86ee272de210bbae8236d45a4cebda3021afe2e9
[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/user_namespace.h>
36 #include <linux/slab.h>
37
38 #include <net/sock.h>
39 #include "util.h"
40
41 #define MQUEUE_MAGIC    0x19800202
42 #define DIRENT_SIZE     20
43 #define FILENT_SIZE     80
44
45 #define SEND            0
46 #define RECV            1
47
48 #define STATE_NONE      0
49 #define STATE_PENDING   1
50 #define STATE_READY     2
51
52 struct ext_wait_queue {         /* queue of sleeping tasks */
53         struct task_struct *task;
54         struct list_head list;
55         struct msg_msg *msg;    /* ptr of loaded message */
56         int state;              /* one of STATE_* values */
57 };
58
59 struct mqueue_inode_info {
60         spinlock_t lock;
61         struct inode vfs_inode;
62         wait_queue_head_t wait_q;
63
64         struct msg_msg **messages;
65         struct mq_attr attr;
66
67         struct sigevent notify;
68         struct pid* notify_owner;
69         struct user_struct *user;       /* user who created, for accounting */
70         struct sock *notify_sock;
71         struct sk_buff *notify_cookie;
72
73         /* for tasks waiting for free space and messages, respectively */
74         struct ext_wait_queue e_wait_q[2];
75
76         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
77 };
78
79 static const struct inode_operations mqueue_dir_inode_operations;
80 static const struct file_operations mqueue_file_operations;
81 static const struct super_operations mqueue_super_ops;
82 static void remove_notification(struct mqueue_inode_info *info);
83
84 static struct kmem_cache *mqueue_inode_cachep;
85
86 static struct ctl_table_header * mq_sysctl_table;
87
88 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
89 {
90         return container_of(inode, struct mqueue_inode_info, vfs_inode);
91 }
92
93 /*
94  * This routine should be called with the mq_lock held.
95  */
96 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
97 {
98         return get_ipc_ns(inode->i_sb->s_fs_info);
99 }
100
101 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
102 {
103         struct ipc_namespace *ns;
104
105         spin_lock(&mq_lock);
106         ns = __get_ns_from_inode(inode);
107         spin_unlock(&mq_lock);
108         return ns;
109 }
110
111 static struct inode *mqueue_get_inode(struct super_block *sb,
112                 struct ipc_namespace *ipc_ns, umode_t mode,
113                 struct mq_attr *attr)
114 {
115         struct user_struct *u = current_user();
116         struct inode *inode;
117         int ret = -ENOMEM;
118
119         inode = new_inode(sb);
120         if (!inode)
121                 goto err;
122
123         inode->i_ino = get_next_ino();
124         inode->i_mode = mode;
125         inode->i_uid = current_fsuid();
126         inode->i_gid = current_fsgid();
127         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
128
129         if (S_ISREG(mode)) {
130                 struct mqueue_inode_info *info;
131                 unsigned long mq_bytes, mq_msg_tblsz;
132
133                 inode->i_fop = &mqueue_file_operations;
134                 inode->i_size = FILENT_SIZE;
135                 /* mqueue specific info */
136                 info = MQUEUE_I(inode);
137                 spin_lock_init(&info->lock);
138                 init_waitqueue_head(&info->wait_q);
139                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
140                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
141                 info->notify_owner = NULL;
142                 info->qsize = 0;
143                 info->user = NULL;      /* set when all is ok */
144                 memset(&info->attr, 0, sizeof(info->attr));
145                 info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
146                 info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
147                 if (attr) {
148                         info->attr.mq_maxmsg = attr->mq_maxmsg;
149                         info->attr.mq_msgsize = attr->mq_msgsize;
150                 }
151                 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
152                 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
153                 if (!info->messages)
154                         goto out_inode;
155
156                 mq_bytes = (mq_msg_tblsz +
157                         (info->attr.mq_maxmsg * info->attr.mq_msgsize));
158
159                 spin_lock(&mq_lock);
160                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
161                     u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
162                         spin_unlock(&mq_lock);
163                         /* mqueue_evict_inode() releases info->messages */
164                         ret = -EMFILE;
165                         goto out_inode;
166                 }
167                 u->mq_bytes += mq_bytes;
168                 spin_unlock(&mq_lock);
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         iput(inode);
183 err:
184         return ERR_PTR(ret);
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;
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 (IS_ERR(inode)) {
201                 error = PTR_ERR(inode);
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                 goto out;
210         }
211         error = 0;
212
213 out:
214         return error;
215 }
216
217 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
218                          int flags, const char *dev_name,
219                          void *data)
220 {
221         if (!(flags & MS_KERNMOUNT))
222                 data = current->nsproxy->ipc_ns;
223         return mount_ns(fs_type, flags, data, mqueue_fill_super);
224 }
225
226 static void init_once(void *foo)
227 {
228         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
229
230         inode_init_once(&p->vfs_inode);
231 }
232
233 static struct inode *mqueue_alloc_inode(struct super_block *sb)
234 {
235         struct mqueue_inode_info *ei;
236
237         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
238         if (!ei)
239                 return NULL;
240         return &ei->vfs_inode;
241 }
242
243 static void mqueue_i_callback(struct rcu_head *head)
244 {
245         struct inode *inode = container_of(head, struct inode, i_rcu);
246         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
247 }
248
249 static void mqueue_destroy_inode(struct inode *inode)
250 {
251         call_rcu(&inode->i_rcu, mqueue_i_callback);
252 }
253
254 static void mqueue_evict_inode(struct inode *inode)
255 {
256         struct mqueue_inode_info *info;
257         struct user_struct *user;
258         unsigned long mq_bytes;
259         int i;
260         struct ipc_namespace *ipc_ns;
261
262         end_writeback(inode);
263
264         if (S_ISDIR(inode->i_mode))
265                 return;
266
267         ipc_ns = get_ns_from_inode(inode);
268         info = MQUEUE_I(inode);
269         spin_lock(&info->lock);
270         for (i = 0; i < info->attr.mq_curmsgs; i++)
271                 free_msg(info->messages[i]);
272         kfree(info->messages);
273         spin_unlock(&info->lock);
274
275         /* Total amount of bytes accounted for the mqueue */
276         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
277             + info->attr.mq_msgsize);
278         user = info->user;
279         if (user) {
280                 spin_lock(&mq_lock);
281                 user->mq_bytes -= mq_bytes;
282                 /*
283                  * get_ns_from_inode() ensures that the
284                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
285                  * to which we now hold a reference, or it is NULL.
286                  * We can't put it here under mq_lock, though.
287                  */
288                 if (ipc_ns)
289                         ipc_ns->mq_queues_count--;
290                 spin_unlock(&mq_lock);
291                 free_uid(user);
292         }
293         if (ipc_ns)
294                 put_ipc_ns(ipc_ns);
295 }
296
297 static int mqueue_create(struct inode *dir, struct dentry *dentry,
298                                 umode_t mode, struct nameidata *nd)
299 {
300         struct inode *inode;
301         struct mq_attr *attr = dentry->d_fsdata;
302         int error;
303         struct ipc_namespace *ipc_ns;
304
305         spin_lock(&mq_lock);
306         ipc_ns = __get_ns_from_inode(dir);
307         if (!ipc_ns) {
308                 error = -EACCES;
309                 goto out_unlock;
310         }
311         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
312                         !capable(CAP_SYS_RESOURCE)) {
313                 error = -ENOSPC;
314                 goto out_unlock;
315         }
316         ipc_ns->mq_queues_count++;
317         spin_unlock(&mq_lock);
318
319         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
320         if (IS_ERR(inode)) {
321                 error = PTR_ERR(inode);
322                 spin_lock(&mq_lock);
323                 ipc_ns->mq_queues_count--;
324                 goto out_unlock;
325         }
326
327         put_ipc_ns(ipc_ns);
328         dir->i_size += DIRENT_SIZE;
329         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
330
331         d_instantiate(dentry, inode);
332         dget(dentry);
333         return 0;
334 out_unlock:
335         spin_unlock(&mq_lock);
336         if (ipc_ns)
337                 put_ipc_ns(ipc_ns);
338         return error;
339 }
340
341 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
342 {
343         struct inode *inode = dentry->d_inode;
344
345         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
346         dir->i_size -= DIRENT_SIZE;
347         drop_nlink(inode);
348         dput(dentry);
349         return 0;
350 }
351
352 /*
353 *       This is routine for system read from queue file.
354 *       To avoid mess with doing here some sort of mq_receive we allow
355 *       to read only queue size & notification info (the only values
356 *       that are interesting from user point of view and aren't accessible
357 *       through std routines)
358 */
359 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
360                                 size_t count, loff_t *off)
361 {
362         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
363         char buffer[FILENT_SIZE];
364         ssize_t ret;
365
366         spin_lock(&info->lock);
367         snprintf(buffer, sizeof(buffer),
368                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
369                         info->qsize,
370                         info->notify_owner ? info->notify.sigev_notify : 0,
371                         (info->notify_owner &&
372                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
373                                 info->notify.sigev_signo : 0,
374                         pid_vnr(info->notify_owner));
375         spin_unlock(&info->lock);
376         buffer[sizeof(buffer)-1] = '\0';
377
378         ret = simple_read_from_buffer(u_data, count, off, buffer,
379                                 strlen(buffer));
380         if (ret <= 0)
381                 return ret;
382
383         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
384         return ret;
385 }
386
387 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
388 {
389         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
390
391         spin_lock(&info->lock);
392         if (task_tgid(current) == info->notify_owner)
393                 remove_notification(info);
394
395         spin_unlock(&info->lock);
396         return 0;
397 }
398
399 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
400 {
401         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
402         int retval = 0;
403
404         poll_wait(filp, &info->wait_q, poll_tab);
405
406         spin_lock(&info->lock);
407         if (info->attr.mq_curmsgs)
408                 retval = POLLIN | POLLRDNORM;
409
410         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
411                 retval |= POLLOUT | POLLWRNORM;
412         spin_unlock(&info->lock);
413
414         return retval;
415 }
416
417 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
418 static void wq_add(struct mqueue_inode_info *info, int sr,
419                         struct ext_wait_queue *ewp)
420 {
421         struct ext_wait_queue *walk;
422
423         ewp->task = current;
424
425         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
426                 if (walk->task->static_prio <= current->static_prio) {
427                         list_add_tail(&ewp->list, &walk->list);
428                         return;
429                 }
430         }
431         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
432 }
433
434 /*
435  * Puts current task to sleep. Caller must hold queue lock. After return
436  * lock isn't held.
437  * sr: SEND or RECV
438  */
439 static int wq_sleep(struct mqueue_inode_info *info, int sr,
440                     ktime_t *timeout, struct ext_wait_queue *ewp)
441 {
442         int retval;
443         signed long time;
444
445         wq_add(info, sr, ewp);
446
447         for (;;) {
448                 set_current_state(TASK_INTERRUPTIBLE);
449
450                 spin_unlock(&info->lock);
451                 time = schedule_hrtimeout_range_clock(timeout, 0,
452                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
453
454                 while (ewp->state == STATE_PENDING)
455                         cpu_relax();
456
457                 if (ewp->state == STATE_READY) {
458                         retval = 0;
459                         goto out;
460                 }
461                 spin_lock(&info->lock);
462                 if (ewp->state == STATE_READY) {
463                         retval = 0;
464                         goto out_unlock;
465                 }
466                 if (signal_pending(current)) {
467                         retval = -ERESTARTSYS;
468                         break;
469                 }
470                 if (time == 0) {
471                         retval = -ETIMEDOUT;
472                         break;
473                 }
474         }
475         list_del(&ewp->list);
476 out_unlock:
477         spin_unlock(&info->lock);
478 out:
479         return retval;
480 }
481
482 /*
483  * Returns waiting task that should be serviced first or NULL if none exists
484  */
485 static struct ext_wait_queue *wq_get_first_waiter(
486                 struct mqueue_inode_info *info, int sr)
487 {
488         struct list_head *ptr;
489
490         ptr = info->e_wait_q[sr].list.prev;
491         if (ptr == &info->e_wait_q[sr].list)
492                 return NULL;
493         return list_entry(ptr, struct ext_wait_queue, list);
494 }
495
496 /* Auxiliary functions to manipulate messages' list */
497 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
498 {
499         int k;
500
501         k = info->attr.mq_curmsgs - 1;
502         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
503                 info->messages[k + 1] = info->messages[k];
504                 k--;
505         }
506         info->attr.mq_curmsgs++;
507         info->qsize += ptr->m_ts;
508         info->messages[k + 1] = ptr;
509 }
510
511 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
512 {
513         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
514         return info->messages[info->attr.mq_curmsgs];
515 }
516
517 static inline void set_cookie(struct sk_buff *skb, char code)
518 {
519         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
520 }
521
522 /*
523  * The next function is only to split too long sys_mq_timedsend
524  */
525 static void __do_notify(struct mqueue_inode_info *info)
526 {
527         /* notification
528          * invoked when there is registered process and there isn't process
529          * waiting synchronously for message AND state of queue changed from
530          * empty to not empty. Here we are sure that no one is waiting
531          * synchronously. */
532         if (info->notify_owner &&
533             info->attr.mq_curmsgs == 1) {
534                 struct siginfo sig_i;
535                 switch (info->notify.sigev_notify) {
536                 case SIGEV_NONE:
537                         break;
538                 case SIGEV_SIGNAL:
539                         /* sends signal */
540
541                         sig_i.si_signo = info->notify.sigev_signo;
542                         sig_i.si_errno = 0;
543                         sig_i.si_code = SI_MESGQ;
544                         sig_i.si_value = info->notify.sigev_value;
545                         /* map current pid/uid into info->owner's namespaces */
546                         rcu_read_lock();
547                         sig_i.si_pid = task_tgid_nr_ns(current,
548                                                 ns_of_pid(info->notify_owner));
549                         sig_i.si_uid = user_ns_map_uid(info->user->user_ns,
550                                                 current_cred(), current_uid());
551                         rcu_read_unlock();
552
553                         kill_pid_info(info->notify.sigev_signo,
554                                       &sig_i, info->notify_owner);
555                         break;
556                 case SIGEV_THREAD:
557                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
558                         netlink_sendskb(info->notify_sock, info->notify_cookie);
559                         break;
560                 }
561                 /* after notification unregisters process */
562                 put_pid(info->notify_owner);
563                 info->notify_owner = NULL;
564         }
565         wake_up(&info->wait_q);
566 }
567
568 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
569                            ktime_t *expires, struct timespec *ts)
570 {
571         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
572                 return -EFAULT;
573         if (!timespec_valid(ts))
574                 return -EINVAL;
575
576         *expires = timespec_to_ktime(*ts);
577         return 0;
578 }
579
580 static void remove_notification(struct mqueue_inode_info *info)
581 {
582         if (info->notify_owner != NULL &&
583             info->notify.sigev_notify == SIGEV_THREAD) {
584                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
585                 netlink_sendskb(info->notify_sock, info->notify_cookie);
586         }
587         put_pid(info->notify_owner);
588         info->notify_owner = NULL;
589 }
590
591 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
592 {
593         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
594                 return 0;
595         if (capable(CAP_SYS_RESOURCE)) {
596                 if (attr->mq_maxmsg > HARD_MSGMAX)
597                         return 0;
598         } else {
599                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
600                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
601                         return 0;
602         }
603         /* check for overflow */
604         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
605                 return 0;
606         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
607             + sizeof (struct msg_msg *))) <
608             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
609                 return 0;
610         return 1;
611 }
612
613 /*
614  * Invoked when creating a new queue via sys_mq_open
615  */
616 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
617                         struct dentry *dentry, int oflag, umode_t mode,
618                         struct mq_attr *attr)
619 {
620         const struct cred *cred = current_cred();
621         struct file *result;
622         int ret;
623
624         if (attr) {
625                 if (!mq_attr_ok(ipc_ns, attr)) {
626                         ret = -EINVAL;
627                         goto out;
628                 }
629                 /* store for use during create */
630                 dentry->d_fsdata = attr;
631         }
632
633         mode &= ~current_umask();
634         ret = mnt_want_write(ipc_ns->mq_mnt);
635         if (ret)
636                 goto out;
637         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
638         dentry->d_fsdata = NULL;
639         if (ret)
640                 goto out_drop_write;
641
642         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
643         /*
644          * dentry_open() took a persistent mnt_want_write(),
645          * so we can now drop this one.
646          */
647         mnt_drop_write(ipc_ns->mq_mnt);
648         return result;
649
650 out_drop_write:
651         mnt_drop_write(ipc_ns->mq_mnt);
652 out:
653         dput(dentry);
654         mntput(ipc_ns->mq_mnt);
655         return ERR_PTR(ret);
656 }
657
658 /* Opens existing queue */
659 static struct file *do_open(struct ipc_namespace *ipc_ns,
660                                 struct dentry *dentry, int oflag)
661 {
662         int ret;
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                 ret = -EINVAL;
670                 goto err;
671         }
672
673         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
674                 ret = -EACCES;
675                 goto err;
676         }
677
678         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
679
680 err:
681         dput(dentry);
682         mntput(ipc_ns->mq_mnt);
683         return ERR_PTR(ret);
684 }
685
686 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
687                 struct mq_attr __user *, u_attr)
688 {
689         struct dentry *dentry;
690         struct file *filp;
691         char *name;
692         struct mq_attr attr;
693         int fd, error;
694         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
695
696         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
697                 return -EFAULT;
698
699         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
700
701         if (IS_ERR(name = getname(u_name)))
702                 return PTR_ERR(name);
703
704         fd = get_unused_fd_flags(O_CLOEXEC);
705         if (fd < 0)
706                 goto out_putname;
707
708         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
709         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
710         if (IS_ERR(dentry)) {
711                 error = PTR_ERR(dentry);
712                 goto out_putfd;
713         }
714         mntget(ipc_ns->mq_mnt);
715
716         if (oflag & O_CREAT) {
717                 if (dentry->d_inode) {  /* entry already exists */
718                         audit_inode(name, dentry);
719                         if (oflag & O_EXCL) {
720                                 error = -EEXIST;
721                                 goto out;
722                         }
723                         filp = do_open(ipc_ns, dentry, oflag);
724                 } else {
725                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
726                                                 dentry, oflag, mode,
727                                                 u_attr ? &attr : NULL);
728                 }
729         } else {
730                 if (!dentry->d_inode) {
731                         error = -ENOENT;
732                         goto out;
733                 }
734                 audit_inode(name, dentry);
735                 filp = do_open(ipc_ns, dentry, oflag);
736         }
737
738         if (IS_ERR(filp)) {
739                 error = PTR_ERR(filp);
740                 goto out_putfd;
741         }
742
743         fd_install(fd, filp);
744         goto out_upsem;
745
746 out:
747         dput(dentry);
748         mntput(ipc_ns->mq_mnt);
749 out_putfd:
750         put_unused_fd(fd);
751         fd = error;
752 out_upsem:
753         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
754 out_putname:
755         putname(name);
756         return fd;
757 }
758
759 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
760 {
761         int err;
762         char *name;
763         struct dentry *dentry;
764         struct inode *inode = NULL;
765         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
766
767         name = getname(u_name);
768         if (IS_ERR(name))
769                 return PTR_ERR(name);
770
771         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
772                         I_MUTEX_PARENT);
773         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
774         if (IS_ERR(dentry)) {
775                 err = PTR_ERR(dentry);
776                 goto out_unlock;
777         }
778
779         if (!dentry->d_inode) {
780                 err = -ENOENT;
781                 goto out_err;
782         }
783
784         inode = dentry->d_inode;
785         if (inode)
786                 ihold(inode);
787         err = mnt_want_write(ipc_ns->mq_mnt);
788         if (err)
789                 goto out_err;
790         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
791         mnt_drop_write(ipc_ns->mq_mnt);
792 out_err:
793         dput(dentry);
794
795 out_unlock:
796         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
797         putname(name);
798         if (inode)
799                 iput(inode);
800
801         return err;
802 }
803
804 /* Pipelined send and receive functions.
805  *
806  * If a receiver finds no waiting message, then it registers itself in the
807  * list of waiting receivers. A sender checks that list before adding the new
808  * message into the message array. If there is a waiting receiver, then it
809  * bypasses the message array and directly hands the message over to the
810  * receiver.
811  * The receiver accepts the message and returns without grabbing the queue
812  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
813  * are necessary. The same algorithm is used for sysv semaphores, see
814  * ipc/sem.c for more details.
815  *
816  * The same algorithm is used for senders.
817  */
818
819 /* pipelined_send() - send a message directly to the task waiting in
820  * sys_mq_timedreceive() (without inserting message into a queue).
821  */
822 static inline void pipelined_send(struct mqueue_inode_info *info,
823                                   struct msg_msg *message,
824                                   struct ext_wait_queue *receiver)
825 {
826         receiver->msg = message;
827         list_del(&receiver->list);
828         receiver->state = STATE_PENDING;
829         wake_up_process(receiver->task);
830         smp_wmb();
831         receiver->state = STATE_READY;
832 }
833
834 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
835  * gets its message and put to the queue (we have one free place for sure). */
836 static inline void pipelined_receive(struct mqueue_inode_info *info)
837 {
838         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
839
840         if (!sender) {
841                 /* for poll */
842                 wake_up_interruptible(&info->wait_q);
843                 return;
844         }
845         msg_insert(sender->msg, info);
846         list_del(&sender->list);
847         sender->state = STATE_PENDING;
848         wake_up_process(sender->task);
849         smp_wmb();
850         sender->state = STATE_READY;
851 }
852
853 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
854                 size_t, msg_len, unsigned int, msg_prio,
855                 const struct timespec __user *, u_abs_timeout)
856 {
857         struct file *filp;
858         struct inode *inode;
859         struct ext_wait_queue wait;
860         struct ext_wait_queue *receiver;
861         struct msg_msg *msg_ptr;
862         struct mqueue_inode_info *info;
863         ktime_t expires, *timeout = NULL;
864         struct timespec ts;
865         int ret;
866
867         if (u_abs_timeout) {
868                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
869                 if (res)
870                         return res;
871                 timeout = &expires;
872         }
873
874         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
875                 return -EINVAL;
876
877         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
878
879         filp = fget(mqdes);
880         if (unlikely(!filp)) {
881                 ret = -EBADF;
882                 goto out;
883         }
884
885         inode = filp->f_path.dentry->d_inode;
886         if (unlikely(filp->f_op != &mqueue_file_operations)) {
887                 ret = -EBADF;
888                 goto out_fput;
889         }
890         info = MQUEUE_I(inode);
891         audit_inode(NULL, filp->f_path.dentry);
892
893         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
894                 ret = -EBADF;
895                 goto out_fput;
896         }
897
898         if (unlikely(msg_len > info->attr.mq_msgsize)) {
899                 ret = -EMSGSIZE;
900                 goto out_fput;
901         }
902
903         /* First try to allocate memory, before doing anything with
904          * existing queues. */
905         msg_ptr = load_msg(u_msg_ptr, msg_len);
906         if (IS_ERR(msg_ptr)) {
907                 ret = PTR_ERR(msg_ptr);
908                 goto out_fput;
909         }
910         msg_ptr->m_ts = msg_len;
911         msg_ptr->m_type = msg_prio;
912
913         spin_lock(&info->lock);
914
915         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
916                 if (filp->f_flags & O_NONBLOCK) {
917                         spin_unlock(&info->lock);
918                         ret = -EAGAIN;
919                 } else {
920                         wait.task = current;
921                         wait.msg = (void *) msg_ptr;
922                         wait.state = STATE_NONE;
923                         ret = wq_sleep(info, SEND, timeout, &wait);
924                 }
925                 if (ret < 0)
926                         free_msg(msg_ptr);
927         } else {
928                 receiver = wq_get_first_waiter(info, RECV);
929                 if (receiver) {
930                         pipelined_send(info, msg_ptr, receiver);
931                 } else {
932                         /* adds message to the queue */
933                         msg_insert(msg_ptr, info);
934                         __do_notify(info);
935                 }
936                 inode->i_atime = inode->i_mtime = inode->i_ctime =
937                                 CURRENT_TIME;
938                 spin_unlock(&info->lock);
939                 ret = 0;
940         }
941 out_fput:
942         fput(filp);
943 out:
944         return ret;
945 }
946
947 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
948                 size_t, msg_len, unsigned int __user *, u_msg_prio,
949                 const struct timespec __user *, u_abs_timeout)
950 {
951         ssize_t ret;
952         struct msg_msg *msg_ptr;
953         struct file *filp;
954         struct inode *inode;
955         struct mqueue_inode_info *info;
956         struct ext_wait_queue wait;
957         ktime_t expires, *timeout = NULL;
958         struct timespec ts;
959
960         if (u_abs_timeout) {
961                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
962                 if (res)
963                         return res;
964                 timeout = &expires;
965         }
966
967         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
968
969         filp = fget(mqdes);
970         if (unlikely(!filp)) {
971                 ret = -EBADF;
972                 goto out;
973         }
974
975         inode = filp->f_path.dentry->d_inode;
976         if (unlikely(filp->f_op != &mqueue_file_operations)) {
977                 ret = -EBADF;
978                 goto out_fput;
979         }
980         info = MQUEUE_I(inode);
981         audit_inode(NULL, filp->f_path.dentry);
982
983         if (unlikely(!(filp->f_mode & FMODE_READ))) {
984                 ret = -EBADF;
985                 goto out_fput;
986         }
987
988         /* checks if buffer is big enough */
989         if (unlikely(msg_len < info->attr.mq_msgsize)) {
990                 ret = -EMSGSIZE;
991                 goto out_fput;
992         }
993
994         spin_lock(&info->lock);
995         if (info->attr.mq_curmsgs == 0) {
996                 if (filp->f_flags & O_NONBLOCK) {
997                         spin_unlock(&info->lock);
998                         ret = -EAGAIN;
999                 } else {
1000                         wait.task = current;
1001                         wait.state = STATE_NONE;
1002                         ret = wq_sleep(info, RECV, timeout, &wait);
1003                         msg_ptr = wait.msg;
1004                 }
1005         } else {
1006                 msg_ptr = msg_get(info);
1007
1008                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1009                                 CURRENT_TIME;
1010
1011                 /* There is now free space in queue. */
1012                 pipelined_receive(info);
1013                 spin_unlock(&info->lock);
1014                 ret = 0;
1015         }
1016         if (ret == 0) {
1017                 ret = msg_ptr->m_ts;
1018
1019                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1020                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1021                         ret = -EFAULT;
1022                 }
1023                 free_msg(msg_ptr);
1024         }
1025 out_fput:
1026         fput(filp);
1027 out:
1028         return ret;
1029 }
1030
1031 /*
1032  * Notes: the case when user wants us to deregister (with NULL as pointer)
1033  * and he isn't currently owner of notification, will be silently discarded.
1034  * It isn't explicitly defined in the POSIX.
1035  */
1036 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1037                 const struct sigevent __user *, u_notification)
1038 {
1039         int ret;
1040         struct file *filp;
1041         struct sock *sock;
1042         struct inode *inode;
1043         struct sigevent notification;
1044         struct mqueue_inode_info *info;
1045         struct sk_buff *nc;
1046
1047         if (u_notification) {
1048                 if (copy_from_user(&notification, u_notification,
1049                                         sizeof(struct sigevent)))
1050                         return -EFAULT;
1051         }
1052
1053         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1054
1055         nc = NULL;
1056         sock = NULL;
1057         if (u_notification != NULL) {
1058                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1059                              notification.sigev_notify != SIGEV_SIGNAL &&
1060                              notification.sigev_notify != SIGEV_THREAD))
1061                         return -EINVAL;
1062                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1063                         !valid_signal(notification.sigev_signo)) {
1064                         return -EINVAL;
1065                 }
1066                 if (notification.sigev_notify == SIGEV_THREAD) {
1067                         long timeo;
1068
1069                         /* create the notify skb */
1070                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1071                         if (!nc) {
1072                                 ret = -ENOMEM;
1073                                 goto out;
1074                         }
1075                         if (copy_from_user(nc->data,
1076                                         notification.sigev_value.sival_ptr,
1077                                         NOTIFY_COOKIE_LEN)) {
1078                                 ret = -EFAULT;
1079                                 goto out;
1080                         }
1081
1082                         /* TODO: add a header? */
1083                         skb_put(nc, NOTIFY_COOKIE_LEN);
1084                         /* and attach it to the socket */
1085 retry:
1086                         filp = fget(notification.sigev_signo);
1087                         if (!filp) {
1088                                 ret = -EBADF;
1089                                 goto out;
1090                         }
1091                         sock = netlink_getsockbyfilp(filp);
1092                         fput(filp);
1093                         if (IS_ERR(sock)) {
1094                                 ret = PTR_ERR(sock);
1095                                 sock = NULL;
1096                                 goto out;
1097                         }
1098
1099                         timeo = MAX_SCHEDULE_TIMEOUT;
1100                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1101                         if (ret == 1)
1102                                 goto retry;
1103                         if (ret) {
1104                                 sock = NULL;
1105                                 nc = NULL;
1106                                 goto out;
1107                         }
1108                 }
1109         }
1110
1111         filp = fget(mqdes);
1112         if (!filp) {
1113                 ret = -EBADF;
1114                 goto out;
1115         }
1116
1117         inode = filp->f_path.dentry->d_inode;
1118         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1119                 ret = -EBADF;
1120                 goto out_fput;
1121         }
1122         info = MQUEUE_I(inode);
1123
1124         ret = 0;
1125         spin_lock(&info->lock);
1126         if (u_notification == NULL) {
1127                 if (info->notify_owner == task_tgid(current)) {
1128                         remove_notification(info);
1129                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1130                 }
1131         } else if (info->notify_owner != NULL) {
1132                 ret = -EBUSY;
1133         } else {
1134                 switch (notification.sigev_notify) {
1135                 case SIGEV_NONE:
1136                         info->notify.sigev_notify = SIGEV_NONE;
1137                         break;
1138                 case SIGEV_THREAD:
1139                         info->notify_sock = sock;
1140                         info->notify_cookie = nc;
1141                         sock = NULL;
1142                         nc = NULL;
1143                         info->notify.sigev_notify = SIGEV_THREAD;
1144                         break;
1145                 case SIGEV_SIGNAL:
1146                         info->notify.sigev_signo = notification.sigev_signo;
1147                         info->notify.sigev_value = notification.sigev_value;
1148                         info->notify.sigev_notify = SIGEV_SIGNAL;
1149                         break;
1150                 }
1151
1152                 info->notify_owner = get_pid(task_tgid(current));
1153                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1154         }
1155         spin_unlock(&info->lock);
1156 out_fput:
1157         fput(filp);
1158 out:
1159         if (sock) {
1160                 netlink_detachskb(sock, nc);
1161         } else if (nc) {
1162                 dev_kfree_skb(nc);
1163         }
1164         return ret;
1165 }
1166
1167 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1168                 const struct mq_attr __user *, u_mqstat,
1169                 struct mq_attr __user *, u_omqstat)
1170 {
1171         int ret;
1172         struct mq_attr mqstat, omqstat;
1173         struct file *filp;
1174         struct inode *inode;
1175         struct mqueue_inode_info *info;
1176
1177         if (u_mqstat != NULL) {
1178                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1179                         return -EFAULT;
1180                 if (mqstat.mq_flags & (~O_NONBLOCK))
1181                         return -EINVAL;
1182         }
1183
1184         filp = fget(mqdes);
1185         if (!filp) {
1186                 ret = -EBADF;
1187                 goto out;
1188         }
1189
1190         inode = filp->f_path.dentry->d_inode;
1191         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1192                 ret = -EBADF;
1193                 goto out_fput;
1194         }
1195         info = MQUEUE_I(inode);
1196
1197         spin_lock(&info->lock);
1198
1199         omqstat = info->attr;
1200         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1201         if (u_mqstat) {
1202                 audit_mq_getsetattr(mqdes, &mqstat);
1203                 spin_lock(&filp->f_lock);
1204                 if (mqstat.mq_flags & O_NONBLOCK)
1205                         filp->f_flags |= O_NONBLOCK;
1206                 else
1207                         filp->f_flags &= ~O_NONBLOCK;
1208                 spin_unlock(&filp->f_lock);
1209
1210                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1211         }
1212
1213         spin_unlock(&info->lock);
1214
1215         ret = 0;
1216         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1217                                                 sizeof(struct mq_attr)))
1218                 ret = -EFAULT;
1219
1220 out_fput:
1221         fput(filp);
1222 out:
1223         return ret;
1224 }
1225
1226 static const struct inode_operations mqueue_dir_inode_operations = {
1227         .lookup = simple_lookup,
1228         .create = mqueue_create,
1229         .unlink = mqueue_unlink,
1230 };
1231
1232 static const struct file_operations mqueue_file_operations = {
1233         .flush = mqueue_flush_file,
1234         .poll = mqueue_poll_file,
1235         .read = mqueue_read_file,
1236         .llseek = default_llseek,
1237 };
1238
1239 static const struct super_operations mqueue_super_ops = {
1240         .alloc_inode = mqueue_alloc_inode,
1241         .destroy_inode = mqueue_destroy_inode,
1242         .evict_inode = mqueue_evict_inode,
1243         .statfs = simple_statfs,
1244 };
1245
1246 static struct file_system_type mqueue_fs_type = {
1247         .name = "mqueue",
1248         .mount = mqueue_mount,
1249         .kill_sb = kill_litter_super,
1250 };
1251
1252 int mq_init_ns(struct ipc_namespace *ns)
1253 {
1254         ns->mq_queues_count  = 0;
1255         ns->mq_queues_max    = DFLT_QUEUESMAX;
1256         ns->mq_msg_max       = DFLT_MSGMAX;
1257         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1258
1259         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1260         if (IS_ERR(ns->mq_mnt)) {
1261                 int err = PTR_ERR(ns->mq_mnt);
1262                 ns->mq_mnt = NULL;
1263                 return err;
1264         }
1265         return 0;
1266 }
1267
1268 void mq_clear_sbinfo(struct ipc_namespace *ns)
1269 {
1270         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1271 }
1272
1273 void mq_put_mnt(struct ipc_namespace *ns)
1274 {
1275         kern_unmount(ns->mq_mnt);
1276 }
1277
1278 static int __init init_mqueue_fs(void)
1279 {
1280         int error;
1281
1282         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1283                                 sizeof(struct mqueue_inode_info), 0,
1284                                 SLAB_HWCACHE_ALIGN, init_once);
1285         if (mqueue_inode_cachep == NULL)
1286                 return -ENOMEM;
1287
1288         /* ignore failures - they are not fatal */
1289         mq_sysctl_table = mq_register_sysctl_table();
1290
1291         error = register_filesystem(&mqueue_fs_type);
1292         if (error)
1293                 goto out_sysctl;
1294
1295         spin_lock_init(&mq_lock);
1296
1297         error = mq_init_ns(&init_ipc_ns);
1298         if (error)
1299                 goto out_filesystem;
1300
1301         return 0;
1302
1303 out_filesystem:
1304         unregister_filesystem(&mqueue_fs_type);
1305 out_sysctl:
1306         if (mq_sysctl_table)
1307                 unregister_sysctl_table(mq_sysctl_table);
1308         kmem_cache_destroy(mqueue_inode_cachep);
1309         return error;
1310 }
1311
1312 __initcall(init_mqueue_fs);