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;
48 bool newns; /* Set if newly created ipc namespace */
51 #define MQUEUE_MAGIC 0x19800202
52 #define DIRENT_SIZE 20
53 #define FILENT_SIZE 80
61 struct posix_msg_tree_node {
62 struct rb_node rb_node;
63 struct list_head msg_list;
70 * Accesses to a message queue are synchronized by acquiring info->lock.
72 * There are two notable exceptions:
73 * - The actual wakeup of a sleeping task is performed using the wake_q
74 * framework. info->lock is already released when wake_up_q is called.
75 * - The exit codepaths after sleeping check ext_wait_queue->state without
76 * any locks. If it is STATE_READY, then the syscall is completed without
77 * acquiring info->lock.
80 * To achieve proper release/acquire memory barrier pairing, the state is set to
81 * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
82 * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
84 * This prevents the following races:
86 * 1) With the simple wake_q_add(), the task could be gone already before
87 * the increase of the reference happens
90 * WRITE_ONCE(wait.state, STATE_NONE);
91 * schedule_hrtimeout()
93 * if (cmpxchg()) // success
94 * ->state = STATE_READY (reordered)
96 * if (wait.state == STATE_READY) return;
97 * sysret to user space
99 * get_task_struct() // UaF
101 * Solution: Use wake_q_add_safe() and perform the get_task_struct() before
102 * the smp_store_release() that does ->state = STATE_READY.
104 * 2) Without proper _release/_acquire barriers, the woken up task
105 * could read stale data
110 * WRITE_ONCE(wait.state, STATE_NONE);
111 * schedule_hrtimeout()
112 * state = STATE_READY;
114 * if (wait.state == STATE_READY) return;
115 * msg_ptr = wait.msg; // Access to stale data!
116 * receiver->msg = message; (reordered)
118 * Solution: use _release and _acquire barriers.
120 * 3) There is intentionally no barrier when setting current->state
121 * to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
122 * release memory barrier, and the wakeup is triggered when holding
123 * info->lock, i.e. spin_lock(&info->lock) provided a pairing
124 * acquire memory barrier.
127 struct ext_wait_queue { /* queue of sleeping tasks */
128 struct task_struct *task;
129 struct list_head list;
130 struct msg_msg *msg; /* ptr of loaded message */
131 int state; /* one of STATE_* values */
134 struct mqueue_inode_info {
136 struct inode vfs_inode;
137 wait_queue_head_t wait_q;
139 struct rb_root msg_tree;
140 struct rb_node *msg_tree_rightmost;
141 struct posix_msg_tree_node *node_cache;
144 struct sigevent notify;
145 struct pid *notify_owner;
146 u32 notify_self_exec_id;
147 struct user_namespace *notify_user_ns;
148 struct ucounts *ucounts; /* user who created, for accounting */
149 struct sock *notify_sock;
150 struct sk_buff *notify_cookie;
152 /* for tasks waiting for free space and messages, respectively */
153 struct ext_wait_queue e_wait_q[2];
155 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
158 static struct file_system_type mqueue_fs_type;
159 static const struct inode_operations mqueue_dir_inode_operations;
160 static const struct file_operations mqueue_file_operations;
161 static const struct super_operations mqueue_super_ops;
162 static const struct fs_context_operations mqueue_fs_context_ops;
163 static void remove_notification(struct mqueue_inode_info *info);
165 static struct kmem_cache *mqueue_inode_cachep;
167 static struct ctl_table_header *mq_sysctl_table;
169 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
171 return container_of(inode, struct mqueue_inode_info, vfs_inode);
175 * This routine should be called with the mq_lock held.
177 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
179 return get_ipc_ns(inode->i_sb->s_fs_info);
182 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
184 struct ipc_namespace *ns;
187 ns = __get_ns_from_inode(inode);
188 spin_unlock(&mq_lock);
192 /* Auxiliary functions to manipulate messages' list */
193 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
195 struct rb_node **p, *parent = NULL;
196 struct posix_msg_tree_node *leaf;
197 bool rightmost = true;
199 p = &info->msg_tree.rb_node;
202 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
204 if (likely(leaf->priority == msg->m_type))
206 else if (msg->m_type < leaf->priority) {
212 if (info->node_cache) {
213 leaf = info->node_cache;
214 info->node_cache = NULL;
216 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
219 INIT_LIST_HEAD(&leaf->msg_list);
221 leaf->priority = msg->m_type;
224 info->msg_tree_rightmost = &leaf->rb_node;
226 rb_link_node(&leaf->rb_node, parent, p);
227 rb_insert_color(&leaf->rb_node, &info->msg_tree);
229 info->attr.mq_curmsgs++;
230 info->qsize += msg->m_ts;
231 list_add_tail(&msg->m_list, &leaf->msg_list);
235 static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
236 struct mqueue_inode_info *info)
238 struct rb_node *node = &leaf->rb_node;
240 if (info->msg_tree_rightmost == node)
241 info->msg_tree_rightmost = rb_prev(node);
243 rb_erase(node, &info->msg_tree);
244 if (info->node_cache)
247 info->node_cache = leaf;
250 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
252 struct rb_node *parent = NULL;
253 struct posix_msg_tree_node *leaf;
258 * During insert, low priorities go to the left and high to the
259 * right. On receive, we want the highest priorities first, so
260 * walk all the way to the right.
262 parent = info->msg_tree_rightmost;
264 if (info->attr.mq_curmsgs) {
265 pr_warn_once("Inconsistency in POSIX message queue, "
266 "no tree element, but supposedly messages "
268 info->attr.mq_curmsgs = 0;
272 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
273 if (unlikely(list_empty(&leaf->msg_list))) {
274 pr_warn_once("Inconsistency in POSIX message queue, "
275 "empty leaf node but we haven't implemented "
276 "lazy leaf delete!\n");
277 msg_tree_erase(leaf, info);
280 msg = list_first_entry(&leaf->msg_list,
281 struct msg_msg, m_list);
282 list_del(&msg->m_list);
283 if (list_empty(&leaf->msg_list)) {
284 msg_tree_erase(leaf, info);
287 info->attr.mq_curmsgs--;
288 info->qsize -= msg->m_ts;
292 static struct inode *mqueue_get_inode(struct super_block *sb,
293 struct ipc_namespace *ipc_ns, umode_t mode,
294 struct mq_attr *attr)
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->ucounts = 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;
374 info->ucounts = get_ucounts(current_ucounts());
379 msgqueue = inc_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
380 if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) {
381 dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
382 spin_unlock(&mq_lock);
383 put_ucounts(info->ucounts);
384 info->ucounts = NULL;
385 /* mqueue_evict_inode() releases info->messages */
389 spin_unlock(&mq_lock);
391 } else if (S_ISDIR(mode)) {
393 /* Some things misbehave if size == 0 on a directory */
394 inode->i_size = 2 * DIRENT_SIZE;
395 inode->i_op = &mqueue_dir_inode_operations;
396 inode->i_fop = &simple_dir_operations;
406 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
409 struct ipc_namespace *ns = sb->s_fs_info;
411 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
412 sb->s_blocksize = PAGE_SIZE;
413 sb->s_blocksize_bits = PAGE_SHIFT;
414 sb->s_magic = MQUEUE_MAGIC;
415 sb->s_op = &mqueue_super_ops;
417 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
419 return PTR_ERR(inode);
421 sb->s_root = d_make_root(inode);
427 static int mqueue_get_tree(struct fs_context *fc)
429 struct mqueue_fs_context *ctx = fc->fs_private;
432 * With a newly created ipc namespace, we don't need to do a search
433 * for an ipc namespace match, but we still need to set s_fs_info.
436 fc->s_fs_info = ctx->ipc_ns;
437 return get_tree_nodev(fc, mqueue_fill_super);
439 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
442 static void mqueue_fs_context_free(struct fs_context *fc)
444 struct mqueue_fs_context *ctx = fc->fs_private;
446 put_ipc_ns(ctx->ipc_ns);
450 static int mqueue_init_fs_context(struct fs_context *fc)
452 struct mqueue_fs_context *ctx;
454 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
458 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
459 put_user_ns(fc->user_ns);
460 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
461 fc->fs_private = ctx;
462 fc->ops = &mqueue_fs_context_ops;
467 * mq_init_ns() is currently the only caller of mq_create_mount().
468 * So the ns parameter is always a newly created ipc namespace.
470 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
472 struct mqueue_fs_context *ctx;
473 struct fs_context *fc;
474 struct vfsmount *mnt;
476 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
480 ctx = fc->fs_private;
482 put_ipc_ns(ctx->ipc_ns);
483 ctx->ipc_ns = get_ipc_ns(ns);
484 put_user_ns(fc->user_ns);
485 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
492 static void init_once(void *foo)
494 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
496 inode_init_once(&p->vfs_inode);
499 static struct inode *mqueue_alloc_inode(struct super_block *sb)
501 struct mqueue_inode_info *ei;
503 ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL);
506 return &ei->vfs_inode;
509 static void mqueue_free_inode(struct inode *inode)
511 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
514 static void mqueue_evict_inode(struct inode *inode)
516 struct mqueue_inode_info *info;
517 struct ipc_namespace *ipc_ns;
518 struct msg_msg *msg, *nmsg;
523 if (S_ISDIR(inode->i_mode))
526 ipc_ns = get_ns_from_inode(inode);
527 info = MQUEUE_I(inode);
528 spin_lock(&info->lock);
529 while ((msg = msg_get(info)) != NULL)
530 list_add_tail(&msg->m_list, &tmp_msg);
531 kfree(info->node_cache);
532 spin_unlock(&info->lock);
534 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
535 list_del(&msg->m_list);
540 unsigned long mq_bytes, mq_treesize;
542 /* Total amount of bytes accounted for the mqueue */
543 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
544 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
545 sizeof(struct posix_msg_tree_node);
547 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
548 info->attr.mq_msgsize);
551 dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
553 * get_ns_from_inode() ensures that the
554 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
555 * to which we now hold a reference, or it is NULL.
556 * We can't put it here under mq_lock, though.
559 ipc_ns->mq_queues_count--;
560 spin_unlock(&mq_lock);
561 put_ucounts(info->ucounts);
562 info->ucounts = NULL;
568 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
570 struct inode *dir = dentry->d_parent->d_inode;
572 struct mq_attr *attr = arg;
574 struct ipc_namespace *ipc_ns;
577 ipc_ns = __get_ns_from_inode(dir);
583 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
584 !capable(CAP_SYS_RESOURCE)) {
588 ipc_ns->mq_queues_count++;
589 spin_unlock(&mq_lock);
591 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
593 error = PTR_ERR(inode);
595 ipc_ns->mq_queues_count--;
600 dir->i_size += DIRENT_SIZE;
601 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
603 d_instantiate(dentry, inode);
607 spin_unlock(&mq_lock);
613 static int mqueue_create(struct user_namespace *mnt_userns, struct inode *dir,
614 struct dentry *dentry, umode_t mode, bool excl)
616 return mqueue_create_attr(dentry, mode, NULL);
619 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
621 struct inode *inode = d_inode(dentry);
623 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
624 dir->i_size -= DIRENT_SIZE;
631 * This is routine for system read from queue file.
632 * To avoid mess with doing here some sort of mq_receive we allow
633 * to read only queue size & notification info (the only values
634 * that are interesting from user point of view and aren't accessible
635 * through std routines)
637 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
638 size_t count, loff_t *off)
640 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
641 char buffer[FILENT_SIZE];
644 spin_lock(&info->lock);
645 snprintf(buffer, sizeof(buffer),
646 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
648 info->notify_owner ? info->notify.sigev_notify : 0,
649 (info->notify_owner &&
650 info->notify.sigev_notify == SIGEV_SIGNAL) ?
651 info->notify.sigev_signo : 0,
652 pid_vnr(info->notify_owner));
653 spin_unlock(&info->lock);
654 buffer[sizeof(buffer)-1] = '\0';
656 ret = simple_read_from_buffer(u_data, count, off, buffer,
661 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
665 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
667 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
669 spin_lock(&info->lock);
670 if (task_tgid(current) == info->notify_owner)
671 remove_notification(info);
673 spin_unlock(&info->lock);
677 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
679 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
682 poll_wait(filp, &info->wait_q, poll_tab);
684 spin_lock(&info->lock);
685 if (info->attr.mq_curmsgs)
686 retval = EPOLLIN | EPOLLRDNORM;
688 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
689 retval |= EPOLLOUT | EPOLLWRNORM;
690 spin_unlock(&info->lock);
695 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
696 static void wq_add(struct mqueue_inode_info *info, int sr,
697 struct ext_wait_queue *ewp)
699 struct ext_wait_queue *walk;
701 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
702 if (walk->task->prio <= current->prio) {
703 list_add_tail(&ewp->list, &walk->list);
707 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
711 * Puts current task to sleep. Caller must hold queue lock. After return
715 static int wq_sleep(struct mqueue_inode_info *info, int sr,
716 ktime_t *timeout, struct ext_wait_queue *ewp)
717 __releases(&info->lock)
722 wq_add(info, sr, ewp);
725 /* memory barrier not required, we hold info->lock */
726 __set_current_state(TASK_INTERRUPTIBLE);
728 spin_unlock(&info->lock);
729 time = schedule_hrtimeout_range_clock(timeout, 0,
730 HRTIMER_MODE_ABS, CLOCK_REALTIME);
732 if (READ_ONCE(ewp->state) == STATE_READY) {
733 /* see MQ_BARRIER for purpose/pairing */
734 smp_acquire__after_ctrl_dep();
738 spin_lock(&info->lock);
740 /* we hold info->lock, so no memory barrier required */
741 if (READ_ONCE(ewp->state) == STATE_READY) {
745 if (signal_pending(current)) {
746 retval = -ERESTARTSYS;
754 list_del(&ewp->list);
756 spin_unlock(&info->lock);
762 * Returns waiting task that should be serviced first or NULL if none exists
764 static struct ext_wait_queue *wq_get_first_waiter(
765 struct mqueue_inode_info *info, int sr)
767 struct list_head *ptr;
769 ptr = info->e_wait_q[sr].list.prev;
770 if (ptr == &info->e_wait_q[sr].list)
772 return list_entry(ptr, struct ext_wait_queue, list);
776 static inline void set_cookie(struct sk_buff *skb, char code)
778 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
782 * The next function is only to split too long sys_mq_timedsend
784 static void __do_notify(struct mqueue_inode_info *info)
787 * invoked when there is registered process and there isn't process
788 * waiting synchronously for message AND state of queue changed from
789 * empty to not empty. Here we are sure that no one is waiting
791 if (info->notify_owner &&
792 info->attr.mq_curmsgs == 1) {
793 switch (info->notify.sigev_notify) {
797 struct kernel_siginfo sig_i;
798 struct task_struct *task;
800 /* do_mq_notify() accepts sigev_signo == 0, why?? */
801 if (!info->notify.sigev_signo)
804 clear_siginfo(&sig_i);
805 sig_i.si_signo = info->notify.sigev_signo;
807 sig_i.si_code = SI_MESGQ;
808 sig_i.si_value = info->notify.sigev_value;
810 /* map current pid/uid into info->owner's namespaces */
811 sig_i.si_pid = task_tgid_nr_ns(current,
812 ns_of_pid(info->notify_owner));
813 sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
816 * We can't use kill_pid_info(), this signal should
817 * bypass check_kill_permission(). It is from kernel
818 * but si_fromuser() can't know this.
819 * We do check the self_exec_id, to avoid sending
820 * signals to programs that don't expect them.
822 task = pid_task(info->notify_owner, PIDTYPE_TGID);
823 if (task && task->self_exec_id ==
824 info->notify_self_exec_id) {
825 do_send_sig_info(info->notify.sigev_signo,
826 &sig_i, task, PIDTYPE_TGID);
832 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
833 netlink_sendskb(info->notify_sock, info->notify_cookie);
836 /* after notification unregisters process */
837 put_pid(info->notify_owner);
838 put_user_ns(info->notify_user_ns);
839 info->notify_owner = NULL;
840 info->notify_user_ns = NULL;
842 wake_up(&info->wait_q);
845 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
846 struct timespec64 *ts)
848 if (get_timespec64(ts, u_abs_timeout))
850 if (!timespec64_valid(ts))
855 static void remove_notification(struct mqueue_inode_info *info)
857 if (info->notify_owner != NULL &&
858 info->notify.sigev_notify == SIGEV_THREAD) {
859 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
860 netlink_sendskb(info->notify_sock, info->notify_cookie);
862 put_pid(info->notify_owner);
863 put_user_ns(info->notify_user_ns);
864 info->notify_owner = NULL;
865 info->notify_user_ns = NULL;
868 static int prepare_open(struct dentry *dentry, int oflag, int ro,
869 umode_t mode, struct filename *name,
870 struct mq_attr *attr)
872 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
873 MAY_READ | MAY_WRITE };
876 if (d_really_is_negative(dentry)) {
877 if (!(oflag & O_CREAT))
881 audit_inode_parent_hidden(name, dentry->d_parent);
882 return vfs_mkobj(dentry, mode & ~current_umask(),
883 mqueue_create_attr, attr);
885 /* it already existed */
886 audit_inode(name, dentry, 0);
887 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
889 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
891 acc = oflag2acc[oflag & O_ACCMODE];
892 return inode_permission(&init_user_ns, d_inode(dentry), acc);
895 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
896 struct mq_attr *attr)
898 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
899 struct dentry *root = mnt->mnt_root;
900 struct filename *name;
905 audit_mq_open(oflag, mode, attr);
907 if (IS_ERR(name = getname(u_name)))
908 return PTR_ERR(name);
910 fd = get_unused_fd_flags(O_CLOEXEC);
914 ro = mnt_want_write(mnt); /* we'll drop it in any case */
915 inode_lock(d_inode(root));
916 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
917 if (IS_ERR(path.dentry)) {
918 error = PTR_ERR(path.dentry);
921 path.mnt = mntget(mnt);
922 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
924 struct file *file = dentry_open(&path, oflag, current_cred());
926 fd_install(fd, file);
928 error = PTR_ERR(file);
936 inode_unlock(d_inode(root));
944 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
945 struct mq_attr __user *, u_attr)
948 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
951 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
954 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
957 struct filename *name;
958 struct dentry *dentry;
959 struct inode *inode = NULL;
960 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
961 struct vfsmount *mnt = ipc_ns->mq_mnt;
963 name = getname(u_name);
965 return PTR_ERR(name);
967 audit_inode_parent_hidden(name, mnt->mnt_root);
968 err = mnt_want_write(mnt);
971 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
972 dentry = lookup_one_len(name->name, mnt->mnt_root,
974 if (IS_ERR(dentry)) {
975 err = PTR_ERR(dentry);
979 inode = d_inode(dentry);
984 err = vfs_unlink(&init_user_ns, d_inode(dentry->d_parent),
990 inode_unlock(d_inode(mnt->mnt_root));
1000 /* Pipelined send and receive functions.
1002 * If a receiver finds no waiting message, then it registers itself in the
1003 * list of waiting receivers. A sender checks that list before adding the new
1004 * message into the message array. If there is a waiting receiver, then it
1005 * bypasses the message array and directly hands the message over to the
1006 * receiver. The receiver accepts the message and returns without grabbing the
1009 * - Set pointer to message.
1010 * - Queue the receiver task for later wakeup (without the info->lock).
1011 * - Update its state to STATE_READY. Now the receiver can continue.
1012 * - Wake up the process after the lock is dropped. Should the process wake up
1013 * before this wakeup (due to a timeout or a signal) it will either see
1014 * STATE_READY and continue or acquire the lock to check the state again.
1016 * The same algorithm is used for senders.
1019 static inline void __pipelined_op(struct wake_q_head *wake_q,
1020 struct mqueue_inode_info *info,
1021 struct ext_wait_queue *this)
1023 struct task_struct *task;
1025 list_del(&this->list);
1026 task = get_task_struct(this->task);
1028 /* see MQ_BARRIER for purpose/pairing */
1029 smp_store_release(&this->state, STATE_READY);
1030 wake_q_add_safe(wake_q, task);
1033 /* pipelined_send() - send a message directly to the task waiting in
1034 * sys_mq_timedreceive() (without inserting message into a queue).
1036 static inline void pipelined_send(struct wake_q_head *wake_q,
1037 struct mqueue_inode_info *info,
1038 struct msg_msg *message,
1039 struct ext_wait_queue *receiver)
1041 receiver->msg = message;
1042 __pipelined_op(wake_q, info, receiver);
1045 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1046 * gets its message and put to the queue (we have one free place for sure). */
1047 static inline void pipelined_receive(struct wake_q_head *wake_q,
1048 struct mqueue_inode_info *info)
1050 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1054 wake_up_interruptible(&info->wait_q);
1057 if (msg_insert(sender->msg, info))
1060 __pipelined_op(wake_q, info, sender);
1063 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1064 size_t msg_len, unsigned int msg_prio,
1065 struct timespec64 *ts)
1068 struct inode *inode;
1069 struct ext_wait_queue wait;
1070 struct ext_wait_queue *receiver;
1071 struct msg_msg *msg_ptr;
1072 struct mqueue_inode_info *info;
1073 ktime_t expires, *timeout = NULL;
1074 struct posix_msg_tree_node *new_leaf = NULL;
1076 DEFINE_WAKE_Q(wake_q);
1078 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1082 expires = timespec64_to_ktime(*ts);
1086 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1089 if (unlikely(!f.file)) {
1094 inode = file_inode(f.file);
1095 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1099 info = MQUEUE_I(inode);
1102 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1107 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1112 /* First try to allocate memory, before doing anything with
1113 * existing queues. */
1114 msg_ptr = load_msg(u_msg_ptr, msg_len);
1115 if (IS_ERR(msg_ptr)) {
1116 ret = PTR_ERR(msg_ptr);
1119 msg_ptr->m_ts = msg_len;
1120 msg_ptr->m_type = msg_prio;
1123 * msg_insert really wants us to have a valid, spare node struct so
1124 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1125 * fall back to that if necessary.
1127 if (!info->node_cache)
1128 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1130 spin_lock(&info->lock);
1132 if (!info->node_cache && new_leaf) {
1133 /* Save our speculative allocation into the cache */
1134 INIT_LIST_HEAD(&new_leaf->msg_list);
1135 info->node_cache = new_leaf;
1141 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1142 if (f.file->f_flags & O_NONBLOCK) {
1145 wait.task = current;
1146 wait.msg = (void *) msg_ptr;
1148 /* memory barrier not required, we hold info->lock */
1149 WRITE_ONCE(wait.state, STATE_NONE);
1150 ret = wq_sleep(info, SEND, timeout, &wait);
1152 * wq_sleep must be called with info->lock held, and
1153 * returns with the lock released
1158 receiver = wq_get_first_waiter(info, RECV);
1160 pipelined_send(&wake_q, info, msg_ptr, receiver);
1162 /* adds message to the queue */
1163 ret = msg_insert(msg_ptr, info);
1168 inode->i_atime = inode->i_mtime = inode->i_ctime =
1169 current_time(inode);
1172 spin_unlock(&info->lock);
1183 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1184 size_t msg_len, unsigned int __user *u_msg_prio,
1185 struct timespec64 *ts)
1188 struct msg_msg *msg_ptr;
1190 struct inode *inode;
1191 struct mqueue_inode_info *info;
1192 struct ext_wait_queue wait;
1193 ktime_t expires, *timeout = NULL;
1194 struct posix_msg_tree_node *new_leaf = NULL;
1197 expires = timespec64_to_ktime(*ts);
1201 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1204 if (unlikely(!f.file)) {
1209 inode = file_inode(f.file);
1210 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1214 info = MQUEUE_I(inode);
1217 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1222 /* checks if buffer is big enough */
1223 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1229 * msg_insert really wants us to have a valid, spare node struct so
1230 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1231 * fall back to that if necessary.
1233 if (!info->node_cache)
1234 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1236 spin_lock(&info->lock);
1238 if (!info->node_cache && new_leaf) {
1239 /* Save our speculative allocation into the cache */
1240 INIT_LIST_HEAD(&new_leaf->msg_list);
1241 info->node_cache = new_leaf;
1246 if (info->attr.mq_curmsgs == 0) {
1247 if (f.file->f_flags & O_NONBLOCK) {
1248 spin_unlock(&info->lock);
1251 wait.task = current;
1253 /* memory barrier not required, we hold info->lock */
1254 WRITE_ONCE(wait.state, STATE_NONE);
1255 ret = wq_sleep(info, RECV, timeout, &wait);
1259 DEFINE_WAKE_Q(wake_q);
1261 msg_ptr = msg_get(info);
1263 inode->i_atime = inode->i_mtime = inode->i_ctime =
1264 current_time(inode);
1266 /* There is now free space in queue. */
1267 pipelined_receive(&wake_q, info);
1268 spin_unlock(&info->lock);
1273 ret = msg_ptr->m_ts;
1275 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1276 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1287 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1288 size_t, msg_len, unsigned int, msg_prio,
1289 const struct __kernel_timespec __user *, u_abs_timeout)
1291 struct timespec64 ts, *p = NULL;
1292 if (u_abs_timeout) {
1293 int res = prepare_timeout(u_abs_timeout, &ts);
1298 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1301 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1302 size_t, msg_len, unsigned int __user *, u_msg_prio,
1303 const struct __kernel_timespec __user *, u_abs_timeout)
1305 struct timespec64 ts, *p = NULL;
1306 if (u_abs_timeout) {
1307 int res = prepare_timeout(u_abs_timeout, &ts);
1312 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1316 * Notes: the case when user wants us to deregister (with NULL as pointer)
1317 * and he isn't currently owner of notification, will be silently discarded.
1318 * It isn't explicitly defined in the POSIX.
1320 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1325 struct inode *inode;
1326 struct mqueue_inode_info *info;
1329 audit_mq_notify(mqdes, notification);
1333 if (notification != NULL) {
1334 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1335 notification->sigev_notify != SIGEV_SIGNAL &&
1336 notification->sigev_notify != SIGEV_THREAD))
1338 if (notification->sigev_notify == SIGEV_SIGNAL &&
1339 !valid_signal(notification->sigev_signo)) {
1342 if (notification->sigev_notify == SIGEV_THREAD) {
1345 /* create the notify skb */
1346 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1350 if (copy_from_user(nc->data,
1351 notification->sigev_value.sival_ptr,
1352 NOTIFY_COOKIE_LEN)) {
1357 /* TODO: add a header? */
1358 skb_put(nc, NOTIFY_COOKIE_LEN);
1359 /* and attach it to the socket */
1361 f = fdget(notification->sigev_signo);
1366 sock = netlink_getsockbyfilp(f.file);
1369 ret = PTR_ERR(sock);
1373 timeo = MAX_SCHEDULE_TIMEOUT;
1374 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1390 inode = file_inode(f.file);
1391 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1395 info = MQUEUE_I(inode);
1398 spin_lock(&info->lock);
1399 if (notification == NULL) {
1400 if (info->notify_owner == task_tgid(current)) {
1401 remove_notification(info);
1402 inode->i_atime = inode->i_ctime = current_time(inode);
1404 } else if (info->notify_owner != NULL) {
1407 switch (notification->sigev_notify) {
1409 info->notify.sigev_notify = SIGEV_NONE;
1412 info->notify_sock = sock;
1413 info->notify_cookie = nc;
1416 info->notify.sigev_notify = SIGEV_THREAD;
1419 info->notify.sigev_signo = notification->sigev_signo;
1420 info->notify.sigev_value = notification->sigev_value;
1421 info->notify.sigev_notify = SIGEV_SIGNAL;
1422 info->notify_self_exec_id = current->self_exec_id;
1426 info->notify_owner = get_pid(task_tgid(current));
1427 info->notify_user_ns = get_user_ns(current_user_ns());
1428 inode->i_atime = inode->i_ctime = current_time(inode);
1430 spin_unlock(&info->lock);
1435 netlink_detachskb(sock, nc);
1443 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1444 const struct sigevent __user *, u_notification)
1446 struct sigevent n, *p = NULL;
1447 if (u_notification) {
1448 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1452 return do_mq_notify(mqdes, p);
1455 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1458 struct inode *inode;
1459 struct mqueue_inode_info *info;
1461 if (new && (new->mq_flags & (~O_NONBLOCK)))
1468 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1473 inode = file_inode(f.file);
1474 info = MQUEUE_I(inode);
1476 spin_lock(&info->lock);
1480 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1483 audit_mq_getsetattr(mqdes, new);
1484 spin_lock(&f.file->f_lock);
1485 if (new->mq_flags & O_NONBLOCK)
1486 f.file->f_flags |= O_NONBLOCK;
1488 f.file->f_flags &= ~O_NONBLOCK;
1489 spin_unlock(&f.file->f_lock);
1491 inode->i_atime = inode->i_ctime = current_time(inode);
1494 spin_unlock(&info->lock);
1499 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1500 const struct mq_attr __user *, u_mqstat,
1501 struct mq_attr __user *, u_omqstat)
1504 struct mq_attr mqstat, omqstat;
1505 struct mq_attr *new = NULL, *old = NULL;
1509 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1515 ret = do_mq_getsetattr(mqdes, new, old);
1519 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1524 #ifdef CONFIG_COMPAT
1526 struct compat_mq_attr {
1527 compat_long_t mq_flags; /* message queue flags */
1528 compat_long_t mq_maxmsg; /* maximum number of messages */
1529 compat_long_t mq_msgsize; /* maximum message size */
1530 compat_long_t mq_curmsgs; /* number of messages currently queued */
1531 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1534 static inline int get_compat_mq_attr(struct mq_attr *attr,
1535 const struct compat_mq_attr __user *uattr)
1537 struct compat_mq_attr v;
1539 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1542 memset(attr, 0, sizeof(*attr));
1543 attr->mq_flags = v.mq_flags;
1544 attr->mq_maxmsg = v.mq_maxmsg;
1545 attr->mq_msgsize = v.mq_msgsize;
1546 attr->mq_curmsgs = v.mq_curmsgs;
1550 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1551 struct compat_mq_attr __user *uattr)
1553 struct compat_mq_attr v;
1555 memset(&v, 0, sizeof(v));
1556 v.mq_flags = attr->mq_flags;
1557 v.mq_maxmsg = attr->mq_maxmsg;
1558 v.mq_msgsize = attr->mq_msgsize;
1559 v.mq_curmsgs = attr->mq_curmsgs;
1560 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1565 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1566 int, oflag, compat_mode_t, mode,
1567 struct compat_mq_attr __user *, u_attr)
1569 struct mq_attr attr, *p = NULL;
1570 if (u_attr && oflag & O_CREAT) {
1572 if (get_compat_mq_attr(&attr, u_attr))
1575 return do_mq_open(u_name, oflag, mode, p);
1578 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1579 const struct compat_sigevent __user *, u_notification)
1581 struct sigevent n, *p = NULL;
1582 if (u_notification) {
1583 if (get_compat_sigevent(&n, u_notification))
1585 if (n.sigev_notify == SIGEV_THREAD)
1586 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1589 return do_mq_notify(mqdes, p);
1592 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1593 const struct compat_mq_attr __user *, u_mqstat,
1594 struct compat_mq_attr __user *, u_omqstat)
1597 struct mq_attr mqstat, omqstat;
1598 struct mq_attr *new = NULL, *old = NULL;
1602 if (get_compat_mq_attr(new, u_mqstat))
1608 ret = do_mq_getsetattr(mqdes, new, old);
1612 if (put_compat_mq_attr(old, u_omqstat))
1618 #ifdef CONFIG_COMPAT_32BIT_TIME
1619 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1620 struct timespec64 *ts)
1622 if (get_old_timespec32(ts, p))
1624 if (!timespec64_valid(ts))
1629 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1630 const char __user *, u_msg_ptr,
1631 unsigned int, msg_len, unsigned int, msg_prio,
1632 const struct old_timespec32 __user *, u_abs_timeout)
1634 struct timespec64 ts, *p = NULL;
1635 if (u_abs_timeout) {
1636 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1641 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1644 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1645 char __user *, u_msg_ptr,
1646 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1647 const struct old_timespec32 __user *, u_abs_timeout)
1649 struct timespec64 ts, *p = NULL;
1650 if (u_abs_timeout) {
1651 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1656 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1660 static const struct inode_operations mqueue_dir_inode_operations = {
1661 .lookup = simple_lookup,
1662 .create = mqueue_create,
1663 .unlink = mqueue_unlink,
1666 static const struct file_operations mqueue_file_operations = {
1667 .flush = mqueue_flush_file,
1668 .poll = mqueue_poll_file,
1669 .read = mqueue_read_file,
1670 .llseek = default_llseek,
1673 static const struct super_operations mqueue_super_ops = {
1674 .alloc_inode = mqueue_alloc_inode,
1675 .free_inode = mqueue_free_inode,
1676 .evict_inode = mqueue_evict_inode,
1677 .statfs = simple_statfs,
1680 static const struct fs_context_operations mqueue_fs_context_ops = {
1681 .free = mqueue_fs_context_free,
1682 .get_tree = mqueue_get_tree,
1685 static struct file_system_type mqueue_fs_type = {
1687 .init_fs_context = mqueue_init_fs_context,
1688 .kill_sb = kill_litter_super,
1689 .fs_flags = FS_USERNS_MOUNT,
1692 int mq_init_ns(struct ipc_namespace *ns)
1696 ns->mq_queues_count = 0;
1697 ns->mq_queues_max = DFLT_QUEUESMAX;
1698 ns->mq_msg_max = DFLT_MSGMAX;
1699 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1700 ns->mq_msg_default = DFLT_MSG;
1701 ns->mq_msgsize_default = DFLT_MSGSIZE;
1703 m = mq_create_mount(ns);
1710 void mq_clear_sbinfo(struct ipc_namespace *ns)
1712 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1715 void mq_put_mnt(struct ipc_namespace *ns)
1717 kern_unmount(ns->mq_mnt);
1720 static int __init init_mqueue_fs(void)
1724 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1725 sizeof(struct mqueue_inode_info), 0,
1726 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1727 if (mqueue_inode_cachep == NULL)
1730 /* ignore failures - they are not fatal */
1731 mq_sysctl_table = mq_register_sysctl_table();
1733 error = register_filesystem(&mqueue_fs_type);
1737 spin_lock_init(&mq_lock);
1739 error = mq_init_ns(&init_ipc_ns);
1741 goto out_filesystem;
1746 unregister_filesystem(&mqueue_fs_type);
1748 if (mq_sysctl_table)
1749 unregister_sysctl_table(mq_sysctl_table);
1750 kmem_cache_destroy(mqueue_inode_cachep);
1754 device_initcall(init_mqueue_fs);