projects/linux·blob v4.13-rc2 · refs

1646 lines · 40.1 KiB

1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak ([email protected])
5 * Michal Wronski ([email protected])
6 *
7 * Spinlocks: Mohamed Abbas ([email protected])
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul ([email protected])
10 *
11 * Audit: George Wilson ([email protected])
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#include <linux/sched/wake_q.h>
39#include <linux/sched/signal.h>
40#include <linux/sched/user.h>
41
42#include <net/sock.h>
43#include "util.h"
44
45#define MQUEUE_MAGIC 0x19800202
46#define DIRENT_SIZE 20
47#define FILENT_SIZE 80
48
49#define SEND 0
50#define RECV 1
51
52#define STATE_NONE 0
53#define STATE_READY 1
54
55struct posix_msg_tree_node {
56 struct rb_node rb_node;
57 struct list_head msg_list;
58 int priority;
59};
60
61struct ext_wait_queue { /* queue of sleeping tasks */
62 struct task_struct *task;
63 struct list_head list;
64 struct msg_msg *msg; /* ptr of loaded message */
65 int state; /* one of STATE_* values */
66};
67
68struct mqueue_inode_info {
69 spinlock_t lock;
70 struct inode vfs_inode;
71 wait_queue_head_t wait_q;
72
73 struct rb_root msg_tree;
74 struct posix_msg_tree_node *node_cache;
75 struct mq_attr attr;
76
77 struct sigevent notify;
78 struct pid *notify_owner;
79 struct user_namespace *notify_user_ns;
80 struct user_struct *user; /* user who created, for accounting */
81 struct sock *notify_sock;
82 struct sk_buff *notify_cookie;
83
84 /* for tasks waiting for free space and messages, respectively */
85 struct ext_wait_queue e_wait_q[2];
86
87 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
88};
89
90static const struct inode_operations mqueue_dir_inode_operations;
91static const struct file_operations mqueue_file_operations;
92static const struct super_operations mqueue_super_ops;
93static void remove_notification(struct mqueue_inode_info *info);
94
95static struct kmem_cache *mqueue_inode_cachep;
96
97static struct ctl_table_header *mq_sysctl_table;
98
99static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100{
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
102}
103
104/*
105 * This routine should be called with the mq_lock held.
106 */
107static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
108{
109 return get_ipc_ns(inode->i_sb->s_fs_info);
110}
111
112static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
113{
114 struct ipc_namespace *ns;
115
116 spin_lock(&mq_lock);
117 ns = __get_ns_from_inode(inode);
118 spin_unlock(&mq_lock);
119 return ns;
120}
121
122/* Auxiliary functions to manipulate messages' list */
123static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
124{
125 struct rb_node **p, *parent = NULL;
126 struct posix_msg_tree_node *leaf;
127
128 p = &info->msg_tree.rb_node;
129 while (*p) {
130 parent = *p;
131 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
132
133 if (likely(leaf->priority == msg->m_type))
134 goto insert_msg;
135 else if (msg->m_type < leaf->priority)
136 p = &(*p)->rb_left;
137 else
138 p = &(*p)->rb_right;
139 }
140 if (info->node_cache) {
141 leaf = info->node_cache;
142 info->node_cache = NULL;
143 } else {
144 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
145 if (!leaf)
146 return -ENOMEM;
147 INIT_LIST_HEAD(&leaf->msg_list);
148 }
149 leaf->priority = msg->m_type;
150 rb_link_node(&leaf->rb_node, parent, p);
151 rb_insert_color(&leaf->rb_node, &info->msg_tree);
152insert_msg:
153 info->attr.mq_curmsgs++;
154 info->qsize += msg->m_ts;
155 list_add_tail(&msg->m_list, &leaf->msg_list);
156 return 0;
157}
158
159static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160{
161 struct rb_node **p, *parent = NULL;
162 struct posix_msg_tree_node *leaf;
163 struct msg_msg *msg;
164
165try_again:
166 p = &info->msg_tree.rb_node;
167 while (*p) {
168 parent = *p;
169 /*
170 * During insert, low priorities go to the left and high to the
171 * right. On receive, we want the highest priorities first, so
172 * walk all the way to the right.
173 */
174 p = &(*p)->rb_right;
175 }
176 if (!parent) {
177 if (info->attr.mq_curmsgs) {
178 pr_warn_once("Inconsistency in POSIX message queue, "
179 "no tree element, but supposedly messages "
180 "should exist!\n");
181 info->attr.mq_curmsgs = 0;
182 }
183 return NULL;
184 }
185 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186 if (unlikely(list_empty(&leaf->msg_list))) {
187 pr_warn_once("Inconsistency in POSIX message queue, "
188 "empty leaf node but we haven't implemented "
189 "lazy leaf delete!\n");
190 rb_erase(&leaf->rb_node, &info->msg_tree);
191 if (info->node_cache) {
192 kfree(leaf);
193 } else {
194 info->node_cache = leaf;
195 }
196 goto try_again;
197 } else {
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
203 if (info->node_cache) {
204 kfree(leaf);
205 } else {
206 info->node_cache = leaf;
207 }
208 }
209 }
210 info->attr.mq_curmsgs--;
211 info->qsize -= msg->m_ts;
212 return msg;
213}
214
215static struct inode *mqueue_get_inode(struct super_block *sb,
216 struct ipc_namespace *ipc_ns, umode_t mode,
217 struct mq_attr *attr)
218{
219 struct user_struct *u = current_user();
220 struct inode *inode;
221 int ret = -ENOMEM;
222
223 inode = new_inode(sb);
224 if (!inode)
225 goto err;
226
227 inode->i_ino = get_next_ino();
228 inode->i_mode = mode;
229 inode->i_uid = current_fsuid();
230 inode->i_gid = current_fsgid();
231 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
232
233 if (S_ISREG(mode)) {
234 struct mqueue_inode_info *info;
235 unsigned long mq_bytes, mq_treesize;
236
237 inode->i_fop = &mqueue_file_operations;
238 inode->i_size = FILENT_SIZE;
239 /* mqueue specific info */
240 info = MQUEUE_I(inode);
241 spin_lock_init(&info->lock);
242 init_waitqueue_head(&info->wait_q);
243 INIT_LIST_HEAD(&info->e_wait_q[0].list);
244 INIT_LIST_HEAD(&info->e_wait_q[1].list);
245 info->notify_owner = NULL;
246 info->notify_user_ns = NULL;
247 info->qsize = 0;
248 info->user = NULL; /* set when all is ok */
249 info->msg_tree = RB_ROOT;
250 info->node_cache = NULL;
251 memset(&info->attr, 0, sizeof(info->attr));
252 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253 ipc_ns->mq_msg_default);
254 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255 ipc_ns->mq_msgsize_default);
256 if (attr) {
257 info->attr.mq_maxmsg = attr->mq_maxmsg;
258 info->attr.mq_msgsize = attr->mq_msgsize;
259 }
260 /*
261 * We used to allocate a static array of pointers and account
262 * the size of that array as well as one msg_msg struct per
263 * possible message into the queue size. That's no longer
264 * accurate as the queue is now an rbtree and will grow and
265 * shrink depending on usage patterns. We can, however, still
266 * account one msg_msg struct per message, but the nodes are
267 * allocated depending on priority usage, and most programs
268 * only use one, or a handful, of priorities. However, since
269 * this is pinned memory, we need to assume worst case, so
270 * that means the min(mq_maxmsg, max_priorities) * struct
271 * posix_msg_tree_node.
272 */
273 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
274 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
275 sizeof(struct posix_msg_tree_node);
276
277 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
278 info->attr.mq_msgsize);
279
280 spin_lock(&mq_lock);
281 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
282 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
283 spin_unlock(&mq_lock);
284 /* mqueue_evict_inode() releases info->messages */
285 ret = -EMFILE;
286 goto out_inode;
287 }
288 u->mq_bytes += mq_bytes;
289 spin_unlock(&mq_lock);
290
291 /* all is ok */
292 info->user = get_uid(u);
293 } else if (S_ISDIR(mode)) {
294 inc_nlink(inode);
295 /* Some things misbehave if size == 0 on a directory */
296 inode->i_size = 2 * DIRENT_SIZE;
297 inode->i_op = &mqueue_dir_inode_operations;
298 inode->i_fop = &simple_dir_operations;
299 }
300
301 return inode;
302out_inode:
303 iput(inode);
304err:
305 return ERR_PTR(ret);
306}
307
308static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
309{
310 struct inode *inode;
311 struct ipc_namespace *ns = sb->s_fs_info;
312
313 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
314 sb->s_blocksize = PAGE_SIZE;
315 sb->s_blocksize_bits = PAGE_SHIFT;
316 sb->s_magic = MQUEUE_MAGIC;
317 sb->s_op = &mqueue_super_ops;
318
319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
320 if (IS_ERR(inode))
321 return PTR_ERR(inode);
322
323 sb->s_root = d_make_root(inode);
324 if (!sb->s_root)
325 return -ENOMEM;
326 return 0;
327}
328
329static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330 int flags, const char *dev_name,
331 void *data)
332{
333 struct ipc_namespace *ns;
334 if (flags & MS_KERNMOUNT) {
335 ns = data;
336 data = NULL;
337 } else {
338 ns = current->nsproxy->ipc_ns;
339 }
340 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
341}
342
343static void init_once(void *foo)
344{
345 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
346
347 inode_init_once(&p->vfs_inode);
348}
349
350static struct inode *mqueue_alloc_inode(struct super_block *sb)
351{
352 struct mqueue_inode_info *ei;
353
354 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
355 if (!ei)
356 return NULL;
357 return &ei->vfs_inode;
358}
359
360static void mqueue_i_callback(struct rcu_head *head)
361{
362 struct inode *inode = container_of(head, struct inode, i_rcu);
363 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
364}
365
366static void mqueue_destroy_inode(struct inode *inode)
367{
368 call_rcu(&inode->i_rcu, mqueue_i_callback);
369}
370
371static void mqueue_evict_inode(struct inode *inode)
372{
373 struct mqueue_inode_info *info;
374 struct user_struct *user;
375 unsigned long mq_bytes, mq_treesize;
376 struct ipc_namespace *ipc_ns;
377 struct msg_msg *msg;
378
379 clear_inode(inode);
380
381 if (S_ISDIR(inode->i_mode))
382 return;
383
384 ipc_ns = get_ns_from_inode(inode);
385 info = MQUEUE_I(inode);
386 spin_lock(&info->lock);
387 while ((msg = msg_get(info)) != NULL)
388 free_msg(msg);
389 kfree(info->node_cache);
390 spin_unlock(&info->lock);
391
392 /* Total amount of bytes accounted for the mqueue */
393 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
394 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
395 sizeof(struct posix_msg_tree_node);
396
397 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
398 info->attr.mq_msgsize);
399
400 user = info->user;
401 if (user) {
402 spin_lock(&mq_lock);
403 user->mq_bytes -= mq_bytes;
404 /*
405 * get_ns_from_inode() ensures that the
406 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
407 * to which we now hold a reference, or it is NULL.
408 * We can't put it here under mq_lock, though.
409 */
410 if (ipc_ns)
411 ipc_ns->mq_queues_count--;
412 spin_unlock(&mq_lock);
413 free_uid(user);
414 }
415 if (ipc_ns)
416 put_ipc_ns(ipc_ns);
417}
418
419static int mqueue_create(struct inode *dir, struct dentry *dentry,
420 umode_t mode, bool excl)
421{
422 struct inode *inode;
423 struct mq_attr *attr = dentry->d_fsdata;
424 int error;
425 struct ipc_namespace *ipc_ns;
426
427 spin_lock(&mq_lock);
428 ipc_ns = __get_ns_from_inode(dir);
429 if (!ipc_ns) {
430 error = -EACCES;
431 goto out_unlock;
432 }
433
434 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
435 !capable(CAP_SYS_RESOURCE)) {
436 error = -ENOSPC;
437 goto out_unlock;
438 }
439 ipc_ns->mq_queues_count++;
440 spin_unlock(&mq_lock);
441
442 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
443 if (IS_ERR(inode)) {
444 error = PTR_ERR(inode);
445 spin_lock(&mq_lock);
446 ipc_ns->mq_queues_count--;
447 goto out_unlock;
448 }
449
450 put_ipc_ns(ipc_ns);
451 dir->i_size += DIRENT_SIZE;
452 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
453
454 d_instantiate(dentry, inode);
455 dget(dentry);
456 return 0;
457out_unlock:
458 spin_unlock(&mq_lock);
459 if (ipc_ns)
460 put_ipc_ns(ipc_ns);
461 return error;
462}
463
464static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
465{
466 struct inode *inode = d_inode(dentry);
467
468 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
469 dir->i_size -= DIRENT_SIZE;
470 drop_nlink(inode);
471 dput(dentry);
472 return 0;
473}
474
475/*
476* This is routine for system read from queue file.
477* To avoid mess with doing here some sort of mq_receive we allow
478* to read only queue size & notification info (the only values
479* that are interesting from user point of view and aren't accessible
480* through std routines)
481*/
482static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
483 size_t count, loff_t *off)
484{
485 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
486 char buffer[FILENT_SIZE];
487 ssize_t ret;
488
489 spin_lock(&info->lock);
490 snprintf(buffer, sizeof(buffer),
491 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
492 info->qsize,
493 info->notify_owner ? info->notify.sigev_notify : 0,
494 (info->notify_owner &&
495 info->notify.sigev_notify == SIGEV_SIGNAL) ?
496 info->notify.sigev_signo : 0,
497 pid_vnr(info->notify_owner));
498 spin_unlock(&info->lock);
499 buffer[sizeof(buffer)-1] = '\0';
500
501 ret = simple_read_from_buffer(u_data, count, off, buffer,
502 strlen(buffer));
503 if (ret <= 0)
504 return ret;
505
506 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
507 return ret;
508}
509
510static int mqueue_flush_file(struct file *filp, fl_owner_t id)
511{
512 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
513
514 spin_lock(&info->lock);
515 if (task_tgid(current) == info->notify_owner)
516 remove_notification(info);
517
518 spin_unlock(&info->lock);
519 return 0;
520}
521
522static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
523{
524 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
525 int retval = 0;
526
527 poll_wait(filp, &info->wait_q, poll_tab);
528
529 spin_lock(&info->lock);
530 if (info->attr.mq_curmsgs)
531 retval = POLLIN | POLLRDNORM;
532
533 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
534 retval |= POLLOUT | POLLWRNORM;
535 spin_unlock(&info->lock);
536
537 return retval;
538}
539
540/* Adds current to info->e_wait_q[sr] before element with smaller prio */
541static void wq_add(struct mqueue_inode_info *info, int sr,
542 struct ext_wait_queue *ewp)
543{
544 struct ext_wait_queue *walk;
545
546 ewp->task = current;
547
548 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
549 if (walk->task->static_prio <= current->static_prio) {
550 list_add_tail(&ewp->list, &walk->list);
551 return;
552 }
553 }
554 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
555}
556
557/*
558 * Puts current task to sleep. Caller must hold queue lock. After return
559 * lock isn't held.
560 * sr: SEND or RECV
561 */
562static int wq_sleep(struct mqueue_inode_info *info, int sr,
563 ktime_t *timeout, struct ext_wait_queue *ewp)
564 __releases(&info->lock)
565{
566 int retval;
567 signed long time;
568
569 wq_add(info, sr, ewp);
570
571 for (;;) {
572 __set_current_state(TASK_INTERRUPTIBLE);
573
574 spin_unlock(&info->lock);
575 time = schedule_hrtimeout_range_clock(timeout, 0,
576 HRTIMER_MODE_ABS, CLOCK_REALTIME);
577
578 if (ewp->state == STATE_READY) {
579 retval = 0;
580 goto out;
581 }
582 spin_lock(&info->lock);
583 if (ewp->state == STATE_READY) {
584 retval = 0;
585 goto out_unlock;
586 }
587 if (signal_pending(current)) {
588 retval = -ERESTARTSYS;
589 break;
590 }
591 if (time == 0) {
592 retval = -ETIMEDOUT;
593 break;
594 }
595 }
596 list_del(&ewp->list);
597out_unlock:
598 spin_unlock(&info->lock);
599out:
600 return retval;
601}
602
603/*
604 * Returns waiting task that should be serviced first or NULL if none exists
605 */
606static struct ext_wait_queue *wq_get_first_waiter(
607 struct mqueue_inode_info *info, int sr)
608{
609 struct list_head *ptr;
610
611 ptr = info->e_wait_q[sr].list.prev;
612 if (ptr == &info->e_wait_q[sr].list)
613 return NULL;
614 return list_entry(ptr, struct ext_wait_queue, list);
615}
616
617
618static inline void set_cookie(struct sk_buff *skb, char code)
619{
620 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
621}
622
623/*
624 * The next function is only to split too long sys_mq_timedsend
625 */
626static void __do_notify(struct mqueue_inode_info *info)
627{
628 /* notification
629 * invoked when there is registered process and there isn't process
630 * waiting synchronously for message AND state of queue changed from
631 * empty to not empty. Here we are sure that no one is waiting
632 * synchronously. */
633 if (info->notify_owner &&
634 info->attr.mq_curmsgs == 1) {
635 struct siginfo sig_i;
636 switch (info->notify.sigev_notify) {
637 case SIGEV_NONE:
638 break;
639 case SIGEV_SIGNAL:
640 /* sends signal */
641
642 sig_i.si_signo = info->notify.sigev_signo;
643 sig_i.si_errno = 0;
644 sig_i.si_code = SI_MESGQ;
645 sig_i.si_value = info->notify.sigev_value;
646 /* map current pid/uid into info->owner's namespaces */
647 rcu_read_lock();
648 sig_i.si_pid = task_tgid_nr_ns(current,
649 ns_of_pid(info->notify_owner));
650 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
651 rcu_read_unlock();
652
653 kill_pid_info(info->notify.sigev_signo,
654 &sig_i, info->notify_owner);
655 break;
656 case SIGEV_THREAD:
657 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
658 netlink_sendskb(info->notify_sock, info->notify_cookie);
659 break;
660 }
661 /* after notification unregisters process */
662 put_pid(info->notify_owner);
663 put_user_ns(info->notify_user_ns);
664 info->notify_owner = NULL;
665 info->notify_user_ns = NULL;
666 }
667 wake_up(&info->wait_q);
668}
669
670static int prepare_timeout(const struct timespec __user *u_abs_timeout,
671 struct timespec *ts)
672{
673 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
674 return -EFAULT;
675 if (!timespec_valid(ts))
676 return -EINVAL;
677 return 0;
678}
679
680static void remove_notification(struct mqueue_inode_info *info)
681{
682 if (info->notify_owner != NULL &&
683 info->notify.sigev_notify == SIGEV_THREAD) {
684 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
685 netlink_sendskb(info->notify_sock, info->notify_cookie);
686 }
687 put_pid(info->notify_owner);
688 put_user_ns(info->notify_user_ns);
689 info->notify_owner = NULL;
690 info->notify_user_ns = NULL;
691}
692
693static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
694{
695 int mq_treesize;
696 unsigned long total_size;
697
698 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
699 return -EINVAL;
700 if (capable(CAP_SYS_RESOURCE)) {
701 if (attr->mq_maxmsg > HARD_MSGMAX ||
702 attr->mq_msgsize > HARD_MSGSIZEMAX)
703 return -EINVAL;
704 } else {
705 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
706 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
707 return -EINVAL;
708 }
709 /* check for overflow */
710 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
711 return -EOVERFLOW;
712 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
713 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
714 sizeof(struct posix_msg_tree_node);
715 total_size = attr->mq_maxmsg * attr->mq_msgsize;
716 if (total_size + mq_treesize < total_size)
717 return -EOVERFLOW;
718 return 0;
719}
720
721/*
722 * Invoked when creating a new queue via sys_mq_open
723 */
724static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
725 struct path *path, int oflag, umode_t mode,
726 struct mq_attr *attr)
727{
728 const struct cred *cred = current_cred();
729 int ret;
730
731 if (attr) {
732 ret = mq_attr_ok(ipc_ns, attr);
733 if (ret)
734 return ERR_PTR(ret);
735 /* store for use during create */
736 path->dentry->d_fsdata = attr;
737 } else {
738 struct mq_attr def_attr;
739
740 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
741 ipc_ns->mq_msg_default);
742 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
743 ipc_ns->mq_msgsize_default);
744 ret = mq_attr_ok(ipc_ns, &def_attr);
745 if (ret)
746 return ERR_PTR(ret);
747 }
748
749 mode &= ~current_umask();
750 ret = vfs_create(dir, path->dentry, mode, true);
751 path->dentry->d_fsdata = NULL;
752 if (ret)
753 return ERR_PTR(ret);
754 return dentry_open(path, oflag, cred);
755}
756
757/* Opens existing queue */
758static struct file *do_open(struct path *path, int oflag)
759{
760 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
761 MAY_READ | MAY_WRITE };
762 int acc;
763 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
764 return ERR_PTR(-EINVAL);
765 acc = oflag2acc[oflag & O_ACCMODE];
766 if (inode_permission(d_inode(path->dentry), acc))
767 return ERR_PTR(-EACCES);
768 return dentry_open(path, oflag, current_cred());
769}
770
771static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
772 struct mq_attr *attr)
773{
774 struct path path;
775 struct file *filp;
776 struct filename *name;
777 int fd, error;
778 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
779 struct vfsmount *mnt = ipc_ns->mq_mnt;
780 struct dentry *root = mnt->mnt_root;
781 int ro;
782
783 audit_mq_open(oflag, mode, attr);
784
785 if (IS_ERR(name = getname(u_name)))
786 return PTR_ERR(name);
787
788 fd = get_unused_fd_flags(O_CLOEXEC);
789 if (fd < 0)
790 goto out_putname;
791
792 ro = mnt_want_write(mnt); /* we'll drop it in any case */
793 error = 0;
794 inode_lock(d_inode(root));
795 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
796 if (IS_ERR(path.dentry)) {
797 error = PTR_ERR(path.dentry);
798 goto out_putfd;
799 }
800 path.mnt = mntget(mnt);
801
802 if (oflag & O_CREAT) {
803 if (d_really_is_positive(path.dentry)) { /* entry already exists */
804 audit_inode(name, path.dentry, 0);
805 if (oflag & O_EXCL) {
806 error = -EEXIST;
807 goto out;
808 }
809 filp = do_open(&path, oflag);
810 } else {
811 if (ro) {
812 error = ro;
813 goto out;
814 }
815 audit_inode_parent_hidden(name, root);
816 filp = do_create(ipc_ns, d_inode(root), &path,
817 oflag, mode, attr);
818 }
819 } else {
820 if (d_really_is_negative(path.dentry)) {
821 error = -ENOENT;
822 goto out;
823 }
824 audit_inode(name, path.dentry, 0);
825 filp = do_open(&path, oflag);
826 }
827
828 if (!IS_ERR(filp))
829 fd_install(fd, filp);
830 else
831 error = PTR_ERR(filp);
832out:
833 path_put(&path);
834out_putfd:
835 if (error) {
836 put_unused_fd(fd);
837 fd = error;
838 }
839 inode_unlock(d_inode(root));
840 if (!ro)
841 mnt_drop_write(mnt);
842out_putname:
843 putname(name);
844 return fd;
845}
846
847SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
848 struct mq_attr __user *, u_attr)
849{
850 struct mq_attr attr;
851 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
852 return -EFAULT;
853
854 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
855}
856
857SYSCALL_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
891out_unlock:
892 inode_unlock(d_inode(mnt->mnt_root));
893 if (inode)
894 iput(inode);
895 mnt_drop_write(mnt);
896out_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 */
924static 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). */
945static 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
963static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
964 size_t msg_len, unsigned int msg_prio,
965 struct timespec *ts)
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 posix_msg_tree_node *new_leaf = NULL;
975 int ret = 0;
976 DEFINE_WAKE_Q(wake_q);
977
978 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
979 return -EINVAL;
980
981 if (ts) {
982 expires = timespec_to_ktime(*ts);
983 timeout = &expires;
984 }
985
986 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
987
988 f = fdget(mqdes);
989 if (unlikely(!f.file)) {
990 ret = -EBADF;
991 goto out;
992 }
993
994 inode = file_inode(f.file);
995 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
996 ret = -EBADF;
997 goto out_fput;
998 }
999 info = MQUEUE_I(inode);
1000 audit_file(f.file);
1001
1002 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1003 ret = -EBADF;
1004 goto out_fput;
1005 }
1006
1007 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1008 ret = -EMSGSIZE;
1009 goto out_fput;
1010 }
1011
1012 /* First try to allocate memory, before doing anything with
1013 * existing queues. */
1014 msg_ptr = load_msg(u_msg_ptr, msg_len);
1015 if (IS_ERR(msg_ptr)) {
1016 ret = PTR_ERR(msg_ptr);
1017 goto out_fput;
1018 }
1019 msg_ptr->m_ts = msg_len;
1020 msg_ptr->m_type = msg_prio;
1021
1022 /*
1023 * msg_insert really wants us to have a valid, spare node struct so
1024 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1025 * fall back to that if necessary.
1026 */
1027 if (!info->node_cache)
1028 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1029
1030 spin_lock(&info->lock);
1031
1032 if (!info->node_cache && new_leaf) {
1033 /* Save our speculative allocation into the cache */
1034 INIT_LIST_HEAD(&new_leaf->msg_list);
1035 info->node_cache = new_leaf;
1036 new_leaf = NULL;
1037 } else {
1038 kfree(new_leaf);
1039 }
1040
1041 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1042 if (f.file->f_flags & O_NONBLOCK) {
1043 ret = -EAGAIN;
1044 } else {
1045 wait.task = current;
1046 wait.msg = (void *) msg_ptr;
1047 wait.state = STATE_NONE;
1048 ret = wq_sleep(info, SEND, timeout, &wait);
1049 /*
1050 * wq_sleep must be called with info->lock held, and
1051 * returns with the lock released
1052 */
1053 goto out_free;
1054 }
1055 } else {
1056 receiver = wq_get_first_waiter(info, RECV);
1057 if (receiver) {
1058 pipelined_send(&wake_q, info, msg_ptr, receiver);
1059 } else {
1060 /* adds message to the queue */
1061 ret = msg_insert(msg_ptr, info);
1062 if (ret)
1063 goto out_unlock;
1064 __do_notify(info);
1065 }
1066 inode->i_atime = inode->i_mtime = inode->i_ctime =
1067 current_time(inode);
1068 }
1069out_unlock:
1070 spin_unlock(&info->lock);
1071 wake_up_q(&wake_q);
1072out_free:
1073 if (ret)
1074 free_msg(msg_ptr);
1075out_fput:
1076 fdput(f);
1077out:
1078 return ret;
1079}
1080
1081static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1082 size_t msg_len, unsigned int __user *u_msg_prio,
1083 struct timespec *ts)
1084{
1085 ssize_t ret;
1086 struct msg_msg *msg_ptr;
1087 struct fd f;
1088 struct inode *inode;
1089 struct mqueue_inode_info *info;
1090 struct ext_wait_queue wait;
1091 ktime_t expires, *timeout = NULL;
1092 struct posix_msg_tree_node *new_leaf = NULL;
1093
1094 if (ts) {
1095 expires = timespec_to_ktime(*ts);
1096 timeout = &expires;
1097 }
1098
1099 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1100
1101 f = fdget(mqdes);
1102 if (unlikely(!f.file)) {
1103 ret = -EBADF;
1104 goto out;
1105 }
1106
1107 inode = file_inode(f.file);
1108 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1109 ret = -EBADF;
1110 goto out_fput;
1111 }
1112 info = MQUEUE_I(inode);
1113 audit_file(f.file);
1114
1115 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1116 ret = -EBADF;
1117 goto out_fput;
1118 }
1119
1120 /* checks if buffer is big enough */
1121 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1122 ret = -EMSGSIZE;
1123 goto out_fput;
1124 }
1125
1126 /*
1127 * msg_insert really wants us to have a valid, spare node struct so
1128 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1129 * fall back to that if necessary.
1130 */
1131 if (!info->node_cache)
1132 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1133
1134 spin_lock(&info->lock);
1135
1136 if (!info->node_cache && new_leaf) {
1137 /* Save our speculative allocation into the cache */
1138 INIT_LIST_HEAD(&new_leaf->msg_list);
1139 info->node_cache = new_leaf;
1140 } else {
1141 kfree(new_leaf);
1142 }
1143
1144 if (info->attr.mq_curmsgs == 0) {
1145 if (f.file->f_flags & O_NONBLOCK) {
1146 spin_unlock(&info->lock);
1147 ret = -EAGAIN;
1148 } else {
1149 wait.task = current;
1150 wait.state = STATE_NONE;
1151 ret = wq_sleep(info, RECV, timeout, &wait);
1152 msg_ptr = wait.msg;
1153 }
1154 } else {
1155 DEFINE_WAKE_Q(wake_q);
1156
1157 msg_ptr = msg_get(info);
1158
1159 inode->i_atime = inode->i_mtime = inode->i_ctime =
1160 current_time(inode);
1161
1162 /* There is now free space in queue. */
1163 pipelined_receive(&wake_q, info);
1164 spin_unlock(&info->lock);
1165 wake_up_q(&wake_q);
1166 ret = 0;
1167 }
1168 if (ret == 0) {
1169 ret = msg_ptr->m_ts;
1170
1171 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1172 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1173 ret = -EFAULT;
1174 }
1175 free_msg(msg_ptr);
1176 }
1177out_fput:
1178 fdput(f);
1179out:
1180 return ret;
1181}
1182
1183SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1184 size_t, msg_len, unsigned int, msg_prio,
1185 const struct timespec __user *, u_abs_timeout)
1186{
1187 struct timespec ts, *p = NULL;
1188 if (u_abs_timeout) {
1189 int res = prepare_timeout(u_abs_timeout, &ts);
1190 if (res)
1191 return res;
1192 p = &ts;
1193 }
1194 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1195}
1196
1197SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1198 size_t, msg_len, unsigned int __user *, u_msg_prio,
1199 const struct timespec __user *, u_abs_timeout)
1200{
1201 struct timespec ts, *p = NULL;
1202 if (u_abs_timeout) {
1203 int res = prepare_timeout(u_abs_timeout, &ts);
1204 if (res)
1205 return res;
1206 p = &ts;
1207 }
1208 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1209}
1210
1211/*
1212 * Notes: the case when user wants us to deregister (with NULL as pointer)
1213 * and he isn't currently owner of notification, will be silently discarded.
1214 * It isn't explicitly defined in the POSIX.
1215 */
1216static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1217{
1218 int ret;
1219 struct fd f;
1220 struct sock *sock;
1221 struct inode *inode;
1222 struct mqueue_inode_info *info;
1223 struct sk_buff *nc;
1224
1225 audit_mq_notify(mqdes, notification);
1226
1227 nc = NULL;
1228 sock = NULL;
1229 if (notification != NULL) {
1230 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1231 notification->sigev_notify != SIGEV_SIGNAL &&
1232 notification->sigev_notify != SIGEV_THREAD))
1233 return -EINVAL;
1234 if (notification->sigev_notify == SIGEV_SIGNAL &&
1235 !valid_signal(notification->sigev_signo)) {
1236 return -EINVAL;
1237 }
1238 if (notification->sigev_notify == SIGEV_THREAD) {
1239 long timeo;
1240
1241 /* create the notify skb */
1242 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1243 if (!nc) {
1244 ret = -ENOMEM;
1245 goto out;
1246 }
1247 if (copy_from_user(nc->data,
1248 notification->sigev_value.sival_ptr,
1249 NOTIFY_COOKIE_LEN)) {
1250 ret = -EFAULT;
1251 goto out;
1252 }
1253
1254 /* TODO: add a header? */
1255 skb_put(nc, NOTIFY_COOKIE_LEN);
1256 /* and attach it to the socket */
1257retry:
1258 f = fdget(notification->sigev_signo);
1259 if (!f.file) {
1260 ret = -EBADF;
1261 goto out;
1262 }
1263 sock = netlink_getsockbyfilp(f.file);
1264 fdput(f);
1265 if (IS_ERR(sock)) {
1266 ret = PTR_ERR(sock);
1267 sock = NULL;
1268 goto out;
1269 }
1270
1271 timeo = MAX_SCHEDULE_TIMEOUT;
1272 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1273 if (ret == 1) {
1274 sock = NULL;
1275 goto retry;
1276 }
1277 if (ret) {
1278 sock = NULL;
1279 nc = NULL;
1280 goto out;
1281 }
1282 }
1283 }
1284
1285 f = fdget(mqdes);
1286 if (!f.file) {
1287 ret = -EBADF;
1288 goto out;
1289 }
1290
1291 inode = file_inode(f.file);
1292 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1293 ret = -EBADF;
1294 goto out_fput;
1295 }
1296 info = MQUEUE_I(inode);
1297
1298 ret = 0;
1299 spin_lock(&info->lock);
1300 if (notification == NULL) {
1301 if (info->notify_owner == task_tgid(current)) {
1302 remove_notification(info);
1303 inode->i_atime = inode->i_ctime = current_time(inode);
1304 }
1305 } else if (info->notify_owner != NULL) {
1306 ret = -EBUSY;
1307 } else {
1308 switch (notification->sigev_notify) {
1309 case SIGEV_NONE:
1310 info->notify.sigev_notify = SIGEV_NONE;
1311 break;
1312 case SIGEV_THREAD:
1313 info->notify_sock = sock;
1314 info->notify_cookie = nc;
1315 sock = NULL;
1316 nc = NULL;
1317 info->notify.sigev_notify = SIGEV_THREAD;
1318 break;
1319 case SIGEV_SIGNAL:
1320 info->notify.sigev_signo = notification->sigev_signo;
1321 info->notify.sigev_value = notification->sigev_value;
1322 info->notify.sigev_notify = SIGEV_SIGNAL;
1323 break;
1324 }
1325
1326 info->notify_owner = get_pid(task_tgid(current));
1327 info->notify_user_ns = get_user_ns(current_user_ns());
1328 inode->i_atime = inode->i_ctime = current_time(inode);
1329 }
1330 spin_unlock(&info->lock);
1331out_fput:
1332 fdput(f);
1333out:
1334 if (sock)
1335 netlink_detachskb(sock, nc);
1336 else if (nc)
1337 dev_kfree_skb(nc);
1338
1339 return ret;
1340}
1341
1342SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1343 const struct sigevent __user *, u_notification)
1344{
1345 struct sigevent n, *p = NULL;
1346 if (u_notification) {
1347 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1348 return -EFAULT;
1349 p = &n;
1350 }
1351 return do_mq_notify(mqdes, p);
1352}
1353
1354static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1355{
1356 struct fd f;
1357 struct inode *inode;
1358 struct mqueue_inode_info *info;
1359
1360 if (new && (new->mq_flags & (~O_NONBLOCK)))
1361 return -EINVAL;
1362
1363 f = fdget(mqdes);
1364 if (!f.file)
1365 return -EBADF;
1366
1367 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1368 fdput(f);
1369 return -EBADF;
1370 }
1371
1372 inode = file_inode(f.file);
1373 info = MQUEUE_I(inode);
1374
1375 spin_lock(&info->lock);
1376
1377 if (old) {
1378 *old = info->attr;
1379 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1380 }
1381 if (new) {
1382 audit_mq_getsetattr(mqdes, new);
1383 spin_lock(&f.file->f_lock);
1384 if (new->mq_flags & O_NONBLOCK)
1385 f.file->f_flags |= O_NONBLOCK;
1386 else
1387 f.file->f_flags &= ~O_NONBLOCK;
1388 spin_unlock(&f.file->f_lock);
1389
1390 inode->i_atime = inode->i_ctime = current_time(inode);
1391 }
1392
1393 spin_unlock(&info->lock);
1394 fdput(f);
1395 return 0;
1396}
1397
1398SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1399 const struct mq_attr __user *, u_mqstat,
1400 struct mq_attr __user *, u_omqstat)
1401{
1402 int ret;
1403 struct mq_attr mqstat, omqstat;
1404 struct mq_attr *new = NULL, *old = NULL;
1405
1406 if (u_mqstat) {
1407 new = &mqstat;
1408 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1409 return -EFAULT;
1410 }
1411 if (u_omqstat)
1412 old = &omqstat;
1413
1414 ret = do_mq_getsetattr(mqdes, new, old);
1415 if (ret || !old)
1416 return ret;
1417
1418 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1419 return -EFAULT;
1420 return 0;
1421}
1422
1423#ifdef CONFIG_COMPAT
1424
1425struct compat_mq_attr {
1426 compat_long_t mq_flags; /* message queue flags */
1427 compat_long_t mq_maxmsg; /* maximum number of messages */
1428 compat_long_t mq_msgsize; /* maximum message size */
1429 compat_long_t mq_curmsgs; /* number of messages currently queued */
1430 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1431};
1432
1433static inline int get_compat_mq_attr(struct mq_attr *attr,
1434 const struct compat_mq_attr __user *uattr)
1435{
1436 struct compat_mq_attr v;
1437
1438 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1439 return -EFAULT;
1440
1441 memset(attr, 0, sizeof(*attr));
1442 attr->mq_flags = v.mq_flags;
1443 attr->mq_maxmsg = v.mq_maxmsg;
1444 attr->mq_msgsize = v.mq_msgsize;
1445 attr->mq_curmsgs = v.mq_curmsgs;
1446 return 0;
1447}
1448
1449static inline int put_compat_mq_attr(const struct mq_attr *attr,
1450 struct compat_mq_attr __user *uattr)
1451{
1452 struct compat_mq_attr v;
1453
1454 memset(&v, 0, sizeof(v));
1455 v.mq_flags = attr->mq_flags;
1456 v.mq_maxmsg = attr->mq_maxmsg;
1457 v.mq_msgsize = attr->mq_msgsize;
1458 v.mq_curmsgs = attr->mq_curmsgs;
1459 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1460 return -EFAULT;
1461 return 0;
1462}
1463
1464COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1465 int, oflag, compat_mode_t, mode,
1466 struct compat_mq_attr __user *, u_attr)
1467{
1468 struct mq_attr attr, *p = NULL;
1469 if (u_attr && oflag & O_CREAT) {
1470 p = &attr;
1471 if (get_compat_mq_attr(&attr, u_attr))
1472 return -EFAULT;
1473 }
1474 return do_mq_open(u_name, oflag, mode, p);
1475}
1476
1477static int compat_prepare_timeout(const struct compat_timespec __user *p,
1478 struct timespec *ts)
1479{
1480 if (compat_get_timespec(ts, p))
1481 return -EFAULT;
1482 if (!timespec_valid(ts))
1483 return -EINVAL;
1484 return 0;
1485}
1486
1487COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1488 const char __user *, u_msg_ptr,
1489 compat_size_t, msg_len, unsigned int, msg_prio,
1490 const struct compat_timespec __user *, u_abs_timeout)
1491{
1492 struct timespec ts, *p = NULL;
1493 if (u_abs_timeout) {
1494 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1495 if (res)
1496 return res;
1497 p = &ts;
1498 }
1499 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1500}
1501
1502COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1503 char __user *, u_msg_ptr,
1504 compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1505 const struct compat_timespec __user *, u_abs_timeout)
1506{
1507 struct timespec ts, *p = NULL;
1508 if (u_abs_timeout) {
1509 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1510 if (res)
1511 return res;
1512 p = &ts;
1513 }
1514 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1515}
1516
1517COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1518 const struct compat_sigevent __user *, u_notification)
1519{
1520 struct sigevent n, *p = NULL;
1521 if (u_notification) {
1522 if (get_compat_sigevent(&n, u_notification))
1523 return -EFAULT;
1524 if (n.sigev_notify == SIGEV_THREAD)
1525 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1526 p = &n;
1527 }
1528 return do_mq_notify(mqdes, p);
1529}
1530
1531COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1532 const struct compat_mq_attr __user *, u_mqstat,
1533 struct compat_mq_attr __user *, u_omqstat)
1534{
1535 int ret;
1536 struct mq_attr mqstat, omqstat;
1537 struct mq_attr *new = NULL, *old = NULL;
1538
1539 if (u_mqstat) {
1540 new = &mqstat;
1541 if (get_compat_mq_attr(new, u_mqstat))
1542 return -EFAULT;
1543 }
1544 if (u_omqstat)
1545 old = &omqstat;
1546
1547 ret = do_mq_getsetattr(mqdes, new, old);
1548 if (ret || !old)
1549 return ret;
1550
1551 if (put_compat_mq_attr(old, u_omqstat))
1552 return -EFAULT;
1553 return 0;
1554}
1555#endif
1556
1557static const struct inode_operations mqueue_dir_inode_operations = {
1558 .lookup = simple_lookup,
1559 .create = mqueue_create,
1560 .unlink = mqueue_unlink,
1561};
1562
1563static const struct file_operations mqueue_file_operations = {
1564 .flush = mqueue_flush_file,
1565 .poll = mqueue_poll_file,
1566 .read = mqueue_read_file,
1567 .llseek = default_llseek,
1568};
1569
1570static const struct super_operations mqueue_super_ops = {
1571 .alloc_inode = mqueue_alloc_inode,
1572 .destroy_inode = mqueue_destroy_inode,
1573 .evict_inode = mqueue_evict_inode,
1574 .statfs = simple_statfs,
1575};
1576
1577static struct file_system_type mqueue_fs_type = {
1578 .name = "mqueue",
1579 .mount = mqueue_mount,
1580 .kill_sb = kill_litter_super,
1581 .fs_flags = FS_USERNS_MOUNT,
1582};
1583
1584int mq_init_ns(struct ipc_namespace *ns)
1585{
1586 ns->mq_queues_count = 0;
1587 ns->mq_queues_max = DFLT_QUEUESMAX;
1588 ns->mq_msg_max = DFLT_MSGMAX;
1589 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1590 ns->mq_msg_default = DFLT_MSG;
1591 ns->mq_msgsize_default = DFLT_MSGSIZE;
1592
1593 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1594 if (IS_ERR(ns->mq_mnt)) {
1595 int err = PTR_ERR(ns->mq_mnt);
1596 ns->mq_mnt = NULL;
1597 return err;
1598 }
1599 return 0;
1600}
1601
1602void mq_clear_sbinfo(struct ipc_namespace *ns)
1603{
1604 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1605}
1606
1607void mq_put_mnt(struct ipc_namespace *ns)
1608{
1609 kern_unmount(ns->mq_mnt);
1610}
1611
1612static int __init init_mqueue_fs(void)
1613{
1614 int error;
1615
1616 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1617 sizeof(struct mqueue_inode_info), 0,
1618 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1619 if (mqueue_inode_cachep == NULL)
1620 return -ENOMEM;
1621
1622 /* ignore failures - they are not fatal */
1623 mq_sysctl_table = mq_register_sysctl_table();
1624
1625 error = register_filesystem(&mqueue_fs_type);
1626 if (error)
1627 goto out_sysctl;
1628
1629 spin_lock_init(&mq_lock);
1630
1631 error = mq_init_ns(&init_ipc_ns);
1632 if (error)
1633 goto out_filesystem;
1634
1635 return 0;
1636
1637out_filesystem:
1638 unregister_filesystem(&mqueue_fs_type);
1639out_sysctl:
1640 if (mq_sysctl_table)
1641 unregister_sysctl_table(mq_sysctl_table);
1642 kmem_cache_destroy(mqueue_inode_cachep);
1643 return error;
1644}
1645
1646device_initcall(init_mqueue_fs);