Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[pandora-kernel.git] / ipc / mqueue.c
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/vmalloc.h>
28 #include <linux/netlink.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/signal.h>
32 #include <linux/mutex.h>
33 #include <linux/nsproxy.h>
34 #include <linux/pid.h>
35 #include <linux/ipc_namespace.h>
36 #include <linux/user_namespace.h>
37 #include <linux/slab.h>
38
39 #include <net/sock.h>
40 #include "util.h"
41
42 #define MQUEUE_MAGIC    0x19800202
43 #define DIRENT_SIZE     20
44 #define FILENT_SIZE     80
45
46 #define SEND            0
47 #define RECV            1
48
49 #define STATE_NONE      0
50 #define STATE_PENDING   1
51 #define STATE_READY     2
52
53 struct posix_msg_tree_node {
54         struct rb_node          rb_node;
55         struct list_head        msg_list;
56         int                     priority;
57 };
58
59 struct ext_wait_queue {         /* queue of sleeping tasks */
60         struct task_struct *task;
61         struct list_head list;
62         struct msg_msg *msg;    /* ptr of loaded message */
63         int state;              /* one of STATE_* values */
64 };
65
66 struct mqueue_inode_info {
67         spinlock_t lock;
68         struct inode vfs_inode;
69         wait_queue_head_t wait_q;
70
71         struct rb_root msg_tree;
72         struct posix_msg_tree_node *node_cache;
73         struct mq_attr attr;
74
75         struct sigevent notify;
76         struct pid* notify_owner;
77         struct user_namespace *notify_user_ns;
78         struct user_struct *user;       /* user who created, for accounting */
79         struct sock *notify_sock;
80         struct sk_buff *notify_cookie;
81
82         /* for tasks waiting for free space and messages, respectively */
83         struct ext_wait_queue e_wait_q[2];
84
85         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
86 };
87
88 static const struct inode_operations mqueue_dir_inode_operations;
89 static const struct file_operations mqueue_file_operations;
90 static const struct super_operations mqueue_super_ops;
91 static void remove_notification(struct mqueue_inode_info *info);
92
93 static struct kmem_cache *mqueue_inode_cachep;
94
95 static struct ctl_table_header * mq_sysctl_table;
96
97 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98 {
99         return container_of(inode, struct mqueue_inode_info, vfs_inode);
100 }
101
102 /*
103  * This routine should be called with the mq_lock held.
104  */
105 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
106 {
107         return get_ipc_ns(inode->i_sb->s_fs_info);
108 }
109
110 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
111 {
112         struct ipc_namespace *ns;
113
114         spin_lock(&mq_lock);
115         ns = __get_ns_from_inode(inode);
116         spin_unlock(&mq_lock);
117         return ns;
118 }
119
120 /* Auxiliary functions to manipulate messages' list */
121 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122 {
123         struct rb_node **p, *parent = NULL;
124         struct posix_msg_tree_node *leaf;
125
126         p = &info->msg_tree.rb_node;
127         while (*p) {
128                 parent = *p;
129                 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130
131                 if (likely(leaf->priority == msg->m_type))
132                         goto insert_msg;
133                 else if (msg->m_type < leaf->priority)
134                         p = &(*p)->rb_left;
135                 else
136                         p = &(*p)->rb_right;
137         }
138         if (info->node_cache) {
139                 leaf = info->node_cache;
140                 info->node_cache = NULL;
141         } else {
142                 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
143                 if (!leaf)
144                         return -ENOMEM;
145                 rb_init_node(&leaf->rb_node);
146                 INIT_LIST_HEAD(&leaf->msg_list);
147                 info->qsize += sizeof(*leaf);
148         }
149         leaf->priority = msg->m_type;
150         rb_link_node(&leaf->rb_node, parent, p);
151         rb_insert_color(&leaf->rb_node, &info->msg_tree);
152 insert_msg:
153         info->attr.mq_curmsgs++;
154         info->qsize += msg->m_ts;
155         list_add_tail(&msg->m_list, &leaf->msg_list);
156         return 0;
157 }
158
159 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160 {
161         struct rb_node **p, *parent = NULL;
162         struct posix_msg_tree_node *leaf;
163         struct msg_msg *msg;
164
165 try_again:
166         p = &info->msg_tree.rb_node;
167         while (*p) {
168                 parent = *p;
169                 /*
170                  * During insert, low priorities go to the left and high to the
171                  * right.  On receive, we want the highest priorities first, so
172                  * walk all the way to the right.
173                  */
174                 p = &(*p)->rb_right;
175         }
176         if (!parent) {
177                 if (info->attr.mq_curmsgs) {
178                         pr_warn_once("Inconsistency in POSIX message queue, "
179                                      "no tree element, but supposedly messages "
180                                      "should exist!\n");
181                         info->attr.mq_curmsgs = 0;
182                 }
183                 return NULL;
184         }
185         leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186         if (unlikely(list_empty(&leaf->msg_list))) {
187                 pr_warn_once("Inconsistency in POSIX message queue, "
188                              "empty leaf node but we haven't implemented "
189                              "lazy leaf delete!\n");
190                 rb_erase(&leaf->rb_node, &info->msg_tree);
191                 if (info->node_cache) {
192                         info->qsize -= sizeof(*leaf);
193                         kfree(leaf);
194                 } else {
195                         info->node_cache = leaf;
196                 }
197                 goto try_again;
198         } else {
199                 msg = list_first_entry(&leaf->msg_list,
200                                        struct msg_msg, m_list);
201                 list_del(&msg->m_list);
202                 if (list_empty(&leaf->msg_list)) {
203                         rb_erase(&leaf->rb_node, &info->msg_tree);
204                         if (info->node_cache) {
205                                 info->qsize -= sizeof(*leaf);
206                                 kfree(leaf);
207                         } else {
208                                 info->node_cache = leaf;
209                         }
210                 }
211         }
212         info->attr.mq_curmsgs--;
213         info->qsize -= msg->m_ts;
214         return msg;
215 }
216
217 static struct inode *mqueue_get_inode(struct super_block *sb,
218                 struct ipc_namespace *ipc_ns, umode_t mode,
219                 struct mq_attr *attr)
220 {
221         struct user_struct *u = current_user();
222         struct inode *inode;
223         int ret = -ENOMEM;
224
225         inode = new_inode(sb);
226         if (!inode)
227                 goto err;
228
229         inode->i_ino = get_next_ino();
230         inode->i_mode = mode;
231         inode->i_uid = current_fsuid();
232         inode->i_gid = current_fsgid();
233         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
234
235         if (S_ISREG(mode)) {
236                 struct mqueue_inode_info *info;
237                 unsigned long mq_bytes, mq_treesize;
238
239                 inode->i_fop = &mqueue_file_operations;
240                 inode->i_size = FILENT_SIZE;
241                 /* mqueue specific info */
242                 info = MQUEUE_I(inode);
243                 spin_lock_init(&info->lock);
244                 init_waitqueue_head(&info->wait_q);
245                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
246                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
247                 info->notify_owner = NULL;
248                 info->notify_user_ns = NULL;
249                 info->qsize = 0;
250                 info->user = NULL;      /* set when all is ok */
251                 info->msg_tree = RB_ROOT;
252                 info->node_cache = NULL;
253                 memset(&info->attr, 0, sizeof(info->attr));
254                 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
255                                            ipc_ns->mq_msg_default);
256                 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
257                                             ipc_ns->mq_msgsize_default);
258                 if (attr) {
259                         info->attr.mq_maxmsg = attr->mq_maxmsg;
260                         info->attr.mq_msgsize = attr->mq_msgsize;
261                 }
262                 /*
263                  * We used to allocate a static array of pointers and account
264                  * the size of that array as well as one msg_msg struct per
265                  * possible message into the queue size. That's no longer
266                  * accurate as the queue is now an rbtree and will grow and
267                  * shrink depending on usage patterns.  We can, however, still
268                  * account one msg_msg struct per message, but the nodes are
269                  * allocated depending on priority usage, and most programs
270                  * only use one, or a handful, of priorities.  However, since
271                  * this is pinned memory, we need to assume worst case, so
272                  * that means the min(mq_maxmsg, max_priorities) * struct
273                  * posix_msg_tree_node.
274                  */
275                 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
276                         min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
277                         sizeof(struct posix_msg_tree_node);
278
279                 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
280                                           info->attr.mq_msgsize);
281
282                 spin_lock(&mq_lock);
283                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
284                     u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
285                         spin_unlock(&mq_lock);
286                         /* mqueue_evict_inode() releases info->messages */
287                         ret = -EMFILE;
288                         goto out_inode;
289                 }
290                 u->mq_bytes += mq_bytes;
291                 spin_unlock(&mq_lock);
292
293                 /* all is ok */
294                 info->user = get_uid(u);
295         } else if (S_ISDIR(mode)) {
296                 inc_nlink(inode);
297                 /* Some things misbehave if size == 0 on a directory */
298                 inode->i_size = 2 * DIRENT_SIZE;
299                 inode->i_op = &mqueue_dir_inode_operations;
300                 inode->i_fop = &simple_dir_operations;
301         }
302
303         return inode;
304 out_inode:
305         iput(inode);
306 err:
307         return ERR_PTR(ret);
308 }
309
310 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
311 {
312         struct inode *inode;
313         struct ipc_namespace *ns = data;
314
315         sb->s_blocksize = PAGE_CACHE_SIZE;
316         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
317         sb->s_magic = MQUEUE_MAGIC;
318         sb->s_op = &mqueue_super_ops;
319
320         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
321         if (IS_ERR(inode))
322                 return PTR_ERR(inode);
323
324         sb->s_root = d_make_root(inode);
325         if (!sb->s_root)
326                 return -ENOMEM;
327         return 0;
328 }
329
330 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
331                          int flags, const char *dev_name,
332                          void *data)
333 {
334         if (!(flags & MS_KERNMOUNT))
335                 data = current->nsproxy->ipc_ns;
336         return mount_ns(fs_type, flags, data, mqueue_fill_super);
337 }
338
339 static void init_once(void *foo)
340 {
341         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
342
343         inode_init_once(&p->vfs_inode);
344 }
345
346 static struct inode *mqueue_alloc_inode(struct super_block *sb)
347 {
348         struct mqueue_inode_info *ei;
349
350         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
351         if (!ei)
352                 return NULL;
353         return &ei->vfs_inode;
354 }
355
356 static void mqueue_i_callback(struct rcu_head *head)
357 {
358         struct inode *inode = container_of(head, struct inode, i_rcu);
359         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
360 }
361
362 static void mqueue_destroy_inode(struct inode *inode)
363 {
364         call_rcu(&inode->i_rcu, mqueue_i_callback);
365 }
366
367 static void mqueue_evict_inode(struct inode *inode)
368 {
369         struct mqueue_inode_info *info;
370         struct user_struct *user;
371         unsigned long mq_bytes, mq_treesize;
372         struct ipc_namespace *ipc_ns;
373         struct msg_msg *msg;
374
375         clear_inode(inode);
376
377         if (S_ISDIR(inode->i_mode))
378                 return;
379
380         ipc_ns = get_ns_from_inode(inode);
381         info = MQUEUE_I(inode);
382         spin_lock(&info->lock);
383         while ((msg = msg_get(info)) != NULL)
384                 free_msg(msg);
385         kfree(info->node_cache);
386         spin_unlock(&info->lock);
387
388         /* Total amount of bytes accounted for the mqueue */
389         mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
390                 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
391                 sizeof(struct posix_msg_tree_node);
392
393         mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
394                                   info->attr.mq_msgsize);
395
396         user = info->user;
397         if (user) {
398                 spin_lock(&mq_lock);
399                 user->mq_bytes -= mq_bytes;
400                 /*
401                  * get_ns_from_inode() ensures that the
402                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
403                  * to which we now hold a reference, or it is NULL.
404                  * We can't put it here under mq_lock, though.
405                  */
406                 if (ipc_ns)
407                         ipc_ns->mq_queues_count--;
408                 spin_unlock(&mq_lock);
409                 free_uid(user);
410         }
411         if (ipc_ns)
412                 put_ipc_ns(ipc_ns);
413 }
414
415 static int mqueue_create(struct inode *dir, struct dentry *dentry,
416                                 umode_t mode, bool excl)
417 {
418         struct inode *inode;
419         struct mq_attr *attr = dentry->d_fsdata;
420         int error;
421         struct ipc_namespace *ipc_ns;
422
423         spin_lock(&mq_lock);
424         ipc_ns = __get_ns_from_inode(dir);
425         if (!ipc_ns) {
426                 error = -EACCES;
427                 goto out_unlock;
428         }
429         if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
430             (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
431              !capable(CAP_SYS_RESOURCE))) {
432                 error = -ENOSPC;
433                 goto out_unlock;
434         }
435         ipc_ns->mq_queues_count++;
436         spin_unlock(&mq_lock);
437
438         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
439         if (IS_ERR(inode)) {
440                 error = PTR_ERR(inode);
441                 spin_lock(&mq_lock);
442                 ipc_ns->mq_queues_count--;
443                 goto out_unlock;
444         }
445
446         put_ipc_ns(ipc_ns);
447         dir->i_size += DIRENT_SIZE;
448         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
449
450         d_instantiate(dentry, inode);
451         dget(dentry);
452         return 0;
453 out_unlock:
454         spin_unlock(&mq_lock);
455         if (ipc_ns)
456                 put_ipc_ns(ipc_ns);
457         return error;
458 }
459
460 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
461 {
462         struct inode *inode = dentry->d_inode;
463
464         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
465         dir->i_size -= DIRENT_SIZE;
466         drop_nlink(inode);
467         dput(dentry);
468         return 0;
469 }
470
471 /*
472 *       This is routine for system read from queue file.
473 *       To avoid mess with doing here some sort of mq_receive we allow
474 *       to read only queue size & notification info (the only values
475 *       that are interesting from user point of view and aren't accessible
476 *       through std routines)
477 */
478 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
479                                 size_t count, loff_t *off)
480 {
481         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
482         char buffer[FILENT_SIZE];
483         ssize_t ret;
484
485         spin_lock(&info->lock);
486         snprintf(buffer, sizeof(buffer),
487                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
488                         info->qsize,
489                         info->notify_owner ? info->notify.sigev_notify : 0,
490                         (info->notify_owner &&
491                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
492                                 info->notify.sigev_signo : 0,
493                         pid_vnr(info->notify_owner));
494         spin_unlock(&info->lock);
495         buffer[sizeof(buffer)-1] = '\0';
496
497         ret = simple_read_from_buffer(u_data, count, off, buffer,
498                                 strlen(buffer));
499         if (ret <= 0)
500                 return ret;
501
502         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
503         return ret;
504 }
505
506 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
507 {
508         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
509
510         spin_lock(&info->lock);
511         if (task_tgid(current) == info->notify_owner)
512                 remove_notification(info);
513
514         spin_unlock(&info->lock);
515         return 0;
516 }
517
518 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
519 {
520         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
521         int retval = 0;
522
523         poll_wait(filp, &info->wait_q, poll_tab);
524
525         spin_lock(&info->lock);
526         if (info->attr.mq_curmsgs)
527                 retval = POLLIN | POLLRDNORM;
528
529         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
530                 retval |= POLLOUT | POLLWRNORM;
531         spin_unlock(&info->lock);
532
533         return retval;
534 }
535
536 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
537 static void wq_add(struct mqueue_inode_info *info, int sr,
538                         struct ext_wait_queue *ewp)
539 {
540         struct ext_wait_queue *walk;
541
542         ewp->task = current;
543
544         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
545                 if (walk->task->static_prio <= current->static_prio) {
546                         list_add_tail(&ewp->list, &walk->list);
547                         return;
548                 }
549         }
550         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
551 }
552
553 /*
554  * Puts current task to sleep. Caller must hold queue lock. After return
555  * lock isn't held.
556  * sr: SEND or RECV
557  */
558 static int wq_sleep(struct mqueue_inode_info *info, int sr,
559                     ktime_t *timeout, struct ext_wait_queue *ewp)
560 {
561         int retval;
562         signed long time;
563
564         wq_add(info, sr, ewp);
565
566         for (;;) {
567                 set_current_state(TASK_INTERRUPTIBLE);
568
569                 spin_unlock(&info->lock);
570                 time = schedule_hrtimeout_range_clock(timeout, 0,
571                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
572
573                 while (ewp->state == STATE_PENDING)
574                         cpu_relax();
575
576                 if (ewp->state == STATE_READY) {
577                         retval = 0;
578                         goto out;
579                 }
580                 spin_lock(&info->lock);
581                 if (ewp->state == STATE_READY) {
582                         retval = 0;
583                         goto out_unlock;
584                 }
585                 if (signal_pending(current)) {
586                         retval = -ERESTARTSYS;
587                         break;
588                 }
589                 if (time == 0) {
590                         retval = -ETIMEDOUT;
591                         break;
592                 }
593         }
594         list_del(&ewp->list);
595 out_unlock:
596         spin_unlock(&info->lock);
597 out:
598         return retval;
599 }
600
601 /*
602  * Returns waiting task that should be serviced first or NULL if none exists
603  */
604 static struct ext_wait_queue *wq_get_first_waiter(
605                 struct mqueue_inode_info *info, int sr)
606 {
607         struct list_head *ptr;
608
609         ptr = info->e_wait_q[sr].list.prev;
610         if (ptr == &info->e_wait_q[sr].list)
611                 return NULL;
612         return list_entry(ptr, struct ext_wait_queue, list);
613 }
614
615
616 static inline void set_cookie(struct sk_buff *skb, char code)
617 {
618         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
619 }
620
621 /*
622  * The next function is only to split too long sys_mq_timedsend
623  */
624 static void __do_notify(struct mqueue_inode_info *info)
625 {
626         /* notification
627          * invoked when there is registered process and there isn't process
628          * waiting synchronously for message AND state of queue changed from
629          * empty to not empty. Here we are sure that no one is waiting
630          * synchronously. */
631         if (info->notify_owner &&
632             info->attr.mq_curmsgs == 1) {
633                 struct siginfo sig_i;
634                 switch (info->notify.sigev_notify) {
635                 case SIGEV_NONE:
636                         break;
637                 case SIGEV_SIGNAL:
638                         /* sends signal */
639
640                         sig_i.si_signo = info->notify.sigev_signo;
641                         sig_i.si_errno = 0;
642                         sig_i.si_code = SI_MESGQ;
643                         sig_i.si_value = info->notify.sigev_value;
644                         /* map current pid/uid into info->owner's namespaces */
645                         rcu_read_lock();
646                         sig_i.si_pid = task_tgid_nr_ns(current,
647                                                 ns_of_pid(info->notify_owner));
648                         sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
649                         rcu_read_unlock();
650
651                         kill_pid_info(info->notify.sigev_signo,
652                                       &sig_i, info->notify_owner);
653                         break;
654                 case SIGEV_THREAD:
655                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
656                         netlink_sendskb(info->notify_sock, info->notify_cookie);
657                         break;
658                 }
659                 /* after notification unregisters process */
660                 put_pid(info->notify_owner);
661                 put_user_ns(info->notify_user_ns);
662                 info->notify_owner = NULL;
663                 info->notify_user_ns = NULL;
664         }
665         wake_up(&info->wait_q);
666 }
667
668 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
669                            ktime_t *expires, struct timespec *ts)
670 {
671         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
672                 return -EFAULT;
673         if (!timespec_valid(ts))
674                 return -EINVAL;
675
676         *expires = timespec_to_ktime(*ts);
677         return 0;
678 }
679
680 static void remove_notification(struct mqueue_inode_info *info)
681 {
682         if (info->notify_owner != NULL &&
683             info->notify.sigev_notify == SIGEV_THREAD) {
684                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
685                 netlink_sendskb(info->notify_sock, info->notify_cookie);
686         }
687         put_pid(info->notify_owner);
688         put_user_ns(info->notify_user_ns);
689         info->notify_owner = NULL;
690         info->notify_user_ns = NULL;
691 }
692
693 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
694 {
695         int mq_treesize;
696         unsigned long total_size;
697
698         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
699                 return -EINVAL;
700         if (capable(CAP_SYS_RESOURCE)) {
701                 if (attr->mq_maxmsg > HARD_MSGMAX ||
702                     attr->mq_msgsize > HARD_MSGSIZEMAX)
703                         return -EINVAL;
704         } else {
705                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
706                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
707                         return -EINVAL;
708         }
709         /* check for overflow */
710         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
711                 return -EOVERFLOW;
712         mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
713                 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
714                 sizeof(struct posix_msg_tree_node);
715         total_size = attr->mq_maxmsg * attr->mq_msgsize;
716         if (total_size + mq_treesize < total_size)
717                 return -EOVERFLOW;
718         return 0;
719 }
720
721 /*
722  * Invoked when creating a new queue via sys_mq_open
723  */
724 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
725                         struct path *path, int oflag, umode_t mode,
726                         struct mq_attr *attr)
727 {
728         const struct cred *cred = current_cred();
729         struct file *result;
730         int ret;
731
732         if (attr) {
733                 ret = mq_attr_ok(ipc_ns, attr);
734                 if (ret)
735                         return ERR_PTR(ret);
736                 /* store for use during create */
737                 path->dentry->d_fsdata = attr;
738         } else {
739                 struct mq_attr def_attr;
740
741                 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
742                                          ipc_ns->mq_msg_default);
743                 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
744                                           ipc_ns->mq_msgsize_default);
745                 ret = mq_attr_ok(ipc_ns, &def_attr);
746                 if (ret)
747                         return ERR_PTR(ret);
748         }
749
750         mode &= ~current_umask();
751         ret = mnt_want_write(path->mnt);
752         if (ret)
753                 return ERR_PTR(ret);
754         ret = vfs_create(dir, path->dentry, mode, true);
755         path->dentry->d_fsdata = NULL;
756         if (!ret)
757                 result = dentry_open(path, oflag, cred);
758         else
759                 result = ERR_PTR(ret);
760         /*
761          * dentry_open() took a persistent mnt_want_write(),
762          * so we can now drop this one.
763          */
764         mnt_drop_write(path->mnt);
765         return result;
766 }
767
768 /* Opens existing queue */
769 static struct file *do_open(struct path *path, int oflag)
770 {
771         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
772                                                   MAY_READ | MAY_WRITE };
773         int acc;
774         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
775                 return ERR_PTR(-EINVAL);
776         acc = oflag2acc[oflag & O_ACCMODE];
777         if (inode_permission(path->dentry->d_inode, acc))
778                 return ERR_PTR(-EACCES);
779         return dentry_open(path, oflag, current_cred());
780 }
781
782 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
783                 struct mq_attr __user *, u_attr)
784 {
785         struct path path;
786         struct file *filp;
787         char *name;
788         struct mq_attr attr;
789         int fd, error;
790         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
791         struct dentry *root = ipc_ns->mq_mnt->mnt_root;
792
793         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
794                 return -EFAULT;
795
796         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
797
798         if (IS_ERR(name = getname(u_name)))
799                 return PTR_ERR(name);
800
801         fd = get_unused_fd_flags(O_CLOEXEC);
802         if (fd < 0)
803                 goto out_putname;
804
805         error = 0;
806         mutex_lock(&root->d_inode->i_mutex);
807         path.dentry = lookup_one_len(name, root, strlen(name));
808         if (IS_ERR(path.dentry)) {
809                 error = PTR_ERR(path.dentry);
810                 goto out_putfd;
811         }
812         path.mnt = mntget(ipc_ns->mq_mnt);
813
814         if (oflag & O_CREAT) {
815                 if (path.dentry->d_inode) {     /* entry already exists */
816                         audit_inode(name, path.dentry);
817                         if (oflag & O_EXCL) {
818                                 error = -EEXIST;
819                                 goto out;
820                         }
821                         filp = do_open(&path, oflag);
822                 } else {
823                         filp = do_create(ipc_ns, root->d_inode,
824                                                 &path, oflag, mode,
825                                                 u_attr ? &attr : NULL);
826                 }
827         } else {
828                 if (!path.dentry->d_inode) {
829                         error = -ENOENT;
830                         goto out;
831                 }
832                 audit_inode(name, path.dentry);
833                 filp = do_open(&path, oflag);
834         }
835
836         if (!IS_ERR(filp))
837                 fd_install(fd, filp);
838         else
839                 error = PTR_ERR(filp);
840 out:
841         path_put(&path);
842 out_putfd:
843         if (error) {
844                 put_unused_fd(fd);
845                 fd = error;
846         }
847         mutex_unlock(&root->d_inode->i_mutex);
848 out_putname:
849         putname(name);
850         return fd;
851 }
852
853 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
854 {
855         int err;
856         char *name;
857         struct dentry *dentry;
858         struct inode *inode = NULL;
859         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
860
861         name = getname(u_name);
862         if (IS_ERR(name))
863                 return PTR_ERR(name);
864
865         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
866                         I_MUTEX_PARENT);
867         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
868         if (IS_ERR(dentry)) {
869                 err = PTR_ERR(dentry);
870                 goto out_unlock;
871         }
872
873         if (!dentry->d_inode) {
874                 err = -ENOENT;
875                 goto out_err;
876         }
877
878         inode = dentry->d_inode;
879         if (inode)
880                 ihold(inode);
881         err = mnt_want_write(ipc_ns->mq_mnt);
882         if (err)
883                 goto out_err;
884         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
885         mnt_drop_write(ipc_ns->mq_mnt);
886 out_err:
887         dput(dentry);
888
889 out_unlock:
890         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
891         putname(name);
892         if (inode)
893                 iput(inode);
894
895         return err;
896 }
897
898 /* Pipelined send and receive functions.
899  *
900  * If a receiver finds no waiting message, then it registers itself in the
901  * list of waiting receivers. A sender checks that list before adding the new
902  * message into the message array. If there is a waiting receiver, then it
903  * bypasses the message array and directly hands the message over to the
904  * receiver.
905  * The receiver accepts the message and returns without grabbing the queue
906  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
907  * are necessary. The same algorithm is used for sysv semaphores, see
908  * ipc/sem.c for more details.
909  *
910  * The same algorithm is used for senders.
911  */
912
913 /* pipelined_send() - send a message directly to the task waiting in
914  * sys_mq_timedreceive() (without inserting message into a queue).
915  */
916 static inline void pipelined_send(struct mqueue_inode_info *info,
917                                   struct msg_msg *message,
918                                   struct ext_wait_queue *receiver)
919 {
920         receiver->msg = message;
921         list_del(&receiver->list);
922         receiver->state = STATE_PENDING;
923         wake_up_process(receiver->task);
924         smp_wmb();
925         receiver->state = STATE_READY;
926 }
927
928 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
929  * gets its message and put to the queue (we have one free place for sure). */
930 static inline void pipelined_receive(struct mqueue_inode_info *info)
931 {
932         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
933
934         if (!sender) {
935                 /* for poll */
936                 wake_up_interruptible(&info->wait_q);
937                 return;
938         }
939         if (msg_insert(sender->msg, info))
940                 return;
941         list_del(&sender->list);
942         sender->state = STATE_PENDING;
943         wake_up_process(sender->task);
944         smp_wmb();
945         sender->state = STATE_READY;
946 }
947
948 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
949                 size_t, msg_len, unsigned int, msg_prio,
950                 const struct timespec __user *, u_abs_timeout)
951 {
952         struct file *filp;
953         struct inode *inode;
954         struct ext_wait_queue wait;
955         struct ext_wait_queue *receiver;
956         struct msg_msg *msg_ptr;
957         struct mqueue_inode_info *info;
958         ktime_t expires, *timeout = NULL;
959         struct timespec ts;
960         struct posix_msg_tree_node *new_leaf = NULL;
961         int ret = 0;
962
963         if (u_abs_timeout) {
964                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
965                 if (res)
966                         return res;
967                 timeout = &expires;
968         }
969
970         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
971                 return -EINVAL;
972
973         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
974
975         filp = fget(mqdes);
976         if (unlikely(!filp)) {
977                 ret = -EBADF;
978                 goto out;
979         }
980
981         inode = filp->f_path.dentry->d_inode;
982         if (unlikely(filp->f_op != &mqueue_file_operations)) {
983                 ret = -EBADF;
984                 goto out_fput;
985         }
986         info = MQUEUE_I(inode);
987         audit_inode(NULL, filp->f_path.dentry);
988
989         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
990                 ret = -EBADF;
991                 goto out_fput;
992         }
993
994         if (unlikely(msg_len > info->attr.mq_msgsize)) {
995                 ret = -EMSGSIZE;
996                 goto out_fput;
997         }
998
999         /* First try to allocate memory, before doing anything with
1000          * existing queues. */
1001         msg_ptr = load_msg(u_msg_ptr, msg_len);
1002         if (IS_ERR(msg_ptr)) {
1003                 ret = PTR_ERR(msg_ptr);
1004                 goto out_fput;
1005         }
1006         msg_ptr->m_ts = msg_len;
1007         msg_ptr->m_type = msg_prio;
1008
1009         /*
1010          * msg_insert really wants us to have a valid, spare node struct so
1011          * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1012          * fall back to that if necessary.
1013          */
1014         if (!info->node_cache)
1015                 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1016
1017         spin_lock(&info->lock);
1018
1019         if (!info->node_cache && new_leaf) {
1020                 /* Save our speculative allocation into the cache */
1021                 rb_init_node(&new_leaf->rb_node);
1022                 INIT_LIST_HEAD(&new_leaf->msg_list);
1023                 info->node_cache = new_leaf;
1024                 info->qsize += sizeof(*new_leaf);
1025                 new_leaf = NULL;
1026         } else {
1027                 kfree(new_leaf);
1028         }
1029
1030         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1031                 if (filp->f_flags & O_NONBLOCK) {
1032                         ret = -EAGAIN;
1033                 } else {
1034                         wait.task = current;
1035                         wait.msg = (void *) msg_ptr;
1036                         wait.state = STATE_NONE;
1037                         ret = wq_sleep(info, SEND, timeout, &wait);
1038                         /*
1039                          * wq_sleep must be called with info->lock held, and
1040                          * returns with the lock released
1041                          */
1042                         goto out_free;
1043                 }
1044         } else {
1045                 receiver = wq_get_first_waiter(info, RECV);
1046                 if (receiver) {
1047                         pipelined_send(info, msg_ptr, receiver);
1048                 } else {
1049                         /* adds message to the queue */
1050                         ret = msg_insert(msg_ptr, info);
1051                         if (ret)
1052                                 goto out_unlock;
1053                         __do_notify(info);
1054                 }
1055                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1056                                 CURRENT_TIME;
1057         }
1058 out_unlock:
1059         spin_unlock(&info->lock);
1060 out_free:
1061         if (ret)
1062                 free_msg(msg_ptr);
1063 out_fput:
1064         fput(filp);
1065 out:
1066         return ret;
1067 }
1068
1069 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1070                 size_t, msg_len, unsigned int __user *, u_msg_prio,
1071                 const struct timespec __user *, u_abs_timeout)
1072 {
1073         ssize_t ret;
1074         struct msg_msg *msg_ptr;
1075         struct file *filp;
1076         struct inode *inode;
1077         struct mqueue_inode_info *info;
1078         struct ext_wait_queue wait;
1079         ktime_t expires, *timeout = NULL;
1080         struct timespec ts;
1081         struct posix_msg_tree_node *new_leaf = NULL;
1082
1083         if (u_abs_timeout) {
1084                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1085                 if (res)
1086                         return res;
1087                 timeout = &expires;
1088         }
1089
1090         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1091
1092         filp = fget(mqdes);
1093         if (unlikely(!filp)) {
1094                 ret = -EBADF;
1095                 goto out;
1096         }
1097
1098         inode = filp->f_path.dentry->d_inode;
1099         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1100                 ret = -EBADF;
1101                 goto out_fput;
1102         }
1103         info = MQUEUE_I(inode);
1104         audit_inode(NULL, filp->f_path.dentry);
1105
1106         if (unlikely(!(filp->f_mode & FMODE_READ))) {
1107                 ret = -EBADF;
1108                 goto out_fput;
1109         }
1110
1111         /* checks if buffer is big enough */
1112         if (unlikely(msg_len < info->attr.mq_msgsize)) {
1113                 ret = -EMSGSIZE;
1114                 goto out_fput;
1115         }
1116
1117         /*
1118          * msg_insert really wants us to have a valid, spare node struct so
1119          * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1120          * fall back to that if necessary.
1121          */
1122         if (!info->node_cache)
1123                 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1124
1125         spin_lock(&info->lock);
1126
1127         if (!info->node_cache && new_leaf) {
1128                 /* Save our speculative allocation into the cache */
1129                 rb_init_node(&new_leaf->rb_node);
1130                 INIT_LIST_HEAD(&new_leaf->msg_list);
1131                 info->node_cache = new_leaf;
1132                 info->qsize += sizeof(*new_leaf);
1133         } else {
1134                 kfree(new_leaf);
1135         }
1136
1137         if (info->attr.mq_curmsgs == 0) {
1138                 if (filp->f_flags & O_NONBLOCK) {
1139                         spin_unlock(&info->lock);
1140                         ret = -EAGAIN;
1141                 } else {
1142                         wait.task = current;
1143                         wait.state = STATE_NONE;
1144                         ret = wq_sleep(info, RECV, timeout, &wait);
1145                         msg_ptr = wait.msg;
1146                 }
1147         } else {
1148                 msg_ptr = msg_get(info);
1149
1150                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1151                                 CURRENT_TIME;
1152
1153                 /* There is now free space in queue. */
1154                 pipelined_receive(info);
1155                 spin_unlock(&info->lock);
1156                 ret = 0;
1157         }
1158         if (ret == 0) {
1159                 ret = msg_ptr->m_ts;
1160
1161                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1162                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1163                         ret = -EFAULT;
1164                 }
1165                 free_msg(msg_ptr);
1166         }
1167 out_fput:
1168         fput(filp);
1169 out:
1170         return ret;
1171 }
1172
1173 /*
1174  * Notes: the case when user wants us to deregister (with NULL as pointer)
1175  * and he isn't currently owner of notification, will be silently discarded.
1176  * It isn't explicitly defined in the POSIX.
1177  */
1178 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1179                 const struct sigevent __user *, u_notification)
1180 {
1181         int ret;
1182         struct file *filp;
1183         struct sock *sock;
1184         struct inode *inode;
1185         struct sigevent notification;
1186         struct mqueue_inode_info *info;
1187         struct sk_buff *nc;
1188
1189         if (u_notification) {
1190                 if (copy_from_user(&notification, u_notification,
1191                                         sizeof(struct sigevent)))
1192                         return -EFAULT;
1193         }
1194
1195         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1196
1197         nc = NULL;
1198         sock = NULL;
1199         if (u_notification != NULL) {
1200                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1201                              notification.sigev_notify != SIGEV_SIGNAL &&
1202                              notification.sigev_notify != SIGEV_THREAD))
1203                         return -EINVAL;
1204                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1205                         !valid_signal(notification.sigev_signo)) {
1206                         return -EINVAL;
1207                 }
1208                 if (notification.sigev_notify == SIGEV_THREAD) {
1209                         long timeo;
1210
1211                         /* create the notify skb */
1212                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1213                         if (!nc) {
1214                                 ret = -ENOMEM;
1215                                 goto out;
1216                         }
1217                         if (copy_from_user(nc->data,
1218                                         notification.sigev_value.sival_ptr,
1219                                         NOTIFY_COOKIE_LEN)) {
1220                                 ret = -EFAULT;
1221                                 goto out;
1222                         }
1223
1224                         /* TODO: add a header? */
1225                         skb_put(nc, NOTIFY_COOKIE_LEN);
1226                         /* and attach it to the socket */
1227 retry:
1228                         filp = fget(notification.sigev_signo);
1229                         if (!filp) {
1230                                 ret = -EBADF;
1231                                 goto out;
1232                         }
1233                         sock = netlink_getsockbyfilp(filp);
1234                         fput(filp);
1235                         if (IS_ERR(sock)) {
1236                                 ret = PTR_ERR(sock);
1237                                 sock = NULL;
1238                                 goto out;
1239                         }
1240
1241                         timeo = MAX_SCHEDULE_TIMEOUT;
1242                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1243                         if (ret == 1)
1244                                 goto retry;
1245                         if (ret) {
1246                                 sock = NULL;
1247                                 nc = NULL;
1248                                 goto out;
1249                         }
1250                 }
1251         }
1252
1253         filp = fget(mqdes);
1254         if (!filp) {
1255                 ret = -EBADF;
1256                 goto out;
1257         }
1258
1259         inode = filp->f_path.dentry->d_inode;
1260         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1261                 ret = -EBADF;
1262                 goto out_fput;
1263         }
1264         info = MQUEUE_I(inode);
1265
1266         ret = 0;
1267         spin_lock(&info->lock);
1268         if (u_notification == NULL) {
1269                 if (info->notify_owner == task_tgid(current)) {
1270                         remove_notification(info);
1271                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1272                 }
1273         } else if (info->notify_owner != NULL) {
1274                 ret = -EBUSY;
1275         } else {
1276                 switch (notification.sigev_notify) {
1277                 case SIGEV_NONE:
1278                         info->notify.sigev_notify = SIGEV_NONE;
1279                         break;
1280                 case SIGEV_THREAD:
1281                         info->notify_sock = sock;
1282                         info->notify_cookie = nc;
1283                         sock = NULL;
1284                         nc = NULL;
1285                         info->notify.sigev_notify = SIGEV_THREAD;
1286                         break;
1287                 case SIGEV_SIGNAL:
1288                         info->notify.sigev_signo = notification.sigev_signo;
1289                         info->notify.sigev_value = notification.sigev_value;
1290                         info->notify.sigev_notify = SIGEV_SIGNAL;
1291                         break;
1292                 }
1293
1294                 info->notify_owner = get_pid(task_tgid(current));
1295                 info->notify_user_ns = get_user_ns(current_user_ns());
1296                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1297         }
1298         spin_unlock(&info->lock);
1299 out_fput:
1300         fput(filp);
1301 out:
1302         if (sock) {
1303                 netlink_detachskb(sock, nc);
1304         } else if (nc) {
1305                 dev_kfree_skb(nc);
1306         }
1307         return ret;
1308 }
1309
1310 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1311                 const struct mq_attr __user *, u_mqstat,
1312                 struct mq_attr __user *, u_omqstat)
1313 {
1314         int ret;
1315         struct mq_attr mqstat, omqstat;
1316         struct file *filp;
1317         struct inode *inode;
1318         struct mqueue_inode_info *info;
1319
1320         if (u_mqstat != NULL) {
1321                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1322                         return -EFAULT;
1323                 if (mqstat.mq_flags & (~O_NONBLOCK))
1324                         return -EINVAL;
1325         }
1326
1327         filp = fget(mqdes);
1328         if (!filp) {
1329                 ret = -EBADF;
1330                 goto out;
1331         }
1332
1333         inode = filp->f_path.dentry->d_inode;
1334         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1335                 ret = -EBADF;
1336                 goto out_fput;
1337         }
1338         info = MQUEUE_I(inode);
1339
1340         spin_lock(&info->lock);
1341
1342         omqstat = info->attr;
1343         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1344         if (u_mqstat) {
1345                 audit_mq_getsetattr(mqdes, &mqstat);
1346                 spin_lock(&filp->f_lock);
1347                 if (mqstat.mq_flags & O_NONBLOCK)
1348                         filp->f_flags |= O_NONBLOCK;
1349                 else
1350                         filp->f_flags &= ~O_NONBLOCK;
1351                 spin_unlock(&filp->f_lock);
1352
1353                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1354         }
1355
1356         spin_unlock(&info->lock);
1357
1358         ret = 0;
1359         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1360                                                 sizeof(struct mq_attr)))
1361                 ret = -EFAULT;
1362
1363 out_fput:
1364         fput(filp);
1365 out:
1366         return ret;
1367 }
1368
1369 static const struct inode_operations mqueue_dir_inode_operations = {
1370         .lookup = simple_lookup,
1371         .create = mqueue_create,
1372         .unlink = mqueue_unlink,
1373 };
1374
1375 static const struct file_operations mqueue_file_operations = {
1376         .flush = mqueue_flush_file,
1377         .poll = mqueue_poll_file,
1378         .read = mqueue_read_file,
1379         .llseek = default_llseek,
1380 };
1381
1382 static const struct super_operations mqueue_super_ops = {
1383         .alloc_inode = mqueue_alloc_inode,
1384         .destroy_inode = mqueue_destroy_inode,
1385         .evict_inode = mqueue_evict_inode,
1386         .statfs = simple_statfs,
1387 };
1388
1389 static struct file_system_type mqueue_fs_type = {
1390         .name = "mqueue",
1391         .mount = mqueue_mount,
1392         .kill_sb = kill_litter_super,
1393 };
1394
1395 int mq_init_ns(struct ipc_namespace *ns)
1396 {
1397         ns->mq_queues_count  = 0;
1398         ns->mq_queues_max    = DFLT_QUEUESMAX;
1399         ns->mq_msg_max       = DFLT_MSGMAX;
1400         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1401         ns->mq_msg_default   = DFLT_MSG;
1402         ns->mq_msgsize_default  = DFLT_MSGSIZE;
1403
1404         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1405         if (IS_ERR(ns->mq_mnt)) {
1406                 int err = PTR_ERR(ns->mq_mnt);
1407                 ns->mq_mnt = NULL;
1408                 return err;
1409         }
1410         return 0;
1411 }
1412
1413 void mq_clear_sbinfo(struct ipc_namespace *ns)
1414 {
1415         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1416 }
1417
1418 void mq_put_mnt(struct ipc_namespace *ns)
1419 {
1420         kern_unmount(ns->mq_mnt);
1421 }
1422
1423 static int __init init_mqueue_fs(void)
1424 {
1425         int error;
1426
1427         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1428                                 sizeof(struct mqueue_inode_info), 0,
1429                                 SLAB_HWCACHE_ALIGN, init_once);
1430         if (mqueue_inode_cachep == NULL)
1431                 return -ENOMEM;
1432
1433         /* ignore failures - they are not fatal */
1434         mq_sysctl_table = mq_register_sysctl_table();
1435
1436         error = register_filesystem(&mqueue_fs_type);
1437         if (error)
1438                 goto out_sysctl;
1439
1440         spin_lock_init(&mq_lock);
1441
1442         error = mq_init_ns(&init_ipc_ns);
1443         if (error)
1444                 goto out_filesystem;
1445
1446         return 0;
1447
1448 out_filesystem:
1449         unregister_filesystem(&mqueue_fs_type);
1450 out_sysctl:
1451         if (mq_sysctl_table)
1452                 unregister_sysctl_table(mq_sysctl_table);
1453         kmem_cache_destroy(mqueue_inode_cachep);
1454         return error;
1455 }
1456
1457 __initcall(init_mqueue_fs);