]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - ipc/mqueue.c
9a142a290749f2af819c47ac487f2192c6502675
[~shefty/rdma-dev.git] / ipc / mqueue.c
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5  *                          Michal Wronski          (michal.wronski@gmail.com)
6  *
7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8  * Lockless receive & send, fd based notify:
9  *                          Manfred Spraul          (manfred@colorfullife.com)
10  *
11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12  *
13  * This file is released under the GPL.
14  */
15
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32 #include <linux/nsproxy.h>
33 #include <linux/pid.h>
34 #include <linux/ipc_namespace.h>
35 #include <linux/slab.h>
36
37 #include <net/sock.h>
38 #include "util.h"
39
40 #define MQUEUE_MAGIC    0x19800202
41 #define DIRENT_SIZE     20
42 #define FILENT_SIZE     80
43
44 #define SEND            0
45 #define RECV            1
46
47 #define STATE_NONE      0
48 #define STATE_PENDING   1
49 #define STATE_READY     2
50
51 struct ext_wait_queue {         /* queue of sleeping tasks */
52         struct task_struct *task;
53         struct list_head list;
54         struct msg_msg *msg;    /* ptr of loaded message */
55         int state;              /* one of STATE_* values */
56 };
57
58 struct mqueue_inode_info {
59         spinlock_t lock;
60         struct inode vfs_inode;
61         wait_queue_head_t wait_q;
62
63         struct msg_msg **messages;
64         struct mq_attr attr;
65
66         struct sigevent notify;
67         struct pid* notify_owner;
68         struct user_struct *user;       /* user who created, for accounting */
69         struct sock *notify_sock;
70         struct sk_buff *notify_cookie;
71
72         /* for tasks waiting for free space and messages, respectively */
73         struct ext_wait_queue e_wait_q[2];
74
75         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
76 };
77
78 static const struct inode_operations mqueue_dir_inode_operations;
79 static const struct file_operations mqueue_file_operations;
80 static const struct super_operations mqueue_super_ops;
81 static void remove_notification(struct mqueue_inode_info *info);
82
83 static struct kmem_cache *mqueue_inode_cachep;
84
85 static struct ctl_table_header * mq_sysctl_table;
86
87 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
88 {
89         return container_of(inode, struct mqueue_inode_info, vfs_inode);
90 }
91
92 /*
93  * This routine should be called with the mq_lock held.
94  */
95 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
96 {
97         return get_ipc_ns(inode->i_sb->s_fs_info);
98 }
99
100 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
101 {
102         struct ipc_namespace *ns;
103
104         spin_lock(&mq_lock);
105         ns = __get_ns_from_inode(inode);
106         spin_unlock(&mq_lock);
107         return ns;
108 }
109
110 static struct inode *mqueue_get_inode(struct super_block *sb,
111                 struct ipc_namespace *ipc_ns, umode_t mode,
112                 struct mq_attr *attr)
113 {
114         struct user_struct *u = current_user();
115         struct inode *inode;
116         int ret = -ENOMEM;
117
118         inode = new_inode(sb);
119         if (!inode)
120                 goto err;
121
122         inode->i_ino = get_next_ino();
123         inode->i_mode = mode;
124         inode->i_uid = current_fsuid();
125         inode->i_gid = current_fsgid();
126         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
127
128         if (S_ISREG(mode)) {
129                 struct mqueue_inode_info *info;
130                 struct task_struct *p = current;
131                 unsigned long mq_bytes, mq_msg_tblsz;
132
133                 inode->i_fop = &mqueue_file_operations;
134                 inode->i_size = FILENT_SIZE;
135                 /* mqueue specific info */
136                 info = MQUEUE_I(inode);
137                 spin_lock_init(&info->lock);
138                 init_waitqueue_head(&info->wait_q);
139                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
140                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
141                 info->notify_owner = NULL;
142                 info->qsize = 0;
143                 info->user = NULL;      /* set when all is ok */
144                 memset(&info->attr, 0, sizeof(info->attr));
145                 info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
146                 info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
147                 if (attr) {
148                         info->attr.mq_maxmsg = attr->mq_maxmsg;
149                         info->attr.mq_msgsize = attr->mq_msgsize;
150                 }
151                 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
152                 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
153                 if (!info->messages)
154                         goto out_inode;
155
156                 mq_bytes = (mq_msg_tblsz +
157                         (info->attr.mq_maxmsg * info->attr.mq_msgsize));
158
159                 spin_lock(&mq_lock);
160                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
161                     u->mq_bytes + mq_bytes > task_rlimit(p, RLIMIT_MSGQUEUE)) {
162                         spin_unlock(&mq_lock);
163                         /* mqueue_evict_inode() releases info->messages */
164                         ret = -EMFILE;
165                         goto out_inode;
166                 }
167                 u->mq_bytes += mq_bytes;
168                 spin_unlock(&mq_lock);
169
170                 /* all is ok */
171                 info->user = get_uid(u);
172         } else if (S_ISDIR(mode)) {
173                 inc_nlink(inode);
174                 /* Some things misbehave if size == 0 on a directory */
175                 inode->i_size = 2 * DIRENT_SIZE;
176                 inode->i_op = &mqueue_dir_inode_operations;
177                 inode->i_fop = &simple_dir_operations;
178         }
179
180         return inode;
181 out_inode:
182         iput(inode);
183 err:
184         return ERR_PTR(ret);
185 }
186
187 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
188 {
189         struct inode *inode;
190         struct ipc_namespace *ns = data;
191         int error;
192
193         sb->s_blocksize = PAGE_CACHE_SIZE;
194         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
195         sb->s_magic = MQUEUE_MAGIC;
196         sb->s_op = &mqueue_super_ops;
197
198         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
199                                 NULL);
200         if (IS_ERR(inode)) {
201                 error = PTR_ERR(inode);
202                 goto out;
203         }
204
205         sb->s_root = d_alloc_root(inode);
206         if (!sb->s_root) {
207                 iput(inode);
208                 error = -ENOMEM;
209                 goto out;
210         }
211         error = 0;
212
213 out:
214         return error;
215 }
216
217 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
218                          int flags, const char *dev_name,
219                          void *data)
220 {
221         if (!(flags & MS_KERNMOUNT))
222                 data = current->nsproxy->ipc_ns;
223         return mount_ns(fs_type, flags, data, mqueue_fill_super);
224 }
225
226 static void init_once(void *foo)
227 {
228         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
229
230         inode_init_once(&p->vfs_inode);
231 }
232
233 static struct inode *mqueue_alloc_inode(struct super_block *sb)
234 {
235         struct mqueue_inode_info *ei;
236
237         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
238         if (!ei)
239                 return NULL;
240         return &ei->vfs_inode;
241 }
242
243 static void mqueue_i_callback(struct rcu_head *head)
244 {
245         struct inode *inode = container_of(head, struct inode, i_rcu);
246         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
247 }
248
249 static void mqueue_destroy_inode(struct inode *inode)
250 {
251         call_rcu(&inode->i_rcu, mqueue_i_callback);
252 }
253
254 static void mqueue_evict_inode(struct inode *inode)
255 {
256         struct mqueue_inode_info *info;
257         struct user_struct *user;
258         unsigned long mq_bytes;
259         int i;
260         struct ipc_namespace *ipc_ns;
261
262         end_writeback(inode);
263
264         if (S_ISDIR(inode->i_mode))
265                 return;
266
267         ipc_ns = get_ns_from_inode(inode);
268         info = MQUEUE_I(inode);
269         spin_lock(&info->lock);
270         for (i = 0; i < info->attr.mq_curmsgs; i++)
271                 free_msg(info->messages[i]);
272         kfree(info->messages);
273         spin_unlock(&info->lock);
274
275         /* Total amount of bytes accounted for the mqueue */
276         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
277             + info->attr.mq_msgsize);
278         user = info->user;
279         if (user) {
280                 spin_lock(&mq_lock);
281                 user->mq_bytes -= mq_bytes;
282                 /*
283                  * get_ns_from_inode() ensures that the
284                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
285                  * to which we now hold a reference, or it is NULL.
286                  * We can't put it here under mq_lock, though.
287                  */
288                 if (ipc_ns)
289                         ipc_ns->mq_queues_count--;
290                 spin_unlock(&mq_lock);
291                 free_uid(user);
292         }
293         if (ipc_ns)
294                 put_ipc_ns(ipc_ns);
295 }
296
297 static int mqueue_create(struct inode *dir, struct dentry *dentry,
298                                 umode_t mode, struct nameidata *nd)
299 {
300         struct inode *inode;
301         struct mq_attr *attr = dentry->d_fsdata;
302         int error;
303         struct ipc_namespace *ipc_ns;
304
305         spin_lock(&mq_lock);
306         ipc_ns = __get_ns_from_inode(dir);
307         if (!ipc_ns) {
308                 error = -EACCES;
309                 goto out_unlock;
310         }
311         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
312                         !capable(CAP_SYS_RESOURCE)) {
313                 error = -ENOSPC;
314                 goto out_unlock;
315         }
316         ipc_ns->mq_queues_count++;
317         spin_unlock(&mq_lock);
318
319         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
320         if (IS_ERR(inode)) {
321                 error = PTR_ERR(inode);
322                 spin_lock(&mq_lock);
323                 ipc_ns->mq_queues_count--;
324                 goto out_unlock;
325         }
326
327         put_ipc_ns(ipc_ns);
328         dir->i_size += DIRENT_SIZE;
329         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
330
331         d_instantiate(dentry, inode);
332         dget(dentry);
333         return 0;
334 out_unlock:
335         spin_unlock(&mq_lock);
336         if (ipc_ns)
337                 put_ipc_ns(ipc_ns);
338         return error;
339 }
340
341 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
342 {
343         struct inode *inode = dentry->d_inode;
344
345         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
346         dir->i_size -= DIRENT_SIZE;
347         drop_nlink(inode);
348         dput(dentry);
349         return 0;
350 }
351
352 /*
353 *       This is routine for system read from queue file.
354 *       To avoid mess with doing here some sort of mq_receive we allow
355 *       to read only queue size & notification info (the only values
356 *       that are interesting from user point of view and aren't accessible
357 *       through std routines)
358 */
359 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
360                                 size_t count, loff_t *off)
361 {
362         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
363         char buffer[FILENT_SIZE];
364         ssize_t ret;
365
366         spin_lock(&info->lock);
367         snprintf(buffer, sizeof(buffer),
368                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
369                         info->qsize,
370                         info->notify_owner ? info->notify.sigev_notify : 0,
371                         (info->notify_owner &&
372                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
373                                 info->notify.sigev_signo : 0,
374                         pid_vnr(info->notify_owner));
375         spin_unlock(&info->lock);
376         buffer[sizeof(buffer)-1] = '\0';
377
378         ret = simple_read_from_buffer(u_data, count, off, buffer,
379                                 strlen(buffer));
380         if (ret <= 0)
381                 return ret;
382
383         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
384         return ret;
385 }
386
387 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
388 {
389         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
390
391         spin_lock(&info->lock);
392         if (task_tgid(current) == info->notify_owner)
393                 remove_notification(info);
394
395         spin_unlock(&info->lock);
396         return 0;
397 }
398
399 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
400 {
401         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
402         int retval = 0;
403
404         poll_wait(filp, &info->wait_q, poll_tab);
405
406         spin_lock(&info->lock);
407         if (info->attr.mq_curmsgs)
408                 retval = POLLIN | POLLRDNORM;
409
410         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
411                 retval |= POLLOUT | POLLWRNORM;
412         spin_unlock(&info->lock);
413
414         return retval;
415 }
416
417 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
418 static void wq_add(struct mqueue_inode_info *info, int sr,
419                         struct ext_wait_queue *ewp)
420 {
421         struct ext_wait_queue *walk;
422
423         ewp->task = current;
424
425         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
426                 if (walk->task->static_prio <= current->static_prio) {
427                         list_add_tail(&ewp->list, &walk->list);
428                         return;
429                 }
430         }
431         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
432 }
433
434 /*
435  * Puts current task to sleep. Caller must hold queue lock. After return
436  * lock isn't held.
437  * sr: SEND or RECV
438  */
439 static int wq_sleep(struct mqueue_inode_info *info, int sr,
440                     ktime_t *timeout, struct ext_wait_queue *ewp)
441 {
442         int retval;
443         signed long time;
444
445         wq_add(info, sr, ewp);
446
447         for (;;) {
448                 set_current_state(TASK_INTERRUPTIBLE);
449
450                 spin_unlock(&info->lock);
451                 time = schedule_hrtimeout_range_clock(timeout, 0,
452                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
453
454                 while (ewp->state == STATE_PENDING)
455                         cpu_relax();
456
457                 if (ewp->state == STATE_READY) {
458                         retval = 0;
459                         goto out;
460                 }
461                 spin_lock(&info->lock);
462                 if (ewp->state == STATE_READY) {
463                         retval = 0;
464                         goto out_unlock;
465                 }
466                 if (signal_pending(current)) {
467                         retval = -ERESTARTSYS;
468                         break;
469                 }
470                 if (time == 0) {
471                         retval = -ETIMEDOUT;
472                         break;
473                 }
474         }
475         list_del(&ewp->list);
476 out_unlock:
477         spin_unlock(&info->lock);
478 out:
479         return retval;
480 }
481
482 /*
483  * Returns waiting task that should be serviced first or NULL if none exists
484  */
485 static struct ext_wait_queue *wq_get_first_waiter(
486                 struct mqueue_inode_info *info, int sr)
487 {
488         struct list_head *ptr;
489
490         ptr = info->e_wait_q[sr].list.prev;
491         if (ptr == &info->e_wait_q[sr].list)
492                 return NULL;
493         return list_entry(ptr, struct ext_wait_queue, list);
494 }
495
496 /* Auxiliary functions to manipulate messages' list */
497 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
498 {
499         int k;
500
501         k = info->attr.mq_curmsgs - 1;
502         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
503                 info->messages[k + 1] = info->messages[k];
504                 k--;
505         }
506         info->attr.mq_curmsgs++;
507         info->qsize += ptr->m_ts;
508         info->messages[k + 1] = ptr;
509 }
510
511 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
512 {
513         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
514         return info->messages[info->attr.mq_curmsgs];
515 }
516
517 static inline void set_cookie(struct sk_buff *skb, char code)
518 {
519         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
520 }
521
522 /*
523  * The next function is only to split too long sys_mq_timedsend
524  */
525 static void __do_notify(struct mqueue_inode_info *info)
526 {
527         /* notification
528          * invoked when there is registered process and there isn't process
529          * waiting synchronously for message AND state of queue changed from
530          * empty to not empty. Here we are sure that no one is waiting
531          * synchronously. */
532         if (info->notify_owner &&
533             info->attr.mq_curmsgs == 1) {
534                 struct siginfo sig_i;
535                 switch (info->notify.sigev_notify) {
536                 case SIGEV_NONE:
537                         break;
538                 case SIGEV_SIGNAL:
539                         /* sends signal */
540
541                         sig_i.si_signo = info->notify.sigev_signo;
542                         sig_i.si_errno = 0;
543                         sig_i.si_code = SI_MESGQ;
544                         sig_i.si_value = info->notify.sigev_value;
545                         sig_i.si_pid = task_tgid_nr_ns(current,
546                                                 ns_of_pid(info->notify_owner));
547                         sig_i.si_uid = current_uid();
548
549                         kill_pid_info(info->notify.sigev_signo,
550                                       &sig_i, info->notify_owner);
551                         break;
552                 case SIGEV_THREAD:
553                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
554                         netlink_sendskb(info->notify_sock, info->notify_cookie);
555                         break;
556                 }
557                 /* after notification unregisters process */
558                 put_pid(info->notify_owner);
559                 info->notify_owner = NULL;
560         }
561         wake_up(&info->wait_q);
562 }
563
564 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
565                            ktime_t *expires, struct timespec *ts)
566 {
567         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
568                 return -EFAULT;
569         if (!timespec_valid(ts))
570                 return -EINVAL;
571
572         *expires = timespec_to_ktime(*ts);
573         return 0;
574 }
575
576 static void remove_notification(struct mqueue_inode_info *info)
577 {
578         if (info->notify_owner != NULL &&
579             info->notify.sigev_notify == SIGEV_THREAD) {
580                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
581                 netlink_sendskb(info->notify_sock, info->notify_cookie);
582         }
583         put_pid(info->notify_owner);
584         info->notify_owner = NULL;
585 }
586
587 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
588 {
589         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
590                 return 0;
591         if (capable(CAP_SYS_RESOURCE)) {
592                 if (attr->mq_maxmsg > HARD_MSGMAX)
593                         return 0;
594         } else {
595                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
596                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
597                         return 0;
598         }
599         /* check for overflow */
600         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
601                 return 0;
602         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
603             + sizeof (struct msg_msg *))) <
604             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
605                 return 0;
606         return 1;
607 }
608
609 /*
610  * Invoked when creating a new queue via sys_mq_open
611  */
612 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
613                         struct dentry *dentry, int oflag, umode_t mode,
614                         struct mq_attr *attr)
615 {
616         const struct cred *cred = current_cred();
617         struct file *result;
618         int ret;
619
620         if (attr) {
621                 if (!mq_attr_ok(ipc_ns, attr)) {
622                         ret = -EINVAL;
623                         goto out;
624                 }
625                 /* store for use during create */
626                 dentry->d_fsdata = attr;
627         }
628
629         mode &= ~current_umask();
630         ret = mnt_want_write(ipc_ns->mq_mnt);
631         if (ret)
632                 goto out;
633         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
634         dentry->d_fsdata = NULL;
635         if (ret)
636                 goto out_drop_write;
637
638         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
639         /*
640          * dentry_open() took a persistent mnt_want_write(),
641          * so we can now drop this one.
642          */
643         mnt_drop_write(ipc_ns->mq_mnt);
644         return result;
645
646 out_drop_write:
647         mnt_drop_write(ipc_ns->mq_mnt);
648 out:
649         dput(dentry);
650         mntput(ipc_ns->mq_mnt);
651         return ERR_PTR(ret);
652 }
653
654 /* Opens existing queue */
655 static struct file *do_open(struct ipc_namespace *ipc_ns,
656                                 struct dentry *dentry, int oflag)
657 {
658         int ret;
659         const struct cred *cred = current_cred();
660
661         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
662                                                   MAY_READ | MAY_WRITE };
663
664         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
665                 ret = -EINVAL;
666                 goto err;
667         }
668
669         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
670                 ret = -EACCES;
671                 goto err;
672         }
673
674         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
675
676 err:
677         dput(dentry);
678         mntput(ipc_ns->mq_mnt);
679         return ERR_PTR(ret);
680 }
681
682 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
683                 struct mq_attr __user *, u_attr)
684 {
685         struct dentry *dentry;
686         struct file *filp;
687         char *name;
688         struct mq_attr attr;
689         int fd, error;
690         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
691
692         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
693                 return -EFAULT;
694
695         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
696
697         if (IS_ERR(name = getname(u_name)))
698                 return PTR_ERR(name);
699
700         fd = get_unused_fd_flags(O_CLOEXEC);
701         if (fd < 0)
702                 goto out_putname;
703
704         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
705         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
706         if (IS_ERR(dentry)) {
707                 error = PTR_ERR(dentry);
708                 goto out_putfd;
709         }
710         mntget(ipc_ns->mq_mnt);
711
712         if (oflag & O_CREAT) {
713                 if (dentry->d_inode) {  /* entry already exists */
714                         audit_inode(name, dentry);
715                         if (oflag & O_EXCL) {
716                                 error = -EEXIST;
717                                 goto out;
718                         }
719                         filp = do_open(ipc_ns, dentry, oflag);
720                 } else {
721                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
722                                                 dentry, oflag, mode,
723                                                 u_attr ? &attr : NULL);
724                 }
725         } else {
726                 if (!dentry->d_inode) {
727                         error = -ENOENT;
728                         goto out;
729                 }
730                 audit_inode(name, dentry);
731                 filp = do_open(ipc_ns, dentry, oflag);
732         }
733
734         if (IS_ERR(filp)) {
735                 error = PTR_ERR(filp);
736                 goto out_putfd;
737         }
738
739         fd_install(fd, filp);
740         goto out_upsem;
741
742 out:
743         dput(dentry);
744         mntput(ipc_ns->mq_mnt);
745 out_putfd:
746         put_unused_fd(fd);
747         fd = error;
748 out_upsem:
749         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
750 out_putname:
751         putname(name);
752         return fd;
753 }
754
755 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
756 {
757         int err;
758         char *name;
759         struct dentry *dentry;
760         struct inode *inode = NULL;
761         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
762
763         name = getname(u_name);
764         if (IS_ERR(name))
765                 return PTR_ERR(name);
766
767         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
768                         I_MUTEX_PARENT);
769         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
770         if (IS_ERR(dentry)) {
771                 err = PTR_ERR(dentry);
772                 goto out_unlock;
773         }
774
775         if (!dentry->d_inode) {
776                 err = -ENOENT;
777                 goto out_err;
778         }
779
780         inode = dentry->d_inode;
781         if (inode)
782                 ihold(inode);
783         err = mnt_want_write(ipc_ns->mq_mnt);
784         if (err)
785                 goto out_err;
786         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
787         mnt_drop_write(ipc_ns->mq_mnt);
788 out_err:
789         dput(dentry);
790
791 out_unlock:
792         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
793         putname(name);
794         if (inode)
795                 iput(inode);
796
797         return err;
798 }
799
800 /* Pipelined send and receive functions.
801  *
802  * If a receiver finds no waiting message, then it registers itself in the
803  * list of waiting receivers. A sender checks that list before adding the new
804  * message into the message array. If there is a waiting receiver, then it
805  * bypasses the message array and directly hands the message over to the
806  * receiver.
807  * The receiver accepts the message and returns without grabbing the queue
808  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
809  * are necessary. The same algorithm is used for sysv semaphores, see
810  * ipc/sem.c for more details.
811  *
812  * The same algorithm is used for senders.
813  */
814
815 /* pipelined_send() - send a message directly to the task waiting in
816  * sys_mq_timedreceive() (without inserting message into a queue).
817  */
818 static inline void pipelined_send(struct mqueue_inode_info *info,
819                                   struct msg_msg *message,
820                                   struct ext_wait_queue *receiver)
821 {
822         receiver->msg = message;
823         list_del(&receiver->list);
824         receiver->state = STATE_PENDING;
825         wake_up_process(receiver->task);
826         smp_wmb();
827         receiver->state = STATE_READY;
828 }
829
830 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
831  * gets its message and put to the queue (we have one free place for sure). */
832 static inline void pipelined_receive(struct mqueue_inode_info *info)
833 {
834         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
835
836         if (!sender) {
837                 /* for poll */
838                 wake_up_interruptible(&info->wait_q);
839                 return;
840         }
841         msg_insert(sender->msg, info);
842         list_del(&sender->list);
843         sender->state = STATE_PENDING;
844         wake_up_process(sender->task);
845         smp_wmb();
846         sender->state = STATE_READY;
847 }
848
849 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
850                 size_t, msg_len, unsigned int, msg_prio,
851                 const struct timespec __user *, u_abs_timeout)
852 {
853         struct file *filp;
854         struct inode *inode;
855         struct ext_wait_queue wait;
856         struct ext_wait_queue *receiver;
857         struct msg_msg *msg_ptr;
858         struct mqueue_inode_info *info;
859         ktime_t expires, *timeout = NULL;
860         struct timespec ts;
861         int ret;
862
863         if (u_abs_timeout) {
864                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
865                 if (res)
866                         return res;
867                 timeout = &expires;
868         }
869
870         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
871                 return -EINVAL;
872
873         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
874
875         filp = fget(mqdes);
876         if (unlikely(!filp)) {
877                 ret = -EBADF;
878                 goto out;
879         }
880
881         inode = filp->f_path.dentry->d_inode;
882         if (unlikely(filp->f_op != &mqueue_file_operations)) {
883                 ret = -EBADF;
884                 goto out_fput;
885         }
886         info = MQUEUE_I(inode);
887         audit_inode(NULL, filp->f_path.dentry);
888
889         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
890                 ret = -EBADF;
891                 goto out_fput;
892         }
893
894         if (unlikely(msg_len > info->attr.mq_msgsize)) {
895                 ret = -EMSGSIZE;
896                 goto out_fput;
897         }
898
899         /* First try to allocate memory, before doing anything with
900          * existing queues. */
901         msg_ptr = load_msg(u_msg_ptr, msg_len);
902         if (IS_ERR(msg_ptr)) {
903                 ret = PTR_ERR(msg_ptr);
904                 goto out_fput;
905         }
906         msg_ptr->m_ts = msg_len;
907         msg_ptr->m_type = msg_prio;
908
909         spin_lock(&info->lock);
910
911         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
912                 if (filp->f_flags & O_NONBLOCK) {
913                         spin_unlock(&info->lock);
914                         ret = -EAGAIN;
915                 } else {
916                         wait.task = current;
917                         wait.msg = (void *) msg_ptr;
918                         wait.state = STATE_NONE;
919                         ret = wq_sleep(info, SEND, timeout, &wait);
920                 }
921                 if (ret < 0)
922                         free_msg(msg_ptr);
923         } else {
924                 receiver = wq_get_first_waiter(info, RECV);
925                 if (receiver) {
926                         pipelined_send(info, msg_ptr, receiver);
927                 } else {
928                         /* adds message to the queue */
929                         msg_insert(msg_ptr, info);
930                         __do_notify(info);
931                 }
932                 inode->i_atime = inode->i_mtime = inode->i_ctime =
933                                 CURRENT_TIME;
934                 spin_unlock(&info->lock);
935                 ret = 0;
936         }
937 out_fput:
938         fput(filp);
939 out:
940         return ret;
941 }
942
943 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
944                 size_t, msg_len, unsigned int __user *, u_msg_prio,
945                 const struct timespec __user *, u_abs_timeout)
946 {
947         ssize_t ret;
948         struct msg_msg *msg_ptr;
949         struct file *filp;
950         struct inode *inode;
951         struct mqueue_inode_info *info;
952         struct ext_wait_queue wait;
953         ktime_t expires, *timeout = NULL;
954         struct timespec ts;
955
956         if (u_abs_timeout) {
957                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
958                 if (res)
959                         return res;
960                 timeout = &expires;
961         }
962
963         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
964
965         filp = fget(mqdes);
966         if (unlikely(!filp)) {
967                 ret = -EBADF;
968                 goto out;
969         }
970
971         inode = filp->f_path.dentry->d_inode;
972         if (unlikely(filp->f_op != &mqueue_file_operations)) {
973                 ret = -EBADF;
974                 goto out_fput;
975         }
976         info = MQUEUE_I(inode);
977         audit_inode(NULL, filp->f_path.dentry);
978
979         if (unlikely(!(filp->f_mode & FMODE_READ))) {
980                 ret = -EBADF;
981                 goto out_fput;
982         }
983
984         /* checks if buffer is big enough */
985         if (unlikely(msg_len < info->attr.mq_msgsize)) {
986                 ret = -EMSGSIZE;
987                 goto out_fput;
988         }
989
990         spin_lock(&info->lock);
991         if (info->attr.mq_curmsgs == 0) {
992                 if (filp->f_flags & O_NONBLOCK) {
993                         spin_unlock(&info->lock);
994                         ret = -EAGAIN;
995                 } else {
996                         wait.task = current;
997                         wait.state = STATE_NONE;
998                         ret = wq_sleep(info, RECV, timeout, &wait);
999                         msg_ptr = wait.msg;
1000                 }
1001         } else {
1002                 msg_ptr = msg_get(info);
1003
1004                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1005                                 CURRENT_TIME;
1006
1007                 /* There is now free space in queue. */
1008                 pipelined_receive(info);
1009                 spin_unlock(&info->lock);
1010                 ret = 0;
1011         }
1012         if (ret == 0) {
1013                 ret = msg_ptr->m_ts;
1014
1015                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1016                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1017                         ret = -EFAULT;
1018                 }
1019                 free_msg(msg_ptr);
1020         }
1021 out_fput:
1022         fput(filp);
1023 out:
1024         return ret;
1025 }
1026
1027 /*
1028  * Notes: the case when user wants us to deregister (with NULL as pointer)
1029  * and he isn't currently owner of notification, will be silently discarded.
1030  * It isn't explicitly defined in the POSIX.
1031  */
1032 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1033                 const struct sigevent __user *, u_notification)
1034 {
1035         int ret;
1036         struct file *filp;
1037         struct sock *sock;
1038         struct inode *inode;
1039         struct sigevent notification;
1040         struct mqueue_inode_info *info;
1041         struct sk_buff *nc;
1042
1043         if (u_notification) {
1044                 if (copy_from_user(&notification, u_notification,
1045                                         sizeof(struct sigevent)))
1046                         return -EFAULT;
1047         }
1048
1049         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1050
1051         nc = NULL;
1052         sock = NULL;
1053         if (u_notification != NULL) {
1054                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1055                              notification.sigev_notify != SIGEV_SIGNAL &&
1056                              notification.sigev_notify != SIGEV_THREAD))
1057                         return -EINVAL;
1058                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1059                         !valid_signal(notification.sigev_signo)) {
1060                         return -EINVAL;
1061                 }
1062                 if (notification.sigev_notify == SIGEV_THREAD) {
1063                         long timeo;
1064
1065                         /* create the notify skb */
1066                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1067                         if (!nc) {
1068                                 ret = -ENOMEM;
1069                                 goto out;
1070                         }
1071                         if (copy_from_user(nc->data,
1072                                         notification.sigev_value.sival_ptr,
1073                                         NOTIFY_COOKIE_LEN)) {
1074                                 ret = -EFAULT;
1075                                 goto out;
1076                         }
1077
1078                         /* TODO: add a header? */
1079                         skb_put(nc, NOTIFY_COOKIE_LEN);
1080                         /* and attach it to the socket */
1081 retry:
1082                         filp = fget(notification.sigev_signo);
1083                         if (!filp) {
1084                                 ret = -EBADF;
1085                                 goto out;
1086                         }
1087                         sock = netlink_getsockbyfilp(filp);
1088                         fput(filp);
1089                         if (IS_ERR(sock)) {
1090                                 ret = PTR_ERR(sock);
1091                                 sock = NULL;
1092                                 goto out;
1093                         }
1094
1095                         timeo = MAX_SCHEDULE_TIMEOUT;
1096                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1097                         if (ret == 1)
1098                                 goto retry;
1099                         if (ret) {
1100                                 sock = NULL;
1101                                 nc = NULL;
1102                                 goto out;
1103                         }
1104                 }
1105         }
1106
1107         filp = fget(mqdes);
1108         if (!filp) {
1109                 ret = -EBADF;
1110                 goto out;
1111         }
1112
1113         inode = filp->f_path.dentry->d_inode;
1114         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1115                 ret = -EBADF;
1116                 goto out_fput;
1117         }
1118         info = MQUEUE_I(inode);
1119
1120         ret = 0;
1121         spin_lock(&info->lock);
1122         if (u_notification == NULL) {
1123                 if (info->notify_owner == task_tgid(current)) {
1124                         remove_notification(info);
1125                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1126                 }
1127         } else if (info->notify_owner != NULL) {
1128                 ret = -EBUSY;
1129         } else {
1130                 switch (notification.sigev_notify) {
1131                 case SIGEV_NONE:
1132                         info->notify.sigev_notify = SIGEV_NONE;
1133                         break;
1134                 case SIGEV_THREAD:
1135                         info->notify_sock = sock;
1136                         info->notify_cookie = nc;
1137                         sock = NULL;
1138                         nc = NULL;
1139                         info->notify.sigev_notify = SIGEV_THREAD;
1140                         break;
1141                 case SIGEV_SIGNAL:
1142                         info->notify.sigev_signo = notification.sigev_signo;
1143                         info->notify.sigev_value = notification.sigev_value;
1144                         info->notify.sigev_notify = SIGEV_SIGNAL;
1145                         break;
1146                 }
1147
1148                 info->notify_owner = get_pid(task_tgid(current));
1149                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1150         }
1151         spin_unlock(&info->lock);
1152 out_fput:
1153         fput(filp);
1154 out:
1155         if (sock) {
1156                 netlink_detachskb(sock, nc);
1157         } else if (nc) {
1158                 dev_kfree_skb(nc);
1159         }
1160         return ret;
1161 }
1162
1163 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1164                 const struct mq_attr __user *, u_mqstat,
1165                 struct mq_attr __user *, u_omqstat)
1166 {
1167         int ret;
1168         struct mq_attr mqstat, omqstat;
1169         struct file *filp;
1170         struct inode *inode;
1171         struct mqueue_inode_info *info;
1172
1173         if (u_mqstat != NULL) {
1174                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1175                         return -EFAULT;
1176                 if (mqstat.mq_flags & (~O_NONBLOCK))
1177                         return -EINVAL;
1178         }
1179
1180         filp = fget(mqdes);
1181         if (!filp) {
1182                 ret = -EBADF;
1183                 goto out;
1184         }
1185
1186         inode = filp->f_path.dentry->d_inode;
1187         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1188                 ret = -EBADF;
1189                 goto out_fput;
1190         }
1191         info = MQUEUE_I(inode);
1192
1193         spin_lock(&info->lock);
1194
1195         omqstat = info->attr;
1196         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1197         if (u_mqstat) {
1198                 audit_mq_getsetattr(mqdes, &mqstat);
1199                 spin_lock(&filp->f_lock);
1200                 if (mqstat.mq_flags & O_NONBLOCK)
1201                         filp->f_flags |= O_NONBLOCK;
1202                 else
1203                         filp->f_flags &= ~O_NONBLOCK;
1204                 spin_unlock(&filp->f_lock);
1205
1206                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1207         }
1208
1209         spin_unlock(&info->lock);
1210
1211         ret = 0;
1212         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1213                                                 sizeof(struct mq_attr)))
1214                 ret = -EFAULT;
1215
1216 out_fput:
1217         fput(filp);
1218 out:
1219         return ret;
1220 }
1221
1222 static const struct inode_operations mqueue_dir_inode_operations = {
1223         .lookup = simple_lookup,
1224         .create = mqueue_create,
1225         .unlink = mqueue_unlink,
1226 };
1227
1228 static const struct file_operations mqueue_file_operations = {
1229         .flush = mqueue_flush_file,
1230         .poll = mqueue_poll_file,
1231         .read = mqueue_read_file,
1232         .llseek = default_llseek,
1233 };
1234
1235 static const struct super_operations mqueue_super_ops = {
1236         .alloc_inode = mqueue_alloc_inode,
1237         .destroy_inode = mqueue_destroy_inode,
1238         .evict_inode = mqueue_evict_inode,
1239         .statfs = simple_statfs,
1240 };
1241
1242 static struct file_system_type mqueue_fs_type = {
1243         .name = "mqueue",
1244         .mount = mqueue_mount,
1245         .kill_sb = kill_litter_super,
1246 };
1247
1248 int mq_init_ns(struct ipc_namespace *ns)
1249 {
1250         ns->mq_queues_count  = 0;
1251         ns->mq_queues_max    = DFLT_QUEUESMAX;
1252         ns->mq_msg_max       = DFLT_MSGMAX;
1253         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1254
1255         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1256         if (IS_ERR(ns->mq_mnt)) {
1257                 int err = PTR_ERR(ns->mq_mnt);
1258                 ns->mq_mnt = NULL;
1259                 return err;
1260         }
1261         return 0;
1262 }
1263
1264 void mq_clear_sbinfo(struct ipc_namespace *ns)
1265 {
1266         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1267 }
1268
1269 void mq_put_mnt(struct ipc_namespace *ns)
1270 {
1271         kern_unmount(ns->mq_mnt);
1272 }
1273
1274 static int __init init_mqueue_fs(void)
1275 {
1276         int error;
1277
1278         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1279                                 sizeof(struct mqueue_inode_info), 0,
1280                                 SLAB_HWCACHE_ALIGN, init_once);
1281         if (mqueue_inode_cachep == NULL)
1282                 return -ENOMEM;
1283
1284         /* ignore failures - they are not fatal */
1285         mq_sysctl_table = mq_register_sysctl_table();
1286
1287         error = register_filesystem(&mqueue_fs_type);
1288         if (error)
1289                 goto out_sysctl;
1290
1291         spin_lock_init(&mq_lock);
1292
1293         error = mq_init_ns(&init_ipc_ns);
1294         if (error)
1295                 goto out_filesystem;
1296
1297         return 0;
1298
1299 out_filesystem:
1300         unregister_filesystem(&mqueue_fs_type);
1301 out_sysctl:
1302         if (mq_sysctl_table)
1303                 unregister_sysctl_table(mq_sysctl_table);
1304         kmem_cache_destroy(mqueue_inode_cachep);
1305         return error;
1306 }
1307
1308 __initcall(init_mqueue_fs);