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 posix_msg_tree_node *node_cache;
82 struct sigevent notify;
83 struct pid *notify_owner;
84 struct user_namespace *notify_user_ns;
85 struct user_struct *user; /* user who created, for accounting */
86 struct sock *notify_sock;
87 struct sk_buff *notify_cookie;
89 /* for tasks waiting for free space and messages, respectively */
90 struct ext_wait_queue e_wait_q[2];
92 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
95 static struct file_system_type mqueue_fs_type;
96 static const struct inode_operations mqueue_dir_inode_operations;
97 static const struct file_operations mqueue_file_operations;
98 static const struct super_operations mqueue_super_ops;
99 static const struct fs_context_operations mqueue_fs_context_ops;
100 static void remove_notification(struct mqueue_inode_info *info);
102 static struct kmem_cache *mqueue_inode_cachep;
104 static struct ctl_table_header *mq_sysctl_table;
106 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
108 return container_of(inode, struct mqueue_inode_info, vfs_inode);
112 * This routine should be called with the mq_lock held.
114 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
116 return get_ipc_ns(inode->i_sb->s_fs_info);
119 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
121 struct ipc_namespace *ns;
124 ns = __get_ns_from_inode(inode);
125 spin_unlock(&mq_lock);
129 /* Auxiliary functions to manipulate messages' list */
130 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
132 struct rb_node **p, *parent = NULL;
133 struct posix_msg_tree_node *leaf;
135 p = &info->msg_tree.rb_node;
138 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
140 if (likely(leaf->priority == msg->m_type))
142 else if (msg->m_type < leaf->priority)
147 if (info->node_cache) {
148 leaf = info->node_cache;
149 info->node_cache = NULL;
151 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
154 INIT_LIST_HEAD(&leaf->msg_list);
156 leaf->priority = msg->m_type;
157 rb_link_node(&leaf->rb_node, parent, p);
158 rb_insert_color(&leaf->rb_node, &info->msg_tree);
160 info->attr.mq_curmsgs++;
161 info->qsize += msg->m_ts;
162 list_add_tail(&msg->m_list, &leaf->msg_list);
166 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
168 struct rb_node **p, *parent = NULL;
169 struct posix_msg_tree_node *leaf;
173 p = &info->msg_tree.rb_node;
177 * During insert, low priorities go to the left and high to the
178 * right. On receive, we want the highest priorities first, so
179 * walk all the way to the right.
184 if (info->attr.mq_curmsgs) {
185 pr_warn_once("Inconsistency in POSIX message queue, "
186 "no tree element, but supposedly messages "
188 info->attr.mq_curmsgs = 0;
192 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
193 if (unlikely(list_empty(&leaf->msg_list))) {
194 pr_warn_once("Inconsistency in POSIX message queue, "
195 "empty leaf node but we haven't implemented "
196 "lazy leaf delete!\n");
197 rb_erase(&leaf->rb_node, &info->msg_tree);
198 if (info->node_cache) {
201 info->node_cache = leaf;
205 msg = list_first_entry(&leaf->msg_list,
206 struct msg_msg, m_list);
207 list_del(&msg->m_list);
208 if (list_empty(&leaf->msg_list)) {
209 rb_erase(&leaf->rb_node, &info->msg_tree);
210 if (info->node_cache) {
213 info->node_cache = leaf;
217 info->attr.mq_curmsgs--;
218 info->qsize -= msg->m_ts;
222 static struct inode *mqueue_get_inode(struct super_block *sb,
223 struct ipc_namespace *ipc_ns, umode_t mode,
224 struct mq_attr *attr)
226 struct user_struct *u = current_user();
230 inode = new_inode(sb);
234 inode->i_ino = get_next_ino();
235 inode->i_mode = mode;
236 inode->i_uid = current_fsuid();
237 inode->i_gid = current_fsgid();
238 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
241 struct mqueue_inode_info *info;
242 unsigned long mq_bytes, mq_treesize;
244 inode->i_fop = &mqueue_file_operations;
245 inode->i_size = FILENT_SIZE;
246 /* mqueue specific info */
247 info = MQUEUE_I(inode);
248 spin_lock_init(&info->lock);
249 init_waitqueue_head(&info->wait_q);
250 INIT_LIST_HEAD(&info->e_wait_q[0].list);
251 INIT_LIST_HEAD(&info->e_wait_q[1].list);
252 info->notify_owner = NULL;
253 info->notify_user_ns = NULL;
255 info->user = NULL; /* set when all is ok */
256 info->msg_tree = RB_ROOT;
257 info->node_cache = NULL;
258 memset(&info->attr, 0, sizeof(info->attr));
259 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
260 ipc_ns->mq_msg_default);
261 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
262 ipc_ns->mq_msgsize_default);
264 info->attr.mq_maxmsg = attr->mq_maxmsg;
265 info->attr.mq_msgsize = attr->mq_msgsize;
268 * We used to allocate a static array of pointers and account
269 * the size of that array as well as one msg_msg struct per
270 * possible message into the queue size. That's no longer
271 * accurate as the queue is now an rbtree and will grow and
272 * shrink depending on usage patterns. We can, however, still
273 * account one msg_msg struct per message, but the nodes are
274 * allocated depending on priority usage, and most programs
275 * only use one, or a handful, of priorities. However, since
276 * this is pinned memory, we need to assume worst case, so
277 * that means the min(mq_maxmsg, max_priorities) * struct
278 * posix_msg_tree_node.
282 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
284 if (capable(CAP_SYS_RESOURCE)) {
285 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
286 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
289 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
290 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
294 /* check for overflow */
295 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
297 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
298 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
299 sizeof(struct posix_msg_tree_node);
300 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
301 if (mq_bytes + mq_treesize < mq_bytes)
303 mq_bytes += mq_treesize;
305 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
306 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
307 spin_unlock(&mq_lock);
308 /* mqueue_evict_inode() releases info->messages */
312 u->mq_bytes += mq_bytes;
313 spin_unlock(&mq_lock);
316 info->user = get_uid(u);
317 } else if (S_ISDIR(mode)) {
319 /* Some things misbehave if size == 0 on a directory */
320 inode->i_size = 2 * DIRENT_SIZE;
321 inode->i_op = &mqueue_dir_inode_operations;
322 inode->i_fop = &simple_dir_operations;
332 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
335 struct ipc_namespace *ns = sb->s_fs_info;
337 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
338 sb->s_blocksize = PAGE_SIZE;
339 sb->s_blocksize_bits = PAGE_SHIFT;
340 sb->s_magic = MQUEUE_MAGIC;
341 sb->s_op = &mqueue_super_ops;
343 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
345 return PTR_ERR(inode);
347 sb->s_root = d_make_root(inode);
353 static int mqueue_get_tree(struct fs_context *fc)
355 struct mqueue_fs_context *ctx = fc->fs_private;
357 put_user_ns(fc->user_ns);
358 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
359 fc->s_fs_info = ctx->ipc_ns;
360 return vfs_get_super(fc, vfs_get_keyed_super, mqueue_fill_super);
363 static void mqueue_fs_context_free(struct fs_context *fc)
365 struct mqueue_fs_context *ctx = fc->fs_private;
368 put_ipc_ns(ctx->ipc_ns);
372 static int mqueue_init_fs_context(struct fs_context *fc)
374 struct mqueue_fs_context *ctx;
376 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
380 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
381 fc->fs_private = ctx;
382 fc->ops = &mqueue_fs_context_ops;
386 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
388 struct mqueue_fs_context *ctx;
389 struct fs_context *fc;
390 struct vfsmount *mnt;
392 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
396 ctx = fc->fs_private;
397 put_ipc_ns(ctx->ipc_ns);
398 ctx->ipc_ns = get_ipc_ns(ns);
405 static void init_once(void *foo)
407 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
409 inode_init_once(&p->vfs_inode);
412 static struct inode *mqueue_alloc_inode(struct super_block *sb)
414 struct mqueue_inode_info *ei;
416 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
419 return &ei->vfs_inode;
422 static void mqueue_free_inode(struct inode *inode)
424 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
427 static void mqueue_evict_inode(struct inode *inode)
429 struct mqueue_inode_info *info;
430 struct user_struct *user;
431 unsigned long mq_bytes, mq_treesize;
432 struct ipc_namespace *ipc_ns;
433 struct msg_msg *msg, *nmsg;
438 if (S_ISDIR(inode->i_mode))
441 ipc_ns = get_ns_from_inode(inode);
442 info = MQUEUE_I(inode);
443 spin_lock(&info->lock);
444 while ((msg = msg_get(info)) != NULL)
445 list_add_tail(&msg->m_list, &tmp_msg);
446 kfree(info->node_cache);
447 spin_unlock(&info->lock);
449 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
450 list_del(&msg->m_list);
454 /* Total amount of bytes accounted for the mqueue */
455 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
456 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
457 sizeof(struct posix_msg_tree_node);
459 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
460 info->attr.mq_msgsize);
465 user->mq_bytes -= mq_bytes;
467 * get_ns_from_inode() ensures that the
468 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
469 * to which we now hold a reference, or it is NULL.
470 * We can't put it here under mq_lock, though.
473 ipc_ns->mq_queues_count--;
474 spin_unlock(&mq_lock);
481 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
483 struct inode *dir = dentry->d_parent->d_inode;
485 struct mq_attr *attr = arg;
487 struct ipc_namespace *ipc_ns;
490 ipc_ns = __get_ns_from_inode(dir);
496 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
497 !capable(CAP_SYS_RESOURCE)) {
501 ipc_ns->mq_queues_count++;
502 spin_unlock(&mq_lock);
504 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
506 error = PTR_ERR(inode);
508 ipc_ns->mq_queues_count--;
513 dir->i_size += DIRENT_SIZE;
514 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
516 d_instantiate(dentry, inode);
520 spin_unlock(&mq_lock);
526 static int mqueue_create(struct inode *dir, struct dentry *dentry,
527 umode_t mode, bool excl)
529 return mqueue_create_attr(dentry, mode, NULL);
532 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
534 struct inode *inode = d_inode(dentry);
536 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
537 dir->i_size -= DIRENT_SIZE;
544 * This is routine for system read from queue file.
545 * To avoid mess with doing here some sort of mq_receive we allow
546 * to read only queue size & notification info (the only values
547 * that are interesting from user point of view and aren't accessible
548 * through std routines)
550 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
551 size_t count, loff_t *off)
553 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
554 char buffer[FILENT_SIZE];
557 spin_lock(&info->lock);
558 snprintf(buffer, sizeof(buffer),
559 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
561 info->notify_owner ? info->notify.sigev_notify : 0,
562 (info->notify_owner &&
563 info->notify.sigev_notify == SIGEV_SIGNAL) ?
564 info->notify.sigev_signo : 0,
565 pid_vnr(info->notify_owner));
566 spin_unlock(&info->lock);
567 buffer[sizeof(buffer)-1] = '\0';
569 ret = simple_read_from_buffer(u_data, count, off, buffer,
574 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
578 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
580 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
582 spin_lock(&info->lock);
583 if (task_tgid(current) == info->notify_owner)
584 remove_notification(info);
586 spin_unlock(&info->lock);
590 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
592 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
595 poll_wait(filp, &info->wait_q, poll_tab);
597 spin_lock(&info->lock);
598 if (info->attr.mq_curmsgs)
599 retval = EPOLLIN | EPOLLRDNORM;
601 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
602 retval |= EPOLLOUT | EPOLLWRNORM;
603 spin_unlock(&info->lock);
608 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
609 static void wq_add(struct mqueue_inode_info *info, int sr,
610 struct ext_wait_queue *ewp)
612 struct ext_wait_queue *walk;
614 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
615 if (walk->task->prio <= current->prio) {
616 list_add_tail(&ewp->list, &walk->list);
620 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
624 * Puts current task to sleep. Caller must hold queue lock. After return
628 static int wq_sleep(struct mqueue_inode_info *info, int sr,
629 ktime_t *timeout, struct ext_wait_queue *ewp)
630 __releases(&info->lock)
635 wq_add(info, sr, ewp);
638 __set_current_state(TASK_INTERRUPTIBLE);
640 spin_unlock(&info->lock);
641 time = schedule_hrtimeout_range_clock(timeout, 0,
642 HRTIMER_MODE_ABS, CLOCK_REALTIME);
644 if (ewp->state == STATE_READY) {
648 spin_lock(&info->lock);
649 if (ewp->state == STATE_READY) {
653 if (signal_pending(current)) {
654 retval = -ERESTARTSYS;
662 list_del(&ewp->list);
664 spin_unlock(&info->lock);
670 * Returns waiting task that should be serviced first or NULL if none exists
672 static struct ext_wait_queue *wq_get_first_waiter(
673 struct mqueue_inode_info *info, int sr)
675 struct list_head *ptr;
677 ptr = info->e_wait_q[sr].list.prev;
678 if (ptr == &info->e_wait_q[sr].list)
680 return list_entry(ptr, struct ext_wait_queue, list);
684 static inline void set_cookie(struct sk_buff *skb, char code)
686 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
690 * The next function is only to split too long sys_mq_timedsend
692 static void __do_notify(struct mqueue_inode_info *info)
695 * invoked when there is registered process and there isn't process
696 * waiting synchronously for message AND state of queue changed from
697 * empty to not empty. Here we are sure that no one is waiting
699 if (info->notify_owner &&
700 info->attr.mq_curmsgs == 1) {
701 struct kernel_siginfo sig_i;
702 switch (info->notify.sigev_notify) {
708 clear_siginfo(&sig_i);
709 sig_i.si_signo = info->notify.sigev_signo;
711 sig_i.si_code = SI_MESGQ;
712 sig_i.si_value = info->notify.sigev_value;
713 /* map current pid/uid into info->owner's namespaces */
715 sig_i.si_pid = task_tgid_nr_ns(current,
716 ns_of_pid(info->notify_owner));
717 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
720 kill_pid_info(info->notify.sigev_signo,
721 &sig_i, info->notify_owner);
724 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
725 netlink_sendskb(info->notify_sock, info->notify_cookie);
728 /* after notification unregisters process */
729 put_pid(info->notify_owner);
730 put_user_ns(info->notify_user_ns);
731 info->notify_owner = NULL;
732 info->notify_user_ns = NULL;
734 wake_up(&info->wait_q);
737 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
738 struct timespec64 *ts)
740 if (get_timespec64(ts, u_abs_timeout))
742 if (!timespec64_valid(ts))
747 static void remove_notification(struct mqueue_inode_info *info)
749 if (info->notify_owner != NULL &&
750 info->notify.sigev_notify == SIGEV_THREAD) {
751 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
752 netlink_sendskb(info->notify_sock, info->notify_cookie);
754 put_pid(info->notify_owner);
755 put_user_ns(info->notify_user_ns);
756 info->notify_owner = NULL;
757 info->notify_user_ns = NULL;
760 static int prepare_open(struct dentry *dentry, int oflag, int ro,
761 umode_t mode, struct filename *name,
762 struct mq_attr *attr)
764 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
765 MAY_READ | MAY_WRITE };
768 if (d_really_is_negative(dentry)) {
769 if (!(oflag & O_CREAT))
773 audit_inode_parent_hidden(name, dentry->d_parent);
774 return vfs_mkobj(dentry, mode & ~current_umask(),
775 mqueue_create_attr, attr);
777 /* it already existed */
778 audit_inode(name, dentry, 0);
779 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
781 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
783 acc = oflag2acc[oflag & O_ACCMODE];
784 return inode_permission(d_inode(dentry), acc);
787 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
788 struct mq_attr *attr)
790 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
791 struct dentry *root = mnt->mnt_root;
792 struct filename *name;
797 audit_mq_open(oflag, mode, attr);
799 if (IS_ERR(name = getname(u_name)))
800 return PTR_ERR(name);
802 fd = get_unused_fd_flags(O_CLOEXEC);
806 ro = mnt_want_write(mnt); /* we'll drop it in any case */
807 inode_lock(d_inode(root));
808 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
809 if (IS_ERR(path.dentry)) {
810 error = PTR_ERR(path.dentry);
813 path.mnt = mntget(mnt);
814 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
816 struct file *file = dentry_open(&path, oflag, current_cred());
818 fd_install(fd, file);
820 error = PTR_ERR(file);
828 inode_unlock(d_inode(root));
836 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
837 struct mq_attr __user *, u_attr)
840 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
843 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
846 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
849 struct filename *name;
850 struct dentry *dentry;
851 struct inode *inode = NULL;
852 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
853 struct vfsmount *mnt = ipc_ns->mq_mnt;
855 name = getname(u_name);
857 return PTR_ERR(name);
859 audit_inode_parent_hidden(name, mnt->mnt_root);
860 err = mnt_want_write(mnt);
863 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
864 dentry = lookup_one_len(name->name, mnt->mnt_root,
866 if (IS_ERR(dentry)) {
867 err = PTR_ERR(dentry);
871 inode = d_inode(dentry);
876 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
881 inode_unlock(d_inode(mnt->mnt_root));
891 /* Pipelined send and receive functions.
893 * If a receiver finds no waiting message, then it registers itself in the
894 * list of waiting receivers. A sender checks that list before adding the new
895 * message into the message array. If there is a waiting receiver, then it
896 * bypasses the message array and directly hands the message over to the
897 * receiver. The receiver accepts the message and returns without grabbing the
900 * - Set pointer to message.
901 * - Queue the receiver task for later wakeup (without the info->lock).
902 * - Update its state to STATE_READY. Now the receiver can continue.
903 * - Wake up the process after the lock is dropped. Should the process wake up
904 * before this wakeup (due to a timeout or a signal) it will either see
905 * STATE_READY and continue or acquire the lock to check the state again.
907 * The same algorithm is used for senders.
910 /* pipelined_send() - send a message directly to the task waiting in
911 * sys_mq_timedreceive() (without inserting message into a queue).
913 static inline void pipelined_send(struct wake_q_head *wake_q,
914 struct mqueue_inode_info *info,
915 struct msg_msg *message,
916 struct ext_wait_queue *receiver)
918 receiver->msg = message;
919 list_del(&receiver->list);
920 wake_q_add(wake_q, receiver->task);
922 * Rely on the implicit cmpxchg barrier from wake_q_add such
923 * that we can ensure that updating receiver->state is the last
924 * write operation: As once set, the receiver can continue,
925 * and if we don't have the reference count from the wake_q,
926 * yet, at that point we can later have a use-after-free
927 * condition and bogus wakeup.
929 receiver->state = STATE_READY;
932 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
933 * gets its message and put to the queue (we have one free place for sure). */
934 static inline void pipelined_receive(struct wake_q_head *wake_q,
935 struct mqueue_inode_info *info)
937 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
941 wake_up_interruptible(&info->wait_q);
944 if (msg_insert(sender->msg, info))
947 list_del(&sender->list);
948 wake_q_add(wake_q, sender->task);
949 sender->state = STATE_READY;
952 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
953 size_t msg_len, unsigned int msg_prio,
954 struct timespec64 *ts)
958 struct ext_wait_queue wait;
959 struct ext_wait_queue *receiver;
960 struct msg_msg *msg_ptr;
961 struct mqueue_inode_info *info;
962 ktime_t expires, *timeout = NULL;
963 struct posix_msg_tree_node *new_leaf = NULL;
965 DEFINE_WAKE_Q(wake_q);
967 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
971 expires = timespec64_to_ktime(*ts);
975 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
978 if (unlikely(!f.file)) {
983 inode = file_inode(f.file);
984 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
988 info = MQUEUE_I(inode);
991 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
996 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1001 /* First try to allocate memory, before doing anything with
1002 * existing queues. */
1003 msg_ptr = load_msg(u_msg_ptr, msg_len);
1004 if (IS_ERR(msg_ptr)) {
1005 ret = PTR_ERR(msg_ptr);
1008 msg_ptr->m_ts = msg_len;
1009 msg_ptr->m_type = msg_prio;
1012 * msg_insert really wants us to have a valid, spare node struct so
1013 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1014 * fall back to that if necessary.
1016 if (!info->node_cache)
1017 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1019 spin_lock(&info->lock);
1021 if (!info->node_cache && new_leaf) {
1022 /* Save our speculative allocation into the cache */
1023 INIT_LIST_HEAD(&new_leaf->msg_list);
1024 info->node_cache = new_leaf;
1030 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1031 if (f.file->f_flags & O_NONBLOCK) {
1034 wait.task = current;
1035 wait.msg = (void *) msg_ptr;
1036 wait.state = STATE_NONE;
1037 ret = wq_sleep(info, SEND, timeout, &wait);
1039 * wq_sleep must be called with info->lock held, and
1040 * returns with the lock released
1045 receiver = wq_get_first_waiter(info, RECV);
1047 pipelined_send(&wake_q, info, msg_ptr, receiver);
1049 /* adds message to the queue */
1050 ret = msg_insert(msg_ptr, info);
1055 inode->i_atime = inode->i_mtime = inode->i_ctime =
1056 current_time(inode);
1059 spin_unlock(&info->lock);
1070 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1071 size_t msg_len, unsigned int __user *u_msg_prio,
1072 struct timespec64 *ts)
1075 struct msg_msg *msg_ptr;
1077 struct inode *inode;
1078 struct mqueue_inode_info *info;
1079 struct ext_wait_queue wait;
1080 ktime_t expires, *timeout = NULL;
1081 struct posix_msg_tree_node *new_leaf = NULL;
1084 expires = timespec64_to_ktime(*ts);
1088 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1091 if (unlikely(!f.file)) {
1096 inode = file_inode(f.file);
1097 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1101 info = MQUEUE_I(inode);
1104 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1109 /* checks if buffer is big enough */
1110 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1116 * msg_insert really wants us to have a valid, spare node struct so
1117 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1118 * fall back to that if necessary.
1120 if (!info->node_cache)
1121 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1123 spin_lock(&info->lock);
1125 if (!info->node_cache && new_leaf) {
1126 /* Save our speculative allocation into the cache */
1127 INIT_LIST_HEAD(&new_leaf->msg_list);
1128 info->node_cache = new_leaf;
1133 if (info->attr.mq_curmsgs == 0) {
1134 if (f.file->f_flags & O_NONBLOCK) {
1135 spin_unlock(&info->lock);
1138 wait.task = current;
1139 wait.state = STATE_NONE;
1140 ret = wq_sleep(info, RECV, timeout, &wait);
1144 DEFINE_WAKE_Q(wake_q);
1146 msg_ptr = msg_get(info);
1148 inode->i_atime = inode->i_mtime = inode->i_ctime =
1149 current_time(inode);
1151 /* There is now free space in queue. */
1152 pipelined_receive(&wake_q, info);
1153 spin_unlock(&info->lock);
1158 ret = msg_ptr->m_ts;
1160 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1161 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1172 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1173 size_t, msg_len, unsigned int, msg_prio,
1174 const struct __kernel_timespec __user *, u_abs_timeout)
1176 struct timespec64 ts, *p = NULL;
1177 if (u_abs_timeout) {
1178 int res = prepare_timeout(u_abs_timeout, &ts);
1183 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1186 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1187 size_t, msg_len, unsigned int __user *, u_msg_prio,
1188 const struct __kernel_timespec __user *, u_abs_timeout)
1190 struct timespec64 ts, *p = NULL;
1191 if (u_abs_timeout) {
1192 int res = prepare_timeout(u_abs_timeout, &ts);
1197 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1201 * Notes: the case when user wants us to deregister (with NULL as pointer)
1202 * and he isn't currently owner of notification, will be silently discarded.
1203 * It isn't explicitly defined in the POSIX.
1205 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1210 struct inode *inode;
1211 struct mqueue_inode_info *info;
1214 audit_mq_notify(mqdes, notification);
1218 if (notification != NULL) {
1219 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1220 notification->sigev_notify != SIGEV_SIGNAL &&
1221 notification->sigev_notify != SIGEV_THREAD))
1223 if (notification->sigev_notify == SIGEV_SIGNAL &&
1224 !valid_signal(notification->sigev_signo)) {
1227 if (notification->sigev_notify == SIGEV_THREAD) {
1230 /* create the notify skb */
1231 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1236 if (copy_from_user(nc->data,
1237 notification->sigev_value.sival_ptr,
1238 NOTIFY_COOKIE_LEN)) {
1243 /* TODO: add a header? */
1244 skb_put(nc, NOTIFY_COOKIE_LEN);
1245 /* and attach it to the socket */
1247 f = fdget(notification->sigev_signo);
1252 sock = netlink_getsockbyfilp(f.file);
1255 ret = PTR_ERR(sock);
1260 timeo = MAX_SCHEDULE_TIMEOUT;
1261 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1280 inode = file_inode(f.file);
1281 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1285 info = MQUEUE_I(inode);
1288 spin_lock(&info->lock);
1289 if (notification == NULL) {
1290 if (info->notify_owner == task_tgid(current)) {
1291 remove_notification(info);
1292 inode->i_atime = inode->i_ctime = current_time(inode);
1294 } else if (info->notify_owner != NULL) {
1297 switch (notification->sigev_notify) {
1299 info->notify.sigev_notify = SIGEV_NONE;
1302 info->notify_sock = sock;
1303 info->notify_cookie = nc;
1306 info->notify.sigev_notify = SIGEV_THREAD;
1309 info->notify.sigev_signo = notification->sigev_signo;
1310 info->notify.sigev_value = notification->sigev_value;
1311 info->notify.sigev_notify = SIGEV_SIGNAL;
1315 info->notify_owner = get_pid(task_tgid(current));
1316 info->notify_user_ns = get_user_ns(current_user_ns());
1317 inode->i_atime = inode->i_ctime = current_time(inode);
1319 spin_unlock(&info->lock);
1324 netlink_detachskb(sock, nc);
1331 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1332 const struct sigevent __user *, u_notification)
1334 struct sigevent n, *p = NULL;
1335 if (u_notification) {
1336 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1340 return do_mq_notify(mqdes, p);
1343 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1346 struct inode *inode;
1347 struct mqueue_inode_info *info;
1349 if (new && (new->mq_flags & (~O_NONBLOCK)))
1356 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1361 inode = file_inode(f.file);
1362 info = MQUEUE_I(inode);
1364 spin_lock(&info->lock);
1368 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1371 audit_mq_getsetattr(mqdes, new);
1372 spin_lock(&f.file->f_lock);
1373 if (new->mq_flags & O_NONBLOCK)
1374 f.file->f_flags |= O_NONBLOCK;
1376 f.file->f_flags &= ~O_NONBLOCK;
1377 spin_unlock(&f.file->f_lock);
1379 inode->i_atime = inode->i_ctime = current_time(inode);
1382 spin_unlock(&info->lock);
1387 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1388 const struct mq_attr __user *, u_mqstat,
1389 struct mq_attr __user *, u_omqstat)
1392 struct mq_attr mqstat, omqstat;
1393 struct mq_attr *new = NULL, *old = NULL;
1397 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1403 ret = do_mq_getsetattr(mqdes, new, old);
1407 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1412 #ifdef CONFIG_COMPAT
1414 struct compat_mq_attr {
1415 compat_long_t mq_flags; /* message queue flags */
1416 compat_long_t mq_maxmsg; /* maximum number of messages */
1417 compat_long_t mq_msgsize; /* maximum message size */
1418 compat_long_t mq_curmsgs; /* number of messages currently queued */
1419 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1422 static inline int get_compat_mq_attr(struct mq_attr *attr,
1423 const struct compat_mq_attr __user *uattr)
1425 struct compat_mq_attr v;
1427 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1430 memset(attr, 0, sizeof(*attr));
1431 attr->mq_flags = v.mq_flags;
1432 attr->mq_maxmsg = v.mq_maxmsg;
1433 attr->mq_msgsize = v.mq_msgsize;
1434 attr->mq_curmsgs = v.mq_curmsgs;
1438 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1439 struct compat_mq_attr __user *uattr)
1441 struct compat_mq_attr v;
1443 memset(&v, 0, sizeof(v));
1444 v.mq_flags = attr->mq_flags;
1445 v.mq_maxmsg = attr->mq_maxmsg;
1446 v.mq_msgsize = attr->mq_msgsize;
1447 v.mq_curmsgs = attr->mq_curmsgs;
1448 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1453 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1454 int, oflag, compat_mode_t, mode,
1455 struct compat_mq_attr __user *, u_attr)
1457 struct mq_attr attr, *p = NULL;
1458 if (u_attr && oflag & O_CREAT) {
1460 if (get_compat_mq_attr(&attr, u_attr))
1463 return do_mq_open(u_name, oflag, mode, p);
1466 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1467 const struct compat_sigevent __user *, u_notification)
1469 struct sigevent n, *p = NULL;
1470 if (u_notification) {
1471 if (get_compat_sigevent(&n, u_notification))
1473 if (n.sigev_notify == SIGEV_THREAD)
1474 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1477 return do_mq_notify(mqdes, p);
1480 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1481 const struct compat_mq_attr __user *, u_mqstat,
1482 struct compat_mq_attr __user *, u_omqstat)
1485 struct mq_attr mqstat, omqstat;
1486 struct mq_attr *new = NULL, *old = NULL;
1490 if (get_compat_mq_attr(new, u_mqstat))
1496 ret = do_mq_getsetattr(mqdes, new, old);
1500 if (put_compat_mq_attr(old, u_omqstat))
1506 #ifdef CONFIG_COMPAT_32BIT_TIME
1507 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1508 struct timespec64 *ts)
1510 if (get_old_timespec32(ts, p))
1512 if (!timespec64_valid(ts))
1517 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1518 const char __user *, u_msg_ptr,
1519 unsigned int, msg_len, unsigned int, msg_prio,
1520 const struct old_timespec32 __user *, u_abs_timeout)
1522 struct timespec64 ts, *p = NULL;
1523 if (u_abs_timeout) {
1524 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1529 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1532 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1533 char __user *, u_msg_ptr,
1534 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1535 const struct old_timespec32 __user *, u_abs_timeout)
1537 struct timespec64 ts, *p = NULL;
1538 if (u_abs_timeout) {
1539 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1544 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1548 static const struct inode_operations mqueue_dir_inode_operations = {
1549 .lookup = simple_lookup,
1550 .create = mqueue_create,
1551 .unlink = mqueue_unlink,
1554 static const struct file_operations mqueue_file_operations = {
1555 .flush = mqueue_flush_file,
1556 .poll = mqueue_poll_file,
1557 .read = mqueue_read_file,
1558 .llseek = default_llseek,
1561 static const struct super_operations mqueue_super_ops = {
1562 .alloc_inode = mqueue_alloc_inode,
1563 .free_inode = mqueue_free_inode,
1564 .evict_inode = mqueue_evict_inode,
1565 .statfs = simple_statfs,
1568 static const struct fs_context_operations mqueue_fs_context_ops = {
1569 .free = mqueue_fs_context_free,
1570 .get_tree = mqueue_get_tree,
1573 static struct file_system_type mqueue_fs_type = {
1575 .init_fs_context = mqueue_init_fs_context,
1576 .kill_sb = kill_litter_super,
1577 .fs_flags = FS_USERNS_MOUNT,
1580 int mq_init_ns(struct ipc_namespace *ns)
1584 ns->mq_queues_count = 0;
1585 ns->mq_queues_max = DFLT_QUEUESMAX;
1586 ns->mq_msg_max = DFLT_MSGMAX;
1587 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1588 ns->mq_msg_default = DFLT_MSG;
1589 ns->mq_msgsize_default = DFLT_MSGSIZE;
1591 m = mq_create_mount(ns);
1598 void mq_clear_sbinfo(struct ipc_namespace *ns)
1600 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1603 void mq_put_mnt(struct ipc_namespace *ns)
1605 kern_unmount(ns->mq_mnt);
1608 static int __init init_mqueue_fs(void)
1612 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1613 sizeof(struct mqueue_inode_info), 0,
1614 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1615 if (mqueue_inode_cachep == NULL)
1618 /* ignore failures - they are not fatal */
1619 mq_sysctl_table = mq_register_sysctl_table();
1621 error = register_filesystem(&mqueue_fs_type);
1625 spin_lock_init(&mq_lock);
1627 error = mq_init_ns(&init_ipc_ns);
1629 goto out_filesystem;
1634 unregister_filesystem(&mqueue_fs_type);
1636 if (mq_sysctl_table)
1637 unregister_sysctl_table(mq_sysctl_table);
1638 kmem_cache_destroy(mqueue_inode_cachep);
1642 device_initcall(init_mqueue_fs);