?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
signal: Ensure generic siginfos the kernel sends have all bits initialized
[cris-mirror.git] / ipc / mqueue.c
blob17bc8b874d925b5ff0fbe041567cce99fba04e73
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 #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
55 struct posix_msg_tree_node {
56 struct rb_node rb_node;
57 struct list_head msg_list;
58 int priority;
59 };
60
61 struct 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
68 struct 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
90 static const struct inode_operations mqueue_dir_inode_operations;
91 static const struct file_operations mqueue_file_operations;
92 static const struct super_operations mqueue_super_ops;
93 static void remove_notification(struct mqueue_inode_info *info);
94
95 static struct kmem_cache *mqueue_inode_cachep;
96
97 static struct ctl_table_header *mq_sysctl_table;
98
99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100 {
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
102 }
103
104 /*
105 * This routine should be called with the mq_lock held.
106 */
107 static 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
112 static 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 */
123 static 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);
152 insert_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
159 static 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
165 try_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
215 static 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;
302 out_inode:
303 iput(inode);
304 err:
305 return ERR_PTR(ret);
306 }
307
308 static 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
329 static 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 & SB_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
343 static 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
350 static 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
360 static 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
366 static void mqueue_destroy_inode(struct inode *inode)
367 {
368 call_rcu(&inode->i_rcu, mqueue_i_callback);
369 }
370
371 static 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
419 static 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;
457 out_unlock:
458 spin_unlock(&mq_lock);
459 if (ipc_ns)
460 put_ipc_ns(ipc_ns);
461 return error;
462 }
463
464 static 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 */
482 static 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
510 static 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
522 static 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 */
541 static 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 */
562 static 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);
597 out_unlock:
598 spin_unlock(&info->lock);
599 out:
600 return retval;
601 }
602
603 /*
604 * Returns waiting task that should be serviced first or NULL if none exists
605 */
606 static 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
618 static 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 */
626 static 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 clear_siginfo(&sig_i);
643 sig_i.si_signo = info->notify.sigev_signo;
644 sig_i.si_errno = 0;
645 sig_i.si_code = SI_MESGQ;
646 sig_i.si_value = info->notify.sigev_value;
647 /* map current pid/uid into info->owner's namespaces */
648 rcu_read_lock();
649 sig_i.si_pid = task_tgid_nr_ns(current,
650 ns_of_pid(info->notify_owner));
651 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
652 rcu_read_unlock();
653
654 kill_pid_info(info->notify.sigev_signo,
655 &sig_i, info->notify_owner);
656 break;
657 case SIGEV_THREAD:
658 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
659 netlink_sendskb(info->notify_sock, info->notify_cookie);
660 break;
661 }
662 /* after notification unregisters process */
663 put_pid(info->notify_owner);
664 put_user_ns(info->notify_user_ns);
665 info->notify_owner = NULL;
666 info->notify_user_ns = NULL;
667 }
668 wake_up(&info->wait_q);
669 }
670
671 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
672 struct timespec64 *ts)
673 {
674 if (get_timespec64(ts, u_abs_timeout))
675 return -EFAULT;
676 if (!timespec64_valid(ts))
677 return -EINVAL;
678 return 0;
679 }
680
681 static void remove_notification(struct mqueue_inode_info *info)
682 {
683 if (info->notify_owner != NULL &&
684 info->notify.sigev_notify == SIGEV_THREAD) {
685 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
686 netlink_sendskb(info->notify_sock, info->notify_cookie);
687 }
688 put_pid(info->notify_owner);
689 put_user_ns(info->notify_user_ns);
690 info->notify_owner = NULL;
691 info->notify_user_ns = NULL;
692 }
693
694 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
695 {
696 int mq_treesize;
697 unsigned long total_size;
698
699 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
700 return -EINVAL;
701 if (capable(CAP_SYS_RESOURCE)) {
702 if (attr->mq_maxmsg > HARD_MSGMAX ||
703 attr->mq_msgsize > HARD_MSGSIZEMAX)
704 return -EINVAL;
705 } else {
706 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
707 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
708 return -EINVAL;
709 }
710 /* check for overflow */
711 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
712 return -EOVERFLOW;
713 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
714 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
715 sizeof(struct posix_msg_tree_node);
716 total_size = attr->mq_maxmsg * attr->mq_msgsize;
717 if (total_size + mq_treesize < total_size)
718 return -EOVERFLOW;
719 return 0;
720 }
721
722 /*
723 * Invoked when creating a new queue via sys_mq_open
724 */
725 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
726 struct path *path, int oflag, umode_t mode,
727 struct mq_attr *attr)
728 {
729 const struct cred *cred = current_cred();
730 int ret;
731
732 if (attr) {
733 ret = mq_attr_ok(ipc_ns, attr);
734 if (ret)
735 return ERR_PTR(ret);
736 /* store for use during create */
737 path->dentry->d_fsdata = attr;
738 } else {
739 struct mq_attr def_attr;
740
741 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
742 ipc_ns->mq_msg_default);
743 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
744 ipc_ns->mq_msgsize_default);
745 ret = mq_attr_ok(ipc_ns, &def_attr);
746 if (ret)
747 return ERR_PTR(ret);
748 }
749
750 mode &= ~current_umask();
751 ret = vfs_create(dir, path->dentry, mode, true);
752 path->dentry->d_fsdata = NULL;
753 if (ret)
754 return ERR_PTR(ret);
755 return dentry_open(path, oflag, cred);
756 }
757
758 /* Opens existing queue */
759 static struct file *do_open(struct path *path, int oflag)
760 {
761 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
762 MAY_READ | MAY_WRITE };
763 int acc;
764 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
765 return ERR_PTR(-EINVAL);
766 acc = oflag2acc[oflag & O_ACCMODE];
767 if (inode_permission(d_inode(path->dentry), acc))
768 return ERR_PTR(-EACCES);
769 return dentry_open(path, oflag, current_cred());
770 }
771
772 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
773 struct mq_attr *attr)
774 {
775 struct path path;
776 struct file *filp;
777 struct filename *name;
778 int fd, error;
779 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
780 struct vfsmount *mnt = ipc_ns->mq_mnt;
781 struct dentry *root = mnt->mnt_root;
782 int ro;
783
784 audit_mq_open(oflag, mode, attr);
785
786 if (IS_ERR(name = getname(u_name)))
787 return PTR_ERR(name);
788
789 fd = get_unused_fd_flags(O_CLOEXEC);
790 if (fd < 0)
791 goto out_putname;
792
793 ro = mnt_want_write(mnt); /* we'll drop it in any case */
794 error = 0;
795 inode_lock(d_inode(root));
796 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
797 if (IS_ERR(path.dentry)) {
798 error = PTR_ERR(path.dentry);
799 goto out_putfd;
800 }
801 path.mnt = mntget(mnt);
802
803 if (oflag & O_CREAT) {
804 if (d_really_is_positive(path.dentry)) { /* entry already exists */
805 audit_inode(name, path.dentry, 0);
806 if (oflag & O_EXCL) {
807 error = -EEXIST;
808 goto out;
809 }
810 filp = do_open(&path, oflag);
811 } else {
812 if (ro) {
813 error = ro;
814 goto out;
815 }
816 audit_inode_parent_hidden(name, root);
817 filp = do_create(ipc_ns, d_inode(root), &path,
818 oflag, mode, attr);
819 }
820 } else {
821 if (d_really_is_negative(path.dentry)) {
822 error = -ENOENT;
823 goto out;
824 }
825 audit_inode(name, path.dentry, 0);
826 filp = do_open(&path, oflag);
827 }
828
829 if (!IS_ERR(filp))
830 fd_install(fd, filp);
831 else
832 error = PTR_ERR(filp);
833 out:
834 path_put(&path);
835 out_putfd:
836 if (error) {
837 put_unused_fd(fd);
838 fd = error;
839 }
840 inode_unlock(d_inode(root));
841 if (!ro)
842 mnt_drop_write(mnt);
843 out_putname:
844 putname(name);
845 return fd;
846 }
847
848 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
849 struct mq_attr __user *, u_attr)
850 {
851 struct mq_attr attr;
852 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
853 return -EFAULT;
854
855 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
856 }
857
858 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
859 {
860 int err;
861 struct filename *name;
862 struct dentry *dentry;
863 struct inode *inode = NULL;
864 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
865 struct vfsmount *mnt = ipc_ns->mq_mnt;
866
867 name = getname(u_name);
868 if (IS_ERR(name))
869 return PTR_ERR(name);
870
871 audit_inode_parent_hidden(name, mnt->mnt_root);
872 err = mnt_want_write(mnt);
873 if (err)
874 goto out_name;
875 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
876 dentry = lookup_one_len(name->name, mnt->mnt_root,
877 strlen(name->name));
878 if (IS_ERR(dentry)) {
879 err = PTR_ERR(dentry);
880 goto out_unlock;
881 }
882
883 inode = d_inode(dentry);
884 if (!inode) {
885 err = -ENOENT;
886 } else {
887 ihold(inode);
888 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
889 }
890 dput(dentry);
891
892 out_unlock:
893 inode_unlock(d_inode(mnt->mnt_root));
894 if (inode)
895 iput(inode);
896 mnt_drop_write(mnt);
897 out_name:
898 putname(name);
899
900 return err;
901 }
902
903 /* Pipelined send and receive functions.
904 *
905 * If a receiver finds no waiting message, then it registers itself in the
906 * list of waiting receivers. A sender checks that list before adding the new
907 * message into the message array. If there is a waiting receiver, then it
908 * bypasses the message array and directly hands the message over to the
909 * receiver. The receiver accepts the message and returns without grabbing the
910 * queue spinlock:
911 *
912 * - Set pointer to message.
913 * - Queue the receiver task for later wakeup (without the info->lock).
914 * - Update its state to STATE_READY. Now the receiver can continue.
915 * - Wake up the process after the lock is dropped. Should the process wake up
916 * before this wakeup (due to a timeout or a signal) it will either see
917 * STATE_READY and continue or acquire the lock to check the state again.
918 *
919 * The same algorithm is used for senders.
920 */
921
922 /* pipelined_send() - send a message directly to the task waiting in
923 * sys_mq_timedreceive() (without inserting message into a queue).
924 */
925 static inline void pipelined_send(struct wake_q_head *wake_q,
926 struct mqueue_inode_info *info,
927 struct msg_msg *message,
928 struct ext_wait_queue *receiver)
929 {
930 receiver->msg = message;
931 list_del(&receiver->list);
932 wake_q_add(wake_q, receiver->task);
933 /*
934 * Rely on the implicit cmpxchg barrier from wake_q_add such
935 * that we can ensure that updating receiver->state is the last
936 * write operation: As once set, the receiver can continue,
937 * and if we don't have the reference count from the wake_q,
938 * yet, at that point we can later have a use-after-free
939 * condition and bogus wakeup.
940 */
941 receiver->state = STATE_READY;
942 }
943
944 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
945 * gets its message and put to the queue (we have one free place for sure). */
946 static inline void pipelined_receive(struct wake_q_head *wake_q,
947 struct mqueue_inode_info *info)
948 {
949 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
950
951 if (!sender) {
952 /* for poll */
953 wake_up_interruptible(&info->wait_q);
954 return;
955 }
956 if (msg_insert(sender->msg, info))
957 return;
958
959 list_del(&sender->list);
960 wake_q_add(wake_q, sender->task);
961 sender->state = STATE_READY;
962 }
963
964 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
965 size_t msg_len, unsigned int msg_prio,
966 struct timespec64 *ts)
967 {
968 struct fd f;
969 struct inode *inode;
970 struct ext_wait_queue wait;
971 struct ext_wait_queue *receiver;
972 struct msg_msg *msg_ptr;
973 struct mqueue_inode_info *info;
974 ktime_t expires, *timeout = NULL;
975 struct posix_msg_tree_node *new_leaf = NULL;
976 int ret = 0;
977 DEFINE_WAKE_Q(wake_q);
978
979 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
980 return -EINVAL;
981
982 if (ts) {
983 expires = timespec64_to_ktime(*ts);
984 timeout = &expires;
985 }
986
987 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
988
989 f = fdget(mqdes);
990 if (unlikely(!f.file)) {
991 ret = -EBADF;
992 goto out;
993 }
994
995 inode = file_inode(f.file);
996 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
997 ret = -EBADF;
998 goto out_fput;
999 }
1000 info = MQUEUE_I(inode);
1001 audit_file(f.file);
1002
1003 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1004 ret = -EBADF;
1005 goto out_fput;
1006 }
1007
1008 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1009 ret = -EMSGSIZE;
1010 goto out_fput;
1011 }
1012
1013 /* First try to allocate memory, before doing anything with
1014 * existing queues. */
1015 msg_ptr = load_msg(u_msg_ptr, msg_len);
1016 if (IS_ERR(msg_ptr)) {
1017 ret = PTR_ERR(msg_ptr);
1018 goto out_fput;
1019 }
1020 msg_ptr->m_ts = msg_len;
1021 msg_ptr->m_type = msg_prio;
1022
1023 /*
1024 * msg_insert really wants us to have a valid, spare node struct so
1025 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1026 * fall back to that if necessary.
1027 */
1028 if (!info->node_cache)
1029 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1030
1031 spin_lock(&info->lock);
1032
1033 if (!info->node_cache && new_leaf) {
1034 /* Save our speculative allocation into the cache */
1035 INIT_LIST_HEAD(&new_leaf->msg_list);
1036 info->node_cache = new_leaf;
1037 new_leaf = NULL;
1038 } else {
1039 kfree(new_leaf);
1040 }
1041
1042 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1043 if (f.file->f_flags & O_NONBLOCK) {
1044 ret = -EAGAIN;
1045 } else {
1046 wait.task = current;
1047 wait.msg = (void *) msg_ptr;
1048 wait.state = STATE_NONE;
1049 ret = wq_sleep(info, SEND, timeout, &wait);
1050 /*
1051 * wq_sleep must be called with info->lock held, and
1052 * returns with the lock released
1053 */
1054 goto out_free;
1055 }
1056 } else {
1057 receiver = wq_get_first_waiter(info, RECV);
1058 if (receiver) {
1059 pipelined_send(&wake_q, info, msg_ptr, receiver);
1060 } else {
1061 /* adds message to the queue */
1062 ret = msg_insert(msg_ptr, info);
1063 if (ret)
1064 goto out_unlock;
1065 __do_notify(info);
1066 }
1067 inode->i_atime = inode->i_mtime = inode->i_ctime =
1068 current_time(inode);
1069 }
1070 out_unlock:
1071 spin_unlock(&info->lock);
1072 wake_up_q(&wake_q);
1073 out_free:
1074 if (ret)
1075 free_msg(msg_ptr);
1076 out_fput:
1077 fdput(f);
1078 out:
1079 return ret;
1080 }
1081
1082 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1083 size_t msg_len, unsigned int __user *u_msg_prio,
1084 struct timespec64 *ts)
1085 {
1086 ssize_t ret;
1087 struct msg_msg *msg_ptr;
1088 struct fd f;
1089 struct inode *inode;
1090 struct mqueue_inode_info *info;
1091 struct ext_wait_queue wait;
1092 ktime_t expires, *timeout = NULL;
1093 struct posix_msg_tree_node *new_leaf = NULL;
1094
1095 if (ts) {
1096 expires = timespec64_to_ktime(*ts);
1097 timeout = &expires;
1098 }
1099
1100 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1101
1102 f = fdget(mqdes);
1103 if (unlikely(!f.file)) {
1104 ret = -EBADF;
1105 goto out;
1106 }
1107
1108 inode = file_inode(f.file);
1109 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1110 ret = -EBADF;
1111 goto out_fput;
1112 }
1113 info = MQUEUE_I(inode);
1114 audit_file(f.file);
1115
1116 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1117 ret = -EBADF;
1118 goto out_fput;
1119 }
1120
1121 /* checks if buffer is big enough */
1122 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1123 ret = -EMSGSIZE;
1124 goto out_fput;
1125 }
1126
1127 /*
1128 * msg_insert really wants us to have a valid, spare node struct so
1129 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1130 * fall back to that if necessary.
1131 */
1132 if (!info->node_cache)
1133 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1134
1135 spin_lock(&info->lock);
1136
1137 if (!info->node_cache && new_leaf) {
1138 /* Save our speculative allocation into the cache */
1139 INIT_LIST_HEAD(&new_leaf->msg_list);
1140 info->node_cache = new_leaf;
1141 } else {
1142 kfree(new_leaf);
1143 }
1144
1145 if (info->attr.mq_curmsgs == 0) {
1146 if (f.file->f_flags & O_NONBLOCK) {
1147 spin_unlock(&info->lock);
1148 ret = -EAGAIN;
1149 } else {
1150 wait.task = current;
1151 wait.state = STATE_NONE;
1152 ret = wq_sleep(info, RECV, timeout, &wait);
1153 msg_ptr = wait.msg;
1154 }
1155 } else {
1156 DEFINE_WAKE_Q(wake_q);
1157
1158 msg_ptr = msg_get(info);
1159
1160 inode->i_atime = inode->i_mtime = inode->i_ctime =
1161 current_time(inode);
1162
1163 /* There is now free space in queue. */
1164 pipelined_receive(&wake_q, info);
1165 spin_unlock(&info->lock);
1166 wake_up_q(&wake_q);
1167 ret = 0;
1168 }
1169 if (ret == 0) {
1170 ret = msg_ptr->m_ts;
1171
1172 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1173 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1174 ret = -EFAULT;
1175 }
1176 free_msg(msg_ptr);
1177 }
1178 out_fput:
1179 fdput(f);
1180 out:
1181 return ret;
1182 }
1183
1184 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1185 size_t, msg_len, unsigned int, msg_prio,
1186 const struct timespec __user *, u_abs_timeout)
1187 {
1188 struct timespec64 ts, *p = NULL;
1189 if (u_abs_timeout) {
1190 int res = prepare_timeout(u_abs_timeout, &ts);
1191 if (res)
1192 return res;
1193 p = &ts;
1194 }
1195 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1196 }
1197
1198 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1199 size_t, msg_len, unsigned int __user *, u_msg_prio,
1200 const struct timespec __user *, u_abs_timeout)
1201 {
1202 struct timespec64 ts, *p = NULL;
1203 if (u_abs_timeout) {
1204 int res = prepare_timeout(u_abs_timeout, &ts);
1205 if (res)
1206 return res;
1207 p = &ts;
1208 }
1209 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1210 }
1211
1212 /*
1213 * Notes: the case when user wants us to deregister (with NULL as pointer)
1214 * and he isn't currently owner of notification, will be silently discarded.
1215 * It isn't explicitly defined in the POSIX.
1216 */
1217 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1218 {
1219 int ret;
1220 struct fd f;
1221 struct sock *sock;
1222 struct inode *inode;
1223 struct mqueue_inode_info *info;
1224 struct sk_buff *nc;
1225
1226 audit_mq_notify(mqdes, notification);
1227
1228 nc = NULL;
1229 sock = NULL;
1230 if (notification != NULL) {
1231 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1232 notification->sigev_notify != SIGEV_SIGNAL &&
1233 notification->sigev_notify != SIGEV_THREAD))
1234 return -EINVAL;
1235 if (notification->sigev_notify == SIGEV_SIGNAL &&
1236 !valid_signal(notification->sigev_signo)) {
1237 return -EINVAL;
1238 }
1239 if (notification->sigev_notify == SIGEV_THREAD) {
1240 long timeo;
1241
1242 /* create the notify skb */
1243 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1244 if (!nc) {
1245 ret = -ENOMEM;
1246 goto out;
1247 }
1248 if (copy_from_user(nc->data,
1249 notification->sigev_value.sival_ptr,
1250 NOTIFY_COOKIE_LEN)) {
1251 ret = -EFAULT;
1252 goto out;
1253 }
1254
1255 /* TODO: add a header? */
1256 skb_put(nc, NOTIFY_COOKIE_LEN);
1257 /* and attach it to the socket */
1258 retry:
1259 f = fdget(notification->sigev_signo);
1260 if (!f.file) {
1261 ret = -EBADF;
1262 goto out;
1263 }
1264 sock = netlink_getsockbyfilp(f.file);
1265 fdput(f);
1266 if (IS_ERR(sock)) {
1267 ret = PTR_ERR(sock);
1268 sock = NULL;
1269 goto out;
1270 }
1271
1272 timeo = MAX_SCHEDULE_TIMEOUT;
1273 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1274 if (ret == 1) {
1275 sock = NULL;
1276 goto retry;
1277 }
1278 if (ret) {
1279 sock = NULL;
1280 nc = NULL;
1281 goto out;
1282 }
1283 }
1284 }
1285
1286 f = fdget(mqdes);
1287 if (!f.file) {
1288 ret = -EBADF;
1289 goto out;
1290 }
1291
1292 inode = file_inode(f.file);
1293 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1294 ret = -EBADF;
1295 goto out_fput;
1296 }
1297 info = MQUEUE_I(inode);
1298
1299 ret = 0;
1300 spin_lock(&info->lock);
1301 if (notification == NULL) {
1302 if (info->notify_owner == task_tgid(current)) {
1303 remove_notification(info);
1304 inode->i_atime = inode->i_ctime = current_time(inode);
1305 }
1306 } else if (info->notify_owner != NULL) {
1307 ret = -EBUSY;
1308 } else {
1309 switch (notification->sigev_notify) {
1310 case SIGEV_NONE:
1311 info->notify.sigev_notify = SIGEV_NONE;
1312 break;
1313 case SIGEV_THREAD:
1314 info->notify_sock = sock;
1315 info->notify_cookie = nc;
1316 sock = NULL;
1317 nc = NULL;
1318 info->notify.sigev_notify = SIGEV_THREAD;
1319 break;
1320 case SIGEV_SIGNAL:
1321 info->notify.sigev_signo = notification->sigev_signo;
1322 info->notify.sigev_value = notification->sigev_value;
1323 info->notify.sigev_notify = SIGEV_SIGNAL;
1324 break;
1325 }
1326
1327 info->notify_owner = get_pid(task_tgid(current));
1328 info->notify_user_ns = get_user_ns(current_user_ns());
1329 inode->i_atime = inode->i_ctime = current_time(inode);
1330 }
1331 spin_unlock(&info->lock);
1332 out_fput:
1333 fdput(f);
1334 out:
1335 if (sock)
1336 netlink_detachskb(sock, nc);
1337 else if (nc)
1338 dev_kfree_skb(nc);
1339
1340 return ret;
1341 }
1342
1343 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1344 const struct sigevent __user *, u_notification)
1345 {
1346 struct sigevent n, *p = NULL;
1347 if (u_notification) {
1348 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1349 return -EFAULT;
1350 p = &n;
1351 }
1352 return do_mq_notify(mqdes, p);
1353 }
1354
1355 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1356 {
1357 struct fd f;
1358 struct inode *inode;
1359 struct mqueue_inode_info *info;
1360
1361 if (new && (new->mq_flags & (~O_NONBLOCK)))
1362 return -EINVAL;
1363
1364 f = fdget(mqdes);
1365 if (!f.file)
1366 return -EBADF;
1367
1368 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1369 fdput(f);
1370 return -EBADF;
1371 }
1372
1373 inode = file_inode(f.file);
1374 info = MQUEUE_I(inode);
1375
1376 spin_lock(&info->lock);
1377
1378 if (old) {
1379 *old = info->attr;
1380 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1381 }
1382 if (new) {
1383 audit_mq_getsetattr(mqdes, new);
1384 spin_lock(&f.file->f_lock);
1385 if (new->mq_flags & O_NONBLOCK)
1386 f.file->f_flags |= O_NONBLOCK;
1387 else
1388 f.file->f_flags &= ~O_NONBLOCK;
1389 spin_unlock(&f.file->f_lock);
1390
1391 inode->i_atime = inode->i_ctime = current_time(inode);
1392 }
1393
1394 spin_unlock(&info->lock);
1395 fdput(f);
1396 return 0;
1397 }
1398
1399 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1400 const struct mq_attr __user *, u_mqstat,
1401 struct mq_attr __user *, u_omqstat)
1402 {
1403 int ret;
1404 struct mq_attr mqstat, omqstat;
1405 struct mq_attr *new = NULL, *old = NULL;
1406
1407 if (u_mqstat) {
1408 new = &mqstat;
1409 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1410 return -EFAULT;
1411 }
1412 if (u_omqstat)
1413 old = &omqstat;
1414
1415 ret = do_mq_getsetattr(mqdes, new, old);
1416 if (ret || !old)
1417 return ret;
1418
1419 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1420 return -EFAULT;
1421 return 0;
1422 }
1423
1424 #ifdef CONFIG_COMPAT
1425
1426 struct compat_mq_attr {
1427 compat_long_t mq_flags; /* message queue flags */
1428 compat_long_t mq_maxmsg; /* maximum number of messages */
1429 compat_long_t mq_msgsize; /* maximum message size */
1430 compat_long_t mq_curmsgs; /* number of messages currently queued */
1431 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1432 };
1433
1434 static inline int get_compat_mq_attr(struct mq_attr *attr,
1435 const struct compat_mq_attr __user *uattr)
1436 {
1437 struct compat_mq_attr v;
1438
1439 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1440 return -EFAULT;
1441
1442 memset(attr, 0, sizeof(*attr));
1443 attr->mq_flags = v.mq_flags;
1444 attr->mq_maxmsg = v.mq_maxmsg;
1445 attr->mq_msgsize = v.mq_msgsize;
1446 attr->mq_curmsgs = v.mq_curmsgs;
1447 return 0;
1448 }
1449
1450 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1451 struct compat_mq_attr __user *uattr)
1452 {
1453 struct compat_mq_attr v;
1454
1455 memset(&v, 0, sizeof(v));
1456 v.mq_flags = attr->mq_flags;
1457 v.mq_maxmsg = attr->mq_maxmsg;
1458 v.mq_msgsize = attr->mq_msgsize;
1459 v.mq_curmsgs = attr->mq_curmsgs;
1460 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1461 return -EFAULT;
1462 return 0;
1463 }
1464
1465 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1466 int, oflag, compat_mode_t, mode,
1467 struct compat_mq_attr __user *, u_attr)
1468 {
1469 struct mq_attr attr, *p = NULL;
1470 if (u_attr && oflag & O_CREAT) {
1471 p = &attr;
1472 if (get_compat_mq_attr(&attr, u_attr))
1473 return -EFAULT;
1474 }
1475 return do_mq_open(u_name, oflag, mode, p);
1476 }
1477
1478 static int compat_prepare_timeout(const struct compat_timespec __user *p,
1479 struct timespec64 *ts)
1480 {
1481 if (compat_get_timespec64(ts, p))
1482 return -EFAULT;
1483 if (!timespec64_valid(ts))
1484 return -EINVAL;
1485 return 0;
1486 }
1487
1488 COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1489 const char __user *, u_msg_ptr,
1490 compat_size_t, msg_len, unsigned int, msg_prio,
1491 const struct compat_timespec __user *, u_abs_timeout)
1492 {
1493 struct timespec64 ts, *p = NULL;
1494 if (u_abs_timeout) {
1495 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1496 if (res)
1497 return res;
1498 p = &ts;
1499 }
1500 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1501 }
1502
1503 COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1504 char __user *, u_msg_ptr,
1505 compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1506 const struct compat_timespec __user *, u_abs_timeout)
1507 {
1508 struct timespec64 ts, *p = NULL;
1509 if (u_abs_timeout) {
1510 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1511 if (res)
1512 return res;
1513 p = &ts;
1514 }
1515 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1516 }
1517
1518 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1519 const struct compat_sigevent __user *, u_notification)
1520 {
1521 struct sigevent n, *p = NULL;
1522 if (u_notification) {
1523 if (get_compat_sigevent(&n, u_notification))
1524 return -EFAULT;
1525 if (n.sigev_notify == SIGEV_THREAD)
1526 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1527 p = &n;
1528 }
1529 return do_mq_notify(mqdes, p);
1530 }
1531
1532 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1533 const struct compat_mq_attr __user *, u_mqstat,
1534 struct compat_mq_attr __user *, u_omqstat)
1535 {
1536 int ret;
1537 struct mq_attr mqstat, omqstat;
1538 struct mq_attr *new = NULL, *old = NULL;
1539
1540 if (u_mqstat) {
1541 new = &mqstat;
1542 if (get_compat_mq_attr(new, u_mqstat))
1543 return -EFAULT;
1544 }
1545 if (u_omqstat)
1546 old = &omqstat;
1547
1548 ret = do_mq_getsetattr(mqdes, new, old);
1549 if (ret || !old)
1550 return ret;
1551
1552 if (put_compat_mq_attr(old, u_omqstat))
1553 return -EFAULT;
1554 return 0;
1555 }
1556 #endif
1557
1558 static const struct inode_operations mqueue_dir_inode_operations = {
1559 .lookup = simple_lookup,
1560 .create = mqueue_create,
1561 .unlink = mqueue_unlink,
1562 };
1563
1564 static const struct file_operations mqueue_file_operations = {
1565 .flush = mqueue_flush_file,
1566 .poll = mqueue_poll_file,
1567 .read = mqueue_read_file,
1568 .llseek = default_llseek,
1569 };
1570
1571 static const struct super_operations mqueue_super_ops = {
1572 .alloc_inode = mqueue_alloc_inode,
1573 .destroy_inode = mqueue_destroy_inode,
1574 .evict_inode = mqueue_evict_inode,
1575 .statfs = simple_statfs,
1576 };
1577
1578 static struct file_system_type mqueue_fs_type = {
1579 .name = "mqueue",
1580 .mount = mqueue_mount,
1581 .kill_sb = kill_litter_super,
1582 .fs_flags = FS_USERNS_MOUNT,
1583 };
1584
1585 int mq_init_ns(struct ipc_namespace *ns)
1586 {
1587 ns->mq_queues_count = 0;
1588 ns->mq_queues_max = DFLT_QUEUESMAX;
1589 ns->mq_msg_max = DFLT_MSGMAX;
1590 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1591 ns->mq_msg_default = DFLT_MSG;
1592 ns->mq_msgsize_default = DFLT_MSGSIZE;
1593
1594 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1595 if (IS_ERR(ns->mq_mnt)) {
1596 int err = PTR_ERR(ns->mq_mnt);
1597 ns->mq_mnt = NULL;
1598 return err;
1599 }
1600 return 0;
1601 }
1602
1603 void mq_clear_sbinfo(struct ipc_namespace *ns)
1604 {
1605 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1606 }
1607
1608 void mq_put_mnt(struct ipc_namespace *ns)
1609 {
1610 kern_unmount(ns->mq_mnt);
1611 }
1612
1613 static int __init init_mqueue_fs(void)
1614 {
1615 int error;
1616
1617 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1618 sizeof(struct mqueue_inode_info), 0,
1619 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1620 if (mqueue_inode_cachep == NULL)
1621 return -ENOMEM;
1622
1623 /* ignore failures - they are not fatal */
1624 mq_sysctl_table = mq_register_sysctl_table();
1625
1626 error = register_filesystem(&mqueue_fs_type);
1627 if (error)
1628 goto out_sysctl;
1629
1630 spin_lock_init(&mq_lock);
1631
1632 error = mq_init_ns(&init_ipc_ns);
1633 if (error)
1634 goto out_filesystem;
1635
1636 return 0;
1637
1638 out_filesystem:
1639 unregister_filesystem(&mqueue_fs_type);
1640 out_sysctl:
1641 if (mq_sysctl_table)
1642 unregister_sysctl_table(mq_sysctl_table);
1643 kmem_cache_destroy(mqueue_inode_cachep);
1644 return error;
1645 }
1646
1647 device_initcall(init_mqueue_fs);
CRIS linux kernel tree