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/fs_context.h>
22 #include <linux/namei.h>
23 #include <linux/sysctl.h>
24 #include <linux/poll.h>
25 #include <linux/mqueue.h>
26 #include <linux/msg.h>
27 #include <linux/skbuff.h>
28 #include <linux/vmalloc.h>
29 #include <linux/netlink.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
32 #include <linux/signal.h>
33 #include <linux/mutex.h>
34 #include <linux/nsproxy.h>
35 #include <linux/pid.h>
36 #include <linux/ipc_namespace.h>
37 #include <linux/user_namespace.h>
38 #include <linux/slab.h>
39 #include <linux/sched/wake_q.h>
40 #include <linux/sched/signal.h>
41 #include <linux/sched/user.h>
46 struct mqueue_fs_context {
47 struct ipc_namespace *ipc_ns;
50 #define MQUEUE_MAGIC 0x19800202
51 #define DIRENT_SIZE 20
52 #define FILENT_SIZE 80
60 struct posix_msg_tree_node {
61 struct rb_node rb_node;
62 struct list_head msg_list;
69 * Accesses to a message queue are synchronized by acquiring info->lock.
71 * There are two notable exceptions:
72 * - The actual wakeup of a sleeping task is performed using the wake_q
73 * framework. info->lock is already released when wake_up_q is called.
74 * - The exit codepaths after sleeping check ext_wait_queue->state without
75 * any locks. If it is STATE_READY, then the syscall is completed without
76 * acquiring info->lock.
79 * To achieve proper release/acquire memory barrier pairing, the state is set to
80 * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
81 * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
83 * This prevents the following races:
85 * 1) With the simple wake_q_add(), the task could be gone already before
86 * the increase of the reference happens
89 * WRITE_ONCE(wait.state, STATE_NONE);
90 * schedule_hrtimeout()
92 * if (cmpxchg()) // success
93 * ->state = STATE_READY (reordered)
95 * if (wait.state == STATE_READY) return;
96 * sysret to user space
98 * get_task_struct() // UaF
100 * Solution: Use wake_q_add_safe() and perform the get_task_struct() before
101 * the smp_store_release() that does ->state = STATE_READY.
103 * 2) Without proper _release/_acquire barriers, the woken up task
104 * could read stale data
109 * WRITE_ONCE(wait.state, STATE_NONE);
110 * schedule_hrtimeout()
111 * state = STATE_READY;
113 * if (wait.state == STATE_READY) return;
114 * msg_ptr = wait.msg; // Access to stale data!
115 * receiver->msg = message; (reordered)
117 * Solution: use _release and _acquire barriers.
119 * 3) There is intentionally no barrier when setting current->state
120 * to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
121 * release memory barrier, and the wakeup is triggered when holding
122 * info->lock, i.e. spin_lock(&info->lock) provided a pairing
123 * acquire memory barrier.
126 struct ext_wait_queue { /* queue of sleeping tasks */
127 struct task_struct *task;
128 struct list_head list;
129 struct msg_msg *msg; /* ptr of loaded message */
130 int state; /* one of STATE_* values */
133 struct mqueue_inode_info {
135 struct inode vfs_inode;
136 wait_queue_head_t wait_q;
138 struct rb_root msg_tree;
139 struct rb_node *msg_tree_rightmost;
140 struct posix_msg_tree_node *node_cache;
143 struct sigevent notify;
144 struct pid *notify_owner;
145 u32 notify_self_exec_id;
146 struct user_namespace *notify_user_ns;
147 struct user_struct *user; /* user who created, for accounting */
148 struct sock *notify_sock;
149 struct sk_buff *notify_cookie;
151 /* for tasks waiting for free space and messages, respectively */
152 struct ext_wait_queue e_wait_q[2];
154 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
157 static struct file_system_type mqueue_fs_type;
158 static const struct inode_operations mqueue_dir_inode_operations;
159 static const struct file_operations mqueue_file_operations;
160 static const struct super_operations mqueue_super_ops;
161 static const struct fs_context_operations mqueue_fs_context_ops;
162 static void remove_notification(struct mqueue_inode_info *info);
164 static struct kmem_cache *mqueue_inode_cachep;
166 static struct ctl_table_header *mq_sysctl_table;
168 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
170 return container_of(inode, struct mqueue_inode_info, vfs_inode);
174 * This routine should be called with the mq_lock held.
176 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
178 return get_ipc_ns(inode->i_sb->s_fs_info);
181 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
183 struct ipc_namespace *ns;
186 ns = __get_ns_from_inode(inode);
187 spin_unlock(&mq_lock);
191 /* Auxiliary functions to manipulate messages' list */
192 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
194 struct rb_node **p, *parent = NULL;
195 struct posix_msg_tree_node *leaf;
196 bool rightmost = true;
198 p = &info->msg_tree.rb_node;
201 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
203 if (likely(leaf->priority == msg->m_type))
205 else if (msg->m_type < leaf->priority) {
211 if (info->node_cache) {
212 leaf = info->node_cache;
213 info->node_cache = NULL;
215 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
218 INIT_LIST_HEAD(&leaf->msg_list);
220 leaf->priority = msg->m_type;
223 info->msg_tree_rightmost = &leaf->rb_node;
225 rb_link_node(&leaf->rb_node, parent, p);
226 rb_insert_color(&leaf->rb_node, &info->msg_tree);
228 info->attr.mq_curmsgs++;
229 info->qsize += msg->m_ts;
230 list_add_tail(&msg->m_list, &leaf->msg_list);
234 static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
235 struct mqueue_inode_info *info)
237 struct rb_node *node = &leaf->rb_node;
239 if (info->msg_tree_rightmost == node)
240 info->msg_tree_rightmost = rb_prev(node);
242 rb_erase(node, &info->msg_tree);
243 if (info->node_cache)
246 info->node_cache = leaf;
249 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
251 struct rb_node *parent = NULL;
252 struct posix_msg_tree_node *leaf;
257 * During insert, low priorities go to the left and high to the
258 * right. On receive, we want the highest priorities first, so
259 * walk all the way to the right.
261 parent = info->msg_tree_rightmost;
263 if (info->attr.mq_curmsgs) {
264 pr_warn_once("Inconsistency in POSIX message queue, "
265 "no tree element, but supposedly messages "
267 info->attr.mq_curmsgs = 0;
271 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
272 if (unlikely(list_empty(&leaf->msg_list))) {
273 pr_warn_once("Inconsistency in POSIX message queue, "
274 "empty leaf node but we haven't implemented "
275 "lazy leaf delete!\n");
276 msg_tree_erase(leaf, info);
279 msg = list_first_entry(&leaf->msg_list,
280 struct msg_msg, m_list);
281 list_del(&msg->m_list);
282 if (list_empty(&leaf->msg_list)) {
283 msg_tree_erase(leaf, info);
286 info->attr.mq_curmsgs--;
287 info->qsize -= msg->m_ts;
291 static struct inode *mqueue_get_inode(struct super_block *sb,
292 struct ipc_namespace *ipc_ns, umode_t mode,
293 struct mq_attr *attr)
295 struct user_struct *u = current_user();
299 inode = new_inode(sb);
303 inode->i_ino = get_next_ino();
304 inode->i_mode = mode;
305 inode->i_uid = current_fsuid();
306 inode->i_gid = current_fsgid();
307 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
310 struct mqueue_inode_info *info;
311 unsigned long mq_bytes, mq_treesize;
313 inode->i_fop = &mqueue_file_operations;
314 inode->i_size = FILENT_SIZE;
315 /* mqueue specific info */
316 info = MQUEUE_I(inode);
317 spin_lock_init(&info->lock);
318 init_waitqueue_head(&info->wait_q);
319 INIT_LIST_HEAD(&info->e_wait_q[0].list);
320 INIT_LIST_HEAD(&info->e_wait_q[1].list);
321 info->notify_owner = NULL;
322 info->notify_user_ns = NULL;
324 info->user = NULL; /* set when all is ok */
325 info->msg_tree = RB_ROOT;
326 info->msg_tree_rightmost = NULL;
327 info->node_cache = NULL;
328 memset(&info->attr, 0, sizeof(info->attr));
329 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
330 ipc_ns->mq_msg_default);
331 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
332 ipc_ns->mq_msgsize_default);
334 info->attr.mq_maxmsg = attr->mq_maxmsg;
335 info->attr.mq_msgsize = attr->mq_msgsize;
338 * We used to allocate a static array of pointers and account
339 * the size of that array as well as one msg_msg struct per
340 * possible message into the queue size. That's no longer
341 * accurate as the queue is now an rbtree and will grow and
342 * shrink depending on usage patterns. We can, however, still
343 * account one msg_msg struct per message, but the nodes are
344 * allocated depending on priority usage, and most programs
345 * only use one, or a handful, of priorities. However, since
346 * this is pinned memory, we need to assume worst case, so
347 * that means the min(mq_maxmsg, max_priorities) * struct
348 * posix_msg_tree_node.
352 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
354 if (capable(CAP_SYS_RESOURCE)) {
355 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
356 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
359 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
360 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
364 /* check for overflow */
365 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
367 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
368 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
369 sizeof(struct posix_msg_tree_node);
370 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
371 if (mq_bytes + mq_treesize < mq_bytes)
373 mq_bytes += mq_treesize;
375 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
376 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
377 spin_unlock(&mq_lock);
378 /* mqueue_evict_inode() releases info->messages */
382 u->mq_bytes += mq_bytes;
383 spin_unlock(&mq_lock);
386 info->user = get_uid(u);
387 } else if (S_ISDIR(mode)) {
389 /* Some things misbehave if size == 0 on a directory */
390 inode->i_size = 2 * DIRENT_SIZE;
391 inode->i_op = &mqueue_dir_inode_operations;
392 inode->i_fop = &simple_dir_operations;
402 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
405 struct ipc_namespace *ns = sb->s_fs_info;
407 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
408 sb->s_blocksize = PAGE_SIZE;
409 sb->s_blocksize_bits = PAGE_SHIFT;
410 sb->s_magic = MQUEUE_MAGIC;
411 sb->s_op = &mqueue_super_ops;
413 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
415 return PTR_ERR(inode);
417 sb->s_root = d_make_root(inode);
423 static int mqueue_get_tree(struct fs_context *fc)
425 struct mqueue_fs_context *ctx = fc->fs_private;
427 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
430 static void mqueue_fs_context_free(struct fs_context *fc)
432 struct mqueue_fs_context *ctx = fc->fs_private;
434 put_ipc_ns(ctx->ipc_ns);
438 static int mqueue_init_fs_context(struct fs_context *fc)
440 struct mqueue_fs_context *ctx;
442 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
446 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
447 put_user_ns(fc->user_ns);
448 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
449 fc->fs_private = ctx;
450 fc->ops = &mqueue_fs_context_ops;
454 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
456 struct mqueue_fs_context *ctx;
457 struct fs_context *fc;
458 struct vfsmount *mnt;
460 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
464 ctx = fc->fs_private;
465 put_ipc_ns(ctx->ipc_ns);
466 ctx->ipc_ns = get_ipc_ns(ns);
467 put_user_ns(fc->user_ns);
468 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
475 static void init_once(void *foo)
477 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
479 inode_init_once(&p->vfs_inode);
482 static struct inode *mqueue_alloc_inode(struct super_block *sb)
484 struct mqueue_inode_info *ei;
486 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
489 return &ei->vfs_inode;
492 static void mqueue_free_inode(struct inode *inode)
494 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
497 static void mqueue_evict_inode(struct inode *inode)
499 struct mqueue_inode_info *info;
500 struct user_struct *user;
501 struct ipc_namespace *ipc_ns;
502 struct msg_msg *msg, *nmsg;
507 if (S_ISDIR(inode->i_mode))
510 ipc_ns = get_ns_from_inode(inode);
511 info = MQUEUE_I(inode);
512 spin_lock(&info->lock);
513 while ((msg = msg_get(info)) != NULL)
514 list_add_tail(&msg->m_list, &tmp_msg);
515 kfree(info->node_cache);
516 spin_unlock(&info->lock);
518 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
519 list_del(&msg->m_list);
525 unsigned long mq_bytes, mq_treesize;
527 /* Total amount of bytes accounted for the mqueue */
528 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
529 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
530 sizeof(struct posix_msg_tree_node);
532 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
533 info->attr.mq_msgsize);
536 user->mq_bytes -= mq_bytes;
538 * get_ns_from_inode() ensures that the
539 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
540 * to which we now hold a reference, or it is NULL.
541 * We can't put it here under mq_lock, though.
544 ipc_ns->mq_queues_count--;
545 spin_unlock(&mq_lock);
552 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
554 struct inode *dir = dentry->d_parent->d_inode;
556 struct mq_attr *attr = arg;
558 struct ipc_namespace *ipc_ns;
561 ipc_ns = __get_ns_from_inode(dir);
567 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
568 !capable(CAP_SYS_RESOURCE)) {
572 ipc_ns->mq_queues_count++;
573 spin_unlock(&mq_lock);
575 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
577 error = PTR_ERR(inode);
579 ipc_ns->mq_queues_count--;
584 dir->i_size += DIRENT_SIZE;
585 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
587 d_instantiate(dentry, inode);
591 spin_unlock(&mq_lock);
597 static int mqueue_create(struct inode *dir, struct dentry *dentry,
598 umode_t mode, bool excl)
600 return mqueue_create_attr(dentry, mode, NULL);
603 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
605 struct inode *inode = d_inode(dentry);
607 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
608 dir->i_size -= DIRENT_SIZE;
615 * This is routine for system read from queue file.
616 * To avoid mess with doing here some sort of mq_receive we allow
617 * to read only queue size & notification info (the only values
618 * that are interesting from user point of view and aren't accessible
619 * through std routines)
621 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
622 size_t count, loff_t *off)
624 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
625 char buffer[FILENT_SIZE];
628 spin_lock(&info->lock);
629 snprintf(buffer, sizeof(buffer),
630 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
632 info->notify_owner ? info->notify.sigev_notify : 0,
633 (info->notify_owner &&
634 info->notify.sigev_notify == SIGEV_SIGNAL) ?
635 info->notify.sigev_signo : 0,
636 pid_vnr(info->notify_owner));
637 spin_unlock(&info->lock);
638 buffer[sizeof(buffer)-1] = '\0';
640 ret = simple_read_from_buffer(u_data, count, off, buffer,
645 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
649 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
651 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
653 spin_lock(&info->lock);
654 if (task_tgid(current) == info->notify_owner)
655 remove_notification(info);
657 spin_unlock(&info->lock);
661 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
663 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
666 poll_wait(filp, &info->wait_q, poll_tab);
668 spin_lock(&info->lock);
669 if (info->attr.mq_curmsgs)
670 retval = EPOLLIN | EPOLLRDNORM;
672 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
673 retval |= EPOLLOUT | EPOLLWRNORM;
674 spin_unlock(&info->lock);
679 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
680 static void wq_add(struct mqueue_inode_info *info, int sr,
681 struct ext_wait_queue *ewp)
683 struct ext_wait_queue *walk;
685 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
686 if (walk->task->prio <= current->prio) {
687 list_add_tail(&ewp->list, &walk->list);
691 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
695 * Puts current task to sleep. Caller must hold queue lock. After return
699 static int wq_sleep(struct mqueue_inode_info *info, int sr,
700 ktime_t *timeout, struct ext_wait_queue *ewp)
701 __releases(&info->lock)
706 wq_add(info, sr, ewp);
709 /* memory barrier not required, we hold info->lock */
710 __set_current_state(TASK_INTERRUPTIBLE);
712 spin_unlock(&info->lock);
713 time = schedule_hrtimeout_range_clock(timeout, 0,
714 HRTIMER_MODE_ABS, CLOCK_REALTIME);
716 if (READ_ONCE(ewp->state) == STATE_READY) {
717 /* see MQ_BARRIER for purpose/pairing */
718 smp_acquire__after_ctrl_dep();
722 spin_lock(&info->lock);
724 /* we hold info->lock, so no memory barrier required */
725 if (READ_ONCE(ewp->state) == STATE_READY) {
729 if (signal_pending(current)) {
730 retval = -ERESTARTSYS;
738 list_del(&ewp->list);
740 spin_unlock(&info->lock);
746 * Returns waiting task that should be serviced first or NULL if none exists
748 static struct ext_wait_queue *wq_get_first_waiter(
749 struct mqueue_inode_info *info, int sr)
751 struct list_head *ptr;
753 ptr = info->e_wait_q[sr].list.prev;
754 if (ptr == &info->e_wait_q[sr].list)
756 return list_entry(ptr, struct ext_wait_queue, list);
760 static inline void set_cookie(struct sk_buff *skb, char code)
762 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
766 * The next function is only to split too long sys_mq_timedsend
768 static void __do_notify(struct mqueue_inode_info *info)
771 * invoked when there is registered process and there isn't process
772 * waiting synchronously for message AND state of queue changed from
773 * empty to not empty. Here we are sure that no one is waiting
775 if (info->notify_owner &&
776 info->attr.mq_curmsgs == 1) {
777 switch (info->notify.sigev_notify) {
781 struct kernel_siginfo sig_i;
782 struct task_struct *task;
784 /* do_mq_notify() accepts sigev_signo == 0, why?? */
785 if (!info->notify.sigev_signo)
788 clear_siginfo(&sig_i);
789 sig_i.si_signo = info->notify.sigev_signo;
791 sig_i.si_code = SI_MESGQ;
792 sig_i.si_value = info->notify.sigev_value;
794 /* map current pid/uid into info->owner's namespaces */
795 sig_i.si_pid = task_tgid_nr_ns(current,
796 ns_of_pid(info->notify_owner));
797 sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
800 * We can't use kill_pid_info(), this signal should
801 * bypass check_kill_permission(). It is from kernel
802 * but si_fromuser() can't know this.
803 * We do check the self_exec_id, to avoid sending
804 * signals to programs that don't expect them.
806 task = pid_task(info->notify_owner, PIDTYPE_TGID);
807 if (task && task->self_exec_id ==
808 info->notify_self_exec_id) {
809 do_send_sig_info(info->notify.sigev_signo,
810 &sig_i, task, PIDTYPE_TGID);
816 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
817 netlink_sendskb(info->notify_sock, info->notify_cookie);
820 /* after notification unregisters process */
821 put_pid(info->notify_owner);
822 put_user_ns(info->notify_user_ns);
823 info->notify_owner = NULL;
824 info->notify_user_ns = NULL;
826 wake_up(&info->wait_q);
829 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
830 struct timespec64 *ts)
832 if (get_timespec64(ts, u_abs_timeout))
834 if (!timespec64_valid(ts))
839 static void remove_notification(struct mqueue_inode_info *info)
841 if (info->notify_owner != NULL &&
842 info->notify.sigev_notify == SIGEV_THREAD) {
843 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
844 netlink_sendskb(info->notify_sock, info->notify_cookie);
846 put_pid(info->notify_owner);
847 put_user_ns(info->notify_user_ns);
848 info->notify_owner = NULL;
849 info->notify_user_ns = NULL;
852 static int prepare_open(struct dentry *dentry, int oflag, int ro,
853 umode_t mode, struct filename *name,
854 struct mq_attr *attr)
856 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
857 MAY_READ | MAY_WRITE };
860 if (d_really_is_negative(dentry)) {
861 if (!(oflag & O_CREAT))
865 audit_inode_parent_hidden(name, dentry->d_parent);
866 return vfs_mkobj(dentry, mode & ~current_umask(),
867 mqueue_create_attr, attr);
869 /* it already existed */
870 audit_inode(name, dentry, 0);
871 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
873 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
875 acc = oflag2acc[oflag & O_ACCMODE];
876 return inode_permission(d_inode(dentry), acc);
879 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
880 struct mq_attr *attr)
882 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
883 struct dentry *root = mnt->mnt_root;
884 struct filename *name;
889 audit_mq_open(oflag, mode, attr);
891 if (IS_ERR(name = getname(u_name)))
892 return PTR_ERR(name);
894 fd = get_unused_fd_flags(O_CLOEXEC);
898 ro = mnt_want_write(mnt); /* we'll drop it in any case */
899 inode_lock(d_inode(root));
900 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
901 if (IS_ERR(path.dentry)) {
902 error = PTR_ERR(path.dentry);
905 path.mnt = mntget(mnt);
906 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
908 struct file *file = dentry_open(&path, oflag, current_cred());
910 fd_install(fd, file);
912 error = PTR_ERR(file);
920 inode_unlock(d_inode(root));
928 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
929 struct mq_attr __user *, u_attr)
932 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
935 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
938 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
941 struct filename *name;
942 struct dentry *dentry;
943 struct inode *inode = NULL;
944 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
945 struct vfsmount *mnt = ipc_ns->mq_mnt;
947 name = getname(u_name);
949 return PTR_ERR(name);
951 audit_inode_parent_hidden(name, mnt->mnt_root);
952 err = mnt_want_write(mnt);
955 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
956 dentry = lookup_one_len(name->name, mnt->mnt_root,
958 if (IS_ERR(dentry)) {
959 err = PTR_ERR(dentry);
963 inode = d_inode(dentry);
968 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
973 inode_unlock(d_inode(mnt->mnt_root));
983 /* Pipelined send and receive functions.
985 * If a receiver finds no waiting message, then it registers itself in the
986 * list of waiting receivers. A sender checks that list before adding the new
987 * message into the message array. If there is a waiting receiver, then it
988 * bypasses the message array and directly hands the message over to the
989 * receiver. The receiver accepts the message and returns without grabbing the
992 * - Set pointer to message.
993 * - Queue the receiver task for later wakeup (without the info->lock).
994 * - Update its state to STATE_READY. Now the receiver can continue.
995 * - Wake up the process after the lock is dropped. Should the process wake up
996 * before this wakeup (due to a timeout or a signal) it will either see
997 * STATE_READY and continue or acquire the lock to check the state again.
999 * The same algorithm is used for senders.
1002 static inline void __pipelined_op(struct wake_q_head *wake_q,
1003 struct mqueue_inode_info *info,
1004 struct ext_wait_queue *this)
1006 list_del(&this->list);
1007 get_task_struct(this->task);
1009 /* see MQ_BARRIER for purpose/pairing */
1010 smp_store_release(&this->state, STATE_READY);
1011 wake_q_add_safe(wake_q, this->task);
1014 /* pipelined_send() - send a message directly to the task waiting in
1015 * sys_mq_timedreceive() (without inserting message into a queue).
1017 static inline void pipelined_send(struct wake_q_head *wake_q,
1018 struct mqueue_inode_info *info,
1019 struct msg_msg *message,
1020 struct ext_wait_queue *receiver)
1022 receiver->msg = message;
1023 __pipelined_op(wake_q, info, receiver);
1026 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1027 * gets its message and put to the queue (we have one free place for sure). */
1028 static inline void pipelined_receive(struct wake_q_head *wake_q,
1029 struct mqueue_inode_info *info)
1031 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1035 wake_up_interruptible(&info->wait_q);
1038 if (msg_insert(sender->msg, info))
1041 __pipelined_op(wake_q, info, sender);
1044 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1045 size_t msg_len, unsigned int msg_prio,
1046 struct timespec64 *ts)
1049 struct inode *inode;
1050 struct ext_wait_queue wait;
1051 struct ext_wait_queue *receiver;
1052 struct msg_msg *msg_ptr;
1053 struct mqueue_inode_info *info;
1054 ktime_t expires, *timeout = NULL;
1055 struct posix_msg_tree_node *new_leaf = NULL;
1057 DEFINE_WAKE_Q(wake_q);
1059 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1063 expires = timespec64_to_ktime(*ts);
1067 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1070 if (unlikely(!f.file)) {
1075 inode = file_inode(f.file);
1076 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1080 info = MQUEUE_I(inode);
1083 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1088 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1093 /* First try to allocate memory, before doing anything with
1094 * existing queues. */
1095 msg_ptr = load_msg(u_msg_ptr, msg_len);
1096 if (IS_ERR(msg_ptr)) {
1097 ret = PTR_ERR(msg_ptr);
1100 msg_ptr->m_ts = msg_len;
1101 msg_ptr->m_type = msg_prio;
1104 * msg_insert really wants us to have a valid, spare node struct so
1105 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1106 * fall back to that if necessary.
1108 if (!info->node_cache)
1109 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1111 spin_lock(&info->lock);
1113 if (!info->node_cache && new_leaf) {
1114 /* Save our speculative allocation into the cache */
1115 INIT_LIST_HEAD(&new_leaf->msg_list);
1116 info->node_cache = new_leaf;
1122 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1123 if (f.file->f_flags & O_NONBLOCK) {
1126 wait.task = current;
1127 wait.msg = (void *) msg_ptr;
1129 /* memory barrier not required, we hold info->lock */
1130 WRITE_ONCE(wait.state, STATE_NONE);
1131 ret = wq_sleep(info, SEND, timeout, &wait);
1133 * wq_sleep must be called with info->lock held, and
1134 * returns with the lock released
1139 receiver = wq_get_first_waiter(info, RECV);
1141 pipelined_send(&wake_q, info, msg_ptr, receiver);
1143 /* adds message to the queue */
1144 ret = msg_insert(msg_ptr, info);
1149 inode->i_atime = inode->i_mtime = inode->i_ctime =
1150 current_time(inode);
1153 spin_unlock(&info->lock);
1164 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1165 size_t msg_len, unsigned int __user *u_msg_prio,
1166 struct timespec64 *ts)
1169 struct msg_msg *msg_ptr;
1171 struct inode *inode;
1172 struct mqueue_inode_info *info;
1173 struct ext_wait_queue wait;
1174 ktime_t expires, *timeout = NULL;
1175 struct posix_msg_tree_node *new_leaf = NULL;
1178 expires = timespec64_to_ktime(*ts);
1182 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1185 if (unlikely(!f.file)) {
1190 inode = file_inode(f.file);
1191 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1195 info = MQUEUE_I(inode);
1198 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1203 /* checks if buffer is big enough */
1204 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1210 * msg_insert really wants us to have a valid, spare node struct so
1211 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1212 * fall back to that if necessary.
1214 if (!info->node_cache)
1215 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1217 spin_lock(&info->lock);
1219 if (!info->node_cache && new_leaf) {
1220 /* Save our speculative allocation into the cache */
1221 INIT_LIST_HEAD(&new_leaf->msg_list);
1222 info->node_cache = new_leaf;
1227 if (info->attr.mq_curmsgs == 0) {
1228 if (f.file->f_flags & O_NONBLOCK) {
1229 spin_unlock(&info->lock);
1232 wait.task = current;
1234 /* memory barrier not required, we hold info->lock */
1235 WRITE_ONCE(wait.state, STATE_NONE);
1236 ret = wq_sleep(info, RECV, timeout, &wait);
1240 DEFINE_WAKE_Q(wake_q);
1242 msg_ptr = msg_get(info);
1244 inode->i_atime = inode->i_mtime = inode->i_ctime =
1245 current_time(inode);
1247 /* There is now free space in queue. */
1248 pipelined_receive(&wake_q, info);
1249 spin_unlock(&info->lock);
1254 ret = msg_ptr->m_ts;
1256 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1257 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1268 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1269 size_t, msg_len, unsigned int, msg_prio,
1270 const struct __kernel_timespec __user *, u_abs_timeout)
1272 struct timespec64 ts, *p = NULL;
1273 if (u_abs_timeout) {
1274 int res = prepare_timeout(u_abs_timeout, &ts);
1279 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1282 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1283 size_t, msg_len, unsigned int __user *, u_msg_prio,
1284 const struct __kernel_timespec __user *, u_abs_timeout)
1286 struct timespec64 ts, *p = NULL;
1287 if (u_abs_timeout) {
1288 int res = prepare_timeout(u_abs_timeout, &ts);
1293 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1297 * Notes: the case when user wants us to deregister (with NULL as pointer)
1298 * and he isn't currently owner of notification, will be silently discarded.
1299 * It isn't explicitly defined in the POSIX.
1301 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1306 struct inode *inode;
1307 struct mqueue_inode_info *info;
1310 audit_mq_notify(mqdes, notification);
1314 if (notification != NULL) {
1315 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1316 notification->sigev_notify != SIGEV_SIGNAL &&
1317 notification->sigev_notify != SIGEV_THREAD))
1319 if (notification->sigev_notify == SIGEV_SIGNAL &&
1320 !valid_signal(notification->sigev_signo)) {
1323 if (notification->sigev_notify == SIGEV_THREAD) {
1326 /* create the notify skb */
1327 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1331 if (copy_from_user(nc->data,
1332 notification->sigev_value.sival_ptr,
1333 NOTIFY_COOKIE_LEN)) {
1338 /* TODO: add a header? */
1339 skb_put(nc, NOTIFY_COOKIE_LEN);
1340 /* and attach it to the socket */
1342 f = fdget(notification->sigev_signo);
1347 sock = netlink_getsockbyfilp(f.file);
1350 ret = PTR_ERR(sock);
1354 timeo = MAX_SCHEDULE_TIMEOUT;
1355 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1371 inode = file_inode(f.file);
1372 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1376 info = MQUEUE_I(inode);
1379 spin_lock(&info->lock);
1380 if (notification == NULL) {
1381 if (info->notify_owner == task_tgid(current)) {
1382 remove_notification(info);
1383 inode->i_atime = inode->i_ctime = current_time(inode);
1385 } else if (info->notify_owner != NULL) {
1388 switch (notification->sigev_notify) {
1390 info->notify.sigev_notify = SIGEV_NONE;
1393 info->notify_sock = sock;
1394 info->notify_cookie = nc;
1397 info->notify.sigev_notify = SIGEV_THREAD;
1400 info->notify.sigev_signo = notification->sigev_signo;
1401 info->notify.sigev_value = notification->sigev_value;
1402 info->notify.sigev_notify = SIGEV_SIGNAL;
1403 info->notify_self_exec_id = current->self_exec_id;
1407 info->notify_owner = get_pid(task_tgid(current));
1408 info->notify_user_ns = get_user_ns(current_user_ns());
1409 inode->i_atime = inode->i_ctime = current_time(inode);
1411 spin_unlock(&info->lock);
1416 netlink_detachskb(sock, nc);
1424 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1425 const struct sigevent __user *, u_notification)
1427 struct sigevent n, *p = NULL;
1428 if (u_notification) {
1429 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1433 return do_mq_notify(mqdes, p);
1436 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1439 struct inode *inode;
1440 struct mqueue_inode_info *info;
1442 if (new && (new->mq_flags & (~O_NONBLOCK)))
1449 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1454 inode = file_inode(f.file);
1455 info = MQUEUE_I(inode);
1457 spin_lock(&info->lock);
1461 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1464 audit_mq_getsetattr(mqdes, new);
1465 spin_lock(&f.file->f_lock);
1466 if (new->mq_flags & O_NONBLOCK)
1467 f.file->f_flags |= O_NONBLOCK;
1469 f.file->f_flags &= ~O_NONBLOCK;
1470 spin_unlock(&f.file->f_lock);
1472 inode->i_atime = inode->i_ctime = current_time(inode);
1475 spin_unlock(&info->lock);
1480 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1481 const struct mq_attr __user *, u_mqstat,
1482 struct mq_attr __user *, u_omqstat)
1485 struct mq_attr mqstat, omqstat;
1486 struct mq_attr *new = NULL, *old = NULL;
1490 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1496 ret = do_mq_getsetattr(mqdes, new, old);
1500 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1505 #ifdef CONFIG_COMPAT
1507 struct compat_mq_attr {
1508 compat_long_t mq_flags; /* message queue flags */
1509 compat_long_t mq_maxmsg; /* maximum number of messages */
1510 compat_long_t mq_msgsize; /* maximum message size */
1511 compat_long_t mq_curmsgs; /* number of messages currently queued */
1512 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1515 static inline int get_compat_mq_attr(struct mq_attr *attr,
1516 const struct compat_mq_attr __user *uattr)
1518 struct compat_mq_attr v;
1520 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1523 memset(attr, 0, sizeof(*attr));
1524 attr->mq_flags = v.mq_flags;
1525 attr->mq_maxmsg = v.mq_maxmsg;
1526 attr->mq_msgsize = v.mq_msgsize;
1527 attr->mq_curmsgs = v.mq_curmsgs;
1531 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1532 struct compat_mq_attr __user *uattr)
1534 struct compat_mq_attr v;
1536 memset(&v, 0, sizeof(v));
1537 v.mq_flags = attr->mq_flags;
1538 v.mq_maxmsg = attr->mq_maxmsg;
1539 v.mq_msgsize = attr->mq_msgsize;
1540 v.mq_curmsgs = attr->mq_curmsgs;
1541 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1546 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1547 int, oflag, compat_mode_t, mode,
1548 struct compat_mq_attr __user *, u_attr)
1550 struct mq_attr attr, *p = NULL;
1551 if (u_attr && oflag & O_CREAT) {
1553 if (get_compat_mq_attr(&attr, u_attr))
1556 return do_mq_open(u_name, oflag, mode, p);
1559 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1560 const struct compat_sigevent __user *, u_notification)
1562 struct sigevent n, *p = NULL;
1563 if (u_notification) {
1564 if (get_compat_sigevent(&n, u_notification))
1566 if (n.sigev_notify == SIGEV_THREAD)
1567 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1570 return do_mq_notify(mqdes, p);
1573 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1574 const struct compat_mq_attr __user *, u_mqstat,
1575 struct compat_mq_attr __user *, u_omqstat)
1578 struct mq_attr mqstat, omqstat;
1579 struct mq_attr *new = NULL, *old = NULL;
1583 if (get_compat_mq_attr(new, u_mqstat))
1589 ret = do_mq_getsetattr(mqdes, new, old);
1593 if (put_compat_mq_attr(old, u_omqstat))
1599 #ifdef CONFIG_COMPAT_32BIT_TIME
1600 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1601 struct timespec64 *ts)
1603 if (get_old_timespec32(ts, p))
1605 if (!timespec64_valid(ts))
1610 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1611 const char __user *, u_msg_ptr,
1612 unsigned int, msg_len, unsigned int, msg_prio,
1613 const struct old_timespec32 __user *, u_abs_timeout)
1615 struct timespec64 ts, *p = NULL;
1616 if (u_abs_timeout) {
1617 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1622 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1625 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1626 char __user *, u_msg_ptr,
1627 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1628 const struct old_timespec32 __user *, u_abs_timeout)
1630 struct timespec64 ts, *p = NULL;
1631 if (u_abs_timeout) {
1632 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1637 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1641 static const struct inode_operations mqueue_dir_inode_operations = {
1642 .lookup = simple_lookup,
1643 .create = mqueue_create,
1644 .unlink = mqueue_unlink,
1647 static const struct file_operations mqueue_file_operations = {
1648 .flush = mqueue_flush_file,
1649 .poll = mqueue_poll_file,
1650 .read = mqueue_read_file,
1651 .llseek = default_llseek,
1654 static const struct super_operations mqueue_super_ops = {
1655 .alloc_inode = mqueue_alloc_inode,
1656 .free_inode = mqueue_free_inode,
1657 .evict_inode = mqueue_evict_inode,
1658 .statfs = simple_statfs,
1661 static const struct fs_context_operations mqueue_fs_context_ops = {
1662 .free = mqueue_fs_context_free,
1663 .get_tree = mqueue_get_tree,
1666 static struct file_system_type mqueue_fs_type = {
1668 .init_fs_context = mqueue_init_fs_context,
1669 .kill_sb = kill_litter_super,
1670 .fs_flags = FS_USERNS_MOUNT,
1673 int mq_init_ns(struct ipc_namespace *ns)
1677 ns->mq_queues_count = 0;
1678 ns->mq_queues_max = DFLT_QUEUESMAX;
1679 ns->mq_msg_max = DFLT_MSGMAX;
1680 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1681 ns->mq_msg_default = DFLT_MSG;
1682 ns->mq_msgsize_default = DFLT_MSGSIZE;
1684 m = mq_create_mount(ns);
1691 void mq_clear_sbinfo(struct ipc_namespace *ns)
1693 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1696 void mq_put_mnt(struct ipc_namespace *ns)
1698 kern_unmount(ns->mq_mnt);
1701 static int __init init_mqueue_fs(void)
1705 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1706 sizeof(struct mqueue_inode_info), 0,
1707 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1708 if (mqueue_inode_cachep == NULL)
1711 /* ignore failures - they are not fatal */
1712 mq_sysctl_table = mq_register_sysctl_table();
1714 error = register_filesystem(&mqueue_fs_type);
1718 spin_lock_init(&mq_lock);
1720 error = mq_init_ns(&init_ipc_ns);
1722 goto out_filesystem;
1727 unregister_filesystem(&mqueue_fs_type);
1729 if (mq_sysctl_table)
1730 unregister_sysctl_table(mq_sysctl_table);
1731 kmem_cache_destroy(mqueue_inode_cachep);
1735 device_initcall(init_mqueue_fs);