ipc/shm.c: make 2 functions static
[linux-2.6.git] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@gmail.com)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *                          Manfred Spraul          (manfred@colorfullife.com)
10  *
11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12  *
13  * This file is released under the GPL.
14  */
15
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32
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(struct kmem_cache *cachep, void *foo)
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, info->notify_cookie);
525                         break;
526                 }
527                 /* after notification unregisters process */
528                 put_pid(info->notify_owner);
529                 info->notify_owner = NULL;
530         }
531         wake_up(&info->wait_q);
532 }
533
534 static long prepare_timeout(const struct timespec __user *u_arg)
535 {
536         struct timespec ts, nowts;
537         long timeout;
538
539         if (u_arg) {
540                 if (unlikely(copy_from_user(&ts, u_arg,
541                                         sizeof(struct timespec))))
542                         return -EFAULT;
543
544                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
545                         || ts.tv_nsec >= NSEC_PER_SEC))
546                         return -EINVAL;
547                 nowts = CURRENT_TIME;
548                 /* first subtract as jiffies can't be too big */
549                 ts.tv_sec -= nowts.tv_sec;
550                 if (ts.tv_nsec < nowts.tv_nsec) {
551                         ts.tv_nsec += NSEC_PER_SEC;
552                         ts.tv_sec--;
553                 }
554                 ts.tv_nsec -= nowts.tv_nsec;
555                 if (ts.tv_sec < 0)
556                         return 0;
557
558                 timeout = timespec_to_jiffies(&ts) + 1;
559         } else
560                 return MAX_SCHEDULE_TIMEOUT;
561
562         return timeout;
563 }
564
565 static void remove_notification(struct mqueue_inode_info *info)
566 {
567         if (info->notify_owner != NULL &&
568             info->notify.sigev_notify == SIGEV_THREAD) {
569                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
570                 netlink_sendskb(info->notify_sock, info->notify_cookie);
571         }
572         put_pid(info->notify_owner);
573         info->notify_owner = NULL;
574 }
575
576 static int mq_attr_ok(struct mq_attr *attr)
577 {
578         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
579                 return 0;
580         if (capable(CAP_SYS_RESOURCE)) {
581                 if (attr->mq_maxmsg > HARD_MSGMAX)
582                         return 0;
583         } else {
584                 if (attr->mq_maxmsg > msg_max ||
585                                 attr->mq_msgsize > msgsize_max)
586                         return 0;
587         }
588         /* check for overflow */
589         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
590                 return 0;
591         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
592             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
593             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
594                 return 0;
595         return 1;
596 }
597
598 /*
599  * Invoked when creating a new queue via sys_mq_open
600  */
601 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
602                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
603 {
604         struct mq_attr attr;
605         int ret;
606
607         if (u_attr) {
608                 ret = -EFAULT;
609                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
610                         goto out;
611                 ret = -EINVAL;
612                 if (!mq_attr_ok(&attr))
613                         goto out;
614                 /* store for use during create */
615                 dentry->d_fsdata = &attr;
616         }
617
618         mode &= ~current->fs->umask;
619         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
620         dentry->d_fsdata = NULL;
621         if (ret)
622                 goto out;
623
624         return dentry_open(dentry, mqueue_mnt, oflag);
625
626 out:
627         dput(dentry);
628         mntput(mqueue_mnt);
629         return ERR_PTR(ret);
630 }
631
632 /* Opens existing queue */
633 static struct file *do_open(struct dentry *dentry, int oflag)
634 {
635 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
636                                         MAY_READ | MAY_WRITE };
637
638         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
639                 dput(dentry);
640                 mntput(mqueue_mnt);
641                 return ERR_PTR(-EINVAL);
642         }
643
644         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
645                 dput(dentry);
646                 mntput(mqueue_mnt);
647                 return ERR_PTR(-EACCES);
648         }
649
650         return dentry_open(dentry, mqueue_mnt, oflag);
651 }
652
653 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
654                                 struct mq_attr __user *u_attr)
655 {
656         struct dentry *dentry;
657         struct file *filp;
658         char *name;
659         int fd, error;
660
661         error = audit_mq_open(oflag, mode, u_attr);
662         if (error != 0)
663                 return error;
664
665         if (IS_ERR(name = getname(u_name)))
666                 return PTR_ERR(name);
667
668         fd = get_unused_fd();
669         if (fd < 0)
670                 goto out_putname;
671
672         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
673         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
674         if (IS_ERR(dentry)) {
675                 error = PTR_ERR(dentry);
676                 goto out_err;
677         }
678         mntget(mqueue_mnt);
679
680         if (oflag & O_CREAT) {
681                 if (dentry->d_inode) {  /* entry already exists */
682                         audit_inode(name, dentry->d_inode);
683                         error = -EEXIST;
684                         if (oflag & O_EXCL)
685                                 goto out;
686                         filp = do_open(dentry, oflag);
687                 } else {
688                         filp = do_create(mqueue_mnt->mnt_root, dentry,
689                                                 oflag, mode, u_attr);
690                 }
691         } else {
692                 error = -ENOENT;
693                 if (!dentry->d_inode)
694                         goto out;
695                 audit_inode(name, dentry->d_inode);
696                 filp = do_open(dentry, oflag);
697         }
698
699         if (IS_ERR(filp)) {
700                 error = PTR_ERR(filp);
701                 goto out_putfd;
702         }
703
704         set_close_on_exec(fd, 1);
705         fd_install(fd, filp);
706         goto out_upsem;
707
708 out:
709         dput(dentry);
710         mntput(mqueue_mnt);
711 out_putfd:
712         put_unused_fd(fd);
713 out_err:
714         fd = error;
715 out_upsem:
716         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
717 out_putname:
718         putname(name);
719         return fd;
720 }
721
722 asmlinkage long sys_mq_unlink(const char __user *u_name)
723 {
724         int err;
725         char *name;
726         struct dentry *dentry;
727         struct inode *inode = NULL;
728
729         name = getname(u_name);
730         if (IS_ERR(name))
731                 return PTR_ERR(name);
732
733         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
734                         I_MUTEX_PARENT);
735         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
736         if (IS_ERR(dentry)) {
737                 err = PTR_ERR(dentry);
738                 goto out_unlock;
739         }
740
741         if (!dentry->d_inode) {
742                 err = -ENOENT;
743                 goto out_err;
744         }
745
746         inode = dentry->d_inode;
747         if (inode)
748                 atomic_inc(&inode->i_count);
749
750         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
751 out_err:
752         dput(dentry);
753
754 out_unlock:
755         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
756         putname(name);
757         if (inode)
758                 iput(inode);
759
760         return err;
761 }
762
763 /* Pipelined send and receive functions.
764  *
765  * If a receiver finds no waiting message, then it registers itself in the
766  * list of waiting receivers. A sender checks that list before adding the new
767  * message into the message array. If there is a waiting receiver, then it
768  * bypasses the message array and directly hands the message over to the
769  * receiver.
770  * The receiver accepts the message and returns without grabbing the queue
771  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
772  * are necessary. The same algorithm is used for sysv semaphores, see
773  * ipc/sem.c for more details.
774  *
775  * The same algorithm is used for senders.
776  */
777
778 /* pipelined_send() - send a message directly to the task waiting in
779  * sys_mq_timedreceive() (without inserting message into a queue).
780  */
781 static inline void pipelined_send(struct mqueue_inode_info *info,
782                                   struct msg_msg *message,
783                                   struct ext_wait_queue *receiver)
784 {
785         receiver->msg = message;
786         list_del(&receiver->list);
787         receiver->state = STATE_PENDING;
788         wake_up_process(receiver->task);
789         smp_wmb();
790         receiver->state = STATE_READY;
791 }
792
793 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
794  * gets its message and put to the queue (we have one free place for sure). */
795 static inline void pipelined_receive(struct mqueue_inode_info *info)
796 {
797         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
798
799         if (!sender) {
800                 /* for poll */
801                 wake_up_interruptible(&info->wait_q);
802                 return;
803         }
804         msg_insert(sender->msg, info);
805         list_del(&sender->list);
806         sender->state = STATE_PENDING;
807         wake_up_process(sender->task);
808         smp_wmb();
809         sender->state = STATE_READY;
810 }
811
812 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
813         size_t msg_len, unsigned int msg_prio,
814         const struct timespec __user *u_abs_timeout)
815 {
816         struct file *filp;
817         struct inode *inode;
818         struct ext_wait_queue wait;
819         struct ext_wait_queue *receiver;
820         struct msg_msg *msg_ptr;
821         struct mqueue_inode_info *info;
822         long timeout;
823         int ret;
824
825         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
826         if (ret != 0)
827                 return ret;
828
829         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
830                 return -EINVAL;
831
832         timeout = prepare_timeout(u_abs_timeout);
833
834         ret = -EBADF;
835         filp = fget(mqdes);
836         if (unlikely(!filp))
837                 goto out;
838
839         inode = filp->f_path.dentry->d_inode;
840         if (unlikely(filp->f_op != &mqueue_file_operations))
841                 goto out_fput;
842         info = MQUEUE_I(inode);
843         audit_inode(NULL, inode);
844
845         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
846                 goto out_fput;
847
848         if (unlikely(msg_len > info->attr.mq_msgsize)) {
849                 ret = -EMSGSIZE;
850                 goto out_fput;
851         }
852
853         /* First try to allocate memory, before doing anything with
854          * existing queues. */
855         msg_ptr = load_msg(u_msg_ptr, msg_len);
856         if (IS_ERR(msg_ptr)) {
857                 ret = PTR_ERR(msg_ptr);
858                 goto out_fput;
859         }
860         msg_ptr->m_ts = msg_len;
861         msg_ptr->m_type = msg_prio;
862
863         spin_lock(&info->lock);
864
865         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
866                 if (filp->f_flags & O_NONBLOCK) {
867                         spin_unlock(&info->lock);
868                         ret = -EAGAIN;
869                 } else if (unlikely(timeout < 0)) {
870                         spin_unlock(&info->lock);
871                         ret = timeout;
872                 } else {
873                         wait.task = current;
874                         wait.msg = (void *) msg_ptr;
875                         wait.state = STATE_NONE;
876                         ret = wq_sleep(info, SEND, timeout, &wait);
877                 }
878                 if (ret < 0)
879                         free_msg(msg_ptr);
880         } else {
881                 receiver = wq_get_first_waiter(info, RECV);
882                 if (receiver) {
883                         pipelined_send(info, msg_ptr, receiver);
884                 } else {
885                         /* adds message to the queue */
886                         msg_insert(msg_ptr, info);
887                         __do_notify(info);
888                 }
889                 inode->i_atime = inode->i_mtime = inode->i_ctime =
890                                 CURRENT_TIME;
891                 spin_unlock(&info->lock);
892                 ret = 0;
893         }
894 out_fput:
895         fput(filp);
896 out:
897         return ret;
898 }
899
900 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
901         size_t msg_len, unsigned int __user *u_msg_prio,
902         const struct timespec __user *u_abs_timeout)
903 {
904         long timeout;
905         ssize_t ret;
906         struct msg_msg *msg_ptr;
907         struct file *filp;
908         struct inode *inode;
909         struct mqueue_inode_info *info;
910         struct ext_wait_queue wait;
911
912         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
913         if (ret != 0)
914                 return ret;
915
916         timeout = prepare_timeout(u_abs_timeout);
917
918         ret = -EBADF;
919         filp = fget(mqdes);
920         if (unlikely(!filp))
921                 goto out;
922
923         inode = filp->f_path.dentry->d_inode;
924         if (unlikely(filp->f_op != &mqueue_file_operations))
925                 goto out_fput;
926         info = MQUEUE_I(inode);
927         audit_inode(NULL, inode);
928
929         if (unlikely(!(filp->f_mode & FMODE_READ)))
930                 goto out_fput;
931
932         /* checks if buffer is big enough */
933         if (unlikely(msg_len < info->attr.mq_msgsize)) {
934                 ret = -EMSGSIZE;
935                 goto out_fput;
936         }
937
938         spin_lock(&info->lock);
939         if (info->attr.mq_curmsgs == 0) {
940                 if (filp->f_flags & O_NONBLOCK) {
941                         spin_unlock(&info->lock);
942                         ret = -EAGAIN;
943                         msg_ptr = NULL;
944                 } else if (unlikely(timeout < 0)) {
945                         spin_unlock(&info->lock);
946                         ret = timeout;
947                         msg_ptr = NULL;
948                 } else {
949                         wait.task = current;
950                         wait.state = STATE_NONE;
951                         ret = wq_sleep(info, RECV, timeout, &wait);
952                         msg_ptr = wait.msg;
953                 }
954         } else {
955                 msg_ptr = msg_get(info);
956
957                 inode->i_atime = inode->i_mtime = inode->i_ctime =
958                                 CURRENT_TIME;
959
960                 /* There is now free space in queue. */
961                 pipelined_receive(info);
962                 spin_unlock(&info->lock);
963                 ret = 0;
964         }
965         if (ret == 0) {
966                 ret = msg_ptr->m_ts;
967
968                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
969                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
970                         ret = -EFAULT;
971                 }
972                 free_msg(msg_ptr);
973         }
974 out_fput:
975         fput(filp);
976 out:
977         return ret;
978 }
979
980 /*
981  * Notes: the case when user wants us to deregister (with NULL as pointer)
982  * and he isn't currently owner of notification, will be silently discarded.
983  * It isn't explicitly defined in the POSIX.
984  */
985 asmlinkage long sys_mq_notify(mqd_t mqdes,
986                                 const struct sigevent __user *u_notification)
987 {
988         int ret;
989         struct file *filp;
990         struct sock *sock;
991         struct inode *inode;
992         struct sigevent notification;
993         struct mqueue_inode_info *info;
994         struct sk_buff *nc;
995
996         ret = audit_mq_notify(mqdes, u_notification);
997         if (ret != 0)
998                 return ret;
999
1000         nc = NULL;
1001         sock = NULL;
1002         if (u_notification != NULL) {
1003                 if (copy_from_user(&notification, u_notification,
1004                                         sizeof(struct sigevent)))
1005                         return -EFAULT;
1006
1007                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1008                              notification.sigev_notify != SIGEV_SIGNAL &&
1009                              notification.sigev_notify != SIGEV_THREAD))
1010                         return -EINVAL;
1011                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1012                         !valid_signal(notification.sigev_signo)) {
1013                         return -EINVAL;
1014                 }
1015                 if (notification.sigev_notify == SIGEV_THREAD) {
1016                         /* create the notify skb */
1017                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1018                         ret = -ENOMEM;
1019                         if (!nc)
1020                                 goto out;
1021                         ret = -EFAULT;
1022                         if (copy_from_user(nc->data,
1023                                         notification.sigev_value.sival_ptr,
1024                                         NOTIFY_COOKIE_LEN)) {
1025                                 goto out;
1026                         }
1027
1028                         /* TODO: add a header? */
1029                         skb_put(nc, NOTIFY_COOKIE_LEN);
1030                         /* and attach it to the socket */
1031 retry:
1032                         filp = fget(notification.sigev_signo);
1033                         ret = -EBADF;
1034                         if (!filp)
1035                                 goto out;
1036                         sock = netlink_getsockbyfilp(filp);
1037                         fput(filp);
1038                         if (IS_ERR(sock)) {
1039                                 ret = PTR_ERR(sock);
1040                                 sock = NULL;
1041                                 goto out;
1042                         }
1043
1044                         ret = netlink_attachskb(sock, nc, 0,
1045                                         MAX_SCHEDULE_TIMEOUT, NULL);
1046                         if (ret == 1)
1047                                 goto retry;
1048                         if (ret) {
1049                                 sock = NULL;
1050                                 nc = NULL;
1051                                 goto out;
1052                         }
1053                 }
1054         }
1055
1056         ret = -EBADF;
1057         filp = fget(mqdes);
1058         if (!filp)
1059                 goto out;
1060
1061         inode = filp->f_path.dentry->d_inode;
1062         if (unlikely(filp->f_op != &mqueue_file_operations))
1063                 goto out_fput;
1064         info = MQUEUE_I(inode);
1065
1066         ret = 0;
1067         spin_lock(&info->lock);
1068         if (u_notification == NULL) {
1069                 if (info->notify_owner == task_tgid(current)) {
1070                         remove_notification(info);
1071                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1072                 }
1073         } else if (info->notify_owner != NULL) {
1074                 ret = -EBUSY;
1075         } else {
1076                 switch (notification.sigev_notify) {
1077                 case SIGEV_NONE:
1078                         info->notify.sigev_notify = SIGEV_NONE;
1079                         break;
1080                 case SIGEV_THREAD:
1081                         info->notify_sock = sock;
1082                         info->notify_cookie = nc;
1083                         sock = NULL;
1084                         nc = NULL;
1085                         info->notify.sigev_notify = SIGEV_THREAD;
1086                         break;
1087                 case SIGEV_SIGNAL:
1088                         info->notify.sigev_signo = notification.sigev_signo;
1089                         info->notify.sigev_value = notification.sigev_value;
1090                         info->notify.sigev_notify = SIGEV_SIGNAL;
1091                         break;
1092                 }
1093
1094                 info->notify_owner = get_pid(task_tgid(current));
1095                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1096         }
1097         spin_unlock(&info->lock);
1098 out_fput:
1099         fput(filp);
1100 out:
1101         if (sock) {
1102                 netlink_detachskb(sock, nc);
1103         } else if (nc) {
1104                 dev_kfree_skb(nc);
1105         }
1106         return ret;
1107 }
1108
1109 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1110                         const struct mq_attr __user *u_mqstat,
1111                         struct mq_attr __user *u_omqstat)
1112 {
1113         int ret;
1114         struct mq_attr mqstat, omqstat;
1115         struct file *filp;
1116         struct inode *inode;
1117         struct mqueue_inode_info *info;
1118
1119         if (u_mqstat != NULL) {
1120                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1121                         return -EFAULT;
1122                 if (mqstat.mq_flags & (~O_NONBLOCK))
1123                         return -EINVAL;
1124         }
1125
1126         ret = -EBADF;
1127         filp = fget(mqdes);
1128         if (!filp)
1129                 goto out;
1130
1131         inode = filp->f_path.dentry->d_inode;
1132         if (unlikely(filp->f_op != &mqueue_file_operations))
1133                 goto out_fput;
1134         info = MQUEUE_I(inode);
1135
1136         spin_lock(&info->lock);
1137
1138         omqstat = info->attr;
1139         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1140         if (u_mqstat) {
1141                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1142                 if (ret != 0)
1143                         goto out;
1144                 if (mqstat.mq_flags & O_NONBLOCK)
1145                         filp->f_flags |= O_NONBLOCK;
1146                 else
1147                         filp->f_flags &= ~O_NONBLOCK;
1148
1149                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1150         }
1151
1152         spin_unlock(&info->lock);
1153
1154         ret = 0;
1155         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1156                                                 sizeof(struct mq_attr)))
1157                 ret = -EFAULT;
1158
1159 out_fput:
1160         fput(filp);
1161 out:
1162         return ret;
1163 }
1164
1165 static const struct inode_operations mqueue_dir_inode_operations = {
1166         .lookup = simple_lookup,
1167         .create = mqueue_create,
1168         .unlink = mqueue_unlink,
1169 };
1170
1171 static const struct file_operations mqueue_file_operations = {
1172         .flush = mqueue_flush_file,
1173         .poll = mqueue_poll_file,
1174         .read = mqueue_read_file,
1175 };
1176
1177 static struct super_operations mqueue_super_ops = {
1178         .alloc_inode = mqueue_alloc_inode,
1179         .destroy_inode = mqueue_destroy_inode,
1180         .statfs = simple_statfs,
1181         .delete_inode = mqueue_delete_inode,
1182         .drop_inode = generic_delete_inode,
1183 };
1184
1185 static struct file_system_type mqueue_fs_type = {
1186         .name = "mqueue",
1187         .get_sb = mqueue_get_sb,
1188         .kill_sb = kill_litter_super,
1189 };
1190
1191 static int msg_max_limit_min = DFLT_MSGMAX;
1192 static int msg_max_limit_max = HARD_MSGMAX;
1193
1194 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1195 static int msg_maxsize_limit_max = INT_MAX;
1196
1197 static ctl_table mq_sysctls[] = {
1198         {
1199                 .ctl_name       = CTL_QUEUESMAX,
1200                 .procname       = "queues_max",
1201                 .data           = &queues_max,
1202                 .maxlen         = sizeof(int),
1203                 .mode           = 0644,
1204                 .proc_handler   = &proc_dointvec,
1205         },
1206         {
1207                 .ctl_name       = CTL_MSGMAX,
1208                 .procname       = "msg_max",
1209                 .data           = &msg_max,
1210                 .maxlen         = sizeof(int),
1211                 .mode           = 0644,
1212                 .proc_handler   = &proc_dointvec_minmax,
1213                 .extra1         = &msg_max_limit_min,
1214                 .extra2         = &msg_max_limit_max,
1215         },
1216         {
1217                 .ctl_name       = CTL_MSGSIZEMAX,
1218                 .procname       = "msgsize_max",
1219                 .data           = &msgsize_max,
1220                 .maxlen         = sizeof(int),
1221                 .mode           = 0644,
1222                 .proc_handler   = &proc_dointvec_minmax,
1223                 .extra1         = &msg_maxsize_limit_min,
1224                 .extra2         = &msg_maxsize_limit_max,
1225         },
1226         { .ctl_name = 0 }
1227 };
1228
1229 static ctl_table mq_sysctl_dir[] = {
1230         {
1231                 .ctl_name       = FS_MQUEUE,
1232                 .procname       = "mqueue",
1233                 .mode           = 0555,
1234                 .child          = mq_sysctls,
1235         },
1236         { .ctl_name = 0 }
1237 };
1238
1239 static ctl_table mq_sysctl_root[] = {
1240         {
1241                 .ctl_name       = CTL_FS,
1242                 .procname       = "fs",
1243                 .mode           = 0555,
1244                 .child          = mq_sysctl_dir,
1245         },
1246         { .ctl_name = 0 }
1247 };
1248
1249 static int __init init_mqueue_fs(void)
1250 {
1251         int error;
1252
1253         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1254                                 sizeof(struct mqueue_inode_info), 0,
1255                                 SLAB_HWCACHE_ALIGN, init_once);
1256         if (mqueue_inode_cachep == NULL)
1257                 return -ENOMEM;
1258
1259         /* ignore failues - they are not fatal */
1260         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1261
1262         error = register_filesystem(&mqueue_fs_type);
1263         if (error)
1264                 goto out_sysctl;
1265
1266         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1267                 error = PTR_ERR(mqueue_mnt);
1268                 goto out_filesystem;
1269         }
1270
1271         /* internal initialization - not common for vfs */
1272         queues_count = 0;
1273         spin_lock_init(&mq_lock);
1274
1275         return 0;
1276
1277 out_filesystem:
1278         unregister_filesystem(&mqueue_fs_type);
1279 out_sysctl:
1280         if (mq_sysctl_table)
1281                 unregister_sysctl_table(mq_sysctl_table);
1282         kmem_cache_destroy(mqueue_inode_cachep);
1283         return error;
1284 }
1285
1286 __initcall(init_mqueue_fs);