[PATCH] r/o bind mounts: elevate write count for rmdir and unlink.
[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
35 #include <net/sock.h>
36 #include "util.h"
37
38 #define MQUEUE_MAGIC    0x19800202
39 #define DIRENT_SIZE     20
40 #define FILENT_SIZE     80
41
42 #define SEND            0
43 #define RECV            1
44
45 #define STATE_NONE      0
46 #define STATE_PENDING   1
47 #define STATE_READY     2
48
49 /* default values */
50 #define DFLT_QUEUESMAX  256     /* max number of message queues */
51 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
52 #define HARD_MSGMAX     (131072/sizeof(void*))
53 #define DFLT_MSGSIZEMAX 8192    /* max message size */
54
55
56 struct ext_wait_queue {         /* queue of sleeping tasks */
57         struct task_struct *task;
58         struct list_head list;
59         struct msg_msg *msg;    /* ptr of loaded message */
60         int state;              /* one of STATE_* values */
61 };
62
63 struct mqueue_inode_info {
64         spinlock_t lock;
65         struct inode vfs_inode;
66         wait_queue_head_t wait_q;
67
68         struct msg_msg **messages;
69         struct mq_attr attr;
70
71         struct sigevent notify;
72         struct pid* notify_owner;
73         struct user_struct *user;       /* user who created, for accounting */
74         struct sock *notify_sock;
75         struct sk_buff *notify_cookie;
76
77         /* for tasks waiting for free space and messages, respectively */
78         struct ext_wait_queue e_wait_q[2];
79
80         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
81 };
82
83 static const struct inode_operations mqueue_dir_inode_operations;
84 static const struct file_operations mqueue_file_operations;
85 static struct super_operations mqueue_super_ops;
86 static void remove_notification(struct mqueue_inode_info *info);
87
88 static spinlock_t mq_lock;
89 static struct kmem_cache *mqueue_inode_cachep;
90 static struct vfsmount *mqueue_mnt;
91
92 static unsigned int queues_count;
93 static unsigned int queues_max  = DFLT_QUEUESMAX;
94 static unsigned int msg_max     = DFLT_MSGMAX;
95 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
96
97 static struct ctl_table_header * mq_sysctl_table;
98
99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100 {
101         return container_of(inode, struct mqueue_inode_info, vfs_inode);
102 }
103
104 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
105                                                         struct mq_attr *attr)
106 {
107         struct inode *inode;
108
109         inode = new_inode(sb);
110         if (inode) {
111                 inode->i_mode = mode;
112                 inode->i_uid = current->fsuid;
113                 inode->i_gid = current->fsgid;
114                 inode->i_blocks = 0;
115                 inode->i_mtime = inode->i_ctime = inode->i_atime =
116                                 CURRENT_TIME;
117
118                 if (S_ISREG(mode)) {
119                         struct mqueue_inode_info *info;
120                         struct task_struct *p = current;
121                         struct user_struct *u = p->user;
122                         unsigned long mq_bytes, mq_msg_tblsz;
123
124                         inode->i_fop = &mqueue_file_operations;
125                         inode->i_size = FILENT_SIZE;
126                         /* mqueue specific info */
127                         info = MQUEUE_I(inode);
128                         spin_lock_init(&info->lock);
129                         init_waitqueue_head(&info->wait_q);
130                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
131                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
132                         info->messages = NULL;
133                         info->notify_owner = NULL;
134                         info->qsize = 0;
135                         info->user = NULL;      /* set when all is ok */
136                         memset(&info->attr, 0, sizeof(info->attr));
137                         info->attr.mq_maxmsg = DFLT_MSGMAX;
138                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
139                         if (attr) {
140                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
141                                 info->attr.mq_msgsize = attr->mq_msgsize;
142                         }
143                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
144                         mq_bytes = (mq_msg_tblsz +
145                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
146
147                         spin_lock(&mq_lock);
148                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
149                             u->mq_bytes + mq_bytes >
150                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
151                                 spin_unlock(&mq_lock);
152                                 goto out_inode;
153                         }
154                         u->mq_bytes += mq_bytes;
155                         spin_unlock(&mq_lock);
156
157                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
158                         if (!info->messages) {
159                                 spin_lock(&mq_lock);
160                                 u->mq_bytes -= mq_bytes;
161                                 spin_unlock(&mq_lock);
162                                 goto out_inode;
163                         }
164                         /* all is ok */
165                         info->user = get_uid(u);
166                 } else if (S_ISDIR(mode)) {
167                         inc_nlink(inode);
168                         /* Some things misbehave if size == 0 on a directory */
169                         inode->i_size = 2 * DIRENT_SIZE;
170                         inode->i_op = &mqueue_dir_inode_operations;
171                         inode->i_fop = &simple_dir_operations;
172                 }
173         }
174         return inode;
175 out_inode:
176         make_bad_inode(inode);
177         iput(inode);
178         return NULL;
179 }
180
181 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
182 {
183         struct inode *inode;
184
185         sb->s_blocksize = PAGE_CACHE_SIZE;
186         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
187         sb->s_magic = MQUEUE_MAGIC;
188         sb->s_op = &mqueue_super_ops;
189
190         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
191         if (!inode)
192                 return -ENOMEM;
193
194         sb->s_root = d_alloc_root(inode);
195         if (!sb->s_root) {
196                 iput(inode);
197                 return -ENOMEM;
198         }
199
200         return 0;
201 }
202
203 static int mqueue_get_sb(struct file_system_type *fs_type,
204                          int flags, const char *dev_name,
205                          void *data, struct vfsmount *mnt)
206 {
207         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
208 }
209
210 static void init_once(struct kmem_cache *cachep, void *foo)
211 {
212         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
213
214         inode_init_once(&p->vfs_inode);
215 }
216
217 static struct inode *mqueue_alloc_inode(struct super_block *sb)
218 {
219         struct mqueue_inode_info *ei;
220
221         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
222         if (!ei)
223                 return NULL;
224         return &ei->vfs_inode;
225 }
226
227 static void mqueue_destroy_inode(struct inode *inode)
228 {
229         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
230 }
231
232 static void mqueue_delete_inode(struct inode *inode)
233 {
234         struct mqueue_inode_info *info;
235         struct user_struct *user;
236         unsigned long mq_bytes;
237         int i;
238
239         if (S_ISDIR(inode->i_mode)) {
240                 clear_inode(inode);
241                 return;
242         }
243         info = MQUEUE_I(inode);
244         spin_lock(&info->lock);
245         for (i = 0; i < info->attr.mq_curmsgs; i++)
246                 free_msg(info->messages[i]);
247         kfree(info->messages);
248         spin_unlock(&info->lock);
249
250         clear_inode(inode);
251
252         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
253                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
254         user = info->user;
255         if (user) {
256                 spin_lock(&mq_lock);
257                 user->mq_bytes -= mq_bytes;
258                 queues_count--;
259                 spin_unlock(&mq_lock);
260                 free_uid(user);
261         }
262 }
263
264 static int mqueue_create(struct inode *dir, struct dentry *dentry,
265                                 int mode, struct nameidata *nd)
266 {
267         struct inode *inode;
268         struct mq_attr *attr = dentry->d_fsdata;
269         int error;
270
271         spin_lock(&mq_lock);
272         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
273                 error = -ENOSPC;
274                 goto out_lock;
275         }
276         queues_count++;
277         spin_unlock(&mq_lock);
278
279         inode = mqueue_get_inode(dir->i_sb, mode, attr);
280         if (!inode) {
281                 error = -ENOMEM;
282                 spin_lock(&mq_lock);
283                 queues_count--;
284                 goto out_lock;
285         }
286
287         dir->i_size += DIRENT_SIZE;
288         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
289
290         d_instantiate(dentry, inode);
291         dget(dentry);
292         return 0;
293 out_lock:
294         spin_unlock(&mq_lock);
295         return error;
296 }
297
298 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
299 {
300         struct inode *inode = dentry->d_inode;
301
302         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
303         dir->i_size -= DIRENT_SIZE;
304         drop_nlink(inode);
305         dput(dentry);
306         return 0;
307 }
308
309 /*
310 *       This is routine for system read from queue file.
311 *       To avoid mess with doing here some sort of mq_receive we allow
312 *       to read only queue size & notification info (the only values
313 *       that are interesting from user point of view and aren't accessible
314 *       through std routines)
315 */
316 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
317                                 size_t count, loff_t * off)
318 {
319         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
320         char buffer[FILENT_SIZE];
321         size_t slen;
322         loff_t o;
323
324         if (!count)
325                 return 0;
326
327         spin_lock(&info->lock);
328         snprintf(buffer, sizeof(buffer),
329                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
330                         info->qsize,
331                         info->notify_owner ? info->notify.sigev_notify : 0,
332                         (info->notify_owner &&
333                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
334                                 info->notify.sigev_signo : 0,
335                         pid_vnr(info->notify_owner));
336         spin_unlock(&info->lock);
337         buffer[sizeof(buffer)-1] = '\0';
338         slen = strlen(buffer)+1;
339
340         o = *off;
341         if (o > slen)
342                 return 0;
343
344         if (o + count > slen)
345                 count = slen - o;
346
347         if (copy_to_user(u_data, buffer + o, count))
348                 return -EFAULT;
349
350         *off = o + count;
351         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
352         return count;
353 }
354
355 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
356 {
357         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
358
359         spin_lock(&info->lock);
360         if (task_tgid(current) == info->notify_owner)
361                 remove_notification(info);
362
363         spin_unlock(&info->lock);
364         return 0;
365 }
366
367 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
368 {
369         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
370         int retval = 0;
371
372         poll_wait(filp, &info->wait_q, poll_tab);
373
374         spin_lock(&info->lock);
375         if (info->attr.mq_curmsgs)
376                 retval = POLLIN | POLLRDNORM;
377
378         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
379                 retval |= POLLOUT | POLLWRNORM;
380         spin_unlock(&info->lock);
381
382         return retval;
383 }
384
385 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
386 static void wq_add(struct mqueue_inode_info *info, int sr,
387                         struct ext_wait_queue *ewp)
388 {
389         struct ext_wait_queue *walk;
390
391         ewp->task = current;
392
393         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
394                 if (walk->task->static_prio <= current->static_prio) {
395                         list_add_tail(&ewp->list, &walk->list);
396                         return;
397                 }
398         }
399         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
400 }
401
402 /*
403  * Puts current task to sleep. Caller must hold queue lock. After return
404  * lock isn't held.
405  * sr: SEND or RECV
406  */
407 static int wq_sleep(struct mqueue_inode_info *info, int sr,
408                         long timeout, struct ext_wait_queue *ewp)
409 {
410         int retval;
411         signed long time;
412
413         wq_add(info, sr, ewp);
414
415         for (;;) {
416                 set_current_state(TASK_INTERRUPTIBLE);
417
418                 spin_unlock(&info->lock);
419                 time = schedule_timeout(timeout);
420
421                 while (ewp->state == STATE_PENDING)
422                         cpu_relax();
423
424                 if (ewp->state == STATE_READY) {
425                         retval = 0;
426                         goto out;
427                 }
428                 spin_lock(&info->lock);
429                 if (ewp->state == STATE_READY) {
430                         retval = 0;
431                         goto out_unlock;
432                 }
433                 if (signal_pending(current)) {
434                         retval = -ERESTARTSYS;
435                         break;
436                 }
437                 if (time == 0) {
438                         retval = -ETIMEDOUT;
439                         break;
440                 }
441         }
442         list_del(&ewp->list);
443 out_unlock:
444         spin_unlock(&info->lock);
445 out:
446         return retval;
447 }
448
449 /*
450  * Returns waiting task that should be serviced first or NULL if none exists
451  */
452 static struct ext_wait_queue *wq_get_first_waiter(
453                 struct mqueue_inode_info *info, int sr)
454 {
455         struct list_head *ptr;
456
457         ptr = info->e_wait_q[sr].list.prev;
458         if (ptr == &info->e_wait_q[sr].list)
459                 return NULL;
460         return list_entry(ptr, struct ext_wait_queue, list);
461 }
462
463 /* Auxiliary functions to manipulate messages' list */
464 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
465 {
466         int k;
467
468         k = info->attr.mq_curmsgs - 1;
469         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
470                 info->messages[k + 1] = info->messages[k];
471                 k--;
472         }
473         info->attr.mq_curmsgs++;
474         info->qsize += ptr->m_ts;
475         info->messages[k + 1] = ptr;
476 }
477
478 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
479 {
480         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
481         return info->messages[info->attr.mq_curmsgs];
482 }
483
484 static inline void set_cookie(struct sk_buff *skb, char code)
485 {
486         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
487 }
488
489 /*
490  * The next function is only to split too long sys_mq_timedsend
491  */
492 static void __do_notify(struct mqueue_inode_info *info)
493 {
494         /* notification
495          * invoked when there is registered process and there isn't process
496          * waiting synchronously for message AND state of queue changed from
497          * empty to not empty. Here we are sure that no one is waiting
498          * synchronously. */
499         if (info->notify_owner &&
500             info->attr.mq_curmsgs == 1) {
501                 struct siginfo sig_i;
502                 switch (info->notify.sigev_notify) {
503                 case SIGEV_NONE:
504                         break;
505                 case SIGEV_SIGNAL:
506                         /* sends signal */
507
508                         sig_i.si_signo = info->notify.sigev_signo;
509                         sig_i.si_errno = 0;
510                         sig_i.si_code = SI_MESGQ;
511                         sig_i.si_value = info->notify.sigev_value;
512                         sig_i.si_pid = task_tgid_vnr(current);
513                         sig_i.si_uid = current->uid;
514
515                         kill_pid_info(info->notify.sigev_signo,
516                                       &sig_i, info->notify_owner);
517                         break;
518                 case SIGEV_THREAD:
519                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
520                         netlink_sendskb(info->notify_sock, info->notify_cookie);
521                         break;
522                 }
523                 /* after notification unregisters process */
524                 put_pid(info->notify_owner);
525                 info->notify_owner = NULL;
526         }
527         wake_up(&info->wait_q);
528 }
529
530 static long prepare_timeout(const struct timespec __user *u_arg)
531 {
532         struct timespec ts, nowts;
533         long timeout;
534
535         if (u_arg) {
536                 if (unlikely(copy_from_user(&ts, u_arg,
537                                         sizeof(struct timespec))))
538                         return -EFAULT;
539
540                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
541                         || ts.tv_nsec >= NSEC_PER_SEC))
542                         return -EINVAL;
543                 nowts = CURRENT_TIME;
544                 /* first subtract as jiffies can't be too big */
545                 ts.tv_sec -= nowts.tv_sec;
546                 if (ts.tv_nsec < nowts.tv_nsec) {
547                         ts.tv_nsec += NSEC_PER_SEC;
548                         ts.tv_sec--;
549                 }
550                 ts.tv_nsec -= nowts.tv_nsec;
551                 if (ts.tv_sec < 0)
552                         return 0;
553
554                 timeout = timespec_to_jiffies(&ts) + 1;
555         } else
556                 return MAX_SCHEDULE_TIMEOUT;
557
558         return timeout;
559 }
560
561 static void remove_notification(struct mqueue_inode_info *info)
562 {
563         if (info->notify_owner != NULL &&
564             info->notify.sigev_notify == SIGEV_THREAD) {
565                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
566                 netlink_sendskb(info->notify_sock, info->notify_cookie);
567         }
568         put_pid(info->notify_owner);
569         info->notify_owner = NULL;
570 }
571
572 static int mq_attr_ok(struct mq_attr *attr)
573 {
574         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
575                 return 0;
576         if (capable(CAP_SYS_RESOURCE)) {
577                 if (attr->mq_maxmsg > HARD_MSGMAX)
578                         return 0;
579         } else {
580                 if (attr->mq_maxmsg > msg_max ||
581                                 attr->mq_msgsize > msgsize_max)
582                         return 0;
583         }
584         /* check for overflow */
585         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
586                 return 0;
587         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
588             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
589             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
590                 return 0;
591         return 1;
592 }
593
594 /*
595  * Invoked when creating a new queue via sys_mq_open
596  */
597 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
598                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
599 {
600         struct mq_attr attr;
601         int ret;
602
603         if (u_attr) {
604                 ret = -EFAULT;
605                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
606                         goto out;
607                 ret = -EINVAL;
608                 if (!mq_attr_ok(&attr))
609                         goto out;
610                 /* store for use during create */
611                 dentry->d_fsdata = &attr;
612         }
613
614         mode &= ~current->fs->umask;
615         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
616         dentry->d_fsdata = NULL;
617         if (ret)
618                 goto out;
619
620         return dentry_open(dentry, mqueue_mnt, oflag);
621
622 out:
623         dput(dentry);
624         mntput(mqueue_mnt);
625         return ERR_PTR(ret);
626 }
627
628 /* Opens existing queue */
629 static struct file *do_open(struct dentry *dentry, int oflag)
630 {
631 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
632                                         MAY_READ | MAY_WRITE };
633
634         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
635                 dput(dentry);
636                 mntput(mqueue_mnt);
637                 return ERR_PTR(-EINVAL);
638         }
639
640         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
641                 dput(dentry);
642                 mntput(mqueue_mnt);
643                 return ERR_PTR(-EACCES);
644         }
645
646         return dentry_open(dentry, mqueue_mnt, oflag);
647 }
648
649 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
650                                 struct mq_attr __user *u_attr)
651 {
652         struct dentry *dentry;
653         struct file *filp;
654         char *name;
655         int fd, error;
656
657         error = audit_mq_open(oflag, mode, u_attr);
658         if (error != 0)
659                 return error;
660
661         if (IS_ERR(name = getname(u_name)))
662                 return PTR_ERR(name);
663
664         fd = get_unused_fd();
665         if (fd < 0)
666                 goto out_putname;
667
668         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
669         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
670         if (IS_ERR(dentry)) {
671                 error = PTR_ERR(dentry);
672                 goto out_err;
673         }
674         mntget(mqueue_mnt);
675
676         if (oflag & O_CREAT) {
677                 if (dentry->d_inode) {  /* entry already exists */
678                         audit_inode(name, dentry);
679                         error = -EEXIST;
680                         if (oflag & O_EXCL)
681                                 goto out;
682                         filp = do_open(dentry, oflag);
683                 } else {
684                         filp = do_create(mqueue_mnt->mnt_root, dentry,
685                                                 oflag, mode, u_attr);
686                 }
687         } else {
688                 error = -ENOENT;
689                 if (!dentry->d_inode)
690                         goto out;
691                 audit_inode(name, dentry);
692                 filp = do_open(dentry, oflag);
693         }
694
695         if (IS_ERR(filp)) {
696                 error = PTR_ERR(filp);
697                 goto out_putfd;
698         }
699
700         set_close_on_exec(fd, 1);
701         fd_install(fd, filp);
702         goto out_upsem;
703
704 out:
705         dput(dentry);
706         mntput(mqueue_mnt);
707 out_putfd:
708         put_unused_fd(fd);
709 out_err:
710         fd = error;
711 out_upsem:
712         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
713 out_putname:
714         putname(name);
715         return fd;
716 }
717
718 asmlinkage long sys_mq_unlink(const char __user *u_name)
719 {
720         int err;
721         char *name;
722         struct dentry *dentry;
723         struct inode *inode = NULL;
724
725         name = getname(u_name);
726         if (IS_ERR(name))
727                 return PTR_ERR(name);
728
729         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
730                         I_MUTEX_PARENT);
731         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
732         if (IS_ERR(dentry)) {
733                 err = PTR_ERR(dentry);
734                 goto out_unlock;
735         }
736
737         if (!dentry->d_inode) {
738                 err = -ENOENT;
739                 goto out_err;
740         }
741
742         inode = dentry->d_inode;
743         if (inode)
744                 atomic_inc(&inode->i_count);
745         err = mnt_want_write(mqueue_mnt);
746         if (err)
747                 goto out_err;
748         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
749         mnt_drop_write(mqueue_mnt);
750 out_err:
751         dput(dentry);
752
753 out_unlock:
754         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
755         putname(name);
756         if (inode)
757                 iput(inode);
758
759         return err;
760 }
761
762 /* Pipelined send and receive functions.
763  *
764  * If a receiver finds no waiting message, then it registers itself in the
765  * list of waiting receivers. A sender checks that list before adding the new
766  * message into the message array. If there is a waiting receiver, then it
767  * bypasses the message array and directly hands the message over to the
768  * receiver.
769  * The receiver accepts the message and returns without grabbing the queue
770  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
771  * are necessary. The same algorithm is used for sysv semaphores, see
772  * ipc/sem.c for more details.
773  *
774  * The same algorithm is used for senders.
775  */
776
777 /* pipelined_send() - send a message directly to the task waiting in
778  * sys_mq_timedreceive() (without inserting message into a queue).
779  */
780 static inline void pipelined_send(struct mqueue_inode_info *info,
781                                   struct msg_msg *message,
782                                   struct ext_wait_queue *receiver)
783 {
784         receiver->msg = message;
785         list_del(&receiver->list);
786         receiver->state = STATE_PENDING;
787         wake_up_process(receiver->task);
788         smp_wmb();
789         receiver->state = STATE_READY;
790 }
791
792 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
793  * gets its message and put to the queue (we have one free place for sure). */
794 static inline void pipelined_receive(struct mqueue_inode_info *info)
795 {
796         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
797
798         if (!sender) {
799                 /* for poll */
800                 wake_up_interruptible(&info->wait_q);
801                 return;
802         }
803         msg_insert(sender->msg, info);
804         list_del(&sender->list);
805         sender->state = STATE_PENDING;
806         wake_up_process(sender->task);
807         smp_wmb();
808         sender->state = STATE_READY;
809 }
810
811 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
812         size_t msg_len, unsigned int msg_prio,
813         const struct timespec __user *u_abs_timeout)
814 {
815         struct file *filp;
816         struct inode *inode;
817         struct ext_wait_queue wait;
818         struct ext_wait_queue *receiver;
819         struct msg_msg *msg_ptr;
820         struct mqueue_inode_info *info;
821         long timeout;
822         int ret;
823
824         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
825         if (ret != 0)
826                 return ret;
827
828         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
829                 return -EINVAL;
830
831         timeout = prepare_timeout(u_abs_timeout);
832
833         ret = -EBADF;
834         filp = fget(mqdes);
835         if (unlikely(!filp))
836                 goto out;
837
838         inode = filp->f_path.dentry->d_inode;
839         if (unlikely(filp->f_op != &mqueue_file_operations))
840                 goto out_fput;
841         info = MQUEUE_I(inode);
842         audit_inode(NULL, filp->f_path.dentry);
843
844         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
845                 goto out_fput;
846
847         if (unlikely(msg_len > info->attr.mq_msgsize)) {
848                 ret = -EMSGSIZE;
849                 goto out_fput;
850         }
851
852         /* First try to allocate memory, before doing anything with
853          * existing queues. */
854         msg_ptr = load_msg(u_msg_ptr, msg_len);
855         if (IS_ERR(msg_ptr)) {
856                 ret = PTR_ERR(msg_ptr);
857                 goto out_fput;
858         }
859         msg_ptr->m_ts = msg_len;
860         msg_ptr->m_type = msg_prio;
861
862         spin_lock(&info->lock);
863
864         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
865                 if (filp->f_flags & O_NONBLOCK) {
866                         spin_unlock(&info->lock);
867                         ret = -EAGAIN;
868                 } else if (unlikely(timeout < 0)) {
869                         spin_unlock(&info->lock);
870                         ret = timeout;
871                 } else {
872                         wait.task = current;
873                         wait.msg = (void *) msg_ptr;
874                         wait.state = STATE_NONE;
875                         ret = wq_sleep(info, SEND, timeout, &wait);
876                 }
877                 if (ret < 0)
878                         free_msg(msg_ptr);
879         } else {
880                 receiver = wq_get_first_waiter(info, RECV);
881                 if (receiver) {
882                         pipelined_send(info, msg_ptr, receiver);
883                 } else {
884                         /* adds message to the queue */
885                         msg_insert(msg_ptr, info);
886                         __do_notify(info);
887                 }
888                 inode->i_atime = inode->i_mtime = inode->i_ctime =
889                                 CURRENT_TIME;
890                 spin_unlock(&info->lock);
891                 ret = 0;
892         }
893 out_fput:
894         fput(filp);
895 out:
896         return ret;
897 }
898
899 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
900         size_t msg_len, unsigned int __user *u_msg_prio,
901         const struct timespec __user *u_abs_timeout)
902 {
903         long timeout;
904         ssize_t ret;
905         struct msg_msg *msg_ptr;
906         struct file *filp;
907         struct inode *inode;
908         struct mqueue_inode_info *info;
909         struct ext_wait_queue wait;
910
911         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
912         if (ret != 0)
913                 return ret;
914
915         timeout = prepare_timeout(u_abs_timeout);
916
917         ret = -EBADF;
918         filp = fget(mqdes);
919         if (unlikely(!filp))
920                 goto out;
921
922         inode = filp->f_path.dentry->d_inode;
923         if (unlikely(filp->f_op != &mqueue_file_operations))
924                 goto out_fput;
925         info = MQUEUE_I(inode);
926         audit_inode(NULL, filp->f_path.dentry);
927
928         if (unlikely(!(filp->f_mode & FMODE_READ)))
929                 goto out_fput;
930
931         /* checks if buffer is big enough */
932         if (unlikely(msg_len < info->attr.mq_msgsize)) {
933                 ret = -EMSGSIZE;
934                 goto out_fput;
935         }
936
937         spin_lock(&info->lock);
938         if (info->attr.mq_curmsgs == 0) {
939                 if (filp->f_flags & O_NONBLOCK) {
940                         spin_unlock(&info->lock);
941                         ret = -EAGAIN;
942                         msg_ptr = NULL;
943                 } else if (unlikely(timeout < 0)) {
944                         spin_unlock(&info->lock);
945                         ret = timeout;
946                         msg_ptr = NULL;
947                 } else {
948                         wait.task = current;
949                         wait.state = STATE_NONE;
950                         ret = wq_sleep(info, RECV, timeout, &wait);
951                         msg_ptr = wait.msg;
952                 }
953         } else {
954                 msg_ptr = msg_get(info);
955
956                 inode->i_atime = inode->i_mtime = inode->i_ctime =
957                                 CURRENT_TIME;
958
959                 /* There is now free space in queue. */
960                 pipelined_receive(info);
961                 spin_unlock(&info->lock);
962                 ret = 0;
963         }
964         if (ret == 0) {
965                 ret = msg_ptr->m_ts;
966
967                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
968                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
969                         ret = -EFAULT;
970                 }
971                 free_msg(msg_ptr);
972         }
973 out_fput:
974         fput(filp);
975 out:
976         return ret;
977 }
978
979 /*
980  * Notes: the case when user wants us to deregister (with NULL as pointer)
981  * and he isn't currently owner of notification, will be silently discarded.
982  * It isn't explicitly defined in the POSIX.
983  */
984 asmlinkage long sys_mq_notify(mqd_t mqdes,
985                                 const struct sigevent __user *u_notification)
986 {
987         int ret;
988         struct file *filp;
989         struct sock *sock;
990         struct inode *inode;
991         struct sigevent notification;
992         struct mqueue_inode_info *info;
993         struct sk_buff *nc;
994
995         ret = audit_mq_notify(mqdes, u_notification);
996         if (ret != 0)
997                 return ret;
998
999         nc = NULL;
1000         sock = NULL;
1001         if (u_notification != NULL) {
1002                 if (copy_from_user(&notification, u_notification,
1003                                         sizeof(struct sigevent)))
1004                         return -EFAULT;
1005
1006                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1007                              notification.sigev_notify != SIGEV_SIGNAL &&
1008                              notification.sigev_notify != SIGEV_THREAD))
1009                         return -EINVAL;
1010                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1011                         !valid_signal(notification.sigev_signo)) {
1012                         return -EINVAL;
1013                 }
1014                 if (notification.sigev_notify == SIGEV_THREAD) {
1015                         long timeo;
1016
1017                         /* create the notify skb */
1018                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1019                         ret = -ENOMEM;
1020                         if (!nc)
1021                                 goto out;
1022                         ret = -EFAULT;
1023                         if (copy_from_user(nc->data,
1024                                         notification.sigev_value.sival_ptr,
1025                                         NOTIFY_COOKIE_LEN)) {
1026                                 goto out;
1027                         }
1028
1029                         /* TODO: add a header? */
1030                         skb_put(nc, NOTIFY_COOKIE_LEN);
1031                         /* and attach it to the socket */
1032 retry:
1033                         filp = fget(notification.sigev_signo);
1034                         ret = -EBADF;
1035                         if (!filp)
1036                                 goto out;
1037                         sock = netlink_getsockbyfilp(filp);
1038                         fput(filp);
1039                         if (IS_ERR(sock)) {
1040                                 ret = PTR_ERR(sock);
1041                                 sock = NULL;
1042                                 goto out;
1043                         }
1044
1045                         timeo = MAX_SCHEDULE_TIMEOUT;
1046                         ret = netlink_attachskb(sock, nc, 0, &timeo, NULL);
1047                         if (ret == 1)
1048                                 goto retry;
1049                         if (ret) {
1050                                 sock = NULL;
1051                                 nc = NULL;
1052                                 goto out;
1053                         }
1054                 }
1055         }
1056
1057         ret = -EBADF;
1058         filp = fget(mqdes);
1059         if (!filp)
1060                 goto out;
1061
1062         inode = filp->f_path.dentry->d_inode;
1063         if (unlikely(filp->f_op != &mqueue_file_operations))
1064                 goto out_fput;
1065         info = MQUEUE_I(inode);
1066
1067         ret = 0;
1068         spin_lock(&info->lock);
1069         if (u_notification == NULL) {
1070                 if (info->notify_owner == task_tgid(current)) {
1071                         remove_notification(info);
1072                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1073                 }
1074         } else if (info->notify_owner != NULL) {
1075                 ret = -EBUSY;
1076         } else {
1077                 switch (notification.sigev_notify) {
1078                 case SIGEV_NONE:
1079                         info->notify.sigev_notify = SIGEV_NONE;
1080                         break;
1081                 case SIGEV_THREAD:
1082                         info->notify_sock = sock;
1083                         info->notify_cookie = nc;
1084                         sock = NULL;
1085                         nc = NULL;
1086                         info->notify.sigev_notify = SIGEV_THREAD;
1087                         break;
1088                 case SIGEV_SIGNAL:
1089                         info->notify.sigev_signo = notification.sigev_signo;
1090                         info->notify.sigev_value = notification.sigev_value;
1091                         info->notify.sigev_notify = SIGEV_SIGNAL;
1092                         break;
1093                 }
1094
1095                 info->notify_owner = get_pid(task_tgid(current));
1096                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1097         }
1098         spin_unlock(&info->lock);
1099 out_fput:
1100         fput(filp);
1101 out:
1102         if (sock) {
1103                 netlink_detachskb(sock, nc);
1104         } else if (nc) {
1105                 dev_kfree_skb(nc);
1106         }
1107         return ret;
1108 }
1109
1110 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1111                         const struct mq_attr __user *u_mqstat,
1112                         struct mq_attr __user *u_omqstat)
1113 {
1114         int ret;
1115         struct mq_attr mqstat, omqstat;
1116         struct file *filp;
1117         struct inode *inode;
1118         struct mqueue_inode_info *info;
1119
1120         if (u_mqstat != NULL) {
1121                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1122                         return -EFAULT;
1123                 if (mqstat.mq_flags & (~O_NONBLOCK))
1124                         return -EINVAL;
1125         }
1126
1127         ret = -EBADF;
1128         filp = fget(mqdes);
1129         if (!filp)
1130                 goto out;
1131
1132         inode = filp->f_path.dentry->d_inode;
1133         if (unlikely(filp->f_op != &mqueue_file_operations))
1134                 goto out_fput;
1135         info = MQUEUE_I(inode);
1136
1137         spin_lock(&info->lock);
1138
1139         omqstat = info->attr;
1140         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1141         if (u_mqstat) {
1142                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1143                 if (ret != 0) {
1144                         spin_unlock(&info->lock);
1145                         goto out_fput;
1146                 }
1147                 if (mqstat.mq_flags & O_NONBLOCK)
1148                         filp->f_flags |= O_NONBLOCK;
1149                 else
1150                         filp->f_flags &= ~O_NONBLOCK;
1151
1152                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1153         }
1154
1155         spin_unlock(&info->lock);
1156
1157         ret = 0;
1158         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1159                                                 sizeof(struct mq_attr)))
1160                 ret = -EFAULT;
1161
1162 out_fput:
1163         fput(filp);
1164 out:
1165         return ret;
1166 }
1167
1168 static const struct inode_operations mqueue_dir_inode_operations = {
1169         .lookup = simple_lookup,
1170         .create = mqueue_create,
1171         .unlink = mqueue_unlink,
1172 };
1173
1174 static const struct file_operations mqueue_file_operations = {
1175         .flush = mqueue_flush_file,
1176         .poll = mqueue_poll_file,
1177         .read = mqueue_read_file,
1178 };
1179
1180 static struct super_operations mqueue_super_ops = {
1181         .alloc_inode = mqueue_alloc_inode,
1182         .destroy_inode = mqueue_destroy_inode,
1183         .statfs = simple_statfs,
1184         .delete_inode = mqueue_delete_inode,
1185         .drop_inode = generic_delete_inode,
1186 };
1187
1188 static struct file_system_type mqueue_fs_type = {
1189         .name = "mqueue",
1190         .get_sb = mqueue_get_sb,
1191         .kill_sb = kill_litter_super,
1192 };
1193
1194 static int msg_max_limit_min = DFLT_MSGMAX;
1195 static int msg_max_limit_max = HARD_MSGMAX;
1196
1197 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1198 static int msg_maxsize_limit_max = INT_MAX;
1199
1200 static ctl_table mq_sysctls[] = {
1201         {
1202                 .procname       = "queues_max",
1203                 .data           = &queues_max,
1204                 .maxlen         = sizeof(int),
1205                 .mode           = 0644,
1206                 .proc_handler   = &proc_dointvec,
1207         },
1208         {
1209                 .procname       = "msg_max",
1210                 .data           = &msg_max,
1211                 .maxlen         = sizeof(int),
1212                 .mode           = 0644,
1213                 .proc_handler   = &proc_dointvec_minmax,
1214                 .extra1         = &msg_max_limit_min,
1215                 .extra2         = &msg_max_limit_max,
1216         },
1217         {
1218                 .procname       = "msgsize_max",
1219                 .data           = &msgsize_max,
1220                 .maxlen         = sizeof(int),
1221                 .mode           = 0644,
1222                 .proc_handler   = &proc_dointvec_minmax,
1223                 .extra1         = &msg_maxsize_limit_min,
1224                 .extra2         = &msg_maxsize_limit_max,
1225         },
1226         { .ctl_name = 0 }
1227 };
1228
1229 static ctl_table mq_sysctl_dir[] = {
1230         {
1231                 .procname       = "mqueue",
1232                 .mode           = 0555,
1233                 .child          = mq_sysctls,
1234         },
1235         { .ctl_name = 0 }
1236 };
1237
1238 static ctl_table mq_sysctl_root[] = {
1239         {
1240                 .ctl_name       = CTL_FS,
1241                 .procname       = "fs",
1242                 .mode           = 0555,
1243                 .child          = mq_sysctl_dir,
1244         },
1245         { .ctl_name = 0 }
1246 };
1247
1248 static int __init init_mqueue_fs(void)
1249 {
1250         int error;
1251
1252         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1253                                 sizeof(struct mqueue_inode_info), 0,
1254                                 SLAB_HWCACHE_ALIGN, init_once);
1255         if (mqueue_inode_cachep == NULL)
1256                 return -ENOMEM;
1257
1258         /* ignore failues - they are not fatal */
1259         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1260
1261         error = register_filesystem(&mqueue_fs_type);
1262         if (error)
1263                 goto out_sysctl;
1264
1265         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1266                 error = PTR_ERR(mqueue_mnt);
1267                 goto out_filesystem;
1268         }
1269
1270         /* internal initialization - not common for vfs */
1271         queues_count = 0;
1272         spin_lock_init(&mq_lock);
1273
1274         return 0;
1275
1276 out_filesystem:
1277         unregister_filesystem(&mqueue_fs_type);
1278 out_sysctl:
1279         if (mq_sysctl_table)
1280                 unregister_sysctl_table(mq_sysctl_table);
1281         kmem_cache_destroy(mqueue_inode_cachep);
1282         return error;
1283 }
1284
1285 __initcall(init_mqueue_fs);