ipc: make shm_get_stat() more robust
[linux-2.6.git] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@gmail.com)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *                          Manfred Spraul          (manfred@colorfullife.com)
10  *
11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12  *
13  * This file is released under the GPL.
14  */
15
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32 #include <linux/nsproxy.h>
33 #include <linux/pid.h>
34
35 #include <net/sock.h>
36 #include "util.h"
37
38 #define MQUEUE_MAGIC    0x19800202
39 #define DIRENT_SIZE     20
40 #define FILENT_SIZE     80
41
42 #define SEND            0
43 #define RECV            1
44
45 #define STATE_NONE      0
46 #define STATE_PENDING   1
47 #define STATE_READY     2
48
49 /* default values */
50 #define DFLT_QUEUESMAX  256     /* max number of message queues */
51 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
52 #define HARD_MSGMAX     (131072/sizeof(void*))
53 #define DFLT_MSGSIZEMAX 8192    /* max message size */
54
55 /*
56  * Define the ranges various user-specified maximum values can
57  * be set to.
58  */
59 #define MIN_MSGMAX      1               /* min value for msg_max */
60 #define MAX_MSGMAX      HARD_MSGMAX     /* max value for msg_max */
61 #define MIN_MSGSIZEMAX  128             /* min value for msgsize_max */
62 #define MAX_MSGSIZEMAX  (8192*128)      /* max value for msgsize_max */
63
64 struct ext_wait_queue {         /* queue of sleeping tasks */
65         struct task_struct *task;
66         struct list_head list;
67         struct msg_msg *msg;    /* ptr of loaded message */
68         int state;              /* one of STATE_* values */
69 };
70
71 struct mqueue_inode_info {
72         spinlock_t lock;
73         struct inode vfs_inode;
74         wait_queue_head_t wait_q;
75
76         struct msg_msg **messages;
77         struct mq_attr attr;
78
79         struct sigevent notify;
80         struct pid* notify_owner;
81         struct user_struct *user;       /* user who created, for accounting */
82         struct sock *notify_sock;
83         struct sk_buff *notify_cookie;
84
85         /* for tasks waiting for free space and messages, respectively */
86         struct ext_wait_queue e_wait_q[2];
87
88         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
89 };
90
91 static const struct inode_operations mqueue_dir_inode_operations;
92 static const struct file_operations mqueue_file_operations;
93 static struct super_operations mqueue_super_ops;
94 static void remove_notification(struct mqueue_inode_info *info);
95
96 static spinlock_t mq_lock;
97 static struct kmem_cache *mqueue_inode_cachep;
98 static struct vfsmount *mqueue_mnt;
99
100 static unsigned int queues_count;
101 static unsigned int queues_max  = DFLT_QUEUESMAX;
102 static unsigned int msg_max     = DFLT_MSGMAX;
103 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
104
105 static struct ctl_table_header * mq_sysctl_table;
106
107 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
108 {
109         return container_of(inode, struct mqueue_inode_info, vfs_inode);
110 }
111
112 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
113                                                         struct mq_attr *attr)
114 {
115         struct user_struct *u = current_user();
116         struct inode *inode;
117
118         inode = new_inode(sb);
119         if (inode) {
120                 inode->i_mode = mode;
121                 inode->i_uid = current_fsuid();
122                 inode->i_gid = current_fsgid();
123                 inode->i_mtime = inode->i_ctime = inode->i_atime =
124                                 CURRENT_TIME;
125
126                 if (S_ISREG(mode)) {
127                         struct mqueue_inode_info *info;
128                         struct task_struct *p = current;
129                         unsigned long mq_bytes, mq_msg_tblsz;
130
131                         inode->i_fop = &mqueue_file_operations;
132                         inode->i_size = FILENT_SIZE;
133                         /* mqueue specific info */
134                         info = MQUEUE_I(inode);
135                         spin_lock_init(&info->lock);
136                         init_waitqueue_head(&info->wait_q);
137                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
138                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
139                         info->messages = NULL;
140                         info->notify_owner = NULL;
141                         info->qsize = 0;
142                         info->user = NULL;      /* set when all is ok */
143                         memset(&info->attr, 0, sizeof(info->attr));
144                         info->attr.mq_maxmsg = msg_max;
145                         info->attr.mq_msgsize = msgsize_max;
146                         if (attr) {
147                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
148                                 info->attr.mq_msgsize = attr->mq_msgsize;
149                         }
150                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
151                         mq_bytes = (mq_msg_tblsz +
152                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
153
154                         spin_lock(&mq_lock);
155                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
156                             u->mq_bytes + mq_bytes >
157                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
158                                 spin_unlock(&mq_lock);
159                                 goto out_inode;
160                         }
161                         u->mq_bytes += mq_bytes;
162                         spin_unlock(&mq_lock);
163
164                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
165                         if (!info->messages) {
166                                 spin_lock(&mq_lock);
167                                 u->mq_bytes -= mq_bytes;
168                                 spin_unlock(&mq_lock);
169                                 goto out_inode;
170                         }
171                         /* all is ok */
172                         info->user = get_uid(u);
173                 } else if (S_ISDIR(mode)) {
174                         inc_nlink(inode);
175                         /* Some things misbehave if size == 0 on a directory */
176                         inode->i_size = 2 * DIRENT_SIZE;
177                         inode->i_op = &mqueue_dir_inode_operations;
178                         inode->i_fop = &simple_dir_operations;
179                 }
180         }
181         return inode;
182 out_inode:
183         make_bad_inode(inode);
184         iput(inode);
185         return NULL;
186 }
187
188 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
189 {
190         struct inode *inode;
191
192         sb->s_blocksize = PAGE_CACHE_SIZE;
193         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
194         sb->s_magic = MQUEUE_MAGIC;
195         sb->s_op = &mqueue_super_ops;
196
197         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
198         if (!inode)
199                 return -ENOMEM;
200
201         sb->s_root = d_alloc_root(inode);
202         if (!sb->s_root) {
203                 iput(inode);
204                 return -ENOMEM;
205         }
206
207         return 0;
208 }
209
210 static int mqueue_get_sb(struct file_system_type *fs_type,
211                          int flags, const char *dev_name,
212                          void *data, struct vfsmount *mnt)
213 {
214         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
215 }
216
217 static void init_once(void *foo)
218 {
219         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
220
221         inode_init_once(&p->vfs_inode);
222 }
223
224 static struct inode *mqueue_alloc_inode(struct super_block *sb)
225 {
226         struct mqueue_inode_info *ei;
227
228         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
229         if (!ei)
230                 return NULL;
231         return &ei->vfs_inode;
232 }
233
234 static void mqueue_destroy_inode(struct inode *inode)
235 {
236         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
237 }
238
239 static void mqueue_delete_inode(struct inode *inode)
240 {
241         struct mqueue_inode_info *info;
242         struct user_struct *user;
243         unsigned long mq_bytes;
244         int i;
245
246         if (S_ISDIR(inode->i_mode)) {
247                 clear_inode(inode);
248                 return;
249         }
250         info = MQUEUE_I(inode);
251         spin_lock(&info->lock);
252         for (i = 0; i < info->attr.mq_curmsgs; i++)
253                 free_msg(info->messages[i]);
254         kfree(info->messages);
255         spin_unlock(&info->lock);
256
257         clear_inode(inode);
258
259         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
260                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
261         user = info->user;
262         if (user) {
263                 spin_lock(&mq_lock);
264                 user->mq_bytes -= mq_bytes;
265                 queues_count--;
266                 spin_unlock(&mq_lock);
267                 free_uid(user);
268         }
269 }
270
271 static int mqueue_create(struct inode *dir, struct dentry *dentry,
272                                 int mode, struct nameidata *nd)
273 {
274         struct inode *inode;
275         struct mq_attr *attr = dentry->d_fsdata;
276         int error;
277
278         spin_lock(&mq_lock);
279         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
280                 error = -ENOSPC;
281                 goto out_lock;
282         }
283         queues_count++;
284         spin_unlock(&mq_lock);
285
286         inode = mqueue_get_inode(dir->i_sb, mode, attr);
287         if (!inode) {
288                 error = -ENOMEM;
289                 spin_lock(&mq_lock);
290                 queues_count--;
291                 goto out_lock;
292         }
293
294         dir->i_size += DIRENT_SIZE;
295         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
296
297         d_instantiate(dentry, inode);
298         dget(dentry);
299         return 0;
300 out_lock:
301         spin_unlock(&mq_lock);
302         return error;
303 }
304
305 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
306 {
307         struct inode *inode = dentry->d_inode;
308
309         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
310         dir->i_size -= DIRENT_SIZE;
311         drop_nlink(inode);
312         dput(dentry);
313         return 0;
314 }
315
316 /*
317 *       This is routine for system read from queue file.
318 *       To avoid mess with doing here some sort of mq_receive we allow
319 *       to read only queue size & notification info (the only values
320 *       that are interesting from user point of view and aren't accessible
321 *       through std routines)
322 */
323 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
324                                 size_t count, loff_t *off)
325 {
326         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
327         char buffer[FILENT_SIZE];
328         ssize_t ret;
329
330         spin_lock(&info->lock);
331         snprintf(buffer, sizeof(buffer),
332                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
333                         info->qsize,
334                         info->notify_owner ? info->notify.sigev_notify : 0,
335                         (info->notify_owner &&
336                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
337                                 info->notify.sigev_signo : 0,
338                         pid_vnr(info->notify_owner));
339         spin_unlock(&info->lock);
340         buffer[sizeof(buffer)-1] = '\0';
341
342         ret = simple_read_from_buffer(u_data, count, off, buffer,
343                                 strlen(buffer));
344         if (ret <= 0)
345                 return ret;
346
347         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
348         return ret;
349 }
350
351 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
352 {
353         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
354
355         spin_lock(&info->lock);
356         if (task_tgid(current) == info->notify_owner)
357                 remove_notification(info);
358
359         spin_unlock(&info->lock);
360         return 0;
361 }
362
363 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
364 {
365         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
366         int retval = 0;
367
368         poll_wait(filp, &info->wait_q, poll_tab);
369
370         spin_lock(&info->lock);
371         if (info->attr.mq_curmsgs)
372                 retval = POLLIN | POLLRDNORM;
373
374         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
375                 retval |= POLLOUT | POLLWRNORM;
376         spin_unlock(&info->lock);
377
378         return retval;
379 }
380
381 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
382 static void wq_add(struct mqueue_inode_info *info, int sr,
383                         struct ext_wait_queue *ewp)
384 {
385         struct ext_wait_queue *walk;
386
387         ewp->task = current;
388
389         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
390                 if (walk->task->static_prio <= current->static_prio) {
391                         list_add_tail(&ewp->list, &walk->list);
392                         return;
393                 }
394         }
395         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
396 }
397
398 /*
399  * Puts current task to sleep. Caller must hold queue lock. After return
400  * lock isn't held.
401  * sr: SEND or RECV
402  */
403 static int wq_sleep(struct mqueue_inode_info *info, int sr,
404                         long timeout, struct ext_wait_queue *ewp)
405 {
406         int retval;
407         signed long time;
408
409         wq_add(info, sr, ewp);
410
411         for (;;) {
412                 set_current_state(TASK_INTERRUPTIBLE);
413
414                 spin_unlock(&info->lock);
415                 time = schedule_timeout(timeout);
416
417                 while (ewp->state == STATE_PENDING)
418                         cpu_relax();
419
420                 if (ewp->state == STATE_READY) {
421                         retval = 0;
422                         goto out;
423                 }
424                 spin_lock(&info->lock);
425                 if (ewp->state == STATE_READY) {
426                         retval = 0;
427                         goto out_unlock;
428                 }
429                 if (signal_pending(current)) {
430                         retval = -ERESTARTSYS;
431                         break;
432                 }
433                 if (time == 0) {
434                         retval = -ETIMEDOUT;
435                         break;
436                 }
437         }
438         list_del(&ewp->list);
439 out_unlock:
440         spin_unlock(&info->lock);
441 out:
442         return retval;
443 }
444
445 /*
446  * Returns waiting task that should be serviced first or NULL if none exists
447  */
448 static struct ext_wait_queue *wq_get_first_waiter(
449                 struct mqueue_inode_info *info, int sr)
450 {
451         struct list_head *ptr;
452
453         ptr = info->e_wait_q[sr].list.prev;
454         if (ptr == &info->e_wait_q[sr].list)
455                 return NULL;
456         return list_entry(ptr, struct ext_wait_queue, list);
457 }
458
459 /* Auxiliary functions to manipulate messages' list */
460 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
461 {
462         int k;
463
464         k = info->attr.mq_curmsgs - 1;
465         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
466                 info->messages[k + 1] = info->messages[k];
467                 k--;
468         }
469         info->attr.mq_curmsgs++;
470         info->qsize += ptr->m_ts;
471         info->messages[k + 1] = ptr;
472 }
473
474 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
475 {
476         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
477         return info->messages[info->attr.mq_curmsgs];
478 }
479
480 static inline void set_cookie(struct sk_buff *skb, char code)
481 {
482         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
483 }
484
485 /*
486  * The next function is only to split too long sys_mq_timedsend
487  */
488 static void __do_notify(struct mqueue_inode_info *info)
489 {
490         /* notification
491          * invoked when there is registered process and there isn't process
492          * waiting synchronously for message AND state of queue changed from
493          * empty to not empty. Here we are sure that no one is waiting
494          * synchronously. */
495         if (info->notify_owner &&
496             info->attr.mq_curmsgs == 1) {
497                 struct siginfo sig_i;
498                 switch (info->notify.sigev_notify) {
499                 case SIGEV_NONE:
500                         break;
501                 case SIGEV_SIGNAL:
502                         /* sends signal */
503
504                         sig_i.si_signo = info->notify.sigev_signo;
505                         sig_i.si_errno = 0;
506                         sig_i.si_code = SI_MESGQ;
507                         sig_i.si_value = info->notify.sigev_value;
508                         sig_i.si_pid = task_tgid_nr_ns(current,
509                                                 ns_of_pid(info->notify_owner));
510                         sig_i.si_uid = current_uid();
511
512                         kill_pid_info(info->notify.sigev_signo,
513                                       &sig_i, info->notify_owner);
514                         break;
515                 case SIGEV_THREAD:
516                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
517                         netlink_sendskb(info->notify_sock, info->notify_cookie);
518                         break;
519                 }
520                 /* after notification unregisters process */
521                 put_pid(info->notify_owner);
522                 info->notify_owner = NULL;
523         }
524         wake_up(&info->wait_q);
525 }
526
527 static long prepare_timeout(struct timespec *p)
528 {
529         struct timespec nowts;
530         long timeout;
531
532         if (p) {
533                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
534                         || p->tv_nsec >= NSEC_PER_SEC))
535                         return -EINVAL;
536                 nowts = CURRENT_TIME;
537                 /* first subtract as jiffies can't be too big */
538                 p->tv_sec -= nowts.tv_sec;
539                 if (p->tv_nsec < nowts.tv_nsec) {
540                         p->tv_nsec += NSEC_PER_SEC;
541                         p->tv_sec--;
542                 }
543                 p->tv_nsec -= nowts.tv_nsec;
544                 if (p->tv_sec < 0)
545                         return 0;
546
547                 timeout = timespec_to_jiffies(p) + 1;
548         } else
549                 return MAX_SCHEDULE_TIMEOUT;
550
551         return timeout;
552 }
553
554 static void remove_notification(struct mqueue_inode_info *info)
555 {
556         if (info->notify_owner != NULL &&
557             info->notify.sigev_notify == SIGEV_THREAD) {
558                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
559                 netlink_sendskb(info->notify_sock, info->notify_cookie);
560         }
561         put_pid(info->notify_owner);
562         info->notify_owner = NULL;
563 }
564
565 static int mq_attr_ok(struct mq_attr *attr)
566 {
567         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
568                 return 0;
569         if (capable(CAP_SYS_RESOURCE)) {
570                 if (attr->mq_maxmsg > HARD_MSGMAX)
571                         return 0;
572         } else {
573                 if (attr->mq_maxmsg > msg_max ||
574                                 attr->mq_msgsize > msgsize_max)
575                         return 0;
576         }
577         /* check for overflow */
578         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
579                 return 0;
580         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
581             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
582             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
583                 return 0;
584         return 1;
585 }
586
587 /*
588  * Invoked when creating a new queue via sys_mq_open
589  */
590 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
591                         int oflag, mode_t mode, struct mq_attr *attr)
592 {
593         const struct cred *cred = current_cred();
594         struct file *result;
595         int ret;
596
597         if (attr) {
598                 ret = -EINVAL;
599                 if (!mq_attr_ok(attr))
600                         goto out;
601                 /* store for use during create */
602                 dentry->d_fsdata = attr;
603         }
604
605         mode &= ~current->fs->umask;
606         ret = mnt_want_write(mqueue_mnt);
607         if (ret)
608                 goto out;
609         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
610         dentry->d_fsdata = NULL;
611         if (ret)
612                 goto out_drop_write;
613
614         result = dentry_open(dentry, mqueue_mnt, oflag, cred);
615         /*
616          * dentry_open() took a persistent mnt_want_write(),
617          * so we can now drop this one.
618          */
619         mnt_drop_write(mqueue_mnt);
620         return result;
621
622 out_drop_write:
623         mnt_drop_write(mqueue_mnt);
624 out:
625         dput(dentry);
626         mntput(mqueue_mnt);
627         return ERR_PTR(ret);
628 }
629
630 /* Opens existing queue */
631 static struct file *do_open(struct dentry *dentry, int oflag)
632 {
633         const struct cred *cred = current_cred();
634
635         static const 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 (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
645                 dput(dentry);
646                 mntput(mqueue_mnt);
647                 return ERR_PTR(-EACCES);
648         }
649
650         return dentry_open(dentry, mqueue_mnt, oflag, cred);
651 }
652
653 SYSCALL_DEFINE4(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         struct mq_attr attr;
660         int fd, error;
661
662         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
663                 return -EFAULT;
664
665         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
666
667         if (IS_ERR(name = getname(u_name)))
668                 return PTR_ERR(name);
669
670         fd = get_unused_fd_flags(O_CLOEXEC);
671         if (fd < 0)
672                 goto out_putname;
673
674         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
675         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
676         if (IS_ERR(dentry)) {
677                 error = PTR_ERR(dentry);
678                 goto out_err;
679         }
680         mntget(mqueue_mnt);
681
682         if (oflag & O_CREAT) {
683                 if (dentry->d_inode) {  /* entry already exists */
684                         audit_inode(name, dentry);
685                         error = -EEXIST;
686                         if (oflag & O_EXCL)
687                                 goto out;
688                         filp = do_open(dentry, oflag);
689                 } else {
690                         filp = do_create(mqueue_mnt->mnt_root, dentry,
691                                                 oflag, mode,
692                                                 u_attr ? &attr : NULL);
693                 }
694         } else {
695                 error = -ENOENT;
696                 if (!dentry->d_inode)
697                         goto out;
698                 audit_inode(name, dentry);
699                 filp = do_open(dentry, oflag);
700         }
701
702         if (IS_ERR(filp)) {
703                 error = PTR_ERR(filp);
704                 goto out_putfd;
705         }
706
707         fd_install(fd, filp);
708         goto out_upsem;
709
710 out:
711         dput(dentry);
712         mntput(mqueue_mnt);
713 out_putfd:
714         put_unused_fd(fd);
715 out_err:
716         fd = error;
717 out_upsem:
718         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
719 out_putname:
720         putname(name);
721         return fd;
722 }
723
724 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
725 {
726         int err;
727         char *name;
728         struct dentry *dentry;
729         struct inode *inode = NULL;
730
731         name = getname(u_name);
732         if (IS_ERR(name))
733                 return PTR_ERR(name);
734
735         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
736                         I_MUTEX_PARENT);
737         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
738         if (IS_ERR(dentry)) {
739                 err = PTR_ERR(dentry);
740                 goto out_unlock;
741         }
742
743         if (!dentry->d_inode) {
744                 err = -ENOENT;
745                 goto out_err;
746         }
747
748         inode = dentry->d_inode;
749         if (inode)
750                 atomic_inc(&inode->i_count);
751         err = mnt_want_write(mqueue_mnt);
752         if (err)
753                 goto out_err;
754         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
755         mnt_drop_write(mqueue_mnt);
756 out_err:
757         dput(dentry);
758
759 out_unlock:
760         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
761         putname(name);
762         if (inode)
763                 iput(inode);
764
765         return err;
766 }
767
768 /* Pipelined send and receive functions.
769  *
770  * If a receiver finds no waiting message, then it registers itself in the
771  * list of waiting receivers. A sender checks that list before adding the new
772  * message into the message array. If there is a waiting receiver, then it
773  * bypasses the message array and directly hands the message over to the
774  * receiver.
775  * The receiver accepts the message and returns without grabbing the queue
776  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
777  * are necessary. The same algorithm is used for sysv semaphores, see
778  * ipc/sem.c for more details.
779  *
780  * The same algorithm is used for senders.
781  */
782
783 /* pipelined_send() - send a message directly to the task waiting in
784  * sys_mq_timedreceive() (without inserting message into a queue).
785  */
786 static inline void pipelined_send(struct mqueue_inode_info *info,
787                                   struct msg_msg *message,
788                                   struct ext_wait_queue *receiver)
789 {
790         receiver->msg = message;
791         list_del(&receiver->list);
792         receiver->state = STATE_PENDING;
793         wake_up_process(receiver->task);
794         smp_wmb();
795         receiver->state = STATE_READY;
796 }
797
798 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
799  * gets its message and put to the queue (we have one free place for sure). */
800 static inline void pipelined_receive(struct mqueue_inode_info *info)
801 {
802         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
803
804         if (!sender) {
805                 /* for poll */
806                 wake_up_interruptible(&info->wait_q);
807                 return;
808         }
809         msg_insert(sender->msg, info);
810         list_del(&sender->list);
811         sender->state = STATE_PENDING;
812         wake_up_process(sender->task);
813         smp_wmb();
814         sender->state = STATE_READY;
815 }
816
817 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
818                 size_t, msg_len, unsigned int, msg_prio,
819                 const struct timespec __user *, u_abs_timeout)
820 {
821         struct file *filp;
822         struct inode *inode;
823         struct ext_wait_queue wait;
824         struct ext_wait_queue *receiver;
825         struct msg_msg *msg_ptr;
826         struct mqueue_inode_info *info;
827         struct timespec ts, *p = NULL;
828         long timeout;
829         int ret;
830
831         if (u_abs_timeout) {
832                 if (copy_from_user(&ts, u_abs_timeout, 
833                                         sizeof(struct timespec)))
834                         return -EFAULT;
835                 p = &ts;
836         }
837
838         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
839                 return -EINVAL;
840
841         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
842         timeout = prepare_timeout(p);
843
844         ret = -EBADF;
845         filp = fget(mqdes);
846         if (unlikely(!filp))
847                 goto out;
848
849         inode = filp->f_path.dentry->d_inode;
850         if (unlikely(filp->f_op != &mqueue_file_operations))
851                 goto out_fput;
852         info = MQUEUE_I(inode);
853         audit_inode(NULL, filp->f_path.dentry);
854
855         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
856                 goto out_fput;
857
858         if (unlikely(msg_len > info->attr.mq_msgsize)) {
859                 ret = -EMSGSIZE;
860                 goto out_fput;
861         }
862
863         /* First try to allocate memory, before doing anything with
864          * existing queues. */
865         msg_ptr = load_msg(u_msg_ptr, msg_len);
866         if (IS_ERR(msg_ptr)) {
867                 ret = PTR_ERR(msg_ptr);
868                 goto out_fput;
869         }
870         msg_ptr->m_ts = msg_len;
871         msg_ptr->m_type = msg_prio;
872
873         spin_lock(&info->lock);
874
875         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
876                 if (filp->f_flags & O_NONBLOCK) {
877                         spin_unlock(&info->lock);
878                         ret = -EAGAIN;
879                 } else if (unlikely(timeout < 0)) {
880                         spin_unlock(&info->lock);
881                         ret = timeout;
882                 } else {
883                         wait.task = current;
884                         wait.msg = (void *) msg_ptr;
885                         wait.state = STATE_NONE;
886                         ret = wq_sleep(info, SEND, timeout, &wait);
887                 }
888                 if (ret < 0)
889                         free_msg(msg_ptr);
890         } else {
891                 receiver = wq_get_first_waiter(info, RECV);
892                 if (receiver) {
893                         pipelined_send(info, msg_ptr, receiver);
894                 } else {
895                         /* adds message to the queue */
896                         msg_insert(msg_ptr, info);
897                         __do_notify(info);
898                 }
899                 inode->i_atime = inode->i_mtime = inode->i_ctime =
900                                 CURRENT_TIME;
901                 spin_unlock(&info->lock);
902                 ret = 0;
903         }
904 out_fput:
905         fput(filp);
906 out:
907         return ret;
908 }
909
910 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
911                 size_t, msg_len, unsigned int __user *, u_msg_prio,
912                 const struct timespec __user *, u_abs_timeout)
913 {
914         long timeout;
915         ssize_t ret;
916         struct msg_msg *msg_ptr;
917         struct file *filp;
918         struct inode *inode;
919         struct mqueue_inode_info *info;
920         struct ext_wait_queue wait;
921         struct timespec ts, *p = NULL;
922
923         if (u_abs_timeout) {
924                 if (copy_from_user(&ts, u_abs_timeout, 
925                                         sizeof(struct timespec)))
926                         return -EFAULT;
927                 p = &ts;
928         }
929
930         audit_mq_sendrecv(mqdes, msg_len, 0, p);
931         timeout = prepare_timeout(p);
932
933         ret = -EBADF;
934         filp = fget(mqdes);
935         if (unlikely(!filp))
936                 goto out;
937
938         inode = filp->f_path.dentry->d_inode;
939         if (unlikely(filp->f_op != &mqueue_file_operations))
940                 goto out_fput;
941         info = MQUEUE_I(inode);
942         audit_inode(NULL, filp->f_path.dentry);
943
944         if (unlikely(!(filp->f_mode & FMODE_READ)))
945                 goto out_fput;
946
947         /* checks if buffer is big enough */
948         if (unlikely(msg_len < info->attr.mq_msgsize)) {
949                 ret = -EMSGSIZE;
950                 goto out_fput;
951         }
952
953         spin_lock(&info->lock);
954         if (info->attr.mq_curmsgs == 0) {
955                 if (filp->f_flags & O_NONBLOCK) {
956                         spin_unlock(&info->lock);
957                         ret = -EAGAIN;
958                         msg_ptr = NULL;
959                 } else if (unlikely(timeout < 0)) {
960                         spin_unlock(&info->lock);
961                         ret = timeout;
962                         msg_ptr = NULL;
963                 } else {
964                         wait.task = current;
965                         wait.state = STATE_NONE;
966                         ret = wq_sleep(info, RECV, timeout, &wait);
967                         msg_ptr = wait.msg;
968                 }
969         } else {
970                 msg_ptr = msg_get(info);
971
972                 inode->i_atime = inode->i_mtime = inode->i_ctime =
973                                 CURRENT_TIME;
974
975                 /* There is now free space in queue. */
976                 pipelined_receive(info);
977                 spin_unlock(&info->lock);
978                 ret = 0;
979         }
980         if (ret == 0) {
981                 ret = msg_ptr->m_ts;
982
983                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
984                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
985                         ret = -EFAULT;
986                 }
987                 free_msg(msg_ptr);
988         }
989 out_fput:
990         fput(filp);
991 out:
992         return ret;
993 }
994
995 /*
996  * Notes: the case when user wants us to deregister (with NULL as pointer)
997  * and he isn't currently owner of notification, will be silently discarded.
998  * It isn't explicitly defined in the POSIX.
999  */
1000 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1001                 const struct sigevent __user *, u_notification)
1002 {
1003         int ret;
1004         struct file *filp;
1005         struct sock *sock;
1006         struct inode *inode;
1007         struct sigevent notification;
1008         struct mqueue_inode_info *info;
1009         struct sk_buff *nc;
1010
1011         if (u_notification) {
1012                 if (copy_from_user(&notification, u_notification,
1013                                         sizeof(struct sigevent)))
1014                         return -EFAULT;
1015         }
1016
1017         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1018
1019         nc = NULL;
1020         sock = NULL;
1021         if (u_notification != NULL) {
1022                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1023                              notification.sigev_notify != SIGEV_SIGNAL &&
1024                              notification.sigev_notify != SIGEV_THREAD))
1025                         return -EINVAL;
1026                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1027                         !valid_signal(notification.sigev_signo)) {
1028                         return -EINVAL;
1029                 }
1030                 if (notification.sigev_notify == SIGEV_THREAD) {
1031                         long timeo;
1032
1033                         /* create the notify skb */
1034                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1035                         ret = -ENOMEM;
1036                         if (!nc)
1037                                 goto out;
1038                         ret = -EFAULT;
1039                         if (copy_from_user(nc->data,
1040                                         notification.sigev_value.sival_ptr,
1041                                         NOTIFY_COOKIE_LEN)) {
1042                                 goto out;
1043                         }
1044
1045                         /* TODO: add a header? */
1046                         skb_put(nc, NOTIFY_COOKIE_LEN);
1047                         /* and attach it to the socket */
1048 retry:
1049                         filp = fget(notification.sigev_signo);
1050                         ret = -EBADF;
1051                         if (!filp)
1052                                 goto out;
1053                         sock = netlink_getsockbyfilp(filp);
1054                         fput(filp);
1055                         if (IS_ERR(sock)) {
1056                                 ret = PTR_ERR(sock);
1057                                 sock = NULL;
1058                                 goto out;
1059                         }
1060
1061                         timeo = MAX_SCHEDULE_TIMEOUT;
1062                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1063                         if (ret == 1)
1064                                 goto retry;
1065                         if (ret) {
1066                                 sock = NULL;
1067                                 nc = NULL;
1068                                 goto out;
1069                         }
1070                 }
1071         }
1072
1073         ret = -EBADF;
1074         filp = fget(mqdes);
1075         if (!filp)
1076                 goto out;
1077
1078         inode = filp->f_path.dentry->d_inode;
1079         if (unlikely(filp->f_op != &mqueue_file_operations))
1080                 goto out_fput;
1081         info = MQUEUE_I(inode);
1082
1083         ret = 0;
1084         spin_lock(&info->lock);
1085         if (u_notification == NULL) {
1086                 if (info->notify_owner == task_tgid(current)) {
1087                         remove_notification(info);
1088                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1089                 }
1090         } else if (info->notify_owner != NULL) {
1091                 ret = -EBUSY;
1092         } else {
1093                 switch (notification.sigev_notify) {
1094                 case SIGEV_NONE:
1095                         info->notify.sigev_notify = SIGEV_NONE;
1096                         break;
1097                 case SIGEV_THREAD:
1098                         info->notify_sock = sock;
1099                         info->notify_cookie = nc;
1100                         sock = NULL;
1101                         nc = NULL;
1102                         info->notify.sigev_notify = SIGEV_THREAD;
1103                         break;
1104                 case SIGEV_SIGNAL:
1105                         info->notify.sigev_signo = notification.sigev_signo;
1106                         info->notify.sigev_value = notification.sigev_value;
1107                         info->notify.sigev_notify = SIGEV_SIGNAL;
1108                         break;
1109                 }
1110
1111                 info->notify_owner = get_pid(task_tgid(current));
1112                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1113         }
1114         spin_unlock(&info->lock);
1115 out_fput:
1116         fput(filp);
1117 out:
1118         if (sock) {
1119                 netlink_detachskb(sock, nc);
1120         } else if (nc) {
1121                 dev_kfree_skb(nc);
1122         }
1123         return ret;
1124 }
1125
1126 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1127                 const struct mq_attr __user *, u_mqstat,
1128                 struct mq_attr __user *, u_omqstat)
1129 {
1130         int ret;
1131         struct mq_attr mqstat, omqstat;
1132         struct file *filp;
1133         struct inode *inode;
1134         struct mqueue_inode_info *info;
1135
1136         if (u_mqstat != NULL) {
1137                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1138                         return -EFAULT;
1139                 if (mqstat.mq_flags & (~O_NONBLOCK))
1140                         return -EINVAL;
1141         }
1142
1143         ret = -EBADF;
1144         filp = fget(mqdes);
1145         if (!filp)
1146                 goto out;
1147
1148         inode = filp->f_path.dentry->d_inode;
1149         if (unlikely(filp->f_op != &mqueue_file_operations))
1150                 goto out_fput;
1151         info = MQUEUE_I(inode);
1152
1153         spin_lock(&info->lock);
1154
1155         omqstat = info->attr;
1156         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1157         if (u_mqstat) {
1158                 audit_mq_getsetattr(mqdes, &mqstat);
1159                 spin_lock(&filp->f_lock);
1160                 if (mqstat.mq_flags & O_NONBLOCK)
1161                         filp->f_flags |= O_NONBLOCK;
1162                 else
1163                         filp->f_flags &= ~O_NONBLOCK;
1164                 spin_unlock(&filp->f_lock);
1165
1166                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1167         }
1168
1169         spin_unlock(&info->lock);
1170
1171         ret = 0;
1172         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1173                                                 sizeof(struct mq_attr)))
1174                 ret = -EFAULT;
1175
1176 out_fput:
1177         fput(filp);
1178 out:
1179         return ret;
1180 }
1181
1182 static const struct inode_operations mqueue_dir_inode_operations = {
1183         .lookup = simple_lookup,
1184         .create = mqueue_create,
1185         .unlink = mqueue_unlink,
1186 };
1187
1188 static const struct file_operations mqueue_file_operations = {
1189         .flush = mqueue_flush_file,
1190         .poll = mqueue_poll_file,
1191         .read = mqueue_read_file,
1192 };
1193
1194 static struct super_operations mqueue_super_ops = {
1195         .alloc_inode = mqueue_alloc_inode,
1196         .destroy_inode = mqueue_destroy_inode,
1197         .statfs = simple_statfs,
1198         .delete_inode = mqueue_delete_inode,
1199         .drop_inode = generic_delete_inode,
1200 };
1201
1202 static struct file_system_type mqueue_fs_type = {
1203         .name = "mqueue",
1204         .get_sb = mqueue_get_sb,
1205         .kill_sb = kill_litter_super,
1206 };
1207
1208 static int msg_max_limit_min = MIN_MSGMAX;
1209 static int msg_max_limit_max = MAX_MSGMAX;
1210
1211 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1212 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1213
1214 static ctl_table mq_sysctls[] = {
1215         {
1216                 .procname       = "queues_max",
1217                 .data           = &queues_max,
1218                 .maxlen         = sizeof(int),
1219                 .mode           = 0644,
1220                 .proc_handler   = &proc_dointvec,
1221         },
1222         {
1223                 .procname       = "msg_max",
1224                 .data           = &msg_max,
1225                 .maxlen         = sizeof(int),
1226                 .mode           = 0644,
1227                 .proc_handler   = &proc_dointvec_minmax,
1228                 .extra1         = &msg_max_limit_min,
1229                 .extra2         = &msg_max_limit_max,
1230         },
1231         {
1232                 .procname       = "msgsize_max",
1233                 .data           = &msgsize_max,
1234                 .maxlen         = sizeof(int),
1235                 .mode           = 0644,
1236                 .proc_handler   = &proc_dointvec_minmax,
1237                 .extra1         = &msg_maxsize_limit_min,
1238                 .extra2         = &msg_maxsize_limit_max,
1239         },
1240         { .ctl_name = 0 }
1241 };
1242
1243 static ctl_table mq_sysctl_dir[] = {
1244         {
1245                 .procname       = "mqueue",
1246                 .mode           = 0555,
1247                 .child          = mq_sysctls,
1248         },
1249         { .ctl_name = 0 }
1250 };
1251
1252 static ctl_table mq_sysctl_root[] = {
1253         {
1254                 .ctl_name       = CTL_FS,
1255                 .procname       = "fs",
1256                 .mode           = 0555,
1257                 .child          = mq_sysctl_dir,
1258         },
1259         { .ctl_name = 0 }
1260 };
1261
1262 static int __init init_mqueue_fs(void)
1263 {
1264         int error;
1265
1266         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1267                                 sizeof(struct mqueue_inode_info), 0,
1268                                 SLAB_HWCACHE_ALIGN, init_once);
1269         if (mqueue_inode_cachep == NULL)
1270                 return -ENOMEM;
1271
1272         /* ignore failues - they are not fatal */
1273         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1274
1275         error = register_filesystem(&mqueue_fs_type);
1276         if (error)
1277                 goto out_sysctl;
1278
1279         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1280                 error = PTR_ERR(mqueue_mnt);
1281                 goto out_filesystem;
1282         }
1283
1284         /* internal initialization - not common for vfs */
1285         queues_count = 0;
1286         spin_lock_init(&mq_lock);
1287
1288         return 0;
1289
1290 out_filesystem:
1291         unregister_filesystem(&mqueue_fs_type);
1292 out_sysctl:
1293         if (mq_sysctl_table)
1294                 unregister_sysctl_table(mq_sysctl_table);
1295         kmem_cache_destroy(mqueue_inode_cachep);
1296         return error;
1297 }
1298
1299 __initcall(init_mqueue_fs);