mqueue: fix mq_open() file descriptor leak on user-space processes
[linux-2.6.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/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
36 #include <net/sock.h>
37 #include "util.h"
38
39 #define MQUEUE_MAGIC    0x19800202
40 #define DIRENT_SIZE     20
41 #define FILENT_SIZE     80
42
43 #define SEND            0
44 #define RECV            1
45
46 #define STATE_NONE      0
47 #define STATE_PENDING   1
48 #define STATE_READY     2
49
50 struct ext_wait_queue {         /* queue of sleeping tasks */
51         struct task_struct *task;
52         struct list_head list;
53         struct msg_msg *msg;    /* ptr of loaded message */
54         int state;              /* one of STATE_* values */
55 };
56
57 struct mqueue_inode_info {
58         spinlock_t lock;
59         struct inode vfs_inode;
60         wait_queue_head_t wait_q;
61
62         struct msg_msg **messages;
63         struct mq_attr attr;
64
65         struct sigevent notify;
66         struct pid* notify_owner;
67         struct user_struct *user;       /* user who created, for accounting */
68         struct sock *notify_sock;
69         struct sk_buff *notify_cookie;
70
71         /* for tasks waiting for free space and messages, respectively */
72         struct ext_wait_queue e_wait_q[2];
73
74         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
75 };
76
77 static const struct inode_operations mqueue_dir_inode_operations;
78 static const struct file_operations mqueue_file_operations;
79 static const struct super_operations mqueue_super_ops;
80 static void remove_notification(struct mqueue_inode_info *info);
81
82 static struct kmem_cache *mqueue_inode_cachep;
83
84 static struct ctl_table_header * mq_sysctl_table;
85
86 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
87 {
88         return container_of(inode, struct mqueue_inode_info, vfs_inode);
89 }
90
91 /*
92  * This routine should be called with the mq_lock held.
93  */
94 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
95 {
96         return get_ipc_ns(inode->i_sb->s_fs_info);
97 }
98
99 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
100 {
101         struct ipc_namespace *ns;
102
103         spin_lock(&mq_lock);
104         ns = __get_ns_from_inode(inode);
105         spin_unlock(&mq_lock);
106         return ns;
107 }
108
109 static struct inode *mqueue_get_inode(struct super_block *sb,
110                 struct ipc_namespace *ipc_ns, int mode,
111                 struct mq_attr *attr)
112 {
113         struct user_struct *u = current_user();
114         struct inode *inode;
115
116         inode = new_inode(sb);
117         if (inode) {
118                 inode->i_mode = mode;
119                 inode->i_uid = current_fsuid();
120                 inode->i_gid = current_fsgid();
121                 inode->i_mtime = inode->i_ctime = inode->i_atime =
122                                 CURRENT_TIME;
123
124                 if (S_ISREG(mode)) {
125                         struct mqueue_inode_info *info;
126                         struct task_struct *p = current;
127                         unsigned long mq_bytes, mq_msg_tblsz;
128
129                         inode->i_fop = &mqueue_file_operations;
130                         inode->i_size = FILENT_SIZE;
131                         /* mqueue specific info */
132                         info = MQUEUE_I(inode);
133                         spin_lock_init(&info->lock);
134                         init_waitqueue_head(&info->wait_q);
135                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
136                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
137                         info->messages = NULL;
138                         info->notify_owner = NULL;
139                         info->qsize = 0;
140                         info->user = NULL;      /* set when all is ok */
141                         memset(&info->attr, 0, sizeof(info->attr));
142                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
143                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
144                         if (attr) {
145                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
146                                 info->attr.mq_msgsize = attr->mq_msgsize;
147                         }
148                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
149                         mq_bytes = (mq_msg_tblsz +
150                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
151
152                         spin_lock(&mq_lock);
153                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
154                             u->mq_bytes + mq_bytes >
155                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
156                                 spin_unlock(&mq_lock);
157                                 goto out_inode;
158                         }
159                         u->mq_bytes += mq_bytes;
160                         spin_unlock(&mq_lock);
161
162                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
163                         if (!info->messages) {
164                                 spin_lock(&mq_lock);
165                                 u->mq_bytes -= mq_bytes;
166                                 spin_unlock(&mq_lock);
167                                 goto out_inode;
168                         }
169                         /* all is ok */
170                         info->user = get_uid(u);
171                 } else if (S_ISDIR(mode)) {
172                         inc_nlink(inode);
173                         /* Some things misbehave if size == 0 on a directory */
174                         inode->i_size = 2 * DIRENT_SIZE;
175                         inode->i_op = &mqueue_dir_inode_operations;
176                         inode->i_fop = &simple_dir_operations;
177                 }
178         }
179         return inode;
180 out_inode:
181         make_bad_inode(inode);
182         iput(inode);
183         return NULL;
184 }
185
186 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
187 {
188         struct inode *inode;
189         struct ipc_namespace *ns = data;
190         int error = 0;
191
192         sb->s_blocksize = PAGE_CACHE_SIZE;
193         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
194         sb->s_magic = MQUEUE_MAGIC;
195         sb->s_op = &mqueue_super_ops;
196
197         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
198                                 NULL);
199         if (!inode) {
200                 error = -ENOMEM;
201                 goto out;
202         }
203
204         sb->s_root = d_alloc_root(inode);
205         if (!sb->s_root) {
206                 iput(inode);
207                 error = -ENOMEM;
208         }
209
210 out:
211         return error;
212 }
213
214 static int mqueue_get_sb(struct file_system_type *fs_type,
215                          int flags, const char *dev_name,
216                          void *data, struct vfsmount *mnt)
217 {
218         if (!(flags & MS_KERNMOUNT))
219                 data = current->nsproxy->ipc_ns;
220         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
221 }
222
223 static void init_once(void *foo)
224 {
225         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
226
227         inode_init_once(&p->vfs_inode);
228 }
229
230 static struct inode *mqueue_alloc_inode(struct super_block *sb)
231 {
232         struct mqueue_inode_info *ei;
233
234         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
235         if (!ei)
236                 return NULL;
237         return &ei->vfs_inode;
238 }
239
240 static void mqueue_destroy_inode(struct inode *inode)
241 {
242         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
243 }
244
245 static void mqueue_delete_inode(struct inode *inode)
246 {
247         struct mqueue_inode_info *info;
248         struct user_struct *user;
249         unsigned long mq_bytes;
250         int i;
251         struct ipc_namespace *ipc_ns;
252
253         if (S_ISDIR(inode->i_mode)) {
254                 clear_inode(inode);
255                 return;
256         }
257         ipc_ns = get_ns_from_inode(inode);
258         info = MQUEUE_I(inode);
259         spin_lock(&info->lock);
260         for (i = 0; i < info->attr.mq_curmsgs; i++)
261                 free_msg(info->messages[i]);
262         kfree(info->messages);
263         spin_unlock(&info->lock);
264
265         clear_inode(inode);
266
267         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
268                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
269         user = info->user;
270         if (user) {
271                 spin_lock(&mq_lock);
272                 user->mq_bytes -= mq_bytes;
273                 /*
274                  * get_ns_from_inode() ensures that the
275                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
276                  * to which we now hold a reference, or it is NULL.
277                  * We can't put it here under mq_lock, though.
278                  */
279                 if (ipc_ns)
280                         ipc_ns->mq_queues_count--;
281                 spin_unlock(&mq_lock);
282                 free_uid(user);
283         }
284         if (ipc_ns)
285                 put_ipc_ns(ipc_ns);
286 }
287
288 static int mqueue_create(struct inode *dir, struct dentry *dentry,
289                                 int mode, struct nameidata *nd)
290 {
291         struct inode *inode;
292         struct mq_attr *attr = dentry->d_fsdata;
293         int error;
294         struct ipc_namespace *ipc_ns;
295
296         spin_lock(&mq_lock);
297         ipc_ns = __get_ns_from_inode(dir);
298         if (!ipc_ns) {
299                 error = -EACCES;
300                 goto out_unlock;
301         }
302         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
303                         !capable(CAP_SYS_RESOURCE)) {
304                 error = -ENOSPC;
305                 goto out_unlock;
306         }
307         ipc_ns->mq_queues_count++;
308         spin_unlock(&mq_lock);
309
310         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
311         if (!inode) {
312                 error = -ENOMEM;
313                 spin_lock(&mq_lock);
314                 ipc_ns->mq_queues_count--;
315                 goto out_unlock;
316         }
317
318         put_ipc_ns(ipc_ns);
319         dir->i_size += DIRENT_SIZE;
320         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
321
322         d_instantiate(dentry, inode);
323         dget(dentry);
324         return 0;
325 out_unlock:
326         spin_unlock(&mq_lock);
327         if (ipc_ns)
328                 put_ipc_ns(ipc_ns);
329         return error;
330 }
331
332 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
333 {
334         struct inode *inode = dentry->d_inode;
335
336         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
337         dir->i_size -= DIRENT_SIZE;
338         drop_nlink(inode);
339         dput(dentry);
340         return 0;
341 }
342
343 /*
344 *       This is routine for system read from queue file.
345 *       To avoid mess with doing here some sort of mq_receive we allow
346 *       to read only queue size & notification info (the only values
347 *       that are interesting from user point of view and aren't accessible
348 *       through std routines)
349 */
350 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
351                                 size_t count, loff_t *off)
352 {
353         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
354         char buffer[FILENT_SIZE];
355         ssize_t ret;
356
357         spin_lock(&info->lock);
358         snprintf(buffer, sizeof(buffer),
359                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
360                         info->qsize,
361                         info->notify_owner ? info->notify.sigev_notify : 0,
362                         (info->notify_owner &&
363                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
364                                 info->notify.sigev_signo : 0,
365                         pid_vnr(info->notify_owner));
366         spin_unlock(&info->lock);
367         buffer[sizeof(buffer)-1] = '\0';
368
369         ret = simple_read_from_buffer(u_data, count, off, buffer,
370                                 strlen(buffer));
371         if (ret <= 0)
372                 return ret;
373
374         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
375         return ret;
376 }
377
378 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
379 {
380         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
381
382         spin_lock(&info->lock);
383         if (task_tgid(current) == info->notify_owner)
384                 remove_notification(info);
385
386         spin_unlock(&info->lock);
387         return 0;
388 }
389
390 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
391 {
392         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
393         int retval = 0;
394
395         poll_wait(filp, &info->wait_q, poll_tab);
396
397         spin_lock(&info->lock);
398         if (info->attr.mq_curmsgs)
399                 retval = POLLIN | POLLRDNORM;
400
401         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
402                 retval |= POLLOUT | POLLWRNORM;
403         spin_unlock(&info->lock);
404
405         return retval;
406 }
407
408 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
409 static void wq_add(struct mqueue_inode_info *info, int sr,
410                         struct ext_wait_queue *ewp)
411 {
412         struct ext_wait_queue *walk;
413
414         ewp->task = current;
415
416         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
417                 if (walk->task->static_prio <= current->static_prio) {
418                         list_add_tail(&ewp->list, &walk->list);
419                         return;
420                 }
421         }
422         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
423 }
424
425 /*
426  * Puts current task to sleep. Caller must hold queue lock. After return
427  * lock isn't held.
428  * sr: SEND or RECV
429  */
430 static int wq_sleep(struct mqueue_inode_info *info, int sr,
431                         long timeout, struct ext_wait_queue *ewp)
432 {
433         int retval;
434         signed long time;
435
436         wq_add(info, sr, ewp);
437
438         for (;;) {
439                 set_current_state(TASK_INTERRUPTIBLE);
440
441                 spin_unlock(&info->lock);
442                 time = schedule_timeout(timeout);
443
444                 while (ewp->state == STATE_PENDING)
445                         cpu_relax();
446
447                 if (ewp->state == STATE_READY) {
448                         retval = 0;
449                         goto out;
450                 }
451                 spin_lock(&info->lock);
452                 if (ewp->state == STATE_READY) {
453                         retval = 0;
454                         goto out_unlock;
455                 }
456                 if (signal_pending(current)) {
457                         retval = -ERESTARTSYS;
458                         break;
459                 }
460                 if (time == 0) {
461                         retval = -ETIMEDOUT;
462                         break;
463                 }
464         }
465         list_del(&ewp->list);
466 out_unlock:
467         spin_unlock(&info->lock);
468 out:
469         return retval;
470 }
471
472 /*
473  * Returns waiting task that should be serviced first or NULL if none exists
474  */
475 static struct ext_wait_queue *wq_get_first_waiter(
476                 struct mqueue_inode_info *info, int sr)
477 {
478         struct list_head *ptr;
479
480         ptr = info->e_wait_q[sr].list.prev;
481         if (ptr == &info->e_wait_q[sr].list)
482                 return NULL;
483         return list_entry(ptr, struct ext_wait_queue, list);
484 }
485
486 /* Auxiliary functions to manipulate messages' list */
487 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
488 {
489         int k;
490
491         k = info->attr.mq_curmsgs - 1;
492         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
493                 info->messages[k + 1] = info->messages[k];
494                 k--;
495         }
496         info->attr.mq_curmsgs++;
497         info->qsize += ptr->m_ts;
498         info->messages[k + 1] = ptr;
499 }
500
501 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
502 {
503         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
504         return info->messages[info->attr.mq_curmsgs];
505 }
506
507 static inline void set_cookie(struct sk_buff *skb, char code)
508 {
509         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
510 }
511
512 /*
513  * The next function is only to split too long sys_mq_timedsend
514  */
515 static void __do_notify(struct mqueue_inode_info *info)
516 {
517         /* notification
518          * invoked when there is registered process and there isn't process
519          * waiting synchronously for message AND state of queue changed from
520          * empty to not empty. Here we are sure that no one is waiting
521          * synchronously. */
522         if (info->notify_owner &&
523             info->attr.mq_curmsgs == 1) {
524                 struct siginfo sig_i;
525                 switch (info->notify.sigev_notify) {
526                 case SIGEV_NONE:
527                         break;
528                 case SIGEV_SIGNAL:
529                         /* sends signal */
530
531                         sig_i.si_signo = info->notify.sigev_signo;
532                         sig_i.si_errno = 0;
533                         sig_i.si_code = SI_MESGQ;
534                         sig_i.si_value = info->notify.sigev_value;
535                         sig_i.si_pid = task_tgid_nr_ns(current,
536                                                 ns_of_pid(info->notify_owner));
537                         sig_i.si_uid = current_uid();
538
539                         kill_pid_info(info->notify.sigev_signo,
540                                       &sig_i, info->notify_owner);
541                         break;
542                 case SIGEV_THREAD:
543                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
544                         netlink_sendskb(info->notify_sock, info->notify_cookie);
545                         break;
546                 }
547                 /* after notification unregisters process */
548                 put_pid(info->notify_owner);
549                 info->notify_owner = NULL;
550         }
551         wake_up(&info->wait_q);
552 }
553
554 static long prepare_timeout(struct timespec *p)
555 {
556         struct timespec nowts;
557         long timeout;
558
559         if (p) {
560                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
561                         || p->tv_nsec >= NSEC_PER_SEC))
562                         return -EINVAL;
563                 nowts = CURRENT_TIME;
564                 /* first subtract as jiffies can't be too big */
565                 p->tv_sec -= nowts.tv_sec;
566                 if (p->tv_nsec < nowts.tv_nsec) {
567                         p->tv_nsec += NSEC_PER_SEC;
568                         p->tv_sec--;
569                 }
570                 p->tv_nsec -= nowts.tv_nsec;
571                 if (p->tv_sec < 0)
572                         return 0;
573
574                 timeout = timespec_to_jiffies(p) + 1;
575         } else
576                 return MAX_SCHEDULE_TIMEOUT;
577
578         return timeout;
579 }
580
581 static void remove_notification(struct mqueue_inode_info *info)
582 {
583         if (info->notify_owner != NULL &&
584             info->notify.sigev_notify == SIGEV_THREAD) {
585                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
586                 netlink_sendskb(info->notify_sock, info->notify_cookie);
587         }
588         put_pid(info->notify_owner);
589         info->notify_owner = NULL;
590 }
591
592 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
593 {
594         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
595                 return 0;
596         if (capable(CAP_SYS_RESOURCE)) {
597                 if (attr->mq_maxmsg > HARD_MSGMAX)
598                         return 0;
599         } else {
600                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
601                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
602                         return 0;
603         }
604         /* check for overflow */
605         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
606                 return 0;
607         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
608             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
609             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
610                 return 0;
611         return 1;
612 }
613
614 /*
615  * Invoked when creating a new queue via sys_mq_open
616  */
617 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
618                         struct dentry *dentry, int oflag, mode_t mode,
619                         struct mq_attr *attr)
620 {
621         const struct cred *cred = current_cred();
622         struct file *result;
623         int ret;
624
625         if (attr) {
626                 ret = -EINVAL;
627                 if (!mq_attr_ok(ipc_ns, attr))
628                         goto out;
629                 /* store for use during create */
630                 dentry->d_fsdata = attr;
631         }
632
633         mode &= ~current_umask();
634         ret = mnt_want_write(ipc_ns->mq_mnt);
635         if (ret)
636                 goto out;
637         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
638         dentry->d_fsdata = NULL;
639         if (ret)
640                 goto out_drop_write;
641
642         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
643         /*
644          * dentry_open() took a persistent mnt_want_write(),
645          * so we can now drop this one.
646          */
647         mnt_drop_write(ipc_ns->mq_mnt);
648         return result;
649
650 out_drop_write:
651         mnt_drop_write(ipc_ns->mq_mnt);
652 out:
653         dput(dentry);
654         mntput(ipc_ns->mq_mnt);
655         return ERR_PTR(ret);
656 }
657
658 /* Opens existing queue */
659 static struct file *do_open(struct ipc_namespace *ipc_ns,
660                                 struct dentry *dentry, int oflag)
661 {
662         const struct cred *cred = current_cred();
663
664         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
665                                                   MAY_READ | MAY_WRITE };
666
667         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
668                 dput(dentry);
669                 mntput(ipc_ns->mq_mnt);
670                 return ERR_PTR(-EINVAL);
671         }
672
673         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
674                 dput(dentry);
675                 mntput(ipc_ns->mq_mnt);
676                 return ERR_PTR(-EACCES);
677         }
678
679         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
680 }
681
682 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
683                 struct mq_attr __user *, u_attr)
684 {
685         struct dentry *dentry;
686         struct file *filp;
687         char *name;
688         struct mq_attr attr;
689         int fd, error;
690         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
691
692         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
693                 return -EFAULT;
694
695         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
696
697         if (IS_ERR(name = getname(u_name)))
698                 return PTR_ERR(name);
699
700         fd = get_unused_fd_flags(O_CLOEXEC);
701         if (fd < 0)
702                 goto out_putname;
703
704         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
705         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
706         if (IS_ERR(dentry)) {
707                 error = PTR_ERR(dentry);
708                 goto out_putfd;
709         }
710         mntget(ipc_ns->mq_mnt);
711
712         if (oflag & O_CREAT) {
713                 if (dentry->d_inode) {  /* entry already exists */
714                         audit_inode(name, dentry);
715                         error = -EEXIST;
716                         if (oflag & O_EXCL)
717                                 goto out;
718                         filp = do_open(ipc_ns, dentry, oflag);
719                 } else {
720                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
721                                                 dentry, oflag, mode,
722                                                 u_attr ? &attr : NULL);
723                 }
724         } else {
725                 error = -ENOENT;
726                 if (!dentry->d_inode)
727                         goto out;
728                 audit_inode(name, dentry);
729                 filp = do_open(ipc_ns, dentry, oflag);
730         }
731
732         if (IS_ERR(filp)) {
733                 error = PTR_ERR(filp);
734                 goto out_putfd;
735         }
736
737         fd_install(fd, filp);
738         goto out_upsem;
739
740 out:
741         dput(dentry);
742         mntput(ipc_ns->mq_mnt);
743 out_putfd:
744         put_unused_fd(fd);
745         fd = error;
746 out_upsem:
747         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
748 out_putname:
749         putname(name);
750         return fd;
751 }
752
753 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
754 {
755         int err;
756         char *name;
757         struct dentry *dentry;
758         struct inode *inode = NULL;
759         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
760
761         name = getname(u_name);
762         if (IS_ERR(name))
763                 return PTR_ERR(name);
764
765         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
766                         I_MUTEX_PARENT);
767         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
768         if (IS_ERR(dentry)) {
769                 err = PTR_ERR(dentry);
770                 goto out_unlock;
771         }
772
773         if (!dentry->d_inode) {
774                 err = -ENOENT;
775                 goto out_err;
776         }
777
778         inode = dentry->d_inode;
779         if (inode)
780                 atomic_inc(&inode->i_count);
781         err = mnt_want_write(ipc_ns->mq_mnt);
782         if (err)
783                 goto out_err;
784         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
785         mnt_drop_write(ipc_ns->mq_mnt);
786 out_err:
787         dput(dentry);
788
789 out_unlock:
790         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
791         putname(name);
792         if (inode)
793                 iput(inode);
794
795         return err;
796 }
797
798 /* Pipelined send and receive functions.
799  *
800  * If a receiver finds no waiting message, then it registers itself in the
801  * list of waiting receivers. A sender checks that list before adding the new
802  * message into the message array. If there is a waiting receiver, then it
803  * bypasses the message array and directly hands the message over to the
804  * receiver.
805  * The receiver accepts the message and returns without grabbing the queue
806  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
807  * are necessary. The same algorithm is used for sysv semaphores, see
808  * ipc/sem.c for more details.
809  *
810  * The same algorithm is used for senders.
811  */
812
813 /* pipelined_send() - send a message directly to the task waiting in
814  * sys_mq_timedreceive() (without inserting message into a queue).
815  */
816 static inline void pipelined_send(struct mqueue_inode_info *info,
817                                   struct msg_msg *message,
818                                   struct ext_wait_queue *receiver)
819 {
820         receiver->msg = message;
821         list_del(&receiver->list);
822         receiver->state = STATE_PENDING;
823         wake_up_process(receiver->task);
824         smp_wmb();
825         receiver->state = STATE_READY;
826 }
827
828 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
829  * gets its message and put to the queue (we have one free place for sure). */
830 static inline void pipelined_receive(struct mqueue_inode_info *info)
831 {
832         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
833
834         if (!sender) {
835                 /* for poll */
836                 wake_up_interruptible(&info->wait_q);
837                 return;
838         }
839         msg_insert(sender->msg, info);
840         list_del(&sender->list);
841         sender->state = STATE_PENDING;
842         wake_up_process(sender->task);
843         smp_wmb();
844         sender->state = STATE_READY;
845 }
846
847 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
848                 size_t, msg_len, unsigned int, msg_prio,
849                 const struct timespec __user *, u_abs_timeout)
850 {
851         struct file *filp;
852         struct inode *inode;
853         struct ext_wait_queue wait;
854         struct ext_wait_queue *receiver;
855         struct msg_msg *msg_ptr;
856         struct mqueue_inode_info *info;
857         struct timespec ts, *p = NULL;
858         long timeout;
859         int ret;
860
861         if (u_abs_timeout) {
862                 if (copy_from_user(&ts, u_abs_timeout, 
863                                         sizeof(struct timespec)))
864                         return -EFAULT;
865                 p = &ts;
866         }
867
868         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
869                 return -EINVAL;
870
871         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
872         timeout = prepare_timeout(p);
873
874         ret = -EBADF;
875         filp = fget(mqdes);
876         if (unlikely(!filp))
877                 goto out;
878
879         inode = filp->f_path.dentry->d_inode;
880         if (unlikely(filp->f_op != &mqueue_file_operations))
881                 goto out_fput;
882         info = MQUEUE_I(inode);
883         audit_inode(NULL, filp->f_path.dentry);
884
885         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
886                 goto out_fput;
887
888         if (unlikely(msg_len > info->attr.mq_msgsize)) {
889                 ret = -EMSGSIZE;
890                 goto out_fput;
891         }
892
893         /* First try to allocate memory, before doing anything with
894          * existing queues. */
895         msg_ptr = load_msg(u_msg_ptr, msg_len);
896         if (IS_ERR(msg_ptr)) {
897                 ret = PTR_ERR(msg_ptr);
898                 goto out_fput;
899         }
900         msg_ptr->m_ts = msg_len;
901         msg_ptr->m_type = msg_prio;
902
903         spin_lock(&info->lock);
904
905         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
906                 if (filp->f_flags & O_NONBLOCK) {
907                         spin_unlock(&info->lock);
908                         ret = -EAGAIN;
909                 } else if (unlikely(timeout < 0)) {
910                         spin_unlock(&info->lock);
911                         ret = timeout;
912                 } else {
913                         wait.task = current;
914                         wait.msg = (void *) msg_ptr;
915                         wait.state = STATE_NONE;
916                         ret = wq_sleep(info, SEND, timeout, &wait);
917                 }
918                 if (ret < 0)
919                         free_msg(msg_ptr);
920         } else {
921                 receiver = wq_get_first_waiter(info, RECV);
922                 if (receiver) {
923                         pipelined_send(info, msg_ptr, receiver);
924                 } else {
925                         /* adds message to the queue */
926                         msg_insert(msg_ptr, info);
927                         __do_notify(info);
928                 }
929                 inode->i_atime = inode->i_mtime = inode->i_ctime =
930                                 CURRENT_TIME;
931                 spin_unlock(&info->lock);
932                 ret = 0;
933         }
934 out_fput:
935         fput(filp);
936 out:
937         return ret;
938 }
939
940 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
941                 size_t, msg_len, unsigned int __user *, u_msg_prio,
942                 const struct timespec __user *, u_abs_timeout)
943 {
944         long timeout;
945         ssize_t ret;
946         struct msg_msg *msg_ptr;
947         struct file *filp;
948         struct inode *inode;
949         struct mqueue_inode_info *info;
950         struct ext_wait_queue wait;
951         struct timespec ts, *p = NULL;
952
953         if (u_abs_timeout) {
954                 if (copy_from_user(&ts, u_abs_timeout, 
955                                         sizeof(struct timespec)))
956                         return -EFAULT;
957                 p = &ts;
958         }
959
960         audit_mq_sendrecv(mqdes, msg_len, 0, p);
961         timeout = prepare_timeout(p);
962
963         ret = -EBADF;
964         filp = fget(mqdes);
965         if (unlikely(!filp))
966                 goto out;
967
968         inode = filp->f_path.dentry->d_inode;
969         if (unlikely(filp->f_op != &mqueue_file_operations))
970                 goto out_fput;
971         info = MQUEUE_I(inode);
972         audit_inode(NULL, filp->f_path.dentry);
973
974         if (unlikely(!(filp->f_mode & FMODE_READ)))
975                 goto out_fput;
976
977         /* checks if buffer is big enough */
978         if (unlikely(msg_len < info->attr.mq_msgsize)) {
979                 ret = -EMSGSIZE;
980                 goto out_fput;
981         }
982
983         spin_lock(&info->lock);
984         if (info->attr.mq_curmsgs == 0) {
985                 if (filp->f_flags & O_NONBLOCK) {
986                         spin_unlock(&info->lock);
987                         ret = -EAGAIN;
988                         msg_ptr = NULL;
989                 } else if (unlikely(timeout < 0)) {
990                         spin_unlock(&info->lock);
991                         ret = timeout;
992                         msg_ptr = NULL;
993                 } else {
994                         wait.task = current;
995                         wait.state = STATE_NONE;
996                         ret = wq_sleep(info, RECV, timeout, &wait);
997                         msg_ptr = wait.msg;
998                 }
999         } else {
1000                 msg_ptr = msg_get(info);
1001
1002                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1003                                 CURRENT_TIME;
1004
1005                 /* There is now free space in queue. */
1006                 pipelined_receive(info);
1007                 spin_unlock(&info->lock);
1008                 ret = 0;
1009         }
1010         if (ret == 0) {
1011                 ret = msg_ptr->m_ts;
1012
1013                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1014                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1015                         ret = -EFAULT;
1016                 }
1017                 free_msg(msg_ptr);
1018         }
1019 out_fput:
1020         fput(filp);
1021 out:
1022         return ret;
1023 }
1024
1025 /*
1026  * Notes: the case when user wants us to deregister (with NULL as pointer)
1027  * and he isn't currently owner of notification, will be silently discarded.
1028  * It isn't explicitly defined in the POSIX.
1029  */
1030 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1031                 const struct sigevent __user *, u_notification)
1032 {
1033         int ret;
1034         struct file *filp;
1035         struct sock *sock;
1036         struct inode *inode;
1037         struct sigevent notification;
1038         struct mqueue_inode_info *info;
1039         struct sk_buff *nc;
1040
1041         if (u_notification) {
1042                 if (copy_from_user(&notification, u_notification,
1043                                         sizeof(struct sigevent)))
1044                         return -EFAULT;
1045         }
1046
1047         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1048
1049         nc = NULL;
1050         sock = NULL;
1051         if (u_notification != NULL) {
1052                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1053                              notification.sigev_notify != SIGEV_SIGNAL &&
1054                              notification.sigev_notify != SIGEV_THREAD))
1055                         return -EINVAL;
1056                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1057                         !valid_signal(notification.sigev_signo)) {
1058                         return -EINVAL;
1059                 }
1060                 if (notification.sigev_notify == SIGEV_THREAD) {
1061                         long timeo;
1062
1063                         /* create the notify skb */
1064                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1065                         ret = -ENOMEM;
1066                         if (!nc)
1067                                 goto out;
1068                         ret = -EFAULT;
1069                         if (copy_from_user(nc->data,
1070                                         notification.sigev_value.sival_ptr,
1071                                         NOTIFY_COOKIE_LEN)) {
1072                                 goto out;
1073                         }
1074
1075                         /* TODO: add a header? */
1076                         skb_put(nc, NOTIFY_COOKIE_LEN);
1077                         /* and attach it to the socket */
1078 retry:
1079                         filp = fget(notification.sigev_signo);
1080                         ret = -EBADF;
1081                         if (!filp)
1082                                 goto out;
1083                         sock = netlink_getsockbyfilp(filp);
1084                         fput(filp);
1085                         if (IS_ERR(sock)) {
1086                                 ret = PTR_ERR(sock);
1087                                 sock = NULL;
1088                                 goto out;
1089                         }
1090
1091                         timeo = MAX_SCHEDULE_TIMEOUT;
1092                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1093                         if (ret == 1)
1094                                 goto retry;
1095                         if (ret) {
1096                                 sock = NULL;
1097                                 nc = NULL;
1098                                 goto out;
1099                         }
1100                 }
1101         }
1102
1103         ret = -EBADF;
1104         filp = fget(mqdes);
1105         if (!filp)
1106                 goto out;
1107
1108         inode = filp->f_path.dentry->d_inode;
1109         if (unlikely(filp->f_op != &mqueue_file_operations))
1110                 goto out_fput;
1111         info = MQUEUE_I(inode);
1112
1113         ret = 0;
1114         spin_lock(&info->lock);
1115         if (u_notification == NULL) {
1116                 if (info->notify_owner == task_tgid(current)) {
1117                         remove_notification(info);
1118                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1119                 }
1120         } else if (info->notify_owner != NULL) {
1121                 ret = -EBUSY;
1122         } else {
1123                 switch (notification.sigev_notify) {
1124                 case SIGEV_NONE:
1125                         info->notify.sigev_notify = SIGEV_NONE;
1126                         break;
1127                 case SIGEV_THREAD:
1128                         info->notify_sock = sock;
1129                         info->notify_cookie = nc;
1130                         sock = NULL;
1131                         nc = NULL;
1132                         info->notify.sigev_notify = SIGEV_THREAD;
1133                         break;
1134                 case SIGEV_SIGNAL:
1135                         info->notify.sigev_signo = notification.sigev_signo;
1136                         info->notify.sigev_value = notification.sigev_value;
1137                         info->notify.sigev_notify = SIGEV_SIGNAL;
1138                         break;
1139                 }
1140
1141                 info->notify_owner = get_pid(task_tgid(current));
1142                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1143         }
1144         spin_unlock(&info->lock);
1145 out_fput:
1146         fput(filp);
1147 out:
1148         if (sock) {
1149                 netlink_detachskb(sock, nc);
1150         } else if (nc) {
1151                 dev_kfree_skb(nc);
1152         }
1153         return ret;
1154 }
1155
1156 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1157                 const struct mq_attr __user *, u_mqstat,
1158                 struct mq_attr __user *, u_omqstat)
1159 {
1160         int ret;
1161         struct mq_attr mqstat, omqstat;
1162         struct file *filp;
1163         struct inode *inode;
1164         struct mqueue_inode_info *info;
1165
1166         if (u_mqstat != NULL) {
1167                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1168                         return -EFAULT;
1169                 if (mqstat.mq_flags & (~O_NONBLOCK))
1170                         return -EINVAL;
1171         }
1172
1173         ret = -EBADF;
1174         filp = fget(mqdes);
1175         if (!filp)
1176                 goto out;
1177
1178         inode = filp->f_path.dentry->d_inode;
1179         if (unlikely(filp->f_op != &mqueue_file_operations))
1180                 goto out_fput;
1181         info = MQUEUE_I(inode);
1182
1183         spin_lock(&info->lock);
1184
1185         omqstat = info->attr;
1186         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1187         if (u_mqstat) {
1188                 audit_mq_getsetattr(mqdes, &mqstat);
1189                 spin_lock(&filp->f_lock);
1190                 if (mqstat.mq_flags & O_NONBLOCK)
1191                         filp->f_flags |= O_NONBLOCK;
1192                 else
1193                         filp->f_flags &= ~O_NONBLOCK;
1194                 spin_unlock(&filp->f_lock);
1195
1196                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1197         }
1198
1199         spin_unlock(&info->lock);
1200
1201         ret = 0;
1202         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1203                                                 sizeof(struct mq_attr)))
1204                 ret = -EFAULT;
1205
1206 out_fput:
1207         fput(filp);
1208 out:
1209         return ret;
1210 }
1211
1212 static const struct inode_operations mqueue_dir_inode_operations = {
1213         .lookup = simple_lookup,
1214         .create = mqueue_create,
1215         .unlink = mqueue_unlink,
1216 };
1217
1218 static const struct file_operations mqueue_file_operations = {
1219         .flush = mqueue_flush_file,
1220         .poll = mqueue_poll_file,
1221         .read = mqueue_read_file,
1222 };
1223
1224 static const struct super_operations mqueue_super_ops = {
1225         .alloc_inode = mqueue_alloc_inode,
1226         .destroy_inode = mqueue_destroy_inode,
1227         .statfs = simple_statfs,
1228         .delete_inode = mqueue_delete_inode,
1229         .drop_inode = generic_delete_inode,
1230 };
1231
1232 static struct file_system_type mqueue_fs_type = {
1233         .name = "mqueue",
1234         .get_sb = mqueue_get_sb,
1235         .kill_sb = kill_litter_super,
1236 };
1237
1238 int mq_init_ns(struct ipc_namespace *ns)
1239 {
1240         ns->mq_queues_count  = 0;
1241         ns->mq_queues_max    = DFLT_QUEUESMAX;
1242         ns->mq_msg_max       = DFLT_MSGMAX;
1243         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1244
1245         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1246         if (IS_ERR(ns->mq_mnt)) {
1247                 int err = PTR_ERR(ns->mq_mnt);
1248                 ns->mq_mnt = NULL;
1249                 return err;
1250         }
1251         return 0;
1252 }
1253
1254 void mq_clear_sbinfo(struct ipc_namespace *ns)
1255 {
1256         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1257 }
1258
1259 void mq_put_mnt(struct ipc_namespace *ns)
1260 {
1261         mntput(ns->mq_mnt);
1262 }
1263
1264 static int __init init_mqueue_fs(void)
1265 {
1266         int error;
1267
1268         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1269                                 sizeof(struct mqueue_inode_info), 0,
1270                                 SLAB_HWCACHE_ALIGN, init_once);
1271         if (mqueue_inode_cachep == NULL)
1272                 return -ENOMEM;
1273
1274         /* ignore failues - they are not fatal */
1275         mq_sysctl_table = mq_register_sysctl_table();
1276
1277         error = register_filesystem(&mqueue_fs_type);
1278         if (error)
1279                 goto out_sysctl;
1280
1281         spin_lock_init(&mq_lock);
1282
1283         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1284         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1285                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1286                 goto out_filesystem;
1287         }
1288
1289         return 0;
1290
1291 out_filesystem:
1292         unregister_filesystem(&mqueue_fs_type);
1293 out_sysctl:
1294         if (mq_sysctl_table)
1295                 unregister_sysctl_table(mq_sysctl_table);
1296         kmem_cache_destroy(mqueue_inode_cachep);
1297         return error;
1298 }
1299
1300 __initcall(init_mqueue_fs);