[PATCH] merge some from Rusty's trivial patches
[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          (wrona@mat.uni.torun.pl)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *                          Manfred Spraul          (manfred@colorfullife.com)
10  *
11  * This file is released under the GPL.
12  */
13
14 #include <linux/init.h>
15 #include <linux/pagemap.h>
16 #include <linux/file.h>
17 #include <linux/mount.h>
18 #include <linux/namei.h>
19 #include <linux/sysctl.h>
20 #include <linux/poll.h>
21 #include <linux/mqueue.h>
22 #include <linux/msg.h>
23 #include <linux/skbuff.h>
24 #include <linux/netlink.h>
25 #include <linux/syscalls.h>
26 #include <linux/signal.h>
27 #include <net/sock.h>
28 #include "util.h"
29
30 #define MQUEUE_MAGIC    0x19800202
31 #define DIRENT_SIZE     20
32 #define FILENT_SIZE     80
33
34 #define SEND            0
35 #define RECV            1
36
37 #define STATE_NONE      0
38 #define STATE_PENDING   1
39 #define STATE_READY     2
40
41 /* used by sysctl */
42 #define FS_MQUEUE       1
43 #define CTL_QUEUESMAX   2
44 #define CTL_MSGMAX      3
45 #define CTL_MSGSIZEMAX  4
46
47 /* default values */
48 #define DFLT_QUEUESMAX  256     /* max number of message queues */
49 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
50 #define HARD_MSGMAX     (131072/sizeof(void*))
51 #define DFLT_MSGSIZEMAX 8192    /* max message size */
52
53 #define NOTIFY_COOKIE_LEN       32
54
55 struct ext_wait_queue {         /* queue of sleeping tasks */
56         struct task_struct *task;
57         struct list_head list;
58         struct msg_msg *msg;    /* ptr of loaded message */
59         int state;              /* one of STATE_* values */
60 };
61
62 struct mqueue_inode_info {
63         spinlock_t lock;
64         struct inode vfs_inode;
65         wait_queue_head_t wait_q;
66
67         struct msg_msg **messages;
68         struct mq_attr attr;
69
70         struct sigevent notify;
71         pid_t notify_owner;
72         struct user_struct *user;       /* user who created, for accounting */
73         struct sock *notify_sock;
74         struct sk_buff *notify_cookie;
75
76         /* for tasks waiting for free space and messages, respectively */
77         struct ext_wait_queue e_wait_q[2];
78
79         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
80 };
81
82 static struct inode_operations mqueue_dir_inode_operations;
83 static struct file_operations mqueue_file_operations;
84 static struct super_operations mqueue_super_ops;
85 static void remove_notification(struct mqueue_inode_info *info);
86
87 static spinlock_t mq_lock;
88 static kmem_cache_t *mqueue_inode_cachep;
89 static struct vfsmount *mqueue_mnt;
90
91 static unsigned int queues_count;
92 static unsigned int queues_max  = DFLT_QUEUESMAX;
93 static unsigned int msg_max     = DFLT_MSGMAX;
94 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
95
96 static struct ctl_table_header * mq_sysctl_table;
97
98 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
99 {
100         return container_of(inode, struct mqueue_inode_info, vfs_inode);
101 }
102
103 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
104                                                         struct mq_attr *attr)
105 {
106         struct inode *inode;
107
108         inode = new_inode(sb);
109         if (inode) {
110                 inode->i_mode = mode;
111                 inode->i_uid = current->fsuid;
112                 inode->i_gid = current->fsgid;
113                 inode->i_blksize = PAGE_CACHE_SIZE;
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 = 0;
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                         inode->i_nlink++;
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 struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
204                                          int flags, const char *dev_name,
205                                          void *data)
206 {
207         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
208 }
209
210 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
211 {
212         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
213
214         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
215                 SLAB_CTOR_CONSTRUCTOR)
216                 inode_init_once(&p->vfs_inode);
217 }
218
219 static struct inode *mqueue_alloc_inode(struct super_block *sb)
220 {
221         struct mqueue_inode_info *ei;
222
223         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
224         if (!ei)
225                 return NULL;
226         return &ei->vfs_inode;
227 }
228
229 static void mqueue_destroy_inode(struct inode *inode)
230 {
231         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
232 }
233
234 static void mqueue_delete_inode(struct inode *inode)
235 {
236         struct mqueue_inode_info *info;
237         struct user_struct *user;
238         unsigned long mq_bytes;
239         int i;
240
241         if (S_ISDIR(inode->i_mode)) {
242                 clear_inode(inode);
243                 return;
244         }
245         info = MQUEUE_I(inode);
246         spin_lock(&info->lock);
247         for (i = 0; i < info->attr.mq_curmsgs; i++)
248                 free_msg(info->messages[i]);
249         kfree(info->messages);
250         spin_unlock(&info->lock);
251
252         clear_inode(inode);
253
254         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
255                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
256         user = info->user;
257         if (user) {
258                 spin_lock(&mq_lock);
259                 user->mq_bytes -= mq_bytes;
260                 queues_count--;
261                 spin_unlock(&mq_lock);
262                 free_uid(user);
263         }
264 }
265
266 static int mqueue_create(struct inode *dir, struct dentry *dentry,
267                                 int mode, struct nameidata *nd)
268 {
269         struct inode *inode;
270         struct mq_attr *attr = dentry->d_fsdata;
271         int error;
272
273         spin_lock(&mq_lock);
274         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
275                 error = -ENOSPC;
276                 goto out_lock;
277         }
278         queues_count++;
279         spin_unlock(&mq_lock);
280
281         inode = mqueue_get_inode(dir->i_sb, mode, attr);
282         if (!inode) {
283                 error = -ENOMEM;
284                 spin_lock(&mq_lock);
285                 queues_count--;
286                 goto out_lock;
287         }
288
289         dir->i_size += DIRENT_SIZE;
290         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
291
292         d_instantiate(dentry, inode);
293         dget(dentry);
294         return 0;
295 out_lock:
296         spin_unlock(&mq_lock);
297         return error;
298 }
299
300 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
301 {
302         struct inode *inode = dentry->d_inode;
303
304         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
305         dir->i_size -= DIRENT_SIZE;
306         inode->i_nlink--;
307         dput(dentry);
308         return 0;
309 }
310
311 /*
312 *       This is routine for system read from queue file.
313 *       To avoid mess with doing here some sort of mq_receive we allow
314 *       to read only queue size & notification info (the only values
315 *       that are interesting from user point of view and aren't accessible
316 *       through std routines)
317 */
318 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
319                                 size_t count, loff_t * off)
320 {
321         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
322         char buffer[FILENT_SIZE];
323         size_t slen;
324         loff_t o;
325
326         if (!count)
327                 return 0;
328
329         spin_lock(&info->lock);
330         snprintf(buffer, sizeof(buffer),
331                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
332                         info->qsize,
333                         info->notify_owner ? info->notify.sigev_notify : 0,
334                         (info->notify_owner &&
335                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
336                                 info->notify.sigev_signo : 0,
337                         info->notify_owner);
338         spin_unlock(&info->lock);
339         buffer[sizeof(buffer)-1] = '\0';
340         slen = strlen(buffer)+1;
341
342         o = *off;
343         if (o > slen)
344                 return 0;
345
346         if (o + count > slen)
347                 count = slen - o;
348
349         if (copy_to_user(u_data, buffer + o, count))
350                 return -EFAULT;
351
352         *off = o + count;
353         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
354         return count;
355 }
356
357 static int mqueue_flush_file(struct file *filp)
358 {
359         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
360
361         spin_lock(&info->lock);
362         if (current->tgid == info->notify_owner)
363                 remove_notification(info);
364
365         spin_unlock(&info->lock);
366         return 0;
367 }
368
369 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
370 {
371         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
372         int retval = 0;
373
374         poll_wait(filp, &info->wait_q, poll_tab);
375
376         spin_lock(&info->lock);
377         if (info->attr.mq_curmsgs)
378                 retval = POLLIN | POLLRDNORM;
379
380         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
381                 retval |= POLLOUT | POLLWRNORM;
382         spin_unlock(&info->lock);
383
384         return retval;
385 }
386
387 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
388 static void wq_add(struct mqueue_inode_info *info, int sr,
389                         struct ext_wait_queue *ewp)
390 {
391         struct ext_wait_queue *walk;
392
393         ewp->task = current;
394
395         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
396                 if (walk->task->static_prio <= current->static_prio) {
397                         list_add_tail(&ewp->list, &walk->list);
398                         return;
399                 }
400         }
401         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
402 }
403
404 /*
405  * Puts current task to sleep. Caller must hold queue lock. After return
406  * lock isn't held.
407  * sr: SEND or RECV
408  */
409 static int wq_sleep(struct mqueue_inode_info *info, int sr,
410                         long timeout, struct ext_wait_queue *ewp)
411 {
412         int retval;
413         signed long time;
414
415         wq_add(info, sr, ewp);
416
417         for (;;) {
418                 set_current_state(TASK_INTERRUPTIBLE);
419
420                 spin_unlock(&info->lock);
421                 time = schedule_timeout(timeout);
422
423                 while (ewp->state == STATE_PENDING)
424                         cpu_relax();
425
426                 if (ewp->state == STATE_READY) {
427                         retval = 0;
428                         goto out;
429                 }
430                 spin_lock(&info->lock);
431                 if (ewp->state == STATE_READY) {
432                         retval = 0;
433                         goto out_unlock;
434                 }
435                 if (signal_pending(current)) {
436                         retval = -ERESTARTSYS;
437                         break;
438                 }
439                 if (time == 0) {
440                         retval = -ETIMEDOUT;
441                         break;
442                 }
443         }
444         list_del(&ewp->list);
445 out_unlock:
446         spin_unlock(&info->lock);
447 out:
448         return retval;
449 }
450
451 /*
452  * Returns waiting task that should be serviced first or NULL if none exists
453  */
454 static struct ext_wait_queue *wq_get_first_waiter(
455                 struct mqueue_inode_info *info, int sr)
456 {
457         struct list_head *ptr;
458
459         ptr = info->e_wait_q[sr].list.prev;
460         if (ptr == &info->e_wait_q[sr].list)
461                 return NULL;
462         return list_entry(ptr, struct ext_wait_queue, list);
463 }
464
465 /* Auxiliary functions to manipulate messages' list */
466 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
467 {
468         int k;
469
470         k = info->attr.mq_curmsgs - 1;
471         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
472                 info->messages[k + 1] = info->messages[k];
473                 k--;
474         }
475         info->attr.mq_curmsgs++;
476         info->qsize += ptr->m_ts;
477         info->messages[k + 1] = ptr;
478 }
479
480 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
481 {
482         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
483         return info->messages[info->attr.mq_curmsgs];
484 }
485
486 static inline void set_cookie(struct sk_buff *skb, char code)
487 {
488         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
489 }
490
491 /*
492  * The next function is only to split too long sys_mq_timedsend
493  */
494 static void __do_notify(struct mqueue_inode_info *info)
495 {
496         /* notification
497          * invoked when there is registered process and there isn't process
498          * waiting synchronously for message AND state of queue changed from
499          * empty to not empty. Here we are sure that no one is waiting
500          * synchronously. */
501         if (info->notify_owner &&
502             info->attr.mq_curmsgs == 1) {
503                 struct siginfo sig_i;
504                 switch (info->notify.sigev_notify) {
505                 case SIGEV_NONE:
506                         break;
507                 case SIGEV_SIGNAL:
508                         /* sends signal */
509
510                         sig_i.si_signo = info->notify.sigev_signo;
511                         sig_i.si_errno = 0;
512                         sig_i.si_code = SI_MESGQ;
513                         sig_i.si_value = info->notify.sigev_value;
514                         sig_i.si_pid = current->tgid;
515                         sig_i.si_uid = current->uid;
516
517                         kill_proc_info(info->notify.sigev_signo,
518                                        &sig_i, info->notify_owner);
519                         break;
520                 case SIGEV_THREAD:
521                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
522                         netlink_sendskb(info->notify_sock,
523                                         info->notify_cookie, 0);
524                         break;
525                 }
526                 /* after notification unregisters process */
527                 info->notify_owner = 0;
528         }
529         wake_up(&info->wait_q);
530 }
531
532 static long prepare_timeout(const struct timespec __user *u_arg)
533 {
534         struct timespec ts, nowts;
535         long timeout;
536
537         if (u_arg) {
538                 if (unlikely(copy_from_user(&ts, u_arg,
539                                         sizeof(struct timespec))))
540                         return -EFAULT;
541
542                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
543                         || ts.tv_nsec >= NSEC_PER_SEC))
544                         return -EINVAL;
545                 nowts = CURRENT_TIME;
546                 /* first subtract as jiffies can't be too big */
547                 ts.tv_sec -= nowts.tv_sec;
548                 if (ts.tv_nsec < nowts.tv_nsec) {
549                         ts.tv_nsec += NSEC_PER_SEC;
550                         ts.tv_sec--;
551                 }
552                 ts.tv_nsec -= nowts.tv_nsec;
553                 if (ts.tv_sec < 0)
554                         return 0;
555
556                 timeout = timespec_to_jiffies(&ts) + 1;
557         } else
558                 return MAX_SCHEDULE_TIMEOUT;
559
560         return timeout;
561 }
562
563 static void remove_notification(struct mqueue_inode_info *info)
564 {
565         if (info->notify_owner != 0 &&
566             info->notify.sigev_notify == SIGEV_THREAD) {
567                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
568                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
569         }
570         info->notify_owner = 0;
571 }
572
573 static int mq_attr_ok(struct mq_attr *attr)
574 {
575         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
576                 return 0;
577         if (capable(CAP_SYS_RESOURCE)) {
578                 if (attr->mq_maxmsg > HARD_MSGMAX)
579                         return 0;
580         } else {
581                 if (attr->mq_maxmsg > msg_max ||
582                                 attr->mq_msgsize > msgsize_max)
583                         return 0;
584         }
585         /* check for overflow */
586         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
587                 return 0;
588         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
589             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
590             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
591                 return 0;
592         return 1;
593 }
594
595 /*
596  * Invoked when creating a new queue via sys_mq_open
597  */
598 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
599                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
600 {
601         struct file *filp;
602         struct mq_attr attr;
603         int ret;
604
605         if (u_attr != NULL) {
606                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
607                         return ERR_PTR(-EFAULT);
608                 if (!mq_attr_ok(&attr))
609                         return ERR_PTR(-EINVAL);
610                 /* store for use during create */
611                 dentry->d_fsdata = &attr;
612         }
613
614         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
615         dentry->d_fsdata = NULL;
616         if (ret)
617                 return ERR_PTR(ret);
618
619         filp = dentry_open(dentry, mqueue_mnt, oflag);
620         if (!IS_ERR(filp))
621                 dget(dentry);
622
623         return filp;
624 }
625
626 /* Opens existing queue */
627 static struct file *do_open(struct dentry *dentry, int oflag)
628 {
629 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
630                                         MAY_READ | MAY_WRITE };
631         struct file *filp;
632
633         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
634                 return ERR_PTR(-EINVAL);
635
636         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
637                 return ERR_PTR(-EACCES);
638
639         filp = dentry_open(dentry, mqueue_mnt, oflag);
640
641         if (!IS_ERR(filp))
642                 dget(dentry);
643
644         return filp;
645 }
646
647 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
648                                 struct mq_attr __user *u_attr)
649 {
650         struct dentry *dentry;
651         struct file *filp;
652         char *name;
653         int fd, error;
654
655         if (IS_ERR(name = getname(u_name)))
656                 return PTR_ERR(name);
657
658         fd = get_unused_fd();
659         if (fd < 0)
660                 goto out_putname;
661
662         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
663         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
664         if (IS_ERR(dentry)) {
665                 error = PTR_ERR(dentry);
666                 goto out_err;
667         }
668         mntget(mqueue_mnt);
669
670         if (oflag & O_CREAT) {
671                 if (dentry->d_inode) {  /* entry already exists */
672                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
673                                         do_open(dentry, oflag);
674                 } else {
675                         filp = do_create(mqueue_mnt->mnt_root, dentry,
676                                                 oflag, mode, u_attr);
677                 }
678         } else
679                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
680                                         ERR_PTR(-ENOENT);
681
682         dput(dentry);
683
684         if (IS_ERR(filp)) {
685                 error = PTR_ERR(filp);
686                 goto out_putfd;
687         }
688
689         set_close_on_exec(fd, 1);
690         fd_install(fd, filp);
691         goto out_upsem;
692
693 out_putfd:
694         mntput(mqueue_mnt);
695         put_unused_fd(fd);
696 out_err:
697         fd = error;
698 out_upsem:
699         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
700 out_putname:
701         putname(name);
702         return fd;
703 }
704
705 asmlinkage long sys_mq_unlink(const char __user *u_name)
706 {
707         int err;
708         char *name;
709         struct dentry *dentry;
710         struct inode *inode = NULL;
711
712         name = getname(u_name);
713         if (IS_ERR(name))
714                 return PTR_ERR(name);
715
716         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
717         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
718         if (IS_ERR(dentry)) {
719                 err = PTR_ERR(dentry);
720                 goto out_unlock;
721         }
722
723         if (!dentry->d_inode) {
724                 err = -ENOENT;
725                 goto out_err;
726         }
727
728         inode = dentry->d_inode;
729         if (inode)
730                 atomic_inc(&inode->i_count);
731
732         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
733 out_err:
734         dput(dentry);
735
736 out_unlock:
737         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
738         putname(name);
739         if (inode)
740                 iput(inode);
741
742         return err;
743 }
744
745 /* Pipelined send and receive functions.
746  *
747  * If a receiver finds no waiting message, then it registers itself in the
748  * list of waiting receivers. A sender checks that list before adding the new
749  * message into the message array. If there is a waiting receiver, then it
750  * bypasses the message array and directly hands the message over to the
751  * receiver.
752  * The receiver accepts the message and returns without grabbing the queue
753  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
754  * are necessary. The same algorithm is used for sysv semaphores, see
755  * ipc/sem.c fore more details.
756  *
757  * The same algorithm is used for senders.
758  */
759
760 /* pipelined_send() - send a message directly to the task waiting in
761  * sys_mq_timedreceive() (without inserting message into a queue).
762  */
763 static inline void pipelined_send(struct mqueue_inode_info *info,
764                                   struct msg_msg *message,
765                                   struct ext_wait_queue *receiver)
766 {
767         receiver->msg = message;
768         list_del(&receiver->list);
769         receiver->state = STATE_PENDING;
770         wake_up_process(receiver->task);
771         smp_wmb();
772         receiver->state = STATE_READY;
773 }
774
775 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
776  * gets its message and put to the queue (we have one free place for sure). */
777 static inline void pipelined_receive(struct mqueue_inode_info *info)
778 {
779         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
780
781         if (!sender) {
782                 /* for poll */
783                 wake_up_interruptible(&info->wait_q);
784                 return;
785         }
786         msg_insert(sender->msg, info);
787         list_del(&sender->list);
788         sender->state = STATE_PENDING;
789         wake_up_process(sender->task);
790         smp_wmb();
791         sender->state = STATE_READY;
792 }
793
794 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
795         size_t msg_len, unsigned int msg_prio,
796         const struct timespec __user *u_abs_timeout)
797 {
798         struct file *filp;
799         struct inode *inode;
800         struct ext_wait_queue wait;
801         struct ext_wait_queue *receiver;
802         struct msg_msg *msg_ptr;
803         struct mqueue_inode_info *info;
804         long timeout;
805         int ret;
806
807         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
808                 return -EINVAL;
809
810         timeout = prepare_timeout(u_abs_timeout);
811
812         ret = -EBADF;
813         filp = fget(mqdes);
814         if (unlikely(!filp))
815                 goto out;
816
817         inode = filp->f_dentry->d_inode;
818         if (unlikely(filp->f_op != &mqueue_file_operations))
819                 goto out_fput;
820         info = MQUEUE_I(inode);
821
822         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
823                 goto out_fput;
824
825         if (unlikely(msg_len > info->attr.mq_msgsize)) {
826                 ret = -EMSGSIZE;
827                 goto out_fput;
828         }
829
830         /* First try to allocate memory, before doing anything with
831          * existing queues. */
832         msg_ptr = load_msg(u_msg_ptr, msg_len);
833         if (IS_ERR(msg_ptr)) {
834                 ret = PTR_ERR(msg_ptr);
835                 goto out_fput;
836         }
837         msg_ptr->m_ts = msg_len;
838         msg_ptr->m_type = msg_prio;
839
840         spin_lock(&info->lock);
841
842         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
843                 if (filp->f_flags & O_NONBLOCK) {
844                         spin_unlock(&info->lock);
845                         ret = -EAGAIN;
846                 } else if (unlikely(timeout < 0)) {
847                         spin_unlock(&info->lock);
848                         ret = timeout;
849                 } else {
850                         wait.task = current;
851                         wait.msg = (void *) msg_ptr;
852                         wait.state = STATE_NONE;
853                         ret = wq_sleep(info, SEND, timeout, &wait);
854                 }
855                 if (ret < 0)
856                         free_msg(msg_ptr);
857         } else {
858                 receiver = wq_get_first_waiter(info, RECV);
859                 if (receiver) {
860                         pipelined_send(info, msg_ptr, receiver);
861                 } else {
862                         /* adds message to the queue */
863                         msg_insert(msg_ptr, info);
864                         __do_notify(info);
865                 }
866                 inode->i_atime = inode->i_mtime = inode->i_ctime =
867                                 CURRENT_TIME;
868                 spin_unlock(&info->lock);
869                 ret = 0;
870         }
871 out_fput:
872         fput(filp);
873 out:
874         return ret;
875 }
876
877 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
878         size_t msg_len, unsigned int __user *u_msg_prio,
879         const struct timespec __user *u_abs_timeout)
880 {
881         long timeout;
882         ssize_t ret;
883         struct msg_msg *msg_ptr;
884         struct file *filp;
885         struct inode *inode;
886         struct mqueue_inode_info *info;
887         struct ext_wait_queue wait;
888
889         timeout = prepare_timeout(u_abs_timeout);
890
891         ret = -EBADF;
892         filp = fget(mqdes);
893         if (unlikely(!filp))
894                 goto out;
895
896         inode = filp->f_dentry->d_inode;
897         if (unlikely(filp->f_op != &mqueue_file_operations))
898                 goto out_fput;
899         info = MQUEUE_I(inode);
900
901         if (unlikely(!(filp->f_mode & FMODE_READ)))
902                 goto out_fput;
903
904         /* checks if buffer is big enough */
905         if (unlikely(msg_len < info->attr.mq_msgsize)) {
906                 ret = -EMSGSIZE;
907                 goto out_fput;
908         }
909
910         spin_lock(&info->lock);
911         if (info->attr.mq_curmsgs == 0) {
912                 if (filp->f_flags & O_NONBLOCK) {
913                         spin_unlock(&info->lock);
914                         ret = -EAGAIN;
915                         msg_ptr = NULL;
916                 } else if (unlikely(timeout < 0)) {
917                         spin_unlock(&info->lock);
918                         ret = timeout;
919                         msg_ptr = NULL;
920                 } else {
921                         wait.task = current;
922                         wait.state = STATE_NONE;
923                         ret = wq_sleep(info, RECV, timeout, &wait);
924                         msg_ptr = wait.msg;
925                 }
926         } else {
927                 msg_ptr = msg_get(info);
928
929                 inode->i_atime = inode->i_mtime = inode->i_ctime =
930                                 CURRENT_TIME;
931
932                 /* There is now free space in queue. */
933                 pipelined_receive(info);
934                 spin_unlock(&info->lock);
935                 ret = 0;
936         }
937         if (ret == 0) {
938                 ret = msg_ptr->m_ts;
939
940                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
941                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
942                         ret = -EFAULT;
943                 }
944                 free_msg(msg_ptr);
945         }
946 out_fput:
947         fput(filp);
948 out:
949         return ret;
950 }
951
952 /*
953  * Notes: the case when user wants us to deregister (with NULL as pointer)
954  * and he isn't currently owner of notification, will be silently discarded.
955  * It isn't explicitly defined in the POSIX.
956  */
957 asmlinkage long sys_mq_notify(mqd_t mqdes,
958                                 const struct sigevent __user *u_notification)
959 {
960         int ret;
961         struct file *filp;
962         struct sock *sock;
963         struct inode *inode;
964         struct sigevent notification;
965         struct mqueue_inode_info *info;
966         struct sk_buff *nc;
967
968         nc = NULL;
969         sock = NULL;
970         if (u_notification != NULL) {
971                 if (copy_from_user(&notification, u_notification,
972                                         sizeof(struct sigevent)))
973                         return -EFAULT;
974
975                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
976                              notification.sigev_notify != SIGEV_SIGNAL &&
977                              notification.sigev_notify != SIGEV_THREAD))
978                         return -EINVAL;
979                 if (notification.sigev_notify == SIGEV_SIGNAL &&
980                         !valid_signal(notification.sigev_signo)) {
981                         return -EINVAL;
982                 }
983                 if (notification.sigev_notify == SIGEV_THREAD) {
984                         /* create the notify skb */
985                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
986                         ret = -ENOMEM;
987                         if (!nc)
988                                 goto out;
989                         ret = -EFAULT;
990                         if (copy_from_user(nc->data,
991                                         notification.sigev_value.sival_ptr,
992                                         NOTIFY_COOKIE_LEN)) {
993                                 goto out;
994                         }
995
996                         /* TODO: add a header? */
997                         skb_put(nc, NOTIFY_COOKIE_LEN);
998                         /* and attach it to the socket */
999 retry:
1000                         filp = fget(notification.sigev_signo);
1001                         ret = -EBADF;
1002                         if (!filp)
1003                                 goto out;
1004                         sock = netlink_getsockbyfilp(filp);
1005                         fput(filp);
1006                         if (IS_ERR(sock)) {
1007                                 ret = PTR_ERR(sock);
1008                                 sock = NULL;
1009                                 goto out;
1010                         }
1011
1012                         ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
1013                         if (ret == 1)
1014                                 goto retry;
1015                         if (ret) {
1016                                 sock = NULL;
1017                                 nc = NULL;
1018                                 goto out;
1019                         }
1020                 }
1021         }
1022
1023         ret = -EBADF;
1024         filp = fget(mqdes);
1025         if (!filp)
1026                 goto out;
1027
1028         inode = filp->f_dentry->d_inode;
1029         if (unlikely(filp->f_op != &mqueue_file_operations))
1030                 goto out_fput;
1031         info = MQUEUE_I(inode);
1032
1033         ret = 0;
1034         spin_lock(&info->lock);
1035         if (u_notification == NULL) {
1036                 if (info->notify_owner == current->tgid) {
1037                         remove_notification(info);
1038                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1039                 }
1040         } else if (info->notify_owner != 0) {
1041                 ret = -EBUSY;
1042         } else {
1043                 switch (notification.sigev_notify) {
1044                 case SIGEV_NONE:
1045                         info->notify.sigev_notify = SIGEV_NONE;
1046                         break;
1047                 case SIGEV_THREAD:
1048                         info->notify_sock = sock;
1049                         info->notify_cookie = nc;
1050                         sock = NULL;
1051                         nc = NULL;
1052                         info->notify.sigev_notify = SIGEV_THREAD;
1053                         break;
1054                 case SIGEV_SIGNAL:
1055                         info->notify.sigev_signo = notification.sigev_signo;
1056                         info->notify.sigev_value = notification.sigev_value;
1057                         info->notify.sigev_notify = SIGEV_SIGNAL;
1058                         break;
1059                 }
1060                 info->notify_owner = current->tgid;
1061                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1062         }
1063         spin_unlock(&info->lock);
1064 out_fput:
1065         fput(filp);
1066 out:
1067         if (sock) {
1068                 netlink_detachskb(sock, nc);
1069         } else if (nc) {
1070                 dev_kfree_skb(nc);
1071         }
1072         return ret;
1073 }
1074
1075 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1076                         const struct mq_attr __user *u_mqstat,
1077                         struct mq_attr __user *u_omqstat)
1078 {
1079         int ret;
1080         struct mq_attr mqstat, omqstat;
1081         struct file *filp;
1082         struct inode *inode;
1083         struct mqueue_inode_info *info;
1084
1085         if (u_mqstat != NULL) {
1086                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1087                         return -EFAULT;
1088                 if (mqstat.mq_flags & (~O_NONBLOCK))
1089                         return -EINVAL;
1090         }
1091
1092         ret = -EBADF;
1093         filp = fget(mqdes);
1094         if (!filp)
1095                 goto out;
1096
1097         inode = filp->f_dentry->d_inode;
1098         if (unlikely(filp->f_op != &mqueue_file_operations))
1099                 goto out_fput;
1100         info = MQUEUE_I(inode);
1101
1102         spin_lock(&info->lock);
1103
1104         omqstat = info->attr;
1105         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1106         if (u_mqstat) {
1107                 if (mqstat.mq_flags & O_NONBLOCK)
1108                         filp->f_flags |= O_NONBLOCK;
1109                 else
1110                         filp->f_flags &= ~O_NONBLOCK;
1111
1112                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1113         }
1114
1115         spin_unlock(&info->lock);
1116
1117         ret = 0;
1118         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1119                                                 sizeof(struct mq_attr)))
1120                 ret = -EFAULT;
1121
1122 out_fput:
1123         fput(filp);
1124 out:
1125         return ret;
1126 }
1127
1128 static struct inode_operations mqueue_dir_inode_operations = {
1129         .lookup = simple_lookup,
1130         .create = mqueue_create,
1131         .unlink = mqueue_unlink,
1132 };
1133
1134 static struct file_operations mqueue_file_operations = {
1135         .flush = mqueue_flush_file,
1136         .poll = mqueue_poll_file,
1137         .read = mqueue_read_file,
1138 };
1139
1140 static struct super_operations mqueue_super_ops = {
1141         .alloc_inode = mqueue_alloc_inode,
1142         .destroy_inode = mqueue_destroy_inode,
1143         .statfs = simple_statfs,
1144         .delete_inode = mqueue_delete_inode,
1145         .drop_inode = generic_delete_inode,
1146 };
1147
1148 static struct file_system_type mqueue_fs_type = {
1149         .name = "mqueue",
1150         .get_sb = mqueue_get_sb,
1151         .kill_sb = kill_litter_super,
1152 };
1153
1154 static int msg_max_limit_min = DFLT_MSGMAX;
1155 static int msg_max_limit_max = HARD_MSGMAX;
1156
1157 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1158 static int msg_maxsize_limit_max = INT_MAX;
1159
1160 static ctl_table mq_sysctls[] = {
1161         {
1162                 .ctl_name       = CTL_QUEUESMAX,
1163                 .procname       = "queues_max",
1164                 .data           = &queues_max,
1165                 .maxlen         = sizeof(int),
1166                 .mode           = 0644,
1167                 .proc_handler   = &proc_dointvec,
1168         },
1169         {
1170                 .ctl_name       = CTL_MSGMAX,
1171                 .procname       = "msg_max",
1172                 .data           = &msg_max,
1173                 .maxlen         = sizeof(int),
1174                 .mode           = 0644,
1175                 .proc_handler   = &proc_dointvec_minmax,
1176                 .extra1         = &msg_max_limit_min,
1177                 .extra2         = &msg_max_limit_max,
1178         },
1179         {
1180                 .ctl_name       = CTL_MSGSIZEMAX,
1181                 .procname       = "msgsize_max",
1182                 .data           = &msgsize_max,
1183                 .maxlen         = sizeof(int),
1184                 .mode           = 0644,
1185                 .proc_handler   = &proc_dointvec_minmax,
1186                 .extra1         = &msg_maxsize_limit_min,
1187                 .extra2         = &msg_maxsize_limit_max,
1188         },
1189         { .ctl_name = 0 }
1190 };
1191
1192 static ctl_table mq_sysctl_dir[] = {
1193         {
1194                 .ctl_name       = FS_MQUEUE,
1195                 .procname       = "mqueue",
1196                 .mode           = 0555,
1197                 .child          = mq_sysctls,
1198         },
1199         { .ctl_name = 0 }
1200 };
1201
1202 static ctl_table mq_sysctl_root[] = {
1203         {
1204                 .ctl_name       = CTL_FS,
1205                 .procname       = "fs",
1206                 .mode           = 0555,
1207                 .child          = mq_sysctl_dir,
1208         },
1209         { .ctl_name = 0 }
1210 };
1211
1212 static int __init init_mqueue_fs(void)
1213 {
1214         int error;
1215
1216         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1217                                 sizeof(struct mqueue_inode_info), 0,
1218                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1219         if (mqueue_inode_cachep == NULL)
1220                 return -ENOMEM;
1221
1222         /* ignore failues - they are not fatal */
1223         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1224
1225         error = register_filesystem(&mqueue_fs_type);
1226         if (error)
1227                 goto out_sysctl;
1228
1229         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1230                 error = PTR_ERR(mqueue_mnt);
1231                 goto out_filesystem;
1232         }
1233
1234         /* internal initialization - not common for vfs */
1235         queues_count = 0;
1236         spin_lock_init(&mq_lock);
1237
1238         return 0;
1239
1240 out_filesystem:
1241         unregister_filesystem(&mqueue_fs_type);
1242 out_sysctl:
1243         if (mq_sysctl_table)
1244                 unregister_sysctl_table(mq_sysctl_table);
1245         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1246                 printk(KERN_INFO
1247                         "mqueue_inode_cache: not all structures were freed\n");
1248         }
1249         return error;
1250 }
1251
1252 __initcall(init_mqueue_fs);