[PATCH] mqueue: nested locking annotation
[linux-2.6.git] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@gmail.com)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *                          Manfred Spraul          (manfred@colorfullife.com)
10  *
11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12  *
13  * This file is released under the GPL.
14  */
15
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32
33 #include <net/sock.h>
34 #include "util.h"
35
36 #define MQUEUE_MAGIC    0x19800202
37 #define DIRENT_SIZE     20
38 #define FILENT_SIZE     80
39
40 #define SEND            0
41 #define RECV            1
42
43 #define STATE_NONE      0
44 #define STATE_PENDING   1
45 #define STATE_READY     2
46
47 /* used by sysctl */
48 #define FS_MQUEUE       1
49 #define CTL_QUEUESMAX   2
50 #define CTL_MSGMAX      3
51 #define CTL_MSGSIZEMAX  4
52
53 /* default values */
54 #define DFLT_QUEUESMAX  256     /* max number of message queues */
55 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
56 #define HARD_MSGMAX     (131072/sizeof(void*))
57 #define DFLT_MSGSIZEMAX 8192    /* max message size */
58
59
60 struct ext_wait_queue {         /* queue of sleeping tasks */
61         struct task_struct *task;
62         struct list_head list;
63         struct msg_msg *msg;    /* ptr of loaded message */
64         int state;              /* one of STATE_* values */
65 };
66
67 struct mqueue_inode_info {
68         spinlock_t lock;
69         struct inode vfs_inode;
70         wait_queue_head_t wait_q;
71
72         struct msg_msg **messages;
73         struct mq_attr attr;
74
75         struct sigevent notify;
76         struct pid* notify_owner;
77         struct user_struct *user;       /* user who created, for accounting */
78         struct sock *notify_sock;
79         struct sk_buff *notify_cookie;
80
81         /* for tasks waiting for free space and messages, respectively */
82         struct ext_wait_queue e_wait_q[2];
83
84         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
85 };
86
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
91
92 static spinlock_t mq_lock;
93 static struct kmem_cache *mqueue_inode_cachep;
94 static struct vfsmount *mqueue_mnt;
95
96 static unsigned int queues_count;
97 static unsigned int queues_max  = DFLT_QUEUESMAX;
98 static unsigned int msg_max     = DFLT_MSGMAX;
99 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
100
101 static struct ctl_table_header * mq_sysctl_table;
102
103 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
104 {
105         return container_of(inode, struct mqueue_inode_info, vfs_inode);
106 }
107
108 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
109                                                         struct mq_attr *attr)
110 {
111         struct inode *inode;
112
113         inode = new_inode(sb);
114         if (inode) {
115                 inode->i_mode = mode;
116                 inode->i_uid = current->fsuid;
117                 inode->i_gid = current->fsgid;
118                 inode->i_blocks = 0;
119                 inode->i_mtime = inode->i_ctime = inode->i_atime =
120                                 CURRENT_TIME;
121
122                 if (S_ISREG(mode)) {
123                         struct mqueue_inode_info *info;
124                         struct task_struct *p = current;
125                         struct user_struct *u = p->user;
126                         unsigned long mq_bytes, mq_msg_tblsz;
127
128                         inode->i_fop = &mqueue_file_operations;
129                         inode->i_size = FILENT_SIZE;
130                         /* mqueue specific info */
131                         info = MQUEUE_I(inode);
132                         spin_lock_init(&info->lock);
133                         init_waitqueue_head(&info->wait_q);
134                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
135                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
136                         info->messages = NULL;
137                         info->notify_owner = NULL;
138                         info->qsize = 0;
139                         info->user = NULL;      /* set when all is ok */
140                         memset(&info->attr, 0, sizeof(info->attr));
141                         info->attr.mq_maxmsg = DFLT_MSGMAX;
142                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
143                         if (attr) {
144                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
145                                 info->attr.mq_msgsize = attr->mq_msgsize;
146                         }
147                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
148                         mq_bytes = (mq_msg_tblsz +
149                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
150
151                         spin_lock(&mq_lock);
152                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
153                             u->mq_bytes + mq_bytes >
154                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
155                                 spin_unlock(&mq_lock);
156                                 goto out_inode;
157                         }
158                         u->mq_bytes += mq_bytes;
159                         spin_unlock(&mq_lock);
160
161                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
162                         if (!info->messages) {
163                                 spin_lock(&mq_lock);
164                                 u->mq_bytes -= mq_bytes;
165                                 spin_unlock(&mq_lock);
166                                 goto out_inode;
167                         }
168                         /* all is ok */
169                         info->user = get_uid(u);
170                 } else if (S_ISDIR(mode)) {
171                         inc_nlink(inode);
172                         /* Some things misbehave if size == 0 on a directory */
173                         inode->i_size = 2 * DIRENT_SIZE;
174                         inode->i_op = &mqueue_dir_inode_operations;
175                         inode->i_fop = &simple_dir_operations;
176                 }
177         }
178         return inode;
179 out_inode:
180         make_bad_inode(inode);
181         iput(inode);
182         return NULL;
183 }
184
185 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
186 {
187         struct inode *inode;
188
189         sb->s_blocksize = PAGE_CACHE_SIZE;
190         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
191         sb->s_magic = MQUEUE_MAGIC;
192         sb->s_op = &mqueue_super_ops;
193
194         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
195         if (!inode)
196                 return -ENOMEM;
197
198         sb->s_root = d_alloc_root(inode);
199         if (!sb->s_root) {
200                 iput(inode);
201                 return -ENOMEM;
202         }
203
204         return 0;
205 }
206
207 static int mqueue_get_sb(struct file_system_type *fs_type,
208                          int flags, const char *dev_name,
209                          void *data, struct vfsmount *mnt)
210 {
211         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
212 }
213
214 static void init_once(void *foo, struct kmem_cache * cachep, unsigned long flags)
215 {
216         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
217
218         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
219                 SLAB_CTOR_CONSTRUCTOR)
220                 inode_init_once(&p->vfs_inode);
221 }
222
223 static struct inode *mqueue_alloc_inode(struct super_block *sb)
224 {
225         struct mqueue_inode_info *ei;
226
227         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
228         if (!ei)
229                 return NULL;
230         return &ei->vfs_inode;
231 }
232
233 static void mqueue_destroy_inode(struct inode *inode)
234 {
235         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
236 }
237
238 static void mqueue_delete_inode(struct inode *inode)
239 {
240         struct mqueue_inode_info *info;
241         struct user_struct *user;
242         unsigned long mq_bytes;
243         int i;
244
245         if (S_ISDIR(inode->i_mode)) {
246                 clear_inode(inode);
247                 return;
248         }
249         info = MQUEUE_I(inode);
250         spin_lock(&info->lock);
251         for (i = 0; i < info->attr.mq_curmsgs; i++)
252                 free_msg(info->messages[i]);
253         kfree(info->messages);
254         spin_unlock(&info->lock);
255
256         clear_inode(inode);
257
258         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
259                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
260         user = info->user;
261         if (user) {
262                 spin_lock(&mq_lock);
263                 user->mq_bytes -= mq_bytes;
264                 queues_count--;
265                 spin_unlock(&mq_lock);
266                 free_uid(user);
267         }
268 }
269
270 static int mqueue_create(struct inode *dir, struct dentry *dentry,
271                                 int mode, struct nameidata *nd)
272 {
273         struct inode *inode;
274         struct mq_attr *attr = dentry->d_fsdata;
275         int error;
276
277         spin_lock(&mq_lock);
278         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
279                 error = -ENOSPC;
280                 goto out_lock;
281         }
282         queues_count++;
283         spin_unlock(&mq_lock);
284
285         inode = mqueue_get_inode(dir->i_sb, mode, attr);
286         if (!inode) {
287                 error = -ENOMEM;
288                 spin_lock(&mq_lock);
289                 queues_count--;
290                 goto out_lock;
291         }
292
293         dir->i_size += DIRENT_SIZE;
294         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
295
296         d_instantiate(dentry, inode);
297         dget(dentry);
298         return 0;
299 out_lock:
300         spin_unlock(&mq_lock);
301         return error;
302 }
303
304 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
305 {
306         struct inode *inode = dentry->d_inode;
307
308         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
309         dir->i_size -= DIRENT_SIZE;
310         drop_nlink(inode);
311         dput(dentry);
312         return 0;
313 }
314
315 /*
316 *       This is routine for system read from queue file.
317 *       To avoid mess with doing here some sort of mq_receive we allow
318 *       to read only queue size & notification info (the only values
319 *       that are interesting from user point of view and aren't accessible
320 *       through std routines)
321 */
322 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
323                                 size_t count, loff_t * off)
324 {
325         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
326         char buffer[FILENT_SIZE];
327         size_t slen;
328         loff_t o;
329
330         if (!count)
331                 return 0;
332
333         spin_lock(&info->lock);
334         snprintf(buffer, sizeof(buffer),
335                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
336                         info->qsize,
337                         info->notify_owner ? info->notify.sigev_notify : 0,
338                         (info->notify_owner &&
339                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
340                                 info->notify.sigev_signo : 0,
341                         pid_nr(info->notify_owner));
342         spin_unlock(&info->lock);
343         buffer[sizeof(buffer)-1] = '\0';
344         slen = strlen(buffer)+1;
345
346         o = *off;
347         if (o > slen)
348                 return 0;
349
350         if (o + count > slen)
351                 count = slen - o;
352
353         if (copy_to_user(u_data, buffer + o, count))
354                 return -EFAULT;
355
356         *off = o + count;
357         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
358         return count;
359 }
360
361 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
362 {
363         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
364
365         spin_lock(&info->lock);
366         if (task_tgid(current) == info->notify_owner)
367                 remove_notification(info);
368
369         spin_unlock(&info->lock);
370         return 0;
371 }
372
373 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
374 {
375         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
376         int retval = 0;
377
378         poll_wait(filp, &info->wait_q, poll_tab);
379
380         spin_lock(&info->lock);
381         if (info->attr.mq_curmsgs)
382                 retval = POLLIN | POLLRDNORM;
383
384         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
385                 retval |= POLLOUT | POLLWRNORM;
386         spin_unlock(&info->lock);
387
388         return retval;
389 }
390
391 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
392 static void wq_add(struct mqueue_inode_info *info, int sr,
393                         struct ext_wait_queue *ewp)
394 {
395         struct ext_wait_queue *walk;
396
397         ewp->task = current;
398
399         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
400                 if (walk->task->static_prio <= current->static_prio) {
401                         list_add_tail(&ewp->list, &walk->list);
402                         return;
403                 }
404         }
405         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
406 }
407
408 /*
409  * Puts current task to sleep. Caller must hold queue lock. After return
410  * lock isn't held.
411  * sr: SEND or RECV
412  */
413 static int wq_sleep(struct mqueue_inode_info *info, int sr,
414                         long timeout, struct ext_wait_queue *ewp)
415 {
416         int retval;
417         signed long time;
418
419         wq_add(info, sr, ewp);
420
421         for (;;) {
422                 set_current_state(TASK_INTERRUPTIBLE);
423
424                 spin_unlock(&info->lock);
425                 time = schedule_timeout(timeout);
426
427                 while (ewp->state == STATE_PENDING)
428                         cpu_relax();
429
430                 if (ewp->state == STATE_READY) {
431                         retval = 0;
432                         goto out;
433                 }
434                 spin_lock(&info->lock);
435                 if (ewp->state == STATE_READY) {
436                         retval = 0;
437                         goto out_unlock;
438                 }
439                 if (signal_pending(current)) {
440                         retval = -ERESTARTSYS;
441                         break;
442                 }
443                 if (time == 0) {
444                         retval = -ETIMEDOUT;
445                         break;
446                 }
447         }
448         list_del(&ewp->list);
449 out_unlock:
450         spin_unlock(&info->lock);
451 out:
452         return retval;
453 }
454
455 /*
456  * Returns waiting task that should be serviced first or NULL if none exists
457  */
458 static struct ext_wait_queue *wq_get_first_waiter(
459                 struct mqueue_inode_info *info, int sr)
460 {
461         struct list_head *ptr;
462
463         ptr = info->e_wait_q[sr].list.prev;
464         if (ptr == &info->e_wait_q[sr].list)
465                 return NULL;
466         return list_entry(ptr, struct ext_wait_queue, list);
467 }
468
469 /* Auxiliary functions to manipulate messages' list */
470 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
471 {
472         int k;
473
474         k = info->attr.mq_curmsgs - 1;
475         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
476                 info->messages[k + 1] = info->messages[k];
477                 k--;
478         }
479         info->attr.mq_curmsgs++;
480         info->qsize += ptr->m_ts;
481         info->messages[k + 1] = ptr;
482 }
483
484 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
485 {
486         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
487         return info->messages[info->attr.mq_curmsgs];
488 }
489
490 static inline void set_cookie(struct sk_buff *skb, char code)
491 {
492         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
493 }
494
495 /*
496  * The next function is only to split too long sys_mq_timedsend
497  */
498 static void __do_notify(struct mqueue_inode_info *info)
499 {
500         /* notification
501          * invoked when there is registered process and there isn't process
502          * waiting synchronously for message AND state of queue changed from
503          * empty to not empty. Here we are sure that no one is waiting
504          * synchronously. */
505         if (info->notify_owner &&
506             info->attr.mq_curmsgs == 1) {
507                 struct siginfo sig_i;
508                 switch (info->notify.sigev_notify) {
509                 case SIGEV_NONE:
510                         break;
511                 case SIGEV_SIGNAL:
512                         /* sends signal */
513
514                         sig_i.si_signo = info->notify.sigev_signo;
515                         sig_i.si_errno = 0;
516                         sig_i.si_code = SI_MESGQ;
517                         sig_i.si_value = info->notify.sigev_value;
518                         sig_i.si_pid = current->tgid;
519                         sig_i.si_uid = current->uid;
520
521                         kill_pid_info(info->notify.sigev_signo,
522                                       &sig_i, info->notify_owner);
523                         break;
524                 case SIGEV_THREAD:
525                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
526                         netlink_sendskb(info->notify_sock,
527                                         info->notify_cookie, 0);
528                         break;
529                 }
530                 /* after notification unregisters process */
531                 put_pid(info->notify_owner);
532                 info->notify_owner = NULL;
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 != NULL &&
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         put_pid(info->notify_owner);
576         info->notify_owner = NULL;
577 }
578
579 static int mq_attr_ok(struct mq_attr *attr)
580 {
581         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
582                 return 0;
583         if (capable(CAP_SYS_RESOURCE)) {
584                 if (attr->mq_maxmsg > HARD_MSGMAX)
585                         return 0;
586         } else {
587                 if (attr->mq_maxmsg > msg_max ||
588                                 attr->mq_msgsize > msgsize_max)
589                         return 0;
590         }
591         /* check for overflow */
592         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
593                 return 0;
594         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
595             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
596             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
597                 return 0;
598         return 1;
599 }
600
601 /*
602  * Invoked when creating a new queue via sys_mq_open
603  */
604 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
605                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
606 {
607         struct mq_attr attr;
608         int ret;
609
610         if (u_attr) {
611                 ret = -EFAULT;
612                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
613                         goto out;
614                 ret = -EINVAL;
615                 if (!mq_attr_ok(&attr))
616                         goto out;
617                 /* store for use during create */
618                 dentry->d_fsdata = &attr;
619         }
620
621         mode &= ~current->fs->umask;
622         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
623         dentry->d_fsdata = NULL;
624         if (ret)
625                 goto out;
626
627         return dentry_open(dentry, mqueue_mnt, oflag);
628
629 out:
630         dput(dentry);
631         mntput(mqueue_mnt);
632         return ERR_PTR(ret);
633 }
634
635 /* Opens existing queue */
636 static struct file *do_open(struct dentry *dentry, int oflag)
637 {
638 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
639                                         MAY_READ | MAY_WRITE };
640
641         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
642                 dput(dentry);
643                 mntput(mqueue_mnt);
644                 return ERR_PTR(-EINVAL);
645         }
646
647         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
648                 dput(dentry);
649                 mntput(mqueue_mnt);
650                 return ERR_PTR(-EACCES);
651         }
652
653         return dentry_open(dentry, mqueue_mnt, oflag);
654 }
655
656 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
657                                 struct mq_attr __user *u_attr)
658 {
659         struct dentry *dentry;
660         struct file *filp;
661         char *name;
662         int fd, error;
663
664         error = audit_mq_open(oflag, mode, u_attr);
665         if (error != 0)
666                 return error;
667
668         if (IS_ERR(name = getname(u_name)))
669                 return PTR_ERR(name);
670
671         fd = get_unused_fd();
672         if (fd < 0)
673                 goto out_putname;
674
675         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
676         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
677         if (IS_ERR(dentry)) {
678                 error = PTR_ERR(dentry);
679                 goto out_err;
680         }
681         mntget(mqueue_mnt);
682
683         if (oflag & O_CREAT) {
684                 if (dentry->d_inode) {  /* entry already exists */
685                         error = -EEXIST;
686                         if (oflag & O_EXCL)
687                                 goto out;
688                         filp = do_open(dentry, oflag);
689                 } else {
690                         filp = do_create(mqueue_mnt->mnt_root, dentry,
691                                                 oflag, mode, u_attr);
692                 }
693         } else {
694                 error = -ENOENT;
695                 if (!dentry->d_inode)
696                         goto out;
697                 filp = do_open(dentry, oflag);
698         }
699
700         if (IS_ERR(filp)) {
701                 error = PTR_ERR(filp);
702                 goto out_putfd;
703         }
704
705         set_close_on_exec(fd, 1);
706         fd_install(fd, filp);
707         goto out_upsem;
708
709 out:
710         dput(dentry);
711         mntput(mqueue_mnt);
712 out_putfd:
713         put_unused_fd(fd);
714 out_err:
715         fd = error;
716 out_upsem:
717         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
718 out_putname:
719         putname(name);
720         return fd;
721 }
722
723 asmlinkage long sys_mq_unlink(const char __user *u_name)
724 {
725         int err;
726         char *name;
727         struct dentry *dentry;
728         struct inode *inode = NULL;
729
730         name = getname(u_name);
731         if (IS_ERR(name))
732                 return PTR_ERR(name);
733
734         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
735                         I_MUTEX_PARENT);
736         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
737         if (IS_ERR(dentry)) {
738                 err = PTR_ERR(dentry);
739                 goto out_unlock;
740         }
741
742         if (!dentry->d_inode) {
743                 err = -ENOENT;
744                 goto out_err;
745         }
746
747         inode = dentry->d_inode;
748         if (inode)
749                 atomic_inc(&inode->i_count);
750
751         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
752 out_err:
753         dput(dentry);
754
755 out_unlock:
756         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
757         putname(name);
758         if (inode)
759                 iput(inode);
760
761         return err;
762 }
763
764 /* Pipelined send and receive functions.
765  *
766  * If a receiver finds no waiting message, then it registers itself in the
767  * list of waiting receivers. A sender checks that list before adding the new
768  * message into the message array. If there is a waiting receiver, then it
769  * bypasses the message array and directly hands the message over to the
770  * receiver.
771  * The receiver accepts the message and returns without grabbing the queue
772  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
773  * are necessary. The same algorithm is used for sysv semaphores, see
774  * ipc/sem.c for more details.
775  *
776  * The same algorithm is used for senders.
777  */
778
779 /* pipelined_send() - send a message directly to the task waiting in
780  * sys_mq_timedreceive() (without inserting message into a queue).
781  */
782 static inline void pipelined_send(struct mqueue_inode_info *info,
783                                   struct msg_msg *message,
784                                   struct ext_wait_queue *receiver)
785 {
786         receiver->msg = message;
787         list_del(&receiver->list);
788         receiver->state = STATE_PENDING;
789         wake_up_process(receiver->task);
790         smp_wmb();
791         receiver->state = STATE_READY;
792 }
793
794 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
795  * gets its message and put to the queue (we have one free place for sure). */
796 static inline void pipelined_receive(struct mqueue_inode_info *info)
797 {
798         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
799
800         if (!sender) {
801                 /* for poll */
802                 wake_up_interruptible(&info->wait_q);
803                 return;
804         }
805         msg_insert(sender->msg, info);
806         list_del(&sender->list);
807         sender->state = STATE_PENDING;
808         wake_up_process(sender->task);
809         smp_wmb();
810         sender->state = STATE_READY;
811 }
812
813 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
814         size_t msg_len, unsigned int msg_prio,
815         const struct timespec __user *u_abs_timeout)
816 {
817         struct file *filp;
818         struct inode *inode;
819         struct ext_wait_queue wait;
820         struct ext_wait_queue *receiver;
821         struct msg_msg *msg_ptr;
822         struct mqueue_inode_info *info;
823         long timeout;
824         int ret;
825
826         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
827         if (ret != 0)
828                 return ret;
829
830         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
831                 return -EINVAL;
832
833         timeout = prepare_timeout(u_abs_timeout);
834
835         ret = -EBADF;
836         filp = fget(mqdes);
837         if (unlikely(!filp))
838                 goto out;
839
840         inode = filp->f_path.dentry->d_inode;
841         if (unlikely(filp->f_op != &mqueue_file_operations))
842                 goto out_fput;
843         info = MQUEUE_I(inode);
844
845         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
846                 goto out_fput;
847
848         if (unlikely(msg_len > info->attr.mq_msgsize)) {
849                 ret = -EMSGSIZE;
850                 goto out_fput;
851         }
852
853         /* First try to allocate memory, before doing anything with
854          * existing queues. */
855         msg_ptr = load_msg(u_msg_ptr, msg_len);
856         if (IS_ERR(msg_ptr)) {
857                 ret = PTR_ERR(msg_ptr);
858                 goto out_fput;
859         }
860         msg_ptr->m_ts = msg_len;
861         msg_ptr->m_type = msg_prio;
862
863         spin_lock(&info->lock);
864
865         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
866                 if (filp->f_flags & O_NONBLOCK) {
867                         spin_unlock(&info->lock);
868                         ret = -EAGAIN;
869                 } else if (unlikely(timeout < 0)) {
870                         spin_unlock(&info->lock);
871                         ret = timeout;
872                 } else {
873                         wait.task = current;
874                         wait.msg = (void *) msg_ptr;
875                         wait.state = STATE_NONE;
876                         ret = wq_sleep(info, SEND, timeout, &wait);
877                 }
878                 if (ret < 0)
879                         free_msg(msg_ptr);
880         } else {
881                 receiver = wq_get_first_waiter(info, RECV);
882                 if (receiver) {
883                         pipelined_send(info, msg_ptr, receiver);
884                 } else {
885                         /* adds message to the queue */
886                         msg_insert(msg_ptr, info);
887                         __do_notify(info);
888                 }
889                 inode->i_atime = inode->i_mtime = inode->i_ctime =
890                                 CURRENT_TIME;
891                 spin_unlock(&info->lock);
892                 ret = 0;
893         }
894 out_fput:
895         fput(filp);
896 out:
897         return ret;
898 }
899
900 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
901         size_t msg_len, unsigned int __user *u_msg_prio,
902         const struct timespec __user *u_abs_timeout)
903 {
904         long timeout;
905         ssize_t ret;
906         struct msg_msg *msg_ptr;
907         struct file *filp;
908         struct inode *inode;
909         struct mqueue_inode_info *info;
910         struct ext_wait_queue wait;
911
912         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
913         if (ret != 0)
914                 return ret;
915
916         timeout = prepare_timeout(u_abs_timeout);
917
918         ret = -EBADF;
919         filp = fget(mqdes);
920         if (unlikely(!filp))
921                 goto out;
922
923         inode = filp->f_path.dentry->d_inode;
924         if (unlikely(filp->f_op != &mqueue_file_operations))
925                 goto out_fput;
926         info = MQUEUE_I(inode);
927
928         if (unlikely(!(filp->f_mode & FMODE_READ)))
929                 goto out_fput;
930
931         /* checks if buffer is big enough */
932         if (unlikely(msg_len < info->attr.mq_msgsize)) {
933                 ret = -EMSGSIZE;
934                 goto out_fput;
935         }
936
937         spin_lock(&info->lock);
938         if (info->attr.mq_curmsgs == 0) {
939                 if (filp->f_flags & O_NONBLOCK) {
940                         spin_unlock(&info->lock);
941                         ret = -EAGAIN;
942                         msg_ptr = NULL;
943                 } else if (unlikely(timeout < 0)) {
944                         spin_unlock(&info->lock);
945                         ret = timeout;
946                         msg_ptr = NULL;
947                 } else {
948                         wait.task = current;
949                         wait.state = STATE_NONE;
950                         ret = wq_sleep(info, RECV, timeout, &wait);
951                         msg_ptr = wait.msg;
952                 }
953         } else {
954                 msg_ptr = msg_get(info);
955
956                 inode->i_atime = inode->i_mtime = inode->i_ctime =
957                                 CURRENT_TIME;
958
959                 /* There is now free space in queue. */
960                 pipelined_receive(info);
961                 spin_unlock(&info->lock);
962                 ret = 0;
963         }
964         if (ret == 0) {
965                 ret = msg_ptr->m_ts;
966
967                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
968                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
969                         ret = -EFAULT;
970                 }
971                 free_msg(msg_ptr);
972         }
973 out_fput:
974         fput(filp);
975 out:
976         return ret;
977 }
978
979 /*
980  * Notes: the case when user wants us to deregister (with NULL as pointer)
981  * and he isn't currently owner of notification, will be silently discarded.
982  * It isn't explicitly defined in the POSIX.
983  */
984 asmlinkage long sys_mq_notify(mqd_t mqdes,
985                                 const struct sigevent __user *u_notification)
986 {
987         int ret;
988         struct file *filp;
989         struct sock *sock;
990         struct inode *inode;
991         struct sigevent notification;
992         struct mqueue_inode_info *info;
993         struct sk_buff *nc;
994
995         ret = audit_mq_notify(mqdes, u_notification);
996         if (ret != 0)
997                 return ret;
998
999         nc = NULL;
1000         sock = NULL;
1001         if (u_notification != NULL) {
1002                 if (copy_from_user(&notification, u_notification,
1003                                         sizeof(struct sigevent)))
1004                         return -EFAULT;
1005
1006                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1007                              notification.sigev_notify != SIGEV_SIGNAL &&
1008                              notification.sigev_notify != SIGEV_THREAD))
1009                         return -EINVAL;
1010                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1011                         !valid_signal(notification.sigev_signo)) {
1012                         return -EINVAL;
1013                 }
1014                 if (notification.sigev_notify == SIGEV_THREAD) {
1015                         /* create the notify skb */
1016                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1017                         ret = -ENOMEM;
1018                         if (!nc)
1019                                 goto out;
1020                         ret = -EFAULT;
1021                         if (copy_from_user(nc->data,
1022                                         notification.sigev_value.sival_ptr,
1023                                         NOTIFY_COOKIE_LEN)) {
1024                                 goto out;
1025                         }
1026
1027                         /* TODO: add a header? */
1028                         skb_put(nc, NOTIFY_COOKIE_LEN);
1029                         /* and attach it to the socket */
1030 retry:
1031                         filp = fget(notification.sigev_signo);
1032                         ret = -EBADF;
1033                         if (!filp)
1034                                 goto out;
1035                         sock = netlink_getsockbyfilp(filp);
1036                         fput(filp);
1037                         if (IS_ERR(sock)) {
1038                                 ret = PTR_ERR(sock);
1039                                 sock = NULL;
1040                                 goto out;
1041                         }
1042
1043                         ret = netlink_attachskb(sock, nc, 0,
1044                                         MAX_SCHEDULE_TIMEOUT, NULL);
1045                         if (ret == 1)
1046                                 goto retry;
1047                         if (ret) {
1048                                 sock = NULL;
1049                                 nc = NULL;
1050                                 goto out;
1051                         }
1052                 }
1053         }
1054
1055         ret = -EBADF;
1056         filp = fget(mqdes);
1057         if (!filp)
1058                 goto out;
1059
1060         inode = filp->f_path.dentry->d_inode;
1061         if (unlikely(filp->f_op != &mqueue_file_operations))
1062                 goto out_fput;
1063         info = MQUEUE_I(inode);
1064
1065         ret = 0;
1066         spin_lock(&info->lock);
1067         if (u_notification == NULL) {
1068                 if (info->notify_owner == task_tgid(current)) {
1069                         remove_notification(info);
1070                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1071                 }
1072         } else if (info->notify_owner != NULL) {
1073                 ret = -EBUSY;
1074         } else {
1075                 switch (notification.sigev_notify) {
1076                 case SIGEV_NONE:
1077                         info->notify.sigev_notify = SIGEV_NONE;
1078                         break;
1079                 case SIGEV_THREAD:
1080                         info->notify_sock = sock;
1081                         info->notify_cookie = nc;
1082                         sock = NULL;
1083                         nc = NULL;
1084                         info->notify.sigev_notify = SIGEV_THREAD;
1085                         break;
1086                 case SIGEV_SIGNAL:
1087                         info->notify.sigev_signo = notification.sigev_signo;
1088                         info->notify.sigev_value = notification.sigev_value;
1089                         info->notify.sigev_notify = SIGEV_SIGNAL;
1090                         break;
1091                 }
1092
1093                 info->notify_owner = get_pid(task_tgid(current));
1094                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1095         }
1096         spin_unlock(&info->lock);
1097 out_fput:
1098         fput(filp);
1099 out:
1100         if (sock) {
1101                 netlink_detachskb(sock, nc);
1102         } else if (nc) {
1103                 dev_kfree_skb(nc);
1104         }
1105         return ret;
1106 }
1107
1108 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1109                         const struct mq_attr __user *u_mqstat,
1110                         struct mq_attr __user *u_omqstat)
1111 {
1112         int ret;
1113         struct mq_attr mqstat, omqstat;
1114         struct file *filp;
1115         struct inode *inode;
1116         struct mqueue_inode_info *info;
1117
1118         if (u_mqstat != NULL) {
1119                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1120                         return -EFAULT;
1121                 if (mqstat.mq_flags & (~O_NONBLOCK))
1122                         return -EINVAL;
1123         }
1124
1125         ret = -EBADF;
1126         filp = fget(mqdes);
1127         if (!filp)
1128                 goto out;
1129
1130         inode = filp->f_path.dentry->d_inode;
1131         if (unlikely(filp->f_op != &mqueue_file_operations))
1132                 goto out_fput;
1133         info = MQUEUE_I(inode);
1134
1135         spin_lock(&info->lock);
1136
1137         omqstat = info->attr;
1138         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1139         if (u_mqstat) {
1140                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1141                 if (ret != 0)
1142                         goto out;
1143                 if (mqstat.mq_flags & O_NONBLOCK)
1144                         filp->f_flags |= O_NONBLOCK;
1145                 else
1146                         filp->f_flags &= ~O_NONBLOCK;
1147
1148                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1149         }
1150
1151         spin_unlock(&info->lock);
1152
1153         ret = 0;
1154         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1155                                                 sizeof(struct mq_attr)))
1156                 ret = -EFAULT;
1157
1158 out_fput:
1159         fput(filp);
1160 out:
1161         return ret;
1162 }
1163
1164 static const struct inode_operations mqueue_dir_inode_operations = {
1165         .lookup = simple_lookup,
1166         .create = mqueue_create,
1167         .unlink = mqueue_unlink,
1168 };
1169
1170 static const struct file_operations mqueue_file_operations = {
1171         .flush = mqueue_flush_file,
1172         .poll = mqueue_poll_file,
1173         .read = mqueue_read_file,
1174 };
1175
1176 static struct super_operations mqueue_super_ops = {
1177         .alloc_inode = mqueue_alloc_inode,
1178         .destroy_inode = mqueue_destroy_inode,
1179         .statfs = simple_statfs,
1180         .delete_inode = mqueue_delete_inode,
1181         .drop_inode = generic_delete_inode,
1182 };
1183
1184 static struct file_system_type mqueue_fs_type = {
1185         .name = "mqueue",
1186         .get_sb = mqueue_get_sb,
1187         .kill_sb = kill_litter_super,
1188 };
1189
1190 static int msg_max_limit_min = DFLT_MSGMAX;
1191 static int msg_max_limit_max = HARD_MSGMAX;
1192
1193 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1194 static int msg_maxsize_limit_max = INT_MAX;
1195
1196 static ctl_table mq_sysctls[] = {
1197         {
1198                 .ctl_name       = CTL_QUEUESMAX,
1199                 .procname       = "queues_max",
1200                 .data           = &queues_max,
1201                 .maxlen         = sizeof(int),
1202                 .mode           = 0644,
1203                 .proc_handler   = &proc_dointvec,
1204         },
1205         {
1206                 .ctl_name       = CTL_MSGMAX,
1207                 .procname       = "msg_max",
1208                 .data           = &msg_max,
1209                 .maxlen         = sizeof(int),
1210                 .mode           = 0644,
1211                 .proc_handler   = &proc_dointvec_minmax,
1212                 .extra1         = &msg_max_limit_min,
1213                 .extra2         = &msg_max_limit_max,
1214         },
1215         {
1216                 .ctl_name       = CTL_MSGSIZEMAX,
1217                 .procname       = "msgsize_max",
1218                 .data           = &msgsize_max,
1219                 .maxlen         = sizeof(int),
1220                 .mode           = 0644,
1221                 .proc_handler   = &proc_dointvec_minmax,
1222                 .extra1         = &msg_maxsize_limit_min,
1223                 .extra2         = &msg_maxsize_limit_max,
1224         },
1225         { .ctl_name = 0 }
1226 };
1227
1228 static ctl_table mq_sysctl_dir[] = {
1229         {
1230                 .ctl_name       = FS_MQUEUE,
1231                 .procname       = "mqueue",
1232                 .mode           = 0555,
1233                 .child          = mq_sysctls,
1234         },
1235         { .ctl_name = 0 }
1236 };
1237
1238 static ctl_table mq_sysctl_root[] = {
1239         {
1240                 .ctl_name       = CTL_FS,
1241                 .procname       = "fs",
1242                 .mode           = 0555,
1243                 .child          = mq_sysctl_dir,
1244         },
1245         { .ctl_name = 0 }
1246 };
1247
1248 static int __init init_mqueue_fs(void)
1249 {
1250         int error;
1251
1252         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1253                                 sizeof(struct mqueue_inode_info), 0,
1254                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1255         if (mqueue_inode_cachep == NULL)
1256                 return -ENOMEM;
1257
1258         /* ignore failues - they are not fatal */
1259         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1260
1261         error = register_filesystem(&mqueue_fs_type);
1262         if (error)
1263                 goto out_sysctl;
1264
1265         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1266                 error = PTR_ERR(mqueue_mnt);
1267                 goto out_filesystem;
1268         }
1269
1270         /* internal initialization - not common for vfs */
1271         queues_count = 0;
1272         spin_lock_init(&mq_lock);
1273
1274         return 0;
1275
1276 out_filesystem:
1277         unregister_filesystem(&mqueue_fs_type);
1278 out_sysctl:
1279         if (mq_sysctl_table)
1280                 unregister_sysctl_table(mq_sysctl_table);
1281         kmem_cache_destroy(mqueue_inode_cachep);
1282         return error;
1283 }
1284
1285 __initcall(init_mqueue_fs);