ipc/sem.c: rewrite undo list locking
[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         struct file *result;
602         int ret;
603
604         if (u_attr) {
605                 ret = -EFAULT;
606                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
607                         goto out;
608                 ret = -EINVAL;
609                 if (!mq_attr_ok(&attr))
610                         goto out;
611                 /* store for use during create */
612                 dentry->d_fsdata = &attr;
613         }
614
615         mode &= ~current->fs->umask;
616         ret = mnt_want_write(mqueue_mnt);
617         if (ret)
618                 goto out;
619         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
620         dentry->d_fsdata = NULL;
621         if (ret)
622                 goto out_drop_write;
623
624         result = dentry_open(dentry, mqueue_mnt, oflag);
625         /*
626          * dentry_open() took a persistent mnt_want_write(),
627          * so we can now drop this one.
628          */
629         mnt_drop_write(mqueue_mnt);
630         return result;
631
632 out_drop_write:
633         mnt_drop_write(mqueue_mnt);
634 out:
635         dput(dentry);
636         mntput(mqueue_mnt);
637         return ERR_PTR(ret);
638 }
639
640 /* Opens existing queue */
641 static struct file *do_open(struct dentry *dentry, int oflag)
642 {
643 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
644                                         MAY_READ | MAY_WRITE };
645
646         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
647                 dput(dentry);
648                 mntput(mqueue_mnt);
649                 return ERR_PTR(-EINVAL);
650         }
651
652         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
653                 dput(dentry);
654                 mntput(mqueue_mnt);
655                 return ERR_PTR(-EACCES);
656         }
657
658         return dentry_open(dentry, mqueue_mnt, oflag);
659 }
660
661 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
662                                 struct mq_attr __user *u_attr)
663 {
664         struct dentry *dentry;
665         struct file *filp;
666         char *name;
667         int fd, error;
668
669         error = audit_mq_open(oflag, mode, u_attr);
670         if (error != 0)
671                 return error;
672
673         if (IS_ERR(name = getname(u_name)))
674                 return PTR_ERR(name);
675
676         fd = get_unused_fd_flags(O_CLOEXEC);
677         if (fd < 0)
678                 goto out_putname;
679
680         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
681         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
682         if (IS_ERR(dentry)) {
683                 error = PTR_ERR(dentry);
684                 goto out_err;
685         }
686         mntget(mqueue_mnt);
687
688         if (oflag & O_CREAT) {
689                 if (dentry->d_inode) {  /* entry already exists */
690                         audit_inode(name, dentry);
691                         error = -EEXIST;
692                         if (oflag & O_EXCL)
693                                 goto out;
694                         filp = do_open(dentry, oflag);
695                 } else {
696                         filp = do_create(mqueue_mnt->mnt_root, dentry,
697                                                 oflag, mode, u_attr);
698                 }
699         } else {
700                 error = -ENOENT;
701                 if (!dentry->d_inode)
702                         goto out;
703                 audit_inode(name, dentry);
704                 filp = do_open(dentry, oflag);
705         }
706
707         if (IS_ERR(filp)) {
708                 error = PTR_ERR(filp);
709                 goto out_putfd;
710         }
711
712         fd_install(fd, filp);
713         goto out_upsem;
714
715 out:
716         dput(dentry);
717         mntput(mqueue_mnt);
718 out_putfd:
719         put_unused_fd(fd);
720 out_err:
721         fd = error;
722 out_upsem:
723         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
724 out_putname:
725         putname(name);
726         return fd;
727 }
728
729 asmlinkage long sys_mq_unlink(const char __user *u_name)
730 {
731         int err;
732         char *name;
733         struct dentry *dentry;
734         struct inode *inode = NULL;
735
736         name = getname(u_name);
737         if (IS_ERR(name))
738                 return PTR_ERR(name);
739
740         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
741                         I_MUTEX_PARENT);
742         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
743         if (IS_ERR(dentry)) {
744                 err = PTR_ERR(dentry);
745                 goto out_unlock;
746         }
747
748         if (!dentry->d_inode) {
749                 err = -ENOENT;
750                 goto out_err;
751         }
752
753         inode = dentry->d_inode;
754         if (inode)
755                 atomic_inc(&inode->i_count);
756         err = mnt_want_write(mqueue_mnt);
757         if (err)
758                 goto out_err;
759         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
760         mnt_drop_write(mqueue_mnt);
761 out_err:
762         dput(dentry);
763
764 out_unlock:
765         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
766         putname(name);
767         if (inode)
768                 iput(inode);
769
770         return err;
771 }
772
773 /* Pipelined send and receive functions.
774  *
775  * If a receiver finds no waiting message, then it registers itself in the
776  * list of waiting receivers. A sender checks that list before adding the new
777  * message into the message array. If there is a waiting receiver, then it
778  * bypasses the message array and directly hands the message over to the
779  * receiver.
780  * The receiver accepts the message and returns without grabbing the queue
781  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
782  * are necessary. The same algorithm is used for sysv semaphores, see
783  * ipc/sem.c for more details.
784  *
785  * The same algorithm is used for senders.
786  */
787
788 /* pipelined_send() - send a message directly to the task waiting in
789  * sys_mq_timedreceive() (without inserting message into a queue).
790  */
791 static inline void pipelined_send(struct mqueue_inode_info *info,
792                                   struct msg_msg *message,
793                                   struct ext_wait_queue *receiver)
794 {
795         receiver->msg = message;
796         list_del(&receiver->list);
797         receiver->state = STATE_PENDING;
798         wake_up_process(receiver->task);
799         smp_wmb();
800         receiver->state = STATE_READY;
801 }
802
803 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
804  * gets its message and put to the queue (we have one free place for sure). */
805 static inline void pipelined_receive(struct mqueue_inode_info *info)
806 {
807         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
808
809         if (!sender) {
810                 /* for poll */
811                 wake_up_interruptible(&info->wait_q);
812                 return;
813         }
814         msg_insert(sender->msg, info);
815         list_del(&sender->list);
816         sender->state = STATE_PENDING;
817         wake_up_process(sender->task);
818         smp_wmb();
819         sender->state = STATE_READY;
820 }
821
822 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
823         size_t msg_len, unsigned int msg_prio,
824         const struct timespec __user *u_abs_timeout)
825 {
826         struct file *filp;
827         struct inode *inode;
828         struct ext_wait_queue wait;
829         struct ext_wait_queue *receiver;
830         struct msg_msg *msg_ptr;
831         struct mqueue_inode_info *info;
832         long timeout;
833         int ret;
834
835         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
836         if (ret != 0)
837                 return ret;
838
839         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
840                 return -EINVAL;
841
842         timeout = prepare_timeout(u_abs_timeout);
843
844         ret = -EBADF;
845         filp = fget(mqdes);
846         if (unlikely(!filp))
847                 goto out;
848
849         inode = filp->f_path.dentry->d_inode;
850         if (unlikely(filp->f_op != &mqueue_file_operations))
851                 goto out_fput;
852         info = MQUEUE_I(inode);
853         audit_inode(NULL, filp->f_path.dentry);
854
855         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
856                 goto out_fput;
857
858         if (unlikely(msg_len > info->attr.mq_msgsize)) {
859                 ret = -EMSGSIZE;
860                 goto out_fput;
861         }
862
863         /* First try to allocate memory, before doing anything with
864          * existing queues. */
865         msg_ptr = load_msg(u_msg_ptr, msg_len);
866         if (IS_ERR(msg_ptr)) {
867                 ret = PTR_ERR(msg_ptr);
868                 goto out_fput;
869         }
870         msg_ptr->m_ts = msg_len;
871         msg_ptr->m_type = msg_prio;
872
873         spin_lock(&info->lock);
874
875         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
876                 if (filp->f_flags & O_NONBLOCK) {
877                         spin_unlock(&info->lock);
878                         ret = -EAGAIN;
879                 } else if (unlikely(timeout < 0)) {
880                         spin_unlock(&info->lock);
881                         ret = timeout;
882                 } else {
883                         wait.task = current;
884                         wait.msg = (void *) msg_ptr;
885                         wait.state = STATE_NONE;
886                         ret = wq_sleep(info, SEND, timeout, &wait);
887                 }
888                 if (ret < 0)
889                         free_msg(msg_ptr);
890         } else {
891                 receiver = wq_get_first_waiter(info, RECV);
892                 if (receiver) {
893                         pipelined_send(info, msg_ptr, receiver);
894                 } else {
895                         /* adds message to the queue */
896                         msg_insert(msg_ptr, info);
897                         __do_notify(info);
898                 }
899                 inode->i_atime = inode->i_mtime = inode->i_ctime =
900                                 CURRENT_TIME;
901                 spin_unlock(&info->lock);
902                 ret = 0;
903         }
904 out_fput:
905         fput(filp);
906 out:
907         return ret;
908 }
909
910 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
911         size_t msg_len, unsigned int __user *u_msg_prio,
912         const struct timespec __user *u_abs_timeout)
913 {
914         long timeout;
915         ssize_t ret;
916         struct msg_msg *msg_ptr;
917         struct file *filp;
918         struct inode *inode;
919         struct mqueue_inode_info *info;
920         struct ext_wait_queue wait;
921
922         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
923         if (ret != 0)
924                 return ret;
925
926         timeout = prepare_timeout(u_abs_timeout);
927
928         ret = -EBADF;
929         filp = fget(mqdes);
930         if (unlikely(!filp))
931                 goto out;
932
933         inode = filp->f_path.dentry->d_inode;
934         if (unlikely(filp->f_op != &mqueue_file_operations))
935                 goto out_fput;
936         info = MQUEUE_I(inode);
937         audit_inode(NULL, filp->f_path.dentry);
938
939         if (unlikely(!(filp->f_mode & FMODE_READ)))
940                 goto out_fput;
941
942         /* checks if buffer is big enough */
943         if (unlikely(msg_len < info->attr.mq_msgsize)) {
944                 ret = -EMSGSIZE;
945                 goto out_fput;
946         }
947
948         spin_lock(&info->lock);
949         if (info->attr.mq_curmsgs == 0) {
950                 if (filp->f_flags & O_NONBLOCK) {
951                         spin_unlock(&info->lock);
952                         ret = -EAGAIN;
953                         msg_ptr = NULL;
954                 } else if (unlikely(timeout < 0)) {
955                         spin_unlock(&info->lock);
956                         ret = timeout;
957                         msg_ptr = NULL;
958                 } else {
959                         wait.task = current;
960                         wait.state = STATE_NONE;
961                         ret = wq_sleep(info, RECV, timeout, &wait);
962                         msg_ptr = wait.msg;
963                 }
964         } else {
965                 msg_ptr = msg_get(info);
966
967                 inode->i_atime = inode->i_mtime = inode->i_ctime =
968                                 CURRENT_TIME;
969
970                 /* There is now free space in queue. */
971                 pipelined_receive(info);
972                 spin_unlock(&info->lock);
973                 ret = 0;
974         }
975         if (ret == 0) {
976                 ret = msg_ptr->m_ts;
977
978                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
979                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
980                         ret = -EFAULT;
981                 }
982                 free_msg(msg_ptr);
983         }
984 out_fput:
985         fput(filp);
986 out:
987         return ret;
988 }
989
990 /*
991  * Notes: the case when user wants us to deregister (with NULL as pointer)
992  * and he isn't currently owner of notification, will be silently discarded.
993  * It isn't explicitly defined in the POSIX.
994  */
995 asmlinkage long sys_mq_notify(mqd_t mqdes,
996                                 const struct sigevent __user *u_notification)
997 {
998         int ret;
999         struct file *filp;
1000         struct sock *sock;
1001         struct inode *inode;
1002         struct sigevent notification;
1003         struct mqueue_inode_info *info;
1004         struct sk_buff *nc;
1005
1006         ret = audit_mq_notify(mqdes, u_notification);
1007         if (ret != 0)
1008                 return ret;
1009
1010         nc = NULL;
1011         sock = NULL;
1012         if (u_notification != NULL) {
1013                 if (copy_from_user(&notification, u_notification,
1014                                         sizeof(struct sigevent)))
1015                         return -EFAULT;
1016
1017                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1018                              notification.sigev_notify != SIGEV_SIGNAL &&
1019                              notification.sigev_notify != SIGEV_THREAD))
1020                         return -EINVAL;
1021                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1022                         !valid_signal(notification.sigev_signo)) {
1023                         return -EINVAL;
1024                 }
1025                 if (notification.sigev_notify == SIGEV_THREAD) {
1026                         long timeo;
1027
1028                         /* create the notify skb */
1029                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1030                         ret = -ENOMEM;
1031                         if (!nc)
1032                                 goto out;
1033                         ret = -EFAULT;
1034                         if (copy_from_user(nc->data,
1035                                         notification.sigev_value.sival_ptr,
1036                                         NOTIFY_COOKIE_LEN)) {
1037                                 goto out;
1038                         }
1039
1040                         /* TODO: add a header? */
1041                         skb_put(nc, NOTIFY_COOKIE_LEN);
1042                         /* and attach it to the socket */
1043 retry:
1044                         filp = fget(notification.sigev_signo);
1045                         ret = -EBADF;
1046                         if (!filp)
1047                                 goto out;
1048                         sock = netlink_getsockbyfilp(filp);
1049                         fput(filp);
1050                         if (IS_ERR(sock)) {
1051                                 ret = PTR_ERR(sock);
1052                                 sock = NULL;
1053                                 goto out;
1054                         }
1055
1056                         timeo = MAX_SCHEDULE_TIMEOUT;
1057                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1058                         if (ret == 1)
1059                                 goto retry;
1060                         if (ret) {
1061                                 sock = NULL;
1062                                 nc = NULL;
1063                                 goto out;
1064                         }
1065                 }
1066         }
1067
1068         ret = -EBADF;
1069         filp = fget(mqdes);
1070         if (!filp)
1071                 goto out;
1072
1073         inode = filp->f_path.dentry->d_inode;
1074         if (unlikely(filp->f_op != &mqueue_file_operations))
1075                 goto out_fput;
1076         info = MQUEUE_I(inode);
1077
1078         ret = 0;
1079         spin_lock(&info->lock);
1080         if (u_notification == NULL) {
1081                 if (info->notify_owner == task_tgid(current)) {
1082                         remove_notification(info);
1083                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1084                 }
1085         } else if (info->notify_owner != NULL) {
1086                 ret = -EBUSY;
1087         } else {
1088                 switch (notification.sigev_notify) {
1089                 case SIGEV_NONE:
1090                         info->notify.sigev_notify = SIGEV_NONE;
1091                         break;
1092                 case SIGEV_THREAD:
1093                         info->notify_sock = sock;
1094                         info->notify_cookie = nc;
1095                         sock = NULL;
1096                         nc = NULL;
1097                         info->notify.sigev_notify = SIGEV_THREAD;
1098                         break;
1099                 case SIGEV_SIGNAL:
1100                         info->notify.sigev_signo = notification.sigev_signo;
1101                         info->notify.sigev_value = notification.sigev_value;
1102                         info->notify.sigev_notify = SIGEV_SIGNAL;
1103                         break;
1104                 }
1105
1106                 info->notify_owner = get_pid(task_tgid(current));
1107                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1108         }
1109         spin_unlock(&info->lock);
1110 out_fput:
1111         fput(filp);
1112 out:
1113         if (sock) {
1114                 netlink_detachskb(sock, nc);
1115         } else if (nc) {
1116                 dev_kfree_skb(nc);
1117         }
1118         return ret;
1119 }
1120
1121 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1122                         const struct mq_attr __user *u_mqstat,
1123                         struct mq_attr __user *u_omqstat)
1124 {
1125         int ret;
1126         struct mq_attr mqstat, omqstat;
1127         struct file *filp;
1128         struct inode *inode;
1129         struct mqueue_inode_info *info;
1130
1131         if (u_mqstat != NULL) {
1132                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1133                         return -EFAULT;
1134                 if (mqstat.mq_flags & (~O_NONBLOCK))
1135                         return -EINVAL;
1136         }
1137
1138         ret = -EBADF;
1139         filp = fget(mqdes);
1140         if (!filp)
1141                 goto out;
1142
1143         inode = filp->f_path.dentry->d_inode;
1144         if (unlikely(filp->f_op != &mqueue_file_operations))
1145                 goto out_fput;
1146         info = MQUEUE_I(inode);
1147
1148         spin_lock(&info->lock);
1149
1150         omqstat = info->attr;
1151         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1152         if (u_mqstat) {
1153                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1154                 if (ret != 0) {
1155                         spin_unlock(&info->lock);
1156                         goto out_fput;
1157                 }
1158                 if (mqstat.mq_flags & O_NONBLOCK)
1159                         filp->f_flags |= O_NONBLOCK;
1160                 else
1161                         filp->f_flags &= ~O_NONBLOCK;
1162
1163                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1164         }
1165
1166         spin_unlock(&info->lock);
1167
1168         ret = 0;
1169         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1170                                                 sizeof(struct mq_attr)))
1171                 ret = -EFAULT;
1172
1173 out_fput:
1174         fput(filp);
1175 out:
1176         return ret;
1177 }
1178
1179 static const struct inode_operations mqueue_dir_inode_operations = {
1180         .lookup = simple_lookup,
1181         .create = mqueue_create,
1182         .unlink = mqueue_unlink,
1183 };
1184
1185 static const struct file_operations mqueue_file_operations = {
1186         .flush = mqueue_flush_file,
1187         .poll = mqueue_poll_file,
1188         .read = mqueue_read_file,
1189 };
1190
1191 static struct super_operations mqueue_super_ops = {
1192         .alloc_inode = mqueue_alloc_inode,
1193         .destroy_inode = mqueue_destroy_inode,
1194         .statfs = simple_statfs,
1195         .delete_inode = mqueue_delete_inode,
1196         .drop_inode = generic_delete_inode,
1197 };
1198
1199 static struct file_system_type mqueue_fs_type = {
1200         .name = "mqueue",
1201         .get_sb = mqueue_get_sb,
1202         .kill_sb = kill_litter_super,
1203 };
1204
1205 static int msg_max_limit_min = DFLT_MSGMAX;
1206 static int msg_max_limit_max = HARD_MSGMAX;
1207
1208 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1209 static int msg_maxsize_limit_max = INT_MAX;
1210
1211 static ctl_table mq_sysctls[] = {
1212         {
1213                 .procname       = "queues_max",
1214                 .data           = &queues_max,
1215                 .maxlen         = sizeof(int),
1216                 .mode           = 0644,
1217                 .proc_handler   = &proc_dointvec,
1218         },
1219         {
1220                 .procname       = "msg_max",
1221                 .data           = &msg_max,
1222                 .maxlen         = sizeof(int),
1223                 .mode           = 0644,
1224                 .proc_handler   = &proc_dointvec_minmax,
1225                 .extra1         = &msg_max_limit_min,
1226                 .extra2         = &msg_max_limit_max,
1227         },
1228         {
1229                 .procname       = "msgsize_max",
1230                 .data           = &msgsize_max,
1231                 .maxlen         = sizeof(int),
1232                 .mode           = 0644,
1233                 .proc_handler   = &proc_dointvec_minmax,
1234                 .extra1         = &msg_maxsize_limit_min,
1235                 .extra2         = &msg_maxsize_limit_max,
1236         },
1237         { .ctl_name = 0 }
1238 };
1239
1240 static ctl_table mq_sysctl_dir[] = {
1241         {
1242                 .procname       = "mqueue",
1243                 .mode           = 0555,
1244                 .child          = mq_sysctls,
1245         },
1246         { .ctl_name = 0 }
1247 };
1248
1249 static ctl_table mq_sysctl_root[] = {
1250         {
1251                 .ctl_name       = CTL_FS,
1252                 .procname       = "fs",
1253                 .mode           = 0555,
1254                 .child          = mq_sysctl_dir,
1255         },
1256         { .ctl_name = 0 }
1257 };
1258
1259 static int __init init_mqueue_fs(void)
1260 {
1261         int error;
1262
1263         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1264                                 sizeof(struct mqueue_inode_info), 0,
1265                                 SLAB_HWCACHE_ALIGN, init_once);
1266         if (mqueue_inode_cachep == NULL)
1267                 return -ENOMEM;
1268
1269         /* ignore failues - they are not fatal */
1270         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1271
1272         error = register_filesystem(&mqueue_fs_type);
1273         if (error)
1274                 goto out_sysctl;
1275
1276         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1277                 error = PTR_ERR(mqueue_mnt);
1278                 goto out_filesystem;
1279         }
1280
1281         /* internal initialization - not common for vfs */
1282         queues_count = 0;
1283         spin_lock_init(&mq_lock);
1284
1285         return 0;
1286
1287 out_filesystem:
1288         unregister_filesystem(&mqueue_fs_type);
1289 out_sysctl:
1290         if (mq_sysctl_table)
1291                 unregister_sysctl_table(mq_sysctl_table);
1292         kmem_cache_destroy(mqueue_inode_cachep);
1293         return error;
1294 }
1295
1296 __initcall(init_mqueue_fs);