Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / 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
33 #include <net/sock.h>
34 #include "util.h"
35
36 #define MQUEUE_MAGIC    0x19800202
37 #define DIRENT_SIZE     20
38 #define FILENT_SIZE     80
39
40 #define SEND            0
41 #define RECV            1
42
43 #define STATE_NONE      0
44 #define STATE_PENDING   1
45 #define STATE_READY     2
46
47 /* used by sysctl */
48 #define FS_MQUEUE       1
49 #define CTL_QUEUESMAX   2
50 #define CTL_MSGMAX      3
51 #define CTL_MSGSIZEMAX  4
52
53 /* default values */
54 #define DFLT_QUEUESMAX  256     /* max number of message queues */
55 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
56 #define HARD_MSGMAX     (131072/sizeof(void*))
57 #define DFLT_MSGSIZEMAX 8192    /* max message size */
58
59
60 struct ext_wait_queue {         /* queue of sleeping tasks */
61         struct task_struct *task;
62         struct list_head list;
63         struct msg_msg *msg;    /* ptr of loaded message */
64         int state;              /* one of STATE_* values */
65 };
66
67 struct mqueue_inode_info {
68         spinlock_t lock;
69         struct inode vfs_inode;
70         wait_queue_head_t wait_q;
71
72         struct msg_msg **messages;
73         struct mq_attr attr;
74
75         struct sigevent notify;
76         struct pid* notify_owner;
77         struct user_struct *user;       /* user who created, for accounting */
78         struct sock *notify_sock;
79         struct sk_buff *notify_cookie;
80
81         /* for tasks waiting for free space and messages, respectively */
82         struct ext_wait_queue e_wait_q[2];
83
84         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
85 };
86
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
91
92 static spinlock_t mq_lock;
93 static struct kmem_cache *mqueue_inode_cachep;
94 static struct vfsmount *mqueue_mnt;
95
96 static unsigned int queues_count;
97 static unsigned int queues_max  = DFLT_QUEUESMAX;
98 static unsigned int msg_max     = DFLT_MSGMAX;
99 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
100
101 static struct ctl_table_header * mq_sysctl_table;
102
103 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
104 {
105         return container_of(inode, struct mqueue_inode_info, vfs_inode);
106 }
107
108 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
109                                                         struct mq_attr *attr)
110 {
111         struct inode *inode;
112
113         inode = new_inode(sb);
114         if (inode) {
115                 inode->i_mode = mode;
116                 inode->i_uid = current->fsuid;
117                 inode->i_gid = current->fsgid;
118                 inode->i_blocks = 0;
119                 inode->i_mtime = inode->i_ctime = inode->i_atime =
120                                 CURRENT_TIME;
121
122                 if (S_ISREG(mode)) {
123                         struct mqueue_inode_info *info;
124                         struct task_struct *p = current;
125                         struct user_struct *u = p->user;
126                         unsigned long mq_bytes, mq_msg_tblsz;
127
128                         inode->i_fop = &mqueue_file_operations;
129                         inode->i_size = FILENT_SIZE;
130                         /* mqueue specific info */
131                         info = MQUEUE_I(inode);
132                         spin_lock_init(&info->lock);
133                         init_waitqueue_head(&info->wait_q);
134                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
135                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
136                         info->messages = NULL;
137                         info->notify_owner = NULL;
138                         info->qsize = 0;
139                         info->user = NULL;      /* set when all is ok */
140                         memset(&info->attr, 0, sizeof(info->attr));
141                         info->attr.mq_maxmsg = DFLT_MSGMAX;
142                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
143                         if (attr) {
144                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
145                                 info->attr.mq_msgsize = attr->mq_msgsize;
146                         }
147                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
148                         mq_bytes = (mq_msg_tblsz +
149                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
150
151                         spin_lock(&mq_lock);
152                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
153                             u->mq_bytes + mq_bytes >
154                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
155                                 spin_unlock(&mq_lock);
156                                 goto out_inode;
157                         }
158                         u->mq_bytes += mq_bytes;
159                         spin_unlock(&mq_lock);
160
161                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
162                         if (!info->messages) {
163                                 spin_lock(&mq_lock);
164                                 u->mq_bytes -= mq_bytes;
165                                 spin_unlock(&mq_lock);
166                                 goto out_inode;
167                         }
168                         /* all is ok */
169                         info->user = get_uid(u);
170                 } else if (S_ISDIR(mode)) {
171                         inc_nlink(inode);
172                         /* Some things misbehave if size == 0 on a directory */
173                         inode->i_size = 2 * DIRENT_SIZE;
174                         inode->i_op = &mqueue_dir_inode_operations;
175                         inode->i_fop = &simple_dir_operations;
176                 }
177         }
178         return inode;
179 out_inode:
180         make_bad_inode(inode);
181         iput(inode);
182         return NULL;
183 }
184
185 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
186 {
187         struct inode *inode;
188
189         sb->s_blocksize = PAGE_CACHE_SIZE;
190         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
191         sb->s_magic = MQUEUE_MAGIC;
192         sb->s_op = &mqueue_super_ops;
193
194         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
195         if (!inode)
196                 return -ENOMEM;
197
198         sb->s_root = d_alloc_root(inode);
199         if (!sb->s_root) {
200                 iput(inode);
201                 return -ENOMEM;
202         }
203
204         return 0;
205 }
206
207 static int mqueue_get_sb(struct file_system_type *fs_type,
208                          int flags, const char *dev_name,
209                          void *data, struct vfsmount *mnt)
210 {
211         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
212 }
213
214 static void init_once(void *foo, struct kmem_cache * cachep, unsigned long flags)
215 {
216         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
217
218         if (flags & SLAB_CTOR_CONSTRUCTOR)
219                 inode_init_once(&p->vfs_inode);
220 }
221
222 static struct inode *mqueue_alloc_inode(struct super_block *sb)
223 {
224         struct mqueue_inode_info *ei;
225
226         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
227         if (!ei)
228                 return NULL;
229         return &ei->vfs_inode;
230 }
231
232 static void mqueue_destroy_inode(struct inode *inode)
233 {
234         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
235 }
236
237 static void mqueue_delete_inode(struct inode *inode)
238 {
239         struct mqueue_inode_info *info;
240         struct user_struct *user;
241         unsigned long mq_bytes;
242         int i;
243
244         if (S_ISDIR(inode->i_mode)) {
245                 clear_inode(inode);
246                 return;
247         }
248         info = MQUEUE_I(inode);
249         spin_lock(&info->lock);
250         for (i = 0; i < info->attr.mq_curmsgs; i++)
251                 free_msg(info->messages[i]);
252         kfree(info->messages);
253         spin_unlock(&info->lock);
254
255         clear_inode(inode);
256
257         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
258                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
259         user = info->user;
260         if (user) {
261                 spin_lock(&mq_lock);
262                 user->mq_bytes -= mq_bytes;
263                 queues_count--;
264                 spin_unlock(&mq_lock);
265                 free_uid(user);
266         }
267 }
268
269 static int mqueue_create(struct inode *dir, struct dentry *dentry,
270                                 int mode, struct nameidata *nd)
271 {
272         struct inode *inode;
273         struct mq_attr *attr = dentry->d_fsdata;
274         int error;
275
276         spin_lock(&mq_lock);
277         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
278                 error = -ENOSPC;
279                 goto out_lock;
280         }
281         queues_count++;
282         spin_unlock(&mq_lock);
283
284         inode = mqueue_get_inode(dir->i_sb, mode, attr);
285         if (!inode) {
286                 error = -ENOMEM;
287                 spin_lock(&mq_lock);
288                 queues_count--;
289                 goto out_lock;
290         }
291
292         dir->i_size += DIRENT_SIZE;
293         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
294
295         d_instantiate(dentry, inode);
296         dget(dentry);
297         return 0;
298 out_lock:
299         spin_unlock(&mq_lock);
300         return error;
301 }
302
303 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
304 {
305         struct inode *inode = dentry->d_inode;
306
307         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
308         dir->i_size -= DIRENT_SIZE;
309         drop_nlink(inode);
310         dput(dentry);
311         return 0;
312 }
313
314 /*
315 *       This is routine for system read from queue file.
316 *       To avoid mess with doing here some sort of mq_receive we allow
317 *       to read only queue size & notification info (the only values
318 *       that are interesting from user point of view and aren't accessible
319 *       through std routines)
320 */
321 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
322                                 size_t count, loff_t * off)
323 {
324         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
325         char buffer[FILENT_SIZE];
326         size_t slen;
327         loff_t o;
328
329         if (!count)
330                 return 0;
331
332         spin_lock(&info->lock);
333         snprintf(buffer, sizeof(buffer),
334                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
335                         info->qsize,
336                         info->notify_owner ? info->notify.sigev_notify : 0,
337                         (info->notify_owner &&
338                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
339                                 info->notify.sigev_signo : 0,
340                         pid_nr(info->notify_owner));
341         spin_unlock(&info->lock);
342         buffer[sizeof(buffer)-1] = '\0';
343         slen = strlen(buffer)+1;
344
345         o = *off;
346         if (o > slen)
347                 return 0;
348
349         if (o + count > slen)
350                 count = slen - o;
351
352         if (copy_to_user(u_data, buffer + o, count))
353                 return -EFAULT;
354
355         *off = o + count;
356         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
357         return count;
358 }
359
360 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
361 {
362         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
363
364         spin_lock(&info->lock);
365         if (task_tgid(current) == info->notify_owner)
366                 remove_notification(info);
367
368         spin_unlock(&info->lock);
369         return 0;
370 }
371
372 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
373 {
374         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
375         int retval = 0;
376
377         poll_wait(filp, &info->wait_q, poll_tab);
378
379         spin_lock(&info->lock);
380         if (info->attr.mq_curmsgs)
381                 retval = POLLIN | POLLRDNORM;
382
383         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
384                 retval |= POLLOUT | POLLWRNORM;
385         spin_unlock(&info->lock);
386
387         return retval;
388 }
389
390 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
391 static void wq_add(struct mqueue_inode_info *info, int sr,
392                         struct ext_wait_queue *ewp)
393 {
394         struct ext_wait_queue *walk;
395
396         ewp->task = current;
397
398         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
399                 if (walk->task->static_prio <= current->static_prio) {
400                         list_add_tail(&ewp->list, &walk->list);
401                         return;
402                 }
403         }
404         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
405 }
406
407 /*
408  * Puts current task to sleep. Caller must hold queue lock. After return
409  * lock isn't held.
410  * sr: SEND or RECV
411  */
412 static int wq_sleep(struct mqueue_inode_info *info, int sr,
413                         long timeout, struct ext_wait_queue *ewp)
414 {
415         int retval;
416         signed long time;
417
418         wq_add(info, sr, ewp);
419
420         for (;;) {
421                 set_current_state(TASK_INTERRUPTIBLE);
422
423                 spin_unlock(&info->lock);
424                 time = schedule_timeout(timeout);
425
426                 while (ewp->state == STATE_PENDING)
427                         cpu_relax();
428
429                 if (ewp->state == STATE_READY) {
430                         retval = 0;
431                         goto out;
432                 }
433                 spin_lock(&info->lock);
434                 if (ewp->state == STATE_READY) {
435                         retval = 0;
436                         goto out_unlock;
437                 }
438                 if (signal_pending(current)) {
439                         retval = -ERESTARTSYS;
440                         break;
441                 }
442                 if (time == 0) {
443                         retval = -ETIMEDOUT;
444                         break;
445                 }
446         }
447         list_del(&ewp->list);
448 out_unlock:
449         spin_unlock(&info->lock);
450 out:
451         return retval;
452 }
453
454 /*
455  * Returns waiting task that should be serviced first or NULL if none exists
456  */
457 static struct ext_wait_queue *wq_get_first_waiter(
458                 struct mqueue_inode_info *info, int sr)
459 {
460         struct list_head *ptr;
461
462         ptr = info->e_wait_q[sr].list.prev;
463         if (ptr == &info->e_wait_q[sr].list)
464                 return NULL;
465         return list_entry(ptr, struct ext_wait_queue, list);
466 }
467
468 /* Auxiliary functions to manipulate messages' list */
469 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
470 {
471         int k;
472
473         k = info->attr.mq_curmsgs - 1;
474         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
475                 info->messages[k + 1] = info->messages[k];
476                 k--;
477         }
478         info->attr.mq_curmsgs++;
479         info->qsize += ptr->m_ts;
480         info->messages[k + 1] = ptr;
481 }
482
483 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
484 {
485         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
486         return info->messages[info->attr.mq_curmsgs];
487 }
488
489 static inline void set_cookie(struct sk_buff *skb, char code)
490 {
491         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
492 }
493
494 /*
495  * The next function is only to split too long sys_mq_timedsend
496  */
497 static void __do_notify(struct mqueue_inode_info *info)
498 {
499         /* notification
500          * invoked when there is registered process and there isn't process
501          * waiting synchronously for message AND state of queue changed from
502          * empty to not empty. Here we are sure that no one is waiting
503          * synchronously. */
504         if (info->notify_owner &&
505             info->attr.mq_curmsgs == 1) {
506                 struct siginfo sig_i;
507                 switch (info->notify.sigev_notify) {
508                 case SIGEV_NONE:
509                         break;
510                 case SIGEV_SIGNAL:
511                         /* sends signal */
512
513                         sig_i.si_signo = info->notify.sigev_signo;
514                         sig_i.si_errno = 0;
515                         sig_i.si_code = SI_MESGQ;
516                         sig_i.si_value = info->notify.sigev_value;
517                         sig_i.si_pid = current->tgid;
518                         sig_i.si_uid = current->uid;
519
520                         kill_pid_info(info->notify.sigev_signo,
521                                       &sig_i, info->notify_owner);
522                         break;
523                 case SIGEV_THREAD:
524                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
525                         netlink_sendskb(info->notify_sock,
526                                         info->notify_cookie, 0);
527                         break;
528                 }
529                 /* after notification unregisters process */
530                 put_pid(info->notify_owner);
531                 info->notify_owner = NULL;
532         }
533         wake_up(&info->wait_q);
534 }
535
536 static long prepare_timeout(const struct timespec __user *u_arg)
537 {
538         struct timespec ts, nowts;
539         long timeout;
540
541         if (u_arg) {
542                 if (unlikely(copy_from_user(&ts, u_arg,
543                                         sizeof(struct timespec))))
544                         return -EFAULT;
545
546                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
547                         || ts.tv_nsec >= NSEC_PER_SEC))
548                         return -EINVAL;
549                 nowts = CURRENT_TIME;
550                 /* first subtract as jiffies can't be too big */
551                 ts.tv_sec -= nowts.tv_sec;
552                 if (ts.tv_nsec < nowts.tv_nsec) {
553                         ts.tv_nsec += NSEC_PER_SEC;
554                         ts.tv_sec--;
555                 }
556                 ts.tv_nsec -= nowts.tv_nsec;
557                 if (ts.tv_sec < 0)
558                         return 0;
559
560                 timeout = timespec_to_jiffies(&ts) + 1;
561         } else
562                 return MAX_SCHEDULE_TIMEOUT;
563
564         return timeout;
565 }
566
567 static void remove_notification(struct mqueue_inode_info *info)
568 {
569         if (info->notify_owner != NULL &&
570             info->notify.sigev_notify == SIGEV_THREAD) {
571                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
572                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
573         }
574         put_pid(info->notify_owner);
575         info->notify_owner = NULL;
576 }
577
578 static int mq_attr_ok(struct mq_attr *attr)
579 {
580         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
581                 return 0;
582         if (capable(CAP_SYS_RESOURCE)) {
583                 if (attr->mq_maxmsg > HARD_MSGMAX)
584                         return 0;
585         } else {
586                 if (attr->mq_maxmsg > msg_max ||
587                                 attr->mq_msgsize > msgsize_max)
588                         return 0;
589         }
590         /* check for overflow */
591         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
592                 return 0;
593         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
594             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
595             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
596                 return 0;
597         return 1;
598 }
599
600 /*
601  * Invoked when creating a new queue via sys_mq_open
602  */
603 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
604                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
605 {
606         struct mq_attr attr;
607         int ret;
608
609         if (u_attr) {
610                 ret = -EFAULT;
611                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
612                         goto out;
613                 ret = -EINVAL;
614                 if (!mq_attr_ok(&attr))
615                         goto out;
616                 /* store for use during create */
617                 dentry->d_fsdata = &attr;
618         }
619
620         mode &= ~current->fs->umask;
621         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
622         dentry->d_fsdata = NULL;
623         if (ret)
624                 goto out;
625
626         return dentry_open(dentry, mqueue_mnt, oflag);
627
628 out:
629         dput(dentry);
630         mntput(mqueue_mnt);
631         return ERR_PTR(ret);
632 }
633
634 /* Opens existing queue */
635 static struct file *do_open(struct dentry *dentry, int oflag)
636 {
637 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
638                                         MAY_READ | MAY_WRITE };
639
640         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
641                 dput(dentry);
642                 mntput(mqueue_mnt);
643                 return ERR_PTR(-EINVAL);
644         }
645
646         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
647                 dput(dentry);
648                 mntput(mqueue_mnt);
649                 return ERR_PTR(-EACCES);
650         }
651
652         return dentry_open(dentry, mqueue_mnt, oflag);
653 }
654
655 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
656                                 struct mq_attr __user *u_attr)
657 {
658         struct dentry *dentry;
659         struct file *filp;
660         char *name;
661         int fd, error;
662
663         error = audit_mq_open(oflag, mode, u_attr);
664         if (error != 0)
665                 return error;
666
667         if (IS_ERR(name = getname(u_name)))
668                 return PTR_ERR(name);
669
670         fd = get_unused_fd();
671         if (fd < 0)
672                 goto out_putname;
673
674         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
675         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
676         if (IS_ERR(dentry)) {
677                 error = PTR_ERR(dentry);
678                 goto out_err;
679         }
680         mntget(mqueue_mnt);
681
682         if (oflag & O_CREAT) {
683                 if (dentry->d_inode) {  /* entry already exists */
684                         audit_inode(name, dentry->d_inode);
685                         error = -EEXIST;
686                         if (oflag & O_EXCL)
687                                 goto out;
688                         filp = do_open(dentry, oflag);
689                 } else {
690                         filp = do_create(mqueue_mnt->mnt_root, dentry,
691                                                 oflag, mode, u_attr);
692                 }
693         } else {
694                 error = -ENOENT;
695                 if (!dentry->d_inode)
696                         goto out;
697                 audit_inode(name, dentry->d_inode);
698                 filp = do_open(dentry, oflag);
699         }
700
701         if (IS_ERR(filp)) {
702                 error = PTR_ERR(filp);
703                 goto out_putfd;
704         }
705
706         set_close_on_exec(fd, 1);
707         fd_install(fd, filp);
708         goto out_upsem;
709
710 out:
711         dput(dentry);
712         mntput(mqueue_mnt);
713 out_putfd:
714         put_unused_fd(fd);
715 out_err:
716         fd = error;
717 out_upsem:
718         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
719 out_putname:
720         putname(name);
721         return fd;
722 }
723
724 asmlinkage long sys_mq_unlink(const char __user *u_name)
725 {
726         int err;
727         char *name;
728         struct dentry *dentry;
729         struct inode *inode = NULL;
730
731         name = getname(u_name);
732         if (IS_ERR(name))
733                 return PTR_ERR(name);
734
735         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
736                         I_MUTEX_PARENT);
737         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
738         if (IS_ERR(dentry)) {
739                 err = PTR_ERR(dentry);
740                 goto out_unlock;
741         }
742
743         if (!dentry->d_inode) {
744                 err = -ENOENT;
745                 goto out_err;
746         }
747
748         inode = dentry->d_inode;
749         if (inode)
750                 atomic_inc(&inode->i_count);
751
752         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
753 out_err:
754         dput(dentry);
755
756 out_unlock:
757         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
758         putname(name);
759         if (inode)
760                 iput(inode);
761
762         return err;
763 }
764
765 /* Pipelined send and receive functions.
766  *
767  * If a receiver finds no waiting message, then it registers itself in the
768  * list of waiting receivers. A sender checks that list before adding the new
769  * message into the message array. If there is a waiting receiver, then it
770  * bypasses the message array and directly hands the message over to the
771  * receiver.
772  * The receiver accepts the message and returns without grabbing the queue
773  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
774  * are necessary. The same algorithm is used for sysv semaphores, see
775  * ipc/sem.c for more details.
776  *
777  * The same algorithm is used for senders.
778  */
779
780 /* pipelined_send() - send a message directly to the task waiting in
781  * sys_mq_timedreceive() (without inserting message into a queue).
782  */
783 static inline void pipelined_send(struct mqueue_inode_info *info,
784                                   struct msg_msg *message,
785                                   struct ext_wait_queue *receiver)
786 {
787         receiver->msg = message;
788         list_del(&receiver->list);
789         receiver->state = STATE_PENDING;
790         wake_up_process(receiver->task);
791         smp_wmb();
792         receiver->state = STATE_READY;
793 }
794
795 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
796  * gets its message and put to the queue (we have one free place for sure). */
797 static inline void pipelined_receive(struct mqueue_inode_info *info)
798 {
799         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
800
801         if (!sender) {
802                 /* for poll */
803                 wake_up_interruptible(&info->wait_q);
804                 return;
805         }
806         msg_insert(sender->msg, info);
807         list_del(&sender->list);
808         sender->state = STATE_PENDING;
809         wake_up_process(sender->task);
810         smp_wmb();
811         sender->state = STATE_READY;
812 }
813
814 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
815         size_t msg_len, unsigned int msg_prio,
816         const struct timespec __user *u_abs_timeout)
817 {
818         struct file *filp;
819         struct inode *inode;
820         struct ext_wait_queue wait;
821         struct ext_wait_queue *receiver;
822         struct msg_msg *msg_ptr;
823         struct mqueue_inode_info *info;
824         long timeout;
825         int ret;
826
827         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
828         if (ret != 0)
829                 return ret;
830
831         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
832                 return -EINVAL;
833
834         timeout = prepare_timeout(u_abs_timeout);
835
836         ret = -EBADF;
837         filp = fget(mqdes);
838         if (unlikely(!filp))
839                 goto out;
840
841         inode = filp->f_path.dentry->d_inode;
842         if (unlikely(filp->f_op != &mqueue_file_operations))
843                 goto out_fput;
844         info = MQUEUE_I(inode);
845         audit_inode(NULL, inode);
846
847         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
848                 goto out_fput;
849
850         if (unlikely(msg_len > info->attr.mq_msgsize)) {
851                 ret = -EMSGSIZE;
852                 goto out_fput;
853         }
854
855         /* First try to allocate memory, before doing anything with
856          * existing queues. */
857         msg_ptr = load_msg(u_msg_ptr, msg_len);
858         if (IS_ERR(msg_ptr)) {
859                 ret = PTR_ERR(msg_ptr);
860                 goto out_fput;
861         }
862         msg_ptr->m_ts = msg_len;
863         msg_ptr->m_type = msg_prio;
864
865         spin_lock(&info->lock);
866
867         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
868                 if (filp->f_flags & O_NONBLOCK) {
869                         spin_unlock(&info->lock);
870                         ret = -EAGAIN;
871                 } else if (unlikely(timeout < 0)) {
872                         spin_unlock(&info->lock);
873                         ret = timeout;
874                 } else {
875                         wait.task = current;
876                         wait.msg = (void *) msg_ptr;
877                         wait.state = STATE_NONE;
878                         ret = wq_sleep(info, SEND, timeout, &wait);
879                 }
880                 if (ret < 0)
881                         free_msg(msg_ptr);
882         } else {
883                 receiver = wq_get_first_waiter(info, RECV);
884                 if (receiver) {
885                         pipelined_send(info, msg_ptr, receiver);
886                 } else {
887                         /* adds message to the queue */
888                         msg_insert(msg_ptr, info);
889                         __do_notify(info);
890                 }
891                 inode->i_atime = inode->i_mtime = inode->i_ctime =
892                                 CURRENT_TIME;
893                 spin_unlock(&info->lock);
894                 ret = 0;
895         }
896 out_fput:
897         fput(filp);
898 out:
899         return ret;
900 }
901
902 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
903         size_t msg_len, unsigned int __user *u_msg_prio,
904         const struct timespec __user *u_abs_timeout)
905 {
906         long timeout;
907         ssize_t ret;
908         struct msg_msg *msg_ptr;
909         struct file *filp;
910         struct inode *inode;
911         struct mqueue_inode_info *info;
912         struct ext_wait_queue wait;
913
914         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
915         if (ret != 0)
916                 return ret;
917
918         timeout = prepare_timeout(u_abs_timeout);
919
920         ret = -EBADF;
921         filp = fget(mqdes);
922         if (unlikely(!filp))
923                 goto out;
924
925         inode = filp->f_path.dentry->d_inode;
926         if (unlikely(filp->f_op != &mqueue_file_operations))
927                 goto out_fput;
928         info = MQUEUE_I(inode);
929         audit_inode(NULL, inode);
930
931         if (unlikely(!(filp->f_mode & FMODE_READ)))
932                 goto out_fput;
933
934         /* checks if buffer is big enough */
935         if (unlikely(msg_len < info->attr.mq_msgsize)) {
936                 ret = -EMSGSIZE;
937                 goto out_fput;
938         }
939
940         spin_lock(&info->lock);
941         if (info->attr.mq_curmsgs == 0) {
942                 if (filp->f_flags & O_NONBLOCK) {
943                         spin_unlock(&info->lock);
944                         ret = -EAGAIN;
945                         msg_ptr = NULL;
946                 } else if (unlikely(timeout < 0)) {
947                         spin_unlock(&info->lock);
948                         ret = timeout;
949                         msg_ptr = NULL;
950                 } else {
951                         wait.task = current;
952                         wait.state = STATE_NONE;
953                         ret = wq_sleep(info, RECV, timeout, &wait);
954                         msg_ptr = wait.msg;
955                 }
956         } else {
957                 msg_ptr = msg_get(info);
958
959                 inode->i_atime = inode->i_mtime = inode->i_ctime =
960                                 CURRENT_TIME;
961
962                 /* There is now free space in queue. */
963                 pipelined_receive(info);
964                 spin_unlock(&info->lock);
965                 ret = 0;
966         }
967         if (ret == 0) {
968                 ret = msg_ptr->m_ts;
969
970                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
971                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
972                         ret = -EFAULT;
973                 }
974                 free_msg(msg_ptr);
975         }
976 out_fput:
977         fput(filp);
978 out:
979         return ret;
980 }
981
982 /*
983  * Notes: the case when user wants us to deregister (with NULL as pointer)
984  * and he isn't currently owner of notification, will be silently discarded.
985  * It isn't explicitly defined in the POSIX.
986  */
987 asmlinkage long sys_mq_notify(mqd_t mqdes,
988                                 const struct sigevent __user *u_notification)
989 {
990         int ret;
991         struct file *filp;
992         struct sock *sock;
993         struct inode *inode;
994         struct sigevent notification;
995         struct mqueue_inode_info *info;
996         struct sk_buff *nc;
997
998         ret = audit_mq_notify(mqdes, u_notification);
999         if (ret != 0)
1000                 return ret;
1001
1002         nc = NULL;
1003         sock = NULL;
1004         if (u_notification != NULL) {
1005                 if (copy_from_user(&notification, u_notification,
1006                                         sizeof(struct sigevent)))
1007                         return -EFAULT;
1008
1009                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1010                              notification.sigev_notify != SIGEV_SIGNAL &&
1011                              notification.sigev_notify != SIGEV_THREAD))
1012                         return -EINVAL;
1013                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1014                         !valid_signal(notification.sigev_signo)) {
1015                         return -EINVAL;
1016                 }
1017                 if (notification.sigev_notify == SIGEV_THREAD) {
1018                         /* create the notify skb */
1019                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1020                         ret = -ENOMEM;
1021                         if (!nc)
1022                                 goto out;
1023                         ret = -EFAULT;
1024                         if (copy_from_user(nc->data,
1025                                         notification.sigev_value.sival_ptr,
1026                                         NOTIFY_COOKIE_LEN)) {
1027                                 goto out;
1028                         }
1029
1030                         /* TODO: add a header? */
1031                         skb_put(nc, NOTIFY_COOKIE_LEN);
1032                         /* and attach it to the socket */
1033 retry:
1034                         filp = fget(notification.sigev_signo);
1035                         ret = -EBADF;
1036                         if (!filp)
1037                                 goto out;
1038                         sock = netlink_getsockbyfilp(filp);
1039                         fput(filp);
1040                         if (IS_ERR(sock)) {
1041                                 ret = PTR_ERR(sock);
1042                                 sock = NULL;
1043                                 goto out;
1044                         }
1045
1046                         ret = netlink_attachskb(sock, nc, 0,
1047                                         MAX_SCHEDULE_TIMEOUT, NULL);
1048                         if (ret == 1)
1049                                 goto retry;
1050                         if (ret) {
1051                                 sock = NULL;
1052                                 nc = NULL;
1053                                 goto out;
1054                         }
1055                 }
1056         }
1057
1058         ret = -EBADF;
1059         filp = fget(mqdes);
1060         if (!filp)
1061                 goto out;
1062
1063         inode = filp->f_path.dentry->d_inode;
1064         if (unlikely(filp->f_op != &mqueue_file_operations))
1065                 goto out_fput;
1066         info = MQUEUE_I(inode);
1067
1068         ret = 0;
1069         spin_lock(&info->lock);
1070         if (u_notification == NULL) {
1071                 if (info->notify_owner == task_tgid(current)) {
1072                         remove_notification(info);
1073                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1074                 }
1075         } else if (info->notify_owner != NULL) {
1076                 ret = -EBUSY;
1077         } else {
1078                 switch (notification.sigev_notify) {
1079                 case SIGEV_NONE:
1080                         info->notify.sigev_notify = SIGEV_NONE;
1081                         break;
1082                 case SIGEV_THREAD:
1083                         info->notify_sock = sock;
1084                         info->notify_cookie = nc;
1085                         sock = NULL;
1086                         nc = NULL;
1087                         info->notify.sigev_notify = SIGEV_THREAD;
1088                         break;
1089                 case SIGEV_SIGNAL:
1090                         info->notify.sigev_signo = notification.sigev_signo;
1091                         info->notify.sigev_value = notification.sigev_value;
1092                         info->notify.sigev_notify = SIGEV_SIGNAL;
1093                         break;
1094                 }
1095
1096                 info->notify_owner = get_pid(task_tgid(current));
1097                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1098         }
1099         spin_unlock(&info->lock);
1100 out_fput:
1101         fput(filp);
1102 out:
1103         if (sock) {
1104                 netlink_detachskb(sock, nc);
1105         } else if (nc) {
1106                 dev_kfree_skb(nc);
1107         }
1108         return ret;
1109 }
1110
1111 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1112                         const struct mq_attr __user *u_mqstat,
1113                         struct mq_attr __user *u_omqstat)
1114 {
1115         int ret;
1116         struct mq_attr mqstat, omqstat;
1117         struct file *filp;
1118         struct inode *inode;
1119         struct mqueue_inode_info *info;
1120
1121         if (u_mqstat != NULL) {
1122                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1123                         return -EFAULT;
1124                 if (mqstat.mq_flags & (~O_NONBLOCK))
1125                         return -EINVAL;
1126         }
1127
1128         ret = -EBADF;
1129         filp = fget(mqdes);
1130         if (!filp)
1131                 goto out;
1132
1133         inode = filp->f_path.dentry->d_inode;
1134         if (unlikely(filp->f_op != &mqueue_file_operations))
1135                 goto out_fput;
1136         info = MQUEUE_I(inode);
1137
1138         spin_lock(&info->lock);
1139
1140         omqstat = info->attr;
1141         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1142         if (u_mqstat) {
1143                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1144                 if (ret != 0)
1145                         goto out;
1146                 if (mqstat.mq_flags & O_NONBLOCK)
1147                         filp->f_flags |= O_NONBLOCK;
1148                 else
1149                         filp->f_flags &= ~O_NONBLOCK;
1150
1151                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1152         }
1153
1154         spin_unlock(&info->lock);
1155
1156         ret = 0;
1157         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1158                                                 sizeof(struct mq_attr)))
1159                 ret = -EFAULT;
1160
1161 out_fput:
1162         fput(filp);
1163 out:
1164         return ret;
1165 }
1166
1167 static const struct inode_operations mqueue_dir_inode_operations = {
1168         .lookup = simple_lookup,
1169         .create = mqueue_create,
1170         .unlink = mqueue_unlink,
1171 };
1172
1173 static const struct file_operations mqueue_file_operations = {
1174         .flush = mqueue_flush_file,
1175         .poll = mqueue_poll_file,
1176         .read = mqueue_read_file,
1177 };
1178
1179 static struct super_operations mqueue_super_ops = {
1180         .alloc_inode = mqueue_alloc_inode,
1181         .destroy_inode = mqueue_destroy_inode,
1182         .statfs = simple_statfs,
1183         .delete_inode = mqueue_delete_inode,
1184         .drop_inode = generic_delete_inode,
1185 };
1186
1187 static struct file_system_type mqueue_fs_type = {
1188         .name = "mqueue",
1189         .get_sb = mqueue_get_sb,
1190         .kill_sb = kill_litter_super,
1191 };
1192
1193 static int msg_max_limit_min = DFLT_MSGMAX;
1194 static int msg_max_limit_max = HARD_MSGMAX;
1195
1196 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1197 static int msg_maxsize_limit_max = INT_MAX;
1198
1199 static ctl_table mq_sysctls[] = {
1200         {
1201                 .ctl_name       = CTL_QUEUESMAX,
1202                 .procname       = "queues_max",
1203                 .data           = &queues_max,
1204                 .maxlen         = sizeof(int),
1205                 .mode           = 0644,
1206                 .proc_handler   = &proc_dointvec,
1207         },
1208         {
1209                 .ctl_name       = CTL_MSGMAX,
1210                 .procname       = "msg_max",
1211                 .data           = &msg_max,
1212                 .maxlen         = sizeof(int),
1213                 .mode           = 0644,
1214                 .proc_handler   = &proc_dointvec_minmax,
1215                 .extra1         = &msg_max_limit_min,
1216                 .extra2         = &msg_max_limit_max,
1217         },
1218         {
1219                 .ctl_name       = CTL_MSGSIZEMAX,
1220                 .procname       = "msgsize_max",
1221                 .data           = &msgsize_max,
1222                 .maxlen         = sizeof(int),
1223                 .mode           = 0644,
1224                 .proc_handler   = &proc_dointvec_minmax,
1225                 .extra1         = &msg_maxsize_limit_min,
1226                 .extra2         = &msg_maxsize_limit_max,
1227         },
1228         { .ctl_name = 0 }
1229 };
1230
1231 static ctl_table mq_sysctl_dir[] = {
1232         {
1233                 .ctl_name       = FS_MQUEUE,
1234                 .procname       = "mqueue",
1235                 .mode           = 0555,
1236                 .child          = mq_sysctls,
1237         },
1238         { .ctl_name = 0 }
1239 };
1240
1241 static ctl_table mq_sysctl_root[] = {
1242         {
1243                 .ctl_name       = CTL_FS,
1244                 .procname       = "fs",
1245                 .mode           = 0555,
1246                 .child          = mq_sysctl_dir,
1247         },
1248         { .ctl_name = 0 }
1249 };
1250
1251 static int __init init_mqueue_fs(void)
1252 {
1253         int error;
1254
1255         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1256                                 sizeof(struct mqueue_inode_info), 0,
1257                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1258         if (mqueue_inode_cachep == NULL)
1259                 return -ENOMEM;
1260
1261         /* ignore failues - they are not fatal */
1262         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1263
1264         error = register_filesystem(&mqueue_fs_type);
1265         if (error)
1266                 goto out_sysctl;
1267
1268         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1269                 error = PTR_ERR(mqueue_mnt);
1270                 goto out_filesystem;
1271         }
1272
1273         /* internal initialization - not common for vfs */
1274         queues_count = 0;
1275         spin_lock_init(&mq_lock);
1276
1277         return 0;
1278
1279 out_filesystem:
1280         unregister_filesystem(&mqueue_fs_type);
1281 out_sysctl:
1282         if (mq_sysctl_table)
1283                 unregister_sysctl_table(mq_sysctl_table);
1284         kmem_cache_destroy(mqueue_inode_cachep);
1285         return error;
1286 }
1287
1288 __initcall(init_mqueue_fs);