache)
1035 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1037 spin_lock(&info->lock);
1039 if (!info->node_cache && new_leaf) {
1040 /* Save our speculative allocation into the cache */
1041 INIT_LIST_HEAD(&new_leaf->msg_list);
1042 info->node_cache = new_leaf;
1048 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1049 if (f.file->f_flags & O_NONBLOCK) {
1052 wait.task = current;
1053 wait.msg = (void *) msg_ptr;
1054 wait.state = STATE_NONE;
1055 ret = wq_sleep(info, SEND, timeout, &wait);
1057 * wq_sleep must be called with info->lock held, and
1058 * returns with the lock released
1063 receiver = wq_get_first_waiter(info, RECV);
1065 pipelined_send(&wake_q, info, msg_ptr, receiver);
1067 /* adds message to the queue */
1068 ret = msg_insert(msg_ptr, info);
1073 inode->i_atime = inode->i_mtime = inode->i_ctime =
1074 current_time(inode);
1077 spin_unlock(&info->lock);
1088 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1089 size_t msg_len, unsigned int __user *u_msg_prio,
1090 struct timespec64 *ts)
1093 struct msg_msg *msg_ptr;
1095 struct inode *inode;
1096 struct mqueue_inode_info *info;
1097 struct ext_wait_queue wait;
1098 ktime_t expires, *timeout = NULL;
1099 struct posix_msg_tree_node *new_leaf = NULL;
1102 expires = timespec64_to_ktime(*ts);
1106 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1109 if (unlikely(!f.file)) {
1114 inode = file_inode(f.file);
1115 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1119 info = MQUEUE_I(inode);
1122 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1127 /* checks if buffer is big enough */
1128 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1134 * msg_insert really wants us to have a valid, spare node struct so
1135 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1136 * fall back to that if necessary.
1138 if (!info->node_cache)
1139 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1141 spin_lock(&info->lock);
1143 if (!info->node_cache && new_leaf) {
1144 /* Save our speculative allocation into the cache */
1145 INIT_LIST_HEAD(&new_leaf->msg_list);
1146 info->node_cache = new_leaf;
1151 if (info->attr.mq_curmsgs == 0) {
1152 if (f.file->f_flags & O_NONBLOCK) {
1153 spin_unlock(&info->lock);
1156 wait.task = current;
1157 wait.state = STATE_NONE;
1158 ret = wq_sleep(info, RECV, timeout, &wait);
1162 DEFINE_WAKE_Q(wake_q);
1164 msg_ptr = msg_get(info);
1166 inode->i_atime = inode->i_mtime = inode->i_ctime =
1167 current_time(inode);
1169 /* There is now free space in queue. */
1170 pipelined_receive(&wake_q, info);
1171 spin_unlock(&info->lock);
1176 ret = msg_ptr->m_ts;
1178 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1179 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1190 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1191 size_t, msg_len, unsigned int, msg_prio,
1192 const struct timespec __user *, u_abs_timeout)
1194 struct timespec64 ts, *p = NULL;
1195 if (u_abs_timeout) {
1196 int res = prepare_timeout(u_abs_timeout, &ts);
1201 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1204 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1205 size_t, msg_len, unsigned int __user *, u_msg_prio,
1206 const struct timespec __user *, u_abs_timeout)
1208 struct timespec64 ts, *p = NULL;
1209 if (u_abs_timeout) {
1210 int res = prepare_timeout(u_abs_timeout, &ts);
1215 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1219 * Notes: the case when user wants us to deregister (with NULL as pointer)
1220 * and he isn't currently owner of notification, will be silently discarded.
1221 * It isn't explicitly defined in the POSIX.
1223 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1228 struct inode *inode;
1229 struct mqueue_inode_info *info;
1232 audit_mq_notify(mqdes, notification);
1236 if (notification != NULL) {
1237 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1238 notification->sigev_notify != SIGEV_SIGNAL &&
1239 notification->sigev_notify != SIGEV_THREAD))
1241 if (notification->sigev_notify == SIGEV_SIGNAL &&
1242 !valid_signal(notification->sigev_signo)) {
1245 if (notification->sigev_notify == SIGEV_THREAD) {
1248 /* create the notify skb */
1249 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1254 if (copy_from_user(nc->data,
1255 notification->sigev_value.sival_ptr,
1256 NOTIFY_COOKIE_LEN)) {
1261 /* TODO: add a header? */
1262 skb_put(nc, NOTIFY_COOKIE_LEN);
1263 /* and attach it to the socket */
1265 f = fdget(notification->sigev_signo);
1270 sock = netlink_getsockbyfilp(f.file);
1273 ret = PTR_ERR(sock);
1278 timeo = MAX_SCHEDULE_TIMEOUT;
1279 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1298 inode = file_inode(f.file);
1299 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1303 info = MQUEUE_I(inode);
1306 spin_lock(&info->lock);
1307 if (notification == NULL) {
1308 if (info->notify_owner == task_tgid(current)) {
1309 remove_notification(info);
1310 inode->i_atime = inode->i_ctime = current_time(inode);
1312 } else if (info->notify_owner != NULL) {
1315 switch (notification->sigev_notify) {
1317 info->notify.sigev_notify = SIGEV_NONE;
1320 info->notify_sock = sock;
1321 info->notify_cookie = nc;
1324 info->notify.sigev_notify = SIGEV_THREAD;
1327 info->notify.sigev_signo = notification->sigev_signo;
1328 info->notify.sigev_value = notification->sigev_value;
1329 info->notify.sigev_notify = SIGEV_SIGNAL;
1333 info->notify_owner = get_pid(task_tgid(current));
1334 info->notify_user_ns = get_user_ns(current_user_ns());
1335 inode->i_atime = inode->i_ctime = current_time(inode);
1337 spin_unlock(&info->lock);
1342 netlink_detachskb(sock, nc);
1349 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1350 const struct sigevent __user *, u_notification)
1352 struct sigevent n, *p = NULL;
1353 if (u_notification) {
1354 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1358 return do_mq_notify(mqdes, p);
1361 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1364 struct inode *inode;
1365 struct mqueue_inode_info *info;
1367 if (new && (new->mq_flags & (~O_NONBLOCK)))
1374 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1379 inode = file_inode(f.file);
1380 info = MQUEUE_I(inode);
1382 spin_lock(&info->lock);
1386 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1389 audit_mq_getsetattr(mqdes, new);
1390 spin_lock(&f.file->f_lock);
1391 if (new->mq_flags & O_NONBLOCK)
1392 f.file->f_flags |= O_NONBLOCK;
1394 f.file->f_flags &= ~O_NONBLOCK;
1395 spin_unlock(&f.file->f_lock);
1397 inode->i_atime = inode->i_ctime = current_time(inode);
1400 spin_unlock(&info->lock);
1405 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1406 const struct mq_attr __user *, u_mqstat,
1407 struct mq_attr __user *, u_omqstat)
1410 struct mq_attr mqstat, omqstat;
1411 struct mq_attr *new = NULL, *old = NULL;
1415 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1421 ret = do_mq_getsetattr(mqdes, new, old);
1425 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1430 #ifdef CONFIG_COMPAT
1432 struct compat_mq_attr {
1433 compat_long_t mq_flags; /* message queue flags */
1434 compat_long_t mq_maxmsg; /* maximum number of messages */
1435 compat_long_t mq_msgsize; /* maximum message size */
1436 compat_long_t mq_curmsgs; /* number of messages currently queued */
1437 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1440 static inline int get_compat_mq_attr(struct mq_attr *attr,
1441 const struct compat_mq_attr __user *uattr)
1443 struct compat_mq_attr v;
1445 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1448 memset(attr, 0, sizeof(*attr));
1449 attr->mq_flags = v.mq_flags;
1450 attr->mq_maxmsg = v.mq_maxmsg;
1451 attr->mq_msgsize = v.mq_msgsize;
1452 attr->mq_curmsgs = v.mq_curmsgs;
1456 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1457 struct compat_mq_attr __user *uattr)
1459 struct compat_mq_attr v;
1461 memset(&v, 0, sizeof(v));
1462 v.mq_flags = attr->mq_flags;
1463 v.mq_maxmsg = attr->mq_maxmsg;
1464 v.mq_msgsize = attr->mq_msgsize;
1465 v.mq_curmsgs = attr->mq_curmsgs;
1466 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1471 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1472 int, oflag, compat_mode_t, mode,
1473 struct compat_mq_attr __user *, u_attr)
1475 struct mq_attr attr, *p = NULL;
1476 if (u_attr && oflag & O_CREAT) {
1478 if (get_compat_mq_attr(&attr, u_attr))
1481 return do_mq_open(u_name, oflag, mode, p);
1484 static int compat_prepare_timeout(const struct compat_timespec __user *p,
1485 struct timespec64 *ts)
1487 if (compat_get_timespec64(ts, p))
1489 if (!timespec64_valid(ts))
1494 COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1495 const char __user *, u_msg_ptr,
1496 compat_size_t, msg_len, unsigned int, msg_prio,
1497 const struct compat_timespec __user *, u_abs_timeout)
1499 struct timespec64 ts, *p = NULL;
1500 if (u_abs_timeout) {
1501 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1506 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1509 COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1510 char __user *, u_msg_ptr,
1511 compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1512 const struct compat_timespec __user *, u_abs_timeout)
1514 struct timespec64 ts, *p = NULL;
1515 if (u_abs_timeout) {
1516 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1521 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1524 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1525 const struct compat_sigevent __user *, u_notification)
1527 struct sigevent n, *p = NULL;
1528 if (u_notification) {
1529 if (get_compat_sigevent(&n, u_notification))
1531 if (n.sigev_notify == SIGEV_THREAD)
1532 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1535 return do_mq_notify(mqdes, p);
1538 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1539 const struct compat_mq_attr __user *, u_mqstat,
1540 struct compat_mq_attr __user *, u_omqstat)
1543 struct mq_attr mqstat, omqstat;
1544 struct mq_attr *new = NULL, *old = NULL;
1548 if (get_compat_mq_attr(new, u_mqstat))
1554 ret = do_mq_getsetattr(mqdes, new, old);
1558 if (put_compat_mq_attr(old, u_omqstat))
1564 static const struct inode_operations mqueue_dir_inode_operations = {
1565 .lookup = simple_lookup,
1566 .create = mqueue_create,
1567 .unlink = mqueue_unlink,
1570 static const struct file_operations mqueue_file_operations = {
1571 .flush = mqueue_flush_file,
1572 .poll = mqueue_poll_file,
1573 .read = mqueue_read_file,
1574 .llseek = default_llseek,
1577 static const struct super_operations mqueue_super_ops = {
1578 .alloc_inode = mqueue_alloc_inode,
1579 .destroy_inode = mqueue_destroy_inode,
1580 .evict_inode = mqueue_evict_inode,
1581 .statfs = simple_statfs,
1584 static struct file_system_type mqueue_fs_type = {
1586 .mount = mqueue_mount,
1587 .kill_sb = kill_litter_super,
1588 .fs_flags = FS_USERNS_MOUNT,
1591 int mq_init_ns(struct ipc_namespace *ns)
1593 ns->mq_queues_count = 0;
1594 ns->mq_queues_max = DFLT_QUEUESMAX;
1595 ns->mq_msg_max = DFLT_MSGMAX;
1596 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1597 ns->mq_msg_default = DFLT_MSG;
1598 ns->mq_msgsize_default = DFLT_MSGSIZE;
1600 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1601 if (IS_ERR(ns->mq_mnt)) {
1602 int err = PTR_ERR(ns->mq_mnt);
1609 void mq_clear_sbinfo(struct ipc_namespace *ns)
1611 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1614 void mq_put_mnt(struct ipc_namespace *ns)
1616 kern_unmount(ns->mq_mnt);
1619 static int __init init_mqueue_fs(void)
1623 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1624 sizeof(struct mqueue_inode_info), 0,
1625 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1626 if (mqueue_inode_cachep == NULL)
1629 /* ignore failures - they are not fatal */
1630 mq_sysctl_table = mq_register_sysctl_table();
1632 error = register_filesystem(&mqueue_fs_type);
1636 spin_lock_init(&mq_lock);
1638 error = mq_init_ns(&init_ipc_ns);
1640 goto out_filesystem;
1645 unregister_filesystem(&mqueue_fs_type);
1647 if (mq_sysctl_table)
1648 unregister_sysctl_table(mq_sysctl_table);
1649 kmem_cache_destroy(mqueue_inode_cachep);
1653 device_initcall(init_mqueue_fs);