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