Merge branch 'for-2.6.31' of git://linux-nfs.org/~bfields/linux
[linux-2.6] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@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/ima.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 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->messages = NULL;
139                         info->notify_owner = NULL;
140                         info->qsize = 0;
141                         info->user = NULL;      /* set when all is ok */
142                         memset(&info->attr, 0, sizeof(info->attr));
143                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
144                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
145                         if (attr) {
146                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
147                                 info->attr.mq_msgsize = attr->mq_msgsize;
148                         }
149                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
150                         mq_bytes = (mq_msg_tblsz +
151                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
152
153                         spin_lock(&mq_lock);
154                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
155                             u->mq_bytes + mq_bytes >
156                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
157                                 spin_unlock(&mq_lock);
158                                 goto out_inode;
159                         }
160                         u->mq_bytes += mq_bytes;
161                         spin_unlock(&mq_lock);
162
163                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
164                         if (!info->messages) {
165                                 spin_lock(&mq_lock);
166                                 u->mq_bytes -= mq_bytes;
167                                 spin_unlock(&mq_lock);
168                                 goto out_inode;
169                         }
170                         /* all is ok */
171                         info->user = get_uid(u);
172                 } else if (S_ISDIR(mode)) {
173                         inc_nlink(inode);
174                         /* Some things misbehave if size == 0 on a directory */
175                         inode->i_size = 2 * DIRENT_SIZE;
176                         inode->i_op = &mqueue_dir_inode_operations;
177                         inode->i_fop = &simple_dir_operations;
178                 }
179         }
180         return inode;
181 out_inode:
182         make_bad_inode(inode);
183         iput(inode);
184         return NULL;
185 }
186
187 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
188 {
189         struct inode *inode;
190         struct ipc_namespace *ns = data;
191         int error = 0;
192
193         sb->s_blocksize = PAGE_CACHE_SIZE;
194         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
195         sb->s_magic = MQUEUE_MAGIC;
196         sb->s_op = &mqueue_super_ops;
197
198         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
199                                 NULL);
200         if (!inode) {
201                 error = -ENOMEM;
202                 goto out;
203         }
204
205         sb->s_root = d_alloc_root(inode);
206         if (!sb->s_root) {
207                 iput(inode);
208                 error = -ENOMEM;
209         }
210
211 out:
212         return error;
213 }
214
215 static int mqueue_get_sb(struct file_system_type *fs_type,
216                          int flags, const char *dev_name,
217                          void *data, struct vfsmount *mnt)
218 {
219         if (!(flags & MS_KERNMOUNT))
220                 data = current->nsproxy->ipc_ns;
221         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
222 }
223
224 static void init_once(void *foo)
225 {
226         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
227
228         inode_init_once(&p->vfs_inode);
229 }
230
231 static struct inode *mqueue_alloc_inode(struct super_block *sb)
232 {
233         struct mqueue_inode_info *ei;
234
235         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
236         if (!ei)
237                 return NULL;
238         return &ei->vfs_inode;
239 }
240
241 static void mqueue_destroy_inode(struct inode *inode)
242 {
243         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
244 }
245
246 static void mqueue_delete_inode(struct inode *inode)
247 {
248         struct mqueue_inode_info *info;
249         struct user_struct *user;
250         unsigned long mq_bytes;
251         int i;
252         struct ipc_namespace *ipc_ns;
253
254         if (S_ISDIR(inode->i_mode)) {
255                 clear_inode(inode);
256                 return;
257         }
258         ipc_ns = get_ns_from_inode(inode);
259         info = MQUEUE_I(inode);
260         spin_lock(&info->lock);
261         for (i = 0; i < info->attr.mq_curmsgs; i++)
262                 free_msg(info->messages[i]);
263         kfree(info->messages);
264         spin_unlock(&info->lock);
265
266         clear_inode(inode);
267
268         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
269                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
270         user = info->user;
271         if (user) {
272                 spin_lock(&mq_lock);
273                 user->mq_bytes -= mq_bytes;
274                 /*
275                  * get_ns_from_inode() ensures that the
276                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
277                  * to which we now hold a reference, or it is NULL.
278                  * We can't put it here under mq_lock, though.
279                  */
280                 if (ipc_ns)
281                         ipc_ns->mq_queues_count--;
282                 spin_unlock(&mq_lock);
283                 free_uid(user);
284         }
285         if (ipc_ns)
286                 put_ipc_ns(ipc_ns);
287 }
288
289 static int mqueue_create(struct inode *dir, struct dentry *dentry,
290                                 int mode, struct nameidata *nd)
291 {
292         struct inode *inode;
293         struct mq_attr *attr = dentry->d_fsdata;
294         int error;
295         struct ipc_namespace *ipc_ns;
296
297         spin_lock(&mq_lock);
298         ipc_ns = __get_ns_from_inode(dir);
299         if (!ipc_ns) {
300                 error = -EACCES;
301                 goto out_unlock;
302         }
303         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
304                         !capable(CAP_SYS_RESOURCE)) {
305                 error = -ENOSPC;
306                 goto out_unlock;
307         }
308         ipc_ns->mq_queues_count++;
309         spin_unlock(&mq_lock);
310
311         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
312         if (!inode) {
313                 error = -ENOMEM;
314                 spin_lock(&mq_lock);
315                 ipc_ns->mq_queues_count--;
316                 goto out_unlock;
317         }
318
319         put_ipc_ns(ipc_ns);
320         dir->i_size += DIRENT_SIZE;
321         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
322
323         d_instantiate(dentry, inode);
324         dget(dentry);
325         return 0;
326 out_unlock:
327         spin_unlock(&mq_lock);
328         if (ipc_ns)
329                 put_ipc_ns(ipc_ns);
330         return error;
331 }
332
333 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
334 {
335         struct inode *inode = dentry->d_inode;
336
337         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
338         dir->i_size -= DIRENT_SIZE;
339         drop_nlink(inode);
340         dput(dentry);
341         return 0;
342 }
343
344 /*
345 *       This is routine for system read from queue file.
346 *       To avoid mess with doing here some sort of mq_receive we allow
347 *       to read only queue size & notification info (the only values
348 *       that are interesting from user point of view and aren't accessible
349 *       through std routines)
350 */
351 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
352                                 size_t count, loff_t *off)
353 {
354         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
355         char buffer[FILENT_SIZE];
356         ssize_t ret;
357
358         spin_lock(&info->lock);
359         snprintf(buffer, sizeof(buffer),
360                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
361                         info->qsize,
362                         info->notify_owner ? info->notify.sigev_notify : 0,
363                         (info->notify_owner &&
364                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
365                                 info->notify.sigev_signo : 0,
366                         pid_vnr(info->notify_owner));
367         spin_unlock(&info->lock);
368         buffer[sizeof(buffer)-1] = '\0';
369
370         ret = simple_read_from_buffer(u_data, count, off, buffer,
371                                 strlen(buffer));
372         if (ret <= 0)
373                 return ret;
374
375         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
376         return ret;
377 }
378
379 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
380 {
381         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
382
383         spin_lock(&info->lock);
384         if (task_tgid(current) == info->notify_owner)
385                 remove_notification(info);
386
387         spin_unlock(&info->lock);
388         return 0;
389 }
390
391 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
392 {
393         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
394         int retval = 0;
395
396         poll_wait(filp, &info->wait_q, poll_tab);
397
398         spin_lock(&info->lock);
399         if (info->attr.mq_curmsgs)
400                 retval = POLLIN | POLLRDNORM;
401
402         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
403                 retval |= POLLOUT | POLLWRNORM;
404         spin_unlock(&info->lock);
405
406         return retval;
407 }
408
409 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
410 static void wq_add(struct mqueue_inode_info *info, int sr,
411                         struct ext_wait_queue *ewp)
412 {
413         struct ext_wait_queue *walk;
414
415         ewp->task = current;
416
417         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
418                 if (walk->task->static_prio <= current->static_prio) {
419                         list_add_tail(&ewp->list, &walk->list);
420                         return;
421                 }
422         }
423         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
424 }
425
426 /*
427  * Puts current task to sleep. Caller must hold queue lock. After return
428  * lock isn't held.
429  * sr: SEND or RECV
430  */
431 static int wq_sleep(struct mqueue_inode_info *info, int sr,
432                         long timeout, struct ext_wait_queue *ewp)
433 {
434         int retval;
435         signed long time;
436
437         wq_add(info, sr, ewp);
438
439         for (;;) {
440                 set_current_state(TASK_INTERRUPTIBLE);
441
442                 spin_unlock(&info->lock);
443                 time = schedule_timeout(timeout);
444
445                 while (ewp->state == STATE_PENDING)
446                         cpu_relax();
447
448                 if (ewp->state == STATE_READY) {
449                         retval = 0;
450                         goto out;
451                 }
452                 spin_lock(&info->lock);
453                 if (ewp->state == STATE_READY) {
454                         retval = 0;
455                         goto out_unlock;
456                 }
457                 if (signal_pending(current)) {
458                         retval = -ERESTARTSYS;
459                         break;
460                 }
461                 if (time == 0) {
462                         retval = -ETIMEDOUT;
463                         break;
464                 }
465         }
466         list_del(&ewp->list);
467 out_unlock:
468         spin_unlock(&info->lock);
469 out:
470         return retval;
471 }
472
473 /*
474  * Returns waiting task that should be serviced first or NULL if none exists
475  */
476 static struct ext_wait_queue *wq_get_first_waiter(
477                 struct mqueue_inode_info *info, int sr)
478 {
479         struct list_head *ptr;
480
481         ptr = info->e_wait_q[sr].list.prev;
482         if (ptr == &info->e_wait_q[sr].list)
483                 return NULL;
484         return list_entry(ptr, struct ext_wait_queue, list);
485 }
486
487 /* Auxiliary functions to manipulate messages' list */
488 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
489 {
490         int k;
491
492         k = info->attr.mq_curmsgs - 1;
493         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
494                 info->messages[k + 1] = info->messages[k];
495                 k--;
496         }
497         info->attr.mq_curmsgs++;
498         info->qsize += ptr->m_ts;
499         info->messages[k + 1] = ptr;
500 }
501
502 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
503 {
504         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
505         return info->messages[info->attr.mq_curmsgs];
506 }
507
508 static inline void set_cookie(struct sk_buff *skb, char code)
509 {
510         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
511 }
512
513 /*
514  * The next function is only to split too long sys_mq_timedsend
515  */
516 static void __do_notify(struct mqueue_inode_info *info)
517 {
518         /* notification
519          * invoked when there is registered process and there isn't process
520          * waiting synchronously for message AND state of queue changed from
521          * empty to not empty. Here we are sure that no one is waiting
522          * synchronously. */
523         if (info->notify_owner &&
524             info->attr.mq_curmsgs == 1) {
525                 struct siginfo sig_i;
526                 switch (info->notify.sigev_notify) {
527                 case SIGEV_NONE:
528                         break;
529                 case SIGEV_SIGNAL:
530                         /* sends signal */
531
532                         sig_i.si_signo = info->notify.sigev_signo;
533                         sig_i.si_errno = 0;
534                         sig_i.si_code = SI_MESGQ;
535                         sig_i.si_value = info->notify.sigev_value;
536                         sig_i.si_pid = task_tgid_nr_ns(current,
537                                                 ns_of_pid(info->notify_owner));
538                         sig_i.si_uid = current_uid();
539
540                         kill_pid_info(info->notify.sigev_signo,
541                                       &sig_i, info->notify_owner);
542                         break;
543                 case SIGEV_THREAD:
544                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
545                         netlink_sendskb(info->notify_sock, info->notify_cookie);
546                         break;
547                 }
548                 /* after notification unregisters process */
549                 put_pid(info->notify_owner);
550                 info->notify_owner = NULL;
551         }
552         wake_up(&info->wait_q);
553 }
554
555 static long prepare_timeout(struct timespec *p)
556 {
557         struct timespec nowts;
558         long timeout;
559
560         if (p) {
561                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
562                         || p->tv_nsec >= NSEC_PER_SEC))
563                         return -EINVAL;
564                 nowts = CURRENT_TIME;
565                 /* first subtract as jiffies can't be too big */
566                 p->tv_sec -= nowts.tv_sec;
567                 if (p->tv_nsec < nowts.tv_nsec) {
568                         p->tv_nsec += NSEC_PER_SEC;
569                         p->tv_sec--;
570                 }
571                 p->tv_nsec -= nowts.tv_nsec;
572                 if (p->tv_sec < 0)
573                         return 0;
574
575                 timeout = timespec_to_jiffies(p) + 1;
576         } else
577                 return MAX_SCHEDULE_TIMEOUT;
578
579         return timeout;
580 }
581
582 static void remove_notification(struct mqueue_inode_info *info)
583 {
584         if (info->notify_owner != NULL &&
585             info->notify.sigev_notify == SIGEV_THREAD) {
586                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
587                 netlink_sendskb(info->notify_sock, info->notify_cookie);
588         }
589         put_pid(info->notify_owner);
590         info->notify_owner = NULL;
591 }
592
593 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
594 {
595         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
596                 return 0;
597         if (capable(CAP_SYS_RESOURCE)) {
598                 if (attr->mq_maxmsg > HARD_MSGMAX)
599                         return 0;
600         } else {
601                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
602                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
603                         return 0;
604         }
605         /* check for overflow */
606         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
607                 return 0;
608         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
609             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
610             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
611                 return 0;
612         return 1;
613 }
614
615 /*
616  * Invoked when creating a new queue via sys_mq_open
617  */
618 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
619                         struct dentry *dentry, int oflag, mode_t mode,
620                         struct mq_attr *attr)
621 {
622         const struct cred *cred = current_cred();
623         struct file *result;
624         int ret;
625
626         if (attr) {
627                 ret = -EINVAL;
628                 if (!mq_attr_ok(ipc_ns, attr))
629                         goto out;
630                 /* store for use during create */
631                 dentry->d_fsdata = attr;
632         }
633
634         mode &= ~current_umask();
635         ret = mnt_want_write(ipc_ns->mq_mnt);
636         if (ret)
637                 goto out;
638         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
639         dentry->d_fsdata = NULL;
640         if (ret)
641                 goto out_drop_write;
642
643         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
644         /*
645          * dentry_open() took a persistent mnt_want_write(),
646          * so we can now drop this one.
647          */
648         mnt_drop_write(ipc_ns->mq_mnt);
649         return result;
650
651 out_drop_write:
652         mnt_drop_write(ipc_ns->mq_mnt);
653 out:
654         dput(dentry);
655         mntput(ipc_ns->mq_mnt);
656         return ERR_PTR(ret);
657 }
658
659 /* Opens existing queue */
660 static struct file *do_open(struct ipc_namespace *ipc_ns,
661                                 struct dentry *dentry, int oflag)
662 {
663         const struct cred *cred = current_cred();
664
665         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
666                                                   MAY_READ | MAY_WRITE };
667
668         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
669                 dput(dentry);
670                 mntput(ipc_ns->mq_mnt);
671                 return ERR_PTR(-EINVAL);
672         }
673
674         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
675                 dput(dentry);
676                 mntput(ipc_ns->mq_mnt);
677                 return ERR_PTR(-EACCES);
678         }
679
680         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
681 }
682
683 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
684                 struct mq_attr __user *, u_attr)
685 {
686         struct dentry *dentry;
687         struct file *filp;
688         char *name;
689         struct mq_attr attr;
690         int fd, error;
691         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
692
693         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
694                 return -EFAULT;
695
696         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
697
698         if (IS_ERR(name = getname(u_name)))
699                 return PTR_ERR(name);
700
701         fd = get_unused_fd_flags(O_CLOEXEC);
702         if (fd < 0)
703                 goto out_putname;
704
705         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
706         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
707         if (IS_ERR(dentry)) {
708                 error = PTR_ERR(dentry);
709                 goto out_err;
710         }
711         mntget(ipc_ns->mq_mnt);
712
713         if (oflag & O_CREAT) {
714                 if (dentry->d_inode) {  /* entry already exists */
715                         audit_inode(name, dentry);
716                         error = -EEXIST;
717                         if (oflag & O_EXCL)
718                                 goto out;
719                         filp = do_open(ipc_ns, dentry, oflag);
720                 } else {
721                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
722                                                 dentry, oflag, mode,
723                                                 u_attr ? &attr : NULL);
724                 }
725         } else {
726                 error = -ENOENT;
727                 if (!dentry->d_inode)
728                         goto out;
729                 audit_inode(name, dentry);
730                 filp = do_open(ipc_ns, dentry, oflag);
731         }
732
733         if (IS_ERR(filp)) {
734                 error = PTR_ERR(filp);
735                 goto out_putfd;
736         }
737         ima_counts_get(filp);
738
739         fd_install(fd, filp);
740         goto out_upsem;
741
742 out:
743         dput(dentry);
744         mntput(ipc_ns->mq_mnt);
745 out_putfd:
746         put_unused_fd(fd);
747 out_err:
748         fd = error;
749 out_upsem:
750         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
751 out_putname:
752         putname(name);
753         return fd;
754 }
755
756 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
757 {
758         int err;
759         char *name;
760         struct dentry *dentry;
761         struct inode *inode = NULL;
762         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
763
764         name = getname(u_name);
765         if (IS_ERR(name))
766                 return PTR_ERR(name);
767
768         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
769                         I_MUTEX_PARENT);
770         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
771         if (IS_ERR(dentry)) {
772                 err = PTR_ERR(dentry);
773                 goto out_unlock;
774         }
775
776         if (!dentry->d_inode) {
777                 err = -ENOENT;
778                 goto out_err;
779         }
780
781         inode = dentry->d_inode;
782         if (inode)
783                 atomic_inc(&inode->i_count);
784         err = mnt_want_write(ipc_ns->mq_mnt);
785         if (err)
786                 goto out_err;
787         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
788         mnt_drop_write(ipc_ns->mq_mnt);
789 out_err:
790         dput(dentry);
791
792 out_unlock:
793         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
794         putname(name);
795         if (inode)
796                 iput(inode);
797
798         return err;
799 }
800
801 /* Pipelined send and receive functions.
802  *
803  * If a receiver finds no waiting message, then it registers itself in the
804  * list of waiting receivers. A sender checks that list before adding the new
805  * message into the message array. If there is a waiting receiver, then it
806  * bypasses the message array and directly hands the message over to the
807  * receiver.
808  * The receiver accepts the message and returns without grabbing the queue
809  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
810  * are necessary. The same algorithm is used for sysv semaphores, see
811  * ipc/sem.c for more details.
812  *
813  * The same algorithm is used for senders.
814  */
815
816 /* pipelined_send() - send a message directly to the task waiting in
817  * sys_mq_timedreceive() (without inserting message into a queue).
818  */
819 static inline void pipelined_send(struct mqueue_inode_info *info,
820                                   struct msg_msg *message,
821                                   struct ext_wait_queue *receiver)
822 {
823         receiver->msg = message;
824         list_del(&receiver->list);
825         receiver->state = STATE_PENDING;
826         wake_up_process(receiver->task);
827         smp_wmb();
828         receiver->state = STATE_READY;
829 }
830
831 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
832  * gets its message and put to the queue (we have one free place for sure). */
833 static inline void pipelined_receive(struct mqueue_inode_info *info)
834 {
835         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
836
837         if (!sender) {
838                 /* for poll */
839                 wake_up_interruptible(&info->wait_q);
840                 return;
841         }
842         msg_insert(sender->msg, info);
843         list_del(&sender->list);
844         sender->state = STATE_PENDING;
845         wake_up_process(sender->task);
846         smp_wmb();
847         sender->state = STATE_READY;
848 }
849
850 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
851                 size_t, msg_len, unsigned int, msg_prio,
852                 const struct timespec __user *, u_abs_timeout)
853 {
854         struct file *filp;
855         struct inode *inode;
856         struct ext_wait_queue wait;
857         struct ext_wait_queue *receiver;
858         struct msg_msg *msg_ptr;
859         struct mqueue_inode_info *info;
860         struct timespec ts, *p = NULL;
861         long timeout;
862         int ret;
863
864         if (u_abs_timeout) {
865                 if (copy_from_user(&ts, u_abs_timeout, 
866                                         sizeof(struct timespec)))
867                         return -EFAULT;
868                 p = &ts;
869         }
870
871         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
872                 return -EINVAL;
873
874         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
875         timeout = prepare_timeout(p);
876
877         ret = -EBADF;
878         filp = fget(mqdes);
879         if (unlikely(!filp))
880                 goto out;
881
882         inode = filp->f_path.dentry->d_inode;
883         if (unlikely(filp->f_op != &mqueue_file_operations))
884                 goto out_fput;
885         info = MQUEUE_I(inode);
886         audit_inode(NULL, filp->f_path.dentry);
887
888         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
889                 goto out_fput;
890
891         if (unlikely(msg_len > info->attr.mq_msgsize)) {
892                 ret = -EMSGSIZE;
893                 goto out_fput;
894         }
895
896         /* First try to allocate memory, before doing anything with
897          * existing queues. */
898         msg_ptr = load_msg(u_msg_ptr, msg_len);
899         if (IS_ERR(msg_ptr)) {
900                 ret = PTR_ERR(msg_ptr);
901                 goto out_fput;
902         }
903         msg_ptr->m_ts = msg_len;
904         msg_ptr->m_type = msg_prio;
905
906         spin_lock(&info->lock);
907
908         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
909                 if (filp->f_flags & O_NONBLOCK) {
910                         spin_unlock(&info->lock);
911                         ret = -EAGAIN;
912                 } else if (unlikely(timeout < 0)) {
913                         spin_unlock(&info->lock);
914                         ret = timeout;
915                 } else {
916                         wait.task = current;
917                         wait.msg = (void *) msg_ptr;
918                         wait.state = STATE_NONE;
919                         ret = wq_sleep(info, SEND, timeout, &wait);
920                 }
921                 if (ret < 0)
922                         free_msg(msg_ptr);
923         } else {
924                 receiver = wq_get_first_waiter(info, RECV);
925                 if (receiver) {
926                         pipelined_send(info, msg_ptr, receiver);
927                 } else {
928                         /* adds message to the queue */
929                         msg_insert(msg_ptr, info);
930                         __do_notify(info);
931                 }
932                 inode->i_atime = inode->i_mtime = inode->i_ctime =
933                                 CURRENT_TIME;
934                 spin_unlock(&info->lock);
935                 ret = 0;
936         }
937 out_fput:
938         fput(filp);
939 out:
940         return ret;
941 }
942
943 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
944                 size_t, msg_len, unsigned int __user *, u_msg_prio,
945                 const struct timespec __user *, u_abs_timeout)
946 {
947         long timeout;
948         ssize_t ret;
949         struct msg_msg *msg_ptr;
950         struct file *filp;
951         struct inode *inode;
952         struct mqueue_inode_info *info;
953         struct ext_wait_queue wait;
954         struct timespec ts, *p = NULL;
955
956         if (u_abs_timeout) {
957                 if (copy_from_user(&ts, u_abs_timeout, 
958                                         sizeof(struct timespec)))
959                         return -EFAULT;
960                 p = &ts;
961         }
962
963         audit_mq_sendrecv(mqdes, msg_len, 0, p);
964         timeout = prepare_timeout(p);
965
966         ret = -EBADF;
967         filp = fget(mqdes);
968         if (unlikely(!filp))
969                 goto out;
970
971         inode = filp->f_path.dentry->d_inode;
972         if (unlikely(filp->f_op != &mqueue_file_operations))
973                 goto out_fput;
974         info = MQUEUE_I(inode);
975         audit_inode(NULL, filp->f_path.dentry);
976
977         if (unlikely(!(filp->f_mode & FMODE_READ)))
978                 goto out_fput;
979
980         /* checks if buffer is big enough */
981         if (unlikely(msg_len < info->attr.mq_msgsize)) {
982                 ret = -EMSGSIZE;
983                 goto out_fput;
984         }
985
986         spin_lock(&info->lock);
987         if (info->attr.mq_curmsgs == 0) {
988                 if (filp->f_flags & O_NONBLOCK) {
989                         spin_unlock(&info->lock);
990                         ret = -EAGAIN;
991                         msg_ptr = NULL;
992                 } else if (unlikely(timeout < 0)) {
993                         spin_unlock(&info->lock);
994                         ret = timeout;
995                         msg_ptr = NULL;
996                 } else {
997                         wait.task = current;
998                         wait.state = STATE_NONE;
999                         ret = wq_sleep(info, RECV, timeout, &wait);
1000                         msg_ptr = wait.msg;
1001                 }
1002         } else {
1003                 msg_ptr = msg_get(info);
1004
1005                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1006                                 CURRENT_TIME;
1007
1008                 /* There is now free space in queue. */
1009                 pipelined_receive(info);
1010                 spin_unlock(&info->lock);
1011                 ret = 0;
1012         }
1013         if (ret == 0) {
1014                 ret = msg_ptr->m_ts;
1015
1016                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1017                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1018                         ret = -EFAULT;
1019                 }
1020                 free_msg(msg_ptr);
1021         }
1022 out_fput:
1023         fput(filp);
1024 out:
1025         return ret;
1026 }
1027
1028 /*
1029  * Notes: the case when user wants us to deregister (with NULL as pointer)
1030  * and he isn't currently owner of notification, will be silently discarded.
1031  * It isn't explicitly defined in the POSIX.
1032  */
1033 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1034                 const struct sigevent __user *, u_notification)
1035 {
1036         int ret;
1037         struct file *filp;
1038         struct sock *sock;
1039         struct inode *inode;
1040         struct sigevent notification;
1041         struct mqueue_inode_info *info;
1042         struct sk_buff *nc;
1043
1044         if (u_notification) {
1045                 if (copy_from_user(&notification, u_notification,
1046                                         sizeof(struct sigevent)))
1047                         return -EFAULT;
1048         }
1049
1050         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1051
1052         nc = NULL;
1053         sock = NULL;
1054         if (u_notification != NULL) {
1055                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1056                              notification.sigev_notify != SIGEV_SIGNAL &&
1057                              notification.sigev_notify != SIGEV_THREAD))
1058                         return -EINVAL;
1059                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1060                         !valid_signal(notification.sigev_signo)) {
1061                         return -EINVAL;
1062                 }
1063                 if (notification.sigev_notify == SIGEV_THREAD) {
1064                         long timeo;
1065
1066                         /* create the notify skb */
1067                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1068                         ret = -ENOMEM;
1069                         if (!nc)
1070                                 goto out;
1071                         ret = -EFAULT;
1072                         if (copy_from_user(nc->data,
1073                                         notification.sigev_value.sival_ptr,
1074                                         NOTIFY_COOKIE_LEN)) {
1075                                 goto out;
1076                         }
1077
1078                         /* TODO: add a header? */
1079                         skb_put(nc, NOTIFY_COOKIE_LEN);
1080                         /* and attach it to the socket */
1081 retry:
1082                         filp = fget(notification.sigev_signo);
1083                         ret = -EBADF;
1084                         if (!filp)
1085                                 goto out;
1086                         sock = netlink_getsockbyfilp(filp);
1087                         fput(filp);
1088                         if (IS_ERR(sock)) {
1089                                 ret = PTR_ERR(sock);
1090                                 sock = NULL;
1091                                 goto out;
1092                         }
1093
1094                         timeo = MAX_SCHEDULE_TIMEOUT;
1095                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1096                         if (ret == 1)
1097                                 goto retry;
1098                         if (ret) {
1099                                 sock = NULL;
1100                                 nc = NULL;
1101                                 goto out;
1102                         }
1103                 }
1104         }
1105
1106         ret = -EBADF;
1107         filp = fget(mqdes);
1108         if (!filp)
1109                 goto out;
1110
1111         inode = filp->f_path.dentry->d_inode;
1112         if (unlikely(filp->f_op != &mqueue_file_operations))
1113                 goto out_fput;
1114         info = MQUEUE_I(inode);
1115
1116         ret = 0;
1117         spin_lock(&info->lock);
1118         if (u_notification == NULL) {
1119                 if (info->notify_owner == task_tgid(current)) {
1120                         remove_notification(info);
1121                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1122                 }
1123         } else if (info->notify_owner != NULL) {
1124                 ret = -EBUSY;
1125         } else {
1126                 switch (notification.sigev_notify) {
1127                 case SIGEV_NONE:
1128                         info->notify.sigev_notify = SIGEV_NONE;
1129                         break;
1130                 case SIGEV_THREAD:
1131                         info->notify_sock = sock;
1132                         info->notify_cookie = nc;
1133                         sock = NULL;
1134                         nc = NULL;
1135                         info->notify.sigev_notify = SIGEV_THREAD;
1136                         break;
1137                 case SIGEV_SIGNAL:
1138                         info->notify.sigev_signo = notification.sigev_signo;
1139                         info->notify.sigev_value = notification.sigev_value;
1140                         info->notify.sigev_notify = SIGEV_SIGNAL;
1141                         break;
1142                 }
1143
1144                 info->notify_owner = get_pid(task_tgid(current));
1145                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1146         }
1147         spin_unlock(&info->lock);
1148 out_fput:
1149         fput(filp);
1150 out:
1151         if (sock) {
1152                 netlink_detachskb(sock, nc);
1153         } else if (nc) {
1154                 dev_kfree_skb(nc);
1155         }
1156         return ret;
1157 }
1158
1159 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1160                 const struct mq_attr __user *, u_mqstat,
1161                 struct mq_attr __user *, u_omqstat)
1162 {
1163         int ret;
1164         struct mq_attr mqstat, omqstat;
1165         struct file *filp;
1166         struct inode *inode;
1167         struct mqueue_inode_info *info;
1168
1169         if (u_mqstat != NULL) {
1170                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1171                         return -EFAULT;
1172                 if (mqstat.mq_flags & (~O_NONBLOCK))
1173                         return -EINVAL;
1174         }
1175
1176         ret = -EBADF;
1177         filp = fget(mqdes);
1178         if (!filp)
1179                 goto out;
1180
1181         inode = filp->f_path.dentry->d_inode;
1182         if (unlikely(filp->f_op != &mqueue_file_operations))
1183                 goto out_fput;
1184         info = MQUEUE_I(inode);
1185
1186         spin_lock(&info->lock);
1187
1188         omqstat = info->attr;
1189         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1190         if (u_mqstat) {
1191                 audit_mq_getsetattr(mqdes, &mqstat);
1192                 spin_lock(&filp->f_lock);
1193                 if (mqstat.mq_flags & O_NONBLOCK)
1194                         filp->f_flags |= O_NONBLOCK;
1195                 else
1196                         filp->f_flags &= ~O_NONBLOCK;
1197                 spin_unlock(&filp->f_lock);
1198
1199                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1200         }
1201
1202         spin_unlock(&info->lock);
1203
1204         ret = 0;
1205         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1206                                                 sizeof(struct mq_attr)))
1207                 ret = -EFAULT;
1208
1209 out_fput:
1210         fput(filp);
1211 out:
1212         return ret;
1213 }
1214
1215 static const struct inode_operations mqueue_dir_inode_operations = {
1216         .lookup = simple_lookup,
1217         .create = mqueue_create,
1218         .unlink = mqueue_unlink,
1219 };
1220
1221 static const struct file_operations mqueue_file_operations = {
1222         .flush = mqueue_flush_file,
1223         .poll = mqueue_poll_file,
1224         .read = mqueue_read_file,
1225 };
1226
1227 static struct super_operations mqueue_super_ops = {
1228         .alloc_inode = mqueue_alloc_inode,
1229         .destroy_inode = mqueue_destroy_inode,
1230         .statfs = simple_statfs,
1231         .delete_inode = mqueue_delete_inode,
1232         .drop_inode = generic_delete_inode,
1233 };
1234
1235 static struct file_system_type mqueue_fs_type = {
1236         .name = "mqueue",
1237         .get_sb = mqueue_get_sb,
1238         .kill_sb = kill_litter_super,
1239 };
1240
1241 int mq_init_ns(struct ipc_namespace *ns)
1242 {
1243         ns->mq_queues_count  = 0;
1244         ns->mq_queues_max    = DFLT_QUEUESMAX;
1245         ns->mq_msg_max       = DFLT_MSGMAX;
1246         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1247
1248         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1249         if (IS_ERR(ns->mq_mnt)) {
1250                 int err = PTR_ERR(ns->mq_mnt);
1251                 ns->mq_mnt = NULL;
1252                 return err;
1253         }
1254         return 0;
1255 }
1256
1257 void mq_clear_sbinfo(struct ipc_namespace *ns)
1258 {
1259         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1260 }
1261
1262 void mq_put_mnt(struct ipc_namespace *ns)
1263 {
1264         mntput(ns->mq_mnt);
1265 }
1266
1267 static int __init init_mqueue_fs(void)
1268 {
1269         int error;
1270
1271         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1272                                 sizeof(struct mqueue_inode_info), 0,
1273                                 SLAB_HWCACHE_ALIGN, init_once);
1274         if (mqueue_inode_cachep == NULL)
1275                 return -ENOMEM;
1276
1277         /* ignore failues - they are not fatal */
1278         mq_sysctl_table = mq_register_sysctl_table();
1279
1280         error = register_filesystem(&mqueue_fs_type);
1281         if (error)
1282                 goto out_sysctl;
1283
1284         spin_lock_init(&mq_lock);
1285
1286         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1287         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1288                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1289                 goto out_filesystem;
1290         }
1291
1292         return 0;
1293
1294 out_filesystem:
1295         unregister_filesystem(&mqueue_fs_type);
1296 out_sysctl:
1297         if (mq_sysctl_table)
1298                 unregister_sysctl_table(mq_sysctl_table);
1299         kmem_cache_destroy(mqueue_inode_cachep);
1300         return error;
1301 }
1302
1303 __initcall(init_mqueue_fs);