mqueue.c revision 709a643d
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5 *                          Michal Wronski          (michal.wronski@gmail.com)
6 *
7 * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 *			    Manfred Spraul	    (manfred@colorfullife.com)
10 *
11 * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12 *
13 * This file is released under the GPL.
14 */
15
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>
42
43#include <net/sock.h>
44#include "util.h"
45
46struct mqueue_fs_context {
47	struct ipc_namespace	*ipc_ns;
48};
49
50#define MQUEUE_MAGIC	0x19800202
51#define DIRENT_SIZE	20
52#define FILENT_SIZE	80
53
54#define SEND		0
55#define RECV		1
56
57#define STATE_NONE	0
58#define STATE_READY	1
59
60struct posix_msg_tree_node {
61	struct rb_node		rb_node;
62	struct list_head	msg_list;
63	int			priority;
64};
65
66struct 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 */
71};
72
73struct mqueue_inode_info {
74	spinlock_t lock;
75	struct inode vfs_inode;
76	wait_queue_head_t wait_q;
77
78	struct rb_root msg_tree;
79	struct rb_node *msg_tree_rightmost;
80	struct posix_msg_tree_node *node_cache;
81	struct mq_attr attr;
82
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;
89
90	/* for tasks waiting for free space and messages, respectively */
91	struct ext_wait_queue e_wait_q[2];
92
93	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
94};
95
96static struct file_system_type mqueue_fs_type;
97static const struct inode_operations mqueue_dir_inode_operations;
98static const struct file_operations mqueue_file_operations;
99static const struct super_operations mqueue_super_ops;
100static const struct fs_context_operations mqueue_fs_context_ops;
101static void remove_notification(struct mqueue_inode_info *info);
102
103static struct kmem_cache *mqueue_inode_cachep;
104
105static struct ctl_table_header *mq_sysctl_table;
106
107static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
108{
109	return container_of(inode, struct mqueue_inode_info, vfs_inode);
110}
111
112/*
113 * This routine should be called with the mq_lock held.
114 */
115static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
116{
117	return get_ipc_ns(inode->i_sb->s_fs_info);
118}
119
120static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
121{
122	struct ipc_namespace *ns;
123
124	spin_lock(&mq_lock);
125	ns = __get_ns_from_inode(inode);
126	spin_unlock(&mq_lock);
127	return ns;
128}
129
130/* Auxiliary functions to manipulate messages' list */
131static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
132{
133	struct rb_node **p, *parent = NULL;
134	struct posix_msg_tree_node *leaf;
135	bool rightmost = true;
136
137	p = &info->msg_tree.rb_node;
138	while (*p) {
139		parent = *p;
140		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
141
142		if (likely(leaf->priority == msg->m_type))
143			goto insert_msg;
144		else if (msg->m_type < leaf->priority) {
145			p = &(*p)->rb_left;
146			rightmost = false;
147		} else
148			p = &(*p)->rb_right;
149	}
150	if (info->node_cache) {
151		leaf = info->node_cache;
152		info->node_cache = NULL;
153	} else {
154		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
155		if (!leaf)
156			return -ENOMEM;
157		INIT_LIST_HEAD(&leaf->msg_list);
158	}
159	leaf->priority = msg->m_type;
160
161	if (rightmost)
162		info->msg_tree_rightmost = &leaf->rb_node;
163
164	rb_link_node(&leaf->rb_node, parent, p);
165	rb_insert_color(&leaf->rb_node, &info->msg_tree);
166insert_msg:
167	info->attr.mq_curmsgs++;
168	info->qsize += msg->m_ts;
169	list_add_tail(&msg->m_list, &leaf->msg_list);
170	return 0;
171}
172
173static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
174				  struct mqueue_inode_info *info)
175{
176	struct rb_node *node = &leaf->rb_node;
177
178	if (info->msg_tree_rightmost == node)
179		info->msg_tree_rightmost = rb_prev(node);
180
181	rb_erase(node, &info->msg_tree);
182	if (info->node_cache) {
183		kfree(leaf);
184	} else {
185		info->node_cache = leaf;
186	}
187}
188
189static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
190{
191	struct rb_node *parent = NULL;
192	struct posix_msg_tree_node *leaf;
193	struct msg_msg *msg;
194
195try_again:
196	/*
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.
200	 */
201	parent = info->msg_tree_rightmost;
202	if (!parent) {
203		if (info->attr.mq_curmsgs) {
204			pr_warn_once("Inconsistency in POSIX message queue, "
205				     "no tree element, but supposedly messages "
206				     "should exist!\n");
207			info->attr.mq_curmsgs = 0;
208		}
209		return NULL;
210	}
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);
217		goto try_again;
218	} else {
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);
224		}
225	}
226	info->attr.mq_curmsgs--;
227	info->qsize -= msg->m_ts;
228	return msg;
229}
230
231static struct inode *mqueue_get_inode(struct super_block *sb,
232		struct ipc_namespace *ipc_ns, umode_t mode,
233		struct mq_attr *attr)
234{
235	struct user_struct *u = current_user();
236	struct inode *inode;
237	int ret = -ENOMEM;
238
239	inode = new_inode(sb);
240	if (!inode)
241		goto err;
242
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);
248
249	if (S_ISREG(mode)) {
250		struct mqueue_inode_info *info;
251		unsigned long mq_bytes, mq_treesize;
252
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;
263		info->qsize = 0;
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);
273		if (attr) {
274			info->attr.mq_maxmsg = attr->mq_maxmsg;
275			info->attr.mq_msgsize = attr->mq_msgsize;
276		}
277		/*
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.
289		 */
290
291		ret = -EINVAL;
292		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
293			goto out_inode;
294		if (capable(CAP_SYS_RESOURCE)) {
295			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
296			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
297				goto out_inode;
298		} else {
299			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
300					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
301				goto out_inode;
302		}
303		ret = -EOVERFLOW;
304		/* check for overflow */
305		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
306			goto out_inode;
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)
312			goto out_inode;
313		mq_bytes += mq_treesize;
314		spin_lock(&mq_lock);
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 */
319			ret = -EMFILE;
320			goto out_inode;
321		}
322		u->mq_bytes += mq_bytes;
323		spin_unlock(&mq_lock);
324
325		/* all is ok */
326		info->user = get_uid(u);
327	} else if (S_ISDIR(mode)) {
328		inc_nlink(inode);
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;
333	}
334
335	return inode;
336out_inode:
337	iput(inode);
338err:
339	return ERR_PTR(ret);
340}
341
342static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
343{
344	struct inode *inode;
345	struct ipc_namespace *ns = sb->s_fs_info;
346
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;
352
353	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
354	if (IS_ERR(inode))
355		return PTR_ERR(inode);
356
357	sb->s_root = d_make_root(inode);
358	if (!sb->s_root)
359		return -ENOMEM;
360	return 0;
361}
362
363static int mqueue_get_tree(struct fs_context *fc)
364{
365	struct mqueue_fs_context *ctx = fc->fs_private;
366
367	fc->s_fs_info = ctx->ipc_ns;
368	return vfs_get_super(fc, vfs_get_keyed_super, mqueue_fill_super);
369}
370
371static void mqueue_fs_context_free(struct fs_context *fc)
372{
373	struct mqueue_fs_context *ctx = fc->fs_private;
374
375	put_ipc_ns(ctx->ipc_ns);
376	kfree(ctx);
377}
378
379static int mqueue_init_fs_context(struct fs_context *fc)
380{
381	struct mqueue_fs_context *ctx;
382
383	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
384	if (!ctx)
385		return -ENOMEM;
386
387	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
388	put_user_ns(fc->user_ns);
389	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
390	fc->fs_private = ctx;
391	fc->ops = &mqueue_fs_context_ops;
392	return 0;
393}
394
395static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
396{
397	struct mqueue_fs_context *ctx;
398	struct fs_context *fc;
399	struct vfsmount *mnt;
400
401	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
402	if (IS_ERR(fc))
403		return ERR_CAST(fc);
404
405	ctx = fc->fs_private;
406	put_ipc_ns(ctx->ipc_ns);
407	ctx->ipc_ns = get_ipc_ns(ns);
408	put_user_ns(fc->user_ns);
409	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
410
411	mnt = fc_mount(fc);
412	put_fs_context(fc);
413	return mnt;
414}
415
416static void init_once(void *foo)
417{
418	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
419
420	inode_init_once(&p->vfs_inode);
421}
422
423static struct inode *mqueue_alloc_inode(struct super_block *sb)
424{
425	struct mqueue_inode_info *ei;
426
427	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
428	if (!ei)
429		return NULL;
430	return &ei->vfs_inode;
431}
432
433static void mqueue_free_inode(struct inode *inode)
434{
435	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
436}
437
438static void mqueue_evict_inode(struct inode *inode)
439{
440	struct mqueue_inode_info *info;
441	struct user_struct *user;
442	unsigned long mq_bytes, mq_treesize;
443	struct ipc_namespace *ipc_ns;
444	struct msg_msg *msg, *nmsg;
445	LIST_HEAD(tmp_msg);
446
447	clear_inode(inode);
448
449	if (S_ISDIR(inode->i_mode))
450		return;
451
452	ipc_ns = get_ns_from_inode(inode);
453	info = MQUEUE_I(inode);
454	spin_lock(&info->lock);
455	while ((msg = msg_get(info)) != NULL)
456		list_add_tail(&msg->m_list, &tmp_msg);
457	kfree(info->node_cache);
458	spin_unlock(&info->lock);
459
460	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
461		list_del(&msg->m_list);
462		free_msg(msg);
463	}
464
465	/* Total amount of bytes accounted for the mqueue */
466	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
467		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
468		sizeof(struct posix_msg_tree_node);
469
470	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
471				  info->attr.mq_msgsize);
472
473	user = info->user;
474	if (user) {
475		spin_lock(&mq_lock);
476		user->mq_bytes -= mq_bytes;
477		/*
478		 * get_ns_from_inode() ensures that the
479		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
480		 * to which we now hold a reference, or it is NULL.
481		 * We can't put it here under mq_lock, though.
482		 */
483		if (ipc_ns)
484			ipc_ns->mq_queues_count--;
485		spin_unlock(&mq_lock);
486		free_uid(user);
487	}
488	if (ipc_ns)
489		put_ipc_ns(ipc_ns);
490}
491
492static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
493{
494	struct inode *dir = dentry->d_parent->d_inode;
495	struct inode *inode;
496	struct mq_attr *attr = arg;
497	int error;
498	struct ipc_namespace *ipc_ns;
499
500	spin_lock(&mq_lock);
501	ipc_ns = __get_ns_from_inode(dir);
502	if (!ipc_ns) {
503		error = -EACCES;
504		goto out_unlock;
505	}
506
507	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
508	    !capable(CAP_SYS_RESOURCE)) {
509		error = -ENOSPC;
510		goto out_unlock;
511	}
512	ipc_ns->mq_queues_count++;
513	spin_unlock(&mq_lock);
514
515	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
516	if (IS_ERR(inode)) {
517		error = PTR_ERR(inode);
518		spin_lock(&mq_lock);
519		ipc_ns->mq_queues_count--;
520		goto out_unlock;
521	}
522
523	put_ipc_ns(ipc_ns);
524	dir->i_size += DIRENT_SIZE;
525	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
526
527	d_instantiate(dentry, inode);
528	dget(dentry);
529	return 0;
530out_unlock:
531	spin_unlock(&mq_lock);
532	if (ipc_ns)
533		put_ipc_ns(ipc_ns);
534	return error;
535}
536
537static int mqueue_create(struct inode *dir, struct dentry *dentry,
538				umode_t mode, bool excl)
539{
540	return mqueue_create_attr(dentry, mode, NULL);
541}
542
543static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
544{
545	struct inode *inode = d_inode(dentry);
546
547	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
548	dir->i_size -= DIRENT_SIZE;
549	drop_nlink(inode);
550	dput(dentry);
551	return 0;
552}
553
554/*
555*	This is routine for system read from queue file.
556*	To avoid mess with doing here some sort of mq_receive we allow
557*	to read only queue size & notification info (the only values
558*	that are interesting from user point of view and aren't accessible
559*	through std routines)
560*/
561static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
562				size_t count, loff_t *off)
563{
564	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
565	char buffer[FILENT_SIZE];
566	ssize_t ret;
567
568	spin_lock(&info->lock);
569	snprintf(buffer, sizeof(buffer),
570			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
571			info->qsize,
572			info->notify_owner ? info->notify.sigev_notify : 0,
573			(info->notify_owner &&
574			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
575				info->notify.sigev_signo : 0,
576			pid_vnr(info->notify_owner));
577	spin_unlock(&info->lock);
578	buffer[sizeof(buffer)-1] = '\0';
579
580	ret = simple_read_from_buffer(u_data, count, off, buffer,
581				strlen(buffer));
582	if (ret <= 0)
583		return ret;
584
585	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
586	return ret;
587}
588
589static int mqueue_flush_file(struct file *filp, fl_owner_t id)
590{
591	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
592
593	spin_lock(&info->lock);
594	if (task_tgid(current) == info->notify_owner)
595		remove_notification(info);
596
597	spin_unlock(&info->lock);
598	return 0;
599}
600
601static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
602{
603	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
604	__poll_t retval = 0;
605
606	poll_wait(filp, &info->wait_q, poll_tab);
607
608	spin_lock(&info->lock);
609	if (info->attr.mq_curmsgs)
610		retval = EPOLLIN | EPOLLRDNORM;
611
612	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
613		retval |= EPOLLOUT | EPOLLWRNORM;
614	spin_unlock(&info->lock);
615
616	return retval;
617}
618
619/* Adds current to info->e_wait_q[sr] before element with smaller prio */
620static void wq_add(struct mqueue_inode_info *info, int sr,
621			struct ext_wait_queue *ewp)
622{
623	struct ext_wait_queue *walk;
624
625	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
626		if (walk->task->prio <= current->prio) {
627			list_add_tail(&ewp->list, &walk->list);
628			return;
629		}
630	}
631	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
632}
633
634/*
635 * Puts current task to sleep. Caller must hold queue lock. After return
636 * lock isn't held.
637 * sr: SEND or RECV
638 */
639static int wq_sleep(struct mqueue_inode_info *info, int sr,
640		    ktime_t *timeout, struct ext_wait_queue *ewp)
641	__releases(&info->lock)
642{
643	int retval;
644	signed long time;
645
646	wq_add(info, sr, ewp);
647
648	for (;;) {
649		__set_current_state(TASK_INTERRUPTIBLE);
650
651		spin_unlock(&info->lock);
652		time = schedule_hrtimeout_range_clock(timeout, 0,
653			HRTIMER_MODE_ABS, CLOCK_REALTIME);
654
655		if (ewp->state == STATE_READY) {
656			retval = 0;
657			goto out;
658		}
659		spin_lock(&info->lock);
660		if (ewp->state == STATE_READY) {
661			retval = 0;
662			goto out_unlock;
663		}
664		if (signal_pending(current)) {
665			retval = -ERESTARTSYS;
666			break;
667		}
668		if (time == 0) {
669			retval = -ETIMEDOUT;
670			break;
671		}
672	}
673	list_del(&ewp->list);
674out_unlock:
675	spin_unlock(&info->lock);
676out:
677	return retval;
678}
679
680/*
681 * Returns waiting task that should be serviced first or NULL if none exists
682 */
683static struct ext_wait_queue *wq_get_first_waiter(
684		struct mqueue_inode_info *info, int sr)
685{
686	struct list_head *ptr;
687
688	ptr = info->e_wait_q[sr].list.prev;
689	if (ptr == &info->e_wait_q[sr].list)
690		return NULL;
691	return list_entry(ptr, struct ext_wait_queue, list);
692}
693
694
695static inline void set_cookie(struct sk_buff *skb, char code)
696{
697	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
698}
699
700/*
701 * The next function is only to split too long sys_mq_timedsend
702 */
703static void __do_notify(struct mqueue_inode_info *info)
704{
705	/* notification
706	 * invoked when there is registered process and there isn't process
707	 * waiting synchronously for message AND state of queue changed from
708	 * empty to not empty. Here we are sure that no one is waiting
709	 * synchronously. */
710	if (info->notify_owner &&
711	    info->attr.mq_curmsgs == 1) {
712		struct kernel_siginfo sig_i;
713		switch (info->notify.sigev_notify) {
714		case SIGEV_NONE:
715			break;
716		case SIGEV_SIGNAL:
717			/* sends signal */
718
719			clear_siginfo(&sig_i);
720			sig_i.si_signo = info->notify.sigev_signo;
721			sig_i.si_errno = 0;
722			sig_i.si_code = SI_MESGQ;
723			sig_i.si_value = info->notify.sigev_value;
724			/* map current pid/uid into info->owner's namespaces */
725			rcu_read_lock();
726			sig_i.si_pid = task_tgid_nr_ns(current,
727						ns_of_pid(info->notify_owner));
728			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
729			rcu_read_unlock();
730
731			kill_pid_info(info->notify.sigev_signo,
732				      &sig_i, info->notify_owner);
733			break;
734		case SIGEV_THREAD:
735			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
736			netlink_sendskb(info->notify_sock, info->notify_cookie);
737			break;
738		}
739		/* after notification unregisters process */
740		put_pid(info->notify_owner);
741		put_user_ns(info->notify_user_ns);
742		info->notify_owner = NULL;
743		info->notify_user_ns = NULL;
744	}
745	wake_up(&info->wait_q);
746}
747
748static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
749			   struct timespec64 *ts)
750{
751	if (get_timespec64(ts, u_abs_timeout))
752		return -EFAULT;
753	if (!timespec64_valid(ts))
754		return -EINVAL;
755	return 0;
756}
757
758static void remove_notification(struct mqueue_inode_info *info)
759{
760	if (info->notify_owner != NULL &&
761	    info->notify.sigev_notify == SIGEV_THREAD) {
762		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
763		netlink_sendskb(info->notify_sock, info->notify_cookie);
764	}
765	put_pid(info->notify_owner);
766	put_user_ns(info->notify_user_ns);
767	info->notify_owner = NULL;
768	info->notify_user_ns = NULL;
769}
770
771static int prepare_open(struct dentry *dentry, int oflag, int ro,
772			umode_t mode, struct filename *name,
773			struct mq_attr *attr)
774{
775	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
776						  MAY_READ | MAY_WRITE };
777	int acc;
778
779	if (d_really_is_negative(dentry)) {
780		if (!(oflag & O_CREAT))
781			return -ENOENT;
782		if (ro)
783			return ro;
784		audit_inode_parent_hidden(name, dentry->d_parent);
785		return vfs_mkobj(dentry, mode & ~current_umask(),
786				  mqueue_create_attr, attr);
787	}
788	/* it already existed */
789	audit_inode(name, dentry, 0);
790	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
791		return -EEXIST;
792	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
793		return -EINVAL;
794	acc = oflag2acc[oflag & O_ACCMODE];
795	return inode_permission(d_inode(dentry), acc);
796}
797
798static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
799		      struct mq_attr *attr)
800{
801	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
802	struct dentry *root = mnt->mnt_root;
803	struct filename *name;
804	struct path path;
805	int fd, error;
806	int ro;
807
808	audit_mq_open(oflag, mode, attr);
809
810	if (IS_ERR(name = getname(u_name)))
811		return PTR_ERR(name);
812
813	fd = get_unused_fd_flags(O_CLOEXEC);
814	if (fd < 0)
815		goto out_putname;
816
817	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
818	inode_lock(d_inode(root));
819	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
820	if (IS_ERR(path.dentry)) {
821		error = PTR_ERR(path.dentry);
822		goto out_putfd;
823	}
824	path.mnt = mntget(mnt);
825	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
826	if (!error) {
827		struct file *file = dentry_open(&path, oflag, current_cred());
828		if (!IS_ERR(file))
829			fd_install(fd, file);
830		else
831			error = PTR_ERR(file);
832	}
833	path_put(&path);
834out_putfd:
835	if (error) {
836		put_unused_fd(fd);
837		fd = error;
838	}
839	inode_unlock(d_inode(root));
840	if (!ro)
841		mnt_drop_write(mnt);
842out_putname:
843	putname(name);
844	return fd;
845}
846
847SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
848		struct mq_attr __user *, u_attr)
849{
850	struct mq_attr attr;
851	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
852		return -EFAULT;
853
854	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
855}
856
857SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
858{
859	int err;
860	struct filename *name;
861	struct dentry *dentry;
862	struct inode *inode = NULL;
863	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
864	struct vfsmount *mnt = ipc_ns->mq_mnt;
865
866	name = getname(u_name);
867	if (IS_ERR(name))
868		return PTR_ERR(name);
869
870	audit_inode_parent_hidden(name, mnt->mnt_root);
871	err = mnt_want_write(mnt);
872	if (err)
873		goto out_name;
874	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
875	dentry = lookup_one_len(name->name, mnt->mnt_root,
876				strlen(name->name));
877	if (IS_ERR(dentry)) {
878		err = PTR_ERR(dentry);
879		goto out_unlock;
880	}
881
882	inode = d_inode(dentry);
883	if (!inode) {
884		err = -ENOENT;
885	} else {
886		ihold(inode);
887		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
888	}
889	dput(dentry);
890
891out_unlock:
892	inode_unlock(d_inode(mnt->mnt_root));
893	if (inode)
894		iput(inode);
895	mnt_drop_write(mnt);
896out_name:
897	putname(name);
898
899	return err;
900}
901
902/* Pipelined send and receive functions.
903 *
904 * If a receiver finds no waiting message, then it registers itself in the
905 * list of waiting receivers. A sender checks that list before adding the new
906 * message into the message array. If there is a waiting receiver, then it
907 * bypasses the message array and directly hands the message over to the
908 * receiver. The receiver accepts the message and returns without grabbing the
909 * queue spinlock:
910 *
911 * - Set pointer to message.
912 * - Queue the receiver task for later wakeup (without the info->lock).
913 * - Update its state to STATE_READY. Now the receiver can continue.
914 * - Wake up the process after the lock is dropped. Should the process wake up
915 *   before this wakeup (due to a timeout or a signal) it will either see
916 *   STATE_READY and continue or acquire the lock to check the state again.
917 *
918 * The same algorithm is used for senders.
919 */
920
921/* pipelined_send() - send a message directly to the task waiting in
922 * sys_mq_timedreceive() (without inserting message into a queue).
923 */
924static inline void pipelined_send(struct wake_q_head *wake_q,
925				  struct mqueue_inode_info *info,
926				  struct msg_msg *message,
927				  struct ext_wait_queue *receiver)
928{
929	receiver->msg = message;
930	list_del(&receiver->list);
931	wake_q_add(wake_q, receiver->task);
932	/*
933	 * Rely on the implicit cmpxchg barrier from wake_q_add such
934	 * that we can ensure that updating receiver->state is the last
935	 * write operation: As once set, the receiver can continue,
936	 * and if we don't have the reference count from the wake_q,
937	 * yet, at that point we can later have a use-after-free
938	 * condition and bogus wakeup.
939	 */
940	receiver->state = STATE_READY;
941}
942
943/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
944 * gets its message and put to the queue (we have one free place for sure). */
945static inline void pipelined_receive(struct wake_q_head *wake_q,
946				     struct mqueue_inode_info *info)
947{
948	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
949
950	if (!sender) {
951		/* for poll */
952		wake_up_interruptible(&info->wait_q);
953		return;
954	}
955	if (msg_insert(sender->msg, info))
956		return;
957
958	list_del(&sender->list);
959	wake_q_add(wake_q, sender->task);
960	sender->state = STATE_READY;
961}
962
963static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
964		size_t msg_len, unsigned int msg_prio,
965		struct timespec64 *ts)
966{
967	struct fd f;
968	struct inode *inode;
969	struct ext_wait_queue wait;
970	struct ext_wait_queue *receiver;
971	struct msg_msg *msg_ptr;
972	struct mqueue_inode_info *info;
973	ktime_t expires, *timeout = NULL;
974	struct posix_msg_tree_node *new_leaf = NULL;
975	int ret = 0;
976	DEFINE_WAKE_Q(wake_q);
977
978	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
979		return -EINVAL;
980
981	if (ts) {
982		expires = timespec64_to_ktime(*ts);
983		timeout = &expires;
984	}
985
986	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
987
988	f = fdget(mqdes);
989	if (unlikely(!f.file)) {
990		ret = -EBADF;
991		goto out;
992	}
993
994	inode = file_inode(f.file);
995	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
996		ret = -EBADF;
997		goto out_fput;
998	}
999	info = MQUEUE_I(inode);
1000	audit_file(f.file);
1001
1002	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1003		ret = -EBADF;
1004		goto out_fput;
1005	}
1006
1007	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1008		ret = -EMSGSIZE;
1009		goto out_fput;
1010	}
1011
1012	/* First try to allocate memory, before doing anything with
1013	 * existing queues. */
1014	msg_ptr = load_msg(u_msg_ptr, msg_len);
1015	if (IS_ERR(msg_ptr)) {
1016		ret = PTR_ERR(msg_ptr);
1017		goto out_fput;
1018	}
1019	msg_ptr->m_ts = msg_len;
1020	msg_ptr->m_type = msg_prio;
1021
1022	/*
1023	 * msg_insert really wants us to have a valid, spare node struct so
1024	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1025	 * fall back to that if necessary.
1026	 */
1027	if (!info->node_cache)
1028		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1029
1030	spin_lock(&info->lock);
1031
1032	if (!info->node_cache && new_leaf) {
1033		/* Save our speculative allocation into the cache */
1034		INIT_LIST_HEAD(&new_leaf->msg_list);
1035		info->node_cache = new_leaf;
1036		new_leaf = NULL;
1037	} else {
1038		kfree(new_leaf);
1039	}
1040
1041	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1042		if (f.file->f_flags & O_NONBLOCK) {
1043			ret = -EAGAIN;
1044		} else {
1045			wait.task = current;
1046			wait.msg = (void *) msg_ptr;
1047			wait.state = STATE_NONE;
1048			ret = wq_sleep(info, SEND, timeout, &wait);
1049			/*
1050			 * wq_sleep must be called with info->lock held, and
1051			 * returns with the lock released
1052			 */
1053			goto out_free;
1054		}
1055	} else {
1056		receiver = wq_get_first_waiter(info, RECV);
1057		if (receiver) {
1058			pipelined_send(&wake_q, info, msg_ptr, receiver);
1059		} else {
1060			/* adds message to the queue */
1061			ret = msg_insert(msg_ptr, info);
1062			if (ret)
1063				goto out_unlock;
1064			__do_notify(info);
1065		}
1066		inode->i_atime = inode->i_mtime = inode->i_ctime =
1067				current_time(inode);
1068	}
1069out_unlock:
1070	spin_unlock(&info->lock);
1071	wake_up_q(&wake_q);
1072out_free:
1073	if (ret)
1074		free_msg(msg_ptr);
1075out_fput:
1076	fdput(f);
1077out:
1078	return ret;
1079}
1080
1081static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1082		size_t msg_len, unsigned int __user *u_msg_prio,
1083		struct timespec64 *ts)
1084{
1085	ssize_t ret;
1086	struct msg_msg *msg_ptr;
1087	struct fd f;
1088	struct inode *inode;
1089	struct mqueue_inode_info *info;
1090	struct ext_wait_queue wait;
1091	ktime_t expires, *timeout = NULL;
1092	struct posix_msg_tree_node *new_leaf = NULL;
1093
1094	if (ts) {
1095		expires = timespec64_to_ktime(*ts);
1096		timeout = &expires;
1097	}
1098
1099	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1100
1101	f = fdget(mqdes);
1102	if (unlikely(!f.file)) {
1103		ret = -EBADF;
1104		goto out;
1105	}
1106
1107	inode = file_inode(f.file);
1108	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1109		ret = -EBADF;
1110		goto out_fput;
1111	}
1112	info = MQUEUE_I(inode);
1113	audit_file(f.file);
1114
1115	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1116		ret = -EBADF;
1117		goto out_fput;
1118	}
1119
1120	/* checks if buffer is big enough */
1121	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1122		ret = -EMSGSIZE;
1123		goto out_fput;
1124	}
1125
1126	/*
1127	 * msg_insert really wants us to have a valid, spare node struct so
1128	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1129	 * fall back to that if necessary.
1130	 */
1131	if (!info->node_cache)
1132		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1133
1134	spin_lock(&info->lock);
1135
1136	if (!info->node_cache && new_leaf) {
1137		/* Save our speculative allocation into the cache */
1138		INIT_LIST_HEAD(&new_leaf->msg_list);
1139		info->node_cache = new_leaf;
1140	} else {
1141		kfree(new_leaf);
1142	}
1143
1144	if (info->attr.mq_curmsgs == 0) {
1145		if (f.file->f_flags & O_NONBLOCK) {
1146			spin_unlock(&info->lock);
1147			ret = -EAGAIN;
1148		} else {
1149			wait.task = current;
1150			wait.state = STATE_NONE;
1151			ret = wq_sleep(info, RECV, timeout, &wait);
1152			msg_ptr = wait.msg;
1153		}
1154	} else {
1155		DEFINE_WAKE_Q(wake_q);
1156
1157		msg_ptr = msg_get(info);
1158
1159		inode->i_atime = inode->i_mtime = inode->i_ctime =
1160				current_time(inode);
1161
1162		/* There is now free space in queue. */
1163		pipelined_receive(&wake_q, info);
1164		spin_unlock(&info->lock);
1165		wake_up_q(&wake_q);
1166		ret = 0;
1167	}
1168	if (ret == 0) {
1169		ret = msg_ptr->m_ts;
1170
1171		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1172			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1173			ret = -EFAULT;
1174		}
1175		free_msg(msg_ptr);
1176	}
1177out_fput:
1178	fdput(f);
1179out:
1180	return ret;
1181}
1182
1183SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1184		size_t, msg_len, unsigned int, msg_prio,
1185		const struct __kernel_timespec __user *, u_abs_timeout)
1186{
1187	struct timespec64 ts, *p = NULL;
1188	if (u_abs_timeout) {
1189		int res = prepare_timeout(u_abs_timeout, &ts);
1190		if (res)
1191			return res;
1192		p = &ts;
1193	}
1194	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1195}
1196
1197SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1198		size_t, msg_len, unsigned int __user *, u_msg_prio,
1199		const struct __kernel_timespec __user *, u_abs_timeout)
1200{
1201	struct timespec64 ts, *p = NULL;
1202	if (u_abs_timeout) {
1203		int res = prepare_timeout(u_abs_timeout, &ts);
1204		if (res)
1205			return res;
1206		p = &ts;
1207	}
1208	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1209}
1210
1211/*
1212 * Notes: the case when user wants us to deregister (with NULL as pointer)
1213 * and he isn't currently owner of notification, will be silently discarded.
1214 * It isn't explicitly defined in the POSIX.
1215 */
1216static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1217{
1218	int ret;
1219	struct fd f;
1220	struct sock *sock;
1221	struct inode *inode;
1222	struct mqueue_inode_info *info;
1223	struct sk_buff *nc;
1224
1225	audit_mq_notify(mqdes, notification);
1226
1227	nc = NULL;
1228	sock = NULL;
1229	if (notification != NULL) {
1230		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1231			     notification->sigev_notify != SIGEV_SIGNAL &&
1232			     notification->sigev_notify != SIGEV_THREAD))
1233			return -EINVAL;
1234		if (notification->sigev_notify == SIGEV_SIGNAL &&
1235			!valid_signal(notification->sigev_signo)) {
1236			return -EINVAL;
1237		}
1238		if (notification->sigev_notify == SIGEV_THREAD) {
1239			long timeo;
1240
1241			/* create the notify skb */
1242			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1243			if (!nc) {
1244				ret = -ENOMEM;
1245				goto out;
1246			}
1247			if (copy_from_user(nc->data,
1248					notification->sigev_value.sival_ptr,
1249					NOTIFY_COOKIE_LEN)) {
1250				ret = -EFAULT;
1251				goto out;
1252			}
1253
1254			/* TODO: add a header? */
1255			skb_put(nc, NOTIFY_COOKIE_LEN);
1256			/* and attach it to the socket */
1257retry:
1258			f = fdget(notification->sigev_signo);
1259			if (!f.file) {
1260				ret = -EBADF;
1261				goto out;
1262			}
1263			sock = netlink_getsockbyfilp(f.file);
1264			fdput(f);
1265			if (IS_ERR(sock)) {
1266				ret = PTR_ERR(sock);
1267				sock = NULL;
1268				goto out;
1269			}
1270
1271			timeo = MAX_SCHEDULE_TIMEOUT;
1272			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1273			if (ret == 1) {
1274				sock = NULL;
1275				goto retry;
1276			}
1277			if (ret) {
1278				sock = NULL;
1279				nc = NULL;
1280				goto out;
1281			}
1282		}
1283	}
1284
1285	f = fdget(mqdes);
1286	if (!f.file) {
1287		ret = -EBADF;
1288		goto out;
1289	}
1290
1291	inode = file_inode(f.file);
1292	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1293		ret = -EBADF;
1294		goto out_fput;
1295	}
1296	info = MQUEUE_I(inode);
1297
1298	ret = 0;
1299	spin_lock(&info->lock);
1300	if (notification == NULL) {
1301		if (info->notify_owner == task_tgid(current)) {
1302			remove_notification(info);
1303			inode->i_atime = inode->i_ctime = current_time(inode);
1304		}
1305	} else if (info->notify_owner != NULL) {
1306		ret = -EBUSY;
1307	} else {
1308		switch (notification->sigev_notify) {
1309		case SIGEV_NONE:
1310			info->notify.sigev_notify = SIGEV_NONE;
1311			break;
1312		case SIGEV_THREAD:
1313			info->notify_sock = sock;
1314			info->notify_cookie = nc;
1315			sock = NULL;
1316			nc = NULL;
1317			info->notify.sigev_notify = SIGEV_THREAD;
1318			break;
1319		case SIGEV_SIGNAL:
1320			info->notify.sigev_signo = notification->sigev_signo;
1321			info->notify.sigev_value = notification->sigev_value;
1322			info->notify.sigev_notify = SIGEV_SIGNAL;
1323			break;
1324		}
1325
1326		info->notify_owner = get_pid(task_tgid(current));
1327		info->notify_user_ns = get_user_ns(current_user_ns());
1328		inode->i_atime = inode->i_ctime = current_time(inode);
1329	}
1330	spin_unlock(&info->lock);
1331out_fput:
1332	fdput(f);
1333out:
1334	if (sock)
1335		netlink_detachskb(sock, nc);
1336	else if (nc)
1337		dev_kfree_skb(nc);
1338
1339	return ret;
1340}
1341
1342SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1343		const struct sigevent __user *, u_notification)
1344{
1345	struct sigevent n, *p = NULL;
1346	if (u_notification) {
1347		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1348			return -EFAULT;
1349		p = &n;
1350	}
1351	return do_mq_notify(mqdes, p);
1352}
1353
1354static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1355{
1356	struct fd f;
1357	struct inode *inode;
1358	struct mqueue_inode_info *info;
1359
1360	if (new && (new->mq_flags & (~O_NONBLOCK)))
1361		return -EINVAL;
1362
1363	f = fdget(mqdes);
1364	if (!f.file)
1365		return -EBADF;
1366
1367	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1368		fdput(f);
1369		return -EBADF;
1370	}
1371
1372	inode = file_inode(f.file);
1373	info = MQUEUE_I(inode);
1374
1375	spin_lock(&info->lock);
1376
1377	if (old) {
1378		*old = info->attr;
1379		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1380	}
1381	if (new) {
1382		audit_mq_getsetattr(mqdes, new);
1383		spin_lock(&f.file->f_lock);
1384		if (new->mq_flags & O_NONBLOCK)
1385			f.file->f_flags |= O_NONBLOCK;
1386		else
1387			f.file->f_flags &= ~O_NONBLOCK;
1388		spin_unlock(&f.file->f_lock);
1389
1390		inode->i_atime = inode->i_ctime = current_time(inode);
1391	}
1392
1393	spin_unlock(&info->lock);
1394	fdput(f);
1395	return 0;
1396}
1397
1398SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1399		const struct mq_attr __user *, u_mqstat,
1400		struct mq_attr __user *, u_omqstat)
1401{
1402	int ret;
1403	struct mq_attr mqstat, omqstat;
1404	struct mq_attr *new = NULL, *old = NULL;
1405
1406	if (u_mqstat) {
1407		new = &mqstat;
1408		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1409			return -EFAULT;
1410	}
1411	if (u_omqstat)
1412		old = &omqstat;
1413
1414	ret = do_mq_getsetattr(mqdes, new, old);
1415	if (ret || !old)
1416		return ret;
1417
1418	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1419		return -EFAULT;
1420	return 0;
1421}
1422
1423#ifdef CONFIG_COMPAT
1424
1425struct compat_mq_attr {
1426	compat_long_t mq_flags;      /* message queue flags		     */
1427	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1428	compat_long_t mq_msgsize;    /* maximum message size		     */
1429	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1430	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1431};
1432
1433static inline int get_compat_mq_attr(struct mq_attr *attr,
1434			const struct compat_mq_attr __user *uattr)
1435{
1436	struct compat_mq_attr v;
1437
1438	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1439		return -EFAULT;
1440
1441	memset(attr, 0, sizeof(*attr));
1442	attr->mq_flags = v.mq_flags;
1443	attr->mq_maxmsg = v.mq_maxmsg;
1444	attr->mq_msgsize = v.mq_msgsize;
1445	attr->mq_curmsgs = v.mq_curmsgs;
1446	return 0;
1447}
1448
1449static inline int put_compat_mq_attr(const struct mq_attr *attr,
1450			struct compat_mq_attr __user *uattr)
1451{
1452	struct compat_mq_attr v;
1453
1454	memset(&v, 0, sizeof(v));
1455	v.mq_flags = attr->mq_flags;
1456	v.mq_maxmsg = attr->mq_maxmsg;
1457	v.mq_msgsize = attr->mq_msgsize;
1458	v.mq_curmsgs = attr->mq_curmsgs;
1459	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1460		return -EFAULT;
1461	return 0;
1462}
1463
1464COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1465		       int, oflag, compat_mode_t, mode,
1466		       struct compat_mq_attr __user *, u_attr)
1467{
1468	struct mq_attr attr, *p = NULL;
1469	if (u_attr && oflag & O_CREAT) {
1470		p = &attr;
1471		if (get_compat_mq_attr(&attr, u_attr))
1472			return -EFAULT;
1473	}
1474	return do_mq_open(u_name, oflag, mode, p);
1475}
1476
1477COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1478		       const struct compat_sigevent __user *, u_notification)
1479{
1480	struct sigevent n, *p = NULL;
1481	if (u_notification) {
1482		if (get_compat_sigevent(&n, u_notification))
1483			return -EFAULT;
1484		if (n.sigev_notify == SIGEV_THREAD)
1485			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1486		p = &n;
1487	}
1488	return do_mq_notify(mqdes, p);
1489}
1490
1491COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1492		       const struct compat_mq_attr __user *, u_mqstat,
1493		       struct compat_mq_attr __user *, u_omqstat)
1494{
1495	int ret;
1496	struct mq_attr mqstat, omqstat;
1497	struct mq_attr *new = NULL, *old = NULL;
1498
1499	if (u_mqstat) {
1500		new = &mqstat;
1501		if (get_compat_mq_attr(new, u_mqstat))
1502			return -EFAULT;
1503	}
1504	if (u_omqstat)
1505		old = &omqstat;
1506
1507	ret = do_mq_getsetattr(mqdes, new, old);
1508	if (ret || !old)
1509		return ret;
1510
1511	if (put_compat_mq_attr(old, u_omqstat))
1512		return -EFAULT;
1513	return 0;
1514}
1515#endif
1516
1517#ifdef CONFIG_COMPAT_32BIT_TIME
1518static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1519				   struct timespec64 *ts)
1520{
1521	if (get_old_timespec32(ts, p))
1522		return -EFAULT;
1523	if (!timespec64_valid(ts))
1524		return -EINVAL;
1525	return 0;
1526}
1527
1528SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1529		const char __user *, u_msg_ptr,
1530		unsigned int, msg_len, unsigned int, msg_prio,
1531		const struct old_timespec32 __user *, u_abs_timeout)
1532{
1533	struct timespec64 ts, *p = NULL;
1534	if (u_abs_timeout) {
1535		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1536		if (res)
1537			return res;
1538		p = &ts;
1539	}
1540	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1541}
1542
1543SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1544		char __user *, u_msg_ptr,
1545		unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1546		const struct old_timespec32 __user *, u_abs_timeout)
1547{
1548	struct timespec64 ts, *p = NULL;
1549	if (u_abs_timeout) {
1550		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1551		if (res)
1552			return res;
1553		p = &ts;
1554	}
1555	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1556}
1557#endif
1558
1559static const struct inode_operations mqueue_dir_inode_operations = {
1560	.lookup = simple_lookup,
1561	.create = mqueue_create,
1562	.unlink = mqueue_unlink,
1563};
1564
1565static const struct file_operations mqueue_file_operations = {
1566	.flush = mqueue_flush_file,
1567	.poll = mqueue_poll_file,
1568	.read = mqueue_read_file,
1569	.llseek = default_llseek,
1570};
1571
1572static const struct super_operations mqueue_super_ops = {
1573	.alloc_inode = mqueue_alloc_inode,
1574	.free_inode = mqueue_free_inode,
1575	.evict_inode = mqueue_evict_inode,
1576	.statfs = simple_statfs,
1577};
1578
1579static const struct fs_context_operations mqueue_fs_context_ops = {
1580	.free		= mqueue_fs_context_free,
1581	.get_tree	= mqueue_get_tree,
1582};
1583
1584static struct file_system_type mqueue_fs_type = {
1585	.name			= "mqueue",
1586	.init_fs_context	= mqueue_init_fs_context,
1587	.kill_sb		= kill_litter_super,
1588	.fs_flags		= FS_USERNS_MOUNT,
1589};
1590
1591int mq_init_ns(struct ipc_namespace *ns)
1592{
1593	struct vfsmount *m;
1594
1595	ns->mq_queues_count  = 0;
1596	ns->mq_queues_max    = DFLT_QUEUESMAX;
1597	ns->mq_msg_max       = DFLT_MSGMAX;
1598	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1599	ns->mq_msg_default   = DFLT_MSG;
1600	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1601
1602	m = mq_create_mount(ns);
1603	if (IS_ERR(m))
1604		return PTR_ERR(m);
1605	ns->mq_mnt = m;
1606	return 0;
1607}
1608
1609void mq_clear_sbinfo(struct ipc_namespace *ns)
1610{
1611	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1612}
1613
1614void mq_put_mnt(struct ipc_namespace *ns)
1615{
1616	kern_unmount(ns->mq_mnt);
1617}
1618
1619static int __init init_mqueue_fs(void)
1620{
1621	int error;
1622
1623	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1624				sizeof(struct mqueue_inode_info), 0,
1625				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1626	if (mqueue_inode_cachep == NULL)
1627		return -ENOMEM;
1628
1629	/* ignore failures - they are not fatal */
1630	mq_sysctl_table = mq_register_sysctl_table();
1631
1632	error = register_filesystem(&mqueue_fs_type);
1633	if (error)
1634		goto out_sysctl;
1635
1636	spin_lock_init(&mq_lock);
1637
1638	error = mq_init_ns(&init_ipc_ns);
1639	if (error)
1640		goto out_filesystem;
1641
1642	return 0;
1643
1644out_filesystem:
1645	unregister_filesystem(&mqueue_fs_type);
1646out_sysctl:
1647	if (mq_sysctl_table)
1648		unregister_sysctl_table(mq_sysctl_table);
1649	kmem_cache_destroy(mqueue_inode_cachep);
1650	return error;
1651}
1652
1653device_initcall(init_mqueue_fs);
1654