?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
PCI: Fix pci_slot_release() NULL pointer dereference
[linux/fpc-iii.git] / ipc / mqueue.c
blob3f7dc5f341f7ed5d20610b6835229fc73e4d47a0
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/vmalloc.h>
28 #include <linux/netlink.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/signal.h>
32 #include <linux/mutex.h>
33 #include <linux/nsproxy.h>
34 #include <linux/pid.h>
35 #include <linux/ipc_namespace.h>
36 #include <linux/user_namespace.h>
37 #include <linux/slab.h>
38
39 #include <net/sock.h>
40 #include "util.h"
41
42 #define MQUEUE_MAGIC 0x19800202
43 #define DIRENT_SIZE 20
44 #define FILENT_SIZE 80
45
46 #define SEND 0
47 #define RECV 1
48
49 #define STATE_NONE 0
50 #define STATE_READY 1
51
52 struct posix_msg_tree_node {
53 struct rb_node rb_node;
54 struct list_head msg_list;
55 int priority;
56 };
57
58 struct ext_wait_queue { /* queue of sleeping tasks */
59 struct task_struct *task;
60 struct list_head list;
61 struct msg_msg *msg; /* ptr of loaded message */
62 int state; /* one of STATE_* values */
63 };
64
65 struct mqueue_inode_info {
66 spinlock_t lock;
67 struct inode vfs_inode;
68 wait_queue_head_t wait_q;
69
70 struct rb_root msg_tree;
71 struct posix_msg_tree_node *node_cache;
72 struct mq_attr attr;
73
74 struct sigevent notify;
75 struct pid *notify_owner;
76 struct user_namespace *notify_user_ns;
77 struct user_struct *user; /* user who created, for accounting */
78 struct sock *notify_sock;
79 struct sk_buff *notify_cookie;
80
81 /* for tasks waiting for free space and messages, respectively */
82 struct ext_wait_queue e_wait_q[2];
83
84 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
85 };
86
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static const struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
91
92 static struct kmem_cache *mqueue_inode_cachep;
93
94 static struct ctl_table_header *mq_sysctl_table;
95
96 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
97 {
98 return container_of(inode, struct mqueue_inode_info, vfs_inode);
99 }
100
101 /*
102 * This routine should be called with the mq_lock held.
103 */
104 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
105 {
106 return get_ipc_ns(inode->i_sb->s_fs_info);
107 }
108
109 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
110 {
111 struct ipc_namespace *ns;
112
113 spin_lock(&mq_lock);
114 ns = __get_ns_from_inode(inode);
115 spin_unlock(&mq_lock);
116 return ns;
117 }
118
119 /* Auxiliary functions to manipulate messages' list */
120 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
121 {
122 struct rb_node **p, *parent = NULL;
123 struct posix_msg_tree_node *leaf;
124
125 p = &info->msg_tree.rb_node;
126 while (*p) {
127 parent = *p;
128 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
129
130 if (likely(leaf->priority == msg->m_type))
131 goto insert_msg;
132 else if (msg->m_type < leaf->priority)
133 p = &(*p)->rb_left;
134 else
135 p = &(*p)->rb_right;
136 }
137 if (info->node_cache) {
138 leaf = info->node_cache;
139 info->node_cache = NULL;
140 } else {
141 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
142 if (!leaf)
143 return -ENOMEM;
144 INIT_LIST_HEAD(&leaf->msg_list);
145 }
146 leaf->priority = msg->m_type;
147 rb_link_node(&leaf->rb_node, parent, p);
148 rb_insert_color(&leaf->rb_node, &info->msg_tree);
149 insert_msg:
150 info->attr.mq_curmsgs++;
151 info->qsize += msg->m_ts;
152 list_add_tail(&msg->m_list, &leaf->msg_list);
153 return 0;
154 }
155
156 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
157 {
158 struct rb_node **p, *parent = NULL;
159 struct posix_msg_tree_node *leaf;
160 struct msg_msg *msg;
161
162 try_again:
163 p = &info->msg_tree.rb_node;
164 while (*p) {
165 parent = *p;
166 /*
167 * During insert, low priorities go to the left and high to the
168 * right. On receive, we want the highest priorities first, so
169 * walk all the way to the right.
170 */
171 p = &(*p)->rb_right;
172 }
173 if (!parent) {
174 if (info->attr.mq_curmsgs) {
175 pr_warn_once("Inconsistency in POSIX message queue, "
176 "no tree element, but supposedly messages "
177 "should exist!\n");
178 info->attr.mq_curmsgs = 0;
179 }
180 return NULL;
181 }
182 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
183 if (unlikely(list_empty(&leaf->msg_list))) {
184 pr_warn_once("Inconsistency in POSIX message queue, "
185 "empty leaf node but we haven't implemented "
186 "lazy leaf delete!\n");
187 rb_erase(&leaf->rb_node, &info->msg_tree);
188 if (info->node_cache) {
189 kfree(leaf);
190 } else {
191 info->node_cache = leaf;
192 }
193 goto try_again;
194 } else {
195 msg = list_first_entry(&leaf->msg_list,
196 struct msg_msg, m_list);
197 list_del(&msg->m_list);
198 if (list_empty(&leaf->msg_list)) {
199 rb_erase(&leaf->rb_node, &info->msg_tree);
200 if (info->node_cache) {
201 kfree(leaf);
202 } else {
203 info->node_cache = leaf;
204 }
205 }
206 }
207 info->attr.mq_curmsgs--;
208 info->qsize -= msg->m_ts;
209 return msg;
210 }
211
212 static struct inode *mqueue_get_inode(struct super_block *sb,
213 struct ipc_namespace *ipc_ns, umode_t mode,
214 struct mq_attr *attr)
215 {
216 struct user_struct *u = current_user();
217 struct inode *inode;
218 int ret = -ENOMEM;
219
220 inode = new_inode(sb);
221 if (!inode)
222 goto err;
223
224 inode->i_ino = get_next_ino();
225 inode->i_mode = mode;
226 inode->i_uid = current_fsuid();
227 inode->i_gid = current_fsgid();
228 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
229
230 if (S_ISREG(mode)) {
231 struct mqueue_inode_info *info;
232 unsigned long mq_bytes, mq_treesize;
233
234 inode->i_fop = &mqueue_file_operations;
235 inode->i_size = FILENT_SIZE;
236 /* mqueue specific info */
237 info = MQUEUE_I(inode);
238 spin_lock_init(&info->lock);
239 init_waitqueue_head(&info->wait_q);
240 INIT_LIST_HEAD(&info->e_wait_q[0].list);
241 INIT_LIST_HEAD(&info->e_wait_q[1].list);
242 info->notify_owner = NULL;
243 info->notify_user_ns = NULL;
244 info->qsize = 0;
245 info->user = NULL; /* set when all is ok */
246 info->msg_tree = RB_ROOT;
247 info->node_cache = NULL;
248 memset(&info->attr, 0, sizeof(info->attr));
249 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
250 ipc_ns->mq_msg_default);
251 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
252 ipc_ns->mq_msgsize_default);
253 if (attr) {
254 info->attr.mq_maxmsg = attr->mq_maxmsg;
255 info->attr.mq_msgsize = attr->mq_msgsize;
256 }
257 /*
258 * We used to allocate a static array of pointers and account
259 * the size of that array as well as one msg_msg struct per
260 * possible message into the queue size. That's no longer
261 * accurate as the queue is now an rbtree and will grow and
262 * shrink depending on usage patterns. We can, however, still
263 * account one msg_msg struct per message, but the nodes are
264 * allocated depending on priority usage, and most programs
265 * only use one, or a handful, of priorities. However, since
266 * this is pinned memory, we need to assume worst case, so
267 * that means the min(mq_maxmsg, max_priorities) * struct
268 * posix_msg_tree_node.
269 */
270 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
271 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
272 sizeof(struct posix_msg_tree_node);
273
274 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
275 info->attr.mq_msgsize);
276
277 spin_lock(&mq_lock);
278 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
279 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
280 spin_unlock(&mq_lock);
281 /* mqueue_evict_inode() releases info->messages */
282 ret = -EMFILE;
283 goto out_inode;
284 }
285 u->mq_bytes += mq_bytes;
286 spin_unlock(&mq_lock);
287
288 /* all is ok */
289 info->user = get_uid(u);
290 } else if (S_ISDIR(mode)) {
291 inc_nlink(inode);
292 /* Some things misbehave if size == 0 on a directory */
293 inode->i_size = 2 * DIRENT_SIZE;
294 inode->i_op = &mqueue_dir_inode_operations;
295 inode->i_fop = &simple_dir_operations;
296 }
297
298 return inode;
299 out_inode:
300 iput(inode);
301 err:
302 return ERR_PTR(ret);
303 }
304
305 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
306 {
307 struct inode *inode;
308 struct ipc_namespace *ns = sb->s_fs_info;
309
310 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
311 sb->s_blocksize = PAGE_SIZE;
312 sb->s_blocksize_bits = PAGE_SHIFT;
313 sb->s_magic = MQUEUE_MAGIC;
314 sb->s_op = &mqueue_super_ops;
315
316 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
317 if (IS_ERR(inode))
318 return PTR_ERR(inode);
319
320 sb->s_root = d_make_root(inode);
321 if (!sb->s_root)
322 return -ENOMEM;
323 return 0;
324 }
325
326 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
327 int flags, const char *dev_name,
328 void *data)
329 {
330 struct ipc_namespace *ns;
331 if (flags & MS_KERNMOUNT) {
332 ns = data;
333 data = NULL;
334 } else {
335 ns = current->nsproxy->ipc_ns;
336 }
337 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
338 }
339
340 static void init_once(void *foo)
341 {
342 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
343
344 inode_init_once(&p->vfs_inode);
345 }
346
347 static struct inode *mqueue_alloc_inode(struct super_block *sb)
348 {
349 struct mqueue_inode_info *ei;
350
351 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
352 if (!ei)
353 return NULL;
354 return &ei->vfs_inode;
355 }
356
357 static void mqueue_i_callback(struct rcu_head *head)
358 {
359 struct inode *inode = container_of(head, struct inode, i_rcu);
360 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
361 }
362
363 static void mqueue_destroy_inode(struct inode *inode)
364 {
365 call_rcu(&inode->i_rcu, mqueue_i_callback);
366 }
367
368 static void mqueue_evict_inode(struct inode *inode)
369 {
370 struct mqueue_inode_info *info;
371 struct user_struct *user;
372 struct ipc_namespace *ipc_ns;
373 struct msg_msg *msg, *nmsg;
374 LIST_HEAD(tmp_msg);
375
376 clear_inode(inode);
377
378 if (S_ISDIR(inode->i_mode))
379 return;
380
381 ipc_ns = get_ns_from_inode(inode);
382 info = MQUEUE_I(inode);
383 spin_lock(&info->lock);
384 while ((msg = msg_get(info)) != NULL)
385 list_add_tail(&msg->m_list, &tmp_msg);
386 kfree(info->node_cache);
387 spin_unlock(&info->lock);
388
389 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
390 list_del(&msg->m_list);
391 free_msg(msg);
392 }
393
394 user = info->user;
395 if (user) {
396 unsigned long mq_bytes, mq_treesize;
397
398 /* Total amount of bytes accounted for the mqueue */
399 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
400 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
401 sizeof(struct posix_msg_tree_node);
402
403 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
404 info->attr.mq_msgsize);
405
406 spin_lock(&mq_lock);
407 user->mq_bytes -= mq_bytes;
408 /*
409 * get_ns_from_inode() ensures that the
410 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
411 * to which we now hold a reference, or it is NULL.
412 * We can't put it here under mq_lock, though.
413 */
414 if (ipc_ns)
415 ipc_ns->mq_queues_count--;
416 spin_unlock(&mq_lock);
417 free_uid(user);
418 }
419 if (ipc_ns)
420 put_ipc_ns(ipc_ns);
421 }
422
423 static int mqueue_create(struct inode *dir, struct dentry *dentry,
424 umode_t mode, bool excl)
425 {
426 struct inode *inode;
427 struct mq_attr *attr = dentry->d_fsdata;
428 int error;
429 struct ipc_namespace *ipc_ns;
430
431 spin_lock(&mq_lock);
432 ipc_ns = __get_ns_from_inode(dir);
433 if (!ipc_ns) {
434 error = -EACCES;
435 goto out_unlock;
436 }
437
438 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
439 !capable(CAP_SYS_RESOURCE)) {
440 error = -ENOSPC;
441 goto out_unlock;
442 }
443 ipc_ns->mq_queues_count++;
444 spin_unlock(&mq_lock);
445
446 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
447 if (IS_ERR(inode)) {
448 error = PTR_ERR(inode);
449 spin_lock(&mq_lock);
450 ipc_ns->mq_queues_count--;
451 goto out_unlock;
452 }
453
454 put_ipc_ns(ipc_ns);
455 dir->i_size += DIRENT_SIZE;
456 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
457
458 d_instantiate(dentry, inode);
459 dget(dentry);
460 return 0;
461 out_unlock:
462 spin_unlock(&mq_lock);
463 if (ipc_ns)
464 put_ipc_ns(ipc_ns);
465 return error;
466 }
467
468 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
469 {
470 struct inode *inode = d_inode(dentry);
471
472 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
473 dir->i_size -= DIRENT_SIZE;
474 drop_nlink(inode);
475 dput(dentry);
476 return 0;
477 }
478
479 /*
480 * This is routine for system read from queue file.
481 * To avoid mess with doing here some sort of mq_receive we allow
482 * to read only queue size & notification info (the only values
483 * that are interesting from user point of view and aren't accessible
484 * through std routines)
485 */
486 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
487 size_t count, loff_t *off)
488 {
489 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
490 char buffer[FILENT_SIZE];
491 ssize_t ret;
492
493 spin_lock(&info->lock);
494 snprintf(buffer, sizeof(buffer),
495 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
496 info->qsize,
497 info->notify_owner ? info->notify.sigev_notify : 0,
498 (info->notify_owner &&
499 info->notify.sigev_notify == SIGEV_SIGNAL) ?
500 info->notify.sigev_signo : 0,
501 pid_vnr(info->notify_owner));
502 spin_unlock(&info->lock);
503 buffer[sizeof(buffer)-1] = '\0';
504
505 ret = simple_read_from_buffer(u_data, count, off, buffer,
506 strlen(buffer));
507 if (ret <= 0)
508 return ret;
509
510 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
511 return ret;
512 }
513
514 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
515 {
516 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
517
518 spin_lock(&info->lock);
519 if (task_tgid(current) == info->notify_owner)
520 remove_notification(info);
521
522 spin_unlock(&info->lock);
523 return 0;
524 }
525
526 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
527 {
528 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
529 int retval = 0;
530
531 poll_wait(filp, &info->wait_q, poll_tab);
532
533 spin_lock(&info->lock);
534 if (info->attr.mq_curmsgs)
535 retval = POLLIN | POLLRDNORM;
536
537 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
538 retval |= POLLOUT | POLLWRNORM;
539 spin_unlock(&info->lock);
540
541 return retval;
542 }
543
544 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
545 static void wq_add(struct mqueue_inode_info *info, int sr,
546 struct ext_wait_queue *ewp)
547 {
548 struct ext_wait_queue *walk;
549
550 ewp->task = current;
551
552 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
553 if (walk->task->static_prio <= current->static_prio) {
554 list_add_tail(&ewp->list, &walk->list);
555 return;
556 }
557 }
558 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
559 }
560
561 /*
562 * Puts current task to sleep. Caller must hold queue lock. After return
563 * lock isn't held.
564 * sr: SEND or RECV
565 */
566 static int wq_sleep(struct mqueue_inode_info *info, int sr,
567 ktime_t *timeout, struct ext_wait_queue *ewp)
568 {
569 int retval;
570 signed long time;
571
572 wq_add(info, sr, ewp);
573
574 for (;;) {
575 __set_current_state(TASK_INTERRUPTIBLE);
576
577 spin_unlock(&info->lock);
578 time = schedule_hrtimeout_range_clock(timeout, 0,
579 HRTIMER_MODE_ABS, CLOCK_REALTIME);
580
581 if (ewp->state == STATE_READY) {
582 retval = 0;
583 goto out;
584 }
585 spin_lock(&info->lock);
586 if (ewp->state == STATE_READY) {
587 retval = 0;
588 goto out_unlock;
589 }
590 if (signal_pending(current)) {
591 retval = -ERESTARTSYS;
592 break;
593 }
594 if (time == 0) {
595 retval = -ETIMEDOUT;
596 break;
597 }
598 }
599 list_del(&ewp->list);
600 out_unlock:
601 spin_unlock(&info->lock);
602 out:
603 return retval;
604 }
605
606 /*
607 * Returns waiting task that should be serviced first or NULL if none exists
608 */
609 static struct ext_wait_queue *wq_get_first_waiter(
610 struct mqueue_inode_info *info, int sr)
611 {
612 struct list_head *ptr;
613
614 ptr = info->e_wait_q[sr].list.prev;
615 if (ptr == &info->e_wait_q[sr].list)
616 return NULL;
617 return list_entry(ptr, struct ext_wait_queue, list);
618 }
619
620
621 static inline void set_cookie(struct sk_buff *skb, char code)
622 {
623 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
624 }
625
626 /*
627 * The next function is only to split too long sys_mq_timedsend
628 */
629 static void __do_notify(struct mqueue_inode_info *info)
630 {
631 /* notification
632 * invoked when there is registered process and there isn't process
633 * waiting synchronously for message AND state of queue changed from
634 * empty to not empty. Here we are sure that no one is waiting
635 * synchronously. */
636 if (info->notify_owner &&
637 info->attr.mq_curmsgs == 1) {
638 struct siginfo sig_i;
639 switch (info->notify.sigev_notify) {
640 case SIGEV_NONE:
641 break;
642 case SIGEV_SIGNAL:
643 /* sends signal */
644
645 sig_i.si_signo = info->notify.sigev_signo;
646 sig_i.si_errno = 0;
647 sig_i.si_code = SI_MESGQ;
648 sig_i.si_value = info->notify.sigev_value;
649 /* map current pid/uid into info->owner's namespaces */
650 rcu_read_lock();
651 sig_i.si_pid = task_tgid_nr_ns(current,
652 ns_of_pid(info->notify_owner));
653 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
654 rcu_read_unlock();
655
656 kill_pid_info(info->notify.sigev_signo,
657 &sig_i, info->notify_owner);
658 break;
659 case SIGEV_THREAD:
660 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
661 netlink_sendskb(info->notify_sock, info->notify_cookie);
662 break;
663 }
664 /* after notification unregisters process */
665 put_pid(info->notify_owner);
666 put_user_ns(info->notify_user_ns);
667 info->notify_owner = NULL;
668 info->notify_user_ns = NULL;
669 }
670 wake_up(&info->wait_q);
671 }
672
673 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
674 ktime_t *expires, struct timespec *ts)
675 {
676 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
677 return -EFAULT;
678 if (!timespec_valid(ts))
679 return -EINVAL;
680
681 *expires = timespec_to_ktime(*ts);
682 return 0;
683 }
684
685 static void remove_notification(struct mqueue_inode_info *info)
686 {
687 if (info->notify_owner != NULL &&
688 info->notify.sigev_notify == SIGEV_THREAD) {
689 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
690 netlink_sendskb(info->notify_sock, info->notify_cookie);
691 }
692 put_pid(info->notify_owner);
693 put_user_ns(info->notify_user_ns);
694 info->notify_owner = NULL;
695 info->notify_user_ns = NULL;
696 }
697
698 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
699 {
700 int mq_treesize;
701 unsigned long total_size;
702
703 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
704 return -EINVAL;
705 if (capable(CAP_SYS_RESOURCE)) {
706 if (attr->mq_maxmsg > HARD_MSGMAX ||
707 attr->mq_msgsize > HARD_MSGSIZEMAX)
708 return -EINVAL;
709 } else {
710 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
711 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
712 return -EINVAL;
713 }
714 /* check for overflow */
715 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
716 return -EOVERFLOW;
717 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
718 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
719 sizeof(struct posix_msg_tree_node);
720 total_size = attr->mq_maxmsg * attr->mq_msgsize;
721 if (total_size + mq_treesize < total_size)
722 return -EOVERFLOW;
723 return 0;
724 }
725
726 /*
727 * Invoked when creating a new queue via sys_mq_open
728 */
729 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
730 struct path *path, int oflag, umode_t mode,
731 struct mq_attr *attr)
732 {
733 const struct cred *cred = current_cred();
734 int ret;
735
736 if (attr) {
737 ret = mq_attr_ok(ipc_ns, attr);
738 if (ret)
739 return ERR_PTR(ret);
740 /* store for use during create */
741 path->dentry->d_fsdata = attr;
742 } else {
743 struct mq_attr def_attr;
744
745 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
746 ipc_ns->mq_msg_default);
747 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
748 ipc_ns->mq_msgsize_default);
749 ret = mq_attr_ok(ipc_ns, &def_attr);
750 if (ret)
751 return ERR_PTR(ret);
752 }
753
754 mode &= ~current_umask();
755 ret = vfs_create(dir, path->dentry, mode, true);
756 path->dentry->d_fsdata = NULL;
757 if (ret)
758 return ERR_PTR(ret);
759 return dentry_open(path, oflag, cred);
760 }
761
762 /* Opens existing queue */
763 static struct file *do_open(struct path *path, int oflag)
764 {
765 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
766 MAY_READ | MAY_WRITE };
767 int acc;
768 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
769 return ERR_PTR(-EINVAL);
770 acc = oflag2acc[oflag & O_ACCMODE];
771 if (inode_permission(d_inode(path->dentry), acc))
772 return ERR_PTR(-EACCES);
773 return dentry_open(path, oflag, current_cred());
774 }
775
776 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
777 struct mq_attr __user *, u_attr)
778 {
779 struct path path;
780 struct file *filp;
781 struct filename *name;
782 struct mq_attr attr;
783 int fd, error;
784 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
785 struct vfsmount *mnt = ipc_ns->mq_mnt;
786 struct dentry *root = mnt->mnt_root;
787 int ro;
788
789 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
790 return -EFAULT;
791
792 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
793
794 if (IS_ERR(name = getname(u_name)))
795 return PTR_ERR(name);
796
797 fd = get_unused_fd_flags(O_CLOEXEC);
798 if (fd < 0)
799 goto out_putname;
800
801 ro = mnt_want_write(mnt); /* we'll drop it in any case */
802 error = 0;
803 inode_lock(d_inode(root));
804 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
805 if (IS_ERR(path.dentry)) {
806 error = PTR_ERR(path.dentry);
807 goto out_putfd;
808 }
809 path.mnt = mntget(mnt);
810
811 if (oflag & O_CREAT) {
812 if (d_really_is_positive(path.dentry)) { /* entry already exists */
813 audit_inode(name, path.dentry, 0);
814 if (oflag & O_EXCL) {
815 error = -EEXIST;
816 goto out;
817 }
818 filp = do_open(&path, oflag);
819 } else {
820 if (ro) {
821 error = ro;
822 goto out;
823 }
824 audit_inode_parent_hidden(name, root);
825 filp = do_create(ipc_ns, d_inode(root),
826 &path, oflag, mode,
827 u_attr ? &attr : NULL);
828 }
829 } else {
830 if (d_really_is_negative(path.dentry)) {
831 error = -ENOENT;
832 goto out;
833 }
834 audit_inode(name, path.dentry, 0);
835 filp = do_open(&path, oflag);
836 }
837
838 if (!IS_ERR(filp))
839 fd_install(fd, filp);
840 else
841 error = PTR_ERR(filp);
842 out:
843 path_put(&path);
844 out_putfd:
845 if (error) {
846 put_unused_fd(fd);
847 fd = error;
848 }
849 inode_unlock(d_inode(root));
850 if (!ro)
851 mnt_drop_write(mnt);
852 out_putname:
853 putname(name);
854 return fd;
855 }
856
857 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
858 {
859 int err;
860 struct filename *name;
861 struct dentry *dentry;
862 struct inode *inode = NULL;
863 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
864 struct vfsmount *mnt = ipc_ns->mq_mnt;
865
866 name = getname(u_name);
867 if (IS_ERR(name))
868 return PTR_ERR(name);
869
870 audit_inode_parent_hidden(name, mnt->mnt_root);
871 err = mnt_want_write(mnt);
872 if (err)
873 goto out_name;
874 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
875 dentry = lookup_one_len(name->name, mnt->mnt_root,
876 strlen(name->name));
877 if (IS_ERR(dentry)) {
878 err = PTR_ERR(dentry);
879 goto out_unlock;
880 }
881
882 inode = d_inode(dentry);
883 if (!inode) {
884 err = -ENOENT;
885 } else {
886 ihold(inode);
887 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
888 }
889 dput(dentry);
890
891 out_unlock:
892 inode_unlock(d_inode(mnt->mnt_root));
893 if (inode)
894 iput(inode);
895 mnt_drop_write(mnt);
896 out_name:
897 putname(name);
898
899 return err;
900 }
901
902 /* Pipelined send and receive functions.
903 *
904 * If a receiver finds no waiting message, then it registers itself in the
905 * list of waiting receivers. A sender checks that list before adding the new
906 * message into the message array. If there is a waiting receiver, then it
907 * bypasses the message array and directly hands the message over to the
908 * receiver. The receiver accepts the message and returns without grabbing the
909 * queue spinlock:
910 *
911 * - Set pointer to message.
912 * - Queue the receiver task for later wakeup (without the info->lock).
913 * - Update its state to STATE_READY. Now the receiver can continue.
914 * - Wake up the process after the lock is dropped. Should the process wake up
915 * before this wakeup (due to a timeout or a signal) it will either see
916 * STATE_READY and continue or acquire the lock to check the state again.
917 *
918 * The same algorithm is used for senders.
919 */
920
921 /* pipelined_send() - send a message directly to the task waiting in
922 * sys_mq_timedreceive() (without inserting message into a queue).
923 */
924 static inline void pipelined_send(struct wake_q_head *wake_q,
925 struct mqueue_inode_info *info,
926 struct msg_msg *message,
927 struct ext_wait_queue *receiver)
928 {
929 receiver->msg = message;
930 list_del(&receiver->list);
931 wake_q_add(wake_q, receiver->task);
932 /*
933 * Rely on the implicit cmpxchg barrier from wake_q_add such
934 * that we can ensure that updating receiver->state is the last
935 * write operation: As once set, the receiver can continue,
936 * and if we don't have the reference count from the wake_q,
937 * yet, at that point we can later have a use-after-free
938 * condition and bogus wakeup.
939 */
940 receiver->state = STATE_READY;
941 }
942
943 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
944 * gets its message and put to the queue (we have one free place for sure). */
945 static inline void pipelined_receive(struct wake_q_head *wake_q,
946 struct mqueue_inode_info *info)
947 {
948 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
949
950 if (!sender) {
951 /* for poll */
952 wake_up_interruptible(&info->wait_q);
953 return;
954 }
955 if (msg_insert(sender->msg, info))
956 return;
957
958 list_del(&sender->list);
959 wake_q_add(wake_q, sender->task);
960 sender->state = STATE_READY;
961 }
962
963 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
964 size_t, msg_len, unsigned int, msg_prio,
965 const struct timespec __user *, u_abs_timeout)
966 {
967 struct fd f;
968 struct inode *inode;
969 struct ext_wait_queue wait;
970 struct ext_wait_queue *receiver;
971 struct msg_msg *msg_ptr;
972 struct mqueue_inode_info *info;
973 ktime_t expires, *timeout = NULL;
974 struct timespec ts;
975 struct posix_msg_tree_node *new_leaf = NULL;
976 int ret = 0;
977 WAKE_Q(wake_q);
978
979 if (u_abs_timeout) {
980 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
981 if (res)
982 return res;
983 timeout = &expires;
984 }
985
986 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
987 return -EINVAL;
988
989 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
990
991 f = fdget(mqdes);
992 if (unlikely(!f.file)) {
993 ret = -EBADF;
994 goto out;
995 }
996
997 inode = file_inode(f.file);
998 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
999 ret = -EBADF;
1000 goto out_fput;
1001 }
1002 info = MQUEUE_I(inode);
1003 audit_file(f.file);
1004
1005 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1006 ret = -EBADF;
1007 goto out_fput;
1008 }
1009
1010 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1011 ret = -EMSGSIZE;
1012 goto out_fput;
1013 }
1014
1015 /* First try to allocate memory, before doing anything with
1016 * existing queues. */
1017 msg_ptr = load_msg(u_msg_ptr, msg_len);
1018 if (IS_ERR(msg_ptr)) {
1019 ret = PTR_ERR(msg_ptr);
1020 goto out_fput;
1021 }
1022 msg_ptr->m_ts = msg_len;
1023 msg_ptr->m_type = msg_prio;
1024
1025 /*
1026 * msg_insert really wants us to have a valid, spare node struct so
1027 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1028 * fall back to that if necessary.
1029 */
1030 if (!info->node_cache)
1031 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1032
1033 spin_lock(&info->lock);
1034
1035 if (!info->node_cache && new_leaf) {
1036 /* Save our speculative allocation into the cache */
1037 INIT_LIST_HEAD(&new_leaf->msg_list);
1038 info->node_cache = new_leaf;
1039 new_leaf = NULL;
1040 } else {
1041 kfree(new_leaf);
1042 }
1043
1044 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1045 if (f.file->f_flags & O_NONBLOCK) {
1046 ret = -EAGAIN;
1047 } else {
1048 wait.task = current;
1049 wait.msg = (void *) msg_ptr;
1050 wait.state = STATE_NONE;
1051 ret = wq_sleep(info, SEND, timeout, &wait);
1052 /*
1053 * wq_sleep must be called with info->lock held, and
1054 * returns with the lock released
1055 */
1056 goto out_free;
1057 }
1058 } else {
1059 receiver = wq_get_first_waiter(info, RECV);
1060 if (receiver) {
1061 pipelined_send(&wake_q, info, msg_ptr, receiver);
1062 } else {
1063 /* adds message to the queue */
1064 ret = msg_insert(msg_ptr, info);
1065 if (ret)
1066 goto out_unlock;
1067 __do_notify(info);
1068 }
1069 inode->i_atime = inode->i_mtime = inode->i_ctime =
1070 current_time(inode);
1071 }
1072 out_unlock:
1073 spin_unlock(&info->lock);
1074 wake_up_q(&wake_q);
1075 out_free:
1076 if (ret)
1077 free_msg(msg_ptr);
1078 out_fput:
1079 fdput(f);
1080 out:
1081 return ret;
1082 }
1083
1084 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1085 size_t, msg_len, unsigned int __user *, u_msg_prio,
1086 const struct timespec __user *, u_abs_timeout)
1087 {
1088 ssize_t ret;
1089 struct msg_msg *msg_ptr;
1090 struct fd f;
1091 struct inode *inode;
1092 struct mqueue_inode_info *info;
1093 struct ext_wait_queue wait;
1094 ktime_t expires, *timeout = NULL;
1095 struct timespec ts;
1096 struct posix_msg_tree_node *new_leaf = NULL;
1097
1098 if (u_abs_timeout) {
1099 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1100 if (res)
1101 return res;
1102 timeout = &expires;
1103 }
1104
1105 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1106
1107 f = fdget(mqdes);
1108 if (unlikely(!f.file)) {
1109 ret = -EBADF;
1110 goto out;
1111 }
1112
1113 inode = file_inode(f.file);
1114 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1115 ret = -EBADF;
1116 goto out_fput;
1117 }
1118 info = MQUEUE_I(inode);
1119 audit_file(f.file);
1120
1121 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1122 ret = -EBADF;
1123 goto out_fput;
1124 }
1125
1126 /* checks if buffer is big enough */
1127 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1128 ret = -EMSGSIZE;
1129 goto out_fput;
1130 }
1131
1132 /*
1133 * msg_insert really wants us to have a valid, spare node struct so
1134 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1135 * fall back to that if necessary.
1136 */
1137 if (!info->node_cache)
1138 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1139
1140 spin_lock(&info->lock);
1141
1142 if (!info->node_cache && new_leaf) {
1143 /* Save our speculative allocation into the cache */
1144 INIT_LIST_HEAD(&new_leaf->msg_list);
1145 info->node_cache = new_leaf;
1146 } else {
1147 kfree(new_leaf);
1148 }
1149
1150 if (info->attr.mq_curmsgs == 0) {
1151 if (f.file->f_flags & O_NONBLOCK) {
1152 spin_unlock(&info->lock);
1153 ret = -EAGAIN;
1154 } else {
1155 wait.task = current;
1156 wait.state = STATE_NONE;
1157 ret = wq_sleep(info, RECV, timeout, &wait);
1158 msg_ptr = wait.msg;
1159 }
1160 } else {
1161 WAKE_Q(wake_q);
1162
1163 msg_ptr = msg_get(info);
1164
1165 inode->i_atime = inode->i_mtime = inode->i_ctime =
1166 current_time(inode);
1167
1168 /* There is now free space in queue. */
1169 pipelined_receive(&wake_q, info);
1170 spin_unlock(&info->lock);
1171 wake_up_q(&wake_q);
1172 ret = 0;
1173 }
1174 if (ret == 0) {
1175 ret = msg_ptr->m_ts;
1176
1177 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1178 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1179 ret = -EFAULT;
1180 }
1181 free_msg(msg_ptr);
1182 }
1183 out_fput:
1184 fdput(f);
1185 out:
1186 return ret;
1187 }
1188
1189 /*
1190 * Notes: the case when user wants us to deregister (with NULL as pointer)
1191 * and he isn't currently owner of notification, will be silently discarded.
1192 * It isn't explicitly defined in the POSIX.
1193 */
1194 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1195 const struct sigevent __user *, u_notification)
1196 {
1197 int ret;
1198 struct fd f;
1199 struct sock *sock;
1200 struct inode *inode;
1201 struct sigevent notification;
1202 struct mqueue_inode_info *info;
1203 struct sk_buff *nc;
1204
1205 if (u_notification) {
1206 if (copy_from_user(&notification, u_notification,
1207 sizeof(struct sigevent)))
1208 return -EFAULT;
1209 }
1210
1211 audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1212
1213 nc = NULL;
1214 sock = NULL;
1215 if (u_notification != NULL) {
1216 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1217 notification.sigev_notify != SIGEV_SIGNAL &&
1218 notification.sigev_notify != SIGEV_THREAD))
1219 return -EINVAL;
1220 if (notification.sigev_notify == SIGEV_SIGNAL &&
1221 !valid_signal(notification.sigev_signo)) {
1222 return -EINVAL;
1223 }
1224 if (notification.sigev_notify == SIGEV_THREAD) {
1225 long timeo;
1226
1227 /* create the notify skb */
1228 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1229 if (!nc) {
1230 ret = -ENOMEM;
1231 goto out;
1232 }
1233 if (copy_from_user(nc->data,
1234 notification.sigev_value.sival_ptr,
1235 NOTIFY_COOKIE_LEN)) {
1236 ret = -EFAULT;
1237 goto out;
1238 }
1239
1240 /* TODO: add a header? */
1241 skb_put(nc, NOTIFY_COOKIE_LEN);
1242 /* and attach it to the socket */
1243 retry:
1244 f = fdget(notification.sigev_signo);
1245 if (!f.file) {
1246 ret = -EBADF;
1247 goto out;
1248 }
1249 sock = netlink_getsockbyfilp(f.file);
1250 fdput(f);
1251 if (IS_ERR(sock)) {
1252 ret = PTR_ERR(sock);
1253 sock = NULL;
1254 goto out;
1255 }
1256
1257 timeo = MAX_SCHEDULE_TIMEOUT;
1258 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1259 if (ret == 1) {
1260 sock = NULL;
1261 goto retry;
1262 }
1263 if (ret) {
1264 sock = NULL;
1265 nc = NULL;
1266 goto out;
1267 }
1268 }
1269 }
1270
1271 f = fdget(mqdes);
1272 if (!f.file) {
1273 ret = -EBADF;
1274 goto out;
1275 }
1276
1277 inode = file_inode(f.file);
1278 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1279 ret = -EBADF;
1280 goto out_fput;
1281 }
1282 info = MQUEUE_I(inode);
1283
1284 ret = 0;
1285 spin_lock(&info->lock);
1286 if (u_notification == NULL) {
1287 if (info->notify_owner == task_tgid(current)) {
1288 remove_notification(info);
1289 inode->i_atime = inode->i_ctime = current_time(inode);
1290 }
1291 } else if (info->notify_owner != NULL) {
1292 ret = -EBUSY;
1293 } else {
1294 switch (notification.sigev_notify) {
1295 case SIGEV_NONE:
1296 info->notify.sigev_notify = SIGEV_NONE;
1297 break;
1298 case SIGEV_THREAD:
1299 info->notify_sock = sock;
1300 info->notify_cookie = nc;
1301 sock = NULL;
1302 nc = NULL;
1303 info->notify.sigev_notify = SIGEV_THREAD;
1304 break;
1305 case SIGEV_SIGNAL:
1306 info->notify.sigev_signo = notification.sigev_signo;
1307 info->notify.sigev_value = notification.sigev_value;
1308 info->notify.sigev_notify = SIGEV_SIGNAL;
1309 break;
1310 }
1311
1312 info->notify_owner = get_pid(task_tgid(current));
1313 info->notify_user_ns = get_user_ns(current_user_ns());
1314 inode->i_atime = inode->i_ctime = current_time(inode);
1315 }
1316 spin_unlock(&info->lock);
1317 out_fput:
1318 fdput(f);
1319 out:
1320 if (sock)
1321 netlink_detachskb(sock, nc);
1322 else if (nc)
1323 dev_kfree_skb(nc);
1324
1325 return ret;
1326 }
1327
1328 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1329 const struct mq_attr __user *, u_mqstat,
1330 struct mq_attr __user *, u_omqstat)
1331 {
1332 int ret;
1333 struct mq_attr mqstat, omqstat;
1334 struct fd f;
1335 struct inode *inode;
1336 struct mqueue_inode_info *info;
1337
1338 if (u_mqstat != NULL) {
1339 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1340 return -EFAULT;
1341 if (mqstat.mq_flags & (~O_NONBLOCK))
1342 return -EINVAL;
1343 }
1344
1345 f = fdget(mqdes);
1346 if (!f.file) {
1347 ret = -EBADF;
1348 goto out;
1349 }
1350
1351 inode = file_inode(f.file);
1352 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1353 ret = -EBADF;
1354 goto out_fput;
1355 }
1356 info = MQUEUE_I(inode);
1357
1358 spin_lock(&info->lock);
1359
1360 omqstat = info->attr;
1361 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1362 if (u_mqstat) {
1363 audit_mq_getsetattr(mqdes, &mqstat);
1364 spin_lock(&f.file->f_lock);
1365 if (mqstat.mq_flags & O_NONBLOCK)
1366 f.file->f_flags |= O_NONBLOCK;
1367 else
1368 f.file->f_flags &= ~O_NONBLOCK;
1369 spin_unlock(&f.file->f_lock);
1370
1371 inode->i_atime = inode->i_ctime = current_time(inode);
1372 }
1373
1374 spin_unlock(&info->lock);
1375
1376 ret = 0;
1377 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1378 sizeof(struct mq_attr)))
1379 ret = -EFAULT;
1380
1381 out_fput:
1382 fdput(f);
1383 out:
1384 return ret;
1385 }
1386
1387 static const struct inode_operations mqueue_dir_inode_operations = {
1388 .lookup = simple_lookup,
1389 .create = mqueue_create,
1390 .unlink = mqueue_unlink,
1391 };
1392
1393 static const struct file_operations mqueue_file_operations = {
1394 .flush = mqueue_flush_file,
1395 .poll = mqueue_poll_file,
1396 .read = mqueue_read_file,
1397 .llseek = default_llseek,
1398 };
1399
1400 static const struct super_operations mqueue_super_ops = {
1401 .alloc_inode = mqueue_alloc_inode,
1402 .destroy_inode = mqueue_destroy_inode,
1403 .evict_inode = mqueue_evict_inode,
1404 .statfs = simple_statfs,
1405 };
1406
1407 static struct file_system_type mqueue_fs_type = {
1408 .name = "mqueue",
1409 .mount = mqueue_mount,
1410 .kill_sb = kill_litter_super,
1411 .fs_flags = FS_USERNS_MOUNT,
1412 };
1413
1414 int mq_init_ns(struct ipc_namespace *ns)
1415 {
1416 ns->mq_queues_count = 0;
1417 ns->mq_queues_max = DFLT_QUEUESMAX;
1418 ns->mq_msg_max = DFLT_MSGMAX;
1419 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1420 ns->mq_msg_default = DFLT_MSG;
1421 ns->mq_msgsize_default = DFLT_MSGSIZE;
1422
1423 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1424 if (IS_ERR(ns->mq_mnt)) {
1425 int err = PTR_ERR(ns->mq_mnt);
1426 ns->mq_mnt = NULL;
1427 return err;
1428 }
1429 return 0;
1430 }
1431
1432 void mq_clear_sbinfo(struct ipc_namespace *ns)
1433 {
1434 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1435 }
1436
1437 void mq_put_mnt(struct ipc_namespace *ns)
1438 {
1439 kern_unmount(ns->mq_mnt);
1440 }
1441
1442 static int __init init_mqueue_fs(void)
1443 {
1444 int error;
1445
1446 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1447 sizeof(struct mqueue_inode_info), 0,
1448 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1449 if (mqueue_inode_cachep == NULL)
1450 return -ENOMEM;
1451
1452 /* ignore failures - they are not fatal */
1453 mq_sysctl_table = mq_register_sysctl_table();
1454
1455 error = register_filesystem(&mqueue_fs_type);
1456 if (error)
1457 goto out_sysctl;
1458
1459 spin_lock_init(&mq_lock);
1460
1461 error = mq_init_ns(&init_ipc_ns);
1462 if (error)
1463 goto out_filesystem;
1464
1465 return 0;
1466
1467 out_filesystem:
1468 unregister_filesystem(&mqueue_fs_type);
1469 out_sysctl:
1470 if (mq_sysctl_table)
1471 unregister_sysctl_table(mq_sysctl_table);
1472 kmem_cache_destroy(mqueue_inode_cachep);
1473 return error;
1474 }
1475
1476 device_initcall(init_mqueue_fs);
Linux with added FPC-III support