Pull ia64-mutex-primitives into release branch
[linux-2.6] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (Michal.Wronski@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  * This file is released under the GPL.
12  */
13
14 #include <linux/capability.h>
15 #include <linux/init.h>
16 #include <linux/pagemap.h>
17 #include <linux/file.h>
18 #include <linux/mount.h>
19 #include <linux/namei.h>
20 #include <linux/sysctl.h>
21 #include <linux/poll.h>
22 #include <linux/mqueue.h>
23 #include <linux/msg.h>
24 #include <linux/skbuff.h>
25 #include <linux/netlink.h>
26 #include <linux/syscalls.h>
27 #include <linux/signal.h>
28 #include <net/sock.h>
29 #include "util.h"
30
31 #define MQUEUE_MAGIC    0x19800202
32 #define DIRENT_SIZE     20
33 #define FILENT_SIZE     80
34
35 #define SEND            0
36 #define RECV            1
37
38 #define STATE_NONE      0
39 #define STATE_PENDING   1
40 #define STATE_READY     2
41
42 /* used by sysctl */
43 #define FS_MQUEUE       1
44 #define CTL_QUEUESMAX   2
45 #define CTL_MSGMAX      3
46 #define CTL_MSGSIZEMAX  4
47
48 /* default values */
49 #define DFLT_QUEUESMAX  256     /* max number of message queues */
50 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
51 #define HARD_MSGMAX     (131072/sizeof(void*))
52 #define DFLT_MSGSIZEMAX 8192    /* max message size */
53
54 #define NOTIFY_COOKIE_LEN       32
55
56 struct ext_wait_queue {         /* queue of sleeping tasks */
57         struct task_struct *task;
58         struct list_head list;
59         struct msg_msg *msg;    /* ptr of loaded message */
60         int state;              /* one of STATE_* values */
61 };
62
63 struct mqueue_inode_info {
64         spinlock_t lock;
65         struct inode vfs_inode;
66         wait_queue_head_t wait_q;
67
68         struct msg_msg **messages;
69         struct mq_attr attr;
70
71         struct sigevent notify;
72         pid_t notify_owner;
73         struct user_struct *user;       /* user who created, for accounting */
74         struct sock *notify_sock;
75         struct sk_buff *notify_cookie;
76
77         /* for tasks waiting for free space and messages, respectively */
78         struct ext_wait_queue e_wait_q[2];
79
80         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
81 };
82
83 static struct inode_operations mqueue_dir_inode_operations;
84 static struct file_operations mqueue_file_operations;
85 static struct super_operations mqueue_super_ops;
86 static void remove_notification(struct mqueue_inode_info *info);
87
88 static spinlock_t mq_lock;
89 static kmem_cache_t *mqueue_inode_cachep;
90 static struct vfsmount *mqueue_mnt;
91
92 static unsigned int queues_count;
93 static unsigned int queues_max  = DFLT_QUEUESMAX;
94 static unsigned int msg_max     = DFLT_MSGMAX;
95 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
96
97 static struct ctl_table_header * mq_sysctl_table;
98
99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100 {
101         return container_of(inode, struct mqueue_inode_info, vfs_inode);
102 }
103
104 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
105                                                         struct mq_attr *attr)
106 {
107         struct inode *inode;
108
109         inode = new_inode(sb);
110         if (inode) {
111                 inode->i_mode = mode;
112                 inode->i_uid = current->fsuid;
113                 inode->i_gid = current->fsgid;
114                 inode->i_blksize = PAGE_CACHE_SIZE;
115                 inode->i_blocks = 0;
116                 inode->i_mtime = inode->i_ctime = inode->i_atime =
117                                 CURRENT_TIME;
118
119                 if (S_ISREG(mode)) {
120                         struct mqueue_inode_info *info;
121                         struct task_struct *p = current;
122                         struct user_struct *u = p->user;
123                         unsigned long mq_bytes, mq_msg_tblsz;
124
125                         inode->i_fop = &mqueue_file_operations;
126                         inode->i_size = FILENT_SIZE;
127                         /* mqueue specific info */
128                         info = MQUEUE_I(inode);
129                         spin_lock_init(&info->lock);
130                         init_waitqueue_head(&info->wait_q);
131                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
132                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
133                         info->messages = NULL;
134                         info->notify_owner = 0;
135                         info->qsize = 0;
136                         info->user = NULL;      /* set when all is ok */
137                         memset(&info->attr, 0, sizeof(info->attr));
138                         info->attr.mq_maxmsg = DFLT_MSGMAX;
139                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
140                         if (attr) {
141                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
142                                 info->attr.mq_msgsize = attr->mq_msgsize;
143                         }
144                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
145                         mq_bytes = (mq_msg_tblsz +
146                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
147
148                         spin_lock(&mq_lock);
149                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
150                             u->mq_bytes + mq_bytes >
151                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
152                                 spin_unlock(&mq_lock);
153                                 goto out_inode;
154                         }
155                         u->mq_bytes += mq_bytes;
156                         spin_unlock(&mq_lock);
157
158                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
159                         if (!info->messages) {
160                                 spin_lock(&mq_lock);
161                                 u->mq_bytes -= mq_bytes;
162                                 spin_unlock(&mq_lock);
163                                 goto out_inode;
164                         }
165                         /* all is ok */
166                         info->user = get_uid(u);
167                 } else if (S_ISDIR(mode)) {
168                         inode->i_nlink++;
169                         /* Some things misbehave if size == 0 on a directory */
170                         inode->i_size = 2 * DIRENT_SIZE;
171                         inode->i_op = &mqueue_dir_inode_operations;
172                         inode->i_fop = &simple_dir_operations;
173                 }
174         }
175         return inode;
176 out_inode:
177         make_bad_inode(inode);
178         iput(inode);
179         return NULL;
180 }
181
182 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
183 {
184         struct inode *inode;
185
186         sb->s_blocksize = PAGE_CACHE_SIZE;
187         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
188         sb->s_magic = MQUEUE_MAGIC;
189         sb->s_op = &mqueue_super_ops;
190
191         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
192         if (!inode)
193                 return -ENOMEM;
194
195         sb->s_root = d_alloc_root(inode);
196         if (!sb->s_root) {
197                 iput(inode);
198                 return -ENOMEM;
199         }
200
201         return 0;
202 }
203
204 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
205                                          int flags, const char *dev_name,
206                                          void *data)
207 {
208         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
209 }
210
211 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
212 {
213         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
214
215         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
216                 SLAB_CTOR_CONSTRUCTOR)
217                 inode_init_once(&p->vfs_inode);
218 }
219
220 static struct inode *mqueue_alloc_inode(struct super_block *sb)
221 {
222         struct mqueue_inode_info *ei;
223
224         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
225         if (!ei)
226                 return NULL;
227         return &ei->vfs_inode;
228 }
229
230 static void mqueue_destroy_inode(struct inode *inode)
231 {
232         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
233 }
234
235 static void mqueue_delete_inode(struct inode *inode)
236 {
237         struct mqueue_inode_info *info;
238         struct user_struct *user;
239         unsigned long mq_bytes;
240         int i;
241
242         if (S_ISDIR(inode->i_mode)) {
243                 clear_inode(inode);
244                 return;
245         }
246         info = MQUEUE_I(inode);
247         spin_lock(&info->lock);
248         for (i = 0; i < info->attr.mq_curmsgs; i++)
249                 free_msg(info->messages[i]);
250         kfree(info->messages);
251         spin_unlock(&info->lock);
252
253         clear_inode(inode);
254
255         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
256                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
257         user = info->user;
258         if (user) {
259                 spin_lock(&mq_lock);
260                 user->mq_bytes -= mq_bytes;
261                 queues_count--;
262                 spin_unlock(&mq_lock);
263                 free_uid(user);
264         }
265 }
266
267 static int mqueue_create(struct inode *dir, struct dentry *dentry,
268                                 int mode, struct nameidata *nd)
269 {
270         struct inode *inode;
271         struct mq_attr *attr = dentry->d_fsdata;
272         int error;
273
274         spin_lock(&mq_lock);
275         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
276                 error = -ENOSPC;
277                 goto out_lock;
278         }
279         queues_count++;
280         spin_unlock(&mq_lock);
281
282         inode = mqueue_get_inode(dir->i_sb, mode, attr);
283         if (!inode) {
284                 error = -ENOMEM;
285                 spin_lock(&mq_lock);
286                 queues_count--;
287                 goto out_lock;
288         }
289
290         dir->i_size += DIRENT_SIZE;
291         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
292
293         d_instantiate(dentry, inode);
294         dget(dentry);
295         return 0;
296 out_lock:
297         spin_unlock(&mq_lock);
298         return error;
299 }
300
301 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
302 {
303         struct inode *inode = dentry->d_inode;
304
305         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
306         dir->i_size -= DIRENT_SIZE;
307         inode->i_nlink--;
308         dput(dentry);
309         return 0;
310 }
311
312 /*
313 *       This is routine for system read from queue file.
314 *       To avoid mess with doing here some sort of mq_receive we allow
315 *       to read only queue size & notification info (the only values
316 *       that are interesting from user point of view and aren't accessible
317 *       through std routines)
318 */
319 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
320                                 size_t count, loff_t * off)
321 {
322         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
323         char buffer[FILENT_SIZE];
324         size_t slen;
325         loff_t o;
326
327         if (!count)
328                 return 0;
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                         info->notify_owner);
339         spin_unlock(&info->lock);
340         buffer[sizeof(buffer)-1] = '\0';
341         slen = strlen(buffer)+1;
342
343         o = *off;
344         if (o > slen)
345                 return 0;
346
347         if (o + count > slen)
348                 count = slen - o;
349
350         if (copy_to_user(u_data, buffer + o, count))
351                 return -EFAULT;
352
353         *off = o + count;
354         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
355         return count;
356 }
357
358 static int mqueue_flush_file(struct file *filp)
359 {
360         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
361
362         spin_lock(&info->lock);
363         if (current->tgid == info->notify_owner)
364                 remove_notification(info);
365
366         spin_unlock(&info->lock);
367         return 0;
368 }
369
370 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
371 {
372         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
373         int retval = 0;
374
375         poll_wait(filp, &info->wait_q, poll_tab);
376
377         spin_lock(&info->lock);
378         if (info->attr.mq_curmsgs)
379                 retval = POLLIN | POLLRDNORM;
380
381         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
382                 retval |= POLLOUT | POLLWRNORM;
383         spin_unlock(&info->lock);
384
385         return retval;
386 }
387
388 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
389 static void wq_add(struct mqueue_inode_info *info, int sr,
390                         struct ext_wait_queue *ewp)
391 {
392         struct ext_wait_queue *walk;
393
394         ewp->task = current;
395
396         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
397                 if (walk->task->static_prio <= current->static_prio) {
398                         list_add_tail(&ewp->list, &walk->list);
399                         return;
400                 }
401         }
402         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
403 }
404
405 /*
406  * Puts current task to sleep. Caller must hold queue lock. After return
407  * lock isn't held.
408  * sr: SEND or RECV
409  */
410 static int wq_sleep(struct mqueue_inode_info *info, int sr,
411                         long timeout, struct ext_wait_queue *ewp)
412 {
413         int retval;
414         signed long time;
415
416         wq_add(info, sr, ewp);
417
418         for (;;) {
419                 set_current_state(TASK_INTERRUPTIBLE);
420
421                 spin_unlock(&info->lock);
422                 time = schedule_timeout(timeout);
423
424                 while (ewp->state == STATE_PENDING)
425                         cpu_relax();
426
427                 if (ewp->state == STATE_READY) {
428                         retval = 0;
429                         goto out;
430                 }
431                 spin_lock(&info->lock);
432                 if (ewp->state == STATE_READY) {
433                         retval = 0;
434                         goto out_unlock;
435                 }
436                 if (signal_pending(current)) {
437                         retval = -ERESTARTSYS;
438                         break;
439                 }
440                 if (time == 0) {
441                         retval = -ETIMEDOUT;
442                         break;
443                 }
444         }
445         list_del(&ewp->list);
446 out_unlock:
447         spin_unlock(&info->lock);
448 out:
449         return retval;
450 }
451
452 /*
453  * Returns waiting task that should be serviced first or NULL if none exists
454  */
455 static struct ext_wait_queue *wq_get_first_waiter(
456                 struct mqueue_inode_info *info, int sr)
457 {
458         struct list_head *ptr;
459
460         ptr = info->e_wait_q[sr].list.prev;
461         if (ptr == &info->e_wait_q[sr].list)
462                 return NULL;
463         return list_entry(ptr, struct ext_wait_queue, list);
464 }
465
466 /* Auxiliary functions to manipulate messages' list */
467 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
468 {
469         int k;
470
471         k = info->attr.mq_curmsgs - 1;
472         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
473                 info->messages[k + 1] = info->messages[k];
474                 k--;
475         }
476         info->attr.mq_curmsgs++;
477         info->qsize += ptr->m_ts;
478         info->messages[k + 1] = ptr;
479 }
480
481 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
482 {
483         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
484         return info->messages[info->attr.mq_curmsgs];
485 }
486
487 static inline void set_cookie(struct sk_buff *skb, char code)
488 {
489         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
490 }
491
492 /*
493  * The next function is only to split too long sys_mq_timedsend
494  */
495 static void __do_notify(struct mqueue_inode_info *info)
496 {
497         /* notification
498          * invoked when there is registered process and there isn't process
499          * waiting synchronously for message AND state of queue changed from
500          * empty to not empty. Here we are sure that no one is waiting
501          * synchronously. */
502         if (info->notify_owner &&
503             info->attr.mq_curmsgs == 1) {
504                 struct siginfo sig_i;
505                 switch (info->notify.sigev_notify) {
506                 case SIGEV_NONE:
507                         break;
508                 case SIGEV_SIGNAL:
509                         /* sends signal */
510
511                         sig_i.si_signo = info->notify.sigev_signo;
512                         sig_i.si_errno = 0;
513                         sig_i.si_code = SI_MESGQ;
514                         sig_i.si_value = info->notify.sigev_value;
515                         sig_i.si_pid = current->tgid;
516                         sig_i.si_uid = current->uid;
517
518                         kill_proc_info(info->notify.sigev_signo,
519                                        &sig_i, info->notify_owner);
520                         break;
521                 case SIGEV_THREAD:
522                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
523                         netlink_sendskb(info->notify_sock,
524                                         info->notify_cookie, 0);
525                         break;
526                 }
527                 /* after notification unregisters process */
528                 info->notify_owner = 0;
529         }
530         wake_up(&info->wait_q);
531 }
532
533 static long prepare_timeout(const struct timespec __user *u_arg)
534 {
535         struct timespec ts, nowts;
536         long timeout;
537
538         if (u_arg) {
539                 if (unlikely(copy_from_user(&ts, u_arg,
540                                         sizeof(struct timespec))))
541                         return -EFAULT;
542
543                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
544                         || ts.tv_nsec >= NSEC_PER_SEC))
545                         return -EINVAL;
546                 nowts = CURRENT_TIME;
547                 /* first subtract as jiffies can't be too big */
548                 ts.tv_sec -= nowts.tv_sec;
549                 if (ts.tv_nsec < nowts.tv_nsec) {
550                         ts.tv_nsec += NSEC_PER_SEC;
551                         ts.tv_sec--;
552                 }
553                 ts.tv_nsec -= nowts.tv_nsec;
554                 if (ts.tv_sec < 0)
555                         return 0;
556
557                 timeout = timespec_to_jiffies(&ts) + 1;
558         } else
559                 return MAX_SCHEDULE_TIMEOUT;
560
561         return timeout;
562 }
563
564 static void remove_notification(struct mqueue_inode_info *info)
565 {
566         if (info->notify_owner != 0 &&
567             info->notify.sigev_notify == SIGEV_THREAD) {
568                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
569                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
570         }
571         info->notify_owner = 0;
572 }
573
574 static int mq_attr_ok(struct mq_attr *attr)
575 {
576         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
577                 return 0;
578         if (capable(CAP_SYS_RESOURCE)) {
579                 if (attr->mq_maxmsg > HARD_MSGMAX)
580                         return 0;
581         } else {
582                 if (attr->mq_maxmsg > msg_max ||
583                                 attr->mq_msgsize > msgsize_max)
584                         return 0;
585         }
586         /* check for overflow */
587         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
588                 return 0;
589         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
590             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
591             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
592                 return 0;
593         return 1;
594 }
595
596 /*
597  * Invoked when creating a new queue via sys_mq_open
598  */
599 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
600                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
601 {
602         struct mq_attr attr;
603         int ret;
604
605         if (u_attr) {
606                 ret = -EFAULT;
607                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
608                         goto out;
609                 ret = -EINVAL;
610                 if (!mq_attr_ok(&attr))
611                         goto out;
612                 /* store for use during create */
613                 dentry->d_fsdata = &attr;
614         }
615
616         mode &= ~current->fs->umask;
617         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
618         dentry->d_fsdata = NULL;
619         if (ret)
620                 goto out;
621
622         return dentry_open(dentry, mqueue_mnt, oflag);
623
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 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
634                                         MAY_READ | MAY_WRITE };
635
636         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
637                 dput(dentry);
638                 mntput(mqueue_mnt);
639                 return ERR_PTR(-EINVAL);
640         }
641
642         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
643                 dput(dentry);
644                 mntput(mqueue_mnt);
645                 return ERR_PTR(-EACCES);
646         }
647
648         return dentry_open(dentry, mqueue_mnt, oflag);
649 }
650
651 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
652                                 struct mq_attr __user *u_attr)
653 {
654         struct dentry *dentry;
655         struct file *filp;
656         char *name;
657         int fd, error;
658
659         if (IS_ERR(name = getname(u_name)))
660                 return PTR_ERR(name);
661
662         fd = get_unused_fd();
663         if (fd < 0)
664                 goto out_putname;
665
666         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
667         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
668         if (IS_ERR(dentry)) {
669                 error = PTR_ERR(dentry);
670                 goto out_err;
671         }
672         mntget(mqueue_mnt);
673
674         if (oflag & O_CREAT) {
675                 if (dentry->d_inode) {  /* entry already exists */
676                         error = -EEXIST;
677                         if (oflag & O_EXCL)
678                                 goto out;
679                         filp = do_open(dentry, oflag);
680                 } else {
681                         filp = do_create(mqueue_mnt->mnt_root, dentry,
682                                                 oflag, mode, u_attr);
683                 }
684         } else {
685                 error = -ENOENT;
686                 if (!dentry->d_inode)
687                         goto out;
688                 filp = do_open(dentry, oflag);
689         }
690
691         if (IS_ERR(filp)) {
692                 error = PTR_ERR(filp);
693                 goto out_putfd;
694         }
695
696         set_close_on_exec(fd, 1);
697         fd_install(fd, filp);
698         goto out_upsem;
699
700 out:
701         dput(dentry);
702         mntput(mqueue_mnt);
703 out_putfd:
704         put_unused_fd(fd);
705 out_err:
706         fd = error;
707 out_upsem:
708         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
709 out_putname:
710         putname(name);
711         return fd;
712 }
713
714 asmlinkage long sys_mq_unlink(const char __user *u_name)
715 {
716         int err;
717         char *name;
718         struct dentry *dentry;
719         struct inode *inode = NULL;
720
721         name = getname(u_name);
722         if (IS_ERR(name))
723                 return PTR_ERR(name);
724
725         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
726         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
727         if (IS_ERR(dentry)) {
728                 err = PTR_ERR(dentry);
729                 goto out_unlock;
730         }
731
732         if (!dentry->d_inode) {
733                 err = -ENOENT;
734                 goto out_err;
735         }
736
737         inode = dentry->d_inode;
738         if (inode)
739                 atomic_inc(&inode->i_count);
740
741         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
742 out_err:
743         dput(dentry);
744
745 out_unlock:
746         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
747         putname(name);
748         if (inode)
749                 iput(inode);
750
751         return err;
752 }
753
754 /* Pipelined send and receive functions.
755  *
756  * If a receiver finds no waiting message, then it registers itself in the
757  * list of waiting receivers. A sender checks that list before adding the new
758  * message into the message array. If there is a waiting receiver, then it
759  * bypasses the message array and directly hands the message over to the
760  * receiver.
761  * The receiver accepts the message and returns without grabbing the queue
762  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
763  * are necessary. The same algorithm is used for sysv semaphores, see
764  * ipc/sem.c fore more details.
765  *
766  * The same algorithm is used for senders.
767  */
768
769 /* pipelined_send() - send a message directly to the task waiting in
770  * sys_mq_timedreceive() (without inserting message into a queue).
771  */
772 static inline void pipelined_send(struct mqueue_inode_info *info,
773                                   struct msg_msg *message,
774                                   struct ext_wait_queue *receiver)
775 {
776         receiver->msg = message;
777         list_del(&receiver->list);
778         receiver->state = STATE_PENDING;
779         wake_up_process(receiver->task);
780         smp_wmb();
781         receiver->state = STATE_READY;
782 }
783
784 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
785  * gets its message and put to the queue (we have one free place for sure). */
786 static inline void pipelined_receive(struct mqueue_inode_info *info)
787 {
788         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
789
790         if (!sender) {
791                 /* for poll */
792                 wake_up_interruptible(&info->wait_q);
793                 return;
794         }
795         msg_insert(sender->msg, info);
796         list_del(&sender->list);
797         sender->state = STATE_PENDING;
798         wake_up_process(sender->task);
799         smp_wmb();
800         sender->state = STATE_READY;
801 }
802
803 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
804         size_t msg_len, unsigned int msg_prio,
805         const struct timespec __user *u_abs_timeout)
806 {
807         struct file *filp;
808         struct inode *inode;
809         struct ext_wait_queue wait;
810         struct ext_wait_queue *receiver;
811         struct msg_msg *msg_ptr;
812         struct mqueue_inode_info *info;
813         long timeout;
814         int ret;
815
816         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
817                 return -EINVAL;
818
819         timeout = prepare_timeout(u_abs_timeout);
820
821         ret = -EBADF;
822         filp = fget(mqdes);
823         if (unlikely(!filp))
824                 goto out;
825
826         inode = filp->f_dentry->d_inode;
827         if (unlikely(filp->f_op != &mqueue_file_operations))
828                 goto out_fput;
829         info = MQUEUE_I(inode);
830
831         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
832                 goto out_fput;
833
834         if (unlikely(msg_len > info->attr.mq_msgsize)) {
835                 ret = -EMSGSIZE;
836                 goto out_fput;
837         }
838
839         /* First try to allocate memory, before doing anything with
840          * existing queues. */
841         msg_ptr = load_msg(u_msg_ptr, msg_len);
842         if (IS_ERR(msg_ptr)) {
843                 ret = PTR_ERR(msg_ptr);
844                 goto out_fput;
845         }
846         msg_ptr->m_ts = msg_len;
847         msg_ptr->m_type = msg_prio;
848
849         spin_lock(&info->lock);
850
851         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
852                 if (filp->f_flags & O_NONBLOCK) {
853                         spin_unlock(&info->lock);
854                         ret = -EAGAIN;
855                 } else if (unlikely(timeout < 0)) {
856                         spin_unlock(&info->lock);
857                         ret = timeout;
858                 } else {
859                         wait.task = current;
860                         wait.msg = (void *) msg_ptr;
861                         wait.state = STATE_NONE;
862                         ret = wq_sleep(info, SEND, timeout, &wait);
863                 }
864                 if (ret < 0)
865                         free_msg(msg_ptr);
866         } else {
867                 receiver = wq_get_first_waiter(info, RECV);
868                 if (receiver) {
869                         pipelined_send(info, msg_ptr, receiver);
870                 } else {
871                         /* adds message to the queue */
872                         msg_insert(msg_ptr, info);
873                         __do_notify(info);
874                 }
875                 inode->i_atime = inode->i_mtime = inode->i_ctime =
876                                 CURRENT_TIME;
877                 spin_unlock(&info->lock);
878                 ret = 0;
879         }
880 out_fput:
881         fput(filp);
882 out:
883         return ret;
884 }
885
886 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
887         size_t msg_len, unsigned int __user *u_msg_prio,
888         const struct timespec __user *u_abs_timeout)
889 {
890         long timeout;
891         ssize_t ret;
892         struct msg_msg *msg_ptr;
893         struct file *filp;
894         struct inode *inode;
895         struct mqueue_inode_info *info;
896         struct ext_wait_queue wait;
897
898         timeout = prepare_timeout(u_abs_timeout);
899
900         ret = -EBADF;
901         filp = fget(mqdes);
902         if (unlikely(!filp))
903                 goto out;
904
905         inode = filp->f_dentry->d_inode;
906         if (unlikely(filp->f_op != &mqueue_file_operations))
907                 goto out_fput;
908         info = MQUEUE_I(inode);
909
910         if (unlikely(!(filp->f_mode & FMODE_READ)))
911                 goto out_fput;
912
913         /* checks if buffer is big enough */
914         if (unlikely(msg_len < info->attr.mq_msgsize)) {
915                 ret = -EMSGSIZE;
916                 goto out_fput;
917         }
918
919         spin_lock(&info->lock);
920         if (info->attr.mq_curmsgs == 0) {
921                 if (filp->f_flags & O_NONBLOCK) {
922                         spin_unlock(&info->lock);
923                         ret = -EAGAIN;
924                         msg_ptr = NULL;
925                 } else if (unlikely(timeout < 0)) {
926                         spin_unlock(&info->lock);
927                         ret = timeout;
928                         msg_ptr = NULL;
929                 } else {
930                         wait.task = current;
931                         wait.state = STATE_NONE;
932                         ret = wq_sleep(info, RECV, timeout, &wait);
933                         msg_ptr = wait.msg;
934                 }
935         } else {
936                 msg_ptr = msg_get(info);
937
938                 inode->i_atime = inode->i_mtime = inode->i_ctime =
939                                 CURRENT_TIME;
940
941                 /* There is now free space in queue. */
942                 pipelined_receive(info);
943                 spin_unlock(&info->lock);
944                 ret = 0;
945         }
946         if (ret == 0) {
947                 ret = msg_ptr->m_ts;
948
949                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
950                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
951                         ret = -EFAULT;
952                 }
953                 free_msg(msg_ptr);
954         }
955 out_fput:
956         fput(filp);
957 out:
958         return ret;
959 }
960
961 /*
962  * Notes: the case when user wants us to deregister (with NULL as pointer)
963  * and he isn't currently owner of notification, will be silently discarded.
964  * It isn't explicitly defined in the POSIX.
965  */
966 asmlinkage long sys_mq_notify(mqd_t mqdes,
967                                 const struct sigevent __user *u_notification)
968 {
969         int ret;
970         struct file *filp;
971         struct sock *sock;
972         struct inode *inode;
973         struct sigevent notification;
974         struct mqueue_inode_info *info;
975         struct sk_buff *nc;
976
977         nc = NULL;
978         sock = NULL;
979         if (u_notification != NULL) {
980                 if (copy_from_user(&notification, u_notification,
981                                         sizeof(struct sigevent)))
982                         return -EFAULT;
983
984                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
985                              notification.sigev_notify != SIGEV_SIGNAL &&
986                              notification.sigev_notify != SIGEV_THREAD))
987                         return -EINVAL;
988                 if (notification.sigev_notify == SIGEV_SIGNAL &&
989                         !valid_signal(notification.sigev_signo)) {
990                         return -EINVAL;
991                 }
992                 if (notification.sigev_notify == SIGEV_THREAD) {
993                         /* create the notify skb */
994                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
995                         ret = -ENOMEM;
996                         if (!nc)
997                                 goto out;
998                         ret = -EFAULT;
999                         if (copy_from_user(nc->data,
1000                                         notification.sigev_value.sival_ptr,
1001                                         NOTIFY_COOKIE_LEN)) {
1002                                 goto out;
1003                         }
1004
1005                         /* TODO: add a header? */
1006                         skb_put(nc, NOTIFY_COOKIE_LEN);
1007                         /* and attach it to the socket */
1008 retry:
1009                         filp = fget(notification.sigev_signo);
1010                         ret = -EBADF;
1011                         if (!filp)
1012                                 goto out;
1013                         sock = netlink_getsockbyfilp(filp);
1014                         fput(filp);
1015                         if (IS_ERR(sock)) {
1016                                 ret = PTR_ERR(sock);
1017                                 sock = NULL;
1018                                 goto out;
1019                         }
1020
1021                         ret = netlink_attachskb(sock, nc, 0,
1022                                         MAX_SCHEDULE_TIMEOUT, NULL);
1023                         if (ret == 1)
1024                                 goto retry;
1025                         if (ret) {
1026                                 sock = NULL;
1027                                 nc = NULL;
1028                                 goto out;
1029                         }
1030                 }
1031         }
1032
1033         ret = -EBADF;
1034         filp = fget(mqdes);
1035         if (!filp)
1036                 goto out;
1037
1038         inode = filp->f_dentry->d_inode;
1039         if (unlikely(filp->f_op != &mqueue_file_operations))
1040                 goto out_fput;
1041         info = MQUEUE_I(inode);
1042
1043         ret = 0;
1044         spin_lock(&info->lock);
1045         if (u_notification == NULL) {
1046                 if (info->notify_owner == current->tgid) {
1047                         remove_notification(info);
1048                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1049                 }
1050         } else if (info->notify_owner != 0) {
1051                 ret = -EBUSY;
1052         } else {
1053                 switch (notification.sigev_notify) {
1054                 case SIGEV_NONE:
1055                         info->notify.sigev_notify = SIGEV_NONE;
1056                         break;
1057                 case SIGEV_THREAD:
1058                         info->notify_sock = sock;
1059                         info->notify_cookie = nc;
1060                         sock = NULL;
1061                         nc = NULL;
1062                         info->notify.sigev_notify = SIGEV_THREAD;
1063                         break;
1064                 case SIGEV_SIGNAL:
1065                         info->notify.sigev_signo = notification.sigev_signo;
1066                         info->notify.sigev_value = notification.sigev_value;
1067                         info->notify.sigev_notify = SIGEV_SIGNAL;
1068                         break;
1069                 }
1070                 info->notify_owner = current->tgid;
1071                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1072         }
1073         spin_unlock(&info->lock);
1074 out_fput:
1075         fput(filp);
1076 out:
1077         if (sock) {
1078                 netlink_detachskb(sock, nc);
1079         } else if (nc) {
1080                 dev_kfree_skb(nc);
1081         }
1082         return ret;
1083 }
1084
1085 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1086                         const struct mq_attr __user *u_mqstat,
1087                         struct mq_attr __user *u_omqstat)
1088 {
1089         int ret;
1090         struct mq_attr mqstat, omqstat;
1091         struct file *filp;
1092         struct inode *inode;
1093         struct mqueue_inode_info *info;
1094
1095         if (u_mqstat != NULL) {
1096                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1097                         return -EFAULT;
1098                 if (mqstat.mq_flags & (~O_NONBLOCK))
1099                         return -EINVAL;
1100         }
1101
1102         ret = -EBADF;
1103         filp = fget(mqdes);
1104         if (!filp)
1105                 goto out;
1106
1107         inode = filp->f_dentry->d_inode;
1108         if (unlikely(filp->f_op != &mqueue_file_operations))
1109                 goto out_fput;
1110         info = MQUEUE_I(inode);
1111
1112         spin_lock(&info->lock);
1113
1114         omqstat = info->attr;
1115         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1116         if (u_mqstat) {
1117                 if (mqstat.mq_flags & O_NONBLOCK)
1118                         filp->f_flags |= O_NONBLOCK;
1119                 else
1120                         filp->f_flags &= ~O_NONBLOCK;
1121
1122                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1123         }
1124
1125         spin_unlock(&info->lock);
1126
1127         ret = 0;
1128         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1129                                                 sizeof(struct mq_attr)))
1130                 ret = -EFAULT;
1131
1132 out_fput:
1133         fput(filp);
1134 out:
1135         return ret;
1136 }
1137
1138 static struct inode_operations mqueue_dir_inode_operations = {
1139         .lookup = simple_lookup,
1140         .create = mqueue_create,
1141         .unlink = mqueue_unlink,
1142 };
1143
1144 static struct file_operations mqueue_file_operations = {
1145         .flush = mqueue_flush_file,
1146         .poll = mqueue_poll_file,
1147         .read = mqueue_read_file,
1148 };
1149
1150 static struct super_operations mqueue_super_ops = {
1151         .alloc_inode = mqueue_alloc_inode,
1152         .destroy_inode = mqueue_destroy_inode,
1153         .statfs = simple_statfs,
1154         .delete_inode = mqueue_delete_inode,
1155         .drop_inode = generic_delete_inode,
1156 };
1157
1158 static struct file_system_type mqueue_fs_type = {
1159         .name = "mqueue",
1160         .get_sb = mqueue_get_sb,
1161         .kill_sb = kill_litter_super,
1162 };
1163
1164 static int msg_max_limit_min = DFLT_MSGMAX;
1165 static int msg_max_limit_max = HARD_MSGMAX;
1166
1167 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1168 static int msg_maxsize_limit_max = INT_MAX;
1169
1170 static ctl_table mq_sysctls[] = {
1171         {
1172                 .ctl_name       = CTL_QUEUESMAX,
1173                 .procname       = "queues_max",
1174                 .data           = &queues_max,
1175                 .maxlen         = sizeof(int),
1176                 .mode           = 0644,
1177                 .proc_handler   = &proc_dointvec,
1178         },
1179         {
1180                 .ctl_name       = CTL_MSGMAX,
1181                 .procname       = "msg_max",
1182                 .data           = &msg_max,
1183                 .maxlen         = sizeof(int),
1184                 .mode           = 0644,
1185                 .proc_handler   = &proc_dointvec_minmax,
1186                 .extra1         = &msg_max_limit_min,
1187                 .extra2         = &msg_max_limit_max,
1188         },
1189         {
1190                 .ctl_name       = CTL_MSGSIZEMAX,
1191                 .procname       = "msgsize_max",
1192                 .data           = &msgsize_max,
1193                 .maxlen         = sizeof(int),
1194                 .mode           = 0644,
1195                 .proc_handler   = &proc_dointvec_minmax,
1196                 .extra1         = &msg_maxsize_limit_min,
1197                 .extra2         = &msg_maxsize_limit_max,
1198         },
1199         { .ctl_name = 0 }
1200 };
1201
1202 static ctl_table mq_sysctl_dir[] = {
1203         {
1204                 .ctl_name       = FS_MQUEUE,
1205                 .procname       = "mqueue",
1206                 .mode           = 0555,
1207                 .child          = mq_sysctls,
1208         },
1209         { .ctl_name = 0 }
1210 };
1211
1212 static ctl_table mq_sysctl_root[] = {
1213         {
1214                 .ctl_name       = CTL_FS,
1215                 .procname       = "fs",
1216                 .mode           = 0555,
1217                 .child          = mq_sysctl_dir,
1218         },
1219         { .ctl_name = 0 }
1220 };
1221
1222 static int __init init_mqueue_fs(void)
1223 {
1224         int error;
1225
1226         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1227                                 sizeof(struct mqueue_inode_info), 0,
1228                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1229         if (mqueue_inode_cachep == NULL)
1230                 return -ENOMEM;
1231
1232         /* ignore failues - they are not fatal */
1233         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1234
1235         error = register_filesystem(&mqueue_fs_type);
1236         if (error)
1237                 goto out_sysctl;
1238
1239         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1240                 error = PTR_ERR(mqueue_mnt);
1241                 goto out_filesystem;
1242         }
1243
1244         /* internal initialization - not common for vfs */
1245         queues_count = 0;
1246         spin_lock_init(&mq_lock);
1247
1248         return 0;
1249
1250 out_filesystem:
1251         unregister_filesystem(&mqueue_fs_type);
1252 out_sysctl:
1253         if (mq_sysctl_table)
1254                 unregister_sysctl_table(mq_sysctl_table);
1255         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1256                 printk(KERN_INFO
1257                         "mqueue_inode_cache: not all structures were freed\n");
1258         }
1259         return error;
1260 }
1261
1262 __initcall(init_mqueue_fs);