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