3853116a2ef8ba29a5946ab55a180ac1c57c93ce
[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
36 #include <net/sock.h>
37 #include "util.h"
38
39 #define MQUEUE_MAGIC    0x19800202
40 #define DIRENT_SIZE     20
41 #define FILENT_SIZE     80
42
43 #define SEND            0
44 #define RECV            1
45
46 #define STATE_NONE      0
47 #define STATE_PENDING   1
48 #define STATE_READY     2
49
50 struct ext_wait_queue {         /* queue of sleeping tasks */
51         struct task_struct *task;
52         struct list_head list;
53         struct msg_msg *msg;    /* ptr of loaded message */
54         int state;              /* one of STATE_* values */
55 };
56
57 struct mqueue_inode_info {
58         spinlock_t lock;
59         struct inode vfs_inode;
60         wait_queue_head_t wait_q;
61
62         struct msg_msg **messages;
63         struct mq_attr attr;
64
65         struct sigevent notify;
66         struct pid* notify_owner;
67         struct user_struct *user;       /* user who created, for accounting */
68         struct sock *notify_sock;
69         struct sk_buff *notify_cookie;
70
71         /* for tasks waiting for free space and messages, respectively */
72         struct ext_wait_queue e_wait_q[2];
73
74         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
75 };
76
77 static const struct inode_operations mqueue_dir_inode_operations;
78 static const struct file_operations mqueue_file_operations;
79 static const struct super_operations mqueue_super_ops;
80 static void remove_notification(struct mqueue_inode_info *info);
81
82 static struct kmem_cache *mqueue_inode_cachep;
83
84 static struct ctl_table_header * mq_sysctl_table;
85
86 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
87 {
88         return container_of(inode, struct mqueue_inode_info, vfs_inode);
89 }
90
91 /*
92  * This routine should be called with the mq_lock held.
93  */
94 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
95 {
96         return get_ipc_ns(inode->i_sb->s_fs_info);
97 }
98
99 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
100 {
101         struct ipc_namespace *ns;
102
103         spin_lock(&mq_lock);
104         ns = __get_ns_from_inode(inode);
105         spin_unlock(&mq_lock);
106         return ns;
107 }
108
109 static struct inode *mqueue_get_inode(struct super_block *sb,
110                 struct ipc_namespace *ipc_ns, int mode,
111                 struct mq_attr *attr)
112 {
113         struct user_struct *u = current_user();
114         struct inode *inode;
115
116         inode = new_inode(sb);
117         if (inode) {
118                 inode->i_mode = mode;
119                 inode->i_uid = current_fsuid();
120                 inode->i_gid = current_fsgid();
121                 inode->i_mtime = inode->i_ctime = inode->i_atime =
122                                 CURRENT_TIME;
123
124                 if (S_ISREG(mode)) {
125                         struct mqueue_inode_info *info;
126                         struct task_struct *p = current;
127                         unsigned long mq_bytes, mq_msg_tblsz;
128
129                         inode->i_fop = &mqueue_file_operations;
130                         inode->i_size = FILENT_SIZE;
131                         /* mqueue specific info */
132                         info = MQUEUE_I(inode);
133                         spin_lock_init(&info->lock);
134                         init_waitqueue_head(&info->wait_q);
135                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
136                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
137                         info->notify_owner = NULL;
138                         info->qsize = 0;
139                         info->user = NULL;      /* set when all is ok */
140                         memset(&info->attr, 0, sizeof(info->attr));
141                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
142                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
143                         if (attr) {
144                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
145                                 info->attr.mq_msgsize = attr->mq_msgsize;
146                         }
147                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
148                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
149                         if (!info->messages)
150                                 goto out_inode;
151
152                         mq_bytes = (mq_msg_tblsz +
153                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
154
155                         spin_lock(&mq_lock);
156                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
157                             u->mq_bytes + mq_bytes >
158                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
159                                 spin_unlock(&mq_lock);
160                                 kfree(info->messages);
161                                 goto out_inode;
162                         }
163                         u->mq_bytes += mq_bytes;
164                         spin_unlock(&mq_lock);
165
166                         /* all is ok */
167                         info->user = get_uid(u);
168                 } else if (S_ISDIR(mode)) {
169                         inc_nlink(inode);
170                         /* Some things misbehave if size == 0 on a directory */
171                         inode->i_size = 2 * DIRENT_SIZE;
172                         inode->i_op = &mqueue_dir_inode_operations;
173                         inode->i_fop = &simple_dir_operations;
174                 }
175         }
176         return inode;
177 out_inode:
178         make_bad_inode(inode);
179         iput(inode);
180         return NULL;
181 }
182
183 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
184 {
185         struct inode *inode;
186         struct ipc_namespace *ns = data;
187         int error = 0;
188
189         sb->s_blocksize = PAGE_CACHE_SIZE;
190         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
191         sb->s_magic = MQUEUE_MAGIC;
192         sb->s_op = &mqueue_super_ops;
193
194         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
195                                 NULL);
196         if (!inode) {
197                 error = -ENOMEM;
198                 goto out;
199         }
200
201         sb->s_root = d_alloc_root(inode);
202         if (!sb->s_root) {
203                 iput(inode);
204                 error = -ENOMEM;
205         }
206
207 out:
208         return error;
209 }
210
211 static int mqueue_get_sb(struct file_system_type *fs_type,
212                          int flags, const char *dev_name,
213                          void *data, struct vfsmount *mnt)
214 {
215         if (!(flags & MS_KERNMOUNT))
216                 data = current->nsproxy->ipc_ns;
217         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
218 }
219
220 static void init_once(void *foo)
221 {
222         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
223
224         inode_init_once(&p->vfs_inode);
225 }
226
227 static struct inode *mqueue_alloc_inode(struct super_block *sb)
228 {
229         struct mqueue_inode_info *ei;
230
231         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
232         if (!ei)
233                 return NULL;
234         return &ei->vfs_inode;
235 }
236
237 static void mqueue_destroy_inode(struct inode *inode)
238 {
239         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
240 }
241
242 static void mqueue_delete_inode(struct inode *inode)
243 {
244         struct mqueue_inode_info *info;
245         struct user_struct *user;
246         unsigned long mq_bytes;
247         int i;
248         struct ipc_namespace *ipc_ns;
249
250         if (S_ISDIR(inode->i_mode)) {
251                 clear_inode(inode);
252                 return;
253         }
254         ipc_ns = get_ns_from_inode(inode);
255         info = MQUEUE_I(inode);
256         spin_lock(&info->lock);
257         for (i = 0; i < info->attr.mq_curmsgs; i++)
258                 free_msg(info->messages[i]);
259         kfree(info->messages);
260         spin_unlock(&info->lock);
261
262         clear_inode(inode);
263
264         /* Total amount of bytes accounted for the mqueue */
265         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
266             + info->attr.mq_msgsize);
267         user = info->user;
268         if (user) {
269                 spin_lock(&mq_lock);
270                 user->mq_bytes -= mq_bytes;
271                 /*
272                  * get_ns_from_inode() ensures that the
273                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
274                  * to which we now hold a reference, or it is NULL.
275                  * We can't put it here under mq_lock, though.
276                  */
277                 if (ipc_ns)
278                         ipc_ns->mq_queues_count--;
279                 spin_unlock(&mq_lock);
280                 free_uid(user);
281         }
282         if (ipc_ns)
283                 put_ipc_ns(ipc_ns);
284 }
285
286 static int mqueue_create(struct inode *dir, struct dentry *dentry,
287                                 int mode, struct nameidata *nd)
288 {
289         struct inode *inode;
290         struct mq_attr *attr = dentry->d_fsdata;
291         int error;
292         struct ipc_namespace *ipc_ns;
293
294         spin_lock(&mq_lock);
295         ipc_ns = __get_ns_from_inode(dir);
296         if (!ipc_ns) {
297                 error = -EACCES;
298                 goto out_unlock;
299         }
300         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
301                         !capable(CAP_SYS_RESOURCE)) {
302                 error = -ENOSPC;
303                 goto out_unlock;
304         }
305         ipc_ns->mq_queues_count++;
306         spin_unlock(&mq_lock);
307
308         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
309         if (!inode) {
310                 error = -ENOMEM;
311                 spin_lock(&mq_lock);
312                 ipc_ns->mq_queues_count--;
313                 goto out_unlock;
314         }
315
316         put_ipc_ns(ipc_ns);
317         dir->i_size += DIRENT_SIZE;
318         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
319
320         d_instantiate(dentry, inode);
321         dget(dentry);
322         return 0;
323 out_unlock:
324         spin_unlock(&mq_lock);
325         if (ipc_ns)
326                 put_ipc_ns(ipc_ns);
327         return error;
328 }
329
330 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
331 {
332         struct inode *inode = dentry->d_inode;
333
334         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
335         dir->i_size -= DIRENT_SIZE;
336         drop_nlink(inode);
337         dput(dentry);
338         return 0;
339 }
340
341 /*
342 *       This is routine for system read from queue file.
343 *       To avoid mess with doing here some sort of mq_receive we allow
344 *       to read only queue size & notification info (the only values
345 *       that are interesting from user point of view and aren't accessible
346 *       through std routines)
347 */
348 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
349                                 size_t count, loff_t *off)
350 {
351         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
352         char buffer[FILENT_SIZE];
353         ssize_t ret;
354
355         spin_lock(&info->lock);
356         snprintf(buffer, sizeof(buffer),
357                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
358                         info->qsize,
359                         info->notify_owner ? info->notify.sigev_notify : 0,
360                         (info->notify_owner &&
361                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
362                                 info->notify.sigev_signo : 0,
363                         pid_vnr(info->notify_owner));
364         spin_unlock(&info->lock);
365         buffer[sizeof(buffer)-1] = '\0';
366
367         ret = simple_read_from_buffer(u_data, count, off, buffer,
368                                 strlen(buffer));
369         if (ret <= 0)
370                 return ret;
371
372         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
373         return ret;
374 }
375
376 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
377 {
378         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
379
380         spin_lock(&info->lock);
381         if (task_tgid(current) == info->notify_owner)
382                 remove_notification(info);
383
384         spin_unlock(&info->lock);
385         return 0;
386 }
387
388 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
389 {
390         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
391         int retval = 0;
392
393         poll_wait(filp, &info->wait_q, poll_tab);
394
395         spin_lock(&info->lock);
396         if (info->attr.mq_curmsgs)
397                 retval = POLLIN | POLLRDNORM;
398
399         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
400                 retval |= POLLOUT | POLLWRNORM;
401         spin_unlock(&info->lock);
402
403         return retval;
404 }
405
406 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
407 static void wq_add(struct mqueue_inode_info *info, int sr,
408                         struct ext_wait_queue *ewp)
409 {
410         struct ext_wait_queue *walk;
411
412         ewp->task = current;
413
414         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
415                 if (walk->task->static_prio <= current->static_prio) {
416                         list_add_tail(&ewp->list, &walk->list);
417                         return;
418                 }
419         }
420         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
421 }
422
423 /*
424  * Puts current task to sleep. Caller must hold queue lock. After return
425  * lock isn't held.
426  * sr: SEND or RECV
427  */
428 static int wq_sleep(struct mqueue_inode_info *info, int sr,
429                         long timeout, struct ext_wait_queue *ewp)
430 {
431         int retval;
432         signed long time;
433
434         wq_add(info, sr, ewp);
435
436         for (;;) {
437                 set_current_state(TASK_INTERRUPTIBLE);
438
439                 spin_unlock(&info->lock);
440                 time = schedule_timeout(timeout);
441
442                 while (ewp->state == STATE_PENDING)
443                         cpu_relax();
444
445                 if (ewp->state == STATE_READY) {
446                         retval = 0;
447                         goto out;
448                 }
449                 spin_lock(&info->lock);
450                 if (ewp->state == STATE_READY) {
451                         retval = 0;
452                         goto out_unlock;
453                 }
454                 if (signal_pending(current)) {
455                         retval = -ERESTARTSYS;
456                         break;
457                 }
458                 if (time == 0) {
459                         retval = -ETIMEDOUT;
460                         break;
461                 }
462         }
463         list_del(&ewp->list);
464 out_unlock:
465         spin_unlock(&info->lock);
466 out:
467         return retval;
468 }
469
470 /*
471  * Returns waiting task that should be serviced first or NULL if none exists
472  */
473 static struct ext_wait_queue *wq_get_first_waiter(
474                 struct mqueue_inode_info *info, int sr)
475 {
476         struct list_head *ptr;
477
478         ptr = info->e_wait_q[sr].list.prev;
479         if (ptr == &info->e_wait_q[sr].list)
480                 return NULL;
481         return list_entry(ptr, struct ext_wait_queue, list);
482 }
483
484 /* Auxiliary functions to manipulate messages' list */
485 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
486 {
487         int k;
488
489         k = info->attr.mq_curmsgs - 1;
490         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
491                 info->messages[k + 1] = info->messages[k];
492                 k--;
493         }
494         info->attr.mq_curmsgs++;
495         info->qsize += ptr->m_ts;
496         info->messages[k + 1] = ptr;
497 }
498
499 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
500 {
501         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
502         return info->messages[info->attr.mq_curmsgs];
503 }
504
505 static inline void set_cookie(struct sk_buff *skb, char code)
506 {
507         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
508 }
509
510 /*
511  * The next function is only to split too long sys_mq_timedsend
512  */
513 static void __do_notify(struct mqueue_inode_info *info)
514 {
515         /* notification
516          * invoked when there is registered process and there isn't process
517          * waiting synchronously for message AND state of queue changed from
518          * empty to not empty. Here we are sure that no one is waiting
519          * synchronously. */
520         if (info->notify_owner &&
521             info->attr.mq_curmsgs == 1) {
522                 struct siginfo sig_i;
523                 switch (info->notify.sigev_notify) {
524                 case SIGEV_NONE:
525                         break;
526                 case SIGEV_SIGNAL:
527                         /* sends signal */
528
529                         sig_i.si_signo = info->notify.sigev_signo;
530                         sig_i.si_errno = 0;
531                         sig_i.si_code = SI_MESGQ;
532                         sig_i.si_value = info->notify.sigev_value;
533                         sig_i.si_pid = task_tgid_nr_ns(current,
534                                                 ns_of_pid(info->notify_owner));
535                         sig_i.si_uid = current_uid();
536
537                         kill_pid_info(info->notify.sigev_signo,
538                                       &sig_i, info->notify_owner);
539                         break;
540                 case SIGEV_THREAD:
541                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
542                         netlink_sendskb(info->notify_sock, info->notify_cookie);
543                         break;
544                 }
545                 /* after notification unregisters process */
546                 put_pid(info->notify_owner);
547                 info->notify_owner = NULL;
548         }
549         wake_up(&info->wait_q);
550 }
551
552 static long prepare_timeout(struct timespec *p)
553 {
554         struct timespec nowts;
555         long timeout;
556
557         if (p) {
558                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
559                         || p->tv_nsec >= NSEC_PER_SEC))
560                         return -EINVAL;
561                 nowts = CURRENT_TIME;
562                 /* first subtract as jiffies can't be too big */
563                 p->tv_sec -= nowts.tv_sec;
564                 if (p->tv_nsec < nowts.tv_nsec) {
565                         p->tv_nsec += NSEC_PER_SEC;
566                         p->tv_sec--;
567                 }
568                 p->tv_nsec -= nowts.tv_nsec;
569                 if (p->tv_sec < 0)
570                         return 0;
571
572                 timeout = timespec_to_jiffies(p) + 1;
573         } else
574                 return MAX_SCHEDULE_TIMEOUT;
575
576         return timeout;
577 }
578
579 static void remove_notification(struct mqueue_inode_info *info)
580 {
581         if (info->notify_owner != NULL &&
582             info->notify.sigev_notify == SIGEV_THREAD) {
583                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
584                 netlink_sendskb(info->notify_sock, info->notify_cookie);
585         }
586         put_pid(info->notify_owner);
587         info->notify_owner = NULL;
588 }
589
590 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
591 {
592         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
593                 return 0;
594         if (capable(CAP_SYS_RESOURCE)) {
595                 if (attr->mq_maxmsg > HARD_MSGMAX)
596                         return 0;
597         } else {
598                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
599                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
600                         return 0;
601         }
602         /* check for overflow */
603         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
604                 return 0;
605         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
606             + sizeof (struct msg_msg *))) <
607             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
608                 return 0;
609         return 1;
610 }
611
612 /*
613  * Invoked when creating a new queue via sys_mq_open
614  */
615 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
616                         struct dentry *dentry, int oflag, mode_t mode,
617                         struct mq_attr *attr)
618 {
619         const struct cred *cred = current_cred();
620         struct file *result;
621         int ret;
622
623         if (attr) {
624                 ret = -EINVAL;
625                 if (!mq_attr_ok(ipc_ns, attr))
626                         goto out;
627                 /* store for use during create */
628                 dentry->d_fsdata = attr;
629         }
630
631         mode &= ~current_umask();
632         ret = mnt_want_write(ipc_ns->mq_mnt);
633         if (ret)
634                 goto out;
635         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
636         dentry->d_fsdata = NULL;
637         if (ret)
638                 goto out_drop_write;
639
640         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
641         /*
642          * dentry_open() took a persistent mnt_want_write(),
643          * so we can now drop this one.
644          */
645         mnt_drop_write(ipc_ns->mq_mnt);
646         return result;
647
648 out_drop_write:
649         mnt_drop_write(ipc_ns->mq_mnt);
650 out:
651         dput(dentry);
652         mntput(ipc_ns->mq_mnt);
653         return ERR_PTR(ret);
654 }
655
656 /* Opens existing queue */
657 static struct file *do_open(struct ipc_namespace *ipc_ns,
658                                 struct dentry *dentry, int oflag)
659 {
660         int ret;
661         const struct cred *cred = current_cred();
662
663         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
664                                                   MAY_READ | MAY_WRITE };
665
666         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
667                 ret = -EINVAL;
668                 goto err;
669         }
670
671         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
672                 ret = -EACCES;
673                 goto err;
674         }
675
676         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
677
678 err:
679         dput(dentry);
680         mntput(ipc_ns->mq_mnt);
681         return ERR_PTR(ret);
682 }
683
684 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
685                 struct mq_attr __user *, u_attr)
686 {
687         struct dentry *dentry;
688         struct file *filp;
689         char *name;
690         struct mq_attr attr;
691         int fd, error;
692         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
693
694         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
695                 return -EFAULT;
696
697         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
698
699         if (IS_ERR(name = getname(u_name)))
700                 return PTR_ERR(name);
701
702         fd = get_unused_fd_flags(O_CLOEXEC);
703         if (fd < 0)
704                 goto out_putname;
705
706         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
707         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
708         if (IS_ERR(dentry)) {
709                 error = PTR_ERR(dentry);
710                 goto out_putfd;
711         }
712         mntget(ipc_ns->mq_mnt);
713
714         if (oflag & O_CREAT) {
715                 if (dentry->d_inode) {  /* entry already exists */
716                         audit_inode(name, dentry);
717                         error = -EEXIST;
718                         if (oflag & O_EXCL)
719                                 goto out;
720                         filp = do_open(ipc_ns, dentry, oflag);
721                 } else {
722                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
723                                                 dentry, oflag, mode,
724                                                 u_attr ? &attr : NULL);
725                 }
726         } else {
727                 error = -ENOENT;
728                 if (!dentry->d_inode)
729                         goto out;
730                 audit_inode(name, dentry);
731                 filp = do_open(ipc_ns, dentry, oflag);
732         }
733
734         if (IS_ERR(filp)) {
735                 error = PTR_ERR(filp);
736                 goto out_putfd;
737         }
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         fd = error;
748 out_upsem:
749         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
750 out_putname:
751         putname(name);
752         return fd;
753 }
754
755 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
756 {
757         int err;
758         char *name;
759         struct dentry *dentry;
760         struct inode *inode = NULL;
761         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
762
763         name = getname(u_name);
764         if (IS_ERR(name))
765                 return PTR_ERR(name);
766
767         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
768                         I_MUTEX_PARENT);
769         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
770         if (IS_ERR(dentry)) {
771                 err = PTR_ERR(dentry);
772                 goto out_unlock;
773         }
774
775         if (!dentry->d_inode) {
776                 err = -ENOENT;
777                 goto out_err;
778         }
779
780         inode = dentry->d_inode;
781         if (inode)
782                 atomic_inc(&inode->i_count);
783         err = mnt_want_write(ipc_ns->mq_mnt);
784         if (err)
785                 goto out_err;
786         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
787         mnt_drop_write(ipc_ns->mq_mnt);
788 out_err:
789         dput(dentry);
790
791 out_unlock:
792         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
793         putname(name);
794         if (inode)
795                 iput(inode);
796
797         return err;
798 }
799
800 /* Pipelined send and receive functions.
801  *
802  * If a receiver finds no waiting message, then it registers itself in the
803  * list of waiting receivers. A sender checks that list before adding the new
804  * message into the message array. If there is a waiting receiver, then it
805  * bypasses the message array and directly hands the message over to the
806  * receiver.
807  * The receiver accepts the message and returns without grabbing the queue
808  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
809  * are necessary. The same algorithm is used for sysv semaphores, see
810  * ipc/sem.c for more details.
811  *
812  * The same algorithm is used for senders.
813  */
814
815 /* pipelined_send() - send a message directly to the task waiting in
816  * sys_mq_timedreceive() (without inserting message into a queue).
817  */
818 static inline void pipelined_send(struct mqueue_inode_info *info,
819                                   struct msg_msg *message,
820                                   struct ext_wait_queue *receiver)
821 {
822         receiver->msg = message;
823         list_del(&receiver->list);
824         receiver->state = STATE_PENDING;
825         wake_up_process(receiver->task);
826         smp_wmb();
827         receiver->state = STATE_READY;
828 }
829
830 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
831  * gets its message and put to the queue (we have one free place for sure). */
832 static inline void pipelined_receive(struct mqueue_inode_info *info)
833 {
834         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
835
836         if (!sender) {
837                 /* for poll */
838                 wake_up_interruptible(&info->wait_q);
839                 return;
840         }
841         msg_insert(sender->msg, info);
842         list_del(&sender->list);
843         sender->state = STATE_PENDING;
844         wake_up_process(sender->task);
845         smp_wmb();
846         sender->state = STATE_READY;
847 }
848
849 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
850                 size_t, msg_len, unsigned int, msg_prio,
851                 const struct timespec __user *, u_abs_timeout)
852 {
853         struct file *filp;
854         struct inode *inode;
855         struct ext_wait_queue wait;
856         struct ext_wait_queue *receiver;
857         struct msg_msg *msg_ptr;
858         struct mqueue_inode_info *info;
859         struct timespec ts, *p = NULL;
860         long timeout;
861         int ret;
862
863         if (u_abs_timeout) {
864                 if (copy_from_user(&ts, u_abs_timeout, 
865                                         sizeof(struct timespec)))
866                         return -EFAULT;
867                 p = &ts;
868         }
869
870         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
871                 return -EINVAL;
872
873         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
874         timeout = prepare_timeout(p);
875
876         ret = -EBADF;
877         filp = fget(mqdes);
878         if (unlikely(!filp))
879                 goto out;
880
881         inode = filp->f_path.dentry->d_inode;
882         if (unlikely(filp->f_op != &mqueue_file_operations))
883                 goto out_fput;
884         info = MQUEUE_I(inode);
885         audit_inode(NULL, filp->f_path.dentry);
886
887         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
888                 goto out_fput;
889
890         if (unlikely(msg_len > info->attr.mq_msgsize)) {
891                 ret = -EMSGSIZE;
892                 goto out_fput;
893         }
894
895         /* First try to allocate memory, before doing anything with
896          * existing queues. */
897         msg_ptr = load_msg(u_msg_ptr, msg_len);
898         if (IS_ERR(msg_ptr)) {
899                 ret = PTR_ERR(msg_ptr);
900                 goto out_fput;
901         }
902         msg_ptr->m_ts = msg_len;
903         msg_ptr->m_type = msg_prio;
904
905         spin_lock(&info->lock);
906
907         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
908                 if (filp->f_flags & O_NONBLOCK) {
909                         spin_unlock(&info->lock);
910                         ret = -EAGAIN;
911                 } else if (unlikely(timeout < 0)) {
912                         spin_unlock(&info->lock);
913                         ret = timeout;
914                 } else {
915                         wait.task = current;
916                         wait.msg = (void *) msg_ptr;
917                         wait.state = STATE_NONE;
918                         ret = wq_sleep(info, SEND, timeout, &wait);
919                 }
920                 if (ret < 0)
921                         free_msg(msg_ptr);
922         } else {
923                 receiver = wq_get_first_waiter(info, RECV);
924                 if (receiver) {
925                         pipelined_send(info, msg_ptr, receiver);
926                 } else {
927                         /* adds message to the queue */
928                         msg_insert(msg_ptr, info);
929                         __do_notify(info);
930                 }
931                 inode->i_atime = inode->i_mtime = inode->i_ctime =
932                                 CURRENT_TIME;
933                 spin_unlock(&info->lock);
934                 ret = 0;
935         }
936 out_fput:
937         fput(filp);
938 out:
939         return ret;
940 }
941
942 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
943                 size_t, msg_len, unsigned int __user *, u_msg_prio,
944                 const struct timespec __user *, u_abs_timeout)
945 {
946         long timeout;
947         ssize_t ret;
948         struct msg_msg *msg_ptr;
949         struct file *filp;
950         struct inode *inode;
951         struct mqueue_inode_info *info;
952         struct ext_wait_queue wait;
953         struct timespec ts, *p = NULL;
954
955         if (u_abs_timeout) {
956                 if (copy_from_user(&ts, u_abs_timeout, 
957                                         sizeof(struct timespec)))
958                         return -EFAULT;
959                 p = &ts;
960         }
961
962         audit_mq_sendrecv(mqdes, msg_len, 0, p);
963         timeout = prepare_timeout(p);
964
965         ret = -EBADF;
966         filp = fget(mqdes);
967         if (unlikely(!filp))
968                 goto out;
969
970         inode = filp->f_path.dentry->d_inode;
971         if (unlikely(filp->f_op != &mqueue_file_operations))
972                 goto out_fput;
973         info = MQUEUE_I(inode);
974         audit_inode(NULL, filp->f_path.dentry);
975
976         if (unlikely(!(filp->f_mode & FMODE_READ)))
977                 goto out_fput;
978
979         /* checks if buffer is big enough */
980         if (unlikely(msg_len < info->attr.mq_msgsize)) {
981                 ret = -EMSGSIZE;
982                 goto out_fput;
983         }
984
985         spin_lock(&info->lock);
986         if (info->attr.mq_curmsgs == 0) {
987                 if (filp->f_flags & O_NONBLOCK) {
988                         spin_unlock(&info->lock);
989                         ret = -EAGAIN;
990                         msg_ptr = NULL;
991                 } else if (unlikely(timeout < 0)) {
992                         spin_unlock(&info->lock);
993                         ret = timeout;
994                         msg_ptr = NULL;
995                 } else {
996                         wait.task = current;
997                         wait.state = STATE_NONE;
998                         ret = wq_sleep(info, RECV, timeout, &wait);
999                         msg_ptr = wait.msg;
1000                 }
1001         } else {
1002                 msg_ptr = msg_get(info);
1003
1004                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1005                                 CURRENT_TIME;
1006
1007                 /* There is now free space in queue. */
1008                 pipelined_receive(info);
1009                 spin_unlock(&info->lock);
1010                 ret = 0;
1011         }
1012         if (ret == 0) {
1013                 ret = msg_ptr->m_ts;
1014
1015                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1016                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1017                         ret = -EFAULT;
1018                 }
1019                 free_msg(msg_ptr);
1020         }
1021 out_fput:
1022         fput(filp);
1023 out:
1024         return ret;
1025 }
1026
1027 /*
1028  * Notes: the case when user wants us to deregister (with NULL as pointer)
1029  * and he isn't currently owner of notification, will be silently discarded.
1030  * It isn't explicitly defined in the POSIX.
1031  */
1032 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1033                 const struct sigevent __user *, u_notification)
1034 {
1035         int ret;
1036         struct file *filp;
1037         struct sock *sock;
1038         struct inode *inode;
1039         struct sigevent notification;
1040         struct mqueue_inode_info *info;
1041         struct sk_buff *nc;
1042
1043         if (u_notification) {
1044                 if (copy_from_user(&notification, u_notification,
1045                                         sizeof(struct sigevent)))
1046                         return -EFAULT;
1047         }
1048
1049         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1050
1051         nc = NULL;
1052         sock = NULL;
1053         if (u_notification != NULL) {
1054                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1055                              notification.sigev_notify != SIGEV_SIGNAL &&
1056                              notification.sigev_notify != SIGEV_THREAD))
1057                         return -EINVAL;
1058                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1059                         !valid_signal(notification.sigev_signo)) {
1060                         return -EINVAL;
1061                 }
1062                 if (notification.sigev_notify == SIGEV_THREAD) {
1063                         long timeo;
1064
1065                         /* create the notify skb */
1066                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1067                         ret = -ENOMEM;
1068                         if (!nc)
1069                                 goto out;
1070                         ret = -EFAULT;
1071                         if (copy_from_user(nc->data,
1072                                         notification.sigev_value.sival_ptr,
1073                                         NOTIFY_COOKIE_LEN)) {
1074                                 goto out;
1075                         }
1076
1077                         /* TODO: add a header? */
1078                         skb_put(nc, NOTIFY_COOKIE_LEN);
1079                         /* and attach it to the socket */
1080 retry:
1081                         filp = fget(notification.sigev_signo);
1082                         ret = -EBADF;
1083                         if (!filp)
1084                                 goto out;
1085                         sock = netlink_getsockbyfilp(filp);
1086                         fput(filp);
1087                         if (IS_ERR(sock)) {
1088                                 ret = PTR_ERR(sock);
1089                                 sock = NULL;
1090                                 goto out;
1091                         }
1092
1093                         timeo = MAX_SCHEDULE_TIMEOUT;
1094                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1095                         if (ret == 1)
1096                                 goto retry;
1097                         if (ret) {
1098                                 sock = NULL;
1099                                 nc = NULL;
1100                                 goto out;
1101                         }
1102                 }
1103         }
1104
1105         ret = -EBADF;
1106         filp = fget(mqdes);
1107         if (!filp)
1108                 goto out;
1109
1110         inode = filp->f_path.dentry->d_inode;
1111         if (unlikely(filp->f_op != &mqueue_file_operations))
1112                 goto out_fput;
1113         info = MQUEUE_I(inode);
1114
1115         ret = 0;
1116         spin_lock(&info->lock);
1117         if (u_notification == NULL) {
1118                 if (info->notify_owner == task_tgid(current)) {
1119                         remove_notification(info);
1120                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1121                 }
1122         } else if (info->notify_owner != NULL) {
1123                 ret = -EBUSY;
1124         } else {
1125                 switch (notification.sigev_notify) {
1126                 case SIGEV_NONE:
1127                         info->notify.sigev_notify = SIGEV_NONE;
1128                         break;
1129                 case SIGEV_THREAD:
1130                         info->notify_sock = sock;
1131                         info->notify_cookie = nc;
1132                         sock = NULL;
1133                         nc = NULL;
1134                         info->notify.sigev_notify = SIGEV_THREAD;
1135                         break;
1136                 case SIGEV_SIGNAL:
1137                         info->notify.sigev_signo = notification.sigev_signo;
1138                         info->notify.sigev_value = notification.sigev_value;
1139                         info->notify.sigev_notify = SIGEV_SIGNAL;
1140                         break;
1141                 }
1142
1143                 info->notify_owner = get_pid(task_tgid(current));
1144                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1145         }
1146         spin_unlock(&info->lock);
1147 out_fput:
1148         fput(filp);
1149 out:
1150         if (sock) {
1151                 netlink_detachskb(sock, nc);
1152         } else if (nc) {
1153                 dev_kfree_skb(nc);
1154         }
1155         return ret;
1156 }
1157
1158 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1159                 const struct mq_attr __user *, u_mqstat,
1160                 struct mq_attr __user *, u_omqstat)
1161 {
1162         int ret;
1163         struct mq_attr mqstat, omqstat;
1164         struct file *filp;
1165         struct inode *inode;
1166         struct mqueue_inode_info *info;
1167
1168         if (u_mqstat != NULL) {
1169                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1170                         return -EFAULT;
1171                 if (mqstat.mq_flags & (~O_NONBLOCK))
1172                         return -EINVAL;
1173         }
1174
1175         ret = -EBADF;
1176         filp = fget(mqdes);
1177         if (!filp)
1178                 goto out;
1179
1180         inode = filp->f_path.dentry->d_inode;
1181         if (unlikely(filp->f_op != &mqueue_file_operations))
1182                 goto out_fput;
1183         info = MQUEUE_I(inode);
1184
1185         spin_lock(&info->lock);
1186
1187         omqstat = info->attr;
1188         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1189         if (u_mqstat) {
1190                 audit_mq_getsetattr(mqdes, &mqstat);
1191                 spin_lock(&filp->f_lock);
1192                 if (mqstat.mq_flags & O_NONBLOCK)
1193                         filp->f_flags |= O_NONBLOCK;
1194                 else
1195                         filp->f_flags &= ~O_NONBLOCK;
1196                 spin_unlock(&filp->f_lock);
1197
1198                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1199         }
1200
1201         spin_unlock(&info->lock);
1202
1203         ret = 0;
1204         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1205                                                 sizeof(struct mq_attr)))
1206                 ret = -EFAULT;
1207
1208 out_fput:
1209         fput(filp);
1210 out:
1211         return ret;
1212 }
1213
1214 static const struct inode_operations mqueue_dir_inode_operations = {
1215         .lookup = simple_lookup,
1216         .create = mqueue_create,
1217         .unlink = mqueue_unlink,
1218 };
1219
1220 static const struct file_operations mqueue_file_operations = {
1221         .flush = mqueue_flush_file,
1222         .poll = mqueue_poll_file,
1223         .read = mqueue_read_file,
1224 };
1225
1226 static const struct super_operations mqueue_super_ops = {
1227         .alloc_inode = mqueue_alloc_inode,
1228         .destroy_inode = mqueue_destroy_inode,
1229         .statfs = simple_statfs,
1230         .delete_inode = mqueue_delete_inode,
1231         .drop_inode = generic_delete_inode,
1232 };
1233
1234 static struct file_system_type mqueue_fs_type = {
1235         .name = "mqueue",
1236         .get_sb = mqueue_get_sb,
1237         .kill_sb = kill_litter_super,
1238 };
1239
1240 int mq_init_ns(struct ipc_namespace *ns)
1241 {
1242         ns->mq_queues_count  = 0;
1243         ns->mq_queues_max    = DFLT_QUEUESMAX;
1244         ns->mq_msg_max       = DFLT_MSGMAX;
1245         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1246
1247         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1248         if (IS_ERR(ns->mq_mnt)) {
1249                 int err = PTR_ERR(ns->mq_mnt);
1250                 ns->mq_mnt = NULL;
1251                 return err;
1252         }
1253         return 0;
1254 }
1255
1256 void mq_clear_sbinfo(struct ipc_namespace *ns)
1257 {
1258         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1259 }
1260
1261 void mq_put_mnt(struct ipc_namespace *ns)
1262 {
1263         mntput(ns->mq_mnt);
1264 }
1265
1266 static int __init init_mqueue_fs(void)
1267 {
1268         int error;
1269
1270         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1271                                 sizeof(struct mqueue_inode_info), 0,
1272                                 SLAB_HWCACHE_ALIGN, init_once);
1273         if (mqueue_inode_cachep == NULL)
1274                 return -ENOMEM;
1275
1276         /* ignore failues - they are not fatal */
1277         mq_sysctl_table = mq_register_sysctl_table();
1278
1279         error = register_filesystem(&mqueue_fs_type);
1280         if (error)
1281                 goto out_sysctl;
1282
1283         spin_lock_init(&mq_lock);
1284
1285         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1286         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1287                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1288                 goto out_filesystem;
1289         }
1290
1291         return 0;
1292
1293 out_filesystem:
1294         unregister_filesystem(&mqueue_fs_type);
1295 out_sysctl:
1296         if (mq_sysctl_table)
1297                 unregister_sysctl_table(mq_sysctl_table);
1298         kmem_cache_destroy(mqueue_inode_cachep);
1299         return error;
1300 }
1301
1302 __initcall(init_mqueue_fs);