mqueue.c revision 6f686574
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/namei.h>
22#include <linux/sysctl.h>
23#include <linux/poll.h>
24#include <linux/mqueue.h>
25#include <linux/msg.h>
26#include <linux/skbuff.h>
27#include <linux/netlink.h>
28#include <linux/syscalls.h>
29#include <linux/audit.h>
30#include <linux/signal.h>
31#include <linux/mutex.h>
32#include <linux/nsproxy.h>
33#include <linux/pid.h>
34#include <linux/ipc_namespace.h>
35#include <linux/slab.h>
36
37#include <net/sock.h>
38#include "util.h"
39
40#define MQUEUE_MAGIC	0x19800202
41#define DIRENT_SIZE	20
42#define FILENT_SIZE	80
43
44#define SEND		0
45#define RECV		1
46
47#define STATE_NONE	0
48#define STATE_PENDING	1
49#define STATE_READY	2
50
51struct ext_wait_queue {		/* queue of sleeping tasks */
52	struct task_struct *task;
53	struct list_head list;
54	struct msg_msg *msg;	/* ptr of loaded message */
55	int state;		/* one of STATE_* values */
56};
57
58struct mqueue_inode_info {
59	spinlock_t lock;
60	struct inode vfs_inode;
61	wait_queue_head_t wait_q;
62
63	struct msg_msg **messages;
64	struct mq_attr attr;
65
66	struct sigevent notify;
67	struct pid* notify_owner;
68	struct user_struct *user;	/* user who created, for accounting */
69	struct sock *notify_sock;
70	struct sk_buff *notify_cookie;
71
72	/* for tasks waiting for free space and messages, respectively */
73	struct ext_wait_queue e_wait_q[2];
74
75	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
76};
77
78static const struct inode_operations mqueue_dir_inode_operations;
79static const struct file_operations mqueue_file_operations;
80static const struct super_operations mqueue_super_ops;
81static void remove_notification(struct mqueue_inode_info *info);
82
83static struct kmem_cache *mqueue_inode_cachep;
84
85static struct ctl_table_header * mq_sysctl_table;
86
87static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
88{
89	return container_of(inode, struct mqueue_inode_info, vfs_inode);
90}
91
92/*
93 * This routine should be called with the mq_lock held.
94 */
95static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
96{
97	return get_ipc_ns(inode->i_sb->s_fs_info);
98}
99
100static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
101{
102	struct ipc_namespace *ns;
103
104	spin_lock(&mq_lock);
105	ns = __get_ns_from_inode(inode);
106	spin_unlock(&mq_lock);
107	return ns;
108}
109
110static struct inode *mqueue_get_inode(struct super_block *sb,
111		struct ipc_namespace *ipc_ns, int mode,
112		struct mq_attr *attr)
113{
114	struct user_struct *u = current_user();
115	struct inode *inode;
116	int ret = -ENOMEM;
117
118	inode = new_inode(sb);
119	if (!inode)
120		goto err;
121
122	inode->i_ino = get_next_ino();
123	inode->i_mode = mode;
124	inode->i_uid = current_fsuid();
125	inode->i_gid = current_fsgid();
126	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
127
128	if (S_ISREG(mode)) {
129		struct mqueue_inode_info *info;
130		struct task_struct *p = current;
131		unsigned long mq_bytes, mq_msg_tblsz;
132
133		inode->i_fop = &mqueue_file_operations;
134		inode->i_size = FILENT_SIZE;
135		/* mqueue specific info */
136		info = MQUEUE_I(inode);
137		spin_lock_init(&info->lock);
138		init_waitqueue_head(&info->wait_q);
139		INIT_LIST_HEAD(&info->e_wait_q[0].list);
140		INIT_LIST_HEAD(&info->e_wait_q[1].list);
141		info->notify_owner = NULL;
142		info->qsize = 0;
143		info->user = NULL;	/* set when all is ok */
144		memset(&info->attr, 0, sizeof(info->attr));
145		info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
146		info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
147		if (attr) {
148			info->attr.mq_maxmsg = attr->mq_maxmsg;
149			info->attr.mq_msgsize = attr->mq_msgsize;
150		}
151		mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
152		info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
153		if (!info->messages)
154			goto out_inode;
155
156		mq_bytes = (mq_msg_tblsz +
157			(info->attr.mq_maxmsg * info->attr.mq_msgsize));
158
159		spin_lock(&mq_lock);
160		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
161		    u->mq_bytes + mq_bytes > task_rlimit(p, RLIMIT_MSGQUEUE)) {
162			spin_unlock(&mq_lock);
163			/* mqueue_evict_inode() releases info->messages */
164			ret = -EMFILE;
165			goto out_inode;
166		}
167		u->mq_bytes += mq_bytes;
168		spin_unlock(&mq_lock);
169
170		/* all is ok */
171		info->user = get_uid(u);
172	} else if (S_ISDIR(mode)) {
173		inc_nlink(inode);
174		/* Some things misbehave if size == 0 on a directory */
175		inode->i_size = 2 * DIRENT_SIZE;
176		inode->i_op = &mqueue_dir_inode_operations;
177		inode->i_fop = &simple_dir_operations;
178	}
179
180	return inode;
181out_inode:
182	iput(inode);
183err:
184	return ERR_PTR(ret);
185}
186
187static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
188{
189	struct inode *inode;
190	struct ipc_namespace *ns = data;
191	int error;
192
193	sb->s_blocksize = PAGE_CACHE_SIZE;
194	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
195	sb->s_magic = MQUEUE_MAGIC;
196	sb->s_op = &mqueue_super_ops;
197
198	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
199				NULL);
200	if (IS_ERR(inode)) {
201		error = PTR_ERR(inode);
202		goto out;
203	}
204
205	sb->s_root = d_alloc_root(inode);
206	if (!sb->s_root) {
207		iput(inode);
208		error = -ENOMEM;
209		goto out;
210	}
211	error = 0;
212
213out:
214	return error;
215}
216
217static struct dentry *mqueue_mount(struct file_system_type *fs_type,
218			 int flags, const char *dev_name,
219			 void *data)
220{
221	if (!(flags & MS_KERNMOUNT))
222		data = current->nsproxy->ipc_ns;
223	return mount_ns(fs_type, flags, data, mqueue_fill_super);
224}
225
226static void init_once(void *foo)
227{
228	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
229
230	inode_init_once(&p->vfs_inode);
231}
232
233static struct inode *mqueue_alloc_inode(struct super_block *sb)
234{
235	struct mqueue_inode_info *ei;
236
237	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
238	if (!ei)
239		return NULL;
240	return &ei->vfs_inode;
241}
242
243static void mqueue_i_callback(struct rcu_head *head)
244{
245	struct inode *inode = container_of(head, struct inode, i_rcu);
246	INIT_LIST_HEAD(&inode->i_dentry);
247	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
248}
249
250static void mqueue_destroy_inode(struct inode *inode)
251{
252	call_rcu(&inode->i_rcu, mqueue_i_callback);
253}
254
255static void mqueue_evict_inode(struct inode *inode)
256{
257	struct mqueue_inode_info *info;
258	struct user_struct *user;
259	unsigned long mq_bytes;
260	int i;
261	struct ipc_namespace *ipc_ns;
262
263	end_writeback(inode);
264
265	if (S_ISDIR(inode->i_mode))
266		return;
267
268	ipc_ns = get_ns_from_inode(inode);
269	info = MQUEUE_I(inode);
270	spin_lock(&info->lock);
271	for (i = 0; i < info->attr.mq_curmsgs; i++)
272		free_msg(info->messages[i]);
273	kfree(info->messages);
274	spin_unlock(&info->lock);
275
276	/* Total amount of bytes accounted for the mqueue */
277	mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
278	    + info->attr.mq_msgsize);
279	user = info->user;
280	if (user) {
281		spin_lock(&mq_lock);
282		user->mq_bytes -= mq_bytes;
283		/*
284		 * get_ns_from_inode() ensures that the
285		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
286		 * to which we now hold a reference, or it is NULL.
287		 * We can't put it here under mq_lock, though.
288		 */
289		if (ipc_ns)
290			ipc_ns->mq_queues_count--;
291		spin_unlock(&mq_lock);
292		free_uid(user);
293	}
294	if (ipc_ns)
295		put_ipc_ns(ipc_ns);
296}
297
298static int mqueue_create(struct inode *dir, struct dentry *dentry,
299				int mode, struct nameidata *nd)
300{
301	struct inode *inode;
302	struct mq_attr *attr = dentry->d_fsdata;
303	int error;
304	struct ipc_namespace *ipc_ns;
305
306	spin_lock(&mq_lock);
307	ipc_ns = __get_ns_from_inode(dir);
308	if (!ipc_ns) {
309		error = -EACCES;
310		goto out_unlock;
311	}
312	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
313			!capable(CAP_SYS_RESOURCE)) {
314		error = -ENOSPC;
315		goto out_unlock;
316	}
317	ipc_ns->mq_queues_count++;
318	spin_unlock(&mq_lock);
319
320	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
321	if (IS_ERR(inode)) {
322		error = PTR_ERR(inode);
323		spin_lock(&mq_lock);
324		ipc_ns->mq_queues_count--;
325		goto out_unlock;
326	}
327
328	put_ipc_ns(ipc_ns);
329	dir->i_size += DIRENT_SIZE;
330	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
331
332	d_instantiate(dentry, inode);
333	dget(dentry);
334	return 0;
335out_unlock:
336	spin_unlock(&mq_lock);
337	if (ipc_ns)
338		put_ipc_ns(ipc_ns);
339	return error;
340}
341
342static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
343{
344  	struct inode *inode = dentry->d_inode;
345
346	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
347	dir->i_size -= DIRENT_SIZE;
348  	drop_nlink(inode);
349  	dput(dentry);
350  	return 0;
351}
352
353/*
354*	This is routine for system read from queue file.
355*	To avoid mess with doing here some sort of mq_receive we allow
356*	to read only queue size & notification info (the only values
357*	that are interesting from user point of view and aren't accessible
358*	through std routines)
359*/
360static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
361				size_t count, loff_t *off)
362{
363	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
364	char buffer[FILENT_SIZE];
365	ssize_t ret;
366
367	spin_lock(&info->lock);
368	snprintf(buffer, sizeof(buffer),
369			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
370			info->qsize,
371			info->notify_owner ? info->notify.sigev_notify : 0,
372			(info->notify_owner &&
373			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
374				info->notify.sigev_signo : 0,
375			pid_vnr(info->notify_owner));
376	spin_unlock(&info->lock);
377	buffer[sizeof(buffer)-1] = '\0';
378
379	ret = simple_read_from_buffer(u_data, count, off, buffer,
380				strlen(buffer));
381	if (ret <= 0)
382		return ret;
383
384	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
385	return ret;
386}
387
388static int mqueue_flush_file(struct file *filp, fl_owner_t id)
389{
390	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
391
392	spin_lock(&info->lock);
393	if (task_tgid(current) == info->notify_owner)
394		remove_notification(info);
395
396	spin_unlock(&info->lock);
397	return 0;
398}
399
400static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
401{
402	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
403	int retval = 0;
404
405	poll_wait(filp, &info->wait_q, poll_tab);
406
407	spin_lock(&info->lock);
408	if (info->attr.mq_curmsgs)
409		retval = POLLIN | POLLRDNORM;
410
411	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
412		retval |= POLLOUT | POLLWRNORM;
413	spin_unlock(&info->lock);
414
415	return retval;
416}
417
418/* Adds current to info->e_wait_q[sr] before element with smaller prio */
419static void wq_add(struct mqueue_inode_info *info, int sr,
420			struct ext_wait_queue *ewp)
421{
422	struct ext_wait_queue *walk;
423
424	ewp->task = current;
425
426	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
427		if (walk->task->static_prio <= current->static_prio) {
428			list_add_tail(&ewp->list, &walk->list);
429			return;
430		}
431	}
432	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
433}
434
435/*
436 * Puts current task to sleep. Caller must hold queue lock. After return
437 * lock isn't held.
438 * sr: SEND or RECV
439 */
440static int wq_sleep(struct mqueue_inode_info *info, int sr,
441		    ktime_t *timeout, struct ext_wait_queue *ewp)
442{
443	int retval;
444	signed long time;
445
446	wq_add(info, sr, ewp);
447
448	for (;;) {
449		set_current_state(TASK_INTERRUPTIBLE);
450
451		spin_unlock(&info->lock);
452		time = schedule_hrtimeout_range_clock(timeout, 0,
453			HRTIMER_MODE_ABS, CLOCK_REALTIME);
454
455		while (ewp->state == STATE_PENDING)
456			cpu_relax();
457
458		if (ewp->state == STATE_READY) {
459			retval = 0;
460			goto out;
461		}
462		spin_lock(&info->lock);
463		if (ewp->state == STATE_READY) {
464			retval = 0;
465			goto out_unlock;
466		}
467		if (signal_pending(current)) {
468			retval = -ERESTARTSYS;
469			break;
470		}
471		if (time == 0) {
472			retval = -ETIMEDOUT;
473			break;
474		}
475	}
476	list_del(&ewp->list);
477out_unlock:
478	spin_unlock(&info->lock);
479out:
480	return retval;
481}
482
483/*
484 * Returns waiting task that should be serviced first or NULL if none exists
485 */
486static struct ext_wait_queue *wq_get_first_waiter(
487		struct mqueue_inode_info *info, int sr)
488{
489	struct list_head *ptr;
490
491	ptr = info->e_wait_q[sr].list.prev;
492	if (ptr == &info->e_wait_q[sr].list)
493		return NULL;
494	return list_entry(ptr, struct ext_wait_queue, list);
495}
496
497/* Auxiliary functions to manipulate messages' list */
498static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
499{
500	int k;
501
502	k = info->attr.mq_curmsgs - 1;
503	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
504		info->messages[k + 1] = info->messages[k];
505		k--;
506	}
507	info->attr.mq_curmsgs++;
508	info->qsize += ptr->m_ts;
509	info->messages[k + 1] = ptr;
510}
511
512static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
513{
514	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
515	return info->messages[info->attr.mq_curmsgs];
516}
517
518static inline void set_cookie(struct sk_buff *skb, char code)
519{
520	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
521}
522
523/*
524 * The next function is only to split too long sys_mq_timedsend
525 */
526static void __do_notify(struct mqueue_inode_info *info)
527{
528	/* notification
529	 * invoked when there is registered process and there isn't process
530	 * waiting synchronously for message AND state of queue changed from
531	 * empty to not empty. Here we are sure that no one is waiting
532	 * synchronously. */
533	if (info->notify_owner &&
534	    info->attr.mq_curmsgs == 1) {
535		struct siginfo sig_i;
536		switch (info->notify.sigev_notify) {
537		case SIGEV_NONE:
538			break;
539		case SIGEV_SIGNAL:
540			/* sends signal */
541
542			sig_i.si_signo = info->notify.sigev_signo;
543			sig_i.si_errno = 0;
544			sig_i.si_code = SI_MESGQ;
545			sig_i.si_value = info->notify.sigev_value;
546			sig_i.si_pid = task_tgid_nr_ns(current,
547						ns_of_pid(info->notify_owner));
548			sig_i.si_uid = current_uid();
549
550			kill_pid_info(info->notify.sigev_signo,
551				      &sig_i, info->notify_owner);
552			break;
553		case SIGEV_THREAD:
554			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
555			netlink_sendskb(info->notify_sock, info->notify_cookie);
556			break;
557		}
558		/* after notification unregisters process */
559		put_pid(info->notify_owner);
560		info->notify_owner = NULL;
561	}
562	wake_up(&info->wait_q);
563}
564
565static int prepare_timeout(const struct timespec __user *u_abs_timeout,
566			   ktime_t *expires, struct timespec *ts)
567{
568	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
569		return -EFAULT;
570	if (!timespec_valid(ts))
571		return -EINVAL;
572
573	*expires = timespec_to_ktime(*ts);
574	return 0;
575}
576
577static void remove_notification(struct mqueue_inode_info *info)
578{
579	if (info->notify_owner != NULL &&
580	    info->notify.sigev_notify == SIGEV_THREAD) {
581		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
582		netlink_sendskb(info->notify_sock, info->notify_cookie);
583	}
584	put_pid(info->notify_owner);
585	info->notify_owner = NULL;
586}
587
588static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
589{
590	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
591		return 0;
592	if (capable(CAP_SYS_RESOURCE)) {
593		if (attr->mq_maxmsg > HARD_MSGMAX)
594			return 0;
595	} else {
596		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
597				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
598			return 0;
599	}
600	/* check for overflow */
601	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
602		return 0;
603	if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
604	    + sizeof (struct msg_msg *))) <
605	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
606		return 0;
607	return 1;
608}
609
610/*
611 * Invoked when creating a new queue via sys_mq_open
612 */
613static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
614			struct dentry *dentry, int oflag, mode_t mode,
615			struct mq_attr *attr)
616{
617	const struct cred *cred = current_cred();
618	struct file *result;
619	int ret;
620
621	if (attr) {
622		if (!mq_attr_ok(ipc_ns, attr)) {
623			ret = -EINVAL;
624			goto out;
625		}
626		/* store for use during create */
627		dentry->d_fsdata = attr;
628	}
629
630	mode &= ~current_umask();
631	ret = mnt_want_write(ipc_ns->mq_mnt);
632	if (ret)
633		goto out;
634	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
635	dentry->d_fsdata = NULL;
636	if (ret)
637		goto out_drop_write;
638
639	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
640	/*
641	 * dentry_open() took a persistent mnt_want_write(),
642	 * so we can now drop this one.
643	 */
644	mnt_drop_write(ipc_ns->mq_mnt);
645	return result;
646
647out_drop_write:
648	mnt_drop_write(ipc_ns->mq_mnt);
649out:
650	dput(dentry);
651	mntput(ipc_ns->mq_mnt);
652	return ERR_PTR(ret);
653}
654
655/* Opens existing queue */
656static struct file *do_open(struct ipc_namespace *ipc_ns,
657				struct dentry *dentry, int oflag)
658{
659	int ret;
660	const struct cred *cred = current_cred();
661
662	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
663						  MAY_READ | MAY_WRITE };
664
665	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
666		ret = -EINVAL;
667		goto err;
668	}
669
670	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
671		ret = -EACCES;
672		goto err;
673	}
674
675	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
676
677err:
678	dput(dentry);
679	mntput(ipc_ns->mq_mnt);
680	return ERR_PTR(ret);
681}
682
683SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
684		struct mq_attr __user *, u_attr)
685{
686	struct dentry *dentry;
687	struct file *filp;
688	char *name;
689	struct mq_attr attr;
690	int fd, error;
691	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
692
693	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
694		return -EFAULT;
695
696	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
697
698	if (IS_ERR(name = getname(u_name)))
699		return PTR_ERR(name);
700
701	fd = get_unused_fd_flags(O_CLOEXEC);
702	if (fd < 0)
703		goto out_putname;
704
705	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
706	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
707	if (IS_ERR(dentry)) {
708		error = PTR_ERR(dentry);
709		goto out_putfd;
710	}
711	mntget(ipc_ns->mq_mnt);
712
713	if (oflag & O_CREAT) {
714		if (dentry->d_inode) {	/* entry already exists */
715			audit_inode(name, dentry);
716			if (oflag & O_EXCL) {
717				error = -EEXIST;
718				goto out;
719			}
720			filp = do_open(ipc_ns, dentry, oflag);
721		} else {
722			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
723						dentry, oflag, mode,
724						u_attr ? &attr : NULL);
725		}
726	} else {
727		if (!dentry->d_inode) {
728			error = -ENOENT;
729			goto out;
730		}
731		audit_inode(name, dentry);
732		filp = do_open(ipc_ns, dentry, oflag);
733	}
734
735	if (IS_ERR(filp)) {
736		error = PTR_ERR(filp);
737		goto out_putfd;
738	}
739
740	fd_install(fd, filp);
741	goto out_upsem;
742
743out:
744	dput(dentry);
745	mntput(ipc_ns->mq_mnt);
746out_putfd:
747	put_unused_fd(fd);
748	fd = error;
749out_upsem:
750	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
751out_putname:
752	putname(name);
753	return fd;
754}
755
756SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
757{
758	int err;
759	char *name;
760	struct dentry *dentry;
761	struct inode *inode = NULL;
762	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
763
764	name = getname(u_name);
765	if (IS_ERR(name))
766		return PTR_ERR(name);
767
768	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
769			I_MUTEX_PARENT);
770	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
771	if (IS_ERR(dentry)) {
772		err = PTR_ERR(dentry);
773		goto out_unlock;
774	}
775
776	if (!dentry->d_inode) {
777		err = -ENOENT;
778		goto out_err;
779	}
780
781	inode = dentry->d_inode;
782	if (inode)
783		ihold(inode);
784	err = mnt_want_write(ipc_ns->mq_mnt);
785	if (err)
786		goto out_err;
787	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
788	mnt_drop_write(ipc_ns->mq_mnt);
789out_err:
790	dput(dentry);
791
792out_unlock:
793	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
794	putname(name);
795	if (inode)
796		iput(inode);
797
798	return err;
799}
800
801/* Pipelined send and receive functions.
802 *
803 * If a receiver finds no waiting message, then it registers itself in the
804 * list of waiting receivers. A sender checks that list before adding the new
805 * message into the message array. If there is a waiting receiver, then it
806 * bypasses the message array and directly hands the message over to the
807 * receiver.
808 * The receiver accepts the message and returns without grabbing the queue
809 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
810 * are necessary. The same algorithm is used for sysv semaphores, see
811 * ipc/sem.c for more details.
812 *
813 * The same algorithm is used for senders.
814 */
815
816/* pipelined_send() - send a message directly to the task waiting in
817 * sys_mq_timedreceive() (without inserting message into a queue).
818 */
819static inline void pipelined_send(struct mqueue_inode_info *info,
820				  struct msg_msg *message,
821				  struct ext_wait_queue *receiver)
822{
823	receiver->msg = message;
824	list_del(&receiver->list);
825	receiver->state = STATE_PENDING;
826	wake_up_process(receiver->task);
827	smp_wmb();
828	receiver->state = STATE_READY;
829}
830
831/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
832 * gets its message and put to the queue (we have one free place for sure). */
833static inline void pipelined_receive(struct mqueue_inode_info *info)
834{
835	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
836
837	if (!sender) {
838		/* for poll */
839		wake_up_interruptible(&info->wait_q);
840		return;
841	}
842	msg_insert(sender->msg, info);
843	list_del(&sender->list);
844	sender->state = STATE_PENDING;
845	wake_up_process(sender->task);
846	smp_wmb();
847	sender->state = STATE_READY;
848}
849
850SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
851		size_t, msg_len, unsigned int, msg_prio,
852		const struct timespec __user *, u_abs_timeout)
853{
854	struct file *filp;
855	struct inode *inode;
856	struct ext_wait_queue wait;
857	struct ext_wait_queue *receiver;
858	struct msg_msg *msg_ptr;
859	struct mqueue_inode_info *info;
860	ktime_t expires, *timeout = NULL;
861	struct timespec ts;
862	int ret;
863
864	if (u_abs_timeout) {
865		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
866		if (res)
867			return res;
868		timeout = &expires;
869	}
870
871	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
872		return -EINVAL;
873
874	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
875
876	filp = fget(mqdes);
877	if (unlikely(!filp)) {
878		ret = -EBADF;
879		goto out;
880	}
881
882	inode = filp->f_path.dentry->d_inode;
883	if (unlikely(filp->f_op != &mqueue_file_operations)) {
884		ret = -EBADF;
885		goto out_fput;
886	}
887	info = MQUEUE_I(inode);
888	audit_inode(NULL, filp->f_path.dentry);
889
890	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
891		ret = -EBADF;
892		goto out_fput;
893	}
894
895	if (unlikely(msg_len > info->attr.mq_msgsize)) {
896		ret = -EMSGSIZE;
897		goto out_fput;
898	}
899
900	/* First try to allocate memory, before doing anything with
901	 * existing queues. */
902	msg_ptr = load_msg(u_msg_ptr, msg_len);
903	if (IS_ERR(msg_ptr)) {
904		ret = PTR_ERR(msg_ptr);
905		goto out_fput;
906	}
907	msg_ptr->m_ts = msg_len;
908	msg_ptr->m_type = msg_prio;
909
910	spin_lock(&info->lock);
911
912	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
913		if (filp->f_flags & O_NONBLOCK) {
914			spin_unlock(&info->lock);
915			ret = -EAGAIN;
916		} else {
917			wait.task = current;
918			wait.msg = (void *) msg_ptr;
919			wait.state = STATE_NONE;
920			ret = wq_sleep(info, SEND, timeout, &wait);
921		}
922		if (ret < 0)
923			free_msg(msg_ptr);
924	} else {
925		receiver = wq_get_first_waiter(info, RECV);
926		if (receiver) {
927			pipelined_send(info, msg_ptr, receiver);
928		} else {
929			/* adds message to the queue */
930			msg_insert(msg_ptr, info);
931			__do_notify(info);
932		}
933		inode->i_atime = inode->i_mtime = inode->i_ctime =
934				CURRENT_TIME;
935		spin_unlock(&info->lock);
936		ret = 0;
937	}
938out_fput:
939	fput(filp);
940out:
941	return ret;
942}
943
944SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
945		size_t, msg_len, unsigned int __user *, u_msg_prio,
946		const struct timespec __user *, u_abs_timeout)
947{
948	ssize_t ret;
949	struct msg_msg *msg_ptr;
950	struct file *filp;
951	struct inode *inode;
952	struct mqueue_inode_info *info;
953	struct ext_wait_queue wait;
954	ktime_t expires, *timeout = NULL;
955	struct timespec ts;
956
957	if (u_abs_timeout) {
958		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
959		if (res)
960			return res;
961		timeout = &expires;
962	}
963
964	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
965
966	filp = fget(mqdes);
967	if (unlikely(!filp)) {
968		ret = -EBADF;
969		goto out;
970	}
971
972	inode = filp->f_path.dentry->d_inode;
973	if (unlikely(filp->f_op != &mqueue_file_operations)) {
974		ret = -EBADF;
975		goto out_fput;
976	}
977	info = MQUEUE_I(inode);
978	audit_inode(NULL, filp->f_path.dentry);
979
980	if (unlikely(!(filp->f_mode & FMODE_READ))) {
981		ret = -EBADF;
982		goto out_fput;
983	}
984
985	/* checks if buffer is big enough */
986	if (unlikely(msg_len < info->attr.mq_msgsize)) {
987		ret = -EMSGSIZE;
988		goto out_fput;
989	}
990
991	spin_lock(&info->lock);
992	if (info->attr.mq_curmsgs == 0) {
993		if (filp->f_flags & O_NONBLOCK) {
994			spin_unlock(&info->lock);
995			ret = -EAGAIN;
996		} else {
997			wait.task = current;
998			wait.state = STATE_NONE;
999			ret = wq_sleep(info, RECV, timeout, &wait);
1000			msg_ptr = wait.msg;
1001		}
1002	} else {
1003		msg_ptr = msg_get(info);
1004
1005		inode->i_atime = inode->i_mtime = inode->i_ctime =
1006				CURRENT_TIME;
1007
1008		/* There is now free space in queue. */
1009		pipelined_receive(info);
1010		spin_unlock(&info->lock);
1011		ret = 0;
1012	}
1013	if (ret == 0) {
1014		ret = msg_ptr->m_ts;
1015
1016		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1017			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1018			ret = -EFAULT;
1019		}
1020		free_msg(msg_ptr);
1021	}
1022out_fput:
1023	fput(filp);
1024out:
1025	return ret;
1026}
1027
1028/*
1029 * Notes: the case when user wants us to deregister (with NULL as pointer)
1030 * and he isn't currently owner of notification, will be silently discarded.
1031 * It isn't explicitly defined in the POSIX.
1032 */
1033SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1034		const struct sigevent __user *, u_notification)
1035{
1036	int ret;
1037	struct file *filp;
1038	struct sock *sock;
1039	struct inode *inode;
1040	struct sigevent notification;
1041	struct mqueue_inode_info *info;
1042	struct sk_buff *nc;
1043
1044	if (u_notification) {
1045		if (copy_from_user(&notification, u_notification,
1046					sizeof(struct sigevent)))
1047			return -EFAULT;
1048	}
1049
1050	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1051
1052	nc = NULL;
1053	sock = NULL;
1054	if (u_notification != NULL) {
1055		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1056			     notification.sigev_notify != SIGEV_SIGNAL &&
1057			     notification.sigev_notify != SIGEV_THREAD))
1058			return -EINVAL;
1059		if (notification.sigev_notify == SIGEV_SIGNAL &&
1060			!valid_signal(notification.sigev_signo)) {
1061			return -EINVAL;
1062		}
1063		if (notification.sigev_notify == SIGEV_THREAD) {
1064			long timeo;
1065
1066			/* create the notify skb */
1067			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1068			if (!nc) {
1069				ret = -ENOMEM;
1070				goto out;
1071			}
1072			if (copy_from_user(nc->data,
1073					notification.sigev_value.sival_ptr,
1074					NOTIFY_COOKIE_LEN)) {
1075				ret = -EFAULT;
1076				goto out;
1077			}
1078
1079			/* TODO: add a header? */
1080			skb_put(nc, NOTIFY_COOKIE_LEN);
1081			/* and attach it to the socket */
1082retry:
1083			filp = fget(notification.sigev_signo);
1084			if (!filp) {
1085				ret = -EBADF;
1086				goto out;
1087			}
1088			sock = netlink_getsockbyfilp(filp);
1089			fput(filp);
1090			if (IS_ERR(sock)) {
1091				ret = PTR_ERR(sock);
1092				sock = NULL;
1093				goto out;
1094			}
1095
1096			timeo = MAX_SCHEDULE_TIMEOUT;
1097			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1098			if (ret == 1)
1099				goto retry;
1100			if (ret) {
1101				sock = NULL;
1102				nc = NULL;
1103				goto out;
1104			}
1105		}
1106	}
1107
1108	filp = fget(mqdes);
1109	if (!filp) {
1110		ret = -EBADF;
1111		goto out;
1112	}
1113
1114	inode = filp->f_path.dentry->d_inode;
1115	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1116		ret = -EBADF;
1117		goto out_fput;
1118	}
1119	info = MQUEUE_I(inode);
1120
1121	ret = 0;
1122	spin_lock(&info->lock);
1123	if (u_notification == NULL) {
1124		if (info->notify_owner == task_tgid(current)) {
1125			remove_notification(info);
1126			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1127		}
1128	} else if (info->notify_owner != NULL) {
1129		ret = -EBUSY;
1130	} else {
1131		switch (notification.sigev_notify) {
1132		case SIGEV_NONE:
1133			info->notify.sigev_notify = SIGEV_NONE;
1134			break;
1135		case SIGEV_THREAD:
1136			info->notify_sock = sock;
1137			info->notify_cookie = nc;
1138			sock = NULL;
1139			nc = NULL;
1140			info->notify.sigev_notify = SIGEV_THREAD;
1141			break;
1142		case SIGEV_SIGNAL:
1143			info->notify.sigev_signo = notification.sigev_signo;
1144			info->notify.sigev_value = notification.sigev_value;
1145			info->notify.sigev_notify = SIGEV_SIGNAL;
1146			break;
1147		}
1148
1149		info->notify_owner = get_pid(task_tgid(current));
1150		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1151	}
1152	spin_unlock(&info->lock);
1153out_fput:
1154	fput(filp);
1155out:
1156	if (sock) {
1157		netlink_detachskb(sock, nc);
1158	} else if (nc) {
1159		dev_kfree_skb(nc);
1160	}
1161	return ret;
1162}
1163
1164SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1165		const struct mq_attr __user *, u_mqstat,
1166		struct mq_attr __user *, u_omqstat)
1167{
1168	int ret;
1169	struct mq_attr mqstat, omqstat;
1170	struct file *filp;
1171	struct inode *inode;
1172	struct mqueue_inode_info *info;
1173
1174	if (u_mqstat != NULL) {
1175		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1176			return -EFAULT;
1177		if (mqstat.mq_flags & (~O_NONBLOCK))
1178			return -EINVAL;
1179	}
1180
1181	filp = fget(mqdes);
1182	if (!filp) {
1183		ret = -EBADF;
1184		goto out;
1185	}
1186
1187	inode = filp->f_path.dentry->d_inode;
1188	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1189		ret = -EBADF;
1190		goto out_fput;
1191	}
1192	info = MQUEUE_I(inode);
1193
1194	spin_lock(&info->lock);
1195
1196	omqstat = info->attr;
1197	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1198	if (u_mqstat) {
1199		audit_mq_getsetattr(mqdes, &mqstat);
1200		spin_lock(&filp->f_lock);
1201		if (mqstat.mq_flags & O_NONBLOCK)
1202			filp->f_flags |= O_NONBLOCK;
1203		else
1204			filp->f_flags &= ~O_NONBLOCK;
1205		spin_unlock(&filp->f_lock);
1206
1207		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1208	}
1209
1210	spin_unlock(&info->lock);
1211
1212	ret = 0;
1213	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1214						sizeof(struct mq_attr)))
1215		ret = -EFAULT;
1216
1217out_fput:
1218	fput(filp);
1219out:
1220	return ret;
1221}
1222
1223static const struct inode_operations mqueue_dir_inode_operations = {
1224	.lookup = simple_lookup,
1225	.create = mqueue_create,
1226	.unlink = mqueue_unlink,
1227};
1228
1229static const struct file_operations mqueue_file_operations = {
1230	.flush = mqueue_flush_file,
1231	.poll = mqueue_poll_file,
1232	.read = mqueue_read_file,
1233	.llseek = default_llseek,
1234};
1235
1236static const struct super_operations mqueue_super_ops = {
1237	.alloc_inode = mqueue_alloc_inode,
1238	.destroy_inode = mqueue_destroy_inode,
1239	.evict_inode = mqueue_evict_inode,
1240	.statfs = simple_statfs,
1241};
1242
1243static struct file_system_type mqueue_fs_type = {
1244	.name = "mqueue",
1245	.mount = mqueue_mount,
1246	.kill_sb = kill_litter_super,
1247};
1248
1249int mq_init_ns(struct ipc_namespace *ns)
1250{
1251	ns->mq_queues_count  = 0;
1252	ns->mq_queues_max    = DFLT_QUEUESMAX;
1253	ns->mq_msg_max       = DFLT_MSGMAX;
1254	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1255
1256	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1257	if (IS_ERR(ns->mq_mnt)) {
1258		int err = PTR_ERR(ns->mq_mnt);
1259		ns->mq_mnt = NULL;
1260		return err;
1261	}
1262	return 0;
1263}
1264
1265void mq_clear_sbinfo(struct ipc_namespace *ns)
1266{
1267	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1268}
1269
1270void mq_put_mnt(struct ipc_namespace *ns)
1271{
1272	kern_unmount(ns->mq_mnt);
1273}
1274
1275static int __init init_mqueue_fs(void)
1276{
1277	int error;
1278
1279	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1280				sizeof(struct mqueue_inode_info), 0,
1281				SLAB_HWCACHE_ALIGN, init_once);
1282	if (mqueue_inode_cachep == NULL)
1283		return -ENOMEM;
1284
1285	/* ignore failures - they are not fatal */
1286	mq_sysctl_table = mq_register_sysctl_table();
1287
1288	error = register_filesystem(&mqueue_fs_type);
1289	if (error)
1290		goto out_sysctl;
1291
1292	spin_lock_init(&mq_lock);
1293
1294	error = mq_init_ns(&init_ipc_ns);
1295	if (error)
1296		goto out_filesystem;
1297
1298	return 0;
1299
1300out_filesystem:
1301	unregister_filesystem(&mqueue_fs_type);
1302out_sysctl:
1303	if (mq_sysctl_table)
1304		unregister_sysctl_table(mq_sysctl_table);
1305	kmem_cache_destroy(mqueue_inode_cachep);
1306	return error;
1307}
1308
1309__initcall(init_mqueue_fs);
1310