new helper: ihold()
[linux-2.6.git] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@gmail.com)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *                          Manfred Spraul          (manfred@colorfullife.com)
10  *
11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12  *
13  * This file is released under the GPL.
14  */
15
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32 #include <linux/nsproxy.h>
33 #include <linux/pid.h>
34 #include <linux/ipc_namespace.h>
35 #include <linux/slab.h>
36
37 #include <net/sock.h>
38 #include "util.h"
39
40 #define MQUEUE_MAGIC    0x19800202
41 #define DIRENT_SIZE     20
42 #define FILENT_SIZE     80
43
44 #define SEND            0
45 #define RECV            1
46
47 #define STATE_NONE      0
48 #define STATE_PENDING   1
49 #define STATE_READY     2
50
51 struct ext_wait_queue {         /* queue of sleeping tasks */
52         struct task_struct *task;
53         struct list_head list;
54         struct msg_msg *msg;    /* ptr of loaded message */
55         int state;              /* one of STATE_* values */
56 };
57
58 struct mqueue_inode_info {
59         spinlock_t lock;
60         struct inode vfs_inode;
61         wait_queue_head_t wait_q;
62
63         struct msg_msg **messages;
64         struct mq_attr attr;
65
66         struct sigevent notify;
67         struct pid* notify_owner;
68         struct user_struct *user;       /* user who created, for accounting */
69         struct sock *notify_sock;
70         struct sk_buff *notify_cookie;
71
72         /* for tasks waiting for free space and messages, respectively */
73         struct ext_wait_queue e_wait_q[2];
74
75         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
76 };
77
78 static const struct inode_operations mqueue_dir_inode_operations;
79 static const struct file_operations mqueue_file_operations;
80 static const struct super_operations mqueue_super_ops;
81 static void remove_notification(struct mqueue_inode_info *info);
82
83 static struct kmem_cache *mqueue_inode_cachep;
84
85 static struct ctl_table_header * mq_sysctl_table;
86
87 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
88 {
89         return container_of(inode, struct mqueue_inode_info, vfs_inode);
90 }
91
92 /*
93  * This routine should be called with the mq_lock held.
94  */
95 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
96 {
97         return get_ipc_ns(inode->i_sb->s_fs_info);
98 }
99
100 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
101 {
102         struct ipc_namespace *ns;
103
104         spin_lock(&mq_lock);
105         ns = __get_ns_from_inode(inode);
106         spin_unlock(&mq_lock);
107         return ns;
108 }
109
110 static struct inode *mqueue_get_inode(struct super_block *sb,
111                 struct ipc_namespace *ipc_ns, int mode,
112                 struct mq_attr *attr)
113 {
114         struct user_struct *u = current_user();
115         struct inode *inode;
116
117         inode = new_inode(sb);
118         if (inode) {
119                 inode->i_mode = mode;
120                 inode->i_uid = current_fsuid();
121                 inode->i_gid = current_fsgid();
122                 inode->i_mtime = inode->i_ctime = inode->i_atime =
123                                 CURRENT_TIME;
124
125                 if (S_ISREG(mode)) {
126                         struct mqueue_inode_info *info;
127                         struct task_struct *p = current;
128                         unsigned long mq_bytes, mq_msg_tblsz;
129
130                         inode->i_fop = &mqueue_file_operations;
131                         inode->i_size = FILENT_SIZE;
132                         /* mqueue specific info */
133                         info = MQUEUE_I(inode);
134                         spin_lock_init(&info->lock);
135                         init_waitqueue_head(&info->wait_q);
136                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
137                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
138                         info->notify_owner = NULL;
139                         info->qsize = 0;
140                         info->user = NULL;      /* set when all is ok */
141                         memset(&info->attr, 0, sizeof(info->attr));
142                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
143                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
144                         if (attr) {
145                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
146                                 info->attr.mq_msgsize = attr->mq_msgsize;
147                         }
148                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
149                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
150                         if (!info->messages)
151                                 goto out_inode;
152
153                         mq_bytes = (mq_msg_tblsz +
154                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
155
156                         spin_lock(&mq_lock);
157                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
158                             u->mq_bytes + mq_bytes >
159                             task_rlimit(p, RLIMIT_MSGQUEUE)) {
160                                 spin_unlock(&mq_lock);
161                                 /* mqueue_evict_inode() releases info->messages */
162                                 goto out_inode;
163                         }
164                         u->mq_bytes += mq_bytes;
165                         spin_unlock(&mq_lock);
166
167                         /* all is ok */
168                         info->user = get_uid(u);
169                 } else if (S_ISDIR(mode)) {
170                         inc_nlink(inode);
171                         /* Some things misbehave if size == 0 on a directory */
172                         inode->i_size = 2 * DIRENT_SIZE;
173                         inode->i_op = &mqueue_dir_inode_operations;
174                         inode->i_fop = &simple_dir_operations;
175                 }
176         }
177         return inode;
178 out_inode:
179         iput(inode);
180         return NULL;
181 }
182
183 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
184 {
185         struct inode *inode;
186         struct ipc_namespace *ns = data;
187         int error;
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, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
195                                 NULL);
196         if (!inode) {
197                 error = -ENOMEM;
198                 goto out;
199         }
200
201         sb->s_root = d_alloc_root(inode);
202         if (!sb->s_root) {
203                 iput(inode);
204                 error = -ENOMEM;
205                 goto out;
206         }
207         error = 0;
208
209 out:
210         return error;
211 }
212
213 static int mqueue_get_sb(struct file_system_type *fs_type,
214                          int flags, const char *dev_name,
215                          void *data, struct vfsmount *mnt)
216 {
217         if (!(flags & MS_KERNMOUNT))
218                 data = current->nsproxy->ipc_ns;
219         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
220 }
221
222 static void init_once(void *foo)
223 {
224         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
225
226         inode_init_once(&p->vfs_inode);
227 }
228
229 static struct inode *mqueue_alloc_inode(struct super_block *sb)
230 {
231         struct mqueue_inode_info *ei;
232
233         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
234         if (!ei)
235                 return NULL;
236         return &ei->vfs_inode;
237 }
238
239 static void mqueue_destroy_inode(struct inode *inode)
240 {
241         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
242 }
243
244 static void mqueue_evict_inode(struct inode *inode)
245 {
246         struct mqueue_inode_info *info;
247         struct user_struct *user;
248         unsigned long mq_bytes;
249         int i;
250         struct ipc_namespace *ipc_ns;
251
252         end_writeback(inode);
253
254         if (S_ISDIR(inode->i_mode))
255                 return;
256
257         ipc_ns = get_ns_from_inode(inode);
258         info = MQUEUE_I(inode);
259         spin_lock(&info->lock);
260         for (i = 0; i < info->attr.mq_curmsgs; i++)
261                 free_msg(info->messages[i]);
262         kfree(info->messages);
263         spin_unlock(&info->lock);
264
265         /* Total amount of bytes accounted for the mqueue */
266         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
267             + info->attr.mq_msgsize);
268         user = info->user;
269         if (user) {
270                 spin_lock(&mq_lock);
271                 user->mq_bytes -= mq_bytes;
272                 /*
273                  * get_ns_from_inode() ensures that the
274                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
275                  * to which we now hold a reference, or it is NULL.
276                  * We can't put it here under mq_lock, though.
277                  */
278                 if (ipc_ns)
279                         ipc_ns->mq_queues_count--;
280                 spin_unlock(&mq_lock);
281                 free_uid(user);
282         }
283         if (ipc_ns)
284                 put_ipc_ns(ipc_ns);
285 }
286
287 static int mqueue_create(struct inode *dir, struct dentry *dentry,
288                                 int mode, struct nameidata *nd)
289 {
290         struct inode *inode;
291         struct mq_attr *attr = dentry->d_fsdata;
292         int error;
293         struct ipc_namespace *ipc_ns;
294
295         spin_lock(&mq_lock);
296         ipc_ns = __get_ns_from_inode(dir);
297         if (!ipc_ns) {
298                 error = -EACCES;
299                 goto out_unlock;
300         }
301         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
302                         !capable(CAP_SYS_RESOURCE)) {
303                 error = -ENOSPC;
304                 goto out_unlock;
305         }
306         ipc_ns->mq_queues_count++;
307         spin_unlock(&mq_lock);
308
309         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
310         if (!inode) {
311                 error = -ENOMEM;
312                 spin_lock(&mq_lock);
313                 ipc_ns->mq_queues_count--;
314                 goto out_unlock;
315         }
316
317         put_ipc_ns(ipc_ns);
318         dir->i_size += DIRENT_SIZE;
319         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
320
321         d_instantiate(dentry, inode);
322         dget(dentry);
323         return 0;
324 out_unlock:
325         spin_unlock(&mq_lock);
326         if (ipc_ns)
327                 put_ipc_ns(ipc_ns);
328         return error;
329 }
330
331 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
332 {
333         struct inode *inode = dentry->d_inode;
334
335         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
336         dir->i_size -= DIRENT_SIZE;
337         drop_nlink(inode);
338         dput(dentry);
339         return 0;
340 }
341
342 /*
343 *       This is routine for system read from queue file.
344 *       To avoid mess with doing here some sort of mq_receive we allow
345 *       to read only queue size & notification info (the only values
346 *       that are interesting from user point of view and aren't accessible
347 *       through std routines)
348 */
349 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
350                                 size_t count, loff_t *off)
351 {
352         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
353         char buffer[FILENT_SIZE];
354         ssize_t ret;
355
356         spin_lock(&info->lock);
357         snprintf(buffer, sizeof(buffer),
358                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
359                         info->qsize,
360                         info->notify_owner ? info->notify.sigev_notify : 0,
361                         (info->notify_owner &&
362                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
363                                 info->notify.sigev_signo : 0,
364                         pid_vnr(info->notify_owner));
365         spin_unlock(&info->lock);
366         buffer[sizeof(buffer)-1] = '\0';
367
368         ret = simple_read_from_buffer(u_data, count, off, buffer,
369                                 strlen(buffer));
370         if (ret <= 0)
371                 return ret;
372
373         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
374         return ret;
375 }
376
377 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
378 {
379         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
380
381         spin_lock(&info->lock);
382         if (task_tgid(current) == info->notify_owner)
383                 remove_notification(info);
384
385         spin_unlock(&info->lock);
386         return 0;
387 }
388
389 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
390 {
391         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
392         int retval = 0;
393
394         poll_wait(filp, &info->wait_q, poll_tab);
395
396         spin_lock(&info->lock);
397         if (info->attr.mq_curmsgs)
398                 retval = POLLIN | POLLRDNORM;
399
400         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
401                 retval |= POLLOUT | POLLWRNORM;
402         spin_unlock(&info->lock);
403
404         return retval;
405 }
406
407 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
408 static void wq_add(struct mqueue_inode_info *info, int sr,
409                         struct ext_wait_queue *ewp)
410 {
411         struct ext_wait_queue *walk;
412
413         ewp->task = current;
414
415         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
416                 if (walk->task->static_prio <= current->static_prio) {
417                         list_add_tail(&ewp->list, &walk->list);
418                         return;
419                 }
420         }
421         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
422 }
423
424 /*
425  * Puts current task to sleep. Caller must hold queue lock. After return
426  * lock isn't held.
427  * sr: SEND or RECV
428  */
429 static int wq_sleep(struct mqueue_inode_info *info, int sr,
430                     ktime_t *timeout, struct ext_wait_queue *ewp)
431 {
432         int retval;
433         signed long time;
434
435         wq_add(info, sr, ewp);
436
437         for (;;) {
438                 set_current_state(TASK_INTERRUPTIBLE);
439
440                 spin_unlock(&info->lock);
441                 time = schedule_hrtimeout_range_clock(timeout,
442                     HRTIMER_MODE_ABS, 0, CLOCK_REALTIME);
443
444                 while (ewp->state == STATE_PENDING)
445                         cpu_relax();
446
447                 if (ewp->state == STATE_READY) {
448                         retval = 0;
449                         goto out;
450                 }
451                 spin_lock(&info->lock);
452                 if (ewp->state == STATE_READY) {
453                         retval = 0;
454                         goto out_unlock;
455                 }
456                 if (signal_pending(current)) {
457                         retval = -ERESTARTSYS;
458                         break;
459                 }
460                 if (time == 0) {
461                         retval = -ETIMEDOUT;
462                         break;
463                 }
464         }
465         list_del(&ewp->list);
466 out_unlock:
467         spin_unlock(&info->lock);
468 out:
469         return retval;
470 }
471
472 /*
473  * Returns waiting task that should be serviced first or NULL if none exists
474  */
475 static struct ext_wait_queue *wq_get_first_waiter(
476                 struct mqueue_inode_info *info, int sr)
477 {
478         struct list_head *ptr;
479
480         ptr = info->e_wait_q[sr].list.prev;
481         if (ptr == &info->e_wait_q[sr].list)
482                 return NULL;
483         return list_entry(ptr, struct ext_wait_queue, list);
484 }
485
486 /* Auxiliary functions to manipulate messages' list */
487 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
488 {
489         int k;
490
491         k = info->attr.mq_curmsgs - 1;
492         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
493                 info->messages[k + 1] = info->messages[k];
494                 k--;
495         }
496         info->attr.mq_curmsgs++;
497         info->qsize += ptr->m_ts;
498         info->messages[k + 1] = ptr;
499 }
500
501 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
502 {
503         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
504         return info->messages[info->attr.mq_curmsgs];
505 }
506
507 static inline void set_cookie(struct sk_buff *skb, char code)
508 {
509         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
510 }
511
512 /*
513  * The next function is only to split too long sys_mq_timedsend
514  */
515 static void __do_notify(struct mqueue_inode_info *info)
516 {
517         /* notification
518          * invoked when there is registered process and there isn't process
519          * waiting synchronously for message AND state of queue changed from
520          * empty to not empty. Here we are sure that no one is waiting
521          * synchronously. */
522         if (info->notify_owner &&
523             info->attr.mq_curmsgs == 1) {
524                 struct siginfo sig_i;
525                 switch (info->notify.sigev_notify) {
526                 case SIGEV_NONE:
527                         break;
528                 case SIGEV_SIGNAL:
529                         /* sends signal */
530
531                         sig_i.si_signo = info->notify.sigev_signo;
532                         sig_i.si_errno = 0;
533                         sig_i.si_code = SI_MESGQ;
534                         sig_i.si_value = info->notify.sigev_value;
535                         sig_i.si_pid = task_tgid_nr_ns(current,
536                                                 ns_of_pid(info->notify_owner));
537                         sig_i.si_uid = current_uid();
538
539                         kill_pid_info(info->notify.sigev_signo,
540                                       &sig_i, info->notify_owner);
541                         break;
542                 case SIGEV_THREAD:
543                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
544                         netlink_sendskb(info->notify_sock, info->notify_cookie);
545                         break;
546                 }
547                 /* after notification unregisters process */
548                 put_pid(info->notify_owner);
549                 info->notify_owner = NULL;
550         }
551         wake_up(&info->wait_q);
552 }
553
554 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
555                            ktime_t *expires, struct timespec *ts)
556 {
557         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
558                 return -EFAULT;
559         if (!timespec_valid(ts))
560                 return -EINVAL;
561
562         *expires = timespec_to_ktime(*ts);
563         return 0;
564 }
565
566 static void remove_notification(struct mqueue_inode_info *info)
567 {
568         if (info->notify_owner != NULL &&
569             info->notify.sigev_notify == SIGEV_THREAD) {
570                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
571                 netlink_sendskb(info->notify_sock, info->notify_cookie);
572         }
573         put_pid(info->notify_owner);
574         info->notify_owner = NULL;
575 }
576
577 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
578 {
579         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
580                 return 0;
581         if (capable(CAP_SYS_RESOURCE)) {
582                 if (attr->mq_maxmsg > HARD_MSGMAX)
583                         return 0;
584         } else {
585                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
586                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
587                         return 0;
588         }
589         /* check for overflow */
590         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
591                 return 0;
592         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
593             + sizeof (struct msg_msg *))) <
594             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
595                 return 0;
596         return 1;
597 }
598
599 /*
600  * Invoked when creating a new queue via sys_mq_open
601  */
602 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
603                         struct dentry *dentry, int oflag, mode_t mode,
604                         struct mq_attr *attr)
605 {
606         const struct cred *cred = current_cred();
607         struct file *result;
608         int ret;
609
610         if (attr) {
611                 if (!mq_attr_ok(ipc_ns, attr)) {
612                         ret = -EINVAL;
613                         goto out;
614                 }
615                 /* store for use during create */
616                 dentry->d_fsdata = attr;
617         }
618
619         mode &= ~current_umask();
620         ret = mnt_want_write(ipc_ns->mq_mnt);
621         if (ret)
622                 goto out;
623         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
624         dentry->d_fsdata = NULL;
625         if (ret)
626                 goto out_drop_write;
627
628         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
629         /*
630          * dentry_open() took a persistent mnt_want_write(),
631          * so we can now drop this one.
632          */
633         mnt_drop_write(ipc_ns->mq_mnt);
634         return result;
635
636 out_drop_write:
637         mnt_drop_write(ipc_ns->mq_mnt);
638 out:
639         dput(dentry);
640         mntput(ipc_ns->mq_mnt);
641         return ERR_PTR(ret);
642 }
643
644 /* Opens existing queue */
645 static struct file *do_open(struct ipc_namespace *ipc_ns,
646                                 struct dentry *dentry, int oflag)
647 {
648         int ret;
649         const struct cred *cred = current_cred();
650
651         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
652                                                   MAY_READ | MAY_WRITE };
653
654         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
655                 ret = -EINVAL;
656                 goto err;
657         }
658
659         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
660                 ret = -EACCES;
661                 goto err;
662         }
663
664         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
665
666 err:
667         dput(dentry);
668         mntput(ipc_ns->mq_mnt);
669         return ERR_PTR(ret);
670 }
671
672 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
673                 struct mq_attr __user *, u_attr)
674 {
675         struct dentry *dentry;
676         struct file *filp;
677         char *name;
678         struct mq_attr attr;
679         int fd, error;
680         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
681
682         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
683                 return -EFAULT;
684
685         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
686
687         if (IS_ERR(name = getname(u_name)))
688                 return PTR_ERR(name);
689
690         fd = get_unused_fd_flags(O_CLOEXEC);
691         if (fd < 0)
692                 goto out_putname;
693
694         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
695         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
696         if (IS_ERR(dentry)) {
697                 error = PTR_ERR(dentry);
698                 goto out_putfd;
699         }
700         mntget(ipc_ns->mq_mnt);
701
702         if (oflag & O_CREAT) {
703                 if (dentry->d_inode) {  /* entry already exists */
704                         audit_inode(name, dentry);
705                         if (oflag & O_EXCL) {
706                                 error = -EEXIST;
707                                 goto out;
708                         }
709                         filp = do_open(ipc_ns, dentry, oflag);
710                 } else {
711                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
712                                                 dentry, oflag, mode,
713                                                 u_attr ? &attr : NULL);
714                 }
715         } else {
716                 if (!dentry->d_inode) {
717                         error = -ENOENT;
718                         goto out;
719                 }
720                 audit_inode(name, dentry);
721                 filp = do_open(ipc_ns, dentry, oflag);
722         }
723
724         if (IS_ERR(filp)) {
725                 error = PTR_ERR(filp);
726                 goto out_putfd;
727         }
728
729         fd_install(fd, filp);
730         goto out_upsem;
731
732 out:
733         dput(dentry);
734         mntput(ipc_ns->mq_mnt);
735 out_putfd:
736         put_unused_fd(fd);
737         fd = error;
738 out_upsem:
739         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
740 out_putname:
741         putname(name);
742         return fd;
743 }
744
745 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
746 {
747         int err;
748         char *name;
749         struct dentry *dentry;
750         struct inode *inode = NULL;
751         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
752
753         name = getname(u_name);
754         if (IS_ERR(name))
755                 return PTR_ERR(name);
756
757         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
758                         I_MUTEX_PARENT);
759         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
760         if (IS_ERR(dentry)) {
761                 err = PTR_ERR(dentry);
762                 goto out_unlock;
763         }
764
765         if (!dentry->d_inode) {
766                 err = -ENOENT;
767                 goto out_err;
768         }
769
770         inode = dentry->d_inode;
771         if (inode)
772                 ihold(inode);
773         err = mnt_want_write(ipc_ns->mq_mnt);
774         if (err)
775                 goto out_err;
776         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
777         mnt_drop_write(ipc_ns->mq_mnt);
778 out_err:
779         dput(dentry);
780
781 out_unlock:
782         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
783         putname(name);
784         if (inode)
785                 iput(inode);
786
787         return err;
788 }
789
790 /* Pipelined send and receive functions.
791  *
792  * If a receiver finds no waiting message, then it registers itself in the
793  * list of waiting receivers. A sender checks that list before adding the new
794  * message into the message array. If there is a waiting receiver, then it
795  * bypasses the message array and directly hands the message over to the
796  * receiver.
797  * The receiver accepts the message and returns without grabbing the queue
798  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
799  * are necessary. The same algorithm is used for sysv semaphores, see
800  * ipc/sem.c for more details.
801  *
802  * The same algorithm is used for senders.
803  */
804
805 /* pipelined_send() - send a message directly to the task waiting in
806  * sys_mq_timedreceive() (without inserting message into a queue).
807  */
808 static inline void pipelined_send(struct mqueue_inode_info *info,
809                                   struct msg_msg *message,
810                                   struct ext_wait_queue *receiver)
811 {
812         receiver->msg = message;
813         list_del(&receiver->list);
814         receiver->state = STATE_PENDING;
815         wake_up_process(receiver->task);
816         smp_wmb();
817         receiver->state = STATE_READY;
818 }
819
820 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
821  * gets its message and put to the queue (we have one free place for sure). */
822 static inline void pipelined_receive(struct mqueue_inode_info *info)
823 {
824         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
825
826         if (!sender) {
827                 /* for poll */
828                 wake_up_interruptible(&info->wait_q);
829                 return;
830         }
831         msg_insert(sender->msg, info);
832         list_del(&sender->list);
833         sender->state = STATE_PENDING;
834         wake_up_process(sender->task);
835         smp_wmb();
836         sender->state = STATE_READY;
837 }
838
839 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
840                 size_t, msg_len, unsigned int, msg_prio,
841                 const struct timespec __user *, u_abs_timeout)
842 {
843         struct file *filp;
844         struct inode *inode;
845         struct ext_wait_queue wait;
846         struct ext_wait_queue *receiver;
847         struct msg_msg *msg_ptr;
848         struct mqueue_inode_info *info;
849         ktime_t expires, *timeout = NULL;
850         struct timespec ts;
851         int ret;
852
853         if (u_abs_timeout) {
854                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
855                 if (res)
856                         return res;
857                 timeout = &expires;
858         }
859
860         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
861                 return -EINVAL;
862
863         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
864
865         filp = fget(mqdes);
866         if (unlikely(!filp)) {
867                 ret = -EBADF;
868                 goto out;
869         }
870
871         inode = filp->f_path.dentry->d_inode;
872         if (unlikely(filp->f_op != &mqueue_file_operations)) {
873                 ret = -EBADF;
874                 goto out_fput;
875         }
876         info = MQUEUE_I(inode);
877         audit_inode(NULL, filp->f_path.dentry);
878
879         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
880                 ret = -EBADF;
881                 goto out_fput;
882         }
883
884         if (unlikely(msg_len > info->attr.mq_msgsize)) {
885                 ret = -EMSGSIZE;
886                 goto out_fput;
887         }
888
889         /* First try to allocate memory, before doing anything with
890          * existing queues. */
891         msg_ptr = load_msg(u_msg_ptr, msg_len);
892         if (IS_ERR(msg_ptr)) {
893                 ret = PTR_ERR(msg_ptr);
894                 goto out_fput;
895         }
896         msg_ptr->m_ts = msg_len;
897         msg_ptr->m_type = msg_prio;
898
899         spin_lock(&info->lock);
900
901         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
902                 if (filp->f_flags & O_NONBLOCK) {
903                         spin_unlock(&info->lock);
904                         ret = -EAGAIN;
905                 } else {
906                         wait.task = current;
907                         wait.msg = (void *) msg_ptr;
908                         wait.state = STATE_NONE;
909                         ret = wq_sleep(info, SEND, timeout, &wait);
910                 }
911                 if (ret < 0)
912                         free_msg(msg_ptr);
913         } else {
914                 receiver = wq_get_first_waiter(info, RECV);
915                 if (receiver) {
916                         pipelined_send(info, msg_ptr, receiver);
917                 } else {
918                         /* adds message to the queue */
919                         msg_insert(msg_ptr, info);
920                         __do_notify(info);
921                 }
922                 inode->i_atime = inode->i_mtime = inode->i_ctime =
923                                 CURRENT_TIME;
924                 spin_unlock(&info->lock);
925                 ret = 0;
926         }
927 out_fput:
928         fput(filp);
929 out:
930         return ret;
931 }
932
933 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
934                 size_t, msg_len, unsigned int __user *, u_msg_prio,
935                 const struct timespec __user *, u_abs_timeout)
936 {
937         ssize_t ret;
938         struct msg_msg *msg_ptr;
939         struct file *filp;
940         struct inode *inode;
941         struct mqueue_inode_info *info;
942         struct ext_wait_queue wait;
943         ktime_t expires, *timeout = NULL;
944         struct timespec ts;
945
946         if (u_abs_timeout) {
947                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
948                 if (res)
949                         return res;
950                 timeout = &expires;
951         }
952
953         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
954
955         filp = fget(mqdes);
956         if (unlikely(!filp)) {
957                 ret = -EBADF;
958                 goto out;
959         }
960
961         inode = filp->f_path.dentry->d_inode;
962         if (unlikely(filp->f_op != &mqueue_file_operations)) {
963                 ret = -EBADF;
964                 goto out_fput;
965         }
966         info = MQUEUE_I(inode);
967         audit_inode(NULL, filp->f_path.dentry);
968
969         if (unlikely(!(filp->f_mode & FMODE_READ))) {
970                 ret = -EBADF;
971                 goto out_fput;
972         }
973
974         /* checks if buffer is big enough */
975         if (unlikely(msg_len < info->attr.mq_msgsize)) {
976                 ret = -EMSGSIZE;
977                 goto out_fput;
978         }
979
980         spin_lock(&info->lock);
981         if (info->attr.mq_curmsgs == 0) {
982                 if (filp->f_flags & O_NONBLOCK) {
983                         spin_unlock(&info->lock);
984                         ret = -EAGAIN;
985                 } else {
986                         wait.task = current;
987                         wait.state = STATE_NONE;
988                         ret = wq_sleep(info, RECV, timeout, &wait);
989                         msg_ptr = wait.msg;
990                 }
991         } else {
992                 msg_ptr = msg_get(info);
993
994                 inode->i_atime = inode->i_mtime = inode->i_ctime =
995                                 CURRENT_TIME;
996
997                 /* There is now free space in queue. */
998                 pipelined_receive(info);
999                 spin_unlock(&info->lock);
1000                 ret = 0;
1001         }
1002         if (ret == 0) {
1003                 ret = msg_ptr->m_ts;
1004
1005                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1006                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1007                         ret = -EFAULT;
1008                 }
1009                 free_msg(msg_ptr);
1010         }
1011 out_fput:
1012         fput(filp);
1013 out:
1014         return ret;
1015 }
1016
1017 /*
1018  * Notes: the case when user wants us to deregister (with NULL as pointer)
1019  * and he isn't currently owner of notification, will be silently discarded.
1020  * It isn't explicitly defined in the POSIX.
1021  */
1022 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1023                 const struct sigevent __user *, u_notification)
1024 {
1025         int ret;
1026         struct file *filp;
1027         struct sock *sock;
1028         struct inode *inode;
1029         struct sigevent notification;
1030         struct mqueue_inode_info *info;
1031         struct sk_buff *nc;
1032
1033         if (u_notification) {
1034                 if (copy_from_user(&notification, u_notification,
1035                                         sizeof(struct sigevent)))
1036                         return -EFAULT;
1037         }
1038
1039         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1040
1041         nc = NULL;
1042         sock = NULL;
1043         if (u_notification != NULL) {
1044                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1045                              notification.sigev_notify != SIGEV_SIGNAL &&
1046                              notification.sigev_notify != SIGEV_THREAD))
1047                         return -EINVAL;
1048                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1049                         !valid_signal(notification.sigev_signo)) {
1050                         return -EINVAL;
1051                 }
1052                 if (notification.sigev_notify == SIGEV_THREAD) {
1053                         long timeo;
1054
1055                         /* create the notify skb */
1056                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1057                         if (!nc) {
1058                                 ret = -ENOMEM;
1059                                 goto out;
1060                         }
1061                         if (copy_from_user(nc->data,
1062                                         notification.sigev_value.sival_ptr,
1063                                         NOTIFY_COOKIE_LEN)) {
1064                                 ret = -EFAULT;
1065                                 goto out;
1066                         }
1067
1068                         /* TODO: add a header? */
1069                         skb_put(nc, NOTIFY_COOKIE_LEN);
1070                         /* and attach it to the socket */
1071 retry:
1072                         filp = fget(notification.sigev_signo);
1073                         if (!filp) {
1074                                 ret = -EBADF;
1075                                 goto out;
1076                         }
1077                         sock = netlink_getsockbyfilp(filp);
1078                         fput(filp);
1079                         if (IS_ERR(sock)) {
1080                                 ret = PTR_ERR(sock);
1081                                 sock = NULL;
1082                                 goto out;
1083                         }
1084
1085                         timeo = MAX_SCHEDULE_TIMEOUT;
1086                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1087                         if (ret == 1)
1088                                 goto retry;
1089                         if (ret) {
1090                                 sock = NULL;
1091                                 nc = NULL;
1092                                 goto out;
1093                         }
1094                 }
1095         }
1096
1097         filp = fget(mqdes);
1098         if (!filp) {
1099                 ret = -EBADF;
1100                 goto out;
1101         }
1102
1103         inode = filp->f_path.dentry->d_inode;
1104         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1105                 ret = -EBADF;
1106                 goto out_fput;
1107         }
1108         info = MQUEUE_I(inode);
1109
1110         ret = 0;
1111         spin_lock(&info->lock);
1112         if (u_notification == NULL) {
1113                 if (info->notify_owner == task_tgid(current)) {
1114                         remove_notification(info);
1115                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1116                 }
1117         } else if (info->notify_owner != NULL) {
1118                 ret = -EBUSY;
1119         } else {
1120                 switch (notification.sigev_notify) {
1121                 case SIGEV_NONE:
1122                         info->notify.sigev_notify = SIGEV_NONE;
1123                         break;
1124                 case SIGEV_THREAD:
1125                         info->notify_sock = sock;
1126                         info->notify_cookie = nc;
1127                         sock = NULL;
1128                         nc = NULL;
1129                         info->notify.sigev_notify = SIGEV_THREAD;
1130                         break;
1131                 case SIGEV_SIGNAL:
1132                         info->notify.sigev_signo = notification.sigev_signo;
1133                         info->notify.sigev_value = notification.sigev_value;
1134                         info->notify.sigev_notify = SIGEV_SIGNAL;
1135                         break;
1136                 }
1137
1138                 info->notify_owner = get_pid(task_tgid(current));
1139                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1140         }
1141         spin_unlock(&info->lock);
1142 out_fput:
1143         fput(filp);
1144 out:
1145         if (sock) {
1146                 netlink_detachskb(sock, nc);
1147         } else if (nc) {
1148                 dev_kfree_skb(nc);
1149         }
1150         return ret;
1151 }
1152
1153 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1154                 const struct mq_attr __user *, u_mqstat,
1155                 struct mq_attr __user *, u_omqstat)
1156 {
1157         int ret;
1158         struct mq_attr mqstat, omqstat;
1159         struct file *filp;
1160         struct inode *inode;
1161         struct mqueue_inode_info *info;
1162
1163         if (u_mqstat != NULL) {
1164                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1165                         return -EFAULT;
1166                 if (mqstat.mq_flags & (~O_NONBLOCK))
1167                         return -EINVAL;
1168         }
1169
1170         filp = fget(mqdes);
1171         if (!filp) {
1172                 ret = -EBADF;
1173                 goto out;
1174         }
1175
1176         inode = filp->f_path.dentry->d_inode;
1177         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1178                 ret = -EBADF;
1179                 goto out_fput;
1180         }
1181         info = MQUEUE_I(inode);
1182
1183         spin_lock(&info->lock);
1184
1185         omqstat = info->attr;
1186         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1187         if (u_mqstat) {
1188                 audit_mq_getsetattr(mqdes, &mqstat);
1189                 spin_lock(&filp->f_lock);
1190                 if (mqstat.mq_flags & O_NONBLOCK)
1191                         filp->f_flags |= O_NONBLOCK;
1192                 else
1193                         filp->f_flags &= ~O_NONBLOCK;
1194                 spin_unlock(&filp->f_lock);
1195
1196                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1197         }
1198
1199         spin_unlock(&info->lock);
1200
1201         ret = 0;
1202         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1203                                                 sizeof(struct mq_attr)))
1204                 ret = -EFAULT;
1205
1206 out_fput:
1207         fput(filp);
1208 out:
1209         return ret;
1210 }
1211
1212 static const struct inode_operations mqueue_dir_inode_operations = {
1213         .lookup = simple_lookup,
1214         .create = mqueue_create,
1215         .unlink = mqueue_unlink,
1216 };
1217
1218 static const struct file_operations mqueue_file_operations = {
1219         .flush = mqueue_flush_file,
1220         .poll = mqueue_poll_file,
1221         .read = mqueue_read_file,
1222         .llseek = default_llseek,
1223 };
1224
1225 static const struct super_operations mqueue_super_ops = {
1226         .alloc_inode = mqueue_alloc_inode,
1227         .destroy_inode = mqueue_destroy_inode,
1228         .evict_inode = mqueue_evict_inode,
1229         .statfs = simple_statfs,
1230 };
1231
1232 static struct file_system_type mqueue_fs_type = {
1233         .name = "mqueue",
1234         .get_sb = mqueue_get_sb,
1235         .kill_sb = kill_litter_super,
1236 };
1237
1238 int mq_init_ns(struct ipc_namespace *ns)
1239 {
1240         ns->mq_queues_count  = 0;
1241         ns->mq_queues_max    = DFLT_QUEUESMAX;
1242         ns->mq_msg_max       = DFLT_MSGMAX;
1243         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1244
1245         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1246         if (IS_ERR(ns->mq_mnt)) {
1247                 int err = PTR_ERR(ns->mq_mnt);
1248                 ns->mq_mnt = NULL;
1249                 return err;
1250         }
1251         return 0;
1252 }
1253
1254 void mq_clear_sbinfo(struct ipc_namespace *ns)
1255 {
1256         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1257 }
1258
1259 void mq_put_mnt(struct ipc_namespace *ns)
1260 {
1261         mntput(ns->mq_mnt);
1262 }
1263
1264 static int __init init_mqueue_fs(void)
1265 {
1266         int error;
1267
1268         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1269                                 sizeof(struct mqueue_inode_info), 0,
1270                                 SLAB_HWCACHE_ALIGN, init_once);
1271         if (mqueue_inode_cachep == NULL)
1272                 return -ENOMEM;
1273
1274         /* ignore failures - they are not fatal */
1275         mq_sysctl_table = mq_register_sysctl_table();
1276
1277         error = register_filesystem(&mqueue_fs_type);
1278         if (error)
1279                 goto out_sysctl;
1280
1281         spin_lock_init(&mq_lock);
1282
1283         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1284         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1285                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1286                 goto out_filesystem;
1287         }
1288
1289         return 0;
1290
1291 out_filesystem:
1292         unregister_filesystem(&mqueue_fs_type);
1293 out_sysctl:
1294         if (mq_sysctl_table)
1295                 unregister_sysctl_table(mq_sysctl_table);
1296         kmem_cache_destroy(mqueue_inode_cachep);
1297         return error;
1298 }
1299
1300 __initcall(init_mqueue_fs);