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;
66 struct ext_wait_queue { /* queue of sleeping tasks */
67 struct task_struct *task;
68 struct list_head list;
69 struct msg_msg *msg; /* ptr of loaded message */
70 int state; /* one of STATE_* values */
73 struct mqueue_inode_info {
75 struct inode vfs_inode;
76 wait_queue_head_t wait_q;
78 struct rb_root msg_tree;
79 struct rb_node *msg_tree_rightmost;
80 struct posix_msg_tree_node *node_cache;
83 struct sigevent notify;
84 struct pid *notify_owner;
85 struct user_namespace *notify_user_ns;
86 struct user_struct *user; /* user who created, for accounting */
87 struct sock *notify_sock;
88 struct sk_buff *notify_cookie;
90 /* for tasks waiting for free space and messages, respectively */
91 struct ext_wait_queue e_wait_q[2];
93 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
96 static struct file_system_type mqueue_fs_type;
97 static const struct inode_operations mqueue_dir_inode_operations;
98 static const struct file_operations mqueue_file_operations;
99 static const struct super_operations mqueue_super_ops;
100 static const struct fs_context_operations mqueue_fs_context_ops;
101 static void remove_notification(struct mqueue_inode_info *info);
103 static struct kmem_cache *mqueue_inode_cachep;
105 static struct ctl_table_header *mq_sysctl_table;
107 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
109 return container_of(inode, struct mqueue_inode_info, vfs_inode);
113 * This routine should be called with the mq_lock held.
115 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
117 return get_ipc_ns(inode->i_sb->s_fs_info);
120 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
122 struct ipc_namespace *ns;
125 ns = __get_ns_from_inode(inode);
126 spin_unlock(&mq_lock);
130 /* Auxiliary functions to manipulate messages' list */
131 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
133 struct rb_node **p, *parent = NULL;
134 struct posix_msg_tree_node *leaf;
135 bool rightmost = true;
137 p = &info->msg_tree.rb_node;
140 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
142 if (likely(leaf->priority == msg->m_type))
144 else if (msg->m_type < leaf->priority) {
150 if (info->node_cache) {
151 leaf = info->node_cache;
152 info->node_cache = NULL;
154 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
157 INIT_LIST_HEAD(&leaf->msg_list);
159 leaf->priority = msg->m_type;
162 info->msg_tree_rightmost = &leaf->rb_node;
164 rb_link_node(&leaf->rb_node, parent, p);
165 rb_insert_color(&leaf->rb_node, &info->msg_tree);
167 info->attr.mq_curmsgs++;
168 info->qsize += msg->m_ts;
169 list_add_tail(&msg->m_list, &leaf->msg_list);
173 static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
174 struct mqueue_inode_info *info)
176 struct rb_node *node = &leaf->rb_node;
178 if (info->msg_tree_rightmost == node)
179 info->msg_tree_rightmost = rb_prev(node);
181 rb_erase(node, &info->msg_tree);
182 if (info->node_cache) {
185 info->node_cache = leaf;
189 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
191 struct rb_node *parent = NULL;
192 struct posix_msg_tree_node *leaf;
197 * During insert, low priorities go to the left and high to the
198 * right. On receive, we want the highest priorities first, so
199 * walk all the way to the right.
201 parent = info->msg_tree_rightmost;
203 if (info->attr.mq_curmsgs) {
204 pr_warn_once("Inconsistency in POSIX message queue, "
205 "no tree element, but supposedly messages "
207 info->attr.mq_curmsgs = 0;
211 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
212 if (unlikely(list_empty(&leaf->msg_list))) {
213 pr_warn_once("Inconsistency in POSIX message queue, "
214 "empty leaf node but we haven't implemented "
215 "lazy leaf delete!\n");
216 msg_tree_erase(leaf, info);
219 msg = list_first_entry(&leaf->msg_list,
220 struct msg_msg, m_list);
221 list_del(&msg->m_list);
222 if (list_empty(&leaf->msg_list)) {
223 msg_tree_erase(leaf, info);
226 info->attr.mq_curmsgs--;
227 info->qsize -= msg->m_ts;
231 static struct inode *mqueue_get_inode(struct super_block *sb,
232 struct ipc_namespace *ipc_ns, umode_t mode,
233 struct mq_attr *attr)
235 struct user_struct *u = current_user();
239 inode = new_inode(sb);
243 inode->i_ino = get_next_ino();
244 inode->i_mode = mode;
245 inode->i_uid = current_fsuid();
246 inode->i_gid = current_fsgid();
247 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
250 struct mqueue_inode_info *info;
251 unsigned long mq_bytes, mq_treesize;
253 inode->i_fop = &mqueue_file_operations;
254 inode->i_size = FILENT_SIZE;
255 /* mqueue specific info */
256 info = MQUEUE_I(inode);
257 spin_lock_init(&info->lock);
258 init_waitqueue_head(&info->wait_q);
259 INIT_LIST_HEAD(&info->e_wait_q[0].list);
260 INIT_LIST_HEAD(&info->e_wait_q[1].list);
261 info->notify_owner = NULL;
262 info->notify_user_ns = NULL;
264 info->user = NULL; /* set when all is ok */
265 info->msg_tree = RB_ROOT;
266 info->msg_tree_rightmost = NULL;
267 info->node_cache = NULL;
268 memset(&info->attr, 0, sizeof(info->attr));
269 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
270 ipc_ns->mq_msg_default);
271 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
272 ipc_ns->mq_msgsize_default);
274 info->attr.mq_maxmsg = attr->mq_maxmsg;
275 info->attr.mq_msgsize = attr->mq_msgsize;
278 * We used to allocate a static array of pointers and account
279 * the size of that array as well as one msg_msg struct per
280 * possible message into the queue size. That's no longer
281 * accurate as the queue is now an rbtree and will grow and
282 * shrink depending on usage patterns. We can, however, still
283 * account one msg_msg struct per message, but the nodes are
284 * allocated depending on priority usage, and most programs
285 * only use one, or a handful, of priorities. However, since
286 * this is pinned memory, we need to assume worst case, so
287 * that means the min(mq_maxmsg, max_priorities) * struct
288 * posix_msg_tree_node.
292 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
294 if (capable(CAP_SYS_RESOURCE)) {
295 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
296 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
299 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
300 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
304 /* check for overflow */
305 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
307 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
308 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
309 sizeof(struct posix_msg_tree_node);
310 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
311 if (mq_bytes + mq_treesize < mq_bytes)
313 mq_bytes += mq_treesize;
315 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
316 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
317 spin_unlock(&mq_lock);
318 /* mqueue_evict_inode() releases info->messages */
322 u->mq_bytes += mq_bytes;
323 spin_unlock(&mq_lock);
326 info->user = get_uid(u);
327 } else if (S_ISDIR(mode)) {
329 /* Some things misbehave if size == 0 on a directory */
330 inode->i_size = 2 * DIRENT_SIZE;
331 inode->i_op = &mqueue_dir_inode_operations;
332 inode->i_fop = &simple_dir_operations;
342 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
345 struct ipc_namespace *ns = sb->s_fs_info;
347 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
348 sb->s_blocksize = PAGE_SIZE;
349 sb->s_blocksize_bits = PAGE_SHIFT;
350 sb->s_magic = MQUEUE_MAGIC;
351 sb->s_op = &mqueue_super_ops;
353 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
355 return PTR_ERR(inode);
357 sb->s_root = d_make_root(inode);
363 static int mqueue_get_tree(struct fs_context *fc)
365 struct mqueue_fs_context *ctx = fc->fs_private;
367 put_user_ns(fc->user_ns);
368 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
369 fc->s_fs_info = ctx->ipc_ns;
370 return vfs_get_super(fc, vfs_get_keyed_super, mqueue_fill_super);
373 static void mqueue_fs_context_free(struct fs_context *fc)
375 struct mqueue_fs_context *ctx = fc->fs_private;
378 put_ipc_ns(ctx->ipc_ns);
382 static int mqueue_init_fs_context(struct fs_context *fc)
384 struct mqueue_fs_context *ctx;
386 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
390 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
391 fc->fs_private = ctx;
392 fc->ops = &mqueue_fs_context_ops;
396 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
398 struct mqueue_fs_context *ctx;
399 struct fs_context *fc;
400 struct vfsmount *mnt;
402 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
406 ctx = fc->fs_private;
407 put_ipc_ns(ctx->ipc_ns);
408 ctx->ipc_ns = get_ipc_ns(ns);
415 static void init_once(void *foo)
417 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
419 inode_init_once(&p->vfs_inode);
422 static struct inode *mqueue_alloc_inode(struct super_block *sb)
424 struct mqueue_inode_info *ei;
426 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
429 return &ei->vfs_inode;
432 static void mqueue_free_inode(struct inode *inode)
434 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
437 static void mqueue_evict_inode(struct inode *inode)
439 struct mqueue_inode_info *info;
440 struct user_struct *user;
441 unsigned long mq_bytes, mq_treesize;
442 struct ipc_namespace *ipc_ns;
443 struct msg_msg *msg, *nmsg;
448 if (S_ISDIR(inode->i_mode))
451 ipc_ns = get_ns_from_inode(inode);
452 info = MQUEUE_I(inode);
453 spin_lock(&info->lock);
454 while ((msg = msg_get(info)) != NULL)
455 list_add_tail(&msg->m_list, &tmp_msg);
456 kfree(info->node_cache);
457 spin_unlock(&info->lock);
459 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
460 list_del(&msg->m_list);
464 /* Total amount of bytes accounted for the mqueue */
465 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
466 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
467 sizeof(struct posix_msg_tree_node);
469 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
470 info->attr.mq_msgsize);
475 user->mq_bytes -= mq_bytes;
477 * get_ns_from_inode() ensures that the
478 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
479 * to which we now hold a reference, or it is NULL.
480 * We can't put it here under mq_lock, though.
483 ipc_ns->mq_queues_count--;
484 spin_unlock(&mq_lock);
491 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
493 struct inode *dir = dentry->d_parent->d_inode;
495 struct mq_attr *attr = arg;
497 struct ipc_namespace *ipc_ns;
500 ipc_ns = __get_ns_from_inode(dir);
506 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
507 !capable(CAP_SYS_RESOURCE)) {
511 ipc_ns->mq_queues_count++;
512 spin_unlock(&mq_lock);
514 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
516 error = PTR_ERR(inode);
518 ipc_ns->mq_queues_count--;
523 dir->i_size += DIRENT_SIZE;
524 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
526 d_instantiate(dentry, inode);
530 spin_unlock(&mq_lock);
536 static int mqueue_create(struct inode *dir, struct dentry *dentry,
537 umode_t mode, bool excl)
539 return mqueue_create_attr(dentry, mode, NULL);
542 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
544 struct inode *inode = d_inode(dentry);
546 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
547 dir->i_size -= DIRENT_SIZE;
554 * This is routine for system read from queue file.
555 * To avoid mess with doing here some sort of mq_receive we allow
556 * to read only queue size & notification info (the only values
557 * that are interesting from user point of view and aren't accessible
558 * through std routines)
560 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
561 size_t count, loff_t *off)
563 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
564 char buffer[FILENT_SIZE];
567 spin_lock(&info->lock);
568 snprintf(buffer, sizeof(buffer),
569 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
571 info->notify_owner ? info->notify.sigev_notify : 0,
572 (info->notify_owner &&
573 info->notify.sigev_notify == SIGEV_SIGNAL) ?
574 info->notify.sigev_signo : 0,
575 pid_vnr(info->notify_owner));
576 spin_unlock(&info->lock);
577 buffer[sizeof(buffer)-1] = '\0';
579 ret = simple_read_from_buffer(u_data, count, off, buffer,
584 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
588 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
590 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
592 spin_lock(&info->lock);
593 if (task_tgid(current) == info->notify_owner)
594 remove_notification(info);
596 spin_unlock(&info->lock);
600 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
602 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
605 poll_wait(filp, &info->wait_q, poll_tab);
607 spin_lock(&info->lock);
608 if (info->attr.mq_curmsgs)
609 retval = EPOLLIN | EPOLLRDNORM;
611 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
612 retval |= EPOLLOUT | EPOLLWRNORM;
613 spin_unlock(&info->lock);
618 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
619 static void wq_add(struct mqueue_inode_info *info, int sr,
620 struct ext_wait_queue *ewp)
622 struct ext_wait_queue *walk;
624 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
625 if (walk->task->prio <= current->prio) {
626 list_add_tail(&ewp->list, &walk->list);
630 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
634 * Puts current task to sleep. Caller must hold queue lock. After return
638 static int wq_sleep(struct mqueue_inode_info *info, int sr,
639 ktime_t *timeout, struct ext_wait_queue *ewp)
640 __releases(&info->lock)
645 wq_add(info, sr, ewp);
648 __set_current_state(TASK_INTERRUPTIBLE);
650 spin_unlock(&info->lock);
651 time = schedule_hrtimeout_range_clock(timeout, 0,
652 HRTIMER_MODE_ABS, CLOCK_REALTIME);
654 if (ewp->state == STATE_READY) {
658 spin_lock(&info->lock);
659 if (ewp->state == STATE_READY) {
663 if (signal_pending(current)) {
664 retval = -ERESTARTSYS;
672 list_del(&ewp->list);
674 spin_unlock(&info->lock);
680 * Returns waiting task that should be serviced first or NULL if none exists
682 static struct ext_wait_queue *wq_get_first_waiter(
683 struct mqueue_inode_info *info, int sr)
685 struct list_head *ptr;
687 ptr = info->e_wait_q[sr].list.prev;
688 if (ptr == &info->e_wait_q[sr].list)
690 return list_entry(ptr, struct ext_wait_queue, list);
694 static inline void set_cookie(struct sk_buff *skb, char code)
696 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
700 * The next function is only to split too long sys_mq_timedsend
702 static void __do_notify(struct mqueue_inode_info *info)
705 * invoked when there is registered process and there isn't process
706 * waiting synchronously for message AND state of queue changed from
707 * empty to not empty. Here we are sure that no one is waiting
709 if (info->notify_owner &&
710 info->attr.mq_curmsgs == 1) {
711 struct kernel_siginfo sig_i;
712 switch (info->notify.sigev_notify) {
718 clear_siginfo(&sig_i);
719 sig_i.si_signo = info->notify.sigev_signo;
721 sig_i.si_code = SI_MESGQ;
722 sig_i.si_value = info->notify.sigev_value;
723 /* map current pid/uid into info->owner's namespaces */
725 sig_i.si_pid = task_tgid_nr_ns(current,
726 ns_of_pid(info->notify_owner));
727 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
730 kill_pid_info(info->notify.sigev_signo,
731 &sig_i, info->notify_owner);
734 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
735 netlink_sendskb(info->notify_sock, info->notify_cookie);
738 /* after notification unregisters process */
739 put_pid(info->notify_owner);
740 put_user_ns(info->notify_user_ns);
741 info->notify_owner = NULL;
742 info->notify_user_ns = NULL;
744 wake_up(&info->wait_q);
747 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
748 struct timespec64 *ts)
750 if (get_timespec64(ts, u_abs_timeout))
752 if (!timespec64_valid(ts))
757 static void remove_notification(struct mqueue_inode_info *info)
759 if (info->notify_owner != NULL &&
760 info->notify.sigev_notify == SIGEV_THREAD) {
761 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
762 netlink_sendskb(info->notify_sock, info->notify_cookie);
764 put_pid(info->notify_owner);
765 put_user_ns(info->notify_user_ns);
766 info->notify_owner = NULL;
767 info->notify_user_ns = NULL;
770 static int prepare_open(struct dentry *dentry, int oflag, int ro,
771 umode_t mode, struct filename *name,
772 struct mq_attr *attr)
774 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
775 MAY_READ | MAY_WRITE };
778 if (d_really_is_negative(dentry)) {
779 if (!(oflag & O_CREAT))
783 audit_inode_parent_hidden(name, dentry->d_parent);
784 return vfs_mkobj(dentry, mode & ~current_umask(),
785 mqueue_create_attr, attr);
787 /* it already existed */
788 audit_inode(name, dentry, 0);
789 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
791 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
793 acc = oflag2acc[oflag & O_ACCMODE];
794 return inode_permission(d_inode(dentry), acc);
797 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
798 struct mq_attr *attr)
800 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
801 struct dentry *root = mnt->mnt_root;
802 struct filename *name;
807 audit_mq_open(oflag, mode, attr);
809 if (IS_ERR(name = getname(u_name)))
810 return PTR_ERR(name);
812 fd = get_unused_fd_flags(O_CLOEXEC);
816 ro = mnt_want_write(mnt); /* we'll drop it in any case */
817 inode_lock(d_inode(root));
818 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
819 if (IS_ERR(path.dentry)) {
820 error = PTR_ERR(path.dentry);
823 path.mnt = mntget(mnt);
824 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
826 struct file *file = dentry_open(&path, oflag, current_cred());
828 fd_install(fd, file);
830 error = PTR_ERR(file);
838 inode_unlock(d_inode(root));
846 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
847 struct mq_attr __user *, u_attr)
850 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
853 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
856 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
859 struct filename *name;
860 struct dentry *dentry;
861 struct inode *inode = NULL;
862 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
863 struct vfsmount *mnt = ipc_ns->mq_mnt;
865 name = getname(u_name);
867 return PTR_ERR(name);
869 audit_inode_parent_hidden(name, mnt->mnt_root);
870 err = mnt_want_write(mnt);
873 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
874 dentry = lookup_one_len(name->name, mnt->mnt_root,
876 if (IS_ERR(dentry)) {
877 err = PTR_ERR(dentry);
881 inode = d_inode(dentry);
886 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
891 inode_unlock(d_inode(mnt->mnt_root));
901 /* Pipelined send and receive functions.
903 * If a receiver finds no waiting message, then it registers itself in the
904 * list of waiting receivers. A sender checks that list before adding the new
905 * message into the message array. If there is a waiting receiver, then it
906 * bypasses the message array and directly hands the message over to the
907 * receiver. The receiver accepts the message and returns without grabbing the
910 * - Set pointer to message.
911 * - Queue the receiver task for later wakeup (without the info->lock).
912 * - Update its state to STATE_READY. Now the receiver can continue.
913 * - Wake up the process after the lock is dropped. Should the process wake up
914 * before this wakeup (due to a timeout or a signal) it will either see
915 * STATE_READY and continue or acquire the lock to check the state again.
917 * The same algorithm is used for senders.
920 /* pipelined_send() - send a message directly to the task waiting in
921 * sys_mq_timedreceive() (without inserting message into a queue).
923 static inline void pipelined_send(struct wake_q_head *wake_q,
924 struct mqueue_inode_info *info,
925 struct msg_msg *message,
926 struct ext_wait_queue *receiver)
928 receiver->msg = message;
929 list_del(&receiver->list);
930 wake_q_add(wake_q, receiver->task);
932 * Rely on the implicit cmpxchg barrier from wake_q_add such
933 * that we can ensure that updating receiver->state is the last
934 * write operation: As once set, the receiver can continue,
935 * and if we don't have the reference count from the wake_q,
936 * yet, at that point we can later have a use-after-free
937 * condition and bogus wakeup.
939 receiver->state = STATE_READY;
942 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
943 * gets its message and put to the queue (we have one free place for sure). */
944 static inline void pipelined_receive(struct wake_q_head *wake_q,
945 struct mqueue_inode_info *info)
947 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
951 wake_up_interruptible(&info->wait_q);
954 if (msg_insert(sender->msg, info))
957 list_del(&sender->list);
958 wake_q_add(wake_q, sender->task);
959 sender->state = STATE_READY;
962 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
963 size_t msg_len, unsigned int msg_prio,
964 struct timespec64 *ts)
968 struct ext_wait_queue wait;
969 struct ext_wait_queue *receiver;
970 struct msg_msg *msg_ptr;
971 struct mqueue_inode_info *info;
972 ktime_t expires, *timeout = NULL;
973 struct posix_msg_tree_node *new_leaf = NULL;
975 DEFINE_WAKE_Q(wake_q);
977 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
981 expires = timespec64_to_ktime(*ts);
985 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
988 if (unlikely(!f.file)) {
993 inode = file_inode(f.file);
994 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
998 info = MQUEUE_I(inode);
1001 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1006 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1011 /* First try to allocate memory, before doing anything with
1012 * existing queues. */
1013 msg_ptr = load_msg(u_msg_ptr, msg_len);
1014 if (IS_ERR(msg_ptr)) {
1015 ret = PTR_ERR(msg_ptr);
1018 msg_ptr->m_ts = msg_len;
1019 msg_ptr->m_type = msg_prio;
1022 * msg_insert really wants us to have a valid, spare node struct so
1023 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1024 * fall back to that if necessary.
1026 if (!info->node_cache)
1027 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1029 spin_lock(&info->lock);
1031 if (!info->node_cache && new_leaf) {
1032 /* Save our speculative allocation into the cache */
1033 INIT_LIST_HEAD(&new_leaf->msg_list);
1034 info->node_cache = new_leaf;
1040 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1041 if (f.file->f_flags & O_NONBLOCK) {
1044 wait.task = current;
1045 wait.msg = (void *) msg_ptr;
1046 wait.state = STATE_NONE;
1047 ret = wq_sleep(info, SEND, timeout, &wait);
1049 * wq_sleep must be called with info->lock held, and
1050 * returns with the lock released
1055 receiver = wq_get_first_waiter(info, RECV);
1057 pipelined_send(&wake_q, info, msg_ptr, receiver);
1059 /* adds message to the queue */
1060 ret = msg_insert(msg_ptr, info);
1065 inode->i_atime = inode->i_mtime = inode->i_ctime =
1066 current_time(inode);
1069 spin_unlock(&info->lock);
1080 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1081 size_t msg_len, unsigned int __user *u_msg_prio,
1082 struct timespec64 *ts)
1085 struct msg_msg *msg_ptr;
1087 struct inode *inode;
1088 struct mqueue_inode_info *info;
1089 struct ext_wait_queue wait;
1090 ktime_t expires, *timeout = NULL;
1091 struct posix_msg_tree_node *new_leaf = NULL;
1094 expires = timespec64_to_ktime(*ts);
1098 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1101 if (unlikely(!f.file)) {
1106 inode = file_inode(f.file);
1107 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1111 info = MQUEUE_I(inode);
1114 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1119 /* checks if buffer is big enough */
1120 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1126 * msg_insert really wants us to have a valid, spare node struct so
1127 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1128 * fall back to that if necessary.
1130 if (!info->node_cache)
1131 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1133 spin_lock(&info->lock);
1135 if (!info->node_cache && new_leaf) {
1136 /* Save our speculative allocation into the cache */
1137 INIT_LIST_HEAD(&new_leaf->msg_list);
1138 info->node_cache = new_leaf;
1143 if (info->attr.mq_curmsgs == 0) {
1144 if (f.file->f_flags & O_NONBLOCK) {
1145 spin_unlock(&info->lock);
1148 wait.task = current;
1149 wait.state = STATE_NONE;
1150 ret = wq_sleep(info, RECV, timeout, &wait);
1154 DEFINE_WAKE_Q(wake_q);
1156 msg_ptr = msg_get(info);
1158 inode->i_atime = inode->i_mtime = inode->i_ctime =
1159 current_time(inode);
1161 /* There is now free space in queue. */
1162 pipelined_receive(&wake_q, info);
1163 spin_unlock(&info->lock);
1168 ret = msg_ptr->m_ts;
1170 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1171 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1182 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1183 size_t, msg_len, unsigned int, msg_prio,
1184 const struct __kernel_timespec __user *, u_abs_timeout)
1186 struct timespec64 ts, *p = NULL;
1187 if (u_abs_timeout) {
1188 int res = prepare_timeout(u_abs_timeout, &ts);
1193 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1196 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1197 size_t, msg_len, unsigned int __user *, u_msg_prio,
1198 const struct __kernel_timespec __user *, u_abs_timeout)
1200 struct timespec64 ts, *p = NULL;
1201 if (u_abs_timeout) {
1202 int res = prepare_timeout(u_abs_timeout, &ts);
1207 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1211 * Notes: the case when user wants us to deregister (with NULL as pointer)
1212 * and he isn't currently owner of notification, will be silently discarded.
1213 * It isn't explicitly defined in the POSIX.
1215 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1220 struct inode *inode;
1221 struct mqueue_inode_info *info;
1224 audit_mq_notify(mqdes, notification);
1228 if (notification != NULL) {
1229 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1230 notification->sigev_notify != SIGEV_SIGNAL &&
1231 notification->sigev_notify != SIGEV_THREAD))
1233 if (notification->sigev_notify == SIGEV_SIGNAL &&
1234 !valid_signal(notification->sigev_signo)) {
1237 if (notification->sigev_notify == SIGEV_THREAD) {
1240 /* create the notify skb */
1241 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1246 if (copy_from_user(nc->data,
1247 notification->sigev_value.sival_ptr,
1248 NOTIFY_COOKIE_LEN)) {
1253 /* TODO: add a header? */
1254 skb_put(nc, NOTIFY_COOKIE_LEN);
1255 /* and attach it to the socket */
1257 f = fdget(notification->sigev_signo);
1262 sock = netlink_getsockbyfilp(f.file);
1265 ret = PTR_ERR(sock);
1270 timeo = MAX_SCHEDULE_TIMEOUT;
1271 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1290 inode = file_inode(f.file);
1291 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1295 info = MQUEUE_I(inode);
1298 spin_lock(&info->lock);
1299 if (notification == NULL) {
1300 if (info->notify_owner == task_tgid(current)) {
1301 remove_notification(info);
1302 inode->i_atime = inode->i_ctime = current_time(inode);
1304 } else if (info->notify_owner != NULL) {
1307 switch (notification->sigev_notify) {
1309 info->notify.sigev_notify = SIGEV_NONE;
1312 info->notify_sock = sock;
1313 info->notify_cookie = nc;
1316 info->notify.sigev_notify = SIGEV_THREAD;
1319 info->notify.sigev_signo = notification->sigev_signo;
1320 info->notify.sigev_value = notification->sigev_value;
1321 info->notify.sigev_notify = SIGEV_SIGNAL;
1325 info->notify_owner = get_pid(task_tgid(current));
1326 info->notify_user_ns = get_user_ns(current_user_ns());
1327 inode->i_atime = inode->i_ctime = current_time(inode);
1329 spin_unlock(&info->lock);
1334 netlink_detachskb(sock, nc);
1341 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1342 const struct sigevent __user *, u_notification)
1344 struct sigevent n, *p = NULL;
1345 if (u_notification) {
1346 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1350 return do_mq_notify(mqdes, p);
1353 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1356 struct inode *inode;
1357 struct mqueue_inode_info *info;
1359 if (new && (new->mq_flags & (~O_NONBLOCK)))
1366 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1371 inode = file_inode(f.file);
1372 info = MQUEUE_I(inode);
1374 spin_lock(&info->lock);
1378 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1381 audit_mq_getsetattr(mqdes, new);
1382 spin_lock(&f.file->f_lock);
1383 if (new->mq_flags & O_NONBLOCK)
1384 f.file->f_flags |= O_NONBLOCK;
1386 f.file->f_flags &= ~O_NONBLOCK;
1387 spin_unlock(&f.file->f_lock);
1389 inode->i_atime = inode->i_ctime = current_time(inode);
1392 spin_unlock(&info->lock);
1397 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1398 const struct mq_attr __user *, u_mqstat,
1399 struct mq_attr __user *, u_omqstat)
1402 struct mq_attr mqstat, omqstat;
1403 struct mq_attr *new = NULL, *old = NULL;
1407 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1413 ret = do_mq_getsetattr(mqdes, new, old);
1417 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1422 #ifdef CONFIG_COMPAT
1424 struct compat_mq_attr {
1425 compat_long_t mq_flags; /* message queue flags */
1426 compat_long_t mq_maxmsg; /* maximum number of messages */
1427 compat_long_t mq_msgsize; /* maximum message size */
1428 compat_long_t mq_curmsgs; /* number of messages currently queued */
1429 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1432 static inline int get_compat_mq_attr(struct mq_attr *attr,
1433 const struct compat_mq_attr __user *uattr)
1435 struct compat_mq_attr v;
1437 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1440 memset(attr, 0, sizeof(*attr));
1441 attr->mq_flags = v.mq_flags;
1442 attr->mq_maxmsg = v.mq_maxmsg;
1443 attr->mq_msgsize = v.mq_msgsize;
1444 attr->mq_curmsgs = v.mq_curmsgs;
1448 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1449 struct compat_mq_attr __user *uattr)
1451 struct compat_mq_attr v;
1453 memset(&v, 0, sizeof(v));
1454 v.mq_flags = attr->mq_flags;
1455 v.mq_maxmsg = attr->mq_maxmsg;
1456 v.mq_msgsize = attr->mq_msgsize;
1457 v.mq_curmsgs = attr->mq_curmsgs;
1458 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1463 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1464 int, oflag, compat_mode_t, mode,
1465 struct compat_mq_attr __user *, u_attr)
1467 struct mq_attr attr, *p = NULL;
1468 if (u_attr && oflag & O_CREAT) {
1470 if (get_compat_mq_attr(&attr, u_attr))
1473 return do_mq_open(u_name, oflag, mode, p);
1476 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1477 const struct compat_sigevent __user *, u_notification)
1479 struct sigevent n, *p = NULL;
1480 if (u_notification) {
1481 if (get_compat_sigevent(&n, u_notification))
1483 if (n.sigev_notify == SIGEV_THREAD)
1484 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1487 return do_mq_notify(mqdes, p);
1490 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1491 const struct compat_mq_attr __user *, u_mqstat,
1492 struct compat_mq_attr __user *, u_omqstat)
1495 struct mq_attr mqstat, omqstat;
1496 struct mq_attr *new = NULL, *old = NULL;
1500 if (get_compat_mq_attr(new, u_mqstat))
1506 ret = do_mq_getsetattr(mqdes, new, old);
1510 if (put_compat_mq_attr(old, u_omqstat))
1516 #ifdef CONFIG_COMPAT_32BIT_TIME
1517 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1518 struct timespec64 *ts)
1520 if (get_old_timespec32(ts, p))
1522 if (!timespec64_valid(ts))
1527 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1528 const char __user *, u_msg_ptr,
1529 unsigned int, msg_len, unsigned int, msg_prio,
1530 const struct old_timespec32 __user *, u_abs_timeout)
1532 struct timespec64 ts, *p = NULL;
1533 if (u_abs_timeout) {
1534 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1539 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1542 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1543 char __user *, u_msg_ptr,
1544 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1545 const struct old_timespec32 __user *, u_abs_timeout)
1547 struct timespec64 ts, *p = NULL;
1548 if (u_abs_timeout) {
1549 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1554 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1558 static const struct inode_operations mqueue_dir_inode_operations = {
1559 .lookup = simple_lookup,
1560 .create = mqueue_create,
1561 .unlink = mqueue_unlink,
1564 static const struct file_operations mqueue_file_operations = {
1565 .flush = mqueue_flush_file,
1566 .poll = mqueue_poll_file,
1567 .read = mqueue_read_file,
1568 .llseek = default_llseek,
1571 static const struct super_operations mqueue_super_ops = {
1572 .alloc_inode = mqueue_alloc_inode,
1573 .free_inode = mqueue_free_inode,
1574 .evict_inode = mqueue_evict_inode,
1575 .statfs = simple_statfs,
1578 static const struct fs_context_operations mqueue_fs_context_ops = {
1579 .free = mqueue_fs_context_free,
1580 .get_tree = mqueue_get_tree,
1583 static struct file_system_type mqueue_fs_type = {
1585 .init_fs_context = mqueue_init_fs_context,
1586 .kill_sb = kill_litter_super,
1587 .fs_flags = FS_USERNS_MOUNT,
1590 int mq_init_ns(struct ipc_namespace *ns)
1594 ns->mq_queues_count = 0;
1595 ns->mq_queues_max = DFLT_QUEUESMAX;
1596 ns->mq_msg_max = DFLT_MSGMAX;
1597 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1598 ns->mq_msg_default = DFLT_MSG;
1599 ns->mq_msgsize_default = DFLT_MSGSIZE;
1601 m = mq_create_mount(ns);
1608 void mq_clear_sbinfo(struct ipc_namespace *ns)
1610 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1613 void mq_put_mnt(struct ipc_namespace *ns)
1615 kern_unmount(ns->mq_mnt);
1618 static int __init init_mqueue_fs(void)
1622 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1623 sizeof(struct mqueue_inode_info), 0,
1624 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1625 if (mqueue_inode_cachep == NULL)
1628 /* ignore failures - they are not fatal */
1629 mq_sysctl_table = mq_register_sysctl_table();
1631 error = register_filesystem(&mqueue_fs_type);
1635 spin_lock_init(&mq_lock);
1637 error = mq_init_ns(&init_ipc_ns);
1639 goto out_filesystem;
1644 unregister_filesystem(&mqueue_fs_type);
1646 if (mq_sysctl_table)
1647 unregister_sysctl_table(mq_sysctl_table);
1648 kmem_cache_destroy(mqueue_inode_cachep);
1652 device_initcall(init_mqueue_fs);